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.1000962 | Genome-Wide Copy Number Variation in Epilepsy: Novel Susceptibility Loci in Idiopathic Generalized and Focal Epilepsies | Epilepsy is one of the most common neurological disorders in humans with a prevalence of 1% and a lifetime incidence of 3%. Several genes have been identified in rare autosomal dominant and severe sporadic forms of epilepsy, but the genetic cause is unknown in the vast majority of cases. Copy number variants (CNVs) are known to play an important role in the genetic etiology of many neurodevelopmental disorders, including intellectual disability (ID), autism, and schizophrenia. Genome-wide studies of copy number variation in epilepsy have not been performed. We have applied whole-genome oligonucleotide array comparative genomic hybridization to a cohort of 517 individuals with various idiopathic, non-lesional epilepsies. We detected one or more rare genic CNVs in 8.9% of affected individuals that are not present in 2,493 controls; five individuals had two rare CNVs. We identified CNVs in genes previously implicated in other neurodevelopmental disorders, including two deletions in AUTS2 and one deletion in CNTNAP2. Therefore, our findings indicate that rare CNVs are likely to contribute to a broad range of generalized and focal epilepsies. In addition, we find that 2.9% of patients carry deletions at 15q11.2, 15q13.3, or 16p13.11, genomic hotspots previously associated with ID, autism, or schizophrenia. In summary, our findings suggest common etiological factors for seemingly diverse diseases such as ID, autism, schizophrenia, and epilepsy.
| Epilepsy, a common neurological disorder characterized by recurrent seizures, affects up to 3% of the population. In some cases, the epilepsy has a clear cause such as an abnormality in the brain or a head injury. However, in many cases there is no obvious cause. Numerous studies have shown that genetic factors are important in these types of epilepsy, but although several epilepsy genes are known, we can still only identify the genetic cause in a very small fraction of cases. In order to identify new genes that contribute to the genetic causes of epilepsy, we searched the human genome for deletions (missing copies) and duplications (extra copies) of genes in ∼500 patients with epilepsy that are not found in control individuals. Using this approach, we identified several large deletions that are important in at least 3% of epilepsy cases. Furthermore, we found new candidate genes, some of which are also thought to play a role in other related disorders such as autism and intellectual disability. These genes are candidates for further studies in patients with epilepsy.
| Epilepsy is one of the most common neurological disorders in humans with a prevalence of ∼1% and a lifetime incidence of up to 3% [1]. The epilepsies present with a broad range of clinical features, and over 50 distinct epilepsy syndromes are now recognized. Particularly in a pediatric setting, a broad range of different epilepsy syndromes can be distinguished. Seizure disorders can roughly be divided into idiopathic or symptomatic epilepsies. While symptomatic epilepsies are due to an identifiable cause such as metabolic disorders, brain trauma or intracranial tumors, idiopathic seizure disorders occur in the absence of identifiable causal factors and are thought to have a strong genetic contribution.
Although it has long been observed that the idiopathic epilepsies have a genetic component, the genetic etiology of only a small fraction of cases can be determined. The role of copy number variants (CNVs) in intellectual disability (ID) [2]–[8], autism [9]–[14] and schizophrenia [15]–[19] has been extensively investigated. It has become increasingly clear that, collectively, rare variants contribute significantly to the etiology of these common diseases–following the rare variant common disease hypothesis. We hypothesize this can be extended to other neurological disorders and that rare CNVs significantly contribute to the genetic etiology of epilepsy.
Recently, in a study targeted to six genomic regions, recurrent microdeletions on chromosome 15q13.3, 16p13.11 and 15q11.2 were identified as important genetic factors predisposing to idiopathic generalized epilepsy (IGE) [20]–[22]. Here, we carry out whole-genome array comparative genomic hybridization (CGH) in a cohort of 517 individuals with mixed types of idiopathic epilepsy in order to discover novel copy number changes associated with epilepsy. We find recurrent microdeletions of 15q13.3, 16p13.11 and 15q11.2 each in ∼1% of affected individuals, confirming previous studies [20]–[22]. In addition to recurrent rearrangements at rearrangement-prone regions, we show that, overall, 8.9% of affected individuals have one or more rare copy number changes involving at least one gene.
We performed genome-wide array CGH to detect copy number changes in 517 patients with mixed types of epilepsy. Of these, 399 have idiopathic generalized epilepsy (IGE), 50 have benign epilepsy with centrotemporal spikes (BECTS) and 68 have other types of idiopathic seizure disorders (Table 1). We used a custom microarray with high-density targeted coverage of 107 regions of the genome flanked by large, highly homologous duplications, termed rearrangement hotspots [23]. In addition, probes were evenly spaced throughout the remainder of the genome with average probe spacing of ∼38 kb. Overall, we find that 46 probands (8.9%) carry one or more rare CNVs not previously reported in the 2493 unrelated controls [24]. The rare CNVs detected in our cohort range in size from 13 kb to 15.9 Mb (average 1.2 Mb; median 600 kb), and the majority (69%) are deletion events.
We first evaluated rearrangement hotspots for copy number changes. We found 20 probands (3.9%) with copy number changes at known rearrangement hotspots including 15q13.3 deletions (n = 5), 16p13.11 deletions (n = 5), 15q11.2 BP1–BP2 deletions (n = 5), 1q21.1 deletions (n = 2), a 16p12.1 deletion (n = 1), a 16p11.2 duplication (n = 1) and a more distal 16p11.2 deletion (n = 1) (Table 2, Figure 1). We also identified four individuals with duplications of 15q11.2 BP1–BP2; because duplications of this region are frequent in the general population, we classified these duplications as polymorphic events. These results confirm our previous studies and emphasize the importance of deletions of 15q13.3, 16p13.11 and 15q11.2 BP1–BP2 as frequent genetic susceptibility factors in epilepsy [20]–[22]. All three regions have also been associated with ID, autism and/or schizophrenia [15], [17], [25]–[32], as have deletions at 1q21.1 [33], [34], two distinct regions of 16p11.2 [10], [14], [35]–[37] and 16p12 [38], which were also detected in our cohort. Deletions of 16p13.11 (5/517 vs 0/2493 controls, p = 0.00014, Fisher's exact test), 15q13.3 (5/517 vs 0/2493, p = 0.00014) and 15q11.2 (5/517 vs. 4/2493, p = 0.010) are significantly enriched in our epilepsy cohort and together account for 2.9% of cases.
We next focused on non-hotspot CNVs that overlap one or more genes and are not present in the control cohort of 2493 individuals [24]. We identified 28 individuals with at least one rare gene-containing deletion or duplication, and five individuals each carry two rare CNVs (Table 2). Fifteen of the events we detected involve a single gene. Two genes were altered in two patients each: AUTS2 deletions were identified in one proband with juvenile myoclonic epilepsy (JME) and one proband with unclassified non-lesional epilepsy with features of atypical benign partial epilepsy (ABPE) [39]. Deletions involving CTYSB (SPECC1) were identified in two probands with IGE. All other single-gene CNVs were seen only once. Seventeen events involved multiple genes, one of which was observed in two different individuals with JME (duplication of 18q11, Table 2).
We found five individuals with two rare CNVs (Figure 2). Two patients with JME and a deletion of 16p13.11 (EMJ071 and EMJ117) each have a second rare deletion. EMJ071 has a large deletion on chromosome 13 that removes the SLITRK6 gene, a member of the SLITRK gene family involved in controlling neurite outgrowth; individual EMJ117 also has a deletion involving the CTYSB gene. Case ND05260 (childhood absence epilepsy, CAE) carries a 647-kb deletion within the GRID2 gene, which encodes a glutamate receptor expressed in the cerebellum, and a 1-Mb duplication of 9q31. Though both are maternally inherited, neither has been reported in controls. Case EPI 51 (idiopathic West syndrome) has two apparently independent duplications of chromosome 5q35, each containing several genes. Finally, we identified one proband with neonatal convulsions (NC) carrying a deletion within the CNTNAP2 gene that spans exons 2–4 as well as a 370-kb deletion of 17p13 involving 7 genes.
DNA from one of more family members was available for analysis in 14 cases. Inheritance, if determined, is shown in Table 2. In twelve cases, we determined that one or both CNVs in the proband were inherited; in three cases the transmitting parent is also affected. In one case (EP007.1), the CNV was not found in the mother, but the father was unavailable. In another case (K047), parents were unavailable, but a brother was found to carry the same CNV suggesting one of the parents carries the same CNV.
In this study, we performed whole-genome array CGH in a series of 517 individuals with a presenting diagnosis of idiopathic epilepsy in order to discover novel copy number changes associated with epilepsy. While our previous studies were targeted to specific genomic regions in probands with IGE [21], [22], here we present data from whole-genome analysis on probands with IGE and extend our analysis to other idiopathic epilepsy syndromes. In total, we identified 46 individuals (8.9%) with 51 rearrangements that may be pathogenic as they were not found in controls or were significantly enriched in our epilepsy cohort.
Rearrangements at several genomic hotspots have been associated with a range of neurocognitive disorders. In our cohort of 517 probands with epilepsy, we find deletions at 15q13.3, 16p13.11 and 15q11.2 in 2.9% of our cases. Interestingly, all of the deletions of 15q13.3 (n = 5) and 4/5 deletions at 16p13.11 and 15q11.2 were in probands with IGE, accounting for 3.3% of the patients with IGE in our cohort confirming our previous findings. While it is possible that deletions of 15q13.3 are also predisposing to non-IGE epilepsy syndromes, we did not find this to be the case in our series (n = 118). Additional large cohorts of patients with focal epilepsy or epileptic encephalopathy will be required to determine whether these deletions also play a significant role in other subtypes of epilepsy.
Deletions of 16p13.11 have previously been associated with intellectual disability +/− congenital anomalies in one study [26]. Three of four probands with 16p13.11 deletions in that series had epilepsy; two further fetal cases had brain abnormalities. The findings in this cohort and one previous study of IGE [20] suggest that deletions of 16p13.11 are more frequent in epilepsy (0.5–1% of cases) than in other phenotypes including ID and autism [26], [27], [32], and may be as frequent as 15q13.3 deletions in individuals with IGE. Deletions and duplications of this region have also been reported in schizophrenia, though the associations have not been statistically significant [16], [29].
Deletions of 15q13.3, detected in five individuals with IGE in our series, have been associated with a wide range of phenotypes including ID, autism, epilepsy and schizophrenia [15], [17], [20]–[22], [25], [28], [30], [31], [40]. The gene within the 15q13.3 region that is most likely responsible for the epilepsy phenotype is CHRNA7, a subunit of the nicotinic acetylcholine receptor. At least two small studies have failed to identify causal point mutations in the CHRNA7 gene in autosomal dominant nocturnal frontal lobe epilepsy [41] and JME [42], but additional studies should be performed to further evaluate affected individuals for mutations. A recent publication identifying atypical rearrangements with exclusive deletions of CHRNA7 further emphasizes the importance of CHRNA7 as the main candidate gene in this region [43].
Compared to the above structural genomic variants, copy number variation at 15q11.2 between breakpoints BP1 and BP2 of the Prader-Willi and Angelman syndrome region is more common in the general population with the BP1–BP2 deletion present in 0.2% of unaffected individuals. Despite this, deletions between BP1 and BP2 have now been reported as enriched in patients with schizophrenia [16], [17], ID [27] and epilepsy [20]. Furthermore, there is evidence that patients with Prader-Willi or Angelman syndrome who have deletions including BP1–BP2 are more severely affected [44]–[46]. In this study, we also find enrichment of deletions at this locus in affected individuals. Together, these studies suggest that deletion of the 15q11.2 BP1–BP2 region confers susceptibility to a wide range of neuropsychiatric conditions, albeit with incomplete penetrance.
Two patients in our series, one each with JME and BECTS, have deletions of 1q21.1, which have been previously associated with a wide range of phenotypes, including intellectually disability and developmental delay [33], [34], schizophrenia [15], [17], [18], congenital heart disease [47], [48] and cataracts [34], [49]. In two large studies of patients who present primarily with cognitive or developmental delay, 5/42 (11.9%) patients also had seizures [33], [34]; 1 of 10 patients with schizophrenia and a 1q21.1 deletion also had epilepsy [15]. Identifying 1q21.1 microdeletions in patients with idiopathic generalized and idiopathic focal epilepsies suggests that variation at this locus predisposes to a broad range of seizure disorders crossing traditional diagnostic boundaries.
In addition, we identified one patient (EMJ162) with JME and a duplication of 16p11.2 (chr16: 29.5–30.2 Mb), which has been associated with autism, developmental delay and schizophrenia [10]–[12], [14], [27], [35], [37]. Finally, we identified one individual with severe idiopathic generalized epilepsy of infancy (SIGEI) (K027) with a more distal deletion of 16p11.2 (chr16: 27.7–28.9 Mb), recently associated with severe early-onset obesity and ID [36], and one patient with BECTS (K105) and a deletion of 16p12.1 (chr16: 20.2–20.8 Mb), also associated with ID and other neurodevelopmental defects [38]. Thus, our data adds to the phenotypic spectrum associated with rearrangements at several genomic hotspot regions. In particular, we identify hotspot deletions in two patients with BECTS. Gene identification in BECTS, despite representing the most common focal epilepsy syndrome of childhood, has been elusive so far. Here, we suggest that some recurrent hotspot deletions might predispose to both idiopathic generalized and focal epilepsies.
We detected 18 deletions and 16 duplications that are not associated with rearrangement hotspots. Fifteen events involve a single gene; of these, 12 are deletions. Although all of the CNVs reported here are not found in our control set of 2493 individuals, it is possible that some are rare but benign CNVs. However, many of the CNVs we identified contain one of more plausible candidate genes for epilepsy (Table 2).
We identified a deletion of exons 2–4 in the CNTNAP2 gene in a proband with neonatal seizures. CNTNAP2 has been identified as a candidate gene for autism [50]–[52], and heterozygous deletions involving the gene were reported in three patients with schizophrenia and autism [53]. The deletion is predicted to cause an in-frame deletion of 153 amino acids in the resulting protein. The same patient has a 370-kb deletion of 17p13 that deletes seven genes and has not been seen in our control cohort. We also identified a patient with a duplication encompassing a related gene, CNTNAP4. Finally, two individuals in our cohort have overlapping deletions within AUTS2. This gene is disrupted by de novo balanced translocations in three unrelated individuals with mental retardation [54] and a pair of twins with autism and mental retardation [55], suggesting a role for AUTS2 in normal cognitive development. The two deletions we detected are intragenic and overlapping.
Previous studies of CNVs in epilepsy have focused on probands with IGE. It is known from studies of families with autosomal dominant epilepsy that a wide range of seizure types can be caused by the same single-gene mutation. For example, Dravet syndrome, a severe early-onset disorder associated with poor cognitive outcome, and the milder generalized epilepsy with febrile seizures plus (GEFS+) syndrome are both caused by mutations in the SCN1A gene [56]–[58]. Therefore, we included probands with common idiopathic focal epilepsies and non-lesional, idiopathic epilepsies. Some of our probands were diagnosed with specific epilepsy syndromes, including myoclonic astatic epilepsy (Doose Syndrome), atypical benign partial epilepsy [39], Landau-Kleffner syndrome, idiopathic West syndrome, severe idiopathic generalized epilepsy of infancy [59] and benign neonatal or infantile seizures. These particular epilepsy syndromes are usually associated with normal MRI results. We find that 6.6% of probands with IGE and 7.9% of those with idiopathic focal epilepsy harbor rare CNVs that may underlie their epilepsy phenotype. Notably, 12.7% of patients with other, often more severe forms of epilepsy in our series carry one or more rare CNVs. In our series, the vast majority of patients with deletions of 15q13.3, 16p13.11 and 15q11.2 BP1–BP2 were in the IGE cohort, accounting for 3.3% of cases. In the non-IGE patients, a deletion of 15q11.2 was found in a single patient with infantile seizures and a deletion of 16p13.11 was found in one patient with BECTS, suggesting that deletions at these three genomic hotspots confer greater risk for IGE than other types of epilepsy.
In summary, we find that 46/517 probands (8.9%) with various forms of idiopathic epilepsy carry one or more rare CNVs that may predispose to seizures, a frequency similar to that in studies of patients who present with other neurocognitive phenotypes, including ID, autism and schizophrenia. Furthermore, we identified CNVs involving genes and genomic regions previously identified in patients with the neurocognitive phenotypes listed above, suggesting common genetic etiological factors for these disorders. Our data suggest that rare CNVs are important in many subtypes of idiopathic epilepsies, including idiopathic generalized and idiopathic focal epilepsies as well as specific idiopathic, non-lesional epilepsy syndromes. The genomic regions and genes identified in this study are potential novel candidate genes for epilepsy.
Patients were collected at five centers after appropriate human subjects approval and informed consent at each site.
Patients were collected at five centers: (1) 140 probands with a primary diagnosis of JME, CAE, absence epilepsy, IGE or idiopathic epilepsy were selected from the NINDS repository (http://ccr.coriell.org/ninds); (2) 160 patients are probands with a primary diagnosis of JME from Switzerland. Patients from cohorts (1) and (2) were previously analyzed using MLPA for the CHRNA7 gene [60], and two probands (EMJ001 and EMJ020) were determined to have 15q13.3 microdeletions by that method; they were not previously analyzed for any other copy number changes. (3) 186 German patients came from two cohorts: 76 patients from a population-based cohort from Northern Germany (POPGEN cohort) and 110 patients with childhood-onset epilepsy collected at the University of Kiel. Finally, 41 patients with various idiopathic generalized epilepsies collected at (4) the University of Iowa and (5) at Washington University, St. Louis. DNA from the NINDS repository was derived from cell lines; DNA from all other cohorts was directly from blood. Patients were diagnosed according to the widely used 1989 ILAE classification [61]. In addition, several pediatric patients were diagnosed with specific syndromes not yet recognized in the ILAE classification (Table 1). Patients with non-lesional, idiopathic epilepsies in which diagnostic criteria of the recent ILAE classification for particular epilepsy syndromes were not met were labeled as “unclassified”.
Array CGH was performed using either custom or commercially available oligonucleotide arrays containing 135,000 isothermal probes (Roche NimbleGen, Inc.). Customized arrays (459 samples) were designed with higher density probe coverage in known rearrangement hotspot regions (average probe spacing 2.5 kb) with lower density whole-genome backbone coverage (average probe spacing 38 kb). A subset of samples (n = 62) was analyzed using a commercially available whole-genome array (Roche NimbleGen 12×135 k whole-genome tiling array) with average probe spacing throughout the genome of 21 kb.
Data were analyzed according to manufacturer's instructions using NimbleScan software to generate normalized log2 fluorescence intensity ratios. Then, for each sample, normalized log intensity ratios are transformed into z-scores using the chromosome-specific mean and standard deviation. Z-scores are subsequently used to classify probes as “increased”, “normal” and “decreased” copy-number using a three-state Hidden Markov Model (HMM). The HMM was implemented using HMMSeg [62], which assumes Gaussian emission probabilities. The “increased” and “decreased” states are defined to have the same standard deviation as the “normal” state but with mean z-score two standard deviations above and below the mean, respectively. Probe-by-probe HMM state assignments are merged into segments according to the following criteria: consecutive probes of the same state less than 50 kb apart are merged, and if two segments of the same state are separated by an intervening sequence of ≤5 probes and ≤10 kb, both segments and intervening sequence are called as a single variant. CNV calls are filtered to eliminate (i) events containing <5 probes, (ii) CNVs with >50% overlap in a series of 2493 control individuals [24] and (iii) events that had no overlap with RefSeq genes. In addition, when comparing CNV calls to control CNVs, we eliminated calls for which there was insufficient probe coverage (<5 probes) in the control data to identify the same or similar CNV. Filtered copy number changes are also visually inspected in a genome browser.
|
10.1371/journal.pcbi.1000548 | Gene Circuit Analysis of the Terminal Gap Gene huckebein | The early embryo of Drosophila melanogaster provides a powerful model system to study the role of genes in pattern formation. The gap gene network constitutes the first zygotic regulatory tier in the hierarchy of the segmentation genes involved in specifying the position of body segments. Here, we use an integrative, systems-level approach to investigate the regulatory effect of the terminal gap gene huckebein (hkb) on gap gene expression. We present quantitative expression data for the Hkb protein, which enable us to include hkb in gap gene circuit models. Gap gene circuits are mathematical models of gene networks used as computational tools to extract regulatory information from spatial expression data. This is achieved by fitting the model to gap gene expression patterns, in order to obtain estimates for regulatory parameters which predict a specific network topology. We show how considering variability in the data combined with analysis of parameter determinability significantly improves the biological relevance and consistency of the approach. Our models are in agreement with earlier results, which they extend in two important respects: First, we show that Hkb is involved in the regulation of the posterior hunchback (hb) domain, but does not have any other essential function. Specifically, Hkb is required for the anterior shift in the posterior border of this domain, which is now reproduced correctly in our models. Second, gap gene circuits presented here are able to reproduce mutants of terminal gap genes, while previously published models were unable to reproduce any null mutants correctly. As a consequence, our models now capture the expression dynamics of all posterior gap genes and some variational properties of the system correctly. This is an important step towards a better, quantitative understanding of the developmental and evolutionary dynamics of the gap gene network.
| Currently, there are two very different approaches to the study of pattern formation: Traditional developmental genetics investigates the role of particular factors in great mechanistic detail, while newly developed systems-biology methods study many factors in parallel but usually remain rather general in their conclusions. Here, we attempt to bridge the gap between the two by studying the expression pattern and function of a particular developmental gene—the terminal gap gene huckebein (hkb) in the fruit fly Drosophila melanogaster—in great quantitative detail using a systems-level approach called the gene circuit method. Gene circuits are mathematical models which allow us to reconstitute a developmental process in the computer. This allows us to study the function of the hkb gene in its wild-type regulatory context with unprecedented accuracy and resolution. Our results confirm earlier, qualitative evidence, and show that hkb plays a small, but crucial role in gap gene regulation. Understanding hkb's regulatory contributions is essential for our wider understanding of dynamic shifts in the position of gap gene expression domains which play important roles during both development and evolution.
| How genes contribute to pattern formation is one of the central questions of modern developmental biology. Traditionally, this question has been addressed using genetic and molecular approaches. Although very powerful, these approaches have several important limitations: First, it is difficult to study expression features which are not specifically affected by a particular mutation (see below). Second, there is always some remaining ambiguity whether an interaction is direct or not [1]. And finally, it is difficult to establish whether known regulatory interactions are not only necessary, but also sufficient to account for patterning in the wild-type system [2]. It is important to develop complementary approaches that help us to overcome these limitations. Here, we show how such an approach can be used to investigate the patterning function of a particular gene in its wild-type context in a rigorous and quantitative manner.
The patterning system we study is the gap gene network of Drosophila. Gap genes constitute the first zygotic step in a regulatory cascade which leads to the determination of body segments along the major (anterior-posterior, A–P) body axis during the blastoderm stage, shortly before the onset of gastrulation [3],[4]. They are involved in the regulation of pair-rule and segment-polarity genes. The latter establish a segmental pre-pattern of gene expression by gastrulation time. Gap genes such as hunchback (hb), Krüppel (Kr), giant (gt) and knirps (kni) are expressed in broad, overlapping domains. These domains are established by spatial gradients of the maternal co-ordinate proteins Bicoid (Bcd), Hb, and Caudal (Cad) (reviewed in [5]). Later these expression patterns are maintained and refined through gap-gap cross-regulation (see [1], and references therein), as well as regulation by the terminal maternal system acting through the terminal gap genes tailless (tll) and huckebein (hkb) (reviewed in [6]). In this report, we focus on hkb and its role in gap gene regulation.
The expression domains of gap genes in the posterior region of the embryo shift towards the anterior over time [7]. These shifts are independent of maternal factors or gap protein diffusion. Instead, they are caused by an asymmetric cascade of cross-repressive interactions between gap genes with overlapping expression domains (reviewed in [8]): the posterior hb domain is established during late blastoderm stage; this leads to the repression of Hb's anterior neighbour gt; Gt then represses its anterior neighbour kni, whose protein product in turn represses its anterior neighbour Kr. In contrast, anterior neighbours never repress their posterior neighbours. Note that a qualitatively similar, but less specific, mechanism for domain shifts has been predicted previously based on theoretical considerations [9].
This mechanism suggests that the posterior hb domain plays a central role in the initiation and regulation of gap domain shifts. However, our understanding of hb regulation in this domain is poor and incomplete. In particular, the position of its posterior boundary itself shifts over time[10], but the regulatory mechanism by which this is achieved remains unknown. In this paper, we use the gene circuit method—a data-driven modelling approach [11],[12]—to investigate the role of Hkb in the establishment and subsequent shift of the posterior hb domain.
The gene circuit method uses mathematical models of gene networks as computational tools to extract regulatory information from quantitative, spatial gene expression data (Figure 1A). We obtained such data for hkb expression using a slightly modified version of an established data-processing pipeline (see [13], and references therein): (1) A polyclonal antibody against Hkb protein was raised and used to stain blastoderm stage Drosophila embryos. (2) Embryo images were acquired using a confocal laser scanning microscope. (3) Image segmentation was applied to obtain numerical tables of average protein concentrations per nucleus. (4) Embryos were sorted into time classes—each covering about 7 min of developmental time—based on Eve expression and morphological markers. (5) Non-specific background staining was removed and (6) data were averaged across all embryos stained for Hkb at a given time point. This yielded an integrated, quantitative time-series of Hkb expression patterns, which we combined with equivalent data for other gap genes from the FlyEx data base [14],[15] for modelling and model fitting (see below).
To simulate the dynamics of gap gene expression, we use gene circuit models (see Methods for equations) [11],[12]. Such models have been successfully used in the past to investigate gap gene expression and regulation [1], [7], [16]–[22]. Gap gene circuits consist of a row of dividing nuclei along the A–P axis of the embryo. Between nuclear divisions, gap proteins are synthesised and decay within each nucleus. In addition, gap proteins diffuse between neighbouring nuclei which are not yet separated by cell membranes at this stage [23]. The model incorporates a few basic assumptions about eukaryotic transcriptional regulation: Regulatory input is fed into a sigmoid regulation-expression function. We assume that each regulatory interaction can be either repressive (if it is negative), absent (if it is close to zero) or activating (if positive), and hence can be represented by a single number or parameter in the model. For simplicity, we assume that regulatory inputs are additive and independent of regulatory context (i. e. the presence or absence of other regulators).
Previous gene circuit models included the gap genes hb, Kr, kni, gt and tll as well as the maternal co-ordinate gene cad (6-gene models; Figure 1B, left) [1], [7], [17]–[20]. All of these genes regulate and are regulated by other genes in the model. However, it is known from the experimental literature that neither tll nor the maternal contribution to cad are affected by gap genes (zygotic cad expression is repressed by Hb, but does not play a role in gap gene regulation) [24]–[29]. This can create modelling artifacts—inconsistent with experimental data—such as an expansion of tll expression which influences gap gene expression in the central region of the embryo [1],[21],[22]. It also leads to problems with the determinability of parameters involved in tll and cad regulation, which in turn affects determinability of regulatory parameters for other gap genes (see below) [19]. Finally, the absence of Hkb in these models results in incorrect expression and regulation of the posterior hb domain [1].
To avoid such problems, we use a revised model—first introduced in [21],[22]—which represents tll and cad as time-variable external inputs. This model only considers hb, Kr, kni and gt as core regulators of the network (4-gene models). Protein concentrations of the products of these genes constitute the state variables of the system, while levels of Tll and Cad are now calculated from data. It is assumed that they regulate, but are not themselves regulated by gap genes. These published models have a constant Bcd gradient and did not consider Cad data from late time points just before the onset of gastrulation [21],[22]. In contrast, we implement Bcd as a time-variable input, and use late Cad expression data to represent the rapidly changing expression dynamics of these two genes at that stage. Bcd starts being rapidly degraded right before the onset of gastrulation [10]. At the same time, Cad disappears from the central region of the embryo and refines into a posterior stripe of zygotic expression which has a homeotic, rather than maternal co-ordinate function [30]. Finally, the most important addition to the model in the context of this paper is that of the terminal gap gene hkb. Similar to tll, it is not regulated by gap genes itself [26],[28] and is included as yet another external input factor. Core regulatory genes and external inputs in our current 4-gene models are summarised in Figure 1B (right).
The modelling framework outlined above does not predetermine any specific regulatory interactions within the gene network. Instead, these interactions—and hence the regulatory topology of the network—are obtained by fitting the model to the data (Figure 1A). This is achieved by minimising a cost function that measures the difference between the two. Previous studies using gap gene circuits used a cost function based on the sum of squared differences between gap protein levels in the model and the data (ordinary least squares, OLS) [1], [7], [17]–[22]. However, the OLS cost function is an appropriate measure under certain assumptions only: all errors in the data have to be independent of each other, and are assumed to follow a normal distribution with zero mean and constant standard deviation. The latter condition clearly does not hold for our data set, since standard deviations vary for each gene over space and time (Figure S1) [10],[31]. Generally, standard deviations become smaller at late time points. They are also relatively small around domain boundaries, and almost negligible in non-expressing regions, indicating that domain position is determined with little variation towards the onset of gastrulation [10]. Therefore, it is more appropriate to consider data variability for model fitting by using a weighted least squares (WLS) cost function for optimisation (Maximum Likelihood Estimation, [32]), in which each squared difference between model and data is weighted inversely proportional to the standard deviation of the corresponding data point. In other words, data points with little embryo-to-embryo variability contribute more to the measured difference between model and data than those with a high variability between embryos. Here, we compare results obtained by both OLS and WLS fits to demonstrate that indeed, WLS is a more suitable measure than OLS not only in theory, but also in practice.
The resulting models are analysed in various ways to gain new biological insights. Analysis of the dynamical behaviour of our models allows us to associate specific regulatory interactions and mechanisms with specific features of gene expression (such as the establishment of a new expression domain or the formation, sharpening or shift of an expression domain boundary). This can either be achieved by graphical examination of specific interactions in the model [1],[2],[7], or by characterising the convergence of the system towards its various dynamical attractors [21],[22]. In addition, we can test how reliably our models predict a specific regulatory network topology, by statistical determinability analysis of our parameter estimates. This is achieved by calculating confidence intervals around our estimated solutions, which give us a range of values in which the true solution of our optimisation problem lies with a given probability (see Methods and [19],[33], for details). If these intervals do not range across several regulatory categories (‘activation’, ‘repression’, or ‘no interaction’), the parameter is well-determined. In contrast, if they cover more than one regulatory category, the parameter is only weakly determined, or not determined at all. It has been shown that biological network models always contain at least a few parameters which cannot be determined, and that this is usually due to parameter correlations [34]. Here, as in a previous study [19], we analyse such parameter dependencies by calculating an average correlation matrix across solutions.
In the sections that follow, we analyse the protein expression pattern of hkb in a quantitative manner. We then use these quantitative expression data as external input to new gap gene circuit models. We obtain parameter estimates for these models (and hence a predicted regulatory topology for the gap gene system) using fits with both OLS and WLS cost functions. We show that the latter produces more consistent and well-determined parameter estimates. In contrast to earlier models, these circuits now reproduce expression dynamics in the posterior hb domain correctly. In particular, they show a correct anterior shift in this expression domain, and thus correct shifts in all gap domains in the posterior region of the embryo. We analyse the dynamical behaviour of our model to show that this is due to the repressive influence of Hkb on hb. We further establish that this is the only significant contribution hkb makes to pattern formation by gap genes. The role of hkb as revealed by our models is entirely consistent with evidence from the experimental literature. Finally, we discuss its implications for gap domain shifts, segment determination and the evolution of the gap gene system.
Polyclonal antiserum against Hkb protein was raised as follows: A full-length cDNA clone of hkb (FlyBase ID: FBgn0001204) was obtained from the Drosophila gene collection (http://www.fruitfly.org/DGC), and recombined into a pET-DEST42 GATEWAY expression vector (Invitrogen). The resulting construct was auto-induced in E. coli strain BL21(DE3) using Overnight Express medium (Novagen/Merck). 6xHis-tagged Hkb protein was purified according to [35]: The most prominent protein band was excised from a preparative SDS-PAGE gel and recovered by electroelution followed by dialysis against double distilled water. Antibodies were raised in two rats using of protein per rat (Eurogentec).
Blastoderm stage embryos of Drosophila melanogaster (collected 1–4 hrs after egg laying) were stained against Hkb (dilution: 1∶100), Eve (1∶2000) and either Hb (1∶1000) or Kni (1∶400), using antisera described above (for ), in [36] (for ) and in [35] (for , and ). Eve is used for time classification [13]. As secondary antibodies, we used , and (Molecular Probes) at a dilution of 1∶4000. Nuclei were counter-stained using Hoechst 34580 (Invitrogen). Laterally oriented embryos were scanned using a water-immersion objective on a Leica SP5 confocal scanning laser microscope. Fluorescent dyes were excited with a single wavelength at a time to prevent bleed-through between channels. The following wavelength windows were used for detection: 410–485 nm (with the 405 nm blue diode laser line), 495–555 nm (488 nm Argon), 565–625 nm (561 nm DPSS), and 640–720 nm (633 nm HeNe). To ensure reproducibility of measurements, scans were performed using identical detector gain and offset for all embryos on a slide. Images of dorsal nuclear and membrane morphology for time classification were obtained using differential interference contrast (DIC) with a water-immersion objective.
Embryo images were processed to yield integrated expression data as described in the Introduction and in [13] (and references therein), with the following exceptions: (1) Images of embryos at early blastoderm stage (comprising cleavage cycles 9 to 13 (C9–C13); cleavage cycle is the period between mitoses and [23]) were segmented using a threshold-based algorithm: Images were de-speckled using a median filter; a top-hat transformation was used to remove uneven background; automated thresholding (using Otsu's method) was corrected interactively wherever necessary until all nuclei in an image were captured by the algorithm; finally, a watershed segmentation algorithm was applied to the distance transform of the thresholded image to avoid fused nuclei [37]. (2) Images of embryos at late blastoderm stage (cleavage cycle 14A (C14A)) were segmented using a watershed algorithm combined with nuclear edge detection as described in [38]. To reduce over-segmentation, we introduced an extended-minima transform before the watershed algorithm was applied [37]. (3) Expression data were not registered, as registration based on expression features in the central region fails at the termini where hkb is expressed, and not enough replacement features were available in that region of the embryo. (4) Due to its low signal-to-noise ratio, Hkb serum had to be used at a relatively high concentration (see above) to elicit a clearly detectable signal. This created high levels of non-specific background staining in the central region of the embryo, which our background removal procedure failed to completely remove. The residual central signal is clearly separated from the two expression domains at the termini. It does not seem to represent any real expression, and has not been observed in any previous study of hkb [26],[28],[39],[40]. To avoid modelling artifacts like those described for Tll in the Introduction, this signal was removed from integrated data by setting Hkb levels in the central region to zero. Moreover, integrated Hkb data were scaled (by an arbitrary factor of 3 across all time classes) to facilitate visual comparison (in Figure 2, right column) and to reduce numerical stability problems when solving the model (see below). Hkb expression data will be integrated into the FlyEx database, available at http://urchin.spbcas.ru/flyex or http://flyex.ams.sunysb.edu/flyex [14],[15].
Quantitative integrated expression data for Bcd, Cad, Hb, Kr, Kni, Gt and Tll are taken from the FlyEx database. Concentration measurements were taken at C13, as well as eight regularly spaced time points during C14A (T1–T8) [13]. The data set used for model fitting consists of averaged nuclear protein concentrations. Averaging is achieved by collecting measurements from individual embryos into a number of bins along the A–P axis. Each integrated expression pattern at a given time point is based on data from 9–62 individual embryos (with the exception of Kni at C13, which is represented by 4 embryos only). Each embryo contributes measurements from multiple nuclei to a bin to be averaged. Therefore, the number of measurements used in the computation of the averaged concentration value per nucleus (the sample mean) is usually much larger than the number of embryos per time point. Based on this and the Central Limit Theorem [41], we assume that concentration values in averaged bins are approximately normally distributed. As it is not known how measurements are correlated, we take them to be independent of each other. Figure S1 shows integrated gap gene expression data with their associated standard deviations.
Gene circuits are hybrid dynamical models with two continuous and one discrete rule: (1) interphase, (2) mitosis and (3) division [11]. During interphase, the change in concentration for each gap gene product in each nucleus over time is described by the following system of ordinary differential equations (ODEs):(1)The three terms on the right-hand side of the equation represent regulated protein synthesis, protein diffusion and protein decay. Integer indices and refer to regulated gap genes and regulators respectively, and refers to external regulators. is the number of gap genes in the model (hb, Kr, kni and gt), is the number of external regulatory inputs (provided by bcd, cad, tll and hkb, genes which regulate gap genes but are not regulated by gap genes themselves). represents the total regulatory input to gene . and are genetic interconnectivity matrices (for state variables and external inputs respectively, of size and ) whose elements (called regulatory weights) each define one particular regulatory interaction in the gap gene network. is a threshold parameter (which represents the influence of uniform maternal factors on the expression of gene ) for the sigmoid regulation-expression function(2)Negative regulatory input leads to increasing repression (with leakage), while positive regulatory input leads to increasing activation until saturation of gene expression at maximum production rate . is a diffusion rate that depends on the distance between nuclei, which halves at every nuclear division ( is the number of previous divisions). is the rate of decay for the product of gene . It is related to the half-life of the protein by . During mitosis, protein synthesis is shut down. Nuclei divide instantaneously at the end of mitosis and the protein concentrations from each mother nucleus are copied to its two daughters. We use the same division schedule as in Figure 2 of [1], which is based on [23],[42].
Gap gene circuits include cleavage cycles 13 and 14A (ending at the onset of gastrulation; ) and cover the region from 35% to 92% along the A–P axis of the embryo (where 0% is the anterior pole). This includes and nuclei at C13 and C14A, respectively. As a consequence, system (1) consists of 120 and 232 ODEs during C13 and C14A respectively. At the boundary points and we replace the diffusion term in the right-hand side of (1) by and respectively, implementing homogeneous Neumann (no-flux) boundary conditions.
Kr, Kni, Gt, Tll and Hkb proteins are not present at significant levels before C13 (see Results and [10]). Thus, we use zero initial conditions for these. Non-zero initial conditions for Bcd, Cad and Hb are obtained by linear interpolation of integrated expression data at C12 () and C13 (). Moreover, to solve (1) one needs concentration levels for external inputs at all time points . This is achieved by linear inter- or extrapolation from data points at ( denotes the single time point in C13). Higher-order inter-/extrapolation is prone to produce artifacts due to fluctuations in the expression data, and is therefore not used here [21]. Because it is not clear whether integrated Bcd profiles at T7 and T8 have non-specific background properly removed, we used linear extrapolation based on T5/6 for these time points. This results in a rapid decay of the Bcd gradient just before the onset of gastrulation qualitatively similar to that described in [10]. Negative extrapolated concentration values were reset to zero wherever necessary.
Equation (1) contains parameters (parameter vector containing , , , , and ), whose values we seek to determine by fitting the model to the data. We denote each measurement in our data set by , specified by the time when the concentration of gene product in nucleus was measured. The corresponding model value obtained from (1) is denoted by . The estimation of unknown parameters in (1) amounts to minimising the cost function(3)where are positive weights, is the number of gap genes, is the number of time classes, and is the number of nuclei (which depends on the number of preceding mitoses ) for which we have data. When all weights in (3) are equal to one, (3) represents an ordinary least squares (OLS) fit, which was the cost function used in all previous studies using gene circuit models [1], [7], [17]–[22]. When the weights are taken to be inversely proportional to the corresponding variances in the data, the cost function becomes the weighted least squares (WLS) distance and its minimum is the Maximum Likelihood Estimate [32].
The quality of a fit of the model to the data is measured by the root mean square (RMS) given by(4)where is the total number of all measurements. A solution is considered to be ‘good’ if its and if there are no visible pattern defects in the model response [1].
We used a two-step optimisation algorithm to minimise the cost function (3): Global optimisation by the parallel Lam Simulated Annealing (pLSA) algorithm [43]–[45] was performed on the Darwin cluster at the High-Performance Computing (HPC) centre of the University of Cambridge (http://www.hpc.cam.ac.uk) as described previously [1],[7],[21],[22]. pLSA solutions were used as starting points for local search by the Levenberg-Marquardt (LM) method [46],[47] as described in [19],[33]. The complete set of estimated parameter values can be found in Table S1. For numerical solution of the model during pLSA optimisation, we use a Runge-Kutta Cash-Karp (Rkck) adaptive-step-size solver set to high accuracy to avoid numerical instability [48]. During local optimisation by LM the model is solved using an implicit multistep Backward Differentiation Formula (BDF) as previously described in [19],[33].
Based on previous studies using gap gene circuits [1], [7], [18]–[22], we define our search space for parameter estimation by the linear constraints , , (), and by the following non-linear penalty function for regulatory parameters and (5)where and are the maximum concentration values in our data set for proteins and , respectively. Previous work has shown that fixing the values of parameters improves parameter determinability without affecting the overall quality of the fits [19]. Therefore, we take in all simulations, which leaves us with unknown parameters in (1) to be estimated.
Here, we only provide a brief overview of the equations used for calculating confidence intervals and parameter correlations (see Introduction). For more detailed explanations of these statistical quantities and their derivations, we refer the reader to [19],[33] (and references therein).
Model optimisation results in a vector with the estimated parameter values as its elements. The ellipsoidal confidence region around , in which the ‘true’ parameter vector lies with a certain probability (defined as 95% in our case) is defined by(6)where and are the number of parameters and measurements, respectively. is the Jacobian (or sensitivity) matrix of size , defined as where is the vector of weighted differences between model and data. Each entry in shows how sensitive the model response is at the data point for a change in the parameter. is the upper part of Fisher's distribution with and degrees of freedom. From (6) one can derive dependent and independent confidence intervals for parameter estimates (). These are, respectively,(7)and(8)Here and are obtained from the Singular Value Decomposition of [48],[49] and .
The correlation coefficient between and is given by(9)where .
We quantified expression levels of Hkb protein in blastoderm stage embryos of Drosophila as described in Methods. Our analysis closely follows that of tll in [10], and focuses on the last two cleavage cycles before gastrulation (C13 and C14A; cleavage cycles and time classes are defined in Methods) [23]. Representative embryo images and quantified expression patterns from those individual embryos are shown for all time classes (T1–T8) of C14A in Figure 2, left and middle column. Scaled, integrated expression data for Hkb are compared to other gap gene expression patterns in Figure 2, right column, which also indicates the number of embryos used to construct the data set.
Hkb protein can first be detected in both its anterior and posterior domain at C13 (data not shown). Protein levels rapidly increase during early C14A (T1–T3). At this stage, peak levels are very similar in both domains, although the anterior is very slightly weaker than the posterior one. Subsequently, the anterior domain gradually weakens (T5–T8), while protein levels in the posterior domain remain more or less constant (although there may be a slight decrease in concentration at T8). The peaks of both domains remain at a constant position throughout (5% A–P position for the anterior, 95% for the posterior domain). Similarly, the width of both domains remains approximately constant: the anterior domain extends back to about 10–15% A–P position, while the posterior domain reaches as far as 85–90%, both domains covering about 10–15% A–P position in each terminal region. None of the two Hkb domains show any discernible D–V asymmetry at any point in time before gastrulation.
Our quantitative hkb expression data enabled us to include this gene in gap gene circuit models. We used both OLS and WLS cost functions for fitting 4-gene models (Figure 1B, right) to quantitative expression data (Figure S1). For the OLS cost function, we performed 740 independent optimisation runs (combined global and local search). The quality of a fit is assessed using the root mean square (RMS) score (defined in Methods). About 80% of the resulting parameter sets have good-scoring RMS values (). This residual error is below the level of variation in the expression data [10],[31]. However, a closer look at the patterns for good-scoring sets reveals that most of them have a slight, but significant, patterning defect in common: model output shows an artifactual hump of Kr expression posterior to its central domain (data not shown). This problem has also been noticed in an earlier study with gap gene circuits without hkb (Manu, Stony Brook University, New York, USA: personal communication). In these circuits, Gt represses hb and the small ectopic Kr domain is required to down-regulate gt to allow initiation of posterior hb expression. This is both incompatible with experimental evidence [50]–[56] and previously published models of the gap gene system [1], [7], [17]–[19],[21],[22]. Therefore, we exclude these solutions from our analysis. Although a large majority of circuits obtained by OLS fits show the small ectopic Kr domain, we found 39 low-scoring parameter sets that do not have this patterning defect (Figure S2). These circuits were selected for further analysis. Their values vary between and .
Local search with the WLS cost function was performed using selected OLS parameter estimates as starting points: the 39 solutions without, and the lowest-scoring 90 solutions with defective Kr expression. In addition, we performed 80 independent optimisation runs using WLS both for global and local search. For our analysis, we selected 117 (out of 209) parameter sets with the lowest WLS scores varying uniformly between and . This corresponds to RMS values between and , which are slightly higher than those for OLS runs since WLS solutions tolerate larger residual errors at early stages of gap gene expression. None of these low-scoring parameter sets show any major patterning defects (Figures 3 and S3), while most solutions with larger WLS scores do (data not shown). In particular, we observed no ectopic expression of Kr in any of these solutions. This is not surprising as standard deviations in the data are small in regions where protein concentration is low. Thus, the corresponding weights for the WLS cost function are large, which prevents the presence of any ectopic expression domains (even if they are small) in low-scoring solutions.
Gap gene expression patterns produced by circuits from the selected OLS and WLS fits are similar, although variability between different models is somewhat larger for OLS (compare Figures S2 and S3). As expected, WLS solutions generally show slightly better fits at late stages. Most visible defects occur early. The posterior borders of the central Kr and the posterior gt domain become established at a slightly different position than in the data (Figures 3, arrowhead, and S3). In addition, there are irregularities in the shape of anterior expression boundaries of the posterior gt domain (WLS only; Figure S3), the central domain of Kr, and the posterior domain of hb (OLS and WLS; asterisks in Figure 3). Although such irregularities in boundary shape lie well within the variability of the integrated data (cf. Figure S1), they are never observed in quantitative expression profiles extracted from individual embryos [10]. Similar problems with the posterior domains of gt and hb have been observed in earlier models of the gap gene system [1],[21].
On the other hand, the dynamic expression of hb in its posterior domain is reproduced correctly. Earlier models exhibited defects in the timing and positioning of the posterior boundary of this domain (see dark grey Hb profile in Figure 3), while the circuits presented here accurately reproduce the establishment and subsequent anterior shift of this expression border (arrows in Figure 3).
Estimates of regulatory weights obtained by both OLS and WLS fits were classified into the following three categories: ‘activation’ (parameter values ), ‘repression’ () and ‘no interaction’ (between and ) [1],[18],[19]. This leads to a predicted regulatory topology of the network based on which category a majority of parameter estimates falls into (summarised in Figure 4). If a threshold of is chosen instead, the predicted network topology remains largely unchanged, with two notable exceptions: the activating effects of both Cad and Tll on hb change to the ‘no interaction’ category indicating that these predicted interactions are very weak, and may not be significant (see Discussion).
Apart from only two interactions, the predicted regulatory topologies agree between OLS and WLS fits. In the case of OLS, Hkb activates gt and represses kni, while for WLS it is the other way around (Figure 4). Strikingly, the more consistent expression patterns between WLS solutions are also reflected by more consistent predictions of network structure. While many parameters fall into different categories in different OLS solutions, only one interaction (regulation of kni by Hkb) shows this type of ambiguity in the case of WLS (Figure 4). This means that WLS solutions are not only more tightly clustered in terms of their expression patterns, but also in terms of the distribution of their parameter values.
A similar pattern can be observed when comparing our new 4-gene models with earlier 6-gene circuits (cf. Figure 1B). Although the predicted regulatory structure is largely in agreement between these two types of model, consistency of the prediction is improved considerably in 4-gene models (even in the case of the OLS solutions presented here). Repression of Kr and gt by Hb, of kni by Gt, of Kr by Kni and of gt by Tll are now present in all parameter sets, while previous results for the 6-gene case showed no interaction for these weights in many solutions [1],[18],[19]. Weak activation of hb by Tll is now predicted by a large majority of parameter sets. Some previous models had predicted this interaction [17], while most showed repression or no interaction between the two genes [1],[18],[19]. Another activating interaction which is now consistently predicted is that between Kni and gt. Finally, there is no auto-activation of gt in a very large majority of our parameter sets.
The regulatory structure of the gap gene system shown in Figure 4 is based solely on the classification of estimated parameters into regulatory categories. To assess the quality of the parameter estimates more rigorously, we computed dependent and independent confidence intervals for each parameter set (see Methods and [19],[33]). We then checked if these confidence intervals fall entirely into negative (‘repression’), or positive (‘activation’) ranges of parameter values, or whether they cluster tightly around zero (‘no interaction’).
Results in Figure 4 are fully confirmed when only dependent confidence intervals (which tend to underestimate the extent of the confidence region) are taken into account. In contrast, not all of our conclusions from Figure 4 are supported when independent confidence intervals (which tend to overestimate the extent of the confidence region) are considered. For example, Figure 5A shows the confidence intervals for interactions between Gt and Kr (left; parameter: ), Bcd and hb (middle; ), as well as Tll and kni (right; ) for all 39 selected OLS fits. Independent confidence intervals for lie in the negative part of the plane for almost all parameter estimates and therefore, repression predicted for this weight in Figure 4 is confirmed by statistical analysis. In other words, this parameter is determinable. Independent confidence intervals for , on the other hand, slightly extend into the negative part of the plane. Therefore, the model only predicts that Bcd does not repress hb. Note that this is a weaker conclusion than predicting activation for this weight from Figure 4. Hence, this parameter is only weakly determinable. In contrast, we cannot draw any conclusions about , since independent confidence intervals extend from the negative into the positive part of the plane. Thus, statistical analysis cannot confirm the repression of kni by Tll inferred from Figure 4, and this parameter is not determinable.
Parameter determinability analysis based on independent confidence intervals for OLS and WLS fits is summarised in Figures 5B and 5C, respectively. We focus on regulatory parameters since, just as in earlier studies [19], promoter strengths , diffusion coefficients and decay rates have extremely large independent confidence intervals meaning that none of these parameters are determinable (data not shown). Confidence intervals for all regulatory weights are shown in Figures S4 (for OLS) and S5 (for WLS fits). It is evident that conclusions from this analysis are generally weaker than those drawn from classifying parameter values only (compare Figures 5B,C with Figure 4).
11 and 12 (out of 32) regulatory parameters cannot be determined for OLS and WLS fits, respectively. Among them are several of the interactions predicted to fall into the ‘no interaction’ category in Figure 4 (, and ) if a threshold of is chosen for the analysis. However, independent confidence intervals of these interactions are all very small and cluster tightly around zero (Figures S4 and S5). Furthermore, their intervals are completely within the ‘no interaction’ category if the threshold is extended to . For these reasons, we consider them to be determinable in Figure 5B and C. This lowers the number of non-determinable regulatory parameters to 10 for both OLS and WLS fits. Out of the remaining 22 regulatory weights, 2 are only weakly determinable (for both OLS and WLS fits), while the regulatory category for the other 20 is confirmed by statistical analysis. Which regulatory parameters are not determinable differs significantly between OLS and WLS solutions and does not follow any obvious pattern, apart from the fact that most interactions by terminal gap genes tll and hkb are affected (Figure 5B,C).
Previous quantitative analyses of the gap gene system suggested a set of basic regulatory mechanisms based on broad activation of gap genes by maternal co-ordinate proteins, and spatially specific gap-gap cross-repression [1],[7]. In addition, they revealed significant anterior shifts in the position of posterior gap domains after their initial establishment during C13 [7],[10]. These shifts are caused by asymmetric repressive interactions as described in the Introduction and in [1],[7],[22]. Parameter analysis (Figures 4 and 5), as well as graphical inspection of regulatory interactions across space and time (data not shown; analysis performed as in [1],[7]) show that our current 4-gene models implement exactly the same regulatory principles as those seen in previous 6-gene circuits.
In addition, our current gap gene circuits now accurately reproduce expression in the posterior hb domain, while shift and establishment of this domain were incorrect in previous models [1], [7], [17]–[22] (Figure 3). To investigate how the inclusion of Hkb affects this domain, we have performed a detailed graphical analysis of hb regulation in the posterior region of the embryo (Figure 6). This analysis reveals the following regulatory principles.
The posterior hb domain is the last gap domain to form in the posterior region of the embryo. Expression is initiated during cleavage cycle 13 and the domain retracts from the posterior pole in early cycle 14A (T2) [10],[57],[58]. Later during cycle 14A, expression levels increase, domain boundaries sharpen and shift further towards the anterior (see Figures 2 and 6, left column).
The late initiation of hb expression in the posterior region can be explained by residual amounts of Kni protein being present in the region during C13 and early cycle C14A (Figure 6, T2, left and middle panel). Kni is a very strong repressor of hb. Kni is increasingly repressed in the most posterior region of the embryo by the gradual accumulation of Gt protein (data not shown). In the model, combined activating inputs by Cad and Tll induce hb expression where Kni levels have fallen to a low-enough level (Figure 6, T2, middle and right panel). At later stages, hb auto-activation gradually supplements and replaces activation by other factors (Figure 6, T5/T8, middle). The posterior boundary of the posterior hb domain is set by Hkb repression (Figure 6, T2–T8, middle). The accumulation of Hkb in this region causes an increase in both levels and extent of this repression over time. This in turn leads to an anterior shift in the region where hb is expressed, such that Hb protein is only actively produced in the anterior part of its domain, while protein degradation dominates further posterior (Figure 6, T5–T8, right). At this level, the mechanism underlying the shift in the posterior hb domain is equivalent to those of other gap domains [7]: expression can extend anteriorly due to the lack of repression by the adjacent domain (posterior gt), while it becomes increasingly repressed posteriorly (by Hkb, in this case).
Our analysis of parameter determinability indicates that those parameters with particularly large confidence intervals could be fixed to specific values—within the non-empty intersections of their dependent intervals—without affecting the quality of the fits. Diffusion rates, for example, show large confidence intervals, despite not being significantly correlated with other parameters (see also below). Therefore, fixing their values during optimisation (to averaged values based on previously found estimates: , , and ) will not change the determinability of the remaining parameters but will reduce the size of the optimisation problem. On the other hand, regulatory weights describing the effect of Hkb on Kr, gt and kni have large confidence intervals (see Figures S4 and S5) because of correlations to other parameters, in particular the regulatory effects of Tll on the same targets (data not shown). This indicates a certain level of redundancy. Since a large majority of the dependent confidence intervals for these weights cover negative and positive values, we have set all of them to zero during optimisation. This leaves us with 37 parameters to be re-estimated.
We used local search with 60 initial parameter sets arbitrarily chosen from the previously found 117 WLS parameter sets. Additionally, we performed 20 global optimisation runs with these parameters fixed. From the resulting solutions, we selected 66 circuits which have low WLS values (about ). As expected, expression patterns produced by these models are very similar to those for WLS fits (data not shown).
The network topology shown for WLS runs in Figure 4 remains absolutely unchanged for the new estimates (with the obvious exception of the regulatory parameters for regulation of Kr, gt, and kni by Hkb which have been set to zero; data not shown). We calculated confidence intervals for these solutions to test whether more parameters are determinable in these models than in OLS and WLS fits with Hkb weights included (Figure S6). Our analysis, based on independent confidence intervals, is summarised in Figure 4D. It is immediately evident that determinability of regulatory parameters has significantly improved in these circuits compared to WLS fits. Only 2 weights ( and ) remain non-determinable, 4 show weak determinability (, , and ), while for the other 23 the confidence intervals confirm the type of regulation revealed by parameter classification. This is a significant improvement compared to circuits which include all regulatory weights for Hkb (compare Figure 4B,C with 4D).
The occurrence of non-determinable parameters is often caused by correlations between parameters [19],[34]. We have analysed these correlations for WLS models with fixed Hkb regulatory parameters, by calculating the mean correlation matrix for all parameters across solutions (see Methods and Figure S7). Note that for all significant entries of the mean correlation matrix the standard deviation is negligible, meaning that those correlations are present in all individual correlation matrices. This revealed the following correlations for parameters which are not or only weakly determinable in these models: Activation of hb by Bcd is negatively correlated with the activating effects of Cad (also weakly determined) and Gt, which indicates a certain level of redundancy of these interactions in the model. Repression of Kr by Tll is negatively correlated with activation of gt by Cad, indicating that the more Gt there is in the posterior (through increased activation of gt by Cad), the less repression by Tll is required to keep Kr expression off in the posterior of the embryo. The repression of hb by Hkb is negatively correlated with activation of hb by Tll, which indicates that a balance needs to be maintained between these interactions to enable correct posterior hb expression.
Finally, the last two interactions which are only weakly determined are the activation of kni by Bcd (negatively correlated with repression of kni by Hb) and the repression of Kr by Hb (negatively correlated with activation of Kr by Bcd; Figures 4D and S7). A similar correlation between Bcd activation and Hb repression can also be seen for gt, but does not lead to reduced determinability in this case. Similar correlations were also found in earlier 6-gene models [19]. They corroborate results which indicate that a delicate balance between activation and repression is essential for correct gap gene expression in the trunk region of the embryo [2]. In addition, we find similar negative correlations between Tll repression and Cad activation for the posterior gap genes gt and kni (Figure S7). These do not affect parameter determinability in our current models, but did so in earlier 6-gene models [19]. This indicates that balance between activation and repression through different maternal systems is crucial in the posterior region of the embryo as well.
After regulatory weights of gap gene circuits have been estimated based on wild-type expression data, analysis of mutants can be conducted in silico [59]. Null mutants of any regulator (or ) can be simulated by setting regulatory weights (or ) to zero for all regulated genes (while leaving all other parameter values unchanged). Similar to earlier gap gene circuit models [1], our current models do not reproduce expression patterns in mutant backgrounds for hb, Kr, gt or kni correctly (data not shown). In contrast, we were more successful at simulating null mutants of the terminal gap genes tll and hkb.
The only known alteration of gap gene expression in hkb mutants is the failure of posterior hb to retract from the posterior pole [26],[60]. This is reproduced correctly in both OLS and WLS solutions (arrows in Figure 7, upper and middle row). In addition, however, many OLS solutions show de-repression of gt and kni in posterior regions of the embryo (asterisks in Figure 7), which is inconsistent with the evidence. We never observed such defects in WLS circuits.
Embryos mutant for tll show more severe patterning defects: Both the posterior domain of gt and the abdominal domain of kni are expanded posteriorly [26],[53],[54],[61],[62], while the posterior hb domain is reduced or absent in these embryos [26],[60],[63]. Only Kr does not seem to be affected [63],[64]. Most OLS and WLS solutions show mutant expression patterns which are inconsistent with this evidence (data not shown). Surprisingly, however, circuits obtained by WLS with fixed diffusion rates and Hkb regulatory parameters, reproduce these defects correctly: there is no posterior hb expression (arrowhead in Figure 7), while gt and, to a lesser degree, kni are de-repressed in the posterior region of the embryo (arrows in Figure 7, bottom row).
Our results constitute a comprehensive, integrative analysis of the expression and function of the terminal gap gene hkb in the blastoderm embryo of Drosophila. On one hand, we have characterised the expression of hkb in a quantitative manner. On the other, we used a systems-level approach—the gene circuit method—to show how Hkb exerts its effect on the expression of hb in its wild-type genetic context, and to demonstrate that it does not have any non-redundant function in gap gene regulation beyond that. But before we discuss these biological insights in more detail below, we highlight two significant improvements in the gene circuit methodology, which have important implications for reverse engineering biological networks in general.
First, we were able to increase the efficiency of optimisation, and the consistency of parameter estimates, by using weighted least squares (WLS) instead of ordinary least squares (OLS) for optimisation. The use of a WLS cost function also reduces the need for human intervention when selecting solutions for analysis, since it prevents the occurrence of minor (but biologically significant) patterning defects such as the ectopic Kr domain observed in most OLS solutions. Out of 740 optimisation attempts with OLS, we only obtained 39 biologically realistic models. In contrast, none of the WLS solutions exhibited this problem, and thus a much larger proportion of them were suitable for analysis. This constitutes a very drastic increase in overall computational efficiency and biological relevance of the obtained fits. Furthermore, OLS solutions showed much larger variability in expression patterns and parameter values than those obtained with WLS. This indicates that fitting with WLS to data with non-constant standard deviations not only leads to biologically more relevant, but also to more consistent results across optimisation runs.
Second, analysis of parameter determinability [19],[34] allows us to eliminate parameters from the optimisation problem, thereby considerably reducing the complexity of the problem. Our models have 48 parameters, a number which we managed to reduce to 37 by fixing non-determinable parameters to specific values (see also [19]). Statistical analysis based on confidence intervals not only gives us an indication of which parameters to fix, but also of which values to fix them to (see Results). This was used successfully here for both diffusion rates and regulatory parameters representing the regulatory effect of Hkb on its targets Kr, gt and kni. Not only were we able to reduce the computational effort for optimisation, but fixing parameters also significantly improved parameter determinability, such that only 2 out of 29 regulatory parameters now remain non-determinable. This is a vast improvement over previous, 6-gene models [19].
In terms of the biology, we first discuss the expression and regulation of hkb. Our quantitative analysis of hkb expression confirms and extends results from earlier studies. Both Hkb domains cover about 10–15% A–P position in the anterior and posterior terminal region of the embryo [26],[39],[40]. Their borders coincide with the limits of the invaginating mesoderm in the ventral furrow during gastrulation [39]. Expression of hkb is more restricted to the terminal regions of the embryo than expression of tll (see Figure 2, right column). This difference is very clear at all time points for the posterior domains. In contrast, the early anterior domains of Hkb and Tll are very similar in extent, and only diverge at mid C14A (from about T5 onward), when the anterior Tll domain retracts from the pole. There are other, more subtle differences as well, revealed by a comparison with the quantitative analysis of Tll in [10]: The anterior domain of Hkb appears before that of Tll, which can only be detected during early cycle 14A. Hkb levels in this domain also decrease much earlier again (from T5 onward) than those of Tll in its anterior domain, whose peak levels remain constant until right before the onset of gastrulation (T7/8). Finally, the anterior domain of Hkb does not show any D–V asymmetry before gastrulation, while the corresponding domain of Tll retracts from the anterior pole and becomes increasingly dorsal during late cycle 14A (T5–T8). In contrast, dynamics of the maximum protein level in the posterior Hkb domain closely follows that of Tll, with the only potential difference being that Hkb persists very slightly longer in this region than Tll right before the onset of gastrulation (T8).
These results are entirely consistent with what we know about hkb regulation. The expression of hkb is completely independent of any other gap genes (including tll) [26],[28]. Both hkb domains depend on higher levels of Torso signalling from the terminal maternal system than those of tll, explaining their more restricted spatial extent [39], [65]–[67]. In addition, the anterior domain also requires the presence of Bcd [68]. These activating inputs are enabled through local relief of strong repression mediated by ubiquitous maternal factors such as Dead ringer (Dri) and Groucho (Gro) in the terminal regions of the embryo [69]. Interestingly, hkb is also regulated by the D–V maternal system, which is required for the ventral shift of the anterior hkb domain during gastrulation [28],[69]. Our results clearly indicate that this interaction is not significant before gastrulation as we can detect no D–V asymmetry in any of the two hkb domains at this stage (Figure 2).
But how does Hkb affect regulation of other gap genes? The regulatory mechanisms for the expression of the trunk gap genes hb, Kr, gt and kni predicted by our models are summarised in Figure 8: (1) Gap genes are broadly activated by the maternal gradients of Bcd and Cad. (2) Auto-activation is involved in maintenance and sharpening of boundaries in the anterior domain of hb, the central domain of Kr and the abdominal domain of kni. (3) The basic staggered arrangement of gap domains is provided by mutual repression between non-overlapping gap genes hb and kni, as well as gt and Kr. (4) Asymmetric repression between overlapping gap genes leads to anterior shifts in domain positions. (5) Terminal gap genes tll and hkb repress gap gene expression in the posterior terminal region of the embryo. These regulatory principles largely confirm results from previous studies using gap gene circuits [1], [7], [17]–[19],[21],[22].
The most significant improvement of our models over earlier ones is that they now correctly reproduce the expression and shift of the posterior hb domain (Figure 3). This means that our current models now reproduce the dynamic shifts of all posterior gap domains correctly [10]. Our analysis suggests that the appearance of this domain depends on the retraction of Kni—through increasing repression by Gt—from the posterior terminal region of the embryo in early cycle 14A (Figure 6). Its posterior boundary is set and subsequently shifted by increasing Hkb repression. These regulatory mechanisms are strongly supported by experimental evidence. Kni has been shown to repress hb: mis-expression of kni leads to a reduction of hb expression in the affected regions [70]–[72], and the posterior hb domain expands anteriorly in kni mutants [72]. Moreover, the abdominal domain of kni is expanded posteriorly in gt mutants [53], and reduced in embryos over-expressing gt [73]. Finally, repression of hb by Hkb is supported by the fact that the posterior hb domain fails to retract from the posterior pole in hkb mutant embryos [26],[60].
While our models reproduce repressive effects on posterior hb expression in a way consistent with experimental evidence, there is not much convincing data supporting the activating inputs responsible for posterior hb expression predicted by our models. Accordingly, we have omitted them from our regulatory summary in Figure 8. Both activation of hb by Cad and by Tll are predicted to be weak in the model. In the case of Cad, there is no evidence for any interaction with hb, as hb is expressed normally in mutants lacking both zygotic and maternal cad [29], as well as in embryos over-expressing cad [74]. Activation by Tll seems to be supported by the fact that posterior expression of hb is strongly reduced or absent in tll mutant embryos [26],[60],[63],[75], while the posterior hb domain expands anteriorly in embryos over-expressing tll [75]–[77]. In addition, there are several predicted Tll binding sites in the regulatory element responsible for posterior hb expression [78]. On the other hand, there is strong evidence that activation of hb by Tll is largely indirect—via repression of kni by Tll—as posterior hb expression is present in tll kni double mutants [77]. Finally, there is some evidence against a role of hb auto-activation in the posterior region. Mutants that express a non-functional Hb protein show no obvious defects in posterior hb expression [79]. Moreover, the expression of hb reporter constructs in the posterior hb domain is broadened and more intense in a hb mutant background compared to wild-type, while it is strongly reduced in embryos over-expressing hb [75]. In the model, none of these activating contributions provide any spatial specificity to posterior hb expression, which is mainly due to repression by Kni and Hkb (Figure 6). Taken together, this suggests that hb may be activated by an unknown, uniformly expressed maternal factor in this region.
There is another unresolved question concerning the posterior hb domain: Translation of hb is repressed by the posterior gradient of the maternal co-ordinate protein Nanos (Nos) and its co-factors [80]–[82]. These factors act through a Nos-response element, which is present in both maternal and zygotic transcripts of hb [81]. It remains unclear how this translational repression is overcome during mid cycle 14A. Either, the Nos gradient has disappeared (or is disappearing) by this time (this has never been assessed), or enough hb transcripts must accumulate to overcome Nos' repressive effect on translation. Quantitative studies of the Nos gradient will be required to resolve this issue.
For the posterior domain of hkb, our results show conclusively that its effect on hb expression is the only role it plays in gap gene regulation in the wild-type embryo. Excluding interactions of Hkb with Kr, kni and gt has no effect on any of these genes in the model. In fact, parameter determinability and prediction of tll mutant gene expression patterns improve significantly if these interactions are excluded (Figures 5 and 7). However, there is some evidence suggesting that Hkb does repress Kr and gt: The central Kr domain expands further posterior in embryos mutant for the maternal gene vasa (vas), tll and hkb than in those mutant for vas and tll alone [83]. Similarly, the posterior domain of gt expands further posterior in tll hkb double mutants than in embryos mutant for tll alone [26],[28]. Furthermore, the posterior gt domain is absent in embryos over-expressing hkb [28],[39]. Note that all of this evidence comes from over-expression experiments or embryos mutant for multiple genes, including tll. This suggests that there are two main reasons why interactions of Hkb with Kr, gt and kni do not play a role in the wild-type embryo: First, expression of hkb never overlaps its potential target genes (with the exception of gt; Figure 2). And second, its repressive input seems to be completely redundant with the corresponding repressive contributions by Tll. This is confirmed by our analysis of parameter determinability.
Apart from the regulation of the posterior hb domain, there are only two predicted interactions that differ in our 4-gene models compared to those in earlier 6-gene models. First, there is no auto-activation of gt in a large majority of our parameter sets. Although this interaction was present in earlier models [1], [17]–[19], gap gene auto-activation in general is not required for correct gap gene expression [17]. Second, activation of gt by Kni is supported by the fact that the posterior domain of gt is weakened and its posterior border fails to form properly in kni mutant embryos [52]–[54]. Fits in which this interaction is fixed to zero all show the ectopic expression of Kr described for OLS fits in the Results section, indicating that it is necessary for correct regulation of gt in the model (data not shown). However, it remains unclear whether its inclusion is an improvement over previous models. The experimental evidence remains ambiguous (effects are weak and the affected posterior border of gt occurs in a region where kni is not expressed), and activation of gt by Kni causes the transient patterning defect observed in the anterior border of the posterior gt domain in our current models (Figure S3). In summary, this suggests that neither of these two differences significantly affect the biological relevance of the models.
Little is known about the function and effect of the anterior hkb domain. In particular, it is not known why anterior hkb does not seem to have a repressive effect on hb, as both genes are co-expressed in this region. Unfortunately, we have not been able to include this domain in our analysis since our models currently do not include head gap genes, which are essential for patterning in the anterior region of the embryo.
Apart from correct posterior hb expression, the second major improvement of the models reported here is that they are able to reproduce null mutants of the terminal gap genes tll and hkb (see Figure 7). A theoretical study previously established that, in principle, it is possible to predict mutant patterns based on gene circuit fits to wild-type data only [59]. However, earlier gene circuit models—optimised against real, noisy expression data—failed to correctly reproduce any gap gene null mutants so far (including tll mutants) [1]. Our models provide an important first step towards the solution of this problem.
Apart from mutations in hkb and tll, gap gene circuit models have been shown to correctly reproduce gap gene expression (and its variational properties) in the presence of fluctuations in the Bcd gradient [21],[22]. All of these perturbations affect the gap gene network in a feed-forward manner. Neither bcd, hkb nor tll are regulated by gap genes themselves.
On the other hand, our current models still cannot accurately reproduce null mutants of the trunk gap genes hb, Kr, gt and kni (data not shown). All of these genes regulate and are regulated by other gap genes. This indicates that the problem is connected with feedback regulation within the model. Various potential reasons for this have been proposed in the past: over-simplified representation of transcriptional regulation in the model, missing production delays, scaling problems in the data, over-fitting to noisy expression data, or missing factors in the model which are redundant in the wild-type, but become important in a mutant background [1]. Further systematic studies will be necessary to elucidate which of these factors affect feedback regulation in our models in a way which makes them fail to reproduce such mutant expression patterns.
Why is all this important? After all, our results establish that hkb plays a very minor role in gap gene regulation. Yet, understanding the regulatory function of hkb is crucial for a better understanding of both the developmental and evolutionary dynamics of the gap gene system. Our current models are the first to reproduce all shifts of posterior gap domains correctly. There is evidence suggesting that the mechanism underlying these shifts is an emergent property of the entire gap gene network [7],[22]. If this is correct, we cannot understand gap domain shifts completely without understanding how all of these domains are regulated.
This view is supported by the following: First, there are no known mutants that affect any of the gap domain shifts individually. Moreover, evidence from an analysis of the dynamical behaviour of gap gene circuits suggests that all trunk gap genes participate in the shift mechanism in an integrated way [7],[22]. Repression between overlapping gap domains (as described in the Introduction) interacts in complex ways with the mutually repressive interactions between Kr and gt as well as hb and kni. In addition, terminal gap genes contribute to domain shifts as well, as we have established in this study. We are far from understanding the causal flow of regulatory information in this system. Our analysis suggests that the posterior hb domain may play a central role in it. All posterior nuclei in the system converge towards an attractor state in which hb is expressed at high level [22]. Moreover, the delayed establishment of its posterior domain coincides precisely with the phase of development when domain shifts occur. Further evidence from tll mutant embryos, which lack a posterior hb domain, will be required to better understand the causal role of this domain in gap gene regulation.
Changes in the regulation of the posterior hb domain also play an important role in the evolution of the gap gene system in dipteran insects (flies, midges, and mosquitoes). Primitive, nematoceran flies such as the psychodid midge Clogmia albipunctata lack posterior hb expression before gastrulation [84], while the posterior domains of gt and hb appear to have swapped positions in mosquitoes [85]. It will be interesting to investigate whether gap gene regulation in these embryos requires Hkb, and how the absence (or change in position) of posterior gap domains affects boundary shifts and their regulation compared to Drosophila. It has been noted previously that shifting gap domains are reminiscent of travelling waves of gene expression in animals with sequential segment determination [1], which is widely assumed to be the ancestral state of segment determination (reviewed in [86]). This suggests that shifting domains are ancestral as well. Understanding how regulatory changes in posterior hb expression affect these shifts in various dipteran species will not only help us understand how the gap gene network performs its patterning function, but also how it evolved. In view of this, our models are an important first step towards an integrative, systems-level understanding of the developmental and evolutionary dynamics of the gap gene network.
|
10.1371/journal.pcbi.1000188 | Cavities and Atomic Packing in Protein Structures and Interfaces | A comparative analysis of cavities enclosed in a tertiary structure of proteins and interfaces formed by the interaction of two protein subunits in obligate and non-obligate categories (represented by homodimeric molecules and heterocomplexes, respectively) is presented. The total volume of cavities increases with the size of the protein (or the interface), though the exact relationship may vary in different cases. Likewise, for individual cavities also there is quantitative dependence of the volume on the number of atoms (or residues) lining the cavity. The larger cavities tend to be less spherical, solvated, and the interfaces are enriched in these. On average 15 Å3 of cavity volume is found to accommodate single water, with another 40–45 Å3 needed for each additional solvent molecule. Polar atoms/residues have a higher propensity to line solvated cavities. Relative to the frequency of occurrence in the whole structure (or interface), residues in β-strands are found more often lining the cavities, and those in turn and loop the least. Any depression in one chain not complemented by a protrusion in the other results in a cavity in the protein–protein interface. Through the use of the Voronoi volume, the packing of residues involved in protein–protein interaction has been compared to that in the protein interior. For a comparable number of atoms the interface has about twice the number of cavities relative to the tertiary structure.
| During protein folding a polypeptide chain takes up a three-dimensional structure that is characterized by close packing of atoms. For cellular processes proteins need to interact, and the binding is also characterized by packing of complementary surfaces. Two types of binding can be envisaged—obligate and non-obligate—the former is exhibited by homodimeric molecules (in which two polypeptide chains are held together in permanent association) and the latter by protein–protein complexes (such as antigen–antibody, enzyme–inhibitor, etc.), which are more transient in nature. Cavities are observed as defects in atomic packing. We present an analysis of cavities within the structure of a protein chain, as well as interfaces formed by the association of two protein chains. For a comparable number of atoms the interface has about twice the number of cavities relative to the tertiary structure. The interfaces contain a higher percentage of larger cavities, which tend to be solvated. We have determined the relationships between the protein volume and the total volume of all the cavities in the structure, the volume of the cavity and the number of atoms (residues) lining it, and the size of the cavity and the number of waters in it.
| Close atomic packing is an important metric for characterizing protein structures—the average packing density for the interior of proteins is similar to that for crystals of small organic molecules [1]. While the average value of packing density in the protein interior is close to 0.75, the efficiency may not be uniform over the whole structure, the density varying in the range 0.66 to 0.84 [2]–[4]. The localized defects in packing show up as cavities [5], and when present they can reduce the stability of the structure [6]. Protein binding has many similar features common to folding, such as the presence of a core in the interface region [7],[8] and complementarities of chemical characteristics of residues in contact across the interface and the nature of the specific interactions linking them [9]–[11]. Although the surfaces that form the interface in protein-protein interaction have complementary shape [12],[13], an issue that has not been addressed is whether the interface can harbor cavities, and their features relative to those present in the protein interior.
Voronoi [14] procedure has been used to assign a unique volume to individual atoms in a collection of atoms, such as in proteins. This has been used to calculate the volumes occupied by amino acid residues and their variation at individual sites [3], [15]–[17]. As to associate a volume of space to an atom the procedure relies on the location of all its neighbors, it works well when applied to atoms in the protein interior. Distinct from the surface atoms those in the interface have surrounding atoms from the interacting protein chains, and as such one can calculate the Voronoi volumes associated to interface atoms [18], and compare these to those in the protein core.
Cavities in structures have also been looked into from the perspective of protein hydration [19]–[21]. Water molecules are also located in the interfaces [22]. Buried water molecules are often conserved among members in a homologous family and are integral structural component of these proteins [23]. When located in cavities they can compensate for the destabilization of reduced hydrophobic and van der Waals interactions [24]. It is of importance to know the average volume occupied by water molecules in protein interior and interfaces and the nature of their interactions with the surrounding protein atoms.
Clefts or pockets on the surface are important for molecular recognition and protein function [25]. Distinct from them are the cavities, defined as enclosed space in the interior of the protein. However, they are sometimes considered together, for example, for defining interior and surface packing densities [26]. Internal cavities have been analyzed separately for monomeric proteins [19],[21], as well as protein interfaces [27], but no attempt has been made to generalize their features from a common perspective. In this work we use a vastly enlarged repertoire of structures to quantify the geometrical characteristics of the cavities found in protein tertiary structures and interfaces—the latter being of two types—those involving obligate homodimeric assemblies and the non-obligate protein-protein heterocomplexes. Other features studied are the occurrence of solvent molecules in the cavities and their hydrogen bonding, the participation of different secondary structural elements in the cavities, the environment of cavity water in the interface, etc. Rather than being mere “packing defects” cavities are also known to play a role in assisting conformational changes between domains or subunit interfaces and in controlling binding and catalysis [28]. Thus a comprehensive analysis of cavities would provide insight into our understanding of protein structure and function.
Cavities and atoms considered in the analysis are defined in Figure 1. For an interface cavity at least 20% of the CL (cavity-lining) atoms ought to belong to a different subunit. If the contribution is less, the cavity is assumed to be part of the tertiary structure of a subunit. The cavities of a single subunit of homodimers and those of monomers were pulled together to constitute the Ter_str (tertiary structure) cavities. The reasons for leaving out individual subunits of protein-protein heterocomplexes are: in some complexes one of the protein components may itself be multimeric; in a few, one component may be rather small to be considered a ‘typical’ protein.
A total of 3384 Ter_str cavities were detected within 219 individual subunits (5 had no cavity) (Table S1). The homodimer dataset contained 4438 cavities, out of which 615 (14%) are Inter_H (occurring between the homodimer subunits) and the remaining 3823 (86%) occur within (contributing to Ter_str). About 8% interfaces and 2% subunits in homodimers are without any detectable cavity. Protein-Protein complexes had 3944 cavities in total, of which 431 (11%) are Inter_C (occurring in heterocomplex interface). Consideration of volume indicates that 24% and 14% of the total cavity volume in the homodimer and complex datasets, respectively, are contributed by interface cavities. ∼20% interfaces in protein-protein complexes are devoid of detectable cavities. The detailed information on cavities in individual PDB entries is provided in Table S2.
On average ∼15 cavities occur in the individual subunits in the monomer and the homodimer datasets and consequently these two categories were considered together to represent Ter_str (Table 1). Compared to the tertiary structure the interfaces have about a third and a sixth number of cavities in homodimers and complexes, respectively. However, the total cavity volume is reduced to two-third and one-fifth in the two types of interfaces, indicating that the cavities in interfaces are larger in size than those in the tertiary structure.
Proteins are of different sizes; besides there is a lot of variation in the numbers of atoms constituting the tertiary structure and the interface. Consequently, we have also expressed the number of cavities and their total volume relative to an average-sized protein of 2000 atoms. Compared to Ter_str the interfaces have about 1.6 times the number of cavities, but the increase in total volume is 3.4 and 2.1 times in homodimers and heteocomplexes, respectively, again indicating the larger size of cavities in the interface, especially for homodimers. In general, the packing of residues in the interface leaves more cavities compared to that in the tertiary structure.
The total volume of Ter_str cavities in a subunit is well correlated to the size, as given by the protein volume (or the number of atoms) (Figure 2A). If the size of an interface is defined by the number of atoms belonging to it, the correlation with the total cavity volume is poor for interface cavities (Figure 2B and 2C). However, when individual cavities are considered the correlation of volume is very strong with both the numbers of cavity lining atoms and residues in all the three categories of cavities (Figure S1). Both linear relationships and equations using power law can fit the data equally well (Table 2), but the former would suggest a negative cavity volume when the number of CL atoms/residues is <4. Using the latter set of equations about 5 atoms or 4 residues are needed to enclose a volume (∼11.5 Å3) large enough to accommodate one water molecule.
The histogram of the distribution of volume in three different classes of cavities is shown in Figure 3A. Interfaces contain higher percentage of larger cavities (14.3% Inter_H and 10.2% of Inter_C with volume>100 Å3) than tertiary structure. The cavities were further divided into empty and solvated cavities and the distribution of their volume (Figure 3B and 3C) indicates that the large cavities (volume>50 Å3) are usually solvated. 50% of all Ter_str cavities are solvated, the corresponding values for Inter_H and Inter_C being 61% and 62%, respectively. The percentages calculated based on volume are 61, 83, and 79%, respectively for the above three categories. Examples of water molecules in cavities belonging to the tertiary structure and interface can be seen in Figure 4B–D.
Rvs (defined in Methods) indicates how spherical a cavity is—for a perfect sphere the value would be 1.0, and would reduce in value as the cavity deviates from being spherical. The distribution of Rvs (Figure 5A) indicates that more than 50% of all cavities are nearly spherical. To investigate if the aspherical shape of the cavity can result from the size we plotted the distribution for the cavities having volume greater than 100 Å3 (Figure 5B). A peak near 0.75 indicates that such cavities are quite irregular in shape. Two small, spherical cavities (labels: 1 and 2) are illustrated in Figure 4A and 4B, in which the largest cavity (label, 5) deviates from the spherical shape.
As cavities are embedded within the protein structure (or interface) we compared the distribution of the CL residue types with that observed over all the proteins (or interfaces). Amino acid preferences for the CL and NCNS (non-cavity-non-surface) regions across all the three classes are shown in Figure 6A and 6B. A large, positive (or negative) value indicates preference (or avoidance), and a value close to zero suggests an occurrence close to the general population. Charged residues (Lys, Glu, Asp, and Arg) are avoided in general. Ter_str cavities prefer hydrophobic residues, such as Cys, Leu, Ile, Phe, Met, and Val. The preference for the branched aliphatic side chains seems to be the common feature for all the categories of cavities. However, in contrast to Ter_str, interface cavities avoid Cys, Phe and Trp, and prefer Thr, Gln, Gly—possibly due to a higher percentage of interface cavities being solvated. Unlike the CL region the trend in propensities is quite similar in the NCNS region in both types of interfaces and tertiary structure.
The amino acid preference for solvated and empty cavities across all three cavity classes is shown in Figure 6C and 6D. Residues which are preferred in empty cavities are Leu, Ile, Met, Phe, and Val, while Gly, Thr, and Tyr are more preferred in solvated cavities. Additionally, Ser is also found in greater number in the solvated cavities of interfaces in heterocomplexes, and His in those of homodimers.
The propensities of different atom types to occur in the CL, NCNS regions, solvated and empty cavities in the three cavity classes are shown in Figure 7. There is a distinct pattern in the atom preference for Ter_str cavities—main-chain atoms (C, N, and CA) are disfavored and the side-chain atoms (aromatic carbon, hydroxyl oxygen and amide nitrogen) are favored in CL as compared to NCNS regions. In the NCNS region all three categories exhibit similar features, for example, polar side-chain atoms (Oa, Oh, Na and Nc) are not preferred and the preference is for C, N, CA and aliphatic carbon (Cc). Polar atoms like oxygen and nitrogen are more preferred in solvated cavities than empty cavities (Figure 7C and 7D). Figure 4D illustrates a cavity with two water molecules, and out of six CL atoms 2 are oxygen and 3 nitrogen.
There is a considerable variation in the cavity volume as a function of the number of water molecules contained in it. To discern any underlying trend we considered the Ter_str cavities, averaged the cavity volume containing a particular number of the solvent molecule (Figure 8) and based on the average numbers one can derive a linear relationship. Roughly, one water molecule can be accommodated in a volume of 15 Å3 (observed value), and an increment of ∼40–45 Å3 is needed for each additional molecule. On average a water molecule participates in 3.4 hydrogen bonds (the number includes those to other water molecules also; if hydrogen bonds to only protein atoms are considered the number is 2.6 for Ter_str cavities and 2.3 for interfaces). 15 Å3 is about the smallest volume that can enclose a water molecule, and such a volume would need about 5 CL atoms (based on equations in Table 2). Figure 4D and 4E, however, provides an example where a rather small cavity had six CL atoms, which could enclose two water molecules that participated in 4 and 2 hydrogen bonds, respectively.
We first compare the Voronoi volume of the NCNS atoms in the interface to those in the protein tertiary structure (values are provided in Table S3). Most of the 13 atom types show an increase of value in the interface, though the change is usually <5% (Figure 9B). It should be mentioned here that for simplification we have grouped atoms together, for example all the aromatic atoms as Cr. However, it is known [17] that there can be some variation between the volumes of these atoms within a given aromatic residue or between any two of them. As such the result would be affected by the atom composition in the datasets. Under these limitations, cases where the difference is more are worth mentioning. In complexes, S and aromatic atoms have smaller values in the interface, indicating that these are better packed relative to the tertiary structure. Fleming and Richards [4] observed that in protein structures Cys and aromatic residues are better packed than the aliphatic ones. It appears that these residues (containing atoms types Cr and S) are still better packed in interfaces. On the other hand, N of Lys and Arg are lesser packed. For homodimers, CB atoms that link the main chain to the functional part of the side chain are also packed less efficiently.
As expected, if we compare the CL atoms instead of the NCNS atoms, there is an increase in volume (11–35%) compared to the atoms in the tertiary structure (Figure 9A). From Figure 9C one can see the difference in the Voronoi volume of CL atoms in solvated cavities, calculated including and excluding water molecules (blue and red bars, respectively). The difference in the volume of the polar atoms is to the extent of 12–25% as compared to 5–7% by the non-polar atoms. However, on including water the values of the polar atoms come to within 5%, indicating that the cavity water molecules are located closer to these CL atoms. One would have expected the bars corresponding to the empty cavities should match with the ones calculated without considering waters for the solvated cavities. But this is not quite correct, as they tend to have different sizes (solvated ones are bigger) and the propensities of atom-types (for example, compare Cr in Figure 7C and 7D) to occur in them are also different.
The percentage composition of occurrence of CL atoms, as well as the ones in the whole data set, in three types of secondary structural elements is provided in Figure S3 and the propensities calculated from these numbers are shown in Figure 10. Strands are preferred in all three cavity classes, more so in Ter_str and Inter_H. Structures other than helices and strands are less inclined to form cavities. Two examples of cavities being located on top of β-sheets can be seen in Figures 11 and 12A. Even the structure shown in Figure 4C has 18 cavities (out of a total of 52) having more than 50% CL atoms coming from β-sheet. There is not much distinction between the cavity types based on the occurrence of the main- and side-chain atoms—Figure S4 indicates that when a helix or sheet contributes to a cavity, ∼70% of the atoms are from the side-chain; however, for ‘Others’ the value comes down to the range 56–63%.
Hubbard & Argos [27] analyzed three classes of cavities: within domains, between domains and between protein subunits. Ter_str cavities considered here would include the first two classes, whereas the interfaces between subunits in obligate homodimers and protein-protein heterocomplexes substituted the last class. Interdomain and intersubunit cavities were on average found to be larger than those located within domains [27]. Our results (Figure 3A) comparing cavities in the tertiary structures and interfaces indicate that interface cavities, especially Inter_H, are indeed larger. However, if we distinguish between intradomain and interdomain cavities (of the 219 individual subunits considered by us 158 were single domain proteins and the rest multidomain) there is not much difference in the distribution (Figure S2). It can be mentioned that cavities with atomic surface components arising from more than one domain were deemed to be interdomain in the earlier study; however, we used a more stringent criterion of having at least 20% of CL atoms coming from a different domain. Nevertheless, Figure 2A indicates that the total cavity volume is a function of the protein volume, irrespective of the number of domains present.
It has been reported that the wide cavities form 0.002–1.55% of the volume of a protein (in a dataset of 75 monomeric proteins); however, no quantitative relationship could be established linking the two [21]. A linear relationship was noted between the number of voids and pockets plotted against the number of residues in each protein, although the total pocket volume did not correlate well [26],[29], possibly because no distinction was made between single and multiple subunit proteins, the latter containing tunnels or holes of large size [26]. From our analysis we could derive a linear relationship between the total volume of the cavities present and the total protein volume (Figure 2A). From this one can derive that for two proteins of volume 30,000 and 50,000 Å3, the cavities will constitute 0.56 and 0.99% of the volume (the two values are 0.61 and 0.91, using the power law). The observed minimum and maximum values were 0.06 and 2.26%, respectively. Cavities are usually located close to the protein surface—considering the CL atoms, most of them belong to the surface of the molecule.
Cavities were found to cover 10% of a typical interface [27]. Comparing the number of CL atoms to the total (Table S1b) we find that 5.5% atoms of the tertiary structure and 13.8 and 10.5% of homodimeric and hetercomplex interfaces form cavities. That for a given number of atoms the interfaces have about twice the number of cavities as the tertiary structure can also be seen from Table 1.
For some structures the resolution of the data may be rather low, or the quality of the electron density too poor for the bound water molecules to be seen. Figure S5 indicates that there is an increase in the number of solvated cavities as the resolution improves from 2.5 Å till about 1.8 Å. If the water molecule is partially or completely disordered it cannot be located. Even with high resolution data detection of water molecules in cavities, especially if they are mobile due to the hydrophobic nature of the cavity, is rather tricky by conventional crystallographic analysis that neglects low resolution data [30] and as such, the average number of water molecules obtained could be an underestimate. Nevertheless, the average number of hydrogen bonds involving water in the solvated cavities—the number is 2.6 with protein atoms, and 3.4 if hydrogen bonding with other water molecules is also included—matches with the typical value of 3 hydrogen bonds made by a buried water molecule reported in literature [20],[21],[31]. The cavity volume needed to enclose one water molecule is ∼15 Å3, however, each additional water requires an extra volume of ∼40–45 Å3 (Figure 8).
The propensity of the secondary structural elements to be associated with cavities indicates that β-strands have a high tendency and the non-regular regions (‘Others’) are disfavored (Figure 10). Interestingly, the packing densities of residues in turns, helices and strands were found to be 0.794, 0.744, and 0.723, respectively [4], indicating the β-strands to be packed least efficiently, possibly due to the greater occurrence of cavities associated with them, examples of which can be seen in Figures 11 and 12A. Loops and turns with higher flexibility can adjust the structure locally to avoid/minimize any local packing defects.
It has been suggested that Cβ atoms do not cover an antiparallel β-sheet by a tightly packed layer, leaving holes equivalent to the size of a methyl group or water molecule [32]. These holes are possibly not included in our analysis because of the volume cut-off used in the definition of cavities. Additionally, these would have had all the CL atoms residing on the β-sheet; however, the percentage of cavities exclusively lined by β-sheet atoms is very low (<5%). The higher involvement of β-sheet residues in lining the cavities may have implications for the energetics of interaction. It has been observed that for protein-protein interactions, those having interfaces mostly made up of β-sheet have, on average lower free energy of binding compared to those having α- or αβ (mixed) classes of interfaces (Guharoy and Chakrabarti, unpublished). This observation may be understood in terms of the lowest packing efficiency of interfacial β structures, leading to lower van der Waals contacts and therefore lower binding free energies as well.
Figure 4A shows that when there is a surface groove that is not matched by a bulge on the surface of the interacting protein this would result in the formation of a cavity in the interface. Water molecules in the groove cannot be squeezed out and remains trapped inside the interface. When we analyzed if the water molecules can have direct hydrogen bond contact with both the subunits (Table 3) we observed that such molecules are just 37% and 51% in Inter_H and Inter_C, respectively, a smaller number (10% and 5%) of water molecules do not form any bond with either subunit. Indeed, one can see from Figure 12A, where the cavity can be considered as a casket of water molecules, the majority of which form hydrogen bonds between themselves. Even when the cavities contain one or two water molecules, the large size of the cavity may preclude the solvent molecules to interact with both the protein components. However, if we consider contacts (instead of hydrogen bonds) made with both the sides, a greater number (72% and 84%) of water molecules bridge the two subunits.
For the water molecules having direct hydrogen bonding with both the protein subunits we considered the involvement of main- and side-chain atoms and how important the solvents are in neutralizing the destabilizing effect of like-charges from the two subunits coming close to each other. It appears from Table 4 that water molecules sitting between like and opposite charges (in the side chain) occur to similar extent in Inter_H, but these are in 3∶4 ratio in Inter_C. A residue close to the two-fold axis in homodimeric interfaces can be in contact with the same residue from the other subunit – the so-called self contacts [33], which may explain some of the occurrences of like charges around water molecules in Inter_H.
The packing density at interfaces has been computed by comparing the Voronoi volume of the buried atoms in the interface to the reference atomic volume [34]. Such a plot is shown in Figure S6, which also includes the distribution for the atoms in the tertiary structure in individual files—as a reference for the normal distribution. When Vr is larger than unity it indicates that the packing density at interfaces is lower than that in protein interiors, and a smaller value indicates a higher density. The average values of Vr for the interfaces in homodimers and heretocomplexes are slightly higher than 1.01(±0.06) reported in [34] for heterocomplex interfaces. Overall, the volumes of the interface atoms are within 3% of those in the protein interior.
The existence of any small molecule, other than water, in the cavities was found out (Table S4). In about 30% cases only 1–3 atoms of a much larger ligand are found to be inside the cavity, which are usually <20 Å3 in volume. These cavities cannot be considered as having a small molecule entrapped. Heterocomplex interfaces have just two cases where molecules used in the crystallization procedure found their way into the cavity. In general, biologically relevant molecules are not found in interface cavities—only two cases of cofactor molecules are found in Inter_H cavities. In the tertiary structure, there is an example of Mg ion being located in a volume of 16 Å3; cavities having Ca ion usually have a volume in the range of 17–18 Å3 (one example is shown in Figure 12B) and a K ion is observed in 20 Å3. Metals such as Hg, Ni, and iron-sulfur clusters occupy a much larger volume. Water molecules usually accompany the ligand in the cavity; the largest of such a cavity is displayed in Figure 12C.
In summary, in this work we have delineated the total volume expected to be occupied by cavities in a protein or a protein-protein interface of a particular size. A quantitative relationship has been derived for the volume of a cavity and the atoms/residues lining it. Of the secondary structural elements, β-strands have a higher inclination to be associated with cavities. For a comparable ensemble of atoms the interfaces contain about twice the number of cavities relative to the tertiary structure. It has been shown recently that a cavity of an appropriate size is the basis of peptidyl-prolyl-isomerase (PPIase) activity of an important class of enzymes (human FK506-binding protein 12) and that it is possible to create artificial PPIase activity by introducing such a cavity on barnase, a bacterial nuclease [35]. A comprehensive understanding of the features of cavities in protein interiors and interfaces, as presented here, would facilitate such protein design experiments.
Atomic coordinates of the proteins were extracted from the Protein Data Bank (PDB) [36]. The dataset consisted of 97 monomeric proteins [13], 122 homodimers [8] and183 protein-protein complexes [37], mostly determined to a resolution of 2.5 Å or better (only16 structures are in the resolution range 2.5–3.0 Å). 219 independent subunits from the first two categories were used to identify cavities in the tertiary structure. The atoms that lose at least 0.1 Å2 of the accessible surface area (ASA) in the complex/dimer structure as compared to that in the isolated subunit were considered as interface atoms [7],[8]. The calculation of protein volume was done by ProGeom, based on the Alpha-Shape theory (server: http://nook.cs.ucdavis.edu/~koehl/ProShape/download.html).
Quite a few algorithms/softwares exist for the calculation of cavities—VOIDOO [38], MS package [39],[40], VOLBL [41],[42], CAST [29] (now rechristened as CASTp), a Monte Carlo (MC) procedure [43], etc. Of these the last two performed in a more consistent way [43]. For our work the cavities for each protein are identified using the CASTp (Computed Atlas of Surface Topography of proteins) server [44] located at http://sts.bioengr.uic.edu/castp/. The basic ingredients of computational geometry applied in CASTp are: Delaunay triangulation, alpha shape, and discrete flow [45]–[48]. CASTp provides a full description of protein pockets and cavities, including volume, surface area, protein atoms that line the concavity, and features of pocket mouth(s) including identification of mouth atoms as well as measurement of mouth area and circumference. The default probe radius of 1.4 Å has been used for our calculations.
Three classes of cavities were identified: (a) Ter_str (cavities in monomeric proteins and within one subunit of homodimeric proteins); (b) Inter_H (those within homodimer interfaces); and (c) Inter_C (within protein-protein complex interfaces).
Surface pockets and cavities belonging to the subunit or interface are illustrated in Figure 1. Any residue contributing one or more atoms to the cavity-lining (CL) region is considered as a CL residue; the same is true for the NCNS (non-cavity-non-surface) region. Interface cavities should have at least 20% of the cavity-lining atoms from a different subunit. The same condition was also used to identify if any Ter_str cavity belonged to the interdomain region, after identifying the individual domain residues from SCOP [49]. As we have used the option in CASTp that defines cavities based on molecular surface (rather than ASA), a few atoms not identified by us as belonging to the interface were also found lining the interface cavities and these were counted as being associated with the cavity (as well as the interface). Only the cavities with volume>11.5 Å3 (the volume of a probe with radius 1.4 Å) were retained for analysis. Further the cavities were classified as solvated or empty based on the presence or the absence of crystallographically determined water molecules in them. Cavities were considered for the existence of embedded water molecules starting from the smallest one. When two cavities, one small and the other large and irregular have some common CL atoms, there could be ambiguity is assigning a water molecule that may lie close to the shared atom(s). As such the cavities were considered in the ascending order of volume. The location of water in a cavity was found out as follows. (i) It has to be within 4.5 Å of a CL atom. (ii) If such water exists, all CL atoms within 4.5 Å from the water are found. (iii) If the distance from the center of mass of the cavity to the water molecule is less than that to any of the CL atoms in contact (as obtained in ii), the water is assumed to belong to the cavity. The existence of any ligand in a cavity was found in a similar fashion. Hydrogen bonding involving a water molecule (to protein atoms, as well as to other water molecules in the cavity) was determined using HBPLUS [50]. The surface representation of the cavities was made using MSMS [51] and displayed with VMD [52].
Based on chemical characteristics the atoms in the PDB files were grouped into thirteen classes. Following are the atom labels (and their definition). N, CA, C, O, CB, S (sulfur of Met and Cys), Oh (the hydroxyl group of Ser, Thr and Tyr), Oa (both the carboxylate oxygen atoms of Asp and Glu, and the amide oxygen of Asn and Gln), Na (the amide nitrogen of Asn and Gln), Nc (side-chain N atoms of Lys and Arg), Nr (ring N atoms of His and Trp), Cr (aromatic C atoms of Phe, Tyr, His and Trp), Cc (aliphatic C atoms excluding CB of Val, Ile, Leu, Met, Lys, Pro, Gln, Glu, Arg and Thr). In the first 5 cases the labels are the same as the atomic labels used in PDB.
The Voronoi [53] procedure for the determination of volume of atomic groups was applied to proteins by Richards [2]. By constructing the minimally sized polyhedron (called a Voronoi polyhedron) around each atom, this procedure allocates the space within a structure, to its constituent atoms. The original program, as modified and extended by Harpaz et al. [54] and Voss et al. [55] (available at http://www.molmovdb.org/geometry/ [17]), has been used in this study. Two parameters need to be provided for the program – the atomic van der Waals radii and Voronoi plane positioning method (method B used).
The propensity of a residue to be in the CL region is given as ln P, whereNx is the number of atoms of amino acid residue of type X lining the cavities and ∑Nx is its total number in the dataset (consisting of all the subunits for Ter_str, and all the interfaces, for Inter_H and Inter_C); Na and ∑Na are the corresponding numbers considering all residue types together. This method is based on counting the atoms, rather than residues, as it is supposed to provide values that are independent of the size of the residue [56]. The propensity was also calculated in a similar fashion considering different types of atoms (instead of residues), as also for the occurrence of secondary structural elements (helix, strand and the rest, termed ‘Others’) lining the cavities. Secondary structure assignments were made using the DSSP program [57].
Rvs provides an estimate of the surface:volume ratio for a cavity relative to that for a sphere having the same volume as the cavity. The following formula is used for its calculation:
|
10.1371/journal.pgen.1004369 | Recent Mitochondrial DNA Mutations Increase the Risk of Developing Common Late-Onset Human Diseases | Mitochondrial DNA (mtDNA) is highly polymorphic at the population level, and specific mtDNA variants affect mitochondrial function. With emerging evidence that mitochondrial mechanisms are central to common human diseases, it is plausible that mtDNA variants contribute to the “missing heritability” of several complex traits. Given the central role of mtDNA genes in oxidative phosphorylation, the same genetic variants would be expected to alter the risk of developing several different disorders, but this has not been shown to date. Here we studied 38,638 individuals with 11 major diseases, and 17,483 healthy controls. Imputing missing variants from 7,729 complete mitochondrial genomes, we captured 40.41% of European mtDNA variation. We show that mtDNA variants modifying the risk of developing one disease also modify the risk of developing other diseases, thus providing independent replication of a disease association in different case and control cohorts. High-risk alleles were more common than protective alleles, indicating that mtDNA is not at equilibrium in the human population, and that recent mutations interact with nuclear loci to modify the risk of developing multiple common diseases.
| There is a growing body of evidence indicating that mitochondrial dysfunction, a result of genetic variation in the mitochondrial genome, is a critical component in the aetiology of a number of complex traits. Here, we take advantage of recent technical and methodological advances to examine the role of common mitochondrial DNA variants in several complex diseases. By examining over 50,000 individuals, from 11 different diseases we show that mitochondrial DNA variants can both increase or decrease an individual's risk of disease, replicating and expanding upon several previously reported studies. Moreover, by analysing several large disease groups in tandem, we are able to show a commonality of association, with the same mitochondrial DNA variants associated with several distinct disease phenotypes. These shared genetic associations implicate a shared underlying functional effect, likely changing cellular energy, which manifests as distinct phenotypes. Our study confirms the important role that mitochondrial DNA variation plays on complex traits and additionally supports the utility of a GWAS-based approach for analysing mitochondrial genetics.
| Mitochondria are the principal source of cellular adenosine triphosphate (ATP) generated through oxidative phosphorylation (OXPHOS), which is linked to the respiratory chain. In humans, thirteen OXPHOS proteins are synthesised from the 16.5 Kb mitochondrial genome (mtDNA). MtDNA has accumulated genetic variants over time, and being strictly maternally inherited, undergoes negligible intermolecular recombination. As a consequence, ancient variants extant in the human population define haplogroups that have remained geographically or ethnically restricted [1]. Work on European haplogroups has shown that some polymorphic mtDNA variants affect mitochondrial function [2], [3].
Given emerging evidence that mitochondria play a key role in several common diseases, it is likely that variation of mtDNA could alter the risk of developing different human disorders. Early mtDNA genetic association studies were under-powered, and the vast majority have not been replicated [4]. However, some recent large studies have found replicable associations with specific human diseases [5]–[11], most notably in sporadic Parkinson's disease [12]–[14]. These observations implicate mtDNA as part of the “missing heritability” of complex human disease traits.
Ultimately, mtDNA codes for a limited number of proteins that affect the same common pathway of energy production implicated in several human diseases. It is likely, therefore, that functional genetic variation of mtDNA will have impact on more than one disease – but this has not been directly studied before. To test this hypothesis, we analysed mtDNA SNP data from 51,106 subjects from the Wellcome Trust Case Control Consortium, comparing genotypes from 11 major diseases: ankylosing spondylitis (AS, n = 2,005), ischemic stroke (IS, n = 4,205), multiple sclerosis (MS, n = 11,377), Parkinson's disease (PD, n = 2,197), primary biliary cirrhosis (PBC, n = 1,921), psoriasis (PS, n = 2,622), schizophrenia (SP, n = 2,019), ulcerative colitis (UC, n = 2,869), coronary artery disease (CAD, n = 3,215), hypertension (HT, n = 2,943) and type-2 diabetes (T2D, n = 2,975) to three independent control groups genotyped on the same platforms (WTCCC-58C, n = 2997, WTCCC-NBS, n = 2897 and WTCCC2-MetabaloChip, n = 5841).
After applying stringent quality control measures (Supplementary Materials, Table S1 & S2), we initially compared the two healthy control groups using PLINK v2.050 [15] (Supplementary Materials, Figure S1), and found no significant difference in allele frequencies. We therefore merged control groups genotyped on the same platform for all subsequent analyses as follows: WTCCC-Control-1, WTCCC-Control-2 and WTCCC-Control-3 (Supplementary Materials, Table S2).
Cluster plots produced by principle component analysis (PCA) revealed no significant population stratification when comparing either: datasets from the same array or array-specific control datasets (Supplementary Materials, Figure S4).
We then compared genotyped SNPs in each disease group to platform-matched control datasets using PLINK v2.050 [15] (Figure 1 & Supplementary Materials, Table S3). This confirmed previously reported associations at the low-resolution haplogroup level [5], [12], [16], [17], endorsing the methodological approach.
Next we performed lexical tree building to identify new associations with phylogenetically related variants, but without basing our anlysis on any prior assumptions related to the published mtDNA haplogroup structure [18], [19]. This method uses fewer SNPs because individuals with missing SNP data cannot be used, but has greated power, and provides graphical summaries of the combinations of SNPs that are associated with increased or descreased risk of disease (Supplementary Materials, Table S4). Lexical tree analysis identified significant relationships between the mtDNA tree structure and schizophrenia, primary biliary cirrhosis, multiple sclerosis (each at p<10−6), ulcerative colitis (p<10−4), and Parkinson's disease (p = 0.004) (Table 1 and Supplementary Materials Figure S3), independently confirming previous haplogroup based association associations [5], [12], [16], [17], and revealing new mtDNA clades associated with several different diseases. The other case-control trees, and comparisons between the different control populations were not significant at the 1% level.
To determine the functional basis of the associations we imputed missing genotypes across the whole mitochondrial genome using 7,729 complete mtDNA sequences. Subsequent analyses were performed on 35,901 European cases and 15,302 European controls, and captured 40.41% of European mtDNA population genetic variation (Supplementary Materials, Figure S2).
In keeping with our original hypothesis, specific variants with predicted functional consequences conferred either an increased risk (Table 2a) or decreased risk (Table 2b) across several different diseases. In addtion, we identified the same allelic-specific associations for different diseases compared to different platform-specific control groups, re-inforcing these findings. Functional variants associated with an increased risk in two or more diseases were limited to two structural genes: MTCYB (m.14793, m.15218) and MTCO3 (m.9477, m.9667). The only non-synonmous protien encoding variant consistently associated with a reduced risk of disease was in MTND3 (m.10398).
We also found evidence of associations across multiple diseases within the non-coding region (d-loop) of mtDNA, and 16S ribosomal RNA subunit genes (Figure 2 and Table 2 and Supplementary Materials, Table. S3). Intriguingly, the same alleles were not associated with all of the diseases we studied, and for two variants (m.11299, m.16294), the same allele had opposite effects for two different diseases (Table 2c).
Overall, the majority of disease-associated alleles conferred an increased risk (61/99), and not a decreased risk (38/99, P<0.001) (Supplementary Materials, Table S3).
Following stringent quality control, our initial analysis confirmed previous associations between mtDNA haplogroups and common disease in a much larger data set. These findings were independentely supported by lexical tree based analysis at higher levels of statistical significance. Subsequent imputation of missing genotypes captured >40% of European mtDNA population genetic variation in 35,901 European cases and 15,302 European controls. By simultaneously analysing eleven, ostensibly unrelated, diseases we identified several imputed mtDNA variants that were associated with more than one disease. The same associations were seen in different disease groups compared to different control groups. This provided confirmatory independent replication of a disease association, and supports our original hypothesis that the same genetic variants of mtDNA contribute to the risk of developing several common complex diseases.
Variants increasing the risk of two or more diseases were limited to MTCYB (m.14793, m.15218) and MTCO3 (m.9477, m.9667), encoding variants in cytochrome b (H16R, T158A) and subunit 3 of cytochrome c oxidase (complex IV, V91L, N154S). Functional variants of MTCYB have previosly been associated with several human phenotypes [20]–[22], but the most compelling evidence of a prior disease association is the increased risk of developing blindness in subjects harboring the mtDNA mutations in MTND genes known to cause Leber hereditary optic neuropathy (LHON), where they synergistically interact with a primary LHON mutation to cause a defect of OXPHOS complex I activity [23]. On the other hand, the only non-synonmous protien encoding variant associated with a reduced risk of several diseases was m.10398 in the MTND3 variant (complex I, T114A). m.10398 occurs twice on the human mtDNA phylogeny (homoplastic on haplogroups J and K), and has previously been associated with a reduced risk of Parkinson's disease [14], [24]. This variant has been shown to reduce complex I activity, cytosolic calcium levels, and the mitochondrial membrane potential [3], [25], [26] and thus may reduce the level of reactive oxygen species, contributing to the underlying disease mechanim of several disorders.Variants in MTCO3 are typically associated with primary mitochondrial disorders [27], [28], but have been also been indentified as risk factors in Alzheimer's disease [29], [30], migrainous stroke [31] and sporadic optic neuropathy [32]. M.9477 and m.9667 are non-synonmous protien encoding variants which are cladally related; present on haplogroup U sub branches (U5 and U5a1b, respectively). Cybrid studies of haplogroup U show a reduction in mtDNA copy number, resulting in a reduction in mitochondrial protein synthesis and complex IV activity [3], [25], impairing energy production and likely contributing to disease.
We also noted disease associations with substitutions in the non-coding region and ribosomal genes (Table 2 and Supplementary Materials, Table S3). Although highly polymorphic at the population level (Figure 2), there is emerging evidence that both regions can have functional effects either through an effect on mtDNA replication, transcription or translation [33], [34], as proposed in Alzheimer's disease [34].
It is intriguing that there were more functional variants associated with an increased risk, than with a decreased risk of disease (Table 2 and Supplementary Materials, Table S3). This suggests that deleterious, novel sub-haplogroup variants have not yet been removed from the population through natural selection, possibly including the younger d-loop variants. This has been observed in the nuclear genome in the rapidly expanding human population [35], [36], implying that the modern human population is far from equilibrium. An alternative explantion is that mtDNA alleles may escape purifying selection because the associated disease phenotype only becomes manifest after female reproductive life.
For two variants (m.11299, m.16294), the same allele was associated with an increased risk of developing one disease, and a reduced risk of developing another (Table 2). Although differences in the sample size post-QC provide one explanation, these findings raise the possibility that different mtDNA-mediated mechanisms are involved in different contexts, perhaps because some variants have a greater impact on bioenergetics, and others on the generation of reactive oxygen species. Alternatively, it is conceivable that the relevance of specific alleles may be context-specific, only excerting a functional effect on a particular haplogroup background [37]. Substantially larger whole mtDNA genome studies will be required to detect clade-specific epistastic interactions if they exist.
In some instances we observed multiple associations with different variants found within the same phylogenetic cluster. For example m.499 (K1a), m.11485 (K1a4) and m.11840 (K1a4a1) are known to reside within subdivisions of the major haplogroup K, and all associated decreased risk of MS and IS. Conversely, m.310 (U4a2) and m.3197 (U5) are distinct subclades of the U associated with increased risk of PS, MS, IS PD AS and UC. Although reassuring from a technical perpective, this illustrates the challenge of mtDNA association studies, where variants with a close ancestral relationship inevitably co-segregate, making it difficult to determine which alleles are responsible for the disease risk.
Finally, analysis of imputed data also revealed several different mtDNA alleles asssociated with different diseases, often reaching high levels of statistical significance (P<10−10, Supplementary Materials, Table S3). However, these findings should only be considered preliminary and require independent replication in other populations (where specific European haplogroup distributions can vary) and thus do not form the major focus of this report.
In conclusion, these findings underscore the role of mitochondrial mechanisms in the pathogenesis of common diseases, and emphasise the importance of incorporating the mitochondrial genome in comprehensive genetic association studies. Although the strict phylogenetic stucture of maternally inherited mtDNA makes it difficult to identify the precise variants responsible, higher resolution genotyping at the whole mtDNA genome level will cast further light on the genetic mechanisms, particularly if recurrent homoplasies independently associate with phenotypes across several clades.
This study used data generated through the Welcome Trust Case Control Consortium. A full list of the corresponding investigators who generated each dataset is available from http://www.wtccc.org.uk/ccc2/wtccc2_studies.html [38]–[45]. Both case and control datasets were downloaded from the European Genotype Archive (http://www.ebi.ac.uk/ega).
Psoriasis (PS), multiple sclerosis (MS), ischemic stroke (IS), Parkinson's disease (PD), primary biliary sclerosis (PBC) and ankylosing spondylitis (AS) patient cohorts were genotyped using the Illumina 610K quad array (Illumina San Diego California USA) and were compared array specific controls, denoted here as WTCCC-Control-1 (combined WTCCC-58C and WTCCC-NBS) genotyped on the Illumina 1.2M Duo platform (Illumina San Diego California USA). Illumina array systems contain 138 mtDNA variants.
Ulcerative colitis (UC), schizophrenia (SP) and their array-specific controls, denoted here as WTCCC-Control-2 (combined, WTCCC-58C and WTCCC-NBS), were genotyped using the Affymetrix SNP6.0 array (Affymetrix, Santa Clara, CA). The Affymetrix SNP6.0 array system contains 445 mtDNA variants.
Coronary artery disease (CAD), Type-2 diabetes (T2D) and hypertension (HT) cohorts and their array specific controls, denoted here as WTCCC-Control-3 (combined WTCCC-58C and WTCCC-NBS), were genotyped using the MetabaloChip array system [46]. The MetabaloChip array system contains 135 mtDNA variants.
To ensure valid comparisons, each disease sample set wasonlycompared to its corresponding control array counterpart(i.e. SNP6.0 cases were compared to SNP6.0 controls).”
Given the case cohort sample sizes post QC (Supplementary Materials, Table S1), the corresponding control cohorts (Supplementary Materials, Table S1), an expected MAF of 0.01, an α = 3.85×10−3 to 3.97×10−4(averaging 13-126 tests dependent upon specific dataset) and disease prevalences of: psoriasis = 2% [47], multiple sclerosis = 1% [48], ischemic stroke = 1% [49], primary biliary cirrhosis = 0.1% [50], Parkinson's disease = 0.3% [51], ankylosing spondylitis = 0.1% [52], ulcerative colitis = 0.1% [53], schizophrenia = 0.33% [54], Type-2 diabetes = 10% [55], coronary artery disease = 3% [56] and hypertension = 30% [57]; we had >80% power to detect an effect size of >1.2 in each cohort (specifically, psoriasis = 79.8%, multiple sclerosis = 93.2%, ischemic stroke = 84.5%, primary biliary cirrhosis = 79.9%, Parkinson's disease = 85.9%, ankylosing spondylitis = 85.4%, ulcerative colitis = 78.9%, schizophrenia = 80.3%, Type-2 diabetes = 85.3%, coronary artery disease = 82.6% and hypertension = 98.7%). Power calculations were carried out using Genetic Power Calculator [58].
Stringent quality control (QC) was applied to each individual cohort (Table S1) [59]. Briefly, each cohort was pruned of missing phenotypes (defined as -9 in the pedigree/sample files). Poorly performing SNPs (genotyped = 0.1[59]), and subsequenctly, samples were removed (individual missingness = 0.1 [59]) using PLINK v2.050 [15]. Additionally non-European mtDNA sequences (defined with m.8701A, m.8540T and 10873T) were also removed [1], [60], [61]. Finally, to verify the quality of genotypes cluster plots of normalized intensity for each SNP were generated using R (http://www.R-project.org) and inspected.
In order to increase statistical power, WTCCC-58C and WTCCC-NBS control cohorts were merged. Initially, we compared the two healthy control groups (Supplementary Materials, Figure S1), and found no significant difference in allele frequencies. Briefly, each control cohort was merged with its array genotyped counterpart (Supplementary Materials, Table S2). As with individual cohorts, MAF = 0.00001, implemented in PLINK v2.050 [15], was used to remove SNPs with missing genotpyes (i.e. call = 0 0). Poorly performing SNPs (genotyped = 0.1[59]), and subsequenctly, samples were removed (individual missingness = 0.1[59]) using PLINK v2.050 [15]. Finally, to correct for control popualtion stratitification, variant frequency was compared between -58C and –NBS using ‘—assoc' PLINK v2.050 (P threshold = 0.05) [15]. Variants with signifcantly different 58C/NBS frequencies were removed. This QC lead to the formation of 3 merged control cohorts: WTCCC-Control-1, WTCCC-Control-2 and WTCCC-Control-3.
Prior to association testing QC'd case cohorts were merged with corresponding QC'd control cohorts (i.e. Multiple sclerosis versus WTCCC-Control-1). Differential missingness tests, which statistically compare the frequency of ‘missing’ genotype data between cases and controls were performed on each case-control comparison [59]. Variants were removed when missingness was significantly different (P = <10−4) [59]. Allelic association was implemented in PLINK v2.050 [15]. Given the discovery nature of the experiment, statistical significance was defined as P<0.05.
Only ancestral Europeans, determined by mitchondrial DNA genotype, were included in this study [1], [60], [61]. Additionally, population structure in each cohort (post-QC) and combined by array type was assessed by principle component analysis (PCA) of mitochondrial DNA variants [62]. Plots were made of the first two components for each array dataset (Illumina = AS, IS, MS, NBS, PBC, PD, PS, WTCCC-58C and WTCCC-NBS, Affymetrix = SP, UC, WTCCC-58C and WTCCC-NBS and Metabalo = T2D, CAD, HT and controls [previously combined WTCCC-58C and WTCCC-NBS]) and separately for the control cohorts for each platform (Supplementary Figure S3). At this resolution, individual PCA cluster analysis showed no significant stratification differences. All principle component scores were calculated in R using the ‘princomp’ function and plotted in R using ggplot (R Core Team 2013) [63].
Imputation was implemented in PLINK v2.050 [15]. Initially a reference panel was constructed. Whole Human mtDNA genome data, n = 18,114 sequences, were downloaded from the National Centre for Biotechnology Information Nucleotide database (http://www.ncbi.nlm.nih.gov/nuccore/), using the keyword phrase ‘Homo [Organism] AND gene_in_mitochondrion[PROP] AND 14000∶19000[SLEN] NOT pseudogene[All Fields]'. Sequences with pathogenic mtDNA variants (available at www.mitomap.org) were removed (n = 458 sequences), non Homo sapien sequences were removed (n = 7). Similar to genotype QC, non-European mtDNA sequences (defined with m.8701A, m.8540T and 10873T) were also removed (n = 7051). Finally truncated mtDNA sequences (<16,500 bp) were removed (n = 663) leaving a final dataset of n = 9,935 sequences. The sequence dataset was aligned using MUSCLE [64], analysed using Haplogrep [65], [66] and subsequently filtered to match the Major European haplogroups (H, V, J, T, U, K, W, X, I, R and N) leaving a final sequence aplosamples and 2,873 variants, representing 100% of of the genetic varation in the reference dataset.
The reference panel was merged with each QC'd case-control cohort in PLINK (v2.050),[15] invoking ‘—flip-scan' to detect and correct any stranding issues. Imputation association testing was carried out using ‘—proxy-assoc’ and, in order to assess the imputation performance, ‘—proxy-drop’.[15] Significant SNPs associations with >99% of samples imputed, number of proxy SNPS >3, a MAF >0.01 and a content metric >0.8 were retained.[15] Given a popualtion size of 7,729 and total genotypic information of 2,873 as 100%, imputation of alleles with MAF>0.0 captures 40% of total mtDNA genetic variabilty (Figure S2).
Cicularised Manhattan plots were generated using code adapted from http://gettinggeneticsdone.blogspot.co.uk/2013/11/a-mitochondrial-manhattan-plot.html, solarplot.R and ggplot2 (http://ggplot2.org/).
Lexical tree analysis was performed in R (R Core Team 2013) [63] using a custom library (snptree, publically available from http://www.staff.ncl.ac.uk/i.j.wilson/). This analysis was performed on the Illumina 610K quad array, the Affymetrix SNP6.0 and the MetabaloChip datasets independently. An independent stringent QC was performed, removing in order: the SNPs with a call rate of below 95% or a MAF of below 0.5%, the 2% of individuals with the most missing sites, the bottom 50% of SNPs with the most missing samples at that site, and those individuals with any missing data from the remaining SNPs. Finally, those individuals with haplotypes (defined by all the remaining SNPs) that were not present in controls or had a frequency of less than 5 were removed. This left 27054 individuals on 24 SNPs for the Illumina 610K quad array, 10,745 individual at 15 SNPs for the Affymetrix 6.0 chip and 14,484 individuals at 5 SNPs for the MetabaloChip. The SNPs retained and their minor allele frequencies (MAF) in the control populations are shown in Supplementary Materials, Table S4. A tree structure was contructed for haplotypes made from the retained SNPs by initially grouping all individuals at the root of a tree, and then successively considering all retained SNPs in decreasing order of their minor allele frequency (Supplementary Materials, Figure S3). At each stage, the haplotypes at each leaf node are split with those with the wild type being put on the left branch and those with the mutant allele on the right. This creates a tree with all leaves representing complete haplotypes and internal nodes partial haplotypes. Test statistics were then calculated for each node on the tree. An overall test statistic for the tree was calculated by calculating the the sum of the five largest node values that were not ancestors or descendents of each other. The test statistic was tested for significance by 1,000,000 random permutations of the Case/Control labels.
|
10.1371/journal.pcbi.1006082 | Multiscale modelization in a small virus: Mechanism of proton channeling and its role in triggering capsid disassembly | In this work, we assess a previously advanced hypothesis that predicts the existence of ion channels in the capsid of small and non-enveloped icosahedral viruses. With this purpose we examine Triatoma Virus (TrV) as a case study. This virus has a stable capsid under highly acidic conditions but disassembles and releases the genome in alkaline environments. Our calculations range from a subtle sub-atomic proton interchange to the dismantling of a large-scale system representing several million of atoms. Our results provide structure-based explanations for the three roles played by the capsid to enable genome release. First, we observe, for the first time, the formation of a hydrophobic gate in the cavity along the five-fold axis of the wild-type virus capsid, which can be disrupted by an ion located in the pore. Second, the channel enables protons to permeate the capsid through a unidirectional Grotthuss-like mechanism, which is the most likely process through which the capsid senses pH. Finally, assuming that the proton leak promotes a charge imbalance in the interior of the capsid, we model an internal pressure that forces shell cracking using coarse-grained simulations. Although qualitatively, this last step could represent the mechanism of capsid opening that allows RNA release. All of our calculations are in agreement with current experimental data obtained using TrV and describe a cascade of events that could explain the destabilization and disassembly of similar icosahedral viruses.
| Plant and animal small non-enveloped viruses are composed of a capsid shell that encloses the genome. One of the multiple functions played by the capsid is to protect the genome against host defenses and to withstand environmental aggressions, such as dehydration. This highly specialized capsule selectively recognizes and binds to the target tissue infected by the virus. In the viral cycle, the ultimate function of the capsid is to release the genome. Observations of many viruses demonstrate that the pH of the medium can trigger genome release. Nevertheless, the mechanism underlying this process at the atomic level is poorly understood. In this work, we computationally modeled the mechanism by which the capsid senses environmental pH and the destabilization process that permits genome release. Our calculations predict that a cavity that traverses the capsid functions as a hydrophobic gate, a feature already observed in membrane ion channels. Moreover, our results predict that this cavity behaves as a proton diode because the proton transit can only occur from the capsid interior to the exterior. In turn, our calculations describe a cascade of events that could explain the destabilization and dismantling of an insect virus, but this description could also apply to many vertebrate viruses.
| Many small viruses are minimalistic structures composed of a spherical proteinaceous capsid constituted by several replicas of one or few proteins that enclosing the viral genome. The capsid plays many different roles, such as recognizing its own genome and provoking its encapsidation, protecting and transporting the genome in the extracellular space, and specifically recognizing and invading cells of the target tissue.
The viral life cycle culminates in a mature viral structure that is able to infect new cells. During such processes, capsid destabilization is a key structural change that is required to release the genetic material into the cytoplasm of the target cell. This destabilization can be triggered by interactions with cell receptors or promoted by chemical or physical conditions, such as pH and temperature changes.
The Picornavirales order includes many vertebrates, invertebrates, and plant viral families of small non-enveloped viruses containing a +ssRNA genome [1]. The viruses’ common structural characteristics are an icosahedral oligomeric viral capsid made of 60xT repeats (T = 1 or T = 3) and a capsid size of approximately 30 nm in diameter. The structural proteins are named VP1-4, with VP1-3 exposed to the exterior particle surface and VP4 in the interior in contact with the genome. The pentameric sub-unit composed of 5xVP1-4, called the penton or pentamer, constitutes a capsid disassembly and assembly intermediate. Many of the members of this viral order have been studied extensively, either because they are agent diseases in humans, such as poliomyelitis, hepatitis A, and common cold, or in livestock and crops. One member of the picornavirales family that nucleates insect viruses and has economic and epidemiological importance is Dicistroviridae [2]. Within this family is Triatoma virus (TrV), the type species of the genus Triatovirus, which is the focus of this work. Another member is Cricket paralysis virus (CrPV; type species of the Cripavirus genus), the atomic structure of which demonstrated organizational similarities between these viruses and picornaviruses hosted by vertebrates [3,4].
In this work we model the process by which a viral capsid could sense the environmental pH and the mechanism that could promote the destabilization and opening of the protein shell. The experimental data that guide our calculations come from previous structural, stability and disassemble studies done on TrV [5–7].
These data showed that this viral capsid is stable at acidic pHs [6] while disassembles under alkaline conditions, and that the genome is the molecule that drives its own release by cracking the protein cage [7]. Due to the genome is isolated at the capsid interior, these data suggests that some kind of interchange, either of solvent or solutes, may occur through the protein shell that mine the equilibrium reached during the process of virus assembly.
In this report, we explore the hypothesis of the existence of ion channels along the five-fold symmetry axes in the capsid of small icosahedral viruses [8,9]. We tested this hypothesis concerning the structure of the TrV capsid (Protein Data Bank code, 3NAP) using techniques, such as molecular dynamic (MD) simulations for channel hydration, quantum mechanic (QM) simulations to study proton transport and coarse-grained (CG) simulation of the full TrV capsid to assess structural stability. Overall, this report presents a simulation study that was conducted on a macromolecular system encompassing more than six orders of magnitude in time and size, ranging from subatomic distances to a system representing nearly 6 million atoms.
Our study provides a novel and simple model of the concerted processes leading to TrV genome release, i.e., pH sensing by means of a proton channel spanning the viral capsid along the five-fold axes. Draining of protons from the capsid interior disrupts the electrostatic equilibrium, destabilizing the structure and eventually enabling the exit of the genome.
Using the crystal structure of TrV as an initial model [10], an MD simulation of the solvated capsid in an atomistic representation was performed with the atomic and CG model for solvent molecules in different hydration regions (Fig 1A). A 50-ns calculation showed an inhomogeneous hydration pattern across the five-fold axis of the capsid pore. Hence, to further characterize the solvent inside the pore over a longer time scale, a simulation of a single pentamer around the five-fold symmetry axis (Fig 1B) was carried out at an atomistic level, as described in the Materials and Methods section. This 100-ns calculation showed that the cavity had three regions with a solvation pattern that differed from the bulk solvent. These regions spanned the five-fold cavity along the axis by six annuluses of symmetry-related amino acids (Fig 1C). The middle region located at the narrowest portion of the five-fold cavity remained devoid of water molecules (green zone in Fig 1C). This effect was observed within the cavity lumen surrounded by the two rings formed by Gln 3014 and Val 3012. We will refer to this region as the hydrophobic gate. Analysis of the pore radius indicated that the hydrophobic gate coincides with the maximum constriction point of the cavity, featuring a radius of nearly 0.3 nm. From this point towards the capsid exterior and interior, the cavity widened but did not exceed a diameter of 0.9 nm (blue regions in Fig 1C). Along the entire simulation of 100 ns, the hydrophobic gate of ca. 0.4 nm in length remained dehydrated (Fig 2C). A similar calculation, but with a Val3012Ser mutation, confirmed that solvent depletion was caused by the hydrophobic character of the valine annulus. Indeed, this simulation with a serine at position 3012 resulted in a more hydrated pore, even though the pore radius had the same diameter as in the native structure (Fig 2B).
The electron density map of TrV, computed with experimental X-ray diffraction data, showed a spherical bulge on the five-fold symmetry axis at the level of Gln 3014. This electronic density was attributed to the presence of a putative metal ion [10]. Based on these data, we considered whether a magnesium ion, Mg2+, inside the pore would modify the hydration state within the cavity. To determine energetically favorable ion positions along the five-fold axis, we performed a MD simulation with the umbrella sampling method (USM) [11]. The main result was the extraction of the potential of mean force (PMF), which provided the ΔG for the binding/unbinding process. Fig 2D shows that the most favorable position for this ion is near the Gln 3014 ring, since this point corresponds to a free energy minimum (Fig 2D). At this position, the Mg2+ is coordinated to three Gln 3014 side chain carbonyl oxygen atoms (Oε1) plus three water molecules with octahedral symmetry (Fig 2E inset). Under this condition, several water molecules filled the cavity, thereby chaining the external solvent to the solvent at the inner region (Fig 2C and 2E). We will refer to this water structure as water wire.
Concerning the umbrella sampling simulations, a close inspection of each window showed that the ion was fully hydrated only at the minimum of the energy profile. Under this situation, the hydration of the cavity was complete.
Regarding these calculations, ions initially located close to both cavity ends were expelled out within a few nanoseconds towards the bulk solvent. As expected, ions with an initial position close to the ring formed by Gln 3014 remained stable throughout the whole 100 ns of the simulation. The local energy minimum corresponding to this position was approximately 11.3 kcal/mol (Fig 2D).
Simulations in which Val 3012 was mutated by serine and in which Mg2+ was positioned inside the pore in the minimum observed free energy profile showed immediate hydration of the pore (in approximately 5 ns). However, in the simulation of the wild-type pentamer, the pore was dehydrated during the entire 100 ns of simulation. Thus, the water molecules had sufficient time to enter the hydrophobic region during the length of the wild-type simulation.
We chose Mg2+ to model the unassigned electron density within the five-fold cavity of TrV because this ion is the most abundant among biological divalent cations. However, an equivalent MD simulation performed by replacing Mg2+ with Ca2+ provided similar results.
Starting with the water structure determined by the MD simulations and to simulate a basic pH, one proton (H+) from each of the four water molecules located in the outer zone of the channel were removed (Fig 3A), resulting in a ratio of OH-/H2O = 4/38. Geometry optimization was then performed (see Quantum multilayered simulation in Materials and Methods). The sequence shown in Fig 3 corresponds to different steps obtained during the QM simulation. In this process, one of the OH- captures a H+ of the closest water molecule, which in turn is bonded to another innermost water molecule (Fig 3A–3D). The concerted proton transfer continues until the OH- is bound to the Mg2+ cation (Fig 3D). Up to this point, protons transport (or the corresponding proton hole movement in the opposite direction) occurs downhill in energy. However, the proton hole transferred from the OH- bonded to Mg2+ through two waters located towards the capsid interior is not energetically favored (Fig 3E TS), probably because the Mg2+ cation has a higher affinity for OH- species than for H2O molecules. Nevertheless, the calculated activation energy barrier is low, at 1.7 kcal, approximating the kinetic thermal energy of an atom or molecule at room temperature (3/2 KBT~0.9 kcal). The final configuration after this transition state is shown in Fig 3E. The highest energy requirement was clearly associated with the deprotonation of the nonconsecutive water molecule, that is, the formation of OH- at the second water molecule below the Mg2+. From this geometric configuration, the concerted proton hole transfer occurred downhill up to generate an OH- at the end of the water wire immediately at the frontier of the bulk solvent (Fig 3G).
Two similar simulations, a calculation with the outer three OH- in different positions and the other including only a single OH- located at the outer molecule of the water wire, gave the same concerted proton transfer until the OH- bound to the Mg2+ cation (see S1 Fig).
At this point, we must highlight that during the geometric optimization, the three remaining hydroxyls placed at the outer zone of the capsid channel maintained their identities, indicating that none of the OH- captured any protons from vicinal water molecules. Most likely this happened because hydroxyls interact with waters in their hydration sphere through strong hydrogen bonds and breaking one water bond would require more than 100 kcal/mol [12]. Other authors have previously reported this kind of interaction [13–16]. The same solvation was observed at the end of the proton hole transfer when the hydroxyl was generated in the interior of the channel.
The migration of H+ from an inner water molecule to another located toward the capsid exterior agrees well with a Grotthuss-like mechanism called "Proton Holes" [17]. Fig 4A displays this migration in a sequence of seven steps that covers a pathway of approximately 2 nm along the water wire. In the initial portion of the trajectory, the proton hole executes three energy favored jumps (ca. -3.5 kcal each). It reaches the inner most water molecule that is coordinated to the Mg2+ ion. From this point forward, the proton hole must overcome a low barrier (of ca. +1.7 kcal) that covers two units of the water wire. Finally, the proton reaches the water at the internal bulk solvent frontier through two favored jumps (ca. -1.2 kcal each).
The three situations we analyzed here have in common the position of the outer hydroxyl from which starts the proton hole transfer. Then, these situations illustrate one way how a concerted proton hole transport can occur. If the position of the initial hole-donor would be different, most likely the proton hole transfer through the water wire pathway will looks different, certainly displaying slightly different energy proton-jumps.
A similar simulation was performed but with neutralization of the negative hydroxyl charges by placing four Li+ ions in the outer water bulk zone. This new simulation provided the same result obtained without lithium ions, indicating that the proton migration was due to the pH gradient and not to a charge gradient.
Another simulation was performed by replacing Mg2+ with Ca2+. Additionally, the expected increase in the oxygen cation distances was due to the larger size of calcium (0.24 nm compared with 0.21 nm for Mg2+), and both the structure and the same concerted mechanism occurred in a completely analogous fashion.
To study the influence of external acidic pH, a similar model to that employed in the former situation was designed by replacing the external OH- with H3O+. In this condition, the calculation started with the hydronium ions in equivalent positions to the OH- ions shown in Fig 3A. In the first simulation with a basic external pH, a transfer of the H+ hole occurred; in the reverse situation with an acidic external pH, the inverse process of classical Grotthuss happened, that is, the proton from the hydronium jumped twice to the closest internal water molecules (ΔE ~ -4.1 kcal/mol). Nevertheless, proton internalization did not occur because proton migration stopped immediately before reaching the water molecule coordinated to the Mg2+ cation (Fig 4B). The water wire near the metal ion becomes disrupted due to an increase in the distance between the hydronium and the inner water molecule. This distance is approximately 0.36 nm (Fig 4B). This disruption was most likely due to an electrostatic repulsion between the cation and the hydronium.
Upon encapsidation, the viral RNA is highly condensed due to the neutralization action of counterions such as Ca2+, Mg2+, K+ and Na+. Although under physiological conditions (i.e., ~0.15 M of simple electrolytic ions) other negative ions are also present, like Cl- or OH-, within the capsid of a mature virion the charge balance must be close to neutrality. This equilibrium situation would be maintained after completion of the process of RNA encapsidation and while the virus remains in the intracellular space (with a neutral or lightly acidic pH). When the virus abandons the infected cell, the pH of its environment changes, and upon binding to the receptor of the target cell, the capsid may allow ion permeation. If this occurs, the equilibrium between positive counterions and RNA will be altered. Our QM simulation indicates that a high pH at the capsid exterior and Mg2+ inside the cavity would drive a Grotthuss-like effect of proton exit, producing an internal charge imbalance and structural instability. This charge excess can impact to one or more of the multiple molecules located at the capsid interior: OH-s, Cl- and other anions in solution, and deprotonable amino acids and/or RNA bases located at fixed positions. For instance, the amino termini of VP1-3 proteins are part of the internal capsid wall in direct contact with the RNA [10], and these termini could be deprotonated when the pH is of about 8.2 [18]. These 180 titratable groups are located around the three- and five-fold axes, and distributed in a thin spherical shell (see S2 Fig and S1 Table). If we also count the 60 copies of the N-teminus of VP4 peptide that are disordered and in contact with the genome [5], all make a total of 240 groups that can be deprotonated when the internal pH gets alkaline. Such distribution of chargeable groups embedded into an almost solid matrix could produce a big increase in internal pressure even when the repulsion would happen among few groups situated in a small region. Unfortunately, experimental evidence suggesting which molecular species are predominantly affected is not currently available. Thus, we attempted to phenomenologically model this situation at the CG level. For this aim we first ran a CG simulation of an empty capsid using the method presented in ref. [19]. Then, we generated an electrostatic internal repulsive force by placing Cl- ions, simulating internal charges inside the capsid.
We performed a 1.5-μs long simulation, along which the global structure of the capsid experienced a relatively high change in RMSD of the alpha carbon positions (Fig 5A), which was mainly related to a loss of the initial symmetry. This effect was reflected mostly at the global level and, to a lesser extent, on individual pentons. Indeed, calculation of the gyration radius along the simulation revealed deviations from the experimental value of only ~0.1 nm (Fig 5B).
The stabilized CG structure after 1.5 μs of simulation was used as a starting point for a series of simulations with progressive substitution of supra coarse-grained water molecules (WLS; see S3 Fig), with Cl- ions at the capsid interior being located at random positions. This set up attempts to mimics the conditions after alkalinization has happened. It is worth noticing that this process is intended only to roughly mimic the progressive increment in negative charges at the core. Since the intrinsic entropy loss in the coarse-graining process makes impossible a one-to-one energetic comparison, the number of ions used must not be considered as an actual number of charge carriers but rather as a mean to progressive increase the internal electrostatic repulsion. This charge imbalance produces repulsions among ions, generating an internal pressure. The impact of this pressure was evaluated through an MD simulation in terms of the capsid radius of the gyration (RG) increase. Incorporating less than 200 anions did not result in a significant change in the RG. Only a small oscillation was observed, which was likely ascribable to the re-stabilization of the system after the replacement of water with ions in the interior. After this oscillation, the radius returned to its initial value (Fig 5B). However, further increasing the number of ions resulted in a permanent increase in the RG of the particle, which was likely related to the number of ions impinging on the inner surface. Arbitrarily increasing the number of ions over 200 resulted in large distortions associated with a rapid increase in the RG. At these high values of non-neutralized internal charge, the internal pressure opened crevices that enabled the passage of ions (Fig 5C). Notably, comparison of the conservation of native protein-protein contact inter and intra pentons indicated that the capsid cracked preferentially at the interface between pentons, while the pentons showed a preference for maintaining their structure, which was in agreement with mass spectrometry and AFM nanoindentation experimental data previously reported by our group [7]. It is worth to notice that the CG simulations were performed in absence of Mg2+ (see Methods). However, this showed (a posteriori) not relevant artifacts, as the destabilization of the capsid occurs along the inter- and not intra-pentamer protein-protein interfaces, in agreement with our previous experimental results [7]. This is also in agreement with the absence of electronic density along the five-fold axis found in the structure of the empty capsid [20].
The atomic 3-dimensional structure of TrV capsid determined through X-ray crystallographic techniques at a 2.5 Å resolution (PDB id 3NAP), excluding all solvent molecules, was used as starting point [10].
All simulations were performed with the program GROMACS 4.6.5 [21] using a multiscale approach. We used the Gromos53a6 [22] Force-Field and Single Point Charge (SPC) [23] water model for atomistic detail and the Sirah Force-Field for coarse-grain detail [24]. This multiscale approach uses three levels of representation of the solvent (see S3 Fig). To construct the system, we used the X-ray structure of the capsid solvated with 1.2 nm of SPC water shell. After the first atomistic solvation, a thick shell of 2.5 nm CG solvent [25] was added at the exterior and interior of the capsid. In both solvent regions, water molecules were randomly substituted with Na+ and Cl- ions to neutralize the total charge and to mimic an ionic strength of 150 mM. The space remaining in the virus interior and between the capsid exterior and the box was solvated using a WLS model (see S3 Fig). This simulation was compared with a second set of simulations containing only a single penton (i.e., including one single five-fold symmetry axis) solvated with a 1.2-nm-thick shell of SPC water and the rest of the dodecahedral box/cell solvated using CG solvent. Periodic boundary conditions were applied using a dodecahedral with a sufficient size to maintain a distance of 1.5 nm between the proteins and the cell walls.
The multiscale approach used consisted of simulating the protein pentamer and a thin layer of water (1 nm thick) around it at an atomistic level, while the rest of the water molecules were simulated using the WT4 water model of the Sirah Force-Field. The simulations were performed in a physiological solution of NaCl at a concentration of 0.15 mol/liter and MgCl2 at a concentration of 0.015 mol/liter. In the case of the full capsid, a third layer was added beyond the WT4 layer. For this third level with less detail, WLS water molecules from the Sirah Force-Field (www.sirahff.com) were used.
To analyze capsid dismantling, simulations were performed by transforming the X-ray coordinates of the capsid to CG using SIRAH tools [26]. The CG protein was solvated with a 2.5-nm water shell, the ionic strength was set to 150 mM and the empty space was filled with CG water as in the triple layer case described above. The space remaining in the capsid interior and between the exterior and the walls of the computational box were solvated with supra coarse-grained water molecules (WLS; see S3 Fig). To induce capsid destabilization by electrostatic forces, we performed a series of CG simulations, in which we progressively substituted CG solvent molecules on the interior side with Cl- ions and with Na+ on the exterior side of the capsid to maintain the electroneutrality of the simulation box. Within the coarseness of our simulation approach, Mg2+ ions at the five-fold axis were not considered because of two reasons: i) there was no available CG model for ions coordinated to both water molecules and oxygen atoms from amino acids, and ii) the CG model for ions were larger than the atomistic models because they included the first solvation shell and there was no space to harbor a structure such as a Mg2+ coordinated to six molecules within the hydrophobic gate.
A Verlet scheme was used to generate the neighbor search list. Long-range electrostatic interactions were calculated with the Particle-mesh Ewald method [27] and Lennard-Jones repulsion and dispersion terms with a Cut-Off scheme. All systems were minimized using a combination of Steepest Descent and Conjugate Gradient algorithms until the potential energy reached a value equal to or lower than 239 kcal mol-1 nm-1.
After a thermalization step, the system was ready to initiate production simulations in an ensemble at constant temperature and pressure (NPT). Temperatures were maintained at 310 K with a time constant of 0.5 ps, and pressure was maintained at 1 bar with a time constant at 0.4 ps. Atomistic simulations were carried out with a time step of 2 fs, while 20 fs was used for the CG simulations.
The stability of Mg2+ ions within the five-fold axis was assessed by calculating the PMF of the transition of the ion along the pore. To achieve this goal, we used USM. The ion was positioned at regular intervals of 0.1 nm along the reaction coordinate with a 717 kcal mol-1 nm-1 force constant. In some windows, the force constant was raised to 2390 kcal mol-1 nm-1 to prevent ions from leaving the desired position. A total of 40 windows were used. The simulation was performed using a single penton in atomistic detail. The technique g_wham [28] was used to obtain the PMF of the ion from the simulations.
To complete this study, a series of simulations were performed in which Mg2+ was positioned along the reaction coordinate to observe its behavior. These sets of calculations were simulated for 2 ns. Only simulations in which the ion remained inside the pore were continued up to 100 ns.
The Gromacs g_densmap subroutine was used to obtain 2D water densities maps. To achieve this goal, RMSD of the water molecule coordinates of each simulation was analyzed to determine the portions of the simulation that were considered stationary and then used to obtain the maps. The pore radius was measured with the program HOLE [29].
To evaluate proton transport, or hydroxyl generation, through the channel characterized by means of MD simulations, hybrid n-layered integrated different levels of molecular orbital (ONIOM) calculations [30] were performed using the GAUSSIAN 09 (G09) package [31]. These simulations utilized density functional theory (DFT) formalism combined with a semi-empirical (SE) method in two layers (DFT/SE).
The region employed for the QM calculation was a cylindrical zone comprising the amino acids lining the cavity at the five-fold symmetry axis. The inner wall of this region includes 5 rings of five-fold related residues: Val 1166, Thr 1667, Gln 3014, Val 3012 and Thr 3010. A Mg2+ and several water molecules were also included in this model, resulting in a total number of 563 atoms for the simulations. For the QM calculations, the initial atomic coordinates were determined from the SPC-solvated pentameric structure obtained after 100 ns of MD simulation. Given that Gromos is a united atom force field, those implicit hydrogens in the MD simulation were added explicitly by means of the program Gaussian View included in the Gaussian package [31]. On the one hand, the DFT region comprised the Mg2+ cation, 38 water molecules, four OH- ions and 3 glutamine residues coordinated with the Mg2+ ion. Three of the OH- ions were located in the outer bulk solvent at arbitrary positions. The fourth hydroxyl was placed at the position corresponding to a water molecule with a longer residence time per the MD calculations. This last molecule coincided with the outermost water along the water wire formed after the MD simulation. B3LYP hybrid exchange-correlation functional [32–34] and molecular orbitals expanded with the 6-31G** Gaussian basis set were used in this layer. On the other hand, the semi-empirical region (SE) included the remaining amino acids forming the inner wall of the channel. These residues have been described by the AM1 method available in G09 [35,36]. Thus, all atoms involved in the proton channel were modeled using a high-level quantum approach. In contrast, the second layer, which was formed by the nearest neighboring amino acids, was also represented by quantum mechanical calculations (SE region).
Atoms in the DFT region were completely relaxed, while the corresponding ones in the SE region remained static. In these calculations, each molecule was treated separately, either with the DFT or the SE approach, and no special treatment for the DFT/SE boundaries was required. Thus, the DFT/SE boundaries did not include atomic bonds.
To determine the activation energy, the Synchronous Transit-Guided Quasi-Newton (STQN) Method developed by H. B. Schlegel and coworkers [37] was used. For minimizations, it performs optimization by default using redundant internal coordinates. Unlike other methods, STQN does not require a guess for the transition structure; instead, the reactant and product structures are the input. Full vibrational frequency analysis with only one imaginary frequency assures that the obtained geometry corresponds to a saddle point.
The processes for both the energy minimization and TS calculations were repeated until the convergence criteria reached the standard cutoff values, i.e., 0.00045 Hartrees/Bohr for the maximum force component, 0.0003 Hartrees/Bohr for the root-mean square force, 0.0018 Å for the maximum displacement and 0.0012 Å for the root-mean-square displacement.
More than three decades of structural studies on various picorna-like viruses, many of them performed on poliovirus (PV) and rhinovirus (HRV), has resulted in the elucidation of certain aspects of capsid destabilization and RNA release [38–41]. Nevertheless, we are still far from achieving a satisfactory description of the events that trigger and link these two processes in non-enveloped viruses. Several of the questions that remain unanswered are as follows: What are the external and internal factors that destabilize the capsid upon cell attachment or internalization? Does the RNA exit the capsid by being extruded through a small orifice in the capsid, or is it released without unwinding, i.e., in its compact folded form? These questions have been partially addressed in several works using both described picornaviruses as models, and they concluded that the RNA exits the viral capsid through small orifices [38–42]. In accordance with this model of genome externalization, a recent study has shown that the small structural protein VP4 from HRV creates small pores in model lipid membranes [43]. This observation is consistent with the liberation of the RNA by extrusion since the passage of the cell membrane could also occur through the small aperture created by VP4. However, one of the main prerequisites underlying the hypothesis of the RNA extrusion model is the need for the existence of a singular point within the viral capsid. This singular point would anchor the genome in a position in such a way that one of its extremes remains just beneath the capsid hole, thus requiring the icosahedral symmetry of the virion to be broken. This hypothesis remains to be experimentally confirmed.
Our previous studies examining TrV [7,20,44] have shown that the process of genome release is not consistent with the extrusion mechanism observed in PV and HRV [38–41]. This difference was also reported recently in Israeli acute bee paralysis virus [45]. Additionally, experimental data previously reported by our group have shown that an alkaline pH induces weakening of capsid-RNA interactions and that the genome itself can promote capsid opening to be released [7]. Moreover, we also observed that the mechanism of cell permeation by TrV VP4 could differ from that postulated for HRV [43] in the sense that instead of making small perforations, it produces large dynamic pores in model membranes [44]. This study showed that concomitantly with the pH effect on TrV capsid destabilization, the efficiency of VP4 permeation of membranes increased with alkalinization of the medium. In summary, all of our previous observations indicate that the alkaline pH of the solvent triggers key processes in TrV, such as capsid destabilization, genome release, and enhancement of the cell wall disruption.
Our MD calculations allowed the elucidation of the multiple roles of the solvent in the narrow cavity that traverses the TrV capsid. We observed a hydrophobic gate inside the pore that formed along the capsid five-fold axis, and this empty space prevented solvent exchange between the interior and exterior of the virion. This feature has not been described in viral capsids, but it has been observed and widely studied in membrane ion channels and other non-biological systems [46–49].
The assumption based on crystallographic data that a Mg2+ is on the five-fold axis, coordinated to five Gln3014 residues, removes the hydrophobic gate property, allowing for a fully hydrated cavity. Moreover, Mg2+ coordinates with incoming water molecules and the oxygen of neighboring amino acids and creates a water wire that allows for the connection between the external region and the capsid interior. Hence, MD calculations provide strong support to rationalize the function of the putative ion observed not only on both CrPV and TrV atomic structures but also in other icosahedral viruses. Thus, Mg2+ would allow for the outer and the inner capsid regions to become connected through a chain of 8 water molecules that extend approximately 2 nm (Figs 3 and 4). These results suggest the concept of "hydrophobic gaiting” for viruses, which would be controlled by the concentration of Mg2+, the most abundant intracellular divalent cation. Under this scenario, a first step for virus maturation would be gate “opening” by Mg2+ binding, which would be very much dependent on the cell, since it is well-known that different cells maintain different concentrations of Mg2+.
It has been noted that mature (full) TrV wild type capsids have open pores with sufficient space to allocate a cation [10], rather than pores that are partially occluded by Gln3014 as observed in empty TrV capsids [20]. The last observation indicates that TrV structural proteins not necessitate cations to assemble nor to maintain the capsid integrity. As a result of the simulations described in this manuscript, we can infer that the putative cation located in this position would play a functional role in hydration and proton conduction through the five-fold channel, and not a structural role as suggested by other authors [50]. Given the observation that the putative Mg2+ is trapped between two energy barriers, it is unlikely that this cation enters the specified pore center from either pore entrance, reinforcing the idea that the ion is taken up from the medium during the virus assembly process and released upon genome exit.
It has been shown in other systems [47,49] that channel hydration is fundamental for favoring the passage of ions by decreasing the energetic barrier for ion translocation. Computational investigations and several experimental approaches have focused on different aspects of water-mediated proton transfer in biological systems. Some examples are the gramicidin A transmembrane channel [51–53], bacteriorhodopsin [54–56], cytochrome C oxidase [57,58], influenza AM2 proton channel [59], green fluorescent protein (GFP) [60], the voltage-gated proton channel HV1 [61], carbonic anhydrase [62] and so-called bioprotonic devices [63].
One of the major QM limitations in modeling biological systems is to represent the solvent complexity at the atomic level; this limitation is observed not only because of the different ions that can be present but also due to the large number of atoms needed to mimic a real physiological pH. Indeed, solely to represent a system at a neutral pH would require a model of more than 3x107 atoms. Since our model is much smaller, our results do not correspond to a real situation, but we can assume that they qualitatively correspond to the pH situation we aimed to describe.
Our QM analysis showed that through a Grotthuss-like mechanism, proton holes could traverse along the water wire (Figs 3 and 4A). This was found to be a concerted transport that moved stepwise proton holes against the pH gradient and protons in the opposite direction. This movement is energetically favored from the outer bulk solvent up to the water molecule coordinated to the Mg2+. At this point, a low barrier prevented the continuation of the proton hole towards the internal portion of the water wire (Fig 3E, TS, and Fig 4A). Nevertheless, because this barrier is on the same order as the average kinetic energy of a molecule at room temperature, the hole could continue along its way. From this point, the downward transit of proton holes was energetically favored and reached the internal bulk solvent in two more jumps. In this way, the internal OH- concentration could increase to equal the pH at the capsid exterior.
Additionally, the cation and its three coordinated water molecules formed a barrier that impeded the transit of protons into the internal solvent. Thus, the channel structure acted as a 'diode' since it only permitted the exit but not the entrance of protons into the capsid.
It can be argued that unidirectional proton passage is not an intrinsic feature of the channel structure, but the result of the simulation conditions mimicking an alkaline or acidic pH. Thus, the likely solvent pH turns the channel into a ‘diode’, allowing proton passage in only one direction. However, it is clear from our calculations that disruption of the water wire is the factor that impedes protons from entering further into the channel. This disruption of connectivity is a consequence of the electrostatic repulsion exerted by Mg2+ on the hydronium. Thus, we can conclude that the two factors making the five-fold axis cavity to act as a proton diode are the limited space of the hydrophobic neck and the positive ion trapped in it.
In summary, the leakage of protons through the five-fold cavity is unidirectional and a consequence of the presence of the Mg2+. This finding is consistent with the observed sensitivity of the TrV virions to alkaline environments and resistance to acidic pH values [6,7] and constitutes the first mechanistic model to explain the sensitization of viral capsid to alkaline pH values.
The exit of protons through the five-fold cavity permits a plausible mechanism by which the concentration of hydroxyl ions inside the capsid can be increased. Since mature TrV virions are assembled in the cytoplasm of the infected cells, the internal pH condition should be neutral. Once released into the extracellular space, the viral particles can encounter an alkaline environment, after which the capsid proton channels are activated, enabling the internal pH to increase. In one of the possible scenarios, if the pH inside the capsid reaches values of approximately 8.5, which is the pH limit for RNA nucleotides to start to titrate, the hydroxyl ions will compete with the genome for binding the counterions. Under these conditions, RNA is no longer fully neutralized, and the electrostatic repulsion between phosphates induces its unfolding, thus causing the appearance of an internal pressure. In fact, we have previously demonstrated that in virions exposed to an alkaline pH, the RNA is the molecule that promotes capsid destabilization and its disassembly [7]. Although the complete process of genome uncoating most likely occurs in the cell cytoplasm, the extracellular alkaline condition could induce a softening of the viral capsid. This phenomenon has been observed as a prerequisite for effective infection of retroviral particles [64], and for the insect viruses IABPV [45] and Helicoverpa armigera stunt virus (Alphatetraviridae; Omegatetravirus) [65]. Moreover, the transient opening of the capsid could permit the egress of the internal small protein VP4, the hydrophobic peptide that is thought to be responsible to induce host cell membrane permeabilization in human rhinovirus [43] and in TrV [44].
The last step of our simulation process was to evaluate the effect produced on the capsid due to repulsive electrostatic forces generated inside the capsid. Our model consisted of the TrV capsid filled with solvent with increasing amounts of anions, which not surprisingly leads to the disruption of the protein shell. After a certain number of anions were added into the interior of the capsid, the simulation showed a rapid increase in the gyration radius (Fig 5B, 5C and 5D). Moreover, this occurred by opening the capsid through the frontier between pentamers, phenomenologically in agreement with a previous energy calculation and experimental data showing that the disassembly of the capsid occurred while maintaining the integrity of the pentons [7]. Interestingly, the crevices observed during the simulations are sufficiently large to allow the simultaneous passage of several solvated Cl-. This separation between pentons opened breaches that are large enough to allow the escape of RNA while even maintaining its secondary folding and, perhaps, part of its tertiary structure.
Thus, our calculations support capsid cracking as a process that allows genome egress, which is an alternative mechanism to the previously mentioned extrusion model.
Finally, in our modelization the limit after which the capsid is dismantled is achieved when the internal charge overpasses 200 anions (that corresponds to an internal charge density of 2.8x10-5 e-/Å3). However, we notice that such number of negative charges could be provided by deprotonation of the VP1-4 proteins N-termini, a situation that can occur when the internal pH is alkaline. This charge inversion at the capsid internal face could not only contribute to the capsid destabilization, but also to dwindle the forces that operated during genome encapsidation. To relax these protein-RNA interactions is a condition necessary to detach and then release the genome from the viral capsid proteins.
In summary, here we simulated different steps of the disassembly mechanism of the TrV viral capsid by employing quantum, classical and CG modelization approaches. Our model suggests the presence of hydrophobic (water/proton) gates in viral capsid channels. The presented multiscale computational calculation assigns a functional role to the putative metal ion on the five-fold axis and provides a rationale for the mechanism of capsid pH sensitivity, as well a novel mechanism of unidirectional proton conduction. Moreover, our model links this pH sensitization to the appearance of internal electrostatic forces that would drive the destabilization and disassembly of the TrV viral capsid, which are key steps in permitting genome release.
|
10.1371/journal.ppat.1003572 | Acute Neonatal Infections ‘Lock-In’ a Suboptimal CD8+ T Cell Repertoire with Impaired Recall Responses | Microbial infection during various stages of human development produces widely different clinical outcomes, yet the links between age-related changes in the immune compartment and functional immunity remain unclear. The ability of the immune system to respond to specific antigens and mediate protection in early life is closely correlated with the level of diversification of lymphocyte antigen receptors. We have previously shown that the neonatal primary CD8+ T cell response to replication competent virus is significantly constricted compared to the adult response. In the present study, we have analyzed the subsequent formation of neonatal memory CD8+ T cells and their response to secondary infectious challenge. In particular, we asked whether the less diverse CD8+ T cell clonotypes that are elicited by neonatal vaccination with replication competent virus are ‘locked-in’ to the adult memory T cell, and thus may compromise the strength of adult immunity. Here we report that neonatal memory CD8+ T cells mediate poor recall responses compared to adults and are comprised of a repertoire of lower avidity T cells. During a later infectious challenge the neonatal memory CD8+ T cells compete poorly with the fully diverse repertoire of naïve adult CD8+ T cells and are outgrown by the adult primary response. This has important implications for the timing of vaccination in early life.
| Newborns typically have a heightened sensitivity to infectious diseases, the reasons for which are not yet well understood. One contributing factor is the limited diversity of lymphocyte receptors early in life to recognize antigen and control infection. We have previously shown that antigen-specific CD8+ T cell repertoires are significantly constricted in neonates compared with adults. In this study, we addressed the question of whether the developmental stage of the host at the time of vaccination influences the composition of the memory CD8+ T cell repertoire and its ability to mount a robust response to subsequent infections. We observed that the antigen-specific T cell repertoires elicited in the context of an acute neonatal infection, that are less diverse and comprised of lower-avidity T cells, are partially ‘locked-in’ to the adult memory T cell repertoire. However, in the face of a secondary infectious challenge, naïve adult T cells outcompete the lower avidity neonatal memory T cells and raise the diversity of the overall CD8+ T cell response. These results have potential implications for the design of vaccines to be administered in early life.
| The immune system of neonates is generally characterized as immature and more susceptible to infections with various pathogens [1]–[3]. Many of the most debilitating infections are inflicted by intracellular pathogens that are either vertically transmitted or acquired very early on in life (e.g. HIV, CMV, EBV, TB, HSV). Although CD8+ T cells are considered the key players in combating these intracellular pathogens, their capacity to provide protective immunity in neonates is still poorly understood. Importantly, since the timing of infection in some cases affects the subsequent pathogen load and pathogenesis of infection, we wished to understand whether early exposure to infection or vaccination compromises the later ability to control infection as an adult.
The ability of CD8+ T cells to mount a protective response to new pathogens is dependent upon the presence of a broad repertoire of T cells of appropriate immune functionality [4], [5]. Diversification of the repertoire is developmentally regulated and the neonatal T cell repertoire in mice is restricted not only by the reduced number of T cells that are present, but also by the number of unique antigen receptors that are able to be produced. Diversity of T-cell receptor (TCR) usage is accomplished by multiple mechanisms during T-cell maturation in the thymus [6]. Somatic recombination of germline segments identified as variable (V), diversity (D), and joining (J) segments by Rag-1 and Rag-2 proteins results in an enormous number of T cells with distinct antigen binding domains. Further diversification is accomplished by nibbling or loss of germline-encoded nucleotides and the addition of complementary template-dependent (P) and random template-independent (N) nucleotide additions at the junctions between these germline segments prior to ligation [7]. The addition of N regions between germline-encoded segments is mediated entirely by terminal deoxynucleotidyl transferase (TdT) and it has been estimated that 90–95% of the diversity of the T cell repertoire is attributed to this critical step [8].
The expression of TdT is likely to have a significant impact on both the quantity and quality of TCR clonotypes that are able to respond to various pathogens at different stages of development. In mice, TdT is not upregulated in the thymus until 4–5 days after birth, with significant nucleotide additions being observed at day 8 [9], [10]. Thus, in the first week of life we would still expect much of the peripheral T cell repertoire to be comprised of clonotypes that have not been sculpted by TdT and thus be devoid of N-additions. Indeed, we recently have characterized the TCRβ repertoire of CD8+ T cells responding to the immunodominant HSV-1 epitope, gB498–505/Kb (gB-8p) in neonate [11] and adult mice [12] and showed that the gB-8p TCRβ repertoire in neonatal mice is severely restricted and comprised of more germline sequence-rich clonotypes [11]. This restricted TCR repertoire in neonates may have direct effects on the ability of primary neonatal CD8+ T cells to respond to antigen, as well as indirect effects on their ability to transition into the long-lived memory pool. The key question we wished to examine in this report is whether the primary CD8+ T cell response in neonates induces a memory pool of sufficient diversity to later mount a robust secondary response to infection, or whether neonatal infections ‘lock-in’ a poor memory CD8+ T cell population that exhibits impaired recall responses in later life. Here, we demonstrate how the developmental stage of the host at the time of vaccination or primary infection can alter the composition of the long-lived memory CD8+ T cell pool, as well as their ability to respond to subsequent infections.
In this report, we aimed to compare antiviral memory CD8+ T cells that were generated in either neonatal or adult stages of development. Over 90% of the CD8+ T cell response in HSV-1-infected C57BL/6 mice is directed against a single Kb-restricted immunodominant epitope in the glycoprotein B (denoted gB-8p) [13]. To compare the expansion of naïve and memory gB-8p CD8+ T cells in neonatal (7-day old) and adult mice (8–12 week-old), both age groups were acutely infected with vaccinia virus expressing the gB-8p peptide (VACV-gB, i.p.) and challenged 6–8 weeks later with HSV-1 (i.p.). In this way, we were able to mimic neonatal vaccinations with live viral vectors and preferentially prime the neonatal T cells that were available in early life. However, smaller numbers of peripheral T cells are present in neonatal mice compared to adults [14], [15]. Therefore, as in our previous study of the neonatal primary response [11], the dose of VACV-gB was normalized in neonates (2×101 PFU/mouse) and adults (2×105 PFU/mouse) by titrating VACV-gB doses down to the least amount of virus that was required to elicit a comparable relative frequency (∼10%) of gB-8p CD8+ T cells at the peak of the response (Fig. 1A). This difference in viral dose is necessary as an adult dose is lethal to neonates and adult mice administered a neonatal dose clear the virus too rapidly to allow the detection of antigen-specific CD8+ T cells. In addition, as primary vaccinia virus infection is intended to mimic vaccination at different ages, we note that a decreased dose of immunogen is routinely administered to children for a variety of human vaccine protocols. While the total number of gB-8p CD8+ T cells was much higher in adults at the peak of the primary infection (due to increased cellularity), similar levels of gB-8p-specific memory T cells were observed in neonatal- and adult-immunized mice by 6 weeks post-infection (Fig. 1B). All mice were then challenged with 1×106 pfu of HSV-1 (i.p.), and we observed a secondary CD8+ T cell response that was comparable between neonate and adult-vaccinated mice (Fig. 1B).
We recently have compared the clonal composition of the gB-8p-specific TCRβ repertoires involved in the primary CD8+ T cell responses to VACV-gB in neonatal and adult mice [11]. The gB-8p-specific TCRβ repertoires in neonatal mice were found to have the same basic features, in terms of gene usage biases and CDR3β amino acid motif, as in adult mice. However, the significantly less diverse gB-8p-specific Vβ10+ TCRβ repertoires of neonatal mice were predominantly comprised of shorter germline-gene-encoded CDR3β sequences. This published data on the primary response to vaccination in adults and neonates was used as a baseline for comparison with the secondary responses following challenge.
To determine whether the less-developed T cell repertoires involved in the immune responses to neonatal vaccination are ‘locked-in’ to secondary responses to infection, we examined the clonotypic composition of gB-8p-specific TCRβ repertoires involved in secondary CD8+ T cell responses to HSV-1 infection in adult mice that had been previously vaccinated with VACV-gB either as neonates or as adults. The same Vβ10 gene usage bias associated with primary CD8+ T cell responses to the gB-8p epitope was also observed in both neonatal-vaccinated and adult-vaccinated mice at the peak of the secondary immune responses (Fig. 2A). However, the large inter-mouse variability in Vβ10 gene usage observed in the primary responses [11] in neonatal mice was substantially reduced in the secondary immune responses.
Single-cell sequencing was then used to examine in greater depth the composition of the gB-8p-specific Vβ10+ CD8+ TCRβ repertoires involved in the secondary immune responses. The gB-8p-specific Vβ10+ TCRβ repertoire data are summarized in Table 1 and representative repertoires are shown in Fig. S1, and the gB-8p-specific Vβ10+ CD8+ TCRβ repertoire characteristics quantitatively compared in Fig. S2. It is important to mention that while the primary and secondary responses were elicited by two different pathogens to avoid antibody interference, we previously showed that the clonotypic composition of gB-8p-specific Vβ10+ CD8+ cells in the primary response is similar among a wide range of infections, including VACV-gB and HSV1 [16]. In adult mice, the general characteristics of the gB-8p-specific Vβ10+ CD8+ TCRβ repertoires associated with primary responses were largely preserved in the secondary immune responses (Fig. S2).
In neonatal mice, we have previously reported that the primary response contains a restricted repertoire of T cells that is largely germline encoded [11]. Thus, we expected the secondary response in neonatal-vaccinated mice should also comprise a more restricted subset of these cells. Surprisingly, the secondary response in neonatal-vaccinated mice was significantly more diverse and showed a much higher proportion of TCRβ clonotypes requiring N-additions than the previously reported neonatal primary response [11] (Fig. 2B, C). However, despite this diversification of the gB-8p-specific Vβ10+ TCRβ repertoires between primary neonatal responses and secondary responses, TCRβ clonotype diversity remained significantly reduced compared to mice that were primed as adults (Fig. 2C). These results suggest that priming neonatal mice leads to a recall response of intermediate diversity between the neonatal primary response and the ‘normal’ adult secondary response. Thus, we investigated what mechanisms contribute to this partial “locking-in” of the immature neonatal repertoire during secondary responses later in life.
Since we observe a significant change in the gB-8p-specific CD8+ TCRβ repertoire between the neonatal primary and secondary responses, we set out to identify when this diversification occurred. Firstly, it seems possible that only a subset of the neonatal primary response contributes to the secondary response, and that these cells are selectively the more ‘adult-like’ clonotypes. This selection for adult-like clonotypes could occur either during the contraction phase from the primary response, or during the expansion phase from the memory response to the secondary response. To investigate this, we first analyzed the neonatal resting memory compartment (Fig. S2). Although there was a trend for a more diverse (Fig. S2K, L) and less germline encoded (Fig. S2J) TCRβ repertoire in the resting memory population compared with the neonatal primary response, this was not sufficient to explain our observations for the secondary responses in neonatal-vaccinated mice.
A key question is to what extent memory neonatal gB-8p CD8+ T cells participate in a secondary immune response? That is, we would expect neonatal gB-8p memory T cells to be present at higher numbers than adult naive gB-8p T cells, and therefore dominate the recall response. Alternatively, it is possible that the less diverse neonatal T cell clonotypes will be impaired in their ability to compete with a fully developed adult naïve T cell repertoire and thus be underrepresented in the response to a secondary challenge. To examine these possibilities, we adoptively transferred equal numbers of neonatal or adult gB-8p memory CD8+ T cells into different congenic recipient mice (CD45.1) followed by HSV-1 challenge. This allowed us to distinguish between the expansion of naïve (CD45.1) and memory (CD45.2) T cells during the secondary immune response. Donor neonatal and adult gB-8p memory T cells were evaluated at the peak of the recall response following HSV-1 infection, and the proportion of the response contributed by neonatal or adult memory cells was assessed. The magnitudes of the overall response were comparable between the recipients of neonatal memory and adult memory T cells (Fig. 3A). However, approximately 3 fold fewer neonatal memory T cells were observed at the peak of the recall response compared to the adult memory T cells (Fig. 3B). In repeat experiments, we also bled recipient mice at 4 and 6 days post-infection and observed a greater contribution by adult memory CD8+ T cells to the overall response at both time points (Fig. S3). Importantly, these differences were not statistically significant until day 6, suggesting that neonatal memory CD8+ T cell become activated but exhibit an impaired ability to expand and compete in the adult mouse.
The reduced contribution of neonatal memory cells compared with adult memory cells in the secondary response raised the possibility that the majority of cells in the secondary recall response in neonatally-vaccinated mice may actually be primary adult CD8+ T cells, rather than neonatal memory cells. To investigate this, we first looked at the clonotypic differences between donor secondary neonate gB-8p-specific memory and the recipient primary adult gB-8p-specific effector CD8+ T cell populations at 6 days post-infection. Although a significantly smaller proportion of the secondary neonatal gB-8p-specific memory T cells used the Vβ10 gene compared with the primary adult gB-8p-specific effector T cells, Vβ10 gene usage was prevalent in most mice (Fig. 4A). Single cell sorting and sequencing of the gB-8p-specific Vβ10+ TCRβ clonotypes for these two populations (Table 1; Fig. S1 C, D) revealed that a significantly higher proportion of secondary neonate memory TCRβ clonotypes required no nucleotide additions (Fig. 4B) and the secondary neonate memory TCRβ repertoires were significantly less diverse compared with the primary adult gB-8p-specific effector Vβ10+ TCRβ repertoires in the same mouse (Fig. 4C). Furthermore, we verified that the features of the secondary neonate gB-8p-specific memory Vβ10+ TCRβ repertoires were indicative of the resting memory population following VACV-gB infection in neonatal mice (Fig. S2 G–R).
To summarize, when we tracked the fate of adoptively transferred neonatal memory cells in the recall response in adult congenic recipients, these neonatal memory TCRβ repertoires maintained similar features to the neonatal memory population. However, the other major contributor to the responding population were the adult primary cells. When we separately analyzed the repertoire of the adult cells contributing to the response to the secondary challenge, we found that they resembled the normal adult secondary response, and were comprised of a significantly higher proportion of TCRβ clonotypes with N-additions, and were significantly more diverse than the neonatal memory cells in the same response. This suggests that the observed ‘diversification’ of the secondary response in neonatal-vaccinated mice arose not because the neonatal repertoire itself was altered, but because the neonatal recall response was so poor, that it was outcompeted by the adult primary response. Therefore, the combination of the narrow neonatal memory and diverse adult primary repertoires led to the observed intermediate level of TCRβ repertoire diversity in the secondary responses in neonatal vaccinated mice.
Given the less diverse TCR repertoire and poor recall responses exhibited by neonatal memory T cells, we next questioned whether this might be mediated by a neonatal T cell pool is insufficiently broad to select high avidity memory gB-8p CD8+ T cells. This is important since high avidity T cells have been shown to respond more vigorously to infection and kill infected cells faster than low avidity T cells [17]. Our hypothesis was that the responding pool of gB-8p CD8+ T cells in neonates will not include as many ‘best-fit’ T cells and will exhibit much lower TCR avidity than adult gB-8p CD8+ T cells. To test this, TCR:pMHC disassociation kinetics were assessed between neonate and adult memory gB-8p CD8+ T cells. During the steady-state, resting memory phase, neonatal gB-8p CD8+ T cells demonstrated much faster pMHC:TCR off-rates compared to adults (Fig. 5). Collectively, these data suggest that the less diverse neonatal gB-8p memory repertoire undergoes poor recall responses due to lower proportion of high-avidity CD8+ T cells in the memory compartment.
To validate and broaden the significance of our results, we next asked whether other types of infection also give rise to neonatal memory CD8+ T cells with poor recall efficacy. For these experiments, we infected neonatal and adult mice with an attenuated strain of Listeria monocytogenes that expresses the gB-8p peptide (denoted ActA LM-gB). This strain lacks a gene that is required for mobility and is incapable of infecting nearby cells, allowing us to better control for variations in the availability and abundance of antigen and challenge both age groups with the same dose. At six weeks post-infection, we co-transferred equal numbers of neonatal (CD45.2) and adult (CD45.1) gB-specific memory CD8+ T cells into congenically marked recipients (Thy1.1), which were subsequently infected with HSV-1 (1×106 pfu, i.p.). By transferring memory CD8+ T cells from adult or neonatal primed donors into the same host, we could rule out potential environmental differences arising during the response. Importantly, the percentage of neonatal and adult gB-specific memory CD8+ T cells were found to be similar prior to infection, with neonatal donor cells slightly outnumbering adults (68.8%±2.3 vs 30.6%±2.3). At the peak of the response, spleens were harvested and the ratio of neonatal to adult memory CD8+ T cells was calculated. Consistent with our results following vaccinia infection, we again observed a greater contribution of adult memory CD8+ T cells to the secondary response (Fig. 6A). Importantly, this analysis measured the total donor response by IFNg production, and thus included both Vβ10+ and Vβ10- gB-8p specific CD8+ T cells. These findings suggest that limited recall responses are likely a common feature among neonatal memory CD8+ T cells, regardless of how they are initially primed.
Despite the fact that neonatal vaccinated mice exhibited reduced TCR diversity, TCR avidity and lower recall efficiency, a critical remaining question was whether these differences resulted in impaired immune protection. To address this, we compared the ability of neonatal and adult memory CD8+ T cells to clear a high dose infection of recombinant Listeria monocytogenes expressing gB-8p (Lm-gB). We choose to challenge mice with LM-gB instead of HSV-1 for these studies because well-defined sites of infection can be more easily monitored. Neonatal and adult mice were again vaccinated with ActA-/- LM-gB (1×106 cfu, i.p.) and CD8+ T cells were allowed to transition into the memory phase. At 6 weeks post-infection, we transferred the same number of neonatal or adult gB-specific CD8+ T cells into separate recipient mice and challenged both groups with wt LM-gB (5×104 cfu, i.v.). Three days later, the livers of recipient mice were homogenized and the bacterial loads were examined. As shown in Fig. 6B, adult memory CD8+ T cells reduced the bacterial load 3 fold compared to an equivalent number of neonatal memory CD8+ T cells. Indeed, the neonatal memory CD8+ T cells showed no better bacterial control than naïve adult cells (which we have shown contribute significantly to the response in the presence of a neonatal memory response). This data suggests that a suboptimal recall response exhibited by neonatal memory CD8+ T cells could result in reduced immune protection.
The focus of this report was to determine how the composition and responsiveness of the memory CD8+ T cell pool is altered by neonatal vaccination or infection that occurs prior to the diversification of the CD8+ T cell repertoire. Our results demonstrate that the restricted neonatal T cell memory pool induced by early vaccination is in fact comprised of fewer clonotypes that are also of lower avidity than the adult response. However, while vaccination early in life recruits many of these ‘less fit’ clonotypes and allows them to persist in the memory CD8+ T cell compartment, these neonatal memory clonotypes are not efficiently recruited into the proliferative recall response to secondary challenge. In the absence of a strong neonatal memory response mediating early viral control, a robust adult primary response is generated, which ultimately comes to dominate the neonatal memory population. Despite this contribution from the adult repertoire, the secondary response in mice vaccinated as neonates remained significantly restricted, in terms of the diversity of TCR clonotypes, compared with secondary responses in adult-vaccinated mice. These observations describe a situation that is mechanistically similar to the phenomenon of ‘original antigenic sin’, in which prior infections with a related pathogen can “trap” the immune system into responding with less efficient memory clonotypes. However, in this case, the same pathogen may “trap” less efficient clonotypes into the immune reserve simply by priming these T cells during the early stages of development.
One of the most interesting findings of our present study was that there is significantly more recruitment of adult naïve clonotypes into the neonatal secondary response than the adult secondary response. Thus, when neonatal memory CD8+ T cells are faced with competition from a fully-developed adult naïve T cell repertoire, they prove inferior and make a smaller contribution to the overall memory response. Previous reports indicate that new naïve T cells seed the periphery at a relatively constant rate of 1–2×106 cells/day and the number of splenic recent thymic emigrants reach a peak at ∼6 weeks of age [18]. These naïve clonotypes will have also been sculpted by TdT, which should allow for significantly greater opportunities to generate high avidity gB-8p-specific CD8+ T cells. Indeed, our data suggest that more ‘best-fit’ gB-8p-specific CD8+ T cells exist in the adult naïve pool at 6–7 weeks of age compared to those available in the neonate memory pool.
A number of recent reports have examined the role of TdT in generating robust anti-viral CD8+ T cell immunity to acute infections. Mansour Haeryfar et al. showed that the overall magnitude and breadth of the CD8+ T cell responses to influenza and vaccinia virus were reduced in TdT-/- mice and the hierarchy of immunodominant epitopes was altered [19]. The authors proposed that the reshuffling of immunodominant determinants was due to the loss of high affinity clones for some (but not all) viral determinants. In support of this, Kedzierska et. al. showed that the avidity of influenza-specific CD8+ T cells was lower in TdT-/- mice for the NP366 epitope, where the response is public and clonotypically restricted, but not for the PA224 epitope, which elicits a more private and diverse TCR repertoire [20]. Ruckwardt et al. [21] recently reported differences between neonate and adult CD8+ T cell responses to respiratory syncytial virus infection with respect to TCR diversity, functional avidity, precursor frequency and epitope immunodominance hierarchy. However, in terms of the latter, this study suggests that the shifting epitope immunodominance is not associated with TdT. Together, these studies indicate that the relative role of TdT in promoting optimal anti-viral CD8+ T cell immunity may ultimately depend upon the clonal complexity of the T cell response against specific viral determinants being examined.
Although the neonatal repertoire is also comprised of TCRs generated in the absence of TdT, it is important to mention that the neonatal repertoire is potentially even less diverse than adult TdT-/- mice due to lower numbers of T-cells in the neonatal periphery. Based on previous estimates, we would expect ∼2×106 different TCRs (with an average clone size of 10) in adult wild-type mice and ∼1–2×105 different TCRs (with an average clone size of 100) in adult TdT-/- mice [8], [22], [23]. However, there are 10–100 times fewer CD8+ T cells in 7-day old neonatal mice compared to adult mice [24], [25]. Therefore we expect the neonatal repertoire to consist of only a small fraction of the total T cell repertoire that is available in adult TdT-/- mice.
In our report, we elected to prime neonates and adult mice with either an acute virus (VACV-gB) or an attenuated bacterial strain (ActA- LM-gB) so that we could more closely mimic vaccinations and clearly delineate effector and memory CD8+ T cell responses. However, generating sufficiently broad CD8+ T cell repertoires may be even more beneficial in the context of chronic and persistent pathogens. Many of these chronic pathogens (e.g. HIV, HCV) are able to evade the immune response by undergoing a high rate of mutation. Thus, not surprisingly, one key correlate of immune protection against these chronic viral pathogens is diversity in TCR usage [26]–[29]. Responding with a larger number of distinct clonotypes that can recognize multiple epitopes on these pathogens as well as a diverse range of epitope variants has been shown to provide better protection against immune escape. While this has not been rigorously examined in neonatal mice, our prediction would be that the diminished neonatal repertoire would be significantly impaired in limiting the emergence of viral escape mutants.
These results suggest that there are long-term consequences for vaccinations or infections that occur prior to the diversification and maturation of the adult immune system. However, it is important to mention that our results do not rule out the possibility that other developmental factors may contribute to poor neonatal immunity. In this report, we have used TCR analysis as a tool to track neonatal clonotypes, since phenotypic markers alone cannot be used to reliably distinguish naïve and memory CD8+ T cells [30]. Our analysis shows that neonatal clonotypes transition into the adult memory pool, but undergo a limited recall response and confer reduced immunity against a secondary challenge. In regard to these challenge experiments, it is important to point out that pathogen clearance was examined in recipient mice following adoptive transfer of either neonatal or adult memory CD8+ T cells. Thus, only a fraction of the total memory pool is participating in the secondary response. This is an important point since other studies have shown that clearance is dependent upon the number of CD8+ T cells that are adoptively transferred into recipient mice prior to challenge [31]. Under limiting conditions, we observed a statistically significant difference in the ability of neonatal and adult memory CD8+ T cells to clear infection. Importantly, this data does not necessarily indicate that neonatal memory CD8+ T cells are incapable of responding (Fig. 6A clearly shows some contribution of neonatal memory CD8+ T cells to the recall response), but rather that they are less functional compared to adults at the level that was examined. The degree of immune protection by neonatal memory CD8+ T cells will likely vary with the number and type of memory cells that are generated [32]. Given that neonatal T cell clonotypes do in fact gain access to the memory pool in adults, it is now imperative that we fully understand how neonatal memory CD8+ T cells differ than adult memory CD8+ T cells at the cellular and molecular level. These studies would be especially important to consider in the context of tissue resident memory T cells, or the long-lived population of T cells that remain detached at the peripheral sites of initial pathogen encounter (i.e. lung, skin, gut, etc). Knowledge from these studies will provide us with a solid platform to understand how infections early in life may impact the development of T-cell mediated diseases in adulthood, and also to guide rational design of vaccines that can be safely administered to neonates.
C57BL/6 (H-2b) and B6-LY5.2/Cr (H-2b) mice were purchased from NCI (Frederick, MD) and Thy1.1 mice (B6.PL-Thy1a/CyJ) were obtained from The Jackson Laboratory (Bar Harbor, Maine). All mice were maintained under pathogen-free conditions in the animal facility at either the University of Arizona or Cornell University. Pregnant mice were individually housed and monitored daily for births. Neonatal mice were used at 7 days of age. Adult mice were obtained from commercial vendors and used at 2–3 months of age.
All animal experiments were conducted by guidelines set by the University of Arizona Institutional Animal Care and Use Committee (IACUC), under the University of Arizona approved animal protocol #08-059, and in accordance with the U.S. Animal Welfare Act. Recombinant vaccinia virus (VACV) expressing the MHC class I-restricted CTL epitope HSV gB498–505 (SSIEFARL, gB-8p in the text), designated VACV-gB, was generously provided by Dr. S.S. Tevethia (Pennsylvania State University of College Medicine, PA). VACV-gB viral stocks were propagated and quantified in 143B cells. HSV-1 strain 17 was obtained from Dr. D.J. McGeoch (University of Glasgow, Scotland, U.K.), cloned as a syn+ variant and tittered on Vero cells in our laboratory as previously described [33], [34]. Neonatal and adult mice were intraperitoneally infected with either 2×101 or 2×105 PFU, respectively. Recombinant strains of Listeria monocytogenes expressing the gB-8p epitope, designated Lm-gB or ΔActA Lm-gB, were provided by Dr. Sing Sing Way (Cincinnati Children's Hospital Medical Center, OH) and have been previously described [35]. Prior to infection, the bacteria were grown to log phase (OD600 0.1), and mice were either immunized with ΔActA Lm-gB (1×106 CFU, i.p.) or challenged with LM-gB (5×104 CFU i.v.) in 100 ul of PBS.
The gB-8p:Kb tetramer was obtained from the National Institutes of Health Tetramer Core Facility (Emory University, Atlanta, GA). mAbs anti-CD8α (clone 53–6.7), anti-CD4 (RM4-5), anti-CD11a (2D7), anti-Vβ10 (B21.5), anti-Vβ8 (F23.1), anti-CD45.2 (104) were purchased from commercial sources. FCM data was acquired on the custom-made FACS LSRII instrument equipped with four lasers, using Diva software (BD Biosciences), and analysis was performed using FloJo software (Treestar).
To evaluate the degree of TCR avidity, the relative off-rates were determined by a tetramer decay assay. For this, splenocytes were stained with anti-CD8α and gB-8p:Kb tetramer for 1 hour at 4°C. These cells were then washed and incubated in the presence of saturating amounts of anti-Kb antibody (AF6; Biolegend) at room temperature to prevent rebinding. At various times, cells were removed, placed in fixation buffer and the amount of gB-8p:Kb tetramer remaining on the surface was quantified by flow cytometry. These measurements were expressed as a percentage relative to tetramer staining at time t = 0.
Splenocytes were harvested at indicated timepoints following infection. CD8+ T cells were isolated using positive immunomagnetic selection (Miltenyi Biotec, Auburn, CA) and CD8+CD4-gB-8p:Kb+ Vβ10+ lymphocytes were individually sorted using the FACSAria cell sorter system (BD Biosciences) as previously described [12], [35]. Control wells without sorted cells were included on every plate to identify any possible contamination. cDNA synthesis, PCR amplification and sequencing of individual Vβ10 transcripts were performed exactly as previously described [12], [35].
The gB-8p-specific CD8+ TCRβ repertoires were characterized by sequentially aligning each TCRβ sequence with the Vβ10 (TRBV4 in IMGT nomenclature) gene, followed by the best-match Jβ gene and the best-match Dβ gene. This analysis was done using the IMGT reference alleles for the Mus musculus TRB genes [36]. The CDR3β sequence was then identified between, and inclusive of, the conserved cysteine in the Vβ-region and the conserved phenylalanine in the Jβ-region.
The diversities of the CD8+ TCRβ repertoires specific for the gB-8p-epitope in each mouse were evaluated using two different measures of diversity, the number of different TCRβ amino acid sequence clonotypes and Simpson's diversity index [37]. Simpson's diversity index accounts for both the variety of amino acid sequence clonotypes and their clone sizes, and ranges in value from 0 (minimal diversity) to 1 (maximal diversity). To account for differences in the sizes of the TCRβ repertoire samples, TCRβ repertoire diversity was estimated as the median value of 10,000 random draws of subsamples of 48 TCRβ sequences from the total TCRβ repertoire sample [37]. The diversity analysis was performed using Matlab (The Mathworks, Natick, MA).
To evaluate immune protection, 5×104 CFU of wt Lm-gB was administered intravenously as previously described [38]. On day 3 post-infection, livers were harvested into sterile PBS and weighed. Tissues were homogenized mechanically using a Tissue-Tearor electric homogenizer (BioSpec Products, Bartlesville, OK). Serial dilutions were made in sterile PBS and plated onto BHI agar. Plates were incubated overnight at 37°C. The log10 CFU/g of tissue was calculated as: log10 [(CFU/dilution factor)×(organ weight+homogenate volume)/organ weight)].
TCRβ repertoire features of the endogenous CD8+ T cell responses were compared using a Mann-Whitney test for all pairwise comparisons between age/infection groups, with Bonferroni correction for multiple pairwise comparisons. TCRβ repertoire features of the recipient and adoptively transferred CD8+ T cell populations were compared using a Wilcoxon text. For the tetramer decay assay results, exponential decay rates for individual mice were compared between neonatal and adult CD8+ T cells using a Mann-Whitney test. All statistical analyses were performed using GraphPad Prism software (GraphPad Software Inc, San Diego, CA).
|
10.1371/journal.pgen.1008012 | The CPEB translational regulator, Orb, functions together with Par proteins to polarize the Drosophila oocyte | orb is a founding member of the CPEB family of translational regulators and is required at multiple steps during Drosophila oogenesis. Previous studies showed that orb is required during mid-oogenesis for the translation of the posterior/germline determinant oskar mRNA and the dorsal-ventral determinant gurken mRNA. Here, we report that orb also functions upstream of these axes determinants in the polarization of the microtubule network (MT). Prior to oskar and gurken translational activation, the oocyte MT network is repolarized. The MT organizing center at the oocyte posterior is disassembled, and a new MT network is established at the oocyte anterior. Repolarization depends upon cross-regulatory interactions between anterior (apical) and posterior (basal) Par proteins. We show that repolarization of the oocyte also requires orb and that orb is needed for the proper functioning of the Par proteins. orb interacts genetically with aPKC and cdc42 and in egg chambers compromised for orb activity, Par-1 and aPKC protein and aPKC mRNA are mislocalized. Moreover, like cdc42-, the defects in Par protein localization appear to be connected to abnormalities in the cortical actin cytoskeleton. These abnormalities also disrupt the localization of the spectraplakin Shot and the microtubule minus-end binding protein Patronin. These two proteins play a critical role in the repolarization of the MT network.
| The specification of polarity axes in the Drosophila egg and embryo depends on the proper organization of the microtubule (MT) and actin cytoskeleton during mid-oogenesis. During this period, the MT organizing center at the posterior of the oocyte is disassembled and a MT network is established at the anterior and anterior-lateral cortex of the oocyte. We show that the CPEB translation factor orb plays a critical role in the reorganization of the MT network. orb appears to function at several levels during MT reorganization. orb interacts genetically with genes encoding Par proteins, aPKC and cdc42, and disrupts the localization of Par-1 and aPKC within the oocyte. orb also plays an important role in organizing the cortical actin cytoskeleton. The defects in the actin cytoskeleton disrupt the cortical association of Shot and Patronin, which are responsible for nucleating the assembly of the anterior MT network.
| Specification of the anterior-posterior (AP) and dorsal-ventral (DV) axes of the Drosophila embryo depends upon determinants that are localized within the egg during oogenesis [1–4]. For example, expression of the TGF-α cell signaling molecule Gurken (Grk) at the anterior corner of the oocyte during mid-to-late oogenesis establishes the DV axis of the egg and subsequently the embryo by signaling to the overlying somatic follicle cells [5–8]. Factors important in determining the AP axis of the embryo are also localized during this same period. Specification of the posterior axis is mediated by oskar (osk) [9, 10]. osk mRNA is targeted to the posterior cortex of the oocyte, where it is translated and functions in the assembly of the pole plasm and the anchoring of the mRNA encoding the posterior determinant nanos [11, 12]. The anterior axis is specified by the Bicoid transcription factor, and its mRNA is localized to the anterior cortex of the oocyte [13–15].
The proper localization of these determinants within the oocyte during mid-to-late oogenesis depends upon the disassembly of the existing microtubule cytoskeleton (MT) during stage 7 of oogenesis and its subsequent repolarization [7, 8]. The polarity of the MT network in the period prior to stage 7 is established early in oogenesis when the oocyte is initially specified [16]. A microtubule organizing center (MTOC) is assembled at the oocyte cortex just posterior to the oocyte nucleus and it directs the elaboration of the MT network by anchoring the minus-ends of MTs. As a consequence of this polarization of the oocyte, mRNAs encoding determinants critical for the early stages (stage 1–7) of egg chamber development accumulate at the posterior cortex. One of these is gurken (grk) mRNA. Grk protein translated from this localized message signals to the somatic follicle cells covering the posterior of the egg chamber to specify posterior follicle cell fate (PFC) [7, 8]. Subsequently, during stage 7, an unknown signal(s) emanating from the somatic PFCs triggers the repolarization of MT network in the germline. This signal induces the disassembly of the posterior MTOC and the network of MTs extending from the MTOC towards the anterior of the oocyte [7, 8, 17–20]. At the same time, de novo MT assembly is nucleated along the anterior and lateral cortex of the oocyte by a centrosome independent mechanism. This mechanism deploys the tubulin minus-end binding protein Patronin and the actin-MT linker Short Stop (Shot) [21]. Accompanying the repolarization of the MT cytoskeleton, the oocyte nucleus migrates from the posterior end of the oocyte to the anterior corner [22]. grk mRNA also relocates so that it is positioned between the oocyte nucleus and the oocyte cortex. Grk protein expressed from the localized message signals dorsal follicle cell fate and this defines the DV axis of the egg chamber and embryo [6, 23].
In addition to Patronin and Shot, the other factors implicated in oocyte repolarization are the Drosophila homologs of the partioning-defective (Par) group genes, par-1, cdc42 and bazooka (baz/par-3) [24–26]. These three genes together with par-6 and aPKC are also required for the initial polarization of the stage 1 egg chamber [24–30]. These proteins generate cellular asymmetries by inhibitory cross-regulatory interactions that impede association with the cell cortex [25, 31–33]. During MT repolarization, Par-1 becomes enriched along the posterior cortex of the oocyte [34–36]. There is a complementary distribution of Baz, Par-6, aPKC and Cdc42: they are enriched along anterior and anterior-lateral cortex, but not the posterior [24, 25, 37, 38]. The available evidence indicates that the asymmetry in the oocyte generated by the activation of the Par polarity network is upstream of the localization of Shot and Patronin along the anterior and lateral cortex, and thus the Patronin dependent de novo assembly of MTs [21].
In addition to being critical for properly localizing grk, osk and bcd mRNAs, the reorganization of the cytoskeleton also alters the distribution of other mRNAs encoding oocyte-specific proteins. One of these mRNAs is orb, which encodes one of the two fly cytoplasmic polyadenylation element RNA-binding (CPEB) proteins [39, 40]. During early stages of oogenesis, orb mRNA is localized at the posterior of the oocyte. After repolarization orb mRNA disappears from the posterior and becomes concentrated along the anterior-lateral margin of the oocyte [41]. While the rearrangement of orb mRNA within the oocyte is clearly downstream of the steps involved in repolarizing the oocyte MT network, the orb gene plays a central role in the initial formation and subsequent development of the oocyte and thus could be an active participant in determining oocyte polarity.
In ovaries, orb expression is restricted to the germline and is required at multiple steps during oogenesis [39, 40, 42]. In wild type ovaries, a cystoblast, generated by an asymmetric division of a stem cell, undergoes four mitotic divisions with incomplete cytokinesis to produce a 16-cell cyst [1]. In the orb null allele, orb343, the last of these mitotic divisions is not completed and the cyst degenerates [40]. While the strong loss-of-function allele, orb303, forms a 16-cell cyst, the oocyte is not properly specified and egg chambers contain only nurse cells [40]. Unlike orb343 and orb303, the Orb protein expressed by the hypomorphic orb allele, orbmel, is wild type. Instead, orbmel transcripts are incorrectly spliced generating an mRNA lacking sequences from the 5’UTR [42]. The removal of these 5’ sequences alters Orb expression as oogenesis proceeds. Prior to stage 7 the level and localization of Orb in the oocyte is similar to that observed in wild type. However, beginning around stage 7, the amount of Orb drops dramatically and most chambers have little residual protein. As a consequence of this reduction in Orb protein, orbmel females produce eggs that give rise to embryos with a range of phenotypic abnormalities including D-V and A-P patterning defects [42]. These patterning defects arise from a failure in the localization and/or translation of two Orb regulatory targets, grk and osk mRNAs, during mid-to-late oogenesis [43–46].
grk and osk transcripts are not, however, the only mRNAs that could be subject to orb regulation during oogenesis. Several recent studies have identified many other mRNAs that are Orb associated in vivo [47, 48]. Included in this group of potential orb regulatory targets are mRNAs encoding the Par proteins, aPKC, Baz, Par-6 and Cdc42. Moreover, there is evidence connecting the other fly CPEB protein, Orb2, to the functioning of one of the Par family proteins, aPKC, in cell polarization in the embryonic CNS, in testes and in tissue culture cells [49–51]. These observations prompted us to ask whether orb impacts the process of repolarization of the oocyte during mid-stages of oogenesis, and conversely whether the Par proteins, and in particular, aPKC, have any effect on orb activity.
In wild type, osk mRNA is localized in a tight crescent at the posterior pole of the oocyte after repolarization (Fig 1A) [11, 52]. While osk mRNA localization to the posterior is independent of Osk, Osk protein is required to ensure that osk mRNA is properly anchored to the posterior cortex [52]. In osk protein null mutants, osk mRNA is localized at the posterior, but localization is not properly maintained (Fig 1B). While orb is required for osk mRNA translation, it also plays a role in the proper localization of osk message [43, 44]. As shown in Fig 1C, in orbmel/orb303 chambers, the tight localization of osk mRNA at the posterior pole is lost. Instead, osk mRNA puncta are distributed in a halo around the posterior pole while there is a diffuse pattern of mRNA along the anterior margin of the oocyte. As previously reported, even more extreme defects in osk mRNA localization are evident when orbmel is combined with the null allele orb343 (Fig 1D) [42, 43]. In this allelic combination there is little if any osk mRNA at the posterior.
The osk mRNA localization defects in the hypomorphic orb mutant combinations resemble those in staufen mutants. staufen encodes an RNA-binding protein that co-localizes with osk mRNA throughout oogenesis, and in staufenD3/Df, osk mRNA is partially localized to the posterior and also accumulates at the anterior (Fig 1E) [11, 52, 53]. Thus, one explanation for the defects in osk mRNA localization during mid-oogenesis is that orb is also required to transport osk mRNA [53]. To test this possibility we compared the distribution of Orb protein with that of osk mRNA. Prior to stage 7, both osk mRNA and Orb protein are localized at the posterior. When the MT network commences repolarization during stage 7, osk mRNA transiently accumulates in a cloud near the middle of the oocyte (Fig 2A) [54]. If Orb is directly involved in osk mRNA transport, it would be expected to co-localize with osk mRNA in this cloud. However, it does not. Instead, most of the Orb is concentrated in the sub-cortical region at the posterior end of the oocyte and along the lateral margins of the oocyte (Fig 2A). Only later, after osk mRNA is re-localized to the posterior pole (and presumably being translated) does it again overlap with the posterior cap of Orb protein (Fig 2B). Another orb regulatory target is orb mRNA and its pattern of localization differs from that of osk [55]. In stage 7 chambers, when osk mRNA is in the center of the oocyte, orb mRNA has a circumferential subcortical distribution around the anterior of the oocyte (Fig 2C). This distribution is maintained at later stages (Fig 2D).
Another indication that orb is not directly involved in osk mRNA transport comes from the effects of grk mutations. In grk2B/2E12 ovaries, posterior follicle cell (PFC) specification is defective and the oocyte fails to initiate repolarization at stage 7 [7, 8]. As a consequence, osk mRNA (S1 Fig), the transport protein Staufen, and MT plus ends become enriched in the center of the oocyte, while bicoid mRNA localizes to both the anterior and posterior of the oocyte [7, 8]. In this grk mutant combination Orb protein and also orb mRNA accumulate around the circumference of the oocyte, far from osk mRNA (S1 Fig).
An alternative explanation for the mislocalization of osk mRNA in orbmel/orb303 and orbmel/orb343 egg chambers is that the cytoskeleton is not properly reorganized during repolarization in the absence of normal orb function. This possibility was suggested by the studies of Martin et al. ([56]), who showed that in hypomorphic orb mutant alleles the oocyte MT network is disrupted and there is premature oocyte cytoplasmic streaming. Several approaches were used to confirm and extend their findings.
Repolarization of the oocyte during stage 7 is triggered by signals emanating from the somatic posterior follicle cells (PFCs). The production of the repolarization signal depends upon the proper specification of the PFCs and this process is orchestrated by the expression of the Grk ligand at the oocyte posterior [7, 8]. Since grk mRNA is a known orb regulatory target, one explanation for the repolarization defects is that the PFCs are not properly specified when orb activity is compromised. To test this possibility we examined the expression of an EGFR dependent enhancer trap, kekkon-lacZ, that is activated in follicle cells by grk signaling [68–72]. As illustrated for two stage 7 egg chambers in S4A and S4B Fig we found that kekkon-lacZ expression in PFCs in orbmel/orb343 egg chambers resembles that in control egg chambers. This result confirms previous studies which showed that anterior follicle cell fate is not duplicated in orb343/mel egg chambers [7]. While kekkon-lacZ expression is unaffected in orbmel/orb343 prior to repolarization, abnormalities are evident at later stages. As shown for a stage 10 orbmel/orb343 chamber in S4C and S4D Fig, expression of kekkon-lacZ in dorsal follicle cells is severely reduced compared to the control. This is expected since grk signaling to the dorsal follicle cells is known to be disrupted in orbmel/orb343 ovaries [42, 45, 46].
Other observations are also consistent with the idea that the defects in MT organization in orb are downstream of both the grk dependent specification of PFCs and of the subsequent repolarization signal from the PFCs to the oocyte. For example, in grk mutants, bicoid mRNA is localized not only along the anterior-lateral margin, but also at the posterior pole[7, 8]. In contrast, when orb activity is compromised, localization of bcd mRNAs to the posterior pole is not observed (S5 Fig) [42]. The reason for this difference is that in grk mutants the PFCs fail to signal the disassembly of the MTOC at the posterior of the oocyte, whereas the posterior MTOC is disassembled in orb mutants.
One explanation for the failure to repolarize the MT cytoskeleton is that orb activity impacts either directly or indirectly the functioning of the Par proteins. In fact, precedence for an orb-Par connection comes from experiments showing that one of the targets for the other fly CPEB protein, orb2, in spermatid cyst polarization and in asymmetric cell division in the embryo is the message encoding the apical Par protein aPKC [49, 50]. To explore this idea further we took advantage of the fact orb is weakly haploinsufficient for D-V polarity [55]. About 5% of the eggs laid by orb343/+ are ventralized due to defects in translating grk mRNA at the dorsal anterior corner of the oocyte (Fig 4A). The frequency of D-V polarity defects can be enhanced by reducing the activity of other genes that are important for orb function in grk signaling.
We used three different aPKC mutants, a strong allele, k06403, and two hypomorphic alleles, ex48 and ex55, to test for dominant genetic interactions with orb [74, 75]. While the frequency of D-V polarity defects in eggs produced by mothers heterozygous for these three aPKC alleles is similar to WT (S1 Table), these mutations substantially enhanced the frequency of D-V polarity defects when trans to orb343/+. The weak hypomorphic alleles increase the frequency of ventralized eggs four to five fold (20% and 25%), while the frequency is increased nearly nine fold (44%) by the null allele (Fig 4A).
To extend this analysis, we also asked whether there are genetic interactions between orb343 and the cdc42 gene, which, like aPKC, plays an important role in establishing apical cell polarity [26, 76]. In orb343/ cdc421 trans-heterozygotes there was modest increase (three-fold) in the frequency of D-V polarity defects (S1 Table), while in orb343/cdc424 trans-heterozygotes the frequency of D-V polarity defects increased by nearly fifteen fold (Fig 4A). Consistent with the idea that the effects on grk signaling are related, at least indirectly, to the functioning of the Par proteins in the process of repolarization, we also observed genetic interactions between aPKCk06403 and cdc424. Whereas background levels (~1%) of D-V polarity defects are evident in eggs produced by either aPKCk06403 and cdc424 heterozygotes, over 35% of the eggs laid by trans-heterozygous mothers had D-V polarity defects (Fig 4A).
We also examined eggs produced by females triply heterozygous for orb343, aPKCk06403 and cdc424. In this triple heterozygote about 90% of the eggs have D-V polarity defects (Fig 4A). As would be predicted, accumulation of Grk protein at the dorsal anterior corner of the oocyte is clearly reduced (S6A–S6C Fig). Interestingly, the oogenesis defects are not restricted to grk translation. S6D–S6F Fig also shows that the localization of osk mRNA at the posterior pole is also reduced compared to control egg chambers.
Further evidence that orb might work in conjunction with aPKC in the process of repolarization comes from analysis of oocyte nucleus positioning in backgrounds simultaneously compromised for both genes. As described above, oocyte nucleus mispositioning is observed in ~7% of the orbmel/orb343 egg chambers. The frequency of a mispositioned oocyte nucleus increases to nearly 25% of the chambers when orbmel/orb343 females are also heterozygous for aPKC k06403 (Fig 4B). A similar enhancement is observed when aPKC k06403 is combined with orbmel/orb343 HD19G. In orbmel/orb343 HD19G chambers about 20% have a mispositioned oocyte nucleus, while the frequency of oocytes with a mispositioned nucleus increases to nearly 50% when the orbmel/orb343 HD19G females are also heterozygous for aPKC k06403 (Fig 4B). Importantly, aPKC on its own is not haploinsufficient for proper oocyte nucleus migration (Fig 4B).
One plausible explanation for the genetic interactions is that one of the orb functions in repolarization is to regulate aPKC mRNA. To explore this possibility, we examined the effects of compromising orb on the pattern of accumulation of aPKC mRNA. While aPKC mRNA is present in both somatic and germline cells in wild type ovaries, the highest concentrations of mRNA in the germarium and in stage 1–7 egg chambers are found in the oocyte (S7A and S7B Fig). In stage 9 and older chambers, aPKC mRNA is no longer enriched in the oocyte relative to levels in the nurse cells; however, within the oocyte a fraction of the mRNA localizes along the oocyte cortex with the highest levels of aPKC mRNA towards the anterior of the oocyte and lower levels towards the posterior (arrows in Fig 5A and 5B and S7C Fig). Orb protein also localizes along the lateral cortex of the oocyte in wild type egg chambers (see Figs 2 and 6), while in chambers compromised for orb, Orb association with the cortex is substantially reduced (S3 Fig). Consistent with a role for Orb in anchoring aPKC mRNA during mid-oogenesis, we find that the anterior and lateral cortex associated aPKC mRNA is either partially (Fig 5C) or largely (Fig 5D & 5E) lost when orb activity is depleted by RNAi.
As noted in the introduction, aPKC mRNA is one of several thousand mRNAs that are associated with ectopically expressed Orb2 and Orb in tissue culture cells [48]. To determine if aPKC mRNA is bound by Orb in ovary extracts, we used immunoprecipitation to isolate Orb associated RNAs. After reverse transcription using an oligo dT primer, we used quantitative PCR to assay for specific mRNA species. For the positive control, we used primers for orb-RA 3’UTR which contains four canonical cytoplasmic polyadenylation elements (CPEs: UUUUAU or UUUUAAU). Previous studies have shown that Orb binds to the orb mRNA 3’UTR and positively autoregulates its own expression [55]. There are twelve predicted aPKC mRNA species with six different predicted 3’UTRs. Four of the six predicted 3’UTRs have canonical CPE sequences. One of these, aPKC-RA, has a 3’UTR with three canonical CPEs while the remaining aPKC mRNAs (RD, RF RJ, RK, RL and RM) have overlapping UTRs with 2 canonical CPEs. Fig 5F shows that in ovary extracts both types of aPKC 3’UTRs are enriched in Orb immunoprecipitates.
We next examined the pattern of accumulation of aPKC protein. In wild type stage 10–11 oocytes, aPKC protein is localized to the anterior-lateral cortex where it appears to be in close association with the cortical actin network (Fig 6A) [37]. Except for this cortically localized protein, there is little aPKC elsewhere in the oocyte. Orb is localized just interior to the cortical actin-aPKC layer (Fig 6A). aPKC is also localized along the apical surface of the somatic follicle cells facing the germline, and in confocal images the somatic and oocyte aPKC proteins typically appear as a set of parallel tracks along the anterior-lateral cortex (Fig 6A).
The pattern of aPKC localization in the oocyte is altered when orb activity is compromised. Instead of a tightly organized track coincident with cortical actin, aPKC protein distribution becomes irregular and patchy (Fig 6B). In some regions, there are small gaps (Fig 6B: lower panel: blue arrowhead) while in other regions aPKC is missing altogether (Fig 6B, lower panel: red arrowhead). In other cases, the aPKC protein extends from the cortex into the interior of the oocyte (Fig 6B, middle panel: red arrowhead).
The effects of reducing orb activity on aPKC localization within the oocyte, taken together with the genetic interactions between orb, aPKC and cdc42 indicate that orb is required for the proper functioning of anterior Par proteins. It seemed possible that the posterior Par proteins might also be dependent on orb. To test this idea, we examined the localization of a Par-1-GFP fusion protein that is expressed in the germline. In control stage 8–11 oocytes, the Par-1-GFP fusion protein localizes along the oocyte cortex and tends to be enriched towards the posterior of the oocyte. Fig 6C and S8 Fig show that like aPKC, Par-1-GFP localization depends upon orb, and is disrupted when orb activity is compromised. The extent of disruption is correlated with the severity of the reduction in orb activity. In orbmel/orb303, a small percentage of the chambers have an obvious, but not complete loss of Par-1-GFP association with the oocyte cortex (Fig 6C). Even more extensive alterations are observed in orbmel/orb343 and orbmel/orb343 HD19G chambers. In these genetic backgrounds, more than half of the egg chambers show either a reduction (S8B Fig) or complete loss of cortical Par1-GFP (Fig 6C and S8C Fig).
In addition to its functions in Par dependent polarity, the apical Par protein Cdc42 can also activate effectors of the actin cytoskeleton (Cip4, WASp and Arp23). Studies by Leibfried et al. ([26]) have shown that one of the important Cdc42 targets during oocyte repolarization is the actin cytoskeleton. When cdc42 activity is compromised, the organization of cortical actin is disrupted. While the apical Par proteins aPKC and Baz are not thought to have a direct role in modeling the actin cytoskeleton, they are required for Cdc42 localization. As a consequence, aPKC and baz mutants have equivalent defects in the anterior lateral cortical domain. For these reasons, we wondered whether orb function might also impact the organization of the cortical actin cytoskeleton during repolarization.
To address this question we examined the cortical actin cytoskeleton in orbmel/orb343 and in orb RNAi egg chambers. In the experiment in Fig 7, we labeled follicle cells membranes with Cadherin 99C (Cad99C) antibodies, while the actin cytoskeleton was labeled with phalloidin [77]. In wild type, actin is enriched along anterior oocyte margin and the anterior lateral cortex (Fig 7A and 7A’) [26]. The tight association of actin along the oocyte cortex seen in wild type chambers is disrupted when orb activity is compromised in either orbmel/orb343 oocytes (Fig 7B and 7B’) or when orb RNAi is expressed during midstages by a maternal α-tubulin driver (#7062) (Fig 7C and 7C’). In some regions, the actin matrix is displaced from the cortex (Fig 7B, arrow). In other regions, there are “flares” of actin filaments that extend out from the cortical actin matrix into the ooplasm (Fig 7C’, arrow). The matrix can also unravel forming small bubbles (Fig 7C’, arrowhead) or even disappear completely (Fig 7B’). These defects could be due to a failure to properly crosslink the cortical actin bundles.
aPKC association with the oocyte cortex is thought to depend upon the integrity of the cortical actin cytoskeleton [26]. This raises the possibility that the defects in aPKC localization in orb mutants might be connected to abnormalities in the cortical actin cytoskeleton. The results shown in Fig 6B indicate that this is likely to be the case. In regions where the cortical actin matrix is disrupted, aPKC association with the cortex is reduced or lost. There seems to be a similar connection between the severity of the defects in Par-1 localization and the extent of the abnormalities in the cortical actin cytoskeleton (see S8 Fig). The orbmel/orb343 chambers that have most extensive perturbations in the cortical actin cytoskeleton have more pronounced defects in Par-1 localization (S8C Fig) than in chambers in which the cytoskeleton defects are less severe (S8B Fig).
The repolarization of the MT network during mid-oogenesis depends upon the MT binding protein Patronin and its association with the actin-MT linker Shot. Since the integrity of the cortical actin network is disrupted when orb activity is compromised, we wondered whether Patronin and Shot association with the oocyte cortex is also affected. To investigate this possibility, we compared the localization of Shot-YFP expressed from a BAC transgene and YFP-Patronin expressed from a germline specific mattub promoter (Fig 8) ([21]) in wild type egg chambers and in chambers in which orb activity was knocked down by RNAi.
In wild type stage 9–11 oocytes Shot and Patronin are found associated with the oocyte cortex (Fig 8, S9 and S10 Figs) [21]. In the oocyte, Shot and Patronin are localized in a punctate pattern just underneath the cortical actin network (S9 and S10 Figs). Shot-YFP (S9 Fig) and Patronin-YFP (S10 Fig: expressed as an endogenously tagged protein) also localize to the apical surface of the follicle cells, and these two proteins appear as a parallel track along the lateral surface of the oocyte with the cortical actin network in between. Both Shot-YFP and YFP-Patronin are enriched along the anterior and anterior-lateral cortex, while they are absent from the posterior cortex (Fig 8A–8C; arrowheads). When orb activity is knockdown by RNAi, the association of Shot-YFP and YFP-Patronin with the anterior-lateral cortex of the oocyte is disrupted and much of the protein is instead distributed in the ooplasm (Fig 8B and 8D). Similar, though not quite as severe alterations in the cortical association of Shot-YFP and Patronin-YFP are observed in stage 9–11 orb343/orbmel egg chambers (S9B and S9B’ Fig and S10B–S10B” Fig).
Par proteins establish and maintain polarity within a cell by both positive and negative cross-regulatory interactions. For this reason it seemed possible that aPKC and orb function in the oocyte might be mutually interdependent. To explore this possibility we used the mid-oogenesis GAL4 driver maternal α-tubulin (7063) to express aPKC RNAi (35140). In this background, we observed that the oocyte nucleus is mispositioned in 69% of the stage 9–11 egg chambers when aPKC activity is depleted. Accompanying the oocyte nucleus position defects, Gurken protein is mislocalized with the oocyte nucleus (Fig 9A and 9B). Additionally, there are alterations in the pattern of Orb protein accumulation. Instead of being distributed subcortically along the entire surface of the oocyte, high levels of Orb accumulate at the anterior oocyte-nurse cell margin (Fig 9C and 9D). There is also a reduction in the posterior cap of osk mRNA compared to wild type (Fig 9E and 9F). Similar effects on the positioning of the oocyte nucleus and the localization of polarity markers (Staufen and Vasa) have been reported for cdc42 [26]. Moreover, like orb and cdc42, the cortical actin network is also perturbed in the aPKC knockdown (Fig 9H and 9H’).
The alterations in the pattern of Orb protein accumulation in the RNAi knockdown experiments prompted us to ask whether aPKC impacts orb autoregulation. Orb promotes its own expression through sequences in the orb mRNA 3’UTR. When the orb 3’UTR is linked to coding sequences for E. coli β-galactosidase in the HD19 transgene, expression of β-galactosidase becomes dependent upon orb activity [55]. S11A Fig shows that β-galactosidase expression from the HD19 (hsp83: lacZ-orb 3’UTR) transgene is also dependent upon aPKC activity. In the aPKC mutant combination, aPKCk06403/aPKCex48, β-galactosidase expression is reduced about two-fold compared to the control (S11B Fig).
Like CPEB proteins in other species, orb activity is regulated by phosphorylation [78]. In wild type ovaries, there are multiple phosphorylated isoforms. On standard SDS polyacrylamide gels these different Orb isoforms typically resolve into a closely spaced doublet with the more heavily phosphorylated isoforms migrating more slowly (S11C Fig). In S11C Fig, we compared the relative yield of the upper (more phosphorylated) and lower (less phosphorylated) bands in wild type and aPKCk06403/aPKCex48 mutant ovaries. In the aPKCk06403/aPKCex48 the ratio of upper to lower bands is reduced compared to wild type (S11D Fig).
Previous studies have implicated orb in the translational regulation of osk and grk in the stages following the repolarization of the MT network [43–45]. In addition, the proper localization of these mRNAs also depends orb activity [42–45]. This observation led to the idea that in addition to controlling translation, orb might also have a role in transport and/or anchoring of these mRNAs once they were properly localized. While our results argue against a direct role in transport, they support the idea that the mislocalization of osk and grk mRNAs when orb activity is compromised during mid-oogenesis arises at least in part because orb is required for the proper organization of both MTs and the cortical actin cytoskeleton.
The reorganization of the MT network after stage 7 is a multistep process. It begins with a signal from the PFCs that induces the disassembly of the MTOC that is located just posterior to the oocyte nucleus. The production of this somatic signal depends upon the proper specification of the PFCs, and PFC specification requires the expression of Grk protein at the posterior pole of oocyte earlier in oogenesis [7, 8]. Translation of grk mRNA at the posterior pole during stages 1–7 depends upon orb, and consequently it functions upstream of PFC specification. However, in our experiments orb activity prior to stage 7 is not limiting, and sufficient amounts of Grk are expressed to properly specify PFCs [7]. Thus, the defects that we observe in oocyte repolarization when orb activity is compromised during mid-oogenesis are downstream of both the grk signal to the posterior follicle cells and the signal from the PFCs to the germline that induces MTOC disassembly.
Three other findings are consistent with this conclusion. First, when PFCs are not properly specified, the posterior MTOC fails to disassemble [7, 8, 18]. By contrast, when orb activity is compromised during mid-oogenesis the MTOC dissembles as in wild type. Second, the formation of a non-centrosomal cortical based MT network is initiated along the anterior/lateral margin of the oocyte even in the absence of the PFC signal. This is not true in our experiments; the anterior/lateral MT network is not properly established. Third, in the absence of the PFC signal, Staufen protein, Kinesin-β-gal and osk mRNA concentrate in the center of the oocyte, while bcd mRNA is found not only at the anterior but also at the posterior end of the oocyte. In contrast, in orb mutants, osk and also bcd mRNA accumulate at the anterior of the oocyte, while Kinesin-β-gal is unlocalized.
As the posterior MTOC is disassembled, a MT network emanating from the anterior and anterior-lateral cortex of the oocyte is established. The initiation of this non-centrosomal cortical based MT network is mediated by the spectraplakin, Shot, and the minus-end MT binding protein, Patronin [21]. Shot associates with the actin rich anterior and anterior lateral cortex and recruits Patronin. Patronin then nucleates the assembly of the MT network. Nashchekin et al. ([21]) have shown that proper polarization of the MT network by Shot and Patronin depends upon the Par protein Par-1. By an unknown mechanism, Par-1 blocks Shot association with the actin rich cortex. Since Par-1 is enriched around the posterior cortex of the oocyte, this restricts the de novo assembly of MTs to the anterior and anterior-lateral cortex. While Par-1 is required to exclude Shot from the posterior cortex, the de novo assembly of MTs requires Shot association with the anterior and anterior-lateral cortex. This presumably does not happen when aPKC, cdc42 and/or baz are compromised in the germline because the anterior and anterior-lateral cortical actin network is disrupted.
Our results place orb upstream of Shot and Patronin and suggest that the defects in oocyte MT repolarization likely arise for several reasons. One would be defects in the localization and functioning of the Par gene products. When orb activity is compromised, the association of the Par protein Par-1 with the posterior and aPKC with the anterior-lateral cortex is disrupted. In the absence of proper cortical association, the cross-regulatory interactions between the anterior and posterior Par proteins would be expected to be ineffective. Also consistent with a role for orb in the functioning of the Par proteins in MT repolarization are genetic interactions between orb and genes encoding the anterior Par proteins, aPKC and cdc42. orb is weakly haploinsufficient for the grk signaling pathway, and about 5% of the eggs laid by orb343/+ females, are ventralized. This weak haploinsufficiency is enhanced when the orb343 mutation is trans to mutations in either aPKC or cdc42. For the aPKC null allele, aPKC k06403, the frequency of ventralized eggs increases to nearly 50%, while about 70% of the eggs laid by females trans-heteozygous for orb343 and cdc424 are ventralized. Moreover, while females heterozygous for either aPKC k06403 or cdc424 alone do not lay ventralized eggs, nearly 40% of the eggs laid by females trans-heterozygous for these two mutations are ventralized. As we found for orb, the localization of the oocyte nucleus to the dorsal anterior corner of the oocyte depends upon cdc42 and aPKC. Leibfried et al. ([26]) found that the oocyte nucleus is mispositioned in egg chambers homozygous for cdc424, while we have shown here that the oocyte nucleus is mispositioned when aPKC is knocked down by RNAi. Moreover, the frequency of mispositioned nuclei in orbmel/orb343 is enhanced when the females are also heterozygous for mutations in aPKC.
At least some of the effects of orb on the Par proteins could be direct. Thus, aPKC mRNAs contain CPEs in their 3’UTRs and we have found that aPKC mRNA is bound by Orb protein in ovary extracts. Moreover, the distribution of aPKC mRNA within the oocyte is altered when orb activity is compromised. Interestingly, mRNAs encoding the three other anterior Par proteins, cdc42, baz, and par-6 also have CPE motifs in their 3’UTRs and are bound by ectopically expressed Orb in tissue culture cells [48]. Thus, the localization and translation of these Par mRNAs could be regulated by or dependent upon orb. In addition, there appears to be a reciprocal relationship between orb and anterior Par proteins. This is suggested by the synergistic genetic interactions between orb and the Par genes encoding aPKC and cdc42. It also fits with our finding that orb autoregulatory activity and the phosphorylation status of Orb are impacted by aPKC depletion.
There are also likely to be indirect effects on the functioning of the Par proteins that in turn perturb the organization of the MT network. For example, Leibfried et al. ([26]) have shown that there is a mutually interdependent relationship between the Par proteins and the actin cytoskeleton. They found that Cdc42 localization along the anterior and anterior-lateral cortex of the oocyte depends upon the integrity of the cortical actin network. Conversely, the assembly of the cortical actin network requires cdc42, aPKC and baz. In fact, one of the more striking phenotypes in orb mutant oocytes is the disorganization of the cortical actin network. As was observed for cdc42 [26], the disruptions in the actin network are accompanied by the mislocalization of aPKC. Given the interdependence of the Par proteins and the actin network the disruption of the actin cytoskeleton in orb mutants could be due to the misexpression of the Par proteins. However, the Par proteins need not be the only or even the key targets for orb regulation of the actin cytoskeleton. For example, the formation of the cortical actin network during mid-oogenesis depends upon two actin nucleators, capu and spir [61–64]. Mutations in these two genes have a number of phenotypes in common with orb. The actin cytoskeleton is fragmented and this in turn leads to a failure to properly organize the MT network and localize osk and grk mRNAs. Moreover, as has been reported for orb [56], premature cytoplasmic streaming is observed in capu and spir mutant egg chambers. Like the Par proteins, the mRNAs encoding capu and spir are bound by ectopically expressed Orb in tissue culture cells, and thus could be targets for orb regulation. On other the hand, there are some notable differences. In contrast to orb, aPKC and cdc42, capu and spir eggs are dorsalized not ventralized. Additionally, Par-1 localization to the posterior and lateral cortex does not appear to depend upon capu or spir [62] whereas it is disrupted in orb mutant chambers. Moreover, the effects of orb on the actin cytoskeleton need not be limited to these proteins. The mRNA encoding the actin effectors Cip4 and WASp have CPEs in their 3’UTRs and are bound by ectopically expressed Orb in tissue culture cells [48]. Defects in the expression of these proteins would interfere with the remodeling of the anterior/anterior-lateral cortical actin cytoskeleton and consequently disrupt Par dependent MT polarization. Finally, orb could also act downstream of the Par proteins. Like cip4 and WASp mRNAs, the mRNAs encoding the MT assembly factors, shot and patronin, have CPEs in their 3’UTRs and are bound by ectopically expressed Orb in tissue culture cells [48]. Insufficient levels of these factors would be expected to slow or block the de novo assembly of MTs along the anterior-lateral cortex.
Thus, a plausible idea is that the defects in the repolarization of the MT network when orb is depleted during mid-oogenesis are likely the consequence of the cumulative effects of misregulating mRNAs encoding not only Par proteins but also proteins involved in organizing the actin cytoskeleton and assembling MTs. Because the MT and actin cytoskeleton regulators have interdependent functions, even small perturbations in the abundance of multiple players could lead to wide ranging disruptions in cytoskeletal organization. That mRNA localization/translational regulation might impact the reorganization of the egg chamber after stage 7 at multiple levels is supported by recent studies on egalitarian (egl). Sanghavi et al. ([67]) report that knocking down egl just before the MT network in the egg chamber is repolarized induces many of the same phenotypic abnormalities and disruptions in cytoskeletal organization that we have observed when orb activity is compromised during mid-oogenesis. Egl together with the Bicaudal-D (BicD) protein loads mRNAs onto a Dynein motors [79–81]. This mRNA cargo complex is responsible for localizing mRNAs in somatic cells and in developing egg chambers. Like Orb, the Egl-BicD cargo complex interacts with many different mRNA species including orb. For this reason, loss of egl activity is likely to have a global impact on mRNA transport and consequently the localized production of a diverse array of factors needed for the reorganization of the oocyte cytoskeleton during mid-oogenesis.
Endogenously tagged Patronin-YFP, YFP-Patronin expressed from a maternal tubulin promoter and Shot-YFP ([21]) were gifts from Daniel St Johnston and Dmitry Nashchekin; osk54, osk84, stauD3, stauDf, KZ32 (Kinesin-β-gal) are gifts from Elizabeth Gavis; grk2B, grk2E12, BB142 (kekkon-lacZ) are gifts from Trudi Schupbach; aPKC mutant alleles aPKCk06403, aPKCex55, aPKCex48 and mattub-GFP-Par-1-N1S are gifts from Yu-Chiun Wang and Eric Wieschaus; cdc421 and cdc424, aPKC RNAi 35140, maternal alphaTubulin67C Gal4 (7062 and 7063) from Bloomington Stock Center.
Eggs were collected by placing flies of the appropriate genotype into cups and were kept at 18 degrees and given fresh apple juice and yeast paste plates daily. The eggshell phenotypes were scored starting on day 3.
osk FISH probes were a gift from Shawn Little at University of Pennsylvania [54].
orb FISH probes were ordered from Biosearch Technologies, and orb probes and aPKC-com FISH probes (from Xu et al. [49]) were coupled to Atto NHS-Ester 565 or 633 (Sigma) and purified using HPLC.
Antibodies used were as follows: mouse anti-Orb (4H8, 6H4) used 1:30 each, mouse anti-Gurken (1D12) used 1:20, mouse anti-β-gal (401A) used 1:10, mouse anti-Bic-D (1B11, 4C2) used 1:20 each from the Developmental Studies Hybridoma Bank; rabbit anti-Cadherin99C used 1:1000 was a gift of Dorothea Godt; mouse monoclonal anti-α-tubulin-FITC (clone DM1A) from Sigma; rabbit anti-aPKC (clone c-20, sc-216) used 1:1000 from Santa Cruz Biotechnology. Wheat germ agglutinin (Alexa Fluor 633, Molecular Probes), Phalloidin (Alexa Fluor 546 or 633, Molecular Probes) and DAPI (Molecular Probes) were used.
Secondary antibodies used were goat anti-mouse IgG Alexa 488, 546 or 647, goat anti-rabbit Alexa 488, 546 or 647 (Molecular Probes). Samples were mounted using aqua polymount (Polysciences) on slides and visualized on a Leica SP5 or Nikon A1 confocal microscope.
Cytoplasmic movements were imaged in live oocytes in halocarbon oil on a Nikon A1 inverted confocal microscope. An image was collected every 5 seconds for at least 2 minutes to visualize cytoplasmic streaming.
Mouse anti-Orb (4H8 and 6H4) or mouse anti- β-gal (401A) were coupled to A/G agarose beads (Santa Cruz Biotechnology) by incubating overnight at 4 degrees. 250 females were dissected in ice cold 1xPBS and ovaries were transferred to dry ice while dissecting. RNAsin (Promega) was added to ovaries and they were crushed using a plastic pestle to make a paste. The ovary paste was centrifuged at 3000 rpm for 5 minutes at 4 degrees twice, and the supernatant was saved. Half of the supernatant was added to the Orb antibody coupled with beads, and the other half was added to the control antibody beads. CoIP buffer ([55]) and RNasin (Promega) was added to IPs, which were left to rotate for 3 hours at 4 degrees. The beads were pelleted by centrifugation and washed with coIP buffer 5 times.
RNA was released from the beads by adding 10 mM HEPES 1% SDS solution and β-mercaptoethanol, and left in a 65 degree water bath for 15 minutes. Phenol followed by phenol chloroform was used for extraction, and the water phase was ethanol precipitated with glycogen added as a carrier. The pellet was dried and then DNAse (Promega) treated.
The RNA samples were incubated with oligodT (IDT) at 65 degrees for 10 minutes. AMV reverse transcriptase (Promega) reactions were set up, and for each IP a control reaction was set up without reverse transcriptase. The samples went through the following program for reverse transcription on a PCR machine: 55 degrees for 1 min, 48 degrees for 30 min, 55 degrees for 15 min, 95 degrees for 5 min, hold at 4. For quantitative PCR, Power CybrGreen PCR master mix (Life Technologies) was used. Each qPCR reaction was done in triplicate and the average CT was used. The control samples without reverse transcriptase were also run to confirm the DNase treatment worked. The amplification of target 3’UTRs from the Orb IP were compared to the amplification from the control IP and normalized to a control (RPL32) to calculate ΔΔCT.
For the Western blots to measure levels of β-gal expression ovaries were dissected in PBS and frozen on dry ice. Frozen tissue was crushed with a pestle in SDS buffer with urea, boiled and spun down. The extracts were loaded on a 10% SDS-Page gel. Proteins were transferred to a PVDF membrane and the membrane was cut to blot for β-gal and BEAF. For the phosphorylated Orb isoforms, ovaries from two female flies were dissected in 100 ul of 1X PBS. The ovaries were immediately transferred to 40 ul of 2XSDS buffer (100 mM Tris-Cl; 4% SDS, 200 mM DTT and 0.2% bromphenol blue) and boiled. A second set of ovaries were dissected, transferred to the same tube and boiled. The samples were then loaded onto a 7.5% SDS polyacrylamide gel. Image J was used to measure the Orb protein upper:lower band ratio.
|
10.1371/journal.pcbi.1003813 | Protein Surface Softness Is the Origin of Enzyme Cold-Adaptation of Trypsin | Life has effectively colonized most of our planet and extremophilic organisms require specialized enzymes to survive under harsh conditions. Cold-loving organisms (psychrophiles) express heat-labile enzymes that possess a high specific activity and catalytic efficiency at low temperatures. A remarkable universal characteristic of cold-active enzymes is that they show a reduction both in activation enthalpy and entropy, compared to mesophilic orthologs, which makes their reaction rates less sensitive to falling temperature. Despite significant efforts since the early 1970s, the important question of the origin of this effect still largely remains unanswered. Here we use cold- and warm-active trypsins as model systems to investigate the temperature dependence of the reaction rates with extensive molecular dynamics free energy simulations. The calculations quantitatively reproduce the catalytic rates of the two enzymes and further yield high-precision Arrhenius plots, which show the characteristic trends in activation enthalpy and entropy. Detailed structural analysis indicates that the relationship between these parameters and the 3D structure is reflected by significantly different internal protein energy changes during the reaction. The origin of this effect is not localized to the active site, but is found in the outer regions of the protein, where the cold-active enzyme has a higher degree of softness. Several structural mechanisms for softening the protein surface are identified, together with key mutations responsible for this effect. Our simulations further show that single point-mutations can significantly affect the thermodynamic activation parameters, indicating how these can be optimized by evolution.
| Cold-adapted organisms require specialized enzymes to maintain functional integrity at low temperatures, and psychrophiles express heat-labile enzymes that possess a high specific activity and catalytic efficiency at low temperatures. The high catalytic rates are achieved by enzyme adaptations yielding lower activation enthalpies and entropies than for mesophilic homologs, thereby solving the problem of the exponential rate decrease with falling temperature. However, the structural mechanisms behind this universal property of cold-adapted enzymes remain unknown. By extensive computer simulations, which reproduce both the experimental reaction rates and the characteristic temperature dependence of activation free energies, we show that it is the softness of the protein-water surface that regulates the activation enthalpy-entropy balance. Structural mechanisms behind this phenomenon are identified and our simulations show that single mutations can significantly affect the thermodynamic activation parameters, indicating how these can be optimized by evolution.
| One of the most intriguing problems in biology regards the molecular mechanisms involved in adaptive capabilities for life in extreme environments. Cold-adapted organisms have an extraordinary ability to grow in and colonize environments where the temperature is close to the freezing point of water. From the viewpoint of chemical kinetics, a key problem with lowering the temperature is that the enthalpy of activation gives rise to an exponential decrease in enzyme reaction rates according to transition state theory(1)Here, krxn is the reaction rate and T the temperature, κ is a transmission coefficient, k and h are Boltzmann's and Planck's constants, respectively, and ΔG‡ is the free energy of activation. The latter quantity can be decomposed into entropic (−TΔS‡) and enthalpic (ΔH‡) contributions and decreasing the temperature from 37°C to 0°C typically results in a 20–250 fold reduction of the activity of a mesophilic enzyme [1]. Survival at low temperatures thus requires that the enzyme kinetics can be adapted to avoid this problem and also that protein stability is maintained in a cold environment.
As a strategy to cope with the strong temperature dependence of the reaction rates, psychrophiles synthesize heat-labile enzymes possessing a high specific activity and catalytic efficiency at low temperatures [2], [3], [4]. It is thus well established that cold-adapted enzymes generally have reduced thermal stability compared to mesophilic orthologues, presumably to counteract the increase in structural rigidity at lower temperatures [1], [5], [6]. However, the change in structural stability does not seem to follow any general rule, but is rather a combination of several factors [5]. More remarkable, however, is the seemingly universal characteristic that catalyzed reactions of cold-adapted enzymes have a lower enthalpy and a more negative entropy of activation than their mesophilic and thermophilic counterparts [1], [5], [6]. Overall activation free energies, on the other hand, are usually similar around room temperature [5]. The lower activation enthalpy thus makes the rate less temperature dependent (equation (1)) and is believed to be the primary adaption in psychrophilic enzymes [2], [7], [8]. It has long been proposed that cold-adaptation originates from increased flexibility of the active site [6], which could hypothetically yield lower activation enthalpies at the expense of requiring more ordering of substrates and the active site, as the reaction barrier is surmounted (i.e., a more negative ΔS‡). However, there seems to be no strong experimental support for this hypothesis and, e.g., X-ray analysis of cold- and warm-active trypsin did not indicate any overall flexibility differences between the two enzymes [9]. Moreover, recent computer simulations of differently adapted citrate synthases showed that the flexibility of the highly conserved active site residues was virtually identical. Instead it was found that differences in protein stiffness outside of the active site appear to be correlated with differences in thermodynamic activation parameters [10].
The origin of catalytic rate optimization in cold-adapted enzymes, in terms of actual structure-function relationships, thus remains rather obscure. Understanding such relationships would not only provide information regarding the evolutionary adaption processes, but potentially also enable rational design of enzymes adapted to low temperature. Computer simulations could provide a unique way of analyzing the reaction energetics of differently adapted enzyme orthologs. However, in order for such a strategy to be viable several criteria must be met. First, analysis of indirect or circumstantial factors (flexibility, electrostatics, hydrophobicity etc.) alone does not suffice for obtaining conclusive evidence. Instead reliable free energy profiles along the reaction pathway must be obtained with high precision. Second, the crucial activation enthalpy-entropy balance for different enzymes must be reproduced by the simulations and the only way to do this is to computationally obtain Arrhenius plots for the activation free energy versus temperature. This involves calculating a large number of free energy profiles at different temperatures so that activation enthalpies and entropies can also be extracted with high precision. Clearly, such extensive sampling by molecular dynamics (MD) simulations precludes the use of most standard QM/MM approaches, but the empirical valence bond (EVB) model [11], [12]. provides a very efficient method for this purpose. Third, provided that the experimentally observed activation enthalpy-entropy balances are captured by the simulations, it must be possible to decompose these into their underlying energy components and ultimately translate them into differences between the enzyme 3D structures and fluctuations.
Here, we report extensive MD/EVB free energy simulations that yield high precision Arrhenius plots for the reactions of psychrophilic and mesophilic trypsins. The calculations reproduce both experimental rates at room temperature and the characteristic relationships between activation enthalpy and entropy for the orthologous salmon and bovine enzymes. The relationship between these parameters and the 3D enzyme structures is reflected by significantly different internal protein energy changes during the reaction. This effect originates from outside of the active site where the cold-adapted salmon enzyme has a higher degree of softness, which is evident from the corresponding potential energy term. We also identify key residues for which simulations predict significantly altered thermodynamic activation parameters upon mutation.
Atomic coordinates for psycrophilic and mesophilic trypsin were obtained from the crystallographic structures with PDB entries 1BZX [13] and 3BTK [14], respectively. All EVB calculations were performed with the molecular dynamics package Q [15] using the OPLS2005 all-atoms force field [16], [17]. Additional simulations details are given in Text S1. The EVB reaction surface was calibrated using the imidazole catalyzed methanolysis of formamide in water [18] as a reference reaction (Text S1). The EVB free energy profiles were calculated using the free energy perturbation (FEP) umbrella sampling approach described elsewhere [11], [12]. Each enzyme and water reaction free energy profile involved 500 ps of MD simulation and compromised 51 discrete FEP steps. Thermodynamic activation parameters were obtained from Arrhenius plots based on simulations at eight different temperatures (275–310 K). At each temperature point 100 and 150 independent FEP simulations were carried out, resulting in a total simulation time of 408 and 612 ns for salmon and bovine trypsin, respectively. In addition 100 ns simulation time was performed at the reactant and transition state at 300 K for both systems. Enzyme mutations were created using the builder tool in Mastro 9.1 (Schrödinger, LLC, New York, NY, 2011). The mutated residues were relaxed prior to MD simulation with the clean up geometry tool in Maestro. In order to obtain reliable sampling, the simulations were repeated 20–60 times at each temperature (275–310) for the mutated model systems.
Serine proteases are enzymes that catalyze the cleavage of peptide bonds in proteins and peptides and have numerous important physiological functions. They have been extensively studied for many decades and the reaction scheme involves formation of a Michaelis-Menten complex, nucleophilic attack by the characteristic serine residue to form an acyl-enzyme intermediate and subsequent hydrolysis of this intermediate to yield the final products [19], [20]. These enzymes have an invariant catalytic triad, which in trypsin is formed by Ser195, His57 and Asp102. The histidine residue acts as a general base for activating the serine side-chain, while Asp102 is essential for stabilizing the resulting protonated form of the histidine [21]. The rate-limiting step of the reaction is generally considered to be the formation of a transient tetrahedral intermediate, the breakdown of which leads to acylated enzyme. The large rate acceleration compared to uncatalyzed peptide bond hydrolysis is primarily accomplished by facilitating formation of the reactive nucleophile and by transition state stabilization. Here, the so-called oxyanion hole, formed by the backbone NH groups of Gly193 and Ser195, also plays a key role by stabilizing the developing negative charge (oxyanion) of the tetrahedral intermediate [21].
We used the reactions of the mesophilic bovine trypsin (BT) and the psychrophilic anionic salmon trypsin (AST) as models to examine the temperature dependence of reaction rates for differently temperature adapted enzymes. The energetics of the rate-limiting formation of the tetrahedral intermediate, using a Cys-Lys-Ala tripeptide as substrate, was calculated by the MD/EVB approach [11], [12]. The results from these simulations at 300 K are shown in Fig. 1a as free energy profiles along the reaction coordinate for the two enzymes. The corresponding free energy profile for the reference reaction used to calibrate the EVB potential (see Text S1), i.e., imidazole catalyzed formation of the tetrahedral intermediate in water [18], is also shown. In order to attain a sufficiently high precision the calculations were averaged over up to 150 independent runs at each temperature (see below). The calculated activation energies at 300 K of 18.2±0.2 kcal/mol and 19.0±0.2 kcal/mol for AST and BT, respectively, are in excellent agreement with the substrate dependent barrier of 15–20 kcal/mol [19]. This difference in activation free energies translates into a 4-fold increase in kcat for AST when compared to BT, which is in remarkable good agreement with experiments that shows 2- to 4-fold increase depending on the temperature [22]. The simulations also clearly demonstrate the large catalytic effect on the reaction for both enzymes. Compared to the uncatalyzed hydrolysis reaction in water, the transition state is found to be stabilized by over 13 kcal/mol [18], [23]. With respect to the imidazole catalyzed reference reaction in solution the corresponding stabilization is about 7 kcal/mol [18].
Since the catalytic rates of the two trypsins at room temperature are well reproduced by the MD/EVB simulations, we can now turn to examine their temperature dependence. Eight different temperatures were chosen in the range of 275 to 310 K and 100–150 independent free energy profile calculations were carried out at each temperature to obtain high precision Arrhenius plots. Activation entropies and enthalpies were then extracted by linear regression from plots of ΔG‡/T vs. 1/T. The temperature dependence of the activation free energies is shown in Fig. 1b and it can immediately be seen that the psychrophilic enzyme (AST) has a significantly smaller slope than the mesophilic counterpart (BT). The calculated activation parameters for BT are ΔH‡ = 20.4 kcal/mol and ΔS‡ = 3.5 e.u, while the corresponding values for AST are ΔH‡ = 9.9 kcal/mol and ΔS‡ = −27.5 e.u. This is thus a remarkable example of enthalpy-entropy compensation where the large differences in ΔH‡ are balanced by -TΔS‡ contributions at 300 K of −1.4 and +8.3 kcal/mol for BT and AST, respectively (Table 1), to yield similar activation free energies. It should be noted that an increase in the activation free energy of 1 kcal/mol directly translates into a 5-fold decrease in kcat. The fact that both the absolute rates at 300 K and the characteristic balance between activation enthalpy and entropy for the mesophilic and psychrophilic enzymes are reproduced by the computer simulations is also remarkable and raises the question of what the structural origin of this effect really is.
As far as energetics is concerned it is relatively straightforward to identify the source of the difference in activation enthalpy between the two enzymes. Since ΔH‡ = ΔU‡+pΔV‡, and the pressure-volume term is completely negligible, the activation enthalpy is determined by the corresponding change in internal (total) energy of the system. The latter can be decomposed into contributions from the reacting fragments (i.e., the EVB atoms whose interaction parameters change along the reaction), their interactions with the surrounding protein and solvent, and the interactions within the surrounding environment(2)Here, the subscripts r and s denote the reacting fragments and surroundings (the protein and solvent included in the simulations), respectively. The last term of equation (2) involves very large energies, since it pertains to a huge number of interactions within the surrounding protein and solvent, making it practically impossible to obtain a converged value for this quantity directly from the MD simulations. However, since both ΔH‡ and can be evaluated from the trajectories with sufficiently high precision we can still get an accurate estimate of all the terms in equation (2). Table 1 shows this breakdown of the energetics which immediately reveals that the source of the decreased activation enthalpy in the cold-adapted enzyme is not associated with a more favorable term. Instead it is a significantly lower value of that is responsible for the decrease in ΔH‡. Hence, while the internal energy change involving the reacting groups is similar, the contribution from the surroundings is predicted to be about 9 kcal/mol more favorable for AST than BT. It would be desirable to further decompose into protein-protein, protein-water and water-water interaction contributions according to(3)but, again, the energies involved are too large to allow converged direct calculations of these averages. However, from the viewpoint of locality it is reasonable to expect that the two first terms involving protein interactions dominate the reduction in or the cold-adapted enzyme. That is, the protein interactions are likely to respond more strongly to the energy change in the active site, associated with climbing the activation barrier, since the active site is primarily embedded in the protein, which in turn is surrounded by water.
At any rate, we can conclude that the reduction of activation enthalpy in the cold-adapted enzyme originates from interactions outside of the active site. This is perhaps not so strange since all residues surrounding the substrate are conserved between the two proteins, making it more likely that energetic differences are to be found farther away. The fact that the energy cost reflected by the term is lower in the cold-active than the warm-active trypsin further suggests that the surroundings of the active site are effectively softer in the salmon enzyme. In this respect, the term “softness” can be more precisely defined than protein flexibility in general, as it refers to the change in potential energy of the surroundings of the active site as the system moves along the reaction coordinate form reactants to transition state. This potential energy change can thus be viewed as reflecting an effective force constant of the surroundings, which is stiffer in the warm-active enzyme and softer in the cold-active. This brings us back to the possible role of protein flexibility in cold-adaptation.
Cold-adapted enzymes are often assumed to benefit from higher flexibility to deal with the decrease in chemical rates and altered structural rigidity at low temperatures. Since the activation entropy is also more negative than for mesophilic homologs, this could be interpreted in terms of an increased flexibility of the active site in the reactant state [8]. This proposal was, however, not supported by Bjelic et al. who evaluated the positional root-mean-square fluctuations (RMSF) of the key residues in the active site of different temperature-adapted citrate synthases [10]. They demonstrated that the active site and substrate mobilities were virtually identical and found no indication of the cold-adapted enzyme having larger active site RMSFs compared to the heat-adapted enzymes. The fluctuations obtained with a spherical boundary model were also found to be virtually identical to those obtained with a much larger simulation system simulated using periodic boundary conditions. It should be noted that the present calculations were carried out with the entire protein immersed in a spherical droplet of water (Fig. S1).
To further examine the flexibility hypothesis, we carried out additional 100 ns simulations at both the transition and reactant states for BT and AST. As in Refs. [10] and [24], we again find that the mobility of the active site is low and practically identical in the two enzymes (Fig. S2). Furthermore, the overall protein backbone RMSFs are very similar with calculated values of 0.65 Å and 0.66 Å for BT in the reactant and transition state, respectively, while the corresponding values for AST are 0.61 Å and 0.65 Å. A plot of the average backbone positional fluctuations versus amino acid sequence (Fig. 2a), however, shows as expected that there are local differences in mobility and that these mainly are found on the protein surface. For example, Tyr97 and Asp150 in AST are significantly more flexible than their corresponding BT residues. Both Tyr97, situated in the Nβ5-Nβ6 loop, and Asp150 of the so-called autolysis loop are also conserved through different cold-adapted trypsins (Fig. 3). Moreover, further analysis of the backbone RMSFs shows that the prevalence of residues with high mobility, measured radially from the active site, differs significantly between bovine and salmon trypsin (Fig. 2b). That is, while both enzymes become more flexible further away from the active site, the cold-adapted protein has a markedly higher prevalence of residues with high RMSF values beyond 10 Å from the active site. The conclusion is thus that both enzymes have a relatively rigid core and softer outer regions, but that the surface regions of the cold-adapted enzyme are, at least locally, softer than for the warm-adapted protein.
Both the energetic and mobility analysis above strongly suggest that the surface of the cold-adapted enzyme is softer compared to its warm-active counterpart and the key question now is how this could be accomplished. Here, ultra-high resolution (0.75–1.0 Å) crystal structures of both BT [25] and AST [26] turn out to provide valuable clues since a large number of surface bound water molecules are resolved in these structures. Such water molecules very often allow polar surface side-chains, like those of Asn, Gln, Ser and Thr, to engage in extensive H-bond networks back to the protein surface. Charged surface side-chains, on the other hand, preferentially point out into solution due to their stronger requirement for solvation by bulk water. Analysis of the key mutations identified in Fig. 3 with these high-resolution structures [25], [26] reveals some basic principles for how the protein surface can be made softer by point-mutations.
First, the surface can be softened by disrupting a water mediated H-bond network through mutation of a polar to a less polar residue. This is exemplified in the trypsins by mutation of Asn97 in BT to Tyr97 in AST, which largely abolishes such a network (Fig. 4a). Second, a surface H-bonding network can also be disrupted by mutation of a polar residue into a charged one, since the latter may prefer to interact with bulk solvent. This is exemplified by the mutation of Ser110 in BT to Lys110 in AST (Fig. 4b). A single or few mutations may also completely change one surface H-bond network into another such as the structurally correlated Thr21Glu, Ser150Asp and Val154Lys mutations, which drastically affect the conformation of the autolysis loop and make the active site region more solvent accessible in AST (Fig. 4c). Finally, one can also identify mutations that destabilize the packing of hydrophobic surface patches by mutation of a nonpolar residue into a charged one. This is, e.g., the case with the mutation of Val90 in BT to Arg90 in AST (Fig. 4d).
Of the mutations mentioned above, we will focus on Asn97Tyr and Ser150Asp since both of these are strictly conserved in the cold-adapted trypsins (Fig. 3) and are also the ones showing the largest increase in flexibility compared to the warm-adapted enzyme (Fig. 2a). We thus again calculated free energy profiles at different temperatures to obtain Arrhenius plots for the four cross-species mutations BTN97Y, BTS150D, ASTY97N and ASTD150S, in order to be able to predict their thermodynamic activation parameters. The resulting calculated free activation energies remain essentially unchanged by the mutations (Table 1), which underscores the general notion that mutations far away from the active site do not significantly affect catalytic rates [27]. However, what is remarkable is that the calculations predict significant changes in ΔH‡ and ΔS‡ for most mutations, but that these are again nearly perfectly compensating. Thus, both the BTN97Y and BTS150D mutations markedly lower the activation enthalpy and make the entropy more negative compared to the wild-type bovine enzyme and they become more like the cold-adapted AST. For the reverse mutations, ASTD150S renders the cold-adapted enzyme more mesophilic-like with a significantly raised ΔH‡ and a more positive ΔS‡. This is also seen by the predicted effect the autolysis loop structure, which approaches the bovine conformation (Fig. S3). The ASTY97N mutation, on the other hand, yields relatively smaller effects on both the activation parameters and . This probably just reflects the fact that correlated mutations (e.g., ASTS95N) may be needed to build up the native bovine H-bond network involving the Nβ5-Nβ6 loop (Fig. 4a), so that a single point-mutation does not suffice.
Since both of the residues mutated are involved in distinct H-bond networks in the bovine enzyme, which appear to rigidify the surface, it is logical that a single mutation could destroy such a network and make the surface softer. In this respect, it would seem more difficult to conversely rigidify the surface by a single mutation, as in the case of ASTY97N, if that requires the creation of a new H-bond network. It is also noteworthy here, that the single BTN97Y mutation is predicted to yield values of ΔH‡, TΔS‡ and that are almost identical to those of native AST and kcat is predicted to increase 19-fold at 4°C for this mutation. That mutation of residue 97, either from BT to AST or vice versa, has a pronounced effect on the backbone mobility of the Nβ5-Nβ6 loop is also evident (Fig. 5), where Asn consistently reduces positional fluctuations whereas Tyr increases them.
In this work, we have addressed the problem of the structural origin of what appears to be a universal characteristic feature of cold-adapted enzymes, namely a reduced enthalpy and more negative entropy of activation. This was done using Atlantic salmon and bovine trypsin, cold- and warm-active, respectively, as models for the phenomenon. With very extensive all-atom computer simulations, using the EVB method to describe the catalytic reaction, reliable Arrhenius plots for the temperature dependence of the activation free energy could be obtained. It should be pointed out here that there is probably no other feasible way at present to calculate Arrhenius plot from first principles. It is rather remarkable that these simulations clearly reproduce the characteristic balance between activation enthalpy and entropy for cold-adapted versus warm-active enzymes, as well as the higher catalytic rate of AST compared to BT. Nevertheless, the activation free energies are similar at room temperature, thus reflecting a near perfect compensation between the former activation parameters.
The origin of the catalytically beneficial low activation enthalpy at low temperatures (accompanied by a more negative entropy) is found to be associated not with the active site but with the protein surface. From an evolutionary point of view this is perhaps not so surprising since any mutations in or near an optimized active site are likely to cause drastic rate reductions. What is rather surprising, however, is that it is the softness of the protein-water surface that appears to regulate the activation enthalpy-entropy balance. The simple picture that emerges is thus that the enzymes have a relatively rigid core, where the probability of successful adaptive mutations is low, surrounded by a softer outer matrix (Fig. 2b) whose properties can be fine-tuned by evolution.
While our earlier study of citrate synthases [10] also identified the same potential energy terms as responsible for the altered activation enthalpy-entropy balance, the structural origin of the effect remained obscure. Here, with the help of ultra-high resolution crystal structures, the actual structural “mechanisms” by which the surface softness is altered could finally be elucidated. Moreover, computational predictions of the effects of surface mutations were made that strongly support these conclusions. In particular, we identified extensive H-bond networks between polar surface groups and bound water molecules in the mesophilic enzyme that rigidify the surface, and several cold-adaptive mutations soften it by disrupting these networks. However, there are also examples of mutations in the cold-adapted enzyme (e.g., Val90Arg and Val154Lys) that appear to destabilize local hydrophobic surface patches. In view of the above findings, it is understandable that structural bioinformatics analysis has not yielded any consistent common descriptors of cold-adapted enzymes [5]. That is, since we identified several different types of mechanisms for surface destabilization and since the effects are often very local it is unlikely that there are distinct global descriptors that can capture them.
It is also noteworthy that the computer simulations predict that the enzyme ΔH‡ and ΔS‡ activation parameters can change significantly due to a single surface mutation. Such phenomena have, in fact, also been experimentally observed for other orthologous enzymes [28]. At first this may seem as a mysterious long-range effect on catalysis, but it should be emphasized that the activation free energies, and hence the catalytic rates, remain essentially unchanged. Instead it is the surface properties that are altered by such point mutations and mutations that soften the surface can apparently reduce the activation enthalpy of the catalyzed reaction at the expense of a more negative activation entropy. Such surface alterations are, however, beneficial for adaptation to low temperatures as they simultaneously make the rate more temperature insensitive and presumably also counteract the structural rigidity imposed by the reduction in temperature.
It is, of course, important to also try to address the generality of the present findings. In this respect, it should first be recalled that the characteristic trends with regard to activation enthalpy and entropy for cold-adapted enzymes appear to be completely universal, without known exceptions [5]. Two different types of enzymes (citrate synthases [10] and trypsins) have now been analyzed by extensive reaction simulations, which reproduce the observed behavior of warm- and cold-adapted orthologs, and which identify protein surface softness as the key variable. There is also other circumstantial evidence pointing towards surface properties, and flexibility in particular [29], [30]. Furthermore, the active site residues are basically always conserved between highly similar orthologous warm- and cold-adapted enzymes, which makes the idea that active site fluctuations would be substantially different very unlikely. It does therefore not seem far-fetched to assume that the difference in protein surface properties found here is likely to be a general feature of cold-adapted enzymes. It is further noteworthy that significantly altered kinetics and thermostability due to amino acid changes at a few sites distant from the active site have also been observed in dehydrogenases that are rate-limited by conformational changes rather than by chemistry [29], [30], as in the case of trypsin [19].
While the effects discussed herein pertain to the catalytic rates of the enzymes, their influence on thermostability is more difficult to assess. As mentioned, reduced thermostability is also an apparent universal characteristic of enzymes adapted to cold environments The net stability of folded proteins normally on the order of 10 kcal/mol, and is the result of large compensatory contributions. This, of course, makes it difficult to point out a single factor to explain differences in thermostability. However, previous studies of trypsin [31] indicate that the stability of a few loops and the C-terminal helix are important factors when explaining the difference in thermostability of cold- and warm-adapted trypsins. While our results also identify the same loop regions as important for adaptation to cold, experimental characterization is needed to examine whether these amino acid substitutions only change the catalytic rate or whether they affect thermostability as well.
|
10.1371/journal.pntd.0001308 | Melarsoprol Cyclodextrin Inclusion Complexes as Promising Oral Candidates for the Treatment of Human African Trypanosomiasis | Human African trypanosomiasis (HAT), or sleeping sickness, results from infection with the protozoan parasites Trypanosoma brucei (T.b.) gambiense or T.b.rhodesiense and is invariably fatal if untreated. There are 60 million people at risk from the disease throughout sub-Saharan Africa. The infection progresses from the haemolymphatic stage where parasites invade the blood, lymphatics and peripheral organs, to the late encephalitic stage where they enter the central nervous system (CNS) to cause serious neurological disease. The trivalent arsenical drug melarsoprol (Arsobal) is the only currently available treatment for CNS-stage T.b.rhodesiense infection. However, it must be administered intravenously due to the presence of propylene glycol solvent and is associated with numerous adverse reactions. A severe post-treatment reactive encephalopathy occurs in about 10% of treated patients, half of whom die. Thus melarsoprol kills 5% of all patients receiving it. Cyclodextrins have been used to improve the solubility and reduce the toxicity of a wide variety of drugs. We therefore investigated two melarsoprol cyclodextrin inclusion complexes; melarsoprol hydroxypropyl-β-cyclodextrin and melarsoprol randomly-methylated-β-cyclodextrin. We found that these compounds retain trypanocidal properties in vitro and cure CNS-stage murine infections when delivered orally, once per day for 7-days, at a dosage of 0.05 mmol/kg. No overt signs of toxicity were detected. Parasite load within the brain was rapidly reduced following treatment onset and magnetic resonance imaging showed restoration of normal blood-brain barrier integrity on completion of chemotherapy. These findings strongly suggest that complexed melarsoprol could be employed as an oral treatment for CNS-stage HAT, delivering considerable improvements over current parenteral chemotherapy.
| Human African trypanosomiasis (HAT) is caused by infection with either Trypanosoma brucei gambiense or T.b.rhodesiense and is fatal if untreated. In the late stages of the disease the parasites enter the brain, producing severe neurological symptoms. The arsenical drug melarsoprol is the only treatment available for rhodesiense disease once it has reached the brain. Melarsoprol is very poorly soluble in water which severely limits its oral bioavailability. Currently melarsoprol is solubilised in propylene glycol which restricts its administration to the intravenous route and treatment schedules are protracted and extremely painful. Further, this formulation of melarsoprol is toxic and kills 5% of all treated patients through the induction of a severe post-treatment reactive encephalopathy. In this study we show that combining melarsoprol with cyclodextrin molecules increases the oral bioavailability of the drug. In a murine model of late stage HAT oral treatment regimens employing melarsoprol cyclodextrin inclusion complexes rapidly cleared parasites from the brain, restored blood-brain barrier function and reduced the severity of the neuropathological response induced by the infection. If complexed melarsoprol is equally efficacious in patients with HAT this would not only have significant positive socio-economic impact but also constitute a major therapeutic advance in the field.
| Human African trypanosomiasis (HAT), also known as sleeping sickness, is endemic in 36 countries, in sub-Saharan Africa where 60 million people are at risk from infection [1], [2]. The disease is caused by the protozoan parasites Trypanosoma brucei (T.b.) gambiense in West Africa and T.b.rhodesiense in East Africa and is spread by the bite of the tsetse fly vector [1], [2]. Infection with T.b.gambiense usually results in a disease that follows a chronic course which can last for up to several years before death ensues while T.b.rhodesiense infection follows an acute pattern with death occurring in only weeks to months [3]. In both infections the disease progresses in two stages, the early or haemolymphatic stage and the late encephalitic or CNS-stage [3]. During the early-stage the parasites migrate from the site of the tsetse fly bite and spread throughout the body via the blood and lymph, invading the peripheral organs. The trypanosomes then cross the blood-brain barrier (BBB) and migrate into the CNS to cause the characteristic clinical manifestations of CNS-stage disease such as alteration of sleep patterns, neuropsychiatric symptoms and a variety of motor and sensory disturbances [4]. If the disease is diagnosed during the early stage it can be treated with pentamidine (for T.b.gambiense) or suramin (for T.b.rhodesiense) [5]. If the infection has reached the CNS, T.b.gambiense infections can be treated with either a concise 10-day regimen of melarsoprol [6], [7] or the recently developed nifurtimox-eflornithine combination therapy (NECT) [8]. In the case of CNS-stage T.b.rhodesiense infections the only treatment option currently available is a lengthy melarsoprol schedule comprising 3–4 cycles of a series of 3–4 injections, of increasing melarsoprol concentration, separated by a 7–10 day interval between each cycle [6], [9].
Melarsoprol (Figure 1A) is a highly lipophilic molecule that is poorly soluble in water with a log POW of 2.53 and a solubility of only 6 mg/L at 25°C [10]. Despite these properties the drug is a potent trypanocide and has been used for the treatment of HAT since its introduction in 1949 [11]. The limited solubility of melarsoprol precludes its oral delivery as only a small fraction of the drug is absorbed through the gastrointestinal tract. Currently melarsoprol is produced as a 3.6% solution in propylene glycol (Arsobal ®) which restricts its administration to the intravenous route. The treatment schedules employed are protracted, excruciatingly painful and require continuous hospitalization. In addition, treatment with Arsobal® is associated with numerous adverse effects including severe tissue necrosis at the injection site, neuropathy, and gastrointestinal upset [4]. However, the most serious adverse reaction is the development of a post-treatment reactive encephalopathy (PTRE) which occurs in 10% of all treated patients, 50% of whom die as a result. Arsobal® treatment is therefore responsible for the death of 5% of all patients given the drug [1], [12]. Although the pathogenesis of the PTRE remains unclear, several hypotheses have been postulated to explain its occurrence. These include direct arsenical toxicity [13], [14], autoimmune reactions [15] and pro-inflammatory immune response directed against trypanosomes remaining within the CNS following systemic clearance of the parasites [16], [17] or parasite antigen released as a consequence of chemotherapy [18]. The severity of the complications associated with Arsobal® chemotherapy make accurate staging of the disease via cerebrospinal fluid analysis absolutely essential both to manage proven CNS-stage infections appropriately and to prevent unnecessary administration of this highly toxic drug to early stage patients [19].
Cyclodextrins are naturally occurring cyclic oligosaccharide molecules composed of six or more glycopyranose units linked by α-1, 4 gycosidic bonds. They take the form of a truncated cone or torus with a hydrophilic exterior and a hydrophobic interior cavity which can be occupied by various guest molecules [20]. Cyclodextrins have been widely utilized by the pharmaceutical industry to alter the physiochemical properties of a variety of drugs through enhancing their solubility and oral bioavailability and decreasing their toxicity [10], [21]. Complexation of melarsoprol with either hydroxypropyl-β-cyclodextrin (mel/HPβCD) or randomly methylated-β-cyclodextrin (mel/RAMβCD) (Figure 1A) has been shown to increase the inherent solubility of the drug by a factor of 7.2×103, which raises the possibility that the melarsoprol cyclodextrin complexes could be efficacious when delivered via the oral route for the treatment of trypanosomiasis [10].
In the current study the efficacy of the melarsoprol cyclodextrin inclusion complexes was investigated using both in vitro and in vivo methodologies and compared with that of contemporary melarsoprol formulations. The effect of oral drug treatment on the BBB was examined using MRI, and both the CNS parasite load and the CNS neuroinflammatory response monitored throughout the treatment regimen. We show here that melarsoprol cyclodextrin complexes are orally effective and non-toxic in curing CNS-stage trypanosome infections in mice.
Trypanotoxicity was determined using an adapted version of the Alamar Blue assay [22]. Bloodstream form T. brucei brucei (strain 427) were cultivated in HMI-9 medium (BioSera Ltd., UK) [23] supplemented with 2 mM β-mercaptoethanol (Sigma-Aldrich, UK) and 10% fetal calf serum (BioSera Ltd., UK) at 37°C in a humidified 5% CO2 environment. Parasites (100 µl of 1×104 trypanosomes/ml) were added to wells of 96-well plates containing doubling dilutions of the drugs (100 µl) ranging in final concentration from 100 µM to 24 pM and incubated for 48 hours. Alamar Blue reagent (20 µl, 0.49 mM in PBS, pH 7.4; Sigma-Aldrich, UK) was added to each well and, after 24 hours, the fluorescence was measured using a LS 55 luminescence spectrophotometer (PerkinElmer Life and Analytical Sciences, USA) set at excitation and emission wavelengths of 530 nm and 590 nm respectively. Data was analysed and inhibitory concentration (IC50) values determined with Prism 5.0 (GraphPad Software, USA) software. The experiment was performed in duplicate on three independent occasions.
A well established and characterised model of CNS-stage human African trypanosomiasis was employed throughout this investigation. Briefly, female CD-1 mice (Charles River Laboratories) (20–30 g body weight) were infected with 3×104 Trypanosoma brucei brucei GVR35 parasites by intraperitoneal injection. The infection was allowed to progress until day 21 without drug intervention. At this point the parasites have established within the CNS and the infection has entered the encephalitic stage.
Infection was confirmed in all mice prior to drug treatment by examination of a wet blood film for the presence of parasites. To determine whether a treatment regimen was curative, blood smears were examined for the presence of parasites on a weekly basis for a period of 60 days. If the animals relapsed to parasitaemia the regimen was considered unsuccessful and the mice were killed. Mice that remained aparasitaemic for the duration of the monitoring period were killed, the brains excised and lightly homogenised in PBS supplemented with 1.5% glucose w/v and injected intraperitoneally into a clean recipient animal. This mouse was then monitored for the presence of parasites for a further 60 days. If the mouse remained aparasitaemic the treatment regimen was considered successful.
All animal procedures were authorised under the Animals (Scientific Procedures) Act 1986 and approved by the University of Glasgow Ethical review Committee.
Trypanosome load within the brain was determined by real-time quantitative PCR. Briefly, whole brains were homogenised and digested with proteinase K (AppliChem GmbH) and DNA extracted from a 25 mg sample of the homogenate (Qiagen, DNeasy Tissue kit). The concentration of the extracted DNA was assessed by measuring the absorbance and the sample diluted to 20 ng/ml. The reaction mix was comprised of; Taqman Brilliant II master mix (Agilent), 0.05 pmol/µL of each primer, 0.1 pmol/µL probe (Eurofins MWG Operon) and 100 ng DNA template. The amplification was performed on a MxPro 3005 thermocycler (Agilent).
The mel/HPβCD and mel/RAMβCD inclusion complexes were prepared as previously described [10]. Each complex was dissolved in sterile water and administered at dose rates of 0.0125, 0.025, 0.05, 0.1 and 0.2 mmol/kg (equivalent to 4.975, 9.95, 19.9, 39.8, and 79.6 mg/kg) of the active ingredient, melarsoprol. Non-complexed HPβCD and RAMβCD (Sigma) were used as control treatments and administered at dose rates equivalent to 0.1 mmol/kg of the complexed agent.
Contemporary melarsoprol and the melaminophenyl arsine derivatives [24], melarsamine hydrochloride (MelCy) (Cymelarsan®) and melarsonyl potassium (MelW) (Trimelarsen®) were prepared as solutions or fine suspensions in sterile water and administered at a dose of 0.05 mmol/kg.
All drug treatments were delivered orally by gavage, once per day for a period of 7 days beginning on day 21 post-infection.
Body weights were measured in groups of uninfected mice before and after completion of treatment and clinical appearance was monitored using an established visual assessment scale [25] throughout the drug regimens to assess overt signs of drug toxicity.
MRI was performed on two mice at day 21 post-infection prior to drug treatment. The mice were re-scanned at 24 hours, 8 and 15 days after completion of chemotherapy. Uninfected mice (n = 3) were also examined. All scans were performed as described previously [26]. Briefly, mice were anaesthetised and the tail vein was cannulated with a 26 gauge×19 mm cannula to facilitate contrast agent administration during MRI scanning. The animal was placed into a mouse cradle and restrained using ear and tooth bars to minimise head movement. Anaesthesia was maintained throughout the procedures and respiration, heart rate and body temperature were observed. The animal was maintained normothermic by an enclosed warm water circuit.
MRI was performed on a Bruker Biospec 7 T/30 cm system equipped with an inserted gradient coil (121 mm ID, 400 mT/m) and a 72 mm birdcage resonator. A surface coil was used for brain imaging. The scanning protocol consisted of a RARE T2 weighted scan [effective TE (echo time) 76 ms, TR (repetition time) 5362 ms, 25 averages, matrix 176×176, FOV (field of view) 17.6×17.6 mm, 20 contiguous coronal slices of 0.4 mm thickness] followed by a RARE T1 weighted scan (effective TE 9 ms, TR 8000 ms, 20 averages, matrix 176×176, FOV 17.6×17.6 mm, 20 contiguous coronal slices of 0.4 mm thickness). Following the RARE T1 weighted scan 0.1 ml of a solution containing 50 µL gadolinium-diethylenetriamine penta-acetic acid (Gd-DPTA Magnevist®; Bayer) and 50 µL of sterile water was injected via the tail vein cannula. Five minutes later the T1 weighted scan was repeated. Gd-DTPA cannot readily cross the intact blood brain barrier due to its charge and high molecular weight [27]. Extravasation of Gd-DTPA observed within the parenchyma demonstrates an impairment of the BBB integrity.
Images were analysed using Image J software (http://rsbweb.nih.gov/ij/). Contrast enhancement maps were generated from the the per and post-contrast T1 weighted scans according to the equation: Enh = (Spost−Spre)÷Spre where Spost = post contrast agent signal and Spre = pre-contrast agent signal. Regions of interest (ROIs) were manually defined to include the entire brain and meninges. The mean percentage signal change for each brain slice was then calculated and signal enhancement maps generated.
Following sacrifice the brains were excised and fixed in 4% neutral buffered formalin, paraffin wax blocks prepared and sections of 3 µm thickness cut and stained with haematoxylin and eosin. These sections were examined by two independent assessors and the severity of the neuropathological reaction graded on a scale of 0–4 where 0 represented normal pathology with no indications of inflammation and a grade of 4 was characterised by the presence of a severe meningoencephalitis with the presence of inflammatory cells in the brain parenchyma [26], [28] (Table S5).
Immunocytochemistry was performed to detected T-cells (rabbit anti-CD3), B-cells (rat anti-B220) and macrophages (rat anti-F4/80) following a standard peroxidise anti-peroxidase protocol using the Dako® EnVision system and DAB visualisation.
Data were analyzed using analysis of variance methods and the General Linear Model (GLM) procedure in Minitab Version 16 followed by multiple pair wise comparison tests. This identified significant main effect differences between groups of uninfected animals, infected animals and treated animals. In studies with measurements over time the GLM procedure provided tests for treatment and time effects and their interaction. Proportions of mice relapsing in different treatment groups were compared using a Tukey-type multiple comparison test for proportions [29]. Changes in body weight were investigated using a paired t-test. P values of less than 5% were considered to be statistically significant. Where appropriate data were log transformed prior to analysis. Group means were plotted showing means and their standard errors, and the size of treatment effects were estimated using differences between group means and their 95% confidence intervals. Log dose response curves provided estimates of IC50 concentrations.
To determine whether the complexed melarsoprol retains its trypanocidal properties a modification of the Alamar blue assay [22] was used to investigate the inhibitory concentration (IC50) of the complexed melarsoprol molecules in comparison to standard melarsoprol and an unrelated trypanocidal drug, diminazene aceturate, in an in vitro trypanosome culture system. The IC50 values determined for mel/HPβCD and mel/RAMβCD were 21.6 nM and 8.8 nM respectively (Figure 1B). Standard melarsoprol returned an IC50 value of 6.9 nM. Statistical analyses of the Alamar blue data revealed no significant changes in the trypanocidal activity of melarsoprol following complexation when compared to the standard form of the drug (P = 0.2002, P = 0.9999; mel/HPβCD and mel/RAMβCD respectively). The HPβCD and RAMβCD molecules alone did not display any trypanocidal activity (Table S1, Figure 1B).
The ability of the complexed melarsoprol compounds to cure CNS-stage trypanosome infections was investigated in a well established and characterized murine model of the disease. The drugs were administered by oral gavage each day at doses ranging from 0.0125 mmol/kg to 0.2 mmol/kg for a seven day period. All animals became aparasitaemic following the melarsoprol regimens; however, all mice treated at the 0.0125 mmol/kg level relapsed to parasitaemia. A relapse to parasitaemia was also detected in one third of the mice treated with mel/HPβCD and one sixth of the mice given mel/RAMβCD at the 0.025 mmol/kg level. Successful cures were obtained in all mice treated with the 0.05 mmol/kg, 0.1 mmol/kg or 0.2 mmol/kg dosage of either complex, indicating that 0.05 mmol/kg was the minimum dosage necessary to achieve successful cures. Animals given HPβCD or RAMβCD alone remained parasitaemic throughout the procedure (Figure 1C).
Paired t-test analysis detected no evidence of decreased body weight in uninfected mice following 7-days of oral drug treatment. A significant (P = 0.019, 95% confidence interval 0.213 g, 1.954 g) increase was detected between the mean body weight of the pre- and post treatment groups (25.83±0.696 g; 26.92±0.890 g respectively). No adverse clinical signs were detected at any point during the chemotherapy regimen with the mice remaining alert and showing good coat condition.
The efficacy of melarsoprol (MelB) and the water soluble melaminophenyl arsine derivatives [24], melarsamine hydrochloride (MelCy) and melarsonyl potassium (MelW) (Figure 1A) when administered per os at a dose of 0.05 mmol/kg for seven consecutive days, during CNS-stage infections was investigated. No cures were obtained in the mice treated with MelCy or MelW and only 33% of the mice given MelB were successfully cured (Figure 1D).
Taqman real-time PCR was performed (Figure S 1) to determine the parasite numbers present within the brain tissue prior to chemotherapy and at 24 hours after each oral dose of mel/HPβCD or mel/RAMβCD (Figure 2A). Animals killed on day 21 post-infection, prior to receiving any drug treatment showed a mean parasite load of 626±82.8 (mean ± SE). Following a single dose of mel/HPβCD or mel/RAMβCD the parasite numbers detected within the brain were significantly (P<0.001) reduced (68.1±14.7; 66.2±10.8 respectively). The decrease in parasite numbers continued in a stepwise manner with successive treatments until the trypanosomes were completely cleared from the brain (Figure 2B & C, Table S2 & S3). Interaction plots comparing the mean copy numbers detected after each dose of mel/HPβCD and mel/RAMβCD show that there are no significant differences between the clearance rates achieved by either of the drugs (Figure 2D). From the Taqman results it is apparent that both forms of complexed melarsoprol clear the trypanosomes from the brain in a rapid and efficient manner with a reduction of greater than 80% of the parasite load 24 hours after the initial drug treatment.
We determined the effect of oral treatment with mel/HPβCD on BBB function using MRI. Mice were examined prior to treatment, and 24 hours, 8 and 15 days following the chemotherapy regimen (Figure 3A). MRI scans were performed before and after the injection of Magnevist® contrast agent (Gd-DPTA) [27] and signal enhancement maps generated as previously described [26]. Changes in BBB integrity were investigated in two infected mice scanned at day 21 post-infection and compared with scans prepared in the same animals 24 hours, 8 days and 15 days after completing a 7 day oral course of mel/HPβCD as well as those from uninfected mice (n = 3). At day 21 post-infection the BBB was significantly compromised (17.87±1.62) (Figure 3B, Figure 4, Table S4). Signal enhancement was present throughout the brain with highest signal change found in the ventricular region. Changes in signal intensity were also apparent in the cerebral cortex, hypothalamus, hippocampus and median eminence (Figure 4). However, by 24 hours after completion of the chemotherapy (Figure 4) the percentage signal change (7.93±0.455) had dropped significantly (P<0.0001) and was comparable (P = 0.9296) to that seen in uninfected mice (7.11±0.162) (Figure 3B, Figure 4) indicating that by this point the integrity of the BBB had become re-established. The integrity of the barrier was maintained in all subsequent scans performed at 8 days (9.25±0.596) (Figure 3B, Figure 4) and 15 days (6.55±0.463) (Figure 3B, Figure 4) after completion of the treatment schedule (Table S4).
The severity of the neuropathological response to the trypanosome infection and drug treatment was determined in mice killed 15 days after completing the treatment schedule and compared to animals killed at day 21 post-infection prior to receiving chemotherapy using a well established grading scale [28] (Table S5). Pathological examination of the brains prepared from animals prior to drug treatment showed mild neuroinflammatory changes (1.5±0.158) with the presence of an inflammatory cell infiltrate in the meninges and Virchow–Robin spaces (Figure 5). Some perivascular cuffs were also apparent surrounding the blood vessels in the hippocampus (Figure 5). The cellular infiltrate was composed mainly of lymphocytes, and macrophages (Figure 6). A significant (P = 0.0366) resolution of this neuroinflammation (1.083±0.083) was apparent in mice killed 15 days after completion of the oral mel/HPβCD regimen. This represents a mean decrease of 27.8% with a 95% confidence interval (0.032, 0.801). Only a few inflammatory cells could be detected in the meninges of these animals accompanied by very mild perivascular infiltration of the occasional blood vessel in the hippocampus (Figure 5 & 6).
New drugs to treat HAT remain an urgent priority. In spite of some recent positive developments [30] the situation remains precarious as evidenced by the failure, late in clinical trials, of pafuramidine (DB289). Ideally new drugs should be orally available and of equal or better efficacy than current drugs with improved safety. Melarsoprol is the only drug suitable to treat CNS-stage rhodesiense disease and remains in use in some areas for gambiense. Its use, however, is tainted by its being administered parenterally and through its well documented adverse events. The study reported here shows that complexation of melarsoprol with the cyclodextrin molecules does not affect the trypanocidal properties of the compound and appreciably enhances the ability of the drug to cure CNS-stage trypanosome infections when given as an oral chemotherapy regimen. The improved oral bioavailability seen in these investigations is most likely due to the increased solubility and reduced degradation kinetics of the drug following complexation with the cyclodextrin molecules [10], [31]. Further, cyclodextrins can also act as carriers, delivering the drug directly to the intestinal membrane while protected within the cavity of the molecule [32], [33]. Consistent with our findings is the improved oral bioavailability reported with a variety of cyclodextrin drug inclusion complexes including anti-parasitic agents. The potent anti-malarial drug artemisinin has low aqueous solubility that severely limits its absorption following oral administration. Complexation of artemisinin with cyclodextrin molecules has been shown to improve the pharmacokinetic profile of the drug compared with Artemisinin 250® when given per os [34]. This has also been demonstrated for the anti-helminthic drug albendazole [35].
The pathogenesis of the PTRE associated with standard melarsoprol treatment is currently unknown although several hypotheses have been suggested [13]–[18]. However, it is probably caused, at least in part, by an acute toxic reaction to low levels of arsenic within the CNS following delivery of an intravenous bolus of the arsenical drug [13], [14]. On the basis of our combined data the lack of toxicity and the resolution of the CNS inflammatory reaction shown following oral treatment with melarsoprol cyclodextrin inclusion complexes can most easily be explained by the more controlled delivery of the trypanocidal drug to the brain following a sustained absorption from the gut compared with that of an intravenous bolus. Consistent with this explanation are the extremely low levels of arsenic in the brain tissue following chemotherapy which were below the level of detection (<5 ng/mL) of the GC-MS assay employed (unpublished data) and the extremely rapid clearance of the parasites from the CNS following drug administration. This is also reflected by the restoration of BBB integrity detected in the mice 24 hours after completion of the chemotherapy regimen. However, a direct comparison of these criteria following a curative IV regimen of Arsobal® would be required to corroborate this hypothesis. Multiple IV doses of Arsobal® cannot be administered in the murine model due to the severe venous damage caused by the propylene glycol solvent present in the drug preparation. Therefore, data regarding drug levels, parasite clearance and BBB function following IV Arsobal® remain unavailable.
Taken together these findings strongly suggest that mel/HPβCD and mel/RAMβCD could be used to treat patients with CNS-stage HAT. Since these experiments were performed using a T.b.brucei model of infection it is possible that these drug complexes will not show the same activity profile when transferred to human disease. However, since the active trypanocidal component of the complex is melarsoprol, with proven effectiveness against both T.b.rhodesiense and T.b.gambiense infections, this scenario seems highly unlikely. Consequently, in the first instance, these complexes should be tested in subjects with T.b.rhodesiense, even though these comprise the minority of cases of HAT compared with T.b.gambiense, since melarsoprol is currently the only drug that can be effective in rhodesiense disease. The drugs are effective orally at dosages that could be delivered in humans. During the concise 10-day schedule for Arsobal® treatment a 60 kg patient would be given a total dose of 1320 mg of melarsoprol. In the current study, melarsoprol cyclodextrin complexes were curative when administered at 0.05 mmol/kg or 19.9 mg/kg melarsoprol daily for a seven day period. To obtain an approximate human equivalent dose (HED) from this data the dosage must be normalized according to body surface area which can be achieved by dividing the murine dose by a factor of 12 [36]. The HED for the complexed drugs would therefore be approximately 1.6 mg/kg, with a total dosage of 672 mg assuming a seven day course and a 60 kg body weight. This is a considerable reduction in the total amount of arsenical required for each drug course, even when compared with the concise schedule. This decreased arsenical dosage could also be a major factor in the apparent lack of toxicity associated with the oral regimen.
These complexes rapidly clear the trypanosomes from the brain following administration, reduce the severity of the neuropathological response induced by the infection, and also restore BBB integrity following treatment. The availability of an orally administrable drug would preclude both the need for hospitalization of the patient throughout the period of treatment and the provision of highly skilled clinicians to administer the drug by slow intravenous infusion. Further, the pain and fear associated with current melarsoprol therapy would be circumvented and patients would be far more likely to be compliant in finishing the treatment course. This would have a significant positive socio-economic impact in local communities and on the already burdened health care budgets of these regions.
One of the major problems in the management of HAT is that there is no clear consensus on the criteria used to classify an infection as having progressed to the CNS-stage [19], [37]. The current WHO criteria suggest that CSF containing >5 white blood cells (WBC)/µL with or without the presence of trypanosomes indicates CNS-stage infection [9]. However in some T.b.gambiense infections the higher value of >20 WBC/µL has been used before commencing melarsoprol treatment [38], [39]. This has significant implications for choosing the correct chemotherapeutic approach to best manage the infection. Inappropriate administration of melarsoprol to patients with early-stage disease exposes them to unnecessary risks form drug toxicity while failure to use melarsoprol in CNS-stage disease will have inevitably fatal consequences [19]. The use of an alternative treatment strategy without the associated adverse safety profile of the intravenous melarsoprol formulation would also obviate significantly the difficulties associated with the current methods of disease staging [19].
In conclusion, the current chemotherapy options for treatment of CNS-stage HAT are extremely limited and all involve parenteral administration of highly toxic and sometimes ineffective drugs. Moreover there are no new alternative drugs for CNS HAT likely to be used in clinical practice for at least 5–10 years [30]. Only one compound, fexinidazole, is currently in Phase I clinical trials [40]. Due to the high failure rate of novel compounds it is critical to maintain drug development in this area to ensure that effective treatments for both forms of this disease are available in the future. Sir James Black, the Nobel Laureate said ‘the most fruitful basis for the discovery of a new drug is to start with an old drug’ [30]. If melarsoprol cyclodextrin inclusion complexes, given via the oral route, prove equally efficacious in patients with HAT this would constitute one of the most significant therapeutic advances in the long history of the disease. Plans to test these drug complexes in a phase II trial in East African patients with T.b.rhodesiense are currently being formulated.
|
10.1371/journal.pgen.1004822 | The Evolution of Sex Ratio Distorter Suppression Affects a 25 cM Genomic Region in the Butterfly Hypolimnas bolina | Symbionts that distort their host's sex ratio by favouring the production and survival of females are common in arthropods. Their presence produces intense Fisherian selection to return the sex ratio to parity, typified by the rapid spread of host ‘suppressor’ loci that restore male survival/development. In this study, we investigated the genomic impact of a selective event of this kind in the butterfly Hypolimnas bolina. Through linkage mapping, we first identified a genomic region that was necessary for males to survive Wolbachia-induced male-killing. We then investigated the genomic impact of the rapid spread of suppression, which converted the Samoan population of this butterfly from a 100∶1 female-biased sex ratio in 2001 to a 1∶1 sex ratio by 2006. Models of this process revealed the potential for a chromosome-wide effect. To measure the impact of this episode of selection directly, the pattern of genetic variation before and after the spread of suppression was compared. Changes in allele frequencies were observed over a 25 cM region surrounding the suppressor locus, with a reduction in overall diversity observed at loci that co-segregate with the suppressor. These changes exceeded those expected from drift and occurred alongside the generation of linkage disequilibrium. The presence of novel allelic variants in 2006 suggests that the suppressor was likely to have been introduced via immigration rather than through de novo mutation. In addition, further sampling in 2010 indicated that many of the introduced variants were lost or had declined in frequency since 2006. We hypothesize that this loss may have resulted from a period of purifying selection, removing deleterious material that introgressed during the initial sweep. Our observations of the impact of suppression of sex ratio distorting activity reveal a very wide genomic imprint, reflecting its status as one of the strongest selective forces in nature.
| The sex ratio of the offspring produced by an individual can be an evolutionary battleground. In many arthropod species, maternally inherited microbes selectively kill male hosts, and the host may in turn evolve strategies to restore the production or survival of males. When males are rare, the intensity of selection on the host may be extreme. We recently observed one such episode, in which the population sex ratio of the butterfly Hypolimnas bolina shifted from 100 females per male to near parity, through the evolution of a suppressor gene. In our current study, we investigate the hypothesis that the strength of selection in this case was so strong that the genomic impact would go well beyond the suppressor gene itself. After mapping the location of the suppressor within the genome of H. bolina, we examined changes in genetic variation at sites on the same chromosome as the suppressor. We show that a broad region of the genome was affected by the spread of the suppressor. Our data also suggest that the selection may have been sufficiently strong to introduce deleterious material into the population, which was later purged by selection.
| In 1930, Fisher noted that the strength of selection on the sex ratio was frequency dependent, echoing earlier findings of Düsing [1], [2]. As a well-mixed outbreeding population progressively deviates from a 1∶1 sex ratio, selection on individuals to restore the sex ratio to parity becomes stronger. In natural animal populations, a common cause of population sex ratio skew is the presence of sex ratio distorting elements, in the form of either sex chromosome meiotic drive [3], or cytoplasmic symbionts [4]. In some cases, these elements can reach very high prevalence, distorting population sex ratios to as much as 100 females per male [5], and producing intense selection for restoration of the individual sex ratio to 1 female per male. The most common consequence of this selection pressure is the evolution of systems of suppression – host genetic variants that prevent the sex ratio distorting activity from occurring. Suppressor factors are known for a wide range of cytoplasmic symbionts and meiotic drive elements [3], [6], [7].
The evolution of suppression of Wolbachia induced male-killing activity in the butterfly Hypolimnas bolina represents a compelling observation of intense natural selection in the wild. Female H. bolina can carry a maternally inherited Wolbachia symbiont, wBol1, which kills male hosts as embryos [8]. The species also carries an uncharacterised dominant, zygotically acting suppression system that allows males to survive infection [6]. Written records and analysis of museum specimens indicate this symbiont was historically present, and active as a male-killer, across much of the species range, from Hong Kong and Borneo through to Fiji, Samoa and parts of French Polynesia [9]. Evidence from museum specimens also indicates that host suppression of male-killing had a very restricted incidence in the late 19th century, with infected male hosts (the hallmark of suppression) being found in the Philippines but not in other localities tested. By the late 20th century, suppression of male-killing was found throughout SE Asia, but not in Polynesian populations where the male-killing phenotype remained active [10]. The most extreme population was that of Samoa, where 99% of female H. bolina were infected with male-killing Wolbachia, resulting in a sex ratio of around 100 females per male within the population [5]. However, following over 100 years of stasis on Samoa, the rapid spread of suppression of male-killing activity of the bacterium was finally observed between 2001 and 2006, restoring both individual and population sex ratio to parity [11].
When strong selection occurs at a locus, it is expected to leave a genomic imprint beyond the target of selection, as a result of genetic hitch-hiking. A neutral (or even deleterious) variant that is initially present in the haplotype in which the favoured allele arose (i.e. is linked to the site of selection), will also increase in frequency [12]. When selection is very strong, the frequency of linked variants may increase across a broad genomic region [13]. Importantly, the extent of the chromosome over which this effect will occur depends on the selection pressure in the first few generations; before recombination has broken down associations between the target of selection and linked variants. Where sex ratio distorters are common, the selection pressure in these first generations may be very strong indeed (before the sex ratio becomes less biased through spread of the suppressor). It is thus likely that selection on the sex ratio will influence linked material over a broader genomic region compared to many other selective regimes. That is, the episode of selection is likely to have a very wide genomic impact.
In this paper, we first mapped a genomic region in SE Asian butterflies that was required for male survival in the presence of Wolbachia. We then investigated the impact of the recent spread of the suppressor in Samoa on the pattern of variation around this region. To this end, we initially developed theory to predict the impact of suppressor spread on linked genetic variation. We then directly observed changes in the frequency of genetic variants surrounding the suppressor locus by comparing the pattern of genetic variation in H. bolina specimens collected in Samoa before (2001) and after the selective sweep (2006 and 2010). By examining post-sweep samples at two time points we were additionally able to track allele frequency changes following the initial sweep. The data revealed changes in the pattern of genetic variation over a 25 cM region surrounding the suppressor locus. We further suggest that the suppressor was probably derived by immigration, and that the sweep may have introduced deleterious material that was subsequently subject to purifying selection.
Hypolimnas bolina has 31 chromosomes and a total genome size of 435 MB [14], [15]. Previous work established that the rescue of male zygotes from Wolbachia induced killing was dominant, and potentially a single locus trait [6]. Genetic markers spanning the genome were developed using a targeted gene approach informed by conservation of synteny in Lepidoptera, with the sequence of H. bolina orthologs obtained through Roche 454 transcriptome sequencing (see Methods and Materials, NCBI SRA accession: SRP045735). These markers were then tested for co-segregation with suppression in order to identify the linkage groups associated with male host survival. Female butterflies from South East (SE) Asia that carried both Wolbachia and the suppressor allele, were crossed with males from the French Polynesian island Moorea (where suppression is absent). The resulting F1 daughters (who inherited Wolbachia from their SE Asian mother) were then backcrossed to Moorea males to create a female-informative family for identification of loci linked to the suppressor. The absence of recombination in female Lepidoptera means that a SE Asia allele on any chromosome that is necessary for male survival will be present in all of the surviving sons of this female (as if they lack it, they die), but this allele will show normal 1∶1 segregation in her daughters (S1 Figure). Initially 10 loci from across the genome were screened. Of these, one locus orthologous to sequence on chromosome 25 in the moth Bombyx mori showed this unusual pattern of inheritance. For this locus, all 16 sons carried the same maternal allele of SE Asia origin while 8 daughters showed Mendelian segregation (probability of observing this pattern of segregation in sons on the null hypothesis of no association = (1/2)16: p<0.0001).
We then obtained an additional 11 markers in this linkage group. Candidates were identified initially via synteny to B. mori, and then confirmed as showing co-segregation with the original marker and as being associated with male survival, in the female-informative family. In this way, a suite of 12 suppressor-linked markers (A-L) were developed, all of which followed the presumed pattern of inheritance of the suppressor - that of presence in all 16 sons and half of the daughters. The remaining 9 non-suppressor-linked markers (M-U), representing 8 separate linkage groups, were developed to investigate potential genome-wide effects. Marker information and accession numbers are given in S1 Table and S2 Table.
A linkage map for this chromosome, the suppressor linkage group (SLG), was then constructed. The region required for male survival was identified by the exclusion of recombinants. This was achieved by examining the segregation of alleles from sons of the SE Asia x Moorea cross above that were mated to Wolbachia-infected Moorea (non-suppressor) females (creating a male-informative family). 307 recombinant daughters were obtained, which were used to create a linkage map of the 12 suppressor-linked markers (data used to create linkage map in S6 Table). The markers were estimated to cover a 41 cM recombination distance and were syntenic with B. mori (Fig. 1). The suppressor locus was localized to a region within this chromosome by excluding linked loci where the SE Asia derived paternal allele was absent in one or more sons (indicating that the genomic region containing the SE Asia allele was not necessary for male survival). Three suppressor-linked alleles (D, E and F), all in the +11 to +12 region, were retained in all 60 sons, whereas the 9 markers proximal and distal to these were excluded by the presence of one or more recombinants (Fig. 1). The probability of observing retention of a marker in a sample of 60 on the null hypothesis of no association between the +11/+12 genomic region and male survival is 0.560 = 9×10−19. Thus we posit that the suppressor lies between marker C at +8 (excluded by one recombinant) and marker G at +17 (excluded by two recombinants) - a region of approximately 10 cM. Our data also indicate that while this genomic region is necessary for male survival, presence of this locus was not always associated with male survival, with the number of surviving sons obtained being one quarter, rather than one half, of the number of daughters obtained in our cross (60 sons vs 307 daughters).
Our data identified a 10 cM genomic region on chromosome 25 of SE Asian H. bolina that was necessary for a male butterfly to survive Wolbachia induced male-killing. This region was also a focus of selection during the spread of suppression of male-killing between 2001 and 2006 in Samoa. During this episode, patterns of allelic variation were observed to be altered over a 25 cM region of chromosome 25, with increases in frequency of one allele at each locus creating the vast majority of heterogeneity between time points. The largest magnitude of change occurred in markers that co-segregated with suppression in SE Asia, and in this region the overall genetic diversity (as measured by AE and π) declined - the classical signature of a selective sweep. Three further features implicate the role of selection in altering allele frequency across this 25 cM region. First, the changes in allele frequency are too large to be accounted for by drift, even under conservative assumptions for population size and generation time. Second, LD is generated across this region, as predicted under a model of strong selection. Third, 9 markers unlinked to the suppressor linkage group showed no evidence of changes in the frequency of allele variants between 2001 and 2006, implying that demographic factors were not the major force driving changes in allele frequency.
While we observed changes consistent with the operation of selection over a very broad genomic area, the degree of change was less than that predicted from our model. This is true both of the magnitude of allele frequency change at loci located near the suppressor locus, and the breadth of the region of chromosome over which changes in allele frequency occurred. Our model, which presumes a panmictic model and no cost to carrying the suppressor, predicts the suppressor should fix (and take alleles within 5 cM distance to frequency in excess of 87%), and that allele frequency changes should be observed chromosome-wide. In contrast, the swept allele at locus D (which lies within 5 cM of the target of selection) attains a frequency of just 0.67 (n = 172, CI 0.59–0.74) in 2006 and 2010 samples. Further, we observed only very small changes in allele frequency at the most distant locus from the region containing the suppressor, locus L.
We suggest there are three non-mutually exclusive explanations for this lack of fit with the model. First, the suppressor mutation in natural populations diffuses spatially following its initial arrival, and each generation of spatial diffusion is associated with a narrower local sweep. The principle impact of spatial diffusion will be to narrow the genomic region that is affected by selection compared to that predicted in a panmictic model, and to reduce the magnitude of change at loci far from the target of selection. For a locus 25 cM distant from the suppressor, association with the suppressor allele may last just one or two generations, such that changes in allele frequency occur only near the point of origin, and are diluted by absence of any selection on these loci in the majority of the species range. However, spatial diffusion represents a poor explanation for the lower than expected frequency of variants at tightly linked loci post-sweep, which are expected to maintain strong association during the spread of the suppressor across the island, as this occurs in about 10 generations. A second possibility is the involvement of other loci in the genome, as enhancers of suppressor action. Our data indicate that the genomic region in question is necessary for male survival, but do not rule out involvement of other loci in enhancing suppression. If other loci are involved, either as required elements or enhancers of male survival, this would slow suppressor spread, and might account for the narrowness of the sweep observed compared to model predictions. However, the requirement of the genomic region for male survival in the presence of Wolbachia again makes this a poor explanation for the failure of tightly linked loci to reach high frequency. Because it is necessary, it should become fixed in the population, and closely associated allele variants should in consequence attain very high frequency. A third possibility is that there is a cost to being homozygous for the suppressor mutation, either in both sexes, or in female hosts only. A cost such as this could prevent fixation of the suppressor allele, and thus also help account for the decreased magnitude of effect at loci tightly linked to the suppressor. If the suppressor allele reaches >0.75 frequency, then males lacking the suppressor would be sufficiently rare that the population sex ratio would be near parity. Biologically, costs to suppressor carriage may be directly associated with the suppression system itself. Modification of a sex determination gene, for instance, might rescue males but be deleterious in females, or when homozygous. Alternatively, costs may be associated with linked mutations. The presence of deleterious loci in linkage with the suppressor is supported by our observation that some material that had been initially swept into the population was lost between 2006 and 2010. Finer-scale investigation of this linkage group, especially within the region identified as required for male survival, is necessary to illuminate the precise dynamics that occurred during this episode of selection.
In our data, we observed concordance between the position of the suppressor ascertained in SE Asian butterflies, and the genomic region subject to selection during spread of suppression through the Samoan population of the butterfly. This observation has two possible interpretations. First, the suppressor mutation may have been introduced into Samoa by migration. Given that the suppressor is absent in the nearest island groups, American Samoa and Fiji, suppressor introduction would be associated with a long distant migrant. Second, the genomic region identified here may represent a hotspot for suppressor mutation, derived independently in Samoa by de novo mutation. This may be an identical mutation to that found in SE Asia, or an alternative mutation in the same gene, which still confers suppression. Alternatively, there may be a suppression-conferring mutation in a different gene within the region identified as containing the suppressor.
The presence of novel swept alleles at loci linked to the suppressor indicates that migration is the most parsimonious explanation for suppressor origin. Variants not present in the 2001 sample were observed to be the main ‘swept’ allele at 4 of the 11 loci at which significant change was detected (indicated with green arrows in Fig. 3). At two of these loci (A & I), the invading allele was defined by a single nucleotide polymorphism (SNP) being absent from the 2001 sample, whereas the other two alleles represented different combinations of existing SNPs. The four loci were in three genomic locations spaced over 17 cM and showed no evidence of linkage disequilbrium in the 2001 pre-sweep sample, and thus they can be treated as independent from each other (Fig. 5). They therefore support (but do not definitely prove) a migratory origin.
None of the loci tested in this study are likely to be the suppressor locus itself (markers were selected that spanned chromosome 25 and had conserved exon sequence – with several being housekeeping genes). Future research should aim to establish the actual nature of the suppressor mutation in both Samoa and SE Asia through fine-scale genetic mapping. Such a project will allow the source of suppression on Samoa (migration or in situ mutation) to be clarified. Beyond this, it will reveal the actual target of selection in this system. It has been widely conjectured that the evolution of sex determination systems might occur in response to the presence of sex ratio distorting microbes [20]. It is notable that a strong candidate gene – doublesex – resides within the equivalent genomic area in Bombyx mori, and with conservation of synteny being profound in Lepidoptera, is likely to reside in this area in Hypolimnas. Doublesex represents a tempting candidate as it is known that splicing of this gene is altered in the presence of male-killing Wolbachia in another lepidopteran, the moth Ostrinia scapulalis [21].
We utilized high-throughput sequencing of the transcriptome of H. bolina to obtain coding sequence from multiple loci across the genome. Following total RNA extraction from 1 male and 1 female adult H. bolina, mRNA library construction and sequencing using the Roche 454 sequencing platform (http://www.454.com), 450 bp reads were de novo assembled into contigs using the Newbler assembler to create the first set of Expressed Sequence Tags (EST) for H. bolina. The trimmed reads have been deposited as one male, and one female, partial transcriptome datasets in the NCBI SRA database, accession SRP045735.
In the absence of any annotated genome or transcriptome for H. bolina, the moth Bombyx mori was used as a proxy reference genome, this being the only available resource for Lepidoptera at the time of the study. There is a high level of synteny of gene location in the Lepidoptera [22] allowing a targeted gene approach, in which several genes could be selected from each chromosome across the genome. Coding sequence of highly conserved genes such as ribosomal proteins and housekeeping genes from B. mori were initially targeted and then retrieved from NCBI (http://www.ncbi.nlm.nih.gov). To determine putative H. bolina orthologs a local tBLASTx was then performed against the H. bolina EST set. Only genes that returned a single tBLASTx hit were included, reducing the likelihood of the inclusion of paralogs in our marker set. The orthologous H. bolina contigs were then translated into amino acid sequences using the ExPASY online tool (http://web.expasy.org/translate), with the sequence lacking mid-sequence stop codons chosen as the most likely translation. In a final test for paralogs, a reciprocal BLAST was performed of coding sequence from the orthologous H. bolina contigs as queries against the B. mori genome using the INPARANOID8 search tool ([23]; http://inparanoid.sbc.su.se/).
Where present, intronic regions were targeted for marker development, as they are likely to have a higher degree of nucleotide diversity. Again, conservation of synteny in Lepidoptera genome organisation allowed the intron/exon boundaries in H. bolina genes to be inferred using the B. mori genome. Through tBLASTx analysis of the B. mori coding sequence of the targeted gene against the B. mori WGS (Whole Genome Shotgun contigs) database in NCBI, exonic regions were identified (as only these regions will align). The translated orthologous H. bolina contig and the corresponding B. mori amino acid sequence were aligned using ClustalW [24] and the position of the intron/exon boundaries subsequently located.
Once intron/exon boundaries were identified in B. mori genes, and extrapolated to the H. bolina orthologous sequences, primers were designed for H. bolina that spanned introns of size 500–1000 bp (Bombyx size approximation). This size range was chosen to enable successful amplification of the intronic region during PCR. Marker optimisation was performed using three test H. bolina samples and successful PCR products were sequenced using Sanger technology.
In order to investigate the genetic architecture of male-killing suppression in H. bolina and determine markers in linkage with the suppressor locus, we crossed females of a butterfly population (the Philippines) that were Wolbachia-infected and homozygous for the male-killing suppressor allele (SS) to males from a Wolbachia-infected population (Moorea, French Polynesia) that lacked the suppressor (ss), to create suppressor-heterozygous Wolbachia-infected offspring (Ss) (S1 Figure).
Recombination does not occur during female meiosis in the Lepidoptera [25], permitting the progeny of Ss females to be used to identify the linkage group (SLG, Suppressor Linkage group) in which the dominant suppressor allele was carried. To this end, Ss females were crossed with ss males to produce the female-informative families. For inclusion in the SLG, markers linked to the suppressor locus are characterized by being present in all surviving sons of the Ss heterozygous mother, rather than the 50% expectation from Mendelian segregation with random survival. Initially each marker was sequenced in the F1 parents (Ss female×ss male). In each case, SNPs were chosen that were heterozygous in the female and homozygous in the male – following the presumed pattern of the suppressor. These same SNPs were then scored in 16 male and 8 female F2 progeny. Once a marker had been found that was present in half of the daughters (following Mendelian inheritance) but all of the sons (for a son to survive it must have at least one copy of the suppressor, and hence linked marker allele), further markers were developed for that same chromosome based on synteny with B. mori. A final suite of 12 markers that produced clean sequence and that spanned the suppressor-associated chromosome were developed to form the SLG.
Recombination does occur in male H. bolina, and thus crosses of Ss males to ss females (the male-informative families) allow a) mapping of genetic markers within a chromosome relative to each other and b) mapping of the suppressor within the linkage group, in terms of a region of the chromosome that is always present in surviving sons. To this end, the 12 linked markers were sequenced in the female F2 (n = 307) from one male informative cross (Ss male×ss female) and a linkage map created using JoinMap (version 3.0; Haldane mapping function) [26]. To place the suppressor locus within the map F2 males (n = 60) from this cross were analysed using the same 12 markers. Absence of recombinants in a core subset of markers, flanked by markers with an increasing numbers of recombinants, indicated the position of the suppressor locus (Fig. 1).
A population sample of butterflies from three time points (2001: n = 48, 2006: n = 48, 2010: n = 46) were collected from the Samoan island of Upolu. For each individual, DNA was extracted using the Qiagen DNeasy kit (www.qiagen.com), and the suite of 12 suppressor-linked markers amplified using PCR. Following Sanger sequencing of the amplicons through both strands, the resultant marker sequences were alignment in Codoncode (www.codoncode.com/). SNPs present within and between the population samples were then identified and scored for each individual butterfly. Using the SNP data (given in S7 Table), the alleles present at each marker in each population sample were estimated using the haplotype reconstruction software PHASE (version 2.1 [27], [28]) with 1000 iterations, a thinning interval of 100 and 1000 burn-in iterations. Allele frequencies at each marker for each time group could then be calculated and compared. Output was also examined by eye, with alleles identified first where there was no ambiguity (either homozygous, or a SNP separating into two defined alleles). Thereafter, alleles were assumed identical to those already identified where possible. The low allele diversity meant this visual analysis produced very similar result to PHASE output, which can thus be considered robust.
Patterns of genetic differentiation were estimated using GENEPOP [29]. First, heterogeneity of allele frequency distributions between pairs of time points was estimated using a G test based on allele frequency distribution. Where allele distributions were heterogeneous, we ascertained the allele whose frequency change made the greatest contribution to heterogeneity as that with the largest standardized residual within the heterogeneity test [18]. This allele was then removed (it was an allele increasing in frequency in each case), and the data retested to ascertain if the population samples were then homogeneous, or whether there was evidence for a second allele that changed in frequency (a second allele was identified in three cases). We additionally used FST standardized population genetic differentiation to quantify the magnitude of change between allele frequency distributions between the two samples. In each case, the rare individuals where sequence could not be obtained for particular alleles, or not inferred accurately, were coded as missing information.
DNA polymorphism statistics and estimates of nucleotide diversity (number of segregating sites, number of haplotypes, pi, theta, the average number of nucleotide sequences (k), Tajima's D, haplotype diversity (Hd)) for each marker for each time point were conducted in DnaSP (version 5) [30]. These statistics were estimated using sequence data excluding gaps i.e. indel mutations were not used (present in 8 of 9 unlinked markers).
Nine unlinked markers, from 8 different chromosomes, were also sequenced for the 2001 and 2006 population samples to investigate the degree to which changes were observed in the wider genome and as a control for demographic effects. These were tested for the presence of heterogeneity between time points using a G test based on allele frequency distributions, for differentiation using the FST statistic, and several polymorphism statistics as described above for the SLG markers.
We additionally analysed evidence for alteration in the pattern of linkage disequilibrium, again using GENEPOP. The significance of LD between all possible combinations of loci was tested in the 2001 and 2006 samples separately. We do not report the magnitude of LD, as this is not a standardized measure, being dependent on the allele frequency distribution at each locus.
|
10.1371/journal.pgen.1006637 | A genome wide association study identifies a lncRna as risk factor for pathological inflammatory responses in leprosy | Leprosy Type-1 Reactions (T1Rs) are pathological inflammatory responses that afflict a sub-group of leprosy patients and result in peripheral nerve damage. Here, we employed a family-based GWAS in 221 families with 229 T1R-affect offspring with stepwise replication to identify risk factors for T1R. We discovered, replicated and validated T1R-specific associations with SNPs located in chromosome region 10p21.2. Combined analysis across the three independent samples resulted in strong evidence of association of rs1875147 with T1R (p = 4.5x10-8; OR = 1.54, 95% CI = 1.32–1.80). The T1R-risk locus was restricted to a lncRNA-encoding genomic interval with rs1875147 being an eQTL for the lncRNA. Since a genetic overlap between leprosy and inflammatory bowel disease (IBD) has been detected, we evaluated if the shared genetic control could be traced to the T1R endophenotype. Employing the results of a recent IBD GWAS meta-analysis we found that 10.6% of IBD SNPs available in our dataset shared a common risk-allele with T1R (p = 2.4x10-4). This finding points to a substantial overlap in the genetic control of clinically diverse inflammatory disorders.
| Leprosy still affects approximately 200,000 new victims each year. A major challenge of leprosy control is the prevention of permanent disability due to nerve damage. Nerve damage occurs if leprosy remains undiagnosed for extended periods or when patients undergo pathological inflammatory responses termed Type-1 Reactions (T1R). T1R is a rare example where beneficial inflammatory responses are temporal separated from host pathological responses. There is strong experimental evidence that supports a role of host genetic factors in T1R susceptibility. Here, we employed a genome-wide association study (GWAS) to investigate susceptibility factors for T1R in Vietnamese families. We followed up the initial GWAS findings in independent population samples from Vietnam and Brazil and identified a set of cis-eQTL genetic variants for the ENSG00000235140 lncRNA as global risk factors for T1R. To test our proposal that T1R is a strong model for pathological inflammatory responses we evaluated if inflammatory bowel disease (IBD) genetic risk-factors were enriched among T1R risk factors. We observed that more than 10% of IBD-risk loci were nominally associated with risk for T1R suggesting a shared mechanism of excessive inflammatory response in the both disease etiologies.
| A clear temporal separation from the different stages of leprosy pathogenesis identifies the endophenotype Type-1 Reactions (T1Rs) as a well-delineated example for host pathological inflammatory responses in humans. An endophenotype, as defined by John and Lewis in 1966, is a microscopic and internal phenotype that is not easily identified in the presence of an exophenotype, which is the dominating phenotype that is more easily recognized [1]. In the context of our study we refer to the term endophenotype as a condition (T1R) that occurs in some but not all persons displaying the necessary exophenotype (leprosy) diverging from the original concept of John and Lewis. Of note, T1R shares immune-pathological similarities with immune reconstitution inflammatory syndrome of HIV patients undergoing highly active antiretroviral therapy [2], and paradoxical reactions in patients with Buruli ulcer undergoing anti-microbial therapy [3, 4].
T1Rs are a major challenge of current leprosy control since the hyper-inflammatory immune response triggered by Mycobacterium leprae, the etiological agent of leprosy, frequently leads to permanent nerve damage [5]. A prompt identification of T1R cases and rapid clinical intervention are essential to prevent lasting neurological damage [6]. While acute neuritis is a hallmark of T1R, the detailed mechanisms that link hyper-inflammation to neuropathy are not known. Depending on the epidemiological setting, 30% to 50% of leprosy cases develop at least one T1R episode [5, 7–10]. Why only a fraction of leprosy-infected individuals undergo T1R is not known but the description of a transcriptome signature in response to M. leprae antigen strongly supported a genetic predisposition to T1R [11]. In addition, genetic variants in a few number of candidate genes (TLR1, TLR2, NOD2, LRRK2 and TNFSF15/TNFSF8) were found to be associated with T1R [12–17]. Independently, variants in several of these genes had also been implicated in susceptibility to leprosy per se raising the possibility of an overlapping genetic control of intensity of pathway activation between protective and pathological host responses [18]. To contrast the genetic control of leprosy and its clinical subtypes from the genetic control of the pathological immune responses typical for T1R, we designed a genome-wide association scan (GWAS) to identify novel genes or variants associated solely with T1R. This may lead to predictive biomarkers for early recognition of T1R and possibly indicate novel pharmacological interventions that reduce the need for potentially adverse long-term corticoid treatment in T1R.
We evaluated the association of host genetic factors with T1R by conducting a family-based GWAS in 221 families with 229 T1R-affect offspring followed by stepwise replication in independent population-based case-control samples (Fig 1). For the discovery phase, approximately 6.3 million genotyped and imputed variants (SNPs and INDELs) that passed quality control were tested for association in both T1R-affected and T1R-free family sets. In T1R-affected families, a suggestive association with T1R was detected on chromosome region 10p21.2 (Fig 2A and 2B). Among the 103 SNPs located in the interval and strongly associated with T1R leprosy (pDiscovery < 0.001), SNP rs7916086 (pDiscovery = 8.2x10-7) displayed the strongest evidence of association. Applying a linkage disequilibrium (LD) threshold of r2 > 0.9, the 103 SNPs located between the two recombination hot-spots in the 10p21.2 locus could be grouped into seven SNP bins (Fig 2C, S1 Table). None of the SNPs in the 10p21.2 locus located outside this hot spot showed evidence for association below p < 0.001. The tag SNP that presented the lowest p value for the association with T1R in each of the seven SNP bins was selected as the leading variant for its particular bin. When the 220kb region comprising the T1R-risk locus was evaluated in the T1R-free families no signal of association was detected (Fig 2C, S1 Table). The formal heterogeneity test confirmed preferential association of T1R with the seven SNP bins reported in the discovery phase with p Heterogeneity ranging from 0.009 to 5.0x10-04 (S1 Table). Of note, an additional 4372 variants located throughout the genome displayed p < 0.001 in the T1R-affected subset and are given in S2 Table.
A multivariable analysis including the leading variant of each SNP bin (r2 = 0.9) associated with T1R selected rs7916086 as the single signal of association in the 10p21.2 chromosomal region (S1 Table). However, due to high LD among SNPs of the investigated bins, alternative models could not be excluded (S1 Fig). Therefore, we selected the seven leading variants for each of the SNP bins (r2>0.9) described above for further confirmative analyses in independent populations.
The leading SNP in the discovery phase, rs7916086, showed borderline evidence for association with T1R in the Vietnamese replication sample (p = 0.04). However, association of rs7916086 with T1R was not validated in the Brazilian sample (p = 0.26) (S3 Table). The leading SNPs in four additional SNP bins, namely rs10509110, rs11006600, rs10826329 and rs10763614, did not show consistent evidence for significant association across the Vietnamese and Brazilian populations (S3 Table). In contrast, SNP rs1875147 displayed strong replicated and validated evidence of association with T1R (Table 1). SNP allele “C” of rs1875147 was identified as global risk factor for T1R with an odds ratio (OR) = 1.37; confidence interval of the one-sided test (uniCI) 95% = 1.11; p = 0.006, in the Vietnamese replication sample and, OR = 1.47; uniCI 95% = 1.15; p = 0.005, in the Brazilian validation sample (Table 1). In addition, the tag SNP for a second bin, rs10826321, was associated with T1R in the Vietnamese replication (p = 0.003) and the Brazilian validation sample (p = 0.04 Table 1). In Vietnam, SNPs rs1875147 and rs10826321 were highly correlated (r2 ≈ 0.8) capturing the same signal of association with T1R. However, compared to the Vietnamese, the LD between rs1875147 and rs10826321 was lower in Brazilians (r2 = 0.21; S1 Fig). Since the 7 SNP were tested for replication and validation we did not apply a Bonferroni correction.
To investigate the independent effect of rs1875147 and rs10826321 in Brazilians we performed a multivariable analysis. SNP rs1875147 maintained the association with T1R (p = 0.009) while rs10826321 lost significance (p = 0.49). Next, we investigated the combined effect of rs10826321 and rs1875147 by conducting a haplotype analysis in the Brazilian sample. We found that the haplotype with the T1R-risk allele in both SNPs (G-C alleles for rs10826321 and rs1875147 respectively) was significantly associated with T1R (p- = 0.04; S4 Table) consistent with results obtained by multivariable analysis supporting the non-independent association of rs1875147 and rs10826321 with T1R. Interestingly, the haplotype (A—C) containing the alternative allele for rs10826321 and the T1R-risk allele for rs1875147 showed a trend towards association with T1R in Brazilians (p = 0.06; S4 Table). This observation supported rs1875147 as the main cause of association of T1R with the 10p21.2 region. When a combined analysis was performed to summarise all study phases, only SNPs rs1875147 surpassed the genome wide threshold for significant association with T1R (Table 1, Fig 2D). In a fixed-effect meta-analysis SNPs rs1875147 presented an OR = 1.54; CI 95% = 1.32–1.80, p = 4.5x10-08 for the C-allele. As modest levels of population heterogeneity were observed for the T1R-risk SNPs in a complementary fixed-effect model (Table 1; S3 Table), we performed a random-effect meta-analysis. The seven SNPs showed similar levels of significance between the fixed and random-effect (S5 Table). For the rs1875147 the random-effect model resulted in a risk-effect of OR = 1.54; CI 95% = 1.28–1.86, p = 6.4x10-08 for the C-allele.
The locus validated for association with T1R mapped within two recombinational hot spots where a single long non-coding RNA (lncRNA) was located (Fig 2D). The novel lncRNA presented two isoforms, one encoded by the ENSG00000235140 (a.k.a. RP11-135D11.2) gene and another encoded by the uncharacterized LOC105378318 (Fig 2D). The two T1R-risk variants, rs1875147 and rs10826321, are located at 6.5 kb and 8.7 kb, respectively, upstream of the transcription start site of the ENSG00000235140 gene. The rs10826321 variant alters the binding motif of a CTCF transcription factor in a CTCF binding site in 83 cell types (S2A Fig). The rs10826321 T1R-risk G-allele is more commonly observed in CTCF binding than the alternative A-allele (S2A Fig). SNP rs1875147 is reported as an expression quantitative trait locus (eQTL) for ENSG00000235140 in the transverse colon where the T1R-risk allele C is correlated with higher gene expression (S2B Fig) [19]. The eQTL effect for rs1875147 was also nominally significant in the terminal ileum of the small intestine and in the spleen in a smaller sample size (S2C Fig). Both rs1875147 and rs10826321 are conserved loci across species [20].
Certain SNP alleles associated with T1R-risk had previously been shown to be susceptibility factors for inflammatory bowel disease (IBD) [21–23]. To investigate if there was an enrichment of risk alleles between T1R and IBD, we systematically compared evidence of association with T1R in the Vietnamese discovery set with evidence for association in a recent GWAS meta-analysis for IBD [24] (Fig 3A).
Of 232 independent top SNPs that had been associated with IBD by meta-analysis, 208 were available in the T1R-affected and T1R-free GWAS datasets [24]. For 22/208 SNPs (10.6%) the IBD risk allele was associated at the 0.05 level with risk of T1R/leprosy. (Fig 3A, S6 Table). This observed proportion of shared risk-alleles between T1R leprosy and IBD is significantly non-random (p = 2.4x10-4). Importantly, none of the 22 SNPs showed significant evidence of association with T1R-free leprosy while 9 SNPs displayed significant heterogeneity between leprosy and T1R indicating an enrichment of stringently defined T1R SNPs among IBD SNPs (p = 1.9x10-3; Fig 3, S6 Table). Similar analyses in T1R-free families, failed to detect an enrichment of leprosy risk alleles among IBD SNPs. Indeed, while several genes with known overlap of IBD and leprosy were detected (i.e. RIPK2, LACC1 and IL23R), there was no genome-wide statistical enrichment for IBD risk alleles in T1R-free leprosy (p = 0.09; Fig 3A). As additional control, we evaluated three non-immunity phenotypes for which recent GWAS meta-analyses were available (Schizophrenia [25], human height [26] and human blood metabolites [27]) for an overlap of genetic risk factors with T1R. There was no significant enrichment of either leprosy or T1R risk alleles with SNP alleles of any of the three control phenotypes (Fig 3B to 3D).
Among the 22 IBD SNPs associated with T1R leprosy, 17 are cis eQTL for one or more genes (S3 Fig). Similarly, 7 of the 9 SNPs significantly heterogeneous between T1R and leprosy were eQTLs in either whole blood, rs3774937 (NFKB1), rs10065637 (ANKRD55), rs11150589 (ITGAL) and rs2836878 (lncRNA ENSG00000235888) or multiple tissues, rs4664304 (LY75), rs113653754 (HLA-DQB1) and rs4768236 (LRRK2; S4 Fig) [19, 28]. SNPs that were eQTL in multiple tissues displayed some of the strongest associations with T1R (S6 Table). Since the LY75 gene encodes a major endocytic receptor of dendritic cells and HLA-DQB1 gene expression is also modulated by a risk SNP, our results highlight the critical role of antigen presentation in dysregulated immunity of both IBD and T1R.
In summary, we have conducted the first GWAS for pathological inflammatory responses in leprosy using the largest collection of T1R-affected individuals to date. Our stepwise replication study in ethnically independent populations led to the description of an eQTL (rs1875147) for the lncRNA gene (ENSG00000235140) as a global risk-factor for T1R. Moreover, we have observed an enrichment of shared risk-alleles between leprosy/T1R and IBD but not for IBD and leprosy per se.
We have shown previously for the PARK2 gene that testing only the leading SNP of the discovery phase in ethnically independent populations without considering population differences in the LD structure may result in false negative associations [29]. Here, the leading SNP in the Vietnamese discovery phase, rs7916086, could not be validated for the association with T1R; but rather, two SNPs highly correlated with rs7916086 in the Vietnamese population (namely rs1875147 and rs10816321) were T1R-risk factors in Brazilians. The lower LD conservation in Brazilians enabled us to narrow down the T1R association signal in the 10p21.2 region to a single SNP, rs1875147, which presented a pre-established regulatory function. Since we used the 1000 Genomes data to impute SNPs for the analysis and chose a high r2 cut off for SNP bin definition, it is unlikely that another common SNP in strong LD with rs1875147 would provide a stronger signal of association. However, we cannot rule out a combination of rare variants as cause of the association signal. Combined, our results highlight the strength of employing different ethnicities in the validation phase since the genetic effects of rs7916086, rs10826321 and rs1875147 could not be disentangled in the Vietnamese sample.
An association with leprosy was previously reported for chromosome region 10p21.2 [30]. The reported peak of association with leprosy per se encompassed the ADO and EGR2 genes. The leading variant in the ADO/EGR2 locus, rs58600253, is located at approximately three mega bases upstream of the T1R associated locus. When the imputed variant rs58600253 (Info = 0.992) was evaluated in the T1R-affected and T1R-free families we observed no significant signal of association (p = 0.25 and p = 0.22, respectively). Moreover, no correlation of rs58600253 with the T1R signal tagged by rs187514 was detected using the best call genotypes (r2 = 0.04). These results indicated that the T1R locus on region 10p21.2 is independent of the leprosy per se ADO/EGR locus. Moreover, a recent GWAS meta-analysis by Wang et al. identified four novel loci associated with leprosy [31]. While none of the leading SNPs reported by Wang et. al. were significant in our T1R GWAS, we observed independent variants associated with leprosy in two out of the four newly reported loci. The rs4684104 SNP near the PPARG gene (p = 2.4 x 10−6; p Heterogeneity = 5.4 x 10−4) and the rs10239102 near the BBS9 gene (p = 4.2 x 10−4, p Heterogeneity = 0.07) were T1R-specific and T1R-non-specific, respectively.
The functional annotation for the rs1875147 T1R-risk alleles argues that upregulation of ENSG00000235140 transcription may contribute to T1R susceptibility. However, this lncRNA gene has not been found to be commonly expressed in all tissues. The ENSG00000235140 gene was detected mostly in the sexual organs, gastro intestinal tract, and in the lungs of healthy individuals [19, 32]. These tissues usually do not harbor M. leprae, but are a reservoir for other mycobacteria such as M. avium paratuberculosis (colon) and M. tuberculosis (lungs). The limited knowledge about the role of ENSG00000235140 in health and disease limits our understanding of this lncRNA in T1R pathogenesis. Notwithstanding, our data present the ENSG00000235140 gene as a prime candidate to unravel the riddle of pathological immune responses in T1R and possibly inflammatory disorders in general.
An overlap regarding the genetic control of leprosy per se and IBD has been previously suggested [21–23, 33]. Although the SNPs associated with IBD and leprosy are frequently the same the risk-allele are less consistent. This factor hinders the establishment of a shared biological mechanism for IBD and leprosy. As T1R affects a considerable proportion of leprosy cases it is possible that, at least partially, the genetic overlap proposed between IBD and leprosy is due to the T1R phenotype. Here our strategy was to evaluate if T1R and IBD shared additional risk-alleles. Although, our approach focusing only on the leading SNP per IBD locus was conservative, the enrichment for shared risk-alleles in IBD and T1R was strong and may represent only part of the shared biological mechanisms. The results reported here strongly support the view that susceptibility to IBD involves a genetic predisposition to mount dysregulated inflammatory immune responses as exemplified by the T1R phenotype in leprosy.
In complex traits, precise phenotype definition is key for the detection of genetic associations. For example, we have previously shown for variants of the TNFSF15/TNFSF8 genes that leprosy patients with the T1R endophenotype are largely the cause of association with the leprosy exophenotype [16, 17]. Consequently, the replication of the TNFSF15/TNFSF8 association in samples of leprosy patients with a low proportion of T1R is expected to display low power. Equally important, accurate phenotype definition directs the interpretation of detected associations. Assigning genes to the exophenotype leprosy that impact on the endophenotype T1R may lead to wrong conclusions about the pathology of leprosy. Hence, a notable strength of our study is the focus on a well-defined endophenotype which is directly connected to a major problem of current leprosy control. This increases the power for detection of genetic effects while at the same time opening a translational link for control of nerve damage.
Despite these strengths, our study also had limitations. For example, we only tested an additive model, since T1R is highly prevalent in leprosy (30 to 50% of all cases); dominant and recessive models of inheritance could unveil additional novel associations. Moderate levels of population heterogeneity were observed in the combined analysis (I2 values ≈ 30 to 50; Table 1, S3 table). The population heterogeneity was likely driven by a winner's curse phenomenon, a bias that inflates risk estimates for newly identified SNPs when a study lacks statistical power [34]. Because of the possible effect of winner’s curse, the combined risk effect should be consider as a summary of our study and the real risk-effect for variants in the 10p21.2 region are likely closer to the effect of the replication and validation phase. A second limitation is the pleiotropic analysis of IBD and leprosy/T1R. As a consequence of the T1R/leprosy sample size, intermediary to low frequency variants with modest genetic effect would not have been detected by our study. This might have led to an increased type II error and an under-estimation of the true overlap in the genetic control of IBD and T1R/leprosy. Hence, studies employing larger numbers of T1R/leprosy patients might provide better estimates of the overlap in the genetic control of these two inflammatory conditions.
The study was conducted according to the principles expressed in the declaration of Helsinki. Written informed consent was obtained for all adult subjects participating in the study. All minors assented to the study, and a parent or guardian provided the informed consent on their behalf. The study was approved by the regulatory authorities and ethics committees of the participating centers. Namely, Comissão Nacional de Ética em Pesquisa (CONEP; 12638) for Goiania; The Research Ethics Committee at Fiocruz (CEP-Fiocruz Protocol 151/01) for Rio de Janeiro; The Research Ethics Committee at Institute Lauro de Souza Lima for Rondonópolis (172/09); the Research Ethics Board at the RI-MUHC in Montreal (REC98-041), and the regulatory authorities of Ho Chi Minh City (So3813/UB-VX and 4933/UBND-VX) for the Vietnamese population.
The subjects included in the study where followed up for a minimum of three years to confirm the presence or absence of T1R episodes. T1R-affected and T1R-free leprosy cases were mainly selected from the borderline class of Ridley and Jopling clinical scale of leprosy as T1R affects predominantly these cases that present an immunologically unstable immune response against M. leprae infection [7, 35]. For the discovery phase, two sets of families of Vietnamese (Kinh) origin with leprosy-affected offspring were selected: the T1R-affected set comprised of 229 offspring belonging to 221 families and a T1R-free set comprised of 229 offspring in 209 families. The T1R-free set was matched to the T1R-affected set by the offspring’s leprosy clinical subtype. In the discovery phase, a transmission disequilibrium test (TDT) was applied to the T1R-affected and the T1R-free families independently. Next, the results of the individual TDTs were compared to investigate heterogeneity between both samples. The genetic heterogeneity test between T1R-affected and T1R-free subsets was tested by means of the FBATHet statistic and is detailed in the statistical approach section [36]. Variants that were associated in the T1R-affected set and showed heterogeneity with the T1R-free set were considered as T1R-specific and were investigated in the next phases of the study.
The initial association results were followed up employing a replication and a validation phase. The replication sample was of Vietnamese ethnicity and encompassed 253 T1R-affected and 563 T1R-free leprosy patients. The validation sample comprised 471 T1R-affected subjects and 446 T1R-free leprosy patients as controls from the Central-west and South-east regions of Brazil as described previously [16, 37, 38]. In both replication and validation samples, cases and controls were matched for leprosy subtype.
Genotypes of all subjects of the discovery phase were determined using the Illumina Human 660w Quad v1 bead chip. SNPs with call rate < 0.98, more than two Mendelian errors in T1R-affected or T1R-free sets, minor allele frequency (MAF) < 0.01 or presented Hardy-Weinberg equilibrium (p < 1.0 x 10−3) in 763 leprosy unaffected parents were removed from the analyses. Genotypes for the replication and validation phase samples were obtained through high-throughput SEQUENOM platform. The same quality control thresholds from the discovery phase were applied for SNP call rates and MAF exclusion to the replication and validation phase, with the exception of the HWE p value cut off which was restricted to p < 0.05 due to the lower number of tested SNPs compared to the discovery phase.
A total of 38,753 genotyped A/T and C/G SNPs were removed prior to the phasing and imputation. The remaining 495,973 SNPs that passed the quality control filtering in the discovery phase were used to impute additional 11.5 million variants (SNPs and INDELs) in both T1R-affected and T1R-free family sets with SHAPEIT2 [39] and IMPUTE2 [40] software and the 1000 genomes Phase I v3 dataset containing 1092 individuals as the reference panel. Given the exploratory nature of the discovery phase, the threshold of imputation information measure (Info) > 0.5 was applied to capture most of the common variants (MAF > 5%) with reasonable confidence (S5 Fig) [41], MAF > 0.001 and more than 10 informative families in both T1R-affected and T1R-free sets were used as a post-imputation quality control filtering for the association analyses. Imputed variants that were evaluated in the replication and validation phase had their genotypes confirmed in 440 subject of the discovery sample using the high-throughput SEQUENOM platform.
In the discovery phase, a TDT was used to estimate non-random transmission of alleles from heterozygote parents to leprosy-affected offspring in both T1R-affected and T1R-free sets (p Discovery). The analysis was carried out under a log-additive model using FBATdosage v2.6 for genotyped and imputed variants [42]. To contrast the TDT tests from the discovery phase a FBATHet test in T1R-affected and T1R-free sets was used (p Heterogeneity). Briefly, heterogeneity of the allelic transmission rates in an endophenotype can be done in the FBATdosage framework by pooling the two subsets (T1R-affected and T1R-free) and contrasting the presence of the endophenotype T1R (T1 = 1/V1) with the absence of T1R (T2 = −1/V2), where V1 and V2 denote the variance of the FBATdosage statistic for the each sample set, respectively [36].
Population-based association analyses were performed using logistic regression under a log-additive model and adjusting by the co-variables gender and age at leprosy diagnosis using PLINK v1.0.7. The one-sided test was used with the alternative hypothesis that the T1R-risk alleles were also risk factors in the replication and validation samples. Multivariable analysis were performed with stepwise conditional logistic regression in SAS 9.3. The haplotype analysis in the Brazilian sample was performed with THESIS v3.1 [43]. The linkage disequilibrium structure was evaluated with Haploview 4.1 [44].
To summarize the different steps of the study we used an inverse variance–weighted meta-analysis with a fixed-effect model and an alternative random-effect model proposed by Han and Eskin as implemented in the software METAL [45] and METASOFT [46], respectively. To estimate the risk effect for the family-based design the un-transmitted allele from parents to T1R-affected offspring in the TDT was used as a pseudo-sib control. Briefly, up to three unaffected pseudo-sibs were created per family, one for each possible un-transmitted genotype. Subsequently, the original T1R-affected offspring were compared to the T1R-free pseudo-sibs in a matched case-control [47]. Under a log-additive model, TDT and pseudo-sibs analyses are equivalent [47]. Of note, METAL and METASOFT use standard errors and β coefficients to combine the statistics of each studied phase. In contrast to the replication and validation steps, a two-sided test was used in the combined analysis for the Vietnamese and Brazilian samples.
To investigate if there was an enrichment of shared risk alleles between T1R and IBD, we used a hypergeometric test to systematically compare evidence of association with T1R in the Vietnamese. For instance, out of the 6,333,954 variants tested for association in our study 319,671 had p < 0.05 in the T1R-affected subset. Using the observed prior information of the number of variants with p < 0.05, the hypergeometric test calculates the statistical significance of randomly selecting 22 variants with p < 0.05 when 208 variants (number of variants from the IBD GWAS meta-analysis present in the T1R dataset) were randomly drawn from a total of ~6.3 million. Here, the hypergeometric test corresponds to the one-tailed Fisher’s exact test. The same analytical approach was applied for the T1R specific variant in IBD, but in this analysis we used the number of variants with p < 0.05 and p heterogeneity < 0.05 out of a total of ~6.3 million variants of the GWAS. Since we tested for sharing of the same risk allele between T1R, and IBD, CD or Ulcerative Colitis (UC) one-tailed p values are reported. The same strategy was used in the three control phenotypes (schizophrenia, height and blood metabolites. Since we tested for sharing of the same risk allele between T1R, and IBD, CD or one-tailed p values are reported.
IBD meta-analysis data was freely available at the IBDgenetics website (https://www.ibdgenetics.org/) [24, 48]. Briefly, seven CD and eight UC collections with genome-wide data were combined with additional replication samples resulting in a total of 42,950 IBD cases and 53,536 health controls for the IBD meta-analysis [24, 48]. Variants that surpassed p < 5.0 x 10−8 for association with IBD were reported as significant. Functional data for annotated SNPs were extracted from the GTeX (http://www.gtexportal.org/home/) and Haploreg v4 http://www.broadinstitute.org/mammals/haploreg/haploreg.php databases. [19, 20]
The FBAT dosage is available at https://www.hgid.org/index.php?menu=download
|
10.1371/journal.pntd.0005733 | Modeling the risk of transmission of schistosomiasis in Akure North Local Government Area of Ondo State, Nigeria using satellite derived environmental data | Schistosomiasis is a parasitic disease and its distribution, in space and time, can be influenced by environmental factors such as rivers, elevation, slope, land surface temperature, land use/cover and rainfall. The aim of this study is to identify the areas with suitable conditions for schistosomiasis transmission on the basis of physical and environmental factors derived from satellite imagery and spatial analysis for Akure North Local Government Area (LGA) of Ondo State. Nigeria. This was done through methodology multicriteria evaluation (MCE) using Saaty’s analytical hierarchy process (AHP). AHP is a multi-criteria decision method that uses hierarchical structures to represent a problem and makes decisions based on priority scales. In this research AHP was used to obtain the mapping weight or importance of each individual schistosomiasis risk factor. For the purpose of identifying areas of schistosomiasis risk, this study focused on temperature, drainage, elevation, rainfall, slope and land use/land cover as the factors controlling schistosomiasis incidence in the study area. It is by reclassifying and overlaying these factors that areas vulnerable to schistosomiasis were identified. The weighted overlay analysis was done after each factor was given the appropriate weight derived through the analytical hierarchical process. The prevalence of urinary schistosomiasis in the study area was also determined by parasitological analysis of urine samples collected through random sampling. The results showed varying risk of schistosomiasis with a larger portion of the area (82%) falling under the high and very high risk category. The study also showed that one community (Oba Ile) had the lowest risk of schistosomiasis while the risk increased in the four remaining communities (Iju, Igoba, Ita Ogbolu and Ogbese). The predictions made by the model correlated strongly with observations from field study. The high risk zones corresponded to known endemic communities. This study revealed that environmental factors can be used in identifying and predicting the transmission of schistosomiasis as well as effective monitoring of disease risk in newly established rural and agricultural communities.
| Urogenital schistosomiasis is a parasitic disease whose transmission depends on environmental factors. The disease is one of the risk factors of bladder cancer, the second most common urogenital cancer after prostate cancer in Nigeria. This study aims to identify area suitable for the transmission of the disease through the use of environmental factors. This was achieved by using spatial analysis and multicriteria evaluation for Akure North Local Government Area (LGA) of Ondo State, Nigeria. The results showed that a large portion of the study area was highly favourable for the transmission of the disease. Of the five communities in the study area, only one had a low risk of transmission of the disease while the risk of transmission increased in the remaining four communities. The predictions made by the model was comparable to observations from field study. This study confirmed that analysis of environmental factors using relevant techniques can be applied in the study of transmission of parasitic diseases.
| Schistosomiasis is the most important water impounding disease and one of the diseases classified by the World Health Organization (WHO) as neglected. It is the second most prevalent tropical disease in its public health implications. It is widespread in Nigeria and has been estimated that 240 million people required treatment globally [1]. Nigeria has been placed at the top of the list of endemic countries in a more recent estmation [2, 3] as it accounts for 14% of the global number of Schistosoma infections. The highest prevalence of infection was found among children of five to ten years of age. Men were more likely than females to have the disease [4]. S. haematobium is the predominant specie in the country corresponding to 79.8% of all reported cases [5]. It often exceeds the high-endemicity threshold of 50% prevalence. Another recent review points out that there are about 100 new cases of bladder cancer and over 600 deaths annually and the major risk factor is infection with Schistosoma haematobium [6]. Thus, the control of schistosomiasis is one of the major prorities of the country’s health system [4].
Urogenital schistosomiasis is caused by the species of flat worms known as Schistosoma haematobium [7]. Adults of the parasites live in the blood of mammals but their life cycle requires a phase of asexual multiplication within a fresh water snail host mainly snails of Bulinus species. The fluke’s life cycle begins when adult female schistosomes deposit eggs in the veins surrounding the bladder of the mammal host. Eggs are extruded from the veins through the tissues into the bladder and are voided to the exterior in the urine [8]. The eggs that are deposited into fresh water hatch as a result of osmotic stress that causes the shells to rapture and embryos known as miracidia emerge [9]. The ciliated miracidia swim frantically through the water by means of cilia covering the body for several hours until they penetrate a suitable snail. Here, the miracidia undergo further development resulting in the formation of cercariae. The mature carcariae leave at the exit pore located at one extremity and are emitted into the water body by the snail in response to some physico-chemical factors of the water body. The most important of these factors are light and temperature.
Humans acquire an infection through the skin by getting in contact with water containing cercariae shed by the snail hosts. They are attracted to the secretions of the skins showing a strong positive response to arginine [10]. When the larvae come in contact with human skin, they secrete lytic enzyme from postacetabular glands which they use to pierce the skin. The larvae burrow through skin or mucous membrane of a human and once inside shed their forked tail and become schistosomulae. Provided that the human is susceptible to infection, the schistosomulae are carried through the lymphatic system and lungs to the portal venules of the liver. It is primarily in the liver and portal system that larvae develop into adult worms of both sexes [8]. It is also here that they copulate and produce eggs for reproduction of their species [11].
Fresh water snail’s distribution and abundance drive the links between transmission and environment/climate. The distribution of the several Bulinus species and possibly the new spread of B. forskalli, which are two different hosts with two different types of links with the environment, show the impact of environmental factors on the spread of schistosomiasis [12]. This is why temperature, rainfall, proximity to water bodies and other environmental factors may be possible factors and be useful to analyse for risk area determination.
The rapidly changing epidemiology of schistosomiasis necesitates that new approaches are used to study the transmission dynamics of the disease in order to promote prevention and control towards sustainenace of a healthy environmental [13,14]. Of recent, geographic information system/remote sensing (GIS/RS) has become a popular, indispensably available instrument applied in many researches including epidemiology, environmental resource management and programming [15,16,17,14,18,19]. GIS/RS provides an important opportunity of obtaining surface characteristic variables such as humidity, soil temperature and vegetative cover of schistosomiasis endemic areas, fitting these parameters with snail population and or prevalence thereby establishing a credible forecast model of schistosomiasis [13,14,18]. Risk maps are outcomes of transmission models in which environmental information has been merged with data from the fields of epidemiology and vector biology [20]. The potential of the combination of remote sensing and geographical information system based spatial analyses for schistosomiasis risk modeling can be used to determine the geographical limit of disease distribution, understand disease ecology and epidemiology, support prevention, surveillance and control through prioritising areas of disease risk and provide early warning for areas where disease transmission could become established [21].
This study aims at generating a predictive risk model for schistosomiasis at a local level by analysing satellite derived environmental data using different techniques.
Ethical approval was sought and obtained from the Ondo State Ministry of Health, Akure, Ondo State.
The study area comprises the whole of Akure North Local Government Area in Ondo state of Nigeria. The Local Government Area comprises of five major communities; Iju, Ita Ogbolu, Oba Ile, Igoba and Ogbese in Akure North LGA (Fig 1). These communities are located between latitudes 5°45' and 7°52'N and longitudes 4°20' and6°05'E. The population of the area is approximately 198,000. The vegetation type of the local government area is typically rainforest dominated by abundant trees and grasses. The economic activities in the area includes fishing and production of food and tree crops such as cocoa, rubber, oil—palm, cashew, teak, gmeligna and indigenous tree species. The predominant occupations in the communities are farming and trading. The area has a maze of numerous drainages (Ala, Oluwa, and Ogbese River).
Every organism has a range of tolerances for every abiotic condition including an optimum range at which it’s most abundant [22]. It has been stated by [23] that temperature is one of the most important physical influences of any biotope from an ecological perspective. Temperature generally affects snail population size and the fecundity of adult snails. Different snail species require a series of optimal temperature tolerance ranges for survival and reproduction [22] but the most favourable range lies between 18°C and 32°C [24]. The influence of temperature on the schistosomes' intra-molluscan stages as a determinant of the parasites' spatial distribution has also been studies by [25]. Eggs of the parasite hatch at temperatures ranging between 10°C and 30°C [26]. The period from penetration of the miracidium to initial shedding of cercariae by the snail varies with temperature between the minimum of 17 days at 30–35°C and several months at lower temperatures. It can be concluded from these studies that thermal variation influences both parasite biology and the rate of natural increase of snail species [27].
Rainfall is an important abiotic condition in the epidemiology of schistosomiasis as it affects the duration of desiccation [28] and the permanence of habitats through droughts and floods. Rainfall also plays a role by changing schistosomiasis transmission foci and affecting the seasonal patterns of cercarial production. The study by [29] on snails concluded that rainfall is one of the most significant factors affecting snail population fluctuations in tropical areas. Rainfall is important to this study because it is an indicator of the availability of water for disease transmission, especially in rain-fed habitats. A seasonal influx of water may also be positively correlated with infection rates and can change the focality of the disease [30]. It also exerts a strong influence on vegetation cover and acts indirectly on the snails through changes in the surrounding habitat's flora and fauna [31].
Elevation has been linked to current velocity. The influence of geomorphology as a significant determinant of current speed in Mpumalanga and Swaziland has been reported by [27] and found that the distribution of hard rock correlates with snail distribution. In a study by [32], an elevational limit of 800m-2200m above sea level was established as suitable elevation for S.mansoni due to hot water temperatures below 800m and cold water temperature above 2200m in Ethiopia. Altitude also influences temperature; the average temperature decrease with height in free atmosphere [33].
Changes in land use such as deforestation, human settlement, and construction of roads, waters dams, canals and irrigation systems have been accompanied by global increases in morbidity and mortality from parasitic diseases. Amongst the factors that determine the nature and extent of change in the incidence of parasitic disease are changes in land use and settlement and the time interval from one land use to another [34]. In China, the construction of the Three Gorges high dam on the Yangtze River affected the snail distribution and annual prevalence of human schistosomiasis varying with water levels in the Yangzte River [35].
Normalized difference vegetation index (NDVI) has been reported to be one of the most successful environmental factors for snail habitat prediction. Normalized Differentiation Vegetation Index (NDVI) represents the amount of chlorophyll in an area. Freshwater snails are usually associated with macrophytes. These leafy aquatic plants provide the host snails with shelter from solar radiation and currents and egg-laying sites [25]. The snails also feed on decaying plants [36]. Also when these plants provide shade, the vegetation moderates the water temperature. The presence of aquatic vegetation is positively linked to the amount of dissolved oxygen and the consumption of carbon dioxide (CO2) and thereby linked to movement and reproduction of pulmonate snails [37].
The relationship between the slope and schistosomiasis has been very rarely reported. In a study by [38], it was discovered that slope played an important role in the distribution of Oncomelania hupensis, the unique intermediate host snail of Schistosoma japonicum in the mouintaneous regions of China.
The location of a household in relation to the suitability of a water body to transmit the disease has shown to be highly relevant with respect to the level of prevalence [39]. A study in Ghana found that high infection levels were clustered around ponds known to contain intermediate host snails of S. haematobium, whilst prevalence was low in households in close proximity to a non infested river [40]. A study by [41] concluded that the relative location of a house to snail-free or snail-colonised water sources was a key factor explaining the spatial pattern of S. mansoni infection in Brazil. These risk factors were therefore selected, reclassified and standardised into four levels of risk namely, very low, low, moderate and high risk.
The datasets used in this research were obtained from different sources based on availability “Table 1”. The thermal band of the Landsat Operational Land Imager (OLI) sensor was used to derive Land surface temperature (LST) over the study area [19]. The digital numbers of band 11 of Landsat OLI was first converted to spectral radiance. Spectral radiance values were then converted to radiant surface temperature. The calculated radiant surface temperatures were subsequently corrected for emissivity and converted into Celsius. The monthly rainfall data was obtained from the European Meteorology Research Programme (www.euramet.org). The Digital Elevation Model (DEM) of the Advanced Space borne Thermal Emission Radiometer (ASTER) was obtained from the National Aeronautical and Space Agency website (www.nasa.gov), the remotely sensed data was captured with a spatial resolution of 30 metres. The Land use data was acquired from the global land-cover website of the University of Maryland, USA (www.landcover.org). NDVI was generated using the near infra red band and the red Band of the Landsat i.e. band 4 and band 3 respectively using the NDVI equation. The equation is explained by subtracting plants reflectance of red light from near–infrared light then dividing the difference by the addition of the red and near infrared light reflected. The slope image was obtained from the Digital Elevation Model and was converted using the spatial analyst tool in the ArcGIS software. Drainages (rivers) of the study areas were obtained from high resolution imagery and buffered from 0m to 2000m using the buffering tool in ArcGIS.
All the prepared raster data layers (criteria) were set to local coordinate system of WGS 1984 at a spatial resolution of 30m. The suitability levels for each of the criterion layers were defined based on literatures, experts’ knowledge and author’s practical experiences [24, 25, 26, 29, 32, 34, 35, 40, 42] except for slope that was based on arbitrary values. The layers were reclassified into different suitability level using reclassify tool in ArcGIS 10.1 as a base to construct the criteria maps.
This was done through methodology multicriteria evaluation (MCE) using Saaty’s analytical hierarchy process (AHP). AHP is a multi-criteria decision method that uses hierarchical structures to represent a problem and makes decisions based on priority scales. In this research AHP was used to obtain the mapping weight or importance of each individual schistosomiasis risk factor. The process of deriving the weights of each factor involved the following steps:
The risk factors do not have the same role and weight in the modelling of the final schistosomiasis risk zones. In order to designate the importance of each parameter, they were weighted using a pair wise comparison method which is one of the components of AHP. To assist in the weighting process of the pair wise matrix, the Saaty’s pair wise comparison table “Table 2” was used in the research.
After computing the pair wise matrix, a measure of consistency (CI) was used to check if the matrix was derived at an acceptable level of consistency using the formula below:
Cl=ƛmax-n/n-1
(1)
Where n is the dimension of comparison matrix, λmax is the maximum eigenvalue of the comparison matrix. The consistency of a pairwise matrix is interpreted as shown in “Table 3”.
The weight values of selected parameters calculated in Analytic Hierarchy Process were used in Weighted Overlay Analysis in ArcGIS to generate the schistosomiasis risk zones.
Buildings were extracted from high resolution imagery for the study through vectorization. The buildings were geocoded i.e. assigned attributes and one out of every ten buildings was selected for urine collection. The map of the study area was printed for field workers and urine was collected from willing participants in each geocoded building.
House to house sample collection was preceded by an interactive meeting of the researcher and village heads during which the purpose of the survey was explained. Each participant was given a clean 50cm3 wide-mouthed, screw-capped specimen bottles to supply terminal urine produced between 10:00am and 2:00pm. Each bottle was labeled to correspond to the number of the person on a pre-designed epidemiological form for sampled buildings. The samples were preserved on collection by adding 5ml of 10% formalin at the point of collection and carried to the laboratory. In the laboratory, each sample bottle was agitated to suspend the ova evenly in urine after which 10ml of the urine was transferred with a sterile disposable syringe to a centrifuge tube and centrifuged for 5 minutes at 1500rpm. The supernatant was discarded and the sediment was transferred onto a microscope slide. A drop of Lugol’s iodine was added and neatly covered with a cover slip. The slide was examined under the microscope for eggs of S. haematobium. When present, the individual was classified as positive for schistosomiasis.
Chi-square statistical analysis was used to test for significant difference between communities and prevalence of infection.
The reclassification and standardisation of the risk factors is shown in “Table 4”. The suitability levels was represented with intergers and ranked from Very Low to Very high.
The schistosomiasis transmission risk associated with each of the factors that were considered in this research is shown in “Figs 2–8”. “Table 5“ shows the result of the comparison matrix of schistosomiasis risk factors used in the study.
The spatial model used to produce schistosomiasis risk zones from risk factors is (Temperature*33.5) + (Rianfall*31.0) + (Elevation*8.2) + (Land use*9.8) + (NDVI*6.7) + (Slope*2.1) + (Proximity*8.6).
The map resulting from this weighted sum overlay for schistosomiaisis risk zones is shown in “Fig 9”. The final risk map is based on the best outcome in context of the frequency and transmission of schistosomiasis and classifies the study area into four suitability risk classes. About 110.15km2 (15%) of the total study area is subjected to very high schistosomiasis risk, 432.19 km2 (57%) labeled as high schistosomiasis risk, and the remaining 98.17 km2 (26%) and 7.52 (2%) areas have low and very low schsitosomiasis risk level respectively. Thus, it is possible to conclude that more than 82% of the area with 17,541 buildings is under high risk of schistosomiasis “Table 6”. The map also shows that the northern parts of the region have the lowest risk of schistomiasis while the risk increases as one moves to the southern parts.
Out of the 1,574 individuals examined in this study, 324 (20.60%) were infected. Infection by the parasite was recorded in all the communities surveyed in this LGA. The highest prevalence of 34.80% was recorded in Ita Ogbolu community while Oba Ile community had the least infection rate of 4.60%. Prevalence of infection differed significantly between these communities (P < 0.05) “Table 7”. This observation is important in future studies on schistosomiasis aimed at studying schistosomiasis transmission and risk in new settlements.
The schistosomiasis risk map produced by the study is important as it was generated at a micro scale to show variations in schistosomiasis transmission which may not be noticeable in country wide risk maps. It is also based on environmental conditions, geomorphologic data as well as human induced variables such as landuse.
The existence of suitable climatic conditions, lower elevation, proximities to river, vegetation, high temperatures and rainfall plays a great role for the spread of schistosomiasis in the study area. During the course of the study, it was determined that temperature, elevation and NDVI had the strongest link with the risk of schistosomiasis spread. This corroborate previous researches done in Nigeria and in other parts of the world with the conclusion that environmental factors govern the transmission dynamics of schistosomiasis. In the present study, the high risk areas corresponds to areas with temperature >26°C while low risk of infection were found in areas with temperatures <200c. The optimal temperature for snail development and survival is around 25°C [43]. The model also suggests that rainfall cannot be used to predict schistosomiasis transmission in the study area. According to [44], the spatial relationship between rainfall and snail population dynamics and infection transmission is difficult to measure since the effect of rainfall varies depending on species of snail and the geographical location. The strong correlation between the risk of schistosomiasis and elevation agrees with investigations in Tanzania and Egypt where altitude was recognized as an important environmental factor in the prevalence of urinary schistosomiasis [15,45]. Suitability was unaffected by slope in this study even though studies on snails host vectors of schistosomiasis showed that snails prefer a slope of less than 20% [46]. The Normalized Difference Vegetation Index (NDVI) is a major environmental factor that can be used in predicting schistosomiasis in the area. In Tanzania and Egypt, NDVI was reported to be a significant environmental variable in schistosomiasis prediction [15,45]. The comparison between risk and proximity to water body in the study under review shows a strong correlation. The high risk zones were located at close proximity to drainages in the study areas. The increase in prevalence with increasing closeness to water might be due to a corresponding increase in water contact activities for domestic, economic or recreational purposes.This has also been observed in study on schools located near Lake Victoria, Kenya where a positive association existed between prevalence of intestinal schistosomiasis and proximity to the lake shores [19].
The field survey records of schistosomiasis prevalence confirm a similar pattern as the spatial epidemiology of schistosomiasis in the region “Table 7”. The results of the Chi square analysis shows that there is significant difference between the communities and prevalence of schistosomiasis.
This research confirmed that satellite derived environmental data can be used to determine schistosomiasis risk levels in an area using relevant techniques. Predictive risk maps can be generated for different areas with different climatic conditions and combined to form country- wide or regional maps. In this study GIS played a good role in the generation of maps for environmental factors, reclassification, overlaying and identification of risk level.
The model has provided baseline data for schistosomiasis in the study area. It has also highlighted areas of high transmission of schistosomiasis. This is useful in planning and monitoring of control interventions. The results from the field study compared fairly well with the model. The model can be greatly improved by including other significant variables such as snail density in the analysis. The model can also be used to assess control interventions and treatment programs.
|
10.1371/journal.pgen.1002453 | Inference of Population Structure using Dense Haplotype Data | The advent of genome-wide dense variation data provides an opportunity to investigate ancestry in unprecedented detail, but presents new statistical challenges. We propose a novel inference framework that aims to efficiently capture information on population structure provided by patterns of haplotype similarity. Each individual in a sample is considered in turn as a recipient, whose chromosomes are reconstructed using chunks of DNA donated by the other individuals. Results of this “chromosome painting” can be summarized as a “coancestry matrix,” which directly reveals key information about ancestral relationships among individuals. If markers are viewed as independent, we show that this matrix almost completely captures the information used by both standard Principal Components Analysis (PCA) and model-based approaches such as STRUCTURE in a unified manner. Furthermore, when markers are in linkage disequilibrium, the matrix combines information across successive markers to increase the ability to discern fine-scale population structure using PCA. In parallel, we have developed an efficient model-based approach to identify discrete populations using this matrix, which offers advantages over PCA in terms of interpretability and over existing clustering algorithms in terms of speed, number of separable populations, and sensitivity to subtle population structure. We analyse Human Genome Diversity Panel data for 938 individuals and 641,000 markers, and we identify 226 populations reflecting differences on continental, regional, local, and family scales. We present multiple lines of evidence that, while many methods capture similar information among strongly differentiated groups, more subtle population structure in human populations is consistently present at a much finer level than currently available geographic labels and is only captured by the haplotype-based approach. The software used for this article, ChromoPainter and fineSTRUCTURE, is available from http://www.paintmychromosomes.com/.
| The first step in almost every genetic analysis is to establish how sample members are related to each other. High relatedness between individuals can arise if they share a small number of recent ancestors, e.g. if they are distant cousins or a larger number of more distant ones, e.g. if their ancestors come from the same region. The most popular methods for investigating these relationships analyse successive markers independently, simply adding the information they provide. This works well for studies involving hundreds of markers scattered around the genome but is less appropriate now that entire genomes can be sequenced. We describe a “chromosome painting” approach to characterising shared ancestry that takes into account the fact that DNA is transmitted from generation to generation as a linear molecule in chromosomes. We show that the approach increases resolution relative to previous techniques, allowing differences in ancestry profiles among individuals to be resolved at the finest scales yet. We provide mathematical, statistical, and graphical machinery to exploit this new information and to characterize relationships at continental, regional, local, and family scales.
| Technologies such as high density genotyping arrays and next generation resequencing have recently facilitated the production of an enormous quantity of data with which to investigate genetic relationships in humans and in other organisms. These data have the potential to provide a new level of insight into patterns of dispersal and mating, and recent and ancient historical events. However there are challenges, in terms of computational burden and statistical modelling, that are yet to be fully addressed. Two of the most popular approaches to investigate population structure using genetic data are exemplified by principal components analysis (PCA) [1], which is often regarded as a non-parametric approach, and STRUCTURE [2], based on explicitly modelling population structure. It is common to apply both approaches to the same dataset, in order to provide a useful summary of the basic features of the data. The PCA approach is based on analysing a matrix (which can be defined in several different ways, e.g. [3]–[5]) whose entries quantify the genetic similarity between pairs of individuals. The principal components (PCs) of this matrix thus represent directions in sample space that maximally explain the observed pattern of genetic similarity. Visualisation of key patterns of structure in the data can be achieved by plotting successive PCs: clusters of individuals can be interpreted as genetic populations, while admixture of two populations results in sets of individuals lying along a line [6], although other historical events can also produce identical PC signals [4] and other issues can also complicate the interpretation of PCs [4], [7].
Model-based methods attempt to more directly reconstruct historical events. In the simplest version of the STRUCTURE approach [2], individuals are assumed to come from one of discrete populations. Population membership and allele frequencies in each population are jointly estimated from the data via a Bayesian modelling framework. A group of very widely used (e.g. [8]–[10]) current approaches powerfully extend this model by allowing individuals be admixed, i.e. to have ancestry from more than one population (e.g. [2], [11]–[17]). Individuals are assigned ancestry vectors, representing the proportion of their ancestry that comes from each of the populations. Although powerful, these approaches have drawbacks – determination of is difficult despite some technical advances [18], [19], and typically is required for satisfactory convergence, due to issues of computational cost and the presence of distinct local optima, affecting even the fastest methods such as ADMIXTURE [15]. Further, little information is provided about the relationships between inferred populations, though observing how results change with varying can aid insight.
The central issue that we address in this work is the fact that both PCA, and the most popular STRUCTURE-like approaches analyse single mutations individually, and do not use information about the relative positions of these mutations in the genome. However the advent of high-density variation data, together with both computational [20]–[22] and experimental [23], [24] advances in techniques for haplotype phasing offer new opportunities for researchers investigating ancestry, due to the possibility of exploiting correlated variation patterns, at sets of closely positioned markers. Markers on the same chromosome are inherited together unless separated by recombination. At a population level, this results in linkage disequilibrium (LD) between close markers that reflects a shared history of descent, invalidating the independence assumption. Haplotype based analysis has the potential to harness this information [25]–[31], but there is as yet no accepted paradigm for how to utilise shared haplotypes to infer population structure. Methods to explore admixture have been developed that aim to be robust to the presence of LD [14], [32], [33], or directly model LD patterns [34] to identify ancestry segments. However, the latter model-based approach requires representative individuals from the admixing populations to be specified in advance, so does not represent a framework for identifying population structure.
Here we develop and apply both non-model and model based approaches, analogous to the PCA and STRUCTURE approaches described above, that aim to use much of the information present in haplotype structure. Both approaches are based on analysing the same matrix, which we call the coancestry matrix. Although our main aim is to introduce a framework to exploit LD information where present, our methods can also treat markers independently as a limiting case. We show theoretically and in practice that in this setting, the coancestry matrix approximately contains all the information used by both PCA, and the model-based STRUCTURE-like approaches, unifying these apparently different approaches. Moreover, we show in some settings our model based approach can be more sensitive than either STRUCTURE or ADMIXTURE, and is able to reliably infer over 100 populations simultaneously. When dense marker sets are available, our haplotype-based algorithm performs substantially and uniformly better than all methods treating markers independently. We illustrate our approach using the Human Genome Diversity Panel (HGDP) dataset, comprising over 600,000 markers typed on 938 individuals. Worldwide, we show that the use of haplotype information improves separation of groups, and reveals differences in genetic ancestry even among individuals coming from the same labelled population, and not detectable by the non-LD-based equivalent approaches.
Our approach attempts to capture the most relevant genealogical information about ancestry in compact form. We construct and motivate the approach using an example (Figure 1). At each locus within a chromosome, the sample history can be represented by a genealogical tree (Figure 1A), whose structure changes along the genome reflecting ancestral recombination events. First considering a single haplotype, the tree relationship to the other haplotypes is fully represented by the most-recent common ancestor (MRCA) time with each. For every individual haplotype, at each locus there exists one or more closest relative(s), which we denote their “nearest neighbour” haplotype(s) in the sample. Conceptually, we can view our haplotype as the ‘recipient’ of genetic material from a nearest neighbour ‘donor’ haplotype, who donates a contiguous DNA segment, bounded by recombination sites altering the ancestral relationship between the haplotypes (Figure 1B–1C), and thus beginning new segments, from a different ‘donor’. From the point of view of our haplotype, the chain of nearest neighbours along the genome corresponds to the most recent genealogical events, and so we assume it captures most of the information on their current population structure that would be provided by the complete genealogy at the locus. Further, we also assume that different nearest neighbour segments (which correspond to distinct coalescence events in regions unbroken by recombination) provide reasonably independent information on the ancestry of the individual. Finally, we aim to capture information on the joint structure of the entire dataset by constructing donor-recipient relationships for every haplotype, in the same way.
Because the set of genealogies consistent with a given dataset is complex to describe, and typically huge, approximate methods are required in order to make inference computationally practical [21], [22]. We use one such method, the Hidden Markov Model (HMM) introduced by Li and Stephens [35], which explicitly reconstructs the chromosome of a ‘recipient’ individual as a series of chunks from the other ‘donor’ individuals in the sample, using information on the types of the recipient, and potential donors, at each mutation. We assume our dataset consists of biallelic markers. We do not order the haplotypes in the same manner as the ‘Product of Approximate Conditionals’ likelihood used by Li and Stephens. Instead, we use an approach in which a single haplotype within an individual is reconstructed using the haplotypes from all other individuals in the sample as potential donors. This process is repeated for every haplotype in turn, so every individual is ultimately reconstructed in terms of all the other individuals. We interpret the donor of each chunk as representing a nearest neighbour of the recipient haplotype for that stretch, with each chunk representing a different nearest neighbour relationship. In the simulated setting shown in Figure 1A, haplotype 1 actually shares a common ancestor 80 generations ago with haplotype 5 (orange) from positions 0 to 284, and 150 generations ago with haplotype 4 (pink) from positions 421 to 750. In between, there is a stretch where there are multiple nearest neighbour haplotypes (Figure 1C), with the shared ancestor further back in the past. Figure 1D shows three sample reconstructions - or ‘paintings’ of the haplotype, produced by the Li and Stephens algorithm. The algorithm recovers the true genealogical relationships reasonably well, with some uncertainty about boundary regions, and with regions with multiple nearest neighbour relationships showing sampling variability. In addition to producing specific realizations of the painting process, the powerful toolkit associated with HMMs makes it possible to calculate expectations of which haplotype acts as donor to haplotype 1 as a function of position, over an infinite number of such paintings (Figure 1E). Figure 1F shows the expected number of chunks inferred from each donor to haplotype , given the data. Extending this across all individuals, the matrix formed by all recipient rows is called the ‘coancestry’ matrix, and is summed over chromosomes. This matrix forms the basis of our inference procedure, motivated by our assumption that chunks provide independent information about ancestry. Intuitively, the coancestry matrix counts the number of recombination events leading to individual being most closely related to , so gives a natural measure of ancestry sharing. We note that the expected lengths of the chunks donated by donor to haplotype , , and the number of mutations in donated chunks, may provide additional information in principle, but we do not exploit this here. To implement this approach in practice, we require previously phased (e.g. [21]) haplotype data from individuals at a defined set of loci, and (optionally) a previously estimated genetic map of the recombination distance between these loci. The Li and Stephens model has two scaling parameters, the recombination rate and the mutation rate , which we set to be the same for all individuals in the dataset. When analysing markers and using LD information, we estimate using the Expectation Maximisation (EM) algorithm [36]. Following [35], is fixed to Watterson's estimate although the parameter can also be estimated directly from the data using EM. Full details of the algorithm, which is available for download as part of the ChromoPainter package, are provided in Text S1.
One important special case is when markers are widely enough spaced as to be effectively unlinked, i.e. the recombination rate between any pair of markers is infinite. It is straightforward to produce our coancestry matrix in this setting by setting the recombination rate to infinity (full details in Text S1). In this setting, chunks will automatically consist of only a single marker, and thus markers are essentially independent. By painting a single biallelic marker, all potential donor haplotypes carrying the same type as the recipient individual are equally likely to actually be chosen as donors, while potential donor haplotypes carrying the other type will be very unlikely to be donors. If we additionally exclude SNPs that vary in only a single individual, which provide no information in our framework, then this ‘unlinked’ coancestry matrix can be trivially calculated analytically for any given value of . This is a symmetric matrix, and it is advantageously not necessary to obtain haplotypic phase (see Text S4). We therefore implement this as a special case in practice, setting . Importantly, the unlinked coancestry matrix can be calculated for any dataset, even in the case where markers in fact are in LD, in which case we view it as summarising available ancestry information, without utilising LD information. As we will explain below, this interpretation is justifiable, by considering the standard PCA and model-based approaches to analyse structure.
We developed and implemented an approach to perform principal components analysis (PCA), by eigenanalysis of a normalised version of our coancestry matrix (Text S4). Our method can be thought of as a natural extension of the approach of Price et al. [5] to a setting where information is available on the relationships between densely typed markers. Specifically, we show (Text S4, Proposition 1) that as , our coancestry matrix reduces to the symmetric unlinked coancestry matrix described above, is approximately proportional to that used for the Eigenstrat PCA decomposition, and that our approach yields PCs corresponding to those calculated under the Eigenstrat PCA decomposition [5]. Thus, the Eigenstrat method corresponds approximately to a special case of our approach. In the results section, we demonstrate that in practice both methods indeed give almost identical principal components for . Where we analyse data as linked (), we simply apply an identical approach to the unlinked case, and in this case the identified PCs account for LD patterns, so differ.
As stated above, our coancestry matrix estimates the fraction of chunks in the genome that individual 's lineage coalesces with one of the two lineages from (diploid) individual before that of any other individual. Intuitively, if individual and individual are in the same population, or related populations, they are expected to share more recent common ancestors in this manner than are pairs of individuals from historically separated groups, so is expected to be relatively large. Even if individual is only partially admixed with a group closely related to that which belongs to, we expect an inflation, albeit of smaller magnitude. Thus, the coancestry matrix is expected to contain rich information about population relationships. In developing a model-based approach, we have not yet implemented a model directly incorporating admixture, but concentrate on a clustering model (but where we can infer the number of clusters , deal with a very large number of potential clusters, explore relationships between groups, and quantify ancestry sources in each group). The aim of such a model is to partition the dataset into groups with indistinguishable genetic ancestry, which we interpret as individual populations. We utilise a Bayesian approach, employing reversible-jump MCMC.
To formalise this idea, we consider populations characterized by a donor matrix , which can be thought of as a population-level coancestry matrix and gives the underlying proportion of chunks from any individual in population that come from population . A population is a group of individuals where: (i) all individuals within the group are equally related, so receive the same underlying fraction of their chunks from each of the other members of the group, (ii) all individuals within the group share identical relationships with any other population , so receive the same fraction of their chunks from each member of any other population , and so (iii) all individuals within the group donate the same fraction, , of the chunks found in any member of population . Thus, a chunk from any recipient individual within population has an identical donor distribution, and an identical recipient distribution, across the sample.
Our model is now defined by our earlier stated assumption that donated chunks within an individual are independent, and no additional information is carried in their size (which for example determines the number of chunks in the genome). For individuals , in populations and respectively, the likelihood a single chunk is donated to individual from is where if , and when , (because individuals cannot donate to themselves). Since chunks are independent, we may simply multiply the likelihood across chunks. Thus, if there are chunks in total donated from individual to individual , the overall likelihood for individual is . At this point, we make an approximation to the likelihood, which we partially justify later. Specifically, we replace the observed number of chunks with the expected number of chunks given by the coancestry matrix, which although not an integer still allows a well-defined likelihood. We treat chunks in different individuals as independent, so multiply across individuals to give a complete likelihood:(1)Note that this likelihood depends on the data only through the terms of the coancestry matrix, which we later show are approximately sufficient statistics for our inference which aids computational efficiency. In this likelihood, we have divided the chunk counts by a value in order to account for a) non-independence of chunks in practice, and b) our substitution of the expected for the observed number of chunks copied. can be thought of as defining an ‘effective number of independent chunks’, which can be either less than, or greater than, the true average number of chunks - we discuss calculation of later.
In our Bayesian approach, we must model the number and distribution of the underlying populations via a prior for . Given sufficient data, the choice of prior should only weakly affect the results (as discussed in Results, we believe this is an important strength of our approach). We choose a Dirichlet prior where , which is conjugate to the multinomial likelihood in Equation 1. The values are proportional to the a-priori expected value of each , and scaling the vector by a value decreases the variance of all elements of by a factor . From the genealogical process, we would expect excess donor/recipient relationships within a group, i.e. that is larger than with . From these elements we construct the prior as the product of three elements: a shared variance term (analogous to the correlated allele frequency of Falush et al. [12]), a within population increase and an otherwise uniform distribution of the chunks donated by population in total. Specifically,(2)The factors and are adjustments for the fact that individuals do not act as donors to themselves. We wish to infer the parameters and and therefore place on them a broad hyperprior based on Gamma distributions. Finally, the assignment of individuals to populations is given a Dirichlet Process Prior, which is weakly informative and allows for direct estimation of the number of populations . Further details are provided in Text S2.
We have implemented our approach as a software package we refer to as fineSTRUCTURE. Because we have chosen the prior of as conjugate to the likelihood in Equation 1, these population specific parameters can be integrated out analytically. The posterior probability of a population configuration, which we call a partition, is conditional on only global parameters (derived in Text S2). The target of inference is these hyper-parameters ( and ) but primarily the population assignment . This we represent in an unordered form as a list of co-assignments, avoiding the problem of associating labels with populations. Inference for is performed using a Markov chain Monte Carlo (MCMC) algorithm closely related to that of Pella and Masuda [19] and also that implemented in the program STRUCTURAMA [37]. The space of possible partitions is explored using an algorithm which proposes new partitions that are modified versions of the previous one (see Text S3). Specifically, the partition is modified by merging or splitting populations, merging then resplitting, or moving individuals. The proposed partition is accepted, meaning that it replaces the previous one, with a probability that depends on the ratio of the likelihood with the previous partition. and are updated within the algorithm using standard Metropolis-Hastings MCMC updates. In common with other MCMC algorithms, ours is run for a so-called burnin, after which the parameters are periodically recorded. If the algorithm is burned in and run sufficiently long, then the parameter samples converge to the posterior distribution (see e.g. [38]) of the parameters given the data, with variation found between samples reflecting posterior statistical uncertainty of parameter estimates. We test for convergence to the posterior by considering the pairwise assignment of population membership for two runs initialised with different random seeds. If the algorithm is converged then the frequency of coassignment should differ only due to Monte-Carlo error between runs.
The statistical model that we have derived has a likelihood depending on the terms of the coancestry matrix, which are rescaled by dividing each by a factor (see above). The factor can increase or decrease depending on many factors.
Different chunks will not in practice be fully independent of each other, tending to decrease the ‘effective’ number of chunks and therefore increase . A first reason is that if individuals and share a distinctive haplotype tract, then they will both be counted as donors for each other and the same chunk will appear twice in the likelihood, once in and the second in . Secondly, adjacent chunks inferred on the same haplotype may not be fully independent of each other due to limitations of the Li and Stephens algorithm in modelling recombining genealogies [39] and to the non-Markovian nature of genealogical relationships themselves [40]. Thirdly, inaccuracies in the data such as phasing errors may create misleading chunk boundaries.
Conversely, by averaging over chunk assignment uncertainty in the painting step we smooth the chunk count distribution for each individual, decreasing by reducing variability in chunk numbers relative to random draws. The effect is particularly large where there is a great deal of uncertainty about chunk assignment, as is the case for weakly linked or unlinked markers. In Text S4, we show that for the special case of unlinked markers (or more generally when we use the unlinked coancestry matrix for inference), appropriate choice of results in our likelihood being asymptotically (in large datasets not dominated by rare markers) equivalent to that of STRUCTURE, provided population structure is not too strong. See Figures S4 and S5 and Text S6 for how strong structure with truly unlinked loci affects our inference. This validates (for moderate structure) the idea of using a multinomial-form likelihood for the coancestry matrix. Further, we show analytically that the correct value of is in the unlinked case.
Although we have not been able to derive such a formula for linked data, we can estimate empirically. Specifically, we calculate the variance of contributions to the coancestry matrix from non-overlapping chromosomal regions that are large enough that the chunk counts in each will be approximately independent. We choose to match the mean observed variance of these contributions to that predicted by the (rescaled) multinomial model using the average number of chunks in the region. The principle of this approach is to achieve a multinomial likelihood matching the statistical uncertainty in the real coancestry matrix terms. In the case of truly unlinked data, this approach will approximately return the theoretically correct value, . In both this and the linked setting, using extensive empirical validation we find that across a range of settings, our estimation procedure finds a conservative, close to optimal estimate for (Text S6).
Our estimation of is similar in approach to the block jackknife of SmartPCA [41] though differs in many particulars, and in interpretation given we can observe or in practice. Our interpretation of is as an effective number of independent chunks donated from to .
One helpful property of this approach is that by attempting to correct for the true underlying variance of the , modelling deficiencies are at least partially corrected. In particular, we observe in the Results that treating markers as unlinked, by using the unlinked version of our coancestry matrix (), results in robust inference in both simulated and real data – even where strong association between markers in fact exists. This allows us to perform comparisons of the two approaches where we use, and do not use, LD information, on the resolution of fine-scale population structure.
Since the fineSTRUCTURE algorithm can identify fine subdivisions, it is often important in practice to have some indication of historical relationships amongst the inferred populations. We have found that performing inference under the full model using successively reducing values of (as is commonly done in ADMIXTURE and related algorithms) does not always perform well in this setting, e.g. by splitting off highly drifted groups. Instead, we recommend an approach that performs inference at the ‘natural’ (i.e. inferred) value of , and then generates a tree of relationships amongst these populations. We start with the maximum a posteriori (MAP) state, found by taking the MCMC iteration with the highest observed posterior likelihood and then performing a number of additional hill-climbing moves to identify any merges or splits that further improve the posterior probability. Starting from this ‘best’ partition, we successively merge populations, choosing the merge giving the highest probability for the merged group at each step, resulting in a bifurcating tree relating each of the populations together. One of the biggest discriminators between populations is within-population counts, which largely reflect genetic drift occurring after a split from other groups, and are thus uninformative in choosing among group merges. In order to allow populations that contain related individuals (i.e. with high ) to be merged more easily, during the tree creation we replace the count matrix with a modified count matrix with diagonal ‘flattened’ to be the next highest value in the row, where and . Although this ad hoc approach provides a key advantage over inference at specific for locating functional population units, we emphasize that this tree is not based on any model of population differentiation. Results may depend significantly on sample size, and so should be treated as an approximate guide to similarity, rather than a full population history. Despite these caveats, the tree empirically performs well in capturing relationships at multiple cases when the data is approximately hierarchical.
We introduce a new approach, described in detail in Models and Methods and Texts S1, S2, S3, to analyse population structure, designed for application to large datasets, particularly where markers are in strong LD but also in other settings. To summarise, given a dataset of individuals, we construct an matrix , which we term the coancestry matrix, and which forms the basis of all our inference. The element estimates the number of discrete ‘segments’ of the genome of individual that are most closely related to the corresponding part of the genome of individual . This matrix is most powerful when constructed so as to use joint information provided by tightly linked markers that are in LD. However, we can also construct an ‘unlinked coancestry matrix’ corresponding to ignoring this information, which is the correct approach if markers are widely spread across a genome. Results from using the unlinked matrix can be used to compare our approach to existing methods, and to quantify gains in information from taking into account LD information in measuring coancestry.
Given the linked or unlinked coancestry matrix, we have described how this can be used to learn about population structure: firstly, by performing PCA, and secondly, by using a model-based analysis to identify clusters of individuals with similar historical ancestry, corresponding to genetically related populations. In this section, we extensively evaluate properties of our approach in theory and using simulated data, and perform a new analysis of the HGDP dataset. We also explain how in conjunction with the clustering algorithm, analysis of the coancestry matrix reveals both differences, and details of historical interactions, among human populations in unprecedented detail.
To understand the properties and performance of our approach in the simplest possible setting, we begin by analysing the case where markers are treated as unlinked, i.e. our unlinked coancestry matrix. In this setting, markers may be truly unlinked, or there may be LD information being ignored. We began by analysing datasets simulated under a setting where there was no underlying population structure, both with and without tight linkage between markers (Text S6). In this setting, PCA will not give meaningful results, but encouragingly, our model-based procedure, which includes a step to estimate the effective number of chunks in the genome, correctly identified populations (Figure S1). This demonstrates our approach is robust, but we must do more to establish its power to detect structure compared to previously developed methods, and the total information present in the data. First considering the problem mathematically, we related our unlinked coancestry matrix to the matrix used in a standard PCA approach, Eigenstrat [5]. This revealed that even though it has a rather different construction and motivation (based on the Li and Stephens algorithm [35]), our matrix is simply a linearly scaled version of the Eigenstrat matrix (Text S4, Proposition 1), implying our PCA approach in this setting ought to perform almost identically to Eigenstrat, and our coancestry matrix captures the same information as standard PCA.
To compare the PCA approaches in practice, we constructed a simulated dataset designed to represent realistic levels of subtle population structure. We simulated data for 100 individuals according to a model containing 5 populations related in a tree-like manner with three major historical splits forming populations A, B and C two of which subsequently split (Figure 2A–2B). We used this scenario for all simulated-data comparisons, and simulated data with LD between markers. We used forward simulation of up to 200 genetic regions each 5 Mb in size, using the program SFS_CODE [42], with parameters chosen to approximate diversity found within and between European populations (see Text S5), and genetic maps based on real estimates for 10 sampled regions of the human genome [26]. We constructed the unlinked coancestry matrix for these data (which is shown for 150 regions in Figure 2C), and performed PCA both using this matrix, and using Eigenstrat on the raw data, yielding as expected almost indistinguishable results (Figure 2D–2E). These no-linkage approaches both show only incomplete separation of the most closely related pair of populations, B1 and B2; we consider the linked coancestry matrix later.
We next turn to our fineSTRUCTURE model-based analysis, again considering the unlinked coancestry matrix even though strong and variable LD exists in the dataset. We first compared performance of our unlinked model to the popular ADMIXTURE [15] software (Figure 3B and 3D, details in Text S8). Encouragingly, as the number of 5 Mb regions increased from 5 to 200 we saw a monotonic performance increase for the no-linkage model, separating all groups with 200 markers. Further, our approach outperformed ADMIXTURE, with the ADMIXTURE performance levelling at around 60% correlation with the truth. In practice, we observed ADMIXTURE successfully splitting groups A, B and C and mostly splitting C1 and C2, but not B1 and B2, as detailed in Figures S6, S7, S8, S9, S10, S11. ADMIXTURE performs inference under a model where markers are treated as unlinked, and where individuals may have genomes made up of mixtures of inferred source populations, while our simulation incorporated drift between populations, but not admixture. To examine whether violations of both these modelling assumptions explain the different results, we simulated a new dataset with the same underlying population structure of 5 populations as before, but no linkage (i.e. independence) between markers within each population. We analysed these data with STRUCTURE, which uses a similar underlying model to that of ADMIXTURE, but includes a no-admixture model (Text S7). For small datasets, STRUCTURE slightly improved performance relative to our unlinked fineSTRUCTURE model, but for larger SNP numbers, fineSTRUCTURE was able to identify all population splits () while again, STRUCTURE was able to split only populations A, B and C (). Thus, even when LD information is not used (or even present), fineSTRUCTURE can offer advantages in some settings over these existing approaches.
We sought to understand mathematically why our approach, based on only a summary of the original variation data – the unlinked coancestry matrix – equivalent to the matrix used for Eigenstrat's version of PCA, appears to perform so well relative to the earlier approaches, which carefully model each individual SNP marker (Text S4). This revealed that, surprisingly, the formulation of the likelihood of the data used by both STRUCTURE [2] and ADMIXTURE [15] can be viewed as approximately a function of only the terms in the coancestry/PCA matrix (under certain technical assumptions such as large datasets; Proposition 2). Under these assumptions, this result then unifies these apparently different approaches in terms of the underlying information they exploit (and suggests the PCA matrix of Eigenstrat is a particularly ‘good’ choice [43]). Furthermore, we also show that provided structure is weak (if strong, all methods are expected to find it), the multinomial likelihood used by fineSTRUCTURE is approximately the same as that used by STRUCTURE, with correct choice of the normalising parameter (Text S4, Proposition 4), and we find in practice that this ‘correct’ value of is well estimated by the jack-knife procedure described above (Figure S2). This means that at least for datasets with large numbers of loci, and ignoring linkage, we expect fineSTRUCTURE, PCA, and STRUCTURE/ADMIXTURE to all utilise similar information in the data.
What explains the different behaviour of the model-based approaches? We believe it is differences in prior models used. Both STRUCTURE and ADMIXTURE assume all underlying populations undergo separate genetic drift from some original founder group, and so this prior model penalises shared drift, for every individual marker, and so increasingly strongly as the number of loci increases. Our simulation framework (realistically, we believe), incorporates drift separate to each group, but also shared drift common to clusters of populations (caused for example, by being closer geographical neighbours). By using a more flexible prior model of structure, fineSTRUCTURE is able to separate populations C1 from C2, and B1 from B2, which the existing model-based approaches have difficulty separating even with sufficient data. By not assuming any particular form for the population-level coancestry matrix , closely related groups are allowed to share genetic material, as visualised in Figure 2C.
To examine improvements offered by utilising LD information, we used our linked coancestry matrix as the basis of new PCA and model-based analyses. The genetic maps used to simulated the sequence data were also used for inference in the linked model, though we note (not shown) that the conclusions still hold without this requirement. We estimated from the data by averaging estimates for 50 of the simulated regions. Using linkage information reduces the within-population variance of the coancestry matrix relative to the between-population variance (by a factor of nearly 3 in the data shown in Figure 2C) but does not change its qualitative structure. We performed PCA decomposition of the linked coancestry matrix (Figure 2F), yielding consistently tighter clustering of points, and in particular clear separation of populations B1 and B2 by the fourth principal component, compared to not using LD information (Figure 2D–2E).
In the model-based setting, linked fineSTRUCTURE strongly outperforms the unlinked version (Figure 3A), confirming the utility of LD-based inference, with only 75 regions required (Figure 3D) to correctly separate all 5 groups vs. 200 when ignoring linkage. Encouragingly, performance improves more dramatically for fewer regions, when structure is at the limits of detection. Examination of a particular case (Figure 3A–3B) with 150 regions shows only a partial separation using unlinked fineSTRUCTURE of the most similar groups B1 and B2, analogously to the PCA result. ADMIXTURE (Text S8) also fails to identify this population split. In practical applications, given a finite genome size, using linkage information will therefore be expected to allow clear identification of more subtle (‘fine’) structure than is detectable otherwise, as we show in the next section. Figure 3C shows the linked model coancestry matrix averaged over populations (using the model-based assignment of individuals to populations), as well as a tree (which is correct except that population A is not equidistant between B populations and C populations), inferred as described in Models and Methods. We view this coancestry matrix and the tree as the ‘outcome’ of our model-based inference procedure – it details groups found, their inferred relationship, but also shows the inferred extent of haplotype sharing between groups, showing for example groups that share closer genetic relationships. As we explain below, we believe that in practical applications, this representation can reveal interesting features of underlying structure.
We analysed the pattern of population structure in the Human Genome Diversity Project (HGDP) dataset [9] of 640,698 autosomal SNPs typed in 938 individuals sampled from 53 different labelled groups, with 5 to 46 sampled individuals per group. Complete inferred-phase haplotypes ([21], [44]) were downloaded from http://hgdp.uchicago.edu/. Estimated b36 recombination rates [26] were downloaded from the HapMap website (http://www.hapmap.org). Despite the size of the dataset, the fineSTRUCTURE algorithm (Text S10) converges in independent runs (Figure S25) to a solution with 149 populations in the most probable posterior state using the data calculated based on the linked model (Figure 4A). Our tree building algorithm aims to represent the relationships among the groups and in the tree, for which almost half (25 of 53) of the original labelled groups exactly correspond to a single clade in the tree, 9 corresponding exactly to a single inferred population. In other cases, geographically neighbouring groups (e.g. several groups sampled in Pakistan) are not separated, implying sample labels do not perfectly correspond to identifiable ancestry signals. Higher up the tree, branches correspond to large continental-level groups, similar to those seen before [45].
In general, many groups are not related through simple hierarchical ‘tree-like’ drift, but also through complex admixture events. These relationships are captured directly in our representation by the coancestry matrix. Although this is high-dimensional even after clustering individuals into groups, and in future we think it is important to incorporate admixture in our modelling framework, we nevertheless believe the very complex structure of the data itself means visual examination of the coancestry matrix provides important insights using linkage information. Previous analysis of the worldwide HGDP using ADMIXTURE, and to an extent PCA, has identified signals of admixture [9], [28], [45] in certain groups. In practice, the number of groups that these methods can infer is typically limited to or fewer, resulting in limited resolution in identifying the detail of such admixture events. In addition, both PCA and ADMIXTURE analyses do not consistently signal the extent of genetic drift in the dataset. Follow-up ‘regional’ analyses, for example focussing on Europe, partially address these issues for drift and admixture within such regions, but not across larger distances. The linked coancestry matrix allows simultaneous visualisation of drift, and admixture, and fine-scale resolution for both (Figure S14). For example (Figure 4B), previous observations [46] of both Central and East Asian ancestry in the Hazara (from Pakistan) can now be refined. The coancestry matrix demonstrates strong haplotype sharing of the Hazara from other Pakistani groups (e.g, the Pathan) as well as varying continuously in admixture fraction with groups from today's north-east Asia (e.g. the Mongola). This provides direct genetic evidence corroborating historical evidence [47] of ancestry sharing between the Hazara and the Mongols. The Burusho, another Pakistani group showing East Asian admixture, are separated from the Hazara by fineSTRUCTURE, but have relatively less North-East Asian DNA, implying distinct admixture histories for these two groups. Many other HGDP admixture signals could be analysed similarly.
Although fineSTRUCTURE performs well on the global dataset, for easier visualisation of results, we developed an approach analysing structure in only sub-regions of the data, but based on the same (worldwide) coancestry matrix as before. In practice, we found this had the second advantage of a small increase in resolution, while retaining the ability to identify many long-range population relationships. This increase in power is related to our prior model – we assume ancestry proportions are independent across groups, while in fact worldwide historical relationships among populations result in correlations in these vectors. Although the prior is overwhelmed by the data for clear splits (unlike that used by other approaches), our algorithm nevertheless can merge very similar groups. Within a subregion of the world, however, differences in ancestry proportions are much closer to independent, potentially improving precision.
For a regional analysis, we chose to split the dataset into eight regions, approximately corresponding to ‘sub-continents’, based only on the results of the merging algorithm used to produce the population tree (Figure 4A). Each geographic region is analysed individually by fineSTRUCTURE under the full model, with other regions considered only via donation of genetic material when pooled into seven overall counts, corresponding to the total received from each (the number of individuals is also used). This approach is a balance of retaining broad-scale information relating to admixture from external sources, while substantially reducing dimensionality. Figure S15 shows the tree for these results which is broadly similar to Figure 4A though differs in some particulars (for example Maya and Colombian are now split but BantuKenya are not) partly due to different ‘diagonal flattening’ restrictions across subcontinents. 226 populations are now found, many of which may simply be related individuals (e.g. within the Druze) whilst others reflect real but subtle population structure.
We focus on the European results as an example (Figure 5A), with other continents shown in Figures S16, S17, S18, S19, S20, S21, S22, S23, S24. Convergence in all cases was excellent (Figures S26, S27, S28, S29, S30, S31, S32, S33), despite significant uncertainty. The smaller scale of the problem here allows more detail of results to be discussed, but also meaningful comparison with other approaches. We identified populations with fineSTRUCTURE, identifying (and in some cases further splitting) the 8 labelled European groups precisely, apart from one French individual showing an ancestry pattern closer to the Tuscans in the dataset (and visually intermediate from both). Again, examination of the identified coancestry matrix parameters is helpful in revealing relationships among the groups, and with outside populations. For example a large coancestry value within some populations (along diagonal blocks in the coancestry matrix) can be interpreted as strong genetic drift, which appears in some groups (e.g. the island Orcadian and Sardinian populations) but is absent in the French. The multiple populations found for Orcadians, Sardinians and Tuscans, with particular subgroups having significantly elevated coancestry even within the same label, suggests more recent kinship perhaps related to geography (which we do not have additional information on). The Adygei (from the Caucasus) are split into three groups, which instead differ mainly in their levels of Russian admixture within Europe, and of Central and East Asian ancestry from outside. Similarly, Tuscans are split from a different North Italian group, due to a very subtle ‘drift’ signal along the diagonal, but mainly by having more African and Middle Eastern ancestry (corroborating results on mitochondrial DNA [48]). Similar signals are seen across other continents.
We applied ADMIXTURE to the same HGDP European data as analysed by fineSTRUCTURE (Text S9). Although the populations are very subtle and ADMIXTURE cross-validation implies (Figure S13), we still obtained meaningful results with (Figure 5B) and fewer (Figure S12) populations, but noise for higher . As expected for this powerful approach, ADMIXTURE gave useful information on European groups, with clear separation of Adygei, Russian and Basque for example and some, but not all, of the within-population splits represented. Based on this analysis, it is not possible to separate certain groups, e.g. the Tuscans and Italians, where inferred non-admixed and admixed individuals are spread among both groups, neither corresponding to sample labels nor supported by other analyses (including ADMIXTURE at different ), and thus results may reflect modelling uncertainty. More generally, the French, Italians/Tuscans and some Orcadians are closer to lying along an admixture continuum in this analysis, while appearing much more cleanly separated, and homogeneous in ancestry makeup, in the linked coancestry matrix (which has identifiable ‘blocks’ of colour for these groups). As expected from the earlier simulations, the differences with fineSTRUCTURE seem to be concentrated in the more subtle splits, and also in the fact that ADMIXTURE analysis cannot here easily benefit from information on outside genetic contributions, e.g. to distinguish a third Adygei group. Finally, for the subtle structure present here, care clearly must be taken in interpreting ADMIXTURE results – in each of the Orcadian, Italian/Tuscan and Sardinian groups, some individuals appear genetically mixed and others do not, while the coancestry matrix does not support such a genuinely distinct relationship.
In addition to using fineSTRUCTURE, we also used our linked (and unlinked) PCA approaches to analyse the data for Europe and other continents (Figures S34, S35, S36, S37, S38, S39, S40). Results in general were consistent with our simulations and with the model-based analysis, giving better separation of groups for the linked PCA version, e.g. clean separation of Italians and Tuscans only when LD information is utilised (Figure S38). Figure 6 illustrates this improvement for a subset of populations in central East Asia. Only the linked model shows clear separate clusters for Miao, She and Tujia, or any obvious separation of Tujia and Han. The latter group are revealed as lying along a line, much noisier in the unlinked case and suggesting variable levels of coancestry between Han individuals and other Chinese groups, presumed to have occurred during the North to South spread of the Han [49], and directly visible in the coancestry matrix (Figure S10).
The strongest advantage to using the linked model is in separating subtly different groups, and we see many cases in our data where labelled groups are split into smaller populations by fineSTRUCTURE, but although these show features consistent with their representing genuine ancestral differences, we do not have additional information, for example on geography to confirm these populations. We therefore devised a scheme to overcome our incomplete information, using the fact that although completely unlinked, two approximately equally sized halves ‘A’ and ‘B’ of an individual's genome automatically share all sampling details, and thus have the same underlying ancestry. Examining similarity in ancestral profiles for the two halves thus provides an indication of whether ancestry differences observed (from half the genome) are genuine, at the finest possible scale. Specifically, we analysed half of the individuals at a time (splitting the dataset approximately evenly for each label), painting their chromosomes using an identical donor set consisting of the other half of the sample, so chunk counts for individual ‘A’ or ‘B’ halves are comparable across individuals. For each individual ‘A’ half, we paired with the most correlated individual ‘B’ half, and recorded the fraction of times this ‘B’ half came from the same individual (Figure 7), and compared this to random chance when using population or label groupings. The results validate our populations as reflecting genuine ancestral differences, pairing halves within clusters more of the time than using labels alone. Interestingly, we paired up genomic halves within individuals consistently more often than predicted by than our clustering (and uniformly more often using linked than unlinked information) demonstrating that human population structure exists at finer scale than the clustering detects, and is most powerfully identified using linkage information.
Partial or complete barriers to mating create groups with distinct genetic ancestry, or, in the present terminology, populations. In our approach, we assume that chromosomes within a particular population have characteristic probabilities of sharing stretches of similar DNA from individuals in their own and in other populations, and view these probabilities as defining population composition and relationships. To infer groups, we first reduce data dimensionality by estimating the relationships among all pairs individuals using a “coancestry matrix”, which is central to our method and based on ‘painting’ the chromosomes of each individual [35]. Loci can be treated as linked or unlinked in the genome. In the unlinked case, we have shown that in theory and in practice, our model-based (MCMC) and PCA approaches are very closely related to the previous approaches exemplified by STRUCTURE and Eigenstrat [2], [5], and that the parametric and non-parametric approaches can all be thought of as, approximately, interpretations of information present in the coancestry matrix. This helps explain previous observations [50] that structure is frequently detectable using both types of approach, or neither. Other approaches to summarizing matrices, such as sparse value decomposition, might bring out additional features [43].
We have also shown that the linked approach substantively improves performance, where LD information is present among tightly packed markers, achieving a resolution in the HGDP that is to our knowledge unprecedented. Intuitively, we believe that the underlying reason is that using haplotype sharing identifies relationships among individuals in the recent past much more strongly than individual ancient SNP sharing, enabling more subtle, recent population structure to be captured [31]. This does not mean the approach is optimal – additional improvements may be found by utilising information (within our framework) on the size of shared haplotypes, mutations private to particular groups, and haplotypic sharing further back in time. We believe the advantages offered by exploiting haplotypic information will continue to grow as full sequence data becomes predominant [51].
In practical implementation, our approach uses two initial, parallelizable analyses: a phasing step, common in modern population genetic analyses, and a subsequent chromosome painting step, both run once on a given dataset, and feasible for datasets with millions of markers using computer clusters. Subsequent steps using the resulting coancestry matrix have computational time depending only on the number of individuals, which with our efficient algorithmic implementation enable us to, for example, analyse far larger numbers of populations – hundreds in the HGDP - than other approaches that reanalyse each mutation at each iteration. We observed a substantial performance improvement for the linked model, when applied to the HGDP data phased jointly using fastPHASE [21], despite inevitable errors in the haplotypes produced by all such phasing approaches. However, we caution against naively combining and analysing datasets phased separately, or by different approaches, which may introduce spurious differences in haplotype composition.
In the model-based approaches discussed here, we have described how the coancestry matrix captures key relationships among groups. However, future approaches may aid interpretation of results, and power, by explicitly modelling the processes of drift, and subsequent admixture, among identified populations and their effect on this matrix. The theory developed here for the unlinked case suggests a close connection between population level genetic drift and the coancestry matrix. Although this (like average pairwise coalescent times [4]) will not uniquely specify historical events, genetic drift specific to a population will have the effect of elevating the within-population coancestry value, while admixture causes a population to become more similar, both as a donor and recipient to the group it is admixing with. Relating identified groups in this manner and developing new ways of representing population structure are both needed, given both the very fine stratification (into 226 groups) achieved by the approach and the half-matching results demonstrating even more structure present in the data. Allowing individuals to show continuous variation in proportions of ancestry from multiple groups might capture this signal [2]. However, because we observe a drift signal private to most of our identified groups, we believe a necessary but difficult modelling challenge is to incorporate successive rounds of genetic drift, admixture, further genetic drift, and even familial relationships into such models.
Overall, our results demonstrate we have not yet reached the limits of the information available using genetic information, and particularly the precision with which ancestry sources can be determined. As full sequence data and larger sample sizes become increasingly available, we anticipate resolution will improve further beyond the level of countries, to regions within countries in many cases, and this will be of value in a range of settings. The methods described here can produce highly accurate clustering and sensible choices of the number of populations in humans and other species, and can be applied to full genome sequences for thousands of individuals.
The algorithms described in this article have been implemented in computer software packages ChromoPainter and fineSTRUCTURE, which are available at http://www.paintmychromosomes.com/.
|
10.1371/journal.pntd.0004861 | MLST and Whole-Genome-Based Population Analysis of Cryptococcus gattii VGIII Links Clinical, Veterinary and Environmental Strains, and Reveals Divergent Serotype Specific Sub-populations and Distant Ancestors | The emerging pathogen Cryptococcus gattii causes life-threatening disease in immunocompetent and immunocompromised hosts. Of the four major molecular types (VGI-VGIV), the molecular type VGIII has recently emerged as cause of disease in otherwise healthy individuals, prompting a need to investigate its population genetic structure to understand if there are potential genotype-dependent characteristics in its epidemiology, environmental niche(s), host range and clinical features of disease. Multilocus sequence typing (MLST) of 122 clinical, environmental and veterinary C. gattii VGIII isolates from Australia, Colombia, Guatemala, Mexico, New Zealand, Paraguay, USA and Venezuela, and whole genome sequencing (WGS) of 60 isolates representing all established MLST types identified four divergent sub-populations. The majority of the isolates belong to two main clades, corresponding either to serotype B or C, indicating an ongoing species evolution. Both major clades included clinical, environmental and veterinary isolates. The C. gattii VGIII population was genetically highly diverse, with minor differences between countries, isolation source, serotype and mating type. Little to no recombination was found between the two major groups, serotype B and C, at the whole and mitochondrial genome level. C. gattii VGIII is widespread in the Americas, with sporadic cases occurring elsewhere, WGS revealed Mexico and USA as a likely origin of the serotype B VGIII population and Colombia as a possible origin of the serotype C VGIII population. Serotype B isolates are more virulent than serotype C isolates in a murine model of infection, causing predominantly pulmonary cryptococcosis. No specific link between genotype and virulence was observed. Antifungal susceptibility testing against six antifungal drugs revealed that serotype B isolates are more susceptible to azoles than serotype C isolates, highlighting the importance of strain typing to guide effective treatment to improve the disease outcome.
| Cryptococcus gattii, which is classically divided into four major molecular types (VGI-VGIV), and two serotypes B and C, is the second most important cause of cryptococcosis. The rising incidence of human and animal cryptococcosis cases caused by molecular type VGIII highlights the need for increased vigilance. In this study, we characterized a large set of C. gattii VGIII isolates. Genetic analysis revealed four diverging sub-populations, which were primarily associated with serotype B or C, and very likely originated from endemic regions in Colombia, Mexico and the USA. Differences in virulence and antifungal susceptibility between serotypes may result in different disease outcomes since serotype B isolates were more virulent in mice than serotype C isolates, but serotype C isolates were less susceptible to azoles, the primary treatment for uncomplicated cryptococcosis. Identification of cryptococcal serotype and molecular type in clinical practice has the potential to guide treatment regimens and hence reduce morbidity and mortality in both sporadic cases and those associated with outbreaks. Our study significantly contributes to the understanding of the epidemiology, genetics and pathogenesis of Cryptococcus and cryptococcosis.
| The encapsulated basidiomycetous yeast Cryptococcus gattii is the second most important etiological agent of cryptococcosis, next to its sibling species C. neoformans. Both species can cause central nervous system and pulmonary manifestations [1]. However, C. gattii has a more limited geographical distribution and is recovered less frequently [2,3]. Initially considered prevalent only in tropical and subtropical regions, C. gattii emerged in a temperate climatic zone on Vancouver Island, British Columbia, Canada in 1999 and has since extended to the Pacific Northwest and other locations within the USA [4,5]. Despite the fact that the global burden of C. gattii may still be unrecognized [3], the increasing isolation of uncommon molecular types and extension of their geographic spread defines C. gattii as an emerging fungal pathogen.
C. gattii is comprised of two serotypes, B and C. Among them, four major molecular types, (VGI/AFLP4, VGII/AFLP6, VGIII/AFLP5 and VGIV/AFLP7) have been consistently recognized by various molecular techniques, including PCR-fingerprinting [6,7], restriction fragment length polymorphism (RFLP) analysis [8], amplified fragment length polymorphism (AFLP) analysis [9] and multilocus sequence typing (MLST) [10]. More recently, a fifth AFLP type (AFLP10) was described based on a single isolate [11]. These molecular types have been proposed for some time to be recognized as either varieties [12] or, more recently, as distinct species [13]. However, for the purpose of the current study they will be treated as distinct molecular types. Among C. gattii, the molecular type VGII, serotype B, has caused most of the outbreak-related cases, via clonal dispersion of three sub-genotypes, VGIIa, VGIIb and VGIIc [14–17]. Independent cases of infections caused by the molecular types VGI, VGIII and VGIV have been reported less frequently. VGI is the most prevalent molecular type in Australia and Papua New Guinea, where it is considered endemic [18,19]. VGIII is found predominantly in clinical and environmental sources in Mexico, Colombia and the USA [20–24], and VGIV has been mainly reported from India, African and a number of South American countries [8,25,26].
The number of C. gattii infections due to strains of the major molecular type VGIII is not only increasing in the endemic areas in South and Central America but also in the USA, where it is now a major cause of human and animal disease, with increasing cases reported from the Southeastern parts, especially California [17,27–29]. A total of 42 cases of human disease alone have been recorded since 2010, with the obtained genotypes being highly similar to the ones seen in both veterinary and environmental isolates. In contrast to the above-mentioned clonal VGII population, VGIII isolates have been highly diverse as shown by MLST analysis. This diversity has been attributed to the existence of both mating types α and a amongst clinical, veterinary and environmental VGIII isolates, which provides indirect evidence of sexual reproduction and dynamic recombination [17,27]. Furthermore, in South America and India, C. gattii VGIII has been isolated from the environment, although a specific association between its occurrence in the environment in these countries and cases of cryptococcosis has yet to be explored [22,30,31]. In a recent study, however, VGIII isolates recovered from arboreal sources in Southern California were genetically related to those causing human cases in similar locations, suggesting that these environmental isolates were the source of human infections [32].
Besides the C. gattii VGIII infections reported amongst HIV positive patients from Southern California and Mexico [20,21,33], the same molecular type has been recovered from immunocompetent human and veterinary patients. Several cases, including fatal infections, have been reported in patients without predisposing risk factors in Brazil, Colombia, Cuba, Mexico and the USA [17,23,24,34–39]. The emergence of C. gattii infections in immunocompetent patients is a source of public health concern as cryptococcosis is not usually suspected in this group and significant delays in diagnosis are associated with adverse outcomes. In apparently healthy hosts, cryptococcosis typically presents with cerebral involvement and is associated with more severe neurological sequelae, such as stroke, blindness, deafness and permanent neural deficits [37,39–41].
In addition, C. gattii VGIII isolates have been reported to be more susceptible to amphotericin B and 5-flucytosine than isolates of C. neoformans and other C. gattii molecular types [42,43]. Although azoles may have good in vitro activity against VGIII isolates, veterinary isolates of molecular type VGIII exhibited a wide range of minimum inhibitory concentrations (MICs) for fluconazole, with MICs as high as 32 μg/ml [28]. Previous studies have also indicated that VGIII is the most virulent molecular type in a Drosophila model of infection [44].
These differences have brought about the need to evaluate the epidemiology, disease transmission and virulence factors of this emerging pathogen. Therefore, the current study aimed to characterize C. gattii VGIII isolates genetically and phenotypically. In particular, to investigate the population genetic structure of VGIII isolates and to correlate geographic origin, source (clinical, veterinary or environmental), virulence in a mouse model of infection, and antifungal susceptibility. The data obtained from these investigations established a clearer understanding of the epidemiology and pathogenicity of C. gattii VGIII, revealed distinct ancestral populations, and found that there is no specific link between virulence and genotype of the studied isolates on the whole or mitochondrial genome level.
One hundred twenty-two isolates of C. gattii molecular type VGIII from Australia (n = 1), Colombia (n = 37), Guatemala (n = 1), Mexico (n = 14), New Zealand (n = 1), Paraguay (n = 1), the USA (n = 66) and Venezuela (n = 1), were studied. Isolates were stored at -80°C in glycerol in the Westmead Millennium Institute Culture Collection (Australian Medical Mycology Culture Collection) located at the Molecular Mycology Research Laboratory, Westmead Millennium Institute, The University of Sydney, Sydney Medical School, Sydney, Australia (S1 Table). Among these isolates 56 were clinical, 38 veterinary and 28 of environmental origin. Standard strains of the four major molecular types of C. gattii, WM 179 (VGI/AFLP4, serotype B), WM 178 (VGII/AFLP6, serotype B), WM 175 (VGIII/AFLP5, serotype B) and WM 779 (VGIV/AFLP7, serotype C) were included as reference strains for the genotypic analysis [10]. The proposed type culture of the AFLP10 strain [13] was not included, as it was not publicly available at the time this study was undertaken.
Isolates were cultured on Sabouraud dextrose agar and incubated for 48 h at 27°C prior to DNA extraction. Genomic DNA was extracted as described previously [45].
Restriction fragment length polymorphism (RFLP) analysis of the orotidine monophosphate pyrophosphorylase gene (URA5) following double digestion with the enzymes Sau96I and HhaI (New England BioLabs Inc) was used to identify the major molecular type of the isolates as reported previously [8].
Mating type was determined by PCR using the primers MfαU and MfαL for mating type α and MFa2U and MFa2L for mating type a and the PCR and amplification conditions as described previously [46].
The serotype of all isolates was determined using RFLP analysis of the capsular polysaccharide gene (CAP59), digested with the enzyme AgeI (New England BioLabs Inc), as described previously [47]. The serotype of 53 selected isolates (23 of serotype B and 30 of serotype C) was in addition determined using the agglutination test CryptoChek (Iatron Laboratories, Tokyo, Japan) according to the manufacturer’s instructions (S1 Table). In all instances, serotypes were concordant by both methods.
Sequences of 360 bp of the partial region of the CAP59 gene, which is used to determine serotype by RFLP, were extracted from the isolates with Whole Genome Sequencing (WGS) and a maximum likelihood dendrogram of these sequences was constructed to show the clear separation between serotypes B and C isolates (S1 Fig). The alignment and dendrogram were generated using the program MEGA 6.0 [48].
MLST typing was performed using the International Society of Human and Animal Mycology (ISHAM) consensus MLST scheme for C. neoformans and C. gattii, which includes seven genetic loci, CAP59, GPD1, LAC1, PLB1, SOD1, and URA5 genes, and the IGS1 region, as described previously [10]. All loci were amplified independently and the obtained PCR products were purified and commercially sequenced by Macrogen Inc., Seoul, Korea. Sequences were edited and contigs were assembled using Sequencher 5.3 (Gene Codes Corporation, Ann Arbor, USA). Each unique sequence was assigned an allele type (AT) number and the seven allele types per strain were subsequently combined to give a unique sequence type (ST) according to the ISHAM consensus MLST database, accessible at http://mlst.mycologylab.org. Alignments were generated using the program MEGA 6.0 [48]. The dendrogram showing the genetic relationship between the isolates based on the maximum likelihood analysis of the seven concatenated loci was generated using the same program. Haplotype network analyses were performed using the software Network 4.6.1.3 (Fluxus Technologies Ltd., Suffolk, UK). The goeBURST algorithm using the PHILOVIZ software (http://www.phyloviz.net) was used to generate a minimum spanning tree of the concatenated sequences to visualize relatedness of the C. gattii isolates according to the source of isolation and serotype. The diagrams show when the STs differ in a single locus variant (SLV), double locus variant (DLV), and triple locus variant (TLV). A clonal complex (CC) concept was adopted when SLV linkage with the founder ST was present [49].
To estimate the genetic diversity amongst the STs, Simpsons diversity index (D) was calculated for the whole population, as well as by country, source, serotype and mating type of the isolates.
The length of each MLST locus, the number of alleles and their frequency were determined and the genetic diversity of the seven loci was estimated by calculating the average number of nucleotide diversity (π) and the number of polymorphic (segregating) sites (S) using the software DnaSP ver. 5.10.1 [50]. For comparison of the genetic diversity between the VGIII sub-populations, the index θ, which is the Weir's formulation of Wright's fixation index (FST) for population differentiation analysis, was calculated for each locus. FST values of >0.05 generally indicate little inter-population variance, and can range from 0 for identical populations to 1 for populations with no alleles in common. The Index of Association (IA) and rBarD were also calculated to test for recombination between and within the VGIII sub-populations. Since clonal reproduction can mask the effects of recombination, IA and rBarD were calculated using the clone-corrected data for each ST after removal of identical genotypes were removed (haplotypes only). The values of both IA and rBarD are expected to be zero if populations are freely recombining and greater than zero if there is association between alleles. The rBarD statistic takes into consideration the number of loci tested and is considered a more robust measure of association. Values of FST, IA and rBarD were calculated using the program MultiLocus 1.3 [51].
The BEAST v1.8.3 software was used to perform the Bayesian molecular clock analysis of the VGIII MLST sequences [52]. The Tamura Nei model with invariable sites and gamma distribution (TrNef + I + G) used in BEAST was the best model selected from the Bayesian Information Criterion in the software jModelTest 2.1.7 [53]. The stepping-stone sampling marginal likelihood estimator available in MrBayes v3.2 software was used to infer the best-fitting clock model for the dataset [54].
A relaxed lognormal clock was applied to infer the time scale incorporating one internal node calibration of 8.5 million years as the time to most recent common ancestor for VGIII as already described [12]. A normal prior age distribution of 0.25 million years was used in the analysis. The XML file was generated in BEAUTI v1.8.3 with a run of 108 generations, 1 tree sampled per 1,000 generations, and a burn-in of 10% [52]. The LogCombiner v1.8.3, distributed with BEAST, was used to combine the files of two independent runs applying a burn-in of 10%. The results were visualized using the Trace v1.6.0 software distributed with BEAST and showed that the effective sample size was higher than 200 in all analyses. The tree with the highest log clade credibility was selected in the software TreeAnotator v1.8.3 and the tree presenting the posterior mean and 95% confidence intervals of the time to most recent common ancestor was visualized in the FigTree v1.4.3 software (http://tree.bio.ed.ac.uk/software/figtree/).
Sixty isolates, comprising 33 serotype B and 27 serotype C, representing the full diversity of the VGIII MLST genotypes, were selected for WGS. High quality DNA was extracted with the ZR Fungal/Bacterial DNA MiniPrep kit (Catalog N° D6005; Zymo Research, Irvine, CA, USA) following the instructions of the manufacturer.
The samples were sequenced using Illumina HiSeq as previously described [16,55]. DNA samples were prepared for paired-end Illumina sequencing using the Kapa Biosystems Library Preparation with Standard PCR kit (Catalog N° KK8232; Woburn, MA, USA) protocol. Approximately 1μg of double-stranded DNA (dsDNA) was sheared using a Sonicman sonicator (Brooks automation, Spokane, WA, USA) to an average size of 650 bp and DNA libraries were prepared for sequencing as described by the manufacturer. Modified oligonucleotides (Integrated DNA Technologies, Coralville, IA, USA) with 8bp indexing capability [56] were substituted at the appropriate step. Libraries were quantified prior to sequencing with quantitative PCR (qPCR) on the ABI 7900HT (Life Technologies Corporation, Carlsbad, CA, USA) using the Kapa library quantification kit (Catalog N° KK4835; Woburn, MA, USA). Libraries were sequenced to a read length of 100bp on the Illumina HiSeq system. WGS read files were deposited in the NCBI Sequence Read Archive under BioProject PRJNA289249.
All sequenced samples were assembled de novo using the SPAdes v2.5.0 assembler [57]. Read data for all genomes were aligned against the de novo assembly for sample WM 175 using Novoalign 3.00.03 (Novocraft Technologies, Selangor, Malaysia). Single nucleotide polymorphisms (SNPs) were detected using the Genome Analysis Toolkit v2.4 (GATK) [58]. SNP calls were filtered using NASP (http://tgennorth.github.io/NASP/) and had to meet the following criteria per SNP loci to be included in the final matrix: coverage of a minimum 10X and less than 10% variant allele calls. Additionally, reads that mapped to multiple locations within the genome were excluded from the analysis, as were positions located in an insertion or deletion site. The de novo assembly of sample WM 1814 was used as the reference strain for serotype C analyses; otherwise WM 175 was used as the reference strain. Additionally, one isolate of each of the other three C. gattii major molecular types (VGI, VGII, and VGIV), previously sequenced by WGS, were included for phylogenetic analysis [15]. In total, three SNP matrices with different taxa were produced: (i) C. gattii molecular types VGI to VGIV; (ii) C. gattii molecular type VGIII only, and (iii) C. gattii major sub-populations (serotype B and C) together.
Whole-genome SNP typing (WGST) was performed as previously described [16,59,60] for phylogenetic analysis in order to understand genetic relationships between isolates. To put the VGIII population in context with the other C. gattii major molecular types the genomes of the following C. gattii isolates were retrieved from GenBank: WM 179 and WM 276 representing molecular type VGI, WM 178, WM 05.419, WM 04.78, WM 06.12, WM 08.309, CDCR265, CDCR272, B9816 and GT 11.7650 representing molecular type VGII, and WM 779 representing molecular type VGIV [15,60]. Maximum parsimony SNP trees were constructed using PAUP* v.4.0b10 (Sinauer Associates, Inc., Sunderland, MA, USA) and visualized using FigTree v.1.3.1 (http://tree.bio.ed.ac.uk/software/figtree/). The C. gattii VGI to VGIV tree was not rooted. Additionally, the SNP matrix (iii) was examined for recombination using the phi test [61]. The neighbor-joining split tree network was drawn on the SNP matrix (iii) in order to visualize the existing recombination between samples using the program SplitsTree4 [62,63] with the uncorrected P-distance transformation. A maximum likelihood tree with 1,000 bootstrap generations was produced from SNP matrix (ii) using the TVM+ASC+G4 model in IQ-TREEv1.3.10 [64]. The tree was visualized using FigTree v1.3.1.
fineStrucutre analysis [65] was performed on the SNP matrix (ii) and (iii) in order to infer the population structure within VGIII as well as identify admixture events occurring between molecular types. Using the phylogenetic tree produced above, one representative from each clonal clade was selected and the SNP matrix was reduced to a pairwise similarity matrix using Chromopainter, which was run using the linkage model and assuming uniform rates of recombination per base pair of sequence. Populations were determined through fineStructure using the above-mentioned similarity matrix.
Putative gene content comparison was performed using BLAST score ratio (BSR) analysis [66], as previously described [15]. VGIII serotype B and serotype C predicted gene content differences were confirmed by alignment of sequence read data. Putative gene characterization of confirmed gene differences were translated into amino acid sequences and searched against the Pfam database (http://pfam.xfam.org/) in order to identify potential protein functions. Additional characterization for selected putative genes were searched against the NCBI non-redundant protein database (http://www.ncbi.nlm.nih.gov/RefSeq/) using blastp.
In order to assess mitochondrial re-arrangements and mutations, four high-virulence and four low-virulence serotype B samples were assembled using SPAdes3.0 [67]. Mitochondrial genes of the highly virulent VGII CDCR265 strain from the Broad institute were used as reference sequences. Fifteen mitochondrial genes were used to identify the mitochondrial contigs in the eight genome assemblies using BLAT. Contigs containing the 15 mitochondrial genes were pulled from the assemblies and were aligned using progressiveMauve v2.3.1 [68].
Additionally, the 56kb contig containing the 15 mitochondrial genes from the isolate WM09.47 was used as a reference sequence for SNP analysis using NASP as described above.
Based on the MLST results, 17 isolates were chosen to study the pathogenic potential of C. gattii molecular type VGIII in a murine model of pulmonary cryptococcosis. Female BALB/c mice, 6-weeks-old and weighing between 16 to 18 g, were inoculated intranasally with 105 yeast cells suspended in sterile saline. Prior to inoculation, the isolates were grown in Sabouraud dextrose agar at 37°C for 24 hours. Five mice per cryptococcal isolate studied were weighed and anesthetized intraperitoneally by injecting 0.03 to 0.04 ml of a combination of ketamine (44 mg/ml) and midazolam (2.7 mg/ml) (80 mg of ketamine and 5 mg of midazolam in a total volume of 1.8 ml), using an insulin syringe. Following induction of anesthesia, the mice were hung on a silk thread by their incisor teeth, so that the necks were fully extended. By using a pipette, 50 μl of inoculum was slowly instilled directly into each nostril. The well-studied C. gattii strains CDCR265 (VGIIa, highly virulent) and CDCR272 (VGIIb, low virulence) from the Vancouver Island outbreak were included as reference strains for comparison [69]. Five mice were also inoculated with sterile saline as an inoculation control. After inoculation, mice were placed in standard cages with access to water and food ad libitum and weighed and observed daily for signs of infection (e.g. difficulty breathing, neurological signs, ruffled fur, lethargy, poor appetite), until the end point of 60 days. Affected mice were euthanized by CO2 (5%) inhalation immediately upon observation of any signs of distress. Necropsy was performed and the brain and lungs were collected for macroscopic and histopathological examinations, to determine the presence of yeast and lesions. Blood collected aseptically from the heart using an insulin syringe was plated on Sabouraud dextrose agar to check for hematogenous dissemination. After harvesting tissues at necropsy, infected material was autoclaved and disposed of by incineration.
To compare the virulence of selected isolates, survival curves for each isolate were graphed. Median survival times were obtained and differences in survival times were analyzed by the Log-rank (Mantel-Cox) test. Statistical analysis and plots were carried out using GraphPad Prism version 6.0b (La Jolla, CA, USA). In all cases, p-values of <0.05 was considered statistically significant.
Susceptibility testing was carried out using the Sensititre YeastOne plate (Thermo Scientific, USA), which is a colorimetric microdilution test, following the manufacturer’s instructions. Briefly, isolates were grown on Sabouraud dextrose agar and incubated for 24 h at 27°C. Discreet yeast colonies were suspended with a swab into 5 ml of sterile water, adjusted to a density of 0.5 McFarland standard (1–5 × 106 cells/ml), and 20 μl aliquots were transferred into 11 ml of YeastOne inoculum broth for a final concentration of 1.5–8 × 103 CFU/ml. An aliquot of 100 μl of inoculum was placed in each well of the Sensititre YeastOne plate using a multichannel pipette. Plates were sealed, incubated at 35°C and read manually after 72 h of incubation.
The reference strains of Candida krusei ATCC 6258 and Candida parapsilosis ATCC 22019, were used as quality control. Controls were read after 24 h of incubation.
Purity of the cell suspension and colony counts were determined by plating 10 μl of inocula on to Sabouraud dextrose agar. The range of drug concentrations tested by 2-fold serial dilutions was 0.125–8 μg/ml for amphotericin B; 0.125–256 μg/ml for fluconazole; 0.015–16 μg/ml for itraconazole; 0.008–8 μg/ml for voriconazole and posaconazole; and 0.06–64 μg/ml for 5-flucytosine.
The range of minimum inhibitory concentrations (MICs), MIC50, MIC90 and geometric mean MICs of each antifungal drug were estimated. Epidemiologic cutoff values (ECV), defined as the MIC value encompassing at least 95% of the wild-type distribution, were calculated for each antifungal drug. Significant differences in MICs between two groups of isolates were compared using a Mann-Whitney test. Group comparisons for MIC data included serotype B vs. serotype C; mating type α vs. mating type a; clinical vs. veterinary isolates; clinical vs. environmental isolates and veterinary vs. environmental isolates. All analyses were performed with GraphPad Prism version 6.0b (La Jolla, CA, USA); p-values <0.05 were considered significant.
The virulence study was carried out in accordance with the protocol No. 4151-06-09 approved by the Westmead Hospital Animal Ethics Committee (WHAEC) adhering to Australian Code for the Care and Use of Animals for Scientific Purposes 8th Edition 2013 and the Animal Research Act New South Wales 1995.
RFLP analysis of the URA5 gene identified 116 of the 122 isolates as molecular type VGIII and six isolates displayed the restriction pattern of the molecular type VGIV. In silico restriction and alignments of the URA5 sequences of these six isolates, revealed a single nucleotide polymorphism (SNP) in the position 528, which is the restriction site of the enzyme Sau96I, resulting in misidentification of those isolates as VGIV [12]. However, as MLST analysis established that all 122 isolates were related, they were classified as molecular type VGIII.
Serotype and mating type analysis identified 60 (49%) isolates as B/α, 39 (32%) as C/α, 15 (12%) as B/a and 8 (7%) as C/a. The obtained serotype data were confirmed based on CAP59 sequences extracted from the whole genome sequencing (WGS) data (S1 Fig). Detailed descriptions of the mating and serotype results obtained for each isolate are provided in S1 Table.
Among the 122 isolates, 55 sequence types (STs) were identified (S1 Table). Of these, ST75 was the most frequent sequence type (21 serotype B isolates: five clinical and 15 veterinary isolates from the USA and one clinical isolate from Mexico), followed by ST79 (16 serotype C isolates: four clinical and nine environmental isolates from Colombia, two clinical isolates from Mexico and one from the USA). ST116 was the third most common sequence type, containing seven serotype B environmental isolates from Colombia.
Of the remaining 52 STs, 37 were represented by a single isolate each, while 15 were represented by two to five isolates, with ST65 (n = 2) and ST68 (n = 2), ST146 (n = 3) and ST74 (n = 5), each identified in more than one country. ST65, ST146 and ST74 each contained isolates from different source (S1 Table).
Based on maximum likelihood analysis and coalescence analysis of the seven concatenated MLST loci, C. gattii molecular type VGIII isolates separated into two major clusters or sub-populations corresponding mainly to serotype B and C (Figs 1A, 2, S1 Fig). These two sub-populations most likely correlate with the VGIIIa and VGIIIb lineages, respectively, that were recently described in independent MLST studies using a different MLST scheme [20,32], based on the loci that both MLST schemes share (GPD1, LAC1, PLB1 and IGS1) [10,20,32].
Among the serotype B isolates, WM 1811 and WM 1812 (ST 99), were identified as serotype C. However, because they shared most of the MLST alleles with the serotype B isolates, they were considered as such for the purpose of the analysis. Similarly, isolates WM 02.138 (ST95), WM 11.943 (ST140) and WM 1663 (ST94) were identified as serotype B, but were considered as serotype C for the analyses, as they clustered more closely with isolates of the later serotype.
Isolates outside the two major clusters, namely the six isolates misidentified as VGIV by URA5-RFLP, mentioned previously, grouped into two additional small sub-populations, with each corresponding either to the serotype B (n = 4) or C (n = 2) (Figs 1A, 1B and 2). From these atypical strains, we deduced the presence of a novel serotype B, VGIII ancient lineage among C. gattii VGIII isolates, represented by three isolates from Colombia (WM 2004, WM 2041 and WM 2042 (ST64)) and one isolate from the USA (WM 11.32 (ST114)), reported herein for the first time. The coalescence analysis showed that these isolates diverged from the VGIII isolates around 1.91 to 6.53 million years ago. A second ancient lineage that diverged from VGIII isolates around 7.94 to 8.92 million years ago, represented by two serotype C isolates (WM 1802 (ST100) and WM 1804 (ST101)) from Mexico, likely corresponds to the previously described VGIIIc/AFLP10 lineage [11,32], as these isolates share most of their MLST alleles with the published strain CBS11687 (IHEM14941 = RV 63979), with the exception of the SOD1 allele [11,13]. Both lineages appear to be basal to the VGIII clade (Figs 1A, 1B and 2).
The sequences obtained for each allele type were deposited in GenBank under the following accession numbers: CAP59 (JX840782—JX840787), GPD1 (JX840788—JX840795), LAC1 (JX840805—JX840821), PLB1 (JX840822—JX840832), SOD1 (JX840833—JX840840), URA5 (JX840841—JX840851), and IGS1 (JX840796—JX840804) (S2 Table or at the INTERNATIONAL FUNGAL MLST DATABASE website at mlst.mycologylab.org).
The VGIII population was genetically highly diverse (D = 0.061). There were no statistically significant differences among the groups with respect to the country of origin, isolate source (clinical vs. veterinary vs. environmental), or mating type. Only minor differences were identified between serotypes, with serotype C isolates being slightly more diverse than those of serotype B (D = 0.056 vs. 0.105 (p < 0.05)) (S3 Table).
Among the seven loci studied, LAC1 was the most informative locus with 17 alleles, 31 polymorphic sites over 477 bp, a nucleotide diversity of 0.895%, and a fixation index (FST or θ) of 0.5137. Although SOD1 was represented by eight alleles, two more than CAP59, it was the least informative locus with only 16 polymorphic sites over 713 bp and a nucleotide diversity of 0.205% (Table 1). Overall, the seven concatenated loci resulted in an alignment of 4,212 bp with 160 polymorphic sites. The high FST values of all seven loci indicate high genetic diversity between the VGIII sub-populations (serotype B and serotype C clusters, respectively), indicating low genetic flow between them (Table 1). The low number of shared MLST alleles between serotype B and serotype C isolates (S1 Table), and the high values obtained with the tests of linkage disequilibrium (IA = 1.16184 and rBarD = 0.197630 (p = 1.00)), further support low genetic flow within this population structure. Tests of linkage disequilibrium showed some recombination within each sub-population, with serotype B isolates (IA = 0.478372 and rBarD = 0.0962771 (p < 0.01)) recombining less frequently than serotype C isolates (IA = 0.342210 and rBarD = 0.0575250 (p = 0.02)).
Haplotype network analyses per locus revealed a low number of shared MLST alleles between sub-populations, indicating a low level of recombination (Fig 3). The low number of shared alleles between these two C. gattii VGIII sub-populations was also supported by the goeBURST analysis with the concatenated dataset (Fig 4). Overall, only the four atypical serotype C isolates from Mexico (WM 1802 (ST100), WM 1804 (ST101), WM 1811 (ST99) and WM 1812 (ST99)), shared alleles, and were grouped in the serotype B cluster. This analysis also presented 11 clonal complexes (CC) (i.e. 11 groups presenting single locus variant (SLV)) with two of them, CC79 (composed of ST79, ST82, ST80, and ST65 in serotype C group) and CC75 (composed of ST75, ST115, ST138, ST72, ST143, ST139, and ST87 in the serotype B group), appearing to play an important role in the epidemiological distribution of the C. gattii VGIII population due to the wide geographical distribution of the CCs (Fig 4).
Whole genomes from 60 C. gattii VGIII isolates, representing at least one isolate per ST identified in the MLST analysis, were sequenced. Whole genome sequencing (WGS) determined the presence of 572,268 SNPs with 514,098 SNPs being parsimony informative. Within the serotype B and serotype C major sub-populations, 88,337 and 79,945 SNPs were identified, respectively. Maximum parsimony analysis based on whole genome SNP typing (WGST) confirmed the same clustering of the VGIII isolates as obtained by MLST typing, although with much higher resolution (Figs 1B and 5). When the whole genomes of the other major molecular types of C. gattii, VGI, VGII, and VGIV [15] are included (Fig 5), WGST SNP data found 1,347,295 total SNPs with 1,055,552 of them being parsimonious SNPs, with a consistency index (CI) of 0.7934.
Neighbor-joining phylogenetic splits tree network analysis of the WGST SNP data clearly separated the major serotype B and serotype C clusters within VGIII and showed many phylogenetic splits within each sub-population (Fig 6A). These findings indicate a shared genetic history, possibly including sexual recombination events within, but not between the two main serotype groups in the VGIII population, which largely contributes to the genetic diversity found within each serotype. When the Phi test for recombination was performed using the SNP data, the test indicated that recombination was present within each of the two major VGIII serotype groups (p = 0.0).
fineStructure analysis showed that the serotype B isolates shared very few or no genomic regions with the serotype C isolates. However, within sub-populations, there are some shared genome regions, with serotype C isolates having a greater amount of shared regions than serotype B isolates, indicating a significant separation between the two major VGIII serotype groups, but at the same time also suggestive of some level of recombination within each of them (Fig 7). In addition fineStructure analysis indicates incomplete lineage sorting among the atypical strains, accounting for the maintenance of ancestral genome parts (S2 Fig). The VGI genome contributed more to the genomes of the atypical VGIII serotype B strains (WM 1802, WM 1804) and the VGIV genome contributed stronger to the genomes of the atypical VGIII serotype C strains (WM 2004, WM 2041, WM 2042 and WM 11.32) (S2 Fig). If these ancestral groups are isolated (genomically and geographically) they would have very little recombination opportunity and little new variation, and therefore do not “share” their genome with others.
Very few differences in gene content were found between serotype B and serotype C isolates using the BLAST Score Ratio (BSR). An analysis of the presence/absence of genes in the VGIII sub-populations identified two gene clusters that were unique to the serotype C genomes, and one gene cluster that was unique to the serotype B genomes. However, all three gene clusters represented hypothetical proteins of unknown function. Both clusters identified in the serotype C isolates, did not have significant matches (E values 7.00E-51 and 3.00E-109), and although the cluster identified in the serotype B isolates matched with a H-N-H homing endonuclease (E value 8.00E-79), the amino acid identity was only 50% with 96% coverage.
Based on previous findings, implicating changes in mitochondrial morphology and mitochondrial gene expression to an increased virulence in the Vancouver Island Outbreak VGII strains [70], the mitochondrial genomes were bioinformatically extracted from the WGS data set of the 60 C. gattii VGIII isolates, representing at least one isolate of each ST identified in the above mentioned MLST analysis. Interestingly, the estimated mitochondrial genome size of C. gattii VGIII strains was 55 kb, which is much larger than the mitochondrial genome sizes of C. gattii VGII strains (34.7 kb) [70], C. neoformans var. grubii strains (24 kb) and C. neoformans var. neoformans strains (32 kb) [71]. Mitochondrial genome sequencing determined the presence of 577 SNPs with 415 SNPs being parsimony informative, with a consistency index (CI) of 0.36. Neighbor joining phylogenetic splits tree network analysis of the mitochondrial genomes confirmed a similar but not identical topology for the two major VGIII clusters identified in the WGS analysis (Fig 6B). No recombination between the two major clades obtained from the mitochondrial genomes was identified, Phi test (p = 0.09032). However, the Phi test for recombination using the mitochondrial SNP data indicated that recombination was present within each of the two major VGIII serotype groups, serotype B, 498 SNPs, with 320 SNPs being parsimony informative, Phi test (p = 0.0032), and serotype C, 333 SNPs, with 258 being parsimony informative, Phi test (p = 0.000000143), indicating possible sexual recombination events.
Seventeen isolates, widely representative of the identified MLST genotypes, were studied in a mouse model of infection. Five were highly virulent and caused 100% mortality, while 12 did not kill any mice within 60 days of inoculation (Table 2, Fig 8). Of the virulent isolates, WM 11.105 (C/α, ST79, a clinical isolate from Colombia) was the most virulent, even more lethal than CDC R265 (the highly virulent VGIIa reference strain from the Vancouver Island outbreak, which was used as a control [69]) (p = 0.0112), followed by WM 2088 (B/a, ST59, a clinical isolate from Colombia), WM 11.139 (B/a, ST143, a veterinary isolate from the USA), WM 09.47 (B/α, ST74, a veterinary isolate from the USA) and WM 11.118 (B/α, a clinical isolate from Colombia), which was the least virulent (p = 0.0025). Pairwise comparison among the other virulent isolates showed no significant differences (p >0.05). Four of the five virulent isolates were serotype B while only one was serotype C. Of the four serotype B isolates, two human clinical isolates (WM 11.139 and WM 2088) were mating type a and two veterinary isolates (WM 09.47 and WM 11.139) were mating type α. No environmental isolates were virulent in this mouse model.
Macroscopic examination after necropsy revealed multiple granulomata in the lungs of the mice infected with virulent cryptococcal isolates. In contrast, few or no granulomata were observed in lung tissue from mice that survived for at least 60 days post inoculation. Direct microscopy of the lung tissue suspensions stained with Indian ink revealed numerous cryptococci in the lung samples of all infected mice. Lung tissue burdens of cryptococci (number of yeasts per gram) did not differ significantly among the virulent isolates. Brains excised from these mice were macroscopically normal and brain suspensions were culture negative. Direct microscopy of India ink preparations of the brain suspensions revealed not more than four cryptococcal cells. It is possible these represent yeasts originating from cerebral or meningeal blood vessels. Histological examination of lung tissue from mice infected with the five virulent isolates revealed widespread location of cryptococci within the alveoli, interstitial tissue and the airways. C. gattii was recovered from the cardiac blood from 11 out of the 17 isolates inoculated into mice, indicative of dissemination of cryptococcal cells from the lungs to circulation (Table 2). Fig 9 shows multiple granulomata and numerous cryptococci in the lungs of mice infected with the most virulent C. gattii VGIII isolate WM 11.105. Isolates with enhanced virulence caused significant weight loss during the course of infection (Fig 10).
Antifungal susceptibilities of all VGIII isolates to amphotericin-B, 5-flucytosine, posaconazole, voriconazole, intraconazole, and fluconazole, were determined (S1 Table). Minimum inhibitory concentrations (MICs), MIC50, MIC90, geometric mean MICs and epidemiological cut-off values for all isolates are shown in Table 3. One veterinary isolate from the USA (WM 11.937) and two clinical isolates from Colombia (WM 11.105 and WM 11.112) had high fluconazole MICs; the first isolate (WM 11.937) with a MIC of 64 μg/ml and the last two (WM 11.105 and WM 11.112) with MICs of 128 μg/ml. The comparison of MIC distributions for the tested drugs according to the serotype, mating type and source of the isolates is shown in S4 Table. Overall, serotype C isolates had statistically significant higher modal MICs and geometric mean MICs for posaconazole, voriconazole, itraconazole and fluconazole than serotype B isolates, but lower geometric mean MICs for 5-fluorocytosine (p <0.05) (Table 4, S4 and S5 Tables). Statistically, environmental isolates were less susceptible to the tested antifungals, except for amphotericin-B, compared with clinical and veterinary isolates (Table 5 and S4 Table). There was no significant difference in antifungal susceptibility profiles between mating type α or a isolates (p > 0.05) (S4 Table). Susceptibility to amphotericin-B did not vary significantly with the source of the isolates (S4 Table). The epidemiological cut off values (ECVs) were the same as the MIC90 for the tested drugs, except for posaconazole, where the ECV was 0.25 μg/ml, compared with an MIC90 of 0.125 μg/ml (Table 3).
Taking into account the rising importance of VGIII as cause of clinical and veterinary infections [17, 20–24, 27–29], their genotypic and phenotypic epidemiology has been under-investigated, compared with strains of molecular type VGII. Two MLST studies performed on a VGIII population from HIV positive patients from Southern California, identified two major molecular groups, VGIIIa and VGIIIb, that differed in virulence and fertility and a minor VGIIIc cluster represented by only one isolate [20,32]. Shortly after the first study was published, VGIII was found to predominate among the molecular types recovered from both human and veterinary samples outside the Pacific Northwest in the USA [17], and amongst cats in California [28]. It was also identified as the second most common molecular type amongst human cases in Colombia [24]. Based on these limited reports, we conducted a VGIII population analysis of clinical, veterinary and environmental isolates from a broader geographic range, taking into account the previously reported endemic areas and sporadic cases [25]. Isolates were sampled from the USA, Colombia and Mexico, and single cases from Australia, Guatemala, New Zealand, Paraguay and Venezuela, to give a more comprehensive perspective of the epidemiology of the VGIII molecular type and to investigate possible correlations between genotype and virulence and antifungal susceptibility phenotypes.
Although C. gattii VGIII has been recovered infrequently from human cases in Argentina, Guatemala [8], Cuba [37], Western Europe [72] and Korea [73, 74], and from the environment in Argentina [31] and India [30], this molecular type remains an important cause of neglected cryptococcosis cases in Brazil [34, 38], Colombia [23,24] and Mexico [21,35]. In addition, the previously described endemic region of C. gattii VGIII is expanding beyond the borders of the US state of California, with an increase of both clinical and veterinary cases [17,28]. The high incidence of cryptococcosis caused by C. gattii VGIII beyond the tropical and subtropical areas, where it is considered to be endemic, and the emergence of this pathogen in more temperate regions traditionally considered of low risk for the acquisition of this infection are major clinical concerns. C. gattii, including VGIII is mostly isolated from patients without recognized immunologic defects and may be associated with worse clinical outcomes, complications such as permanent neurologic sequelae and the requirement for prolonged periods of antifungal treatment [37,39–41]. Failure to consider the diagnosis or delays in doing so in immunocompetent patients, result in failing to initiate the most appropriate therapy, and consequently increase morbidity and mortality. This is illustrated by recent cases of disseminated cryptococcosis caused by C. gattii VGIII, including two fatal cases reported from Cuba and the USA [37,39].
As also reported in the aforementioned studies, the herein studied VGIII population showed a high level of genetic diversity, with no geographic restriction of genotypes. This is at variance with observations made with VGII subtypes [14–17]. Not only were the VGIII isolates from the endemic areas of Colombia, Mexico and the USA closely related, but they also shared genotypes with most of the sporadic cases from around the world, namely ST68 (found in New Zealand and the USA), and ST65 (found in Venezuela and Colombia) (Figs 1 and 4). Although only a single isolate from Guatemala was identified as ST96, this genotype clustered very closely with isolates from Mexico (Fig 1). Similarly, a single isolate of ST144 found in Australia, clustered very closely with isolates from the USA (Fig 1). Interestingly, there was an association between clinical and veterinary genotypes and those identified from environmental samples, which present the natural reservoirs for C. gattii (Fig 1).
Determination of the serotype of the isolates clearly revealed that the major MLST, WGST and mitochondrial genome clusters within VGIII correspond to either serotype B or serotype C, corresponding most likely to the subgroups VGIIIa and VGIIIb, respectively, which were designated previously, based exclusively on MLST genotypic clustering [20,32]. This discrimination between the serotypes of C. gattii VGIII has been demonstrated previously using Fourier transform infrared spectroscopy, which, in contrast to MLST, WGS and mitochondrial genome sequencing, characterizes phenotype instead of genotype [72]. Phylogenetic and coalescence analyses also revealed two more distant, but basal groups in the VGIII population, which interestingly, have been erroneously designated as VGIV following URA5-RLFP analysis, due to a SNP in the restriction site of Sau96I [12]. In addition, these two minor clusters did not share any of the MLST alleles with the major groups. Notably, they were additionally separated according to serotype, in spite of being represented by only a few isolates each (Figs 1 and 2; S1 Table). Importantly, one of these minor/basal groups, which includes the serotype C isolates WM 1802 and WM 1804 from Mexico, is closely related to the previously described AFLP10 type, which differs in one of the seven MLST loci, the SOD1 locus having allele type (AT) AT51 in strain IHEM14941S compared to AT39 in the herein studied strains WM 1802 and WM 1804 [11,32]. This strain has recently been proposed as a distinct species among C. gattii (C. decagattii) [13]. The second minor group described in this study, which includes the serotype B isolates WM 2004, WM 2041 and WM 2042 (all ST64) from Colombia, and WM 11.32 (ST114) from the USA, could similarly represent another new cryptic species, considering the species concept proposed by Hagen et al. [13]. However, DNA barcoding gap analysis, accounting for all four identified subgroups within the VGIII isolates combined, does not reveal a DNA barcoding gap (S3A Fig). If the newly described minor serotype B subgroup (WM 2004, WM 2041, WM 2042 and WM 11.32) is removed from this analysis, a DNA barcoding gap emerges between the major two serotype groups, B and C, and the minor serotype C subgroup (WM 1802 and WM 1804, similar to AFLP10 [13]) (S3B Fig). As such, the separation of the AFLP10 isolates from the VGIII isolates may be mistaken because of the small sample size. Depending on the number of isolates that are included in the phylogenetic analysis, the species borders can become blurred, indicating ongoing speciation events.
Based solely on the MLST analysis serotype C isolates were slightly more diverse than serotype B isolates and there was a low gene flow between isolates of different serotypes, as reflected by population differentiation analysis (Figs 1, 2 and 3). Tests of linkage disequilibrium showed additionally that serotype C isolates recombine more readily than serotype B isolates. These conclusions were also supported by the phylogenetic network and fineStructure analyses of the whole genomes and the mitochondrial genomes, showed almost no recombination between the two serotypes but recombination within each of the serotypes, with more sharing of genomic content amongst serotype C isolates (Figs 6 and 7). Identification of the two opposite mating types amongst both serotype B and serotype C isolates provides further evidence for sexual recombination within these VGIII sub-populations, which may contribute to their genetic diversity. The occurrence of well-supported sub-populations, which are separated geographically and in time, suggests that recombination and genetic exchange events are not occurring between the two major serotype specific groups of the molecular type VGIII and that this population is going through a process of expansion, divergence and perhaps speciation.
MLST, maximum parsimony-based WGST, and coalescence analyses demonstrated that the two major VGIII sub-populations, serotypes B and C, which share minimal genetic diversity, likely originated from very distant ancestors in the VGIII endemic regions of Colombia, Mexico or the USA (Figs 1A, 1B, 4 and 5). The two distinct atypical populations link the two major VGIII populations (serotype B and C) specifically to the C. gattii lineages VGI and VGIV, confirming findings from comparative WGS studies, which also showed a closer link between VGI, VGIII and VGIV [75] (Fig 5). fineStructure analysis confirmed their ancestral role by indicating shared gene content between VGI and/or VGIV (S2 Fig). This reflects also findings by Farrer at al., that structural genome rearrangements are almost exclusive to the VGI, VGIII and VGIV lineages [75]. Given that atypical isolates may be under-sampled or misidentified as molecular type VGIV using traditional molecular methods (i.e. URA5-RFLP), further studies are required to more accurately infer the ancestors of the VGIII population.
Inclusion of the highly virulent reference strain of the VGIIa subtype (CDCR265) [69] in the mouse model of infection permitted the recognition of VGIII isolates with enhanced or comparable virulence to the VGII Vancouver Island outbreak isolates and very similar diseases patterns (Figs 8, 9 and 10; Table 2). Importantly, the mortality from infection with serotype B isolates was higher than that caused by serotype C isolates. Nevertheless, most of the isolates formed granulomata and direct microscopic examination revealed yeast cells in all lung sections similar to an isolate with increased virulence (WM 11.105) (Fig 9). These findings indicate that as reported in infections mainly caused by VGIIa strains in British Columbia and the Pacific Northwest [36,76], pulmonary cryptococcosis is the predominant clinical manifestation of C. gattii VGIII serotype B infections. Among the highly virulent serotype B isolates identified in this study, the veterinary isolate WM 09.47 shared the same genotype (ST74) with the strain responsible for a fatal case of cryptococcosis reported in New Mexico in 2010 (WM 11.935, B7495) [39] and with a clinical isolate (WM 1819) from Mexico recovered in 1990. This finding suggests that the identification of certain genotypes may be indicative of increased virulence. It is conceivable that these virulent genotypes are circulating but are undocumented. Paradoxically, there were no specific whole or mitochondrial genome differences between low- and high-virulence isolates in the two major groups. This is similar to the findings by Ma et al. in 2009, which showed also no correlation between the major genes coding for known virulence factors and the actual virulence in the VGII Vancouver Island Outbreak strains [70], but identified changes in the mitochondrial morphology and mitochondrial gene expression as major factors of increased intracellular proliferation, corresponding to increased virulence. However, specific comparative mitochondrial genome analysis between high and low virulent VGIII strains conducted herein did not, like the WGS analysis, reveal any specific changes. Gene rearrangement analysis (i.e., progressive Mauve) showed variation among the mitochondrial genomes of the strains included in the virulent study, not specific changes associated with either high or low virulent strains were found (S4 Fig). The variation found within the mitochondrial genomes is in agreement with the observation made previously, attributing the fact that mitochondria are more recombinogenic than their nuclear counterparts, to the ability to change the mitochondrial phenotype [75]. As no specific whole genome or mitochondrial genome differences between high and low virulent have been found herein, the differences in virulence may therefore be related to phenotypic characteristics generated by differences in gene expression, for example, different rates of multiplication at 37°C, the ability to disseminate from the lung to the brain and other sites via the blood and to overcome the host immune response.
Recovery of yeast cells from cardiac blood (heart blood collected at time of euthanasia) suggests that C. gattii VGIII can disseminate to the CNS, but that experimentally-inoculated mice die of cryptococcal pneumonia before establishment of meningoencephalitis, as previously found [77]. We suggest that the small number of yeasts detected by direct examination of brain tissue from mice infected with VGIII isolates represented yeasts within cerebral blood vessels, as there was no clinical or histopathological evidence for infection of the central nervous system.
To date, many in vitro susceptibility studies have been performed on C. neoformans and C. gattii, to elucidate differences between species, serotypes and molecular types. Differences between serotypes have been mostly reported within C. neoformans, where there is a clear correspondence between serotype and molecular type. This close correspondence has not been reported within C. gattii, mainly because the serotype in this species is rarely identified and documented [11,28,42,43,78–81]. Because differences in antifungal susceptibility can influence therapeutic choices in the clinical setting, the findings of this study are of interest. Serotype C isolates were significantly less susceptible to azoles, especially fluconazole, than serotype B isolates (Table 4). In addition, irrespective of serotype, environmental isolates were slightly less susceptible to azoles and 5-fluorocytosine than clinical and veterinary isolates, indicating that their use in establishing ECVs may be misleading (Table 5). However, the association between the source of the isolates (clinical, veterinary and environmental) and antifungal susceptibility profiles remains unclear, as different findings have been reported elsewhere. In a previous Brazilian study, for instance, clinical isolates of C. neoformans were reported to be less susceptible to antifungal drugs than environmental isolates [80], and in general, veterinary isolates of C. gattii collected worldwide were found to be less susceptible to antifungal drugs than clinical and environmental isolates [11].
Overall, C. gattii VGIII strains have been more susceptible to amphotericin B and 5-flucytosine than other C. gattii molecular types and C. neoformans [42,78,81]. In the present study ECVs of both these antifungal agents were higher than reported previously (0.5 μg/ml for amphotericin B and 4 μg/ml for 5-flucytosine). C. gattii has shown variable susceptibility to fluconazole and other azoles [28,43,79]. However, this study is the first to document high values of GM MICs and ECVs amongst C. gattii VGIII isolates, specifically for fluconazole and itraconazole, which are currently recommended as alternative induction therapy for pulmonary cryptococcosis [82]. All of these findings emphasize that recognition of serotype and molecular type in C. gattii isolates can identify isolates with acquired resistance mechanisms, based on the reported ECVs for each drug and may be relevant to the choice of the treatment regimen for a specific cryptococcal infection. Clinical studies correlating these parameters with responses to therapy and patient outcomes are required.
In conclusion, the herein reported study of clinical, veterinary and environmental isolates from the main endemic areas in the world revealed a high genetic diversity within the C. gattii molecular type VGIII population. Two well-supported and divergent lineages were identified, corresponding to serotypes B and C. In addition distant ancestors within the molecular type that are represented by isolates from VGIII endemic areas were revealed in either Mexico or Colombia/USA, linking the two major VGIII populations to the other major molecular types within C. gattii, specifically to VGI and VGIV. The predominant clinical manifestation of C. gattii VGIII infections was pulmonary disease rather than meningitis or encephalitis. No specific associations between the WGS or mitochondrial genome and virulence have been observed. Antifungal susceptibility profiles differed according to serotype. The results of this study reinforce the notion that global cooperation is necessary to more accurately determine the prevalence of C. gattii infection and redefine endemic regions. Additionally, surveillance of antifungal susceptibility patterns and correlation with clinical outcomes is needed to optimize therapeutic guidelines and hence clinical outcomes.
|
10.1371/journal.pntd.0004423 | Phase 1/2a Trial of Plasmodium vivax Malaria Vaccine Candidate VMP001/AS01B in Malaria-Naive Adults: Safety, Immunogenicity, and Efficacy | A vaccine to prevent infection and disease caused by Plasmodium vivax is needed both to reduce the morbidity caused by this parasite and as a key component in efforts to eradicate malaria worldwide. Vivax malaria protein 1 (VMP001), a novel chimeric protein that incorporates the amino- and carboxy- terminal regions of the circumsporozoite protein (CSP) and a truncated repeat region that contains repeat sequences from both the VK210 (type 1) and the VK247 (type 2) parasites, was developed as a vaccine candidate for global use.
We conducted a first-in-human Phase 1 dose escalation vaccine study with controlled human malaria infection (CHMI) of VMP001 formulated in the GSK Adjuvant System AS01B. A total of 30 volunteers divided into 3 groups (10 per group) were given 3 intramuscular injections of 15μg, 30μg, or 60μg respectively of VMP001, all formulated in 500μL of AS01B at each immunization. All vaccinated volunteers participated in a P. vivax CHMI 14 days following the third immunization. Six non-vaccinated subjects served as infectivity controls.
The vaccine was shown to be well tolerated and immunogenic. All volunteers generated robust humoral and cellular immune responses to the vaccine antigen. Vaccination did not induce sterile protection; however, a small but significant delay in time to parasitemia was seen in 59% of vaccinated subjects compared to the control group. An association was identified between levels of anti-type 1 repeat antibodies and prepatent period.
This trial was the first to assess the efficacy of a P. vivax CSP vaccine candidate by CHMI. The association of type 1 repeat-specific antibody responses with delay in the prepatency period suggests that augmenting the immune responses to this domain may improve strain-specific vaccine efficacy. The availability of a P. vivax CHMI model will accelerate the process of P. vivax vaccine development, allowing better selection of candidate vaccines for advancement to field trials.
| Plasmodium vivax malaria has several unique features. Two of the main features are the inability to culture this parasite in vitro and its propensity to form dormant stages within the liver, which can only be treated with a single class of drugs that are contraindicated for a proportion of the population. Therefore, vaccines will play an important role in preventing this geographically widespread malaria species. In this clinical trial, we tested increasing amounts of the vaccine candidate VMP001/AS01B for safety and immunogenicity. In order to test if the vaccine can afford protection, we challenged the volunteers via the bite of infected mosquitoes, the first time such a human infection model has been used to evaluate vaccine efficacy for P. vivax malaria. While the vaccine did not protect any of the vaccinated subjects, this study resulted in some important findings, including the observation that a significant proportion of the subjects displayed a trend towards a delay in infection in individuals that correlated with antibodies to the repeat region of the vaccine antigen.
| Malaria is a devastating parasitic disease transmitted through the bite of infected Anopheles mosquitoes. Outside sub-Saharan Africa, Plasmodium vivax is the most prevalent of all human malarias with approximately 2.48 billion people at risk [1] and an estimated 16 million cases in 2013 (WHO World Malaria Report, 2014). Unlike Plasmodium falciparum, P. vivax produces liver stages (hypnozoites) that, initially dormant, can reactivate several weeks to months after the primary infection causing symptomatic disease [2,3]. This propensity to relapse stands as a significant barrier to efforts to eradicate this species of malaria [3]. Additionally, P. vivax is increasingly reported as the causative agent of symptoms associated with severe malaria as well as chloroquine resistance [4–7]. A vaccine to prevent infection and disease caused by P. vivax is urgently needed to reduce morbidity of the disease and accelerate elimination of this parasite.
The circumsporozoite protein (CSP) is the most abundant sporozoite protein present on the sporozoites of all Plasmodium species and has been shown to have great potential as a vaccine target [8,9]. Antibodies to the repeat region of P. falciparum CSP have been shown to be associated with protection [10–12]. Unlike P. falciparum, the repeat region of P. vivax CSP exhibits sequence heterogeneity resulting in immunologically distinct populations indicating that a vaccine based on one strain may not be sufficient to protect against all circulating strains [13]. To take into account the diversity of P. vivax strains, we developed vivax malaria protein 001 (VMP001) as a candidate vaccine for P. vivax malaria. The vaccine antigen VMP001 is an Escherichia coli produced synthetic chimeric recombinant protein that incorporates the three major domains of CSP but is distinct from the native molecule [14,15]. This synthetic construct includes the amino (N-) and carboxy (C-) terminal parts of CSP and a truncated repeat region that contains repeat sequences from the immunologically divergent VK210 (type 1) and the VK247 (type 2) strains of parasites. The VMP001 antigen was adjuvanted with AS01B, a proprietary liposome-based adjuvant system from GSK Biologicals that contains the immunostimulants monophosphoryl lipid A (MPL) and QS-21, a triterpene glycoside purified from the bark of Quillaja saponaria [9]. This adjuvant system has been used in other malaria vaccine candidates, including RTS,S [9].
We report the results of a first in humans phase 1 clinical trial using VMP001/AS01B in terms of reactogenicity, immunogenicity, and efficacy against a P. vivax sporozoite challenge in healthy, malaria-naive adults.
The study (ClinicalTrials.gov identifier NCT01157897), sponsored by the Office of the Surgeon General, U.S. Army, was conducted following scientific and ethical review by the WRAIR scientific review committee, WRAIR institutional review board (IRB), the USAMRMC’s Human Subjects Research and Review Board as well as the Western IRB and assigned protocol numbers WRAIR 1692, HRPO A-16037. The protocol was conducted under the U.S. Food and Drug Administration (FDA) Investigational New Drug (IND) application #14380.
Written informed consent was obtained from all volunteers prior to screening and enrollment. Subjects were healthy malaria naïve men and women aged 18–55 years. All subjects had normal blood levels of glucose-6-phosphate dehydrogenase (G6PD), and were either homozygous or heterozygous positive for Duffy antigen. Subjects agreed to be available for the duration of study with no plans to travel to a malaria endemic area or outside the Washington, DC area until a treatment course was completed following CHMI.
This was a phase 1, non-randomized, open label, dose-escalation study in 36 adults. Thirty volunteers, divided into 3 cohorts (10 in each group), were vaccinated with 3 doses of VMP001/AS01B. Cohorts 1, 2, and 3 received 15 μg, 30 μg, or 60 μg, respectively, of VMP001in 500 μL of AS01B at each immunization. The first and second immunizations in each cohort were separated by 28 days, and the third dose for all cohorts was normalized such that the interval between the last immunization and day of challenge was 2 weeks (Fig 1). Controlled human malaria infection (CHMI) using P. vivax infected Anopheles dirus mosquitoes was performed in the volunteers from all cohorts that completed all 3 immunizations (n = 27) and a control group (n = 6) who were not administered investigational product (Figs 1 and 2).
The VMP001 antigen [14,15] and the AS01B adjuvant system [9] have been described previously. The VMP001 recombinant subunit protein was produced in and purified from E. coli [15] and reconstituted in 500 μl of AS01B. AS01B is an Adjuvant System containing 50 μg 3-O-desacyl-4’- monophosphoryl lipid A (MPL, produced by GSK) and 50 μg Quillaja saponaria Molina, fraction 21 (QS-21, licensed by GSK from Antigenics Inc, a wholly owned subsidiary of Agenus Inc., a Delaware, USA corporation), in a liposomal formulation [9]. Lyophilized VMP001 was reconstituted with liquid AS01B and administered in doses of 15 μg, 30 μg, or 60 μg (500 μl/dose) by slow intramuscular injection within 1 hour of reconstitution.
Vaccine tolerability was assessed by evaluating local reactogenicity and systemic symptoms, as well as changes in biochemical and hematologic laboratory parameters. Clinical assessments of subjects were performed 30 min after each vaccination and again 1, 2, and 6 days later. Local and systemic solicited adverse events (AEs) were collected during this 7 day period. Severity of adverse reactions was classified as grade 1 (mild), grade 2 (moderate) and grade 3 (severe). Unsolicited AEs were recorded during the 28 days following vaccinations 1 and 2 and during the 14 days after vaccination 3 and reported to a safety monitoring committee (SMC). Biochemical and hematologic laboratory parameters were measured prior to administration and 6 days after each vaccination. Complete blood cell counts with white blood cell differential counts were performed in addition to serum levels of creatinine, alanine aminotransferase, and aspartate aminotransferase.
During the challenge phase, solicited AEs were recorded beginning on the day of CHMI. Clinical assessments continued on days 1 and 3 post-CHMI, and then daily beginning on day 5 until the subject had 3 consecutive negative daily blood smears following the initiation of treatment for malaria infection. All unsolicited events were documented for 28 days beginning on the day of CMHI. Safety laboratories were measured on the day of CHMI, at the time of initiating anti-malaria therapy, and on days 28 and 42 post-CMHI. Serious AEs (SAEs) were reported from the time of subject enrollment until study closure. Episodes of P. vivax relapse were recorded up to the conclusion of the study period, approximately 5 years post-CHMI.
This study was conducted at the WRAIR Clinical Trials Center. A total of 30 subjects received at least one vaccination. Of those, 27 subjects that completed the three dose vaccination regimen, and 6 infectivity control subjects were challenged by the bites of five P. vivax infected mosquitoes. (Figs 1 and 2).
Immunizations of subjects in all three dose cohorts were well tolerated and no safety halting criteria were met. There were no clinically concerning imbalances observed between groups. Similar frequencies of solicited local and solicited general AEs were reported in the three cohorts (Fig 3). There were no AEs in any vaccination cohort that led to withdrawal of any subjects. One SAE (ductal carcinoma in situ of the breast) occurred during the study and it was determined that the event was not related to the study vaccine or CHMI.
Mild (grade 1) to moderate (grade 2) pain in the days following immunization was the most frequently reported local solicited AE, occurring in 100% of subjects who received at least one vaccination. No subjects experienced severe (grade 3) pain. Grade 3 solicited local AEs included erythema at the injection site in 4 subjects, 1 in cohort 2 and 3 in cohort 3. The grade 3 erythema lasted for ≤2266 3 days in 3 of the subjects and in 7 days for one subject. Erythema was not associated with more severe pain or any functional impairment.
Fatigue and headache were the most common solicited systemic AEs following each vaccination, increasing in frequency from vaccination 1 to vaccination 3. The number of subjects experiencing fatigue and headache increased from 17% to 52% and 17% to 45%, respectively, following vaccination 1 to vaccination 3. Myalgia and arthralgia occurred in their highest frequency following vaccination 3, occurring in 41% and 21% of subjects, respectively. Gastrointestinal AEs, to include nausea, diarrhea and abdominal pain, (19%) and fever (10%) were also recorded most frequently following vaccination 3. All other systemic AEs occurred in 10% or less of the subjects following any vaccination. All solicited systemic AEs resolved within the 7-day follow-up period. There was one episode of fever that met the criteria for a grade 3 solicited systemic AE which resolved in less than 24 h. There appeared to be a trend towards increased numbers of solicited AEs associated with each subsequent vaccination. Throughout the vaccination period, eleven mild (grade 1) biochemical or hematologic laboratory adverse events were documented, none of which were determined to be related to vaccination. There were no grade 2 or 3 laboratory abnormalities during this timeframe.
This study represents only the second site to implement a P. vivax CHMI model which, unlike P. falciparum, requires mosquitoes that have been fed on blood directly obtained from an infected human donor rather than from in vitro cultured parasites. To ensure subject safety, donor blood was subjected to transfusion-grade screening for blood-borne infections as well as vector-borne infections. All study volunteers were screened to ensure they had normal concentrations of G6PD to prevent hemolytic anemia during radical cure therapy with primaquine. The challenge was well tolerated with no untoward reactogenicity following mosquito bites and 100% of the subjects that were exposed became parasitemic. No untoward SAEs were observed following the challenge and treatment phases.
On days 65 and 79 days post-CHMI, 2 subjects experienced P. vivax relapse [17] which we hypothesize was associated with their inability to metabolize PQ into sufficiently high enough concentrations of its active form leading to drug failure. One subject experienced a total of two relapses while the second subject experienced three relapses. No further relapses were observed up to the end of the study period, approximately 5 years post CHMI.
Seroconversion to VMP001 following the second vaccination was set as the progression criteria for proceeding to CHMI. Subjects were considered to have seroconverted if, at a serum dilution of 1:100, the OD414 of the test sample obtained two weeks post-2nd immunization was significantly different by paired t-test compared to serum obtained prior to 1st immunization (data not shown). All subjects seroconverted, with anti-VMP001 antibodies detectable in 100% of vaccinees at 2 weeks post vaccine dose 2.
Having met the progression criteria, subsequent ELISA data was reported as antibody titer, defined as the reciprocal of the serum dilution giving an OD414 of 1.0, and responses were measured at 2 weeks following each immunization as well as at 1 and 6 months post challenge. The highest geometric mean titer (GMT) of anti-VMP001 antibody were noted 2 weeks after dose 2 in cohorts 2 and 3 (74,608 and 61,711, respectively) and on the day of challenge (DOC; 2 weeks after dose 3) in cohort 1 (61,203) (Fig 4). Peak GMTs of anti-VMP001 antibody were not significantly different between the groups. While there was a decrease in antibody titers in all three cohorts in the weeks following the 2nd immunization, titers were boosted minimally (for cohort 3) to significantly (for cohort 1), following the 3rd immunization. As a result, GMTs were not statistically significantly different between cohorts 1, 2, and 3 on the DOC (Fig 4). Antibody titers showed a 5 to 8-fold decrease 6 months post-challenge compared to the pre-challenge titers. However, these titers remained significantly higher than those measured following the 1st immunization.
Antibody fine-specificities were evaluated to determine if the antibody responses were skewed to any specific region of the protein. Antibodies were detected to all regions of the protein, i.e. to the N-terminal, central repeat region as well as the C-terminal region. There were no significant differences in the GMT of anti-C term or anti-N term antibody between any groups on the DOC. The DOC GMT of anti-Type 1 repeat antibody was significantly higher in group 2 (GMT 16,554) compared to groups 1 and 3 (GMT 4,412 and 5,922 respectively) (Fig 5). There were no significant differences in GMT of anti-Type 1 repeat antibody between groups 1 and 3 on the DOC.
CD4+ T cell responses to VMP001 were detected in all individuals, with a majority (60%) showing peak response 14 days after the 2nd immunization. All vaccinated individuals produced IL-2, 93% produced TNF-α and 55% produced IFN-γ following stimulation with VMP001. Cytokine positive cells were predominantly IL2+ single positive or IL2+TNF-α+ double positive (Fig 6). Smaller frequencies of triple positive cells that also expressed IFN-γ were also detected (Fig 6). Cytokine profiles did not show marked differences post challenge. The responses were predominantly directed towards the N-term region (90% volunteers). Only 17% of vaccinated subjects responded to the repeat and C-term regions. There were no detectable CD8+ T cell responses in any volunteer at any of the time-points tested.
Following CHMI, all infectivity control subjects (100%) became infected. All 27 (100%) immunized subjects who underwent CHMI also became infected (Fig 7). Vaccine efficacy was 0%. The median prepatent period of all immunized subjects was 11.9 days and that of infectivity controls was 10.7 days. A significant delay in the median prepatency period was noted in all cohorts as compared to the infectivity controls (cohort 1, 10.9 days, p = 0.0261; cohort 2, 12.6 days, p<0.0001; cohort 3, 12.4 days, p = 0.0003). The median prepatency period for cohort 3 was significantly longer than that of cohort 1 (p = 0.035; Log-rank test); otherwise, there was no statistically significant delay in median prepatency periods between the other vaccinated groups.
Vaccinated subjects with a prepatency period greater than 2 standard deviations above the mean of the prepatency period of the infectivity controls were categorized as having a significant delay in the onset of parasitemia. The median time to parasitemia for the delayed group (n = 16) was 12.8 days and 10.9 days in the subjects without delay (n = 11) (Fig 8A).
On the day of the CHMI there was a significant correlation (r = 0.51, p = 0.006) in all vaccinated subjects between anti-type 1 peptide antibody titer and net time to parasitemia (time to parasitemia of vaccinated subjects—time to parasitemia of control subjects). The GMT of anti-type 1 repeat antibody was significantly greater in the delayed group (10,489, 95% CI = 6,154–17,878) as compared to the group who did not experience delay (4,279, 95% CI = 2,684–6,822) (p = 0.025, 2-tailed Mann Whitney t-test) (Fig 8B). There were no other significant differences in the GMT of anti-VMP001, anti-N term, or anti-C term antibody between the delayed and no delay groups.
No significant differences were identified between the cohorts in terms of CMI response at any of the time points and there was no association between CMI and delay in prepatency period.
The development of a controlled human malaria infection challenge model to evaluate P. vivax vaccines is challenging and fraught with numerous technical obstacles. While P. falciparum has been adapted for use in CHMI primarily because of the ability to grow mature gametocytes that infect Anopheles mosquitoes to produce infectious stage sporozoites, the same methodology cannot be used for P. vivax. To date, P. vivax has proven refractory to continuous in vitro culture. Therefore, infection of permissive Anopheline mosquito species (i.e. An. dirus and An. albimanus)—that themselves are difficult to maintain and/or infect in insectaries—relies on the identification and consent of naturally-infected patients to present to medical treatment facilities to donate blood for initiating infection in mosquitoes. Nevertheless, the successful implementation of P. vivax CHMI previously reported in Colombia has opened the door to test vaccine efficacy by incorporating an infected mosquito challenge [19]. In preparation for conducting a vaccine efficacy study we established a CHMI model for P. vivax at WRAIR, in collaboration with AFRIMS, using mosquitoes that were infected in Thailand (Chuang et al. in preparation.
Here we report the initial safety, immunogenicity, and efficacy data for the candidate vaccine VMP001/AS01B. This represents the first human efficacy study incorporating a CHMI for any P. vivax vaccine.
This vaccine formulation, administered in 3 increasing antigen doses (15 μg, 30 μg, or 60 μg) was well tolerated and the adverse event profile was consistent with other vaccines containing the AS01B adjuvant system [10,20,21].
The vaccine induced strong antibody and CD4+ T cell immune responses in all antigen dose groups. While all groups demonstrated greater than 50-fold boosting in antibody titers between the first and second immunization, only the low dose cohort showed a modest 2.8-fold increase in titers between the second and third immunization, thus matching the titers of the other two cohorts, which did not show any boost in antibody titers post third immunization. A possible explanation for the absence of antibody boosting following the second vaccination in the medium- and high-dose cohorts could be that the antibody titers were already saturated for these cohorts and either the vaccine dose or the range of intervals between the 2nd and 3rd vaccination (6 and 4 weeks, respectively) did not allow for sufficient enhancement in titers.
No subjects were sterilely protected following CHMI; however, all dose groups experienced a statistically significant delay in mean prepatency period as compared to the infectivity control group suggesting an anti-parasite effect elicited by the vaccine. A two day delay in prepatent period reflects a significant decrease in liver-stage parasites [22] [23], indicating that the immune responses generated by the vaccine is able to induce partial protection in vaccinated subjects. This would be consistent with the previously described correlation of high P. falciparum CSP repeat-specific antibody titer with sterile protection in malaria naïve adults [10] and children in endemic regions [11,12]. We have previously reported an efficacy study in Aotus monkeys that were immunized with VMP001 formulated in Montanide ISA 720 and CpG. Following an intravenous challenge a vaccine efficacy of 66.7% was observed and this protection was associated with anti-type 1 antibodies [24]. This observation supports the results observed in the current study. The lack of protection in humans may be due to a lower magnitude of anti-type 1 antibody titers in comparison to those observed in the Aotus study which was conducted with a different adjuvant formulation (Yadava, A. manuscript in preparation).
As we consider strategies to improve vaccine efficacy, alternate approaches, such as particulate delivery to improve immunogenicity and epitope-display, and well as alterations in schedule and dosing to improve qualitative and quantitative responses are points to ponder. We have developed CSV-S,S, a particulate formulation containing VMP001 which, similar to the P. falciparum CSP-based vaccine RTS,S, is co-expressed as a hepatitis B fusion particle. Analysis of the fine specificity of antibody responses in serum from rhesus monkeys immunized with CSV-S,S demonstrated significantly higher antibody response to the type 1 repeat peptide as well as greater responses to a smaller AGDR sequence within the type 1 peptide, suggesting that a particle formulation may improve the humoral response to the repeat sequences over soluble protein alone in the presence of adjuvants that are appropriate for human use [16].
The significant association of type 1 repeat-specific antibody responses with delay in prepatency period suggests that new vaccine strategies that enhance immune responses to this region might further improve vaccine efficacy against these strains of P. vivax. In addition to the modulation of immune responses to the repeat region by particulate formulations, an alternate strategy is to increase the number of repeat motifs in the vaccine construct. The resulting increase in epitope density may result in enhanced anti- repeat-specific responses.
Finally, alterations in schedule and dosing to optimize antibody affinity to protective epitopes may improve vaccine efficacy as has been reported for the RTS,S vaccine (Regules et al. in preparation).
The logistical difficulties of performing P. vivax CHMI have slowed the developmental efforts for a vaccine against this parasite. We demonstrate that this challenge model, although complex, is feasible and can provide rapid assessment of vaccine efficacy. An unexpected outcome from the CHMI in this study identified two subjects who experienced multiple relapses from latent hypnozoites parasites despite adherence the optimal radical cure therapy. The investigation into the cause and follow-up of these two subjects has been reported previously and identified an association between the cytochrome P450 isoenzyme 2D6 (CYP2D6) phenotype and the metabolism of PQ. It appears that CYP2D6 poor metabolizers are unable to convert the parent drug PQ into its active metabolite responsible for anti-hypnozoite activity and are, therefore, more likely to experience PQ failure and P. vivax relapse [13]. We propose that in addition to the Duffy blood group and G6PD testing, a laboratory screening test be used to characterize volunteers’ CYP2D6 genotype/phenotype in order to exclude subjects who, based on their genetic background, would be more likely to fail PQ therapy and experience relapse.
Decreases in P. falciparum malaria especially in Southeast Asia have not been associated with commensurate decreases in P. vivax malaria. Recent strategies to eliminate P. vivax by targeting the reservoir of latently infected patients with antimalarial 8-aminoquinolines alone are unlikely to achieve elimination because of both the safety and lack of active drug metabolites in a significant proportion of the population [13]. A highly protective and durable pre-erythrocytic CSP-based P. vivax vaccine would have a dual beneficial effect of preventing not only the initial infection but also secondary relapses from hyponozoites thus inhibiting the establishment of latent infection.
|
10.1371/journal.pgen.1000308 | Shaping Skeletal Growth by Modular Regulatory Elements in the Bmp5 Gene | Cartilage and bone are formed into a remarkable range of shapes and sizes that underlie many anatomical adaptations to different lifestyles in vertebrates. Although the morphological blueprints for individual cartilage and bony structures must somehow be encoded in the genome, we currently know little about the detailed genomic mechanisms that direct precise growth patterns for particular bones. We have carried out large-scale enhancer surveys to identify the regulatory architecture controlling developmental expression of the mouse Bmp5 gene, which encodes a secreted signaling molecule required for normal morphology of specific skeletal features. Although Bmp5 is expressed in many skeletal precursors, different enhancers control expression in individual bones. Remarkably, we show here that different enhancers also exist for highly restricted spatial subdomains along the surface of individual skeletal structures, including ribs and nasal cartilages. Transgenic, null, and regulatory mutations confirm that these anatomy-specific sequences are sufficient to trigger local changes in skeletal morphology and are required for establishing normal growth rates on separate bone surfaces. Our findings suggest that individual bones are composite structures whose detailed growth patterns are built from many smaller lineage and gene expression domains. Individual enhancers in BMP genes provide a genomic mechanism for controlling precise growth domains in particular cartilages and bones, making it possible to separately regulate skeletal anatomy at highly specific locations in the body.
| Every bone in the skeleton has a specific shape and size. These characteristic features must be under separate genetic control, because individual bones can undergo striking morphological changes in different species. Researchers have long postulated that the morphology of individual bones arises from the local activity of many separate growth domains around each bone's surface. Differential growth within such domains could modify size, curvature, and formation of specific processes. Here, we show that local growth domains around individual bones are controlled by independent regulatory sequences in bone morphogenetic protein (BMP) genes. We identify multiple regulatory sequences in the Bmp5 gene that control expression in particular bones, rather than all bones. We show that some of these elements are remarkably specific for individual subdomains around the surface of individual bones. Finally, we show that local BMP signaling is necessary and sufficient to trigger highly localized growth patterns in ribs and nasal cartilages. These results suggest that the detailed pattern of growth of individual skeletal structures is encoded in part by multiple regulatory sequences in BMP genes. Gain and loss of anatomy-specific sequences in BMP genes may provide a flexible genomic mechanism for modifying local skeletal anatomy during vertebrate evolution.
| The vertebrate skeleton is constructed of cartilage and bone tissues that are formed into highly specific shapes, sizes, and repeating arrays during normal development. Individual bones can show striking morphological specializations in different species, suggesting that separate genetic mechanisms must exist for regulating the growth of skeletal tissue at highly specific anatomical sites in the body [1],[2]. Despite the importance of skeletal structures for support, protection, eating, breathing, and movement, the detailed genetic mechanisms controlling the shape and growth of individual bones are still poorly understood.
Over fifty years ago, Bateman proposed that characteristic skeletal shapes are determined by varying patterns of differential growth and erosion that occur in stereotyped positions along the surfaces of each bone [3]. Localized growth at ends of a bone results in long straight structures. Uniform deposition around a bone produces uniform circumferential growth. Preferential deposition and erosion on opposite surfaces of a bone generates lateral displacement or curvature. Localized patches of deposition and erosion may also produce many of the specific processes, ridges, foramina, and articular surfaces that are characteristic of each bone in the body. Although highly localized patterns of deposition and erosion have long been proposed or visualized in the skeleton [4]–[6], little is known about how such stereotyped patterns may be encoded in the genome.
Previous studies demonstrate that secreted signaling molecules in the bone morphogenetic protein (BMP) family play a key role in both formation and repair of skeletal structures [7]. These molecules are expressed both in early skeletal precursors, and in the surface perichondrium and periosteum layers that surround growing cartilage and bone [8]–[13]. Pure recombinant BMPs can induce cartilage and bone formation when implanted at ectopic sites in animals [14],[15]. Conversely, mouse mutants missing members of the BMP family show defects in subsets of bone and cartilage elements. The classical mouse short ear locus encodes one of the mammalian BMP molecules (BMP5) [16]. Mutations at this locus reduce outer ear growth by disrupting the formation and activity of the surface perichondrium surrounding outer ear cartilage [17]. The same locus also controls the presence or absence of processes on specific vertebrae and the fibula, the morphology of the xiphoid process at the end of the sternum, the number of ribs along the vertebral column, and the total volume of the thoracic cavity [18]–[21].
A large number of spontaneous and induced short ear mutations suggest that the Bmp5 locus is surrounded by large regulatory regions required for developmental expression patterns in bones and other tissues [22]–[24]. Here we carry out detailed enhancer surveys to test the regional specificity of regulatory sequences controlling Bmp5 expression in skeletal tissues. Our studies suggest that stereotyped growth patterns along the surface of both ribs and nasal cartilages are controlled by highly specific “anatomy” elements in the Bmp5 gene. These modular enhancers in BMP genes may provide a flexible basis for encoding the detailed growth and form of specific bones in the vertebrate skeleton.
Previous regulatory scans of the Bmp5 locus identified several large regions that could drive expression of a lacZ reporter gene in developing skeletal structures [23],[24]. Expression in developing ribs was observed when two different bacterial artificial clones (BACs) covering non-overlapping regions of the gene (Figure 1A, E, H) were coinjected with a minimal heat shock-lacZ expression construct [24]. BAC199 includes most of the exons and introns of the Bmp5 gene. In contrast, BAC178 includes sequences from a large region three prime (3′) of all Bmp5 coding exons. Chromosome rearrangements in this 3′ region have generated two regulatory alleles of the Bmp5 locus (Bmp5se38DSD and Bmp5se4CHLd, Figure 1A). These alleles confirm that extensive 3′ sequences are required for normal expression and function of the endogenous Bmp5 gene [23].
To compare the lacZ expression in ribs driven by different BACs spanning the Bmp5 locus, we examined a series of coronal sections taken along the dorsoventral axis of the ribs of transgenic embryos beginning near the vertebral column and ending near the sternum. In dorsal sections from a BAC199-lacZ transgenic embryo, β-galactosidase activity was surprisingly restricted to a lateral domain within the rib perichondrium (Figure 1F). This pattern changed as sections progressed ventrally. In later sections, β-galactosidase activity was found in both the lateral and medial rib perichondrium (Figure 1G).
Interestingly, the pattern of lacZ expression controlled by the distal BAC clone, BAC178, was complementary to that seen with proximal BAC199. In dorsal sections from a BAC178-lacZ transgenic embryo, β-galactosidase activity was found in anterior, medial and posterior domains of the rib perichondrium (Figure 1I). More ventral rib sections showed loss of medial expression but retained β-galactosidase activity predominantly in anterior and posterior rib perichondrium (Figure 1J). Thus, BAC178-lacZ rib expression complements BAC199-lacZ rib expression as it changes along the dorsoventral axis (Figure 1F, I and G, J). Taken together, these results suggest that gene expression in different domains of the rib perichondrium is controlled by distinct regulatory elements in the Bmp5 locus. Notably, the complementary rib regulatory regions are separated by over 100 kilobases (kb) (Figure 1A).
Analysis of endogenous Bmp5 expression in wild-type and Bmp5se38DSD regulatory mutants confirms the existence of distinct control regions for different domains of the rib perichondrium. The Bmp5se38DSD regulatory mutation derives from a chromosomal rearrangement whose breakpoint lies near the Bmp5 coding exons [22]. Therefore, this rearrangement is predicted to remove all distal rib control sequences (Figure 1A). In situ hybridization analysis of Bmp5 transcripts in dorsal rib sections shows reduction of anterior, medial and posterior rib domain expression within Bmp5se38DSD ribs as compared to wild-type (Figure 1C, D). In contrast, strong Bmp5 expression is still seen in the lateral rib perichondrium (asterisk in Figure 1D), as expected given the location of lateral control elements upstream of the Bmp5se38DSD breakpoint. Therefore, general Bmp5 expression in rib perichondrium appears to be a composite of smaller, independently regulated expression domains.
To further characterize and localize Bmp5 regulatory sequences, we used sequence alignment programs PipMaker and LAGAN/VISTA to compare human and mouse Bmp5 loci [25]–[27]. This approach revealed numerous evolutionarily conserved non-coding regions (ECRs) scattered across the Bmp5 locus (Figure 2, Figure S1). We then cloned multiple small genomic fragments containing single or multiple ECRs upstream of a minimal heat shock-lacZ reporter cassette, injected them into fertilized mouse eggs, and scored expression patterns in transgenic embryos. The survey of putative enhancer regions extended across the entire 400 kb interval detailed in Figure 1A (see Figure 2A and Figure S1). A 6.2 kb clone from the BAC199 region including 4 ECRs surrounding Bmp5 exon 4 (Ex4r) drove reproducible expression in nasal cartilages, distal limbs, and ribs (Figure 2B). As with the BAC transgenics, a series of coronal sections was taken along the ribs of Ex4r-lacZ transgenics. Dorsal rib sections again revealed expression in a restricted domain along the lateral surface of the developing rib (Figure 2C). To further characterize this peripheral surface domain we hybridized adjacent sections with molecular markers for perichondrium (type I collagen, Col1a1), chondrocytes (type II collagen, Col2a1, and type X collagen, Col10a1), and developing muscle (MyoD1) [28]–[30]. The lacZ-positive region corresponds to a particular sector of the surface perichondrium surrounding the ribs, which otherwise extends in a continuous circle around the developing rib cartilage (Figure 2C–F). Unlike the BAC199-lacZ transgene, the Ex4r sequence did not drive discrete localized expression in a medial perichondrium domain in more ventral rib sections (data not shown). These results demonstrate that anatomical control sequences for lateral perichondrium expression map within the exon 4 region, and that additional sequences are required for the medial rib expression seen with BAC199.
To further narrow the region required for lateral perichondrium expression, we tested a series of smaller genomic fragments and deletions of conserved ECRs from within the Ex4r subclone (Figure S2). This analysis demonstrated that the core sequences necessary for lateral perichondrium expression reside in a 1069 bp peak of conservation at the 3′ end of the Ex4r region, and that other sequences in the Ex4r construct are required for expression in limbs and nasal cartilages.
Bmp5 is expressed in the perichondrium surrounding many other skeletal structures, including the nasal septum and the shelf-like turbinates that project into the nasal cavity [11]. In addition, new micro computerized tomography (MicroCT) analysis of wild-type and Bmp5 mutant skulls shows that the Bmp5 gene is also required for normal development of turbinates in the anterior nasal region (Figure 3A, B), and for normal branching patterns in more posterior nasal regions (Figure 3C, D). To determine whether Bmp5 expression in nasal cartilages is also controlled by separable regulatory sequences, we examined the nasal region in both BAC199-lacZ and Ex4r-lacZ transgenic embryos. The larger BAC199 clone showed widespread β-galactosidase activity throughout the nasal cartilages, including multiple turbinates and the nasal septum (data not shown). In contrast, Ex4r-lacZ transgenics showed activity restricted to a small arc-like domain located on the inner surface of nasal cartilage between turbinate shelves in the anterior nasal cavity and along the neck of the developing turbinates (Figure 3E). Like the lateral rib expression, turbinate expression was seen predominantly in subregions of the surface perichondrium (Figure S3A, C). No expression was seen in the nasal septum, in posterior cartilages or at the tips of the shelf-like projections of the turbinates themselves. Testing fragments of the Ex4r clone demonstrated that sequences directing restricted nasal cartilage expression and restricted lateral rib perichondrium expression are distinct (Figure S2).
Examination of other regions of the Bmp5 locus known to contain skeletal enhancers identified an additional non-overlapping sequence that also gives expression in nasal cartilages. A 17 kb clone (Phage 7 in Figure 1) previously reported to give thyroid cartilage expression [23] also showed strikingly specific nasal cavity expression. β-galactosidase activity was seen at the dorsal tips of the expanding turbinate shelves, colocalizing with Col2a1 in proliferating chondrocytes, but was absent from the ventral tips, the turbinate necks, and the cartilages between turbinate shelves (Figure 3F, Figure S3B, F), a pattern partially complementary to that driven by the Ex4r construct.
The sequences included in Phage 7 are located approximately 100 kb 3′ of the chromosome breakpoint in the Bmp5se38DSD regulatory mutation (Figure 1A). Endogenous Bmp5 expression is dramatically reduced in the dorsal tips of turbinate shelves in Bmp5se38DSD mice compared to wild-type (Figure 3G, H), as well as in the cribriform plate, the structural roof of the nasal cavity. These data confirm that 3′ regulatory sequences are required for Bmp5 expression in the tips of turbinate shelves, but not in the surrounding neck and inter-turbinate perichondrium. Bmp5se38DSD mutant mice also show defects in the cribriform plate and branching alterations in nasal turbinates (data not shown). Thus, in both ribs and nasal cartilage, an apparently continuous layer of perichondrium consists of distinct expression domains controlled by separate regulatory elements in the Bmp5 gene.
Ribs are derived from somitic mesoderm [31]. Previous chick-quail lineage tracing experiments have shown that rib cells arise from different portions of developing somites: the head, neck, and the inner surface of ribs are derived from the posterior compartment of somites (white regions in Figure 4A), while the lateral surface of the mid shaft of the rib arises from the anterior compartment of somites (blue regions of Figure 4A) [32]. We noticed that lacZ expression driven by the Ex4r construct begins some distance from the vertebral column, and is strongest along the midshaft of ribs (Figure 4B), a pattern reminiscent of the anatomical domain thought to arise from anterior somites.
To analyze rib enhancer activity at additional developmental stages, we generated stable transgenic lines for the Ex4r-lacZ construct and collected embryos beginning at embryonic day 10.5. At this early stage of development, the Ex4r-lacZ construct is expressed in the anterior halves of developing somites (Figure 4C). Examination of lacZ localization in rib sections at later stages showed that Ex4r-driven expression was largely missing from the head and neck region of ribs, was present in the lateral perichondrium along the rib shaft, and became symmetric around the rib in sternal portions (Figure 4D–F). Both the somitic expression and the changes in patterns along the length of ribs suggest that the lateral rib expression reflects the dual origin of ribs from separate somite compartments.
Multiple BMP family members are expressed in overlapping patterns in the developing ribs [12]. To test the biological effects of localized increase or decrease in BMP signaling in subdomains of the rib perichondrium, we used the Ex4r sequence to drive the expression of either a constitutively active (caBmprIb) or dominant negative (dnBmprIb) version of BMP receptor IB [33]. We chose BmprIb because it is widely expressed throughout the developing skeleton, including rib perichondrium, and is known to be used by multiple BMP ligands [33]–[38]. Each receptor construct was coinjected with the original Ex4r-lacZ clone to generate transgenic embryos.
Both Ex4r-caBmprIb and Ex4r-dnBmprIb transgenic embryos showed gross changes in rib development at E14.5 when examined by whole-mount skeleton preparations (Figure 5A–C). Increased BMP signaling in the lateral rib domain caused an overgrowth of alcian blue-positive cartilage, beginning midway along the rib shaft (Figure 5B arrow). Sections through Ex4r-caBmprIb embryos that were assayed for β-galactosidase activity showed that rib expansion was accompanied by an excess of lacZ-positive cells in the lateral rib (Figure 5E). This lateral expression marked the outer edge of the rib deformation (Figure S4A, C) and overlays a cartilaginous mass of cells made up predominantly of hypertrophic chondrocytes expressing Col2a1 and Col10a1 (Figure S4E, G) [28],[29]. In contrast, decreased BMP signaling caused a marked deflection in rib trajectory (Figure 5C bracket). Ribs in Ex4r-dnBmprIb transgenic embryos emerged normally from the vertebral column but were deflected inwards along the central region of the rib shaft, producing a more constricted upper thorax. This deformation in trajectory was not accompanied by changes in rib cross section (Figure 5F, Figure S4). Neither construct affected the head or neck of the ribs (Figure 5B, C), as expected from the restricted expression domain of Ex4r control sequences along the rib shaft (Figure 4B).
The highly localized domains of Bmp5 expression in rib perichondrium are reminiscent of previous models suggesting that rib growth occurs by differential activity on the lateral and inner surfaces of the rib [3]. To visualize in vivo patterns of bone deposition in growing ribs, we injected mice twice, at 6 and 7 weeks, with calcein, a fluorescent dye that specifically incorporates into newly formed bone (Figure 6). Dorsal rib cross-sections showed two major growth domains labeled with dye; one visible along the lateral periosteal surface (D1), and a second predominantly along the anterior, medial, and posterior endosteal surfaces of the rib (D2). Each bone deposition front is represented by two calcein labelings, reflecting the two separate injections (Figure 6A). Injections with different dye colors demonstrate that the bone fronts labeled by the initial injection (arrows in Figure 6A) become embedded in bone after a week of growth, and that the new bone fronts labeled by the second injection are found near surfaces (arrowheads in Figure 6A, data not shown). These deposition patterns show striking asymmetry, with bone deposition occurring preferentially in the lateral domain of the outer surface periosteum and in the anterior, medial, and posterior domains of the inner endosteum.
To compare patterns of bone deposition and bone resorption, dorsal rib cross-sections were also examined for tartrate-resistant acid phosphatase activity, an osteoclast marker [39]. Bone resorption was also highly asymmetric, and complementary to the areas of bone deposition (Figure 6B). In the outer periosteum, osteoclast activity was most intense on the anterior, medial, and posterior surfaces of the rib; and was nearly absent along the lateral surface where major bone deposition was occurring. Likewise, along the inner endosteum, osteoclast activity was most intense on the lateral wall, and nearly absent from the anterior, medial and posterior surfaces. During growth, these highly asymmetric patterns of bone deposition and resorption would result in the net lateral displacement of ribs and the expansion of the intrathoracic cavity, while preserving marrow space.
Bmp5 mutant mice are known to have a smaller thoracic volume than wild-type animals [21]. To further characterize detailed bone deposition patterns in Bmp5 mutants, we performed dual calcein injections on Bmp5 null and Bmp5 regulatory mutants, and measured the amount of bone deposition in the different rib domains described above (Figure 6C). Mice with null mutations in the Bmp5 gene show a significant reduction in bone deposition in both major ossification domains, D1 and D2 (Figure 6C). In contrast, regulatory mutant mice missing anterior, medial and posterior but not lateral rib control sequences (Bmp5se4CHLd, Figure 1A) show significantly reduced bone deposition in D2, but not in D1 domains. The Bmp5 gene is thus required for normal rates of bone growth on both the outer and inner surface of the rib, and these two growth domains are controlled independently by different regulatory regions of the Bmp5 locus.
It has long been recognized that cartilage and bone can be molded into a remarkable range of different shapes and sizes. Previous genetic studies show that the morphology of different skeletal elements is controlled by multiple independent genetic factors [2],[40],[41]. Based on studies of jaw and limb morphology in mice, Bailey previously suggested that different subregions of a single bone must be controlled by a large number of independent “morphogenes”, each active in small patches along the surface of a single bone [2]. Despite recent progress identifying genes that regulate formation of all cartilage or all bones, or genes that control skeletal formation in different subdomains along the body axis, little is known about the fine-grained mechanisms that control detailed growth patterns of individual skeletal elements [42]. Here we show that highly defined growth domains in particular bones are controlled by remarkably specific enhancers in the Bmp5 gene (Figure 7). We propose that anatomy-specific enhancers in BMP genes provide a genomic mechanism for independent developmental control of local growth along discrete domains of individual cartilages and bones in the vertebrate skeleton.
When BMP genes were first discovered and assayed for expression in vertebrates, individual members of the family were initially proposed to promote general steps in the differentiation of all skeletal tissue [8]. Although Bmp5 is expressed in a continuous fashion in the perichondrial layer surrounding many developing skeletal structures [11],[12], our enhancer surveys do not show evidence for general enhancers in the Bmp5 gene that drive expression around the surface of all cartilage or all bones. Instead, distinct Bmp5 enhancers regulate expression in individual skeletal structures. Furthermore, separate enhancers also exist for discrete domains around the surface of individual bones, including lateral, anterior, medial, and posterior domains of the rib perichondrium, and tip versus neck and inter-turbinate domains in the nasal cartilages (Figures 1, 3). This remarkably fine control of gene expression is clearly sufficient to alter skeletal morphology at specific locations (Figure 5). Null and regulatory mutations also show that the Bmp5 gene is necessary for normal bone deposition rates along particular surfaces of growing ribs (Figure 6). These results confirm that detailed growth patterns in an individual bone can be encoded by highly specific anatomy enhancers in genes for bone morphogenetic proteins.
Previous studies of HOX genes have shown that expression and function at particular anatomical locations in the body are related to the physical location of genes along the chromosome [43]–[45]. The overall correlation between anatomy and gene position may arise from progressive changes in chromatin structure during body axis development; or from proximity to enhancers that map outside the HOX complex, which have decreasing effects on genes that map at increasing physical distances from the enhancer [45],[46]. In contrast, the Bmp5 skeletal enhancers we have identified to date show no obvious relationship between anatomical position in the body and physical location within the Bmp5 locus. The regulatory elements for discrete surface domains around a single bone clearly map to different regions of the Bmp5 gene (Figure 1). In addition, rib and nose enhancers are interspersed with each other (Figure 7) and with other separate enhancer regions previously identified controlling expression in the sternum, thyroid cartilage, lung, and genital tubercle [23],[24].
The dispersed enhancer pattern seen in Bmp5 may reflect the different roles of BMP and HOX genes in skeletal patterning. Nested sets of HOX gene expression are evolutionarily ancient programs used to pattern basic body axes in both vertebrates and invertebrates [44],[45],[47]. In contrast, both cartilage and bone are more evolutionarily recent, vertebrate-specific tissues that vary widely in form from species to species [1]. For example, respiratory nasal turbinates are thought to have arisen separately in bird and mammals to help conserve water during breathing [48]–[50]. They vary widely in branching structure within mammals, and are reduced or absent in fish, amphibians, and reptiles [48],[51]. Since a variety of studies suggest that BMPs are the endogenous signals used to induce cartilage and bone in vertebrates [7], formation of nasal turbinates and other species-specific skeletal structures presumably occurs through cis- or trans-acting alterations that produce local changes in BMP expression at particular sites in the body. Therefore, the complex architecture of skeletal enhancers in the Bmp5 gene may reflect a historical process of piecemeal gain and loss of regulatory elements controlling local domains of BMP expression.
How are the remarkably specific domains of Bmp5 expression generated along the surface of ribs or nasal cartilages? A variety of data suggests that mechanical forces can give rise to highly localized patterns of bone deposition and erosion [52],[53]. For example, rib cages and skulls both enclose rapidly growing tissues. Outward pressure from soft tissue growth may lead to bone deposition on skeletal surfaces under mechanical tension (the convex outermost surface of ribs or cranial bones), and bone erosion on surfaces under compression (the innermost surface of ribs or cranial bones). Although mechanical tension and compression are clearly coupled to bone remodeling, we do not think that the restricted patterns of expression we observe for Bmp5 enhancers are simply responding to the distribution of mechanical forces on growing skeletal structures. First, there is no obvious relationship between mechanical forces and the contrasting tip and neck expression patterns seen in nasal cartilages. Second, the Bmp5 enhancer that drives expression along the outer surface of ribs is not similarly expressed along the outer surface of either the sternum or the skull, although these bones should be subject to similar mechanical forces from the rapid expansion of underlying tissue. Third, the Ex4r-lacZ construct that drives highly localized patterns of expression in growing ribs also drives compartmentalized expression in developing somites (Figure 4). These results suggest that the remarkably specific Bmp5 domains in ribs are related to the dual origin of ribs from different somite compartments, rather than to simple mechanical forces acting during later growth and expansion of the thoracic cavity.
Previous lineage tracing experiments have shown that the lateral edges of rib shafts are derived from cells in the anterior half of somites [32] (Figure 4A). Response elements for anterior somite transcription factors could provide a simple mechanism for controlling mid-shaft Bmp5 expression in the lateral perichondrial domain. Conversely, response elements for posterior somite expression could provide another simple mechanism for regulating Bmp5 expression in rib head and necks, and in the anterior, medial, and posterior perichondrial domains along the rib shaft, similar to the patterns seen with BAC178 (Figure 1). The current sizes of Bmp5 rib enhancers are still too large to identify particular binding sites for upstream factors. However, future narrowing of the minimal sequences capable of driving rib domain expression may make it possible to link specific somite transcription factors with the different domains of rib expression identified in this study.
The dual origin of axial structures from anterior and posterior halves of adjacent somites produces vertebrae and ribs that form one half segment out of register with the original metameric pattern seen in somites. The functional significance of this shift has been debated for over a hundred years [31], [54]–[56]. Resegmentation causes axial muscles, and many of their origin and insertion points on adjacent vertebrae and ribs, to all be derived from a single somite. Our studies suggest resegmentation also plays a key role in establishing detailed growth patterns in developing ribs (Figure 7). Although ribs are usually thought of as simple tubular structures, they can be extensively modified in different organisms to produce the diverse cross-sectional shapes, as well as the varied curvatures seen in wide- and narrow-bodied animals [1],[57]. It has long been recognized that differential deposition on the lateral surface of ribs must underlie the expansion and ultimate shape of ribs and thoracic cavities [3]. We suggest that resegmentation helps establish the lineage domains that make it possible to independently control cartilage and bone growth in specific rib surface domains. The multiple enhancers present in BMP genes provide an elegant mechanism for linking such lineage domains to actual sites of bone growth, leading to highly detailed patterns of deposition that can be independently controlled along the length and around the circumference of a single bone.
While lineage domains may be used to produce separate lateral versus medial domains of gene expression in developing ribs, we think additional mechanisms must be operating to produce other highly localized patches of Bmp5 expression. For example, our comparison of BAC199 and BAC178 expression suggests at least four different expression domains may exist at certain positions along the ribs (lateral, medial, anterior, posterior; Figure 1). Control sequences for the lateral domain have been mapped to a single 1069 bp peak of sequence conservation within the Bmp5 Ex4r region, but additional sequences responsible for expression in the other domains remain to be identified in the larger regions covered by BAC199 and BAC178.
Highly localized expression patterns are also seen in multiple spatially restricted patches along the necks and tips of nasal cartilages (Figure 3). The elements controlling these patches are distinct from those controlling rib expression. In addition, nasal cartilage development is quite different from rib morphogenesis (Figure 7). For example, the facial bones and cartilages are derived from cranial neural crest that migrates from positions in the developing brain [58]–[60]. HOX genes are not expressed in this cranial region, and transplantation studies have demonstrated a remarkable degree of plasticity in the cranial neural crest populations [61],[62]. Patterning signals are thought to emerge from the local endoderm and ectoderm to control the shape and size of individual facial skeletal structures [61],[63]. Therefore, unlike ribs, we currently know of no lineage compartments that can account for the various separate tip and shelf domains seen during the branching morphogenesis of nasal cartilages.
In Drosophila, branching morphogenesis takes place during tracheal airway development, and is controlled by numerous local patches of breathless/FGF expression. The specific enhancers controlling breathless expression near tips of growing trachea branches have not been isolated, but may respond to different combinations of transcription factors that are themselves expressed in local or intersecting patterns [64]. Multiple, locally acting enhancers in BMP and FGF genes may thus represent a common molecular strategy for molding skeletal tissue or trachea airways into particular shapes in different animals [7],[64].
Further studies of anatomy-specific elements in BMP genes should lead to a molecular understanding of the type of “morphogenes” that have long been postulated to control local growth decisions in different subdomains of particular bones [2]. In addition, gain and loss of regulatory elements in BMP genes may provide a simple genomic mechanism for evolutionary modification of skeletal structures. While null mutations in BMP genes often have pleiotropic defects, adaptive changes in specific regulatory sequences could localize effects to particular skeletal structures, making it possible to alter vertebrate anatomy while preserving viability and fitness [7]. This possibility has taken on renewed interest in light of studies linking changes in BMP expression to different beak shapes in naturally occurring bird species [65]–[67], and to different jaw morphologies in African cichlids [68]. Regulatory lesions are difficult to identify, and it has not yet been possible to track particular bird or fish anatomical changes to specific DNA sequence alterations in BMP genes. Nonetheless we think the kind of modular regulatory architecture we have found for the Bmp5 gene probably exists around many other members of the BMP family [69],[70]. Isolation and characterization of additional anatomy elements from BMP genes will make it possible to test whether anatomical changes in naturally occurring species result from structural and functional modifications in the type of modular enhancer regions identified in this study.
Regulatory (Bmp5se38DSD, Bmp5se4CHLd) and null (Bmp5null) alleles were described previously [11],[22],[23]. All strains used for bone growth assays are on the C57Bl/6J background. The generation of BAC199-lacZ, BAC178-lacZ and Phage7-lacZ transgenics was reported in [23],[24]. All new DNA constructs were prepared for microinjection as previously described [24]. The Ex4r-caBmprIb and Ex4r-dnBmprIb plasmids were coinjected with the Ex4r-lacZ clone at a 4∶1 molar ratio. Pronuclear injection into FVB embryos was carried out by the Stanford Transgenic Facility and Xenogen Biosciences in accordance with protocols approved by the Stanford University Institutional Animal Care and Use Committee.
BAC426K2 (Genbank accession #AC079245) and BAC343K17 (Genbank Accession #AC079244) were isolated from the RPCI-23 Female (C57Bl/6J) Mouse BAC library (Invitrogen) using a 1334 bp EcoRI probe and/or a 591 bp HaeIII probe located 123,536 bp and 225,112 bp, respectively, from the Bmp5 transcriptional start site. Sequences were compiled following designation of BAC426K2 and BAC343K17 as clones of high biomedical interest by the National Human Genome Research Institute and sequencing by the Advanced Center for Genome Technology at the University of Oklahoma, Norman. 5′ mouse sequences were added from BAC429A10 (Genbank accession #AC144940) as they became available. Human Bmp5 genomic sequence was compiled from the following clones: Genbank accession numbers AL589796, AL137178, AL133386, AL590290, AL590406 and AL592426. The human and mouse Bmp5 sequences were masked using RepeatMasker (A.F.A. Smit, R. Hubley and P. Green, unpubl.; http://www.repeatmasker.org/). ECRs were identified using global sequence alignment programs as previously described [69].
The Ex4r-lacZ plasmid was generated by amplifying a 6221 bp fragment corresponding to mouse sequences 93,656–99,876 bp in Figure 2 using primers 622: 5′GGATTGCGGCCGCTATGGACAGCTTTGAAGAGCTTTGGTA3′ and 624: 5′GGATTGCGGCCGCTATTCTAGCCTCTCCTGTAGGATTATG3′. Following NotI digestion, the fragment was cloned into the Not5'hsplacZ vector [23].
To generate Ex4r-caBmprIb and Ex4r-dnBmprIb constructs, the constitutively active (ca) or dominant negative (dn) form of BmprIb was amplified using primers lpf21: 5′CATGCCATGGCCATGCTCTTACGAAGCTCTGGAAAAT3′ and lpf22: 5′GCTCTAGAGCTTAGATCCCCCCTGCCCGGTTATTATTATCAGAGTTTAATGTCCTGGGACTCTG3′. The PCR products were digested with NcoI/XbaI and cloned into the Ex4r-lacZ plasmid that had been digested with NcoI and XbaI (partial), replacing the lacZ cassette.
To generate plasmid Ex4rCD-lacZ, a 3 kb fragment was amplified from the Bmp5 BAC426K2 using primers 624 (above) and 627: 5′GGATTGCGGCCGCTATTCTAGGCTGTTGGAAAGCAAGTCTA3′. The PCR product was digested with NotI and cloned into Not5'hsplacZ.
Construct Ex4rΔC-lacZ was generated using primers 750: 5′ATGTGGCCAAACAGGCCTATTAATGGTCAACCAGATGAATACAGCA3′ and 751: 5′ATGTGGCCTGTTTGGCCTATTATAGAACACATAGAGGCATACCAGG3′ to amplify directly from the Ex4r-lacZ plasmid using the Expand Long Template PCR system (Roche #1681834). The 12.9 kb product was digested with SfiI and the free ends were ligated together. The 5488 bp insert with a 733 bp deletion of ECR C from Ex4r was removed by NotI digestion and recloned into an unamplified Not5'hsplacZ vector.
The Ex4rΔD-lacZ plasmid was generated by amplifying 4677 kb and 454 bp products from BAC426K2 using primer 622 (above) with primer 754: 5′ATGTGGCCTGTTTGGCCTATTCCTTTTGAGAATCTCGGCTTCTAGA3′ and primer 752: 5′ATGTGGCCAAACAGGCCTATTGGCAGGTTAGAGAAAGTAATGATAG3′ with primer 624 (above), respectively. The PCR products were digested with SfiI and ligated together. The resulting 5.1 kb product containing a 1069 bp deletion of ECR D from Ex4r was digested with NotI and ligated into the Not5'hsplacZ vector.
Plasmid ECRD-lacZ was generated by amplifying a 1127 bp fragment from BAC426K2 using primers 690: 5′ATGTGGCCTGTTTGGCCTATTCTTTCTCTAACCTGCCTCTACCCTG3′ and 736: 5′ATGTGGCCAAACAGGCCTATTGAAGCCGAGATTCTCAAAAGGTGGA3′. The PCR product was digested with SfiI and ligated into pSfi-hsplacZ [69].
Embryos collected by Xenogen Biosciences were fixed for 1 hour in 4% paraformaldehyde in 1× PBS at 4°C, placed in cold 1× PBS and shipped overnight on ice.
Whole-mount staining for β-galactosidase activity was performed as described [69] with the following modifications: Embryo fixation times varied with age (E10.5 for 30 minutes, E13.5 for 75 minutes, E14.5 for 90 minutes). E13.5-E14.5 embryos were hemisected after 1 hour. Rib and nasal cartilage cryosections from lacZ whole-mount embryos were collected and counterstained as described [69]. Prior to embedding, samples were equilibrated in embedding solution (15% sucrose, 7.5% gelatin (300 Bloom, Sigma #G2500) in 1× PBS) for 1 hour at 42°C. Ex4r-caBmprIb and Ex4r-dnBmprIb transgenic embryos were frozen in OCT compound (Tissue Tek), cryosectioned at 25 microns and counterstained with Nuclear Fast Red (Vector labs, #H-3403).
β-galactosidase activity on cryosections was assayed by fixing samples in 4% paraformaldehyde in 1× PBS for 5–8 minutes at room temperature. Slides were rinsed 3 X 5 minutes with 1× PBS, washed in lacZ wash buffer (0.1 M sodium phosphate buffer (pH 7.3), 2 mM MgCl, 0.01% deoxycholate, 0.02% Nonidet P-40) for 10 minutes and incubated in lacZ stain (wash buffer supplemented with 4 mM K3Fe(CN)6, 4 mM K4Fe(CN)6⋅3 H2O, 0.1M Tris (pH 7.4) and 1 mg/mL X-gal (Sigma #B4252)) at 37°C for at least 24 hours. Stained sections were rinsed with 1× PBS, fixed for an additional 10 minutes in 4% paraformaldehyde in 1× PBS, and counterstained with Nuclear Fast Red.
The Bmp5, Col10a1, and Col2a1 probes used were described [23],[29],[71]. The MyoD1 probe was generated from a clone ordered from Open Biosystems (clone id 372340). The Col1a1 probe was generated using the pMColI-Bam plasmid (a gift of Dr. Ernst Reichenberger). Timed matings were performed to collect wild-type (C57Bl/6J), Ex4r-lacZ, Phage7-lacZ, and Bmp5se38DSD mutant heads and/or torsos at E13.5-E15.5. Ex4r-caBmprIb and Ex4r-dnBmprIb embryos generated by Xenogen Biosciences were collected at E15.5 and fixed for 1 hour in 4% paraformaldehyde in 1× PBS, bisected, and one half embryo embedded in OCT and one half analyzed for β-galactosidase activity to identify transgenic embryos. 12 micron sections were collected from samples frozen in OCT compound and analyzed for gene expression as previously described [72]; except that the color reagent BM purple (Roche #1442074) was used in place of NBT/BCIP.
E14.5 skeletons were prepared as described [73], with the following modifications: Embryos were placed directly into staining solution after ethanol dehydration. Following potassium hydroxide treatment, embryos were cleared in 50% glycerol overnight, and then stored in 100% glycerol. All steps were done at room temperature.
Two successive intraperitoneal injections of calcein (Sigma # C0875, 2.5 mg/ml in 1× PBS) were performed at postnatal day 43 (p43) and p51 on C57Bl/6J males (10 mg injected/kg body weight). Whole rib cages were collected at p53 and dehydrated in ethanol for at least 1 week at 4°C, then embedded in methylmethacrylate and ground sectioned to obtain 50 micron coronal sections by HMAC (Birmingham, AL). To quantify levels of bone deposition in wild-type and mutant animals, calcein labeled rib cages from six males of each category (C57Bl/6J, Bmp5se4CHLd and Bmp5null) were equilibrated overnight in 15% sucrose in 1× PBS and at least 24 hours in 30% sucrose in 1× PBS, all at 4°C. Rib cages were bisected, and the right half was embedded in OCT. Six 50 micron coronal cryosections were taken approximately 1 mm apart, beginning at the growth plate and moving dorsally. Each section was digitally photographed, and pixel areas between labeled bone deposition fronts were measured with Photoshop. All measurements were taken on the fifth rib. Data are expressed as mean areas±s.e.m. relative to wild-type mice. Differences between groups were evaluated using Student's t-test.
C57Bl/6J male rib cages were collected at p53 into cold 1× PBS, fixed in 4% paraformaldehyde in 1× PBS for 3 days at 4°C, and washed 3 times for 30 minutes in cold 1× PBS. The right halves were embedded in paraffin, sectioned, and stained by HMAC [74].
Scans from 4 wild-type and 5 Bmp5null mutant skulls, aged 4 weeks postnatally, were generated using a Scanco MicroCT-40 operated at a tube potential of 45 kV and tube current of 177 microA using a 0.30 second integration with 2× averaging. All samples had undergone skeletal preparation prior to scanning.
|
10.1371/journal.pgen.1007203 | Chinmo prevents transformer alternative splicing to maintain male sex identity | Reproduction in sexually dimorphic animals relies on successful gamete production, executed by the germline and aided by somatic support cells. Somatic sex identity in Drosophila is instructed by sex-specific isoforms of the DMRT1 ortholog Doublesex (Dsx). Female-specific expression of Sex-lethal (Sxl) causes alternative splicing of transformer (tra) to the female isoform traF. In turn, TraF alternatively splices dsx to the female isoform dsxF. Loss of the transcriptional repressor Chinmo in male somatic stem cells (CySCs) of the testis causes them to “feminize”, resembling female somatic stem cells in the ovary. This somatic sex transformation causes a collapse of germline differentiation and male infertility. We demonstrate this feminization occurs by transcriptional and post-transcriptional regulation of traF. We find that chinmo-deficient CySCs upregulate tra mRNA as well as transcripts encoding tra-splice factors Virilizer (Vir) and Female lethal (2)d (Fl(2)d). traF splicing in chinmo-deficient CySCs leads to the production of DsxF at the expense of the male isoform DsxM, and both TraF and DsxF are required for CySC sex transformation. Surprisingly, CySC feminization upon loss of chinmo does not require Sxl but does require Vir and Fl(2)d. Consistent with this, we show that both Vir and Fl(2)d are required for tra alternative splicing in the female somatic gonad. Our work reveals the need for transcriptional regulation of tra in adult male stem cells and highlights a previously unobserved Sxl-independent mechanism of traF production in vivo. In sum, transcriptional control of the sex determination hierarchy by Chinmo is critical for sex maintenance in sexually dimorphic tissues and is vital in the preservation of fertility.
| Sexually dimorphic adult tissues, like ovaries and testes, require continuous sex-specific instruction for proper function. Establishment of female somatic sex identity in Drosophila is controlled by an alternative splicing cascade wherein Sex-lethal (Sxl) produces the female-specific protein TransformerF (TraF). By contrast, males lack Sxl and undergo default splicing, preventing TraF production. Since TraF expression in males causes sex transformation and impairs tissue function, males must have evolved robust protection against feminization. Here, we investigate the role of a single factor, Chinmo, in protecting male sex identity in the testis: loss of Chinmo in male somatic stem cells causes them to acquire female identity. We demonstrate that this feminization occurs through the induction of TraF and its downstream targets. Surprisingly, Sxl is not induced in these sex transformed cells. Instead, two other alternative splice factors, Virilizer and Female lethal (2)d, are enriched in chinmo-mutant somatic cells and are required for their feminization. Our work demonstrates that transcriptional repression of female-biased alternative splice factors prevents sex transformation in the somatic gonad and that traF production can occur independently of Sxl. Given the importance of sex maintenance in tissue homeostasis, such protective mechanisms may exist in other tissues.
| Sexual dimorphism, or the differences between male and female individuals in a species, is observed in many organisms, including insects, reptiles, and mammals. Sex-specific tissue development is essential for proper gonadogenesis, and sexual dimorphism has also been observed in other tissues such as brain, adipose tissue, and intestine [1–4]. While extensive literature has dissected the mechanism of sex determination in early development, recent studies have demonstrated that maintenance of sex identity is also essential for adult tissue homeostasis [5–7]. It is therefore critical to determine the signals that both specify and maintain sex identity.
Differential gene expression via alternative splicing establishes the sex-specific differences observed in the fruit fly Drosophila melanogaster. In flies, the sex of an organism is determined by its number of X chromosomes [8–10]. In XX flies, a positive autoregulatory mechanism activates and maintains expression of the RNA-recognition motif (RRM) containing protein Sex-lethal (Sxl) [11]. In female somatic cells, Sxl binds directly to a polyuridine (poly(U)) tract upstream of exon 2 in transformer (tra) pre-mRNA [12, 13]. This results in the skipping of exon 2, which contains an early stop codon, and synthesis of full-length Tra (TraF) in females. In XY flies, which lack Sxl, tra mRNA incorporates exon 2, resulting in premature translational termination and a presumptive small peptide with no known function [13]. Several other factors have been shown to act in concert with Sxl in sex-specific alternative splicing, such as Virilizer (Vir), Female lethal (2)d (Fl(2)d), and Spenito (Nito). All three proteins have an RRM and are required for sex-specific and non-sex-specific functions in Drosophila [14–19].
One of the best characterized targets of the RNA-binding protein TraF is doublesex (dsx), which can yield one of two functional isoforms [20]. In XX flies, TraF is required for the alternative splicing of dsx and fruitless (fru) pre-mRNAs, generating female-specific DsxF and preventing Fru synthesis [21, 22]. In XY flies, which lack TraF, dsx and fru pre-mRNA undergo default splicing and generate male-specific DsxM and FruM. The DsxF and DsxM transcription factors regulate the majority of known sex-specific differences in gene expression and external appearance in Drosophila, often by direct transcriptional regulation of critical sex-specific genes [20, 23, 24]. DsxF and DsxM have identical DNA binding sites and bind regulatory sites in many common target genes, and it is generally believed that Dsx isoform association with sex-specific co-factors determines whether the target gene is activated or repressed [20, 25–28].
Loss of sex identity in sexually dimorphic tissues has profound effects on organ development and function [1–4, 29–31]. In the gonad, sex identity is specified autonomously in both the germline and the soma; somatic gonadal cells additionally send essential non-autonomous cues to instruct germline sex identity [29, 31–34]. Proper gonadogenesis is impeded when the sex identity of the germline does not match that of the soma, and such a mismatch frequently causes sterility [31, 32]. Despite the importance of maintaining sex identity for tissue development and homeostasis, regulation of canonical sex determinants at the transcriptional level has remained relatively unexplored.
In Drosophila gonads, germline stem cells (GSCs) divide to produce daughters that ultimately differentiate into sperm and oocytes, respectively. Proper gametogenesis proceeds through the ensheathment of GSC daughters by somatic support cells that exhibit sex-specific differences. In the testis, a niche of quiescent somatic cells termed the hub supports GSCs and somatic cyst stem cells (CySCs), which produce somatic support cells (Fig 1A, left). GSCs divide with oriented mitosis, and daughter cells that are displaced from the niche differentiate through 4 rounds of transit-amplifying mitotic divisions. CySCs are the only mitotically active somatic cells in wild type testes, and they divide to produce post-mitotic cyst cells. Two cyst cells ensheath a single GSC daughter and remain associated with the germ cell cluster throughout its transit-amplifying divisions. During somatic differentiation, cyst cells grow dramatically to accommodate the enlarging spermatogonia [35–38].
In the ovary, GSCs also divide to produce differentiating daughter cells that undergo 4 mitotic divisions to give rise to 16-cell interconnected germ cysts (Fig 1A, right). The developing germ cyst is surrounded by a layer of somatic follicle cells, which are produced by follicle stem cells (FSCs). CySCs and FSCs require similar self-renewal signals, and both male and female somatic gonadal cells exhibit similar cellular behaviors [39–50]. However, their differentiating offspring exhibit distinct behaviors and markers: cyst cells are quiescent as they differentiate, while follicle cells continue to cycle. Additionally, follicle cells form an epithelium to ensheath the germline, while cyst cells grow in volume and express tight junction proteins to encapsulate spermatogonia [35, 37, 38, 51–53].
Sex-specific anatomical differences are achieved by differential expression of transcription factors [2]. In particular, the transcription factor Chinmo is expressed in male but not female somatic gonadal cells [29, 54, 55]. Chinmo contains a Broad, Tramtrack, and Bric-à-brac/Poxvirus and Zinc finger (BTB/POZ) domain and two C2H2-Zinc fingers (ZFs). Many BTB-ZF proteins in Drosophila and mammals have characterized roles as transcriptional repressors [56–58]. However, while clonal loss of chinmo from imaginal tissue leads to ectopic gene expression in a cell-autonomous manner [54], no direct targets of Chinmo have been identified. Congruent with its dimorphic expression in the somatic gonad, chinmo has no apparent requirement in follicle cells but is essential for CySC niche occupancy [54, 55]. Chinmo is also required for the maintenance of male sex identity in CySCs, as loss of chinmo from all CySCs causes them to lose male sex identity, express markers of ovarian follicle cells and adopt an epithelial-like organization [29]. These data have led to a model in which single CySC clones lacking chinmo are outcompeted by wild type CySC neighbors, but chinmo depletion in all CySCs removes this competitive environment and leads to sex transformation [29, 54]. We have also observed that chinmo-mutant CySC clones that lack the JAK/STAT and EGFR pathway inhibitor Socs36E can form aggregates, suggesting that CySCs lacking chinmo can feminize so long as they are given a chance to proliferate [59]. This sex transformation was reportedly due in part to a transcriptional loss at the dsx locus, leading to a loss of DsxM; however, sustained expression of UAS-dsxM could not prevent the acquisition of female sex identity in chinmo-mutant CySCs, indicating that the molecular mechanism by which these cells feminize is still unclear [29].
Our work supports an alternate model whereby male sex identity is maintained not by preventing transcriptional loss of dsxM, but by preventing alternative splicing of dsx pre-mRNA into dsxF. Since TraF is responsible for dsx alternative splicing in canonical sex determination, we investigated a possible role for TraF in CySC feminization upon chinmo loss. Here, we report that Chinmo maintains male sexual identity by preventing the expression of the female sex determinant traF through a two-step mechanism. We first show that Chinmo represses both expression and alternative splicing of tra pre-mRNA. Next, we demonstrate that feminization of chinmo-mutant CySCs does not require Sxl. We instead find that RNA binding proteins Vir and Fl(2)d, which are necessary to alternatively splice traF in the adult ovary, are important for the feminization of chinmo-mutant CySCs. Thus, we uncover a novel mode of sex maintenance involving previously unreported regulation of tra transcription and a Sxl-independent mechanism of traF splicing in the somatic gonad.
We found dimorphic expression of Chinmo in Drosophila gonads. While Chinmo protein was expressed in all cell types of the adult testis stem cell niche (Fig 1B and 1B’, arrowheads; S1I Fig), it was not detectable in somatic cells of the adult ovary (Fig 1C; S1J Fig). We next confirmed that loss of Chinmo expression in CySCs leads to the acquisition of female identity. When chinmo was depleted in the CySC lineage by RNAi using the somatic driver tj-gal4 (tj>chinmoRNAi; S1K Fig), expression of the male sex determinant DsxM was lost (Fig 2A–2C), and the follicle cell marker Castor (Cas), normally absent from the testis, was ectopically expressed (Fig 2D–2F). In wild type testes, Fasciclin 3 (Fas3) was expressed in niche cells but not in CySCs (Fig 2G). However, in tj>chinmoRNAi testes, we observed Fas3-expressing somatic aggregates resembling epithelial follicle cells that eventually organized at the periphery (Fig 2H and 2I). A marker of late-stage follicle cell maturation, Slow border cells (Slbo), was absent from wild type CySCs (S1A and S1B Fig) but was ectopically expressed in tj>chinmoRNAi testes (S1C Fig). Finally, transcripts of the DsxF target Yp1 were upregulated in chinmo-deficient testes (S1D Fig; [20]. This sex transformation phenotype is due to loss of chinmo in the CySC lineage and not the niche, as depletion of chinmo specifically in niche cells produced no overt phenotype (S1E and S1F Fig; [29]).
Because CySCs serve a critical role in maintaining GSCs, as well as producing somatic support cells, the stem cell niche in tj>chinmoRNAi testes frequently becomes agametic even at relatively early time points after depletion (S1G and S1H Fig; [29, 60]). Based upon these observations, we hypothesized that tj>chinmoRNAi males would become sterile. To test this, we mated successively tj>chinmoRNAi males to OregonR virgin females and scored the number of progeny. Upon each of two mating rounds, tj>chinmoRNAi males exhibited a significant reduction in fertility (25% and 55% compared to control males, p<0.05 and p<0.001, respectively) (Fig 2J). By the third successive mating, tj>chinmoRNAi males were completely sterile whereas control males were not (p<0.0001). Taken together, our results align with previous work showing that Chinmo is required in adult CySCs to preserve male sex identity [29]. Additionally, we demonstrate that CySC male identity is essential for fertility.
We next sought to determine the mechanism by which CySCs undergo feminization upon loss of chinmo. According to a previous report, dsxM mis-expression in chinmo-deficient CySCs (c587>chinmoRNAi; >dsxM) delays feminization [29], suggesting that dsxM transcription was reduced in chinmo-mutant CySCs. However, at the time point when all c587>chinmoRNAi testes contained Fas3-positive aggregates, nearly all c587>chinmoRNAi; >dsxM testes were also feminized [29], indicating a delay but not an abrogation of the phenotype. Additionally, depletion of all dsx transcripts using an RNAi transgene (dsxKK111266) targeting the common region of dsxM and dsxF did not recapitulate the defect seen upon loss of chinmo [29]. These data indicate that the loss of DsxM alone cannot fully account for the phenotype of chinmo-mutant CySCs. We reasoned that the loss of DsxM protein observed in chinmo-mutant CySCs could result from alternative splicing of the dsx pre-mRNA into dsxF rather than from a transcriptional decrease at the dsx locus (Fig 3A). If this were true, we would expect to find in chinmo-deficient CySCs: 1) active transcription of the dsx locus; 2) expression of alternatively-spliced dsxF transcripts; and 3) expression of DsxF protein. To assess dsx transcription levels, we surveyed 4 independently-generated dsx transcriptional reporters: two Gal4 knock-in reporters in the dsx locus (dsx-gal4; [2] and dsx-gal4Δ2; [61]), one MiMIC allele at the dsx locus (dsxMI03050-GFSTF.1; [62]) and one Janelia transgene containing a 2.5 kb dsx regulatory element (GMR40A05-gal4; [63]. We selected dsx-gal4 for further use because it was the only line that was robustly expressed in both adult testes and adult ovaries and therefore accurately reflected dsx transcription (Fig 3B and 3C). By contrast, the other 3 lines displayed male-biased or very low expression in gonads (S2A–S2F Fig). We then assessed dsx-gal4 activity as a proxy for transcription of the dsx locus upon chinmo depletion. We used a genetic approach to remove chinmo from the CySC lineage by analyzing testes homozygous for the chinmoST allele [29] in the dsx-gal4 background. While chinmoST/CyO testes express normal levels of Chinmo, chinmoST/chinmoST males lack Chinmo in the CySC lineage [29]. As expected, GFP was expressed in somatic cells in control chinmoST/CyO; dsx-gal4/UAS-GFP testes and ovaries (Fig 3B and 3C). Importantly, GFP was also expressed in chinmoST/chinmoST; dsx-gal4/UAS-GFP mutant testes (Fig 3D), demonstrating that dsx is still transcribed in chinmo-deficient cyst cells. We also visualized dsx transcript abundance in tj>chinmoRNAi testes using semi-quantitative RT-PCR. Primers that recognize both dsx mRNA isoforms (dsxCOMMON, or dsxC) reveal that dsx is still present in tj>chinmoRNAi testes (Fig 3E). We confirmed that dsxF is produced specifically by chinmo-deficient somatic cells by performing RT-PCR on FACS-sorted CySCs and early cyst cells. As expected, a dsxF-specific band was observed in RNA extracts from wild type ovaries (Fig 3F, left lane). We also observed dsxF in FACS-purified chinmo-deficient cyst cells (Fig 3F, right lane). As expected, dsxF was absent from FACS-purified wild type cyst cells (Fig 3F, middle lane).
We next visualized Dsx protein in control tj>+ and tj>chinmoRNAi testes using an antibody that detects both isoforms of Dsx (anti-DsxC; [64]). We observed that Dsx protein is still synthesized in somatic cells lacking chinmo (Fig 3G and 3H); because DsxM is lost from tj>chinmoRNAi testes (Fig 2C), we conclude that the Dsx protein present in chinmo-deficient somatic cells is DsxF. We confirmed that anti-DsxC detects DsxF by staining tj>dsxF ovaries (S3 Fig). These results suggest that DsxM loss in chinmo-deficient CySCs is not due to transcriptional loss of dsxM, but rather alternative splicing that generates the female isoform DsxF.
We next tested whether DsxF production is causal to feminization of CySCs lacking chinmo. We took a genetic approach and blocked dsxF splicing by using mutant alleles dsxD/dsx1. dsxD cannot be alternatively spliced into dsxF but produces normal levels of dsxM, and dsx1 is a null allele. dsxD/dsx1 flies only produce DsxM. XX dsxD/dsx1 animals develop male abdominal pigmentation, genitalia, and sex combs due to a masculinized soma (S4A–S4D Fig; [65]). We introduced dsxD/dsx1 into males homozygous for the chinmoST allele [29]. As expected, control chinmoST/CyO; dsxD/dsx1 sibling testes appeared normal (Fig 4A; Fig 4D, second bar; S1 Table). By contrast, 100% of chinmoST/chinmoST; TM2/TM6B testes at 7 days post-eclosion contained Fas3-positive somatic aggregates outside the hub (Fig 4B; Fig 4D, first bar; S1 Table). Strikingly, only 57% of chinmoST/chinmoST; dsxD/dsx1 testes contained Fas3-positive aggregates (Fig 4C; Fig 4D, purple bar; S1 Table), a significant reduction compared to chinmoST/chinmoST flies (p<0.001). Taken together, our results reveal that (1) acquisition of female identity in chinmo-mutant CySCs occurs by dsx alternative splicing that generates DsxF and (2) dsxF production is required in part for CySC feminization upon loss of chinmo.
Given that DsxF is produced in chinmo-deficient CySCs, these cells must also express a factor that promotes alternative splicing of dsx pre-mRNA. In female somatic cells, this alternative splicing is mediated by TraF [22]. We hypothesized that tj>chinmoRNAi testes express ectopic TraF that produces dsxF. To test this, we performed semi-quantitative (Fig 5A) and quantitative RT-PCR (Fig 5B and 5C) analysis on the tra locus in tj>chinmoRNAi testes. We found that total tra mRNA abundance significantly increased (3.6-fold) in tj>chinmoRNAi testes compared with tj>+ testes (p<0.05) (Fig 5A, blue arrowhead; Fig 5B, compare blue to white bar). Furthermore, traF-specific primers revealed a 6.1-fold enrichment of traF mRNA in tj>chinmoRNAi testes compared to tj>+ controls (p<0.001) (Fig 5A, red arrowhead; Fig 5C, compare blue to white bar). These data demonstrate that the ectopic tra in chinmo-deficient CySCs is indeed spliced into the female traF isoform.
To confirm these results, we monitored tra alternative splicing in vivo using a transgene that yields GFP expression when tra pre-mRNA is alternatively spliced (UAS-traFΔT2AGFP). In this transgene, the third exon of tra (which is adjoined with exon 1 in the female isoform) is replaced by the coding sequences for self-cleaving T2A peptide and GFP (S5 Fig). As expected, we detected little to no GFP expression in wild type (male) cyst cells (Fig 5D), while wild type (female) follicle cells expressed high levels (Fig 5E’, arrowheads). Notably, we also observed high levels of GFP in the soma of tj>chinmoRNAi testes (Fig 5F’, arrowheads), demonstrating that tra pre-mRNA is alternatively spliced to traF in these feminized somatic cells. We conclude that Chinmo normally represses tra transcription and alternative splicing in the male somatic gonad.
To determine if ectopic Chinmo is sufficient to repress tra transcription, we mis-expressed it in adult ovarian follicle cells using tj-gal4. To evade lethality caused by Chinmo mis-expression [54], we used a temperature-sensitive gal80 allele (tj-gal4, tub-gal80TS or tjTS) and reared flies at the permissive temperature (18°C). Adult F1 females were then shifted to the restrictive temperature (29°C) for 5 days before ovaries were homogenized. We observed a 2.2-fold decrease in total tra mRNA abundance (p<0.001) and a 1.8-fold decrease in traF abundance (p<0.001) in tjTS>chinmo ovaries compared with tjTS>+ ovaries (S6A Fig). dsxF was also decreased 5.8-fold (p<0.001) in tjTS>chinmo ovaries compared with tjTS>+ ovaries, presumably as a result of reduced TraF (S6A Fig). Taken together, our results demonstrate that Chinmo is both necessary and sufficient to prevent somatic expression of the female sex determinants traF and dsxF.
These findings suggest that sex transformation in chinmo-deficient cyst cells is due to ectopic TraF. To test this, we concomitantly depleted both tra and chinmo in the somatic lineage of the testis. Whereas 98% of tj>chinmoRNAi testes contained Fas3-positive aggregates outside of the niche, only 48% of tj>traRNAi; chinmoRNAi testes had such aggregates, indicating a significant block in feminization (p<0.0001) (Fig 5I; Fig 5J, purple bar; S1 Table). In these rescued tj>traRNAi; chinmoRNAi testes, CySCs no longer expressed Fas3, and the germline appeared normal (Fig 5I). We also performed epistatic experiments with tra mutant alleles, similar to the dsxD/dsx1 experiment. XX tra1/Df(3L)st-j7 animals develop male somatic structures due to loss of TraF (S4E–S4H Fig; [66]). Whereas 100% of chinmoST/chinmoST testes were feminized as assessed by Fas3-positive aggregates, only 61% of chinmoST/chinmoST; tra1/Df(3L)st-j7 testes were feminized (p<0.001) (Fig 4D, green bar; S1 Table). The phenotype was not sensitive to tra dose as chinmoST/chinmoST; tra/+ testes were still feminized (Fig 4D, yellow bars; S1 Table). These results demonstrate that Chinmo prevents both tra transcription and alternative splicing in CySCs and that feminization of male somatic cells in the absence of chinmo is due to ectopic traF.
Global expression of TraF in XY flies during development causes female somatic differentiation [67]. To test whether TraF expression alone is sufficient to cause male-to-female sex transformation in adult CySCs, we over-expressed traF cDNA in tj-gal4 expressing cells and used gal80TS to restrict expression to only adult CySCs (tjTS). While we observed accumulation of somatic aggregates in tjTS>traF testes, they did not express Fas3 or Cas, in contrast to those in tj>chinmoRNAi testes (compare Fig 5K and 5L to Fig 2I for Fas3 and compare Fig 5N and 5O to Fig 2F for Cas). These data suggest that traF-misexpressing cyst cells have not fully acquired a follicle-like fate. However, we found on average 121.0±8.8 somatic cells expressing Zinc finger homeodomain 1 (Zfh1), which marks CySCs and their earliest differentiating daughters [68], in tjTS>traF testes compared with 40.1±1.6 cells in control tjTS>+ testes (p<0.0001) (S7A, S7B and S7I Fig). Upon somatic traF mis-expression, we also observed accumulation of somatic cells expressing Tj, which marks a broader population of CySCs and early cyst cells [69] (S7C and S7D Fig). tjTS>traF testes contained 158.7±14.5 Tj-positive cells compared with 80.3±3.9 cells in tjTS>+ testes (p<0.001) (S7J Fig). We interpret the accumulation of Zfh1-positive, Tj-positive cells in tjTS>traF testes as a delay in somatic differentiation. Because cyst cells must exit the cell cycle in order to support the developing male germline, there are no somatic cells located away from the niche in wild type testes that are positive for 5-ethynyl-2’-deoxyuridine (EdU), an S-phase marker. We previously showed that when somatic differentiation is delayed, EdU-positive cyst cells are observed several cell diameters away from the niche [36]. Consistent with our prior results, in control tjTS>+ testes, only Tj-positive cells near the hub incorporated EdU (S7E Fig, arrowheads; n = 0/20 testes with EdU-positive cyst cells located away from the niche). By contrast, in tjTS>traF testes we detected EdU-positive somatic cells located many cell diameters away from the niche, suggesting that these cells had delayed differentiation (S7F Fig, arrows; n = 20/26 testes with EdU-positive cyst cells located away from the niche).
Consistent with a defect in somatic differentiation, cyst cells mis-expressing traF were impaired in their ability to support the germline. In tjTS>traF testes, early germ cells accumulated (identified by dot- and dumbbell-shaped α-spectrin-positive fusomes) at the expense of more differentiated spermatogonia, as fewer germ cysts with branched fusomes were observed (S7G and S7H Fig). tjTS>traF testes also contained significantly fewer EdU-positive, 4- and 8-cell spermatogonial cysts than tjTS>+ testes (S7E, S7F and S7K Fig). These results demonstrate that ectopic TraF in CySCs is deleterious to their differentiation, but alone cannot drive CySCs to assume a follicle-like fate. Taken together with our previous finding that tra is downstream of chinmo in CySC feminization, we conclude while TraF induction is important for CySC feminization upon loss of chinmo, it is not sufficient.
Our finding that chinmo-deficient CySCs produce traF (Fig 5A and 5C) reveals that they possess machinery to splice tra pre-mRNA into the female isoform. We considered the possibility that wild type CySCs might be competent to alternatively splice tra. However, somatic mis-expression of UAS-traFΔT2AGFP in wild type somatic cells did not lead to tra alternative splicing, since GFP was absent from the somatic lineage (Fig 5D). Thus, wild type CySCs are intrinsically unable to generate traF mRNA, precluding this model.
It follows, then, that one or more factors are ectopically expressed upon loss of chinmo that alternatively splice tra pre-mRNA into traF. Since Sxl is required for traF production in wild type females (Fig 6A; [12]), we investigated whether Sxl is ectopically expressed in chinmo-mutant CySCs. As expected, Sxl protein was absent from wild type testes but was detectable in wild type ovaries (Fig 6B and 6C). Importantly, we did not observe Sxl in chinmo-mutant testes (Fig 6D). These results were validated by assessing Sxl mRNA isoform abundance in adult gonads. Semi-quantitative RT-PCR demonstrated that control tj>+ testes express male-specific SxlM (Fig 6E, left lane), which contains an early stop codon and encodes no functional protein, while control tj>+ ovaries express female-specific SxlF (Fig 6E, middle lane), which encodes functional Sxl. tj>chinmoRNAi testes still express SxlM (Fig 6E, right lane), consistent with the absence of Sxl protein in these testes (Fig 6D). We also tested whether mis-expression of chinmo in female follicle cells could prevent Sxl alternative splicing; however, both tjTS>+ and tjTS>chinmo ovaries expressed only the female-specific SxlF isoform (S6B Fig).
Furthermore, unlike depletion of tra, depletion of Sxl in feminizing, chinmo-deficient somatic cells did not suppress Fas3 expression or the epithelial organization of somatic cells (Fig 5J, light blue bar; S1 Table). [As expected, somatic depletion of Sxl in an otherwise wild type background produced no testis phenotype (S1 Table). We confirmed that the UAS-Sxl-RNAi line was effective at knockdown because somatic depletion of Sxl in females led to only a rudimentary ovary with 100% penetrance, n = 23.] Consistent with this, none of three distinct mutant alleles of Sxl prevented feminization in chinmoST/chinmoST testes (Fig 6F–6K; S1 Table). Taken together, these data support a model where the ectopic tra pre-mRNA in chinmo-mutant CySCs is alternatively spliced into traF via a non-canonical, Sxl-independent mechanism.
We next examined a potential role for other candidates with known roles in female-specific alternative splicing of tra. We found that vir, fl(2)d, and nito transcripts were 1.5-fold (p<0.05), 3.4-fold (p<0.001), and 5.7-fold (p<0.0001) higher in adult ovaries compared with adult testes, respectively, suggesting sex-biased expression in adult gonads (Fig 7A and 7B). This observation is consistent with ModENCODE RNA-seq data demonstrating that vir, fl(2)d, and nito transcripts are present at very low levels in wild type testes [70]. However, levels of all three transcripts significantly increased (2.3-fold, 1.7-fold, and 1.8-fold for vir, fl(2)d, and nito, respectively) in tj>chinmoRNAi testes compared with control tj>+ testes (Fig 7A and 7B; p<0.05 for vir and nito, p<0.01 for fl(2)d). While depleting vir or fl(2)d had no effect on testis development or spermatogenesis (S8A–S8C Fig), we found that depletion of vir in the female somatic gonad caused severe defects in ovary development. tj>virRNAi females develop some female reproductive structures and contain an oviduct, but lack ovaries (S8D and S8E Fig). Both tj>virRNAi and tj>fl(2)dRNAi females failed to lay fertilized eggs. To test whether vir or fl(2)d are necessary for traF splicing in adult ovaries, we depleted vir or fl(2)d in the female somatic gonad using tjTS, rearing flies at the permissive temperature to prevent vir or fl(2)d knockdown during development. After eclosion, adult females were then reared at the restrictive temperature to allow for vir and fl(2)d depletion. While wild type follicle cells express GFP produced by UAS-traFΔT2AGFP (Fig 7D), GFP is dramatically reduced in the follicle cells of tjTS>virRNAi and tjTS>fl(2)dRNAi ovaries (Fig 7C, 7E and 7F). These results demonstrate that vir and fl(2)d are both female-biased in the adult gonad and are required for traF alternative splicing in follicle cells.
To test whether vir or fl(2)d are required for sex transformation upon somatic loss of chinmo, we depleted each factor concomitantly with chinmo and monitored the frequency of CySC feminization. Depletion of vir or fl(2)d in tj>chinmoRNAi testes significantly reduced the percentage of feminized testes (p<0.001 and p<0.0001, respectively) (Fig 7G and 7H; Fig 5J, green and yellow bars, respectively; S1 Table). In contrast, depletion of nito did not prevent feminization (Fig 7I; Fig 5J, red bar; S1 Table). As expected, somatic depletion of vir, fl(2)d or nito in an otherwise wild type testis had no effect (S1 Table). We also tested the sufficiency of fl(2)d for CySC sex transformation. We found that mis-expression of fl(2)d in the adult CySC lineage (tjTS>fl(2)d) did not cause Fas3-positive aggregates to accumulate (Fig 5M). Furthermore, the follicle cell marker Cas was not induced in tjTS>fl(2)d testes (Fig 5P). Due to the lack of a UAS-vir transgenic Drosophila line, we were unable to test the sufficiency of vir for CySC feminization. Based on these findings, we conclude that Vir and Fl(2)d are epistatic to chinmo and are required, but not sufficient, for feminization of chinmo-mutant CySCs. Taken together with our previous results, this strongly implicates Vir and Fl(2)d in alternative splicing of the ectopic tra pre-mRNA observed in sex-transformed CySCs.
Here, we show that that one single factor, Chinmo, preserves the male identity of adult CySCs in the Drosophila testis by regulating the levels of canonical sex determinants. We demonstrate that CySCs lacking chinmo lose DsxM expression not by transcriptional loss but rather by alternative splicing of dsx pre-mRNA into dsxF. These chinmo-mutant CySCs ectopically express TraF and DsxF, and both factors are required for their feminization. Furthermore, our results demonstrate that tra alternative splicing in cyst cells lacking chinmo is achieved independently of Sxl. Instead, our work strongly suggests that traF production in the absence of chinmo is mediated by splicing factors Vir and Fl(2)d. We propose that male sex identity in CySCs is maintained by a two-step mechanism whereby traF is negatively regulated at both transcriptional and post-transcriptional levels by Chinmo (Fig 8). In this model, loss of chinmo from male somatic stem cells first leads to transcriptional upregulation of tra pre-mRNA as well as of vir and fl(2)d. Then the tra pre-mRNA in these cells is spliced into traF by the ectopic Vir and Fl(2)d proteins. The ectopic TraF in chinmo-deficient CySCs then splices the dsx pre-mRNA into dsxF, resulting in loss of DsxM and gain of DsxF, and finally induction of target genes usually restricted to follicle cells in the ovary.
Chinmo has motifs associated with transcriptional repression and its loss clonally is associated with ectopic transcription [54]. One interpretation of our data is that Chinmo directly represses tra, vir, and fl(2)d in male somatic gonadal cells. As the binding site and potential co-factors of Chinmo are not known, future work will be needed to determine whether Chinmo directly regulates expression of these genes. We also note that ~50% of chinmo-mutant testes still feminize in the genetic absence of tra or dsxF. These latter data indicate that Chinmo regulates male sex identity through another, presumably parallel, mechanism that does not involve canonical sex determinants. However, this tra/dsx-independent mode of sex maintenance downstream of Chinmo is not characterized and will require the identification of direct Chinmo target genes.
We previously showed that JAK/STAT signaling promotes chinmo in several cell types, including CySCs [54]. Since JAK/STAT signaling is itself sex-biased and restricted to the embryonic male gonad, we presume that activated Stat92E establishes chinmo in male somatic gonadal precursors, perhaps as early as they are specified in the embryo [33, 71]. Because loss of Stat92E from CySCs does not result in an apparent sex transformation phenotype [29, 40, 60], we favor the interpretation that Stat92E induces expression of chinmo in CySCs but that other sexually biased factors maintain it. One potential candidate is DsxM, which is expressed specifically in early somatic gonads and at the same time when Stat92E activation is occurring in these cells [72]. In fact, multiple DsxM ChIP-seq peaks were identified in the chinmo locus, suggesting potential regulation of chinmo by DsxM [26]. Taken together with our findings, this suggests a potential autoregulatory feedback loop whereby DsxM preserves its own expression in adult CySCs by maintaining Chinmo expression, which in turn prevents traF and dsxF production.
Recent studies on tissue-specific sex maintenance demonstrate that while the Sxl/Tra/Dsx hierarchy is an obligate and linear circuit during embryonic development, at later stages it is more modular than previously appreciated. For example, Sxl can regulate female-biased genes in a tra-independent manner [73, 74]. Additionally, Sxl and TraF regulate body size and gut plasticity independently of the only known TraF targets, dsx and fru [3, 4]. We find that negative regulation of the TraF-DsxF arm of this cascade is required to preserve male sexual identity in CySCs but unexpectedly is independent of Sxl. Because depletion of Vir or Fl(2)d significantly blocks sex transformation and both are required for tra alternative splicing in the ovary, our work reveals they can alternatively splice tra pre-mRNA even in the absence of Sxl. To the best of our knowledge, this is the first demonstration of Sxl-independent, Tra-dependent feminization. These results raise the broader question of whether other male somatic cells have to safeguard against this novel mechanism. Because recent work has determined that sex maintenance is important in systemic functions regulated by adipose tissue and intestinal stem cells [3, 4], it will be important to determine whether Chinmo represses traF in these settings. Finally, since the transcriptional output of the sex determination pathway is conserved from Drosophila (Dsx) to mammals (DMRT1), it is possible that transcriptional regulation of sex determinants plays a similar role in adult tissue homeostasis and fertility in higher organisms.
The following fly stocks were used and are described in FlyBase: OregonR; yw; tj-gal4; tub-gal80TS; dsx-gal4; dsx-galΔ2; GMR40A05-gal4; dsxMI03050-GFSTF.1; dsx1; dsxD; chinmoST; tra1; Df(3L)st-j7; UAS-GFPnls; UAS-dcr2; UAS-chinmoRNAi (HMS00036); UAS-traRNAi (HMS02830); UAS-traF; UAS-3xHAfl(2)d; UAS-dsxF; UAS-5’UTR-chinmo-3’UTR; UAS-SxlRNAi (HMS00609); Sxlf1; Sxlf2; Sxlf18; UAS-virRNAi (HMC03908); UAS-fl(2)dRNAi (HMC03833); UAS-nitoRNAi (HMS00166).
For RNAi-mediated depletion of chinmo, Sxl, tra, vir, fl(2)d, and nito, flies were reared at an ambient temperature (21°C). Adult males were collected twice a week and aged at 29°C to increase Gal4 activity. For temporal control of gene expression, tj-gal4, tub-gal80TS virgins were crossed to UAS-traF or UAS-3xHAfl(2)d males and progeny were reared at the permissive temperature (18°C) to prevent traF or fl(2)d mis-expression during embryonic, larval, and pupal development. Adult males of the correct genotype were collected twice a week and shifted to the restrictive temperature (29°C) to inactivate Gal80.
In this transgene, most of the third exon of tra is replaced by the coding sequences for self-cleaving T2A peptide and GFP. Specifically, the coding sequences of T2A and GFP were cloned in frame immediately downstream of the 26th nucleotide (nt) of tra exon 3 and immediately upstream of the last 18 nt of this exon. PCR was performed with Q5 high-fidelity polymerase from New England Biolabs (M0491S). The PCR product was digested with EcoRI and XhoI before cloning into the pUASTattb vector [75]. The construct was verified by sequencing, and a transgenic line was established through ΦC-31 integrase mediated transformation (Bestgene, attP site VK05, BDSC#9725).
The following primary antibodies were used: rat anti-Chinmo (1:1000; gift of N. Sokol, Indiana University, IN, USA), goat anti-Vasa (1:50, dC-13, Santa Cruz), rabbit anti-Vasa (1:1500; gift of R. Lehmann, Skirball Institute/NYU School of Medicine, NY, USA), guinea pig anti-Tj (1:5000; gift of D. Godt, University of Toronto, ON, Canada), rabbit anti-Zfh1 (1:5000; gift of R. Lehmann), mouse anti-Fasciclin-3 (1:50; Developmental Studies Hybridoma Bank (DSHB)), mouse anti-Eya (1:20; DSHB), rat anti-DsxM (1:200; gift of B. Oliver, National Institutes of Health, MD, USA), rat anti-DsxC (1:50; gift of M. Arbeitman, Florida State University, FL, USA), rabbit anti-Castor (1:50; gift of W. Odenwald, National Institutes of Health, MD, USA), mouse anti-α-spectrin (1:20, DSHB), mouse anti-SxlM18 (1:5; DSHB), rabbit anti-GFP (1:500; Invitrogen). Secondary antisera used were all raised in donkey (Jackson ImmunoResearch).
Testes and ovaries were dissected in 1x PBS and fixed in 4% paraformaldehyde in 1x PBS for 30 minutes at room temperature (RT). Fixed tissue was washed twice at RT in 0.5% PBST (1x PBS with 0.5% Triton X-100) and blocked in PBTB (1x PBS, 0.2% Triton X-100, 1% BSA) for 1 hour at RT or overnight at 4°C. Primary antibodies were incubated overnight at 4°C and washed off twice at RT in PBTB. Secondary antibodies were incubated for 2 hours at RT in the dark and washed off twice in 0.2% PBST (1x PBS with 0.2% Triton X-100). Tissue was mounted in Vectashield Medium (Vector Laboratories) prior to confocal analysis, and confocal images were captured using a Zeiss LSM 510 confocal microscope, 63x objective.
DIC images of adult female reproductive structures (at 5x) were obtained using a Zeiss Axioplan microscope with a Retiga Evi (QImaging) digital camera and QCapture Pro 6.0 software.
Testes and ovaries were dissected in 1x PBS and fixed in 20% EM-grade paraformaldehyde (Electron Microscopy Sciences) in 1x PBS for 20 minutes at RT. Fixed tissue was washed 3 times for 15 minutes each in TNT (0.1M Tris-HCl, 0.3M NaCl, 0.05% Tween-20) and blocked using Image-iT FX Signal Enhancer (ThermoFisher) for 30 minutes at RT, then washed 3 times for 15 minutes each in TNT. Primary anti-DsxC was incubated overnight at 4°C. After anti-DsxC incubation, tissue was blocked in PBTB for 1 hour and then treated with anti-Vasa and anti-Tj. Primary antibodies were washed twice for 15 minutes each in PBTB, then secondary antibodies were incubated overnight at 4°C in PBTB. Finally, the DsxC signal was amplified by TSA (see below) and testes were mounted in Vectashield prior to analysis.
TSA (Perkin Elmer) was performed to amplify DsxM and DsxC signals. HRP anti-rat (Jackson ImmunoResearch) was used as a secondary antibody and the tertiary Cy3-conjugated tyramide reaction was performed per the manufacturer’s instructions.
To purify CySCs and early cyst cells, the somatic cell lineage was labeled using tj-gal4 to drive UAS-GFPnls expression. Testes were dissociated in trypsin/collagenase for 15 minutes and the cell suspension was passed through 70μm filters (Falcon). GFP-expressing somatic cells were purified from the resulting filtrate by FACS using a Sony SY3200 highly automated parallel sorting (HAPS) cell sorter into TRIzol LS (ThermoFisher), and RNA was extracted according to the manufacturer’s instructions. Post-sort purity of samples was confirmed by immunocytochemistry and the absence of Vasa-positive germ cells.
EdU-labeling of testes was performed using the Click-iT EdU Alexa Fluor 647 Imaging Kit (ThermoFisher). Testes were dissected in S2 cell culture medium (Life Technologies) then incubated in 10 μM EdU for 30 minutes. Testes were then fixed, washed, and stained as described above. The cycloaddition reaction was performed per the manufacturer’s instructions. Testes were mounted in Vectashield prior to confocal analysis.
To detect mRNA levels of canonical sex determinants by PCR, whole ovaries (n = 5–10) or whole testes (n = 55–200) were isolated and homogenized into TRIzol (ThermoFisher). RNA was extracted and DNase-treated (Ambion) per the manufacturer’s instructions. Reverse transcription was performed using Maxima reverse transcriptase (ThermoFisher) according to the manufacturer’s instructions and 1–2 μg of RNA as template. qRT-PCR was performed using SYBR Green PCR Master Mix (ThermoFisher) and a Biorad CFX96 Real-Time PCR Machine. Semi-quantitative RT-PCR was performed on a Biorad iCycler. Because the proportion of somatic cells is significantly increased in tj>chinmoRNAi testes compared to tj>+ controls, the qRT-PCR values were normalized first to tubulin and second to zfh1, an early somatic marker.
total tra: fwd-GAGCCCGCATCGGTATAATC; rev-GACGTGGTAGCCTTTGGTATC
traF: fwd-AACCCAGCATCGAGATTCC; rev-CGAACCTCGTCTGCAAAGTA
dsxC: fwd-GAAAGAACGGCGCCAAT; rev-GGCGTCTGCGTCCTTTAATA
dsxM: fwd-GAGCTGATGCCACTCATGTAT; rev-CTGGGCTACAGTGCGATTTA
dsxF: fwd-GAATGAGTACTCCCGTCAACAT; rev-GGGCAAAGTAGTATTCGTTACTCTA
rpl15: fwd-AGGATGCACTTATGGCAAGC; rev-GCGCAATCCAATACGAGTTC
α-tub84b: fwd-CAACCAGATGGTCAAGTGCG; rev-ACGTCCTTGGGCACAACATC
β-tub56d: fwd-CTCAGTGCTCGATGTTGTCC; rev-GCCAAGGGAGTGTGTGAGTT
SxlJYR: fwd-ACACAAGAAAGTTGAACAGAGG; rev-CATTCCGGATGGCAGAGAATGG
SxlEM: fwd-CGCTGCGAGTCCATTTCC; rev-GTGGTTATCCCCCATATGGC
vir: fwd-CATGAGGAAGTGACGGACATC; rev-GGAAAGTCTGCCTGGACTCG
fl(2)d: fwd-GGCCAACAAGGAGCAAGAA; rev-CGCTCGAACAGGAGATTGAC
nito: fwd-GGTGTACAAGTCCACAACCAGA; rev-CGACGGTGATCCAAAGGAA
The fertility of adult males was assayed by mating individual males with two wild type (OregonR) virgin females (between 5–10 days old) for 48 hours at 25°C. After a 2-day mating period, males were recovered and preserved for subsequent matings using fresh virgin OregonR females. Fertility was scored by counting the number of F1 offspring produced by each individual cross and reported as the average number of F1 offspring for each genotype.
Statistical parameters for each experiment are reported in the figure legends. Data were analyzed using Microsoft Excel and are reported to be statistically significant when p<0.05 by the appropriate statistical test. For qRT-PCR data, significance was determined by two-tailed Student’s t-test. For fertility assays and cyst cell quantifications, significance was determined using single-factor ANOVA. For rescue of CySC feminization (Fas3-positive aggregates), significance was determined using Fisher’s Exact Test.
|
10.1371/journal.pcbi.1003128 | Co-expression Profiling of Autism Genes in the Mouse Brain | Autism spectrum disorder (ASD) is one of the most prevalent and highly heritable neurodevelopmental disorders in humans. There is significant evidence that the onset and severity of ASD is governed in part by complex genetic mechanisms affecting the normal development of the brain. To date, a number of genes have been associated with ASD. However, the temporal and spatial co-expression of these genes in the brain remain unclear. To address this issue, we examined the co-expression network of 26 autism genes from AutDB (http://mindspec.org/autdb.html), in the framework of 3,041 genes whose expression energies have the highest correlation between the coronal and sagittal images from the Allen Mouse Brain Atlas database (http://mouse.brain-map.org). These data were derived from in situ hybridization experiments conducted on male, 56-day old C57BL/6J mice co-registered to the Allen Reference Atlas, and were used to generate a normalized co-expression matrix indicating the cosine similarity between expression vectors of genes in this database. The network formed by the autism-associated genes showed a higher degree of co-expression connectivity than seen for the other genes in this dataset (Kolmogorov–Smirnov P = 5×10−28). Using Monte Carlo simulations, we identified two cliques of co-expressed genes that were significantly enriched with autism genes (A Bonferroni corrected P<0.05). Genes in both these cliques were significantly over-expressed in the cerebellar cortex (P = 1×10−5) suggesting possible implication of this brain region in autism. In conclusion, our study provides a detailed profiling of co-expression patterns of autism genes in the mouse brain, and suggests specific brain regions and new candidate genes that could be involved in autism etiology.
| Autism spectrum disorder (ASD) is a complex neurodevelopmental condition associated with many different genes. However, the neuroanatomical and functional properties of these genes in the brain are largely unknown. Here we examined the co-expression network of 26 genes associated with ASD, using data from the Allen Mouse Brain Atlas, which provides a whole-genome, high-resolution map of gene expression pattern in the adult mouse brain. We discovered that autism genes are significantly more co-expressed than expected by chance, suggesting common neuro-functional properties. We then examined the spatial properties of co-expression modules that are highly enriched with autism genes. Consequently, we found that genes in two of these modules are significantly over-expressed in the cerebellar cortex, and particularly in sections that are predominantly populated by granular cells. These findings provide the essential link between gene networks associated with ASD and specific brain regions, and hence lay out a basis for further exploration of the particular neuronal circuits involved in ASD etiology.
| Autism spectrum disorder (ASD) is one of the most prevalent and highly heritable neurodevelopmental disorders in humans [1]–[3]. There is strong evidence that the onset and severity of ASD is governed in part by complex molecular mechanisms affecting the normal development of the brain [1], [4]. While no major anatomical pathology have been observed in brains of ASD cases [5], various molecular and neuroimaging studies have linked several brain regions to ASD. For example, Voineagu et al. have found differences in gene expression patterns in the cortex of ASD brain [6]. Cortical regions has also been highlighted in neuroimaging studies of autistic brains along the cerebellum and other brain areas [7], [8]. In addition, other studies have pointed to various molecular mechanisms that might be altered in the autistic brain [9]–[11]. In this realm, genes involved in synapse formation and brain circuitry are consistently found to be dysregulated in people with ASD [12]–[14].
Recent genomic advances have led to the discovery of diverse genetic loci linked to ASD, including chromosomal aberrations [15], [16], copy number variations [11], [17], [18] and both common and rare single nucleotide variations (SNVs) [19]–[24]. Consequently, to date, more than 330 candidate genes have been associated with ASD susceptibility [25] and many more are projected to be found. However, despite the plethora of genetic variations associated with ASD, the molecular mechanisms and neuroanatomical structures underlying ASD traits remain largely unclear.
The mouse model system provides a convenient and safe approach to experimentally study neuroanatomical mechanisms and candidate genes for autism susceptibility [26]–[28]. At present, dozens of single gene knockout and transgenic mice models have been used to elucidate neuropathology that might underlie the autism-like behaviors [29]. Despite the obvious genetic and neuroanatomical differences between mouse and human, mouse models are extremely valuable and effectively used in dissecting out the role of specific gene, pathway, neuron subtype, or brain region in a particular abnormal behavior shared by both these mammals. In this realm, the Allen Brain Atlas of the mouse [30]–[33] provides a comprehensive source of genome-wide high-resolution atlas of gene expression throughout the adult mouse brain. In this study, we utilized this database to examine the spatial co-expression characteristics of genes associated with autism susceptibility. Consequently, we identified several co-expression gene networks that are enriched with autism genes highlighting potential candidate genes and brain regions implicated in autism.
The autism gene database, AutDB [34], was used to construct a set of 84 genes, AutRef84, strongly implicated in autism pathogenesis [25], [29]. Genes contained in AutDB were classified into four genetic categories: (1) Rare: rare submicroscopic copy number variants and single gene disruptions or mutations directly linked to ASD; (2) Syndromic: genes implicated in syndromes in which a subset of affected individuals also develop autistic symptoms; (3) Association: small risk-conferring candidate genes with common polymorphisms that have been identified from genome-wide association studies in idiopathic ASD cases; and (4) Functional: functional candidate genes relevant to ASD that are not covered by any of the other genetic categories. AutRef84 was generated by filtering out functional candidate genes that lack any experimentally derived genetic link to ASD as well as genes that solely belong to the Association classification. The resulting dataset consisted of 64 genes classified within the rare classification and 20 within the syndromic classification (Supplementary Table S1).
We utilized gene expression data available from the Allen Brain Atlas (ABA) of the mouse brain (http://mouse.brain-map.org) which contains voxelized expression profiles for ∼20,000 genes derived from in situ hybridization (ISH) experiments conducted on male, 56-day old C57BL/6J mice) [30]–[33]. Gene expression profiles in the ABA were generated for the mouse brains from processed image data for the sagittal sections of a single hemisphere and, for 4,104 genes of high neurobiological interest, coronal and sagittal sections across the whole brain. We focused on genes for which brain-wide data are available. The ISH data were co-registered to the Allen Reference Atlas, which is partitioned into 49,742 cubic voxels of size 200 microns. A gene labeled g has a profile of expression energy presented as a function E(v,g) over the brain, where v is a voxel label. The Allen Mouse Brain Atlas Addiction Database (http://addiction.brainarchitecture.org/) used this expression dataset to create a high quality brain expression dataset, called Abest, which consists of 3,041 genes with the most highly correlated expression energy profiles between the coronal and sagittal sections [35]. This gene-expression dataset was used for the co-expression analysis discussed below.
For the set of 3,041 genes defined above, a gene-by-gene co-expression matrix was computed as follows:(1)where V = 49,742 is the total number of voxels in the brain. The quantity C(g,g') is the cosine similarity between the gene-expression profiles of genes g and g' with values ranging between zero and one by construction. The motivation for choosing the cosine similarity as the co-expression measure in this study, is its connection to the difference between the energies of two brain-wide functions. The proposed co-expression measure (Equation 1) can also be described as a simple decreasing function of the squared energy of the difference between two brain-wide functions. The co-expression matrix C is symmetric, with ones on the diagonal. Its size is 3,041, the number of genes in the full Abest dataset. Given a subset of this dataset, the co-expression matrix of this subset can be obtained by extracting the sub-matrix of C corresponding to the position of the genes in the full dataset.
To determine whether the autism genes are more co-expressed than other genes expected by chance, the cumulative distribution function (CDF) of the entries of the autism-related and Abest co-expression matrices were compared, and a two-sample Kolmogorov–Smirnov test was used to determine whether they were drawn from the same probability distribution. In addition, we compared the connectivity of the co-expression network of the autism genes to co-expression networks of the same size randomly drawn from Abest. For that, Monte Carlo simulations [36] were conducted to generate 100,000 random gene sets sampled from Abest, and the co-expression matrices of these sets of genes where extracted from the co-expression matrix C.
Given the co-expression matrix of G genes of interest, one can consider the underlying weighted graph with nodes corresponding to genes, and the weight of links equals to the co-expression of the nodes [37]. The matrix can be cut at any value ρ between zero and one, resulting in links with weights lower than the threshold ρ. At any value of the threshold, the connected components (sets of connected genes) can be computed using Tarjan's algorithm. In particular, the maximum size M(ρ) and average size A(ρ) of connected components can be calculated for all gene sets over different co-expression values. If Nρ(k) is the number of connected components in the co-expression matrix of G genes defined by ρ threshold and contain exactly k genes out of G genes in the gene set, the maximum and average sizes are expressed as follows:(2)(3)Further, we applied the classical definition of “clique” from graph theory [38], [39] to our co-expression matrix to characterize networks of co-expressed genes such that every gene in the network is connected to all other genes at a co-expression threshold ρ.
Next, we aimed at evaluating the unique anatomical properties of co-expression cliques that are significantly enriched with autism genes. For that, we first identified virtually all cliques in the dataset that contained ng≥2 autism genes. Then, we used Monte Carlo simulations using 100,000 randomly generated gene sets of size G = 26 to compute the likelihood of each clique to contain at least ng autism genes. Finally, for cliques that were significantly enriched with autism genes, we calculated the sum of the normalized expression profiles of the genes in the clique as follow:(4)We then used Sclique, to examine the neuroanatomical properties of the genes included in the clique as follows [40]. For a given brain region ω in the Allen Reference Atlas [41], the fitting score between the expression profile and the region ω is defined as the cosine similarity between the expression profile and the characteristic function χω of the brain region.(5)where χω(v) equals one if voxel labeled v belongs to region ω, and zero otherwise. This fitting score would equal one if the sum of the gene expression in the clique were proportional to the characteristic function of the region ω, and zero if it were entirely supported outside the region. The fitting score defined in the above equation equals the co-expression between a (hypothetical) gene whose expression profile would be Sclique, and a (hypothetical) gene whose expression would coincide exactly with region ω. Fitting scores of a given clique can be computed in all brain regions, and the distribution of these fitting scores can be simulated by repeatedly drawing random cliques of genes (with the same number of genes) from Abest. These analyses were performed using a commercial software package (MATLAB R2011b, The MathWorks Inc., Natick, MA, 2000).
Finally, we used the Bioconductor GOstats package in R software [42] to assess whether genes belonging to a co-expression clique, also share other functional or molecular properties. The absolute list of GO terms were obtained using both a (a) cut-off = 2*ratio (fg/fc) [where fg = frequency of occurrence of a GO term in the given gene set, fc = frequency of its occurrence in the complete list of human genes] and (b) cut-off = median value of ratio (fg/fc). Only significant terms (P<0.01) with an associated gene count> = 5 were considered.
Overall, 26 genes were found in the intersection of the autism-genes dataset and the dataset of high-quality expression genes from the Allen Brain Atlas (ABA) of the mouse brain (AutRef84∩Abest = 26). These autism-related genes showed a higher degree of co-expression connectivity than all other genes in this dataset (Kolmogorov–Smirnov P = 5×10−28). Comparing the empirical distributions of co-expression values of the autism genes to the other genes in the Allen dataset revealed that the largest deviation between these distributions was at co-expression value of 47.53% (Figure 1A). Furthermore, we evaluated the connectivity of genes across different co-expression values. Here too, the average size of connected components (See methods) among autism gene was consistently larger than seen in 1000 randomly generated gene sets from the Allen database (Figure 1B).
Next, we examined cliques (see methods) of co-expressed genes delineated by autism genes inter-connected at co-expression values of ≥47.53%. A total of 59 overlapping cliques were characterized containing on average 563.5 genes and 8.3 autism genes (Figure 2). Finally, using Monte Carlo simulations, we identified ten cliques that were significantly enriched with autism genes at P<0.01 (Figure 3). Of note, the two top ranked cliques remained significant (P<0.05) even after accounting for multiple testing using the conservative Bonferroni correction.
The top-ranked clique in our analysis (hereafter will be referred as Clique I), was delineated by the autism genes: Ptchd1, Galnt13, Dpp6 and Astn2, and included another 29 genes, inter-connected with a co-expression values of ≥70% (Supplementary Table S2). The second top-ranked clique (hereafter will be referred as Clique II) included the autism genes: Rims3, and Astn2, and another four genes, all inter-connected at ≥79% co-expression level (Supplementary Table S3). Examining the neuroanatomical expression properties across a set of 134 brain regions of the left hemisphere (41 of which are cortical, and 93 subcortical) grouped by the 12 main brain regions according to the Allen Reference Atlas, revealed a significant over-expression of genes belonging to Cliques I in the cerebellar cortex (Figures 4A, 4B, Supplementary Figure S1). Genes belonging to Clique II also showed a slight over-expression in the cerebellar cortex (Figures 5A, 5B), as well as in several cortical regions, however these signals were much weaker than the one of clique I. Next, we asked if the over-expression in the cerebellum is a unique property of these two cliques. For that, we examined the neuroanatomical expression of all cliques in Figure 2 and found only seven other cliques showing a similar over-expression in the cerebellar cortex. Interestingly, these cliques had substantial gene overlap with cliques I&II, and were ranked high in their autism gene enrichment scores (Supplementary Table S4), thus supporting the illumination of the cerebellar cortex in this analysis.
Finally, we asked if these two co-expressed cliques are associated with particular cell type or any other functional or cellular property. Examining cell-type-specific microarray data which we have for 64 cell types [43] revealed that 4 of them (stellate basket cells, granule cells, oligodendrocytes and Purkinje cells) are considerably populating the cerebellum (Supplementary Figure S2). Further, we looked through different coronal sections through the cerebellum from the Allen Reference Atlas [41] and visually compared them to sections of the sum of expressions of genes in Cliques I & II. Figure 6 shows the normalized volumetric expression quantities of both cliques along with the closest coronal section of the mouse brain in the Allen Reference Atlas. One can see that the voxels with the most intense expression in both cliques tend to follow the granular layer. Hence, the results of these analyses suggest that genes in both clique I & II tend to be over-expressed in granule cells in the cerebellar cortex. Using the Bioconductor GOstats package in R, we two biological processes: “Transmission of nerve impulse” (P = 0.001842), and “Ion transport” (P = 0.000733) and one cellular component “Vesicles” (P = 0.001134) that were enriched with genes from Clique I (Supplementary Table S5). Unfortunately, the number of genes in Clique II was too small for this analysis.
In this study, we explored the co-expression network 26 autism genes within the framework of 3,041 genes exhibiting the highest-quality expression data in the Allen Mouse Brain Atlas database [30]–[33]. The significantly tighter co-expression connectivity among the 26 autism genes than other genes, implies common functional properties for these genes in the mouse brain. Further investigation into the co-expression patterns of these genes revealed two cliques of co-expressed genes that were significantly dominated by autism genes. Genes in both these cliques shown significant over-expression in the cerebellar cortex, and particularly in sections that are predominantly populated by granular cells. Some regions of the cerebral cortex are also highlighted by the second clique (Figure 5), but to a lesser extent than the cerebellar cortex. Another recent study of our group examining the expression of the same autism gene set (AutRef84) in different human tissues, found a statistically significant enrichment in the frontal cortex [44]. The cerebral cortex was highlighted in other neuroanatomical studies of autism in both human [45], [46] and mouse [47] and is known to play a central role in cognitive and emotional processing [48], which are key deficits in autism and other neuropsychiatric disorders. In addition, a recent neuroimaging study [49] highlighted functional sub-regions in the cerebellum as playing a role in both motor and cognitive tasks. Other genes associated with autism have been shown to play a role in the development of this region [50]–[53]. Our results, provide additional support in the potential involvement of the cerebellum in autism etiology, and suggest additional candidate genes that are also over-expressed in the cerebellar cortex.
Two recent transcriptomic analyses in human brains [6], [54] revealed additional co-expression modules enriched with autism-associated genes. Some of these modules partially overlap with our findings in either gene content or brain regions, suggesting common functional and neuroanatomical properties of autism gene in both human and mouse brains. Together, these studies provide new insights into the specific gene networks and brain regions that could be involved in autism etiology.
A major strength of our study is the utilization of the Allen Mouse Brain Atlas [30]–[33] which comprises a high-resolution genome-wide exploration of gene expression in the adult mouse brain. This data allows one to explore gene expression properties up to a resolution of 200 microns, which provide a good distinction between different brain regions as well as potentially tell apart different sub-regions and cell types. Another advantage of this study is the focus on those genes exhibiting the highest expression correlation between the coronal and sagittal sections [35] as well as restricting the autism gene to a subset demonstrating, to the best of our knowledge, the most compelling associations to autism susceptibility. These strict criteria reduce the chances of erroneous results. Our study has also some pitfalls. First, the analyses were done on data from mouse brains. Since autism is a human condition, one may ask how well finding of this study apply to human brain. A recent study comparing postsynaptic protein composition between mouse and human suggest a high correlation between these two mammals in those matters [55]. Nevertheless, similar analyses in the human brain are still required to provide a finer validity to our findings. In addition, the strict criteria used here, restricted the number of studied genes to 3, 041 and 26 autism genes which are roughly represent 15% and 31% of the genes in the Allen Brain Atlas and AutDB datasets respectively. Such a small number of genes might results in false negatives and hence might miss other co-expression properties and brain regions associated with autism. Hence, larger studies are needed to complement the results of our analysis.
In conclusions, our study provides unique insights into the neuroanatomical co-expression properties of genes associated with autism in the mouse brain and suggest specific regions implicated in autism etiology. Complementing these findings with additional genomics and neuroimaging analyses from both mouse and human brains would help gaining a broader picture of the autistic brain.
|
10.1371/journal.pcbi.1004712 | Spatial Metrics of Tumour Vascular Organisation Predict Radiation Efficacy in a Computational Model | Intratumoural heterogeneity is known to contribute to poor therapeutic response. Variations in oxygen tension in particular have been correlated with changes in radiation response in vitro and at the clinical scale with overall survival. Heterogeneity at the microscopic scale in tumour blood vessel architecture has been described, and is one source of the underlying variations in oxygen tension. We seek to determine whether histologic scale measures of the erratic distribution of blood vessels within a tumour can be used to predict differing radiation response. Using a two-dimensional hybrid cellular automaton model of tumour growth, we evaluate the effect of vessel distribution on cell survival outcomes of simulated radiation therapy. Using the standard equations for the oxygen enhancement ratio for cell survival probability under differing oxygen tensions, we calculate average radiation effect over a range of different vessel densities and organisations. We go on to quantify the vessel distribution heterogeneity and measure spatial organization using Ripley’s L function, a measure designed to detect deviations from complete spatial randomness. We find that under differing regimes of vessel density the correlation coefficient between the measure of spatial organization and radiation effect changes sign. This provides not only a useful way to understand the differences seen in radiation effect for tissues based on vessel architecture, but also an alternate explanation for the vessel normalization hypothesis.
| In this paper we use a mathematical model, called a hybrid cellular automaton, to study the effect of different vessel distributions on radiation therapy outcomes at the cellular level. We show that the correlation between radiation outcome and spatial organization of vessels changes signs between relatively low and high vessel density. Specifically, that for relatively low vessel density, radiation efficacy is decreased when vessels are more homogeneously distributed, and the opposite is true, that radiation efficacy is improved, when vessel organisation is normalised in high densities. This result suggests an alteration to the vessel normalization hypothesis which states that normalisation of vascular beds should improve radio- and chemo-therapeutic response, but has failed to be validated in clinical studies. In this alteration, we show that Ripley’s L function allows discrimination between vascular architectures in different density regimes in which the standard hypothesis holds and does not hold. Further, we find that this information can be used to augment quantitative histologic analysis of tumours to aid radiation dose personalisation.
| It is increasingly recognised that an important aspect of cancers is their heterogeneity [1]. This heterogeneity exists between patients, between different tumours within a single patient [2], within the differing cellular populations in a single tumour and even at the genetic scale between cancer cells originating from the same ancestor [3]. In particular, microenvironmental heterogeneity is becoming widely accepted as a key factor in tumour progression and response to therapy [1]. Nutrients, growth factors, extracellular matrix and other cell types are all part of the normal tissue that surrounds and pervades a solid tumour and has been shown to vary widely across different tumour stages and types. This is, in part, due to the dynamic and heterogeneous interplay between the tumour and its microenvironment.
Radiation biologists have, for many years, understood the importance of cell biological and microenvironmental factors on radiation response. Current radiation therapy dose planning, however, largely neglects this information and is, instead, based on years of clinical experience using intuition and trial and error. As such, there remains limited tailoring of dose planning to an individual patient’s tumour. With the advent of modern quantitative histologic [4] and biological imaging methods [5], however, this paradigm is poised to change.
Research in this area over the last decade [6] has sought to understand the macroscopic spatial distribution of hypoxia within tumours using non-invasive imaging. This information has then been utilised to develop spatially heterogeneous dose plans to improve tumour control. For example, Malinen et al. [7] inferred average oxygen concentrations from radiocontrast concentrations measured by Dynamic Contrast Enhanced (DCE) Magnetic Resonance Imaging (MRI) in a dog sarcoma. Other work to understand the effects of radiation in individual patients has utilized MRI scans in combination with mathematical models of tumour growth. These models have incorporated heterogeneity in cell type by considering a two compartment spatial partial differential equation (PDE) model, separately accounting for proliferation and motility, without consideration of oxygen effects [8] to explain different radiation responses measured by changes in tumour size over time. More recently, cellular automaton models of stem cell driven tumours, comprised of populations with differences in proliferative phenotype [9] were used to compare tumour response to a range of spatially heterogeneous radiation dose plans or to different sequencing of radio- and chemo-therapeutic strategies [10]. What is lacking to date, however, is research into how tissue level microenvironmental heterogeneity can affect radiation response, and how this could be inferred from patient data.
Heterogeneity in tumour oxygenation, in particular the occurrence of hypoxia, is a well-known cause of radiation therapy failure [11, 12]. Work done in vitro to understand differences in radiation effect due to oxygenation differences have been valuable, and have established an empiric relationship called the Oxygen Enhancement Ratio (OER) [13] to understand how radiation efficacy varies with oxygen concentration. These studies do not, however, allow us to understand the effects of radiation in vivo, as they do not consider the heterogeneity of oxygenation at the microscopic, cellular scale.
Beyond the macroscopic changes in oxygenation, vessel and cellular density and metabolism, a number of theoretical studies have suggested that the local microscopic heterogeneity in oxygenation can vary widely in space and time in tumours [14–16] and healthy tissues [17] alike. In addition a large body of work has sought to understand an apparent paradox of therapy directed at angiogenesis, the process of new vessel creation [18]. In short, it was thought that by blocking a cancer’s ability to create new vessels, it would be possible to starve the cancer of nutrients and oxygen, quickly leading to its demise. While effective anti-angiogenesis drugs have been developed, this promise never came to fruition. The leading hypothesis to explain this failure is termed the ‘vascular normalization hypothesis’ [19], which suggests that these drugs, which block vascular endothelial growth factor (VEGF), do not simply inhibit new vessel production, but instead work by normalizing the vascular bed in question by pruning out ineffective vessels and enhancing flow in others. This hypothesis, while well supported by experimental work, has not yet been able to fully explain results at the clinical scale [20]. These studies highlight an opportunity to improve our understanding of how spatio-temporal oxygen dynamics at the cellular scale affect tissue-level response to therapy. To this end, we develop a computational, hybrid cellular automaton model of a tumour growing within a surrounding normal tissue in a vascularized domain which we use to investigate whether spatial statistics gleaned from measures of vascular organisation can be used to predict radiation efficacy. By identifying broad relationships between in silico tissue architecture and radiation response, we aim to progress toward a translatable method of radiation plan optimization using information extracted from biopsies.
The remainder of this paper is structured as follows. We first elucidate our methods by describing the underlying rules and parameters governing the cellular automaton as well as the method of calculating oxygen transport and uptake. In the results section we describe simulation results concerning healthy tissue growth in regularly arranged and then heterogeneous vascular architectures. We then describe tumour growth and invasion in regular architectures and follow this with observations concerning the distribution of oxygen tension in different possible vascular organisations. We then develop a metric, based on Ripley’s L function [21], which allows us to quantify and correlate these patterns with radiation response. We end with a discussion of how this metric reconciles some of the difficulties with the vascular normalisation hypothesis, and suggest how the metric might be used in the clinic to personalise radiation dose planning.
We consider the effect of a heterogeneous microenvironment through the inclusion of en face blood vessels modelled as point sources of oxygen, and through competition of tumour cells with an initial field of healthy cells. While the assumption that blood vessels are point sources, and lie en face may be a simplification of the true biological complexity, it serves as a reasonable, and commonly used [15, 16, 22], starting point for modelling. We begin by creating a hybrid cellular automaton (HCA) model [23] in which we describe cells by individual agents whose states are updated over synchronous discrete time steps of fixed duration on the timescale of the cell cycle (automaton time steps), and which occupy sites on a two-dimensional square lattice representing a slice of tissue. The size of the lattice spacing is chosen such that each automaton element is approximately the size of a single cell. A second, identical lattice is created on which we approximate the continuous concentration of a freely diffusible molecule, representing oxygen which is updated on a finer timescale (oxygen time step—see supplemental information for further description of the relationship between these timescales). While the cells and oxygen distribution are updated on separate lattices, each influences the other. The feedback between the cells and the microenvironment is captured through a partial differential equation (PDE) governing oxygen transport and consumption. Although the parameter values in our model are drawn from data on a particular cancer type, the primary brain tumour glioblastoma, most of the underlying model assumptions are likely to apply to many other vascularised solid tumours.
Cell fate decisions in our model are determined by a number of microenvironmental and cell state-specific thresholds and values. The order in which cells are chosen to decide their fate is computed in a random fashion so as to avoid any order bias. To determine the position of cells within our domain, we tessellate the continuous domain on which the PDE is defined into squares of size Δx × Δx to arrive at a regular lattice occupied by both cells and vessels. While all cells are assumed to be the same size and shape (a single lattice site), we model two cell types, cancerous and healthy cells, each of which can divide to produce two identical daughters. The probabilities and thresholds for these cell fate decisions are listed in Table 1 and Fig 1, respectively, and will be discussed individually in the coming sections.
Having described each of the constituent parts of the model, we now describe how they are coupled to form the HCA. We consider a two-dimensional lattice of size N × N, with each lattice element identified by coordinates (iΔx, jΔx) where i, j ∈ {1, 2, …, N}.
As schematised in Fig 1, at each automaton time step every cell (chosen in random order) in the domain is subject to a series of decisions based on the current automaton state and local oxygen concentration, c(x, t). The specific checks that each cell undergoes, regardless of its phenotype, are:
compare c(x, t) to cap and cp; consume oxygen at rate μi rc, die, or remain quiescent; check number of free neighbouring lattice sites; determine proliferative behaviour: if proliferative constraints are met (space and oxygen), then consume additional oxygen for proliferation [53] and place daughter cell in randomly chosen empty neighboring lattice site, otherwise become quiescent.
As discussed in the previous section, the organisation of vessels in our model can have a significant effect on both the carrying capacity of a domain and also the resulting cellular-oxygen distribution of the cells inhabiting the domain. It is exactly these two values (cell number and surviving fraction) which influence our computation of total surviving cells, a measure critical for comparison across heterogeneous samples, and to understanding radiation efficacy at larger scales, like TCP. How these effects are governed by vessel organisation however, we have not yet elucidated. In this section we will investigate the effects of differing vascular patterns on radiation efficacy.
The vascular normalisation hypothesis [67] suggests that radiotherapy should be more efficacious when applied to tissues with normalised vascular beds. To test this in our model, we introduce a spatial metric by which to understand the overall level of spatial heterogeneity in our simulations.
We will utilise the variance stabilized Ripley’s L function [21] to measure the deviation from homogeneity in our vessel distributions. This measure, which is a function of distance, describes the average number of points within a given distance of any other point. For a complete description of this measure, see the supplemental information.
In order to allow for easy correlation, we first distill Ripley’s L function (which is a function of distance) into a single number by taking the mean value from 0 − 19 cell diameters (from adjacent to a distance beyond the ability to affect one another). We calculate the mean Ripley’s L function for all of the 500 simulations in each vessel number case and plot it against the associated carrying capacity (Fig C in S1 Text). We find a significant negative correlation for the lower vessel densities which loses statistical significance (p-values not shown) as the domains begin to reach confluence for all but the minority of vessel arrangements.
Fig 8 shows scatter plots of Ripley’s L function and surviving cells after radiation in 6 representative families of the 500 simulations with increasing vessel numbers. We find that at low vessel densities there is a strong negative correlation between Ripley’s L function and surviving cells, but this correlation changes sign at high vessel densities, explaining the counter-intuitive change in surviving fraction after Θ = 0.01 in Fig 7. This suggests that in situations where the vessels are more rarefied, normalising existing vasculature could actually make radiation less effective, in contrast to the vessel normalisation hypothesis.
We have used an HCA model of vascular tumour growth in a planar domain to investigate the dependence of cellular oxygenation and radiotherapy efficacy on vascular density and patterning. Our results indicate that simple spatial summary statistics such as Ripley’s L function, which could be easily obtained from biopsy images, may predict, together with more standard measures like vessel density, radiation therapy efficacy and the effect of vascular normalisation on this. While our model is parameterised from glioblastoma data, we anticipate these results to be more widely applicable to other cancer types.
Our results corroborate those of Alarcón et al. [14], who also used a HCA approach to studying vascular tumour growth. However, our work differs in several key ways. Specifically, the vasculature in their model lies in plane rather than en face, a choice which was made consciously in our model to ease future translational validation with patient tissue samples. Further, we have ignored several mechanisms of competition in order to focus on defining summary measures of cellular-oxygenation status. These differences aside, Alarcón et al. found, as we did, that heterogeneity on the scale of oxygen concentration affects cell growth, both in overall speed of tumour growth and also in shape of resulting tissue. Further, they reported significant heterogeneity in steady state oxygen concentration across their domains which was dependent on vascular organisation, but this was never explored in terms of radiation effect. A similar finding was reported by Al-Shammari et al. in a biophysical model of healthy muscle tissue [17], this time in a system utilizing en face oxygen sources, as in our work. We have observed similar changes in cellular-oxygen distributions at equilibrium, and indeed, it is these changes that drive our findings concerning radiation effect.
We have seen that the assumption that mean vessel density, and subsequently mean oxygenation, can be used as a surrogate for the number of surviving cells after radiation is insufficient. Further, we found that the relationship between each of our vessel pattern measures and the surviving cells after radiation exhibits a sign change in the mid vessel density range. This change of sign in correlation means that the patterns, within a given vessel density, that take the extreme values of each measure of vascular organization can represent opposite extremes of radiation response. This suggests that if one were to perturb the vasculature, for example toward a more homogeneous distribution, one could induce opposite effects on radiation response, depending on the vessel density.
With Folkman’s discovery, in 1971, of a master tumour angiogenesis regulating factor [18], the world thought that the suggested method of blocking this factor, by which it was promised that we could ‘starve’ tumours of their oxygen supply, would dominate cancer research. It was thought that a cure for cancer would occur in a short time period. However, early trials of single agent anti-angiogenic drugs failed to produce results [20]. Later trials, with combinations of chemotherapy and anti-angiogenic drugs, however, showed promise, but even this was discordant with the leading hypothesis describing the mechanism of anti-angiogenic therapy, which was thought to entirely starve tumours of blood supply.
It was not until 2001, when Jain suggested the ‘vascular normalisation hypothesis’ [19], that these results could be understood under a single rubric. Jain suggested that anti-angiogenic drugs (sometimes called vascular normalisation therapy, or VNT), instead of entirely blocking new vessel formation, worked to normalise vasculature, pruning inefficient vessels and creating a more regular lattice, thereby improving drug and oxygen delivery. More recent iterations of this hypothesis also include improvement of the efficiency of existing vessels (reviewed by Jain [67, 68]).
While this advance in our thinking has provided a way to explain the counter-intuitive results of many trials, it still does not explain why these combinations of anti-angiogenic drugs with chemotherapeutics (or radiation therapy) do not help all patients. Using our simple model system, we have observed a changing correlation with spatial measures and radiation response. This suggests that in certain cases, ‘improving’ the homogeneity of vascularisation would hurt the radiation effect, whilst in other cases it would help: a heterogeneous response to ‘normalisation’ of vascular patterning. We show that for certain cases, vessel density held constant, normal vascular patterning can respond either better or worse to radiation therapy (Fig 8).
To translate these conclusions to the clinic would first require biological validation of these hypotheses, to include a careful study of variation in spatial vessel patterning on different spatial scales within the same tumor, or possibly between spatially distinct biopsies. Further, there are a number of model limitations to overcome. Specifically, the assumption that all vessels are en face and all vessels are the same in terms of size and efficiency. Validation could be achieved either through in vivo window chamber experiments, or indirectly through examination of post-radiation surgical specimens. Model development, to include the addition of angiogenesis, biologically derived vessel geometries, vessel size heterogeneity and collapse, could be used to extend these predictions to more realistic situations.
If these limitations were addressed, and validation was achieved, the technology exists currently to take advantage of this new idea. Specifically, macroscopic oxygen concentrations could be inferred from DCE MRI (or other advanced imaging) to create optimised dose plans, as suggested by Malinen et al. [7]. This information about putative vessel density could then be coupled with histologic measurements using automated localisation of vessels [27] and an algorithm to calculate Ripley’s L. These two pieces of information could then be incorporated by creating a temporally and a spatially optimised radiation plan whereby the appropriate radiation dose would be delivered before VNT to the area that would suffer after normalisation, and then the final radiation could be delivered after VNT to the area that would benefit from it.
|
10.1371/journal.pmed.1002374 | Harmonization of community health worker programs for HIV: A four-country qualitative study in Southern Africa | Community health worker (CHW) programs are believed to be poorly coordinated, poorly integrated into national health systems, and lacking long-term support. Duplication of services, fragmentation, and resource limitations may have impeded the potential impact of CHWs for achieving HIV goals. This study assesses mediators of a more harmonized approach to implementing large-scale CHW programs for HIV in the context of complex health systems and multiple donors.
We undertook four country case studies in Lesotho, Mozambique, South Africa, and Swaziland between August 2015 and May 2016. We conducted 60 semistructured interviews with donors, government officials, and expert observers involved in CHW programs delivering HIV services. Interviews were triangulated with published literature, country reports, national health plans, and policies. Data were analyzed based on 3 priority areas of harmonization (coordination, integration, and sustainability) and 5 components of a conceptual framework (the health issue, intervention, stakeholders, health system, and context) to assess facilitators and barriers to harmonization of CHW programs. CHWs supporting HIV programs were found to be highly fragmented and poorly integrated into national health systems. Stakeholders generally supported increasing harmonization, although they recognized several challenges and disadvantages to harmonization. Key facilitators to harmonization included (i) a large existing national CHW program and recognition of nongovernmental CHW programs, (ii) use of common incentives and training processes for CHWs, (iii) existence of an organizational structure dedicated to community health initiatives, and (iv) involvement of community leaders in decision-making. Key barriers included a wide range of stakeholders and lack of ownership and accountability of non-governmental CHW programs. Limitations of our study include subjectively selected case studies, our focus on decision-makers, and limited generalizability beyond the countries analyzed.
CHW programs for HIV in Southern Africa are fragmented, poorly integrated, and lack long-term support. We provide 5 policy recommendations to harmonize CHW programs in order to strengthen and sustain the role of CHWs in HIV service delivery.
| Community health workers (CHWs) play an important role in scaling up HIV services in the community and contribute in major ways to achieving HIV goals.
Multiple disparate CHW programs for HIV often exist in a single country.
The lack of coordination between CHW programs and the lack of integration of CHW programs into larger health systems may have impeded the full realization of the potential impact of CHWs in HIV.
A total of 60 interviews with government officials, expert observers, and donors were conducted to investigate the harmonization of CHW programs for HIV in Lesotho, Mozambique, South Africa, and Swaziland.
We provide key policy recommendations targeted to policy-makers and program implementers at the national, regional, and global level to strengthen and sustain the role of CHWs in HIV service delivery.
To increase harmonization, decision-makers in HIV-endemic settings should further develop government CHW programs; officially recognize nongovernmental CHW programs; standardize CHW training, incentives, and services; provide an organizational structure dedicated to community health initiatives; involve the community in decision-making; and provide adequate and long-term resources for CHW programs for HIV.
A range of actions are available for a more harmonized approach to CHW programs for HIV in the context of complex health systems and multiple donors.
Increased coordination, integration, and sustainability of CHW programs for HIV may increase the impact of CHW-led HIV services and improve the health, economic, and social outcomes of their clients.
| The importance of community health workers (CHWs) and their contribution to healthcare and health promotion have garnered increasing attention from governments, donors, health systems researchers, and planners [1]. CHWs have been suggested to play a “transformative” role in scaling up HIV services in communities for achieving HIV goals and improve linkages between those who need care and those who can provide it [2–7]. Two recent reviews of CHWs in HIV care described positive effects on HIV service organization, delivery [8], and cost [9], but highlighted the need for programs to be integrated into the wider health system for sustainability [10,11].
CHW programs are believed to be weak in many countries, poorly coordinated with fragmented attempts at improvement, and, with multiple disparate CHW programs, often observed in a single country [12–14]. There have been few examples of integrating CHW programs into national health systems and human resources for health (HRH) policy and plans. This may be impeding the potential impact of CHWs in HIV [12,14]. Fragmented and vertically supported CHW programs that are not aligned with health systems may manifest in program inefficiencies, poor accountability, and poor coordination that can block effective and sustainable programming, service delivery, training, deployment, and support to CHWs [15,16]. To strengthen and sustain the role of CHWs in HIV service delivery, “harmonizing” CHW-led HIV services (defined as increasing coordination, integration, and sustainability), appears necessary to address commonly encountered challenges of integration within country HRH plans, lack of clarity around ownership of CHW programs, poor linkages with the health system that enable effective team-based care, and inadequate focus on sustained support for CHWs [15–19]. Box 1 provides an overview of the expansion of CHW programs.
In recent years, alignment of CHW initiatives has been promoted through several policy initiatives [13,29–31]. The Global Health Workforce Alliance (GHWA)—an alliance of government leaders, donors, health workers, and civil society—called in 2013 for a coherent and harmonized approach to CHW support within countries [13,32] and articulated 3 guiding principles for such harmonization (the “three ones”): one national strategy, one authority, and one monitoring and accountability framework [13]. The GHWA’s public commitment brought critical attention to the harmonization of CHW programs for HIV because of the history of CHW cadre creation specific for HIV. Harmonization has become increasingly important with expansion of CHW HIV responsibilities for achieving 90-90-90 goals [7]. In February 2017, UNAIDS announced the need to scale up the use of CHWs [33]. These initiatives suggest that disease-specific organizations are investing in scale-up of CHW programs [7,33].
Harmonization of CHW programs can occur along a number of dimensions: within the context of this study, we identify coordination among development partners, integration into the broader health system, and assurance of the program’s sustainability to be 3 priority areas. Among CHW programs, coordination efforts seek to reduce duplication, fragmentation, confusion created by competing models, and overlap of responsibilities of differently trained CHWs in the same geographic areas [13,34]. Integration refers to the absorption of CHW programs into existing networks of larger health systems, primarily Ministries of Health or a large private provider (nongovernmental organization [NGO] or commercial) [13]. Sustainability is a key issue for CHW programs supporting HIV services that have been solely supported by donor programs, such as the US President's Emergency Plan for AIDS Relief or the Global Fund [13,17,35]. S1 Box provides additional details on each of these priority areas.
Harmonization of CHW programs may have been further impeded by lack of understanding of key harmonization issues, experiences, and lessons learned to date [13]. Systematizing collective learning and understanding of CHW program harmonization is a practical and formative step towards building evidence in this field. This study contributes new evidence on the harmonization of CHW programs for HIV in two ways. First, it describes factors that help or hinder efforts toward achieving a more harmonized approach to implementing large-scale CHW programs in the context of a generalized HIV epidemic, complex health systems, and multiple donors. We identify reasons for keeping fragmentation for innovations and for exploring different approaches. Second, the study provides policy recommendations for a more harmonized approach to CHW programs for HIV. The premise of the study was not to suggest that all CHW programs should be harmonized but rather to assess mediators and trade-offs associated with harmonization.
This study was reviewed by the Harvard T.H. Chan School of Public Health Institutional Review Board and considered exempt from full review since it is based on anonymous data with no identifiable information about survey participants. Additionally, the study was approved by the national ethical committees of Lesotho, Mozambique, South Africa, and Swaziland.
We conducted 4 descriptive country case studies between August 2015 and May 2016. We purposively sampled 4 countries heavily affected by the HIV epidemic and that have had some success in integrating donor-supported CHW-led activities into their national health systems: Lesotho, Mozambique, South Africa, and Swaziland. HIV prevalence among adults aged 15–49 ranged from 10.5% in Mozambique to 27.7% in Swaziland [36] (see S2 Text for additional contextual information on CHW programs delivering HIV services in each country). While the selection of case studies in any comparative analysis is to some extent arbitrary, the 4 countries are geographically and historically closely related, which implied that their broad context would be relatively similar compared to other potential countries. By attempting to hold the epidemic and broad context constant among the 4 case studies [37], our aim was to focus on mediators of harmonization at the intervention, stakeholders, and health system levels, which were key elements in our analytic framework [34]. Qualitative data were collected in each country through semistructured face-to-face interviews.
Three categories of respondents were eligible to take part in the study. First, we interviewed government officials, defined as key staff involved in the programmatic and coordination roles of provincial or national CHW programs. This included officials involved in planning and coordinating activities, technical staff involved in the oversight of government-funded CHW activities for HIV, and officials participating in national advisory boards and working groups related to CHW work. Second, we interviewed donors, defined as staff involved in determining which programs to fund, those who oversee grants or loans funding CHW activities for HIV, implementation and evaluation of CHW programs providing HIV services, and those sitting on national advisory boards and working groups related to CHW work. Third, we interviewed expert observers, defined as any staff working with ongoing CHW activities for HIV at a nongovernmental (or faith-based) organization or academic expert within a participating country. From each participating organization, efforts were made to include at least 1 technical staff and 1 staff involved in planning and coordination of activities. Exclusion criteria were age less than 18 years old at the time of the study and inability to provide informed consent.
The interview process was single-staged. We used purposeful sampling to identify interviewees by constructing a preliminary list of 15–20 relevant stakeholders prior to arriving in each country. We chose to interview government officials, expert observers, and donors for three reasons. First, stakeholders were likely well informed about the programmatic and coordination roles of provincial and national CHW programs. We sought participants who were aware of the coordination, planning, sustainability, and operations of CHW-led activities, such as the implementation designs and operational processes of specific CHW programs. Second, we aimed to identify participants likely to interact with multiple CHW programs and who had responsibilities above and beyond the CHW program that they were primarily involved with—in order to provide insights into approaches to better coordinate programmatic activities and governance across CHW programs. In other words, we sought participants likely to have experience with existing efforts to reduce duplication, fragmentation, inefficiencies created by competing models, and overlap of responsibilities of differently trained CHWs in the same geographic areas. Finally, we were interested in assessing power structures and how participants navigated the complex web of political actors and institutions involved with coordination and harmonization of CHW programs at the provincial and national level. We assessed key drivers of political agenda setting and policy diffusion around harmonization (S1 Table: Country data analysis sheet: “political strategies for harmonization”) [38,39]. All participants were involved with national, nongovernmental, or disease-specific CHW programs that deliver HIV services including: education, testing and counseling, and antiretroviral therapy (ART) adherence support. For consistency and completeness, we wrote semistructured interview guides. Interviews lasted approximately 1 hour and aimed to identify likely facilitating and impedimentary contributors to harmonization. Interviews were conducted by data collectors with experience in qualitative data collection; either in English, in a mix of the country’s national language and English, or in the country’s national language with professional interpretation. Each interviewee was informed of the purpose of the study, our intention to take notes, and our process for handling interview data. Interviewees provided verbal consent. Interviews were recorded and transcribed into electronic text.
First, we identified major issues and relationships surrounding harmonization of CHW programs based on reviewed documents. We conducted a narrative review of existing published and grey literature building on recent work on coordination, integration, and sustainability of CHW programs [13,29,40]. We used an analytic framework developed ex ante to map overarching findings and inform the design of our semistructured questionnaires. Additional details on our narrative review and analytical framework are provided in S1 Text and elsewhere [41]. Second, to assess key constructs of our framework for each country, we analyzed data from interviewees. Transcripts were read in their entirety, coded to indicate whether they provided information on harmonization, and extracted data was summarized and grouped under priority areas of harmonization and components adapted from the analytic framework (S1 Table). The 3 priority areas of harmonization (coordination, integration, and sustainability) are interlinked components. We, therefore, additionally assessed how priority areas of harmonization related to each other. Outcome variables for this analysis were coordination, integration, and sustainability. Explanatory variables were mediators of harmonization of CHW programs, categorized by 5 components of our analytic framework.
We interviewed a total of 60 participants including government officials, donors, and expert observers in each of the four countries. Selected characteristics of the study participants are shown in Table 1. The country case studies proceed as follows. First, we identify key themes arising from the interviews related to each priority area of harmonization and categorize them by components of our analytic framework, separated for each country. We include country-specific examples, list success stories and failures, and provide insights into advantages and disadvantages of harmonizing of CHW programs. In Table 2, we show an assessment of the current state of harmonization of CHW programs, separated by country. Second, we list key mediators of harmonization of CHW programs identified by study participants. Finally, we compare all 4 countries and show overarching findings. In Table 3, we show overarching mediators of harmonization. In Table 4, we map overarching findings for priority areas for harmonization in our analytic framework.
Using multicountry case studies, we found that CHWs supporting HIV programs were in general highly fragmented and poorly integrated into national health systems in Southern Africa. We assessed harmonization across 3 priority areas: coordination, integration, and sustainability of CHW programs. Across 4 countries, there are frequently large existing national CHW programs, generally supported in part by the national government. In addition to large-scale government programs, participants often reported a large range of nongovernmental CHW programs delivering HIV services, either alone or in combination with other services. The coexistence of multiple stakeholders supporting CHWs, while generally welcomed (e.g., as an initial emergency response to the HIV epidemic), has often led to multiple parallel systems and fragmentation of services, and could be impeding the potential bigger impact of CHW programs. Frequently mentioned challenges included a lack of oversight and accountability at multiple administrative levels, poor support of CHWs and CHW attrition, and misalignment with a community’s health needs. To our knowledge, this study is the first to develop multicountry case studies to inform decision-making around harmonization of CHW programs for HIV in highly affected countries.
A wide range of facilitating and impedimentary factors were identified to guide policy recommendations and inform decision-making on harmonization of HIV community programs. To make our study’s findings as generalizable as possible to other settings, we extracted overarching mediators of harmonization. Although the countries we assessed are different in many aspects, we found that a large number of mediators were consistent across settings. Key facilitators of harmonization across the study’s countries included (i) a large existing national CHW program and recognition of non-governmental CHW programs, (ii) the use of common incentives and training processes for CHWs, (iii) the existence of an organizational structure dedicated to community health initiatives, and (iv) the involvement of community leaders in decision-making. Other important facilitators of harmonization included the use of local health facilities and political support at all levels of the government. Key barriers to harmonization included (i) a wide range of stakeholders with various guidelines and timelines, (ii) lack of equivalence between training programs of various CHWs, and (iii) lack of ownership and accountability of non-governmental CHW programs, in addition to limited financial and human resources across programs. Future research could explore and test the applicability of each of the mediators identified in our study in other settings. For instance, use of ward-based outreach teams was an important facilitator of harmonization in South Africa [42], but it remains unclear to what extent this reflects an opportunity to achieve harmonization in other countries. Swaziland, for instance, is geographically nested within South Africa and shares many of the health system and contextual characteristics with neighboring KwaZulu Natal province in South Africa. Another important facilitator of harmonization in South Africa that may be of interest to decision-makers in other countries was the use of “clinic committees,” which engaged community leaders and helped create links between communities and health facilities. Conversely, respondents mentioned that Swaziland had integrated specific CHW programs, such as the Red Cross volunteers, shifting from a model of volunteerism to monthly stipends [43]. These experiences may provide important lessons and opportunities for other settings. In Box 2, we show 5 policy recommendations to help achieve harmonization of CHW programs, based on the overarching findings from our 4 country case studies.
Although this study identified many benefits of harmonizing CHW programs for HIV, respondents also brought up disadvantages to harmonization. Indeed, greater central control of CHWs may not necessarily be positive. Integrating CHWs entirely into a single national health system or large private provider, for instance, could reduce flexibility (e.g., in recruiting community members) or reduce trust by community members in CHWs if they do not trust services provided by the government or other large institution [44]. Although most participants largely appeared in favor, some expressed concerns around increased harmonization. Participants raised questions about how financial and technical harmonization of CHW programs would work in practice. The government’s capacity to provide high quality supervision and management may be limited, and quality of services may deteriorate if CHWs are overburdened with tasks. In Mozambique, the multitude of existing CHW responsibilities reduced their ability to complete additional tasks. Some participants also questioned how to continue tailoring HIV services to community needs, as opposed to only what the government might (mis)understand as being their need, and suggested that such a process might take a long time. Some respondents also voiced concerns that older CHWs may lose their positions if they do not fit new criteria in a harmonized system; and that harmonization of incentives could redefine the scope of work expected of CHWs, thereby changing the rules of engagement for CHWs who were recruited as volunteers. Large national CHW programs are increasingly moving away from volunteerism as a reliable motivator for CHWs to deliver substantial and measurable results long-term. Long-term planning among development partners, such as around the need to exit or transition to a larger provider when program funding ends, may also limit more aspirational planning. The “accompaniment approach,” for instance, suggests to support CHW programs for HIV until countries have the capacity to sustain service delivery [45]. Overall, however, participants saw more advantages than disadvantages to harmonization, such as increased impact of CHW programs, and more long-term solutions to community health issues. Stakeholders were generally supportive of a more harmonized approach to implementing CHW programs.
Finally, in addition to the trade-offs associated with harmonization, we note the importance of the broader health system and context supporting CHW programs. As CHW programs are harmonized and scaled up in settings with scarce resources [33,46,47], a critical question is whether the benefits of these investments will be reaped without concurrent effort to strengthen the broader health system. Integration depends on the health system’s capacity to absorb the program—financially and structurally. CHW programs, for instance, may need to be coordinated with local clinics and teaching facilities that have adequate financial and human resources (e.g., nurses, physicians, teachers) to coordinate CHW programs operating in their catchment area. Similarly, the broader context, including “demographic, economic, political, legal, ecological, sociocultural…and technological factors”, can play a critical role in enabling or hindering the adoption of health sector innovations [34]. The donor environment, such as fiduciary requirements imposed on donor agencies by governing structures, influences the health system but extends beyond it.
This study has a number of limitations. First, our findings are based on case studies which were purposively selected based on the epidemiological context and existence of CHW programs. We were also unable to interview many more relevant decision-makers. However, we generally reached saturation through data triangulation and interviewing a large number of key stakeholders who were well informed. Second, we interviewed stakeholders directly involved in management, support structures, or funding of CHW programs. Our intention was to assess the views of policy-makers and program implementers who make decisions regarding CHW programs and are currently at the forefront of planning and engaging with the expansion of CHWs. Future research should include the bottom-up perspectives from CHW clients, CHWs, and community members. The views of CHW clients are of particular interest, as the end goal of health systems reform is to improve health outcomes as opposed to programs or processes. CHWs are at the forefront of delivery and are able to assess whether the identified issues and mediators of harmonization are feasible. Third, although our results are similar across countries, the findings are country-specific. For instance, it is possible that facilitators to harmonization in one country are barriers in another. While we caution against generalizing to other settings, our findings are likely informative for decision-makers involved with CHW programs, particularly in the context of a large, generalized HIV epidemic, complex health systems, and multiple donors. Fourth, the concepts of CHWs and harmonization are multifaceted and can be interpreted differently by different people. Our interviewers, however, began by asking study participants how they defined CHWs and asked specific questions on clearly defined components of harmonization. Fifth, harmonization is in many cases a challenging and gradual process. In settings with decentralized government, for instance, minor, incremental steps may be required to achieve full harmonization of CHW programs for HIV. “Partial” integration may mean the cadres are not necessarily owned by the MOH, but by community groups, NGOs, or private providers and simply more strongly aligned to national systems. Conversely, countries with a stronger central government, fleeting political will, or a substantial existing national CHW program may be able to achieve effective harmonization quickly. The government of Rwanda, for instance, coordinated salary support to CHWs with an NGO [48], and Brazil and Ethiopia placed their CHWs entirely into an existing civil service structure [16], which considerably facilitated harmonization of their community programs. Finally, we note that harmonization is comprised of multiple interlinked components. There is significant overlap of topics across areas of harmonization and elements of the analytic framework. The perception of a CHW program’s effectiveness among community members and policy-makers, for instance, is important for both integration and sustainability of CHW programs [49,50]. Both coordination and integration into the wider health system are also oft-cited facilitators of sustainability (e.g., for their contribution to be sustained, CHW programs may need to be integrated into the wider health system).
Multiple disparate CHW programs for HIV exist in 4 Southern African countries. The lack of coordination between CHW programs and the lack of integration of CHW programs into larger health systems may have impeded the full realization of the potential impact of CHWs in HIV. To strengthen and sustain the role of CHWs in HIV service delivery, decision-makers in HIV-endemic settings should take the following actions: further develop government CHW programs; officially recognize nongovernmental CHW programs; standardize CHW training, incentives, and services; provide a dedicated organizational structure for community health initiatives; involve the community in decision-making; and utilize a favorable political window of opportunity for increased harmonization. Adequate and long-term resources are urgently needed to support a more harmonized approach towards CHW programs for HIV.
|
10.1371/journal.pgen.1006517 | Evolved genetic and phenotypic differences due to mitochondrial-nuclear interactions | The oxidative phosphorylation (OxPhos) pathway is responsible for most aerobic ATP production and is the only pathway with both nuclear and mitochondrial encoded proteins. The importance of the interactions between these two genomes has recently received more attention because of their potential evolutionary effects and how they may affect human health and disease. In many different organisms, healthy nuclear and mitochondrial genome hybrids between species or among distant populations within a species affect fitness and OxPhos functions. However, what is less understood is whether these interactions impact individuals within a single natural population. The significance of this impact depends on the strength of selection for mito-nuclear interactions. We examined whether mito-nuclear interactions alter allele frequencies for ~11,000 nuclear SNPs within a single, natural Fundulus heteroclitus population containing two divergent mitochondrial haplotypes (mt-haplotypes). Between the two mt-haplotypes, there are significant nuclear allele frequency differences for 349 SNPs with a p-value of 1% (236 with 10% FDR). Unlike the rest of the genome, these 349 outlier SNPs form two groups associated with each mt-haplotype, with a minority of individuals having mixed ancestry. We use this mixed ancestry in combination with mt-haplotype as a polygenic factor to explain a significant fraction of the individual OxPhos variation. These data suggest that mito-nuclear interactions affect cardiac OxPhos function. The 349 outlier SNPs occur in genes involved in regulating metabolic processes but are not directly associated with the 79 nuclear OxPhos proteins. Therefore, we postulate that the evolution of mito-nuclear interactions affects OxPhos function by acting upstream of OxPhos.
| Maintaining two distinct mitochondrial haplotypes with functionally important nucleotide variations in a single population is difficult. If interactions with the mitochondria affect the nuclear genome such that some alleles are more advantageous when they occur with a specific mitochondrial genome, then there is an additional genetic burden that also should favor a single mitochondrion. Yet in a single, well mixed Fundulus heteroclitus (teleost fish) population where individuals appear to freely interbreed, there are two mitochondria with five amino acid substitutions between them. We investigate this population and show that individuals with different mitochondria have several hundred loci with extreme allele frequency biases. These differences are not due to neutral or demographic processes and affect cardiac energetics. We conclude that the allele frequency differences are most likely evolving by natural selection and affecting ATP production and energy homeostasis.
| The Oxidative Phosphorylation (OxPhos) pathway is composed of approximately 89 proteins encoded by the two genomes in animal cells: all 13 mitochondrial proteins and 76 nuclear proteins. These proteins form the five OxPhos enzyme complexes and are responsible for most cellular ATP production. Genetic defects in the OxPhos proteins affect at least 1 in 8,000 people and are the cause for the most common inherited metabolic diseases [1–3]. The interactions between mitochondrial and nuclear OxPhos proteins may be equally as important as deleterious mutations within either genome. The importance of these interactions have been demonstrated experimentally in humans, Mus, Drosophila, Tigriopus, Callosobruchus, and Saccharomyce where healthy nuclear and mitochondrial genome hybrids between species or among distant populations within a species affect fitness and OxPhos functions [4–19]. For example, hybrid breakdown due to mito-nuclear incompatibilities among Tigriopus californicus populations occur in F2 individuals [7] and alter ROS production [6], OxPhos enzyme activities [20], ATP production and survival [21]. These mito-nuclear interactions (GxG) are often affected by the environment as demonstrated by Drosophila simulans mitochondria that have pleiotropic effects at high environmental temperatures when substituted into one D. melanogaster genotype but not another [11, 22, 23]. Similarly, in seed beetles, Callosobruchus maculatus, temperature dependent metabolic rates rely on the interactions between the mitochondrial and nuclear genomes [24]. These mito-nuclear interactions that affect OxPhos are biologically important because they affect fitness (egg production, survivorship, and mating success) [5, 7, 10, 11, 21–23, 25, 26]. In general, these data suggest that mito-nuclear interactions among species or divergent populations are likely to affect an organism’s physiology and these interactions are environmentally dependent [11, 26–29].
Mito-nuclear interactions between different species or populations affect biological function [4–19]. However, it is less understood whether these interactions impact individuals within a single natural population. Theoretically, natural selection due to mito-nuclear interactions could alter allele frequencies when one mt-haplotype has greater fitness with a specific nuclear allele [30]. To determine if mito-nuclear interactions affect genotypes in naturally occurring populations, we examined a Fundulus heteroclitus population from a single inter-tidal estuarine creek. This population, just south of the Hudson River in Mantoloking NJ, USA, has two major mt-haplotypes with five non-synonymous substitutions: a “northern” haplotype, common in populations north of the Hudson River and a “southern” haplotype, common in populations south of the Hudson River [31]. The genetic divergence among mt-haplotypes may have been influenced by a historical break at the Hudson River due to the last glaciation [32], enhancing nucleotide pattern differences between northern and southern populations [33]. Populations with two distinct mt-haplotypes are the result of secondary-intergradation, whereby migrants meet where there was once a physical barrier [34]. Importantly, F. heteroclitus has large populations with low migration rates and is adapted to local environmental conditions [35–41]. Northern populations experience temperatures more than 12°C colder than southern populations and have evolved adaptations to temperature in cardiac metabolism and enzyme expression [39, 42]. Because the evolutionary variation in OxPhos genes has been associated with divergence among populations in response to environmental variation [30, 43–45] and has also been proposed to drive speciation [10, 29, 46–48], we might expect the mt-haplotype to affect the nuclear genotype and alter OxPhos function in the admixed population [49].
To explore potentially evolved epistatic interactions between nuclear and mitochondrial genomes, we addressed two questions: are allele frequencies at nuclear loci significantly different between the two specific mt-haplotypes, and if so, do these differences affect OxPhos function? To answer these questions, 155 Mantoloking, NJ F. heteroclitus individuals were genotyped at >11,000 SNPs, and their cardiac OxPhos metabolisms were measured. Individuals with southern and northern mt-haplotypes are present at a 60/40 ratio, respectively. We demonstrate significant allele frequency differences at 349 SNP loci between the two mt-haplotypes, and the different nuclear genotype and mt-haplotype combinations are associated with significant OxPhos metabolic differences.
Two genotyping by sequencing (GBS) [50] datasets were used: 1) MK-specific (individuals from a single Mantoloking, NJ population) to assess nuclear-mitochondrial associations, and 2) 3-population dataset (Maine, MK, and Georgia individuals) to ascertain the effect of recent admixture. After filtering, 11,705 nuclear SNPs were distributed among 10,180 F. heteroclitus genome-scaffolds [51] for the MK-specific dataset, and 10,471 for the 3-population dataset. All SNPs and annotations were derived from the 64bp sequence tags used to call SNPs.
Strong selection at many nuclear loci creates a genetic load that is detrimental to a species' survival [52, 53]. Therefore, it is unlikely that a population could maintain biologically important mito-nuclear interactions at many loci in a panmictic population (where migration = 0.5 of effective population size, Ne) [54–56]. We suggest that selection due to mito-nuclear interactions may occur if there is extensive standing genetic variation and many genes of small effect affect biological traits. To investigate whether these interactions between genomes do affect allele frequencies, we calculated FST values for each of the 11,705 SNPs in the MK-specific dataset, using the two mt-haplotypes as independent groups or populations. FST provides a statistically robust measure of the relative allelic variation between groups versus within groups. We denote this within population value as wFST. To be clear, although we are examining a single population, we use the two mt-haplotypes as artificial populations for wFST calculations. We found that 349 nuclear SNPs have wFST values that are large statistical outliers (p<0.01; Fig 1). Supplemental tables (S1 and S2 Tables) provide information on genome location, read depth, wFST values, p-values and allele frequencies for all 11,706 SNPs, and location, annotation and 64bp tag sequence for each of the 349 outlier SNPs.
SNPs with wFST outliers values are defined as having significantly large wFST values that are unlikely to occur relative to SNPs with similar heterozygosity (He) [57–59]. With 10% or 1% FDR, 236 or 72, respectively, of these 349 nuclear SNPs were significant; with a more conservative Bonferroni’s correction, 44 SNPs were significant (Fig 1). Among the 349 outlier SNPs, none had significant linkage disequilibrium with each other (D’ is not significant, p > 0.1 and r2 <0.3). Differences in minor allele frequency (MiAF) could affect wFST values [60], yet outlier versus non-outlier SNPs have similar MiAF: mean MiAF = 0.132 and 0.162 for 9,440 non-outlier, non-significant SNPs (this excludes SNPs that had significant FST values but not significant outliers) and 349 outlier SNPs respectively (S1 Fig). A separate analysis using allele counts for a Fisher Exact test revealed 229 SNPs with significantly biased allele frequencies (p<0.01). Of these 229 SNPs, 219 were also wFST outliers.
To investigate whether dividing individuals into two arbitrary groups can produce many significant wFST values, we produced a thousand random permutations for 9,440 non-significant SNPs (S2 Fig). None of the 1,000 permutations across 9,440 SNPs produces many wFST values as large as the 349 outlier wFST values as seen in the small overlap in their distributions (S2 Fig). Furthermore, the 99% upper confidence level for the arbitrary wFST values is less than the minimum wFST value for the 349 outlier SNPs (>0.002 and 0.0269, respectively, S2 Fig). Thus, grouping individuals into two arbitrary groups produces few SNPs with significant wFST values, indicating that the 349 outlier SNPs are statistically meaningful.
Each of the 349 outlier SNPs has a wFST value dissimilar from the genome wide wFST value (Fig 1) and is unlikely to occur (S2 Fig). However, even though the 349 outlier wFST values are unlikely, the data could still suffer from type I error. We proceed with our analyses using the 349 outlier SNPs for three reasons. First, to balance type I and type II errors—there are likely to be many more significant SNPs we have not discovered because of the weakness of adaptive tests [59, 61]. Second, the use of different FDR values (1% -10%) yields a large range of significant SNPs (72 to 236), and it has been argued that FDR of 20% or more may be appropriate [62]. Third, and most importantly, we are asking if these 349 outlier SNPs are related to population structure and mitochondrial physiology. Including false positives (type I error) will not bias these tests except to make them less likely to find significant structure or association.
Given the evolutionary history of F. heteroclitus and the observation that the MK population has both mt-haplotypes, recent admixture may bias allele frequencies between individuals with northern and southern mt-haplotypes. In order to ascertain the effect of recent admixture, we used Admixture version 1.3.0 to infer ancestries from the 3-population SNP dataset (Fig 2). For the Admixture analysis, we thinned the 3-population 10K SNPs so that all SNPs were >100bp apart, resulting in 3,700 thinned SNPs. Among these 3,700 SNPs, the MK population’s average admixture was 3.2% and never exceeded 14.7%. The plot of ancestry fraction (Q values) from Admixture clearly distinguishes Maine and Georgia from MK (Fig 2B). These analyses indicate that the MK population is a separate and independent population from Maine and Georgia with little recent ancestral admixture and that allele frequency differences between mt-haplotypes within MK are not due to shared genealogies with mt-DNAs. Significant LDs for 3,700 thinned SNPs in the 3-population data are rare and physically close together (S3 Fig): 5 SNP pairs have significant LD (FDR <0.1) with the largest distance = 222 bp. In comparison, nearly all SNP pairs within a scaffold (98%) are > 1,000 bp apart (S4A Fig).
Even though there is little indication of recent admixture (Fig 2), the outlier wFST values could arise with the recent admixing of distinct populations resulting in SNPs with linkage disequilibrium over large distances [63, 64]. Thus we might expect long distance LD for the 349 outlier SNPs with other nuclear SNPs or with mt-haplotypes. Among the 11K MK SNPs are 221,348 calculated LDs using a 50 SNP sliding window; 6,401 of these are significant with 10% FDR [65], and yet only 66 are significant and >100bp apart (S4B Fig). That is, 99% of all SNPs with significant LDs are less than 100bp apart (S4B Fig). This is also true for the 349 outlier SNPs. For each of the 349 outlier SNPs, there are only 229 SNPs in significant LD (FDR 10%) with any other of the 11K SNPs (S4C Fig). Of these 229, six are between SNPs >100bp apart, and among these six, five are less than 200bp apart. One of the 349 outlier SNPs is in LD with another SNP greater than a million bp apart, yet this SNP lacks significant LD with many other closer SNPs. Between each of the 11K SNPs and the mt-haplotype, none are in LD with any reasonable FDR (minimum FDR 0.57). Thus, none of the 349 outlier SNPs are in LD with the mitochondria. The lack of LD between any of the 349 outlier SNPs and the mitochondria reflects the relatively small allele frequency differences between mt-haplotypes for the 349 outliers SNPs (S1 Fig). Specifically, no SNP is close to fixation between mt-haplotypes (i.e., a difference close to 1). These patterns of LD among nuclear SNPs, including between the 349 outlier SNPs with mitochondrial SNPs, do not support recent admixture. Thus, based on LDs, there is little evidence that the biased allele frequencies in the 349 outlier SNPs are due to recent admixture.
We performed a Tajima’s D analysis on the 11,706 SNPs with a 50bp window using VCFtools [66]. The Tajima’s D value distributions for non-significant and outlier 349 SNPs (S5 Fig) are similar, and the 1% tail of these distributions have equal frequencies of both significant and non-significant SNPs (p-value > 0.2). Tajima’s D compares pair-wises differences to the number of segregating sites where linked SNPs should share large positive values when associated with balancing selection. Our Tajima’s D analysis provides no support for balancing selection. This result likely reflects how SNPs are called: SNPs are called from 64 bp sequenced tags that are typically hundreds of thousands to millions of bps apart (S4A Fig), and only 0.1% of SNPs are in LD with other SNPs (S4B Fig). The limited LD among SNPs within a 64bp tag, and the distance among tags suggest that significant Tajima’s D values are unlikely to occur because the assumption for Tajima’s D analyses is that sites affected by non-neutral processes will affect nearby linked sites. The rarity of SNPs in LD (S4 Fig) and the large MiAF (S1 Fig) suggest that SNPs have existed as long-term standing genetic variation; Tajima’s D analyses are unlikely to detect selection in this case [67].
To confirm and explore if there is any hidden population structure, we applied discriminant analysis of principal components (DAPC) [68]. Using the 349 outlier SNPs, DAPC identified two groups as the most parsimonious grouping: those associated with the northern or southern mt-haplotype (Fig 3A and 3B). Using all 11K SNPs or 9,440 non-outlier, non significant SNPs, revealed a single grouping (S6 Fig). The observation that distinct groups are not seen with all 11K SNPs (S6 Fig) or the 9,440 non-outlier, non-significant SNPs but are seen with the 349 outlier SNPs lends additional evidence that there is a single well mixed population and the outlier SNPs clearly discriminate individuals into two mt-haplotype associated groups.
STRUCTURE analyses [69] with K = 1–5 using MK’s 349 outlier SNPs corroborated the DAPC results (Fig 3C): K = 2 was much more likely than K = 1 and was the best supported K based on ΔK (the rate of change Ln-likelihood [70, 71]). With K = 2, individuals form two distinct clusters associated with each mt-haplotype. Larger Ks did not produce more definitive groups.
To understand the magnitude of the 349 outlier wFST values, we compared these within population fixation index values to those between MK and a population 40 miles further south (Rutgers, NJ). Unexpectedly, MK wFST values were larger than the between population FST values (Fig 4). Rutgers only has the southern mt-haplotype, and between population FST values are a function of mt-haplotype: FST values are smaller for comparisons of nuclear SNPs between Rutgers and just individuals in MK with southern mt-haplotypes than for comparisons of Rutgers to individuals in MK with northern mt-haplotypes (Fig 4, red versus blue curve). Specifically, comparisons using MK individuals with northern mt-haplotypes are right-shifted with more loci with large FST values. These data indicate that the genetic distances of the 349 outlier SNPs between mt-haplotypes are larger within the MK populations than between populations and that the genetic distances between populations for nuclear SNPs are a function of the mt-haplotype.
In summary, the MK population is a well-mixed population with a few hundred unlinked nuclear SNPs that have significant allele frequency differences between the two mt-haplotypes (Fig 1). These biased allele frequencies create large wFST values that are distinct from the rest of the genome and unlikely to occur based on data permutations. The hypothesis that the MK population is well-mixed is supported by A) Admixture analysis on 3-population 3,700 nuclear SNPs (Fig 2), B) the DAPC indicating a single population based on all 11K nuclear SNPs (S6 Fig) and C) the few significant SNPs in LD across the genome or with the mt-haplotype (S3 and S4 Figs). These data, and the observation that MK wFST values among 349 outlier nuclear SNPs are larger than the FST value for the same loci among populations (Fig 4), suggest that that demographic effects including migration would not cause an association between the mitochondrial and nuclear genomes. These 349 outlier nuclear SNPs have an allele frequency difference between mt-haplotypes of 11.19% (95% CI = 10.69 to 11.69%), more than 3 times larger than the 3.28% (95% CI = 3.22% to 3.35%) allele frequency difference for the remaining 9,440 non-outlier, non-significant SNPs (S1 Fig). For the 349 outlier SNPs, this allele frequency difference translates to wFST values >0.054 (Fig 1), compared to the majority of SNPs (9,440, 81%) where 95% of wFST values are <-0.001 and have p-values >0.1 (Fig 1). The evolutionary importance of these 349 outlier SNPs is suggested by wFST values that are not likely to occur relative to other SNPs that share similar He. These differences are not due to different allele frequencies (MiAF = 0.132 for the 349 outlier versus 0.162 for 9,440 remaining SNPs, S1 Fig) or heterozygosity (0.23 and 0.20 for 349 outlier and 9,440 non-significant SNPs, respectively). Yet for the 349 outlier SNPs, as indicated by wFST values, the allele frequency differences between the two mt-haplotypes are significantly larger than the allele frequency variance within these groups; this is unusual relative to 96% of the other SNPs. These data on the 349 outlier nuclear SNPs are surprising given what we know about F. heteroclitus ecology and reproduction: individuals occupy small home ranges in estuaries [36, 72] and share a common reproductive strategy of laying and fertilizing eggs in the upper intertidal zone [73, 74]. We tentatively conclude that these 349 outlier SNPs are most likely evolving by natural selection due to the interactions between the nuclear and mitochondrial genomes.
If these differences in 349 outlier SNPs are meaningful, we would expect differences in biological functions between mito-nuclear genotypes. To determine if the 394 outlier SNPs affect a biological function, cardiac OxPhos metabolism was measured as State 3 respiration (an integrative measure of ADP and substrate dependent mitochondrial respiration) and compared to mito-nuclear genotypes among the MK individuals (Fig 5). In ANCOVA (with Admixture coefficients, acclimation, assay temperature and body mass as covariates), we used four mitochondrial groups as the main effect (Fig 5A). The four mitochondrial groups are based on STRUCTURE analysis of the 349 outlier SNPs (Fig 3C). Most individuals have >70% of nuclear alleles associated with one of the two mt-haplotypes (Fig 3C). However, 21 individuals have mixed ancestry; these individuals shared at least 30% of nuclear alleles with the opposite cluster (northern 349 SNP alleles with southern mt-haplotype or southern 349 SNP alleles with northern mt-haplotype). This defines four groups: the two main structure groups (Fig 3C) and two groups with mixed ancestry from each cluster (individuals with >30% of the alternative allele). Among these four mito-nuclear groups, State 3 is significantly different (Fig 5A, p < 0.0194); admixture was not significant (p >0.8) while mass, acclimation and acute temperature were significant (p < 0.05). State 3 respiration was significantly lower in individuals with the northern mt-haplotype compared to those with the southern mt-haplotype (Fig 5A, Tukey post hoc test). Individuals with “mixed” nuclear backgrounds showed intermediate mitochondrial respiration. For mixed ancestry individuals with a northern mt-haplotype, having a larger number of “southern” associated nuclear alleles increased respiration rates, whereas the opposite effect was observed for individuals with a southern mt-haplotype with a larger number of “northern” associated nuclear alleles. Thus, this analysis indicates that variation in the nuclear genome modulates mt-haplotype effects on OxPhos metabolism.
A second analysis regresses State 3 respiration against the fraction of southern mt-ancestry (Fig 5B) using the same covariates as above (Admixture coefficients, body, mass, acute and acclimation temperature). The fraction of southern mt-ancestry is significant (p < 0.0055) and explains 6% of the variance (Fig 5B). These data indicate that individuals with the greater number of southern alleles with a northern mt-haplotype have greater OxPhos metabolism and individuals with more northern alleles with a southern mt-haplotype have lower OxPhos metabolism. Notice, that we form a polygenic score for each individual using the 349 outlier SNPs, which defines the fraction of ancestry related to southern mt-haplotype. Thus, because the MK 349 SNPs have significantly biased allele frequencies and their inferred ancestry of K = 2 reflects the two mt-haplotype, significant regression indicates a significant mito-nuclear interaction.
The potential genes that could affect State 3 mitochondrial respiration include approximately 89 nuclear and mitochondrial proteins that form the five OxPhos complexes, and approximately 1,500 other nuclear genes involved in mitochondrial functions [75]. Using BLAST [76], we aligned the 64bp sequences containing outlier SNPs against the F. heteroclitus genome. Sixty-four base pairs were used because this is the length of the sequences retained from sequencing and pipeline analysis[77]. Although all outlier sequences aligned to the genome, only 162 aligned to annotated genes. Many of these genes encode transcription factors (e.g., zinc finger proteins), are involved in signaling pathways (e.g., 1-phosphatidylinositol 4,5-bisphosphate, GTPase or receptors), or are trans-membrane proteins. Four genes were identified that likely influence OxPhos metabolism (Table 1). The first, acyl-coenzyme A thioesterase, which is mitochondrially localized, affects intermediates in the citric acid cycle, which forms OxPhos substrates via lipid metabolism regulation [78–80]. A second gene, adenylate kinase, regulates mitochondrial respiration by altering ADP/ATP ratios [81] and creating feedback signal communication. A third gene, NAD-dependent malic enzyme, catalyzes a reaction that forms pyruvate from malate with the reduction of NAD+ to NADH [82–84]. The mitochondrial variant acts as a regulatory enzyme, allosterically activated by fumarate and inhibited by ATP [85]. These substrates, products and allosteric regulators are all involved in OxPhos metabolism. The fourth gene, ribosomal mitochondrial protein (MRP) S35, is responsible for translating the 13 mitochondrial proteins making up the OxPhos pathway. MRPs are linked to human mitochondrial disorders such as Leigh Syndrome, multiple mitochondrial dysfunctions syndrome, Russell-Silver syndrome, Spinocerebellar ataxia with blindness and deafness, Stuve-Wiedemann syndrome, and Usher syndrome [86].
The described genes may play an important metabolic role through mitochondrial OxPhos protein regulation and translation. Surprisingly, none of the 349 outlier sequences aligned to the 76 nuclear encoded proteins that form the OxPhos complexes. There are two possible reasons for this: 1) none of the 11,000 SNP sequences include OxPhos proteins, or 2) SNPs in these proteins are not affected by the mito-nuclear interactions. Five of the 11,000 SNPs mapped to nuclear proteins that directly participate in OxPhos: one in Complex I, two in Complex II, and two in Complex V. Yet, there was no allele frequency bias in these OxPhos proteins. Mito-nuclear interactions may still affect other nuclear OxPhos genes that were not sequenced in our study. However, given our data, we postulate that epistatic selection affects OxPhos functions and is acting upstream of the OxPhos pathway.
We can compare our SNP data to previously published gene expression data on mt-haplotype effects in F. heteroclitus. Flight, et al. [87] published the effects of mt-haplotype, sex and hypoxia on mRNA expression measured in individuals from the same population. We used their published microarray probes to determine the co-localization of our 349 significant outlier SNPs with the oligos used to measure gene expression. Among the 349 significant outlier SNPs, 136 SNPs are within 150Kb of the oligos used by Flight et al. [87] (S3 Table). For these 136 genes, 24 (18%) have a significant p-value (p <0.01) for one of the three main effects: sex, hypoxia or mt-haplotype. One SNP (S10023_121524) had a significant mt-haplotype effect and was ~55Kb distant from a gene annotated by Flight et al., as spectrin beta chain, brain 1. For the four genes listed in Table 1, two co-localized with oligos used by Flight. The SNP associated with acyl-coenzyme A thioesterase 9 is ~12Kb downstream from Flight’s diamine acetyltransferase 1 gene, which has a significant hypoxia effect. The SNP associated with adenylate kinase has the same annotation as in Flight’s microarray, is 32 Kb downstream, and is different between sexes. The significant SNPs are from sequences that cover less than 0.0001% of the genome, and Flight’s microarray investigated ~4,000 genes (~14% of F. heteroclitus annotated genes) [88]. The observation that any of our SNPs are co-localized to any of these support the concept that they are functionally and evolutionarily important.
Although significant associations between nuclear loci and mt-haplotype point towards selection on mito-nuclear interactions, an alternative explanation could be assortative mating. Allelic bias may occur if individuals can recognize mates with similar mt-haplotypes and accordingly, preferentially mate. However, no evidence of this has been documented in F. heteroclitus. Our data is more likely explained by selection on mito-nuclear interactions for two reasons. First, there is a large overlap in 349 significant loci found by using two different methods: Arlequin [57] FST test and Fisher Exact test. It is very unlikely that these significant associations are random. Second, Admixture using 3,700 SNP from the 3-population set (Fig 2) and DAPC using all 11K SNP from the MK SNP set (S6 Fig) indicate little if any population structure within the MK population. Assortative mating would have to be highly selective to maintain allelic bias because the unlinked 349 outlier SNPs would come to equilibrium if only drift and incomplete isolation was responsible. Thus, assortative mating seems unlikely; instead MK seems to be a well-mixed random breeding population.
Difficult questions to answer are how two mt-haplotypes are maintained in a single population and how so many loci are potentially affected by natural selection due to GxG interactions. Theoretically, it is difficult to maintain functionally different mitochondrial haplotypes in a single population due to GxG interactions [54–56]. Mito-nuclear polymorphisms can be maintained with sex-linked loci under restricted conditions [56, 89]. Yet, the 349 outlier wFST loci are distributed over 100s of scaffolds, and thus it seems unlikely that they are limited to sex-linked chromosomes. Mutation-selection balance also seems unlikely because there is a high frequency of the minor alleles: the average heterozygosity for the 349 outlier wFST loci is 0.23 and is similar to neutral loci. It is also difficult to suggest migration or other demographic effects because the allele frequency difference within MK population for the 349 outlier wFST loci is larger than the allele frequency difference among populations (Fig 4). Additionally, the 3-population data set indicates a well-mixed population (Fig 2). The high allele frequencies for both alleles among the 349 outlier wFST loci (S1 Fig) suggest that there is balancing selection, which might arise if the two alleles have different fitness effects in different environments. GxE interactions where allele effects have a high variance among environments could maintain selectively important polymorphisms especially if there is extensive pleiotropy [56] or unpredictable environmental variations [90]. Notice, because these individuals were captured together in the same estuarine creek, it is unreasonable to suggest spatial variation in the environment; instead temporal variation is common in the F. heteroclitus environment and may contribute to maintaining the observed mito-nuclear genetic variation.
What we do know is that there is environmentally dependent, adaptive divergence in OxPhos mRNA expression among populations [39, 91–93], suggesting that GxE interactions are possible. In other species, mito-nuclear interactions have pleiotropic effects [11, 22] and affect genome wide mRNA expression patterns [94, 95]. Thus, although lacking data to specifically address the evolutionary genetics that maintain selectively different mito-nuclear interactions, we suggest that temporal environmental variation affects mito-nuclear polymorphisms that have pleiotropic effects. The hypothesis of pleiotropic effects is supported by the diversity of annotations associated with the 349 outlier wFST, which include transcription factors and signaling pathway genes that are likely to have a wide diversity of phenotypic effects.
Although epistatic interactions between mitochondrial and nuclear genes have been shown to affect overall organismal fitness and metabolic activity [5, 8, 11, 14, 15, 22–24, 96], these studies have used divergent mt-haplotypes and divergent nuclear backgrounds or fail to show an effect on allele frequencies in natural populations. The data we present show that mito-nuclear interactions influence allele frequencies in a natural, freely interbreeding population. We show that 349 outlier SNPs have greater allele frequency differences between mt-haplotypes than within a mt-haplotype, creating large, significant wFST values (wFST values are FST values within a population between two mt-haplotypes). The distribution of wFST values within and FST values between populations for neutral SNPs is different from that of the 349 outlier SNPs. These 349 outlier SNPs were used to form a polygenic factor, where the individual scores affected OxPhos metabolism, supporting the hypothesis that mito-nuclear interactions are evolutionarily important. These observations are difficult to resolve with any neutral or realistic demographic mechanisms. Thus, we tentatively conclude that the most parsimonious explanation is that selection on mito-nuclear interactions is strong enough to alter allele frequencies for 100s of SNPs. The observation that several of these genomic SNPs are for genes that modulate OxPhos supports this hypothesis. The selection for mito-nuclear interactions that modulate OxPhos may occur if there is extensive standing genetic variation and the genes have small effects.
For the sake of clarity, we use two informative names for the two datasets analyzed in this manuscript. “MK-specific” refers to the SNP dataset that is solely based on the 180 individuals from New Jersey. To define admixture, a second data set (“3-population”) includes these MK individuals and individuals from Maine (ME, n = 35) and Georgia (GA, n = 38). SNP discovery pipeline [77] defines SNPs that are polymorphic with specific frequency and read depth, and thus, while many of the SNPs in the MK-specific and 3-population dataset are the same, 46% are unique.
Adult F. heteroclitus were collected during the summer months from Mantoloking, NJ (40.049427°N, -74.065087°W), Wiscassett, ME (43° 57’ 15.10”N, 69° 43’ 13.64”W), and Sapelo Island, GA (31° 27’ 13.39”N 81° 21’47.65”W). All fish were captured in minnow traps with little stress and removed in less than one hour. Fieldwork was completed within publicly available lands, and no permission was required for access. F. heteroclitus does not have endangered or protected status, and small marine minnows do not require collecting permits for non-commercial purposes. All fish were acclimated for 4 weeks to either 12°C or 28°C, temperatures naturally encountered in their natural environment. These two acclimation temperatures were used to explore how chronic (acclimation) and acute temperatures affect physiological functions [97]. Fish were exposed to a 14 hour light cycle, kept at 15ppt salinity and fed twice a day, 7 days a week. Housing, acclimation and non-surgical tissue sampling protocols were in compliance with and approved by the University of Miami Institutional Animal Care and Use Committee (IACUC).
DNA was isolated from fin clips and stored in 270 ul of Chaos (buffer 4.5M guanidinium thiocynate, 2% N-lauroylsarcosine, 50mM EDTA, 25mM Tris-HCL pH 7.5, 0.2% antifoam, 0.1M β-mercaptoethanol) with ~ 1g of silica beads and combined with 130 ul of 10X TE (100mM Tris pH 7.8, 10mM EDTA pH 8.0). Tissue was homogenized using zirconium beads. Supernatant was removed and placed in a new tube with 200 ul of 95% EtOH and mixed. This solution was then quickly added to silica columns for DNA isolation. Loaded columns were centrifuged for 1 minute at 6,000xg, and flow through was discarded. As modified from [98], columns were washed three times with 750 ul of protein wash buffer (70 ml 96% EtOH and 26 ml binding buffer which contained 6M guanidine thiocyanate, 20 mM EDTA pH 8.0, 10 mM Tris-HCl pH 7.5, 4% Triton X-100) followed by centrifugation for 1 minute at 6,000xg. Then, samples were washed with 650 ul wash buffer (60% EtOH, 50mM NaCl, 10mM Tris-HCl pH 7.4, 0.5mM EDTA pH 8.0) and centrifuged for 1 minute at 16,100xg followed by another wash with 650 ul wash buffer and centrifugation for 3 minutes at 16,100xg to dry the silica column. 100 ul of 0.1XTE (10mM Tris, 0.1 mM EDTA) was added to elute the genomic DNA upon centrifugation for 1 minute at 6,000xg.
Mitochondrial haplotypes were defined by PleI, and BstYI restriction digest of ND2 and cytochrome b respectively. The digests were run on a 1% agarose gel to separate DNA fragments. Individuals from Maine and Georgia were used as controls. Both restriction enzymes yielded the same results for each individual. Haplotypes defined by restriction enzymes were the same as mitochondrial SNPs identified by GBS (genotyping by sequencing [50]). There were 19 mitochondrial SNPs that were in complete linkage disequilibrium, LD, (D’ 1.0). A single mitochondrial SNP was imputed for all individuals and used to determine relationships among nuclear-mitochondrial genotypes.
Isolated DNA quality was assessed via gel electrophoresis, and concentrations were quantified using Biotium AccuBlue High Sensitivity dsDNA Quantitative Solution according to manufacturer’s instructions. After quantification, 100 ng of DNA from each sample was dried down in a 96-well plate. Samples were then hydrated overnight with 5 ul of water before AseI restriction enzyme digestion. This digest, based on in silco digest of the F. heteroclitus genome (NCBI accession JXMV00000000.1 [99]), should produce 523,349 fragments with 117,639 <500bp in size. Adaptors with separate barcodes for each individual (0.4 pmol/sample) were ligated to the genomic DNA after digestion with AseI. DNA samples were then pooled and purified using an equal volume of carboxyl coated magnetic beads (Fisher Scientific) in a PEG/salt solution (0.5 g beads in 100 mls of 20% PEG 8000, 2.5 M NaCl). Two bead purifications were used to select fragments between 100 and 400 bp. First, DNA less than 400 bp was separated from larger DNA which is bound to magnetic beads at low NaCl2 concentration (0.87 M), then bead-salt solution was raised (NaCl2 at 1.25M) so that only DNA larger than 100 bp are bound. These beads were washed with 70% EtOH, and DNA was eluted. The size range of purified products was verified using Agilent 2100 Bioanalyzer (Santa Clara, CA). A range of PCR cycles on the 100-400bp genomic fragments was used to optimize the amplification of restriction fragments using primers that anneal to the adapters. The distribution and concentration of the amplified library was verified using Agilent 2100 Bioanalyzer (Santa Clara, CA). DNA from the 18-cycle run formed the GBS library that was sequenced (Illumina Hi Seq 2500, 75bp single end reads; Elim Biopharmaceuticals, Inc., Hayward, CA).
The Java program, TASSEL [77] used the first 64bps of single end sequences and aligned them to the F. heteroclitus genome to call SNPs. The F. heteroclitus genome (NCBI accession JXMV00000000.1), which consists of 10,180 scaffolds plus mitochondria, was used to map sequencing reads. Two GBS datasets were produced: 1) the MK-specific, and 2) the 3-population dataset. For the MK-specific dataset, individuals were removed that had less than 50% of SNPs, reducing the number of individuals from 180 to 155. The data were filtered to remove SNPs with less than 1% minimum allele frequencies that occurred in less than 70% of individuals. Hardy-Weinberg expectation was calculated for individual loci using Arlequin v3.5.1.2 [57], and we excluded 256 SNPs where Ho>He and was significant (p<0.01). This latter filter is used to remove potential SNPs that represent differences between paralogs versus true allelic variants for a single locus [100]. For the 3-population dataset, individuals were removed that had less than 30% of SNPs, reducing the number of individuals from 257 to 234. SNPs that occurred in less than 77% of individuals were removed.
For the MK-specific dataset, allele frequencies were defined using adegenet in R [68], and minor alleles were defined among all 155 NJ individuals. That is, a minor allele was defined across all individuals even when their frequencies >0.5 within a mt-haplotype. Two approaches were used to identify allele frequencies that had a bias relative to mt-haplotype: Fisher-Exact test and outlier-test using Arlequin v3.5.1.2 [57]. The Fisher-Exact test determines the bias in allele frequencies at each locus relative to mt-haplotype using PLINK [101]. Arlequin was used to compare the relative genetic distance between the two mt-haplotype relative to other loci. Specifically, we used an outlier test to define fixation index (FST) values that exceed the expectation based on the observed data. For comparisons between the two mt-haplotypes, we use fixation index (FST), and for clarity we use wFST (within population among mt-haplotypes). To identify SNPs with wFST outlier values, we used Arlequin v3.5.1.2 [57]. Outlier wFST values are based on FDIST [58, 102] as implemented in Arlequin, where coalescent simulations are used to get a null distribution and confidence intervals around the observed values and then tested to determine if observed locus-specific wFST values can be considered as outliers conditioned on the globally observed wFST value.
For Admixture analysis we thinned the 3-population dataset, removing SNPs closer than 100 bp (as suggested by the Admixture manual [103]). Thinning resulted in 3,700 SNPs. These 3,700 SNPs were input into Admixture v.1.3.0 [104] to infer ancestries of ME, GA, and MK individuals and provide an unbiased estimation of overall population structure.
LD was determined for MK individuals in 1) all SNPs in the MK-specific dataset, and 2) among 3.7K thinned SNPs from the 3-population dataset [77]. LDs were determined using a moving 50bp-SNP window providing r2 (correlation coefficient), D’ and p-values associated with pairs of SNPs within and among scaffolds. The significant LD between SNP pairs and each SNP with mt-SNPs are reported as p-values <0.01 and with FDR correction [105]. FDR based on Benjamini & Hochberg [105] and were calculated in R using p.adjust [106].
Tajima’s D was calculated using VCFtools [66] with 50bp non-overlapping windows. VCFtools uses the physical distance (50bp) to calculate Tajima’s D. We used a 50bp window because nearly all SNPs within a 64 bp tag are captured by this window (i.e. SNPs occur at +10 bp in a tag-sequence). Using a 100bp window produced nearly identical results.
STRUCTURE v2.3.4 [69] was used to identify the number of ancestral populations (K) with similar allele frequencies and was also used to predict the magnitude of admixture within the single collection site. CLUMPAK [107] was used to average output from multiple STRUCTURE runs. For the 349 outlier SNPs, models allowing admixture and correlated gene frequencies were used with seven independent runs for each K-value from 1–5. Eleven thousand permutations with 11,121 initial runs (burn-in) were used for each run. The most parsimonious K was defined as that with the most likely K (largest mean Ln-likelihood) and the ΔK was based on the rate of change Ln-likelihood [70] using STRUCTURE HARVESTER [71]. We chose the most likely K if K was equal to 1, and used ΔK for K where the most likely was greater than one because ΔK can only resolve the best K with K >1.
Discrimination analysis of principal components (DAPC) was conducted in R using ‘adegenet’ [68]. DAPC uses the principal components of allele frequencies to infer the number of clusters of genetically related individuals by partitioning into a between-group and within- group component and maximizing discrimination between groups [108].
To compare SNP genome locations (position on specific scaffold) to the location of oligo-nucleotides used by Flight, et al. [87] in the construction of their microarray, we use bwa to align the oligo-nucleotides to the most recent F. heteroclitus genome at NCBI (GCA_000826765.1 Fundulus_heteroclitus-3.0.2 scaffolds). Most oligos used in their microarray are >250bp and few full-length oligos aligned to the genome (most likely due to introns). To overcome this problem, we used non-overlapping 50bp windows for the alignments.
All individuals used for GBS analyses had their cardiac OxPhos metabolism measured as described in [109, 110]. Heart ventricles were dissected, cut into halves, and half was placed into a muscle relaxation solution (10 mM Ca-EGTA buffer, 0.1 µM free calcium, 20 mM imidazole, 20 mM taurine, 50 mM K-MES, 0.5 mM DTT, 6.56 mM MgCl2, 5.77 mM ATP, 15 mM phosphocreatine, pH 7.1) [111]. The other half was saved for future RNA work. Tissues were then permeabilized using 2.5mg/ml saponin solution for 15 minutes, followed by 4 washes in relaxation solution for 5 minutes each [111]. Once permeabilized, tissues were immediately transferred to the respirometry chambers containing a respiration medium (5mM EGTA, 3mM MgCl2.6H2O, 60mM K-lactobionate 20mM Taurine, 10mM KH2PO4, 20mM HEPES, 110mM Sucrose, 1g/l BSA).
The acute effect of temperature on mitochondrial activity was measured at three temperatures: 12°C, 20°C, and 28°C. Activity was measured and analyzed using the Oxygraph 2-k and DatLab software (OROBOROS INSTRUMENTS, Innsbruck, Austria). Population, acclimation temperature, and acute temperature changes were all randomized. All OxPhos determinations were relative to the amount of DNA in the measured tissue. Respiration rates were measured as pmol O2 s-1 ml-1 per ng DNA. The detailed analyses of acclimation and acute effect on OxPhos function within population are lengthy and are the subject of a separate publication [109].
After addition of the tissue to the respiration chamber, state 3 was determined. State 3 is defined as routine oxygen consumption resulting in ATP production in the presence of substrates and ADP. First, the substrates pyruvate (5 mM), glutamate (10mM), and succinate (10mM) were added, followed by ADP addition (5mM, state 3); cytochrome c (10μM) was added to check mitochondrial membrane integrity [111]. The tissue was recovered after respiration assays, and total DNA was quantified using AccuBlue high sensitivity dsDNA quantitation solution (Biotium). All activity was normalized by ng/ul of DNA. OxPhos function is represented as residuals from acclimation, assay temperature, body mass, and percent admixture from the 3-population SNP dataset. Percent admixture from Admixture v1.3.0 had no significant effect on OxPhos function [112].
Adult F. heteroclitus were captured in minnow traps with little stress and removed in less than 1 hour. Fieldwork was completed within publicly available lands, and no permission was required for access. Housing, acclimation and non-surgical tissue sampling protocols were in compliance with and approved by the University of Miami Institutional Animal Care and Use Committee (IACUC).
|
10.1371/journal.ppat.1007469 | Role of a fluid-phase PRR in fighting an intracellular pathogen: PTX3 in Shigella infection | Shigella spp. are pathogenic bacteria that cause bacillary dysentery in humans by invading the colonic and rectal mucosa where they induce dramatic inflammation. Here, we have analyzed the role of the soluble PRR Pentraxin 3 (PTX3), a key component of the humoral arm of innate immunity. Mice that had been intranasally infected with S. flexneri were rescued from death by treatment with recombinant PTX3. In vitro PTX3 exerts the antibacterial activity against Shigella, impairing epithelial cell invasion and contributing to the bactericidal activity of serum.
PTX3 is produced upon LPS-TLR4 stimulation in accordance with the lipid A structure of Shigella. In the plasma of infected patients, the level of PTX3 amount only correlates strongly with symptom severity. These results signal PTX3 as a novel player in Shigella pathogenesis and its potential role in fighting shigellosis. Finally, we suggest that the plasma level of PTX3 in shigellosis patients could act as a biomarker for infection severity.
| Soluble pattern recognition molecules, PRMs, are components of the humoral arm of innate immunity. The long pentraxin 3, PTX3, is a prototypic soluble PRM that is produced in response to primary inflammatory signals. Shigella spp. are human entero-pathogens which invade colonic and rectal mucosa where they cause deleterious inflammation. We show that PTX3 acts as an ante-antibody and contributes to the clearance of extracellular Shigella. As a countermeasure, Shigella uses invasiveness and low-inflammatory LPS to control PTX3 release in infected cells. This study highlights that the extracellular phase of the invasion process can be considered the “Achille heels” of Shigella pathogenesis.
| The first line of immune defense against pathogens is guaranteed by the recognition of Pathogen Associated Molecular Patterns (PAMPs) by Pattern Recognition Receptors (PRRs). The family of PRRs includes secreted, membrane-bound and cytosolic PRRs [1]. Pathogenic organisms use sophisticated strategies to modulate PRR recognition and to control downstream signaling. Accordingly, the human enteropathogen Shigella exploits different mechanisms to hijack the innate immune response. The Shigella genus includes 4 serogroups: S. boydi, S. dysenteriae, S. flexneri and S. sonnei. and S. flexneri are the main serogroups circulating in industrialized and developing countries respectively [2], but most studies centered on the invasion process of Shigella in vitro and in vivo have been carried out with S. flexneri.
Shigella penetrates epithelial cells through a series of effectors secreted via a Type 3 Secretion System (T3SS) [3] encoded by a large virulence plasmid common to all Shigella serogroups. Once inside the colonic mucosa, shigellae either fuel or dampen the inflammatory response, depending on the step of the invasion process. In epithelial cells, Shigella multiplies freely within the cytoplasm [4] where the cytosolic PRR Nod1 recognizes cell-wall peptidoglycan (PGN) and activates NF-κB [5,6]. This leads to CXCL8/IL-8 production. IL-8 attracts neutrophils which are required for the clearance of shigellae but which also participate in the destruction of the epithelial barrier [7]. Within epithelial cells, S. flexneri changes its lipopolysaccharide (LPS) structure from a highly inflammatory hexa-acylated lipid A form to a less inflammatory tetra- and tri-acylated lipid A variant [8]. This low-inflammatory LPS is poorly recognized by the PRR Toll-Like-Receptor 4 (TLR4), making macrophages and neutrophils less able to control the infection. Furthermore, in B lymphocytes Shigella induces TLR2-mediated apoptotic death through a mechanism mediated by T3SS, independent of cell invasion [9]. These immune evasion strategies involve TLR2 and TLR4, which are PRRs present on the surface of many cell populations, suggesting that the extracellular step could be critical for successful infection. To gain insight about this poorly explored aspect of Shigella pathogenesis we analyzed the potential role of humoral PRRs, focusing on the possible involvement of Pentraxin 3 (PTX3), which has served as a paradigm of humoral innate immunity [10].
PTX3 is a key element of the humoral arm of innate immunity, downstream of and complementary to cellular recognition and activation. Pentraxins are an evolutionarily highly conserved superfamily of proteins. PTX3 is the prototypic long pentraxin, while the short pentraxins include C-reactive protein (CRP) and serum amyloid P component (SAP), acute-phase proteins in man and mouse respectively. PTX3 is rapidly produced and released by several cell types, e.g. mononuclear phagocytes, dendritic cells (DCs) and neutrophils [11] in response to primary inflammatory signals (e.g. TLR engagement, TNF-α, IL-1β). PTX3 binds selected microorganisms, including Aspergillus fumigatus, Pseudomonas aeruginosa [12, 13] and uropathogenic Escherichia coli (UPEC) [14]. It also promotes complement activation, thereby facilitating pathogen recognition by phagocytes [15]. All these features prompted us to investigate whether PTX3 could play a role in Shigella pathogenesis and how and to what extent this PRR is released upon Shigella infection. Our findings highlight that PTX3 is a new player in the Shigella cross-talk with the infected tissues and provide novel insights into the mechanisms of Shigella to control the production of this humoral PRR.
Firstly, we addressed the question of whether PTX3 could opsonize Shigella as reported with A. fumigatus, Salmonella typhimurium, P. aeruginosa, Neisseria meningiditis [12,13,16] and uropathogenic E. coli (UPEC) (14). We observed that PTX3 (50 μg/mL, 1,1 μM) opsonized the wild type S. flexneri 5 strain M90T (Fig 1A and 1B) and its plasmidless, non-invasive variant BS176 though to a lesser extent compared to P. aeruginosa (strain PAO1), used as a positive control (12, 17)
Epithelial cell invasion and macrophage death [18] are hallmarks of Shigella pathogenesis. We investigated whether PTX3 binding could affect epithelial-cell penetration by Shigella. M90T was incubated with different concentrations of recombinant PTX3 (0.05, 0.5, 5 and 50 μg/mL) prior to infection of HeLa cells. M90T treated with bovine serum albumin (BSA) (50 μg/mL) or with a rabbit polyclonal antibody raised against the T3SS-invasin IpaD (50 μg/mL), which is involved in bacterial internalization [19] was used in parallel. As shown in Figs 1C and S1A, PTX3 opsonization with only the concentration of 50 μg/mL significantly decreased the number of intracellular bacteria at 1 h and 2 h post-infection (p.i.) (p< 0.001: PTX3-opsonized M90T vs. M90T plus BSA at 2h) to levels like those induced by the anti-IpaD antibody.
In human peripheral blood monocytes-derived macrophages (MoMs), opsonization with recombinant PTX3 at same concentrations as above only determined a significant improvement of bacterial phagocytosis at the concentration of 50 μg/mL (Fig 1D and 1E, S1B Fig) (p< 0.01: PTX3-opsonized M90T vs. M90T plus BSA). Likewise, the rate of MoMs cell death, measured through lactate dehydrogenase (LDH) release, (Fig 1F and S1C Fig) was improved by PTX3 opsonization, as also shown with the anti-IpaD antibody [19, 20]. However, the cell death rate followed the trend as above since it was only increased with PTX3 at the concentration of 50 μg/mL (p< 0.01: PTX3-opsonized-M90T vs.M90T plus BSA) (S1C Fig)
PTX3 is also involved in complement cascade activation and regulation [21]. There is scant and dated literature on Shigella sensitivity to the bactericidal activity of serum [22], so we set up a bactericidal serum assay against S. flexneri M90T. In preliminary experiments using different serum concentrations (10%; 13%; 17%; 67%) for 30 min at 37°C, we found that M90T was sensitive to concentrations of > 10% pooled normal human serum (NHS). Exposure to 17% NHS killed around 60% of bacteria (Fig 1G). Addition of low concentrations of PTX3- 5–0.5 μg/mL in the bactericidal serum assay (17% serum) significantly increased bacterial death rate of the bacteria (Fig 1H).
In conclusion: PTX3-bound bacteria are partially impaired in their ability to invade epithelial cells and are better internalized by macrophages as shown with antibody-bound bacteria [23, 24]. Low doses of PTX3 increase the bactericidal activity of serum/complement.
We therefore passed to analyze a possible contribution of PTX3 in vivo, during infection. Several reports have described the therapeutic effect of recombinant PTX3 in the models of aspergillosis [25, 26, 27]; of P. aeruginosa acute and chronic lung infections [17, 28, 29]; of acute respiratory syndrome [30] and of influenza virus lung infection [31]
Mice are naturally resistant to Shigella oral infection. Therefore, alternative infection models have been developed and used [32). Among them, the intranasal route of infection in mice [33] has been extensively used and validated to study Shigella virulence and to analyze vaccine candidates [32, 33]. After intranasal infection (i.n.) with virulent Shigella, mice develop a dramatic pneumonia and die within a few days [32, 33].
In line with these issues, we investigated whether treating mice with recombinant PTX3 could affect the outcome of intranasal infection with M90T, as shown with the other pathogens as above. Based on previously performed pharmacokinetic analyses and therapeutic approaches in a murine model of lung infection with A. fumigatus [12, 25, 26, 27] and P. aeruginosa [17, 28, 29], we established a treatment schedule of daily intra-peritoneal (i.p.) injections with recombinant human PTX3 (0.5 mg/Kg, 11 μM) or vehicle, starting from the day of the i.n. inoculum with 3 x 108 colony forming units (CFU) of M90T (corresponding to the LD50 dose). The treatment with PTX3 was carried out for 48 hours (three PTX3 injections).
Following this experimental plan, we monitored animal survival/death (in three independent experiments) during 1 week. As expected [6, 34, 35], during the 72 hours, 55% of mice infected with M90T died (12/22) died. By contrast, the animals infected with M90T and treated with PTX3 survived even when the treatment with PTX3 was stopped (19/19), (p< 0.0002, two-tailed Mantel-Cox test) (Fig 2A)
At 72 h post-challenge, the bacterial load in lungs of M90T-infected animals was ~ 106 CFU (per lung), as previously published [6, 34, 35] whereas in PTX3 treated animals the number of CFU (per lung) was ~ 104 CFU, which was significantly (p< 0.0001) reduced with respect to that of infected, untreated animals (Fig 2B).
The local level of pro-inflammatory cytokines and chemokines was measured in BALs (broncho- alveolar lavage) (Fig 2C) and lung homogenates (Fig 2D). We chose some chemokines/cytokines, which have been previously shown to be influenced by PTX3 treatment [12 and 17, 29] and/or to be involved in Shigella infection [6, 34, 8]. CCL5 (RANTES) and IL-1β values were similar in BALs and lung homogenates for untreated and PTX3-treated infected animals, while those of TNF-α CXCL2/MIP-2 and CXCL1/KC were drastically decreased in both homogenates and BALs of PTX3-treated mice compared to infected-untreated mice (for all, p <0.0001, two-tailed Mann-Whitney test). Furthermore, we quantified the PTX3 levels in sera (Fig 2E), lung homogenates (Fig 2F) and BALs (Fig 2G). In lung homogenates of M90T-only-infected animals the PTX3 level was lower than that of PTX3-treated infected and uninfected mice (for both, p < 0.0001 two-tailed Mann-Whitney test). Likewise, in BALs and sera the levels of PTX3 were significantly (p < 0.0001, two-tailed Mann-Whitney test) reduced in M90T-only-infected animals with respect to those treated with PTX3.
At macroscopic examination (S2 Fig Top), lungs of infected mice were enlarged, of rubbery consistency, edematous and dusky red in color due to severe hyperemia. The lungs of infected and PTX3-treated animals were macroscopically like the controls, and they appeared aerated, pinkish and spongy.
In lungs of M90T-infected mice, hematoxylin-eosin staining showed acute pneumonia with severe bronchoalveolitis, alveolar edema and many damaged areas in the parenchyma. Pulmonary phlogosis was characterized by a severe and diffuse neutrophil infiltrate in peribronchial, intraluminal and interstitial areas between alveoli (S2A, S2D and S2G Fig). In contrast, in PTX3-treated infected mice, lungs conserved a physiological architecture with a moderate inflammation of the aerated parenchyma and airways and a low and scattered neutrophilic exudate (S2B, S2E and S2H Fig). Histopatological scores (S1 Table) confirmed these observations. Immunohistochemical staining of PTX3 in tissues of untreated M90T-infected mice (S3A Fig) revealed a diffuse production of PTX3, due to the involvement of bronchial mucosa cells and rare interspersed neutrophils. Strong PTX3-immunostain was observed in lung sections of PTX3-treated infected mice (S3B Fig). In uninfected PTX3-treated mice, PTX3 was barely detectable with only a few scattered PMNs physiologically associated with the bronchial mucus and the BALT (S3C Fig).
As oppose to the protective effect of recombinant PTX3 in the models of P. aeruginosa and S. fumigatus lung infection, Ptx3-deficient mice showed increased mortality and lung colonization [12, 17]. Likewise, Ptx3 -/- mice were more susceptible than the wild type to influenza virus and to UPEC infections [31, 14]. Hence, we assessed whether deficiency of PTX3 could affect the virulence of Shigella. With this aim Ptx3-/- mice were infected via intranasal route with M90T as above and the animals were monitored for a week. We found that Ptx3-/- mice challenged with M90T showed an accelerated death with respect to M90T-infected wild type mice (p < 0.046, two-tailed Mantel-Cox test) as the majority (75%: 20/30) of the animals died after 48 hours p.i. (Fig 3A) At a same time point, only 23% (7/30) wild type infected mice died. The bacterial load in lungs of Ptx3-/- of the infected animals was around ten times more (9,3 x105 vs 1,54 x105) (p <0.0001, two-tailed Mann-Whitney test) than that found in infected wild type animals (Fig 3B). Likewise, the levels of TNF-α, CXCL-1 and CXCL-2 in BALs (Fig 3C) and lung homogenates (Fig 3D) were significantly higher (for all, p <0.0001, two-tailed Mann-Whitney test) than those elicited in infected wild type mice.
Conclusively, these findings suggest that PTX3 is involved in Shigella pathogenesis. Nevertheless, only a genetic rescue with transgene expression or recombinant Ptx3 administration to Ptx3-/- mice could consolidate this result.
We then proceeded to examine whether and to what extent Shigella infection could trigger per se PTX3 release.
DCs are a major source of PTX3, released following triggers such as various inflammatory cytokines or bacterial PAMPs [10]. BMDCs from C57BL/6 mice were infected with the invasive strain M90T, or with the non-invasive isogenic strain BS176, lacking the virulence plasmid. BMDCs were infected with shigellae at the Multiplicity of Infection (MOI) of 10. PTX3 and TNF-α production was monitored in parallel at 1, 3, 6 and 18 h. post-infection (p.i)
BS176 induced a significantly higher release of PTX3 than M90T (Fig 4A), while TNF-α release was equivalent with both strains (S4A Fig). To test if invasiveness was correlated to PTX3 release, we introduced another non-invasive strain into our assay. M90T ΔipaB lacks the T3SS-secreted invasin IpaB and is thus non-invasive in epithelial cells even if it can still construct the T3SS machinery [19]. PTX3 release induced by M90T ΔipaB was similar to that of BS176 (Fig 4A).
To confirm this trend, the infection protocol of BMDCs was modified to abrogate the invasive phenotype of M90T and PTX3 release was measured as above. First, BMDCs were exposed to the invasive and non-invasive strains, previously killed with gentamicin (Fig 4B). M90T- and BS176-killed bacteria potentially could be phagocytized by DCs; however, the invasive ability of M90T was destroyed. The release of PTX3 did not change in BMDCs infected with killed BS176. In contrast, the PTX3 values induced by killed M90T were significantly higher than those recorded with live M90T. Then, we prevented cell internalization of bacteria by treating the cells with cytochalasin D, which disrupts cytoskeleton organization [36]. In contrast to the previous approach, here invasive and non-invasive bacteria could not be internalized by the cells. Under this condition the PTX3 values induced by M90T were comparable to those of BS176 (Fig 4C).
The two different experimental approaches are aimed differently at preventing the invasive phenotype of virulent shigellae. However, under both conditions the internalization of live M90T by the cells was inhibited, leading to a significant increase of PTX3 release.
Therefore, we analyzed the production of PTX3 and TNF-α in parallel in human peripheral blood MoDCs. The difference in PTX3 release observed with invasive and non-invasive Shigella strains in BMDCs was also recorded in MoDCs (Fig 4D). The trend of TNF-α release was similar to that observed with PTX3 (S4B Fig), in contrast to observations in BMDCs where the TNF-α production was equally induced by the two strains.
These results indicate that inhibition of cell invasion—due to either bacteria killing, deficiency of virulence plasmid, or host cell cytoskeleton disruption—leads to higher PTX3 release by human or murine DC.
To evaluate the signaling pathways involved in PTX3 production, we analyzed the putative roles of factors involved in downstream TLR signaling: Myd88, Trif, Cd14 and Irf3 [37]. Infection of Myd88-/- BMDCs significantly reduced the release of PTX3 with respect to wild-type BMDCs (Fig 4E). A similar trend was noted for TNF-α production (S4C Fig). The absence of Trif, Irf3 or Cd14 greatly reduced the PTX3 production as shown in Fig 4F and 4G. Likewise, TNF-α measures were drastically reduced but not totally abrogated in Trif-/-, Irf3-/- and Cd14 -/- BMDCs (S4C and S4D Fig).
As LPS is the unique PAMP engaging the Myd88 and Trif pathways, we investigated whether LPS could trigger PTX3 release, as also shown for UPEC [14].
Shigella modifies LPS composition during intracellular residence in epithelial cells [8]. Intracellular bacteria possess a hypo-acylated lipid A characterized by a blend of lipid A forms; tetra- and tri-acylated variants are more prevalent than the hexa-acylated isoform, which is the main component present in LPS of M90T grown in laboratory media. M90T ΔmsbB1 ΔmsbB2 [38,8] is a M90T mutant that carries penta-acylated (86%) and tetra-acylated lipid A forms. M90T ΔmsbB1 ΔmsbB2 is fully invasive. Purified LPS from this strain showed a low inflammatory potential, in accordance with the lipid A structure [8].
We checked whether LPS composition could impact PTX3 production in DCs treated with three Shigella LPS variants: that extracted from bacteria grown in laboratory medium [acellular (a)LPS], that purified from intracellular bacteria [intracellular (i)LPS] and LPS of M90T ΔmsbB1 ΔmsbB2. E. coli LPS was used in parallel. BMDCs stimulated with iLPS and M90T ΔmsbB1 ΔmsbB2 LPS produced a lower level of PTX3 with respect to aLPS and E. coli LPS, consistent with the degree of lipid A acylation (Fig 4H). TNF-α release reflected the differences between the LPSs (S4E Fig). Then, we examined whether the composition of LPS on live bacteria could affect PTX3 yield using M90T and M90T ΔmsbB1 ΔmsbB2 in an invasion assay as above. However, PTX3 and TNF-α release were similar for M90T and M90T ΔmsbB1 ΔmsbB2, despite different lipid A composition (Fig 4I and S4F Fig).
MoDCs were stimulated with the three Shigella LPS forms (aLPS, iLPS and M90T ΔmsbB1 ΔmsbB2 LPS) following the scheme described for BMDCs. PTX3 yield drastically decreased with iLPS and LPS of M90T ΔmsbB1 ΔmsbB2 with respect to aLPS (Fig 4J) and TNF-α release followed the same trend (S4G Fig). In contrast with results in BMDCs and in accordance with Lipid A composition, the amount of PTX3 and of TNF-α in infected MoDCs was lower with M90T ΔmsbB1 ΔmsbB2 than with M90T (Fig 4K and S4H Fig).
Thus, in addition to invasiveness, LPS composition plays a pivotal role in triggering PTX3 release in human DCs. The effect of acylation degree of LPS on live bacteria was not evident in murine DCs (Fig 4I vs Fig 4K) in line with the different TLR4 lipid A-sensitivity, as reported [39,40].
As macrophages are a further source of PTX3 during infections, we analyzed whether Shigella could promote PTX3 release in macrophages, as shown in DCs.
In mouse BMDMs infected with M90T or BS176, the difference in PTX3 yield between M90T and BS176 was like that observed in BMDCs (Fig 5A). TNF-α release did not follow this trend as the release of this cytokine was similar between the two strains (S5A Fig). However, it has been reported that Shigella-infected macrophages are poorly responsive to Shigella infection in the absence of an adequate pre-stimulation [41].
We then infected LPS-primed BMDMs as described previously [8] to evaluate the influence of LPS modification on PTX3 release. BMDMs were primed with M90T aLPS (mainly hexa-acylated), M90T iLPS (mainly tetra-acylated) and M90T ΔmsbB1 ΔmsbB2 LPS (mainly penta-acylated) and were then infected with M90T at MOI 10. Infected BMDMs primed with iLPS and M90T ΔmsbB1 ΔmsbB2 LPS (both LPSs hypo-acylated) released significantly less PTX3 than those primed with aLPS (Fig 5B), while the amount of TNF-α was similar for all the LPSs, as already shown [8] (S5B Fig). Then, we stimulated the BMDMs (without infection) with the three Shigella LPS (10 ng/mL) as above for 6 h. PTX3 and TNF-α yields were lower with iLPS and M90T ΔmsbB1 ΔmsbB2 LPS (both LPSs hypoacylated) than with aLPS (Fig 5C and S5C Fig). To evaluate the relative role of structural LPS on live bacteria and LPS priming, BMDMs were primed with different LPSs and then infected with M90T ΔmsbB1 ΔmsbB2 (carrying a penta-acylated LPS) instead of M90T (carrying a hexa-acylated LPS) as above. PTX3 release followed the trend observed with M90T, which was consistent with the acylation degree of the LPSs used for priming; however, M90T ΔmsbB1 ΔmsbB2 triggered a significantly lower release of PTX3 compared to M90T for each condition, in accordance with the hypo-acylation degree of its LPS (Fig 5D). For TNF-α, the difference between M90T and M90T ΔmsbB1 ΔmsbB2 was maintained, while no significant difference was induced by priming with the various LPSs as for PTX3(S5D Fig). We then examined whether MoMs (peripheral blood monocyte-derived macrophages) respond to Shigella infection and LPS stimulation as observed with BMDMs, BMDCs and MoDCs. With this aim, MoMs were infected with either: M90T, BS176, M90T ΔmsbB1 ΔmsbB2 or stimulated with the different LPS and both PTX3 and TNF-α production was evaluated. When infected with M90T, BS176 or M90T ΔmsbB1 ΔmsbB2, PTX3 (Fig 5E) and TNF-α yields (S5E Fig) were significantly higher with BS176 than with M90T. This result was consistent with that seen in BMDMs, BMDC and MoDCs. Likewise, M90T ΔmsbB1 ΔmsbB2 induced the lowest PTX3 and TNF-α production in accordance with its LPS structure and as also observed in BMDMs and MoDCs. In MoMs stimulated with the four forms of LPS (M90T iLPS and aLPS, M90T ΔmsbB1 ΔmsbB2 and E. coli LPS), iLPS and M90T ΔmsbB1 ΔmsbB2 LPS induced the lowest production of PTX3 (Fig 5F) and TNF-α, just as for BMDMs (S5F Fig).
Together, these experiments suggest that LPS composition is a major trigger of PTX3 production in macrophages and that PTX3 production is influenced by both structural LPS on infecting bacteria and purified LPS used for stimulation or priming.
Finally, we addressed the question about the TLR4 downstream signaling leading to PTX3 release in BMDMs as shown in BMDCs. We then used Myd88-/-, Trif-/-, Irf3-/- and Cd14-/- BMDMs (Fig 5G) as described for BMDCs. Surprisingly, PTX3 release was abrogated in all knockout BMDMs, indicating that the MyD88-dependent and MyD88-independent pathways are synergistic in the production of PTX3. In accordance with published results [8] we found that the MyD88 pathway was mainly involved in the release of TNF-α (S5G and S5H Fig), with some contribution of the Trif pathway. In Cd14-/- BMDMs, the values of TNF-α were significantly lower than in wild-type cells, while the absence of Irf3 did not impair TNF-α release (S5G Fig). These results highlight some differences in the pathways leading to TNF-α and PTX3 release.
Although S. sonnei is the main serogroup circulating in high-income countries, in recent years S. sonnei has also been observed to prevail over S. flexneri in previously low- income countries where socioeconomic conditions have improved [2]. We evaluated whether S. sonnei could, like S. flexneri M90T, induce PTX3 production. BMDCs, MoDCs and BMDMs and MoMs were infected with S. sonnei as with M90T (S6A, S6B, S6C, S6G and S6I Fig). Release of PTX3 induced by S. sonnei was similar to that induced by M90T in all cell populations. Likewise, TNF-α was similar with both strains under all the conditions (S6D, S6E, S6F, S6H and S6J Fig).
The involvement of PTX3 in shigellosis prompted us to investigate PTX3 levels in plasma of shigellosis patients. Plasma samples were collected from 31 patients in the acute stage (0–7 days after onset) of culture-proven S. sonnei shigellosis and from 19 healthy subjects and PTX3 levels were measured. Mean PTX3 levels were significantly higher in the patient group (10.4 ng/mL) compared to the control group (2.3 ng/mL) (p = 0.003). The highest levels of PTX3 in patients and controls were 50 ng/mL and 11.75 ng/mL, respectively (Fig 6). PTX3 levels were significantly higher within 2 days of disease onset than in samples collected later and were positively associated with signs or symptoms related to the severity of shigellosis such as body temperature, number of watery stools per 24 hours and bloody stools (S2 Table). Among acute cases of S. sonnei shigellosis whose maximal measured body temperature was above 39°C, the mean PTX3 level was much higher than among acute patients whose maximal measured temperature was equal or below 39°C. Similar results were found in regard to presence of blood in stool or when both signs of severity were present, albeit reaching borderline statistical significance (p = 0.05, p = 0.06 respectively).
The role of the humoral arm of the innate immunity has been poorly explored in the context of intracellular bacterial pathogens like Shigella. Here we unveil that phase fluid PRM PTX3 could play a decisive role in resolving Shigella infection as the treatment with recombinant human PTX3 rescues animals from death in the murine model of pulmonary shigellosis. As oppose to the protective effect of recombinant PTX3 Ptx3-/- infected mice showed a defective ability to clear bacteria and an accelerated kinetics of death in the same infection model. This is in line with that observed with P. aeruginosa, A. fumigatus and UPEC (12, 14, 17).
The therapeutic effect of PTX3 has already been reported for the extracellular pathogens P. aeruginosa and A. fumigatus and Streptococcus suis [13, 17, 25, 26, 27 28, 29, 42] This PRM influences pathogen phagocytosis through opsonization of microorganisms and by reinforcing complement activity [11,21].
In lungs of M90T-only infected animals, bacterial load was high as reported [6, 33, 34] while this number was significantly reduced in PTX3-treated animals. In vitro, PTX3 opsonization impairs Shigella internalization in epithelial cells, which are the replicative niche of this pathogen. Likewise, PTX3 binds human cytomegalovirus and inhibits viral-cell fusion and internalization, supposedly by crosslinking of glycoproteins on viral or cellular surface [43]. PTX3 binding to Shigella might inhibit bacteria-host cell contact by interfering sterically with the T3SS machinery thereby affecting bacterial micropinocytosis. Antibodies directed against Shigella surface structures have been shown to prevent epithelial cell internalization [23, 24, 44 and this study] and to promote the opsonophagocytic activity in macrophages [45]. In this light, PTX3 opsonization is reminiscent of some in vitro properties of protective antibodies against Shigella external structures, thus contributing to Shigella eradication before the adaptive immune response is mounted. These features reveal a role of PTX3 as an “ante-antibody” [16, 44] in Shigella infections. At a local level of infection, extracellular Shigella that are no longer protected by the host cell cytoplasm, can be targeted by the complement system and easily internalized by macrophages, thus improving the bacterial clearance by the tissues as we observed in lungs of PTX3-treated mice. Indeed, Shigella is sensitive to the serum bactericidal activity (this study), and low concentrations of PTX3 increase the killing activity of complement. It is unsurprising that PTX3 contributes to complement activity only at low concentrations given that it can play a double role, either by activating the three complement pathways or by negatively regulating them through various mechanisms, thereby limiting complement-mediated inflammation [21].
Moreover, the architecture in lungs of M90T-only infected animals was destroyed by dramatic inflammation mainly characterized by a massive neutrophil infiltration. In these mice, high levels of TNF-α and chemokines such as CXCL1 and CXCL2, which act as potent neutrophil chemoattractants [46], contribute to lung injury. In Ptx3-/- infected mice the inflammatory response seemed to be exacerbated, as the production of TNF-α, CXCL1 and CXCL2 was significantly higher than in wild type animals. In contrast, in lungs of PTX3-treated infected mice low levels of CXCL1 and CXCL2 were associated with few areas filled by a neutrophilic exudate. PTX3 can also directly contribute to local regulation of the inflammatory reaction based on neutrophil infiltrate as it binds to the adhesion molecule P-selectin and inhibits leukocyte rolling in the endothelium [47]. Although these findings sound very encouraging a definitive result could be achieved through a genetic rescue with transgene expression in Ptx3-/- infected mice or alternatively through administration of recombinant Ptx3 to these animals.
In the M90T-infected wild type mice, the serum level of PTX3 was between 10 to 20 ng/mL after 72 h of infection, which was in the range of that observed in the plasma of shigellosis patients and in patients of aspergillosis (12). Likewise, the levels of PTX3 in BAL were in the range of that recorded in BAL (2–10 ng/mL) of mice infected with A. fumigatus (25) or with influenza virus (31) as both pathogens are very sensitive to the protective effect of PTX3.
These findings arise the question of whether PTX3 could play a role during natural infection in humans. In clinical trials with a S. dysenteriae 1-attenuated but still invasive vaccine, no PTX3 increase was observed in the plasma of vaccinated individuals, [48] suggesting that only invasion of the epithelial layer by Shigella is not sufficient to rise the amount of PTX3 at the systemic level. In contrast to the attenuated strains, which are likely confined to the epithelial layer, fully virulent shigellae break the integrity of the epithelial barrier and penetrate into the mucosa where they induce severe inflammation characterized by a huge amount of cytokines and inflammatory factors, which can act as signals to promote PTX3 release. In line with this issue, in plasma of patients with shigellosis (infected with S. sonnei) the levels of PTX3 correlate positively with the severity of symptoms, particularly with high temperature and blood in the stools being a reliable parameter of severity of shigellosis, just as in conditions like sepsis [49] and critical infections [50].
As it seemed that Shigella invasion alone poorly promotes the production of PTX3, we analyzed PTX3 release in vitro upon Shigella infection. Epithelial cells, macrophages, DCs and neutrophils that contribute to Shigella-mediated inflammation have all been described as producers of PTX3.
In vitro intestinal cells such as Caco-2 cells do not release PTX3 and only increase their mRNA expression upon exposure to bacteria and bacterial moieties [51]. Likewise, Shigella-infected Caco-2 cells did not produce detectable levels of PTX3 (S7A Fig). Moreover, bronchial epithelial cells express and produce PTX3 upon TNF-α trigger, but not following bacterial PAMP stimulation (such as LPS) [52]. We could suggest that in natural and experimental shigellosis, epithelial cells cannot or barely produce PTX3 upon initial bacterial contact. PTX3 is likely to be released only when inflammatory signals like TNF-α and especially IL-1β are present. Upon inflammatory activation, neutrophils release about 25% of their pre-formed PTX3 content in lactoferrin granules [53]. In lung sections of infected animals, neutrophils were strongly immunostained for PTX3. Neutrophils recruited by inflammatory mediators could constitute a main source of PTX3 upon Shigella invasion under in vivo conditions.
In contrast to epithelial cells, Shigella-infected macrophages and DCs produce PTX3. However, virulent/invasive shigellae induce lower levels of PTX3 than do non-invasive/avirulent shigellae in these cells, suggesting that phenotypes associated with invasiveness play a major role in controlling PTX3 production.
LPS is a main Shigella trigger in macrophages and DCs, eliciting PTX3 production. In contrast to UPEC-infected cells that secrete PTX3 upon LPS-TLR4/MyD88 activation, both TLR4 mediated pathways, MyD88 and Trif, participate in PTX3 production in Shigella-infected macrophages and DCs. Involvement of the TriF pathway is evident as the absence of Trif or Irf3 abrogates and massively reduces the PTX3 release in BMDMs and BMDCs, respectively. The role of IRF3 in PTX3 production has already been observed in the context of tissue repair and remodeling [54]. In Shigella-infected BMDMs, Irf3 is not involved in TNF-α production, thereby suggesting that the PTX3 and TNF-α downstream signaling pathways diverge upon LPS-TLR4 activation.
The composition of LPS finely modulates PTX3 production, in accordance with the degree of acylation. Hypo-acylated LPS derived from intracellular shigellae stimulates a reduced release of PTX3 with respect to hexa-acylated LPS extracted from bacteria grown under conventional conditions, as also reported for cytokine production [8]. The impact of LPS composition present on live bacteria on PTX3 and TNF-α yield is particularly evident in human cells. Under some experimental conditions, LPS composition tuned PTX3 production more finely than TNF-α production. This result strongly links PTX3 production to LPS composition also in different physiological and pathological contexts.
In conclusion, at a local level of Shigella infection, PTX3 could tip the balance toward bacterial eradication playing a double role: on the bacterial side, it helps to reduce epithelial cell invasion, to implement macrophage phagocytosis and to favor complement activity; on the host tissue side, it contributes to the prevention of the development of the destructive inflammation, which is a main feature of shigellosis. This immune evasion behavior can favor bacterial survival and proliferation, thus allowing tissue colonization during the initial phases of the disease. We suggest that PTX3 could potentially contribute to the eradication of the infection by targeting the poorly explored extracellular phase of the invasion process, which can be considered the “Achille heels” of Shigella invasion process.
M90T: wild type S. flexneri strain (serotype 5a); BS176: plasmidless, non-invasive M90T derivative [55]; M90T ΔipaB: non-invasive T3SS mutant [19]; M90T ΔmsbB1 ΔmsbB2: LPS mutant which lacks both copies of the msbB genes (msbB1 and msbB2), each of which encodes the enzyme myristolyl transferase. M90T ΔmsbB1 ΔmsbB2 carries a hypo-acylated lipid A [8, 38]. M90T-GFP was created by transforming wild type strains with TGI pFpV25.1 vector [56]. The Shigella sonnei wild type strain has been described by Rossi [57]. Bacteria were routinely grown in Trypticase soy broth (TSB) (BBL, Becton Dickinson and Co., Cockeysville, MD) or agar (TSA). TSA containing 100 mg of Congo Red dye (Cr) per liter was used to select virulent clones of Shigella. Streptomycin (Sm) and ampicillin (Ap) were added to cultures at 100 μg/mL.
For the PTX3 binding assay a Pseudomonas aeruginosa wild type laboratory strain, PAO1 was used as a positive control.
The procedure was carried out as originally described with minor modifications [12]. A total of 107 CFU were incubated in 50 μl HBSS (Hank’s Balanced Salt Solution) with 0.5% BSA and biotinylated PTX3 (50 μg/mL) (1.1 μM,) [15] or biotinylated BSA (70 μg/mL). After 1 h at room temperature, samples were extensively washed with HBSS. Samples were incubated in 100 μL HBSS with 0.5% BSA with streptavidin-FITC anti-mouse Ig (1:1000) (BD, Pharmingen). Binding was evaluated by fluorescence-activated cell sorting (FACS).
Normal human serum (NHS) was produced from buffy coats obtained by the blood bank of Sapienza University and was collected from 10 healthy adult volunteers (blood donors) with no history of shigellosis following written informed consent. The blood was allowed to clot, and the serum was subsequently harvested, pooled, and stored at -70°C until used. Heat-inactivated serum (HIS) was generated by incubating NHS for 1 h at 56°C. Exponential phase bacteria were suspended in Dulbecco's Phosphate-Buffered Saline (DPBS) with Mg2+ and Ca2+ and incubated at 37°C for 30 min in the presence of NHS or HIS.
Bacterial survival was calculated as CFU in the presence of NHS/CFU in the presence of HIS x 100.
HeLa cells (ATCC Cell Biology Collection) were seeded on 6-well plates (8 x 104 cells/mL) and allowed to adhere for 12 h. An amount of 108 CFU of wild type M90T in early exponential phase (OD600 0,4–0,6) were incubated in either DMEM (Gibco, Life technologies) only or with PTX3 (50 μg/mL) or anti-IpaD Ab or BSA (50 μg/mL) or with PTX3 and BSA at the same time for 1 h at 37°C and then used to infect HeLa cells at MOI 50. After 60 min, cells were washed three times with PBS and incubated for an additional 1 or 2 h. Cells were washed three more times with PBS, detached with trypsine, counted and lysed with deoxycholic acid (0,5% in H2O). Cell lysate was plated on Congo Red (0,1%) containing TSB agar plates and CFU were counted after 18 h of incubation.
C57BL/6 female mice from Charles River, Calco, Italy, were maintained in a specific pathogen-free animal facility of the University of Camerino and euthanized by cervical dislocation.
Myd88-/-, Trif-/-, Irf3-/- or Cd14-/- were a kind gift of Maria Rescigno (Myd88-/-), IFOM-IEO Campus, Milan, Italy and Francesca Granucci (Trif-/-, Irf3-/- or Cd14-/-), Università degli Studi di Milano-Bicocca, Milan, Italy. Ptx3-/- were generated as described [12]
Wild-type and mutant BMDMs and BMDCs were derived from bone marrow cells collected from five-week-old female mice, as already reported [8] and as detailed as follows.
BMDCs (bone barrow derived dendritic cells) were differentiated for 7 days in RPMI 1640 (Lonza, Italy) containing 10% heat inactivated fetal bovine serum (FBS) (Gibco, Life technologies), 100 μM non-essential amino acids, 1000 U/mL penicillin and 1000 U/mL streptomycin (all from Lonza, Italy), supplemented with 30% R1, containing fibroblast produced GM-CSF, as described [58]. After 7 days, BMDCs were characterized by immunostaining with CD11b, CD11c, CD80, CD86, MHCII monoclonal antibodies (all from BD Pharmingen, Italy) through a flow cytometric analysis on a FACSCalibur cytometer (Becton Dickinson,San José, CA, USA). Data acquisition (104 events for each sample) was performed using CellQuest software (Becton Dickinson, San José, CA, USA). Analysis was performed with FlowJo software (TreeStar Inc., Ashland, USA).
BMDC infections with S. flexneri strains and S. sonnei were carried out at MOI 10. Infected BMDCs were incubated for 2 h before washing and adding gentamicin (60 μg/mL) (Gibco). Cells were incubated for further 1 h, 3 h, 6 h and 18 h post infection (p.i.) and supernatants were collected and analyzed for PTX3 and TNF-α release. When indicated, cytochalasin D was added 1 h before infection (0,4 μg/mL) (Invivogen). Alternatively, bacteria were killed by gentamicin (60 μg/mL) treatment for 1 h and added to BMDCs at MOI 10. The supernatants were then collected for ELISA at different time points (1 h, 3 h, 6 h and 18 h).
For LPS stimulation BMDCs, cells were seeded on 12-well plates (5 x 105 cells/well). LPS stimulation was carried out with: LPS derived from intracellular shigellae (iLPS), shigellae grown in TSB medium (aLPS) and M90T ΔmsbB1msbB2 LPS, as described, [8] and commercial E. coli LPS (LPS ultrapure–EB; InvivoGen) at the concentration of 1 and 10 ng/mL for 18 h. Supernatants were collected for ELISA analysis.
BMDMs (bone barrow derived macrophages) were differentiated for 8 days in RPMI 1640 (Lonza, Italy) containing 10% heat inactivated FBS, 100 μM non-essential amino acids, 1 mM sodium pyruvate, 1000 U/mL penicillin and 1000 U/mL streptomycin (all from Lonza, Italy), supplemented with 30% L929 fibroblast supernatant, containing M-CSF (Macrophage-Colony Stimulating Factor). F4/80 and CD11b double-positive cells were considered as differentiated BMDMs and used in the experiments.
For stimulation assay, differentiated BMDMs were seeded on 24-well plates (5 x 105 cells/well) and exposed to different concentrations (1, 10, or 100 ng/mL) of M90T iLPS, aLPS, and M90T ΔmsbB1msbB2 LPS, as described, [8] and commercial E. coli LPS (LPS ultrapure–EB; InvivoGen). Stimulation was carried out for 6 h and 18 h. Cell supernatants were recovered and stored at -20°C to be used in the ELISA assay.
BMDM infections assay with S. flexneri and S. sonnei strains were performed as reported [8] with minimal differences. Briefly, bacteria were used at MOI 10 on cells pre-treated or not for 4 h with Shigella iLPS, aLPS, ΔmsbB1 ΔmsbB2 LPS and E. coli LPS at the concentration of 10 ng/mL. Infected BMDMs were incubated at 37°C for 1 h, washed twice with PBS, and treated with gentamicin (60 μg/mL) for 3 h. At this time, supernatants were analyzed for PTX3 and TNF-α release.
PBMCs (peripheral blood mononuclear cells) were isolated from buffy coats obtained by the blood bank of Sapienza University from healthy adult volunteers (blood donors) following written informed consent.
PBMCs (peripheral blood mononuclear cells) were obtained from blood of healthy adult volunteers (blood donors), through a density gradient. CD14+ monocytes were isolated using the MACS microbead system (Miltenyi Biotec, Bergisch Gladbach, Germany). The monocytes were cultured for 6 days in RPMI 1640 (Lonza) supplemented with 10% heat-inactivated FBS (Euroclone Fetal Bovine Serum, GE Healthcare Life Sciences,U.S.), 100 μM non-essential amino acids, 1 mM sodium pyruvate, 1000 U/ml penicillin and 1000 U/mL streptomycin (all from Lonza, Italy) and 50 ng/mL GM-CSF (Granulocyte-Macrophage Colony-Stimulating Factor) (Miltenyi Biotec) to obtain human macrophages.
Stimulation assays: MoMs (peripheral blood monocyte-derived macrophages) were seeded in 24-well plates (2,5 x 105 cells/well), exposed to 1 ng/mL of LPS derived from intracellular shigellae (iLPS), shigellae grown in TSB medium (aLPS), Shigella ΔmsbB1 ΔmsbB2 LPS and commercial E. coli LPS (LPS ultrapure–EB; InvivoGen) and incubated for 12 h. Cell supernatants were recovered and processed for ELISA.
MoM infection assays with Shigella M90T strain, its derivatives BS176 and M90T ΔmsbB1 ΔmsbB2 strains, and S. sonnei strain were performed using MOI 0,1 on cells pretreated or not with Shigella iLPS, aLPS, ΔmsbB1msbB2 LPS and commercial E. coli LPS (LPS ultrapure–EB; InvivoGen) for 4 h (0,1 ng/mL). Infected macrophages were incubated at 37°C for 1 h, washed twice with PBS solution, and treated with gentamicin (60 μg/mL) for 3 h. Supernatants were recovered and evaluated by ELISA. For infection with PTX3 treated bacteria, shigellae were incubated with PTX3 or BSA (50 μg/mL) or a polyclonal rabbit anti-IpaD antibody (5 μL) (gift of Abdel Allaoui) for 1 h at 37°C. MOI 5 was used for 2 h. MoDCs were infected with bacteria at MOI of 10.
MoM death evaluation: 108 M90T were opsonized with recombinant PTX3 (50 μg/mL), BSA (50 μg/mL) or anti-IpaD Ab or nothing for 1 h at room temperature and then used to infect MoMs at MOI 5 for 2 h before evaluation of lactate dehydrogenase (LDH) release in supernatant. LDH was measured through the CytoTox 96 Non-Radioactive Cytotoxicity Assay kit (Promega, USA), according to manufacurer’s instruction. To adjust for spontaneous lysis, % release was calculated as follows: (Release in sample–release from non-infected cells)/(maximum release–release from non-infected cells) * 100.
Phagocytosis Assay: percentage of bacterial internalization by MoMs was evaluated through cytofluorimetric analysis. CD14 was used as MoM marker, and cells positive for both GFP and CD14 were considered infected cells. Data acquisition (104 events for each sample) was performed using CellQuest software (Becton Dickinson, San José, CA, USA). Analysis was performed with FlowJo software (TreeStar Inc., Ashland, USA).
Dendritic cells. MoDC (peripheral blood monocyte-derived dendritic cell) culture, stimulation and infection.
PBMCs were obtained from blood of healthy adult volunteers (blood donors), through a density gradient as above. CD14+ monocytes were isolated using the MACS microbead system (Miltenyi Biotec, Bergisch Gladbach, Germany). The monocytes were cultured for 5 days in RPMI 1640 (Lonza) supplemented with 10% heat-inactivated FBS (HyClone Fetal Bovine Serum, GE Healthcare Life Sciences,U.S.), 100 μM non-essential amino acids, 1000 U/mL penicillin and 1000 U/mL streptomycin (all from Lonza, Italy), 20 ng/mL IL-4 and 50 ng/mL GM-CSF (both Miltenyi Biotec) to obtain immature human dendritic cells. MoDCs were characterized by immuno-staining with CD11c, CD14 and CD80 (all from BD Pharmingen, Italy) through a flow cytometric analysis.
For MoDC stimulation, MoDCs were collected, seeded on 12-well plates (5 x 105 cells/well) and exposed to 10 ng/mL of iLPS, aLPS, M90T ΔmsbB1 ΔmsbB2 LPS and E. coli LPS for 12 h.
For MoDC infections, cells were seeded at 5 x 105 cells/well on the morning of infection. Exponential phase bacteria were added at MOI 10 directly to the wells containing MoDCs. Plates were centrifuged for 5 min at 300 x g and incubated at 37° C for 2 h. Fresh medium containing gentamycin (60 μg/mL) was added, and cells were incubated for further 1 h, 3 h, 6 h and 18 h.
Five-week-old (18–20 gr) C57BL/6 female wild type (Charles River, Calco, Italy) or Ptx3-/- mice were maintained in a specific pathogen-free animal facility of the University of Camerino. For all the infections, mice were anesthetized intramuscularly with 50 μL of a solution containing Zoletil (1 mg) (Virbac, Carros, France) and Xilor (2%) (BIO 985, San Lazzaro, Italy) and inoculated intranasally with 20 μL of 0.9% NaCl suspensions containing 3 x 108 CFU of S. flexneri M90T strain [47,6].
When required, infected mice (wild type) were treated once per day for 3 days with recombinant PTX3 (10 μg/mouse intraperitoneally, 0,5 mg/Kg, corresponding to 11 μM) or with sterile saline and examined daily to evaluate the survival during 8 days (n = 22 for M90T-infected-mice, n = 19 for M90T-infected and treated-mice, n = 10 for non-infected PTX3-treated mice, in three separate experiments), or euthanized at 3 days post-infection (n = 10 for M90T-infected-mice and n = 11 for M90T-infected and treated-mice, n = 10 for non-infected PTX3-treated mice, in three separate experiments). At this time, bronchoalveolar lavages (BAL) were performed and BALs were used for cytokine analysis. Lungs were removed, homogenized and plated on TSA plates (1 lung per mice) or analyzed for relevant cytokines or alternatively fixed in formalin. Consecutive sections from the middle of the five lung lobes were used for histological and immunohistochemical examination. In the experiments using Ptx3-/- mice survival was analyzed up to 8 days p.i. (n = 30, in three separate experiments) and their lungs were processed to assess the bacterial load and cytokine production at 48 h p.i. (n = 10 Ptx3-/-; mice n = 13 wild type), in three separate experiments).
Lungs samples were fixed in 4% formaldehyde for 18 h at room temperature and treated for histopathological and immunochemistry studies as described [6]. The samples were gradually dehydrated and then embedded in paraffin. The specimens were cut in 3-μm-thick slices and stained with hematoxylin-eosin (Carlo Erba) or immunostained. Immunohistochemistry was performed by using the following antibodies: rabbit polyclonal anti-human PTX3 [14]. and mouse monoclonal anti-PMNs (MA5-12607, clone BM-2, ThermoFisher Scientific, USA). The sections were incubated with the secondary antibodies (1:200) for 45 minutes and then examined blindly and scored by a pathologist.
Cytokine and chemokine concentrations were determined by commercially available ELISA kits (Duo Set R&D systems). The absorbance was measured on a LT-4000 Microplate reader (Labtech) (Hercules, CA, USA).
The LPSs used in this study are the same as those used by Paciello [8]. Refer to this article for relevant information about the experimental procedures for LPS extraction and purification.
31 patients in the acute stage (0–7 days after onset) of culture-proven S. sonnei shigellosis were recruited for the study. They included 31 children aged 0.2–10 years and one adult. The control group included 19 healthy subjects, 11 adults and 8 children aged 0.5–14 years. Signed informed consent was obtained from the parents of all participating children and from all participating adults. Participants or their parents completed a questionnaire with personal data and details regarding symptoms and onset of disease. Blood samples were collected using EDTA tubes (Geiner Bio-One). Plasma was separated and stored at -80 oC until assayed.
Data was presented as mean ± S.D., and the number of independent experiments is indicated in each legend of the figures. Statistics were performed with GraphPad Prism and data analysis was carried out as follow: Mantel-Cox test was used to compare survival curves; non-parametric Mann-Whitney U test for CFU counts and cytokine/chemokines quantification in mice; Student's t-test for PTX3 and TNF-α release in cell cultures and paired t-test for PTX3 quantification in plasma of patients. P < 0.05 was considered significant.
Mice experiments were conducted according to the ethical requirements of the Animal Care Committee of the University of Camerino (study protocol No 17/2012) upon approval of the Ministero della Salute, Direzione Generale della Sanità Animale e dei Farmaci Veterinari, Ufficio VI (Benessere animale), in line with the Guidelines laid down by the European Communities Council (86/609/ECC) for the care and use of laboratory animals. The study involving shigellosis patients and controls has been has been approved by the IRBs of Hillel Yaffe Medical Center and the Israel Ministry of Health (Study protocol No. AH-382-11). Written signed informed consent was obtained from the parents of all participating children and from all participating adults. PBMCs (peripheral blood mononuclear cells) were isolated from buffy coats obtained by the blood bank of Sapienza University from healthy adult volunteers (blood donors) following written informed consent.
|
10.1371/journal.ppat.1007093 | Prion replication environment defines the fate of prion strain adaptation | The main risk of emergence of prion diseases in humans is associated with a cross-species transmission of prions of zoonotic origin. Prion transmission between species is regulated by a species barrier. Successful cross-species transmission is often accompanied by strain adaptation and result in stable changes of strain-specific disease phenotype. Amino acid sequences of host PrPC and donor PrPSc as well as strain-specific structure of PrPSc are believed to be the main factors that control species barrier and strain adaptation. Yet, despite our knowledge of the primary structures of mammalian prions, predicting the fate of prion strain adaptation is very difficult if possible at all. The current study asked the question whether changes in cofactor environment affect the fate of prions adaptation. To address this question, hamster strain 263K was propagated under normal or RNA-depleted conditions using serial Protein Misfolding Cyclic Amplification (PMCA) conducted first in mouse and then hamster substrates. We found that 263K propagated under normal conditions in mouse and then hamster substrates induced the disease phenotype similar to the original 263K. Surprisingly, 263K that propagated first in RNA-depleted mouse substrate and then normal hamster substrate produced a new disease phenotype upon serial transmission. Moreover, 263K that propagated in RNA-depleted mouse and then RNA-depleted hamster substrates failed to induce clinical diseases for three serial passages despite a gradual increase of PrPSc in animals. To summarize, depletion of RNA in prion replication reactions changed the rate of strain adaptation and the disease phenotype upon subsequent serial passaging of PMCA-derived materials in animals. The current studies suggest that replication environment plays an important role in determining the fate of prion strain adaptation.
| The main risk of emergence of prion diseases in humans is associated with a cross-species transmission of prions of zoonotic origin. Prion transmission between species is regulated by a species barrier. Amino acid sequences of host prion protein and donor prions are believed to be the main factors that control species barrier and strain adaptation. Yet, despite our knowledge of the primary structures of mammalian prions, predicting the fate of prion strain adaptation is very difficult. The current study asked the question whether changes in cofactor environment affect the fate of prions adaptation. To address this question, hamster prion strain was propagated under normal or RNA-depleted conditions in vitro first using mouse and then hamster substrates. This work demonstrated that depletion of RNA in prion replication reactions changed the rate of strain adaptation and the disease phenotype upon subsequent serial passaging in animals. The current studies suggest that replication environment plays an important role in determining the fate of prion strain adaptation.
| Prion diseases are a group of fatal neurodegenerative diseases of humans and other mammals that can arise spontaneously or via transmission [1]. The transmissible agent of prion disease consists of a prion protein in β-sheet rich self-propagating states referred to as PrPSc that template conversion of the same protein in its normal, cellular state (PrPC) into disease-related pathogenic state [2–6].
The main risk of emergence of prion diseases in humans is associated with a cross-species transmission of prions of zoonotic origin. Transmission of prions or PrPSc between species is less efficient in comparison to transmissions within the same species due to the species barrier (reviewed in [7]). According to a traditional view, the species barrier arises because of differences between amino acid sequences of donor PrPSc and host PrPC, whereas the magnitude of a barrier is believed to be determined by the extent to which the strain-specific conformation of donor PrPSc can be accommodated within the primary structure of the new host PrP [8–10]. While the role of amino acid sequences in regulating the species barrier has been supported by numerous studies [7,8,11–13], multiple exceptions have been described over the years, arguing that other yet unknown factors contribute to the barrier. For instance, in certain lines of transgenic mice expressing human PrPC the transmission of a new variant Creutzfeldt-Jakob Disease occurred at full attack rate. Yet in other lines of humanized mice the transmission showed significant barrier as judged from long incubation times, incomplete attack rates or lack of clinical diseases despite identity in amino acid sequences of host PrPC and donor PrPSc [14,15]. Moreover, several studies illustrated that prions from a variety of species could be transmitted very effectively to the bank vole despite differences in amino acid sequences showing very little if any species barrier [16–18]. These studies suggest that the bank vole is a universal host.
Successful cross-species prion transmission is typically accompanied by strain adaptation that might result in stable changes of strain-specific disease phenotype, a phenomenon known as prion strain mutation. Prion strain mutation has been attributed to changes in PrPSc conformation that occur as a result of adaptation to a new host [19,20]. Despite the fact that amino acid sequences of the prion protein are known for many mammalian species, predicting the fate of prion strain adaptation upon cross-species transmission is very difficult if possible at all. A comprehensive mechanism explaining species barrier has yet to be developed.
A number of studies in the last decade provided convincing evidence that cellular molecules of non-protein nature including RNAs and lipids assist prion replication [4,5,21–25]. In vitro studies of prion replication using Protein Misfolding Cyclic Amplification (PMCA) suggested that RNAs and polyanions form favorable biochemical environment that assists replication [4,21]. However, the mechanism by which non-protein co-factors assist prion conversion is not well defined [26–29]. Moreover, the extent to which non-protein co-factors specify strain-specific properties is not clear [23]. Interestingly, the effects of RNAs in facilitating replication of prion were found to be species- and strain-dependent [25,30,31]. These studies suggested that optimal prion replication might require species- and strain-specific cofactors. Bearing this in mind, we hypothesized that cofactor environment might play an important role in defining the fate of prion strain adaptation.
Previously, we showed that a change in co-factor environment alone, and specifically reversible depletion of total cellular RNA, without changes in PrP primary sequence led to stable changes in strain-specific physical features of PrPSc [32]. In the current study, we asked the questions whether the changes in prion replication environment affect the fate of prions adaptation. To address this question, hamster strain 263K was propagated in normal or RNA-depleted conditions in serial PMCA with beads (PMCAb) in mouse brain homogenate, and then re-adapted back to hamster brain homogenate under normal or RNA-depleted conditions. Subsequent serial transmission revealed that PMCAb-derived products formed under RNA-depleted and normal conditions induced different disease phenotypes in animals.
For testing whether the fate of prion adaptation depends on cofactor environment, the following experiments were conducted (Fig 1A). First, hamster strain 263K was propagated in serial PMCAb for thirteen rounds under normal or RNA-depleted conditions using mouse brain homogenate as a substrate (Fig 1A). RNA depletion in brain homogenate was confirmed by an agarose gel (S1 Fig). The products of PMCAb reactions in normal and RNA-depleted mouse brain homogenates will be referred to as 263KM and 263K(M), respectively. Second, 263K(M) were readapted to hamster substrate by propagating in serial PMCAb reactions under normal or RNA-depleted conditions using hamster brain homogenates. The PMCAb products generated in normal and RNA-depleted hamster brain homogenates will be referred to as 263K(M)H and 263K(MH), respectively (Fig 1A). In parallel, 263KM was readapted to hamster substrate in serial PMCAb reactions under normal conditions. The products of this reaction will be referred to as 263KMH (Fig 1A).
Analysis of PK-resistant PMCAb products by Western blotting revealed that propagation of 263K in mouse substrate displayed a phenomenon similar to the transmission barrier (Fig 1B). This result was consistent with the previous studies on adaptation of hamster strains to mouse substrate in PMCA [33,34]. Amplification of 263K in a mouse substrate was observed regardless of presence of RNA in PMCAb (Fig 1B). Upon re-adaptation to hamster substrate, both 263KM and 263K(M) displayed relatively stable replication in hamster brain homogenate without significant barrier suggesting that 263KM and 263K(M) contain conformations compatible with hamster substrate. (Fig 1C). In addition to PK-resistant products of standard size, shorter PK-resistant band of approximately 23 kDa was visible in all three reactions.
To test whether RNA depletion changed the fate of prion adaptation, Syrian hamsters were inoculated with PMCAb-derived 263KMH, 263K(M)H or 263K(MH). By the end of PMCAb experiments, the original 263K brain material was diluted 1027-fold in PMCAb-derived 263KMH, 263K(M)H and 263K(MH); this dilution is approximately 1017 fold higher than the limiting dilution of 263K [35]. None of the animals from three groups developed any notable signs of prion disease for up to 518 days post inoculation (Table 1). Analysis of their brains by Western blotting revealed relatively minor yet variable amounts of PK-resistant material in all three groups (Fig 2A).
Animals with the highest amounts of PK-resistant material from each group were selected for the 2nd passage. Out of the eight animals from the 2nd passage of 263KMH, three developed clinical signs of prion disease at 246 days postinoculation and where euthanized at the terminal stage a few days later, 257 or 272 days postinoculation (Table 1). Animals that developed the disease showed considerably more PrPSc relative to the animals from the same group that were euthanized in the absence of the disease at 503 days postinoculation (Fig 2B). To probe the pace of adaptation in other two groups, two animals from each the 263K(M)H and 263K(MH) groups were euthanized at 347 days postinoculation in the absence of clinical signs (Fig 2B). Notably, by 347 days, the signal intensity in 263K(M)H groups was the same as in the corresponding inoculum, whereas in the 263K(MH) group the signal dropped considerably lower relative to the signal in corresponding inoculum (Fig 2B). This result suggests a low replication and/or a high clearance rate in the 263K(MH) group relative to the other two groups. Nevertheless, all remaining six animals from the 2nd passage of 263K(M)H developed clinical signs after incubation time ranging from 392 to 575 days postinoculation (Table 1). In this group, the clinical signs progressed much more slowly in comparison to the animals from the 2nd passage of 263KMH (Table 1). Large amounts of PrPSc were found in all clinical animals from 263K(M)H group (Fig 2B). None of the animals from the 2nd passage of 263K(MH) developed clinical diseases for up to 503 days postinoculation (Table 1). Lack of disease in this group correlated well with very minor amounts of PK-resistant material observed in their brain (Fig 2B).
Three animals from each group of the 2nd passage were analyzed using histopathology for deposition of PrPSc, reactive astrogliosis and vacuolation. From the 263KMH and 263K(M)H groups, only animals that developed clinical disease were selected for histopathological analysis. Consistent with lack of clinical signs and low amounts of PrPSc observed by Western blots, 263K(MH) animals exhibited minor vacuolation and astrogliosis, and only focal deposits of PrP mostly in the thalamus and hippocampus (Fig 3, S2 and S3 Figs). In contrast to the 263K(MH) group, animals from 263KMH and 263K(M)H groups displayed more substantial deposition of disease-associated PrP and more pronounced astrogliosis with respect to anatomical distribution and degree of changes (Fig 3 and S3 Fig). Disease-associated PrP was found in form of diffuse/synaptic fine deposits, occasional mini-plaques, and less frequent pericellular PrP deposits. In periventricular areas amorphous deposits were seen. Furthermore, subependymal amorphous plaque-like deposits were noted in 263KMH and 263K(M)H groups, but were very mild in 263K(MH) group. Nevertheless, as judged from the lesions and PrP deposition scored across brain regions, 263K(M)H and 263K(MH) groups displayed similar profile shapes for both the lesion and PrP immunoreactivity, albeit the 263K(MH) group was less affected than 263K(M)H group (S3 Fig). The shape of PrP immunoreactivity profile of 263KMH group was notably different when compared to the profiles of 263K(M)H and 263K(MH) suggesting that different strains might be emerging in 263KMH and 263K(M)H / 263K(MH) groups (S3 Fig). Animals from the age-matched control group lacked any deposition of PrPSc or spongiform degeneration (S4 Fig).
For the 3rd passage, one brain with the strongest signal on Western blot was selected from each group. All animals from the 263KMH passage showed clinical signs similar to those of 263K between 44 and 66 days postinoculation (Table 1). In this group, the disease progressed quickly and all animals were euthanized at 71 or 80 days postinoculation. In comparison to the 263KMH group, the animals from the 3rd passage of 263K(M)H showed the first clinical signs after significantly longer incubation time (Table 1). Animals of this group developed an agitated, fidgeting behavior, dry skin, rough and patchy coat. The disease progressed slower in comparison to the 263KMH group and animals were euthanized at the terminal stages between 428 and 477 days postinoculation. Surprisingly, none of the animals from the 3rd passage of 263K(MH) developed clinical signs for up to 614 days postinoculation (Table 1). Interestingly, in both 263KMH and 263K(M)H groups, the intensities of PrPSc signal on Western blot were very similar to the intensities of corresponding brain materials from the 2nd passage used for inoculations (Fig 2C). The 263K(MH) group showed lower amounts of PrPSc in comparison to the 263KMH and 263K(M)H groups, yet substantial increase was observed in the 3rd passage of 263K(MH) relative to the animals from the 2nd passage of 263K(MH) (Fig 2C).
A detailed comparison of the PK-resistance profiles by Western blot revealed that the relative proportions of the mono- and unglycosylated isoforms are higher in 263K(M)H and 263K(MH) groups relative to the 263KMH or 263K groups (Fig 2D). In addition, 263K(M)H and 263K(MH) groups showed slight shifts in the mobility of the PK-resistant products toward the lower molecular weight (Fig 2D). Moreover, both 263K(M)H and 263K(MH) groups showed an additional lower molecular weight PK-resistant product at ~12 kDa (Fig 2C and 2D). Appearance of the lower molecular weight bands under non-denaturing conditions suggests that PrPSc might expose internal PK-cleavage sites in the central region of PrP as reported previously [36], or that alternative PK-resistant states might exist in 263K(M)H and 263K(MH) groups.
To probe further the differences in physical properties of PrPSc, brain materials from four groups were analyzed using two assays: (i) treated with increasing concentrations of PK and (ii) subjected to increasing concentrations of GdnHCl following by PK treatment. The PK-resistance profiles were found to be similar for 263KMH, 263K(M)H and 263K groups (Fig 4A and 4B). In 263K(MH) group, PrPSc was considerably more sensitive to digestion at high concentrations of PK (above 50 μg/ml) relative to PrPSc from other three groups (Fig 4A and 4B). In the experiments on denaturation, PrPSc of 263KMH and 263K groups exhibited very similar GdnHCl-induced denaturation profiles (S5 Fig). A shift of the PK-resistant products toward lower molecular weight bands was observed in both groups at 3 M GdnHCl, an indication that internal PK-cleavage sites have been exposed in PrPSc upon denaturation (S5 Fig). The GdnHCl-induced denaturation profiles of 263K(M)H and 263K(MH) groups were notably different from those of the 263KMH and 263K groups as well as from each other (S5 Fig). In the 263K(M)H group, the lower molecular weight bands were barely visible at the low concentrations of GdnHCl, but increased gradually after 2M of GdnHCl (S5 Fig). In the 263K(MH) groups, the low molecular weight bands were visible well at the low concentrations of GdnHCl; their intensity decreased in parallel with the drop of intensity of standard PK resistant bands at high concentrations of GdnHCl. In summary, 263KMH, 263K(M)H and 263K(MH) groups displayed three different patterns of GdnHCl-induced denaturation, with the pattern of 263KMH being very similar to that of 263K. Comparative analysis of animals from the 2nd and 3rd passages of 263KMH and the original 263K revealed similar histopathological features with respect to spongiform degeneration, reactive astrogliosis or PrP immunoreactivity in all three groups (Fig 5, S6 Fig). In all three groups, immunostaining for PrP displayed the following types of PrP deposits: diffuse/synaptic fine deposits, intra- and perineuronal deposits, plaques and mini-plaques in the subependymal area, small granular accumulations on the ependyma, and amorphous plaque-like deposits in the subpial and subependymal regions (Fig 5). These PrP deposits can be seen under large magnification in S7 Fig. With respect to anatomical distribution, the pathology in the 2nd and 3rd passages of 263KMH was reminiscent yet not identical to that of scrapie 263K (Fig 5, S8 Fig). While the intensity of the PrP deposition in the 3rd passage increased in comparison to the 2nd passage, it did not reach the intensity observed for 263K group in a few brain regions such as the cerebellum and caudate-putamen (Fig 5, S8 Fig). The intensity of reactive astrogliosis correlated well with the intensity of PrP deposition and its anatomical distribution in all three groups (Fig 5). Again, the reactive astrogliosis was more prominent in the animals of the 3rd passage relative to the 2nd passage of 263KMH. Overall, the histopathological analysis confirmed that the disease phenotype of 263KMH was reminiscent yet not completely identical to that of 263K.
Histopathological analysis of the animals from the 3rd passage of 263K(M)H and 263K(MH) revealed that both groups display characteristic features of prion disease including spongiform degeneration, reactive astrogliosis and PrP deposition (Fig 6, S8 Fig). Consistent with the clinical status and analysis of PrPSc by Western blot, 263K(M)H group showed much more pronounced histopathological changes relative to the 263K(MH) group. In 263K(M)H group, significant spongiform changes, reactive astrogliosis and PrP deposition were observed throughout the brain (Fig 6). In 263K(MH) group, the reactive astrogliosis, spongiform degeneration and PrP deposition were very mild across all brain regions (Fig 6, S9 Fig). With respect to the pattern of PrP deposition, unique features were found in both the 263K(M)H and 263K(MH) groups that were not seen in 263KMH or 263K groups. In addition to dot-like and diffuse/synaptic immunoreactivity, 263K(M)H group was characterized by numerous large pial deposits, deposits associated with leptomeningeal blood vessels, and intense perivascular deposits (Fig 7A–7D). The animals of 263K(MH) showed predominantly dot-like deposits and some stellate deposits in hippocampus as well as perivascular deposits and smaller plaques (Fig 7E and 7F). Diffuse/synaptic immunoreactivity in 263K(MH) group was much lower than in 263K(M)H group. In both 263K(M)H and 263K(MH) groups, the cerebellum was less affected in comparison to 263K group. In summary, neuropathological analysis illustrated considerable changes in strain-specific characteristics in both 263K(M)H and 263K(MH) groups in comparison to 263KMH or 263K groups.
The current study revealed that prion replication environment and specifically cellular RNAs play an important role in determining the fate of prion strain adaptation. We found that depletion of RNA in replication reactions changed the rate of strain adaptation and the disease phenotype upon serial passaging of PMCAb-derived material in animals. Serial passaging of 263K propagated under normal conditions in mouse and then hamster substrates (designated as 263KMH) resulted in a disease phenotype similar but not entirely identical to the original 263K. We do not know whether authentic 263K will emerge upon further serial transmission of 263KMH. Surprisingly, 263K propagated first in RNA-depleted mouse substrate and then normal hamster substrate (designated as 263K(M)H) resulted in a new disease phenotype. This disease phenotype was characterized by a longer incubation time and clinical duration of disease relative to the original 263K or 263KMH group and altered neuropathological features. We do not know whether the incubation time of 263K(M)H will shorten upon further serial transmission. 263K propagated first in RNA-depleted mouse and then RNA-depleted hamster substrates (designated as 263K(MH)) failed to produce clinical diseases for three serial passages despite persistent replication of PrPSc during serial transmission. Analysis of the PK-digestion patterns, glycoform ratios and GdnHCl-induced denaturation profiles revealed structural differences in PrPSc from 263KMH, 263K(M)H and 263K(MH) groups. In a manner similar to 263K, the internal sites of PK cleavage were exposed in 263KMH PrPSc only after exposure to 3M GdnHCl. In contrast to 263KMH, the internal PK-cleavage sites were well-accessible in 263K(MH) and mildly accessible in 263K(M)H even under nondenaturing conditions (S5 Fig). In summary, three different outcomes of prion transmission were observed in animals depending on the presence or absence of RNA in PMCAb reactions.
Animals of all three groups 263KMH, 263K(M)H and 263K(MH) were asymptomatic in the first passage. Prior to inoculation into hamsters, 263KMH, 263K(M)H and 263K(MH) were amplified in serial PMCAb using hamster substrates. Therefore, the lack of clinical disease in the first passage cannot be attributed to the differences in the amino acid sequences of the PMCAb-derived inocula and PrPC of the host. Instead, the lack of the disease could be in part due to a decline of prion-specific infectivity during serial PMCAb. Indeed, previous studies reported that the specific prion infectivity of 263K declines gradually during serial PMCA [37]. Nevertheless, it is difficult to attribute three asymptomatic serial passages of 263K(MH) solely to a low specific infectivity of the PMCAb-derived products. In contrast to 263MH and 263K(M)H groups, in 263K(MH) group the amounts of PrPSc increased very slowly during serial transmission. In fact, in the 2nd passage most of the animals of 263MH and 263K(M)H groups showed considerable increase in the intensity of PrPSc signal relative to the signal intensity of the corresponding inocula, whereas the signal intensity for most of the animals of 263K(MH) group was the same as in corresponding inoculum (Fig 2B). This could be due to intrinsically slow replication rate of 263K(MH), its fast clearance, or both. Consistent with this conclusion, brain-derived 263K(MH) was found to be considerably less resistant to proteolytic digestion at high concentrations of PK in comparison to the brain-derived 263MH or 263K(M)H.
The hamster-adapted prion strain 263K used in the current study originated from the natural pool of scrapie that was isolated from the Cheviot breed in 1950 [38]. Since then, the 263K’s ancestor was transmitted through mice and rats before it was finally stabilized to hamsters after multiple serial passages [39]. Considering this history of interspecies passages, it is not surprising that 263K possesses certain level of plasticity and is able of overcoming a hamster-to-mouse species barrier. While 263K was initially regarded as nonpathogenic for mice as it failed to produce clinical disease in the first passage [40], subsequent studies documented slow or silent replication of 263K in mice that could lead to a clinical disease upon serial transmission [41–43]. Moreover, transgenic mice that overexpress mouse PrPC (Tg20) developed clinical diseases in the first passage upon transmission of Sc237, the hamster-adapted strain of the same origin as 263K [34]. The ability of 263K PrPSc to recruit mouse PrPC and convert it into PrPSc pathogenic to mice was illustrated further using in vitro experiments that employed PMCA [33]. Notably, the prion diseases observed in mice upon inoculation of PMCA-derived PrPSc seeded with 263K and amplified in mouse substrate showed unique disease phenotype [33]. Remarkably, other studies demonstrated that mice infected directly with hamster Sc237 PrPSc or PrPSc obtained upon replication of Sc237 in PMCA with mouse substrate produced different disease phenotypes upon serial transmission [34]. Those studies suggested that prion replication environment, whether it is in vitro environment of PMCA or environment of cellular sites of prion replication in a brain, is an important factor that determines disease phenotype upon cross-species transmission.
What is the role of RNA in prion replication? In previous studies, cellular and synthetic RNAs were shown to stimulate replication of prions in vitro [4,21]. The degree of the stimulating effect was found to be species- and strain-dependent [25,30,31]. While RNAs strongly facilitated replication of all hamster strains examined, the effect on replication of mouse strains was considerably less pronounced and strain-dependent [25,30,31]. Considering that the hamster strains are predominantly di-glycosylated, whereas the glycosylation statuses of mouse strains are variable, one can speculate that the species- and strain-dependency of the RNA effect could be due to differences in pattern or density of carbohydrate epitopes on PrPSc surface [44]. It is not clear whether RNAs assist prion replication in vivo. On one hand, convincing evidence have been presented that prion-specific polynucleotides are lacking in PrPSc particles isolated from Sc237-infected animals, the hamster-adapted strain of the same origin as 263K [45]. On the other hand, RNA molecules were found to co-localize with large extracellular PrPSc aggregates in hamsters infected with Sc237 [29]. Moreover, synthetic homopolymeric polynucleotides of sizes above 200 bases were found to stimulate conversion in vitro and form nuclease-resistant complexes with PrP molecules during PMCA reactions [29,46]. The detailed molecular mechanism behind the effects of RNA on prion replication is not known. Yet, it is reasonable to conclude that at least in vitro RNA provides favorable biochemical environment for prion replication. As suggested by previous studies, it is highly unlikely that the stimulating effects could be attributed to specific RNA sequences [4,46]. Instead, it appears that the polyanionic nature of RNA and, perhaps, its unique conformational features are important.
Application of PMCAb for examining the contribution of biochemical environment on fate of prion stain adaptation could be regarded as a main limitation of the current work. At the same time, the experimental design that involves PMCAb provides an opportunity for probing the hypothesis, which otherwise would be difficult, if not impossible, to test. Based on the results presented in the current work, one can speculate that accessibility of the cellular sites of prion replication to RNAs could be one of the factors that contribute to determining the fate of prion strain adaptation upon cross-species transmission. Among other factors that control the outcomes of the cross-species transmission are the transmission route and involvement of secondary lymphoid tissues in prion replication [47]. While it is difficult to fully disentangle strain adaptation in vivo from the adaptation that occurs during PMCAb reactions, the results of our previous studies helped to assess the contribution of the PMCAb technique itself to the apparent strain adaptation [48]. In those experiments, 103-fold diluted 263K brain material was subjected to 24 serial PMCAb rounds with 1:10 dilution between rounds using only hamster NBH as a substrate. The resulting PMCAb-derived 263K was produced after an equal number of PMCAb rounds and dilution fold as in the current study. PMCAb-derived 263K was found to induce prion diseases in hamsters with 263K-specific disease phenotype an incubation time to the terminal stage 106±12 days postinoculation, which was longer than 82±2 days observed in animals inoculated with the equivalent amounts of brain-derived 263K. This experiment is consistent with the previous data that amplification of 263K in PMCAb in hamster substrate reduces specific prion infectivity, yet does not alter the disease phenotype [37]. Nevertheless, the fact that three serial passages were required for 263KMH to exhibit the disease phenotype similar to 263K suggests that additional strain adaptation was taking place in hamsters during serial transmission.
The results of the current study can be discussed within two broad hypotheses—the cloud and deformed templating hypotheses, which are not mutually exclusive. According to the cloud hypothesis, the populations of PrPSc particles are intrinsically heterogeneous within individual strains or isolates due to spontaneous conformational mutations. PrPSc populations might consist of a major and a number of minor structural variants [49,50]. Upon a cross-species transmission, a minor PrPSc variant might become predominant in a new host due to changes in selection criteria and give rise to a new disease phenotype. If the cloud hypothesis accounts for the changes in fate of prion adaptation observed in the current studies, the current results suggest that altering a biochemical environment of prions replication in vitro gives a selective advantage to one of preexisting minor PrPSc variants resulting in an altered disease phenotype. The deformed templating model postulates that a change in replication environment plays an active role in generating new PrPSc variants, in addition to its role in imposing a new selective pressure [51,52]. PrPSc templates that do not fit well to a new environment still can seed altered PrPSc structural variants via deformed templating. While the majority of the newly generated variants might not replicate efficiently in altered environment, a variant that fits well to the new environment will eventually emerge through multiple trial-and-error seeding events. If one assumes that the deformed templating is behind observed effects, then 263K(M)H and 263K(MH) should be considered as two new variants that emerged de novo in RNA-depleted environment of PMCAb. It would be challenging to document what mechanism takes place, as it requires experimental testing of whether a minor PrPSc variant associated with a new disease phenotype pre-existed in the original prion isolate or strain. Nevertheless, our previous work demonstrated that PMCAb replication of 263K in RNA-depleted hamster substrate gave rise to new PrPSc variants that were absent in the original brain-derived 263K seeds [32]. Moreover, deformed templating mechanism describes well the evolution of prion strains of synthetic origin that were induced by recombinant PrP fibrils in animals despite fundamental structural differences between recombinant PrP fibrils and authentic PrPSc [53–56].
Regardless of which hypothesis is correct, the current work highlights a new important role of cofactor environment in prion cross-species transmission and adaptation. How changes in replication environment, and specifically RNA-depletion, can affect the fate of a prion strain? Previously, we suggested that adequate replication environment is necessary for insuring high fidelity of prion strain replication [32]. If this is true, RNA depletion during propagation of RNA-dependent strains could be compensated in part by other cellular polyanions creating heterogeneous replication environments and boosting diversity of the PrPSc variants. The current results support this mechanism, which is speculative at present time. Notably, in previous studies three mouse strains maintained their highly infectious and pathogenic state upon replication in PMCA in the presence of a lipid (phosphatidylethanolamine) as a sole cofactor, yet lost their original strain-specific features and converged into a single new strain [23]. Other studies reported transient changes in strain-specific clinical signs of the disease upon serial passaging of RML PrPSc that was subjected to replication in PMCA under RNA-depleted conditions [31]. In a second passage, the clinical signs of the disease reversed to the original RML-specific signs suggesting that the original RML variant overcompeted the new variant that emerged under the RNA-depleted conditions [31]. In summary, the current and previous studies suggest that maintaining adequate replication environment might be essential for maintaining prion strainness.
Recent study generated several novel prion strains by propagating chronic wasting disease prion isolates in PMCA that utilized recombinant bank vole PrP as a substrate and PrP knock-out mouse brain homogenate as a source of cofactors, and then replacing mouse brain homogenate with different cofactors of polyanionic nature [57]. Remarkably, the same set of PrPSc structures were generated in non-seeded or spontaneous PMCA reactions conducted only in the presence of recombinant bank vole PrP substrate and polyanionic cofactors [57]. These data suggest that cofactors might confine spontaneous PrP misfolding pathways in vitro while guiding it toward limited set of PrPSc structures that give rise to new prion strains upon transmission in animals. The current study tested the effect of RNA during cross-species adaptation and suggests that RNAs might be important for ensuring a high fidelity of prion replication.
In addition to cellular cofactors, what other parameters might be involved in determining the fate of prion cross-species transmission and strain adaptation? N-linked glycans represent the source of enormous diversity with respect to their composition and structure, yet their role in prion pathogenesis remains largely unknown. Our recent work revealed that PrPC sialoglycoforms are recruited into PrPSc selectively in a strain-specific manner [58]. Based on 2D analysis of glycosylation of individual strains, we proposed that individual strain-specific structures of PrPSc govern selection of PrPC sialoglycoforms that can be accommodated within individual structures producing a strain-specific pattern of carbohydrate epitopes on PrPSc surface [44,59]. In addition to a strain-specific structure, the pattern of carbohydrate epitopes is likely to be shaped by a host due to species-specific differences in the spectrum of N-linked glycans synthetized by different hosts. On one hand, transmission to a new host is likely to change carbohydrate patterns due to changes in PrPSc structure and exposure to a new pool of N-linked glycans in a new host. On the other hand, new pool of N-linked glycans might also play a role in selection of minor structural PrPSc variants in a new host. An interplay between the effect of PrPSc structure on selection of sialoglycoforms and conversely the effect of an altered pool of N-linked glycans on selection of PrPSc structural variants might explain the fact that stabilization of a new strain phenotype sometimes requires multiple serial passaging. Nevertheless, the role of N-linked glycans in determining the fate of prion strain adaptation has yet to be explored.
This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The animal protocol was approved by the Institutional Animal Care and Use Committee of the University of Maryland, Baltimore (Assurance Number A32000-01; Permit Number: 0215002).
Four to five week old Golden Syrian hamsters (all males, Harlan Laboratories, Indianapolis, IN) were inoculated intracranically (IC) into the left hemisphere, ~3 mm to the left of the midline and ~3 mm anterior to a line drawn between the ears under 2% isoflurane anesthesia. Each animal received 50 μl of PMCAb-derived materials diluted 10-fold in 1% BSA/PBS. After inoculation, hamsters were observed daily for disease using a ‘blind’ scoring protocol. Animals from the first passage did not develop any clinical symptoms and were euthanized at 518 days post inoculation by asphyxiation with CO2 (Table 1). For the second and third passages, 10% brain homogenates (BH) in PBS were dispersed by 30 sec of sonication immediately before inoculation. Each hamster received 50 μl of 10% BH inoculum IC under 2% isoflurane anesthesia and was observed daily for disease using a ‘blind’ scoring protocol. Animals that developed clinical signs were sacrificed at the terminal stages of the diseases as indicated in Table 1, whereas animals that did not develop clinical signs were sacrificed at 503 days or 614 days post inoculation for the second and third passages, respectively (Table 1).
PMCAb procedure has been described in detail elsewhere [60]. Briefly, healthy hamsters or mice were euthanized and immediately perfused with PBS, pH 7.4, supplemented with 5 mM EDTA. 10% brain homogenate (w/v) was prepared using ice-cold conversion buffer (Ca2+-free and Mg2+-free PBS, pH 7.5, 0.15 M NaCl, 1.0% Triton supplemented with 1 tablet of Complete protease inhibitors cocktail (cat # 1836145, Roche) per 50 mL of buffer) and glass/Teflon homogenizers attached to a cordless 12 V compact drill (Ryobi). The brains were ground at low speed until homogeneous, and then five additional strokes completed the homogenization. The resulting 10% normal brain homogenate (NBH) was used as the substrate in PMCAb reactions.
To produce RNA-depleted NBH, pre-cleared 10% brain homogenate was incubated with 100 μg/mL of RNase A (cat # R-4875, Sigma-Aldrich) for 1 hour at 37°C under gentle rotation prior to its use as substrate in PMCAb. RNA-depletion in RNase-treated NBH was confirmed by agarose gel as previously described [25]. To prepare seeds, 10% 263K brain homogenates in PBS were diluted 10-fold in the conversion buffer, and 10 μL of the dilution was used to seed amplification in 90 μL of fresh substrate. Samples in 0.2 mL thin-wall PCR tubes (Fisher, Cat. No. 14230205) supplemented with 3 Teflon beads (McMaster-Carr, Robbinsville, NJ) were placed in a floating rack inside a Misonix S-4000 microplate horn (Qsonica LLC, Newtown, CT) filled with 350 mL of water. Two coils of rubber tubing attached to a circulating water bath were installed to maintain 37 °C inside the sonicator chamber. The standard sonication program consisted of 30 s sonication pulses delivered at 50% power efficiency applied every 30 min during a 24 h period, consisting each round of 48 cycles. 10-fold dilutions were used between serial rounds in both mouse and hamster substrates.
To analyze PMCAb products, 10 μl of each sample were supplemented with 2.5 μl SDS and 2.5 μl PK (cat. #P8107S, New England BioLabs, Ipswich, MA), to a final concentration of SDS and PK of 0.25% and 50 μg/ml respectively, followed by incubation at 37°C for 1 hour. The digestion was terminated by addition of SDS-sample buffer and boiling for 10 min. Samples were loaded onto NuPAGE 12% BisTris gels, transferred to PVDF membrane, and stained with 3F4, D18 or SAF-84 primary antibody for detecting PrPSc as indicated.
To analyze scrapie BH, an aliquot of 10% brain homogenate was mixed with an equal volume of 4% sarcosyl in PBS, supplemented with 50 mM Tris, pH 7.5, and digested with 20 μg/ml PK (New England BioLabs) for 1 hour at 37°C with 1000 rpm shaking. The digestion was terminated by addition of SDS-sample buffer and boiled for 10 min and loaded onto NuPAGE 12% BisTris gels. After transfer to PVDF membrane, PrPSc was detected with 3F4 or SAF-84 primary antibody as indicated.
For the analysis of PK-resistance profile, 10% BHs of animals from the 3rd passage were diluted 10-fold into 4% sarkosyl in PBS and sonicated for 30 sec, then supplemented with equal volume of 100 mM Tris, pH 7.5. The samples were centrifuged 5 min at 16,000 g to remove debris. The supernatant was digested with PK (New England BioLabs) at increasing concentrations (2, 10, 50, 250, 1,000, and 2,000 μg/ml) for 30 min at 37°C under shaking. The digestion was terminated by addition of SDS-sample buffer and heating for 10 min in a boiling water bath. Analysis of GdnHCl-induced denaturation was performed as previously described [54].
Western blot signal intensity was digitized for densitometry analysis using AlphaView software (ProteinSimple, San Jose, CA). Target bands were selected using a uniform rectangular sampling area that encompassed the band of interest. Background optical density of an equal area from the same blot was determined and subsequently subtracted from the density of the bands. The values were normalized, with value at 2 μg/ml PK taken as 100%. Three independent brains were analyzed for each sample type, for calculating mean and standard deviations. The plots were drawn in Microsoft Excel.
Histopathological studies were performed on three animals per group. Formalin fixed brain halves were divided at the midline. Right hemisphere was frozen, and left hemisphere was fixed in 10% neutral buffered formalin solution. Formalin-fixed hemispheres were paraffin embedded, sliced into 4 μm sections and processed for hematoxylin-eosin stain as well as for immunohistochemistry for PrP using the mouse monoclonal anti-PrP antibody SAF-84 (1:1000, Cayman Chemical, Ann Arbor, Michigan, USA) or anti-glial fibrillar acidic protein (GFAP; 1:3000, Dako, Glostrup, Denmark). Horse radish peroxidase-labeled goat anti-rabbit and anti-mouse antibody (KPL, Milford, MA) were used as secondary antibody for GFAP (rabbit) and SAF-84 (mouse). Detection was performed using DAB Quanto chromogen and substrate (VWR, Radnor, PA). Brains were treated in formic acid (96%) prior to embedding in paraffin to deactivate prion infectivity. For detection of disease-associated PrP, we applied a pretreatment of 30 minutes hydrated autoclaving at 121°C followed by 5 minutes in 96% formic acid. As age-matched normal controls for the histopathology study, Golden Syrian hamsters of 660 days old were used. This age corresponds to the biological age of the oldest experimental group euthanized in the current work.
|
10.1371/journal.pntd.0003459 | Platelet Activating Factor Contributes to Vascular Leak in Acute Dengue Infection | Although plasma leakage is the hallmark of severe dengue
infections, the factors that cause increased vascular permeability have not been identified. As platelet activating factor (PAF) is associated with an increase in vascular permeability in other diseases, we set out to investigate its role in acute dengue infection.
PAF levels were initially assessed in 25 patients with acute dengue infection to determine if they were increased in acute dengue. For investigation of the kinetics of PAF, serial PAF values were assessed in 36 patients. The effect of dengue serum on tight junction protein ZO-1 was determined by using human endothelial cell lines (HUVECs). The effect of dengue serum on and trans-endothelial resistance (TEER) was also measured on HUVECs.
PAF levels were significantly higher in patients with acute dengue (n = 25; p = 0.001) when compared to healthy individuals (n = 12). In further investigation of the kinetics of PAF in serial blood samples of patients (n = 36), PAF levels rose just before the onset of the critical phase. PAF levels were significantly higher in patients with evidence of vascular leak throughout the course of the illness when compared to those with milder disease. Serum from patients with dengue significantly down-regulated expression of tight junction protein, ZO-1 (p = 0.004), HUVECs. This was significantly inhibited (p = 0.004) by use of a PAF receptor (PAFR) blocker. Serum from dengue patients also significantly reduced TEER and this reduction was also significantly (p = 0.02) inhibited by prior incubation with the PAFR blocker.
Our results suggest the PAF is likely to be playing a significant role in inducing vascular leak in acute dengue infection which offers a potential target for therapeutic intervention.
| Although plasma leakage is the hallmark of severe dengue
infections, the factors that cause increased vascular permeability have not been identified. As platelet activating factor (PAF) is associated with an increase in vascular permeability in other diseases, we set out to investigate its role in acute dengue infection. In this study, we found that PAF was significantly increased in patients with DHF, and the PAF levels rose just before the onset of the critical phase of dengue, during which vascular leak is thought to occur. PAF in serum of dengue patients was associated with reduced expression of tight junction proteins (ZO-1) and reduction in trans-endothelial resistance (TEER) of human endothelial cells. Use of PAFR blockers significantly reduced the down regulation of ZO-1 by serum of dengue patients and also the reduction of TEER, suggesting that PAF plays a significant role in inducing vascular leak in acute dengue infections.
| Dengue is thought to infect 390 million individuals per year resulting in approximately 96 million clinically apparent infections[1]. The annual burden of dengue has been estimated to be 750,000 disability adjusted life years (DALYs)[2] which is higher than the global burden of 17 other disease conditions, including upper respiratory tract infections, hepatitis and Japanese Encephalitis[3]. It has been declared a priority infection by the WHO, UNICEF and World Bank[4]. Currently there are no effective antiviral drugs to treat acute infection, nor a licensed vaccine to prevent infection.
Dengue infections are caused by four dengue virus (DENV) serotypes that are highly homologous [5]. Infection with any one of these serotypes can lead to asymptomatic infection disease or may manifest as undifferentiated viral fever, dengue fever or result in severe dengue infections in the form of dengue haemarrohagic fever (DHF), dengue shock syndrome (DSS) or expanded syndrome of dengue infection. Expanded syndrome of dengue infection is characterized by isolated organ involvement such as liver failure, myocarditic or encephalitis [4]. Although the majority of infections are asymptomatic or cause mild clinical disease, DHF and DSS are associated with a high morbidity and often with fatal outcomes. Increased vascular permeability leading to vascular leak is the hallmark of severe dengue infection [6]. Although the exact timing of vascular leak is not fully known, it is thought to occur early during infection and then substantially increase during the critical phase when it can be detected clinically or by laboratory methods [7]. The critical phase of dengue infection is thought to last for 24 to 48 hours following which the leaked fluid is reabsorbed and the patient recovers [4]. Complications as a result of plasma leakage such as shock, pleural effusions, ascites along with other complications such as liver failure and encephalopathy, also occur during the critical phase [4]. Currently the causes of increased vascular permeability are unknown. However, due to the rapid reversibility of increased vascular permeability, endothelial dysfunction rather than necrosis of the endothelium is thought to be the cause of vascular leak [6]. In fact, in postmortem studies, neither dengue NS1 antigen, viral protein or complement components have been detected in the endothelium, suggesting that endothelium dysfunction is likely mediated by host inflammatory mediators [8].
Cytokines and other mediators such as VEGF, TNFα and MCP-1 have been suggested to contribute to endothelial dysfunction and lead to vascular leak in dengue [9–14]; among these, VEGF has been extensively studied[12,13] and it has been documented that plasma VEGF levels correlated with vascular leak [13]. Using Human Endothelial cell lines (HUVECs) Appana et al. have shown that factors causing vascular leak are likely to be present in serum of dengue patients. In their experiments, sera from patients with acute dengue have shown to reduce expression of gap junction proteins and disrupt morphology of HUVECs [15]. Apart from the above, lipid mediators such as platelet activating factor (PAF), are known to play a role in increasing vascular permeability in disease conditions such as sepsis and anaphylaxis [16–19]. PAF is believed to be essential for the increase in vascular permeability and associated inflammatory changes seen in cerebral malaria [20]. PAFR−/− mice have shown to be less susceptible in developing severe dengue than the wild type mice [21]. In addition, thrombocytopenia and haemoconcentration observed in the wild type mice was significantly reversed by use of a PAFR blocker [21]. Platelets have been shown to be highly activated in dengue and platelet-monocyte aggregates were found to correlate with thrombocytopenia and increased vascular permeability [22]. Activation of both platelets and complement and release of inflammatory mediators is proposed as an alternate mechanism that causes vasculopathy leading to the plasma leakage [23].
The role of lipid mediators such as PAF has not been studied in dengue infection in humans. However, as PAF is involved in vascular leak in other diseases such as sepsis and anaphylaxis and since there is evidence of its potential role in causing vascular leak in mouse models, it would be crucial to evaluate the role of PAF in triggering vascular leak in acute dengue infection. In this study we found that PAF was significantly increased in patients with DHF and that the PAF levels rose just before the onset of the critical phase of dengue, during which vascular leak is thought to occur. PAF in serum of dengue patients altered expression pattern of tight junction protein ZO-1 and decreased the integrity of human endothelial cell monolayer, as measured by trans-endothelial resistance (TEER). Prior use of PAFR blocker significantly reduced these effects, suggesting that PAF plays a significant role in inducing vascular leak in acute dengue infection.
In the initial phase of the study, 25 adult patients with clinical features suggestive of dengue infection, admitted to a general medical ward in a tertiary care hospital (Colombo South Teaching Hospital) in Colombo during the year 2013, were enrolled following informed written consent. 16 of these patients developed DHF and 9 had DF. 12 dengue-seropositive healthy individuals were also recruited for the initial assays of lipid mediators.
Another 36 adult patients were enrolled for the second phase of the study, in which serial blood samples were taken in the morning (6 a.m.) and at 1.00p.m., from the time of admission to the time of discharge from hospital. The onset of illness was defined as the time of onset of fever. If the patient was recruited following 3 days of fever, it was considered as the duration of illness to be 72 hours. As patients with DHF were in hospital for a longer time than those with DF, approximately 5 to 7 serial blood samples were collected from them, but only 3 to 4 samples were collected from those with DF.
The study was approved by the Ethics Review Committee of the University of Sri Jayawardanapura. All adult patients provided informed written consent.
All clinical features, such as presence of fever, abdominal pain, vomiting, bleeding manifestations, hepatomegaly, blood pressure, pulse pressure and evidence of fluid leakage were recorded several times each day. The full blood counts, the alanine transaminase (ALT) and aspartate transaminase (ALT) levels were assessed during the course of the illness. Clinical disease severity was classified according to the 2011 WHO dengue diagnostic criteria [4]. The classification of whether the patient had DF or DHF was decided by the attending physician at the time of discharge after carefully reviewing the clinical and laboratory features and complications. Accordingly, patients with a rise in haematocrit above ≥ 20% of the baseline haematocrit or clinical or ultrasound scan evidence of plasma leakage in a patient was classified as having DHF. Shock was defined as having cold clammy skin, along with a narrowing of pulse pressure of ≤ 20 mmHg. According to this definition 25 patients were diagnosed to have DHF and 11 DF. Although some patients had very low platelet counts <25,000 cells/mm3, and high liver enzymes, they were classified as having DF since there was no evidence of fluid leakage.
Quantitative PAF, PAF-acetyl hydrolase and secretory PAF receptor levels were done in duplicate on all serum samples by quantitative ELISA. Levels of PAF, PAF-acetyl hydrolase (PAF-AH) and PAF receptor levels (PAFR) were initially done in the 25 patient samples and also in 12 healthy dengue seropositive individuals to determine if PAF, PAF-AH and PAFR were different in patients and healthy individuals before carrying out these assays in serial serum samples. Following the initial assessment of PAF in the 25 patient samples, the PAF levels were done in duplicate in all 36 serial samples. The levels of PAF, PAF-AH and PAFR levels (Cusabio, China) were carried out according to manufacturers’ instructions.
Acute dengue infection was confirmed in the serum samples using the NS1 early dengue ELISA (Panbio, Australia). All assays were done in duplicate. Dengue was also confirmed in these patients with a commercial capture-IgM and IgG enzyme-linked immunosorbent assay (ELISA) (Panbio, Brisbane, Australia). The ELISA was performed and the results were interpreted according to the manufacturers’ instructions. This ELISA assay has been validated as both sensitive and specific for primary and secondary dengue virus infections [24,25].
Human umbilical vein endothelial cells (HUVECs) (Lonza, Switzerland) were maintained in endothelial cell–based medium 2 (Lonza, Switzerland) supplemented with 10% fetal calf serum and growth factors (human epidermal growth factor, hydrocortisone, human recombinant fibroblast growth factor-beta, vascular endothelial growth factor, Insulin-like growth factor, Ascorbic acid, Heparin, FBS, and Gentamicin/Amphotericin-B) at 37°C at 5% CO2. Cells were grown in culture flasks or culture slides (BD, USA) pre-coated with 0.1% gelatin (Sigma, UK). Pre-coating was carried out by incubating flasks with 100 uL/cm2 of 0.1% gelatin at 37°C for 2 hours. PAF and PAF receptor antagonist (1-(N,N-Dimethylcarbamoyl)-4-ethynyl-3-(3-fluoro-4-((1H-2-methylimidazo[4,5-c]pyridin-1-yl)methyl)benzoyl)-indole, HCl (Calbiochem, Germany), (both from Millipore, Germany) were used for treatments. Both PAF and PAFR blocker were diluted in dH2O according to the manufacture instructions and aliquoted. PAF was stored in - 20°Ca and PAFR antagonist was stored in 4°C until further use.
Endothelial cells were seeded into gelatin-coated eight-well culture slides. The following day, serum samples from dengue patients (diluted in medium at a ratio of 1:3) or PAF were added and incubated for 3 hours at 37°C at 5% CO2. HUVECs were then immunostained for ZO-1 as described below. In experiments with PAF blockage, PAFR blocker was added to the culture medium one hour prior the addition of PAF or dengue serum.
HUVEC cells grown in cell chambers (BD, USA) were fixed with 2% paraformaldehyde (Alfa Aesar, UK) for 10 minutes and permeabilized (0.1% Triton X-100; Sigma, UK) for five minutes at room temperature. The cells were then blocked with 2% BSA and 5% FCS for 45 minutes. Purified rabbit monoclonal anti-human ZO-1 antibody (Life technologies, USA) (1:200 dilution) and secondary Alexa Fluor 488 mouse anti-rabbit IgG (heavy and light chains) (Invitrogen, USA) or Alexa Fluor 568 mouse anti-rabbit IgG (heavy and light chains) were used for staining; NucBlue Live ReadyProbes (Molecular probes, USA) was used to stain nuclei. Cells were then mounted with Mowiol 4–88 fluorescent mounting medium (Sigma, UK), and the data was acquired on a Zeiss LSM 780 Confocal Inverted Microscope; the image analysis performed using Image J software 1.47v (NIH,USA ). Confocal images were analysed using an automated FIJI macro. Discrete ZO-1 staining was isolated from images by taking raw image and subtracting Gaussian smoothed (sigma = 15) duplicate of the image. To make segmentation easier of the background-subtracted images, a Gaussian kernel (sigma = 2.0) was then applied to remove high-frequency noise from the clusters. Images were then threshold and the resulting binary mask skeletonised using the Fiji ‘skeletonize’ function. The binary fragments were then quantified using the ‘Analyse Particles’ plugin and the area summed to give a measure of total tight junction expression per image. To give an expression level per cell, the area sum value per image was divided by the number of cell nuclei present, as indicated by DAPI nuclear staining. All imaging experiments done in triplicate and five image fields per condition were obtained to include in the analysis.
24 well tissue culture plates and cell inserts (BD, USA) were coated with 500μl and 200μl 0.1% gelatin, respectively, and incubated for 2 hours at 37°C before adding the cells. The gelatin was washed with PBS and cell inserts kept in a in a companion plate (353504, BD, USA) and EGM-2 media (Lonza, Switzerland). Both to the insert (200μl) and the companion plate (700μl) were washed before addition of cell suspensions at a concentration of 50,000 cells/50μl of media and cultured overnight at 37°C in 5% CO2. On the following day, the inserts were transferred in to another companion plate with 700μl of EGM-2/well and 250μl of EGM2 was added in the insert prior to measuring TEER using Millicell-ERs (Fisher Scientific, UK). The experiments were carried out when the HUVECs formed a confluent monolayer and plateau of electronic resistance was observed on the 3rd day when the resistance reached 450 ohms. 10 different experiments were carried out in triplicate using serum from healthy individuals and also serum from healthy individuals with PAFR blocker (1-(N,N-Dimethylcarbamoyl)-4-ethynyl-3-(3-fluoro-4-((1H-2-methylimidazo[4,5-c]pyridin-1-yl)methyl)benzoyl)-indole, HCl (Calbiochem, Germany). To determine the effect of dengue sera on TEER, 9 separate experiments with dengue serum in the presence and absence of PAFR blocker were carried out in three biological replicates. In all the experiments the PAFR blocker was added 1 hour prior to adding serum of dengue patients and incubated at 37°C.
Statistical analysis was performed using Graph Pad PRISM version 6. As the data were not normally distributed, differences in means were compared using the Mann-Whitney U test (two tailed); when three or more groups were compared Kruskal Wallis test was used. Receiver-operator characteristic (ROC) curves, showing the area under the curve (AUC) were generated to determine the discriminatory performance of the highest serum PAF level detected with regard to severity of illness.
Serum samples for analysis of PAF, PAF-AH and soluble PAF-Receptor was obtained on day 5–6 of illness. In the initial analysis of all 25 patients with acute dengue infection and healthy individuals, PAF levels were significantly higher in patients (p = 0.002) when compared to healthy individuals. Although not significant (p = 0.15), PAF levels were higher in patients with DHF (median 335.2, Inter quartile range 4.7 to 443.1 ng/ml) when compared to those with DF (median 47.63, IQR 0 to 111.6 ng/ml) (Fig. 1A). Since high PAF values could be either due to increased production or reduced breakdown, we also analysed PAF—acetyl hydrolase (PAF-AH) levels in these patients. PAF-AH is the enzyme that breaks down PAF [26]. PAF-AH levels have been shown to be low in some diseases such as asthma and anaphylaxis and thus thought to contribute to disease pathogenesis by reduced breakdown of PAF [27,28]. We found that PAF-AH levels were significantly higher (p<0.0001) in patients with acute dengue when compared to healthy individuals. The PAF-AH levels were significantly higher (p = 0.01) in those with DHF (median 112.6, IQR 88.35 to 151.1 ng/ml) when compared to those with DF (median 85.35, IQR 73.2 to 98.3 ng/ml) (Fig. 1B) suggesting that high PAF values were not due to reduced breakdown of PAF. We also assessed soluble PAF-Receptor levels in serum and found that there was no difference in PAF-R levels in patients with acute dengue or in healthy individuals (Fig. 1C).
As dengue infection is a very dynamic disease, the patients for determining kinetics of PAF were recruited on a mean of 106.2 (SD±19.2) hours of illness. As we found that PAF levels were significantly higher in patients with acute dengue, we assessed PAF levels in serial blood samples collected from patients throughout the course of the illness. Patients with DHF had significantly higher PAF throughout the course of the illness when compared to those with DF (Fig. 2). However, there was a wide variation in the PAF levels in patients from both groups. Except for 3 patients with DF the PAF levels of all other patients (8/11) with DF never rose to >100 ng/ml throughout the course of the illness. Of these 3 patients who had higher values, one patient had platelet counts that dropped to 30,000 cells/mm3 and she also complained of vaginal bleeding in the absence of her usual menstrual period. However, she was classified as having DF as she did not have any clinical or laboratory evidence of fluid leakage. The second patient with DF whose PAF levels rose to 293.16 ng/ml, also complained of vaginal bleeding in the absence of menstruation. The other DF patient whose PAF levels rose to 123.4 ng/ml only had a mild rise in liver enzymes, no evidence of fluid leakage and no bleeding manifestations. 3/25 patients with DHF had values <100 ng/ml. One of these patients presented to hospital on day 6 of illness and had already progressed to the critical phase. The other 2 patients were not in the critical phase on admission and were admitted to hospital on day 4.
Interestingly, a diurnal variation in PAF levels was observed in the majority of patients with DHF but not in those with DF.
As PAF has shown to reduce expression of ZO-1 in HUVECs and increase endothelial permeability, we initially assessed if similar observations were found in our model. As expected, the use of PAF on HUVECs significantly reduced (p = 0.007) surface expression of ZO-1, which was dose dependent (Fig. 3A). ZO-1 expression of HUVECs was significantly up regulated in a dose dependant manner with the use of a PAFR blocker (Fig. 3B). The PAFR blocker(1-(N,N-Dimethylcarbamoyl)-4-ethynyl-3-(3-fluoro-4-((1H-2-methylimidazo[4,5-c]pyridin-1-yl)methyl)benzoyl)-indole, HCl (Calbiochem, Germany), potentially inhibits binding of PAF to its receptor in a competitive manner in equilibrium binding studies. A non-competitive inhibition is reported if the membrane bound PAFR are pre-incubated with this blocker before adding PAF, which is thought to be due to a slower antagonist dissociation rate (Calbiochem, Germany). Since the inhibition of PAF was most significant when the PAFR blocker was used at a concentration of 500ng/ml, this concentration was used for other blocking experiments.
The above experiments showed that PAF reduces expression of ZO-1 and this effect was significantly inhibited by the use of a PAFR blocker, we then proceeded to determine the effect of serum from dengue patients on expression of ZO-1. We also found that serum from patients with DHF significantly downregulated ZO-1 expression (p = 0.004) (Fig. 3C and Fig. 4). Furthermore, HUVEC cells incubated with serum from DHF patients showed disrupted morphology, reduced ZO-1 expression and widening of gap junctions (Fig. 4). However, the down regulation of ZO-1 expression by dengue sera was significantly inhibited (p = 0.004) by incubating HUVECs with a PAFR blocker (Fig. 3C and Fig. 4).
The HUVECs incubated with serum from dengue patients showed disrupted morphology, reduced ZO-1 expression and widening of gap junctions (Fig. 4). The ZO-1 expression was seen to be restored when HUVECs were pre-incubated with a PAFR antagonist (Fig. 4).
As the above experiments showed that both dengue serum and PAF affected on ZO-1 expressionin a dose dependant manner, which was inhibited by the use of a PAFR blocker, and also the effect of dengue serum of ZO-1 expression was inhibited by a PAFR blocker, we next proceeded to determine the effect of dengue sera on trans-endothelial resistance (TEER). The use of serum from dengue patients significantly reduced TEER (mean −35.82, SD ± 12.93 Ώ) when compared to use of serum from healthy individuals (mean 1.96, SD ± 1.88 Ώ). This reduction in TEER by dengue serum was significantly (p = 0.002) inhibited by the use of a PAFR blocker prior to incubation of the HUVECs with dengue sera (Fig. 5).
In this study we have investigated the role of PAF as a mediator of vascular leak in acute dengue infection. We found that PAF levels were significantly elevated in patients with dengue infection, as well as its principle breakdown enzyme PAF-AH, which suggested that the increase in PAF was likely due to increased production rather than reduced breakdown. We also found that PAF levels were significantly higher in patients with DHF throughout the course of the acute disease when compared to those with DF, although huge inter-individual variations in PAF levels were observed. PAF has been shown to be important in vascular leak in dengue mice models and PAFR−/− mice were shown to have milder clinical disease [21]. The role of PAF in human dengue infection has only been investigated in the context of in vitro studies where mononuclear leucocytes of dengue immune donors were found to produce more PAF than non-immune donors [29].
HUVECs models have been widely used to assess increase in vascular permeability by many mediators and drug molecules as well as to determine changes in trans-endothelial electrical resistance (TEER) [11,15,30]. Experiments carried out by Appanan et al. showed that mediators present in the serum of dengue patients reduced expression of tight junction and adherent junction proteins which are likely to result in increased vascular permeability [15]. We found that PAF reduces expression of ZO-1 in a dose dependent manner and this downregulation of ZO-1 was significantly inhibited by the pre-treatment of HUVECs with a PAFR blocker. Our results further show that similar to the findings of Appanna et al [15], incubation of HUVECs with serum from dengue patients resulted in down regulation of ZO-1 expression but we here show that this was significantly inhibited by pre-treatment with a PAFR blocker. This suggests that PAF present at high concentrations in serum of dengue patients is likely to contribute to vascular leak by reducing expression of tight junction proteins. In addition to our experiments with HUVECs in assessing ZO-1 expression, we also investigated the effect of dengue serum on TEER in HUVECs. We found that serum from dengue patients significantly reduces the TEER in HUVECs when compared to serum from healthy individuals and this reduction of TEER was significantly inhibited if the HUVECs were pre-treated with a PAFR blocker prior to addition of serum from dengue patients. Therefore, these data further confirm that PAF present in serum of dengue patients reduces TEER in the endothelium, as this reduction was significantly ameliorated by a PAFR blocker. However, although the reduction of TEER in HUVECs was significantly inhibited by the use of a PAFR blocker, the TEER still did not return to normal, suggesting that apart from PAF, other mediators in the serum could also contribute to the vascular leak.
PAF is a potent inflammatory lipid mediator rapidly produced by many cells, such as endothelial cells, monocytes, mast cells and leucocytes following cellular stress [31]. It is known to cause hypotension, thrombocytopenia, increased vascular permeability and cardiac dysfunction when experimentally administered to animal models [32–34]. In dengue infection, platelets have been shown to be highly activated and platelet-monocyte aggregates have shown to contribute to the increase in vascular permeability [22]. Although there could be multiple pathways leading to activating of platelets, PAF could be further be contributing significantly to platelet activation and thus the immunopathology of the severe forms of the disease.
In this study we also found that PAF varied in the same patient in samples collected in the morning and the afternoon. Since we sampled patients only twice a day, it is difficult to comment if the variation in PAF levels was diurnal or whether such variations are observed more frequently. It has been shown that human monocytes produce PAF in a bi-phasic pattern when stimulated with LPS, which was shown to be due to the effects cytokines such as TNFα and IL-1β [31,35]. PAF has been shown to activate transcription of NF-κB resulting in expression of many inflammatory cytokines such as TNFα and IL-1β [31,35,36]. Since LPS was the main stimulus that resulted in bi-phasic production of PAF and other cytokines, it is possible that LPS plays a similar role in acute dengue infection. For instance, it has been shown that patients who develop plasma leakage have significantly higher levels of LPS than those who did not have plasma leakage [37]. Therefore, the possibility of LPS driving the production of PAF and other cytokines should be further investigated.
In summary, our results show that PAF levels were significantly higher in more severe forms of dengue and were associated with a reduced expression of tight junction proteins and reduced cell layer integrity that is likely to result in an increased paracellular leak. Use of PAFR blockers significantly reduced these effects; therefore, our results suggest the PAF is likely to be playing a significant role in inducing vascular leak in acute dengue infections; this has implications for the future management of patients such as use of PAFR blocker in acute dengue infection.
|
10.1371/journal.pcbi.1006136 | Design of optimal nonlinear network controllers for Alzheimer's disease | Brain stimulation can modulate the activity of neural circuits impaired by Alzheimer’s disease (AD), having promising clinical benefit. However, all individuals with the same condition currently receive identical brain stimulation, with limited theoretical basis for this generic approach. In this study, we introduce a control theory framework for obtaining exogenous signals that revert pathological electroencephalographic activity in AD at a minimal energetic cost, while reflecting patients’ biological variability. We used anatomical networks obtained from diffusion magnetic resonance images acquired by the Alzheimer’s Disease Neuroimaging Initiative (ADNI) as mediators for the interaction between Duffing oscillators. The nonlinear nature of the brain dynamics is preserved, given that we extend the so-called state-dependent Riccati equation control to reflect the stimulation objective in the high-dimensional neural system. By considering nonlinearities in our model, we identified regions for which control inputs fail to correct abnormal activity. There are changes to the way stimulated regions are ranked in terms of the energetic cost of controlling the entire network, from a linear to a nonlinear approach. We also found that limbic system and basal ganglia structures constitute the top target locations for stimulation in AD. Patients with highly integrated anatomical networks–namely, networks having low average shortest path length, high global efficiency–are the most suitable candidates for the propagation of stimuli and consequent success on the control task. Other diseases associated with alterations in brain dynamics and the self-control mechanisms of the brain can be addressed through our framework.
| This work aims to close the knowledge gap between theory and experiment in brain stimulation. Previous modeling approaches for stimulation have overlooked the nonlinear dynamical nature of the brain and failed to shed light on efficient mechanisms for the exogenous control of the brain. Amid the current efforts for developing personalized medicine, we introduce a framework for producing tailored stimulation signals, based on individual neuroimaging data and innovative modeling. This is the first time, to our knowledge, that brain stimulation for the most common cause of dementia, Alzheimer’s disease, is theoretically addressed. Our approach leads to the identification of potential target regions and subjects to successfully respond to brain stimulation therapies and yields various disease-reverting signals. Although focused on Alzheimer’s in this study, our methodology could be applied to other clinical conditions characterized by abnormalities in brain dynamics, like epilepsy and Parkinson’s, the treatment of which can benefit from the use of optimal control strategies.
| Alzheimer’s disease (AD) is the most common cause of dementia, with classic biomarkers including vascular and glucose metabolism dysregulation, amyloid-β and tau deposition, white matter degeneration, functional impairment, and grey matter atrophy [1]. Because of the complex mechanisms and non-physiological factors that interact in an intricate manner [2], our understanding of the disease and ability to produce efficient therapeutic interventions has been limited.
One area of therapeutic interventions currently being investigated is brain stimulation, which aims to correct (control) pathological activity by steering it towards a trajectory (pattern of brain activity) we consider healthy. Several attempts have been made to assess the potential brain stimulation has for treating AD. For example, cognitive improvement was found immediately after applying repetitive transcranial magnetic stimulation (rTMS) and transcranial direct current stimulation (tDCS) [3]. An ongoing clinical trial has reported reversion of impaired glucose metabolism in the temporal and parietal association cortices along with slowing of cognitive indicators for the progression of AD by applying deep brain stimulation (DBS) in the fornix [4,5]. However, unlike the case of Parkinson’s disease where a decrease in tremor constitutes a short-term measure for the success of the therapy, these AD studies lacked such a biomarker. Consequently, they were unable to guarantee that their stimulation parameters were the optimal for their purposes. By optimal, we understand inputs that make the pathological state disappear while the energy used by the external controlling agent (cost) is minimal. Beyond the use of DBS for AD, most brain stimulation protocols are likely suboptimal in their selection of the signal shape, amplitude, and stimulation sites, since they are set by trial-and-error. Additionally, stimulation treatments are currently identical for all individuals with the same clinical condition, disregarding biological variability [4–8], and highlighting the need for an increased understanding on how to optimize stimulation protocols for individual patients. To address some of these issues, computational modeling techniques have been previously used [9–12].
However, modeling brain stimulation requires the consideration of some well-known facts, such as that the evolution of brain activity is intrinsically related to the subjacent anatomical network and the interplay of various neuronal populations [13]. As in any other network, it is reasonable to assume some elements (or nodes) have an architectural leading role in the self-regulation of that neural system [14,15]. An input feeding into one of these elements has the potential to propagate through the network, influencing the system towards the state desired by the controller. The existence and characteristics of such input signals are then given by the dynamical structure of the system and the way its elements are coupled to the inputs. Systems in which those signals that drive the activity to a desired configuration exist are known as controllable (as opposed to uncontrollable), relating to the property ‘controllability’ [16].
Some studies have focused on identifying the most suitable sites for network controllability from a structural viewpoint only [17] while simplifying the dynamical interactions occurring on top of the connectivity scaffold. Other studies [9–12] used linear dynamics to model neural processes, which are known to be intrinsically nonlinear [13,18–20]. Hence, their predictions on neural network control should be taken with caution. Neglecting the nonlinear nature of the brain for the sake of mathematical simplicity might bias or corrupt the results therein obtained. Taylor et al., in their seminal paper [10], simulated seizure abatement in a nonlinear model by means of the so-called pseudospectral method. Their approach assumed that the entire cortex was stimulated, thus resulting in a readily controllable system. However, a global control-strategy might be not achievable in practice, and may relate to high energy deposition over the brain tissue. Additionally, only stimuli that were independent of the state variables (open-loop) were considered, whereas recent evidence supports the idea of enhanced benefits associated with closed-loop brain stimulation [21–24].
Finally, the identification of optimal (electromagnetic) signals among the universe of those that can be created [6–8] has a paramount importance in terms of patient’s welfare and technological improvement. As such, modeling approaches should be able to predict the brain structures that better respond to targeted stimulation for achieving a control objective over the network. For instance, the surgical implantation of devices (for DBS) could be avoided if theoretical calculations envisage that stimulation of cortical neuronal conglomerates, with, e.g., tDCS, produces comparable results to what is achieved by means of DBS. In the same way, pinpointing optimal control signals likely translates to less exposure for the patient and to a reduction of procedure-related costs in terms of number of sessions required, the shape and amplitude of the signals that are used, etc.
In this work, we attempt to reconcile the theory of neural network control and the true nonlinear nature of the brain and shed light on the development of efficient stimulation therapies for AD. One framework that deals with nonlinearities while optimizing input signals for controlling dynamical systems is the state-dependent Riccati equation control (SDRE) [25,26]. SDRE has several applications in mechanical problems and aerospace engineering [26] though few in the fields of biological and high-dimensional systems, where the above-mentioned simplistic linear approaches have been preferred. We use SDRE to obtain the optimal signals to steer AD systems towards healthy states and to predict the best candidate brain regions and subjects to undergo a stimulation therapy.
We compute anatomical connection density matrices for the interaction between neuronal populations in cortical and subcortical brain regions from diffusion weighted magnetic resonance imaging (DW-MRI) data acquired by the Alzheimer’s Disease Neuroimaging Initiative (ADNI). A nonlinear dynamical module (Duffing oscillator) [27] is then assigned to each area, as the generator of its macroscopic electrical activity. A pathological state is defined as one in which all oscillators present high-amplitude theta-band frequencies. Conversely, in a healthy state, they oscillate with an alpha-band frequency. This designation of the pathological and healthy states seeks to match the slowing of electroencephalograms (EEG) induced by AD [28–34]. The control tasks consist of shifting the pathological activity of the nonlinear system to healthy activity, even though the damage the disease has caused to the patient is irreversible. The optimal signals for accomplishing this control objective are region and subject-specific, and obtained through SDRE. Brain regions can be ranked according to the energetic cost of performing the control task. These orders are nonlinearity-dependent. Interestingly, the regions associated with lower costs for the controller are those more topologically connected to the rest of the nodes in the network. These regions lie mostly in the limbic system or basal ganglia. On the other hand, subjects with compact networks (e.g., low average shortest path length, high clustering coefficient, high global efficiency,) can be controlled with more inputs to individual regions than the rest.
We obtained anatomical connection density matrices, W, from DW-MRI (Methods, Data acquisition and processing) for each of the 41 patients in ADNI (Methods, Study participants; S1 Table; S2 Table). We applied the SDRE optimal control framework (S1 Text) to a system of Duffing oscillators (Methods, Dynamical systems) representing the electrical activity of cortical and subcortical ‘pyramidal neuron’ populations coupled through the anatomical connection matrices. Specifically, we conceived a hypothetical experiment in which all the regions in the anatomical parcellation we used [35] can receive an input, but where each simulation has the input entering only one area.
We calculated the minimal-energy input that converted pathological (theta-band) into healthy (alpha-band) activity, for every region and subject (Methods, Control tasks). The rationale for this is the increase of power in the theta band (4.0–7.5 Hz) of the EEG spectrum, and the decrease of power in the alpha (8.0–12.5 Hz) and beta (13.0–32.0 Hz) bands in AD [28,29]. As reported by several studies, there is a correlation between cognitive impairment and the acuteness of EEG abnormalities [30]. Additionally, the use of cholinergic drugs (which can transiently shift the EEG spectra towards normality) was related to improved memory and attention performances in AD [32]. In another study, Babiloni et al. [33] found that theta sources of the EEG in parietal, occipital, temporal and limbic areas had higher magnitude in AD than in healthy controls, while alpha sources had lower magnitude in AD. Moreover, all the alpha sources showed positive correlations with the Mini Mental State Examination (MMSE) score for global cognitive level, suggesting the favorable impact of a shift towards elevated alpha activity. A different, posterior study also found positive correlations between alpha power and the patients’ MMSE scores [34]. Although a specific causal relationship between EEG rhythms and AD has not been established, brain stimulation that corrects EEG abnormalities will also positively affect the patient’s welfare by restoring cognitive performance, yet the disease is not cured. The persistence of the disease appears in our model through its parameters and the anatomical connection density matrices which is consistent with reports showing abnormalities in the graphs obtained from DW-MRI in AD [15,36,37]. The general scheme of our methodology is presented in Fig 1.
In contrast to conventional ideas on brain stimulation where identical signals are applied [4–8] regardless of subject-to-subject variability, we calculated a broad set of patient-specific signals that revert AD pathological activity, and studied their performance on the control tasks. Fig 2A, 2C and 2E show (respectively) the initial set-up of the temporal solutions of the nonlinear model, their behavior in the last five seconds of the simulated interval, and the optimal control signal, u(t), that hypothetically enters the left pallidum, in this example, and produces a successful control task. This corresponds to a specific subject in ADNI’s database. Fig 2B, 2D and 2F present the same analysis for a second subject. In both cases, the strength of the nonlinearity was γ = 200 s−2mV−2. This is a typical value among the strengths of the nonlinearity we tested (see S3 Table for a complete list of the parameters used throughout the study).
As seen from the temporal evolution of variable x, which represents the postsynaptic potential over one randomly chosen region in the model, the controlled trajectory almost identically matches the desired trajectory (low-amplitude alpha oscillation) by the end of the simulation (Fig 2C and 2D). Please, note the subtle differences in the signals the controller is set to deliver from one subject (Fig 2E) to the other (Fig 2F).
These dissimilarities are mostly due to the generation of subject-dependent minimal-energy signals. The magnitudes of the calculated optimal signals (-0.1–0.1 V, approximately) are around one order lower than the signals that are currently used in DBS for AD (3.0–3.5 V) [4,5]. The magnitude generally decreased with time although the signals possessed complicated shapes. The energetic cost of controlling the full network of oscillators was also computed. Roughly speaking, the energy was defined as the time-integral of the norm of the control input, u(t) (S1 Text). We found that low magnitude signals are associated with reduced costs (Fig 2E and 2F).
If the input was placed over a different region, the system might or might not be controllable. Several subject-dependent cases in which the optimal control framework failed to produce stimulation signals were obtained for nonlinear systems. S1 Fig shows equivalent results to those in Fig 2, although obtained over linear systems (γ = 0 s−2mV−2). No case of uncontrollable systems for any subject was found for signals entering the linear variant of the model though, which seems unrealistic to occur in any practical implementation. Additionally, the magnitude of the control signals obtained was generally lower for linear than for nonlinear systems. These results depend on the anatomical connection matrices and the dynamical model (coupled Duffing oscillators) we have used for the simulations.
We collected the results of all simulations to construct a general picture of the power regions (nodes in the networks) have to control the AD system. Results for the simulations using the same strength of the nonlinearity, γ, were averaged across all the subjects in the study. An uncontrollable system is associated with an infinite energetic cost (whereas the inverse of the cost will be zero) since an infinite input signal would be required to correct pathological EEG activity in such a case. To overcome the presence of uncontrollable systems with proper visualization tools, we chose the inverse of the cost as the variable of interest for ranking the performance of the brain regions in our control tasks. Thus, a region with high inverse of the cost is associated with enhanced optimal control (the full network can be readily controlled with an input entering such region).
Fig 3A shows the brain areas’ ranking for the limit case of a linear system (mean inverse of the cost ± standard error of the mean). Top-ranked areas appear in the leftmost part of the panel. Fig 3B contains a graphical visualization of the brain sites where they are approximately located. The size of the spheres is directly proportional to the inverse of the cost. We found that several of the top-ranked regions are spatially close, with predominance over the left hemisphere. New rankings were obtained when the nonlinearities increased (see Fig 3C and 3D). As the magnitudes of the costs generally grow with the strength of the nonlinearity, the upper limit of the vertical axis in Fig 3C, representing the maximum mean inverse of the cost registered for γ = 200 s−2mV−2, is smaller than the corresponding one in Fig 3A.
Additionally, we assessed the relationship between the rankings of the regions resulting from controlling systems with different nonlinearities. The statistical dependence between the rankings associated with the nonlinearities was measured in terms of Spearman correlation (Pearson correlation between the rankings). The Spearman’s rank correlation coefficient (Spearman’s rho) between the linear system’s order and the corresponding to a nonlinear system with γ = 100 s−2mV−2 was ρ = 0.98 (p < 0.001). It decreased to ρ = 0.87 (p < 0.001) when the nonlinearity was increased to γ = 200 s−2mV−2 and further down to ρ = 0.59 (p < 0.001) for γ = 300 s−2mV−2. The orders corresponding to two consecutive nonlinearities also differ more (Fig 4A).
These ‘expected’ rankings, obtained from looking at the average inverse of the cost only, had similarities in their top and bottom-most components (Fig 4B), suggesting a global privileged/disadvantageous position of some areas in the brain network that transcends the effects of the nonlinearities. We consider it important to note that individual cases of uncontrollable systems were ubiquitously reported when nonlinearities were considered. Only the individual calculation of the minimal-energy control signals, instead of an analysis over the main values as performed in this section, can conduce to a trustable subject-specific selection of stimulation targets.
Nevertheless, it is interesting to note how regions on the top of the mean control order (Fig 3A and 3C) belonged to a clearly defined group with prevalence in the left hemisphere: the left pallidum, left putamen, left amygdala, left hippocampus, right thalamus proper, left insula, left basal forebrain, left fusiform and the caudate nuclei. Overall, these high-ranking regions belong to the limbic system and the basal ganglia. On the other hand, the worst-ranked areas included the right postcentral gyrus, both paracentral lobule, right inferior and superior parietal lobules, left cuneus and the right temporal lobe, which can all be classified as temporoparietal regions. Temporal and parietal cortical areas are affected in AD early in the disease course [4].
Identifying the most suitable candidates for a successful brain stimulation treatment remains a challenge. Even subjects suffering from the same condition are intrinsically different due to genetic and environmental factors [38]. Here, we aimed to create a gold standard for selecting both stimulation sites and individuals most likely to benefit from stimulation therapy based on the subject’s anatomical networks estimated from DW-MRI data. We looked at the relationship between the results of the control tasks and the topological characteristics of the networks, W′s, over which they are performed (S2 Text).
To gain further insight into the regions offering the best optimal control perspectives, we studied the relationship between the mean regional inverse of the costs and average local measures. The chosen topological measures were node strength, si; eccentricity, ei; closeness centrality, qi; betweenness centrality, bi; clustering coefficient, ci and communicability, Mi. In Fig 5A–5F we show the above-mentioned dependences. These topological quantities classify the degree of ‘importance’ a node has in the network. For example, a low eccentricity denotes short paths from a node to the rest of the network, which is also interpreted as a high closeness centrality, whereas communicability counts direct and indirect paths of all lengths between two nodes. The strength accounts for both, the number of connections a node has and the value of the connection weights. On the other hand, high values of betweenness centrality relate to nodes that act as bridges in the network and a high clustering coefficient means tendency to form triangles, or cluster together.
We found significant correlations between the mean inverse of the cost of controlling the brain network from a region and all the local measures (si, ei, qi, bi, ci and Mi). The only decreasing relationship found was with the eccentricity, ei, meaning that regions with a small shortest path length might constitute the more suitable targets for controlling the network. On the other hand, nodes with high si, qi, bi, ci and Mi were associated with high mean inverse of the cost. Analyzing the strength of the correlations revealed an interesting pattern: the three correlation coefficients appearing on the top row of Fig 5 –for quantities strictly related to direct connections–were considerably higher than those on the bottom which relate to measures for quantifying relay nodes, segregation levels and indirect paths, respectively. This suggests that direct links (high weights, small shortest paths) between nodes are what makes a stimulus fully propagate over a network to reach the control objective at a low energetic cost.
As previously expressed, a set of inputs entering certain specific regions for each subject failed to convert theta activity into alpha activity. The number of successful signals thus provides a good estimate of how responsive the patients would be to the tentative treatment herein modeled. Therefore, we studied the relationship between the number of inputs resulting in controlling the systems and global measures of the subjects’ anatomical network. This is shown in Fig 6A–6D (characteristic path length, l; radius, r; average clustering coefficient, C, and global efficiency, Eg, in this order). We found that subjects with small average shortest path length (l) of their anatomical networks were controlled by more inputs–in other words, from a high number of regions. In the same way, the lower the radius was, the more inputs were efficient in the control tasks. More clustered networks yielded the same result. Small average distance between the nodes in the network and high clustering coefficient are attributes associated with the small-worldness property [39], a concept that relates the fast spread of stimuli to the existence of ‘shortcuts’ in a network. Finally, the number of areas from which the AD brain could be controlled per subject was also proportionally related to the global efficiency, a measure that reflects how efficiently information can be exchanged over the network. We did not obtain any significant correlation between the number of controllable dynamical systems and the diameter of the networks.
What is presented here stands for all the strengths of the nonlinearity we tested. All the results in this section corresponded to γ = 200 s−2mV−2. However, the magnitudes of the correlation coefficients were higher as the strength of the nonlinearity decreased. The same analysis presented in Fig 5 can be found in S2 Fig for the linear systems.
The goal of brain stimulation is to exogenously control (i.e., manipulate) the brain’s activity so that it follows a desired pattern associated with a healthy state [11]. The specific characteristics of the signals in brain stimulation experiments/therapies are usually overlooked. Square pulses are translated from the treatment of one condition to the other (e.g., from Parkinson’s to Alzheimer’s) [4,5], sometimes tuned in an exploratory way, and applied identically to every subject without considering individual differences. In a world where medicine is constantly becoming more personalized, treatments which are designed using broad statistical measures and account poorly for interpatient variability are inefficient [38]. Current approaches to modeling brain stimulation present major shortcomings (see the recent review by Bassett et al. [13], for example) and importantly, nonlinearities are known to characterize the brain’s dynamical behavior and should not be excluded from any realistic modeling.
In this work, we introduced a framework for calculating the optimal signals and most suitable regional targets in the brain for controlling AD activity, catered to individual subjects. Unlike other studies [9,11,12], ours considers the existence of nonlinearities in the modeling of brain dynamics by extending the use of the so-called state-dependent Riccati equation control to biological, high-dimensional systems. The calculation of the optimal signals that can propagate over the network and set its temporal dynamics to a desired control state also provides insights into the way neural systems control themselves. If a network node associates with low cost for exogenously controlling the neural system, then that same element must have certain advantaged position for the self-regulation processes occurring there. Importantly, our methodology is not restricted to AD. Any other clinical condition characterized by abnormalities in brain dynamics (and with existing meaningful neuroimaging data for using in the modeling process) could be addressed similarly with: 1) a model for the dynamics is assumed, 2) the model is set to produce pathological and healthy activity, and 3) brain stimulation signals that revert pathological activity at the lowest possible energetic cost are found through SDRE.
We modeled a stimulation-therapy for AD based on the correction of the EEG spectrum towards higher frequencies. As such, we looked for inputs (control signals) to individual areas of the brain that revert pathological activity at the lowest possible energetic cost. Among all the possible signals that were obtained for each subject, the one producing the fastest, least energy-consuming response, can be administered in a brain stimulation procedure. On the other hand, the SDRE-framework allows us to naturally identify those brain regions into which stimulation would not contribute to achieving the control objective (by using the definition of uncontrollable nonlinear system). The right postcentral gyrus, for example, yielded uncontrollable systems for all the subjects in the sample.
We also studied the dependence of the optimal control tasks on the anatomical networks conditioning the dynamics. In essence, we found a strong relationship between the success of the control tasks and the topological features of the anatomical connection density matrices that served as scaffold for the interaction of the cortical and subcortical ‘pyramidal neuron’ populations in the model. Overall, the significant correlations existing between the mean inverse of the cost and the local topological measures suggest that nodes with high connectivity associate with low cost of controlling the full network of oscillators. Our results agree with previous findings that stimulation to strongly connected nodes in brain networks produces low-energy transitions [11,12]. For each subject, we found that the better connected a network is–namely, a network having low average shortest path length, high clustering coefficient and/or high global efficiency–, the more inputs to individual nodes success on the control task and make the system evolve to the predetermined healthy state (subjects having ‘a better-connected network’ are the optimal candidates for AD’s effects-reverting protocols). However, these indications relating optimal nonlinear control and network measures are only an approximation (based on average values), and we recommend the calculation of the optimal signals and targets for each subject to undergo our proposed brain stimulation for AD.
The inclusion of nonlinearities in our model causes several control tasks to fail for each subject, a fact that, to the best of our knowledge, has not been reported for linear brain dynamics. However, we do not expect that inputs to every neuronal conglomerate in real stimulation experiments are able to steer the (AD) brain to the desired state, given its complexity and nonlinear character [18–20]. The order in which areas were ranked according to the energy used for controlling the network, changed with the strength of the nonlinearity, γ. Interestingly, as γ increased, the linear dependence of the expected cost of controlling from a region on its topological characteristics was less obvious (see Fig 5 and S2 Fig)–the higher the γ is, the less the systems look like sets of linear (harmonic) oscillators coupled through the anatomical connection density matrices. This likely denotes competition between the effects of the nonlinearity and the structure of the network for the dynamical interaction, and warrants further investigation. Overall, our findings reveal the importance of using nonlinear realistic modeling to better understand brain stimulation and its accurate design.
When ranking the regions in the brain according to the average cost of controlling the network with a single stimulus, we found that the lowest energetic cost was associated with limbic and basal ganglia areas, or strongly connected to them, such as the thalamus. The role of these areas in motor control, learning, memory and relay of information [40,41] engages them in a wide number of connections, and consequently (see Fig 5), makes them highly desirable targets for stimulation. The globus pallidi sends basal ganglia information to the thalamus which projects back to the cortex [41]. Specifically, the left pallidum–at the top of the nonlinear systems ranking–has been previously identified as having the least overall multifactorial damage by AD [1]. The caudate nuclei and putamen receive and process cortical and thalamic information which is later transmitted to the globus pallidi [41]. On the other hand, the large-scale brain network topology seems to be organized to concentrate information flow in the hippocampal formation [42], structure with a key role in memory processing [43], and also among those associated with better optimal nonlinear network control in this work. Finally, the amygdala has a broad pattern of anatomical connections, especially with other subcortical structures [44], making it another of the top targets for achieving successful control tasks.
The bulk of the poorly-ranked areas comprised temporal and parietal association cortices and sensory and motor cortices structures. Interestingly, most of these bottom-ranking areas are in the right hemisphere. Some experimental evidence supports this finding, such as reports of increased vascular and AD burden (amyloid-β and tau deposition) in the right hemisphere, compared to the left [45]. Additionally, in one of the studies that inspired this work [4], no downstream evoked response in the right hemisphere was recorded for one patient out of six. They performed DBS of the fornix, an axonal bundle that acts as a major output and input tract for the hippocampus and the temporal lobe. The absence of a right-sided response in some subjects while indirectly stimulating several regions simultaneously, along with the recorded worsening of AD in the right hemisphere may explain the low performance of right hemisphere controllers in our work.
Most top-ranked regions were subcortical structures (e.g., pallidum, amygdala, thalamus proper, hippocampus). However, other similarly-ranked areas, such as the insula, are cortical. Current brain stimulation techniques differ in reach, design and degree of invasiveness. In therapeutic practice, either one (subcortical structures) or the other (cortical structures) are targeted [6–8]. In a recent work, noninvasive DBS of the hippocampus in living mice was achieved by Grossman et al. while applying alternating high frequency currents at slightly different frequencies over the scalp [46]. Although the pattern of currents used by Grossman et al. (sinusoidal-like) is simpler than the ones we have obtained (Fig 2), their work shows the possibility of stimulating neuronal sets at any depth by using superficial devices. As such, the most suitable regional target for each patient (either subcortical or cortical) could be reached by using a single device. In a previous study, Terney et al. introduced current stimulation by high-frequency noisy signals [47] with positive results. The temporal profiles of the signals administered in that study somehow resembles the ones we obtained, although theirs have higher frequency, amplitude and seemingly noisier components. Together, these works indicate the feasibility of our proposal in terms of designing a device that delivers tailored signals to any location in the brain. We predict an eventual merging of our theoretical approach with cutting-edge stimulation technology like the ones proposed in the referred studies.
Another issue regarding the future development of optimal nonlinear network control of AD is the possibility of the spilling of stimulation to adjacent nuclei [48,49] as we propose to target single localized regions. Nonetheless, the lack of focality of brain stimulation techniques might be an advantage for their clinical application [49]. Several of the regions from which the desired trajectory was achieved at low energetic cost in our model have physical proximity (see Fig 3B and 3D), and could be reached in a target-specific experiment [48]. We hypothesize that simultaneous stimulation of different structures would produce faster optimal control of the pathological activity. Spilling and the intentional stimulation of selected structures with different signals are other modeling possibilities we will assess in a separate work.
On the other hand, the recently-introduced adaptive deep brain stimulation (aDBS) is gaining support for replacing the conventional constant-parameters brain stimulation in the treatment of Parkinson’s [21–24]. aDBS uses the subthalamic local field potential (LFP) activity recorded directly from the DBS electrode itself as a feedback for tuning the stimulation signal in real time. The level of beta frequency band oscillations in the LFP correlates with motor impairment, in the presence or absence of therapeutic interventions [21]. A brain–computer interface system uses this biomarker to control when the stimulation is applied. Thus, aDBS is a closed-loop technology [23]. Such procedure delivers less energy to the patient (with fewer side effects) and is clinically superior to standard continuous DBS, according to the results reported in several studies [21,22]. Our framework, designed without knowledge of the existence of aDBS, aims to obtain stimulation protocols that are also assembled over the analysis of a feedback signal related to the patient’s clinical condition (see equation (S1.5)). The successful application of aDBS supports the use of closed-loop approaches for stimulation, such as the one we have introduced.
Finally, we would like to point out the pioneering nature of our work and list its methodological limitations in what follows. Further work is to be done in solving those before proceeding to demonstrate the efficacy of our approach in actual brain stimulation experiments for AD (either in animal models or human subjects). The main issue that needs to be addressed is replacing the parameters in our model with real values estimated from the analysis of a patient’s electrical activity. The selection of the dynamical model used in this work was based on its relative mathematical simplicity (it offers the possibility of assessing both linear and nonlinear cases by switching a single parameter) while still resembling broadly used electro-physiologically-inspired neural mass models [18,50]. Several techniques for estimating its parameters are available, with outstanding results emerging from the use of the innovation method based on local linearization filters [51,52].
However, the estimation of the effective connectivities [53] mediating the interaction between neuronal populations (78 × 78 values in our case) might constitute a computationally costly problem. This is why, inspired by previous approaches [10–12], we focused on the ‘structural side’ of connectivity for optimally controlling the AD’s brain. Both functional and effective connectivity correlate to structural connectivity [54,55]. In this work we assumed that the strength of the structural connectivity was proportional to a measure derived from diffusion MRI tractography–the anatomical connection densities [56]. Nevertheless, there is a consensus on the limited performance of tracking algorithms and anatomically-imposed difficulties that suggests prudence in making such assumption [57,58]. An inherent limitation of DW-MRI is its inability to detect the direction of nervous fibers [56], which extends to all current neuroimaging methods [58]. However, a substantial proportion of reciprocal connections has been identified [59], justifying the ubiquitous use of undirected anatomical networks. Additionally, variability across DW-MRI studies and methods [57], constitutes a major issue to deal with for achieving generalization. A finer partitioning of the brain into regions could also result in more localized targets for clinical stimulation.
The controllers we designed have lower magnitude–with lower associated energy deposition–than what has been identified as the safety threshold [4,5] in DBS for AD (3.0–3.5 V) and are still (computationally) successful in the reversion of pathological activity. Whereas the shape of our signals is in-between the low-frequency sinusoidal-like inputs that Grossman et al. [46] applied and the high-frequency random noise stimulation performed by Terney et al. [47] with no side effects, further research must be carried on to address safety concerns. Our controllers might have a negative outcome–as any other current delivered to the brain tissue. Few adverse effects are generally reported when weak electrical currents are administered to the scalp (e.g. through tDCS) [3]. High-frequency invasive stimulation (DBS) causes stronger side effects overall [60]. Thus, the application of our signals should be preceded by a complete in vivo assessment of their impact on the tissue.
Another of the limitations of our model in its current state is assuming that all nodes present the same time constants (S3 Table), yielding to approximately equal oscillation frequencies (~6.4 Hz for the ‘pathological state’, 8.0 Hz for the ‘healthy state’). Variability was the result of the interaction with other regions through the anatomical networks only. Apart from the strength of the nonlinearity, the existing differences in the nodes’ natural frequencies [61] can also affect the outcomes of the control tasks. To address this specific issue, we performed an exploratory analysis over one of the subjects in the database, in which each node was assigned a random natural frequency inside the theta and alpha bands. A more detailed discussion can be found in S3 Text and S3 Fig. In short, the variation did affect the ability of the controllers to revert the disease consequences, by both increasing the energetic cost and producing more tasks to fail. However, randomly generated frequencies also lack physiological meaning. As previously stated, no definitive control strategy can be delineated until the calculations are performed over models that include as much realistic information as possible, i.e., estimating the actual oscillation frequencies from the subject’s electrical activity.
Even with limitations in the modeling at this very primary stage and the need for experimental validation, the results herein reported constitute a progress, and overall, this work might represent a change to the methodology for addressing the control principles of the brain. Our future research intends to use multimodal data to overcome the above-stated imperfections. Our ultimate goal is to design controllers for efficiently and realistically reverting pathological states of each patient’s brain activity.
The study was conducted according to Good Clinical Practice guidelines, the Declaration of Helsinki Principles, US 21CFR Part 50—Protection of Human Subjects, and Part 56—Institutional Review Boards, and pursuant to state and federal HIPAA regulations (adni.loni.usc.edu). Study subjects and/or authorized representatives gave written informed consent at the time of enrollment for sample collection and completed questionnaires approved by each participating sites Institutional Review Board. The authors obtained approval from the ADNI Data Sharing and Publications Committee for data use and publication, see documents http://adni.loni.usc.edu/wp-content/uploads/how_to_apply/ADNI_Data_Use_Agreement.pdf and http://adni.loni.usc.edu/wp-content/uploads/how_to_apply/ADNI_Manuscript_Citations.pdf, respectively.
This study used 41 individual baseline data from ADNI. Structural magnetic resonance images (MRI) and diffusion weighted MRI (DW-MRI) were acquired for each of the ADNI subjects included in the study. We used the individual clinical diagnoses assigned by the ADNI experts, which were based on multiple clinical evaluations. The 41 subjects were diagnosed as Alzheimer’s patients. The ADNI was launched in 2003 as a public-private partnership, led by Principal Investigator Michael W. Weiner, MD. The primary goal of ADNI has been to test whether serial magnetic resonance imaging, positron emission tomography, other biological markers, and clinical and neuropsychological assessment can be combined to measure the progression of mild cognitive impairment and early AD.
See S1 Table for a complete list of the included ADNI subjects and S2 Table for demographic characteristics.
We choose a mathematical model that balances simplicity and physiological reliability. This consists of a set of Duffing oscillators [27], linearly coupled through the anatomical connection density matrices, W ∈ ℝN×N. Here, N is the number of nodes in the network (N = 78). The state variables x are interpreted as excitatory postsynaptic potentials in a neural mass formulation [50,67] (units: mV). In the low activity limit, the sigmoid activation function in these models can be replaced by a third-order approximation [68]. Then, in the model we use, the dynamics in area i is described by:
x˙i=yi
y˙i=−αxi−γxi3+β∑j=1NWjixj,
(1)
x(0)=x0,y(0)=y0
where z = [x,y]T ∈ ℝn (n = 2N) is the state vector and β is the strength of the coupling. The parameter γ is the strength of the nonlinearity. The limit case of a linear system can be readily achieved by making γ = 0. Additionally, the amplitude of the solution grows with the initial condition, x0 and the frequency, with α [27]. Any desired feature of the spectrum related to the system in (1) (healthy or pathological activity in AD) can be simulated by tuning these parameters. We distinguish two different solutions, z, of (1) based on the values of the parameter α (and the initial conditions, z0 = [x0, y0]T). Let us call these solutions zp, if α = αp (produces high-amplitude theta-band oscillations), and zh, if α = αh (produces low-amplitude alpha-band oscillations). Throughout this paper we use subscripts ‘p’ and ‘h’ to denote pathological and healthy states, respectively. The specific set of parameters used in this work appears in S3 Table.
The control task consists of steering the nonlinear system that results from making α = αp in (1) to the one obtained when α = αh by applying an input that enters only one node in the network. Thus, we search for stimuli that make the difference between zp and zh, e =zp − zh, as small as possible [69]. Those control signals revert the increase in the theta-band power registered in AD and steer it to a ‘healthy activity’, even though the underlying pathological system–given in the model by the parameter αp and the affected anatomical networks, W– is unchanged. There is a correlation between abnormalities in EEG spectral measures and severity of dementia; drug-induced transient restoration of EEG normality is related to improved attention and memory performances [30,32]. We include these findings as key elements for modeling optimal brain stimulation for AD.
The dynamical equations for e can be written generically, under certain conditions [26], as:
e˙(t)=A(e)e(t)+Bu(t),e(0)=e0,
(2)
where u(t) symbolizes the external (control) signal and B is a vector whose components are only different from zero at the entry corresponding to the y-variable in the region receiving the input. By changing the position of this non-zero element, we cover all possible ‘stimulations’ to single regions. New, different systems are also generated each time the connectivity matrices, W, are changed. The optimal feedback control signal u(t) is obtained from solving the so-called state-dependent Riccati equation (SDRE) [25,26,69,70]. This input exists if system (2) is observable and controllable. Observability is guaranteed with a careful though simple selection of weights in the SDRE framework [16]. The state-dependent linearization in (2) yields a controllable system in a region ℧ ∈ ℝn if the matrix [B|A(e)B|…|An−1(e)B] has rank n for every e ∈ ℧ [26,69]. See S1 Text for more details on the SDRE theory and the explicit form of matrix A.
All the simulations and analysis in this work were implemented and performed within MATLAB R2017a (The MathWorks Inc., Natick, MA, USA). Visualization of the results was partially performed by means of BrainNet Viewer [71]. The systems in (2) were iteratively solved using a Local Linearization scheme, which is known to be stable and preserves nonlinearities [72,73]. The evaluation of (2) at each iteration conduced to locally linear systems. The solution to the SDRE was obtained by using MATLAB’s ‘lqr’ function. In practice, the controllability condition is checked while the numerical integration of the system is performed. Thus, the assessment of the controllability condition was also pointwise-managed through ‘lqr’, as it returns an error for uncontrollable systems.
We studied the relationship between the mean cost of controlling the entire network from a region and the total number of controllable systems (regions from which the input propagates to the whole network and reverts the pathological activity) per subject, and local and global network topological measures, respectively. The topological quantities we considered were: node strength, eccentricity, closeness centrality, betweenness centrality, clustering coefficient, characteristic path length, radius, diameter, average clustering coefficient and global efficiency. The Brain Connectivity Toolbox [58] was used for calculating the above-mentioned indices. A related measure, the communicability [74], that accounts not only for the shortest path lengths communicating two nodes in a network, but also for indirect connections that permit information to travel, was computed as well. The quantities herein used are weighted [14,75] since the anatomical connection densities can take any value from 0 to 1. Definitions and further information can be found in S2 Text.
|
10.1371/journal.pgen.1004457 | Nucleosomes Shape DNA Polymorphism and Divergence | An estimated 80% of genomic DNA in eukaryotes is packaged as nucleosomes, which, together with the remaining interstitial linker regions, generate higher order chromatin structures [1]. Nucleosome sequences isolated from diverse organisms exhibit ∼10 bp periodic variations in AA, TT and GC dinucleotide frequencies. These sequence elements generate intrinsically curved DNA and help establish the histone-DNA interface. We investigated an important unanswered question concerning the interplay between chromatin organization and genome evolution: do the DNA sequence preferences inherent to the highly conserved histone core exert detectable natural selection on genomic divergence and polymorphism? To address this hypothesis, we isolated nucleosomal DNA sequences from Drosophila melanogaster embryos and examined the underlying genomic variation within and between species. We found that divergence along the D. melanogaster lineage is periodic across nucleosome regions with base changes following preferred nucleotides, providing new evidence for systematic evolutionary forces in the generation and maintenance of nucleosome-associated dinucleotide periodicities. Further, Single Nucleotide Polymorphism (SNP) frequency spectra show striking periodicities across nucleosomal regions, paralleling divergence patterns. Preferred alleles occur at higher frequencies in natural populations, consistent with a central role for natural selection. These patterns are stronger for nucleosomes in introns than in intergenic regions, suggesting selection is stronger in transcribed regions where nucleosomes undergo more displacement, remodeling and functional modification. In addition, we observe a large-scale (∼180 bp) periodic enrichment of AA/TT dinucleotides associated with nucleosome occupancy, while GC dinucleotide frequency peaks in linker regions. Divergence and polymorphism data also support a role for natural selection in the generation and maintenance of these super-nucleosomal patterns. Our results demonstrate that nucleosome-associated sequence periodicities are under selective pressure, implying that structural interactions between nucleosomes and DNA sequence shape sequence evolution, particularly in introns.
| In eukaryotic cells, the majority of DNA is packaged in nucleosomes comprised of ∼147 bp of DNA wound tightly around the highly conserved histone octamer. Nucleosomal DNA from diverse organisms shows an anti-correlated ∼10 bp periodicity of AT-rich and GC-rich dinucleotides. These sequence features influence DNA bending and shape, facilitating structural interactions. We asked whether natural selection mediated through the periodic sequence preferences of nucleosomes shapes the evolution of non-protein-coding regions of D. melanogaster by examining the inter- and intra-species genomic variation relative to these fundamental chromatin building blocks. The sequence changes across nucleosome-bound regions on the melanogaster lineage mirror the observed nucleosome dinucleotide periodicities. Importantly, we show that the frequencies of polymorphisms in natural populations vary across these regions, paralleling divergence, with higher frequencies of preferred alleles. These patterns are most evident for intronic regions and indicate that non-protein coding regions are evolving toward sequences that facilitate the canonical association with the histone core. This result is consistent with the hypothesis that interactions between DNA and the core have systematic impacts on function that are subject to natural selection and are not solely due to mutational bias. These ubiquitous interactions with the histone core partially account for the evolutionary constraint observed in unannotated genomic regions, and may drive broad changes in base composition.
| Sequence-dependent differences in the physical properties of DNA influence its associations with the histone core, as well as the kinetics of nucleosome assembly and stability [2]–[9]. One of the most generalizable sequence affinities of the histone octamer is the periodic variation of dinucleotide frequencies across nucleosomal DNA. Alignments of nucleosomal sequences from diverse eukaryotes display a prominent ∼10 bp periodic enrichment of AT-rich dinucleotides, along with an anti-correlated periodicity of GC-rich dinucleotides [8], [10]–[14]. The ∼10 bp spacing of AA/TT dinucleotides generates intrinsically curved DNA molecules with increased nucleosome binding affinity [5], [8], [14]–[17]. Peaks of AA/TT frequency are found specifically over positions where the minor groove bends interiorly, whereas GC dinucleotides peak where the major groove is facing the histone core. Structural data suggest that DNA shape, in particular the narrowing of the minor groove and the associated lowering of its electrostatic potential at AT-rich sequences facilitate contacts with key histone arginines [9], [18], [19]. GC dinucleotides contract the major groove, which also facilitates the tight winding of DNA around the core [9], [20].
Although these broadly conserved dinucleotide patterns have been cited as evidence for a genomic “code” for nucleosome positioning [10], the role of sequence in nucleosome function remains contested and unresolved [6], [8], [21], [22]. Correlation between in vitro and in vivo nucleosome maps in yeast may reflect the influence of the inherent sequence preferences of the histone core on nucleosome positioning [7], [23]. However, strong experimental evidence suggests that trans-acting factors (e.g. RNA polymerase II, transcription factors and ATP-dependent remodelers) are central to establishing nucleosome positions along genomic DNA (translational positions), particularly in genic regions, with sequence providing a weaker contribution [7], [8], [21], [24], [25]. In cases where DNA sequence does impact translational nucleosome positions, its influence is largely attributed to GC content and anti-nucleosomal sequences, such as poly-dA/dT tracts, rather than dinucleotide patterns [5], [8], [25]–[27].
Dinucleotides are instead thought to play a distinct but integrally connected role in directing and preserving the ‘rotational positioning’ of nucleosomal DNA [8], [9], [20], [28], which refers to the orientation of DNA relative to the core. Due to the structural constraints inherent to nucleosome formation, a given translational position in the genome will assemble with a particular rotational alignment. This determines which bases face the nucleosome interior and exterior, and also the positioning of the major and minor grooves relative to the core. Nucleosomes tend to occupy translational genomic positions which are offset by ∼10 bp increments [13], [28], [29]. Thus, due to the helical structure of DNA, with ∼10.4 bp per turn, the rotational orientation of DNA relative to the core is thought to be unchanged as nucleosomes assume new favored translational positions (Figure 1A). This 10 bp incremental movement leaves the exposure of sites at the surface unchanged [20], [28], and is in agreement with the reported step size of many chromatin remodelers [30]. By influencing the rotational positioning of DNA relative to the histone core, nucleotide changes at particular nucleosome positions (or in flanking regions) could have diverse functional impact, for example on nucleosome assembly, stability, remodeling efficiency, RNA and DNA polymerase processivities and transcription factor binding site access. However, despite considerable evidence that dinucleotide patterns impact nucleosome positioning and dynamics in vitro, in vivo evidence of function has remained elusive.
One approach to discovering function is to look for evidence of natural selection in sequence polymorphism (variation within species) and divergence (variation between species). Individual mutations influencing histone-DNA interactions may have only slight, undetectable phenotypic effects in the laboratory; in contrast, the associated fitness consequences in large natural populations can strongly shape rates of divergence and levels of polymorphism over many generations [31]. Of course, strongly selected variants will go to fixation quickly and be maintained in very high frequency against the weaker force of mutational reversion. However, observations of extensive DNA sequence polymorphism and divergence throughout the genome, including nucleosomal sequences, indicate that such systematic selection is not dominating stochastic effects (mutation, genetic drift, and variation in selection coefficients) in the evolutionary dynamic.
Analysis of codon bias suggests that at equilibrium between selection, mutation and genetic drift, the ratio of the frequencies of two alternative synonymous codons throughout a single genome can be used to estimate the direction and magnitude of selection [32]. The action of natural selection can be inferred when synonymous codon pairs exhibit a strong “bias” towards one state relative to the other. This analysis extends to the distribution of polymorphic allele frequencies in genomes sampled from natural populations [33]–[35]. A similar approach can be applied to alternative nucleotides at particular positions within nucleosomal sequences. As the magnitude of selection increases, the expected frequency of preferred alleles increases. Consequently, the distribution of SNP frequencies (or “site frequency spectrum”) at a given nucleosome position (analogous to a synonymous SNP) is expected to shift towards relatively higher frequencies of preferred alleles.
If the observed dinucleotide patterns reflect selectively favored states, ancestrally unpreferred base pairs across nucleosomes should diverge towards the “preferred” state along species lineages. Further, if the “preferred” divergence patterns reflect the average impact of natural selection, then frequencies of polymorphisms in natural populations should be more skewed at sites experiencing stronger selection. “Unpreferred” variants, specifically substitutions or polymorphisms away from favored nucleotides, such as substitution of an ancestral A with a G at nucleosome positions which are systematically enriched for AA dinucleotides, should diverge more slowly and be rarer when polymorphic in the population. In contrast, “preferred” variants, such as substitution of an ancestral A with a G at positions of enriched for the GC dinucleotide, should diverge more rapidly and be more common when polymorphic.
At the lower resolution of an entire nucleosome and its nearby flanking regions, both divergence and polymorphism are observed to vary [36]–[39], but evidence of a role for natural selection in the underlying evolutionary dynamics remains sparse [40]–[43]. Studies of human SNPs [37], [38] and divergence in humans, yeast and medaka [36], [38], [39], [43] show that both expected heterozygosity and divergence between species are elevated near the central dyad and depressed in the adjacent linker regions, though these patterns appear to differ by substitutional pathway [38]. One possibility is that patterns of variation relative to nucleosomes derive from nucleosome-specific mutational biases. This could result from suppression of mutation by a protective aspect of nucleosome occupancy [44], or it could arise from an interaction between the histone core and DNA damage recognition or repair mechanisms [45]–[48]. Of course, natural selection mediated via DNA:nucleosome interactions may also strongly reshape the patterns of SNP variation and divergence between taxa [40]–[43]. Analysis of the site frequency spectrum promises to distinguish between these two alternatives.
The whole-nucleosome-resolution analyses considered above cannot leverage the specific structural predictions of dinucleotide interactions with the core and their strong mechanistic implications. Examination of polymorphism and divergence at each base pair position across the nucleosomal DNA opens a rich and precise view, as well as powerful tests of alternative mechanisms such as biased mutation and natural selection. We report the discovery of fine-scale periodicities in inter- or intra-species sequence variation relative to nucleosomes and discuss their implications for the role of natural selection mediated through nucleosome function. Our analysis of DNA sequence polymorphism and divergence across isolated nucleosomal fragments from D. melanogaster embryos reveals that nucleosomal sequences are diverging towards “preferred” nucleotides. Regions where the minor groove is interior are becoming more AT-rich, and regions where the major groove is interior are becoming more GC-rich along the melanogaster lineage. Using a new index for quantitating the frequency spectrum (Δπ), we identify clear signals associated with natural selection, which parallel the observed periodicities in divergence. This selection is strongest in intronic regions, where nucleosome assembly and positioning are expected to have greater functional impacts. These findings support the hypothesis that the widely observed sequence affinities of the core octamer have functional consequences that are subject to natural selection. Given the dominant role of nucleosomes in the packaging of the genome and their conserved sequence preferences, their interactions may broadly shape the sequence of melanogaster and other genomes.
To investigate the impact of nucleosomes on DNA sequence variation, we isolated nuclei from D. melanogaster embryos, performed Micrococcal nuclease (MNase) digestion, and used paired-end sequencing to position fragments on the genome (Figure S1). Previous studies in Drosophila identified a range of periodic dinucleotides in association with nucleosomes [10], [11]. Our collections of 276,614 intergenic and 270,998 intronic autosomal 147 bp nucleosomal fragments (hereafter n147, Tables S1 and S2) cover 68.5% of the unique intronic and intergenic euchromatic autosomal genome and display a ∼10 bp periodicity for many dinucleotide frequencies (Figures 1 and S2). In these and subsequent analyses, the 5′-3′ sequence from bases −73 to −1 were joined to the reverse complement of bases 1 to 73, to reflect the dyad symmetry of the nucleosome (see Materials and Methods). AA, TT and GC showed the strongest periodicity of WW and SS (where W = A|T, S = G|C) dinucleotide pairs, respectively (Figure 1B). These same dinucleotides show a distinct overrepresentation in the non-coding regions of the genome as a whole (Figure 1C). As noted in previous studies, AA and TT are similarly periodic and occur where the minor groove is interior (at superhelix locations, SHL, ±(i+0.5); where i is 0, 1,…6). However, noticeable differences between the distributions are apparent. For example, the frequency of TT displays a distinctly smaller peak at ∼SHL 4.5, and AA frequency displays a stronger drop at ∼SHL 2 (Figure 1B). GC frequency across n147 regions is anti-correlated with AA/TT and is characterized by a prominent upward concavity (Figure 1B). These dinucleotide periodicities extend well beyond n147 edges into linker regions, consistent with the proposed translational step size of 10 bp.
Upon examination of the dinucleotide frequencies flanking aligned n147 regions, we discovered an additional large-scale pattern in AA/TT and GC dinucleotide frequencies (Figure 1D). This ∼180 bp periodic variation in frequency tracks with overall nucleosome “occupancy” in the regions flanking the n147. Average AA/TT frequencies (Figure 1D) and overall A/T frequencies (Figure S3) are higher in regions of greater nucleosome “occupancy” and lower in putative “linker” regions. Thus, the AA/TT sequence features that facilitate nucleosome formation are enriched over regions with higher nucleosome “occupancy.” Conversely, GC frequency (and overall G/C frequencies, Figure S3) peaks at the periphery of more nucleosome-dense regions and in “linker” regions. These surprising “super-nucleosomal” periodicities extend the observed n147 patterns to flanking multi-nucleosomal arrays, and suggest a contribution of sequence to translational positioning. Consistent with chemical mapping of nucleosomes, this result suggests that the observed experimental correlation between MNase nucleosome “occupancy” and GC content [1], [8], [22], [26], [49], [50] reflects differential recovery, rather than positional preference [23], [51].
If variations in dinucleotide frequencies relative to nucleosomes result from accumulated sequence divergence, we expect substitution patterns to parallel the observed base preferences. However, the timescale(s) at which these patterns evolve is unknown. Lineage-specific or “polarized” divergence is the proportion of nucleotide sites that are different in melanogaster while identical in its sister taxa simulans (most recent common ancestor 2.5 MYA) and the proximate outgroup (yakuba or erecta; 6–7 MYA, see Materials and Methods). Overall genomic divergence on the melanogaster lineage shows a marked excess of G→A (inferred ancestral G, derived A in melanogaster) and C→T (ancestral C, derived T) substitutions compared to A→G and T→C (Figure 2A). This is in agreement with earlier estimates of divergence on the melanogaster lineage [52], [53] and with the observed two-fold greater mutation rate [54]. We next considered the average divergence at each site across n147 regions, normalized for underlying base frequencies. This analysis revealed a striking ∼10 bp periodicity in transitions (GC→AT and AT→GC) for two estimates of divergence; per-n147 in Figure 2B is weighted by the redundancy in the n147 set, while per-site in Figure S4 weights each site equally. Rates for GC→AT and AT→GC are anti-correlated and track with underlying dinucleotide frequencies. Thus, ancestral GC bases are more likely to become AT in nucleosomal regions where AA/TT dinucleotides are in higher frequency, and AT bases are more likely to become G or C at sites where GC is enriched. GC→AT divergence also shows a marked curvature, with a peak at the dyad axis.
Given the substantial variation in individual substitution rates, we next examined specific pathways to determine their relative contributions. Of all pathways, G→A, C→T, A→G and T→C exhibited the most obvious periodicities in divergence (Figure 2C; see Discussion). In some cases, divergence patterns reflect the subtleties observed for dinucleotide frequency patterns. For example, C→T rates are less peaked at SHL 4.5, the location of the lowest peak in TT frequency (Figure 2C). C→T also shows a greater difference in rates between the n147 periphery and linker regions (compared to G→A). Interestingly, for each ancestral base, the periodicities of substitutions that do not change GC content, appear weaker, perhaps due to both scaling and weaker signal to noise (Figure 2C). A subset of non-overlapping n147 regions showed similar patterns (Figure S5A).
When mapped onto the DNA from the nucleosome structure [55], peaks of intergenic G→A divergence clearly occur within regions where the minor groove is interior and in contact with key arginines of the histone core (Figure 2D). Note also the higher G→A divergence toward the central axis, as reflected by the downward concavity in Figure 2C. This is consistent with analyses of the impact of sequence variation on nucleosome structure, which identified this central region of H3/H4 interactions as most constrained [56]. Conversely, A→G substitution rates are highest in regions where the major groove is interior (Figure S5B). This pattern is consistent with established SS dinucleotide patterns [8], [10]–[14] and the observation that GC rich sequences are disfavored for minor groove compression and favor narrowing of the major groove [9], [18].
Divergence patterns should also reflect the observed nucleosome-scale periodicities in base and dinucleotide frequencies (Figures 1D and S3). To increase signal, we combined complementary substitutions, G→A and C→T (G→A:C→T) and A→G and T→C (A→G:T→C). Aligned n147 regions show substantially lower divergence rates than their immediate flanking sequences (Figure 2E). Rates drop to the local background within ∼500 bp, following the skew of AA/TT dinucleotides (and overall AT content; Figures 1D and S3). In spite of this local variation in rates, due at least in part to MNase preferences (Figure S6), we observe a large-scale (180 bp) periodicity in G→A:C→T divergence surrounding intergenic n147 nucleosomal regions (Figure 2E). Introns showed a similar but weaker pattern, potentially due to the influence of flanking coding regions (Figure 2E). Any periodicity of the A→G:T→C divergence in flanking regions is less obvious (Figure S5C), at least partially due to a 50% lower rate of divergence and thus inherently weaker signal.
These large-scale patterns allow us to resolve general trends in divergence relative to nucleosome occupancy. We find that, on average, G→A:C→T changes along the melanogaster lineage are fixed at higher rates across nucleosomes relative to linkers, mirroring underlying AA/TT dinucleotide frequencies. This is in apparent contrast to the report that the cytosine deamination mutational pathway (a major source of G→A:C→T transitions) and associated divergence is suppressed by nucleosome occupancy [44]. To clarify this discrepancy, we examined the interactions between divergence and “occupancy” of the n147 fragments, as estimated by depth of coverage by 142–152-bp nucleosomal fragments, n142-152. Indeed, we observe a negative correlation between this metric and all substitutional pathways (Table S3, Figure S7A). However, we note that n142-152 coverage is correlated with GC content of the n147 region (Figure S8D), as previously reported in other studies [1], [5], [8], [26], [49], and that correlations between nucleosome fragment GC and divergence are even more striking (Table S3, Figure S7B, S8C). This is also true for 500 bp intergenic windows, independent of nucleosome coverage (Table S3). When we parse n147 by n142-152 “occupancy,” we observe differences in AA/TT frequency, G→A:C→T divergence, and nucleosome phasing in flanking regions (Figure S8A). The periodicities of these features are most obvious surrounding highly occupied (GC-rich) n147 regions, but they do not appear to be unique to them. Thus, we conclude that nucleosome bound regions in D. melanogaster embryos are generally more AT-rich and have higher rates of G→A:C→T substitution than their adjacent “linker” regions, inconsistent with the fundamental claim in Chen, et al. [44] (mentioned above).
The divergence patterns we observe are consistent with known nucleosomal dinucleotide preferences [5], [8], [10]–[17]. This is analogous to observations for codons, where substitutions mirror genome-wide codon usage biases and are attributed to natural selection for preferred codons [34], [57], [58]. However, divergence patterns alone cannot exclude the hypothesis that substitutional patterns result from biased mutation relative to nucleosomes. Mutation rates may vary across nucleosome-bound regions and could lead to compositional variation and different rates of divergence. Nevertheless, once a new selectively neutral allele arises, its dynamics and thus its distribution of frequencies are independent of type (or rate) of mutation [59], [60]. While natural selection influences the probability of fixation (thus the rate of divergence), mild differences in fitness will also shift the site frequency spectra of polymorphic alleles [61]–[63]. Neutral and deleterious mutations tend to spend much of their typically short lives as rare alleles, while weakly favored alleles will be found at higher frequencies as many more drift towards fixation. Although the impacts of varying demographic histories [64] and of linked selection [48], [49] can lead to distributions of selectively neutral polymorphisms that mimic particular forms of selection, they should do so randomly across the genome and not show a positional relationship within nucleosomal sequences.
The hypothesis that nucleosome structure and function impose natural selection on genomic sequence variation predicts periodicities in the frequency spectra. Indeed, the average per-n147 frequencies of G-A and C-T SNPs in a sample of 36 D. melanogaster genomes from Raleigh (North Carolina) exhibit nucleosomal patterns paralleling those observed for dinucleotides and polarized divergence (Figures 3A and S9). Frequencies of A alleles at G-A SNPs show clear periodicity across intergenic and intronic n147 regions, extending into linker regions (Figure 3A). A alleles are relatively more common in SHL ±(i+0.5) regions, and G alleles are higher in regions where the major groove faces the histone core. Removal of singleton SNPs (cases where either allele is observed only once), which can mitigate the impact of possible sequencing errors, raises average A frequencies but does not eliminate the periodicity (Figure S9). Partitioning such SNPs by ancestral state can remove the impact of average mutation rate differences and reveal differences in the patterns of selection. Nucleosomal patterns of the average per-n147 frequencies of derived SNPs, such as G→GA (ancestral G and a derived, polymorphic A), exhibit clear periodicities that generally parallel divergence and nucleosomal dinucleotide frequencies (Figures S10, intergenic, and S11, intronic).
To systematically assess the periodicity in the frequency spectra we calculated a new index, Δπ, (closely related to Tajima's D [59]) across n147 regions for the Raleigh sample [65]. Where p is the frequency of a SNP in the sample and is the estimate of the heterozygosity, we define Δπ as the average (per SNP) deviation in from expectation under equilibrium between genetic drift and mutation to selectively equivalent alleles (see Materials and Methods). The “folding” of the frequency spectrum such that p is equivalent to (1−p) mitigates the impacts of errors in the inference of the ancestral state [66] and emphasizes variation in the midrange of p. Weak positive selection is predicted to skew toward higher values (more positive Δπ), while weak negative selection leads to more negative Δπ. Thus, systematic differences in selective forces at different positions across n147 regions should yield a pattern in Δπ that parallels that observed for divergence. These patterns of Δπ are superimposed on the observed genome-wide average negative skew [65] (Table S4) that can be attributed to strongly deleterious mutations [67], [68], varying demographic history [59], [64] or linked selection (background selection [69] and hitchhiking [70]).
Indeed, when we examined average Δπ for G→GA polymorphisms, we discovered a clear ∼10 bp periodic skew in frequency across nucleosomal regions, mirroring G→A divergence (per-n147 in Figure 3B and per-site in Figure S12). n147 G→GA Δπ are less negative in regions of higher AA dinucleotide frequency. Interestingly, intronic G→GA Δπ shows even more pronounced periodicity in the frequency spectrum, including the prominent drop at SHL 2 observed for AA frequency (Figures 3B and S12). Δπ for C→CT polymorphisms is also periodic in introns (both per-n147 and per-site), with peaks aligning with regions of high TT frequency; while intergenic n147 share a subset of these peaks (Figures 3B and S12), the overall patterns show much weaker periodicity (see below). Although peaks in intronic C→CT Δπ overlap roughly with those for G→GA sites, they show a more convex shape, similar to the C→T divergence. Substitutions in the complementary directions (e.g. A→AG) also show a periodic skew in allele frequencies. Introns display a striking periodicity in A→AG Δπ aligned with GC frequency (Figure 3B, while intergenic n147 A→AG sites show only two peripheral Δπ peaks and several peaks (valleys) that are discordant with the GC dinucleotide periodicity. Like underlying GC frequency, intronic A→AG Δπ has a concave upward shape. We observe weaker but interesting indications of continued periodicity in linker regions, consistent with selection for the preservation of rotational positioning in association with translational repositioning. The patterns of Δπ for 5 non-overlapping subsets of n147 regions were similar (Figure S13). We conclude that for several substitutional pathways there is strong evidence of selection maintaining the observed nucleosomal (di)nucleotide preferences.
The periodicity of nucleosomal Δπ is not limited to the Raleigh population. The strongest of these periodic patterns in Δπ are also apparent in a smaller, independent set of 21 sequenced genomes from a Rwandan (Africa) population [71] (Figures 3C and S14), which also exhibits more negative average Δπ values. This African sample is assumed to represent a larger, more stable population from the center of the species distribution, while the Raleigh sample represents the serial diasporas out-of-Africa and into North America. Notwithstanding differences in average Δπ, these strong and predicted periodicities in nucleosomal in Δπ support our hypothesis that direct interactions between the histone core and DNA sequence polymorphisms yield functional effects with fitness consequences.
An alternative hypothesis to explain these periodicities holds that the sequences evolve independently of natural selection and that the in vivo positions of our isolated nucleosomes reflect the innate preferred rotational positions of the particular genome used. Derived SNPs detected in a single strain are likely to be in high frequency, and thus we might observe periodicity in the frequency spectra at such SNPs in the absence of natural selection. To test for the impact of this hypothesized ascertainment bias on the periodicity of Δπ, we filtered the n147 for those in which the source genome bore the ancestral alleles. Despite the unavoidable thinning of the data, we observed clearly periodic polarized Δπ for those pathways with the strongest initial signals, e.g. intronic G→GA, C→CT and A→AG (Figure S15). These results indicate that the observed periodicities in the frequencies of preferred bases (parallel to the dinucleotide frequencies and the divergence) cannot be attributed to biases in the ascertainment associated with the genotype from which the nucleosomal sequences were prepared.
We next considered the values of Δπ surrounding n147 regions. The observed skew in intergenic G→GA:C→CT Δπ extends into adjacent sequence (Figure 3D), tracking with the periodicity of G→A:C→T divergence. Interestingly, in the ∼500 bp flanking n147 regions, there appear to be major and minor Δπ peaks associated with each divergence peak. Given the shoulder of C→CT Δπ values in linker regions adjacent to n147 (Figure 3A), this could represent a nucleosomal and a linker peak. Intronic regions show higher overall values of G→GA:C→CT Δπ and similar, but weaker, indications of increased G→GA:C→CT Δπ associated with nucleosome occupancy (Figure 3C). Among other interesting patterns in Δπ and contrasts to divergence in these flanking regions are those associated with the complementary set of substitutional paths, A→AG:T→TC, which exhibits peaks over apparent linker regions in Δπ but no parallel pattern in A→G:T→C divergence (Figure S16).
On average (per-n147), G→GA and C→CT are the most common polymorphisms and have among the most positive Δπ, indicating weak positive selection, in addition to being the most rapidly diverging bases (Figure 3D). Although rates of A→G and T→C divergence (and rates of associated polymorphisms) are much lower, these types of polymorphic sites also have high average Δπ (Figure 3E). Thus, substitutions with the most periodic divergence and Δπ also show the least overall negative skew in the frequency spectrum. Relative relationships of n147 average π, Δπ and divergence are quite similar to those of a non-overlapping subset and to the genome-wide averages (Table S4 and Figure S17). These broad genomic patterns appear inconsistent with equilibrium models and may reflect heterogeneity and/or recent (transient) shifts in selective forces [35], [52], [72].
Histones are among the most ubiquitous and highly conserved eukaryotic proteins. Thus, it is not surprising that nucleosomal dinucleotide periodicities, which derive from key structural interactions between DNA sequence and the histone core, are shared widely across species. In spite of the near universality of these patterns among eukaryotes and decades of research, our understanding of their functional impact and evolutionary dynamics remain unsettled. In this work we examined genomic variation across regions defined by isolated nucleosomal DNA fragments. Our goal was to first determine if these regions showed interpretable variation in divergence between species, then to analyze population genomic variation for evidence of a role for natural selection in the generation and maintenance of nucleosome-associated sequence variation.
We find that divergence on the melanogaster lineage mirrors the sequence preferences of the histone core. This periodic variation in substitution rates across nucleosomal regions indicates that interior minor groove regions display more rapid substitution of AT for GC, and that AT base pairs in regions where the major groove faces inward are more likely to become GC rich. These striking patterns align directly with dinucleotide patterns that stabilize associations between DNA and the histone core, as documented in numerous biochemical and structural studies [5], [8], [15], [18], [19], [56]. If nucleosome-bound regions are evolving toward the observed nucleosome sequence preferences, a key question is whether this is the result of mutational bias relative to the positioning of chromatin proteins, or whether it is the consequence of natural selection based on functional differences. The available depth of population data and our new index Δπ allowed us to directly address this question. We find remarkable periodicities in Δπ that parallel the observed patterns of divergence. The spectra of SNP frequencies across n147 regions are variable, with higher Δπ when the inferred ancestral allele is unpreferred, and the derived allele is structurally favored. Therefore, we conclude that selection is, at least in part, driving the maintenance of nucleosome-associated sequence patterns on the melanogaster lineage.
If the fitness differences associated with such histone:DNA interactions are largely arising from nucleosomal dynamics (assembly, disassembly, movement and modification) and rotational positioning of functional elements, then we can further hypothesize that transcribed (intronic) nucleosomal sequences should exhibit stronger periodicity than untranscribed (intergenic) nucleosomal sequences. Consistent with this hypothesis, correlations of lineage specific divergence and Δπ with the relevant underlying dinucleotide frequencies are stronger for intronic sequences. Table 1 shows that in each case where a large difference between intergenic and intronic is apparent, it is the intronic that is larger. The two exceptions, G→A & AA and A→G & GC, are those where both correlations are among the highest. As might be expected given the longer timescale and greater number of variable sites, divergence correlates more strongly with dinucleotide frequencies than Δπ. Interestingly, Figures 2C and 3B show that these intronic vs. intergenic differences in correlation of divergence and Δπ with dinucleotide patterns may be attributable to large deviations from expectation in specific regions of the nucleosomal sequences, while other regions follow the expected periodic patterns.
While natural selection is the most direct interpretation of these results, interactions between chromatin proteins and DNA damage and repair are well documented [44]–[48], [73]–[76]. Contextually biased mutation (substitution) pathways could underlie the observed periodicities in nucleosomal divergence. However, Drosophila does not have a significant level of 5-methylcytosine [77], the deamination of which is thought to drive the strong contextual biases (NpCpG) in vertebrates [78]. Indeed, a recent genomic sequencing study of Drosophila mutation accumulation lines yielded no evidence for contextual biases [54]. Most importantly, such sequence-contextual as well as nucleosome-mediated biases in mutation rates are excluded as an explanation for the observed periodicities in the skew of the SNP frequency spectrum (Δπ), since strictly neutral mutations should display the same frequency distribution across the genome [31], [33], [59]. Support for a role of natural selection maintaining these periodicities is bolstered by the stronger periodicities in intronic nucleosomal sequences, where transcription-associated remodeling and disruption of nucleosome-DNA interactions are more likely to have functional impacts.
There is, however, one potential “selectively neutral” mechanism to explain the observed periodic patterns in Δπ. Biased gene conversion (BGC), a process where heteroduplex regions formed between homologs are repaired in a direction favoring one base, can create SNP dynamics analogous to those of directional selection [79]. BGC systematically favoring GC over AT has been observed in a few species and indirectly implicated in others by associations of local GC content with estimated rates of crossing over [80]. However, evidence for such an association is not observed in Drosophila [81]. Given that the magnitudes of average Δπ and its periodicities for G→GA and C→CT SNPs are comparable to those of A→AG, any explanation of our results invoking BGC would have to involve multiple distinct gene conversional biases that depend on nucleosome position. While this is conceivable and worthy of further investigation, we conclude that the canonical GC-biased gene conversion is not a significant component of the evolutionary dynamics leading to these intricate nucleosomal patterns of polymorphism and divergence.
Whether these periodic patterns are the product of natural selection or BGC, the magnitude of the average force shaping the dynamics of nucleosomal SNPs must be small compared to that affecting the evolution of nonsynonymous variants. The shifts in G→A divergence between peaks and valleys in the n147 are ∼0.001 against a background average of ∼0.01, suggesting relatively weak constraint of 1 in 10 mutations. The non-synonymous rate of divergence on the melanogaster lineage, ∼0.006, is about one tenth of that for synonymous divergence corresponding to 9 out of 10 mutations being selected against [61]. Comparable conclusions could be drawn from the modest magnitudes of periodic fluctuation in expected heterozygosities and, indeed, in the widely observed periodicities in dinucleotide frequencies of nucleosomal sequences. Still, by virtue of its four-fold greater genomic footprint, the net selective impact of just the selection associated with such nucleosomal periodicities could approach the magnitude of non-synonymous variants. As is the case for coding sequence, differences in the relative (average) rates reflect the aggregate impact of selection that must vary substantially among nucleosomes, as well as among sites.
Evidence of natural selection supporting nucleosome-associated sequence periodicities and the implication of their biological impact casts the potential functions of non-protein-coding regions in a new light. Substantial portions of Drosophila, human and other genomes appear to be under evolutionary constraint, yet lack any functional annotation [67], [68]. Further, SNPs identified by genome-wide association studies (GWASs) of interesting human phenotypes often have mild attributable effects and map to unannotated intronic or intergenic regions, where mechanistic hypotheses concerning the impacts of such genomic variation are lacking. We demonstrate that at least part of the constraint in Drosophila arises from interactions between histone proteins and DNA sequence.
Our results suggest dinucleotide periodicities and the rotational positioning that they guide have significant biological consequences. Sequences affecting rotational positioning can influence the binding of transcriptional activators and participate in regulation of expression or gene splicing [25], [28], [82]–[84]. More generally, they impact nucleosome assembly and stability [2]–[7], [9], [17], properties that broadly impact chromatin dynamics and may influence higher order chromatin structures. Further, the observed large-scale periodicities in dinucleotide frequencies (and divergence and Δπ patterns supporting them) demonstrate that sequences that facilitate rotational positioning are specifically enriched relative to adjacent nucleosomes. So, while periodic sequence patterns are considered more relevant to rotational positioning, they clearly interact with the translational positioning of arrayed nucleosomes in Drosophila. Going forward, deeper and more detailed population genomic analyses should provide a unique window into the complex in vivo interactions between DNA sequence and nucleosome function.
The significance of these periodic patterns of polymorphism and divergence is amplified in light of the substantial proportion of the eukaryotic genome packaged in nucleosomes (four-fold greater than that of coding sequence in Drosophila) and the broad conservation of dinucleotide interactions with the histone core. Indeed, no other DNA-protein interaction remotely approaches the genomic density or structural impact of nucleosomes. The striking periodic variation we observe relative to nucleosomes fundamentally changes expectations about divergence and SNP frequency, particularly in non-protein-coding regions. Our results point to a layer of evolutionary forces across entire genomes, emanating from the interactions of DNA sequence variation with the structure and function of the histone core.
Embryos were collected from population cages [85] over a 1 hr period and aged at 25°C for 2–3 hr. Staged embryos were dechorionated in 50% bleach for 2 minutes, washed extensively, and then homogenized on ice in SEC buffer (10 mM HEPES, 150 mM NaCl, 10 mM EDTA 10% glycerol, 1 mM DTT) with Protease Inhibitors (PI) (0.1 mM PMSF and 2X Roche EDTA-free Protease Inhibitor tablets). After lysate filtering and centrifugation, pelleted nuclei were resuspended in CIB (15 mM Tris pH 7.5, 60 mM KCl, 15 mM NaCl. 0.34 M Sucrose, 0.15 mM Spermine, 0.5 mM Spermidine)+PI and then repelleted. Centrifugation of nuclei in CIB was repeated 3 times, and the resultant pellet was flash frozen and stored at −80°C.
Pelleted frozen nuclei were resuspended in CIB+PI, and chromatin was digested with 0.5 U/ml Micrococcal nuclease (Sigma) for 37°C for 15 min. MNase treated nuclei were pelleted, resuspended in 0.1% NP-40 PBS+PI, and incubated at 4°C for 3 hrs to release (primarily) mononucleosomes. Nuclei were re-pelleted, and chromatin from the supernatant was phenol-chloroform extracted. Digestion was analyzed on an agarose gel.
Approximately 882 ng of DNA was used as starting material for paired-end sequencing library construction following the Illumina protocol (PE-102-1001). 10 µl of paired end adapter oligos were ligated to the end-repaired, A-tailed fragments in a 50 µl reaction. The adapter-ligation product was gel-purified to select molecules approximately 150–700 bp in length and re-suspended in 30 µl total volume. 1 µl of size-selected ligation product was used as template for 12 cycles of library enrichment PCR in a 50 µl reaction volume. The enriched library was purified using QIAGEN MinElute columns and sequenced (2×36 cycles) on one lane of a flow cell (FC42JB8) with an Illumina GAIIx running the Illumina software SCS v2.4.135/Pipeline v1.4.0.
Subsequently, 8 µl of the same size-selected ligation product was used as a template for 10 cycles of library enrichment PCR in a 100 µl reaction volume. To enrich for 147 bp fragments, the library was purified using QIAGEN MinElute columns, then size selected on an agarose gel to recover fragments approximately 273 bp in length (as determined by an Agilent Bioanalyzer). This size-selected library was sequenced (2×36 cycles) on four lanes of a flow cell (FC61BGN) with an Illumina GAIIx running the Illumina software SCS v2.5.38/Pipeline v1.5.0.
Reads that passed the Illumina pipeline's quality filters were then aligned to the Berkeley Drosophila Genome Project's Release 5 reference sequence [86] using Version 0.7.0 of the MAQ program [85]. Read pairs that mapped more than 1,000 bp apart and those for which the combined sum of the quality scores of mismatches exceeded 300 were filtered using the maq map –a 1000 -e 300 command. Otherwise, default map parameters were used.
Mapped paired end clones of length 147 bp were then filtered based on Release 5.16 FlyBase Annotation of the D. melanogaster genome and classified as intronic or intergenic. For classification, all bases, including the flanking ±50 bp, were required to map entirely to a contiguous intronic or intergenic region. Heterochromatic reads were removed using cytogenomically-defined boundaries [87]. Downstream analysis was carried out on 276,614 intergenic and 270,998 intronic autosomal 147 bp nucleosomal fragments, referred to as intronic and intergenic n147. These cover 61% and 79% respectively of the target intergenic and intronic regions in the euchromatic autosomes (chr2 and chr3). The coordinates of the n147±50 bp flanking regions are in Tables S1 and S2. Average nucleosomal read depths, where represented, are a pileup representation of a set of similarly processed paired end clones with lengths ranging from 142–152 bp.
In calculating dinucleotide frequencies, divergence, π and Δπ across n147 regions (and, where relevant, ±50 bp flanking), both positional and average calculations took the dyad symmetry into account. Substitutional pathways were switched at the dyad axis, such that positions −73 to −1 were joined to the reverse complement of bases 1–73. Where included, flanking regions (±50 bp) were treated similarly. For larger scale positional divergence and genomic averages, data from complementary substitutional pathways were combined.
The n147 dinucleotide frequencies are the averages (in the reference sequence) over all n147 fragments for each position. Genomic over/underrepresentations of dinucleotides were calculated by dividing the difference between observed and expected frequencies by the expected frequency. Estimates of expected intergenic and intronic dinucleotide frequencies were calculated based on underlying base frequencies. Observed frequencies were computed directly from the reference sequence.
The sequences of the euchromatic portions of 36 D. melanogaster genomes from Raleigh, North Carolina were released by DPGP (http://www.dpgp.org/1K_50genomes.html - Reference_Release_1.0). The sequencing, alignment and assignment of estimated quality scores are described in Langley. et al., 2012 [65]. The sequences of the 22 D. melanogaster genomes from Rwanda, Africa were released by DPGP (http://www.dpgp.org/dpgp2/update_20Jan2012/dpgp2_v2_rg.ID5.nohets.fastq.bz2). The sequencing, alignment and assignment of estimated quality scores are described in Pool et al., 2012 [71]. For both data sets, only bases with a minimum quality score of Q30 or greater were included in the analyses.
Calculations of divergence on the melanogaster lineage, frequency, expected heterozygosity (π) and the index of skew in the frequency spectrum (Δπ) were based solely on sites that could be polarized using a multiple alignment of D. melanogaster, D. simulans, D. yakuba and D. erecta genomes [65], i.e., simulans and yakuba and/or erecta have the identical base. For consideration of the potential impact of ascertainment bias association with the isolation of nucleosomes, sites were subjected to a more stringent polarization (simulans, yakuba and erecta have the identical base) and calculations were done only on sites where the experimental genome had the ancestral allele. These statistics (divergence, π and Δπ) were estimated with two alternative weightings. The first, per-n147, gave weight to genomic sites proportional to their occurrence at nucleosome positions among the n147. Thus, in instances where a particular site was found multiple times in the n147 set, the divergences or SNPs at that site were give proportionally more weight. This in effect weights the signal by nucleosome site “occupancy.” The second, per-site, counted normalized the weighting such that each site (conserved, divergence or SNP) contributed equally, independent of its recurrence in the n147. The per-n147 estimates reflect those nucleosomal sequences that were readily isolated, while the per-site method treats each population genomic variant equally.
A third representation of the mapping of divergence and Δπ consists of random non-overlapping regions sampled from the n147. These analyses are presented to simply address whether the periodicities arise solely from a ∼10 bp periodicity in the overlap of n147 fragments. For divergence, non-overlapping n147 sets (72,710 intronic and 72,859 intergenic sequences) were generated by random sampling of intergenic and intronic n147 without replacement. Newly drawn sequences were added to the non-overlapping subset only if they did not share any positions with prior sampled sequences. These non-overlapping subsets together cover 53% of the target intergenic and intronic regions in the euchromatic autosomes (chr2 and chr3) covered by the full n147. For non-overlapping Δπ, intergenic and intronic n147 (not including the flanking ±50 bp) were first filtered for only those nucleosomal regions containing the relevant SNP (taking dyad symmetry into account). This produced non-overlapping intergenic and intronic sets of ∼90,000 regions each for G→GA and C→CT and ∼55,000 regions for A→AG. These sets were then subjected to random sampling without replacement. A new n147 was added to the set if it did not overlap any already in the set. Non-overlapping G→GA and C→CT intronic and intergenic sets contained ∼46,000 regions each and A→AG sets contained ∼30,000.
All three of these methods for calculating divergence and Δπ yielded similarly periodic patterns reflecting the fact that while the genomic coverage of the n147 is not deep, it is also relatively uniform (cv 0.65) and the periodicities are not arising from a small subset of the n147 or interactions from overlapping n147 sequences.
We require a sample-based index of the skew in the site frequency spectrum. Tajima [59] proffered the test statistic, D, a normalization of d, the difference between two estimates of the same population parameter, 4Nμ, where μ is the mutation rate to selectively neutral alleles and N is the population size. These estimates areandin a sample of size n, where xi is the frequency of one of the two alleles at each of the S segregating sites in an arbitrary genomic segment of length L base pairs. Thus (expected heterozygosity) and are per site (base pair) estimates of 4Nμ.
But here we seek not a test statistic for a genomic segment, but an index of the same deviation, that can be aggregated across heterogeneously sampled data and compared across classes of genomic annotation.
To that end considera simple rescaling of Tajima's (little) d,Tajima [59] also presents the distribution of the proportion of S segregating sites with frequency i/n in the sample, Gn(i)/S. Gn(i) can not be used to compute properties of a sample unless one can argue that the sites evolved independently and are sampled independently. If we choose a set of S segregating sites (assumed to be independently sampled from a population, i.e. no linkage disequilibrium), rather than a genomic segment, we have the expected heterozygosity in a sample of size n,and so our index,This index is thus a measure of the deviation of the population (expected) heterozygosity per segregating site from its predicted value under the assumptions of equilibrium between selectively neutral mutation and genetic drift in a Wright-Fisher population.
To estimate this deviation across sites with different sample sizes, we can calculate the weighted average, weighting by the reciprocal of the variance.
The variance of isNotice that this is the theoretical variance in at a single segregating site in a sample of size n under the assumptions of the neutral model (above).
If the frequencies at different sites are independent then we can estimate the sample variance of Δπ(n) for S sites with sampling depth n, simply as Var[Δπ(n)]/S.
Assuming again that these SNPs are sampled independently both within and over sample sizes (i.e., no linkage disequilibrium), the average Δπ can be estimated by the weighted averagewhere nmin>3 and nmax = largest sample size.
Δπ for each position in the n147 regions (and its average across positions) was calculated over polarized sites with sampling sizes (n) between 32 and 34 in the Raleigh [65] and 18 to 21 for Rwandan data [71]. For n147 average Δπ the n147 regions were trimmed as described above for divergence. Data from complementary pathways were merged for larger scale positional Δπ.
n142-152 coverage of intergenic n147 regions was defined as the sum of the coverage of the region by the larger set of 142–152 bp nucleosomal fragments. This corresponds to what some authors call “occupancy.” GC frequency of intergenic n147 regions was calculated based on the nucleotide frequencies in the D. melanogaster reference sequence from bases 6 to 141 of the n147 regions (to minimize the potential impact of MNase sequence bias). For density plots and Spearman's ρ, only n147 regions with at least 50 (out of 147) intergenic bases polarizable were included in the analyses. As above, the interior 5 bp on each end of n147 regions were trimmed from the analysis. For correlations between divergence and genomic intergenic GC frequencies, Spearman's ρ were reported for non-overlapping 500 bp windows in which at least 166 intergenic bases were polarizable and no more than 250 bp in the reference were “N”.
Plots were generated using R [88]. Prior to plotting, all calculations were symmetrized around the dyad axis. n147 (±50 bp) divergence and Δπ plots (Figure 1D, Figure 2B,C, Figure 3A) were smoothed using running average in a window of 5 bp (weights: 0.125, 0.250, 0.250, 0.250, 0.125). For large scale plots of dinucleotide frequency, divergence and Δπ (Figure 1C, Figure 2D, Figure 3B), flanking regions were smoothed using running average smoothing with a window of 50 bp of equal weights. In those plots, the central n147 regions were smoothed separately using a 30 bp window of equal weights. Regions of 5 bp upstream and downstream of the n147 edges were trimmed prior to smoothing for large-scale plots.
To elucidate the distribution of the smoothed (as above) divergence on DNA from the structure of the nucleosome we colored-coded values of each base pair in a schematic rendering of the DNA strands of pdb1kx5 [55] using PyMOL [89]. Only the “top” turn of the DNA (base pairs 73 to −6) is shown. Bases 73 to 70 were rendered as grey, due to extreme values induced by MNase sequence bias.
|
10.1371/journal.pgen.1001273 | Proteins Encoded in Genomic Regions Associated with Immune-Mediated Disease Physically Interact and Suggest Underlying Biology | Genome-wide association studies (GWAS) have defined over 150 genomic regions unequivocally containing variation predisposing to immune-mediated disease. Inferring disease biology from these observations, however, hinges on our ability to discover the molecular processes being perturbed by these risk variants. It has previously been observed that different genes harboring causal mutations for the same Mendelian disease often physically interact. We sought to evaluate the degree to which this is true of genes within strongly associated loci in complex disease. Using sets of loci defined in rheumatoid arthritis (RA) and Crohn's disease (CD) GWAS, we build protein–protein interaction (PPI) networks for genes within associated loci and find abundant physical interactions between protein products of associated genes. We apply multiple permutation approaches to show that these networks are more densely connected than chance expectation. To confirm biological relevance, we show that the components of the networks tend to be expressed in similar tissues relevant to the phenotypes in question, suggesting the network indicates common underlying processes perturbed by risk loci. Furthermore, we show that the RA and CD networks have predictive power by demonstrating that proteins in these networks, not encoded in the confirmed list of disease associated loci, are significantly enriched for association to the phenotypes in question in extended GWAS analysis. Finally, we test our method in 3 non-immune traits to assess its applicability to complex traits in general. We find that genes in loci associated to height and lipid levels assemble into significantly connected networks but did not detect excess connectivity among Type 2 Diabetes (T2D) loci beyond chance. Taken together, our results constitute evidence that, for many of the complex diseases studied here, common genetic associations implicate regions encoding proteins that physically interact in a preferential manner, in line with observations in Mendelian disease.
| Genome-wide association studies have uncovered hundreds of DNA changes associated with complex disease. The ultimate promise of these studies is the understanding of disease biology; this goal, however, is not easily achieved because each disease has yielded numerous associations, each one pointing to a region of the genome, rather than a specific causal mutation. Presumably, the causal variants affect components of common molecular processes, and a first step in understanding the disease biology perturbed in patients is to identify connections among regions associated to disease. Since it has been reported in numerous Mendelian diseases that protein products of causal genes tend to physically bind each other, we chose to approach this problem using known protein–protein interactions to test whether any of the products of genes in five complex trait-associated loci bind each other. We applied several permutation methods and find robustly significant connectivity within four of the traits. In Crohn's disease and rheumatoid arthritis, we are able to show that these genes are co-expressed and that other proteins emerging in the network are enriched for association to disease. These findings suggest that, for the complex traits studied here, associated loci contain variants that affect common molecular processes, rather than distinct mechanisms specific to each association.
| Common genetic variants in over 150 genomic loci have now been unequivocally associated to immune-mediated diseases by genome-wide association studies (GWAS) [1]–[18]. It is presumed that these associations represent perturbations to a common but limited set of underlying molecular processes that modulate risk to disease. The next challenge – and the great promise of human genetics – is the identification of these disease-causing pathways so they may be targeted for diagnostics and therapeutic intervention.
In identifying such processes, there are difficulties in both (i) identifying the specific genes at (and how they are molecularly impacted by) each association and (ii) inferring disease-causing mechanisms from the set of identified genes. Linkage disequilibrium blocks containing disease-associated SNPs can be hundreds of kilobases in size, and some contain tens of genes to consider. Genes are often informally implicated in pathogenesis by their proximity to the most associated marker, their biological plausibility, or simply their being the only protein-coding gene in the region. In reality, however, it is only a very small subset of confirmed GWAS associations for which specific functional variants have been proven experimentally.
More systematic approaches have been applied to connect genes to a common process with the use of independent data, such as Gene Set Enrichment Analysis (GSEA) and Gene Relationships Across Implicated Loci (GRAIL) [1], [19]–[21]. Both approaches identify connections between genes based on descriptive categories that outline the theorized underlying pathogenesis. However, these concepts are often general, so that specific hypotheses and molecular pathways can be difficult to define and are somewhat limited to established knowledge bases.
Observations of interactions between the products of protein-coding genes offer the most direct route to identifying pathogenic processes. It has been shown in a number of Mendelian diseases that genes causal of a particular phenotype tend to physically interact [22]–[26]. This has been confirmed in the model organism C. elegans, where RNAi knock-down of germline genes correlated highly with their products interacting in yeast-two hybrid experiments [26]. A classic example of a human Mendelian disease that recapitulates this model is Fanconi Anemia (FA), an autosomal recessive disorder linked to at least 13 loci, at least 8 of which function in a DNA repair complex [22]. Protein-protein interaction (PPI) data has also been used to formulate hypotheses about co-expressed genes as well as cancer genes [27], [28]. We note that previous attempts to use PPI data to prioritize candidate genes in Mendelian disorders have been successful as was the case with the published tool Prioritizer [29]. We therefore set out to test such an approach in complex disease.
Investigators have rapidly populated databases of such protein-protein interactions over the past decade. The reported interactions in PPI databases stem from both small, directed investigations and high-throughput experiments, primarily yeast two-hybrid screens and affinity purification followed by mass spectrometry [30]. These data are inherently noisy: beyond technical false positives and negatives, experiments in vitro may report interactions that do not occur in vivo simply because the proteins involved never overlap spatially or temporally. To mitigate the noisiness of PPI databases, we extract networks from “InWeb”, assembled in 2007 by Lage et al [24], [31]. InWeb is a database of 169,810 high-confidence pair-wise interactions involving 12,793 proteins (human proteins and their orthologs). Lage et al. define high-confidence interactions as those seen in multiple independent experiments and reported more often in lower-throughput experiments [24]. To further restrict the data to biologically plausible interactions, we overlay mRNA expression information to confirm co-expression of binding partners; this correlates with co-localization, similar phenotype and participation in a protein complex [31], [32].
Assessing the significance of networks built from PPI data is challenging for two reasons: first, overall connectivity is a function of the binding degree (number of connections in the database for a given protein) of proteins within the network. Thus, the apparent density of a network could simply be due to the lack of specificity with which its constituents bind in vitro. Second, certain processes are more extensively studied, so more connections between proteins involved in them may be reported (see Figure S1; immune proteins are reported in more publications and have a higher mean binding degree). This confounds our effort to assess connectivity of associated loci if there is a bias in the data. From a genetic standpoint, a common randomization method would involve sampling SNPs from the genome matched for the appropriate parameters (such as gene density and protein binding degree). This method becomes highly limited if the disease loci contain genes that are better studied than the randomly sampled SNPs.
Therefore, we apply a permutation method that is robust to non-specific binding and differences in publication density. We perform a within-degree node-label permutation that is carried out as follows: a random network is built that has nearly the exact same structure as the original InWeb network, only the node labels (i.e. the protein names) are randomly re-assigned to nodes of equal binding degree; this method assumes a null distribution of connectivity that is entirely a function of the binding degree of individual proteins. Random networks will have the same size, number of edges and per-protein binding degree as InWeb; we build 50,000 different random networks. With this method, we are able to test the non-randomness of our network conditional on the exact binding degree distribution of our disease proteins.
Others have used PPI data in complex disease to understand epistatic loci or to build a network of interacting proteins from associated loci [33]–[35]. The novelty of our method lies not in the idea that PPI data can be used to help understand genetic loci associated to disease, but rather in that we have developed a broadly-applicable method to statistically evaluate the degree to which non-random PPI networks emerge from loci associated to complex disease and to leverage from this insight about causal proteins in large loci [33], [34]. We show this to be the case in a number of diseases.
Here, we use this methodology to evaluate whether genes in loci associated to five complex traits are significantly connected via protein-protein interactions. We report an algorithm to build and assess PPI networks using the InWeb database and find robust, statistically significant networks underlying associations to RA, CD, height and lipid levels, which we suggest as representative of the underlying pathogenic molecular processes. We then perform several detailed analyses on the RA and CD networks to confirm that they contain true biological insight into disease. We use independent mRNA expression data to show that the prioritized associated proteins we propose as interacting are co-expressed in relevant immune tissues, supporting a plausible biological setting for our findings as well as the validity of the reported protein-protein interactions. Lastly, by analyzing more recent GWAS meta-analysis results, we show that these networks contain components that show significant evidence of further genetic associations: proteins interacting with multiple associated network members and encoded elsewhere in the genome themselves carry an excess of association to disease in the latest meta-analyses of each of these diseases. Our method, available for download, generates an experimentally tractable hypothesis of the molecular underpinnings of pathogenesis.
We construct and evaluate networks of disease loci as outlined in Figure 1. We first define associated proteins as gene products encoded in genomic loci harboring variants associated to disease (Figure 1A, 1B; see Materials and Methods for locus definition). We construct networks of protein-protein interactions representing proteins as nodes connected by an edge if there is in vitro evidence of interaction (InWeb high-confidence interaction set). We build direct networks, in which any two associated proteins can be connected by exactly one edge, and indirect networks, where associated proteins can be connected via common interactor proteins (not encoded in associated loci) with which the associated proteins each share an edge. We restrict direct and indirect interactions to only those between proteins encoded in distinct associated loci.
We then calculate several metrics to evaluate network properties. These metrics can be divided into two categories: an edge metric and node metrics. The edge metric is the direct network connectivity parameter defined as the number of edges in the direct network. We interpret direct network connectivity as the frequency with which different loci harbor proteins that directly bind each other, regardless of how they assemble; direct network connectivity is therefore our most straightforward metric. Node metrics include the following: associated protein direct connectivity and associated protein indirect connectivity which refer to the number of distinct loci an associated protein can be connected to directly and indirectly, respectively, and common interactor connectivity which refers to the average number of proteins in distinct loci bound by common interactors in indirect networks. We interpret all three node metrics as descriptive of the type of network that was constructed: a stream of connections (such as the network A-B-C-D-E) will likely have low and insignificant node metrics despite a significant edge metric, whereas a more tightly clustered network might be enriched for both edge and node metrics. We assess the statistical significance of the various connectivity parameters using a within-degree node-label permutation strategy that controls for variation in the degree to which certain proteins are studied or behave in vitro (Figure 1C; see Text S1 for details on the permutation strategy, evaluation of its ability to distinguish signal from noise and a benchmark analysis of Fanconi Anemia). As we are interested in the processes underlying disease, we also define the gene encoding the top-scoring protein in each locus as most likely to be causal for association (Figure 1D; see Text S1 for prioritization strategy). We then use tissue expression data to test whether our nominated candidate genes are enriched in the same tissue(s) and therefore participate in a network that is biologically feasible (Figure 1E; Text S1). With this approach, we aim to construct plausible models of biological networks underlying pathogenesis.
Our approach controls for biases in the data: using the high-confidence interactions from InWeb addresses laboratory artifacts, and node-label permutation accounts for ascertainment biases due to differing levels of knowledge on biological processes for those proteins present in InWeb (Figure S1). We show empirically that priority scores given to proteins have no correlation with the degree to which they are represented in the database (Figure S2). A fundamental limitation of any functional data is that genes for which data are missing will not be considered. This applies to similar methods, including expression data that can be limited to genes represented on specific arrays or ontology analyses that are restricted to well characterized genes. Here, proteins that are entirely absent from the filtered InWeb data are not considered in our analysis (see Discussion). It is important to note that these genes, listed in Table S1, cannot be ruled out as potentially affected by causal variation since we have no power to make such a conclusion. We note, however, that the loci we have considered here (for the 5 complex traits) have the majority of their genes present in the high-confidence InWeb database (Table S1, median inclusion of 81.5%).
We also tested two additional permutation strategies on RA and CD – one based on random sampling of SNPs from the genome matched for proximal gene content and protein binding degree and the other based on edge permutation – that generally provided equivalent results (Figures S3 and S4); however SNP permutation may not be robust in the presence of extremes of gene density or protein binding degree at some loci, and edge permutation does not preserve the network structure of InWeb (Text S1). This analysis pipeline, which we call Disease Association Protein-Protein Link Evaluator (DAPPLE), is available for download at http://www.broadinstitute.org/~rossin/DAPPLE.
We first tested the method on the Mendelian disease Fanconi Anemia (FA) as a proof of principle. We input 9 of the FA genes and found 23 connections among them; compared to 50,000 random networks, the FA network is enriched for connectivity (direct network connectivity p<<2×10−5, Figure S5, Text S1). This result is consistent with current understanding of how the FA genes code for proteins that are part of the same DNA repair complex [36].
We then set out to test our method on two autoimmune diseases that are both complex traits. Recent GWA studies in autoimmune and inflammatory diseases have been particularly successful at determining loci encoding risk to disease, with over 100 loci described to date [2]–[7], [1]. We investigated rheumatoid arthritis (RA) and Crohn's disease (CD) and built networks from proteins encoded in 25 and 27 gene-containing associated loci, respectively [2], [8]. As described above, we built direct and indirect networks for each set of loci, evaluated the significance of the 4 network metrics to assess the probability that such networks could arise by chance, and we nominated candidate genes by assessing network participation. We followed up our results by assessing tissue co-expression as a test for the biological feasibility.
We were able to connect 20/27 loci for RA and 12/25 loci for CD in direct networks, strongly suggesting functional connections between proteins encoded in the associated regions. When compared to 50,000 random networks, we found that the direct network connectivity (the number of direct network edges) was statistically significant (27 for each disease; PRA = 3×10−4, PCD = 1.11×10−3; Figure 2) as was the associated protein direct connectivity (Figure S6A and S6B, PRA = 0.02, PCD = 0.00305). Thus disease-associated loci encode directly interacting proteins beyond chance expectation, suggesting that risk variants may act on suites of proteins involved in the same process.
We were then able to connect all but one gene-containing associated loci in each disease by expanding our networks to include common interactors (26/27 in RA; 24/25 in CD). The associated protein indirect connectivity was significantly enriched in both diseases (Figure S6A and S6B p = 1×10−5 in RA, p = 4.1×10−4 for CD), as was the common interactor connectivity (Figure S6A and S6B, p = 7×10−5 for RA and p = 1.1×10−4 for CD).
In aggregate, these results suggest that the observations of connectivity in Mendelian diseases are recapitulated in both RA and CD and that common risk variants predisposing to these diseases may impact sets of interacting proteins.
Given the significant connectivity of common interactors in the indirect networks for RA and CD, we speculated that common interactors might themselves be affected by previously undescribed risk variation. To test this, we consulted association data for each disease in the available data from meta-analyses, which for RA was in a newly completed meta-analysis and for CD was the same study that yielded the 30 loci [2], [37]. We assigned each recombination hotspot-bounded linkage-disequilibrium (LD) block in the genome an association score that represents the maximum score in that block corrected for the number independent SNPs therein. Genes were assigned association scores based on the blocks they overlap; this score distribution can then be compared to the scores of all gene-containing blocks in the genome (for both diseases, we removed the MHC from this analysis due to LD properties). Using this method, we found that common interactors expressed in the same tissues as associated proteins in our networks (see below) were encoded in regions with evidence of association significantly in excess to what is expected in gene-containing regions. In RA, the distribution of common interactor scores was skewed toward higher association (one-tailed rank sum p = 1.7×10−5) and in CD, we saw similar enrichment (p = 6.5×10−4). See Text S1 for details of analysis. This observed skew suggests that the common interactors themselves may harbor risk variants; we therefore considered the regions they overlap as candidates for replication (see “Crohn's Network Predicts New Loci” section).
To test whether the observed significant connectivity seen in RA and CD was present in non-immune complex traits, we tested our method on three traits: human height, blood lipid concentration (both LDL and HDL) and Type 2 Diabetes (T2D). We used 37 replicated gene-containing loci associated with human height, 18 with blood lipid levels and 36 with T2D [9]–[17]. The loci associated to height and lipids each contain proteins that assemble into significantly connected direct networks (Figure S6C and S6D, direct network connectivity p = 1×10−4 and p = 1.9×10−4 for each disease, respectively; see Text S1 for significance of other 3 parameters). In the height network, 19/42 loci participated in the direct network and 34/42 participate in the indirect networks, but only the direct network connectivity and the common interactor connectivity were significantly greater than chance. In the lipids network, 11/19 participated in the direct network and 16/19 in the indirect; all node metrics except the common interactor connectivity were significantly enriched. 9/37 T2D loci participated in the direct network and 34/37 in the indirect network; however, 3/4 metrics were not greater than chance expectation and only one was slightly enriched (Figure S6E, network connectivity p = 0.44960; see Text S1 for significance of other 3 parameters).
We therefore conclude that the PPI connectivity seen in two autoimmune diseases can be generalized to other complex trait loci (height- and lipid-associated regions), though we could not confirm the significance of the T2D network.
Our results suggest that functionally connected proteins reside in regions of the genome associated to disease risk. Permutation analysis revealed that these connections are in excess compared to what is expected given the binding profiles of associated proteins. For RA and CD, other proteins interacting with the associated proteins also show evidence of association beyond chance expectation. Cumulatively, these findings suggest that risk to the complex disease/traits studied here is spread over functional groups of proteins, directly analogous to observations in Mendelian traits.
An obstacle to interpreting GWA results stems from the difficulty in identifying the genes within associated regions influenced by risk variants. Candidate genes are often selected based on proximity to most associated markers and miscellaneous forms of previous knowledge. We therefore asked whether our observations could lead us to a principled, data-driven approach to selecting candidate genes by assessing their role in our networks. As shown in Figure 1 and described in detail in the Text S1, we used an iterative optimization method to assign priority scores to associated genes based on the network participation of their encoded proteins. We nominate genes that achieve the best score within their locus as the candidates for influencing disease risk. We describe the results in detail here for RA and CD; see Table S2 for scores assigned to RA, CD, blood lipid level and height genes.
We were able to nominate candidate genes in 12/21 RA loci encoding multiple genes (Table S2; Text S1). Examples of candidate genes in RA were IL2RA, CD40, CD28, PTPN22, CTLA4 and TRAF1. We accomplished the same task in CD, nominating candidate genes in 10/18 multi-genic loci. Candidates included JAK2, STAT3, IL23R/IL12RB2, PTPN2, MST1R and AIRE. For both diseases, genes in single-gene loci are also scored, though they are automatically considered the candidate gene (but not necessarily part of the underlying mechanism). It is important to note that we do not expect high-scoring proteins in every locus; we only expect high scores for those proteins that may participate in the common process(es) detected via enrichment in connections. RA and CD, like most complex diseases, most likely have many underlying processes, not all of which are captured here.
The core networks involving only these candidate genes represent our mechanistic predictions of pathways underlying pathogenesis in RA and CD. From a statistical standpoint the final networks built from candidate proteins account for the excess connectivity that we initially observed: the significance remains if we restrict multi-genic loci to just these genes (Figure S7A–S7D, direct network connectivity p<2×10−5 for RA and CD), while networks built from the remaining non-prioritized genes are less significant (Figure S8, direct network connectivity p = 0.0368 and p = 0.993, for RA and CD respectively). The remaining significance in RA is most likely a sign of additional important proteins that did not make the cutoff. From a biological standpoint, our candidates agree with experimental findings in the few cases where such evidence exists [38]–[45]. We therefore show that the connectivity between associated loci in RA and CD is primarily driven by a small subset of associated proteins encoded in those regions; this observation suggests that the interacting proteins – and the biological pathways they represent – may be the targets of risk variation.
To test the biological plausibility of our nominated core networks, we asked whether the candidate genes are co-enriched in subsets of particularly relevant tissues in a reference microarray dataset consisting of 14,184 transcripts measured in 55 immune, 8 gastro-intestinal, 27 neurological and 36 miscellaneous other tissues (126 total) [46]. These publicly available data are curated: expression intensities were converted to enrichment scores to reflect the enrichment of a gene in a tissue given its expression in all tissues. For each tissue, we compared the expression enrichment of RA and CD candidate genes to the rest of the genes in the genome using a one-tailed rank-sum test, resulting in a p-value for each tissue. A significant difference for a given tissue indicated that the genes in question were enriched for expression in that tissue compared to all genes in the genome. We also performed the same analysis for the remaining non-prioritized genes in associated regions to test whether the network prioritization method identified genes that were enriched in tissues distinct from non-prioritized genes. For discussion purposes, we defined “top” tissues as tissues achieving p<0.1 (Figure 3 depicts the entire distribution of p-values). This analysis led to 3 main conclusions. First, we found that for each disease, enrichment only occurred in immunologically relevant tissues (Figure 3; strikingly, immune tissues are nearly all ranked higher than other tissues). Second, we found that this was not the case to such an extent for non-prioritized genes (Figure 3, black points). Third, we found that the non-prioritized genes had fewer tissues where we could detect enrichment (Figure 3, RA and CD candidate gene tissue scores are more significant than tissue scores of non-prioritized genes). We formally tested this by comparing the p-value distributions for candidate genes and non-prioritized genes using a one-tailed rank-sum test (p = 2.85×10−7 for RA; p = 2.55×10−4 for CD). Of the 11 top tissues for CD candidate genes, 7 are subgroups of T-cell lymphocytes; the analogous list for RA (21 tissues) contains a mix of immune tissues, again dominated by T-cell subgroups (Table 1). The top tissue compartment for both diseases is defined as CD4+ T-cells.
We hypothesized that a subset of proteins connected to the core CD network (Figure 4B, the network built from prioritized genes in CD loci) might be near true causal variation. Having observed significant enrichment for association in the common interactors, we then chose a more conservative approach to propose candidate genes. We selected all proteins that connect directly to the core CD network only (21 genes) and filtered them on expression in the relevant tissues (Table 1). While this manuscript was being prepared, a larger meta-analysis was completed and recently published that reports 39 new loci associated to CD (295 overlapping genes) [47]. Of the 293 genes proposed by our method (small circles, Figure 4B), 10 were in newly associated regions (small red circles). This represents a statistically significant enrichment compared to chance expectation based on random draws from all 21,718 genes (p = 0.001) as well as random draws from genes expressed in at least one of the CD-relevant tissues (p = 0.01).
We performed a similar analysis in RA since the recent meta-analysis discovered 6 new loci (18 new genes) [37]. Of the 610 genes proposed, 1 was among the 18 new genes (Figure 4A, small red circle). This does not represent a statistically significant enrichment.
The networks (Figure 4) suggest pathogenic mechanisms in agreement with current thinking on disease etiology and propose novel roles for candidate proteins in these pathways. The RA network (Figure 4A) appears to represent signaling cascades involved in the inhibition or stimulation of the NF-kB complex, a factor that activates transcription of genes encoding cytokines, antibodies, co-stimulatory molecules and surface receptors [43]. STAT4 encodes a transcription factor that is activated upon engagement of cytokines, such as IL12 and interferon type I, with their receptors [43]. We show that not only does STAT4 show enrichment for connectivity, it is connected indirectly to a number of associated genes encoding surface receptor subunits that also achieve high network scores, such as IL12RB, IL2RA and PTPRC. TNFAIP3 (known as A20 in mice) is a cytoplasmic zinc finger protein that inhibits NF-kB activity, and knockout mice develop widespread and ultimately lethal inflammation, making it a plausible player in RA pathogenesis [48]. Also in the NF-kB pathway is associated protein CD40, which scores highly in our networks and binds TRAF6 and TRAF1 directly. CD40 is normally found on B cells but has also been shown to act as a co-stimulatory molecule on T cells to augment CD28 response and activate NF-kB [49].
PTPN22, a gene with strong genetic support for harboring risk variants (including the strongly associated R620W coding polymorphism), has been shown to act as a negative regulator of TCR but has not yet been definitively linked to a pathogenic mechanism [43], [50]. Here, we place it in context of other highly associated proteins and suggest that it is part of a common mechanism.
Finally, the RA network places a number of other proteins that have not yet been formally studied in the context of the proposed network underlying RA; these include CD2 and CD48, as well as FCG2RA and PRKCQ, genes suspected of being causal but not formally placed in a mechanism with other associations.
In CD the core of the candidate network (IL12B/IL23R/JAK2/STAT3; Figure 4B) corresponds to the interleukin-23 (IL23) signaling pathway. IL12B encodes p40, a component of the heterodimeric IL23. The IL23R gene encodes one half of the also heterodimeric IL23 receptor. This receptor is a cell surface complex found on a variety of immune cells; on activation, it induces Janus Kinase 2 (Jak2) autophosphorylation, which in turn leads to the translocation of STAT3 to the nucleus to activate transcription of various pro-inflammatory cytokines [40]. IL23 signaling is necessary for the activation and maintenance of a subset of CD4+ T cells acting as ‘inflammatory effectors’; these interleukin-17 responsive T-cells (Th17) have been implicated in autoimmune inflammation in CD and experimental models of other autoimmune diseases [40]. We note that IL23 belongs to the interleukin 12 family of cytokines and both ligand and receptor share subunits with the canonical IL12-mediated signaling pathway, which induces activation of regulatory T cells (Treg).
Our CD network suggests that other proteins participate in this pathway, including the tyrosine phosphatases encoded by PTPN2, a gene also associated to other autoimmune diseases [51]. Other proteins that are indirectly connected to this pathway include IRF1, which we score highly and that has separately been reported to activate transcription of IL12RB1 [52]. Furthermore, the common interactors that we prioritize for replication of association given their involvement in the CD network – including JAK1, STAT4, TYK2 and IL2RA – fall into the IL12 and IL23 signaling pathway (TYK2 and IL2RA were of the genes recently found to be in regions of association).
The CD network also generates new hypotheses about potentially important genes. We prioritize AIRE, an associated protein involved in T-cell development, which has not been extensively studied in the context of Crohn's but could plausibly lead to autoimmunity. ZNF365, a gene that achieves a high permutation score, has been assumed to be the causal gene because it is the only gene to reside in the wingspan of its locus; however, it has not been studied as part of the core network described here (IL23R/JAK2/STAT3 pathway). Finally, CSF2, IKZF3 and GRB7 are in the same large locus (17 genes) but achieve significant permutation scores; these genes have been less well studied in the context of CD.
We have shown that proteins encoded in regions associated to RA, CD, height and lipids interact and that the networks they form are significantly connected when compared to random networks. In CD and RA, the genes encoding prioritized proteins are preferentially expressed in immune tissues relevant to the pathogenesis of both diseases, while the rest of the genes in associated loci show less tissue preference. Furthermore, we can connect other associated proteins to these networks via common interactors, which appear to be encoded in genomic regions harboring further risk variants. Newly available data in CD allowed us to confirm that genes predicted to be near causal variation are indeed in regions now known to be associated to CD. We note that the conclusion of connectivity could not be extended to T2D, and we hypothesize that the lack of connectivity may be due to disparate underlying mechanisms that have yet to be well captured genetically. Though our aim was to build and analyze networks that emerge from replicated regions of association, we feel that a promising future direction may be to look more broadly for networks enriched in weaker signals of association. Evidence that this type of analysis may be helpful is that we pointed to a set of weaker CD association signals that were found to be true positives in a larger study.
Our results have several implications for the interpretation of genome-wide association studies: first, our ability to connect the majority of associated loci in a limited number of molecular networks suggests that these represent processes underlying pathogenesis. Second, these networks are unbiased, in the sense that they do not rely on previous classifications of gene function or pathway lists; rather, we assemble our networks from low-level functional genomics data and allow network structure, if any, to emerge. Third, our approach is general; we have demonstrated it using interactions between protein products, but any relationship between genes or other genomic features (non-coding RNAs, enhancer elements, conserved regions etc.) may be used in the same fashion. Even more powerful, approaches combining such orthogonal data types will be rewarding. The limitation to using PPI data from a curated database such as InWeb is that proteins for which no high-confidence interactions exist will be left out of the analysis. As such, our analysis is limited to proteins present in the database. Additionally, while we controlled for the biases we observed, other undetected biases still may exist.
Interestingly, there are certain cases where the method is able to distinguish between proteins that are close in the genome and functionally very similar. In RA, the rs12746613 locus has 3 genes in the PPI database – FCG2RA, FCGR3A and HSP70B. FCG2RA achieved a nominal p-value of 0.00703, whereas FCGR3A achieved p = 0.38296. Similarly, in the large rs3197999 locus in CD, the method gave MST1R a p-value of 0.0066 whereas MST1, the ligand of MST1R, achieved a p-value of 1. In these cases, the method is able to distinguish between functionally similar genes. There are times when it is unable to distinguish between functionally similar genes, however, such as the IL21/IL2 locus in RA, the STAT1/STAT4 locus in RA and the STAT3/STAT5A/STAT5B locus in CD.
We note in passing that the candidate genes we nominate are on average the closest to the most associated SNP in each locus, even though proximity within the LD region was not considered in the PPI analysis (p = 0.005, Figure S9). This supports the theory that the majority of causal variation will be close to the association signal rather than anywhere in the region of LD. We also observed overlap between genes prioritized by this method and GRAIL, a text-mining approach that uses orthogonal data (Table S2) [19]. We depict this information, as well as overlap between prioritized genes and the presence of non-synonymous SNPs, in Figure S10.
In this paper, we have studied 5 complex phenotypes, 4 of which show evidence of abundant PPI connections across loci. Our results therefore allow us to speculate that other complex diseases may behave in the same way and that genetic risk may be spread over the molecular processes that influence disease, rather than a single, catastrophic mutation as in Mendelian inheritance. In order to determine whether what we find here is expandable to complex disease in general, however, we would need to apply our method to the many more diseases and traits to which regions of the genome have been associated. Nonetheless, for the networks that emerge here, our approach identifies sets of proteins plausibly involved in pathogenesis, and the next step will be to identify what the molecular and phenotypic consequences of perturbing such processes are and how they relate to overall disease etiology.
We used a probabilistic database of reported protein-protein interactions described in 2007 by Lage et al [24], [31]. This database contains 428,430 reported interactions, 169,810 of which are deemed high-confidence, non-self interactions across 12,793 proteins. High-confidence is defined by a rigorously tested signal to noise threshold as determined by comparison to well-established interactions [24]. Briefly, InWeb combines reported protein interactions from MINT, BIND, IntAct, KEGG annotated protein-protein interactions (PPrel), KEGG Enzymes involved in neighboring steps (ECrel), Reactome and others as described elsewhere in detail [53]–[61]. All human interactions were pooled and interactions in orthologous protein pairs passing a strict threshold for orthology were included. Each interaction was assigned a probabilistic score based on the neighborhood of the interaction, the scale of the experiment in which the interaction was reported and the number of different publications in which the interaction had been sited. The data we used is available at www.broadinstitute.org/~rossin/PPI/ppi.html.
30 CD SNPs were derived from the first CD meta analysis of which 25 contain genes [2]. 28 RA SNPs were derived from the most recent RA review of which 27 contain genes [8]. 42 Height SNPs were derived from a number of analyses of which 38 contain genes [10], [13], [16]. 19 blood lipid level SNPs were derived from a number of analyses of which all 19 contain genes [11], [15]. Finally, 42 T2D SNPs were derived from a number of analyses of which 37 contain genes [9], [12], [14], [17].
Hotspot and linkage disequilibrium (LD) information were downloaded from www.hapmap.org for CEU hg17 and hg18 to match the version in which associations were reported [62]. We defined the wingspan of a SNP as the region containing SNPs with r2>0.5 to the associated SNP; this region is then extended to the nearest recombination hotspot. We downloaded the Ensembl human gene list from UCSC Genome Browser and collapsed isoforms into single genes [63]. We converted gene IDs from Ensemble to InWeb IDs. A gene's residence in a locus is defined by whether 110 kb upstream and 40 kb downstream (to include regulatory DNA) of the coding region of the gene's largest isoform overlaps the SNP wingspan [64].
All analyses, including building networks and evaluating significance, were carried out in R, Perl and Python and are available at www.broadinstitute.org/~rossin/PPI/ppi.html. Details on the algorithms are available in the Text S1 file.
|
10.1371/journal.ppat.1006797 | Recombinant PrPSc shares structural features with brain-derived PrPSc: Insights from limited proteolysis | Very solid evidence suggests that the core of full length PrPSc is a 4-rung β-solenoid, and that individual PrPSc subunits stack to form amyloid fibers. We recently used limited proteolysis to map the β-strands and connecting loops that make up the PrPSc solenoid. Using high resolution SDS-PAGE followed by epitope analysis, and mass spectrometry, we identified positions ~116/118, 133–134, 141, 152–153, 162, 169 and 179 (murine numbering) as Proteinase K (PK) cleavage sites in PrPSc. Such sites likely define loops and/or borders of β-strands, helping us to predict the threading of the β-solenoid. We have now extended this approach to recombinant PrPSc (recPrPSc). The term recPrPSc refers to bona fide recombinant prions prepared by PMCA, exhibiting infectivity with attack rates of ~100%. Limited proteolysis of mouse and bank vole recPrPSc species yielded N-terminally truncated PK-resistant fragments similar to those seen in brain-derived PrPSc, albeit with varying relative yields. Along with these fragments, doubly N- and C-terminally truncated fragments, in particular ~89/97-152, were detected in some recPrPSc preparations; similar fragments are characteristic of atypical strains of brain-derived PrPSc. Our results suggest a shared architecture of recPrPSc and brain PrPSc prions. The observed differences, in particular the distinct yields of specific PK-resistant fragments, are likely due to differences in threading which result in the specific biochemical characteristics of recPrPSc. Furthermore, recombinant PrPSc offers exciting opportunities for structural studies unachievable with brain-derived PrPSc.
| PrPSc Prions propagate by inducing the refolding of the natively folded normal cellular prion protein (PrPC) into the prion conformation in brain and other mammalian tissues (wild-type). Understanding the structure of PrPSc is essential to understanding how PrPSc prions propagate. The secondary structure of PrPC is composed of four α-helical regions, two very small β-sheets and random coil, while PrPSc is composed entirely of β-sheets and random coil. The β-sheets of PrPSc wind four times to form a spring-like 4-rung β-solenoid. The rungs are connected by stretches of random coil. We have used proteinase K (PK), to cleave these random coil connecters, allowing us to identify the location of the more PK-resistant β-strand stretches within wild-type PrPSc. In this work, we use recombinant PrPSc, (in vitro generated infectious prions) to show that patterns of PK cleavage for recombinant and wild-type PrPSc are very similar, indicating that they share a common architecture. This also means that recombinant PrPSc is a true surrogate for wild-type PrPSc. Since recombinant PrPSc is derived from recombinant PrP, future structural studies employing specific amino acid or stable-isotope labeled amino acid substitutions are easily achievable. Such substitutions are essential for NMR studies.
| Prions are infectious proteins [1, 2]. They propagate by inducing the host’s isosequential normal cellular prion protein (PrPC) to adopt the infecting prion’s conformation [2]. Prions can be transmitted from one organism to another by different means, for example by oral route [2, 3], hence their infectious nature. Prions pique an extraordinary theoretical and experimental interest because they challenge the notion that only nucleic acids are able to transmit heritable information. But they are also of critical practical importance, since some of them are associated with devastating neurodegenerative diseases. In particular, the mammalian PrPSc (prion protein, “scrapie” isoform) is the causal agent of the fatal transmissible spongiform encephalopathies (TSEs) [2–4]. TSEs affect both humans and agriculturally important animals, and while PrPSc prions typically remain contained within a given species, transmission of bovine PrPSc to humans occurred in the aftermath of the massive European bovine spongiform encephalopathy epizootic, killing more than 200 people and generating widespread alarm [5, 6]. Fortunately, through the intervention of regulators, the crisis has largely abated, although transmission of CJD through blood transfusion remains a concern [5]. This leaves only sporadic Creutzfeldt-Jakob disease (CJD), a rare ailment with a yearly incidence of ~1 case per million people, as the main human prion disease [2, 3].
Elucidating the molecular mechanism that governs the propagation of PrPSc, including the aforementioned transmission barriers has been a central issue and a challenge in prion research since these agents were first discovered [1]. This endeavor has been linked to the quest to elucidate the structure of PrPSc, an obvious pre-requisite to understand how such conformation propagates, i.e, how it is copied.
In this respect, it is important to note that most known prions, and in particular, PrPSc, form amyloids [4, 7–9]. Therefore, the main force driving and modulating prion propagation must be templating of an incoming partially or totally unfolded prion precursor protein molecule by the upper and lower surfaces of the amyloid fiber. These contain “sticky” β-strands ready to form an array of hydrogen bonds and thereby induce the formation of a new β-strand-rich layer, thus promoting growth of the amyloid filament in the direction of its axis. A recent cryo-electron microscopy study has determined the outline of the architecture of GPI-anchorless PrPSc, showing that it is a 4-rung β-solenoid [10]. This agrees with prior fiber X-ray diffraction [11, 12], 2D electron crystallography [11, 13] and SAXS-based [14] studies of other brain-derived wild type (wt) PrPSc molecules leading to a similar conclusion. On the other hand, another recent study has shown that shorter PrP sequences, such as PrP23-144, can adopt a flat, in-register amyloid architecture which is also infectious [15].
During propagation of multi-rung β-solenoidal structure, only the upper and lower rungs participate in inter-molecular hydrogen bonding. Identifying the specific amino acid residues that participate in β-strands, in particular those that make up the templating interfaces is essential to understand the details of PrPSc propagation, and, critically, to understand transmission barriers.
In the past, we have used limited proteolysis to probe PrPSc, in an attempt to identify sequential stretches that comprise β-strands vs. those that constitute the random coil loops/turns of PrPSc. It should be noted that, in contrast with early hypotheses, the elegant studies of the Surewicz and Caughey groups [16, 17] demonstrated that no α-helical secondary structure is likely to exist in PrPSc. Data from deuterium/hydrogen exchange studies, and limited proteolysis experiments are incompatible with the presence of any substantial amount of α-helical structure, and a critical reassessment of FTIR studies strongly suggests that absorbance peaks ascribed to α-helices is likely to have been a mis-assignment [9].
Using two analytical approaches, high resolution SDS-PAGE combined with epitope analysis, and mass spectrometry, we identified positions ~116/118, 133–134, 141, 152–153, 162, 169 and 179 (murine numbering) as PK cleavage sites in brain-derived PrPSc. These sites likely define loops and/or borders of β-strands, and are helping us to define the hypothetical threading of the β-solenoid [18].
In this context, recPrPSc is a very attractive tool for structural studies, given that it allows the introduction of the sequence variations, labels and isotopically labeled amino acid residues necessary for rigorous NMR studies. A number of recombinant PrP preparations with different degrees of infectivity have been described since the seminal report by Legname et al.[19–22]. Recently, Wang et al. generated recPrPSc exhibiting incubation times similar to those of brain-derived PrPSc of the same sequence and causing the same pathogenic changes as that of wt prion disease [23, 24]. While incubation times should be considered very cautiously, given that a long incubation time can be the result of low titer but also of a transmission barrier, the study by Wang et al. has led to the definitive acceptance that bona fide, highly infectious recPrPSc can be generated in vitro. As a corollary, the possibility to use the versatile recPrPSc as a convenient model for elucidation of the structure of PrPSc in general was opened.
Here, we report studies to probe the structure of infectious recPrPSc using limited proteolysis. Mouse and bank vole (Myodes glareolus) recPrPSc prions yield an array of N-terminally truncated PK-resistant fragments very similar to that seen after PK treatment of brain-derived PrPSc. This is strongly supportive of shared key architectural elements between both prion types.
The following reagents were obtained from the indicated commercial sources: PNGase F, from New England Biolabs (Ipswich, MA, USA); Tris/Tricine electrophoresis buffer and Broad-Range SDS-PAGE Standards from BioRad (Hercules, CA, USA); Sypro Ruby dye, and Novex Sharp Pre-stained Portein Standard, from Thermo-Fisher (Whaltman, MA, USA); Immobilon P 0,45 μm PVDF membranes, from Millipore (Billerica, MA, USA); Ultra-low Range Molecular Weight Marker, Pefabloc, PMSF and PK, from Sigma-Aldrich (St Louis, MO, USA). All other reagents were from Sigma-Aldrich unless otherwise indicated.
Antibody R1, which recognizes PrP epitope 225–230 [25], was a generous gift from Anna Serban, Institute for Neurodegenerative Diseases, UCSF, and was used at a 1:5000 dilution; antibody #51, which recognizes PrP epitope 92–100 [26] was kindly provided by Lothar Stitz, Fridrich Loeffler Institut, Insel Reims, Germany, and was used undiluted; antibody 3F10, which recognizes PrP epitope 137–151 [27] was a generous gift from Yong-Sun Kim, Hallym University, Republic of Korea, and was used at a 1:5000 dilution. Antibody SAF-84, which recognizes PrP epitope 165–172, was from Thermo Fisher Scientific (Rockford, IL, USA), and as used at a 1:5000 dilution. Secondary antibodies were goat anti-human (Thermo Fisher) and goat anti-mouse (GE Healthcare Life Science, Chicago, IL. USA), used to detect R1, 3F10 and #51, respectively; both were used at a 1:5000 dilution.
Recombinant mouse PrP23-230 (MoPrP23-230) was expressed in E. coli competent cells. Bacteria were harvested by centrifugation at 5.000 g. Bacterial pellets were lysed by incubation for 30 minutes at room temperature with shaking in lysis buffer (50mM Tris-HCl, 5mM EDTA, 1% Triton X-100, 1mM PMSF, 100 μg/ ml lysozyme, pH = 8); MgCl2 and DNase I were then added to 20mM and 5μg/ml final concentrations, respectively, and further incubation at room temperature carried out for 30 minutes. Inclusion bodies thus obtained were collected by centrifugation at 20.000 g at 4°C for 20 minutes, and solubilized with inclusion body solubilization solution (6M Gn/HCl, 10 mM of Tris-HCl, 100 mM Na2HPO4 and 10 mM β-mercaptoethanol, pH = 8.0). The solubilized sample was then filtered through a 0.22 μm filter and loaded to a 5 ml FF Crude His-Trap column (GE Healthcare, Life Sciences (Chicago, IL. USA) connected to a 1200 Series HPLC system (Agilent Technology, Santa Clara, CA, USA). The column was washed with inclusion body solubilization solution and refolded in-column by gradually diminishing the concentration of Gn/HCl and β-mercaptoethanol with a gradient of 10 mM of Tris-HCl, 100 mM Na2HPO4, pH = 8.0 over 100 minutes; recMoPrP23-230 was then eluted with 30 ml of 200 mM imidazole in10 mM Tris-HCl, 100 mM Na3PO4,pH = 8.0). The eluate was dialyzed against10mM NaH2PO4, pH = 5.8, and subsequently against d.i. H2O at 4°C. Dialyzed samples were centrifuged to eliminate any aggregated material present and stored at -80°C until used for conversion to recPrPSc.
Bank vole PrP23-231 (BVPrP23-231) was expressed following the same protocol, and similarly applied to a His-Trap column as described above; however, it was eluted from the column by application of a solution consisting of 10 mM Tris-HCl, 2M Gn/HCl, and 100 mM Na2HPO4, pH = 8.0. Elution fractions containing PrP, as determined by SDS-PAGE with Coomassie staining, were then folded by dialysis against 10 mM sodium acetate, pH = 5. Precipitated material was removed by centrifugation.
Two preparations of recombinant (rec) murine PK-resistant (PK-res) PrP and one of recombinant bank vole PK-res PrP were analyzed in this study; all of them were prepared using different variants of recombinant PMCA (recPMCA). Recombinant MoPrPSc-17kDa was prepared in Grand Rapids (USA) and has been described before [23]. Briefly, it was generated de novo from recMoPrP23-230 by recPMCA in the presence of POPG and RNA and its infectivity has been tested in WT mice resulting a 100% attack rate [23, 24].
A recMoPrPSc (vide infra with regard to its infectivity properties) was generated in Bilbao, Spain, from recMoPrP23-230 expressed, purified and folded as described above, by means of recPMCA seeded with recMoPrPSc-17kDa in the presence of 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoglycerol (POPG) and RNA [23, 24]. This sample was termed recMoPrPSc-950. A sample of recBVPrPSc (vide infra) was prepared by seeded recPMCA using a protocol that will be detailed elsewhere. Briefly, recBVPrP23-231 (109I) prepared, purified and folded as described above, was used as a substrate in a mixture that contains dextran and detergent and that was subjected to several cycles of PMCA; the initial seed used was a small portion of brain homogenate from a bank vole (109I) infected with deer CWD prions.
Transgenic homozygous GPI-anchorless (GPI-) PrP mice (tg44-/-), were obtained by crossing of tg44(+/-) heterozygous (GPI-) PrP mice [29], generously provided by Bruce Chesebro, Rocky Mountain Laboratories, NIH, USA [26]. This GPI-anchorless tg mouse model is the same that we have used in the past [26]; additional GPI-anchorless tg mouse models have been developed [30]. Female mice were inoculated ic at six weeks of age with 20 μl of a 2% RML-infected mouse brain homogenate, kindly provided by Juan María Torres, CISA, Madrid, Spain. After 365 days post inoculation, mice were euthanized, their brains surgically removed, rinsed in PBS, and stored at -80 °C until needed. A 10% w/v, brain homogenate was prepared in PBS, 5% sarkosyl, using a dounce homogenizer (Wheaton Industries Inc, NJ, USA), followed by one pulse of sonication to clarify the homogenate, with an ultrasonic homogenizer probe (Cole Parmer Instrument CO., Chicago IL, USA). The brain homogenate was treated with 25μg/ml of PK for 30 minutes at 37°C, and then deglycosylated with PNGase F following the manufacturer´s recommendations.
RecMoPrPSc was treated with 10 μg/ml PK, at 37°C for 30 minutes. The reaction was quenched by adding 2 mM Pefabloc and incubated for 15 minutes on ice. PK-resistant fragments were then pelleted by centrifugation at 18.000 g at 4°C for 1 hour using a (Microfuge 22R centrifuge, Beckman Coulter). Under these conditions, all PK-resistant fragments are recovered in the pellet (S1 Fig). Pellets were resuspended in 6M Gn/HCl and stored at -20°C until use.
PK-resistant fragments were precipitated with ice-cold 85% MeOH. Pellets were resuspended in MiliQ H2O and Tricine buffer in a ratio 1:2. Reduction was carried out by adding β-mercatoethanol to 2% (v/v). Samples were boiled for 10 minutes. High resolution electrophoresis was carried as described by Vázquez-Fernández et al. [26]. After electrophoresis, gels were washed with miliQ H2O and incubated with fixing solution (10% MeOH, 7% acetic acid) for 1 hour at room temperature. Sypro Ruby staining was then performed by incubation overnight at room temperature in the darkness. Alternatively, the gels were transferred to Immobilon P 0.45 μm PVDF membranes, which were subsequently probed with the antibodies described above.
A 1μL sample of the solution of PK-resistant recPrPSc fragments solubilized in 6M Gnd/HCl (vide supra) was mixed with 49 μL of sinapinic acid solution (10 μg/mL of sinapinic acid dissolved in 30% acetonitrile (ACN) with 0.3% trifluoroacetic acid (TFA) and analyzed by MALDI-TOF. One half μL aliquots were deposited using the dried-droplet method onto a 384 Opti-TOF MALDI plate (Applied Biosystems, Foster City, CA, USA). MALDI analysis was performed in a 4800 MALDI-TOF/TOF analyzer (Applied Biosystems). MS spectra were acquired in linear mode (20 kVsource) with a Nd:YAG, (355 nm) laser, and averaging 500 laser shots. For spectra data analysis of recMoPrPSc samples, an initial external calibration was carried out using insulin (m/z = 5733), ribonuclease A (m/z = 13682) and lysozyme (m/z = 14305), (Sigma-Aldrich) as standards. A peak with m/z = 9390.2 Da, corresponding to fragment N153-S230, was unambiguously identified with a mass error < 1 Da by ESI-TOF analysis (vide infra) of the same sample (S2 Fig). This peak was used as an internal calibrant, and all m/z values in the spectrum corrected accordingly. For recBVPrPSc samples, only external calibration was used. The final resulting m/z values were matched to PrP fragments with the help of GPMAW 6.0 software (Lighthouse, Odense, Denmark). Final experimentally calculated mass data are shown in Tables 1 and 2, and match theoretical values within the experimental error of the MALDI-TOF analysis.
As indicated, 10 μl of PK-resistant recPrPSc were subjected to ESI-TOF analysis. The sample was injected to an Agilent 1100 HPLC system equipped with a Vydac 218TP C-18 column (Vydac, MD, USA). A gradient of ACN over 0.1% formic acid was applied over 60 minutes, at a flow of 0.2 ml/min. The effluent of the column was fed into a Bruker Microtof Focus mass spectrometer (Bruker Daltonik, Billerica, MA, USA) and sprayed into the mass detector. The capillary voltage was set at 4500 V, the pressure of the nebulizer was 2.5 Bar, the drying gas flow 8 L/minute and the drying temperature, 200 °C. The mass range of the detector was 50–3000 m/z.
Using seeded recPMCA, we generated a recMoPrP auto-propagative species that we termed recMoPrPSc-950, and was partially resistant to PK (Fig 1A). In order to assess its infectivity, and therefore its prionic nature, we performed animal bioassays. We inoculated this putative recombinant murine prion, into the brains of 18 Tga20 mice. All 18 inoculated Tga20 mice developed standard clinical signs of prion disease (ataxia, hindlimb paralysis, kyphosis, weight loss) and were eventually euthanized (Fig 2A). Histopathological and immunohistochemical examination of the brains of animals inoculated with PK-resistant MoPrP, showed characteristic TSE spongiform lesions and PrPSc deposits, typical of prion disease (Fig 2B). Furthermore, immunochemical analysis of brain homogenates revealed the presence of PK-resistant PrP (Fig 2C). All this confirms that the inoculated material was a prion, and therefore it could be appropriately referred to as recMoPrPSc-950. In this study we also analyzed a preparation of the recombinant murine prion recMoPrPSc-17kDa [23–24]. Although its infectious nature had already been established [23, 24], we confirmed it under our experimental conditions. We inoculated a group of 5 Tga20 mice, all of which developed signs of prion disease and were humanely euthanized after 119 ± 20 days. Immunohistochemical and immunochemical analysis of brains confirmed prion disease, in agreement with previous published studies [23, 24]
We subjected recMoPrPSc-950 and recMoPrPSc-17kDa to limited proteolysis using 10 μg/ml of PK for 30 minutes at 37 °C. PK-resistant fragments were detected by Sypro Ruby staining after SDS-PAGE. In recPrPSc-17kDa, a clear, intense ~17 kDa fragment was readily apparent, with two additional somewhat fainter and broader bands, one of ~6.5 kDa, and another with a MW between 6.5 and 3.5 kDa (Fig 3A, left). The ~17 kDa band is obviously equivalent to the classic PrP27-30 PK-resistant fragment seen in brain-derived samples, which migrates between 27–30 kDa [1, 3, 16, 26]. Both PK treated recMoPrPSc-17kDa and PrP27-30 lack amino acids 23–90, so the difference in migration between the two molecules is due to the lack of a GPI anchor and lack of glycosylation in the PK-treated recMoPrPSc-17kDa. This 17kDa eponymous fragment has been previously reported for recMoPrPSc-17kDa [23]. The lower MW PK-resistant fragments may have been previously overlooked, since they would run near the front in conventional SDS-PAGE systems.
In PK treated recMoPrPSc-950 a ~16 kDa band was conspicuous. Additional PK-resistant fragments with smaller apparent MWs between ~15 and ~3.5 kDa, several of them with an intensity similar to that of the uppermost ~16 kDa fragment, were also detected (Fig 3A, right). Compared to the pattern obtained from recPrPSc-17kDa, a number of intense bands in the 15–10 kDa range were seen in recMoPrPSc-950 that were extremely faint or altogether absent in recPrPSc-17kDa (Fig 3A).
We used a set of monoclonal antibodies with well characterized linear PrP epitopes to perform epitope mapping of the PK-resistant fragments present in our samples. We used antibodies recognizing “N-terminal” (antibody #51: 92–100), central (antibody 3F10: 137–151) and C-terminal (antibody R1: 225–230) epitopes. The 92–100 epitope is considered here “N-terminal” given the total absence of any PK-resistant species containing the ~23–90 amino-terminal stretch, believed to be flexible in all PrP conformers. Therefore, position ~90 is considered, for simplicity, to be the reference “amino terminus” of PK-treated samples. The amino acid sequences of mouse and bank vole PrP are shown in S3 Fig to facilitate evaluation of the results.
The use of a high resolution Tris/Tricine SDS-PAGE system optimized the separation of PK-resistant fragments, particularly the smaller ones, while allowing us to compare their pattern with that of PK-treated GPI-anchorless PrPSc, previously analyzed by our group [26]. These results are shown in Fig 3B–3D, which shows results in the following order: “N-terminal”, central and C-terminal antibodies.
The “N-terminal” antibody, #51 (epitope 92–100), detected the ~17 and ~16 kDa bands stained by Sypro ruby in gels corresponding to recPrPSc-17kDa and recMoPrPSc-950, respectively (Fig 3B). In agreement with previous studies [26], antibody #51 recognized the predominant ~17 kDa PK-resistant band in a GPI-anchorless PrPSc-containing sample, which corresponds to (81/89-232) (Fig 3B left); it should be mentioned that GPI-anchorless PrP contains two extra C-terminal amino acid residues as a consequence of the way in which the transgene was designed [26].
Considering their apparent sizes, the ~17 and ~16 kDa PK-resistant fragments in the recPrPSc-17kDa and recMoPrPSc-950 samples (Fig 3B) must correspond to partially overlapping collections of PK-resistant fragments with “ragged termini” [31, 32], with a predominance of cleavages around positions ~86–89 in recPrPSc-17kDa, and 92–98 in recMoPrPSc-950. This difference is reminiscent of the difference between Drowsy vs. Hyper or CJD type I vs. type II major PK-resistant PrPSc fragments [31, 32].
The recMoPrPSc-950 sample also contained a broad ~5–7 kDa band, absent in recMoPrPSc-17kDa (Fig 3B center). Considering its apparent MW, it can be concluded that this band necessarily corresponds to doubly N- and C-truncated PK-resistant fragments.
Antibody R1, which recognizes the C-terminal epitope (225–230) detected, as expected, the ~17 and ~16 kDa PK-resistant fragments also seen using Sypro Ruby and antibody #51, in GPI-anchorless PrPSc, recPrPSc-17kDa, and in recMoPrPSc-950, respectively, supporting the notion that these fragments span from ragged ends beginning around position G92 all the way to the C-terminus (Fig 3C). In the GPI-anchorless sample, R1 detected, besides, the 6 additional bands previously described and characterized by Vázquez-Fernández et al. [26] (Fig 3C left). Remarkably, the pattern of bands detected by this antibody was, to a considerable extent, similar in the other two infectious prion samples, i.e., recMoPrPSc-17kDa and recMoPrPSc-950. Namely, the ~14.6, ~13, ~12,~10.2, 8, and ~6.7 kDa bands described by Vázquez-Fernández in GPI-anchorless PrPSc [26] were seen in the GPI-anchorless PrPSc, recMoPrPSc-17kDa and recMoPrPSc-950 samples, although the relative intensities of bands varied from sample to sample (Fig 3C). Considering the extreme C-terminal position of the R1 epitope, which leaves virtually no leeway for alternative sequence combinations leading to a given apparent MW, we can tentatively conclude that there might be a very close identity of these bands between the three samples, which in turn means that cleavage sites are approximately the same. On the other hand, there was one evident difference in the PK digestion pattern of recMoPrPSc-950 with respect to the recMoPrPSc-17kDa and GPI-anchorless samples: two bands, with apparent MWs of ~4.5 and ~3.5 kDa, which are absent in the other samples (Fig 3C). Considering their sizes, they should correspond to novel PK-resistant fragments with N-termini around~G194 and ~E206, respectively.
The central antibody 3F10 (epitope 137–151) should recognize the ~17/16, ~14.6, ~13 and perhaps the ~12 kDa bands recognized by R1 (Fig 3C), given that they also contain the epitope recognized by 3F10: these bands correspond to fragments ~92/98-230,~116–230, ~134–230 and ~138–230, respectively [26]. Indeed, the antibody revealed bands of these sizes in the two prion samples, albeit with different relative intensities (Fig 3D). Antibody 3F10 also detected additional fragments of ~10, ~8, ~7 and ~6 kDa in the recombinant samples that were not present in GPI-anchorless PrPSc. More specifically, the ~10 kDa band was seen in both samples, while the others were seen in recMoPrPSc-950 (Fig 3D). None of these bands coincides with those recognized by R1 (compare Fig 3D with Fig 3C), whether they have or have not the same size, as there is no possible overlap of epitopes for fragments smaller than 12 kDa. Thus, fragments ~152–230, ~162–230 and ~169/179-230, lack the 3F10 epitope. Therefore, the ~10, ~8, ~7, and ~6 kDa fragments recognized by 3F10 must necessarily correspond to doubly N- and C-terminally truncated PK-resistant fragments. Some of these fragments, more specifically some of those seen in the recMoPrPSc-950 sample, also contain the (92–100) epitope recognized by antibody #51 (Fig 3A), while others do not, and therefore their N-termini must lie beyond the 92–100 sequence. In summary, the combined mapping reveals the existence of a number of PK-resistant fragments with a double truncation at both the N- and C-termini. These fragments are particularly prevalent in recMoPrPSc-950, with N-termini, in this case, around~G92. These N-,C-truncated fragments were not seen in PK-treated GPI-anchorless PrPSc (Fig 3D left), in agreement with previous results [26].
We further probed the PK-resistant fragments in PK-treated recMoPrPSc-950 with an additional antibody, SAF-84 (epitope: 166–172, located between those recognized by 3F10 and R1). Results were consistent with the patterns revealed by these two antibodies and are described in detail in S4 Fig.
We sought to confirm the identity of PK-resistant bands in our recombinant samples, approximately revealed by the sizes and pattern of bands surmised from epitope analysis, by means of mass spectrometry. For logistical reasons, we could only obtain data from recMoPrPSc-950. As shown in Fig 4, Table 1, and S2 Fig, MALDI and ESI-TOF analysis of the same sample identified a number of C-terminal peptides PK-resistant peptides, namely, 89–230, 97–230, 116–230, 134–230, 138–230, 141–230, 152–230, 153–230, 162–230 and 179–230. Such peptides coincide quite well with the apparent MWs of C-terminal peptides detected by antibody R1 (Table 1). Also, in agreement with results obtained with epitope analysis, these peptides are equivalent to those obtained after PK treatment of GPI-anchorless PrPSc [26].
We also generated a recBVPrP auto-propagative species that was partially resistant to PK (Fig 1B). To confirm its infectious, prionic nature, we inoculated this putative recBVPrPSc into the brains of a group of 10 bank voles expressing homozygous PrP (109I) [33]. Eight of ten bank voles inoculated with putative recBVPrPSc succumbed to prion disease with a mean survival time of 239 ± 49 days post-inoculation, while the other 2 died of inter-current causes at an early age (206 days post-inoculation), and therefore it cannot be concluded whether they were developing a prion disease or not. Histopathology of brains of these animals confirmed prion disease (S5 Fig). Full details of this prion disease has been reported elsewhere [34]. Therefore, the inoculated material is also a prion and can be appropriately referred to as recBVPrPSc. Treatment of recBVPrPSc with 20 μg/ml of PK resulted in the appearance of a number of PK-resistant fragments, as seen after Coomassie staining (Fig 5). A doublet of closely migrating ~17 kDa fragments was predominant. We reasoned that it might correspond to the entire PrP sequence minus the extremely flexible ~23–89 tail (Bank vole numbering), with two close but slightly different N-terminal cleavage patterns, i.e., the equivalent of the ~17 kDa band of GPI-anchorless PrPSc. These two bands were excised and subjected to in-gel tryptic digestion. MALDI analysis of the digest led to the detection of a number of tryptic peptides from amino-terminal (H111-R136), central (P137-R148) and carboxy-terminal (E221-R229) regions of PrP (considering the expected loss of the extreme amino-terminal flexible tail, up to ~G90, in fact confirmed by the complete absence of peptides from that region). A characteristic peak with m/z of 1820 Da, corresponding to a “ragged end” tryptic peptide G90-K106 [35]was also seen in both samples (S6 Fig). No obvious differences in the spectra from the two bands were identified. Since an obvious possibility was the slight difference in apparent MW between the two bands might be the result of different ragged termini, a thorough search for peaks corresponding to tryptic peptides with different ragged termini was carried out, but yielded no obvious differences between the two bands. Therefore, the nature of the difference between the two bands cannot be explained at this point.
In addition to this doublet, at least three additional PK-resistant fragments were detected in the Coomassie-stained SDS-PAGE gel (Fig 5). These lower bands were also subjected to in-gel tryptic digestion followed by MALDI-TOF analysis of the resulting tryptic fragments; MALDI spectra (S6 Fig) showed signals corresponding to tryptic fragments of different regions of the BVPrP sequence, in different proportions, confirming that these bands are PrP fragments of different sizes. In parallel, direct MALDI analysis of the sample confirmed the presence of fragments with sequences 153–231, 135–231, 133–231, 117–231, 75/83-231, and 83/90-231 (Fig 5 and Table 2). A group of peaks corresponding to peptides with masses between ~7000 and ~8000 Da was also evident (Fig 5 and Table 2); these might correspond to doubly N- and C-terminally truncated PK-resistant peptides, similar to those seen in the recMoPrPSc samples, although this cannot be confirmed at this time.
Recombinant PrPSc will become an invaluable tool for prion structural studies. Recombinant PrPSc is produced by converting bacterially derived PrP into recPrPSc, which means that incorporation of stable isotopes or different natural or unnatural amino acids into PrP sequences is greatly simplified. This will allow researchers to prepare custom-made recPrPSc for NMR-based analyses. However, it will be critical for these efforts to use fully infectious recPrPSc samples. To date, reports have described studies of amyloid recombinant PrP preparations exhibiting very limited infectivity [36, 37], known to have a structure different to that of PrPSc. On the other hand, very rich structural information is being extracted from highly infectious PrP23-144 amyloid fibers, revealing a flat in-register cross-β stack [15, 38] that is also different from the 4-rung β-solenoid that characterizes full length PrPSc [10].
Our recPrPSc prions, made in vitro from bacterially derived recombinant PrP, exhibit full infectivity, with attack rates of 100% and incubation periods comparable to wt prions. They share key architectural features with brain-derived PrPSc, when analyzed by a limited proteolysis-based structural analysis [26, 39, 40]. Limited proteolysis of these recPrPSc species generated a fragmentation pattern consisting of a number of PK-resistant fragments that were the same as or equivalent to those obtained during limited proteolysis of GPI-anchorless MoPrPSc and wt SHaPrPSc. In particular, mass spectrometry-based analysis revealed nicks at positions 89/90, 116/18, 133/134, 141, 152/153, 162 and 179 (Table 1), in excellent agreement with conclusions derived from epitope analysis (Fig 3).
A group of doubly truncated fragments were much more conspicuous in recMoPrPSc-950 than recMoPrPSc-17kDa. Their exact sequence remains uncertain. However, one of the fragments from recMoPrPSc-950 clearly spans the sequence comprising the epitopes of antibodies #51 and 3F10. This result is consistent with a fragment from positions ~89–152. Such a fragment would complement the fragment spanning the sequence 153–231 see in PK treated recMoPrPSc-950 and GPI-anchorless PrPSc. The theoretical MW of such a fragment would be 6.7 kDa, in good agreement with the band recognized by antibody #51 (Fig 3B). The existence of three distinct bands recognized by 3F10 indicates additional cleavage sites between ~89 and ~152. Based on the analysis of other prions, candidates for these N-terminal cleavage sites are ~117/119 and ~134. Such fragments would be recognized by 3F10 but not #51. For the previously stated reasons, we are unable to define the precise sequences of these peptides. Furthermore, the different responses of different antibodies complicate interpretation of the data. Thus, the ~6.5 kDa fragments detected by 3F10 in recMoPrPSc-17kDa seem not to contain the (92–100) epitope, but in the absence of additional information, it is difficult to conclude where exactly their termini are located.
Doubly truncated fragments have not been associated with the majority of “classical” brain derived prions, such as GPI-anchorless PrPSc [26], scrapie MoPrPSc, 263K and Dy SHaPrPSc [39], or CJD PrPSc [41]. In contrast, low MW bands corresponding to doubly truncated PK-resistant fragments are hallmarks of “atypical” PrPSc strains, including Gertsmann-Streussler-Scheinker (GSS)-PrPSc, and atypical scrapie-OvPrPSc strains, such as Nor98 PrPSc [40, 42]. Thus, analysis of brain homogenates from GSSP102L patients showed the presence of two PK-resistant PrP fragments with apparent molecular masses of ~21 and ~8 kDa. The ~21 kDa fragment, similar to the PrP-res type 1 described in CJD (i.e., the classic triad of PrP27-30 fragments with variable glycosylation), is typically found in some cases, whereas the ~8 kDa fragment is found in all cases, and has been taken to represent a pathognomonic characteristic of GSS [31, 43]. However, a similar PK-resistant fragment has been also detected in bank vole-adapted CJD PrPSc, blurring to some extent the distinction between classic and atypical strains of PrPSc [41]. Mass spectrometry-based analysis of the GSS ~8 kDa fragment revealed that it consists of a collection of peptides with ragged termini, spanning from 74/78/80/82 to 147/150/151/152/153 [43]. Furthermore, the ~7 kDa PK-resistant fragment of PrPSc detected in A117V GSS cases was seen, using mass spectrometry analysis, to span from Gly88/Gly90 to Arg148/Glu152/Asn153 [44].
As shown by Pirisinu et al., this pattern is remarkably similar to that of Nor98 atypical PrPSc, treatment of which with PK yields a ~7 kDa resistant fragment whose sequence is 71/79-153 [40]. In contrast, the most resistant part of PrPSc from classical strains is, precisely, the complementary sequence: a 152/153-232 fragment becomes prevalent with increasing treatment time with PK of GPI-anchorless PrPSc [26], and remains folded upon guanidine-induced partial unfolding [26, 45]. All of this suggests that the region around 152/153 marks a “hinge” that connects two stable sub-domains within PrPSc. It is noteworthy that this region signals two halves of the putatively folded region of PrPSc of comparable size; since the flexible loop likely spans to P157 (murine numbering), it would connect two sub-domains of~62 and ~72 residues spanning N- and C-terminally with respect to it.,Higher resistance of either the ~152/153-230 half, typical of “classical” PrPSc strains [26, 39, 45] (but vide supra) or of the ~80/90-152/153 half, characteristic of “atypical strains” [31, 40, 43, 46] might reflect differences in the threading within these specific sub-domains, with differences in the relative content in β-sheet secondary structure (longer or shorter β strands) and packing of the loops connecting them. However, the fact that overall similar nicks are detected in all cases suggests that threading differences are not very large, and that overall, the same elements of secondary structure, likely arranged in the same way, are characteristic of the structures of all three classes of PrPSc. It should also be noted that in any given PrPSc prion isolate, including our recombinant ones, there might exist mixtures of more than one structure. In that case, the relative abundance of specific PK-resistant fragments will reflect the relative contributions of such structural variants.
Our study also provides preliminary evidence, in recMoPrPSc-950, of two additional, C-terminally located, PK cleavage sites not previously detected in brain-derived PrPSc. The strongest evidence of the existence of such cleavage sites was provided by a ~4.5 kDa band in the Sypro Ruby stained gel of recMoPrPSc (Fig 3A). A band with a similar apparent MW detected by the C-terminal antibody R1 (epitope 225–230), and since no similar size bands detected by either #51 (epitope 92–100) or 3F10 (epitope 137–151), it follows that there is a PK-resistant fragment spanning from a position around G194 to the C-terminus. Furthermore, a band detected by R1 (225–230), with an apparent MW of ~3.5, suggests the existence of a second C-terminal PK-resistant fragment, starting around position E206 and spanning to the C-terminus. The absence of a clear equivalent band in the Sypro Ruby-stained gel suggests that the relative abundance of this fragment might be small. Recently, we started to elaborate a generic threading model of PrPSc by distributing PK-cleavage sites, proline residues and other known structural constraints into a 4-rung solenoid [18]. A cleavage site at G194 is compatible with the predicted starting point of the lowermost (C-terminal) rung. However, definitive proof of the identity of these cleavage points should await confirmation by mass spectrometry. Furthermore, given that these cleavage sites have not been detected in GPI-anchorless PrPSc, it remains to be seen whether they are or not a general feature of the architecture of PrPSc.
Ours is not the first structural study of recPrPSc. Recently, Noble et al. probed the structure of an infectious recPrPSc sample by deuterium/hydrogen exchange followed by pepsin digestion and mass spectrometric analysis [47]. They found very substantial protection (i.e., resistance to exchange) in a stretch spanning from position ~89 up to the C-terminus, suggestive of a β-sheet-rich secondary structure. Short stretches exhibiting somewhat lower protection suggest the presence of loops/turns. In particular, the R150-Y154 stretch stands out as the possible location of a loop. The furthermost C-terminal stretch Y224-S230 also shows slightly decreased protection. These results are very similar to those reported by Smirnovas et al. in a similar analysis of GPI-anchorless PrPSc [17]. These authors found substantial protection, indicative of compact, β-sheet-rich structure, from G81 up to the C-terminus, with a lower protection from Y224 to the C-terminus. These results support the notion that the structure of the recPrPSc prepared by Noble et al. is similar to that of GPI-anchorless PrPSc, in agreement with the results reported here. It should be noted that the pattern of exchange of a non-infectious recPrP amyloid sample was very different, with low exchange rates seen only beyond position ~160 [17].
In summary, our studies show that several infectious mouse and bank vole recPrPSc, generated with the concourse of PMCA, exhibit biochemical properties that strongly suggest that they share key architectural elements with brain-derived PrPSc. Furthermore at least in the case of the mouse sample that we have obtained and studied, they seem to feature a mixture of structural properties of “classical” and “atypical” strains of brain PrPSc, although they also show some specific structural nuances. Therefore, we are convinced that such recPrPSc constitute an excellent tool for future additional structural studies. It is noteworthy that cryoEM images of our samples (S7 Fig) as expected, showed fibrils that are very similar to those seen in brain-derived GPI-anchorless PrPSc samples. In summary, recPrPSc samples will be very useful in future structural studies based on the use of NMR.
|
10.1371/journal.ppat.1006964 | Molecularly barcoded Zika virus libraries to probe in vivo evolutionary dynamics | Defining the complex dynamics of Zika virus (ZIKV) infection in pregnancy and during transmission between vertebrate hosts and mosquito vectors is critical for a thorough understanding of viral transmission, pathogenesis, immune evasion, and potential reservoir establishment. Within-host viral diversity in ZIKV infection is low, which makes it difficult to evaluate infection dynamics. To overcome this biological hurdle, we constructed a molecularly barcoded ZIKV. This virus stock consists of a “synthetic swarm” whose members are genetically identical except for a run of eight consecutive degenerate codons, which creates approximately 64,000 theoretical nucleotide combinations that all encode the same amino acids. Deep sequencing this region of the ZIKV genome enables counting of individual barcodes to quantify the number and relative proportions of viral lineages present within a host. Here we used these molecularly barcoded ZIKV variants to study the dynamics of ZIKV infection in pregnant and non-pregnant macaques as well as during mosquito infection/transmission. The barcoded virus had no discernible fitness defects in vivo, and the proportions of individual barcoded virus templates remained stable throughout the duration of acute plasma viremia. ZIKV RNA also was detected in maternal plasma from a pregnant animal infected with barcoded virus for 67 days. The complexity of the virus population declined precipitously 8 days following infection of the dam, consistent with the timing of typical resolution of ZIKV in non-pregnant macaques and remained low for the subsequent duration of viremia. Our approach showed that synthetic swarm viruses can be used to probe the composition of ZIKV populations over time in vivo to understand vertical transmission, persistent reservoirs, bottlenecks, and evolutionary dynamics.
| Understanding the complex dynamics of Zika virus (ZIKV) infection during pregnancy and during transmission to and from vertebrate host and mosquito vector is critical for a thorough understanding of viral transmission, pathogenesis, immune evasion, and reservoir establishment. We sought to develop a virus model system for use in nonhuman primates and mosquitoes that allows for the genetic discrimination of molecularly cloned viruses. This “synthetic swarm” of viruses incorporates a molecular barcode that allows for tracking and monitoring individual viral lineages during infection. Here we infected rhesus macaques with this virus to study the dynamics of ZIKV infection in nonhuman primates as well as during mosquito infection/transmission. We found that the proportions of individual barcoded viruses remained relatively stable during acute infection in pregnant and nonpregnant animals. However, in a pregnant animal, the complexity of the virus population declined precipitously 8 days following infection, consistent with the timing of typical resolution of ZIKV in non-pregnant macaques and remained low for the subsequent duration of viremia.
| Zika virus (ZIKV; Flaviviridae, Flavivirus) infection during pregnancy can cause congenital Zika syndrome (CZS)—a collection of neurological, visual, auditory, and developmental birth defects—in at least 5% of babies [1]. The frequency of vertical transmission is not known, although data suggest that it may be very common, especially if infection occurs during the first trimester [2]. For both pregnant and nonpregnant women, it was previously thought that ZIKV caused an acute self-limiting infection that was resolved in a matter of days. It is now clear that ZIKV can persist for months in other body tissues after it is no longer detectable in blood and in the absence of clinical symptoms [2–7]. During pregnancy, unusually prolonged maternal viremia has been noted, with viral RNA detected in maternal blood up to 107 days after symptom onset [8–11]. The source of virus responsible for prolonged viremia is not known, though it has been speculated that this residual plasma viral load could represent virus genome release from maternal tissues, the placenta, and/or the fetus.
Recently, we established Indian-origin rhesus macaques (Macaca mulatta) as a relevant animal model to understand ZIKV infection during pregnancy, demonstrating that ZIKV can be detected in plasma, CSF, urine, and saliva. In nonpregnant animals viremia was essentially resolved by 10 days post infection [12,13]. In contrast, in pregnant monkeys infected in either the first or third trimester of pregnancy, viremia was prolonged, and was associated with decreased head growth velocity and consistent vertical transmission [2]. Strikingly, significant ocular pathology was noted in fetuses of dams infected with French Polynesian ZIKV during the first trimester [2]. We also showed that viral loads were prolonged in pregnant macaques despite robust maternal antibodies [2]. We therefore aimed to better understand the in vivo replication and evolutionary dynamics of ZIKV infection in this relevant animal model.
To do this, we developed a novel “synthetic swarm” virus based on a pathogenic molecular ZIKV clone that allows for tracking and monitoring of individual viral lineages. The synthetic swarm consists of viruses that are engineered to be genetically identical except for a run of 8 consecutive degenerate nucleotides present in up to ~64,000 theoretical combinations that all encode the same amino acid sequence. This novel barcoded virus is replication competent in vitro and in vivo, and the number and relative proportion of each barcode can be characterized by deep sequencing to determine if the population composition changes among or within hosts. Here and in a companion manuscript by Weger-Lucarelli et al., we demonstrate that this system will provide a useful tool to study the complexity of ZIKV populations within and among hosts; for example, this system can assess bottlenecks following various types of transmission and determine whether non-sterilizing prophylaxis and therapeutics impact the composition of the virus population. Moreover, data from molecularly barcoded viruses will help inform research of ZIKV infection during pregnancy by providing a better understanding of the kinetics of tissue reservoir establishment, maintenance, and reseeding.
Molecular barcoding has been a useful tool to study viruses including simian immunodeficiency virus, influenza virus, poliovirus, Venezuelan equine encephalitis virus, and West Nile virus, establishing conceptual precedent for our approach [14–20]. To generate barcoded ZIKV, we introduced a run of eight consecutive degenerate codons into a region of NS2A (amino acids 144–151) that allows for every possible synonymous mutation to occur in the ZIKV infectious molecular clone (ZIKV-IC) derived from the Puerto Rican isolate ZIKV-PRVABC59 [21]. Following bacteria-free cloning and rolling circle amplification (RCA), linearized and purified RCA reaction products were used for virus production via transfection of Vero cells. All produced virus was collected, pooled, and aliquoted into single-use aliquots, such that single aliquots contain a representative sampling of all genetic variants generated; this barcoded synthetic swarm virus was termed ZIKV-BC-1.0.
We used a multiplex-PCR approach to deep sequence the entire coding genome of the ZIKV-BC-1.0 stock, as well as the ZIKV-IC from which ZIKV-BC-1.0 was derived. For each stock, 1 x 106 viral RNA templates were used in each cDNA synthesis reaction (Table 1), and both stocks were sequenced in duplicate. We identified two nucleotide positions outside of the barcode region that encoded fixed differences between ZIKV-IC and ZIKV-BC-1.0, when compared to the KU501215 reference that we used for mapping. The variant at site 1964 encodes a nonsynonymous change (V to L) in Envelope, and the variant at site 8488 encodes a synonymous substitution in NS5. The variant at site 1964 was also present in our ZIKV-PRVABC59 stock (see [22]), and Genbank contains records for two sequences that match this sequence (accession numbers KX087101 and KX601168) and two that do not (KU501215 and KX377337). In addition, a single nucleotide position in NS5 (site 9581) contained an 80/20 ratio of C-to-T nucleotide substitutions in ZIKV-BC-1.0 but was fixed as a C in ZIKV-IC. The C-to-T change is a synonymous mutation in a leucine codon. There were no other high-frequency variants that differentiate the two stocks outside of the barcode region in the remainder of the genome encoding the polyprotein open reading frame.
We then characterized the diversity of barcode sequences present in the ZIKV-BC-1.0 stock prior to in vitro and in vivo studies. We used three separate approaches to define which barcodes to consider ‘authentic.’ In the first approach (Approach ‘A’), we identified all the distinct non-WT barcodes that were detected in the ZIKV-IC and the ZIKV-BC-1.0 stocks in the region of NS2A encompassing the barcode. We then calculated the arithmetic mean (0.0018%) plus 3 times the standard deviation (0.016%) of the frequency of all the non-WT barcodes present in the two replicates of the ZIKV-IC stock, even if the frequency of a specific barcode in the ZIKV-IC stock was 0%. This threshold frequency was 0.049%. For the second approach (Approach ‘B’), we calculated the arithmetic mean (0.012%) plus 3 times the standard deviation (0.040%) of the frequency of all the non-WT barcodes present only in the ZIKV-IC stock. This threshold frequency was 0.13%. For the final method (Approach ‘C’), we identified the highest frequency of the most common non-WT barcode present in either replicate of the ZIKV-IC stock. This threshold frequency was 0.57%. As this third calculation was the most conservative, we used 0.57% to be the minimum threshold to consider a barcode in ZIKV-BC-1.0 as ‘authentic.’ Using this value, we included 20 sequences in our list of authentic barcodes, and these were followed throughout the study. These barcodes were given independent labels (e.g. Zika_BC01, Zika_BC02, etc.) to simplify reporting. The wild type barcode sequence was also tracked, and it is labeled Zika_WT. All remaining sequences that were detected were labeled as ‘Other’ (see S1–S3 Tables for barcodes identified using all three approaches).
To ascertain whether input RNA template numbers influenced barcode composition, we sequenced a dilution series of viral RNA templates in triplicate (Fig 1 and S4 and S5 Tables). When we used 10,000 or 2000 input vRNA templates, we detected all 20 barcodes. For 500, 250, 100, and 50 input templates, the average number of enumerated barcodes was 17.3 ± 0.9, 14.3 ± 1.2, 12.7 ± 0.5, and 6.7 ± 0.5, respectively. These observations suggest that some barcodes are lost from the population when the number of input templates is reduced.
We also examined diversity and similarity across sequencing replicates in this titration experiment using all the detected sequences, including the sequences labeled as ‘Other.’ Not surprisingly, Simpson’s diversity increased when a greater number of input templates were used, plateauing at 500 input copies (S1 Fig). When comparing similarity across replicates, the samples with 2,000 and 10,000 inputs had the highest Morisita-Horn similarity index (S2 Fig). Unfortunately, it was not possible to obtain a large number of input templates at all timepoints from ZIKV-infected pregnant animals; therefore, the absence of a barcode in sequencing reads from a particular experiment could mean that either the barcode was not present at that timepoint or that it was present in the biological sample but not at a high enough concentration to be detected when sequencing from a small number of templates (S3 Fig).
Prior to use in nonhuman primates, viral infectivity and replication of ZIKV-BC-1.0 was assessed in vitro using Vero, LLC-MK2, C6/36, and Aag2 cells. Viral growth curves were similar between ZIKV-BC-1.0, infectious clone-derived virus (ZIKV-IC), and wild-type ZIKV-PRVABC59 (ZIKV-PR) (S4 Fig and Weger-Lucarelli et al., manuscript submitted). These results suggested that insertion of degenerate nucleotides in the barcode viral genome did not have a significant deleterious effect on either infectivity or replicative capacity in vitro, but we cannot exclude the possibility that different barcodes may have different effects with respect to each other. To confirm that ZIKV-BC-1.0 did not have any replication defects in vivo, we assessed its replication capacity in rhesus macaques. Three rhesus macaques were inoculated subcutaneously with 1 x 104 PFU of ZIKV-BC-1.0. All three animals were productively infected with ZIKV-BC-1.0, with detectable plasma viral loads one day post inoculation (dpi) (Fig 2). Plasma viral loads in all three animals peaked between two and four dpi and ranged from 2.34 x 103 to 9.77 x 104 vRNA copies/ml. Indeed, ZIKV-BC-1.0 displayed viral replication kinetics comparable to ZIKV-IC and ZIKV-PR (Fig 2), and replication kinetics were comparable to previous studies with other strains of ZIKV in nonpregnant rhesus macaques [12,13,23]. To compare overall replication kinetics, the data were log10-transformed and area under the curve (AUC) was calculated. One-way ANOVA then was conducted to compare AUC between groups and the data were not significantly different [F(2,6) = 0.887, p = 0.460] (S6 Table).
We also infected a single pregnant macaque (776301) by subcutaneous inoculation of 1 x 104 PFU of ZIKV-BC-1.0. This animal had been exposed to dengue virus serotype 3 (DENV-3; strain Sleman/78) approximately one year prior to inoculation with ZIKV-BC-1.0. To evaluate cross-reactive neutralizing antibody (nAb) responses elicited by prior exposure to DENV-3 in this animal, serum was obtained prior to inoculation with ZIKV-BC-1.0. Neutralization curves with both DENV-3 and ZIKV revealed that DENV-3 immune sera did not cross-react with ZIKV, whereas DENV-3 was potently neutralized (Fig 3A). The animal then was infected with ZIKV-BC-1.0 at 35 days of gestation (mid-first trimester; rhesus term is 165 ± 10 days) and had detectable plasma viral loads for 67 dpi (Fig 3B); consistent with replication kinetics of wildtype ZIKV in both pregnant macaques [2] and humans [8,9,24]. The animal also had four days of detectable vRNA in urine but no detectable vRNA (Fig 3B) in the amniotic fluid on 22, 36, 50, or 120 dpi (57, 71, 85, 155 days gestation, respectively). By 29 dpi neutralization curves of both viruses revealed a similar profile, indicating the production of a robust maternal nAb response to ZIKV (Fig 3A) coincident with prolonged plasma viral loads, similar to what has been shown previously in other ZIKV-infected pregnant macaques [2]. DENV-3 neutralization curves at 0 and 29 dpi were indistinguishable (Fig 3A).
The pregnancy progressed without adverse outcomes, and at 155 days of gestation, the fetus was surgically delivered, euthanized, and tissues collected. The fetus had no evidence of microcephaly or other abnormalities upon gross examination. Approximately 60 fetal and maternal tissues (see S7 Table for a complete list) were collected for histopathology and vRNA by QRT-PCR. No ZIKV RNA was detected in any samples collected from the fetus. This was surprising, as from seven neonatal macaques we have examined to date (zika.labkey.com), this was the only animal found not to have detectable ZIKV RNA in tissues. Still, ZIKV RNA was detected at the maternal-fetal interface in a section of placental disc (S7 Table). Fetal histology also revealed neutrophilic infiltration of the spleen (Fig 4A), minimal to mild suppurative lymphadenitis of the inguinal lymph node (Fig 4B), minimal multifocal lymphocytic deciduitis, mild multifocal placental infarction with suppurative villositis (Fig 4C and 4D), but normal CNS anatomy, similar to changes noted in previous in utero ZIKV infections [2,25,26]. These data provide indirect evidence that vertical transmission did occur and demonstrate that ZIKV-BC-1.0 is fully functional in vivo with replication kinetics indistinguishable from other ZIKV strains. Thus, inclusion of the barcode did not detectably impair infectivity or replication in adult macaques.
We deep sequenced the viruses replicating in the nonpregnant animals who were infected with ZIKV-BC-1.0 and ZIKV-IC (Fig 5A and 5B, Tables 2 and S8). In each group of three animals, we sequenced viruses at two time points from two animals, and then one time point from a third animal. In animals infected with ZIKV-IC, we found that >95% of sequences in the virus stock and all three animals were wild type across the 24 nucleotides that corresponded to where the barcode was located in ZIKV-BC-1.0.
We counted the number of authentic barcodes detected in the stock and the plasma of the nonpregnant animals infected with ZIKV-BC-1.0. We detected a range of 8 to 20 authentic barcodes in these samples (Fig 5C). We then compared the frequency distribution of the individual barcodes in the plasma of these three animals relative to that in the stock to assess whether there was any evidence for a bottleneck that influenced overall barcode distribution. This was accomplished using two independent statistical approaches. The first compared the frequency distributions by a stochastic equality test, which compares several random pairs of individual values taken from the two samples to test whether there is a significant tendency to get higher values in one over the other [27]. In all three animals, the frequency of each barcode at day two (514982) or at day three (715132 and 688387) was not significantly different from the frequency of each barcode in the stock: p-value based on 1000 bootstrap replications = 0.478, 0.602, and 0.114, respectively. Likewise, the frequency of each barcode at day two (514982) or at day three (715132) was not significantly different from the frequency of each barcode in the stock when compared using the Kolmogorv-Smirnov test: p-value = 0.358 and 0.841. However, this approach did detect a significant difference between the frequency of each barcode in 688387 at day three: p-value = 0.0021, but we believe this to be an artifact of the animal’s viral load at this time point, because the frequency of each barcode at day five did not differ significantly from the frequency of each barcode in the stock, p-value = 0.358. These data in conjunction with the results of the stochastic equality test therefore suggest that there was no evidence for changes in barcode frequency compared to input.
We also examined the sequences outside the barcode region to determine if there were additional nucleotide differences present in the virus population as it replicated in animals. There were small fluctuations in some viral SNPs, but we detected no dramatic shifts in nucleotide frequencies among viruses replicating in vivo, except at site 9581, which is synonymous. In the ZIKV-BC-1.0 stock, there was a mixture of T and C nucleotides (22% and 78% of sequences, respectively) at this site. This position remained a mixture in the animals, but the ratios fluctuated. It dipped to a ratio of 10/90 in animal 688387 at day 5 to as high as 30/70 in animal 715132 at day 5. Overall, there were no new mutations that were detected at greater than 10% frequency in both replicates in the virus populations during the first 5 days after infection in nonpregnant animals. Unfortunately, this site was too far from the barcode (~5000bp) for us to obtain linkage on the same set of paired sequences, making it impossible to know whether this particular nucleotide change was carried on specific barcodes.
We also deep sequenced the barcode in virus populations replicating in the one pregnant animal (776301) infected with ZIKV-BC-1.0. Recognizing that the later time points from this animal had persistent, but low plasma viral loads, we modified our sequencing approach to prepare one tube of cDNA, and then split it into two independent PCR reactions that amplified small fragments (131bp and 178bp) spanning the region containing the barcode (Fig 6A, Tables 3 and S9). We quantified the number of authentic barcodes we detected using the same parameters described in the previous section (Fig 6B). At days 3, 5, and 7, we detected all 20 barcodes. For the remainder of the infection, we detected 8.1 ± 2.3 barcodes. Likewise, barcode diversity, as measured by Simpson’s diversity index, also declined beginning at day 8 and remained low throughout the duration of infection (Fig 6C). Interestingly, some barcodes, such as Zika_BC02, were not detected at later time points, even though it had been present at ~15% during early infection. Other barcodes, such as Zika_BC07, 08, and 09, became more common at later time points, even though they were only present at ~2–5% during early infection. Unfortunately, with such low virus input templates at the late time points, there were differences between replicates indicative of sampling uncertainty. With the exception of two samples (day57_A and day60_B), however, greater than 85% of the sequences matched one of the 20 authentic barcodes.
To begin to understand potential transmission bottlenecks within the vector and the impact they might have on ZIKV population diversity, Aedes aegypti vector competence for ZIKV-BC-1.0 was evaluated at days 7, 13, and 25 days post feeding (PF) from mosquitoes that were exposed to the pregnant macaque at 4 dpi. A single Ae. aegypti (out of 90 tested) was transmission-competent at day 25 PF (Table 4) as measured by plaque assay. Infection efficiency indicates the proportion of mosquitoes with virus-positive bodies among the tested ones. Dissemination efficiency indicates the proportion of mosquitoes with virus-positive legs, and transmission efficiency indicates the proportion of mosquitoes with infectious saliva among the tested ones. All other mosquitoes screened using this methodology were ZIKV-negative. We also found low mosquito infection rates in a previous study exposing mosquitoes to ZIKV-infected rhesus macaques [22]. We deep sequenced virus (viral template numbers added to cDNA synthesis reactions are listed in S10 Table) from all three anatomic compartments from this mosquito (body, leg, and saliva), and we only detected the presence of a single barcode: Zika_BC02. The viral loads in the body, leg, and saliva were 2.57 x 108, 4.73 x 107, and 4.29 x 104 vRNA copies/ml, respectively. Zika_BC02 was present in the pregnant animal’s virus population at ~16% between days 3 and 5 after infection, representing the second most common barcode in the population (Fig 7, Tables 5 and S10).
Mosquito-borne viruses like ZIKV typically exist in hosts as diverse mutant swarms. Defining the way in which stochastic forces within hosts shape these swarms is critical to understanding the evolutionary and adaptive potential of these pathogens and may reveal key insight into transmission, pathogenesis, immune evasion, and reservoir establishment. To date, no attempts have been made to enumerate and characterize individual viral lineages during ZIKV infection. Here, we characterized the dynamics of ZIKV infection in rhesus macaques. Specifically, using a synthetic swarm of molecularly barcoded ZIKV, we tracked the composition of the virus population over time in both pregnant and nonpregnant animals. Our results demonstrated that viral diversity fluctuated in both a spatial and temporal manner as host barriers or selective pressures were encountered and this likely contributed to narrowing of the barcode composition in macaques. For example, the proportions of individual barcoded virus templates remained stable during acute infection, but in the pregnant animal infected with ZIKV-BC-1.0 the complexity of the virus population declined precipitously 8 days following infection of the dam. This was coincident with the timing of typical resolution of ZIKV in non-pregnant macaques (Figs 2 and 3), and after this point the complexity of the virus population remained low for the subsequent duration of viremia (Fig 6C). We speculate that the narrowing of the barcode composition in the pregnant animal was the result of establishment of an anatomic reservoir of ZIKV that is not accessible to maternal neutralizing antibodies, which is shed into maternal plasma at low, but detectable, levels. It also is possible that declining viral barcode diversity was an artifact of a declining viral population size and the consequent effects on sampling, without reservoir establishment. Unfortunately, the absence of ZIKV RNA in the fetus at term prevented us from comparing the barcode composition in the fetus to the barcodes in maternal plasma, so this experiment could not resolve questions related to the potential that the feto-placental unit acts as a tissue reservoir of ZIKV.
There are several factors that could explain the apparent lack of ZIKV RNA in the fetus at term. One possibility is that ZIKV-BC-1.0 was impaired in its ability to traffic to the feto-placental unit due to introduction of the barcode sequence. However, we believe this scenario to be unlikely because the presence of ZIKV-induced pathology in both the fetus and placenta provide indirect evidence for vertical transmission (Fig 4). In addition, the inability to detect ZIKV RNA in affected tissues could be due to the focal nature of infection, assay sensitivity, and/or viral clearance by the time of necropsy. Indeed, resolution of maternal viral loads occurred 91 days prior to necropsy, and Hirsch et al. recently demonstrated that ZIKV infection of the placenta was highly focal and could only be determined by comprehensive biopsy of all placental perfusion domains [26]. Furthermore, although our previous studies suggest high rates of vertical transmission [2], it is unlikely that vertical transmission occurs 100% of the time in humans or macaques. Finally, this animal had pre-existing DENV-3 immunity and it remains unclear what role this may play in subsequent ZIKV infection during pregnancy. Therefore, matching ZIKV barcodes in neonatal tissues with barcodes found in the mother will be important for better understanding vertical transmission. While the ZIKV-BC-1.0 reported here has limited complexity, we have recently developed a new synthetic swarm, ZIKV-BC-2.0, which uses an optimized transfection strategy and has orders of magnitude more putative authentic barcodes. This new virus will be used in future studies in conjunction with deep sequencing techniques that enumerate individual templates with unique molecular identifiers [28]. We therefore expect that future studies of pregnant animals infected with barcoded ZIKV will help distinguish between these possibilities.
In addition to better understanding vertical transmission, synthetic swarm viruses will be useful tools for future studies aimed at understanding persistent reservoirs, bottlenecks, and overall evolutionary dynamics. For example, by using synthetic swarm viruses it should be possible to estimate the effective size of ZIKV populations (Ne) which determines whether selection or genetic drift is the predominant force shaping their genetic structure and evolution [29,30]. Likewise, it should be possible to estimate the number of founder viruses that are required to initiate infection of the fetus during vertical transmission and/or the number of founder viruses required to initiate infection during mosquito-borne versus sexual transmission. Both the number of founder viruses and Ne have not been estimated for any step of the ZIKV infection and/or transmission cycle, but we postulate that a single or limited number of infectious particles likely contribute to the infection of the fetus during vertical transmission. Strong bottlenecks have been observed previously during vertical transmission in plant virus systems [30,31] and during mother-to-offspring transmission of HIV-1 [32,33] and bovine viral diarrhea virus [34]. In these studies, the vast majority of offspring harbored a single or few viral variants, which suggested a stringent population bottleneck associated with vertical transmission. Therefore, knowledge of Ne is of major interest for a better understanding of how virus population structure changes and/or regenerates as it encounters host barriers or selective pressures within and between hosts. Furthermore, barcoded ZIKV will be useful in studies that combine deep sequencing with experimental evolution to observe within host dynamics of ZIKV variants. Barcoded ZIKV is particularly appropriate for studying the effects of evolutionary forces, such as selection and genetic drift, on the emergence of new ZIKV variants that result from host adaptation or that may emerge in the face of new selective pressures: for example, biocontrol strategies, antiviral therapies, immune escape, vaccines, etc. The effects of these evolutionary forces on virus evolution historically have been challenging to address without the inclusion of neutral markers to estimate selection coefficients and Ne.
Although we developed this system to better understand the dynamics of ZIKV infection in the vertebrate host, this approach can be applied to address other questions about ZIKV transmission. For example, ZIKV-BC-1.0 can be used to quantify the bottleneck forces during mosquito infection and transmission. As a result, we also attempted to characterize barcodes present in mosquitoes that fed on the ZIKV-BC-1.0-infected pregnant animal. Consistent with our previous experiments [22], only a single Ae. aegypti became infected with ZIKV-BC-1.0 after feeding on ZIKV-BC-1.0-viremic macaques. This was likely the result of the low amount of infectious virus in macaque blood [35]. We only detected a single barcode during infection of mosquitoes. This is not entirely surprising because mosquitoes ingest small amounts of blood from infected hosts, which limits the size of the viral population founding infection in the vector. For example, it has been previously estimated that as few as 5–42 founder viruses initiate DENV infection of the mosquito midgut [36]. Also, during replication in mosquitoes, flaviviruses undergo population bottlenecks as they traverse physical barriers like the midgut and salivary glands [36,37]. We therefore expected barcode diversity to be low in infected mosquitoes and these data are perhaps indicative of a stringent midgut bottleneck in this individual that limited the variant pool in other anatomic compartments, but this requires further experimental confirmation. Consistent with what we show here, previous work has demonstrated considerable haplotype turnover for West Nile virus in Culex pipiens but not in Ae. aegypti, i.e., haplotypes remained relatively stable as the virus trafficked from the midgut to the saliva [37]. Likewise, Weger-Lucarelli et al., manuscript submitted most often detected only a single barcode in different Ae. aegypti populations that were exposed to ZIKV-BC-1.0 using an artificial membrane feeding system. In sum, our approach showed that synthetic swarm viruses can be used to probe the composition of viral populations over time in vivo to understand vertical transmission, persistent reservoirs, bottlenecks, and evolutionary dynamics.
This study was a proof of concept study designed to examine whether molecularly barcoded ZIKV could be used to elucidate the source of prolonged maternal viremia during pregnancy (Fig 3). Datasets used in this manuscript are publicly available at zika.labkey.com.
This study was approved by the University of Wisconsin-Madison Institutional Animal Care and Use Committee (Animal Care and Use Protocol Number G005401).
Five male and five female Indian-origin rhesus macaques utilized in this study were cared for by the staff at the Wisconsin National Primate Research Center (WNPRC) in accordance with the regulations, guidelines, and recommendations outlined in the Animal Welfare Act, the Guide for the Care and Use of Laboratory Animals, and the Weatherall report. In addition, all macaques utilized in the study were free of Macacine herpesvirus 1, Simian Retrovirus Type D, Simian T-lymphotropic virus Type 1, and Simian Immunodeficiency Virus. For all procedures, animals were anesthetized with an intramuscular dose of ketamine (10ml/kg). Blood samples were obtained using a vacutainer or needle and syringe from the femoral or saphenous vein. The pregnant animal (776301) had a previous history of experimental DENV-3 exposure, approximately one year prior to ZIKV infection.
African Green Monkey kidney cells (Vero; ATCC #CCL-81) were maintained in Dulbecco’s modified Eagle medium (DMEM) supplemented with 10% fetal bovine serum (FBS; Hyclone, Logan, UT), 2 mM L-glutamine, 1.5 g/L sodium bicarbonate, 100 U/ml penicillin, 100 µg/ml of streptomycin, and incubated at 37°C in 5% CO2. Aedes albopictus mosquito cells were (C6/36; ATCC #CRL-1660) were maintained in DMEM supplemented with 10% fetal bovine serum (FBS; Hyclone, Logan, UT), 2 mM L-glutamine, 1.5 g/L sodium bicarbonate, 100 U/ml penicillin, 100 µg/ml of streptomycin, and incubated at 28°C in 5% CO2. ZIKV strain PRVABC59 (ZIKV-PR; GenBank:KU501215), originally isolated from a traveler to Puerto Rico with three rounds of amplification on Vero cells, was obtained from Brandy Russell (CDC, Ft. Collins, CO). Virus stocks were prepared by inoculation onto a confluent monolayer of C6/36 mosquito cells with two rounds of amplification. A single harvest with a titer of 1.58 x 107 plaque forming units (PFU) per ml (equivalent to 2.01 x 1010 vRNA copies per ml) of Zika virus/H.sapiens-tc/PUR/2015/PRVABC59-v3c2 were used for challenges utilizing wild type virus. This virus also served as the backbone upon which the genetically-barcoded virus was generated.
Genetically-barcoded ZIKV was constructed using the ZIKV reverse genetic platform developed by Weger-Lucarelli et al. [21]. The region for the barcode insertion was selected by searching for consecutive codons in which inserting a degenerate nucleotide in the third position would result in a synonymous change. The genetically-barcoded ZIKV clone then was constructed using a novel method called bacteria-free cloning (BFC). First, the genome was amplified as two overlapping pieces from the two-part plasmid system of the reverse genetic platform (see [21]). The CMV promoter was amplified from pcDNA3.1 (Invitrogen). The barcode region was then introduced in the form of an overlapping PCR-amplified oligo (IDT, Iowa, USA). All PCR amplifications were performed with Q5 DNA polymerase (New England Biolabs, Ipswich, MA, USA). Amplified pieces were then gel purified (Macherey-Nagel). The purified overlapping pieces were then assembled using the HiFi DNA assembly master mix (New England Biolabs) and incubated at 50°C for four hours. The Gibson assembly reaction then was treated with Exonuclease I (specific for ssDNA), lambda exonuclease (removes non-circular dsDNA) and DpnI (removes any original bacteria derived plasmid DNA) at 37°C for 30 minutes followed by heat inactivation for 20 minutes at 80°C. Two microliters of this reaction then was used for rolling circle amplification (RCA) using the REPLI-g Mini kit (Qiagen). RCA was performed following the manufacturer's specifications except that 2M trehalose was used in place of water in the reaction mixture because it has been previously shown that this modification reduces secondary amplification products [38]. Reactions were incubated at 30°C for four hours and then inactivated at 65°C for three minutes. Sequence was confirmed by Sanger sequencing.
Virus was prepared in Vero cells transfected with the purified RCA reaction. Briefly, RCA reactions were digested with NruI at 37°C for one hour to linearize the product and remove the branched structure. Generation of an authentic 3’UTR was assured due to the presence of the hepatitis-delta ribozyme immediately following the viral genome [21]. The digested RCA reaction then was purified using a PCR purification kit (Macherey-Nagel) and eluted with molecular-grade water. Purified and digested RCAs were transfected into 80–90% confluent Vero cells using the Xfect transfection reagent (Clontech) following manufacturer’s specifications. Infectious virus was harvested when 50–75% cytopathic effects were observed (6 days post transfection). Viral supernatant then was clarified by centrifugation and supplemented to a final concentration of 20% fetal bovine serum and 10 mM HEPES prior to freezing and storage as single use aliquots. Titer was measured by plaque assay on Vero cells as described in a subsequent section.
The ZIKV-PR stock, ZIKV-IC, and ZIKV-BC-1.0 were thawed, diluted in PBS to 1 x 104 PFU/ml, and loaded into a 3 ml syringe maintained on ice until inoculation. Each of nine nonpregnant Indian-origin rhesus macaques was anesthetized and inoculated subcutaneously over the cranial dorsum with 1 ml ZIKV-PR stock (n = 3), ZIKV-IC stock (n = 3), or ZIKV-BC-1.0 stock (n = 3) containing 1 x 104 PFU. Likewise, the pregnant animal was anesthetized and inoculated via the same route with 1 ml barcoded virus stock containing 1 x 104 PFU. All animals were closely monitored by veterinary and animal care staff for adverse reactions and signs of disease. Nonpregnant animals were examined, and blood and urine were collected from each animal daily from 1 through 10 days, and 14 days post inoculation (dpi). Sampling continued for the pregnant animal until the resolution of viremia.
The Aedes aegypti black-eyed Liverpool (LVP) strain used in this study was obtained from Lyric Bartholomay (University of Wisconsin-Madison, Madison, WI) and maintained at the University of Wisconsin-Madison as previously described [39]. Ae. aegypti LVP are susceptible to ZIKV [40]. Infection, dissemination, and transmission rates were determined for individual mosquitoes and sample sizes were chosen using long established procedures [40–42]. Mosquitoes that fed to repletion on macaques were randomized and separated into cartons in groups of 40–50 and maintained as described in [22]. All samples were screened by plaque assay on Vero cells. Dissemination was indicated by virus-positive legs. Transmission was defined as release of infectious virus with salivary secretions, i.e., the potential to infect another host, and was indicated by virus-positive salivary secretions.
All ZIKV screens from mosquito tissue and titrations for virus quantification from virus stocks were completed by plaque assay on Vero cell cultures. Duplicate wells were infected with 0.1 ml aliquots from serial 10-fold dilutions in growth media and virus was adsorbed for one hour. Following incubation, the inoculum was removed, and monolayers were overlaid with 3 ml containing a 1:1 mixture of 1.2% oxoid agar and 2X DMEM (Gibco, Carlsbad, CA) with 10% (vol/vol) FBS and 2% (vol/vol) penicillin/streptomycin. Cells were incubated at 37 °C in 5% CO2 for four days for plaque development. Cell monolayers then were stained with 3 ml of overlay containing a 1:1 mixture of 1.2% oxoid agar and 2X DMEM with 2% (vol/vol) FBS, 2% (vol/vol) penicillin/streptomycin, and 0.33% neutral red (Gibco). Cells were incubated overnight at 37 °C and plaques were counted.
Macaque serum samples were screened for ZIKV and DENV neutralizing antibody utilizing a plaque reduction neutralization test (PRNT) on Vero cells as described in [43] against ZIKV-PR and DENV-3. Neutralization curves were generated using GraphPad Prism software. The resulting data were analyzed by non-linear regression to estimate the dilution of serum required to inhibit 50% and 90% of infection.
Under real-time ultrasound guidance, a 22-gauge, 3.5-inch Quincke spinal needle was inserted into the amniotic sac. After 1.5–2 ml of fluid were removed and discarded due to potential maternal contamination, an additional 3–4 ml of amniotic fluid were removed for viral qRT-PCR analysis as described elsewhere [2,13]. These samples were obtained at the gestational ages 57, 71, 85, and 155 days. All fluids were free of any blood contamination.
Plasma was isolated from EDTA-anticoagulated whole blood collected the same day by Ficoll density centrifugation at 1860 rcf for 30 minutes. Plasma was removed to a clean 15ml conical tube and centrifuged at 670 rcf for an additional 8 minutes to remove residual cells. Viral RNA was extracted from 300 µL plasma using the Viral Total Nucleic Acid Kit (Promega, Madison, WI) on a Maxwell 16 MDx instrument (Promega, Madison, WI). Tissues were processed with RNAlater (Invitrogen, Carlsbad, CA) according to the manufacturer's protocols. Viral RNA was isolated from the tissues using the Maxwell 16 LEV simplyRNA Tissue Kit (Promega, Madison, WI) on a Maxwell 16 MDx instrument. A range of 20–40 mg of each tissue was homogenized using homogenization buffer from the Maxwell 16 LEV simplyRNA Tissue Kit, the TissueLyser (Qiagen, Hilden, Germany) and two 5 mm stainless steel beads (Qiagen, Hilden, Germany) in a 2 ml snap-cap tube, shaking twice for 3 minutes at 20 Hz each side. The isolation was continued according to the Maxwell 16 LEV simplyRNA Tissue Kit protocol, and samples were eluted into 50 µl RNase free water. RNA was then quantified using quantitative RT-PCR. If a tissue was negative by this method, a duplicate tissue sample was extracted using the Trizol Plus RNA Purification kit (Invitrogen, Carlsbad, CA). Because this purification kit allows for more than twice the weight of tissue starting material, there is an increased likelihood of detecting vRNA in tissues with low viral loads. RNA then was re-quantified using the same quantitative RT-PCR assay. Viral load data from plasma are expressed as vRNA copies/ml. Viral load data from tissues are expressed as vRNA/mg tissue.
At ~155 days gestation, the fetus was removed via surgical uterotomy and maternal tissues were biopsied during laparotomy. These were survival surgeries for the dams. The entire conceptus (fetus, placenta, fetal membranes, umbilical cord, and amniotic fluid) was collected and submitted for necropsy. The fetus was euthanized with an overdose of sodium pentobarbitol (50 mg/kg). Tissues were dissected using sterile instruments that were changed between each organ and tissue type to minimize possible cross contamination. Each organ/tissue was evaluated grossly in situ, removed with sterile instruments, placed in a sterile culture dish, and sectioned for histology, viral burden assay, or banked for future assays. Sampling priority for small or limited fetal tissue volumes (e.g., thyroid gland, eyes) was vRNA followed by histopathology, so not all tissues were available for both analyses. Sampling of all major organ systems and associated biological samples included the CNS (brain, spinal cord, eyes), digestive, urogenital, endocrine, musculoskeletal, cardiovascular, hematopoietic, and respiratory systems as well as amniotic fluid, gastric fluid, bile, and urine. A comprehensive listing of all specific tissues collected and analyzed is presented in S7 Table.
Biopsies of the placental bed (uterine placental attachment site containing deep decidua basalis and myometrium), maternal liver, spleen, and a mesenteric lymph node were collected aseptically during surgery into sterile petri dishes, weighed, and further processed for viral burden and when sufficient sample size was obtained, histology. Maternal decidua was dissected from the maternal surface of the placenta.
Tissues (except neural tissues) were fixed in 4% paraformaldehyde for 24 hours and transferred into 70% ethanol until alcohol processed and embedded in paraffin. Neural tissues were fixed in 10% neutral buffered formalin for 14 days until routinely processed and embedded in paraffin. Paraffin sections (5 µm) were stained with hematoxylin and eosin (H&E). Pathologists were blinded to vRNA findings when tissue sections were evaluated microscopically. Photomicrographs were obtained using a bright light microscope Olympus BX43 and Olympus BX46 (Olympus Inc., Center Valley, PA) with attached Olympus DP72 digital camera (Olympus Inc.) and Spot Flex 152 64 Mp camera (Spot Imaging) and captured using commercially available image-analysis software (cellSens DimensionR, Olympus Inc. and spot software 5.2).
For ZIKV-PR, vRNA from plasma and tissues was quantified by qRT-PCR using primers with a slight modification to those described by Lanciotti et al. to accommodate African lineage ZIKV sequences [44]. The modified primer sequences are: forward 5’-CGYTGCCCAACACAAGG-3’, reverse 5’-CACYAAYGTTCTTTTGCABACAT-3’, and probe 5’-6fam-AGCCTACCTTGAYAAGCARTCAGACACYCAA-BHQ1-3’. The RT-PCR was performed using the SuperScript III Platinum One-Step Quantitative RT-PCR system (Invitrogen, Carlsbad, CA) on a LightCycler 480 instrument (Roche Diagnostics, Indianapolis, IN). The primers and probe were used at final concentrations of 600 nm and 100 nm respectively, along with 150 ng random primers (Promega, Madison, WI). Cycling conditions were as follows: 37°C for 15 min, 50°C for 30 min and 95°C for 2 min, followed by 50 cycles of 95°C for 15 sec and 60°C for 1 min. Viral RNA concentration was determined by interpolation onto an internal standard curve composed of seven 10-fold serial dilutions of a synthetic ZIKV RNA fragment based on a ZIKV strain derived from French Polynesia that shares >99% similarity at the nucleotide level to the Puerto Rican strain used in the infections described in this manuscript.
Virus populations replicating in macaque plasma or mosquito tissues were sequenced in duplicate using a method adapted from Quick et. al. [45]. Viral RNA was isolated from mosquito tissues or plasma using the Maxwell 16 Total Viral Nucleic Acid Purification kit, according to manufacturer’s protocol. Viral RNA then was subjected to RT-PCR using the SuperScript IV Reverse Transcriptase enzyme (Invitrogen, Carlsbad, CA). Theoretical input viral template numbers are shown in Tables 1–3 and 5. For sequencing the entire ZIKV genome, the cDNA was split into two multi-plex PCR reactions using the PCR primers described in Quick et. al with the Q5 High-Fidelity DNA Polymerase enzyme (New England Biolabs, Inc., Ipswich, MA). For sequencing solely the barcode region, individual PCR reactions were performed that either used a primer pair generating a 131 bp amplicon (131F: 5’-TGGTTGGCAATACGAGCGATGGTT-3’; 131R: 5’-CCCCCGCAAGTAGCAAGGCCTG-3’) or a 178bp amplicon (178F: 5’-CCTTGGAAGGCGACCTGATGGTTCT-3’; 178R (same as 131R): 5’-CCCCCGCAAGTAGCAAGGCCTG-3’). Purified PCR products were tagged with the Illumina TruSeq Nano HT kit or the and sequenced with a 2 x 300 kit on an Illumina MiSeq.
Full genome ZIKV sequences generated with the multiplex PCR approach were analyzed using a workflow we termed “Zequencer_2017” (https://bitbucket.org/dholab/zikv_barcode_manuscript_scripts/src). Briefly, sequences were analyzed using a series of custom Python scripts. To characterize the entire ZIKV genome, up to 1000 reads spanning each of the 35 amplicons were extracted from the data set and then mapped to the Zika reference for PRVABC59 (Genbank:KU501215). Variant nucleotides were called using SNPeff [46], using a 5% cutoff. Mapped reads and reference scaffolds were loaded into Geneious Pro (Biomatters, Ltd., Auckland, New Zealand) for intrasample variant calling and differences between each sample and the KU501215 reference were determined. Sequence alignments of the stock viruses can be found in the sequence read archive: ZIKV-IC (accession number: SRX3258286); ZIKV-BC-1.0 (accession number: SRX3258287).
To characterize the barcodes and their frequencies, we developed a workflow called “ZIKV_barcode_analysis” (https://bitbucket.org/dholab/zikv_barcode_manuscript_scripts/src) that makes use of the bbmap suite of tools. Briefly, paired-end reads were merged and quality trimmed using bbmerge. Then, reads containing the barcode were extracted, using bbduk to select reads containing both 20 bp sequences upstream and downstream of the barcode region. These reads were mapped against the Zika reference (GenBank:KU501215) with bbmap, and were then oriented and trimmed so that only the 24 bp barcode remained. Identical barcodes were identified and counted. Custom Python scripts were used to identify authentic barcodes and calculate their frequency in each sample.
The diversities of the sequence populations were evaluated using the Simpson’s diversity index:
Ds=1−∑i=1cni(ni−1)n(n−1)
where ni is the number of copies of the ith unique sequence, c is the number of different unique sequences, and n is the total number of sequences in the sample.
The similarities between pairs of samples were assessed using the Morisita-Horn similarity index:
CMH= 2∑i=1cfigi∑i=1c(fi2+gi2)
where fi = n1i / N1 and gi = n2i / N2, n1i and n2i are the number of copies of the ith unique sequence in samples 1 and 2, and N1 and N2 are the total number of sequences in samples 1 and 2, respectively. The summations in the numerator and the denominator are over the c unique sequences in both samples.
The Simpson’s diversity and Morisita-Horn similarity indices account for both the number of unique sequences and their relative frequencies. These relative diversity and similarity indices range in value from 0 (minimal diversity/similarity) to 1 (maximal diversity/similarity). The Simpson’s diversity index considers a more diverse population as one with a more even distribution of sequence frequencies and the Morisita-Horn similarity index considers populations to be more similar if the higher frequency sequences in both samples are common to both samples and have similar relative frequencies.
Primary data that support the findings of this study are available at the Zika Open-Research Portal (https://zika.labkey.com). Raw FASTQ sequencing data are available at the sequence read archive, accession number: SRP131908. The authors declare that all other data supporting the findings of this study are available within the article and its supplementary information files.
|
10.1371/journal.ppat.1004136 | The Downregulation of GFI1 by the EZH2-NDY1/KDM2B-JARID2 Axis and by Human Cytomegalovirus (HCMV) Associated Factors Allows the Activation of the HCMV Major IE Promoter and the Transition to Productive Infection | Earlier studies had suggested that epigenetic mechanisms play an important role in the control of human cytomegalovirus (HCMV) infection. Here we show that productive HCMV infection is indeed under the control of histone H3K27 trimethylation. The histone H3K27 methyltransferase EZH2, and its regulators JARID2 and NDY1/KDM2B repress GFI1, a transcriptional repressor of the major immediate-early promoter (MIEP) of HCMV. Knocking down EZH2, NDY1/KDM2B or JARID2 relieves the repression and results in the upregulation of GFI1. During infection, the incoming HCMV rapidly downregulates the GFI1 mRNA and protein in both wild-type cells and in cells in which EZH2, NDY1/KDM2B or JARID2 were knocked down. However, since the pre-infection levels of GFI1 in the latter cells are significantly higher, the virus fails to downregulate it to levels permissive for MIEP activation and viral infection. Following the EZH2-NDY1/KDM2B-JARID2-independent downregulation of GFI1 in the early stages of infection, the virus also initiates an EZH2-NDY1/ΚDM2Β-JARID2-dependent program that represses GFI1 throughout the infection cycle. The EZH2 knockdown also delays histone H3K27 trimethylation in the immediate early region of HCMV, which is accompanied by a drop in H3K4 trimethylation that may contribute to the shEZH2-mediated repression of the major immediate early HCMV promoter. These data show that HCMV uses multiple mechanisms to allow the activation of the HCMV MIEP and to prevent cellular mechanisms from blocking the HCMV replication program.
| Human cytomegalovirus (HCMV) is a significant pathogen that belongs to the herpesvirus family. Here we show that the histone H3K27 methyltransferase EZH2 and its regulators JARID2 and NDY1/KDM2B are required for the establishment of productive infection. Mechanistically, the EZH2-NDY1/KDM2B-JARID2 axis downregulates GFI1, a repressor of the HCMV major-immediate-early promoter (MIEP) and inhibition of this axis upregulates GFI1 and interferes with the activation of the MIEP and HCMV infection. GFI1 is rapidly downregulated during infection in both wild-type and EZH2, NDY1/KDM2B, JARID2 knockdown cells. However, since the starting levels of GFI1 in the latter are significantly higher, they remain high despite the virus-induced GFI1 downregulation, preventing the infection. Following the downregulation of GFI1 immediately after virus entry, HCMV initiates an EZH2-NDY1/KDM2B-JARID2-JMJD3-dependent program to maintain the low expression of GFI1 throughout the infection cycle. The knockdown of EZH2 also modulates the accumulation of histone H3K27me3 and H3K4me3 in the immediate-early region of HCMV, and by doing so, it may contribute directly to the MIEP repression induced by the knockdown of EZH2. These data show that HCMV uses multiple mechanisms to allow the activation of the HCMV MIEP and to prevent cellular mechanisms from blocking the HCMV replication program.
| Human cytomegalovirus (HCMV) is a double stranded DNA virus that belongs to the beta-herpesvirus subfamily of the herpesvirus family. Other members of this subfamily are the human herpes viruses 6 and 7 (HHV-6 and HHV-7). HCMV seroprevalence varies widely among populations residing in different geographical regions and among different socioeconomic and age groups [1]. The virus infects many cell types, including fibroblasts, hematopoietic, endothelial, epithelial, smooth muscle and neuronal cells [2]. Most otherwise healthy individuals that are infected with HCMV, experience few if any symptoms. However, some may present symptoms similar to mononucleosis, including fatigue, fever and muscle aches [1]. After the initial infection, the virus enters life-long latency in hematopoietic and endothelial cells, during which the viral genome is maintained as a low-copy number extrachromosomal plasmid. During latency, the productive viral transcription program is almost entirely repressed, with only a subset of latency-associated transcripts being expressed [3]. The Immediate-Early (IE) genes whose expression is a prerequisite for the onset and progression of productive infection remain silenced, and as a result, there is no production of infectious virions. Under specific conditions, the viral genomes can undergo sporadic reactivation, re-initiating a full replicative cycle, which results in virus production and dissemination. Latently-infected individuals are typically asymptomatic. Reactivation of the virus is frequently observed in HIV-infected individuals and in patients undergoing treatment with immunosuppressive or chemotherapeutic drugs [1], [3], [4], although it may also occur in immunocompetent hosts [3]. Virus reactivation may be responsible for debilitating or life-threatening illnesses [1], [3], [4].
The genome of HCMV consists of unique short (US) and unique long (UL) segments both of which are flanked by inverted repeats [1]. Viral gene expression, during HCMV infection, occurs in a temporally regulated manner and it is characterized by three sequential and interdependent waves of transcription. The first wave includes the robust transcription of the immediate-early (IE) genes IE1-72 KDa and IE2-86 KDa, which antagonize and inactivate the host defenses while in addition they induce the expression of the early viral genes. The early genes, expressed in the course of the second wave of transcription, contribute to viral DNA replication, a prerequisite for the activation of the late genes. The latter encode viral structural proteins and are required for virion assembly and virion release from the infected cells. To initiate the transcription of the immediate-early genes, the virus employs cellular transcriptional activators and inhibits cellular transcriptional repressors targeting the major immediate-early promoter (MIEP) [5]. One of the transcriptional repressors targeting this promoter is Growth factor independence 1 (GFI1), a zinc finger protein with a SNAG repressor domain [6], [7]. GFI1 was originally identified as a transcription factor that contributes to the transition of IL-2-dependent T cell lymphoma lines to IL-2 Independence [8]. Today, we know that GFI1 is an important regulator of hematopoietic cell differentiation, contributing to multiple steps in hematopoiesis and lymphopoiesis, (reviewed in [9], [10]). In addition, we know that GFI1 regulates the functional response of macrophages and dendritic cells to Toll like receptor (TLR) signals [11]. At the molecular level, it has been shown that GFI1 is part of a large nuclear complex that includes CoREST, lysine-specific demethylase-1 (LSD1), and HDACs 1 and 2. CoREST and LSD1 associate with GFI1 by binding the GFI1 SNAG repression domain [12].
Immediate-early gene transcription during HCMV infection, or virus reactivation in latently infected cells, depends on the state of differentiation of the target cells [3], [13]-[19]. Undifferentiated cells tend to resist productive infection, suggesting that epigenetic mechanisms, including chromatin modifications and DNA methylation may alter the permissiveness to the virus. Earlier studies addressing this hypothesis, confirmed that immediate-early gene transcription can be altered by the acetylation status of histone H3 associated with the major immediate-early promoter [14], [16], [20]–[25]. In the present study, we focus our attention on the role of histone methylation in HCMV infection and we show that changes in viral infectivity caused by modulation of the chromatin modification machinery of the cell are due to changes in the transcription of the immediate-early genes. Unlike earlier studies however, the present study focuses on the regulation of cellular transcription factors that control the expression the HCMV immediate-early region.
Methylation of histone tails in the promoter region, or the body of a gene, plays a major role in the regulation of gene expression. Histones undergo lysine mono-, di-, or tri-methylation at multiple sites and the functional consequences of histone methylation are site-dependent. Thus, tri-methylation of promoter-associated histone H3 at K4 is a feature of active chromatin, while di-methylation and tri-methylation of histone H3 at K9, or tri-methylation at K27 are features of inactive chromatin. Moreover, mono-, di- and tri-methylation at other sites, such as K36 in the body of a gene, may affect transcriptional elongation and/or RNA splicing (reviewed in [26]). Methylation of core histones at different sites is catalyzed by a host of site-specific methyltransferases. For example, tri-methylation of histone H3 at K27 is catalyzed by EZH2, a component of the polycomb repressor complex 2 (PRC2) [27], [28], whose activity is regulated by several co-factors, including the jumonji domain-containing proteins JARID2 [29] and NDY1/KDM2B [30].
Histone methylation is reversible, with demethylation being catalyzed by a host of site-specific histone demethylases. The first histone demethylase to be identified (LSD1) removes H3K4me1 and H3K4me2 methyl groups, through an oxidative reaction that uses FAD as a co-factor and produces an unstable imine intermediate [31]. The large family of jumonji domain-containing histone demethylases removes lysine methyl groups from a variety of mono-, di- or tri-methylated sites, through an oxidative reaction that uses Iron (FeII) and α-ketoglutarate as co-factors and produces an unstable hydroxymethyl intermediate. Demethylation of histone H3K27me3 is catalyzed by the jumonji domain demethylases, UTX/KDM6A, its homolog UTY, and JMJD3/KDM6B (Reviewed in [32]).
The results of the present study showed that immediate-early gene transcription and HCMV infection of human foreskin fibroblasts (HFFs) depend on histone H3K27 trimethylation, which is under the control of EZH2, JARID2, NDY1/KDM2B and the histone demethylase JMJD3. The EZH2/NDY1/ /JARID2/JMJD3 axis silences GFI1, a repressor of the MIEP of HCMV. Inhibition of this axis therefore, upregulates GFI1 and interferes with the activation of the MIEP and HCMV infection. Immediately after virus entry in virus-infected cells, UV-sensitive virus-associated factors facilitate MIEP activation by promoting the rapid downregulation of GFI1 in both wild-type and NDY1/KDM2B, EZH2 or JARID2 knockdown cells. However, since the levels of GFI1 in the latter cells prior to the infection are significantly higher than in wild-type cells, the HCMV-induced GFI1 degradation fails to downregulate GFI1 to levels permissive for MIEP activation and viral infection. To maintain the silencing of GFI1, the virus also initiates an NDY1/EZH2/JARID2/JMJD3-dependent program, which represses GFI1 throughout the infection cycle. The knockdown of EZH2 may contribute to the repression of the MIEP, also by modulating the accumulation of histone H3K27me3 and H3K4me3 in the immediate early region of HCMV in the first three hours from the start of the viral infection. We conclude that HCMV infection depends on EZH2NDY1/ /JARID2/JMJD3-dependent and independent mechanisms which are activated by the virus and control the expression of GFI1, a transcriptional repressor of the immediate-early region of HCMV. EZH2-dependent mechanisms also control histone modifications in the immediate-early region of HCMV that may contribute to the activation of the MIEP very early in infection.
To determine whether histone methylation plays a role in the efficiency of viral infection and replication, human foreskin fibroblasts (HFFs) were transduced with pLKO.1-based lentiviral constructs of shRNAs of NDY1/KDM2B, EZH2, PHF2 and RBP2/JARID1A/KDM5A, or with the empty vector, prior to infection with HCMV. NDY1/KDM2B, PHF2 and JARID1A/RBP2/KDM5A are Jumonji domain-containing histone demethylases that target histone H3K36me2/me1 (NDY1/KDM2B), K9me1 (PHF2) and K4me3/me2/me1 (RBP2) [32], [33]. EZH2 is a SET domain histone methyltransferase that promotes histone H3K27 tri-methylation [27], [28]. It is important to note that H3K27 trimethylation is also promoted by the demethylase NDY1/KDM2B, which upregulates the expression of EZH2 and contributes to its functional activation [30], [34]. Furthermore, NDY1/KDM2B and EZH2 function in concert on a subset of promoters, which cannot be repressed by either of the two acting alone [30]. The results of the experiment revealed that, whereas the knockdown of PHF2 and RBP2 have no effect on the ability of the virus to infect HFFs, the knockdown of NDY1/KDM2B and EZH2 almost completely block the infection (Fig. 1A). Transduction of HFFs with pBabe-puro-based retroviral constructs of the same histone modifying enzymes, or with the empty vector, had no effect on the efficiency of HCMV infection (Fig. 1B).
To determine whether the resistance of shNDY1/KDM2B and shEZH2-transduced cells to HCMV infection indicates a block or a mere delay of the infection, we monitored the virus titer in the supernatants of infected cells every other day for up to 9 days post-infection. The results confirmed that the knockdown of NDY1/KDM2B or EZH2, but not PHF2 or RBP2, block the infection (Fig. 1C).
The preceding experiments revealed that NDY1/KDM2B and EZH2 are both required for HCMV infection and replication, while other histone modifying enzymes are not. Given that NDY1/KDM2B and EZH2 operate in concert to upregulate EZH2 and histone H3K27 trimethylation [30], [34], we hypothesized that it is histone H3K27 trimethylation that is required for efficient infection by HCMV. To address this hypothesis, human foreskin fibroblasts were transduced with a pBabe-puro-based retroviral construct of the H3K27me3 demethylase JMJD3, or with a pLKO.1-based lentiviral shRNA construct of the same enzyme. Cells transduced with these constructs or with the empty vectors, were infected with HCMV and the efficiency of infection was determined qualitatively, as well as quantitatively, using a plaque assay for virus titration. The results showed that whereas shJMJD3 did not interfere with viral infection and replication, JMJD3, which demethylates histone H3K27me3, did (Fig. 1D).
EZH2 is known to form a complex with the prototype Jumonji domain protein JARID2 and to bind chromatin in concert with JARID2 [29]. Through this interaction JARID2 regulates the EZH2 methyltransferase activity. We therefore proceeded to address the role of JARID2 in HCMV infection and replication in HFFs. The experiments in Figure 1D showed that whereas tranduction with a pLKO.1-based lentiviral construct of shJARID2 interferes with HCMV infection and replication, transduction with a pBabe-based construct of JARID2 does not.
If histone H3K27 tri-methylation is required for HCMV infection, as suggested by the preceding experiments, replacement of the endogenous EZH2 with its SET domain mutant, which lacks histone methytransferase activity, should fail to rescue the permissiveness of shEZH2-transduced HFFs to HCMV. To address this hypothesis, we knocked down the endogenous EZH2 with an shRNA that targets sequences in the 3′ UTR of the endogenous EZH2 mRNA, and we replaced it with wild-type exogenous EZH2 or with a catalytically-inactive ΔSET mutant of EZH2. Titration of HCMV in these cells confirmed the hypothesis (Fig. 1E). We conclude that the histone methyltransferase activity of EZH2 is required for infection of human foreskin fibroblasts by HCMV.
To explore the mechanism by which histone H3K27 tri-methylation controls viral infection and replication, we first examined whether the knockdown of NDY1/KDM2B, EZH2 or JARID2, inhibits viral entry. To this end, HFFs transduced with pLKO.1 or pLKO.1-based shNDY1/KDM2B, shEZH2 or shJARID2 constructs were infected with the recombinant virus UL32-EGFP-HCMV-TB40, which expresses the capsid-associated tegument protein pUL32 (pp150) as a fusion with EGFP (MOI 10 PFU per cell), producing fluorescent virions as well as tegument puncta in the cytoplasm and the nucleus of infected cells [35]. One hour after infection, the cells were fixed and the viral entry was visualized by EGFP fluorescence, which monitors the UL32-EGFP-containing viral particles. Quantitative analysis of the UL32-EGFP showed that the efficiency of viral entry in cells transduced with the empty vector and in cells transduced with the shRNA constructs was similar (Fig. S1A and S1B). In a repeat of the experiment with HCMV AD169, the viral strain used in all other experiments in this study, we examined the efficiency of viral entry by monitoring intracellular pp65 by immunofluorescence. The results confirmed that viral entry is not affected by the knockdown of EZH2, NDY1/KDM2B or JARID2 (Fig. S1C and S1D). Parallel experiments employing quantitative real time PCR to measure HCMV genomic equivalents in nuclear and cytoplasmic lysates of HFFs transduced with the same shRNA constructs and isolated 6 hours after infection, also revealed no differences (Fig. S1E and S1F). The preceding data combined, suggest that H3K27 tri-methylation does not affect viral entry and thus, the resistance to infection caused by the inhibition of H3K27 trimethylation is due to a barrier at a different step of the infection cycle.
To determine whether histone modifying enzymes regulate the activation of the immediate-early gene promoter, cells transduced with shRNAs or expression constructs of these enzymes, or with the corresponding empty vectors, were infected with HCMV and 5 hours later, they were stained for IE1 and counterstained with DAPI. Flow cytometry and fluorescence microscopy of the stained cells illustrated that NDY1/KDM2B and EZH2 are required for HCMV immediate-early gene expression (Fig. 2A and Fig.S2A). The knockdown of PHF2 and RBP2, which are not required for HCMV infection, also had no effect on IE1 promoter activity (Fig. 2A). The results of these experiments suggested that H3K27 trimethylation is required for HCMV immediate-early gene expression. Parallel experiments provided additional support to this conclusion by showing that both the knockdown of JARID2 and the overexpression of the histone H3K27me3 demethylase JMJD3 also inhibit the activation of the HCMV MIEP (Fig. 2B and Fig.S2B).To determine whether blocking histone H3K27 trimethylation inhibits or simply delays the expression of IE1, we knocked down EZH2, or NDY1/KDM2B, in HFFs and we examined the expression of IE1 at multiple time points from the start of the infection. The results demonstrated that these manipulations interfere with the expression of IE1, at all the time points (Fig. 2C). In agreement with these data, treatment of human foreskin fibroblasts with the EZH2 inhibitor 3-Deazaneplanocin A (DZNep) [36]-[38] inhibited both the expression of immediate-early genes and the infection by HCMV (Fig. 2D–F).
In parallel experiments, we monitored the abundance of EGFP-positive cells and the intensity of EGFP fluorescence by fluorescence microscopy, or flow cytometry in HeLa cells transduced with shEZH2 or shNDY1/KDM2B lentiviral constructs, and transfected with a reporter construct expressing EGFP from the MIEP of HCMV. The results showed that the knockdown of either NDY1/KDM2B or EZH2 inhibits the activity of the MIE promoter even in cells not infected with HCMV (Fig.S3). We conclude that H3K27 trimethylation is both necessary and sufficient for the activation of this promoter.
Since H3K27me3 normally represses transcription [39], we hypothesized that it may activate the major immediate-early promoter by repressing a transcriptional repressor. To address this hypothesis, we examined whether the knockdown of NDY1/KDM2B, EZH2, or JARID2 and the overexpression of JMJD3 in HFFs, alter the expression of known transcriptional regulators of the MIEP (Fig.3A). Real time RT-PCR revealed that GFI1 is the only MIEP repressor significantly upregulated in these cells (Fig. 3A). Probing western blots of lysates of the same cells with an anti-GFI1 antibody showed that GFI1 is upregulated also at the protein level (Fig. 3A).
GFI1 represses the MIE promoter of HCMV by binding to two sites within the promoter [6] (Fig.3B, Upper panel). The binding of GFI1 to these sites was tested with experiments using foreskin fibroblasts transduced with pLKO.1 or with the pLKO.1-based shNDY1/KDM2B, shEZH2, or shJARID2 constructs, which de-repress GFI1 in these cells (Fig. 3A). These cells were infected with HCMV AD169. ChIP experiments, using cell lysates harvested 1 hour after infection, confirmed the binding of GFI1 to both GFI1 binding sites in the MIE promoter, but not in the MIE coding region and showed that treatments promoting the upregulation of GFI1 by inhibiting H3K27 trimethylation increase the binding (Fig 3B Lower panel). To determine the functional role of GFI1 binding, we mutated both sites by site-directed mutagenesis (AATC mutated to AACT and AAGT, respectively) of an MIEP-EGFP reporter construct. The wild-type and mutant constructs were transfected into HEK 293T cells that had been stably transduced with pLKO1, shEZH2, shJARID2, shNDY1/KDM2B, GFI1, or shGFI1. Analyzing the cells by fluorescence microscopy, revealed that whereas the knockdown of EZH2, JARID2, or NDY1/KDM2B, and the overexpression of GFI1 inhibit the wild-type promoter, they have no effect on the mutant promoter (Fig 3C). These data combined, confirmed that histone H3K27 tri-methylation represses GFI1. Inhibiting H3K27 trimethylation de-represses GFI1, which binds and represses the MIE promoter of HCMV.
The preceding data raised the question of the mechanism by which the tri-methylation of histone H3 at K27 regulates the GFI1 promoter. This question was addressed with ChIP assays designed to measure the relative abundance of H3K27me3 at five promoter sites (from position −1044 bp to position −209 bp), in cells transduced either with the empty lentiviral vector, or with shRNA lentivirus constructs targeting EZH2, NDY1/KDM2B or JARID2. P16INK4a was used as the positive control. The most 5′ of the five promoter sites (site #1, located between −1044 bp and −956 bp), maps within a repressive domain [40]. These experiments revealed that the knockdown of any of these chromatin regulators induced a significant decrease in the abundance of H3K27me3 at this site (Fig. 3D). We conclude that EZH2, NDY1/KDM2B and JARID2 promote histone H3K27 trimethylation within a negative regulatory domain of the GFI1 promoter and may be responsible for the transcriptional repression function previously mapped within this domain [40].
Based on the data in the preceding paragraphs, we conclude that GFI1 is a direct repressor of the HCMV MIEP. However, it is possible that GFI1 may regulate the MIEP by additional indirect mechanisms. One such mechanism is via p21CIP1/WAF1 which is a known target of the GFI1 transcriptional repressor [41]. The repression of GFI1 by NDY1/EZH2/JARID2 may lead to the upregulation of p21CIP/WAF1, which in turn, may inhibit the progression from the G1 phase to the S phase of the cell cycle. Since the accumulation of cells in G1 favors the expression of the IE genes of HCMV [42], [43], GFI1 may regulate the MIEP not only directly, but also indirectly via p21CIP/WAF1. To address this question, HFFs were transduced with pLKO1-based lentiviral constructs of shEZH2 or shNDY1/KDM2B, or with the empty pLKO1 lentiviral vector. Western blots of lysates of these cells harvested before, and at various time points after infection and probed with an anti-p21CIP/WAF1 antibody, revealed that the expression of p21CIP/WAF1 is not affected by the knockdown of either EZH2 or NDY1 (Fig.S4). These data suggest that p21CIP/WAF1 is not involved in the regulation of the MIEP of HCMV by NDY1/EZH2/JARID2.
Earlier studies had shown that the resistance to HCMV caused by the repression of the MIEP can be overcome by infection at a high moi [44]–[47]. Based on this observation, we predicted that HCMV infection of shEZH2, shNDY1, shJARID2, or JMJD3-transduced HFFs at an moi of 5 could overcome their resistance to infection. The results confirmed the prediction (Fig. S5), providing additional support to the hypothesis that the HCMV phenotype induced by the knockdown or overexpression of these molecules is caused by the repression of the MIEP.
The preceding data showed that by repressing the MIE promoter, GFI1 can block HCMV infection. Based on these data, we hypothesized that HCMV may down-regulate GFI1, to increase the permissiveness of the cells to the incoming virus. Experiments addressing this hypothesis showed that GFI1 is indeed down-regulated rapidly in HCMV-infected cells, both at the RNA and protein levels (Fig. 4A and 4B). The rapid downregulation of GFI1 in HCMV-infected cells suggested that the incoming virus may induce the degradation of both the GFI1 mRNA and protein. This hypothesis was addressed with the experiments in Figure 4B. Monitoring the levels of the GFI1 mRNA in Actinomycin D-treated HFFs by real time RT-PCR, confirmed that virus infection accelerates the degradation of the GFI1 mRNA (Fig. 4C). Similarly, monitoring the levels of the GFI1 protein in MG132-treated cells by western blotting showed that MG132 stabilizes the expression of GFI1 in virus-infected cells (Fig. 4D). These data suggested that the incoming virus renders the cells permissive to infection by rapidly degrading GFI1 both at the RNA and the protein levels, and that protein degradation is mediated through the ubiquitin-proteasome pathway. However, since inhibition of the proteasome is also known to stabilize hDaxx [17], [44], it is possible that MG132 may block the degradation of GFI1 by inhibiting the degradation of hDaxx and viral infection.
The fact that both the GFI1 RNA and the GFI1 protein were degraded rapidly after virus infection, suggests that both processes are initiated by factors entering the cells with the incoming virus. To determine the nature of these factors, we infected the cells with UV-irradiated virus and we examined the expression of GFI1 at 0, 0.5, 1 and 2.5 hours from the start of the exposure to the virus. The results showed that the UV-irradiated virus induced the degradation of hDaxx as expected [44] (FigS6A), but failed to downregulate GFI1 at both the RNA and protein levels (Fig.S6B and S6C). The downregulation of the GFI1 mRNA may be mediated by virion-associated UV-sensitive non-coding RNAs that target GFI1. The downregulation of the GFI1 protein may be mediated by another UV-sensitive virion-associated molecule, whose nature remains to be determined. However, we have not formally excluded that a de novo expressed protein may be responsible for the phenotype.
The regulation of GFI1 by EZH2, NDY1/KDM2B or JARID2, prompted us to investigate whether the HCMV-induced GFI1 downregulation is EZH2/NDY1/JARID2-dependent. To address this question, HFFs were transduced with the lentiviral vector pLKO.1, or with pLKO.1-based constructs of shNDY1, shEZH2, or shJARID2, and 48 hours later, they were infected with HCMV. Western blotting of uninfected and HCMV-infected cell lysates, harvested two hours after the infection, revealed that the downregulation of GFI1 was not prevented by the knockdown of any of these chromatin regulators. However, since the starting levels of GFI1 prior to the infection in shEZH2 shNDY1, or shJARID2-transduced cells were significantly higher than in control cells, GFI1 continued to be expressed, even after the infection (Fig.4E). Given the inhibitory effects of the NDY1/EZH2/JARID2 knockdown on the activity of the MIEP, we conclude that the levels of GFI1 detected in these cells are sufficient to repress the promoter.
Although the rapid downregulation of GFI1 in the initial stages of viral infection may be independent of the NDY1/EZH2/JARID2 axis however, the initial downregulation of GFI1 may be maintained via the activation of this axis throughout the infection cycle. Real time RT-PCR indeed showed that the mRNA levels of NDY1/KDM2B, EZH2 and JARID2 increase gradually, while the RNA levels of JMJD3 decrease in the course of the viral infection (Fig.4F). UV-irradiated virus, which cannot establish a productive infection, had no effect on the expression of these epigenetic regulators (Fig.S7).
To determine whether HCMV infection depends on the down-regulation of GFI1, human foreskin fibroblasts were transduced with pBabe-puro-based retroviral constructs of GFI1 or GFI1B, a GFI1-related gene, also encoding a SNAG domain-containing transcriptional repressor, or with the empty vector. The transduced cells were infected with HCMV. Infection was monitored by light microscopy 5 days later (Fig.S8) and the progeny virus was harvested 12 days later and titrated by a plaque assay (Fig.S8). Alternatively, transduced cells were infected with HCMV and they were stained for IE1 expression 5 hours post-infection. Stained cells were analyzed by flow cytometry (Fig.S8). The results revealed that GFI1 inhibits the activity of the MIEP and HCMV infection as expected, while GFI1B does not. Since GFI1B also represses p21Cip/WAF1 [48], these results provide additional support to the conclusion that p21CIP/WAF1 is not involved in the regulation of the MIEP of HCMV by NDY1/EZH2/JARID2/GFI1 (see above).
In parallel experiments, we used a TRIPZ-based doxycycline-inducible shRNA construct of EZH2 and a pLKO1-based constitutive shRNA construct of GFI1 to knock down these genes in HFFs, separately or in combination. The knockdown of EZH2 and GFI1 were confirmed by western blotting of lysates harvested from these cells prior to HCMV infection, before and after treatment with doxycycline (Fig.5A1). Infection of these cells with HCMV showed that whereas the knockdown of EZH2 inhibits IE1 expression (Fig. 5A2 lanes 1 and 3) and permissiveness to infection (Fig5A3, first and third bar), the knockdown of GFI1 does not affect either (Fig.5A2, lanes 1 and 2 and5A3, bars 1 and 2). However, the knockdown of GFI1, partially restored IE1 expression and permissiveness to viral infection in cells in which EZH2 was also knocked down (Fig.5A2, lanes 3 and 4 and Fig5A3, bars 3 and 4). We conclude that EZH2 contributes to HCMV infection by inhibiting the expression of GFI1. The fact that the restoration of IE1 expression and permissiveness to viral infection were only partial, suggests that EZH2 may have additional GFI1-independet effects on IE1 expression.
Next we examined the effects of the EZH2 knockdown on histone H3K27 and H3K4 trimethylation in the enhancer, the cis repression sequence (crs) and intron 1 in the immediate-early region of HCMV (Fig.6). The peak of H3K27 trimethylation in the enhancer and in intron 1 in HFFs transduced with a lentiviral shControl construct was observed at 1.5 hours from the start of the exposure to the virus and declined to very low levels at the 3 hour time point. H3K27 trimethylation in the crs increased more rapidly (0.5 hours) and remained high throughout the observation period. Knocking down EZH2 delayed H3K27 trimethylation in the crs, and perhaps in intron 1, with low levels of H3K27me3 at the 0.5 hour time point, and in the enhancer, with low levels of H3K27me3 at the 1.5 hour time point. More important, in the shEZH2 cells H3K27 trimethylation remained high at the three hour time point in all three sites. These changes in the abundance of H3K27me3 were associated with parallel changes in the abundance of H3K4me3. In HFFs transduced with the shControl construct, the abundance of H3K4me3 increased in all three sites throughout the three hour observation period. However, in the shEZH2-transduced cells, its abundance in the enhancer region increased more slowly than in control cells. Moreover, in the crs and the intron 1 regions its abundance declined at the three hour time point, with the decline in intron 1, being dramatic. These data are consistent with the chromatin modification data in the IE region of the murine CMV in the immediate-early stage of the infection [49]. In addition, they are consistent with earlier observations suggesting an initial cell-mediated MIEP repression that precedes viral gene expression in HCMV-infected cells [16]. More important, these data suggest that H3K27 and H3K4 trimethylation in the regulatory elements of the IE region of HCMV are co-ordinatelly regulated. However, the rules of their co-ordinate regulation are not yet known and they will require additional work to be determined. One additional question that remains is how the regulatory elements of the IE region of HCMV undergo delayed H3K27 trimethylation, when EZH2 is knocked down. We hypothesize that this may be happening because of residual EZH2 activity, remaining after the EZH2 knockdown. Alternatively, it may be mediated by EZH1. This question will be addressed in future studies.
Data presented in this report, showed that NDY1/KDM2B, EZH2 and JARID2 synergize to repress GFI1, a SNAG domain-containing transcriptional repressor [6]–[8]. In addition, they confirmed that GFI1 represses the MIEP of HCMV by binding to two sites, 159–163 and 105–109 base pairs upstream of the transcription start site, and that the knockdown of NDY1/KDM2B, EZH2 or JARID2 results in the upregulation of GFI1 and in the GFI1-dependent dramatic repression of the MIEP of HCMV. During HCMV infection, the GFI1 protein and mRNA are downregulated rapidly, most likely via degradation, both in control cells and in cells in which NDY1/KDM2B, EZH2 or JARID2 was knocked down. However, the pre-infection levels of GFI1 in the latter cells are significantly higher than in the control cells, and the degradation is not sufficient to extinguish GFI1 expression, which is required for the establishment of HCMV infection. As a result, cells in which NDY1/KDM2B, EZH2 or JARID2 were knocked down, are resistant to HCMV infection. Following the initial degradation of GFI1, the virus reprograms the epigenetic machinery of the cell, by up-regulating NDY1/KDM2B, EZH2 and JARID2 and by down-regulating the histone H3K27me3 demethylase JMJD3. This reprogramming is expected to maintain the expression of GFI1 at low levels throughout the infection cycle (Fig. 7).
The combination of NDY1/KDM2B, EZH2 and JARID2 promotes histone H3K27 trimethylation, a chromatin mark associated with transcriptional repression [50], [51]. EZH2, the enzyme responsible for histone H3K27 trimethylation, binds JARID2, a non-canonical jumonji domain protein that regulates the EZH2 methyltransferase activity [29]. NDY1/KDM2B, which can be induced by growth factors such as FGF2 [30], binds some EZH2 target genes and demethylates histone H3K36(me2) and H3K36(me1) [34]. The latter promotes EZH2 binding to the same genes and transcriptional repression [32]. In this report, we presented evidence that GFI1, a transcriptional repressor of the major immediate-early promoter of HCMV [6], [52], is one of the genes targeted by the H3K27(me3) methyltransferase EZH2 and its regulators JARID2 and NDY1/KDM2B, as well as the histone H3K27(me3) demethylase JMJD3.
HCMV initiates viral infection by targeting GFI1 via multiple mechanisms. Immediately after exposure to the virus, the GFI1 mRNA and protein are rapidly downregulated, most likely via degradation. The rapid drop in the levels of these molecules immediately after exposure to the virus suggested that they may be degraded by virion-associated factor(s). Their downregulation was UV-sensitive, suggesting that it may be due to degradation by virion-associated nucleic acids [53], [54]. The GFI1 mRNA may be a direct target of virion-associated non-coding RNAs [55], [56]. The GFI1 protein may be degraded via the proteasome, which is known to play an important role in the transcription of the HCMV immediate early genes [57], [58]. However, the mechanism by which virion-associated nucleic acids may regulate the proteasomal degradation of the GFI1 protein remains to be determined. Over the years, the main focus of studies addressing the activation of the proteasome by HCMV is on the tegument protein pp71 [59]. pp71 interacts with hDaxx in PML bodies to inhibit hDaxx-mediated silencing by promoting its degradation [60]-[63]. However, GFI1 cannot be a target of pp71, because the latter is not UV-sensitive.
The rapid downregulation of the GFI1 mRNA and protein, which occurs immediately after exposure to the virus, is followed by epigenetic reprogramming, which is expected to downregulate GFI1 throughout the infection cycle. Thus, in the early stages of infection, HCMV employs mechanisms that have not yet been determined to up-regulate NDY1/KDM2B, EZH2 and JARID2 and to down-regulate JMJD3. These chromatin modifiers target an inhibitory domain within the GFI1 promoter and enhance the trimethylation of histone H3 at K27 (Fig.3D). It is not yet known how these chromatin regulators are targeted to the GFI1 promoter. Potentially, they may function as GFI1 co-repressors and they may be targeted to the GFI1 promoter by GFI1 itself. This is suggested by earlier findings showing that GFI1 represses its own promoter [64] and by the observation that within the inhibitory domain of the human GFI1 promoter [40], there is one GFI1 binding site, containing the characteristic AATC core.
The effects of the knockdown of EZH2, NDY1/KDM2B and JARID2, along with the effects of the overexpression of JMJD3, on the activity of the MIEP and viral infection were counterintuitive. One would expect that these genetic manipulations would lead to a decrease in H3K27 trimethylation, both globally and regionally in the MIEP, and that this would result in an increase in MIEP activity. The fact that we see the opposite would suggest that either these genetic manipulations fail to alter the balance of repressive and activating epigenetic marks in the MIEP, or that tipping the balance toward the activating marks is not sufficient to override the effects of the GFI1 repressor. Of course, it is also possible that changes in the pattern of MIEP-associated chromatin modifications, induced by these genetic manipulations, promote the binding of GFI1, facilitating the MIEP repression. To address these questions, we surveyed the effects of the EZH2 knockdown on the abundance of H3K27me3 (a repressive mark) and H3K4me3 (an activating mark) in the enhancer, crs and intron 1 in the IE region of HCMV in the first three hours from the start of the infection. The results showed a delay in H3K27 trimethylation. The slow kinetics of this process resulted in an increase in the abundance of H3K27me3 in the IE enhancer and intron 1 at the three hour time point, when the abundance H3K27me3 normally decreases. The rapid increase in the abundance of H3K27me3, which we observed in HCMV-infected control cells in the very early stages of the infection, is consistent with the results of earlier studies showing that MIEP activation during infection by murine CMV and HCMV is preceded by an increase in the abundance of repressive histone marks. Our data also showed that the increase in the abundance of in H3K27me3 at the three hour time point in shEZH2-transduced cells is paralleled by a significant decrease in the abundance of H3K4me3,.suggesting that repressive and activating marks are co-ordinately regulated. The rules of this coordinate regulation and the potential involvement of these epigenetic modifications in the recruitment of GFI1 remain to be determined.
Overall, the data presented in this report identify a novel pathway of epigenetic regulation of cellular gene expression that regulates the expression of HCMV immediate-early genes and viral infection. Inhibition of the pathway may have preventive or therapeutic applications in viral infection, while selective activation of the pathway may have therapeutic applications in cancer.
Human Foreskin Fibroblasts (HFFs) were used for HCMV infection (kind gift from Dimitrios Iliopoulos, Harvard Medical School) and HEK 293T cells were transfected to package lentivirus and retrovirus constructs. HEK 293T cells or HELA cells were used for transfection of HCMV MIEP reporter constructs in experiments addressing the regulation of the HCMV major immediate-early promoter by chromatin modifying enzymes. All cell lines were maintained in Dulbecco's modified Eagle's minimal essential medium (DMEM) supplemented with 10% fetal bovine serum, penicillin/streptomycin, L-glutamine and non-essential amino acids.
The wild-type laboratory strain of HCMV we used was the AD169 strain. The recombinant UL32-EGFP-HCMV-TB40 virus, which expresses the capsid-associated tegument protein pUL32 (pp150), fused to EGFP [35] was used for some experiments addressing viral entry.
To determine the role of the EZH2 enzymatic activity on immediate-early gene transcription and HCMV infection, virus-infected cells were treated with the EZH2 inhibitor 3-deazaneplanocin A (DZNep) (Cayman Chemical Company, MI), at the final concentration of 10 µM. DZNep was dissolved in dimethyl sulfoxide (DMSO).
To infect HFFs with HCMV, cell monolayers were incubated with the virus at a multiplicity of infection (MOI) of 0.5 PFU/cell, or at variable multiplicities in virus titration experiments.Unless otherwise specified, the cells were exposed to the virus for 2 hours at 37°C. Subsequently, the virus was removed and replaced with fresh medium. Plaque assays for virus titration were performed on HFFs according to standard protocols [65]. To monitor the growth of HCMV in HFFs transduced with shEZH2, shNDY1/KDM2B, shPHF2 and shRBP2 lentiviral constructs, cells were infected with HCMV AD169 at an MOI of 0.5 PFU/cell. Viral supernatants harvested from these cultures every two days for 9 days were titrated, using plaque assays.
To measure the efficiency of viral entry, HFFs transduced with pLKO.1-based lentiviral constructs of shEZH2, shNDY1/KDM2B, shJARID2, or with the empty vector, were infected with the UL32-EGFP-HCMV-TB40 recombinant virus at an MOI of 10 PFU/cell, as previously described [66]. One hour after infection, the cells were fixed and viral entry was visualized by monitoring intracellular EGFP fluorescence via fluorescence microscopy. Fluorescence intensities of UL32-EGFP were calculated with the Zeiss LSM image examiner software. To correct for background fluorescence, we deduced from the fluorescence of infected cells the fluorescence of adjacent non-infected cells. Alternatively, viral entry was monitored by real-time PCR of viral DNA in cell lysates harvested at six hours from the start of the exposure to the wild-type HCMV AD169 virus [45].
The retrovirus and lentivirus constructs we used are listed in Table 1. Human JARID2 was cloned into the pLenti-CMV-puro-DEST vector (Addgene, cat no 17452) using the LR Clonase II Plus enzyme mix (Invitrogen, cat no 12538120) according to the manufacturer's instructions. Human HA-PHF2 was cloned into the EcoRI site of pBABEpuro, and human FLAG-JMJD3-HA was cloned between the BamHI and XhoI sites of the same vector. The rest of the lentiviral and retroviral constructs were either purchased or kindly provided by others (Table 1). shEZH2 was induced in TRIPZ-shEZH2-transduced cells with Doxycycline (1 µg/ml)
Jarid2 cDNA was PCR-amplified using the pCMV-SPORT6-JARID2 (Open Biosystems, cat. no. MHS1010-99622028) as template, cloned in the pENTR/TOPO vector (Invitrogen, cat no K2400-20). The sequence was verified and was recombined to the pLentipuro vector (Addgene, cat no. 17452) using the Gateway LR Clonase II Plus kit (Invitrogen, cat no. 12538-120).
To determine the effects of NDY1/KDM2B, EZH2 JARID2 and GFI1 on the activity of the HCMV MIEP in the absence of viral infection, a MIEP-EGFP reporter construct (pEGFP-C1) (Clontech) was transfected into HEK 293T cells or their derivatives in which NDY1/KDM2B, EZH2 or JARID2 were knocked down or GFI1 was overexpressed and the expression of EGFP was monitored by fluorescence microscopy or flow cytometry. The same cells were also transfected with a derivative of pEGFP-C1 in which the two GFI1 binding sites in the MIEP [7] were inactivated by point mutation. The mutant construct was generated by site-directed mutagenesis, as previously described [67], using primers: CMVmut1: 5'-GGGTGGAGACTTGGAAAGTCCCGTGAGTCAAACCG-3′ and CMVmut2: 5'-ATTTTGGAAAGTCCCGTTAGTTTTGGTGCCAAAACAAAC-3′. All products of PCR mutagenesis were sequenced after cloning, to ensure that no additional mutations were generated.
HEK 293T cells were transiently co-transfected with retroviral constructs and the amphotropic packaging construct (Ampho-pac). Alternatively, HEK 293T cells were transiently co-transfected with lentiviral constructs and pCMV/VSV-G (where VSV-G is vesicular stomatitis virus protein G) and pCMV-dR8.2 dvpr. Transfection was carried out using Fugene 6 (Roche Applied Science).
To transduce HFFs with the packaged viruses, early passage cells were incubated with viral supernatants in the presence of 5 µg/ml polybrene (Sigma-Aldrich, Deisenhofen, Germany) for 24 hours. Forty-eight hours later, cells were selected with puromycin (2 µg/ml) or hygromycin B (200 µg/ml). Cells infected with multiple retrovirus or lentivirus constructs, were selected for these constructs sequentially.
Cells were washed twice in ice-cold PBS and they were lysed in Triton X-100 lysis buffer [50 mM Tris (pH 7.5), 200 mM NaCl, 1% Triton X-100, 0.1% SDS, 10 mM Na3VO4, 50 mMNaF, 1 mM β-glycerophosphate, 1 mM sodium pyrophosphate, 1 mM EDTA, 1 mM EGTA, and 1 mM PMSF supplemented with a mixture of protease inhibitors]. The lysates were sonicated in a Misonix 3000 sonicator for 5 seconds at power level 1.5, and they were centrifuged for 20 min at 13,000×g. Western blots of the supernatants (soluble whole-cell lysates) were probed with the EZH2 rabbit monoclonal antibody (no. 4905, Cell Signaling), the GFI1 mouse monoclonal antibody (2.5D17, Sigma), the KDM2B goat polyclonal antibody (sc-69477, Santa Cruz), the p21WAF1/CIP1 human monoclonal antibody (no. 05-345, Cell Signaling), the RBP2 (JARID1B) rabbit polyclonal antibody (no. ABE239, Millipore), the PHF2 rabbit polyclonal antibody (3497, Cell Singaling), the JMJD3 rabbit polyclonal antibody (no. 3457, Cell Singaling), the JARID2 rabbit polyclonal antibody (ab48137, Abcam), the HA-Tag mouse monoclonal antibody (no. 2367, Cell Signaling), the myc-Tag rabbit monoclonal antibody (no. 2278, Cell Signaling), the pp71 (2H10-9) antibody or the IE1 mouse monoclonal antibody (BS500) [68]. Anti-mouse as well as anti-rabbit horseradish peroxidase-conjugated secondary antibodies, obtained from Sigma, were diluted in 5% milk in TBS-T and incubated with the blots for 1 h at room temperature. The bound secondary antibodies were detected with ECL-plus detection reagent (Amersham Biosciences) or the ECL SuperSignal (Pierce). Digital images of the proteins were acquired using the LAS-4000 luminescent image analyzer (Fujifilm Life Science).
To monitor the activation of the MIE promoter, 0.8×105 HFFs were plated on coverslips and they were infected with HCMV [69]. Five hours later, they were fixed and immunostained with an anti-IE1 monoclonal antibody (BS500) as previously described [65]. Anti-mouse as well as anti-rabbit Alexa 488-conjugated secondary antibodies were purchased from Molecular Probes (Invitrogen). The number of IE1-positive cells/coverslip was determined by epifluorescence microscopy. Each experiment was performed in triplicate.
HFFs cultured in 12-well plates were infected with HCMV AD169. HELA or HEK 293T cells, also seeded in 12-well plates, were transfected with pEGFP-C2 (Clontech) using Fugene 6 (Roche Applied Science). Infected and transfected cells were harvested, using a cell dissociation buffer (Molecular Probes). Harvested cells were fixed in paraformaldehyde (3% vol/vol in PBS). The HCMV infected cells were first permeabilized with 0.1% saponin in PBS, also supplemented with 2% calf serum, and then washed and resuspended in 100 µl of the same buffer, containing a 1/100 dilution of a mouse anti-IE1 antibody (BS500) [68] Following incubation with the antibody at room temperature for 1 h, the cells were washed twice and then incubated with a fluorescein isothiocyanate (FITC)-conjugated sheep anti-mouse secondary antibody (Sigma) (dilution 1∶1000) for 1 h. Transfected HELA cells were stored in PBS supplemented with 2% calf serum, after they were fixed. Virus and mock-infected, as well as transfected and non-transfected samples, were analyzed on a CyAn LX High Performance Flow Cytometer.
ChIP was performed using a Chromatin Immunoprecipitation assay kit (Millipore, cat no. 17-295). Chromatin cross-linking was achieved via a 10 minute treatment of nuclear extracts with 1% formaldehyde at 37°C. Cross-linked lysates were sonicated to shear the DNA to an average length of 300 to 1000 base pairs. Following sonication, the lysates were pre-cleared via incubation with a 50% slurry of salmon sperm DNA/Protein A Agarose for 30 minutes. The pre-cleared supernatants were incubated with the primary antibodies anti-H3K27me3 (no. 9756; Cell Signaling), anti-H3K4me3 (Abcam ab8580) and total anti-H3 (Abcam ab1791) (1∶50 dilution) overnight and with salmon sperm DNA/Protein A Agarose beads at 4°C for 1 h. Following multiple washes, the DNA-protein complexes were eluted and the DNA was recovered by reversing the cross-linking with NaCl and proteinase K. The DNA was then extracted using the Qiaquick PCR Purification Kit (Qiagen, cat. no 28106) and it was analyzed by SYBR-Green real-time qPCR, along with the input DNA. The primer sets used to amplify the GFI1 and the p16Ink4a loci as well as the HCMV MIEP are listed in the Table 2.
Total cell RNA was isolated, using Trizol (Invitrogen). cDNA was synthesized from 1.0 µg of total RNA, using oligo-dT priming and the Retroscript reverse transcription kit (Ambion, cat no. AM1710). The genes analyzed and the primers used are listed in Table 3. Real-time PCR was performed in triplicate using the Universal SYBR Green PCR master mix kit (Exiqon) and a 7500 Real-Time System (Applied Biosystems). mRNA levels were normalized to GAPDH, which was used as an internal control. All data are from 3 independent experiments, each performed in triplicate.
Nuclear and cytoplasmic fractions were isolated from HFFs cells using the Nuclear/Cytosolic Fractionation kit (Cat No AKR-171, Cell Biolabs, Inc.) according to the manufacturer's instructions. Purified DNA from each fraction was amplified by real-time PCR and HCMV genomes were quantified using the CMV Real-TM Quant kit (Cat No V7-100/2FRT, Sacacce).
|
10.1371/journal.pcbi.1000559 | Looking at Cerebellar Malformations through Text-Mined Interactomes of Mice and Humans | We have generated and made publicly available two very large networks of molecular interactions: 49,493 mouse-specific and 52,518 human-specific interactions. These networks were generated through automated analysis of 368,331 full-text research articles and 8,039,972 article abstracts from the PubMed database, using the GeneWays system. Our networks cover a wide spectrum of molecular interactions, such as bind, phosphorylate, glycosylate, and activate; 207 of these interaction types occur more than 1,000 times in our unfiltered, multi-species data set. Because mouse and human genes are linked through an orthological relationship, human and mouse networks are amenable to straightforward, joint computational analysis. Using our newly generated networks and known associations between mouse genes and cerebellar malformation phenotypes, we predicted a number of new associations between genes and five cerebellar phenotypes (small cerebellum, absent cerebellum, cerebellar degeneration, abnormal foliation, and abnormal vermis). Using a battery of statistical tests, we showed that genes that are associated with cerebellar phenotypes tend to form compact network clusters. Further, we observed that cerebellar malformation phenotypes tend to be associated with highly connected genes. This tendency was stronger for developmental phenotypes and weaker for cerebellar degeneration.
| We described and made publicly available the largest existing set of text-mined statements; we also presented its application to an important biological problem. We have extracted and purified two large molecular networks, one for humans and one for mouse. We characterized the data sets, described the methods we used to generate them, and presented a novel biological application of the networks to study the etiology of five cerebellum phenotypes. We demonstrated quantitatively that the development-related malformations differ in their system-level properties from degeneration-related genes. We showed that there is a high degree of overlap among the genes implicated in the developmental malformations, that these genes have a strong tendency to be highly connected within the molecular network, and that they also tend to be clustered together, forming a compact molecular network neighborhood. In contrast, the genes involved in malformations due to degeneration do not have a high degree of connectivity, are not strongly clustered in the network, and do not overlap significantly with the development related genes. In addition, taking into account the above-mentioned system-level properties and the gene-specific network interactions, we made highly confident predictions about novel genes that are likely also involved in the etiology of the analyzed phenotypes.
| A quarter of century ago a (former) Hewlett-Packard executive famously complained: “If only HP knew what HP knows” [1]. This inability to access invaluable “collective wisdom” is by no means specific to a single community. It is felt acutely in every present-day endeavor involving multi-human exploration of complex phenomena. The problem is especially dramatic in the case of the explosively expanding molecular biology literature. There are thousands of existing biological periodicals and millions of potentially useful publications. New journals are emerging on a weekly basis and new articles accumulate as if deposited by an avalanche.
Understandably, no omniscient repository exists that lists all known (published) molecular events (such as protein–protein interactions) detected in human or murine cells. Although current text-mining tools are imperfect in their extraction accuracy and recall, they do help us to process huge amounts of unstructured text in nearly real time (which humans cannot do), moving us a bit closer to total awareness about the current state of knowledge [2].
Here we describe and make available two large new data sets derived through mining one-third of a million full-text research articles and a complete and up-to-date PubMed collection of journal abstracts. These data sets comprise mouse- and human-specific molecular interactions between genes and/or their products. We present here only the subset of text-mined interaction assertions that involve gene or protein names that we can link to unique identifiers in the standard sequence databases. This choice is determined by the goal of making our data immediately useful for applications that would have difficulty handling ambiguity in gene identity. The complete data are available through the Columbia University (http://wiki.c2b2.columbia.edu/workbench) and the University of Chicago (http://anya.igsb.anl.gov/genewaysApp).
We use our newly generated data to analyze genetic variation related to abnormal cerebellum phenotypes in mouse and human. Our analysis results in a compact set of statistically significant predictions that can be tested experimentally.
Text mining with the GeneWays system [3],[4] allows us to capture multiple classes of relationships among biological entities, such as “A phosphorylates B,” “C activates D,” and “E is a part of F.” Table S1 displays the full list of relations that we can extract currently. The system also can recognize multiple classes of biological entities (terms) mentioned in the text: genes, proteins, mRNAs, small molecules, processes (such as cell death and proliferation), tissues, cell types, and phenotypes (such as diabetes and hypertension). While one can immediately think of a wide spectrum of applications where the full diversity of entities must be used, most of the current experimental methods are either gene-centric or genetic loci-centric (e.g., gene expression arrays, ChIP-on-chip, yeast two-hybrid, and genetic linkage or association data).
For this reason, the molecular networks we present here are gene-centric. This means that a given node in the network represents the union of the gene and its products (mRNA(s) and protein(s), if any); we exclude all other types of nodes (such as small molecules and phenotypes). Our practice of collapsing multiple nodes to a single node (gene plus mRNA plus protein) does not lead to a loss of information, because most of the physical interactions are defined for specific types of molecules. For example, in our restricted network relationship, “phosphorylate” can link only a pair of proteins, one acting as a protein kinase and another as the kinase's substrate, but not a gene and an mRNA. Furthermore, each original sentence used to extract the relation is preserved in the data set, along with the extracted fact and the reference to the appropriate paper, so that additional disambiguation can be conducted later, if required. We refer to each pair of extracted relationships and the original snippet of text as an action mention, as opposed to action, which is a relation disconnected from the source text and potentially mapped to multiple distinct action mentions.
A single pair of nodes in our text-mined network can be connected with multiple edges. These edges (interactions) can be undirected (we treat “A binds B” and “B binds A” as identical) or directed (“C activates D” is not the same as “D activates C”). We also subdivide edge types into two groups: logical and physical. Logical interactions include a family of regulatory relations that can be either direct (physical contact between two molecules) or indirect (mediated by one or more other molecules), such as activate, inhibit, and regulate (see Table S1). Physical interactions are by definition direct, such as methylate, bind, glycosylate, and cleave (see Table S1). The distinction between physical and logical interactions is important in understanding the data sets that we describe here. GeneWays ontology [5] includes a number of relationships between molecules that are neither physical nor logical interactions (for example, A is an ortholog of B, or C is part of D). We call this class of relations other.
In typical free text, gene names are dissociated from any references to gene-annotation databases. Furthermore, the “raw” text-mined molecular-interaction data are vast (GeneWays 7.0 comprises more than 8 million action mentions) but rather noisy: the error rate is close to 35% [6]. To get to smaller, cleaner, species-specific networks, we performed the following steps.
First, out of the complete network we retained only those gene names that can be linked to either human or mouse sequence database entries (normalization step) (see “Mapping names to genes” in the Text S1). Second, we filtered out relationships that are not molecular interactions and collapsed multiple edges between two nodes into a single edge. Third, we weeded out “raw” text-mined statements that did not meet our precision threshold (precision is defined as the proportion of correctly extracted statements among all those automatically extracted by a system). The third step was conducted automatically, using our automated curator engine [6], which has near-human curation precision (see the “Quality-of-extraction assessment” section in the Text S1.).
The first step resulted in the H70 and M70 networks (human- and mouse-specific GeneWays 7.0), in which nodes can be connected by multiple directed or undirected edges. The second step led to generation of the H70-PL and M70-PL networks (PL stands for physical and logical), where direction of edges was abandoned. The third step, assigned a precision threshold of 0.9 (90% of action mentions are correct), produced even smaller data sets, H70-PL0.9 and M70-PL0.9. Table 1 provides an overview of these networks at different levels of granularity. All intermediate data sets in this pipeline of data filtering are available for third-party computational analyses (see Datasets S2 to S5)
In addition, we produced networks with non-redundant edges and solely physical interactions, H70-P0.9 and M70-P0.9. As in the previous data sets, to filter these networks we used a precision threshold of 0.9.
To evaluate the quality of the H70-PL0.9 network, we chose two random sets of logical and physical action mentions, a hundred mentions each, and asked an expert to evaluate their correctness. The expert commented on two steps of the process: whether the action mention is correctly extracted by the GeneWays system and, if the answer was “yes,” whether the corresponding gene names were correctly mapped to sequence identifiers. This allowed us to measure the absolute precision of the H70-PL0.9 network, the precision of term mapping, and the overall precision over the information extraction and term mapping stages.
The physical action mentions set indicated a precision of 0.8, with a confidence interval (CI) of [0.71, 0.87]. (We use CI at the 95% level of significance consistently throughout this paper.) The logical action mentions set showed a higher precision of 0.91, CI: [0.84, 0.95]. Because in our data set the number of logical interactions exceeds the number of physical interactions by more than two-to-one (2.49∶1), the overall precision of the HL70-PL0.9 data set is close to the target value of 0.9 (0.88). Term-to-sequence mapping precision was 0.89 (CI: [0.84, 0.93]) and 0.87 (CI: [0.81, 0.91]) for physical and logical action mentions, respectively (see Table 2).
Despite the favorable precision of the GeneWays extraction and the per-term mapping, the precision over both steps is less impressive: 0.66 (CI: [0.56, 0.74]) and 0.69 (CI: [0.59, 0.77]) for the physical and logical datasets, respectively. The reason for the lower overall result is the multiplicative calculus of the probability of not making an error: The overall precision of a term-mapped logical action is a product of the information extraction precision and the precision of two independent term mappings: 0.91×0.87×0.87 = 0.69.
Thus far we have evaluated the quality of extraction and mapping of action mentions. Recall that the same relation (action) between a pair of genes can be independently extracted from multiple sentences, generating distinct action mentions. Intuitively, the precision of an action (because an action is correctly extracted if at least one of its associated action mentions is correctly extracted) should be at least as high (or higher) than precision of the corresponding action mentions. To evaluate this precision, we sampled a hundred random actions from the H70-PL0.9 dataset, asked an expert to evaluate them at the levels of extraction and term mapping, and obtained an estimate of action-level two-stage precision of 0.74, CI: [0.65, 0.82]. This estimate is higher than the estimate of two-stage action mention precision (0.66 or 0.69). We believe that the action-level precision is more relevant to real-life applications in which scientists tend to care primarily about the precision of actions (statements distilled from multiple sources) rather than about their individual instances linked to text.
Note that the precision discussed in this section reflects only properties of our information extraction system and not the verity of published data.
Several publicly accessible databases generated by manual analysis of research literature are available, including the Human Protein Reference Database (HPRD) [7],[8], Reactome [9], the Biomolecular Interaction Network Database [10], and the Database of Interacting Proteins [11]. These four data sets, along with a few others, were carefully compared in a recent study [12]. HPRD is by far the largest of the four.
As another quality control measure for our study, we compared our data with HPRD 7. The HPRD 7 network [7],[8] comprises 9,460 nodes (unique gene identifiers) and 37,081 edges, compared to 7,793 nodes and 52,518 edges in the H70-PL0.9 network. The H70-P0.9 network comprises 5,453 nodes and 16,707 edges; the node-wise and the edge-wise overlaps of H70-P0.9 with the HPRD networks are 4,543 and 4,877, respectively.
The HPRD and H70-PL0.9 networks share 5,945 unique gene-specific nodes. Out of the possible maximum of 17,668,540 interaction pairs between these nodes, the HPRD network has 23,662 and the H70-PL0.9 covers 43,496. We would expect a random overlap of about 58 interactions, while in reality we observe 7,577. The expected and the observed values are so far apart that the p-value (obtained with a hyper-geometric overlap test) is effectively zero—that is, the apparent overlap between the two sets of data is extremely non-random.
Because human-curated databases may still harbor errors [13], we also compared our literature-mined dataset to a small set of high quality interactions produced by careful manual verification of a set of interactions shared by several human-curated databases [13].
In a recent study, Cusick et al. sought to evaluate the ultimate (truth) quality of the molecular interaction datasets generated via manual curation of the literature [13]. The authors selected two sets of curated interactions: one consisted of interactions that were curated in multiple databases and that were supported by multiple manuscripts and the other consisted of interactions supported by a single publication. They then carefully recurated the selected interactions and were able to estimate the corresponding error rates. As a byproduct of the evaluation, the authors produced two relatively small datasets, LC-multiple and LC-single, with 110 and 92 interacting pairs respectively, of exceptionally high-quality curated interactions. The LC-multiple set contained the interactions that were supported by multiple manuscripts even after the recuration and the LC-single set contained the interactions with one supporting manuscript that was confirmed during the recuration. The LC-multiple set subsequently was used as a gold standard for the evaluation of high-throughput yeast two-hybrid assays in a second manuscript by the same group, in which an additional random set of 188 supposedly non-interacting pairs (the Negative set) was selected [14].
We used the LC-multiple, LC-single, and Negative sets as comparison standards for our own literature-mined networks (see Table 3). It is reassuring that our H70-PL network covers nearly 70% (75 out of 110 pairs) and that our most filtered human network, H70-P0.9, covers more than 55% of the well-supported interactions in the LC-multiple set. The more obscure interactions from the LC-single set are not covered as well (i.e., H70-P0.9 contains about 20% of the LC-single set). However, given that we have processed only a small portion of all of the scientific literature with a system that highly favors precision over recall, being able to recover 20% of the interactions supported by a single article is surprisingly high. Finally, our networks do not contain any of the interactions listed in the Negative set. For comparison, the last two lines in Table 3 give the results for the two high-throughput assays MAPPIT and Y2H-CCSB evaluated in Figure 2 of [14].
The performance of text-mining methods is commonly evaluated using two metrics: precision and recall. For information-extraction systems, precision is defined as the proportion of correctly extracted statements among all those automatically extracted by a system. The recall is the ratio between the number of statements correctly extracted by the system and the total number of statements that can be extracted from the original text by a hypothetically perfect system. In a less than perfect system, recall and precision are antagonistic: one is increased at the expense of the other.
In this study we favored precision at the expense of recall: We explicitly used a statement precision threshold as a filtering criterion. We also excluded actions with ambiguous gene names and disqualified some 105 potentially useful instances of text-mined intramolecular relations that fit neither physical nor logical categories (such as contain and is a homolog of), thus worsening recall and improving precision. In addition, we used only those actions that involve either genes or their products (and no other entity classes).
While our human-specific network, which unifies physical and logical interactions (H70-PL0.9), is larger than HPRD 7, the relationship is reversed for the physical-interaction (H70-P0.9) data set and HPRD 7. This is because we filtered out from our data numerous physical action mentions that did not pass our precision threshold. (Note that HPRD 7 incorporates high-throughput interaction data that is probably distinct from the small-scale experimental data published in research papers, in terms of error patterns.) Nevertheless, the HPRD 7 data sets and our data sets are very different. The joint interaction coverage of HPRD 7, H70-P0.9, and M70-P0.9 ortholog data sets is more than twice as large as the coverage of HPRD 7 alone (Figure 1); this is enough to merit the use of a union of these networks in biological applications.
Because we are making the “raw” (unfiltered) statements publicly available, anyone interested in using our data can apply his/her own custom-made filtering process to achieve the desired balance between recall and precision in the output.
Two genes residing in genomes of distinct species can either share a common origin (homology) or be unrelated. Homologs come in at least in two flavors [15]: orthologs and paralogs. Two genes in, say, human and mouse, are orthologs if they were separated by a speciation event. If, in addition to speciation, an intragenomic gene duplication occurred, separating two genes from a common ancestral gene, they are paralogs. For example, human and mouse embryonic β-globins are orthologs, but mouse α-globin is a paralog of human β-globin.
Physical interactions between molecules are not immutable over long evolutionary intervals [16]. Nevertheless, an interaction between two proteins discovered in one species has a reasonable chance of existing between orthologs of these proteins in another species if the two species are closely related. Therefore, if we know of interacting molecules in one species and can identify orthologous molecules in another species, we can formulate hypotheses about the existence of orthologous interactions in the latter species. All such computationally formed hypotheses are subject to experimental validation.
Mouse and human genomes are separated by more than 100 million years of independent evolution [17], but mouse genetics and molecular biology are commonly used to understand human phenotypes in health and disease. Therefore, we decided to compile a molecular-interaction network summarizing the wealth of knowledge for humans and mice. We used orthology-mapping of human and mouse genes to connect the two networks. (Reactome's developers [9] used a similar strategy with their manual compilation of data.) Such a network could potentially have a multitude of practical applications.
We assembled our network by combining mouse- and human-specific networks extracted from the biomedical literature using text-mining tools. We used human-to-mouse gene orthology mapping provided by the Mouse Genome Database [18],[19]. Some of the mouse interactions could not be mapped to corresponding human interactions because at least one of the involved genes lacked known human orthologs. We transferred by mouse-to-human orthology-mapping 49,493 and 16,317 interactions for physical-logical and physical networks, respectively. These orthology-mapped interactions are subsets of the 57,786 and 18,252 interactions in the physical-logical and physical networks, respectively. Although a large number of interactions occur both in humans and mice individually (see Figure 1), the double-confirmed overlap constitutes only about a third of all interactions in the union network (see Figure 1 A and B). Figure 1 C shows a three-way Venn diagram for our text-mined interactomes (human and mouse) and the HPRD dataset. Clearly, all three networks contain a substantial number of unique interactions that merit their joint consideration in biological applications (see Dataset S6).
To illustrate an application of our data to the analysis of mammalian phenotypes, we performed mapping of mouse cerebellar phenotypes (related to ataxia) to the three-data set network.
The word ataxia (αταξια—“lack of order”), in its English usage, refers to a lack of muscular coordination in an animal body. Humans with ataxia often have difficulty walking, talking, maintaining posture and balance, controlling eye movements, holding and manipulating objects, gesturing, and even swallowing food. In a mammalian brain, the cerebellum is predominantly responsible for spatial and temporal coordination of complex neuromuscular processes. Cerebellar function is also essential for cognition sensory discrimination [20]. Most cases of ataxia are associated with either environmental or genetic damage to this brain region. The typical environmental triggers of ataxia include head trauma, viral infections, and exposure to recreational or medicinal poisons, such as alcohol, lithium carbonate, tranquilizers, antipsychotics, and the anticonvulsant carbamazepine. Genetic factors include a diverse spectrum of genomic aberrations that cause abnormal development and/or premature degeneration of the cerebellum. Ataxia can be severely debilitating and, unfortunately, the phenotype is reversible in only a minority of cases (such as those caused by short-term alcohol intake).
Mouse and human geneticists who study brain phenotypes typically group developmental malformations by the anatomical structures that are affected. As brain topology in three-dimensional space does not lend itself readily to verbal description, we provide three projections of a typical mouse brain in Figure 2 (see also the interactive model in Figure S1). Moving front-to-back in the external view of the mouse brain, there are two olfactory bulbs followed by hemispheres of cerebral cortex that are immediately adjacent to the cerebellum and brainstem (see Figure 2 A–C). We focus here on the cerebellum (literally, “little brain”) that contains involuted cortex with narrow leaf-like gyri (“folia,” see Figures 3 A and C). Like the brain itself, the cerebellum has two hemispheres with a worm-like medial structure, the vermis, between them (Figure 3 A and B).
In both humans and mice, a collection of genetic aberrations exist that are known to predispose the bearer to specific cerebellar abnormalities. For computational implementation it is convenient to represent phenotypic variations of cerebellar structure with hierarchically ordered categories in a mammalian phenotype (MP) ontology [21]. We focused on five broad anatomical/cerebellar causes of ataxia which can be observed as structural abnormalities in brain imaging studies (such as MRI scans) or histological analysis. These phenotypes are represented with MP concepts: degeneration (MP:0000876), abnormal foliation (MP:0000857 and MP:0000853), abnormal vermis (MP:0000864), small cerebellum/cerebellar hypoplasia (MP:0000852 and MP:0000851), and absent cerebellum (MP:0000850).
Cerebellar degeneration is abnormal death of cerebellar neurons—the cerebellar folia become narrower over time and are separated by irregular and wider spaces compared with those in a healthy brain (see Figure 3 A). As with other major insults to the cerebellum, degeneration reveals itself in abnormalities in body balance, jerky movements of limbs, unsteady (wide-legged) gait, and irregularities of speech (slurred or slow) and eye movement (nystagmus, or rapid involuntary eye movements). Most defined degenerative ataxias affect the fully mature cerebellum, but a small subset of degenerative ataxias have a developmental onset [22].
Abnormal foliation typically involves the absence of some of the cerebellar folia and irregular shape of those that are present. In normal individuals, cerebellar foliation is stereotypical, with the basic folial pattern conserved between mice and humans. Disruption of foliation disrupts the topographical map of incoming and outgoing neuronal connections [23].
An abnormal vermis is typically reduced (compressed and distorted) compared with a normal one, or it can even be completely missing (see Figure 3 A). Clinical outcome is variable [24]. Dandy-Walker malformation is one of the well-known birth defects in humans and mice that are defined by an abnormal vermis. In addition to an aberrant vermis, Dandy-Walker malformation frequently involves enlargement of the fourth brain ventricle and an increase in fluid-filled space around the brain [25]. It is not uncommon in clinical reports to find an abnormal vermis coupled with other cerebellar malformations [26],[27].
Small cerebellum, or cerebellar hypoplasia, refers to phenotypes in which the cerebellum, while present, does not develop to normal size (see Figure 3 A). In humans, cerebellar hypoplasia can lead to delayed or undeveloped speech, difficulties with walking and maintaining balance, mental retardation, floppy muscle tone, nystagmus, and seizures. In its worst forms, cerebellar hypoplasia can be completely debilitating and even lethal [28].
Absent cerebellum is infrequent in adult humans and mice, possibly because in most cases it causes early death. Rarely, individuals are only mildly affected. For example, a documented brain autopsy of a 38-year-old individual who accidentally electrocuted himself revealed a virtually absent cerebellum [29]. The individual was apparently functional and capable of conducting all common human activities, including gesturing, talking, performing complex manual work, and participating in social interactions. Some have proposed that an absent cerebellum is therefore less disabling than a present, but abnormal cerebellum [30].
Fortunately, the Mouse Genome Database (MGD, [18]) uses the MP ontology to link genetic variation in mouse genes to phenotypic aberrations that are causally related to known genomic changes. We were able to use the MGD to associate 244 human genes (with the help of the human and mouse orthology) with the five ataxia phenotypes described above and with ataxia (MP:0001393) (see Table S7). By integrating mouse (M70-PL0.9), human (H70-PL0.9), and HPRD (Release 7) networks through human–mouse gene orthology mapping, we obtained a larger network of interacting human genes with annotation of ataxia phenotypes generated in mouse studies. The largest connected component of the ataxia graph includes 88 human genes. These 88 genes are connected with 145, 147, and 72 interactions derived from human GeneWays, mouse GeneWays, and HPRD, respectively (see Figure 3 D).
Our analysis of ataxia-related phenotypes in the context of a molecular network generated rather curious and statistically significant results, as described in the following section.
We have provided two very large molecular-interaction sets for human and mouse genes (Datasets S2, S3, S4, S5). The sets were integrated through gene orthology and are immediately applicable to a spectrum of experimental data analysis tasks (Dataset S6). Our analysis of mouse mutant cerebellar phenotypes, with the aid of our text-mined networks, lead to a number of intuitively reasonable and biologically testable predictions.
Our present study shares its spirit, goals, and some methods with other efforts in the field. For example, one of the most recent studies succeeded in integrating a diverse array of approaches to design a tool generating new disease-related hypotheses [57]. This group was able to combine information extraction [58], biomedical ontology mining, statistical analysis of sequences of natural language tokens, probabilistic analysis of error patterns across data types, computational reasoning, understanding of large-scale experimental datasets, and exploratory visualization in one application.
Because we are releasing our complete set of annotated data to public domain, these data might be instrumental for direct comparison for analogous text-mining results produced elsewhere [59]–[67].
Automated reasoning over text-mined, experimental, and machine-deduced data (Reading, Reasoning, and Reporting, as [57] put it), is likely to become a dominant mode of science in the near future, as size of experimental datasets and complexity of natural system under scrutiny grows.
GeneWays is an information extraction (text-mining) system: It ingests computer-encoded full-text research articles or journal abstracts and distills from them a collection of biological relations. The architecture and implementation of the system are described in great detail elsewhere [3], [4], [6], [68]–[71], so here we provide just a brief outline of the system.
In a simplified view, the processing pipeline inside the GeneWays system is a sequence of just two steps. The first step deals with recognition of words or phrases representing important biological entities (such as p53, Alzheimer's disease, or endoplasmic reticulum; computer scientists call this step named entity recognition, NER). The second step deals with detecting relations among the entities (such as p53 activates JAK) and is called information extraction (IE).
Our NER module (MarkIt, [72]) works by following a hierarchy of rules defined by human experts. Our IE module (GENIES, [3],[68]) also is based on a collection of expert-encoded rules, but the underlying mathematical model is a bit more sophisticated (a deterministic context-free grammar). We use MarkIt to identify a spectrum of biological entities, such as disease, process, gene, protein, RNA, small molecule, tissue, and species. We apply GENIES to each individual sentence, trying to reconstruct the most probable steps that led to generation of the sentence. This reconstruction process is called parsing; besides satisfying our academic curiosity, parsing is useful for capturing complex relationships between entities. The results of parsing depend strongly on the formal grammar implemented in the parser.
Most of the relations that we can extract from biomedical texts are directional (A activates B is not the same as B activates A) and binary (only two entities are involved, which we call upstream and downstream, according to their position within the relation). A sentence can contain any number between none and dozens of relations. We can think of a typical binary relation as a quadruplet of values: [negation, upstream entity, action, downstream entity] (see Figure 5). Negation allows us to capture negative statements (“In our experiment, AKT failed to phosphorylate BAD”→[not, AKT, phosphorylate, BAD]) as well as positive statements. Relation type (action) indicates semantic connection between the two entities (bind, activate, methylate, transport, is part of, is homolog of, etc.).
To facilitate automatic reasoning over semantic groups of relations, we arrange them into an acyclic directed graph, where graph arcs represent the relation “is a.” The GeneWays system currently recognizes 391 different action types, 207 of which are frequent (see Table S1).
To generate the molecular networks described in this study, we analyzed 368,331 full-text articles and 8,039,972 article abstracts from PubMed (see Table S8). The system identified 5,890,150 relations in the full text articles and 2,534,299 relations in the abstracts: in total, 5,934,024 unique relationships. The action types with the largest number of relationships are induce (695,615), bind (532,385), inhibit (386,523), associate (370,133), contain (366,654), and activate (332,336); the numbers in parentheses indicate the abundance of relations of each type in the GeneWays 7.0 database.
|
10.1371/journal.pntd.0002736 | A Systematic Approach to Capacity Strengthening of Laboratory Systems for Control of Neglected Tropical Diseases in Ghana, Kenya, Malawi and Sri Lanka | The lack of capacity in laboratory systems is a major barrier to achieving the aims of the London Declaration (2012) on neglected tropical diseases (NTDs). To counter this, capacity strengthening initiatives have been carried out in NTD laboratories worldwide. Many of these initiatives focus on individuals' skills or institutional processes and structures ignoring the crucial interactions between the laboratory and the wider national and international context. Furthermore, rigorous methods to assess these initiatives once they have been implemented are scarce. To address these gaps we developed a set of assessment and monitoring tools that can be used to determine the capacities required and achieved by laboratory systems at the individual, organizational, and national/international levels to support the control of NTDs.
We developed a set of qualitative and quantitative assessment and monitoring tools based on published evidence on optimal laboratory capacity. We implemented the tools with laboratory managers in Ghana, Malawi, Kenya, and Sri Lanka. Using the tools enabled us to identify strengths and gaps in the laboratory systems from the following perspectives: laboratory quality benchmarked against ISO 15189 standards, the potential for the laboratories to provide support to national and regional NTD control programmes, and the laboratory's position within relevant national and international networks and collaborations.
We have developed a set of mixed methods assessment and monitoring tools based on evidence derived from the components needed to strengthen the capacity of laboratory systems to control NTDs. Our tools help to systematically assess and monitor individual, organizational, and wider system level capacity of laboratory systems for NTD control and can be applied in different country contexts.
| Capacity strengthening activities such as technical training for staff, student research project supervision, and equipment provision are being carried out in laboratories worldwide as part of the global effort to control neglected tropical diseases (NTDs). However, these activities often focus on developing the skill sets of an individual and are not being thoroughly monitored and assessed. To address these gaps we developed a set of monitoring and assessment tools that can be used to determine the capacities required and achieved by laboratory systems to support the control of NTDs. The tools simultaneously focus on individuals (e.g., technicians, students, researchers), organisations (e.g., universities, research institutions, clinical facilities), national governments, and international agencies. Using the tools highlighted the strengths and limitations of each laboratory system in addition to the role of the laboratory regionally and internationally. We used the tools in Kenya, Ghana, Malawi and Sri Lanka, and concluded that our tools can be adapted and tailored to use in other countries and laboratories.
| Effective prevention and treatment of neglected tropical diseases (NTDs) requires reliable and efficient laboratories for diagnosis and for supporting disease and entomological mapping surveys and yet laboratory systems are often weak in low and middle-income countries (LMICs) where the majority of this testing is carried out [1], [2]. Neglected tropical diseases consist of 17 microbiological diseases (see Table S1 for a list of the 17 Neglected Tropical Diseases as Classified by WHO) that affect the poorest people in the world. Current estimates suggest that over one billion people are infected with at least one NTD, and that these diseases cause approximately 534,000 deaths and 57 million disability adjusted life years (DALYs) each year [3]. In January 2012, as part of the London Declaration, a number of charities, pharmaceutical companies, and other businesses pledged to work together to improve the lives of people affected by NTDs and ultimately progress towards the elimination or control of 10 NTDs by 2020.
The lack of capacity in NTD laboratory systems in LMICs is a major barrier to monitoring and evaluation of interventions used for control and elimination of NTDs. The DFID funded Centre for Neglected Tropical Disease (CNTD) in the UK is monitoring the impact of mass drug administration (MDA) on the incidence of NTDs. The programme has found that lack of laboratory capacity in the CNTD supported countries is a critical bottleneck to implementing and monitoring community-based elimination interventions. To help the laboratories perform more effectively, the CNTD requested support from the Liverpool School of Tropical Medicine's (LSTM) Capacity Research Unit to design, monitor, and evaluate the capacity development of four laboratories in Ghana, Kenya, Malawi, and Sri Lanka.
Definitions of capacity development vary depending on the sector or particular programme focus, but a common definition is “ability of individuals, organisations or systems to perform appropriate functions effectively, efficiently and sustainably” [4]. Laboratory capacity strengthening is complex; it can require investment in specialised equipment, the support of all cadres of staff including laboratory scientists and researchers, as well as the leadership of the organisation in which the laboratory is housed, and sufficient time for training and embedding new processes, systems and equipment. Our aim was to develop a capacity strengthening programme which used a common approach to assessment and monitoring, but which could be tailored to take account of the different ways laboratories were financed, managed, and operated and their interactions with national programmes and regional collaborators. There are many capacity strengthening initiatives being carried out with laboratories in LMICs [5]; however, many of these initiatives focus on individuals' skills (e.g., technical skill of using microscope) [6] or institutional systems and processes (e.g., quality control office) [7] ignoring wider national and international structures (e.g., national and regional health systems) integral to establishing sustainable capacity.
In addition to the dearth of literature on organizational and national or international structures integral to capacity strengthening, rigorous approaches and methods to evaluate capacity strengthening initiatives are scarce [8]. Measuring the progress and impact of these capacity strengthening efforts is a priority for the international development community [9], but donors and scientists alike are struggling with how to do this well [5]. Evidence-based tools have been developed to help evaluate health research capacity strengthening [8] but in the area of laboratory capacity strengthening for NTD control and elimination specifically, no such tools exist.
The CNTD's goal in relation to laboratory capacity is to strengthen one laboratory in each of the four countries to support intervention activities that aimed to control and eliminate NTDs by 2020. To support this goal, our project aimed to describe and measure the capacities required by each laboratory at the individual (e.g., technicians, students, researchers), organizational (e.g., universities, research institutions, clinical facilities),, and national and international levels. To achieve CNTD's goal, our specific objectives were to a) use available evidence to describe the optimal capacities needed at each of the three levels for each laboratory if they were to achieve the goal, b) develop a set of assessment and monitoring collection tools that would enable us to assess what capacity gaps needed addressing if laboratories were to achieve optimal capacity and c) develop a capacity strengthening action plan to address the gaps and indicators that would enable us to monitor progress as capacity gaps were addressed.
We used a validated framework and theory of change principles to guide the development of our capacity strengthening tools. The framework for designing and evaluating a health research capacity-building programme is based on four phases of capacity strengthening (see Table 1) - awareness, experiential, expansion, and consolidation [10]. Based on this framework an important first step in the awareness phase is to carefully review current capacity against a set of optimal standards and conduct a needs assessment to identify capacity gaps. We focused efforts on engaging all relevant stakeholders to determine the objectives of the capacity strengthening programme, identify capacity gaps and needs, and jointly develop a capacity development action plan. Our approach enabled stakeholders to be actively involved in the assessment and monitoring process. To carry out these activities we recognized that we would require specific assessment and monitoring collection tools and would need to consult various data sources within each laboratory system.
We also draw on theory-based evaluation methods, particularly theory of change evaluation, to develop our approach to laboratory capacity strengthening. We define theory of change as “An on-going process of reflection to explore change and how it happens – and what that means for the part organisations play in a particular context, sector and/or group of people” [11]. Using a theory of change approach involves specifying an explicit theory of how and why a capacity strengthening intervention might cause an effect, and this is used to guide the evaluation [12]. Guided by this, our theory was that strengthening laboratories for NTD control is a complex and non-linear process involving wider systems and actors beyond the institution; we also assumed strengthening capacity in the laboratories would involve strengthening partnerships, organisational development, empowering people, and open communication. We purposely choose to incorporate theory of change in our work in order to determine indicators that could help us explore the relationship between the programme inputs, activities, and outcomes.
Prior to our research, no tools existed for specifically examining the capacities required by laboratory systems at the individual, organizational, and national and international levels to support the control of NTDs, or for capturing information from various data sources within laboratory systems. Therefore we developed our own tools based on evidence concerning the components (i.e., people, skills, systems, resources) needed to strengthen the capacity of laboratory systems. We used a three-stage approach to develop the assessment and monitoring tools. First we searched published evidence concerning laboratory capacity strengthening at the individual, organisation, and national and international system level. We searched the electronic databases of PubMed and Google Scholar, using the keywords “laboratory”, “NTD” and “capacity strengthening”. We also consulted books and published reports concerning capacity strengthening initiatives conducted with medical laboratories. From this information we were able to generate a list of all the components that were necessary for an optimal laboratory system in the domain of NTDs and used this to inform the design of our tools. Specifically, the following documents guided the development of our assessment and monitoring tools; the Global Laboratory Initiative Stepwise Process towards TB Laboratory Accreditation [13] and adapted for NTD laboratories, the EFQM excellence model [14], the SIDA evaluation model of HEPNet [15] and the UNDP Measuring Capacity document [16]. Using all the components in the list of optimal capacities we developed a questionnaire for laboratory managers, a semi-structured interview guide for use with laboratory stakeholders, a capacity gap checklist for use with the laboratory manager and laboratory staff, and a checklist for ISO 15189 to be used for on-site observations (see Table 2). Our intention was to use these tools during a site visit to collect data that would allow us, in collaboration with local stakeholders (e.g., laboratory technicians, laboratory managers, NTD scientists, directors of institutions, Ministry of Health representatives, etc.), to identify capacity gaps and to create a comprehensive capacity development action plan to address the gaps.
We analyse the data generated from all the tools using content and thematic analysis. Specifically, we use an analytic framework to help guide thematic data analysis of the interview and focus group data. The analytic framework consists of a range of apriori codes that help to organize the data generated and includes codes pertaining to quality assurance, institutional collaboration, funding, NTD coverage or focus, research capacity, and organizational resources. Data from the checklists and questionnaire are analysed using content analysis.
We use the findings of the capacity gap analysis to jointly develop with laboratory managers their own unique five-year capacity development strategy to improve their capacity to conduct research and analysis to support NTD control. Gaps in capacity that need to be filled to achieve the strategy are agreed upon during a consensus meeting with invited stakeholders. Priority gaps that require action in the first year are proposed by stakeholders and amalgamated into a one year capacity development action plan with measurable indicators and targets to drive capacity strengthening. The plans are then finalised through Skype and email discussions (e.g., details concerning completion dates) after the completion of each of the visits. These capacity development action plans can also be used to mobilize donor funding as they highlight and provide justification for the priority areas where funding needs to be invested.
Following development of the tools, we implemented them in four of the CNTD/LF programme (2012-16) funded laboratories, including Ghana, Kenya, Malawi, and Sri Lanka. The laboratories in each country were initially selected by CNTD to be a part of their MDA programme because it had been identified that a lack of capacity globally in laboratory systems was a major bottleneck in the monitoring of MDA. Of all of the laboratories in the MDA programme, the laboratories in Ghana, Kenya, Malawi, and Sri Lanka were chosen to be a part of the pilot study because each were seen to be potential regional leaders in the control of NTD and had a potential ability to support NTD laboratories in other countries. See Table 3 for a description of each laboratory involved in the study. Implementation of the tools occurred throughout 2012 during a 5–10 day visit at each institution, with two complementary members (e.g., laboratory specialist, social scientist) of the Capacity Research Unit leading each visit.
A total of 62 semi-structured interviews were conducted, 17 in Malawi, 11 in Ghana, 16 in Kenya, and 18 in Sri Lanka. We interviewed stakeholders from a range of institutions and levels including laboratory scientists, laboratory directors, research staff, WHO staff, ministry representatives, students, human resource and financial staff, donors, and senior academics. For example, key NTD stakeholders in Kenya were drawn from the Eastern and Southern Africa Centre of International Parasite Control NTD laboratory located in the Kenyan Medical Research Institute and the National NTD programme through the office of the Department of Disease Prevention and Control in the Ministry of Health. In addition to the semi-structured interviews, in each country one pre-visit questionnaire and ISO checklist were completed, 2–4 capacity gap checklists were completed, and one focus group was held.
We revised the tools after their implementation in each country by conducting a retrospective analysis of how the tools contributed or not to the awareness phase in the framework for designing and evaluating a health research capacity-building programme that guided the design of our capacity strengthening tools. The analysis was developed through collaborative and candid dialogue by the research partners, using the framework as the basis for deliberation. These analysis meetings with the entire research team reviewing the findings were an important step in establishing rigour in the refinement of the tools. Throughout the analysis, questions were asked such as; “Were all relevant stakeholders at organisation and policy level as well as individuals involved in implementing capacity strengthening cycle engaged?” and “Was there an emphasis on local ownership with defined role for external input?” Results of the retrospective analysis shed light on factors such as how some stakeholders were not participating in the capacity assessment possibly as a result of the work being carried out in a context where being critical could be considered inappropriate, particularly for a junior member of staff.
To address this particular issue, we adapted the methods to include focus group discussions specifically for laboratory staff, where laboratory managers did not participate. These refinements enabled us to gain an increasingly greater depth and breadth of information from laboratory staff. The retrospective analysis also illuminated that the laboratories held varying capacity strengths and gaps and the tools needed to be able to be tailored accordingly. For example, following the work in Malawi, modifications of the tools included re-designing the ISO checklist to enable laboratory staff to bypass sections of questions that were not relevant to their laboratory's stage of development. By analyzing the implementation of the tools in succession in different countries we had time to use systematically lessons we had learnt to revise the tools between each evaluation.
We obtained ethics approval for the capacity strengthening component of the work from the LSTM Research Ethics Committee. The wider DFID funded CNTD programme has ethics approval for all monitoring and evaluation activities scheduled to be implemented in the country laboratories.
Using the rich information collected with the tools we were able to identify strengths and gaps in NTD laboratories' systems capacity (see Table 4). The identified strengths and gaps varied amongst the countries; however, inter-laboratory comparison revealed some similarities. For example, all laboratory systems mentioned that NTDs being recognized as a national priority was a specific strength, which resulted in greater availability of national funding and human resource support for laboratories. The following quote from a stakeholder in Kenya illustrates this finding, “A national multi-year strategic plan for control of NTD was published in 2011”. Furthermore, in all countries the laboratories had strong links to policymakers and existing national and regional collaborations.
In regards to capacity gaps, one common gap was the lack of funding for NTD research, as allocating funding for research was seen as less of a priority than operations and management when health sector funding decisions were being made. Also common to all of the laboratories was a lack of quality assurance documentation and safety systems, a lack of formalized agreements with national NTD programmes, and reliance on external funds. There also was a specific disease focus in each laboratory, without consideration of the broader NTD focus, creating a need for each laboratory to consider how they move beyond their specific focus on malaria or lymphatic filariasis etc. to NTDs as a whole. Finally, there was a lack of research and biostatistics capacity in all of the laboratories, partially due to the fact that research training courses were not accessible to all staff.
Activities were identified for each country to undertake to work towards achieving ISO 15189. As with the strengths and gaps, the identified activities varied amongst the countries; however, inter-laboratory comparison revealed some similarities. The checklist revealed that none of the countries had written safety systems in place (e.g., procedures to follow in event of a biohazardous incident that are essential to achieve quality assurance). Therefore, similar activities that needed to be undertaken in each country included the drafting of full standard operating procedures for all experimental processes, safety, and equipment in the laboratory. Additional gaps in relation to ISO standards included the need to appoint and assign a safety officer and to have job descriptions available for all staff.
The tools generated information about how the NTD laboratories could support national NTD programmes in the region with achieving their aims. The NTD laboratories were found to provide timely and helpful input on country specific issues for topics related to NTDs such as sample diagnostics, vector analysis, and the efficacy of control programmes. For example, in Kenya the tools helped identify the potential for the laboratory to provide support to regional LF control programmes in Tanzania, Zimbabwe, Botswana, and Zambia. Additional potential activities that were identified through our process include confirmation of NTD elimination through implementation of monitoring and evaluation activities, quality control, processing of samples collected through operational research carried out in hotspot areas where transmission of NTD is persisting even after several mass interventions, and support other operational research activities aimed to support implementation. Furthermore, in each country the laboratories were found to provide robust scientific data to support national and regional NTD control programmes, enabling policy makers to make informed decisions that contributed to control and elimination of NTDs in their country and region.
Information about each NTD laboratory's position within national and international networks and collaborations was generated from the set of tools. Findings indicate that the level of technical expertise and experience within the laboratory system enhanced a laboratory's position within their networks as with this expertise the laboratory was seen to be a preferential collaborator. Technical expertise was perceived by stakeholders to be more essential to a laboratory's position within networks than other factors such as geographic proximity. For example, the laboratory scientists in Ghana are highly skilled in using real-time polymerase chain reaction (RT-PCR). Given their expertise the Ghanaian scientists were identified as being able to provide training to other laboratories within the CNTD network.
We have described our systematic process for developing evidence-based, practical ways of assessing and monitoring the capacity of laboratories in LMICs to contribute to NTD control and elimination. The set of tools we have developed help to systematically evaluate individual, organizational and system level capacity of laboratory systems for NTD control. Using the tools enabled the stakeholders and researchers to jointly develop a capacity development action plan that aimed to control or eliminate NTDs in their region. We had multi-level stakeholders involved, including laboratory staff, administrators, international organization representatives, academics, and policy makers. This creation of partnerships with a range of decision makers is known to be an effective strategy to strengthen capacity [18], [19]. The literature in the field highlights that assessment and monitoring is more often driven by those outside of the country such as donors who are often concerned with conducting fiscal assessments [20].
While the importance of individual and institutional capacity has been raised in the literature [21], this study is novel as it explores capacity within laboratory systems at the national and regional levels. The tools enabled us to explore outcomes beyond the individual level such as understanding the strengths and gaps at the organizational level (e.g., relationship between NTD laboratory and College of Medicine in Malawi). Through exploration of capacity at the organization level, it was revealed that there is a need for each laboratory to consider how they can move beyond their one specific NTD focus. This consideration of moving to a broader focus could even include discussion of the integration of NTDs into the control of the big three i.e., tuberculosis (TB), malaria, and HIV. Potential synergies between the Global Fund diseases of malaria, HIV and TB were identified by the NTD community many years ago [22]. They are all diseases of the poor and co-endemic with at least one NTD across the distribution of the WHO focus NTDs. Initially, the focus was on optimizing delivery strategies and building on common features in the supply chain management system to scale up intervention coverage in a highly cost effective way. As NTD laboratories embark on scaling up through inter-sectoral approaches, they could also capitalise on the growing support for reference laboratories, through the Global Fund for AIDS, Tuberculosis and Malaria (GFATM). NTD diagnostics could be included in the activities of these national reference laboratories. Diagnosis for the Global Fund diseases are commonly achieved using rapid diagnostic procedures based on small quantities of finger-prick blood samples that also can be used to test for many NTDs, as can DNA extracted from blood.
Using the tools also gives credence to the idea that capacity resides at different levels, including individual, institutional, national and regional but is best addressed institutionally. Addressing capacity strengthening initiatives at the institutional level is congruent with principles within theory of change evaluation which emphasise that organisations and individuals within them have a key role to play in moving from one state of capacity to another, while also acknowledging the contribution and influence of other actors outside the organisation's control [11]. Taking this systemic view, capacity strengthening can be conceptualised as a process of change within a complex system of unpredictable interactions and inter-relationships between elements and individuals. A small change in one aspect or relationship can have a significant impact on capacity, and the key to success is in observing and capturing these changes which often happen in a non-linear way. Although we only have implemented the tools in four countries thus far, the commonalities across cases suggest that our tools are appropriate for a range of contexts. We found value in transferring the tools from thee different African contexts to a South East Asian context, as the tools were found to be flexible enough to be adapted to the different country context and enabled us to collect relevant data and monitor progress in capacity strengthening. This flexibility in the tools, allowing for adaptation to different contexts, has been shown to enhance capacity strengthening initiatives [10]. We believe therefore that the tools could be used in laboratory systems beyond the scope of NTDs and would encourage further research to examine this.
This study contributed to the literature about how to assess and monitor capacity strengthening in practice. Through using the tools we learnt more about the process of capacity strengthening including the recognition that personal relationships are key to capacity strengthening initiatives. Assessing and monitoring indicators such as relationships amongst stakeholders (e.g., laboratory director and national program) is far less tangible than indicators used in the bulk of capacity strengthening research (e.g., number of people trained) [23]. This finding leads to the recognition of the value of using mixed research methods to measure changes in capacity, rather than the traditional approach of predominantly quantitative measures [24] in order to obtain an in-depth understanding of complex constructs and inter-relationships that operate in health systems.
Our novel set of assessment and monitoring tools provide a practical and field-tested approach for assessing laboratory capacity strengthening initiatives. We have implemented the tools for laboratory system strengthening in NTD laboratory systems in three countries in Africa and one country in South East Asia, but they could be adapted for use in other geographical and laboratory contexts.
|
10.1371/journal.ppat.1003426 | An Apicoplast Localized Ubiquitylation System Is Required for the Import of Nuclear-encoded Plastid Proteins | Apicomplexan parasites are responsible for numerous important human diseases including toxoplasmosis, cryptosporidiosis, and most importantly malaria. There is a constant need for new antimalarials, and one of most keenly pursued drug targets is an ancient algal endosymbiont, the apicoplast. The apicoplast is essential for parasite survival, and several aspects of its metabolism and maintenance have been validated as targets of anti-parasitic drug treatment. Most apicoplast proteins are nuclear encoded and have to be imported into the organelle. Recently, a protein translocon typically required for endoplasmic reticulum associated protein degradation (ERAD) has been proposed to act in apicoplast protein import. Here, we show ubiquitylation to be a conserved and essential component of this process. We identify apicoplast localized ubiquitin activating, conjugating and ligating enzymes in Toxoplasma gondii and Plasmodium falciparum and observe biochemical activity by in vitro reconstitution. Using conditional gene ablation and complementation analysis we link this activity to apicoplast protein import and parasite survival. Our studies suggest ubiquitylation to be a mechanistic requirement of apicoplast protein import independent to the proteasomal degradation pathway.
| The apicoplast is an essential parasite organelle derived from an algal endosymbiont. Most apicoplast proteins are nuclear encoded and post-translationally imported. Part of this journey utilizes the endoplasmic reticulum associated degradation or ERAD system of the algal endosymbiont. Typically, the ERAD system is ubiquitylation-dependent and acts in the retrotranslocation across the ER membrane and proteasomal destruction of misfolded secretory proteins. In the apicoplast, this system has been retooled into a protein importer. The apicoplast ERAD system is broadly conserved between most apicomplexans and surprisingly retains the ubiquitylation machine typically associated with destruction. This study brings together biochemical studies in Plasmodium and genetic studies in Toxoplasma. Together they provide significant mechanistic insight into the process of protein import into the apicoplast. We provide evidence that ubiquitylation may be a mechanistic requirement for import and demonstrate it to be essential to the parasite, thus providing new opportunities for drug development.
| Apicomplexans are eukaryotic pathogens and responsible for important human and animal diseases including malaria and toxoplasmosis. The Apicomplexa evolved from single-celled photosynthetic algae, and their adaptation to animal parasitism likely predates the emergence of animals from water to land. The presence of a plastid, the apicoplast, is the most important remnant of this evolutionary past [1], [2]. While no longer photosynthetic, the organelle synthesizes isoprenoids and fatty acids [3]. The apicoplast is essential for parasite survival, and its metabolism, biogenesis and maintenance are important targets for anti-parasitic drug treatment. The apicoplast was derived by secondary endosymbiosis, where a unicellular red alga was incorporated into a heterotrophic protist. As a consequence of this secondary endosymbiosis the apicoplast is surrounded by four membranes. The organelle carries a genome, yet most of its proteins are nuclear-encoded and imported into the organelle after translation. Targeting depends on a bipartite leader peptide, the first section of which mediates co-translational import into the endoplasmic reticulum, and the second part mediates delivery to the apicoplast, likely through fusion of endosomal vesicles with the outermost membrane of the organelle [4]. Three translocons breaching successive membranes have been proposed to act in the further transport of proteins into the stroma of the apicoplast [5]. The two inner membranes of the apicoplast are homologous to the membranes of the primary chloroplast and protein transport depends on systems derived from the chloroplast TIC and TOC machinery [6], [7], [8], [9]. Insight into the third translocon emerged first in cryptomonads, an algal group that like Apicomplexa harbors a secondary plastid. The secondary plastids of cryptomonads retained a remnant of the algal nucleus, the nucleomorph. Analysis of the gene content of the nucleomorph led to the discovery of plastid proteins that resembled components of the endoplasmic reticulum associated degradation (ERAD) machinery [10]. ERAD is a quality control mechanism that retro-translocates misfolded secretory proteins across the ER membrane [11]. Sommer and colleagues proposed that this mechanism has been adapted for protein import in secondary plastids [10]. There is now significant support for this hypothesis. Homologs of ERAD proteins have been identified and localized to plastids in various algal and apicomplexan species including a core of the membrane protein Der1, the AAA ATPase Cdc48 and its cofactor Ufd1 [10], [12], [13], [14], [15]. Recombinant plastid proteins can complement yeast ERAD mutants [14]. Importantly, genetic ablation of the ERAD component Der1Ap in T. gondii blocks apicoplast protein import, producing a phenotype that closely resembles ablation of the apicoplast TIC component Tic20 [6], [15].
During classical ERAD, protein translocation coincides with ubiquitylation, a process that typically employs a cascade of three enzymes: ubiquitin-activating enzyme (E1), ubiquitin-conjugating enzyme (E2), and ubiquitin ligase (E3) [16], [17]. Consuming ATP, the E1 enzyme adenylates ubiquitin at the C-terminus, creating a mixed anhydride. The sulfhydryl group of the E1 active-site cysteine then attacks the anhydride, which results in the formation of a high-energy thio-ester linking ubiquitin to E1. Ubiquitin is then passed to the active-site cysteine of the E2 enzyme. Lastly, with the aid of an E3 ligase, ubiquitin is transferred from E2 and covalently attached to the ε-amino group of a lysine in the target protein. Although clearly important in mediating ERAD, the role of ubiquitylation in protein import into secondary plastids is unclear. Interestingly, some ERAD-like ubiquitylation factors are observed in the plastids of cryptomonads, diatoms, and Apicomplexa [12], [18], [19].
While protein degradation is the key function of classical ERAD this could seem counterintuitive in the context of apicoplast protein import. However ubiquitin's functions are not limited to proteasomal degradation and extend to a variety of cellular protein trafficking systems [20]. Furthermore, ubiquitylation may be a critical mechanistic requirement of protein transport via the ERAD translocon [11], [21]. Some authors now view the ERAD associated E3 ligase Hrd1 as a favored candidate for the actual protein-conducting pore [22].
In this study, we elucidate the function of ubiquitylation in the apicoplast. We identify and localize a comprehensive set of ubiquitylating components in the apicomplexan parasites P. falciparum and T. gondii. Using recombinant apicoplast enzymes from P. falciparum we reconstitute ubiquitylation in vitro using a variety of heterologous and homologous cofactors. By genetic analysis in T. gondii we demonstrate that loss of the apicoplast-localized ubiquitin-conjugating enzyme leads to loss of apicoplast protein import and parasite demise. Importantly complementation of this mutant depends on an active site cysteine required for enzymatic activity. Taken together our experiments suggest an essential mechanistic role for the ERAD-like ubiquitylation machinery in apicoplast protein import.
Using a combination of computational approaches we identified a comprehensive set of proteins that may act as apicoplast ubiquitylation system (see Materials and Methods). The results of these analyses (summarized in Table S1 in Text S1) identified apicoplast candidates for E1, E2 and E3 enzymes in both P. falciparum and T. gondii. We next determined whether these candidates are indeed targeted to the apicoplast. We targeted the locus of T. gondii TgE1Ap by single homologous integration and placed a haemagglutinin (HA) epitope tag at the C-terminus of the protein. Stable transgenic clones show apicoplast staining when labeled with an anti-HA antibody by immunofluorescence (Fig. 1A, the P. falciparum homolog E1 is also localized to the apicoplast [12]). Our attempts to localize the candidates for apicoplast E2 by tagging the respective genes directly in the locus did not produce viable transgenics in either T. gondii or P. falciparum. Epitope fusion close to the C-terminal active domain may interfere with function and prevent replacement of the native gene. However, the coding sequence of TgE2Ap could be fused to an epitope tag in the context of an ectopic expression plasmid (maintaining the native locus). Parasites expressing this construct show apicoplast labeling indistinguishable from that observed for E1 when probed with an epitope specific antibody. To localize the Plasmodium homolog (and to aid subsequent biochemical analysis) we also expressed a portion of Mal13P1.227 fused to an affinity tag in E. coli and used the purified recombinant protein to raise a specific antiserum. Immunofluorescence assays on P. falciparum parasites with this serum produced labeling that coincides with labeling for the apicoplast marker ACP (Fig. 1 C).
Two putative apicoplast E3 ubiquitin ligases were identified in Plasmodium, PfE3cAp (PFC0740c - PF3D7_0316900) and PfE3wAp (PFC0510w - PF3D7_0312100), and two in Toxoplasma (TGME49_226740 and TGME49_304460). We attempted to tag the proteins by placing different epitopes at the C-terminus through homologous gene targeting but were not successful. In case of PfE3cAp transgenics that showed initial locus targeting were quickly lost upon selection (Fig. S1A in Text S1). However, we recovered viable transgenic parasites tagged in the PfE3wAp locus. Targeted integration of the cassette and transcription of PfE3wAp-GFP was confirmed by PCR and RT-PCR (Fig. S1B–C in Text S1). Immunofluorescence assays showed PfE3wAp-GFP to localize to the apicoplast (Fig. 1D). Finally, using an episomal expression vector, we found that the first 167 amino acids of PfE3cAp target a GFP reporter to the apicoplast (Fig. 1E).
Apicoplast proteins are often processed at the N-terminus removing a leader peptide [4]. We analyzed processing for TgE1Ap, TgE2Ap and PfE2Ap for which suitable reagents were available. TgE1Ap produces the pattern typical for apicoplast proteins, two major bands likely corresponding to the precursor (heavier band) and mature protein (lighter band) Fig. 1F. Interestingly both TgE2Ap and PfE2Ap blots showed additional bands potentially arising from further post-translational modification (Fig. 1 G, H).
While the immunofluorescence assays indicate apicoplast localization of the ubiquitylation enzymes, overlap with luminal markers is only partial (see enlarged insert in Fig. 1A). We fixed and processed TgE2AP-HA parasites for electron microscopy and incubated cryo-sections with an anti-HA antibody. Note that gold particles are found in the membranous periphery of the apicoplast (Fig. 2 and Fig. S4 in Text S1). This labeling is indistinguishable from that previously observed for the apicoplast ERAD-like proteins Der1Ap and Cdc48Ap [15] and the periplastid protein PPP1 [23]. We conclude that the apicoplast has a full complement of E1, E2 and E3 ubiquitylation enzymes localized to the periphery of the organelle, most likely the periplastid compartment as observed for the ERAD-like system in the diatom Phaeodactylum tricornitum [14], [18], [19].
We next sought to establish whether the candidate apicoplast ubiquitylation system is capable of activating and ligating ubiquitin. We amplified or synthesized sequences encoding full length PfE1LAp and PfE2Ap, or the RING domains of PfE3wAp and PfE3cAp respectively, and engineered them to be expressed as recombinant fusion proteins carrying an N-terminal glutathione S-transferase (GST) and/or six-histidine (HIS) affinity tag. Proteins of the expected size could be purified for all four constructs (Fig. 3A, B). We established biochemical ubiquitylation assays using combinations of parasite enzymes and commercially available heterologous components (Fig. 3C, recombinant human factors are shown in red, Plasmodium enzymes in green). Enzymes were incubated with recombinant human ubiquitin in a buffer containing an ATP regenerating system. When analyzed by Western blot, ubiquitin chains can be detected as ladders of high molecular weight bands [24]. Among the numerous human ubiquitin-activating enzymes tested, UBCH5a and UBCH13 were found to be suitable partners for PfE3cAp and PfE3wAp leading to robust ubiquitylation. Note that this activity is strictly dependent on the recombinant parasite E3 and absent in controls (Fig. 3D, E). The pattern obtained differed between the two E2 enzymes and suggested ubiquitylation of the RING domain in the context of only UBCH5a, while interaction with UBCH13 appeared to produce free poly-ubiquitin. Variation of ubiquitylation pattern depending on the E2 partnered with the ligase is frequently observed [25]. To test this independently we probed the in vitro reaction with anti-GST antibody to visualize the E3 and its higher molecular weight ubiquitin adducts. Consistently, this revealed shifts in molecular weight of PfE3cAp and PfE3wAp only when incubated with UBCH5a (Fig. 3F) as free polyubiquitin is not detected in this assay format.
Next we tested whether ubiquitylation activity can be reconstituted entirely with parasite enzymes. When recombinant PfE1LAp and PfE2Ap were incubated with ubiquitin alone (Fig. 3G, left lane), no ubiquitylation was detected. However, upon addition of recombinant E3 ligase PfE3wAp or PfE3cAp, ubiquitylation was readily observed. Lastly we wished to evaluate the activity for native parasite enzymes. Among the reagents generated and tested in this study a custom-made antibody to PfE2Ap was found suitable for immunoprecipitation under native conditions. Often the conjugating and ligating enzymes form a complex and can be co-precipitated and detected by their combined activity [26], [27]. We incubated pull down fractions from parasite lysates with recombinant human UBA1, and biotinylated-ubiquitin (using tagged ubiquitin enhances sensitivity and focuses the assay on only newly ubiquitylated proteins). We observed significant ubiquitylation that was dependent on the immunoprecipitate and UBA1 (Fig. 3H). Taken together our observations provide biochemical support for the notion that the apicoplast ERAD-like system is capable of mediating ubiquitylation.
The apicoplast ERAD system has a critical role in protein import into the organelle [5], [18]. We tested whether ubiquitylation is a mechanistic requirement of this process by genetic ablation of the apicoplast ERAD-like ubiquitylation enzymes. We attempted disruption of the loci of PfE3cAp, PfE3wAp, and PfsUBA1. We isolated strains bearing drug marker insertions in the PfE3wAp gene and documented loss of associated transcription (Fig. S2B, S3C in Text S1). However, we also noted multiple genomic duplications in these strains complicating interpretation (Fig. S3D in Text S1). We did not obtain viable parasites with disrupted PfE3cAP or PfsUBA1 loci. This is consistent with a potentially essential role for these proteins, and we therefore turned to T. gondii where the construction of conditional mutants is feasible.
We engineered a parasite strain where the endogenous promoter of the TgE2Ap gene was replaced by a regulatable promoter in the following referred to as (i)ΔTgE2Ap (Fig. 4A, [23]). This was accomplished by double cross over in the T. gondii TATiΔKu80 background, a parasite line that favors homologous recombination and expresses a transactivator that can be modulated using anhydrotetracycline (ATc). Drug resistant parasite clones were tested by PCR and integration of the promoter was confirmed by Southern blot. We monitored the level of TgE2Ap mRNA in response to ATc by quantitative PCR. Fig. 4D shows down-regulation of the transcript below 10% of its normal level at day four of ATc treatment. We asked whether loss of TgE2Ap affects parasite growth and performed plaque and real-time fluorescence assays. Parasites grow normally in the absence of ATc indicated by formation of plaques, however in the presence of ATc, plaque formation is severely attenuated (Fig. 4F). Similarly, (i)ΔTgE2Ap parasites show significant growth reduction in the fluorescence assay in the presence of ATc (Fig. 4E), preincubation of parasites in ATc abolished growth entirely. We conclude, that TgE2Ap is critical for parasite growth.
We next tested the ability of (i)ΔTgE2Ap parasites to import apicoplast proteins in the absence or presence of ATc and measured the import-dependent lipoylation of the apicoplast pyruvate dehydrogenase E2 subunit [6]. (i)ΔTgE2Ap parasites were treated with ATc for different periods and pulse-labeled for one hour with [35S] methionine/cysteine. For the chase samples the radioactive isotope was removed, and cells were incubated for two additional hours in normal media. The samples were then used for immunoprecipitation with an anti-lipoic acid antibody followed by separation on SDS-PAGE. Treatment of cells with ATc for 2 days resulted in attenuation of import, leading to complete loss after 4 days (Fig. 4G, H). Lipoylation of two mitochondrial enzymes remained unaffected. We also monitored apicoplast loss, a frequent consequence of interference with apicoplast protein import [6], [15]. We observed a drop over time, but note that loss of import significantly precedes plastid loss. Loss of apicoplast protein import has also been shown to result in loss of leader peptide removal and backing up of precursor protein into the ER and other elements of the secretory pathway [6], [15], [28] We therefore measured the levels of precursor and processed mature form of the apicoplast reporter protein FNR-RFP [29].. We grew parasites for 0 to 4 days on ATc and performed Western blots using parasite protein extracts from each day. Probing these blots with an antibody against RFP revealed that precursor levels of FNR-RFP remained unchanged throughout the 4 days, while the mature protein was no longer detected after 2 days on ATc further supporting a strong import defect (Fig. 4I). We also monitored the localization of FNR-RFP in treated and untreated (i)ΔTgE2Ap parasites by immunofluorescence assay. In untreated parasites FRN-RFP is restricted to the apicoplast (Fig. 4K). After 48 hours of ATc treatment 38% of parasite vacuoles also show significant labeling outside of the apicoplast surrounding the nucleus likely representing the ER (Fig. 4J, untreated TgE2Ap or ATc treated wild type parasites showed such labeling in <3% of counted four cell vacuoles, n = 200). We conclude that apicoplast protein import is impaired in the absence of TgE2Ap.
Apicoplast ubiquitylation enzymes are capable of synthesizing ubiquitin chains in vitro, but is this activity required in vivo? To test this we established a complementation assay. The coding sequence of the TgE2Ap gene driven by a constitutive promoter was introduced into the uracil-phosphoribosyltransferase (UPRT) locus of the (i)ΔTgE2Ap mutant (Fig. 5C). Parasites were selected for the loss of UPRT activity using 5-fluorodeoxyuridine [30] and a clonal cell line that now constitutively expressed a second copy of TgE2Ap in the conditional knock down background was isolated. We confirmed correct integration by PCR (Fig. 5D). We tested the ability of this strain to form plaques when expression from the native locus is ablated by ATc treatment, and found that genetic complementation fully rescued growth (Fig. 5E).
Multiple sequence alignment of TgE2Ap and E2 enzymes from a wide range of eukaryotes showed that TgE2Ap shares conserved features, reported earlier to be critical for this class of enzymes. We therefore modelled the C-terminal domain of TgE2Ap onto the structure of UBC4, a well characterize yeast ubiquitin conjugating enzyme [31]. Multiple sequence alignment and homology modelling identified C573 as the presumptive active site cysteine (Fig. 5A, B, see Fig. S3 in Text S1). Most E2 enzymes possess a signature HPN triad proximal to the active site cysteine [32]. The histidine has been previously suggested to be dispensable for E2-catalyzed ubiquitylation, but is important for the folding of the active site in other systems [33]. The asparagine residue on the other hand was consistently found to be important for RING-E3/E2-dependent ubiquitin conjugation [34]. A conserved HXH triad is found at this position in apicomplexans (Fig. 5B). We engineered a series of point mutants in TgE2Ap replacing C573, H563, and H565 with alanine respectively. These genes were then introduced into the (i)ΔTgE2Ap mutant as described above and tested for their ability to complement loss of TgE2Ap upon ATc treatment using plaque assay. Expression of the H563A point mutant fully complemented loss of native TgE2Ap (Fig. 5E) and parasites now grow even in the presence of ATc. In contrast, despite numerous attempts we were unable to establish a stable parasite line expressing H565A, which may suggest dominant effects of this mutation. We were able to isolate mutants expressing C573A, however these strains show no complementation, and are still fully susceptible to ATc treatment (Fig. 5E). We conclude that enzymatic activity is a requirement for TgE2Ap function in vivo and that C573 and H565 residues are critical for the function of the enzyme while H563 is likely dispensable.
Endosymbiosis is a key evolutionary mechanism underlying the emergence and diversification of eukaryotes – in particular for photosynthetic eukaryotes. The acquisition of a eukaryotic red algal symbiont led to the chromalveolates, a large super-phylum of tremendous ecological diversity that includes apicomplexan parasites. The descendent of the algal symbiont, the apicoplast, maintains a highly compartmentalized organization, and nuclear encoded proteins have to overcome four membranes on their journey to the stroma. An apicoplast-localized ERAD-like system appears to play an important role in apicoplast protein import. Recent reports have identified and characterized components of this ERAD–like system in different algal and parasite species [7], [10], [12], [13], [14], [15]. In this study we provide significant biochemical and genetic evidence for the hypothesis that an apicoplast localized ubiquitylation cascade is an essential element of this protein import system. We identify apicoplast ubiquitin activating, conjugating and ligating enzymes in two important apicomplexan parasites, P. falciparum and T. gondii. We show in vitro and in vivo that these proteins have conserved biochemical activities and are capable of ubiquitin transfer. Finally, in genetic studies, we show that TgE2Ap, for which we were able to isolate a conditional mutant, is essential for apicoplast protein import, organellar maintenance and parasite growth. Overall these observations support a direct mechanistic role of ubiquitylation in protein translocation independent of ubiquitin's function in proteasomal degradation [11]. The classical ERAD system is believed to recognize and respond to the folding state of secretory proteins. Interestingly, recent studies show that the transit peptide of apicoplast proteins is primarily unstructured and that this conformation may be critical for proper transport to the organelle [35]. This model would need a distinguishing element to avoid elimination of apicoplast proteins by the classical ERAD. Specific chaperone sets could potentially provide such specificity, but remain to be discovered. A recent study in Arabidopsis has identified a role for ubiquitylation also in primary plastids, however this role appears to be distinct from secondary plastids. In this case ubiquitylation results in degradation of the components of the TOC complex and is thought to more globally regulate chloroplast biogenesis during plant development [36].
The identity of the apicoplast ubiquitin or ubiquitin-like modifier remains a significant unresolved question. Our results demonstrate that apicoplast enzymes are capable of acting on archetypical ubiquitin (recombinant human protein), studies in P. tricornitum show similar activity for a E3 ligase found in the diatom secondary chloroplast [18]. However whether the apicoplast system actually utilizes ubiquitin in vivo remains to be established. As shown in Fig. 1G and H Western blots for TgE2Ap and PfE2Ap show additional bands. It is conceivable that these bands represent ubiquitin or a ubiquitin-like protein covalently bound to the active site of the apicoplast localized E2. However we note that, for TgE2A, none of the bands was affected by reduction of the protein or point mutation of the active site cysteine (data not shown). Alternatively this may indicate an ubiquitin-like protein bound to a residue different from the active site of the enzyme or multiple processing steps as have been observed for some apicoplast membrane proteins [37]. Our efforts to demonstrate ubiquitin bound to apicoplast ubiquitylation enzymes purified from P. falciparum or T. gondii so far did not result in robust detection (using either antibodies or mass spectrometry, data not shown). Furthermore ubiquitylation of plastid-bound cargo proteins is not readily observed in apicomplexans or diatoms. A reasonable candidate for which apicoplast localization has been suggested [12] is an atypical, large ubiquitin-like protein (PF08_0067). Curiously, this protein lacks the di-glycine motif typically required for the formation of the isopeptide bond and a homolog has yet to be identified in the Toxoplasma genome. Similarly, plastid ubiquitin candidates from algae show lack of sequence motifs typically required for polyubiquitylation [19]. It is conceivable that this ubiquitin-like protein could be processed and/or ligated in a novel fashion that does not depend on a di-glycine sequence. Alternatively, its function may resemble that of the HERP protein in the classical ERAD pathway. Like PF08_0067, HERP has an ubiquitin-like domain at the N-terminus followed by transmembrane domains at the C-terminus [38]. HERP is believed to interact with HRD1 and to regulate the ubiquitylation activity of the ERAD translocon in response to folding stress [39]. In that case PF08_0067 is likely not the main substrate for the apicoplast ubiquitylation system and the modifier is yet to be discovered.
Studying the apicoplast ubiquitin faces technical obstacles that so far prevented direct tagging of the candidate ubiquitin and subsequent detection of modified cargo. There are several strong candidates for plastid-localized deubiquitylation enzyme in apicomplexans and diatoms (Table S1 in Text S1, [13], [18]). The activity of these enzymes may dramatically shorten the time ubiquitin remains on proteins and thus prevent the robust detection of ubiquitin adducts [22]. Isolation of mutants lacking apicoplast deubiquitylation might allow testing of this hypothesis and potentially lead to accumulation of modified cargo proteins. While a number of mechanistic details of the apicoplast ubiquitylation system remain to be elucidated, we demonstrate that the system is essential to the organelle and the parasite. Building on a longstanding effort to target ubiquitylation for the development of anti-cancer drugs [40] may potentially lead to new anti-parasitic compounds in the future.
P. falciparum strains 3D7, D10_ACP-(leader)-GFP (MR4, MRA568) and derivatives were cultured in human O+ red blood cells [41]. T. gondii RH strain parasites and derivatives were propagated in human fibroblasts and genetically modified as described [6], [15].
For in vitro ubiquitylation assays recombinant P. falciparum enzymes were incubated with recombinant human or parasite factors. Typically 50–200 µM recombinant ubiquitin, 0.05–0.2 µM E1 enzyme, 1–5 µM E2 enzymes, and 1–12.5 µM of E3 ligases were incubated in 50 mM Tris-HCl, pH 7.4, 1 mM DTT in presence of a re-energizing system (BostonBiochem) containing the ATP and ATP regenerating enzymes to recycle hydrolyzed ATP needed for the assay, for 2 hours at 37°C followed by SDS-PAGE and immunoblotting.
Ex vivo ubiquitylation assays were performed by lysing 3D7 P. falciparum in 20 mM HEPES pH 7.9, 10 mM KCl, 1 mM EDTA, 1 mM EGTA, 1 mM DTT, 0.5 mM AEBSF (Fisher Scientific), 0.65% Igepal v/v, and protease inhibitor cocktail (Roche), or 20 mM HEPES pH 7.9, 0.1 M NaCl, 0.1 mM EDTA, 0.1 mM EGTA, 1.5 mM MgCl2, 1 mM DTT, 1 mM AEBSF and protease inhibitor cocktail (Roche). Supernatants were pooled and proteins were precipitated using the indicated antibodies and magnetic Protein A beads. Proteins bound to beads were mixed with re-energizing buffer, 0.5 µg/µl biotin-conjugated ubiquitin, 5 mM AEBSF and protease inhibitor cocktail. Reactions were incubated at 30°C with gentle agitation for two hours. Samples were eluted with 4× Laemmli buffer and analysed using biotin affinity blots. Human recombinant UBE1 and UBC enzymes, E3 ligases biotin conjugated ubiquitin and re-energizing buffer used in these assays were purchased from Boston Biochem.
T. gondii gene models were tested by 5′- and 3′-RACE. Note that additional exons were identified for TgE2Ap (see genbank JX431938 for correct sequence). A conditional TgE2Ap knock-out was generated by exchanging the native promoter for the tetracycline inducible t7s4 promoter in the TATiΔKu80 parasite background. The targeting construct used 1.2 kb up- and 1.5 kb downstream of the TgE2Ap start codon introduced into vector pDT7S4. Linearized plasmid was transfected into the parental strain followed by pyrimethamine selection. To complement the knock-out, a TgE2Ap minigene was inserted into the UPRT locus under the control of a constitutive sag1 promoter. Transgenics were isolated in 5 µM 5-FUDR and identified by PCR. Parasite growth was measured by fluorescence and plaque assay in the presence and absence of 0.5 µm anhydrotetracycline (ATc). Please refer to the supplement materials for a more detailed description of materials and methods used in this study (including a table of all primers).
|
10.1371/journal.pcbi.0030193 | Fast Pairwise Structural RNA Alignments by Pruning of the Dynamical Programming Matrix | It has become clear that noncoding RNAs (ncRNA) play important roles in cells, and emerging studies indicate that there might be a large number of unknown ncRNAs in mammalian genomes. There exist computational methods that can be used to search for ncRNAs by comparing sequences from different genomes. One main problem with these methods is their computational complexity, and heuristics are therefore employed. Two heuristics are currently very popular: pre-folding and pre-aligning. However, these heuristics are not ideal, as pre-aligning is dependent on sequence similarity that may not be present and pre-folding ignores the comparative information. Here, pruning of the dynamical programming matrix is presented as an alternative novel heuristic constraint. All subalignments that do not exceed a length-dependent minimum score are discarded as the matrix is filled out, thus giving the advantage of providing the constraints dynamically. This has been included in a new implementation of the FOLDALIGN algorithm for pairwise local or global structural alignment of RNA sequences. It is shown that time and memory requirements are dramatically lowered while overall performance is maintained. Furthermore, a new divide and conquer method is introduced to limit the memory requirement during global alignment and backtrack of local alignment. All branch points in the computed RNA structure are found and used to divide the structure into smaller unbranched segments. Each segment is then realigned and backtracked in a normal fashion. Finally, the FOLDALIGN algorithm has also been updated with a better memory implementation and an improved energy model. With these improvements in the algorithm, the FOLDALIGN software package provides the molecular biologist with an efficient and user-friendly tool for searching for new ncRNAs. The software package is available for download at http://foldalign.ku.dk.
| FOLDALIGN is an algorithm for making pairwise structural alignments of RNA sequences. It uses a lightweight energy model and sequence similarity to simultaneously fold and align the sequences. The algorithm can make local and global alignments. The power of structural alignment methods is that they can align sequences where the primary sequences have diverged too much for normal alignment methods to be useful. The structures predicted by structural alignment methods are usually better than the structures predicted by single-sequence folding methods since they can take comparative information into account. The main problem for most structural alignment methods is that they are too computationally expensive. In this paper we introduce the dynamical pruning heuristic that makes the FOLDALIGN method significantly faster without lowering the predictive performance. The memory requirements are also significantly lowered, allowing for the analysis of longer sequences. A user-friendly (still command-line based, though) implementation of the algorithm is available at the Web site: http://foldalign.ku.dk
| Noncoding RNA (ncRNA) genes and regulatory structures have been shown to be both highly abundant and highly diverse parts of the genome [1,2]. One theory is that many of these ncRNAs are part of RNA-based regulatory systems [3]. Recently, several papers about large-scale searches for vertebrate RNA genes or motifs using comparative genomics have been published [4–6]. These large-scale searches indicate that there are potentially many unknown structures still hidden in the genomes.
It has been shown that alignment of ncRNAs requires information about secondary structure when the sequence similarity is below 60% [7]. The reason for this is that compensating mutations change the primary sequence without changing the structure of the molecule. The Sankoff algorithm for simultaneously folding and aligning of RNA sequences can in principle be applied to cope with this [8]. However, the resource requirements of the algorithm are too high even for a few short sequences. For two sequences of length L, the time complexity is O(L6) and the memory complexity is O(L4). Heuristics are therefore needed before the algorithms for folding and aligning RNA sequences become fast enough to be useful.
FOLDALIGN 1.0 was the first simplified implementation of the Sankoff algorithm [9]. It contained a simple scoring scheme with separate substitution matrices for base-paired and single-stranded nucleotides. It had three constraints: (i) the length of the final alignment could not be longer than λ nucleotides; (ii) the maximum length difference between two subsequences being aligned was limited to δ nucleotides, and (iii) it could only align stem-loop structures. The second version of the algorithm uses a combination of substitutions (similar to the RIBOSUM matrices [10]) and a lightweight energy model to align two sequences [11]. FOLDALIGN 2.0 also uses the λ and δ constraints, but it can align branched structures. This algorithm was used in one of the large-scale searches for vertebrate ncRNAs [6].
Variants of two types of heuristics are currently very popular, namely pre-aligning and pre-folding. The pre-aligning methods use sequence similarity to limit the search space by requiring that the final alignment must contain the pre-aligned nucleotides. The length of the pre-aligned subsequences varies from short stretches called anchors [12] to full sequences [13–16]. Methods that require the sequences to be fully aligned before the structure is predicted are not strictly Sankoff-based methods, as they separate the alignment and folding steps completely. Pre-folding uses single-sequence folding to limit the structures that can be found by the comparative algorithms. A popular method of pre-folding is to use base-pairing probabilities found by the single-sequence folding to limit which base pairs can be included in the conserved structure [17,18]. As for align-then-fold methods, methods using pre-folding can be taken to the extreme where the folding and alignment steps are completely separated. One example of this is the combination of the RNAcast and the RNAforester methods [19,20]. Some methods can use both pre-aligning and pre-folding heuristics [21–23].
The currently implemented Sankoff-based methods, for pairwise alignment and secondary-structure prediction of RNA sequences, can be split into two groups: the energy-based methods and the probabilistic methods. The energy-based methods, FOLDALIGN, Dynalign [24], locaRNA [17], and SCARNA [23], are based on minimization of the free energy [25]. Free-energy minimization is based on a physical model of how the different elements of an RNA structure contribute to the free energy. The parameters are partly found experimentally and partly estimated from multiple alignments. The probabilistic models are usually based on Stochastic Context Free Grammars (SCFGs); see [26] for an introduction. These methods include Consan [12] and Stemloc [22]. The Stochastic Context Free Grammars parameters are estimated from multiple alignments. Each of these methods uses different kinds of heuristics. The previous version of FOLDALIGN [11] uses banding (δ, see below), limits the alignment length for local alignments (λ, see below), and limits the number of ways a bifurcation can be calculated (described below). Dynalign [24] uses banding based on pre-alignment using a hidden Markov model. LocaRNA [17] limits the number of potential base pairs by only using base pairs with a single-sequence base-pair probability above a given cutoff. SCARNA [23] uses a similar strategy, but further decouples the left and right sides of the base pairs. Consan [12] uses short stretches of normal sequence alignments to constrain the folding. Stemloc [22] uses the N1 best single-sequence–predicted structures and the N2 best normal-sequence alignments to limit the final combined alignment and structure prediction.
In this paper, dynamical pruning of the dynamic programming matrix is introduced as a new heuristic in the FOLDALIGN algorithm [11]. In all its simplicity, the dynamic pruning discards any subalignment that does not have a score above a length-dependent threshold. This is similar to one of the heuristics used in BLAST [27]. The advantage of the pruning method compared with the pre-aligning methods is that it can be used when there is not enough sequence similarity to make the necessary alignments. The advantage compared with the pre-folding methods is that none of the comparative information is lost in a single-sequence folding step. It is shown empirically that the pruning leads to a huge speed increase while the algorithm retains its good performance. The speed increase makes studies like [6] much more feasible. The method of dynamical pruning is simple and general. It should therefore be possible to use it in many of the other methods available for folding and aligning RNA sequences. As pruning is a feature of the dynamic programming method, it may be used in any algorithm using dynamic programming.
In addition to the dynamical pruning, the FOLDALIGN software package has been significantly updated. The constraint, which speeds up the algorithm by limiting the calculation of branch points, is now also used to lower the memory requirement during the local-alignment stage. During the backtrack stage of the algorithm, more information is needed. To try to limit the memory consumption during this stage, an extra pre-backtrack step is used. The pre-backtrack step locates all branch points in the conserved structure, and these are then used to divide the structure into unbranched segments. The unbranched structures are then realigned and backtracked separately. As the unbranched structures usually are shorter than the full branched structure, the memory consumption is reduced. The use of the divide and conquer method increases the run time of the algorithm, but not by much since the realignments of the segments are unbranched.
In addition to the algorithmic improvements, the energy model has been improved as well. External single-strand nucleotides are scored in a consistent way. Also, more insert base pairs are allowed. These improvements lead to better structure predictions.
FOLDALIGN is a tool for making local or global structural alignments of RNAs [9,11,28–31]. It uses a combination of sequence similarity and structure to make the alignment. The present article shows how the two main inconveniences, the time and memory requirements, can be drastically lowered without losing the ability to make good alignments.
The main improvement of the algorithm is the dynamical pruning. The pruning eliminates subalignments that are so poor that they can be assumed to never be a part of a biologically relevant alignment [27]. The pruning works by eliminating all subalignments with a score below a length-dependent threshold. This not only removes the subalignment itself but also all the longer alignments that the subalignment would have become a part of. Elimination of subalignments also improves the memory performance since it is not necessary to store the eliminated alignments. The dynamical pruning method is general to the dynamic programming method, and it should therefore be trivial to use it in other applications where dynamic programming is used. Note however, that when pruning is used there is no guarantee that the solution is the optimal solution, and in some cases a solution is not found at all. In these rare cases pruning is not feasible, but FOLDALIGN can provide an alignment by realigning without pruning.
The memory requirement of the algorithm is further lowered by exploiting the branch point constraint. In [11] the calculation of branch points was limited to one calculation for equivalent alignments/structures. This speeds up the algorithm significantly. Here, it is also used to lower the memory requirement; see below and the Memory implementation section of Protocol S1.
Much effort has been put into trying, during backtrack, to keep the memory requirement below what is needed during the initial local-alignment scan. This is done by using a divide and conquer scheme that first locates all branch points in the structure, and then uses them to divide the structure into hopefully shorter unbranched segments. These segments are then backtracked normally. While this strategy usually works, there is no guarantee that it will always keep the memory usage low during backtrack. One clear example where it does not work is in the case of an unbranched alignment. A cubic space model similar to the linear space models [32,33] used in sequence alignment could be used to keep the memory requirement below a given cutoff. A method similar to the Treeterbi method [34], of locating and removing memory cells that are not part of final alignment during the alignment step, could also be used to limit the memory consumption during the realignment of stem segments.
The algorithm uses a lightweight energy model based on energy minimization [11,25] and sequence similarity to simultaneously fold and align two sequences. The energy model has five different contexts: stem, hairpin-loop, bulge-loop, internal-loop, and external/bifurcated-loop. In the stem context, two pairs of nucleotides are allowed to base-pair if both pairs can form an A ⇔ U, C ⇔ G, or G ⇔ U base pair. A stem must always start with such a conserved base pair, but if the stem is already at least one base-pair long, then a base pair in one of the sequences can be aligned to two gaps in the other sequence. A stem must contain at least two conserved base pairs. The four other contexts are used to align unpaired nucleotides; see Protocol S1 and [11] for details. The parameters of the energy model have all been multiplied by −10 mol/kcal to allow the maximization of the alignment score [11]. Different sequence similarity substitution scores are used for base-paired nucleotides and single-stranded nucleotides [10,11]. Affine gap penalties are also used.
The alignment of two sequences I and K is initiated by aligning one nucleotide from each sequence to another. These are then expanded into longer alignments using dynamic programming. The central part of the recursion can be seen below. A more complete recursion can be seen in the recursion section of Protocol S1.
Di,j,k,l is the score of an alignment of subsequence (i,j) from sequence I to subsequence (k,l) from sequence K. Sbp through SgrK are the costs of adding one or more nucleotides to the alignment. σ is the structure context. i & j, k & l, are the start and end coordinates of a subalignment in sequence I and K, respectively. ni is the nucleotide at position i (in sequence I). Likewise for nj, nk, and nl. Here, unpaired nucleotides in branched loops are scored like unpaired nucleotides in external loops. Therefore, the alignment score D must be corrected if the context is not the external loop context. The score including this correction is called D′, and it is not necessary to store it in memory since it is easily calculated from D; see Protocol S1. Cmblhelix is a cost for adding extra stems.
Equation 1A adds a base pair in both structures. Equations 1B–1C add base-pair inserts in either of the structures. Equations 1D–1E add aligned unpaired nucleotides in either end of the alignment. Equations 1F–1I add an unpaired nucleotide aligned to a gap to the alignment. Equation 1J is the bifurcation case which joins two substructures into one in each of the structures. Figure 1 gives an overview of the different Equation 1 cases.
In addition to the new pruning constraint (see below), the algorithm employs three old constraints. 1) The maximum motif length constraint λ limits the maximum length of the subsequences in the resulting alignment (the final alignment can only be longer than λ due to gaps). The use of λ allows the program to split the shortest of the sequences into smaller chunks and scan a λ nucleotide–long window along the other. For details see [11]. 2) The maximum length difference between any two subsequences is constrained to δ nucleotides. These two constraints reduce the time complexity to a maximum of O(LILKλ2δ2) and the memory complexity to a maximum of O(λ3δ). LI and LK are the lengths of the sequences. 3) The bifurcation constraint limits the types of substructures that can be joined together in the case of Equation 1J. The constraint requires that the first nucleotide of each of the two (one from each sequence) upstream substructures, Di,m,k,n, are base-paired. Furthermore, the end nucleotides of the downstream structures, Dm+1,j,n+1,l, must form a base pair with each other, see Figure 2. In [11], the bifurcation constraint was used to save time, but here it is also used to save memory. Inspection of Equation 1 shows that the cases A–I only depend on the scores of alignments with either coordinate i or coordinate i + 1. Case J depends on subalignments with start coordinates i and coordinates i < m + 1 < j. Due to the bifurcation constraint, all alignments with start coordinate m + 1 must have the stem context. It is therefore not necessary to keep information about alignments in any other context for coordinates i + 2,...,i + λ. This saves a large amount of memory, but leads to trouble during backtrack, see below. Further details can be found in the Memory implementation section of Protocol S1.
The new implementation of the FOLDALIGN algorithm has a lower time and memory complexity than the old implementation during global alignment. Since a global alignment must include both ends of both sequences, the δ parameter can be used to also limit the start coordinate of a subalignment in the second sequence. In this way the δ parameter becomes similar to the M parameter as used in [35]. The new time complexity is
and the memory complexity is
, where Lmin = min{LI, LK}. The old implementation has a time complexity of
and a memory complexity of
since it used the local-alignment algorithm with λ equal to the sequence lengths.
Figures 4 and 5 show the average time and memory needed to locally align two 1,000 nucleotides–long sequences with different values of λ using different versions of FOLDALIGN. The “real data” curves are for sequence pairs that contain a ∼300 nucleotides–long SRP gene in its genomic context. The “shuffled data” curve is for the same sequence pairs shuffled while conserving the dinucleotide distribution [39]; for details about the dataset see [11]. The curve marked “2.0” is for the previous version of FOLDALIGN. The “No pruning” curve is for the current version using the option -no_pruning which turns off the pruning. The “Pruning” curves are for the current version using the default pruning.
Figure 4 shows that the dynamical pruning has a dramatical impact on the run time of the algorithm. Especially in the case of shuffled sequences where there are no conserved motifs. This is very important during large-scale searches for ncRNAs where a large amount of sequences, many of which do not contain a motif, are aligned [6]. When there is a conserved motif, the speedup is slightly less pronounced, but it is still very drastic. Comparing the “2.0” and the “No pruning” curves, it is also clear that the new implementation is a lot faster even without pruning. The time needed to run the 99 sequence pair dataset used for finding the parameters (pairs of 500 nucleotide–long sequences, see Materials and Methods; λ = 200, δ = 15) dropped from ∼397 CPU hours to ∼9 CPU hours.
Figure 5 shows that pruning also has a significant impact on the required memory. It also shows that it is now possible to align sequences using much larger λ. It is clear from Figure 5 that more memory is required for aligning sequences that contain a conserved structure than for aligning sequences without conserved structures.
The parameters of the algorithm can be split into three groups: Energy, Substitution, and Free. The energy parameters are taken from energy minimization (multiplied by −10 mol/kcal, since we are maximizing a score) [25]; see also the release notes of the software package. The substitution scores are the Ribosum-like log odds scores with one set of scores for single-stranded nucleotides and one for base-paired nucleotides. There are four free parameters: the Ribosum clustering percentage, the gap open, the gap-elongation penalties, and the energy-substitution weight. To simplify the search for the best set of parameters, the gap-elongation parameter was fixed at half the value of the gap-open parameter. The best values for the parameters were found by aligning the 99 sequence dataset using a range of values for each of the free parameters and choosing the values yielding the best performance. The parameters can be seen in Table 1. See Materials and Methods and [11] for details about the data.
To get the background distribution, each sequence was shuffled 20 times conserving the dinucleotide distribution [39] (see Materials and Methods). The extreme value distribution parameters were averaged independently for each of the sequences. The parameters found from the alignment of shuffled sequences were then used to estimate the significance of the alignments from the real sequences pairs. A p-value (see Materials and Methods) cutoff of 0.2 was found to be optimal for selecting the significant alignments. The performances (see Materials and Methods) on the 99 sequence dataset can be seen in Table 2. While this method for determining the parameters of the extreme value distribution is biased in several ways, most notably by the use of short sequences compared with the expected length of a random alignment and the finite length effects, the method does seem to yield reasonable results. Fewer than 20 shufflings can be used with little effect on the performance (unpublished data).
The SRP dataset was used as an independent test set. The extreme value distribution parameters were estimated from the alignment of 20 shufflings of each sequence pair and a p-value cutoff of 0.2 was used. This yielded eight out of eight SRP genes, and three false positive alignments. The estimated p-value of the eight true hits is <0.0005. The p-values of the three false positives are <0.0005, 0.069, and 0.134. The positive predictive value, PPV, is 0.73, and the sensitivity is 1.00. This shows that the algorithm can be used as a general tool for finding new RNA structures.
Initial tests of FOLDALIGN's global-alignment performance while using pruning showed that the simple pruning used during local alignment removes too much during global alignment. Many sequence pairs simply did not produce an alignment. The “problem” is that the global alignment must insert a minimum number of gaps equal to the length difference between the two sequences. When the length difference is large, the cost of inserting the minimum number of gaps is enough to make the algorithm prune away all alignments. To circumvent this problem, a special global-alignment pruning scheme is used; see the previous section called Pruning.
Using the global pruning scheme, most global alignments between related sequences produce an alignment. Unfortunately, this also lowers the efficiency of the pruning significantly. Figure 6 shows the time needed to do an alignment without pruning divided by the time needed to do the same alignment using pruning as a function of the length difference between the two sequences. When the length difference is small, it is significantly faster to use pruning. When the length difference is large, there is only a small speed advantage. At a length difference of 25, the use of pruning is only ∼20% faster than not using pruning.
The global 5S rRNA and tRNA datasets (see Material and Methods) were used to select the parameters of the global score matrix, see Table 1 (gap penalties, substitution versus energy weight, and Ribosum clustering percentage). The global SRP and RNaseP datasets were used as validation datasets. As a performance measure, the Matthews correlation coefficient (MCC) of the base-pair prediction was used [40]. Correctly predicted base pairs are counted as true positives, predicted base pairs that are not found in the annotation are counted as false positives. Annotated base pairs not found in the prediction are counted as a false negative. Positions not predicted to base-pair that are not annotated to base-pair, are counted as true negatives. The average MMCs are: 5S rRNA 0.81, tRNA 0.86, RNaseP 0.50, and SRP 0.49. The results for the RNaseP and SRP datasets indicate that the good performance reported for 5S rRNA and tRNA may be due to overfitting. If this were the case, then it should also be possible to overfit on the RNaseP and SRP datasets. These datasets were therefore used to find alternative sets of parameters. The best MCCs found for both datasets were 0.56. The poor performance therefore does not seem to be due to overfitting. Some of the performance difference is likely to be due to structural inserts in the structures. Some of the sequence pairs in both the RNaseP and the SRP datasets contain stem inserts which FOLDALIGN currently cannot handle. The 5S rRNA and tRNA datasets contain fewer large stem inserts.
Recently Dowell and Eddy published results [12] which showed that while FOLDALIGN makes good alignments, its base-pair prediction sensitivity is slightly lower than that of other methods for folding and aligning RNA sequences. Since the dataset used in [12] also contains more sequence pairs than the Bralibase dataset [7], the dataset [12] is used to test the global-structure prediction performance of the algorithm. Figure 7 shows the performance of the old and new versions of FOLDALIGN and some of the other methods which can make pairwise structural alignments of RNA using the dataset from [12]; see Table 3 for details about the methods. In addition to the methods shown in Figure 7 and Table 3, the combination of RNAcast and RNAforester as described in the main page of the RNAshapes package was also tested. The combined method only produced an alignment for less than 60% of the sequence pairs and the results are therefore not reported. In cases where the combined method did produce an alignment, the alignment usually looked good. The data used is the dataset from Figure 7 in [12]. The performance measure is the MCC (see Materials and Methods). Figure 7 shows that the performance has improved from the old to the new version of FOLDALIGN. It also shows that the pruning constraint does not affect the performance in most ranges of sequence identity. Only in the identity range from 0.1 to 0.2 is there a significant difference. This is due to three sequence pairs (less than 1% of the dataset) for which no alignments are found. In these cases, pruning is not feasible; however, an alignment can still be obtained by running FOLDALIGN without pruning. Table 3 shows how long a time it takes to align the full dataset [12]. The memory requirements are also shown in Table 3. Two types of memory consumption are shown. The first (Max) is the maximum amount of memory (Resident Set Size, RSS) used during the alignment of the entire dataset. The second (Ave) is the average amount of memory needed to align each of the pairs. For “FOLDALIGN 2.1” and “FOLDALIGN 2.1 -no pruning”, the Max number (68 Mb and 323 Mb, respectively) are due to a pair of tRNAs which are predicted to a have a stemloop structure. The divide and conquer algorithm can therefore not be used to split the backtrack into smaller sections, and the memory requirement becomes high. Solutions to the stemloop problem can be to use a cubic space models [32,33] or a method similar to the Treeterbi algorithm [34]. If this one alignment is ignored, the Max memory consumptions are 21 Mb and 77 Mb, respectively. Locarna [17] is the fastest and requires the least amount of memory, but it is also the least accurate. FOLDALIGN is slower and uses more memory, but it is also one of the most accurate. Dynalign [24] also makes accurate alignments but is slower than FOLDALIGN. Consan [12] and Stemloc [22] are slow and use large amounts of memory without being more accurate than Dynalign and FOLDALIGN. The computer used to make these tests runs Linux (kernel 2.6) on two 2.4-GHz 32-bit Intel Xeon CPUs, and it has 4 Gb of memory.
FOLDALIGN is a fast and efficient tool for making pairwise local or global alignments of RNA sequences. Whereas there exist a number of methods able to make global alignments, FOLDALIGN seems still to be one of the very few tools that can do pairwise local structural alignment of RNA sequences. Thus the main motivation was in particular to improve on this aspect of FOLDALIGN. In accordance, this paper described a new way of improving the run time and memory efficiency of dynamical programming methods. Furthermore, there are several improvements to the software package. The main changes are: pruning of the dynamical programming matrix, a better implementation of memory usage, and a better energy model.
The pruning of the dynamical pruning matrix works by requiring that a subalignment must have a score above a length-dependent cutoff. Otherwise, the subalignment is removed from the dynamical programming matrix and cannot be part of any longer subalignment. Pruning efficiently slashes run time and memory requirements without degrading the predictive performance. Using pruning to speed up other dynamical programming applications should be straightforward.
The memory usage of the implementation is further improved in two ways: in the branch point calculation and during backtrack. In the calculation of branch points, two substructures are only added together if the nucleotides at the start and end positions of the downstream substructure are base-paired. Therefore, no downstream subalignment is saved unless its start and end positions base-pair. A similar method was used to speed up the algorithm in [11], but now it is used to both speed up the calculation and to lower the memory consumption. During backtrack, information is needed for every cell in the dynamic programming matrix that will be passed by the traceback algorithm (pruned alignments will not be passed and pruned cells will therefore not be needed). A divide and conquer approach has been implemented which first locates the branch points in the common structure, and then uses them to divide the structure into hopefully smaller unbranched segments. These can then be backtracked separately. While the improvement to the branch point calculation and the divide and conquer approach are more specialized than the pruning heuristic, they are likely to be of use for other RNA alignment methods.
The improvements in speed and memory requirements are important as they make studies like [6] more feasible, and thereby help to elucidate ncRNA genes' and structures' role in molecular biology.
The energy model has been improved in two ways. The first is that insert basepairs are now allowed at any position in a stem except for the first one. The second improvement is that external single-stranded nucleotides are now always scored in a consistent way—namely, as single-stranded nucleotides in a multibranched loop.
Even though the main focus for the FOLDALIGN algorithm is local alignment, the global-alignment test shows that the algorithm is the most accurate method for making global alignments of low-similarity sequences. There are several directions where the resource requirements can be improved upon. Firstly, one could introduce an align-then-fold step, to the aid of finding a lower bound for the pruning threshold. Secondly, a pre-scan for all dinucleotides could be made to initiate all stems. Thirdly, restricting the fold envelope (as introduced by Holmes [22]) by compiling a list of different shapes [19,20] might be used as a further constraint. Finally, adding a Markov model, as in Harmanci et al. [24], for initial restriction of the align envelope might aid in further resource improvements.
The FOLDALIGN software package is released under the GNU public license. It has been tested on Linux operating systems, but should run on any system that can compile standard ANSI C++. Invoking the FOLDALIGN program with the option -help will print a short description of all available options. The package can be downloaded from http://foldalign.ku.dk.
In conclusion, FOLDALIGN constitutes an efficient tool for pairwise local and global structural alignment of RNA sequences. With the introduction of pruning, the time of the genomic screen (ten chromosomes between human and mouse) in [6] was reduced from five months (on 70 CPUs) to only one week.
Two datasets are used to optimize the local-alignment parameters and evaluate the performance. Each sequence pair in the datasets contains at least one conserved RNA structure and the surrounding context. The sequences of the region with this conserved structure are at most 40% identical. The single-strand minimum folding energy of the conserved structure is indistinguishable from the folding energy of the surrounding sequence. The genomic contexts were found in GenBank [41]. The first dataset (used for optimizing the parameters) consists of 99 sequence pairs. The sequence plus context sequence is 500 nucleotides long. The conserved structures are 5S rRNA (two pairs) [42], Purine riboswitches (five pairs) [43–45], THI riboswitches (21 pairs) [45–50], tRNA (65 pairs) [51], or U1 (six pairs) [52]. For some of the sequences, there are other RNAs annotated within the context region (mainly tRNAs). These extra RNAs are annotated in the dataset if the entire RNA sequence is part of the context. Partial RNAs are annotated as intergenic sequence. The total number of the tRNA pairs in the dataset is therefore 277. The second dataset (used to evaluate the performance) consists of eight SRP pairs [53]. For details about the datasets, see [11].
To train and test FOLDALIGN's global-alignment performance, a new dataset has been made. The sequence pairs of the dataset were selected from the 5S rRNA, RNaseP, SRP, and tRNA databases [42,53–55]. Any sequences containing nucleotides other than A, C, G, or U were removed from the databases. A few sequences which obviously did not fit into the databases were removed. Then the sequences in each database were redundancy-reduced to 90% similarity using the Hobohm 2 algorithm [56]. Sequence pairs were selected from the remaining sequences by sorting the pairs by their identity and selecting the pairs with the lowest identity. Each sequence can only be part of one sequence pair. The structures were cleaned by annotating any non A - U, G - C, or G - U base pair as single-stranded. Nucleotides annotated to base pairs with gaps were also reannotated to be single-stranded.
The 5S rRNA database has three separate sections (Eubacteria, Eukaryota, and Archaea). Each section was treated separately before the final datasets were joined. This part of the data contains 215 sequence pairs. From the RNaseP database, only the sequences in the bacterial type A alignment [57] were used, as this alignment seems to have the most sequences and the best annotation. This dataset has 101 sequence pairs. The SRP dataset contains 121 sequence pairs. The pseudo-knot base pairs were removed from the structures. The tRNA dataset contains 1,810 sequence pairs.
The global-alignment dataset made by [12] contains 324 sequence pairs from the tRNA (184 pairs) and 5S rRNA (140 pairs) families of RFAM [45].
By default, FOLDALIGN returns the structure, alignment, and positions of the best-scoring local alignment of a pair of sequences. However, it can also output the score and coordinates of the alignment with the highest score compared with the log of its length for each pair of positions (i,k) in the two sequences [58]. The list of scores and coordinates are turned into a ranked list of non-overlapping alignments. As described in [11], this is done by the following. 1) Find the alignment with the highest score compared with the logarithm of the sequence lengths. 2) Remove all alignments that overlap the alignment found in step 1. 3. If more alignments are available and desired, go back to step 1.
Ideally, the significance of an alignment is found by comparing its score to a large number of scores from shuffled alignments using the extreme value distribution [59,60]. With a method like FOLDALIGN, alignment of thousands of random sequences is not feasible. In [61,62], it is suggested that from each alignment it is possible to use more than just the score of best alignment to find the parameters of the extreme value distribution.
The parameters κ and Λ (Λ is called λ in other texts) of the distribution are found using the method described in [37]:
where A is the mean of all alignments scoring above a cutoff C, and nA>C is the number of alignments scoring above the cutoff. In [11], several values of C were tested, and one assumed to be optimal was chosen. This sometimes led to problems with the distribution being estimated from only one alignment score. We have therefore changed the script to use a fixed value of C = 0, which usually yields good results. The probability of getting a random alignment with a score larger than or equal to the score D is [37]:
The extreme value distribution parameters are estimated for each sequence pair using alignments of 20 shufflings of that pair. The dinucleotide distribution is conserved during the shuffling [39].
The localization performance is measured by counting the number of structure pairs found (Pt), structure pairs missed (Nf), and the number of false positive predictions (Pf) made by the method.
The annotated structure pairs overlapped by a prediction are counted as found (Pt) if at least half the nucleotides covered by the prediction in both separate sequences are annotated as RNA structures. If there is more than one Pt prediction which covers the same pair of structures, then the structure pair is only counted once. Predictions in which at least half of the nucleotides are not annotated as RNA in both sequences are counted as false positives (Pf). If a false positive prediction overlaps a structure from one RNA family and only one family in a sequence, then the prediction gets this family, otherwise it gets the “Unknown” family. If a false positive prediction gets an RNA family for one of the sequences and the “Unknown” family for the other, then the false positive is counted as belonging to the known RNA family. A missed structure pair (Nf) is a pair of annotated structures which is not overlapped by any significant positive predictions. Mixed RNA families, like a tRNA versus a 5S rRNA, are ignored. From the Pt, Pf, and Nf numbers, the positive predictive value PPV = Pt / (Pt + Pf) and the sensitivity Sens = Pt / (Pt + Nf) are calculated.
To compare the predicted structures and the annotated structures, the MCC [40] for structures are used:
Pt is the number of predicted base pairs which are also annotated. Pf is the number of predicted basepairs which are not annotated. Nf is the number of base pairs that are annotated but not predicted. Nt is the number of positions that are both predicted and annotated not to base pair. As Nf is always very large, the approximation described in [30] could have been used. No correction for sliding base pairs is used [35].
|
10.1371/journal.pcbi.1003049 | A Kinetic Platform to Determine the Fate of Nitric Oxide in Escherichia coli | Nitric oxide (NO•) is generated by the innate immune response to neutralize pathogens. NO• and its autoxidation products have an extensive biochemical reaction network that includes reactions with iron-sulfur clusters, DNA, and thiols. The fate of NO• inside a pathogen depends on a kinetic competition among its many targets, and is of critical importance to infection outcomes. Due to the complexity of the NO• biochemical network, where many intermediates are short-lived and at extremely low concentrations, several species can be measured, but stable products are non-unique, and damaged biomolecules are continually repaired or regenerated, kinetic models are required to understand and predict the outcome of NO• treatment. Here, we have constructed a comprehensive kinetic model that encompasses the broad reactivity of NO• in Escherichia coli. The incorporation of spontaneous and enzymatic reactions, as well as damage and repair of biomolecules, allowed for a detailed analysis of how NO• distributes in E. coli cultures. The model was informed with experimental measurements of NO• dynamics, and used to identify control parameters of the NO• distribution. Simulations predicted that NO• dioxygenase (Hmp) functions as a dominant NO• consumption pathway at O2 concentrations as low as 35 µM (microaerobic), and interestingly, loses utility as the NO• delivery rate increases. We confirmed these predictions experimentally by measuring NO• dynamics in wild-type and mutant cultures at different NO• delivery rates and O2 concentrations. These data suggest that the kinetics of NO• metabolism must be considered when assessing the importance of cellular components to NO• tolerance, and that models such as the one described here are necessary to rigorously investigate NO• stress in microbes. This model provides a platform to identify novel strategies to potentiate the effects of NO•, and will serve as a template from which analogous models can be generated for other organisms.
| Nitric oxide (NO•) is a highly reactive metabolite used by immune cells to combat pathogens. Since the biological effects of NO• are governed by its broad reactivity, it is desirable to determine how NO• distributes among its many targets inside a cell. A quantitative understanding of this distribution and how it is controlled will facilitate the development of novel NO•-potentiating therapeutics. Here, we have constructed and experimentally validated a comprehensive kinetic model of NO• biochemistry within Escherichia coli that includes NO• autoxidation, respiratory inhibition, enzymatic detoxification, and damage and repair of biomolecules. Using this model, we investigated the control of NO• dynamics in E. coli cultures, and found that the primary aerobic detoxification system, NO• dioxygenase (Hmp), functions as a dominant NO• consumption pathway under microaerobic conditions (35 µM O2), and loses utility as the NO• delivery rate increases. We confirmed these predictions experimentally, thereby demonstrating the predictive power of the model. This model will serve as a quantitative platform to study nitrosative stress, provide a template from which models for other organisms can be generated, and facilitate the development of antimicrobials that synergize with host-derived NO•.
| NO• is an uncharged, highly diffusible, membrane-permeable metabolite, generated by mammalian NO• synthases (NOS) for use in signaling and defense [1], [2]. The diversity of functions performed by NO•, from pathogen detoxification to vasodilation, reflect its broad reactivity. NO• directly reacts with iron-sulfur ([Fe-S]) clusters, superoxide (O2•−), and O2, whereas its oxidized forms (NO2•, N2O3, and N2O4) damage thiols, tyrosine residues, and DNA bases [2]–[5]. Such widespread activity has made the biological effects of NO• difficult to predict [2]. For instance, if 1,000 NO• molecules entered a cell, what would become of them? How many would disrupt an [Fe-S] cluster to form a protein-bound dinitrosyl-iron complex (DNIC)? How many would autoxidize to form nitrogen dioxide (NO2•) and then react with another NO• to form nitrous anhydride (N2O3)? How many N2O3 would deaminate DNA bases? These questions are representative of one unifying, fundamental question of NO• metabolism: how does NO• distribute within a cell? The answer to this question lies in understanding the kinetic competition of NO• with its many intracellular targets. However, the NO• biochemical network is complex (Figure 1), contains numerous short-lived intermediates at low concentrations [6], converges to only a few stable end-products [4], and involves various damaged biomolecules that are continually digested or repaired [7]. Such complexity has necessitated the use of computational models to both interpret and predict the outcome of NO• treatment [4].
A number of kinetic models have been developed to simulate NO• chemistry in biological contexts [3], [4], [6]–[17]. Many of these models have focused on mammalian systems due to the importance of NO• in human physiology. Nalwaya and Deen [9] calculated steady-state concentration profiles of NO•, CO2, O2•−, and peroxynitrite (ONOO−) in idealized mammalian cell cultures using a reaction–diffusion model, and explored the effect of varying the rates and locations (extracellular, mitochondrial, or cytosolic) of their generation. Their results suggested negligible spatial variation in species concentrations, and identified conditions under which the different cellular compartments serve as dominant sources or sinks. However, their model did not include the reactions of numerous intracellular metabolites that either directly react with NO•, or its autoxidation products (NO2•, N2O3, and N2O4). Lancaster [3] constructed a non-diffusive, but more extensive kinetic model to encompass the complex reaction network of NO• and its autoxidation products with glutathione (GSH) and tyrosine in mammalian systems. This model allowed for predictions regarding the relative importance of the various NO•-consuming pathways under inflammatory and non-inflammatory regimes, and highlighted the dominance of oxidative reactions. Lim et al. [4] built upon the work of Lancaster [3] by incorporating additional antioxidants, as well as a separate membrane compartment to account for partitioning of certain species in the lipid-phase. Their model was developed to be representative of inflamed tissue in vivo and used to estimate steady-state intracellular concentrations of different reactive nitrogen species (RNS), in addition to identifying their major sources and sinks in the cytosol and membrane compartments of mammalian cells. Interestingly, none of these models considered the interaction of NO• and its autoxidation products with [Fe-S] clusters, cytochromes, or DNA, and their treatment of the relevant enzymatic processes was limited to NO• dioxygenase and superoxide dismutase. Recently, Tórtora et al. [7] measured rates of ROS- and RNS-induced damage to the mitochondrial aconitase [4Fe-4S] cluster, and incorporated the reactions into a kinetic model of aconitase inactivation in the presence of O2•− and NO•. Since their focus was specifically on the inactivation of aconitase, they did not consider much of the extensive reaction network of NO•, O2•−, and their products. Bagci et al. [14] merged a mitochondrial apoptotic network [18] with a kinetic model of NO• chemistry [10] and extended treatment to include formation of N2O3, NO2• and ONOO−, as well as their interactions with GSH, non-heme iron, and mitochondrial cytochrome c. However, their attention was primarily on the dynamics of the apoptotic response, and many RNS-related reactions and biological species that were not directly involved in apoptosis were omitted. Though previous models provide a firm foundation for modeling NO• in biological systems, none are sufficiently comprehensive to quantify the distribution of NO• among its many intracellular consumption pathways.
Here, we describe the construction, experimental validation, and utility of a comprehensive model of NO• metabolism in Escherichia coli. This model includes NO• autoxidation, enzymatic detoxification, [Fe-S] damage, thiol and tyrosine nitrosation, DNA base deamination, tyrosine nitration, and the repair steps responsible for regeneration of RNS targets. A model of NO• stress with this level of detail has not been previously recognized for any organism. Using this model, we quantitatively explored the distribution of NO• consumption in E. coli, and predicted that the utility of the major aerobic NO• detoxification system (Hmp) depends on the NO• delivery rate and extends to environments with O2 concentrations as low as 35 µM (microaerobic). We went on to experimentally confirm these predictions, thereby demonstrating the utility of this model to the study of NO• metabolism. This computational model will serve as a platform to quantitatively interrogate the kinetic competition of NO• with its many targets in E. coli, and assess the influence of various parameters on its distribution.
Upon diffusing into E. coli, NO• may be consumed directly through enzymatic detoxification (Hmp, NorV, NrfA), or reactions with [Fe-S] clusters, O2•−, or O2 (Figure 1, Figures S1, S2). Several resulting nitrosative species, including NO2• and N2O3, can further react to deaminate DNA bases, nitrosate protein and low molecular weight thiols, and nitrate tyrosine residues. To quantify how NO• distributes within a cell, we have constructed a comprehensive kinetic model of the NO• biochemical reaction network in E. coli, where autoxidation, detoxification (Hmp, NorV, NrfA), [2Fe-2S] and [4Fe-4S] damage and repair, thiol nitrosation and denitrosation, DNA base deamination and repair, enzyme expression and degradation, tyrosine nitration, and reversible cytochrome inhibition are included (Figure 1). The model consists of 179 reactions, 132 chemical and biochemical species, and 163 kinetic parameters (Tables S1, S2, S3, Text S1). Of the kinetic parameters, 24 have values that are uncertain, either due to variability or unavailability in literature (Table S4). An overview describing the construction of the model is presented in the Materials and methods section, whereas a more detailed description has been presented in Text S1. Due to its scope and completeness, the model is suited to predict the distribution of NO• consumption among the available pathways in E. coli. For example, the fraction of NO• detoxified by Hmp, the amount of NO2•, N2O3, and ONOO− formed, the quantity of [Fe-S] clusters and DNA bases damaged and repaired, the extent and duration of cytochrome inhibition, and amount of thiols nitrosated can all be calculated from model simulations. Further, the model allows parameter variation (for example, enzyme mutation/deletion) and quantification of the impact these alterations have on NO• metabolism. To substantiate the utility of the model, we first validated that the model could reproduce experimentally-measured NO• dynamics and make accurate predictions of experimental outcomes.
We sought to validate that the model could capture NO• dynamics in E. coli cultures. Since extracellular NO• loss, including autoxidation and gas phase transport, was non-negligible, we bridged the intracellular model to the experimental system by adding an extracellular (growth media) compartment that accounted for autoxidation and gas-phase transport (Materials and methods). Kinetic parameters specific to the extracellular compartment (NO• delivery rate, NO• and O2 gas phase mass transfer coefficients, and NO• autoxidation rate) were determined from experimental NO• and O2 measurements in the absence of cells (Materials and methods, Figure S3, Text S1). In the experimental system, exponential-phase wild-type E. coli were treated with 0.5 mM dipropylenetriamine (DPTA) NONOate, and the concentration of NO• in the culture was monitored over time (Materials and methods). The NO• concentration peaked rapidly to 9.7 µM following delivery of DPTA, and decreased at a steady rate for ∼0.6 hours, after which the concentration dropped quickly to submicromolar levels (Figure 2A). Using a nonlinear least squares optimization algorithm, 39 uncertain parameters (24 kinetic constants and 15 species concentrations) from the model were optimized to capture the experimentally-measured NO• concentration profile (Materials and methods, Table S4). Uncertain parameters were defined as those that were absent from literature, or those whose literature values had a high degree of variability. All other parameters were either set to their literature values, or measured independently in our experimental apparatus (Tables S1, S2, S3, Text S1). Given that the optimization algorithm does not guarantee identification of the globally optimal solution, 100 independent sets of random initial parameter values were used (Materials and methods). The optimized parameter set yielding the lowest sum of squared residuals (SSR) between the simulated and experimental [NO•] curve is presented in Figure 2A, and demonstrates the model's ability to capture NO• dynamics in a wild-type E. coli culture. For comparison, we took the three most comparable NO• models [3], [4], [9], adapted them to our experimental conditions, and performed an analogous nonlinear least squares optimization in an attempt to capture the NO• dynamics of wild-type E. coli cultures (Materials and methods). As depicted in Figure S4, none of the alternative models could accurately simulate E. coli NO• dynamics. Quantitatively, the SSRs between the experimental data and the [NO•] curves predicted by the adapted, alternative models of Lim et al. [4], Lancaster [3], and Nalwaya and Deen [9] were, respectively, 200-, 200-, and 70-fold greater than that of the model presented here. These data convincingly demonstrate that the model presented here far exceeds current state-of-the-art kinetic models for simulation of microbial NO• metabolism.
With an ability to simulate NO• dynamics confirmed, we sought to identify which of the 39 parameters adjusted by the nonlinear optimization procedure were informed by the process, and which had a negligible influence under these conditions. We varied each parameter individually and calculated the corresponding increase in SSR, keeping all other parameters at their optimized values (Figure 2B). The analysis revealed that the Hmp NO• binding (kHmp,NO•-on), and Hmp expression (kHmp-exp,max and KHmp-exp,NO•) parameters were the most influential, whereas the oxidation of NorV (kNorV-O2) was of minor significance, but exhibited a greater effect than the remaining parameters, which were all negligible (less than a 5% increase in SSR) (Figure S5). This prompted us to identify the minimum biochemical reaction network necessary to simulate NO• dynamics in aerobic, wild-type E. coli cultures (Materials and methods). As depicted in Table S5, the model presented here can be simplified to include 17 reactions, 18 chemical and biochemical species, and 14 kinetic parameters without exceeding an overall 5% increase in SSR. While this simplified model can capture the NO• dynamics presented in Figure 2A, we note that it is not suitable for the calculation of additional NO• outcomes, such as the degree of [Fe-S] cluster damage or cytochrome inhibition, and it is not generally translatable to other experimental conditions, such as anaerobic environments. The comprehensive model, on the other hand, can perform such calculations and be applied under many more experimental conditions.
The importance of parameters governing Hmp detoxification activity suggested a dominant role for this enzyme in the consumption of NO• under aerobic conditions, a result that is consistent with previous studies of NO• sensitivity in E. coli [19]–[22]. To quantitatively investigate the contribution of Hmp to NO• consumption, we calculated the cumulative, time-dependent distribution (overall and intracellular) of NO• for wild-type E. coli treated with DPTA using the optimized parameter values (Figures 2C and 2D). The simulated distributions predicted that autoxidation of NO• in the media accounts for the majority of NO• removal shortly after DPTA addition, with loss to the gas phase comprising most of the remaining flux. By 45 min after delivery, the model predicted that cellular consumption of NO• had accumulated to match that of gaseous loss, and after 1 h became the primary sink. The predicted concentration of NO• dropped rapidly to submicromolar levels at 43 minutes post-dose, where it remained for the duration of the simulation, as Hmp continued to remove NO• as it was released by DPTA. The majority (78.1%) of the total NO• released by DPTA was predicted to be consumed by the cells, while autoxidation in the media and loss to the gas phase accounted for 13.6% and 8.3% of the total NO• consumption, respectively. Virtually all (99.85%) of the NO• consumed by the cells was predicted to be through Hmp detoxification, with most of the remaining 0.15% through oxidation by O2 and O2•−. Reduction by anaerobic detoxification enzymes (NorV and NrfA) and nitrosylation of [Fe-S] clusters was predicted to account for less than 0.03% and 0.04% of the cellular NO• consumption, respectively. To provide additional experimental evidence in support of these intracellular distributions, we experimentally validated that a mutant lacking the NorV enzyme (ΔnorV) consumed NO• at the same rate as wild-type under the experimental conditions tested (Figure S6).
Given the importance of Hmp to the removal of NO•, we assessed the predictive power of the model by determining whether it could accurately predict NO• dynamics in a Δhmp mutant culture. We simulated a Δhmp mutant by fixing the Hmp expression rate to zero. All other model parameter values were left unchanged. As expected, the removal of Hmp was predicted to have a considerable effect on the cells' ability to remove NO• from the environment (Figure 3A). Although the [NO•] curve simulated for the Δhmp culture closely matched that predicted for wild-type at early times (0 to ∼10 min) after DPTA delivery, it started to diverge rapidly as Hmp began to dominate the consumption of NO• in the wild-type culture (Figure 2C). The model predicted that the wild-type culture would quickly consume NO• to reach a submicromolar NO• concentration by 43 min, while the concentration of NO• in the Δhmp culture would gradually decline, requiring over 6.4 hours to achieve submicromolar levels. In contrast to wild-type cultures where it was predicted that most NO• would be converted to NO3− by Hmp, the model predicted that the majority of NO• in Δhmp cultures would be converted to NO2− through autoxidation (Figure 3C). To experimentally confirm the Δhmp model predictions, we measured the concentration of NO• in a Δhmp culture after treatment with 0.5 mM DPTA under identical conditions as wild-type (Figure 3B). In addition, we measured NO2− and NO3− in the Δhmp culture at 10 h post-dose, when it was predicted that over 99% of the donor had dissociated. The model-predicted NO• concentration curve and final NO2− and NO3− concentrations were in excellent agreement with the experimental data (without further optimization of any parameters) (Figure 3B), validating the ability of the model to make accurate predictions regarding major perturbations to the system.
To further investigate NO• clearance from the Δhmp culture, we simulated the corresponding intracellular distribution of NO• (Figure 3D). In the Δhmp culture, consumption of NO• by cells was predicted to account for less than 1% of the total NO• delivered, compared to the 78.1% for wild-type cells. Over 73% of the NO• that was consumed through intracellular pathways was predicted to be by reaction with O2 or O2•−, while anaerobic enzymatic reduction (NorV and NrfA) and [Fe-S] nitrosylation accounted for the remaining 14.1% and12.6%, respectively (Figure 3D).
After validating the model, we sought to identify parameters that control the NO• distribution in E. coli cultures. We focused on experimentally-accessible model parameters to enable experimental validation of predictions. To identify control parameters, we performed a parametric analysis (Materials and methods) to assess the effect of varying each parameter on the distribution of NO•. Varied parameters included enzyme concentrations or maximum expression rates, initial concentration and release rate of the NO• donor, O2 concentration in the environment, and intracellular concentrations of GSH, amino acids, and energy metabolites (Figure 4A, Table S6). In addition to Hmp expression, the parametric analysis revealed NO• donor concentration and release rate, as well as O2 concentration, as important parameters governing the distribution of NO• consumption. Anaerobic NO• detoxification enzymes became the dominant mode of NO• removal within E. coli at lower O2 concentrations due to the loss of Hmp NO• dioxygenase activity, a decrease in the O2-mediated deactivation of NorV, and reduced repression of NrfA expression. The lower O2 concentration also decreased the rate of NO• autoxidation in the media, leaving intracellular reactions and escape to the gas phase as the two primary modes of NO• removal. Although it had little impact on the total NO• distribution, removing superoxide dismutase activity resulted in a small, but noticeable increase in the fraction of intracellular NO• consumed through reaction with O2•− to form ONOO−. Interestingly, the model predicted that higher donor release rates decrease the utility of Hmp in detoxifying NO• (Figure 4B). This decrease can be attributed to the higher NO• concentrations achieved with faster release rates, which in turn enhance substrate inhibition due to the binding of NO• to the Hmp active site before O2 [23].
To further examine the effect of donor release rate on model dynamics, we simulated delivery of NO• to cultures at an increased rate, where Hmp contribution to NO• consumption was predicted to be largely reduced. The initial concentration of donor was maintained at 0.5 mM, but the release rate was increased from 1.34×10−4 s−1 (1.4 h half-life, DPTA) to 1.35×10−3 s−1 (8.6 min half-life), the measured rate for the NO• donor propylamine propylamine (PAPA) NONOate (Figure S7, Text S1). We performed simulations for wild-type and Δhmp cultures, and generated the corresponding NO• concentration profiles (Figure 4C). The strong influence of NO• delivery kinetics on model dynamics are readily apparent when comparing the NO• concentration profiles simulated for PAPA (Figure 4C) with those for DPTA (Figure 3A). The faster release rate of PAPA predicted a peak NO• concentration nearly four times that of DPTA (34 µM compared to 9 µM, respectively), and a large increase in similarity between the simulated wild-type and Δhmp [NO•] curves was observed. Although the predicted NO• concentration in the PAPA-treated wild-type culture dropped rapidly to submicromolar levels at a time similar to that of DPTA (37 min and 43 min, respectively), Δhmp entered this regime after 1.2 h when treated with PAPA, compared to the 6.4 h predicted for DPTA. We simulated the corresponding NO• distributions for PAPA-treated cultures to examine the participation of the different pathways in NO• removal. The elevated NO• concentrations simulated for the faster-releasing PAPA greatly increased flux through various consumption pathways, where over 99% of the total NO• consumption was predicted to occur within the first hour after dose for both wild-type and Δhmp (Figures 5A and 5C, respectively). The activity of Hmp, however, was attenuated by the higher NO• concentration due to substrate inhibition (see Text S1), reducing its ability to participate in detoxification. When Hmp activity was restored and became the most rapid NO• removal pathway after ∼30 minutes, simulation results showed that over 90% of the total NO• had already been consumed through autoxidation and gas transfer pathways (Figure 5A). As a result, the fraction of total NO• consumed by cellular pathways in the wild-type culture was predicted to decrease by nearly 10-fold (78.1% to 8.4%) due to the increased NO• delivery rate (compare Figures 2C and 5A). When treating with DPTA, the NO• concentration profile and distribution simulated for the Δhmp mutant (Figure 3B) were observed to differ greatly from those of wild-type (Figure 2C), but were significantly more similar to wild-type when using PAPA as the donor (Figures 5A and 5C) due to the large reduction in Hmp-mediated NO• consumption predicted for the wild-type culture. The intracellular distribution simulated for wild-type treated with PAPA (Figure 5B) was still dominated by Hmp, despite its large reduction in activity. However, the proportion of intracellular NO• consumed through pathways other than Hmp was predicted to increase by over 15-fold (0.15% to 2.6%) upon increasing the NO• delivery rate, suggesting that these other pathways maintain activity while Hmp is inhibited. Thus, the reduction in Hmp activity predicts a 15-fold increase in contribution by other intracellular pathways to the removal of NO• within the cell, including damage to biomolecules such as [Fe-S] clusters.
To experimentally validate the prediction that the utility of Hmp decreases as the delivery rate of NO• increases, we measured and compared the ability of wild-type and Δhmp to remove NO• from the culture when dosed with PAPA. We observed excellent agreement between model-predicted and experimentally-measured NO• concentration profiles for the addition of PAPA to wild-type and Δhmp cultures, with no further optimization of model parameters (Figure 6A). The peak concentration of NO• was underestimated by approximately 10%, which was also observed when measuring NO• release from PAPA in media without cells (Figure S7), suggesting that the disagreement was not associated with cellular parameters. In addition, the rate of NO• clearance by the wild-type cells was slightly overestimated. This could originate from the treatment of Hmp expression in the model, where a more extensive implementation of its governing regulatory network may improve the accuracy of the simulated transcriptional response of hmp expression to elevated levels of NO•. As predicted, the measured difference in time required to remove NO• from the culture between wild-type and Δhmp was small for PAPA (0.6 h difference in time to reach submicromolar levels), highlighting the decreased utility of Hmp under conditions of more rapid NO• release. These results demonstrate that the model can accurately identify parameters that control the distribution of NO• in bacterial cultures, and quantify the impact of their manipulation.
In addition to NO• removal from the cell interior and surrounding environment, the model can be used to calculate the extent to which NO• affects various cellular targets, including [Fe-S] nitrosylation [24]–[26] and cytochrome inhibition [27], [28]. Therefore, we utilized the model to evaluate the protective effect of Hmp with respect to [Fe-S] cluster damage (Figure 6B) and cytochrome bd inhibition (Figure 6C). Due to the wide range of reaction rates reported for the nitrosylation of [Fe-S] clusters by NO• (kNO•-[Fe-S]), the parameter value was varied across this range when predicting the extent of [Fe-S] damage. Simulated exposure of wild-type and Δhmp E. coli to 0.5 mM DPTA predicted a 2- to 4-fold reduction in the total concentration of [Fe-S] clusters damaged as a result of Hmp activity. When simulations were repeated for PAPA, however, the total [Fe-S] damage predicted for wild-type and Δhmp cultures differed by a maximum of 5%, in agreement with the predicted dependence of Hmp utility on NO• release rate. Furthermore, the duration of NO•-mediated cytochrome bd inhibition following DPTA treatment was predicted to greatly increase for the Δhmp culture relative to wild-type, requiring over 9 h (compared to 0.7 h for wild-type) for the concentration of NO•-bound cytochromes to drop below 50% of the total. Treatment with PAPA resulted in more similar cytochrome inhibition between the strains, with durations of 0.6 h and 1.5 h predicted for wild-type and Δhmp, respectively. Collectively, the results from these damage descriptors, in addition to the rate and distribution of NO• consumption, predicted a greater similarity in recovery from bacteriostasis between wild-type and Δhmp when treated with PAPA than with DPTA. To test the prediction, we monitored the optical density (OD600) of each strain following treatment with 0.5 mM DPTA or PAPA (Figure 6D). In agreement with the prediction, the duration of NO•-induced stasis was more similar between wild-type and Δhmp strains when using a faster NO• donor. Growth inhibition of Δhmp following PAPA treatment was less severe than that observed for DPTA, where cells exited stasis less than 2 h after wild-type, compared to over 10 h for DPTA.
Hmp is considered the major aerobic enzyme responsible for NO• detoxification [19], [21], [29], whereas NorV is considered the major anaerobic detoxification system [22], [30], [31]. Surprisingly, the parametric analysis suggested that Hmp remains dominant at O2 concentrations as low as 25 µM (∼14% air saturation [32]) (Figure 4A). To experimentally confirm that this was the case, we adjusted the experimental setup by adding N2-bubbling at a rate of 1 ml/s. In the presence of wild-type E. coli at an OD600 of 0.05, an O2 concentration of 35 µM was achieved and maintained constant throughout the time course of a DPTA experiment (Figure S8). This concentration was over 5-fold less than air-saturated media (185 µM), but also above the 25 µM used in the parametric analysis. Due to the adjustment in experimental conditions, the model was similarly optimized for NO• dynamics from microaerobic wild-type E. coli cultures (Material and methods), and found to capture the data very well (Figure 7A). We note that N2-bubbling increased fluctuations in the NO• measurements, but the increased error was minor compared to the range of NO• concentrations investigated. Using the optimized model, we predicted the effect of genetic deletions of norV and hmp on the NO• dynamics. Consistent with the previous parametric analysis, NorV was identified as a negligible consumption pathway under microaerobic conditions (35 µM O2), whereas Hmp was identified as the major NO• sink. These predictions were experimentally validated, and the results are presented in Figures 7B and 7C. These data demonstrate that the model is useful for studying sub-aerobic environmental conditions, and that the switch between Hmp-dominated and NorV-dominated NO• consumption regimes occurs at very low O2 concentrations.
The corresponding extracellular and intracellular NO• distributions for wild-type, Δhmp, and ΔnorV under microaerobic conditions were simulated, and are presented in Figure 7. Loss of NO• to the gas phase was predicted to largely increase for all strains (26% and 27% of the total NO• consumption for wild-type and ΔnorV, respectively, and 95% for Δhmp), due to the increased air-liquid surface area caused by the bubbling of N2 through the culture, as well as the reduced rate of autoxidation. Autoxidation was predicted to have negligible NO• consumption activity compared to the cellular and gas transport pathways (0.5% of the total for wild-type and ΔnorV, and 1.2% for Δhmp), owing to the reduced O2 concentration, as well as the lower peak NO• concentration (∼4.5 µM for all strains) than was achieved under aerobic, non-bubbling conditions using DPTA (∼8–10 µM). Cellular consumption of NO• was still predicted to be the greatest sink of NO• for the wild-type and ΔnorV strains (accounting for 74% and 73% of the total consumption, respectively), but only a minor pathway in the Δhmp culture (3.8%). The intracellular distributions (Figure 7) for wild-type and ΔnorV cultures were still predicted to be dominated by Hmp detoxification (both exceeding 98% of intracellular NO• consumed by Hmp), as was seen under aerobic conditions. The NO• consumed by Δhmp cells, however, was now predicted to occur primarily through NorV reduction (93% of the intracellular NO•), compared to the 14% contribution predicted for Δhmp in aerobic conditions. Overall, the simulation results predicted Hmp to be the primary mode of NO• consumption under O2 concentrations as low as 35 µM, but suggested an increased role of NorV reduction in the event that Hmp detoxification becomes unavailable.
NO• is a critical antimicrobial of the innate immune response whose utility originates from its ability to diffuse through cellular membranes [33], deactivate bacterial enzymes [26], inhibit respiration [28], and react with O2 and O2•− to yield the reactive nitrogen species, NO2•, N2O3, N2O4, and ONOO− [24]. The biochemical reaction network of NO• includes both spontaneous and enzymatic reactions involving many short-lived species that decompose to several common end-products [4]. Increasing the complexity of this system is the continuous degradation and repair of damaged biomolecules, which regenerates targets for NO• and its reactive intermediates [34]. A quantitative description of how NO• distributes among these many pathways is critical to understanding immune function and pathogenesis, as well as to designing NO•-based and NO•-synergizing therapeutics [35]–[37]. However, the complexity of the NO• reaction network renders exhaustive experimental monitoring infeasible, and interpretation of measurements difficult [4], [6]. To address these challenges, experimentally-informed computational models are required to explore the NO• reaction network.
Though several kinetic models have been developed to study the chemistry of NO• in biological systems, of which the majority are mammalian, none have had sufficient breadth and depth to address the full range of effects of NO• exposure [1], [38]. The model presented here is far more comprehensive than those constructed previously, incorporating the damage, modification, and repair of biomolecules, as well as enzymatic detoxification and transcriptional control. These functionalities allow focused investigation of intracellular components of the NO• network, such as [Fe-S] cluster and DNA damage, but also culture-wide prediction of the NO• distribution. We validated the utility of the model by demonstrating that it can reproduce NO• dynamics in a bacterial culture, make accurate predictions regarding large perturbations to the system, and identify parameters that control the distribution of NO• in bacterial cultures. Specifically, model simulations predicted that NO• autoxidation and Hmp-catalyzed detoxification were the primary sinks for NO• consumption in aerobic wild-type E. coli cultures. Oxidation of NO• has been shown in the past to be a major contributor to the consumption of NO• under certain conditions [3], and the dominant role of Hmp in aerobic detoxification is in agreement with previous studies that have demonstrated its importance in tolerating NO• stress [21]–[23]. In addition, we used the model to (1) uncover a novel dependency of Hmp utility on the NO• delivery rate, and (2) discover that Hmp is the dominant cellular NO• detoxification system at dissolved O2 concentrations as low as 35 µM (microaerobic). Both of these predictions were validated experimentally, thereby demonstrating the utility of the model for the study of NO• metabolism. Specifically, when treated with a fast-releasing NO• donor (PAPA), the consumption of NO• and recovery from bacteriostasis was far more similar between wild-type and Δhmp E. coli than with a slower NO• donor (DPTA). This effect arises from substrate inhibition of the Hmp active site caused by high NO•/O2 concentration ratios and the time required to synthesize Hmp [23]. An effect of NO• delivery on its toxicity has been observed previously in a mammalian system [39], [40], where it was shown that killing of human lymphoblastoid cells (TK6 and NH32) was a function of both NO• concentration and cumulative dose. Here, we have demonstrated an influence of NO• delivery rate on the dynamics of NO• consumption and recovery in bacterial cultures, and also offered a detailed, mechanistic description of the observed dependence. In addition, we discovered that Hmp remains the major cellular detoxification system at dissolved O2 concentrations as low as 35 µM. This effect originates from the strong induction of Hmp expression upon NO• exposure even under anaerobic conditions [41], [42], and the rapid O2-mediated deactivation of NorV, the alternative NO• detoxification system that has been previously identified as critical for resisting NO• stress under anaerobic conditions [22], [31]. These data demonstrate the flexibility of this method to different environmental conditions (microaerobic), and provide support for the role of Hmp as a virulence factor [43], [44], since O2 concentrations at infections sites/in macrophages and neutrophils are typically hypoxic (less than 50 µM O2 [4], [45], [46]). Interestingly, both NorV- and Hmp-type enzymes have been found to be virulence factors for numerous organisms [36], [47]–[50], and thus a quantitative understanding of the conditions under which each contributes to NO• clearance would be valuable for the study of their importance to virulence.
The work presented here demonstrates the predictive accuracy and utility of a comprehensive model of NO• metabolism in E. coli. The scope of this model allows for detailed, quantitative exploration of numerous NO• network features and environmental conditions, including future investigations of the roles of O2 concentration and indirect NO• delivery, such as that observed for S-nitrosothiols [41]. Further, this model will prove useful for the optimization of NO•-synergizing and NO•-based therapeutics, which are being investigated as antibiotic alternatives for the treatment of both gram-positive and gram-negative infections, including those caused by Mycobacterium tuberculosis, Staphylococcus aureus, Pseudomonas aeruginosa, E. coli, and Acinetobacter baumannii [35]–[37]. Such therapies include NO•-releasing nanoparticles [36], NO•-releasing dressings [37], and rhodanines, which kill non-replicating mycobacteria through the potentiation of host-derived NO• [35]. Interestingly, the study by Sulemankhil and colleagues identified NO• release rate and dosage as important parameters governing the effectiveness of the examined dressings. The modeling approach presented here could provide a more quantitative understanding of how these potential therapeutics neutralize pathogens, and would prove useful for identifying methods to increase their potency through the quantitative identification of the NO• distribution pathways used by specific organisms. To achieve this potential utility, the modeling method described here must be adapted for use in organisms other than E. coli. To do this, the enzymatic reactions within the model would need to be removed, replaced, or augmented based on the systems harbored by the pathogen of interest, and uncertain parameters would need to be identified by training the model on experimental data, as performed here. In the event that an important reaction is missing from a model, stable NO• end products (such as NO2− and NO3−) would be measured and both metabolic databases and the organism's genome would be mined for model additions capable of capturing the experimental data. Potential additions would then be experimentally validated by measuring in vitro kinetics of samples purified from cultures of interest. Execution of these steps will produce models of NO• metabolism in pathogens, that will mirror utility and capabilities achieved by the kinetic platform described here.
All strains used in this study were E. coli K-12 MG1655. The Δhmp and ΔnorV mutants were obtained from the Keio collection [51], and transferred into the MG1655 background using the P1 phage method. Proper chromosomal integration and absence of gene duplication were checked by PCR. The hmp primers used were 5′-CCGAATCATTGTGCGATAACA-3′ (forward) and 5′-ATGATGGATACTTTCTCGGCAGGAG-3′ (reverse) for accurate integration, and 5′- TCCCTTTACTGGTGGAAACG-3′ (forward) and 5′-CACGCCCAGATCCACTAACT-3′ (reverse) for gene duplication. The norV primers used were 5′-CCAGCACATCAACGGAAAAA-3′ (forward) and 5′-ATGATGGATACTTTCTCGGCAGGAG-3′ (reverse) for accurate integration, and 5′-GACTGGGAAGTGCGTGATTT-3′ (forward) and 5′-CGGAAGCGTAAACCAGTCAT-3′ (reverse) for gene duplication.
NO• donors (Z)-1-[N-(3-aminopropyl)-N-(3-ammoniopropyl)amino]diazen-1-ium-1,2-diolate (DPTA NONOate) and (Z)-1-[N-(3-aminopropyl)-N-(n-propyl)amino]diazen-1-ium-1,2-diolate (PAPA NONOate) were purchased from Cayman Chemical Company. All other chemicals and reagents were purchased from Sigma Aldrich or Fisher Scientific, unless otherwise noted.
E. coli from a frozen −80°C stock were inoculated into 1 ml of fresh LB broth and grown for 4 hours at 37°C and 250 r.p.m. 10 µl of the LB culture were used to inoculate 1 ml of MOPS minimal media (Teknova) containing 10 mM glucose. The minimal glucose culture was grown at 37°C and 250 r.p.m. overnight (16 h) and used to inoculate 20 ml fresh MOPS glucose (10 mM) in a 250 ml baffled shake flask to a final OD600 of 0.01. The flask culture was grown at 37°C and 250 r.p.m. to exponential phase (OD600 = 0.2), at which point 4 ml was transferred to separate microcentrifuge tubes in 1 ml aliquots and centrifuged at 15,000 r.p.m. for 3 min at 37°C. To remove the culture media, 980 µl of the supernatant was removed and cell pellets were resuspended in 1 ml of pre-warmed (37°C) 10 mM MOPS glucose media. Samples were combined in a 15 ml Falcon tube, and returned to the shaker (37°C, 250 r.p.m.). After 5 minutes, the resuspended culture was diluted to an OD600 of 0.03 in fresh, pre-warmed (37°C) MOPS glucose (10 mM) in a 50 ml Falcon tube with a final culture volume of 10 ml. The culture was stirred with a sterilized magnetic stirring bar, and immersed in a stirred water bath to maintain the temperature at 37°C. Growth was monitored until the OD600 reached a value of 0.05 (approximately 45 minutes after diluting to OD600 of 0.03), at which time the NO• donor solution (DPTA or PAPA) was added. On the day of use, the NONOate powder was dissolved in a chilled (4°C), sterile solution of 10 mM NaOH in deionized H2O, and stored on ice prior to delivery. After NONOate delivery, every half-hour (DPTA) or twenty minutes (PAPA), 75 µl aliquots were removed to measure the OD600 (Synergy H1 Microplate Reader, BioTek Instruments, Inc.). The concentration of NO• in the culture was monitored continuously over the course of the experiment using an ISO-NOP NO• sensor (World Precision Instruments, Inc.). The electrode was calibrated daily, prior to use, according to the manufacturer's specifications.
The microaerobic NO• consumption assay was performed using the same procedure, except N2 bubbling was included to reduce the dissolved O2 concentration. Immediately following the dilution of cells to an OD600 of 0.03 in the 50 ml Falcon tube, N2 gas (99.998% pure) was bubbled into the culture through a sterile pipet tip at a constant flow rate of 1 ml/s. The O2 concentration was observed to drop quickly and stabilize at approximately 19% air saturation (35 µM) within 15 minutes of initiating the N2 bubbling, where it remained for the duration of the experiment. The concentration of O2 was monitored continuously to ensure stable conditions throughout the assay (Figure S8).
The concentration of dissolved O2 was measured using the FireStingO2 fiber-optic O2 meter with the OXF1100 fixed needle-type minisensor (PyroScience GmbH). The sensor was calibrated according to the manufacturer's specifications, and the signal was automatically compensated for temperature fluctuations using the TDIP15 temperature sensor (PyroScience GmbH) during all O2 measurements.
The concentration of NO2− and NO3− were measured using the Nitrate/Nitrite Colorimetric Assay Kit from Cayman Chemical Company, following the manufacturer's instructions. Briefly, Griess reagents were added to diluted samples to convert the NO2− to a purple azo compound, and quantified by measuring the absorbance at 540 nm using a microplate reader [52]. A calibration curve was generated using varying dilutions of a standard NO2− solution. The total NO2−+NO3− concentration of the samples were obtained by first converting the NO3− to NO2− using nitrate reductase, and then treating with Griess reagents. The NO3− concentration in the samples was calculated as the difference between the total NO2−+NO3− concentration and the NO2− concentration. All samples were measured in triplicate.
All simulation calculations were performed using Matlab (R2012a). The governing set of differential mass balances was integrated using the stiff numerical ODE integrator (ode15s function).
Optimization of model parameters was performed in Matlab using the lsqcurvefit function, which solves nonlinear least-squares minimization problems. Through an iterative process, the function identified parameter values yielding the lowest sum of squared residuals (SSR) between the experimentally-measured and model-simulated NO• concentration profiles. Since the nonlinearity of the minimization problem gives rise to local minima, we performed 100 independent optimizations, each initialized with a random set of parameter values (within their allowed range).
The parameter optimization procedure was used to determine the values of extracellular parameters specific to our experimental system: NO• donor dissociation (kNONOate), transfer of NO• to the gas phase (kLaNO•), and the rate of NO• autoxidation (kNO•-O2). Cell-free growth media was treated with 0.5 mM DPTA under conditions identical to the aerobic NO• consumption assay, and the resulting NO• concentration profile and final (10 h) NO2− and NO3− concentrations were measured. The optimization yielded values of 1.34×10−4 s−1 (1.4 h half-life), 4.74×10−3 s−1, and 1.80×106 M−2s−1 for kNONOate, kLaNO•, and kNO•-O2, respectively (see Text S1 for further detail). Figure S3 demonstrates excellent agreement between the predicted and measured [NO•] curve and final NO2− and NO3− concentrations when using the optimized parameter values.
Of the cellular-related model parameters, 39 were classified as uncertain due to variability or unavailability in literature (Table S4). A parameter optimization was conducted to identify the set of parameter values yielding the lowest SSR between the simulated and experimentally-measured NO• concentration profile resulting from the addition of 0.5 mM DPTA to an aerobic, exponential-phase culture of wild-type E. coli. The predicted [NO•] curve using the optimal parameter set was in excellent agreement with the experimental data (Figure 2A).
For the microaerobic (35 µM O2) NO• consumption assay, uncertain parameters were re-optimized for the low-O2 environment due to expected changes in cellular properties and the effect of N2 bubbling on gas transfer rates. We note that differences in N2 bubble properties (such as bubble size and lifetime) caused by the presence of cells prevented the use of cell-free NO• measurements in determining extracellular parameters for this experimental setup. Instead, the simulated O2 concentration was fixed to 35 µM based on experimental observations (Figure S8), and the remaining extracellular and uncertain parameters (total of 42 parameters) were simultaneously optimized to best capture the NO• concentration curve measured for wild-type cells treated with DPTA under microaerobic conditions (Table S7). The optimal set of parameter values was able to accurately capture the experimentally-measured NO• dynamics in the microaerobic environment (Figure 7). An individual parametric analysis of the 42 optimized parameters was performed to determine those that had a significant impact (greater than 5% increase in SSR) on the predicted [NO•] curve in the microaerobic environment (Figure S9). Similar to the aerobic parametric analysis, Hmp-associated parameters (kHmp,NO•-on, kHmp-exp,max, and KHmp-exp,NO•) were found to strongly influence the predicted NO• dynamics. Parameters governing the rate of NONOate dissociation (kNONOate) and NO• transfer to the gas phase (kLaNO•) also demonstrated substantial control of the [NO•] curve upon variation. Finally, NorV expression (kNorV-exp,max and KNorV-exp,NO•) and inactivation (kNorV-O2) parameters were found to have a significant impact on the SSR.
Parametric analyses were used to evaluate the influence of model parameters on either the simulated [NO•] curve or the predicted distribution of NO• consumption in the culture. The effect of parameter variation on the [NO•] curve was quantified by the resulting change in SSR between the model-simulated and experimentally-measured NO• concentration profiles. Specifically, parameters were individually varied among 100 evenly-spaced points spanning their allowed range, and the resulting SSR at each parameter value was calculated. The effect of parameter variation on the SSR for aerobic (Figure 2B, Table S4) and microaerobic (Figure S9, Table S7) wild-type E. coli cultures was evaluated.
To quantify the effect of varying experimentally-accessible parameters on the predicted distribution of NO• consumption, parameters were individually varied among five logarithmically-spaced values spanning their permitted range (Table S6). Simulations were run for each different parameter set, and the final distribution of NO• consumption among the available pathways (such as autoxidation, transport to the gas phase, Hmp-mediated detoxification, and [Fe-S] damage) was calculated (Figure 4A).
Three existing models of NO• chemistry, developed by Lim et al. [4], Lancaster [3], and Nalwaya and Deen [9], were individually assessed for their ability to simulate NO• dynamics in a culture of wild-type E. coli. The alternative models were constructed and adapted to our experimental system using the following procedure. Starting with the model presented in this study, all reactions absent in the alternative model were eliminated, except the release of NO• from a NONOate, and the NO• and O2 liquid-gas transport reactions. Reactions present in the alternative model that were not included in the present model (due to the consumption or production of an unknown or nonspecific species, or the simplification of a more complex process in the present model) were added to the adapted model. For Lancaster's model, the NO• formation and disappearance reactions, as well as the disappearance of NO2• and •OH, were not included because the rates of the disappearance reactions are user-defined, and the formation of NO• is accounted for by the NONOate dissociation reaction. The model described by Nalwaya and Deen contains a simplified reaction representing the consumption of NO• by a heme- and flavin-dependent dioxygenase, analogous to Hmp detoxification in E. coli. The reaction was included in the adapted model, and the associated bimolecular rate constant was allowed to vary during parameter optimization. Additionally, the rate parameters governing ONOO− and ONOOH reactions used by Nalwaya and Deen were adjusted for a pH of 7.6, where the fraction of ONOO− in protonated form was calculated to be 12% [9]. Although Nalwaya and Deen do not include NO• autoxidation in their model, it was incorporated into the adapted version, as autoxidation is an important effect under the aerobic experimental conditions used in this study.
Species concentrations in the alternative models were set to the same values or ranges used in the present model, except for a few minor differences. The concentrations of proteins and transition metal centers (Mn+) in the model of Lim et al. were allowed a range of 5–8 mM and 1–500 µM, respectively. The protein concentration range was selected based on typical protein content reported for E. coli [119], while Mn+ was allowed the same concentration range as [Fe-S] clusters in the present model, which assumes ∼5% of proteins contain [Fe-S] clusters [120].
The three adapted models were subjected to a parameter optimization procedure analogous to that used for the model presented here (see “Parameter optimization” section above), where parameters classified as uncertain were varied to minimize the SSR between the predicted and experimental [NO•] curves. Ultimately, none of the three adapted models were able to capture the dynamics of NO• measured in wild-type E. coli cultures, yielding [NO•] curves with SSR values that were 200-fold (Lim et al. and Lancaster) and 70-fold (Nalwaya and Deen) greater than the SSR achieved by the present model (Figure S4).
In order to identify the core set of reactions required to accurately simulate NO• dynamics in aerobic wild-type E. coli cultures (Figure 2A), a systematic reduction of the model reaction network was performed using a two-tier process. In the first tier, reactions were sequentially deleted from the original network in a random order. After each reaction deletion, the SSR between the simulated and experimentally-measured [NO•] curve for DPTA-treated wild-type E. coli was calculated. If the SSR exceeded a 5% increase over the original SSR, the reaction deletion was undone. This process was repeated until no remaining reactions could be removed without exceeding the 5% increase in SSR. The entire model reduction process was repeated for a total of 100 iterations, each following a random sequence of reaction deletions. The reduced reaction network was selected as the set containing the least number of reactions. In the event of two or more minimum sets, the network yielding the lowest SSR was chosen.
In the second tier, the minimal reaction network was further reduced through a similar reaction deletion process, except with the inclusion of a parameter optimization step. After deleting a reaction, any remaining parameters in the reduced model classified as uncertain (Table S4) were re-optimized, following the nonlinear least-squares optimization procedure described above. If the optimization succeeded in decreasing the SSR to within 5% of the original SSR value, the reaction was removed from the final network. The final, minimum biochemical reaction network determined through this process is presented in Table S5.
|
10.1371/journal.pgen.1006437 | Phosphorylation of CBP20 Links MicroRNA to Root Growth in the Ethylene Response | Ethylene is one of the most important hormones for plant developmental processes and stress responses. However, the phosphorylation regulation in the ethylene signaling pathway is largely unknown. Here we report the phosphorylation of cap binding protein 20 (CBP20) at Ser245 is regulated by ethylene, and the phosphorylation is involved in root growth. The constitutive phosphorylation mimic form of CBP20 (CBP20S245E or CBP20S245D), while not the constitutive de-phosphorylation form of CBP20 (CBP20S245A) is able to rescue the root ethylene responsive phenotype of cbp20. By genome wide study with ethylene regulated gene expression and microRNA (miRNA) expression in the roots and shoots of both Col-0 and cbp20, we found miR319b is up regulated in roots while not in shoots, and its target MYB33 is specifically down regulated in roots with ethylene treatment. We described both the phenotypic and molecular consequences of transgenic over-expression of miR319b. Increased levels of miR319b (miR319bOE) leads to enhanced ethylene responsive root phenotype and reduction of MYB33 transcription level in roots; over expression of MYB33, which carrying mutated miR319b target site (mMYB33) in miR319bOE is able to recover both the root phenotype and the expression level of MYB33. Taken together, we proposed that ethylene regulated phosphorylation of CBP20 is involved in the root growth and one pathway is through the regulation of miR319b and its target MYB33 in roots.
| Ethylene is one of the most essential hormones for plant developmental processes and stress responses. However, the phosphorylation regulation in the ethylene signaling pathway is largely unknown. Here we found that ethylene induces the phosphorylation of CBP20 at S245, and the phosphorylation is involved in root growth. Genome wide study on ethylene regulated gene expression and microRNA expression together with genetic validation suggest that ethylene- induced phosphorylation of CBP20 is involved in root growth and one pathway is through the regulation of miR319b and its target gene MYB33. This study provides evidence showing a new link of cap binding protein phosphorylation associated microRNA to root growth in the ethylene response.
| The plant hormone ethylene (C2H4) is essential for a myriad of physiological and developmental processes [1–3]. A linear ethylene signaling pathway has been established [4] that plants perceive ethylene by ER-located receptors, which are similar to the bacterial two component histidine kinases [5, 6]. In the absence of ethylene, the receptors activate a Raf-like protein kinase CONSTITUTIVE TRIPLE RESPONSE 1 (CTR1) [7]. Activated CTR1 inhibit an ER-tethered protein ETHYLENE INSENSITIVE 2 (EIN2) through phosphorylation [8, 9]. EIN2 is degraded and the degradation is mediated by two F-box proteins: ETP1 and ETP2 [10]. In the presence of ethylene, the EIN2 C-terminal (EIN2-C) is dephosphorylated, cleaved and translocated into both the nucleus and P-body [9, 11, 12]. In the nucleus, the EIN2 CEND transduces signals to the transcription factors ETHYLENE INSENSTIVE3 (EIN3) and ETHYLENE INSENSITIVE3-LIKE1 (EIL1), which are sufficient and necessary for activation of all ethylene-response genes [13, 14]. In P-body, EIN2C mediates translation repression of EBF1 and EBF2 [15, 16], which are the two F-box proteins, which target EIN3 for degradation [17, 18]. Recently new study discovered that noncanonical histone acetylation H3K23Ac is involved in ethylene regulated gene activation in an EIN2 and partial EIN3 dependent manner [19].
Protein phosphorylation plays critical roles in ethylene response. Such as ethylene receptors are similar in sequence and structure to bacterial two-component histidine kinases, and ethylene controls autophosphorylation of the histidine kinase domain in ethylene receptor ETR1 [20], the histidine kinase activity of ETR1 is not required for but plays a modulating role in the regulation of ethylene responses [21]. Furthermore, biochemical and functional analysis of CTR1, a protein similar to the Raf family protein kinases that negatively regulates ethylene signaling in Arabidopsis [22]. Recently study has demonstrated that in the absence of ethylene, CTR1 targets to EIN2 C-terminal end for phosphorylation with its kinase domain [9]. However, the phosphorylation regulation in ethylene signaling is still under developed.
In this study, we found that ethylene induces phosphorylation of CBP20 at Ser245. Constitutive phosphorylation mimic form of CBP20S245D or CBP20S245E rescues the root less sensitive phenotype of cbp20 mutant in ethylene, but the constitutive dephosphorylation mimic form of CBP20S245A is unable to rescue cbp20 mutant phenotype. Through small RNA sequencing and mRNA sequencing, we found a set of miRNAs and their targets are specifically regulated in roots by ethylene in a CBP20 dependent manner. Among them, the expressions of miR319b and its potential target MYB33 display anti-correlation pattern in a CBP20 dependent manner in Col-0 roots. Small RNA northern blot in roots shows that miR319 is specific up regulated by ethylene treatment, which requires CBP20 phosphorylation. Genetic study shows that over expression of miR319b leads to the down regulation of MYB33, resulting in enhanced ethylene sensitive phenotype in roots, which is similar to its target myb33 mutant phenotype. The phenotype of miR319bOE is rescued by adding mutated MYB33 (mMYB33), which containing mutation at miR319 targeting site. Furthermore, we provided evidence showing that miR319b, while not miR159 influences the expression of MYB33 in the presence of ethylene in roots. Overall, our results demonstrate that ethylene regulated phosphorylation of CBP20 is involved in the root growth. One model is through the regulation of miR319b and its target MYB33 in roots in response to ethylene, providing a new link of cap binding protein phosphorylation associated microRNA to root growth in the ethylene response.
Previous studies have shown that phosphorylation plays critical roles in ethylene signaling and many ethylene regulated phosphorylation proteins have been identified [11, 23]. By searching the phosphorylation MS/LS data under ethylene treatment, we found that the CBP20, a component of cap-binding complex, is highly phosphorylated at Ser245 site with ethylene treatment (Fig 1A and 1B and S1A Fig). Protein alignment with CBP20s from different species showed that CBP20 is highly conserved, and the Ser245s are all identical through different species examined (S1B Fig), suggesting the function of CBP20 is conserved and the phosphorylation at S245 of CBP20 is potentially important for CBP20.
To study the role of phosphorylation of CBP20 in the regulation of ethylene response, the ethylene response phenotype of cbp20 mutant was examined. We found cbp20 mutant displayed partial reduced ethylene responsive phenotype in the roots, but not in the hypocotyls and apical hooks (Fig 1C–1E). To examine the connection between phosphorylation state of CBP20 and the ethylene responsive phenotype of cbp20, 35S promoter driven phosphorylation mimic form of CBP20 (35S:CBP20S245D and 35S:CBP20S245E) and dephosphorylation mimic form of CBP20 (35S:CBP20S245A) (S2A Fig), were generated and introduced into cbp20 mutant to obtain CBP20S245A/cbp20, CBP20S245D/cbp20 and CBP20S245E/cbp20. Additionally, 35S promoter driven wild type CBP20 (35S:CBP20) was introduced to cbp20 as control (S2A Fig). The full length of CBP20, CBP20S245E and CBP20S245D were able to rescue the cbp20 mutant phenotype in the presence of ethylene. However, CBP20S245A was unable to rescue the phenotype (Fig 1C–1E), which suggests that the phosphorylation of CBP20 is involved in the root growth in the presence of ethylene.
To explore how the phosphorylation of CBP20 is involved in the regulation in ethylene response, we first examined the root phenotype of cbp80 mutant in response to ethylene. We found that cbp80 displays similar phenotype as cbp20 in the presence of ethylene (S2B Fig). We next tested the interaction between wild type CBP80 with CBP20, CBP20S245A, CBP20S245D or CBP20S245E by yeast two-hybrid. In consistent with previous study [24], we were able to detect the interaction between CBP20 and CBP80, however, the interaction was not influenced by the phosphorylation states of CBP20 (S2C Fig). Generally, CBP80 interacts with CBP20 and in assisting CBP20 transfer into nucleus [25], we then examined cellular localization of CBP20S245A-YFP and CBP20S245D-YFP or CBP20S245E-YFP into cbp20 with or without the presence of ethylene. Both wild type CBP20 and mutated CBP20 were mainly localized in the nucleus and their localizations were not altered by ethylene treatment (S2D Fig).
CBC complex has a key role in several gene expression mechanisms [26–28] and CBP20 is essential for miRNA biogenesis [29–31]. We speculated that CBP20 is required for the biogenesis of miRNAs in the root growth in ethylene response. To address this question, we conducted small RNA sequencing using the roots and shoots isolated from 3-day old etiolated seedlings of Col-0 or cbp20 mutant treated with air or ethylene (S3A and S3B Fig). In consistent with previous study [30], most of species of miRNA detected were down regulated in cbp20 (S1–S4 Tables). By comparing the miRNA expressions in the roots and shoots of Col-0 and cbp20 treated with air or 4 hours ethylene gas. We found that ethylene altered miRNA expressions in a tissue specific manner (Fig 2A). As shown in Fig 2A and 2B, almost no shared ethylene induced differential expressed miRNAs were detected in the shoots and roots in Col-0 or cbp20 mutant. Given ethylene regulated miRNA expression is tissue specific, and cbp20 root specific phenotype with ethylene treatment, we speculated that the ethylene responsive root phenotype of cbp20 potentially due to the alteration of miRNAs in roots. Through further analysis, 13 ethylene regulated miRNAs (P<0.05) were identified specifically in Col-0 roots (Fig 2A). Among them, 7 miRNAs were up regulated and 5 were down regulated, and most of them are uncharacterized miRNAs (Fig 2B and S1 Table). Overall these results demonstrate that ethylene alters miRNA expression in a tissue specific manner, and we are able to identify ethylene regulated miRNAs specifically in roots in a CBP20 depend manner.
In plants, the main function of miRNAs appears to be in gene regulation. Therefore, we expected the expression of the potential targets of 12 miRNAs identified above is anti-correlated with their miRNAs in response to ethylene in Col-0 roots. We conducted RNA sequencing use the same tissues as mentioned in small RNA seq with two biology duplications (S4A and S4B Fig). Comparable numbers of ethylene-regulated genes were detected in the roots and shoots of Col-0; however, only about 20–30% of genes were overlapped between these two type tissues (Fig 2C and S1–S4 Datasets), showing the tissue specific in ethylene response. Further GO analysis showed that ethylene related GO terms were enriched in those ethylene regulated genes shared between shoots and roots, and root development related GO terms were enriched in the genes specifically regulated in roots (S5A–S5C Fig). Similarly, in cbp20 mutant the gene regulation also showed tissue specificity in response to ethylene (Fig 2C and S1–S4 Datasets).
We then compared ethylene regulated gene expression in the roots of Col-0 and cbp20. As shown in Fig 2D, about 60% of up regulated genes and 75% of down regulated genes in Col-0 roots were not altered in cbp20 roots in the presence of ethylene (S5 and S6 Datasets), showing CBP20 dependency. We then studied the association between 12 microRNAs and their targets genes in ethylene response. In total 841 potential target genes were identified and 203 with high confidence (T score < = 5) (S7 Fig and S8 Dataset). Among them, only 8 target genes were differentially regulated by ethylene in the roots of Col-0, while their differential expressions were impaired in cbp20 (Fig 2E). By comparing the expressions of ethylene altered miRNAs and their target genes in Col-0 roots, two miRNAs (miR319b, miR863-3p) were identified that up regulated by ethylene and the expression of their potential target genes was down regulated in the presence of ethylene in the roots of Col-0, while not in the roots of cbp20 (Fig 2F).
To validate the function of miRNAs identified above in response to ethylene, and the connection between their expressions with the phosphorylation state of CBP20, we examined mature miR319 by northern blot in cbp20 mutant, CBP20/cbp20, CBP20S245A/cbp20 and CBP20S245E/cbp20 with or without ethylene treatment. In consistent with small RNA sequencing result, miR319 indeed was up regulated by ethylene in the roots of Col-0 (Fig 3A), and the elevation was impaired in the roots of cbp20 mutant. Furthermore, constitutive phosphorylated CBP20S245E, while not dephosphorylated CBP20S245A were able to recover the ethylene induced elevation of miR319 expression (Fig 3A), indicating that ethylene induced phosphorylation of CBP20 potentially required for the elevation of miR319b expression in roots. We next examined the expression of pri-miR319b in the roots of Col-0, cbp20, CBP20/cbp20, CBP20S245A/cbp20, CBP20S245D/cbp20 and CBP20S245E/cbp20 treated with air or 4 hours ethylene gas by quantitative RT-PCR. As shown in Fig 3B, the expression of pri-miR319b was decreased with the ethylene treatment and the down regulation was impaired in cbp20 mutant. The down regulation of pri-miR319b was detected in the roots of CBP20S245D/cbp20 or CBP20S245E/cbp20, while not in that of CBP20S245A/cbp20 (Fig 3B), indicating that the phosphorylation is required for the down regulation of pri-miRNA, further suggesting that the elevation of miR319b in response to ethylene due to the biogenesis of miRNA, while not due to the elevation of pri-miR319b.
To further examine how the phosphorylation of CBP20 influences the gene expression of MYB33 in response to ethylene, we conducted qRT-PCR in the roots of Col-0, CBP20/cbp20, CBP20S245A/cbp20, CBP20S245D/cbp20 and CBP20S245E/cbp20 with or without ethylene treatment. The expression level of MYB33 was indeed decreased by ethylene treatment, which is consistent with RNA-seq result (Fig 3C). In cbp20 and CBP20S245A/cbp20 plants, the down regulation of MYB33 was impaired, however, in CBP20S245D/cbp20 or CBP20S245E/cbp20 (Fig 3C), the expression of MYB33 is recovered as that of in Col-0. Overall, the result shows that the expression of MYB33 is anti-correlated with the expression of miR319b specifically in roots in ethylene response, indicating that ethylene induced phosphorylation of CBP20 inhibits the expression of MYB33, which potentially through CBP20 regulated biogenesis of miR319b in roots.
To further examine whether miR319b plays a role in root growth in ethylene response, we generated the miR319b overexpression (miR319bOE) plants, and examined their phenotype in response to ethylene. As expected, the roots of miR319bOE plants were more sensitive to ethylene than that of wild type (Fig 4A–4C). MYB33 is one of potential targets of miR319b, we therefore obtained myb33 mutant to examine its phenotype in response to ethylene. As expected, the roots of myb33 mutant displayed similar phenotype as that of miR319bOE in the presence of ethylene (Fig 4A–4C and S6A and S6B Fig). Comparing to wild type, the pri-miR319b was increased (Fig 4D), while the expression of MYB33 was decreased in miR319bOE plants (Fig 4E), showing that the elevation of miR319b is not due to the elevation of its precursor in the presence of ethylene, but potentially due to the miRNA biogenesis process in response to ethylene. We then conducted a 5′ RNA Ligase-Mediated (RLM)-Rapid Amplification of cDNA ends (RACE) in both Col-0 and miR319OE plants to evaluate that MYB33 is one of targets of mir319b in vivo, In Col-0, no cleavage event was detected between the 10th nucleotide U and the 11nd nucleotide C from the 5′ end of the miRNA in Col-0. However, in miR31bOE, 5 out of 15 cleavage events were detected between nucleotides 10 and 11 from the 5′ end of the miRNA (S6C Fig), which are in consistent with the published data [32]. Taken all together, these results indicate that miR319b is involved in root growth by targeting MYB33 for degradation in a CBP20 dependent manner.
Because MYB33 is a shared target between miR319 and miR159, we examined the expression of miR159 in the roots of Col-0 and cbp20 treated with air and ethylene by northern blot. Inconsistent with small RNA-seq result, no differential expression of miR159 was detected in Col-0 roots between air and ethylene treatments (Fig 5A). As previous published data [30], we detected the reduction of miR159 in cbp20 comparing to that of in Col-0, which is consistent with published data [30]. However, no ethylene induced alteration for miR159 was detected in the roots of both Col-0 or cbp20 mutant (Fig 5A). In addition, no significant difference of pri-miR159 was detected in both Col-0 and cbp20 roots by ethylene treatment as well (Fig 5B). Furthermore, the phosphorylation mimic forms of CBP20S245D and CBP20S245E, while not dephosphorylation mimic form of CBP20S245A behaved as wild type CBP20 (Fig 5B). We further examined the pri-miR159a in two independent miR319bOE plants and found pri-miR159a was not affected by the overexpression of miR319b (Fig 5C), indicating that in the presence of ethylene, the down regulation of MYB33 is associated with the up regulation of miR319b, while not miR159.
To evaluate whether the phenotype of miR319bOE is caused by down regulation of MYB33, we constructed mutated MYB33 (mMYB33) carrying mutation in miR319 targeting site (S7A Fig) and introduced it into miR319bOE plants to obtain miR319bOE/mMYB33OE. The ethylene responsive phenotypes of both the roots and shoots of miR31bOE were recovered in miR319bOE/mMYB33OE plants (Fig 6A–6C). We then examined the expression of MYB33 in the roots of miR319bOE and miR319bOE/mMYB33OE. In consistent with Fig 5E, the expression of MYB33 was down regulated in miR319bOE plants, and was highly up regulated in miR319bOE/mMYB33OE in comparing to that of in Col-0 (Fig 6D), while the expression of pri-miR319b in miR319bOE and in miR319bOE/mMYB33OE plants were comparable (S7B Fig). TCPs are known targets of miR319, however, no ethylene induced alteration of TCPs was detected in our RNA-seq result, to further confirm the result, we examined the gene expression of TCPs by qRT-PCR in Col-0 treated with or without ethylene. No significant change was detected for those gene expressions in response to ethylene (S7C Fig). In addition, gene expression of TCP24 was not altered in miR319bOE/mMYB33 in comparing to that of in miR319bOE plants (S7D Fig), and the expression of other TCPs displayed similar patterns as TCP24 both in Col-0 and miR319OE plants (S7E Fig). Finally we conducted Agrobacterium-mediated transient co-expression assay with MYB33 or mutated mMYB33 CDS fused to 35S::LUC 3’UTR with or without the miR319b precursor. Comparing to the assay without miR319b precursor, MYB33 expression was significantly lower in the presence of miR319b precursor (Fig 7A and 7B). However, no significant change was detected for mMYB33 expression between with and without the presence of miR319b precursor (Fig 7A and 7B). The similar assay was also conducted using YFP-HA-tagged MYB33 or mutated MYB33 (mMYB33) with or without miR319b precursor to examine how miR319b influence gene expression of MYB33 and its protein level. The gene expression of MYB33, while not mMYB33, is down regulated in the presence of miR319b (S8A Fig). In consistent with gene expression, MYB33 protein is also lower in the assay with the presence of miR319b. However, mMYB33 protein was not altered by miR319b (S8B Fig). Taken all together our data support that miR319b targets to MYB33 for degradation in roots.
It has been well known that protein phosphorylation are involved in many different plant hormones such as phosphorylation regulates the polarity of PIN in auxin [33–37], gibberellins [33, 38, 39], cytokinin [40], ABA [41] and in BR signaling [42–44]. Many studies have demonstrated that MPKKK cascade promotes ACS6 and EIN3 phosphorylation [45, 46]. Recently study has demonstrated that in the absence of ethylene, the receptors activate CTR1, which phosphorylates EIN2 C-terminus [9]. With the presence of ethylene, the EIN2C is dephosphorylated and then cleaved and translocated into nucleus to activate the downstream signaling pathway [11, 12]. However, the phosphorylation regulation in ethylene signaling is still largely unknown. Genome wide phospho-peptide survey in 3-day old etiolated seedlings treated with air or ethylene was done previously [11], we found the phosphorylation of CBP20 is highly regulated by ethylene gas (Fig 1). In the absence of ethylene, no phosphorylated peptides of CBP20 were detected, while in the presence of ethylene gas, 17 spectrum counts of phosphorylation peptide (239aa-253aa) was detected. Further genetics study demonstrated that the phosphorylation of CBP20 is involved in the growth of root in the presence of ethylene (Fig 1C–1E).
CBP20 is a subunit of CBC complex, which is vital for plant development. Previous study has demonstrated that CBP80, the other subunit of CBC is involved in the regulation of hypocotyl in ethylene signaling through regulation the biogenesis of small RNAs [47]. In mammalians, it has demonstrated that growth factors mTORC1 kinase regulated S6 kinases able to phosphorylates CBP80, activating the CBC affinity for 7mG [48, 49]. However, no evidence has shown CBC complex is regulated by phosphorylation in response to hormones. Here, for the first time we provide evidence showing that CBP20 Ser245 site is highly phosphorylated with ethylene treatment. The constitutive phosphorylated CBP20, while not the constitutive non-phosphorylated CBP20 is able to rescue the ethylene root phenotype of cbp20, strongly suggesting that the phosphorylation of CBP20 is involved in ethylene response. Yet, how CBP20 phosphorylation occurs in the presence of ethylene is still undetermined. In our precious study of phosphopeptides in etr1-1 mutant with or without ethylene treatment, no phosphorylation was detected for CBP20, showing that the phosphorylation of CBP20 is ethylene dependent. Therefore, the identification of kinases that regulate CBP20 phosphorylation specifically in the presence of ethylene will be an immediately interest.
CBC complex regulate many aspects of biological processes including transcription regulation, pre-mRNA splicing, pre-mRNA 3’end processing, miRNA biogenesis, mRNA stability, mRNA and snRNA nuclear export, the pioneer round of translation and nonsense-mediated RNA decay [28]. However, no evidence has shown that CBP20 is involved in ethylene response. In our study, through high throughput sequencing for small RNAs and mRNAs in different plants treated with or without ethylene, we identified ethylene regulated miRNAs. In addition, we found that CBP20 regulates many species of miRNA expressions in response to ethylene with a tissue specific manner (Fig 2). miRNAs are involved in many different aspects of plants. Specifically in plant hormones, such as miR160 targets to several ARF family members to activate auxin signaling pathway for root cap formation [50]; miR159 targets to MYB33 to activate ABA signaling pathway for seed germination [30]. In ethylene signaling pathway, it have been reported that EIN3 represses miR164 transcription and up regulates the transcript level of NAC2 to regulate leaf senescence [51]. Here we provide evidence showing for the first time that miRNAs are differentially regulated by ethylene in a tissue specific manner (Fig 2A and 2B), and many of the differential regulations are abolished in cbp20 mutant (Fig 2A and 2B). Previous studies have shown that CBP20 is required for the biogenesis of many miRNAs [30]. Interestingly, our data showed that some miRNA species are down regulated in cbp20 mutant, which indicating non-CBP20 dependent miRNA biogenesis is potentially involved in ethylene response.
By comparing the miRNAs in Col-0 and in that of cbp20, we found many miRNA species are up regulated in cbp20 in the presence of ethylene. One possibility is that the precursors of those miRNAs are elevated by ethylene, resulting in the elevation of their miRNAs. Alternatively, CBP20 independent miRNA biogenesis machinery is elevated in the presence of ethylene, resulting in the increase of the miRNAs. However, recently study has shown that small RNA biogenesis machinery component Dicers are not involved in ethylene response [15]. Therefore, further comprehensive studies will be critical to characterize the newly identified ethylene regulated, while not CBP20 dependent miRNAs and uncover the mechanistic details that how biogenesis occurs specifically in the presence of ethylene.
MicroRNAs in plant are small RNAs, which are approximately 21 nucleotides in length. Normally, they are negative regulators of gene expression through base pairing to the complementary sequence within the target mRNAs, leading to the target mRNA degradation through RISC-mediated cleavage. In comparing to the ethylene altered small RNAs with ethylene regulated genes, we found that miR319b was up-regulated while MYB33 was down-regulated in Col-0 roots with ethylene treatment, and the regulation is CBP20 dependent. Small RNA northern blot shows that miR319 is indeed up-regulated under ethylene treatment and the regulation is dependent on CBP20 phosphorylation. However, it is well known that MYB33 is a shared target between miR159a and miR319b. The miR159 and miR319 families are similar in sequence, but they have distinct target genes: miR159 is specific for MYB transcription factors, mainly MYB33 and MYB65. In contrast, miR319 mainly targets TCP transcription factors, predominantly TCP2 and TCP4. MiR319 also targets MYB33 and MYB65, but due to its low abundance, this regulation is negligible. However, in our study we provided multiple lines of evidence showing that in the ethylene response, miR319b targets MYB33 for degradation specifically in roots, leading to the ethylene regulated cbp20 root phenotype: (1), miR319 was specifically up regulated by ethylene in Col-0 roots, while not in cbp20 mutant (Fig 4A). However, miR159 was not regulated by ethylene (Fig 5A); (2) MYB33 was down regulated in Col-0 roots in response to ethylene, while not in cbp20 roots (Fig 4C). (3) The pri-miR319b was down regulated in Col-0 roots, while not in cbp20 roots (Fig 4B); (4) The pri-miR159a was not regulated by ethylene in Col-0 (Fig 5B); (5) In the over expression miR319b plants, MYB33 was largely down regulated, while miR159 and pri-miR159a were not altered (Figs 4C, 5B and 5C); (6) Overexpression of the mMYB33 containing mutated miR319b target site is able to recover phenotype caused by miR319bOE (Fig 6); (7) Our data (Fig 7) and published data has shown that miR319b targets MYB33 for cleavage [32]. In summary, our study discovered that ethylene regulates the phosphorylation of CBP20, and the phosphorylation is required for the elevation of miR319, which leading to the down regulation of MYB33 expression in roots, resulting in root growth inhibition in the presence of ethylene.
All mutants were in the Columbia-0 (Col-0) background, cbp80 (CS878659), myb33-1 (SALK_065473), myb33-2 (SALKseq_056201) are ordered from ABRC. cbp20 has been described in [52]. Seeds were sterilized with 4% bleach and then washed as least three times with sterilized water, then the seeds were sown on MS medium. Plants were grown in long days (16h light/8h dark) at 22°C on soil.
To construct CBP20 overexpression vectors for complementing cbp20 mutant phenotype, the CBP20 full length and phosphorylation site mutated CDS sequences of CBP20 were amplified using the Phusion High-Fidelity DNA Polymerase (NEB). The PCR products were cut with KpnI and SalI, and then the corresponding fragments were ligated into the KpnI-SalI site of the pCHF3 vector to give rise to 35S:CBP20-YFP, 35S:CBP20S245A-YFP, 35S:CBP20S245E-YFP and 35S:CBP20S245D-YFP.
To construct vectors for yeast two-hybrid, the CDS of CBP80 was amplified using the Phusion High-Fidelity DNA Polymerase. The PCR product was cut with SalI and XbaI, and then the corresponding fragment was ligated into the SalI-SpeI site of the pDBLeu vector (Invitrogen) to give rise to pBD-CBP80. The CDSs of CBP20, CBP20S245A, CBP20S245E and CBP20S245D were amplified using the High-Fidelity DNA Polymerase. The PCR products were digested by SalI and SpeI, and the corresponding fragments were ligated into the SalI-SpeI sites of the pEXP-AD502 vector (Invitrogen) to give rise to pAD-CBP20, pAD-CBP20S245A, pAD-CBP20S245E and pAD-CBP20S245D.
To construct miR319b overexpression vector, a 1kb genomic DNA contain the full length of pri-miR319b was amplified using the Phusion High-Fidelity DNA Polymerase. The PCR product was digested with KpnI and SalI, and then the corresponding fragment was ligated into the KpnI-SalI site of the pCHF3 vector to give rise to pCHF3-miR319b. All the sequences above were verified by sequencing.
The binary constructs were introduced into Agrobacterium tumefaciens strain GV3101 by electroporation and then introduced into Col-0 or cbp20 mutant plants by the floral dip method [53]. Transgenic plants were screened on MS plates in the presence of 50 mg/L kanamycin, and homozygous lines were verified by antibiotic selection. For each construct, multiple independent lines were examined with similar results, and as least one representative line was shown.
The data has been collected from previous study and the calculation was also followed the method as published [11].
Arabidopsis seeds were sown on MS medium plates with or without addition of 1 μM or 10 μM 1-aminocyclopropane-1-carboxylic acid (ACC, Sigma), the biosynthetic precursor of ethylene. After 3 days of cold treatment, the plates were wrapped in foil and kept in 22°C dark chamber for 3 days. The hypocotyls and roots were measured using NIH Image (http://rsb.info.nih.gov/nih-image/).
The yeast two-hybrid assay was performed according to the ProQuest™ Two-Hybrid System (Invitrogen). Briefly, pBD-CBP80 and pAD-CBP20, -CBP20S245A, -CBP20S245E or -CBP20S245D were co-transformed into the yeast strain Mav203 (Invitrogen). The transformants were grown on SD/-Trp-Leu medium or SD/-Trp-Leu-His with 10mM 3AT dropout medium. The transformants growing on SD/-Trp-Leu-His with 10mM 3AT dropout medium indicates interaction between corresponding proteins. Primers used in this assay were listed in S5 Table.
The seedlings of 35S:CBP20-YFP, 35S:CBP20S245A-YFP, 35S:CBP20S245E-YFP and 35S:CBP20S245D-YFP transgenic plants were grown on MS medium with or without addition of 10 μM ACC in dark for 3 days in 22°C. Then the YFP fluorescence of root tips was observed under Zeiss LSM 710 Confocal microscopy.
Arabidopsis seeds were grown on MS medium in the air-tight containers in the dark at 22°C supplied with a flow of hydrocarbon-free air (Zero grade air, AirGas) for 3 days. The plants tissues were harvest after with continually flow of hydrocarbon-free air or hydrocarbon-free air with 10 parts per million (ppm) ethylene gas for 4 hours as previously described [7].
Total RNA was extracted using a RNeasy Plant Kit (Qiagen) from 3 days etiolated seedlings treated with air or 4 hours ethylene gas. First-strand cDNA was synthesized using Superscript III First-Strand cDNA Synthesis Kit (Invitrogen). Real time PCR was performed with the LightCycler 480 SYBR Green I Master (Roche) following the manufacturer’s instructions. PCR reactions were performed in triplicate on a Roche 96 Thermal cycler. The expression level was normalized to UBQ10 control.
Total RNA was isolated from roots or shoots of 3-day old etiolated seedlings treated with air or 4 hours ethylene gas using TRIzol reagent (Invitrogen). For mRNA library construction, in briefly, the mRNA was isolated by NEBNext Poly(A) mRNA Magnetic Isolation Module and fragmented at 94°C for 15mins. Then the cDNA was synthesis by NEBNext Ultra Directional RNA Library Prep Kit for Illumina. The PCR reactions were conducted by using different index primers (NEBNext Multiplex Oligos for Illumina). The PCR products were purified by Agencourt AMPure XP Beads (Beckman Coulter). The quality of the libraries was assessed by Bioanalyzer (Agilent High Sensitivity Chip). The libraries then were sequenced on Hiseq 4000 Systems (Illumina).
For small RNA library construction, in briefly, the cDNAs were synthesized using NEBNext Small RNA Library Prep Set for Illumina. The PCR reaction was amplified by different Index primers and the PCR products were first purified by the Agencourt AMPure XP Beads, and then selected the size using 6% PolyAcrylamide Gel. The ~140 bp bands corresponding to miRNAs were isolated. The library quality was assessed on Bioanalyzer (Agilent High Sensitivity Chip). The libraries were sequenced on Hiseq 4000 Systems (Illumina) after assessed on Bioanalyzer.
miRNA prediction pipeline was written by Python scripting language. High-quality small RNA reads were obtained from raw reads through filtering out poor quality reads and removing adaptor sequences using FASTX toolkit [54]. Adaptor-trimmed unique sequences were aligned to TAIR10 Arabidopsis genome using bowtie [55] and structural RNAs such as tRNA, rRNA, snRNA, and snoRNA were excluded. The perfect matched reads between 18–28 nucleotides (nts) in length were selected. To obtain the precursor sequences, potential miRNA sequences (reads ≥ 50) were extended upstream and downstream of 100 to 500 nts with a step size of 100 nts. Each putative precursor sequence was folded using RNA fold from Vienna RNA software package [56], and the potential miRNA* sequences were selected with mismatch ratio of 0.3 or less. The region of these putative precursor sequences with addition of 15 nts marginal sequences were re-folded using RNA fold to check whether miRNA/miRNA* duplex was suitable for primary criteria for annotation of plant miRNAs [57]. The miRNA candidates were essentially grouped into families by mature sequence similarity and/or loci. Using the miRNA annotation information of Arabidopsis thaliana in miRBase 21 (http://www.mirbase.org), all members of miRNA candidate families of the known miRNAs were selected.
The putative target sites of miRNAs were identified by aligning mature miRNA sequences with the Arabidopsis cDNA sequences using TargetFinder (http://carringtonlab.org/resources/targetfinder). miRNA targets were computationally predicted essentially as described [58–60]. Briefly, potential targets from FASTA searches were scored using a position-dependent, mispairing penalty system. Penalties were assessed for mismatches, bulges, and gaps (+1 per position) and G:U pairs (+0.5 per position). Penalties were doubled if the mismatch, bulge, gap, or G:U pair occurred at positions 2 to 13 relative to the 5’-end of the miRNAs. Only one single-nucleotide bulge or single-nucleotide gap was allowed, and the targets with penalty scores of six or less were considered to be putative miRNA targets.
RNA-seq raw reads were aligned to TAIR10 genome release using Top Hat version 2.0.9 [61] with default parameters. Differential expressed genes were calculated by Cufflinks version 2.2.1 following the workflow with default parameters [62]. Differentially expressed genes were those for which relative fold change values of larger than 1.5 and RPKM value larger than 1 were observed. To evaluate reproducibility of the RNA-seq data, the expression levels between two replicates for each sample and conditions were compared for all genes with FPKM > 0.5 in both replicates. The log2 transformed FPKM values (log2 (FPKM + 1) was calculated, then R scripts were used to analyze the correlation between biological replicates.
Total RNA was isolated from root of 3 days etiolated seedlings treated with air or 4 hours ethylene gas using TRIzol reagent (Invitrogen). 10ug RNA of each sample was separated on 15% denaturing 8M urea-PAGE gel and then transferred and UV crosslinked onto BrightStar®-Plus Positively Charged Nylon Membrane (Ambion). The membrane was pre-hybridized by ULTRAhyb®-Oligo Hybridization Buffer. miRNA probes were end-labelled by T4 Polynucleotide Kinase (NEB) with r-P32 ATP. The membrane was hybridized with probe overnight and then wash by 2xSSC for two times. Then the membrane was exposed to a phosphor imager screen and the relative abundance levels were measured by ImageQuant TL software.
5’RLM-RACE was performed following the manufacturer’s instructions of FirstChoice RLM-RACE Kit (Ambion). Briefly, total RNA (10 μg) from root of Col-0 and miR319b OE line was directly ligated to the 5’RACE Adapter by T4 RNA ligase (Ambion). cDNA was synthesized using Superscript III First-Strand cDNA Synthesis Kit (Invitrogen) use Oligo (dT) primer. Gene-specific reaction was first done with the 5’RACE Outer Primer and gene-specific primer MYB33 SpeI-R. Then the PCR product was purified and performed by the second round of PCR using 5’RACE inner Primer and gene-specific primer MYB33 SpeI-R (S5 Table). The 5RLM-RACE product was gel purified, digested with Sal I and Spe I and then cloned into pDBLeu vector for sequencing.
Transient expression assay in N. benthamiana were performed by infiltrating 4-week-old N. benthamiana plants with Agrobacterium containing MYB33 or mutated MYB33 CDS with or without Agrobacterium harbouring constructs containing the miR319b precursor. Leaf tissue was collected 3 days later for RNA and protein analysis. For luciferase assay, Agrobacterium containing MYB33 or mutated mMYB33 CDS fused to 35S::LUC 3’UTR with or without Agrobacterium harbouring constructs containing the miR319b precursor were injected in to N. benthamiana plants. After 3 days, The leaves were sprayed with 500 μM luciferin (Promega, Madison, Wisconsin) and placed in the dark for 5 min. Luciferase activity was observed using NightOWL LB 983 in vivo Imaging System (Berthold, Oak Ridge, Tennessee). Primers used in this study were listed in S5 Table.
|
10.1371/journal.pbio.1000388 | Defensin-Like ZmES4 Mediates Pollen Tube Burst in Maize via Opening of the Potassium Channel KZM1 | In contrast to animals and lower plant species, sperm cells of flowering plants are non-motile and are transported to the female gametes via the pollen tube, i.e. the male gametophyte. Upon arrival at the female gametophyte two sperm cells are discharged into the receptive synergid cell to execute double fertilization. The first players involved in inter-gametophyte signaling to attract pollen tubes and to arrest their growth have been recently identified. In contrast the physiological mechanisms leading to pollen tube burst and thus sperm discharge remained elusive. Here, we describe the role of polymorphic defensin-like cysteine-rich proteins ZmES1-4 (Zea mays embryo sac) from maize, leading to pollen tube growth arrest, burst, and explosive sperm release. ZmES1-4 genes are exclusively expressed in the cells of the female gametophyte. ZmES4-GFP fusion proteins accumulate in vesicles at the secretory zone of mature synergid cells and are released during the fertilization process. Using RNAi knock-down and synthetic ZmES4 proteins, we found that ZmES4 induces pollen tube burst in a species-preferential manner. Pollen tube plasma membrane depolarization, which occurs immediately after ZmES4 application, as well as channel blocker experiments point to a role of K+-influx in the pollen tube rupture mechanism. Finally, we discovered the intrinsic rectifying K+ channel KZM1 as a direct target of ZmES4. Following ZmES4 application, KZM1 opens at physiological membrane potentials and closes after wash-out. In conclusion, we suggest that vesicles containing ZmES4 are released from the synergid cells upon male-female gametophyte signaling. Subsequent interaction between ZmES4 and KZM1 results in channel opening and K+ influx. We further suggest that K+ influx leads to water uptake and culminates in osmotic tube burst. The species-preferential activity of polymorphic ZmES4 indicates that the mechanism described represents a pre-zygotic hybridization barrier and may be a component of reproductive isolation in plants.
| Sperm cells of animals and lower plants are mobile and can swim to the oocyte or egg cell. In contrast, flowering plants generate immobile sperm encased in a pollen coat to protect them from drying out and are transported via the pollen tube cell towards the egg apparatus to achieve double fertilization. Upon arrival the pollen tube tip bursts to deliver two sperm cells, one fusing with the egg cell to generate the embryo and the other fusing with the central cell to generate the endosperm. Here, we report the mechanisms leading to pollen tube burst and sperm discharge in maize. We found that before fertilization the defensin-like protein ZmES1-4 is stored in the secretory zone of the egg apparatus cells and that pollen tubes cannot discharge sperm in ZmES1-4 knock-down plants. Application of chemically synthesized ZmES4 leads to pollen tube burst within seconds in maize, but not in other plant species, suggesting this mechanism may be species specific. Finally, we identified the pollen tube-expressed potassium channel KZM1 as a target of ZmES4, which opens after ZmES4 treatment and probably leads to K+ influx and sperm release after osmotic burst.
| Flowering plants (angiosperms) emerged some 180–140 MYA [1] and have since inhabitated most ecological environments, which are often far away from humid conditions. Hence, new reproductive mechanisms were acquired to keep cells from drying out and to realize species-specific interactions between male and female reproductive structures, partly over long distances. Adaptive selection led to the reduction of the haploid male gametophyte to a three-cellular pollen grain and pollen tube, respectively, which is able to be transported over long distances and to grow deeply inside female reproductive tissues. As a further consequence the sperm cells lost their motility. The reduced female gametophyte (embryo sac), which is haploid in most plant species, is deeply embedded and protected in the maternal tissues of the ovule and ovary and harbors the female gametes (egg and central cell) as well as some accessory cells (synergid and antipodal cells; Figure 1A). Of these, the synergid cells are involved in pollen tube signaling, sperm delivery, and transport [2]. Extensive cell-cell communication events are likely to take place between both gametophytes, among male and female gametophyte cells, respectively, as well as the surrounding sporophytic tissue [3]. The central events preceding fertilization involve signaling towards the pollen tube to arrest growth and to induce discharge of the two sperm cells, a process first described by the famous German-Polish botanist Eduard Strasburger in 1884 [4]. It took more than 120 years to identify the first molecular players involved in these processes.
Cross-talk between both gametophytes to arrest pollen tube growth has recently been shown to depend on the FER receptor-like kinase (RLK) [5] and the GPI-anchored protein LRE [6] localized at the synergid plasma membrane. Nitric oxide and reactive oxygen species seem to play additional roles in this process [7],[8], and the activity of the Ca2+ pump ACA9 [9] localized at the pollen tube plasma membrane was shown to be required for pollen tube rupture. We were investigating whether small secreted cysteine-rich proteins (CRPs), specifically expressed in the embryo sac cells, play a role in these processes. Plant genomes encode large classes of CRPs accounting for more than 800 genes in Arabidopsis and approximately 600 genes in rice. Among them, defensins/defensin-like proteins (DEFs/DEFLs), lipid transfer proteins (LTPs), rapid alkanization factor (RALF) proteins, and thionins resemble the largest classes [10]. The majority of these genes are expressed in reproductive tissues and a number of genes have been shown to be specifically expressed in the cells of the embryo sac [11],[12],[13],[14],[15],[16]. Here, we report the functional analysis of a small gene-family of four members encoding DEFL proteins in the maize inbred line A188. All four genes are specifically expressed in the embryo sac and have been shown previously to be down-regulated immediately upon fertilization [12].
ZmES1-4 protein localization was studied in the maize embryo sac cells before and after fertilization in plants expressing GFP fused to the C-terminus of ZmES4 under control of its endogenous promoter. All transgenic lines showed fusion protein localization exclusively in the embryo sac cells. Some variation was observed among the cell types expressing ZmES4-GFP: most ovules expressed it in both synergid cells, others in about 80% ovules only in one synergid cell, while others showed expression additionally in egg and central cells. Signals in the antipodal cells were never observed. As shown in Figure 1A–1C and Figure S1N and S1O, the fusion protein was most prominently located in the secretory zone of the two synergids surrounding the filiform apparatus before fertilization. Unlike the ZmEA1-GFP protein, which is secreted from the egg apparatus and plays a role in short range pollen tube guidance [17], GFP fluorescence was not detected in the cell walls of micropylar nucellus cells. Weaker signals were visible in the central cell of some but not all transgenic lines. In a stack of confocal laser scanning microscopy images (Figure 1D) of a very young cellularized embryo sac, displayed fusion protein signals were also found in the endoplasmic reticulum (ER) of the central cell. 15 hours after pollination (hap), around 7 to 8 h after fertilization, ZmES4-GFP was no longer detectable in the degenerated receptive synergid cell and polar localization in the persisting synergid cell was lost (Figure 1E and 1F). Faint fluorescence, accumulating in the ER around the nucleus, was visible in the fertilized egg cell of some transgenic lines. The fusion protein was no longer detectable in all lines 24 hap, indicating that it was actively degraded after fertilization. Time course measurements using a line expressing the fusion protein in about 80% ovules only in one of the two synergid cells showed that signals first appeared after the whole embryo sac is fully differentiated (Figure 1G). During further maturation and embryo sac enlargement, most fluorescence was visible in the ER around the nucleus and the secretary zone of the synergid cell (Figure 1H). In this line, the single ZmES4-GFP expressing synergid cell was the exclusive target of the pollen tube (Figure 1I). Transient transformation of onion epidermis cells was used to investigate the secretion of ZmES proteins. As shown in Figure S1, ZmES4-GFP fusion protein entered the secretory pathway and seemed to be present in golgi stacks, constitutive secretory vesicles, as well as in the cell wall. We suggest, however, that in contrast to onion epidermis cells, in non-degenerate synergid cells, ZmES4-GFP appears to be retained in regulated secretory vesicles, as signals were not observed in the cell wall before fertilization.
In order to study the DEF activity of ZmES proteins, ZmES4 was expressed under the control of the ubiquitously expressed 35S promoter in Arabidopsis. As shown in Figure S2A and S2B, antibacterial activity, which has also been reported for most plant DEFs studied so far [18], was not observed. Although antifungal activity was also not observed at the seedling stage (Figure S2C), infected ZmES4 over-expressing seedlings recovered faster than control seedlings and fungal hyphae were no longer macroscopically visible a few weeks after infection. In contrast, control plants still contained fungal hyphae, displayed chlorosis, and delayed flowering (Figure S2D). These experiments indicated that ZmES proteins still display low DEF activity and are apparently able to bind to fungal targets with a low affinity. Their major physiological function, however, is likely to be different.
To study the role of ZmES proteins during fertilization, the whole gene-family was down-regulated using RNAi-silencing. ZmES4-RNAi plants, which showed weak or lack of the RNAi-transcript, did not show any obvious phenotype. Plants that contained high RNAi-transcript levels displayed female sterility after self-pollination (Figure 2A) but full seed-set after back-crossing to wt plants. Progeny of a single copy RNAi-line back-crossed to wild type plants displayed 43% transmission of the transgene via the pollen (n = 28). Self-pollination showed 50% transmission (n = 30), indicating almost full transmission of the silencing construct via the pollen but low transmission via the female gametophyte.
After pollination with an ACTp:GUS marker line [17], maternal tissues of the ovary and ovule were sectioned to visualize the pollen tube during the fertilization process. As shown in Figure 2B, a wt pollen tube penetrated the embryo sac, released its content, and blue staining of both egg and central cell indicated successful fertilization due to the activity of sperm derived paternal GUS genes inside female gametes. After fertilization, guidance signals apparently no longer existed, as additional pollen tubes failed to penetrate the micropylar nucellus region of the ovule. In contrast, ZmES4-RNAi plants displayed penetration of the micropylar nucellus, but GUS signals were neither detectable inside the gametes nor inside the synergid cell (Figure 2C–2E). Moreover, both synergid cells were still intact and gamete delivery had not occurred. Pollen tube over-growth was partially observed inside the egg apparatus region (Figure 2E) or growth around the egg apparatus (Figure 2G). Here, the very mature egg apparatus cells started to disintegrate. Fertilized wt ovules of the same cob show enlargement of the embryo sac and initiation of embryo and endosperm development (Figure 2F). A summary of fertilization rates is shown in Figure 2H (see also Table S1): two ZmES4-RNAi-lines and a single copy back-crossed line were analyzed for occurrence of fertilization around 1 and 2d after fertilization (27–32 hap and 53–55 hap), respectively. GUS activity in both, egg and central cell, was measured as successful fertilization. In contrast to wt embryo sacs that displayed a fertilization rate of 88%, ZmES4-RNAi lines showed a reduction to 60%. This value is higher than the expected 44% of a genetic null-mutant indicating that either the silencing effect is incomplete and/or additional factors are involved in the release of pollen tube contents.
Due to limitations in visualizing the double fertilization process in planta, we chemically synthesized mature ZmES4 protein and investigated its effect on the growth behavior of in vitro in liquid-germinated and -grown pollen tubes of maize and other plant species. Buffer containing 30 µM ZmES4 induced rapid pollen tube burst at the very tube tip within seconds after application to germinated pollen tubes (Figure 3B). Within less than a minute almost 100% pollen tubes ruptured and tube content was released explosively. Even at 1,000-fold lower concentrations, one-third of pollen tubes bursted within less than 2 min (Figure S3). This reaction was species-preferential as pollen tube rupture in the maize relative Tripsacum dactyloides was essentially delayed (around 1,600–2,300 s versus 35–50 s in maize, independent from the maize genotype used; Figure 3C and 3D) or did not occur at all in other plant species tested, such as tobacco or lily (Figure 3D). In contrast, other extracellular CRPs, trypsin inhibitor from soybean [19], or LURE2 of Torenia fournieri [20] or AFP2 from Raphanus sativus [21] (Figure 3A) did not affect pollen tube growth behavior when applied at the same concentration and the same conditions. This indicates that maize and their relatives possess species-preferential ZmES targets.
Rapid pollen tube rupture pointed to an osmotic process and, consequently, altered solute transport at the pollen tube plasma membrane. In order to obtain a mechanistic clue about ZmES activity, the membrane potentials of growing pollen tubes were recorded, using the micro-electrode impalement technique. Membrane potentials in the range of −70 to −100 mV (n >10) were recorded in fast growing pollen tubes (average −80 mV). In contrast to control substances, ZmES4 application led to a transient membrane depolarization followed by repolarization efforts, but the burst generally occurred within 1 to 2 min before repolarization was achieved (Figure 3E). These measurements indicate that ZmES4 application either leads to Cl− efflux or H+ and/or K+ influx and that ZmES4 target(s) likely represent ion-channel(s).
We aimed to identify ZmES4 target(s) and initially analyzed the secondary and tertiary structure of predicted mature ZmES proteins. Based on structure modeling ZmES proteins are more closely related to plant DEFs such as RsAFP2 and invertebrate venom peptides of various species including scorpions, snakes, sea anemones, spiders, insects, marine cone snails, and worms [22]. Less structural homology to other plant CRPs, such as the S-locus group and LURE1/2 (Figures S4 and S5) or the phylogenetic distinct vertebrate DEFs [23], was evident. Despite limited overall sequence identity, ZmES DEFLs and related animal toxins contain a conserved structure consisting of an α-helix and a triple-stranded antiparallel β-sheet (in a βαββ configuration) that is stabilized by 3–4 intramolecular disulfide bonds (Figure 4A) [22]. Although this core structure is highly conserved, the primary sequence displays little amino acid identity. As a consequence, the surface is highly polymorphic (Figure 4A) and different from each other, indicating that these proteins bind to different targets. Animal DEFs/toxins related to ZmES containing the described βαββ-structure have been shown to modulate either K+ or Na+ ion channels [22] by acting as pore blockers or gating modifiers, supporting the assumption of a role for ion channels in ZmES4-mediated pollen-tube rupture. To test the possibility of CsCl or BaCl2 as ion channel blocking agents [24], these ions were added to the pollen-tube growth medium. While addition of 100 µM MgCl2 (control) and BaCl2 did not show any significant alteration of ZmES4 effects, strong delay of more than eight times of pollen tube rupture was observed when 100 µM of the K+ channel specific inhibitor CsCl [24] was present in PGM (Figure 3D). This indicates that K+ channels could represent ZmES4 target(s).
To study the activity of potassium channels in response to ZmES4 application we expressed several known channels of the plant shaker K+ channel family of maize in Xenopus oocytes. This structurally related family consists of three different subgroups: inward rectifying (K+ uptake) channels, outward rectifying (K+ release) channels, and weakly voltage dependent channels that conduct outward as well as inward K+ currents. The rectification properties of these functionally distinct subgroups are based on their voltage dependence. Whereas outward rectifying channels are activating with a time-dependent kinetic upon depolarization, inward rectifying channels are activating with a time-dependent kinetic at hyperpolarized membrane potentials. The third subgroup consists of channels that are only weakly voltage dependent and thus their activation kinetics appear instantaneously. In maize three shaker-like K+ channels electrophysiologically characterized are shown to be expressed in pollen tubes (Figure S6). The kinetics of the inward rectifying channel ZMK1 and the weakly voltage-dependent channel ZMK2 [25] remained unchanged by ZmES4 application (Figure S7E). In contrast, the inward rectifying Shaker K+ channel KZM1 [26] lost its voltage dependence upon external application of ZmES4 (Figure 4B and 4C). Moreover, KZM1 was inhibited by Cs+ in a voltage dependent manner (Figure S7A), supporting the finding that ZmES4 induced pollen tube burst could be significantly delayed by Cs+ application. In line with the properties of a voltage-independent channel, KZM1-mediated K+ currents appeared with an instantaneous activation kinetic in response to all tested membrane voltages, leading to inward as well as outward K+ currents (Figure 3B and 3C, Figure S7B). Interestingly, at membrane potentials more negative than −120 mV the currents in the presence of ZmES4 were smaller than under control conditions. This observation indicates that ZmES4 exhibits a weak inhibitory effect at very negative voltages in addition to its influence on the gating of KZM1. Due to the weak voltage dependence in the presence of ZmES4, the currents reversed at the Nernst potential for potassium. This confirms that the observed ZmES4-induced currents were due to KZM1. Furthermore, the effect of ZmES4 was fully reversible and KZM1 regained voltage dependence and rectification properties upon protein washout (Figure S7B). In contrast to ZmES4, 100 µg/ml of the CRP RsAFP2 [21] changed neither the activation kinetics nor the voltage dependence of KZM1 (Figure S7C and S7D).
After the first description of fertilization in flowering plants more than 120 y ago [4], we are now beginning to understand the underlying molecular and physiological mechanisms involved in pollen tube growth arrest and burst leading to sperm discharge and gamete fusion. Here, we described the activity of the synergid cell secreted DEFL protein ZmES4, required for species-preferential pollen tube burst in maize and its activity on the pollen tube expressed potassium channel KZM1. ZmES4 application triggered membrane depolarization and opening of the pollen tube expressed potassium channel KZM1 at physiological conditions before tube burst, which occurred on average within less than a minute. ZmES4 action is very fast, suggesting that a rapid decrease in the cytosolic water potential represents the main physiological cause of pollen tube burst. Moreover, ZmES4 activity was significantly delayed in the presence of Cs+, a K+ channel specific blocker, supporting the finding that changes in potassium fluxes play a central role for pollen tube burst in maize. Indeed, osmotic processes represent important forces driving plant growth, movements, and development. Besides sugars, potassium is the major osmotically active solute to maintain plant cell turgor and drives irreversible cell expansion and reversible changes in cell volume [27]. Although the potassium concentration in the filiform apparatus, the entry point of the pollen tube inside the embryo sac, is not known, it was shown that synergid cells contain considerably high potassium concentrations [28]. Once released by the degenerating receptive synergid cell, elevated apoplastic potassium concentrations together with a given hyperpolarized pollen tube membrane potential might provide the necessary driving force allowing rapid potassium influx upon K+ channel opening. This hypothesis is further supported by the observation that high cytosolic potassium concentrations in synergid cells are concomitantly lost with pollen tube discharge [28].
In summary we showed that ZmES4-GFP fusion proteins accumulate before fertilization in vesicles at the secretory zone of mature synergid cells, which have been considered to represent the glandular cells of the female gametophyte [29],[30],[31]. Upon pollen tube arrival, vesicles contents seem to be released and disappear within 24 h. Using an RNAi knock-down approach, we found that pollen tubes were guided towards the female gametes but fail to release their contents and occasionally show overgrowth inside the egg apparatus. Using chemically synthesized ZmES4 protein, we further showed that it induced pollen tube rupture in vitro in a species-preferential manner. Moreover, ZmES4 converts the potassium channel KZM1 from a voltage-dependent inward rectifier in a voltage independent, non-rectifying channel. KZM1 is assumed to represent a major housekeeping channel in many tissues and accounts for K+ homeostasis. KZM1 has also been considered as a mediator of K+ uptake into the phloem as well as guard cells in the leaf epidermis [26]. Pollen tubes have been shown to be especially sensitive to osmotic changes and it is therefore not surprising that a large number of ion channels are involved in osmotic adjustment and K+ homeostasis and are expressed during pollen development and tube growth (see also Figure S6) [32]. KZM1 is strongly expressed in pollen tubes, but it cannot be excluded that ZmES proteins also act at other pollen tube expressed targets including Ca2+ channels—a property reminiscent to animal toxins capable of modulating various ion channels simultaneously [18],[33]. However, all described activities are capable to lead to a rapid increase in osmotic pressure in the pollen tube, which ultimately results in pollen tube rupture at its weakest point, the very tip lacking callose containing cell wall material. As a consequence, sperm cells and pollen tube tip factors required for fertilization processes are released explosively into the receptive synergid, a process described previously already at the cellular level in Torenia fournieri [34]. As described above, this process is species-preferential, as ZmES4 strongly modulated KZM1 activity but obviously showed low activity on ion-channels of related T. dactyloides pollen tubes and no significant effect on tubes of other plants tested. We discovered an up to now unknown additional species-specific reproductive hybridization barrier that could represent a further component of reproductive isolation and, thus, speciation in plants [29],[35],[36]. Moreover, this finding opens the possibility to introduce ZmES related CRPs of the DEFL subclass in embryo sac cells or potassium channels in pollen tubes of various grass species to overcome species-specific crossing barriers.
The surface loop that is responsible for species-specific antifungal activity of the plant DEF RsAFP2 has been mapped as C5-C6-loop [21],[37], the same loop that protrudes from the ZmES1 structural model (arrow in Figure 4A). This region differs in ZmES1 and ZmES2-4 indicating that ZmES CRPs may indeed target different proteins. Additionally, this region indicates a first sub-domain that could be engineered in order to overcome its species-specific properties.
Considering that the egg apparatus secretes a large cocktail of diverse small CRPs [11],[12],[13],[14],[15],[16], perhaps more ion channels and/or receptors are affected simultaneously culminating in explosive tube discharge. This hypothesis is supported by the finding that RNAi silencing of the whole ZmES family did not lead to the expected 50% reduction of the fertilization rate. However, it also cannot be ruled out that the knock-down effect was indeed 100% efficient. Thus a systematic approach is now required to analyze the functions and activities of the various embryo sac expressed DEFL genes on all pollen tube expressed ion channels and other receptors. Interestingly, our findings show that pollen tube growth inhibition and discharge is not only mechanistically related to defense mechanisms against fungal attack, but moreover similar molecular players are involved. While many DEFs are required to protect both plant and animal reproductive tissues from pathogens [29],[38],[39], ZmES proteins obviously evolved a novel function, although still possessing low antifungal activity. As indicated above, a large number of DEFL genes are also expressed in the Arabidopsis and Torenia fournieri female gametophyte [11],[15],[20] indicating that the reported physiological mechanism of tube burst might well be conserved in angiosperms and that more DEFL genes obtained other functions. Hoverer, due to the polymorphic nature of DEFL genes, orthologous genes cannot be predicted and have to be identified experimentally. For example, related DEFL genes LURE1 and LURE2 from Torenia fournieri have recently been shown to encode the female gametophyte secreted attractants of the pollen tube [20] providing further support for the assumption that DEFL genes have evolved specific functions during plant reproduction. Interestingly, a similar evolutionary phenomenon has recently been proposed for the FER RLK: a homozygous fer mutant in Arabidopsis leads to fungal resistance [40], although the function of this RLK in the ovule seems to be restricted to pollen tube growth arrest [5]. Thus, after the identification of the first players involved in pollen tube growth arrest and burst, it will now be exciting to find out how the pathways are connected and how they evolved: for example, do FER or LRE [6] signaling pathways lead to secretion of ZmES1-4 containing vesicles similar to the release of cortical granules of animal species [41]? Do ZmES proteins also act on maize homologs of Arabidopsis ACA9 Ca2+ pumps [9] whose activity is also required for pollen tube burst? And finally, can we switch back the evolutionary clock and engineer “reproductive” DEFLs into “defense” DEFLs and vice versa? Systematic approaches to study female gametophyte expressed DEFL gene families will provide exciting answers during the next few years.
Maize inbred lines A188 and H99 as well as transgenic lines were grown under standard greenhouse conditions at 26°C with 16 h light and a relative air humidity of about 60%. Pollen grains were harvested as follows: ears were shaken in the early morning to remove old pollen grains. At 9–10 a.m. tassels containing mature pollen grains were covered by paper bags and hand shaken to collect fresh pollen that was used to pollinate ears, which were bagged before silk emergence. Silks were cut to 4–8 cm immediately before pollination. Arabidopsis seeds were vernalised in growth chambers for 2 d at 4°C without light and an average air humidity of about 55%–60%. After vernalisation seeds were germinated and grown in growth chambers at long day conditions with 16 h light (22°C and 20°C in the dark). Pseudomonas syringae inoculated Arabidopsis plants were grown in separate growth chambers under short day conditions with 9 h of light (18°C). Onion bulbs for transient transformation were obtained from AGRATA GmbH.
To generate the ZmES4-GFP-fusion construct (ZmES4p:ZmES4-GFP) for maize transformation, 1,620 bp upstream of the ZmES4 ORF was used as the promoter region. The eGFP gene was PCR amplified from the eGFP expression vector pMon30049 [42] with the primers GFPBam (5′-GGATCCGGCCGATGGGCAAGGGC-3′) and GFPXho (5′-CTCGAGTCACTTGTAGAGTTCATCC-3′) and digested with BamHI and XhoI. The NOS-terminator was amplified from pBi121 with primers SANF (5-GTCGACTCGAATTTCCCCGATC-3′) and NOEco (5′-GAATTCCCGATCTAGTAAC-3′) and digested accordingly. The ZmES4 promoter was amplified from maize inbred line A188 genomic DNA with primers PESSpe (5′-ATAGTTATTGATCTACTGGTCATGTAC-3′) and PESr (5′-CTGTGTCAGGCAGTC-3′). The ZmES4 ORF was amplified from cDNA clone ZmEC44/6[12] with primers CESf (5′-GTTCCACCACATTACTTCC-3′) and ESBsp (5′-GCGACTAGTTCCACCACATTAC-3′). The resulting fragments were amplified through overlapping PCR with primers PESSpe (5-ACTAGTTATTGATCTACTGGTCATGTAC-3) and ESBsp (5-GGGCCCTTTTGTCGTGGTGGATGTGC-3), restricted with SpeI and Bsp120, and cloned into the appropriate interfaces of the vector L29GFPeag (DNA Cloning Service). For transient expression studies, the ZmES4 promoter was replaced by the maize ubiquitin promoter present in the vector pLNU-GFP (DNA Cloning Service). The ORF of ZmES4 was amplified from clone ZmEC44/6 [12] with primers ESR-GFP (5′-GTCGGATCCATTTTGTCGTGGTG-3′) and ESF-GFP (5′-CGCGACTAGTTCCACCACATTAC-3′). The resulting fragment was cloned into the corresponding restriction sites (SpeI-BamHI) of the pLNU-GFP vector. This vector was also used as a positive control for transient biolistic transformation experiments.
Epidermal onion cell layers were bombarded with 2–5 µg plasmid DNA according to the procedure described [43], except that inner onion peels (2×2.5 cm) were placed with the concave side up on 0.5% agar plates. The condition of bombardment was 1,100 psi rupture discs under a vacuum of 28 mm Hg with 6 cm target distance using the Particle gun model PDS100/He (BioRad). Bombarded peel-halves were placed after transformation with the concave side down and the cut surface in sterile 0.6% agar Petri dishes for about 17–22 h in the dark before removing the epidermis for observation using a fluorescence microscope.
For stable transformation of immature maize embryos, 5 µg plasmid DNA was precipitated onto gold particles with an average size of 0.4–1.2 mm (Heraeus) following a modified protocol [44] originally described by BioRad. Particle-DNA pellets were first resuspended in 240 µl ethanol and then 3.5 µl was each spread on the macrocarrier for particle bombardment. Immature embryos were isolated 12–13 d after pollination from A188 inbred ears after pollination with pollen from maize inbred line H99. Isolated hybrid embryos were pre-cultured 8–14 d and an additional 4 d in osmotic medium before bombardment. Co-transformation was carried out with a 35Sp:PAT-construct carrying the selectable marker PAT for glufosinate ammonium resistance (to select for BASTA® resistance). Particle bombardment, tissue culture, and selection of transgenic maize plants were carried out as described [45]. To study the role of ZmES genes during fertilization, an RNAi construct was generated to silence the whole gene-family simultaneously. The construct was designed as follows: 400 bp of the clone ZmEC 44/6 was amplified in sense orientation with primers ES4F-Eco (5′- GTCAGAATTCACCACATTACTTCCA-3′) and ES4R-Bam (5′-GACTGGATCCCAAGACATTTACAA-3′) as well as inverse with primers ES4F-Bsr (5′- GCGCTGTACACCACATTACTTCCA-3′) and ES4R-Mlu (5′-GACTACGCGTCCCAAGACATTTACAA-3′). After digestion with respective restriction enzymes, both fragments were cloned into the corresponding splicing sites of the pUbi-iF2 vector that contains a maize Ubi promoter, 1,121 bp of the Fad2 intron, and the NOS terminator. This construct was generated and like the other constructs sequenced by DNA Cloning Service.
To constitutively express ZmES4 in Arabidopsis for pathogenity tests, the vector pBD515.3 [46] as well as the clone ZmEC 44/6 carrying a ZmES4 cDNA were cut with BamHI and KpnI, ligated, and fully sequenced with the primer ZmES5AF (5′-AAAACGAATAATAATCCGGCAATGGAGTCTT-3′). Arabidopsis shoot apical meristems were transformed via the Agrobacterium-mediated vacuum infiltration method previously described [47].
Genomic DNA from maize leaves was isolated based on the described method [48] with some modifications. 200–300 mg leaf material was frozen in liquid nitrogen and ground in 2 ml Eppendorf tubes containing two small steel balls using the Retsch swinging mill MM 2000 (Retsch). Extraction buffer was added to pulverised leaf material and extracted twice with phenol. After centrifugation Na2Ac (3 M, pH 5.2) was added to the aqueous part and precipitated with Isopropanol. Finally, the pellet was washed twice with EtOH (80%) and digested in R40 (10 mm Tris-HCl: pH 8.0, 1 mm EDTA: pH 8.0 and 40 µg/ml RNAse A) for 3–6 h at 37°C. Isolation of genomic DNA from Arabidopsis thaliana was done based on the method described [49]. Capillary Southern and Northern blots as well as labeling, hybridization, washing, and autoradiographic exposures were performed as described [50]. Probes to detect ZmES-GFP transgene integrations were isolated from plasmids that have been used for maize transformations. DNA probes for radiographic detection were generated as described [50] or performed with Digoxigenin-11-dUTP (DIG) by PCR according to protocols provided by Roche Molecular Biochemicals. A GFP-specific probe was prepared as follows: the GFP sequence was amplified in a standard PCR using the primers NOS (5′-CAAGACCGGCAACAGGATTC-3′) and GFPF (5′-GACTATCTTCTTCAAGGATGAC-3′). A ZmES4-specific probe was prepared using the primers 200a (5′-CCCTTGGATTGGATTGGATCG-3′) and 200b (5′-GAAGTCTGTGGTG3′).
In order to study the expression of various potassium channels in germinated pollen tubes of maize, fresh pollen of the maize inbred line A188 was germinated in 30 drops of each 5 µl PGM (pollen germination medium [51]) for 40 min. Drops were transferred into a 1.5 ml centrifuge tube and collected by centrifugation at 100× g for 60 s and immediately frozen in liquid nitrogen. For cDNA synthesis mRNA was isolated using the Dynabeads® mRNA DIRECT™ Micro kit (Dynal® Biotech, Invitrogen) according to the manufacturers' instructions using 2-fold lysis-binding buffer. cDNA was synthesized using Transcriptor High Fidelity cDNA Synthesis Kit (Roche). mRNA annealed to magnetic Dynabeads was transferred using a PickPen™ (Bio-Nobile). Quality and amount of generated cDNAs were checked by PCR using intron-flanking GAPDH-specific primers GAPDHfor (5'-AGGGTGGTGCCAAGAAGGTTG-3') and GAPDHrev (5'-GTAGCCCCACTCGTTGTCGTA-3') as well as Actin 81/83-specific primers ZmAct 81/83fw (5'-GGTGATGGTGTGTCT-3') and ZmAct 81/83rev (5'-ACTGAGCACAATGTTAC-3'). Expression of four known maize K+-channels was analyzed using the following primer pairs: KZM1-fw3 (5'-CGAGCTATTCCATGCTCCTC-3') and KZM1-rv3 (5'-GGTCATGGCTGTTTCCTTGT-3') for KZM1, KZM2-LC-fw3 (5'-TCATGTATATCCACAGCAGAAC-3′) and KZM2-LC-re3 (5′-TGATGATATTGAGCTTCCGT-3') for KZM2, ZMK1-LC-fw (5'-ATAACAATGGGCATACAG-3') and ZMK1-LC-re (5'-TTCCGTCTTTCATTGAG-3') for ZMK1, as well as ZMK2-LC-fw (5'-TCCTCAGAAACCGCAC-3') and ZMK2-LC-re (5'-CGATCAACGCCTTCTC-3') for ZMK2. Gene bank accession numbers are indicated in the legend of Figure S6. RT-PCR was performed using SYBR® Advantage® qPCR Premix (Clontech) in an Eppendorf Mastercycler® ep realplex with initial denaturation for 45 s at 95°C and 40 cycles of amplification with 8 s denaturation at 95°C, 20 s of annealing at 59°C, and 15 s of extension at 72°C followed by melting curve analysis performed from 55–95°C.
With the exception of the bioassay experiments described below, Axiovert 35 M or Axiovert 200 inverted fluorescence microscopes (both Zeiss) with the filter set 01 (FITC filter with excitation at 450–490 nm; emission at >515 nm) or filter set 38 (GFP filter with excitation at 470–495 nm; emission at 525 nm) were used to observe GFP fluorescence in onion epidermal and maize ovule sections. A DAPI filter (Zeiss, excitation at 359–371 nm and emission >397 nm) was used to visualize DNA and cell wall material. Samples were excited with UV-light produced by a HBO 50/Ac lamp and images taken with a Nikon DS-5Mc camera (Nikon). Nikon software EclipseNet was used to obtain and merge fluorescence images. Confocal laser scanning microscopy was performed using the Leica TCS 4D CLSM (Leica-Laser-Technologie, Heidelberg, Germany). Samples were excited by 488 nm with an Argon laser and monitored as described [52].
In vitro pollination was performed as described earlier [12] with some modifications. Unfertilized maize ears (silk length of 5–10 cm) were harvested and, after removing all hull leaves, first cut longitudinally and then transversally to obtain segments of about 5 to 6 cm length. Segments were kept on wet lab paper in a Petri dish (15 cm in diameter) with the ovule containing part of the ear upwards. All silks were orientated towards one direction, cut to the same length, and pollinated with freshly harvested ACTp:GUS pollen. To avoid drying, segments were covered within the Petri dish with wet filter paper. Sections through maize ovules were isolated from ovaries 27–32 hap as well as 53–55 hap, which corresponds to ∼22 h and ∼46 h after fertilization, respectively. Further on, these sections were used for histochemical GUS-assays and have been incubated overnight at 37°C in staining buffer containing X-Gluc (5-bromo-4-chloro-3-indolyl-β-D-glucuronic acid CHA-salt) as described [17]. Stained samples were analyzed by light microscopy using an Axiovert 200 microscope (Zeiss). Documentation was done using a CAMEDIA C-4040 ZOOM digital camera (Olympus), and images were processed with Adobe Photoshop CS2 (Adobe Systems Inc.).
The mature predicted ZmES4 microprotein (61 amino acids) was chemically synthesized with 80%–-90% HPLC-purity by JPT Peptide Technologies GmbH. Successful synthesis of the linear peptide was shown by LC-MS. Intra-molecular disulfide-bridges were introduced via thermodynamically controlled folding. 0.5 mg of the ZmES4 protein was each first dissolved in DMSO and then in double distilled water. Double concentrated PGM [53] was added with a final concentration of DMSO not exceeding 2% and ZmES4 microprotein available in 1× PGM. Trypsin inhibitor from soybean (Sigma), RsAFP2, and LURE2 were each dissolved using the same method described above.
Fresh pollen grains of maize and Tripsacum dactyloides were shaken directly onto 10 µl drops of 1× PGM placed in a 35 m m plastic Petri dish. Pollen of tobacco (Nicotiana benthamiana and Nicotiana tabaccum) and lily (Hemerocallis fulva) was transferred directly from a fresh opened anther onto drops of 1× PGM using forceps. Germination was performed at room temperature for 20 to 120 min, depending on plant species analyzed. Germinated pollen tubes were subjected to biological assays only if germination rates exceeded 75%. During this assay, droplets were observed using an Eclipse TE2000 inverted microscope (Nikon) equipped with a 1.4 Megapixel digital AxioCam MRm camera and AxioVision digital image processing software (both Zeiss).
A 10 µl solution either of ZmES4/PGM or control CRPs/PGM was added to each droplet with germinated pollen using the CellTram® Air (Eppendorf) and a glass micropipette. A time series was started directly after protein addition and mixing, and pollen burst recorded for up to 120 min or stopped earlier when pollen tube burst was completed. In order to block potassium, sodium, or other cation channels, which may be targets of the ZmES4, pollen was germinated in 10 µl droplets of PGM for 15 min containing either 200 µM CsCl, MgCl2, BaCl2, GdCl3, or LaCl3. Then 10 µl ZmES4/PGM solution was added to the droplet and the number of bursted pollen tubes was determined as described above.
A functional analysis of maize K+ channels was performed by two-electrode voltage clamp technique following heterologous expression in Xenopus oocytes. Generation of cRNA and preparation of oocytes have been described previously [54]. Two-electrode voltage-clamp recordings were performed making use of a TURBO TEC 3× amplifier (NPI Electronic GmbH) integrated in an automated perfusion system developed for fast and efficient application of small amounts of drugs/solutions (ScreeningTool, NPI Electronic GmbH [55]). Voltage clamp recordings were performed in external solution containing 1 mM CaCl2, 1 mM MgCl2, 10 mM Tris-MES pH 7.4, 100 mM KCl, and 2% (w/v) DMSO. ZmES4 was applied at a final concentration of 150 µg/ml. Currents were elicited by voltage jumps in the range of −180 to +60 mV (10 mV increments) starting from a holding potential of −10 mV. To study the inhibition by the K+ channel specific inhibitor Cs+ [24], 5 mM CsCl was added to the standard solution. Following an activating voltage pulse of −130 mV, tail currents (It) at t = 0 were extracted from test pulses in the range of −180 to +60 mV (10 mV increments) and plotted against the applied voltage.
Impalement of pollen tube tips growing on ½ MS agar: growing pollen tubes were impaled right below the tip, using a micromanipulator (type 5171; Eppendorf) in combination with a piezo translator (P-280.30, Physik Instrumente). Micro-electrodes were pulled from borosilicate capillaries (OD 1 mm, i.d. 0.58 mm, with filament; Hilgenberg), filled with 300 mM KCl and connected by an Ag/AgCl half cell to the microelectrode amplifier (VF-102; Bio-Logic). Current-clamp measurements were recorded with PULSE software (HEKA Electronics). Data were low pass filtered at 300 Hz (8-pole Bessel-filter type 902; Frequency Devices) and sampled at 1 kHz. ZmES4 or control CRP containing liquid PGM was applied by a Drummond Nanoliter Injector (WPI).
Bacterial inoculations: 4- to 5-wk-old Arabidopsis plants were hand infiltrated with Pseudomonas syringae pv. tomato DC3000 suspensions with 105 cfu/ml. At different time points (0, 1, 3, and 5 days after inoculation [DAI]) bacterial growth was analyzed by macerating two leaf discs per replicate in 10 mM MgCl2 and plating serial dilutions of the bacterial suspension on Kings B medium supplemented with 50 mg/l Rifampicin and 50 mg/l Cycloheximide. Assessment of Peronospora parasitica sporulation: different Arabidopsis ecotypes were sown onto a mixture of three parts of seedling compost and one part of sand in aluminum test tube caps with a diameter of 2.8 cm. After stratification for 2 d at 4°C, seedlings were further cultivated at 18°C in a growth cabinet with a 9 h day/15 h night cycle. 7- to 8-d-old seedlings were spray inoculated with a conidial suspension of the P. parasitica isolate Emwa 1 [56] at concentrations of either 103 or 105 conidia/ml in distilled water. The seedlings were placed in a tray covered with a clear plastic lid to maintain approximately 100% relative humidity. The degree of colonization was monitored 10, 30, and 70 DAI. For the latter time point plantlets were picked individually into 9 cm pots containing potting compost and sand (3:1).
ZmES1-4 sequence data (Genbank accessions NP_001104950, NP_001106003, NP_001106035, and NP_001105620) were compiled and compared online with GenBank, SwissProt, PIR, and PRF databases with default BLAST algorithms [57] and aligned online by ClustalW (http://www.ebi.ac.uk/clustalw/) [58]. Protein alignments were drawn by GeneDoc version 2.6.02 [59]. Prediction of protein localization and processing sites was performed online using PSORT (http://psort.nibb.ac.jp), iPSORT (http://www.HypothesisCreator.net/iPSORT), and SignalP V2.0 (http://www.cbs.dtu.dk/services/SignalP-2.0). 3D-modeling was performed using Discovery Studio 1.5 software (Accelrys Limited) and constructs were generated and drawn using Clone Manager 6 (Sci-Ed Software).
|
10.1371/journal.pntd.0005035 | Noninvasive Diagnosis of Visceral Leishmaniasis: Development and Evaluation of Two Urine-Based Immunoassays for Detection of Leishmania donovani Infection in India | Visceral Leishmaniasis (VL), a severe parasitic disease, could be fatal if diagnosis and treatment is delayed. Post kala-azar dermal leishmaniasis (PKDL), a skin related outcome, is a potential reservoir for the spread of VL. Diagnostic tests available for VL such as tissue aspiration are invasive and painful although they are capable of evaluating the treatment response. Serological tests although less invasive than tissue aspiration are incompetent to assess cure. Parasitological examination of slit-skin smear along with the clinical symptoms is routinely used for diagnosis of PKDL. Therefore, a noninvasive test with acceptable sensitivity and competency, additionally, to decide cure would be an asset in disease management and control.
We describe here, the development of antibody-capture ELISA and field adaptable dipstick test as noninvasive diagnostic tools for VL and PKDL and as a test of cure in VL treatment. Sensitivity and specificity of urine-ELISA were 97.94% (95/97) and 100% (75/75) respectively, for VL. Importantly, dipstick test demonstrated 100% sensitivity (97/97) and specificity (75/75) in VL diagnosis. Degree of agreement of the two methods with tissue aspiration was 98.83% (κ = 0.97) and 100% (κ = 1), for ELISA and dipstick test, respectively. Both the tests had 100% positivity for PKDL (14/14) cases. ELISA and dipstick test illustrated treatment efficacy in about 90% (16/18) VL cases when eventually turned negative after six months of treatment.
ELISA and dipstick test found immensely effective for diagnosis of VL and PKDL through urine samples thus, may substitute the existing invasive diagnostics. Utility of these tests as indirect methods of monitoring parasite clearance can define infected versus cured. Urine-based dipstick test is simple, sensitive and above all noninvasive method that may help not only in active VL case detection but also to ascertain treatment response. It can therefore, be deployed widely for interventions in disease management of VL particularly in poor resource outskirts.
| Visceral leishmaniasis (VL), one of the most prevalent parasitic diseases in the developing world causes serious health concerns. Post kala-azar dermal leishmaniasis (PKDL) is a skin disease which occurs after treatment as a sequel to VL. Parasitological diagnosis involves invasive tissue aspiration which is tedious and painful. Commercially available immunochromatographic rapid diagnostic test such as rK39-RDT is used for field diagnosis of VL, detects antibodies in serum samples. Urine sample is however, much easier in collection, storage and handling than serum and would be a better alternative where collection of tissue aspirate or blood is impractical. In this study, we have developed and evaluated the performance of two urine-based diagnostic assays, ELISA and dipstick test, and compared the results with serological rK39-RDT. Our study shows the capability of urine-based tests in detecting anti-Leishmania antibodies effectively for both VL and PKDL diagnosis. The ability of dipstick test to demonstrate negative results after six months in 90% of the VL cases after treatment could be useful as a test of clinical cure. Urine-based tests can therefore replace the need for invasive practices and ensure better diagnosis under filed settings.
| Visceral Leishmaniasis (VL) is a vector-borne fatal infectious disease disseminated in 88 countries of the world, particularly in remote areas of India, Bangladesh, Sudan, Brazil, Ethiopia and South Sudan where 90% of the cases occur [1]. Around 200 to 400 thousands of new VL cases are reported every year globally [2]. Definitive diagnosis is important in the early phase of infection as drugs available for this disease have severe side effects [3]. Microscopic visualisation of spleen or bone marrow aspirates considered as a gold standard test for the confirmation of VL is conventionally used wherever feasible, although risky, painful and an invasive practice demanding expertise [4]. A molecular diagnosis like polymerase chain reaction (PCR) is constrained to research labs and tertiary hospitals only [5]. The presence of high levels of antibodies in the sera of VL patients was exploited for serological diagnosis using enzyme-linked immune sorbent assay (ELISA), direct agglutination test (DAT) and immunochromatographic rapid diagnostic test (RDT). The fact that ELISA and DAT are time taking, sophisticated and labour intensive, limits these methods for routine diagnosis in the majority of the VL endemic areas [6]. Antigen rk39 (39 amino acid kinesin-related protein) based RDTs are the most commonly used rapid test for sero-diagnosis of VL, especially in the Indian subcontinent where it gave sensitivity and specificity estimates of 97% and 90.2%, respectively [7]. However, their diagnostic performance varies from moderate in Latin America to even low in East African region [8]. Other than rK39, many newer recombinant antigens such as rKE16, rK28, and rKLO8 have been developed in the last decade for serological diagnosis [9–11]. Leishmania-specific antibodies remain within the infected blood for years after treatment. Thus serum-based antibody detection tests cannot serve as a test of cure [12].
To overcome the shortcomings of serum-based diagnostic tests several studies have been conducted to find a noninvasive biological source for VL diagnosis such as use of saliva and urine samples instead of serum [13]. For example, PCR have been found 88%-96.8% sensitive when conducted with DNA extracted from urine samples [14–16]. Saliva as diagnostic sample has been reported with 58.6%-82.5% and 83.3% sensitivity in rK39-RDTs and ELISA, respectively [17,18]. rK39-RDTs for sero-diagnostics were also tested with urine samples in several studies and found sensitivities and specificities of 96.40% and 66.2–100%, respectively [19]. Urine-based latex agglutination test (KAtex) for detecting leishmanial antigen in urine is hampered by low sensitivity (47%-95%) and obligation to boil the samples to increase the test specificity [20]. Apart from KAtex, several urine-based antigen detection assays have been reported in recent years with varying performances [21,22]. Very few antibody capture ELISA have been conducted for diagnosis of VL through urine samples but not in the Indian region [23,24].
Post kala-azar dermal leishmaniasis (PKDL) is a skin disease caused by the same Leishmania donovani parasite responsible for VL [25]. In VL endemic areas serology with clinical presentation such as macules, papules, plaques or nodules, suggest probable PKDL. However, confirmation of PKDL relies on parasitological examination of slit-skin smear (SSS) or biopsy [26].
In the present study, we aimed to develop and evaluate urine-based ELISA and dipstick test as noninvasive diagnostic tools and compared with serum-based commercially available rK39-RDT using parasitologically confirmed VL cases from geographically important endemic areas of India. We have also assessed the performance of ELISA and dipstick test for PKDL diagnosis through urine samples. To determine the presence of VL antibodies of urine in response to therapy, this study was extended to comprise the evaluation of urine-based ELISA and dipstick test after six months of treatment. Our results confirmed the potential of urine samples in ELISA and dipstick test for the noninvasive diagnosis of VL and PKDL and as a test of clinical cure for human VL.
A total of 186 participants were enrolled in the School of Tropical Medicine (STM), Kolkata, Rajendra Memorial Research Institute of Medical Sciences (RMRIMS), Patna, Shri Krishna Medical College and Hospital (SKMCH), Muzaffarpur and Indian Institute of Chemical Biology (IICB), Kolkata during March 2011 to July 2016. Urine samples of 97 parasitologically proven and rK39-RDT (InBios Int. Inc., USA) tested VL cases were obtained before the treatment started. Out of 97 VL samples, 18 were acquired for longitudinal study at different time points, before treatment began (day 0), when the treatment completed (day 30), and after six months since the first sample collected (day>180). Samples were also collected from 75 control subjects including 25 non-endemic healthy controls (NEHC) from IICB, 25 endemic healthy controls (EHC) from RMRIMS and STM and 25 symptomatically similar diseases other than VL (OD) from STM. Fourteen urine samples which were confirmed by skin biopsy or slit-skin smear were collected from PKDL patients. Immediately after collection of urine, sodium azide (NaN3) was added to each sample at a final concentration of 0.1% as a preservative. The samples were stored at 4°C refrigeration until use.
This study was approved by the Ethical Committee of Indian Institute of Chemical Biology, Kolkata and by the respective hospitals. Written informed consent in their local languages was taken from the participants or parents, if not adult prior to the study. The reason of collection and storage of urine samples for evaluation of newly developed immunoassays was clearly mentioned in the consent form.
L. donovani strain AG83 (ATCC PRA-413) parasite was cultured according to the standard protocol [27]. Recombinant proteins of strain AG83 namely, rGP63 (a 63 kDa recombinant glycoprotein) and rCPA (a recombinant cysteine protease) were purified from its clones [28,29]. Soluble leishmanial antigens (SLA) extracted from L. donovani were prepared by the following [30]. Leishmania promastigote membrane antigen (LAg) was extracted as described [31]. Antibody capture ELISA was performed using 96-well flat bottom ELISA plates (Nunc Maxisorp, Denmark). Briefly, 1 μg/well of LAg in phosphate buffer was coated, and plates were kept overnight at 4°C, afterwards blocked with 1% of bovine serum albumin (BSA) (Sigma, USA) in phosphate buffer saline (PBS) for 2 h at 37°C. Subsequently, 1:10 dilution of urine samples followed by 1:4000 diluted HRP-conjugated anti-human IgG (Bangalore GeNei, India) were applied and incubated at 37°C for 1 h. Plates were washed between each step with PBS containing Tween-20. Finally, 5 mg o-phenylenediamine dihydrochloride (OPD, Sigma, USA) and 0.05% H2O2 (Merck, Germany) in 10 ml phosphate-citrate buffer were added for 15 min at room temperature (RT) and the optical densities were read at 492 nm.
Nitrocellulose membrane (Hybond 0.45 μm, GE Healthcare, UK) was cut in the form of rectangular strips and soaked in 25 mM Tris-HCl buffer (pH 7.6). In a semi-dried condition, 1.5 μg of LAg in 2 μl Tris-HCl buffer were coated in the form of a dot. After complete drying, free areas of the strips were blocked with 5% of BSA + 0.1% Tween-20 and 0.01% NaN3 in 100 mM Tris Buffer Saline (TBS) and incubated at 4°C overnight. Next day, strips were washed thrice with 100 mM TBS and 0.05% Tween-20 (TBS-T) followed by drying at RT. Subsequently, urine samples at 1:5 dilution in TBS-T were incubated with the strips for 30 min at RT. After two washes in TBS-T, strips were incubated with 1:2000 diluted peroxidase-conjugated goat anti-human IgG for 30 min at RT. Following, two washes in TBS-T and a final wash in 100 mM TBS (without Tween-20), strips were incubated for 5 min in a freshly prepared substrate solution having 0.05% of 3, 3’-diaminobenzidine tetrahydrochloride (DAB, Sigma, USA) containing 0.05% of H2O2 in 100 mM TBS. The reaction was stopped by dipping the strips in distilled water for 2 min. Dark brown colour spot depicted anti-Leishmania reactivity of urine samples.
Nitrocellulose membrane-based dipstick was prepared to have a test and a control line for visual detection of the disease. In brief, 1.5 μg of LAg/dipstick at the test line and 1: 20 dilution of rabbit anti-human IgG at the control line were bound to the membrane and blocked with 5% BSA at 4°C overnight. Next day after drying, the membrane was adhered to a plastic sheet and stored at RT until the test. Like dot blot assay the dipstick test comprises incubation with urine sample followed by enzyme-conjugated anti-human IgG and then substrate chromogen including washing in each step. Dark brown coloured bands both at the test and control line show VL positivity and a single band at the control line infers VL negativity.
Statistical analyses were performed with Graph Pad Prism version 5.0. Two-tailed, Mann-Whitney U test was used for comparison of ELISA values of different groups and considered statistically significant if the P values<0.05. A receiver-operator characteristic (ROC) curve for ELISA was constructed to determine the cut-off value with 95% confidence intervals (CI) that discriminate between VL-positive and -negative urine samples. To assess the overall performance of the test sensitivity and specificity were calculated and diagnostic accuracy was established by area under the curve (AUC) where AUC = 1 indicates an accurate test. Kappa values (0.8–1, perfect agreement) were estimated to find the degree of agreement for urine-based ELISA and dipstick test with the reference test of tissue aspiration.
To assess Leishmania proteins as potential noninvasive diagnostic candidate antibody capture ELISA was performed as a proof of principle experiment with four leishmanial antigens, rGP63, rCPA, SLA and LAg. LAg showed better recognition of IgG antibodies in VL urine with high specificity compared to other tested antigens (Fig 1). Reactivity of LAg was also investigated against other antibody isotypes, IgA, IgM and IgE present in the urine samples. Presence of high levels of anti-leishmanial IgG antibodies in VL urine samples than non-VL controls distinguishes disease condition better over other antibody isotypes (Fig 2). Parameters such as urine dilution and antigen concentrations were standardised to set optimal ELISA condition for the diagnosis (S1 and S2 Figs). Finally, all urine samples were examined by ELISA and results were interpreted to find urine IgG reactivity of each group against LAg (Fig 3). With the assessment of ROC curve (Fig 4), the test cut-off point, 0.1875 was selected from possible cut-offs at 95% CI. Urine from VL patients illustrated significantly (P <0.0001) stronger recognition against LAg with a high degree of specificity. The test showed positivity in 95 of 97 VL cases, yielding a sensitivity of 97.94% for detection of Leishmania exposure. False negative results were found for 2 (2.06%) of 97 VL samples. All 14 PKDL samples showed antibody titres above the cut-off line thus 100% sensitive for urine ELISA. Overall specificity was calculated using 75 non-VL urine samples of controls and found 100% specificity without any false positives. Anti-LAg antibody titre in VL urine is 27, 12 and 16 times higher than those of NEHC, EHC and OD, respectively (Fig 3). The AUC obtained for the test was 0.9984, depicting its potential to discriminate cases with and without the disease and comprises a high degree of agreement (κ = 0.97) with the gold standard test. Additionally, two VL cases which were parasitologically confirmed but negative to rK39-RDT with serum were found positive in urine ELISA, suggesting their better performance in comparison to serum rK39-RDT.
To determine the urine antibody levels in response to treatment, we selected paired urine samples from 18 VL patients at day 0, day 30 and day >180. There is no significant decline in the titre just after completing the treatment (day 30). Out of 18 urine samples, 16 showed a significant decrease (p<0.0001) in the urine IgG titre only after six months (day > 180) since the treatment started (Fig 5).
Principles of ELISA were adapted in the form of dot blot to develop a nitrocellulose membrane based diagnostic test. Dot blot optimisation was performed with urine samples from different groups to see the qualitative reactivity of the antigen on the membrane. Antigen concentration of 1.5 μg/dot, urine dilution of 1:5 and blocking with 5% BSA were optimised for the assay (S3–S5 Figs). Under these conditions, 10 VL samples and 11 non-VL controls were used in the dot blot experiment. Urine from confirmed VL cases tested positive, and the non-VL controls were all negative in the assay as determined by the visual observation (Fig 6). Anti-human IgG was selected and optimised for the assay as an experimental control which show reactivity with urine IgG irrespective of their specificity (S6 Fig).
Dot blot assay was transformed into a field adaptable dipstick format as discussed in Methods Section for the qualitative detection of total IgG in human urine (S7 Fig). Results of the dipstick test showed positivity in all VL urine samples, acquiring a sensitivity of 100% (97/97). None of the urine samples taken from negative controls showed reactivity with the dipstick assay, resulting in 100% specificity (75/75) of the test. Thus, complete agreement (κ = 1.0) was observed between the dipstick test and the reference test of tissue aspiration for diagnosis of VL. In comparison to urine dipstick test, serum-based rK39-RDT showed 95.87% sensitivity (93/97) and 97.33% specificity (73/75). Two patients with confirmed VL, who were false negative in our ELISA and with serum rK39-RDT, demonstrated positive dipstick test showing its aptitude to detect even low antibody titre in urine. All 14 PKDL urine samples tested had positive results in dipstick assay thus showing 100% sensitivity of the test. The analytical performances of dipstick test and ELISA with rK39-RDT are compared in Table 1, and the representative illustrations of dipstick assay are shown in Fig 7.
Evaluation of dipsticks as a test of cure was conducted with 18 paired VL samples at day 0, 30 and >180 after treatment initiation. Samples after one month of therapy (day 30) showed a decrease in the positivity only in 5 samples as the test line of the dipstick had less intense colour than the samples using before treatment started (day 0). Out of 18 follow-up samples, 16 were entirely negative on and after six months of the treatment initiation (day >180). Two samples remained positive after 6 months of treatment till the data reported (S1 Table).
Noninvasive diagnosis of VL has been a challenge for long, and to date no accepted urine-based rapid test is available. WHO in their 3rd report on neglected tropical diseases in 2015 recommended the need of improved diagnostic tests for VL, PKDL and test of cure. We report here, ELISA and a field adaptable dipstick test which make use of urine samples instead of serum for diagnosis of VL as well as PKDL diseases. To the best of our knowledge, this is the first study where a dedicated dipstick test was developed for the detection of leishmanial antigen-specific antibodies from human urine samples and 100% sensitivity, and specificity was achieved. We have also reported an ELISA for the first time with urine of PKDL patients, where it gives 100% sensitivity. Additionally, ELISA and dipstick assay show clearance of the antibodies from 90% of the urine samples six months after start of treatment. Unlike the blood stream where antibodies persist for several years after treatment, our data show prospects in the use of these urine tests for monitoring prognosis following treatment.
Employing LAg as the antigen, the respective sensitivities and specificities were found to be 97.94% and 100% in ELISA, and 100% and 100%, in dipstick assay for VL. These findings are consistent with our earlier reports of diagnostic ability of LAg in detecting serum antibodies [32]. Single antigens so far were not found very beneficial for diagnosis of VL. LAg is a mixture of at least ten urine-reactive antigenic polypeptides present in the membrane of the parasite L. donovani. Reactivity of each polypeptide for urine antibodies gives synergistically better diagnostic performance than an antigen alone thus justifying its better diagnostic aptitude over cloned antigens.
Urine as diagnostic tool is a thrust in VL research. Several established serological methods such as ELISA, DAT and RDTs have been tested with urine samples in recent studies. For example, DAT when conducted with urine samples showed 90.7% and 96.4% sensitivity and specificity, respectively [33]. Attempts have been made in the past few years to identify protein biomarkers in urine during VL [34,35]. A latex agglutination test such as KAtex has suboptimal sensitivity in detecting urine antigen in VL diagnosis [20]. A prototype ELISA kit detecting urinary antigen was developed and evaluated in Ethiopia, Sudan, Bangladesh and Brazil. The overall sensitivities of the kit ranged from 88.4 to 100% with 100% specificity [21]. Leishmania infantum antigens which are excreted in urine of VL patients were also identified and cloned [36]. All VL urine samples were reported to be positive in capture ELISA when three leishmanial antigen assays were combined [22].
Antibody detection in urine could be beneficial for the noninvasive diagnosis of VL. However, only few were performed in this area. Islam et al in Bangladesh have investigated urine IgG using L. donovani crude antigen in ELISA and reported 93.5% sensitivity and 89.3% specificity [23]. The same group later showed 94% and 99.6% sensitivity and specificity, respectively with kinesin-related antigen rKRP42 [24]. In these studies VL follow-ups and PKDL patients were not included and the studies have been only at the ELISA level. Very recently, in Bangladesh Ghosh et al have observed only a marginal drop in the sensitivity and >96% specificity when urine samples were used and compared with serum by antibody-capture ELISA. In the study antigen rK28 demonstrated 95.4% sensitivity with urine than 98.9% with serum samples[37]. We have reported urine-ELISA for the first time with the Indian VL patients and the first to report for PKDL patients. Here the studies were conducted in three major centres of India which cover the key endemic areas of VL in this region.
There is no rapid test available for urine-antibody detection from VL and PKDL patients. Serum-based rK39-RDTs were tested with urine samples in a study in India and found very low specificity ranging from 66.2% to 77.08% only, thus not considered suitable for VL diagnosis in this format [38]. However, similar studies in Bangladesh and two other groups in India have shown sensitivities and specificities ranging from 95% to 100% and 86.33% to 100%, respectively [39–41]. Few studies also reported the reactivity of rK39-RDT with PKDL urine samples though the number of samples used was very few and the results were contradictory. One of our co-authors Goswami et al [40] showed PKDL positivity with rK39-RDT in six urine samples while Mohapatra et al [13] reported 3 out of 3 negative PKDL results with urine samples. We observed 100% urine positivity of PKDL cases (14/14) both in ELISA and dipstick tests. Diagnosis of PKDL is of utmost importance as it may harbour Leishmania parasites in the skin and provide another reservoir for Leishmania infection. However, PKDL is not a fatal infection and often evades VL surveillance programmes. Noninvasive diagnosis of PKDL can help in early identification of PKDL sufferers.
Except the invasive tissue aspiration, present serological diagnostic tests are not good enough for use as ‘test of cure’ [42]. Decline in antibody response against K39 antigen have been reported in VL patients but the time to become sero-negative varied [43]. A study showed antibodies against rK39 antigen remains unaltered in serum after 180 days of treatment. However, antibodies against rK26 and rK18 demonstrated significant decline in antibody titre of serum samples after 6 months of treatment [44]. Recently, urine antigen response after VL therapy has been investigated through ELISA. In the study, Leishmania Antigen Detect ELISA showed 100% negativity of urine sample at 180 days of treatment [21]. Antibody detection through serum-based rK39-RDT has been tested with urine of treated VL individuals in two separate studies. Urine samples of 40 VL follow-ups after 1–3 months and 11 urine samples just after treatment completion were positive with rK39-RDT thus demonstrating its failure as a test of cure [13,40]. We explored for the first time a significant (16/18) fall of urine antibody titre in ELISA after 180 days of treatment. Moreover, results of dipsticks tests which were positive before treatment turned completely negative (16/18) within six months of treatment. Positivity in two samples even after six months might be an indication of relapse in future and needs to be monitored for disease. Our results demonstrating the prognostic use of urine-based ELISA and dipstick assay warrant further detailed longitudinal studies.
Collection, storage and handling of urine samples are safe, noninvasive and advantageous over serum samples. The advantage of this is particularly to infants with Leishmania infection, from whom the collection of blood is difficult. Although, dried blood spot on filter paper has been proven successful in diagnosis and epidemiological studies of VL using DAT, however, DAT requires central laboratory facility [6,45]. Regarding urine dipstick test, it is a ready-to-use device which is stored at room temperature and can be used for at least one year with simple desiccation. It is also easier in handling and does not involve much expertise or sophisticated instruments, thus suitable for outreach diagnosis. The assay does not require pretreatment of the urine samples as in KAtex test. Urine samples were tested and found stable and reproducible up to more than six years when stored with 0.1% azide at 4°C [24].
This study, however, has limitations. The VL subjects chosen for the studies were clinically confirmed thus cannot give an idea about infection versus disease. Population-based study will be required to evaluate the performance of dipstick test with VL suspected individuals and asymptomatic carriers in defined VL endemic areas. Moreover, to compare treatment response, blood samples could not be collected from follow-up patients who came in the hospital outpatient department for routine checkup after at least 6 months of treatment. However, previous reports suggest persistence of anti-Leishmania antibodies up to six months of VL treatment [43,44]. This dipstick is currently based on enzyme-catalysed colorimetric reaction which takes 90 min to complete (S8 Fig). It could be reduced to 5 min with the existing gold-tagged lateral flow technology (already started).
Our findings highlight the potentials of urine-based ELISA and dipstick tests that can offer an efficient and convenient alternative to invasive diagnostics of VL as well as PKDL. Dipstick may help to overcome the need for invasive tissue aspiration particularly in primary health care centres where it is unlikely to be feasible.
Like in the Indian subcontinent, use of rK39-RDT with strict clinical case definition has marked a good impact in screening VL cases at the primary level. In East African regions where performance of rK39-RDT is not good this urine-based dipstick could be valuable in VL diagnosis at field settings. Moreover, dipstick test can also suggest the treatment response thus can be used as a test of cure. The collection of urine is comparatively easier and painless than withdrawing blood so it can help to screen Leishmania exposure at remote areas. Subsequently it can contribute in the control programs for VL management and eradication.
|
10.1371/journal.pcbi.1002187 | Heat Shock Partially Dissociates the Overlapping Modules of the Yeast Protein-Protein Interaction Network: A Systems Level Model of Adaptation | Network analysis became a powerful tool giving new insights to the understanding of cellular behavior. Heat shock, the archetype of stress responses, is a well-characterized and simple model of cellular dynamics. S. cerevisiae is an appropriate model organism, since both its protein-protein interaction network (interactome) and stress response at the gene expression level have been well characterized. However, the analysis of the reorganization of the yeast interactome during stress has not been investigated yet. We calculated the changes of the interaction-weights of the yeast interactome from the changes of mRNA expression levels upon heat shock. The major finding of our study is that heat shock induced a significant decrease in both the overlaps and connections of yeast interactome modules. In agreement with this the weighted diameter of the yeast interactome had a 4.9-fold increase in heat shock. Several key proteins of the heat shock response became centers of heat shock-induced local communities, as well as bridges providing a residual connection of modules after heat shock. The observed changes resemble to a ‘stratus-cumulus’ type transition of the interactome structure, since the unstressed yeast interactome had a globally connected organization, similar to that of stratus clouds, whereas the heat shocked interactome had a multifocal organization, similar to that of cumulus clouds. Our results showed that heat shock induces a partial disintegration of the global organization of the yeast interactome. This change may be rather general occurring in many types of stresses. Moreover, other complex systems, such as single proteins, social networks and ecosystems may also decrease their inter-modular links, thus develop more compact modules, and display a partial disintegration of their global structure in the initial phase of crisis. Thus, our work may provide a model of a general, system-level adaptation mechanism to environmental changes.
| In the last two decades our knowledge on stress-induced changes has been expanded rapidly. As a part of this work a large number of key proteins and biological processes of cellular adaptation to stress have been uncovered. However, we know relatively little on the systems level changes of the cell in stress. In our study we used the network approach to study the changes of the yeast protein-protein interaction network (interactome) in the archetype of stress, heat shock. The major finding of our study is that heat shock induced a marked decrease in the inter-community connections of the yeast interactome. The observed changes resembled to a ‘stratus-cumulus’ type transition of the interactome structure, since the unstressed yeast interactome had a globally connected organization, similar to that of stratus clouds, whereas the heat shocked interactome had a multifocal organization, similar to that of cumulus clouds. Our results indicated that heat shock induces a partial disintegration of the global protein-protein network structure of yeast cells. This change may be rather general occurring at the initial phase of crises in many complex systems, such as proteins in physical stretch, ecosystems in abrupt environmental changes or social networks in economic crisis.
| In the last decade due to the advance of high-throughput technologies system level inquiries became widespread. The network approach emerged as a versatile tool to assess the background of the regulation and changes of cellular functions. Analysis of protein-protein interaction (PPI) networks gives particularly rich system level information to understand the functional organization of living cells [1]–[6]. Determination of network modules (i.e. network groups, or communities) became a focal point of the analysis of network topology leading to more than a hundred independent methods to solve this challenging problem. In protein-protein interaction networks tight modules are corresponding to large protein complexes. However, more extensive, pervasively overlapping modules detected by recent methods, including ours, revealed a deeper insight to the multi-functionality of cellular proteins [7]–[9]. Despite of the widespread studies on network modules, the overlaps of interactome modules have not been studied yet in detail.
Network dynamics received an increasing attention in recent years. The stress response, inducing a genome-wide up- and down-regulation of gene expression after an abrupt environmental stimulus, is a particularly good model of the reorganization of cellular networks, where the observed changes have a paramount importance in survival, adaptation and evolution [10]–[13]. Yeast is an appropriate model organism for studying the system-level changes after stress, since we have an extensive knowledge on the organization of the yeast PPI network (interactome) [14]–[17], and stress-induced changes in the yeast gene expression pattern have also been studied in detail [18], [19]. Despite of major interest in key biological examples of network dynamics, changes of protein-protein interaction networks in stress have not been analyzed yet.
There are two main ways to integrate gene expression data with interactome, identifying active subnetworks [20]–[22] or analysing the whole interactome under genomic responses [15], [16], [23]. In the current study we used the latter approach and assessed the changes of the yeast interactome after the archetype of stress, heat shock. Upon heat shock the yeast PPI network became a much ‘larger world’: heat shock induced a close to 5-fold increase in the weighted diameter and a significant, but partial disintegration of the modular structure of the yeast interactome. The decrease of inter-modular protein-protein contacts may enable a ‘post heat shock’ re-integration of the yeast protein-protein interaction network communities, where the slightly different inter-modular contacts may provide a cost-efficient adaptation response to the changed environment.
To investigate the changes of the yeast interactome topology in heat shock, a well-characterized form of stress, we calculated the weight of each protein-protein interaction both in resting state and after heat shock. We used the physical protein-protein interaction subset of the BioGRID database [24], combining the benefits of this comprehensive, literature curated database with the more reliable, direct relationship of physical interactions. (We also extended our studies to a high-confidence PPI dataset, and found similar results as described in Materials and Methods.) Link weights of both basal state and heat shocked yeast cells were approximated using mRNA levels, since large-scale, complete datasets for protein abundances are currently missing (see Materials and Methods). We chose heat shock, as the form of stress we studied in detail, since it is considered to be a ‘severe stress’, where a good correlation between the transcriptome and the translatome has been demonstrated [25]. Interaction weights of the yeast PPI network were generated by averaging of the mRNA abundances of the two interacting proteins. Baseline and 15 min, 37°C heat shocked mRNA levels were obtained from the Holstege- [26] and Gasch-datasets [19], respectively, as described in the Materials and Methods section in detail.
The distribution of interaction weights showed a significant decrease upon heat shock (Figure S1 of Text S1; Wilcoxon paired test, p<2.2*10−16). To interpret this change we note, that the PPI networks of ‘resting’ and heat shocked yeast cells had the same links. However, the two interactomes had a largely different weight structure due to the differences in mRNA expression pattern upon heat shock. Table 1 shows a few main attributes of the interactome topology of unstressed and heat shocked yeast cells. In agreement with the significant change in weight distribution, the median weight of interactions had a 14% decrease in heat shock yeast cells. Interestingly, in unstressed yeast cells larger mRNA levels were predominantly associated with larger unweighted degrees, while in heat shocked yeast cells larger mRNA levels were predominantly associated with lower unweighted degrees. Thus, heat shock induces a shift of connection weights from hub-like proteins to non-hubs (see Figure S2 of Text S1), which may indicate a partial uncoupling of the local segments of yeast interactome upon heat shock.
The most remarkable change was the close to 5-fold (491%) increase of weighted diameter (Table 1). This was a rather suprising finding, which reflected that the interactome became a much ‘larger world’ after heat shock. The increase of weighted diameter was accompanied by shift in the distribution of weighted shortest path lengths (based on Dijkstra's algorithm [27]) towards longer paths, causing a significant difference (Wilcoxon paired test, p<2.2*10−16). Similarly to these findings, the average weighted shortest path length also showed a large increase (47.1 in unstressed versus 263.8 in heat shocked yeast cells). The distribution of ‘effective weighted degrees’ showed a scale-free like pattern, and a significant shift towards lower degrees after heat shock (Figure S3 of Text S1; Wilcoxon paired test, p<2.2*10−16). We note, that the ‘effective weighted degree’ captures the total number of fractional weighted connections of a node to another (see Materials and Methods and [8] for details). The shift towards lower weighted degrees was reflected by the decrease in both the median weighted degree and the number of hubs (14% and 22% decrease, respectively; Table 1).
The decrease of median interaction weights, median weighted degree and number of hubs indicated that heat shocked yeast cells developed a generally less intensive, ‘resource-sparing’ interactome. The ‘resource-sparing’ character is in agreement with the close to 5-fold increase of weighted diameter showing that the yeast interactome preferably ‘spares’ the shortcuts, and becomes much less integrated upon stress.
Visual inspection of stress-induced changes of the entire yeast interactome is of limited value, since the multitude of interactions makes the comparison difficult. However, there are comprehensible subnetworks allowing an easy, pair-wise assessment. We show the subnetworks of the strongest and weakest links on Figure 1. The subnetwork of strongest links (cf. Figure 1A and Figure 1B) of unstressed yeast cells contained a highly connected ribosomal protein complex (see Figure 1A, inset) and an additional center of carbohydrate metabolism (see Figure 1A, right bottom). Both centers are crucial for the fast cell divisions characteristic to unstressed yeast cells. Please note that the number of links is the same in both panels. Therefore, the link-density of the two major centers is much larger than the apparent density shown on Figure 1A. Upon heat shock several locally dense regions appeared, which were centered on heat-shock proteins (see circles on Figure 1B). This structure showed a re-organization of the interactome around proteins crucial in cell survival and recovery including dehydrogenases, proteins of glucose metabolism, a key player of protein degradation (polyubiquitin), as well as the molecular chaperones, Hsp70 and Hsp104 as detailed in the legend of Figure 1. The subnetwork of network-integrating weakest links [1]–[3], [6], [28] had a uniform link-density in basal state (Figure 1C). After heat shock a very densely connected twin-centre of nucleolar proteins emerged (see the right side of Figure 1D) responsible for rRNA processing and ribosome biogenesis (∼80 and ∼90% of genes by GO term, respectively; p<10−30 in both cases by hypergeometric test). This is in agreement with the key role of nucleolar protein complexes in cell survival [29]. In these representations the unstressed yeast interactome was closer to an organization resembling to the flat, dense, dark and low-lying stratus clouds, whereas the interactome after heat shock was closer to a multifocal structure, resembling to puffy and white cumulonimbus clouds. In former studies ‘stratus’ and ‘cumulus’ forms were described as alternative structures of the general form of yeast interactome [30]. Stratus- and cumulus-type organizations may be differing topology classes in many types of networks, such as in protein structure networks as we proposed recently [31].
In summary, the general network parameters suggested a partial disintegration of the interactome of heat shocked yeast cells represented by the large increase in weighted diameter (Table 1), and by the emergence of a cumulus-like global organization of the subnetworks of strongest and weakest links (Figure 1). Interestingly, metabolic networks of the symbiont, Buchnera aphidicola [32] and the free-living bacterium, Escherichia coli (Figure S4 of Text S1) displayed similar patterns like the interactomes of unstressed and heat shocked yeast cells. Metabolic pathways of B. aphidicola (Figure S4A of Text S1) showed a rather compact organization similar to a ‘stratus-type’ structure, whereas E. coli (Figure S4B of Text S1) had a more multifocal structure similar to a ‘cumulus-type’ network. The latter, cumulus-like structure may show that adaptation to a variable environment resulted in a multifocal pathway structure of E. coli, while the stratus-like structure of the B. aphidicola metabolism may be a consequence of a more stable environment. These assumptions are supported by the larger modularity of metabolic networks in organisms living in variable environment than that evolved under more constant conditions [33].
After our first results suggesting a partial disintegration of the yeast interactome in heat shock exemplified by the increased network weighted diameter and the emergence of a multifocal-like structure of the subnetworks of strongest and weakest links, we examined the heat shock-induced changes of yeast PPI network modules. For the determination of yeast interactome modules we used our recently developed ModuLand framework [8], since it detects pervasive overlaps like other recent methods [34], and therefore gives a more detailed description of PPI network modules than other modularization techniques [8], [34]. Moreover, the ModuLand method introduces community centrality, which is a measure of the overall influence of the whole network to one of its nodes or links. Community centrality enables an easy discrimination of module cores, containing the most central proteins of interactome modules, and makes the functional annotation of PPI network modules rather easy [8]. These modular cores are the hill-tops of the 3D representation of the interactome on Figure 2. On Figure 2 the horizontal plane corresponds to a conventional 2D network layout of the yeast interactome, while the vertical scale shows the community centrality value of yeast proteins. Functional annotations of the most central interactome modules are listed in Table S1 of Text S1 and Table S2 of Text S1. In the unstressed condition (Figure 2A) the central position was occupied by two ribosomal modules showing the overwhelming influence of protein synthesis on yeast cellular functions in exponentially growing yeast cells. Though this module pair was overlapping, their cores were different. Moreover, upon heat shock the two ribosomal modules showed different alterations. The third central module contained proteins of carbohydrate metabolism reflecting the importance of energy supply in yeast growth and proliferation. The additional modules recovered several modules identified before (e.g. the proteasome, ribosome biogenesis and the nuclear pore complex, see [8]). The larger functional diversity of the modules here than that obtained in our preliminary investigations using a much smaller, un-weighted dataset [8] showed the advantages of using a large dataset and interaction weights.
In contrast with the unstressed situation, the ribosomal modules had a much smaller community centrality upon heat shock (Figure 2B), which is in agreement with the inhibition of translation after heat shock. The carbohydrate metabolism module kept its central position (Table S1 of Text S1 and Table S2 of Text S1). A novel central module emerged containing proteins involved in the regulation of autophagy, a key process in cellular survival. Several other interactome communities also increased their community centrality, such as modules of heat shock proteins containing several major molecular chaperones and their co-chaperones (e.g.: Sti1, Hsp70, Hsp82 and Hsp104), which all play a key role in sequestering and refolding misfolded proteins after heat shock. Another module of growing centrality was the trehalose synthase module providing an important chemical chaperone for yeast survival (Table S2 of Text S1). Finally, a module of negative regulators of cellular processes (such as that of Bhm1 and Bhm2) also gained centrality (Table S2 of Text S1), exemplifying the energy-saving efforts of the yeast cell in heat shock. The more multifocal modular structure of the yeast cell after heat shock (Figure 2B) compared to the more centralized, compact modular structure of resting cells (Figure 2A) is in agreement with the partial disintegration of the yeast interactome suggested by the increasing weighted diameter (Table 1) and changes of subnetworks containing the strongest and weakest links (Figure 1).
To analyze the changes of yeast interactome modules after heat shock further, we compared the modular distribution of proteins in unstressed and heat shocked yeast cells. Figure 3A shows the cumulative distribution of the ‘effective number of modules’. The ‘effective number of modules’ measure efficiently captures the cumulative number of all modular fractions, where a protein belongs to (see Materials and Methods and [8] for details). After heat shock yeast proteins belonged to a significantly fewer number of interactome modules (Wilcoxon paired test, p<2.2*10−16). In other words this means that modules of the yeast interactome had a smaller overlap after heat shock than in the unstressed state, since there were less proteins belonging to multiple modules, i.e. modular overlaps.
Assessing the modular structure one level higher, where modules were treated as elements of a coarse-grained network [8], we compared the effective degree of modules of unstressed and heat shocked yeast cells (Figure 3B). The effective degree captures the total number of fractional weighted connections of a module to another (for details, see Materials and Methods). Upon heat shock interactome modules were connected to significantly smaller number of other modules (Mann-Whitney U test, p = 0.02299). Since a link between modules is related to the overlap between them ([8], for details see Materials and Methods), the decrease of inter-modular contacts upon heat shock reflects once again a smaller overlap between the interactome communities.
The decrease of modular overlap was similar in other stress conditions (e.g. in oxidative stress, reductive stress, osmotic stress, nutrient limitation, see Figure S5 of Text S1), although the heterogeneity of these conditions did not allow to create a coherent picture in every details. The partial decoupling of the interactome modules of stressed yeast cells (Figure 3) is in agreement with the increase of weighted network diameter (Table 1) and with the appearance of a larger multifocality in both the subnetworks of strongest and weakest links (Figure 1), as well as in the 3D image of modular structure (Figure 2). All these findings show a partial disintegration of the yeast interactome upon heat shock.
Prompted by our data showing a partial disintegration of the yeast interactome after heat shock, we became interested to assess those proteins, which preserve the residual integration of the interactome upon heat shock. First, we assessed the community centrality changes of yeast proteins after heat shock, since high community centrality values characterize those yeast proteins, which receive a large influence from others [8], and thus integrate the responses of the yeast interactome. As a second step, we studied the bridges, i.e. the inter-modular proteins playing a key role in the remaining connection of interactome modules after heat shock.
Figure 4 shows the comparison of the community centrality values [8] of yeast proteins before and after heat shock highlighting five markedly different behaviors. Group A proteins increased their community centrality upon heat shock, Groups B and C contain proteins, which had a continuously high community centrality, while those proteins, which decreased their community centrality are in Group D. Finally, Group E proteins had a continuously low community centrality. Table S3 of Text S1 lists the proteins of the various groups of Figure 4 with their name and functional annotation. Proteins increasing their community centrality (Group A) upon heat shock included major molecular chaperones sequestering, disaggregating and refolding misfolded proteins (Hsp42 and Hsp104), as well as stabilizing cellular membranes (Hsp12) [35]. Group A proteins were also involved in stress signaling and in stress response regulation (e. g. Psr2 phosphatase, Rsp5 ubiquitin ligase) [36], [37], in autophagy regulation (Tor1, Tor2), in the reorganization of the cytoskeleton (Las17 actin assembly factor) [38] and also in yeast carbohydrate metabolism (Glk1 glucokinase, Hxt6 and Hxt7 glucose transporters). These proteins were all heat shock proteins, since they showed increased mRNA expression upon heat shock. Yeast proteins with continuously high community centrality (Group B) included ubiquitin, a ribosome associated, constitutive form of Hsp70 and several key enzymes of carbohydrate metabolism. Proteins having a high, but decreasing importance upon heat shock (Group C) were constituents of the ribosome. Importantly, enzymes and proteins involved in pre-rRNA processing, thus in the synthesis of new ribosomes, showed a large decrease in their community centrality and formed a major part of Group D. These changes reflected the down-regulation of protein synthesis and cell proliferation, which are hallmarks of the heat shock response. Group E proteins with a continuously low importance included several proteins with yet unknown functions, which is understandable knowing the minor role of these proteins both in unstressed and heat shocked yeast cells.
In summary, chaperones, proteins of stress signaling and other heat shock proteins, redirecting yeast carbohydrate metabolism in heat-shock, became key players in the residual integration of yeast protein-protein interaction network after heat shock. On the contrary, those proteins, which had been major integrators of the non-stressed yeast interactome (such as proteins of the ribosome or ribosome synthesis) lost their integrating function, and contributed to the partial, modular disintegration of yeast interactome after heat shock.
Next, we selected Group A through C proteins as they had large community centrality value in heat shocked conditions, and examined their localization in the subnetwork of the yeast interactome containing the strongest links (Figure 5). Considering that Group A proteins had low community centrality values in unstressed condition, it is not surprising that only one of Group A protein was visible in the subnetwork containing the top 4% of strongest links (Figure 5A). Group A proteins (small→large community centrality) appeared as nodes having strongest links, and occupied rather dispersed locations after heat shock (Figure 5B). Group B proteins (large→large community centrality) were accumulated in one of the two alternative centers of the subnetwork in unstressed condition, and became more dispersed after heat shock (cf. Figure 5C and Figure 5D). Group C proteins (extra large→large community centrality) occupied the other alternative center, the dense core of the subnetwork in unstressed yeast cells, while, similarly to the other groups, they became more dispersed after heat shock (cf. Figure 5E and Figure 5F).
In summary, proteins with large community centralities had rather condensed positions in the interactomes of unstressed yeast cells, while they occupied more scattered, dispersed positions after heat shock. This reflects well the key role of the proteins with large community centralities to maintain the integration of the cumulus-type, multifocal interactome of heat shocked yeast cells at multiple positions.
As a first inquiry to assess the role of bridges in the maintenance of interactome integrity after heat shock, we highlight a group of four proteins (Table 2; Hsp42, Hsp70, Hsp104 and glycogen phosphorylase). These proteins, beyond their very remarkable increase in community centrality values, were the only proteins, which had a parallel increase in their modular overlap upon heat shock (where the latter was defined as the effective number of their modules, the measure used already in Figure 3A). We note that this behavior was peculiar, since the modular overlap had a general decrease after heat shock (see Figure 3). Therefore it was plausible to claim that the 4 proteins listed in Table 2 were not only central, but also behaved as bridges, connecting parts of the partially disintegrated interactome after heat shock. It is noteworthy that 3 out of the 4 proteins are molecular chaperones (Hsp42, Hsp70, Hsp104), while glycogen phosphorylase is a key enzyme of energy mobilization, a necessity in stress. This finding is in agreement with the results of previous studies and assumptions [39], [40].
As a second inquiry to study the role of bridges in the interactome of unstressed and heat shocked yeast cells, we examined changes of bridgeness of yeast proteins. Figure 6 plots the bridgeness of yeast proteins before and after heat shock. Bridgeness was defined as before [8], involving the smaller of the two modular assignments of a node in two adjacent modules summed up for every module pairs. This value is high, if the node belongs more equally to two adjacent modules in many cases, i.e. it behaves as a bridge between a single pair, or between multiple pairs of modules. Such bridging positions correspond to saddles between the ‘community-hills’ of the 3D interactome community landscape shown on Figure 2. Note that the bridgeness measure characterizes an inter-modular position of the node between adjacent modules, while the modular overlap measure reveals the simultaneous involvement of the node in multiple modules.
The highlighted zones of Figure 6 show that the importance of 9 bridges increased, that of 7 bridges remained fairly unchanged, while the importance of only 3 bridges decreased upon heat shock. The increase of the number of key bridging proteins upon heat shock shows the increased importance of a few interactome-intergating proteins after stress (a very strong tendency for a significant change, with p = 0.051 by Mann-Whitney U test, between the highlighted bridges of Figure 6 having a value larger than 10). The position of the 7 persistently high bridges and the 9 heat shock-induced bridges in the subnetwork of the yeast interactome containing the strongest links is shown on Figure S6 of Text S1. Bridges appeared in this subnetwork at a larger ratio (31% compared to 69% before and after heat shock, respectively), and were re-organized to more inter-modular positions in the interactome of the strongest links after heat shock (Figure S6 of Text S1). Name and function of key bridges are listed in Table S4 of Text S1. The 5 bridges present in both the unstressed condition and after heat shock in the strongly linked subnetwork were Srp1, Yef3, Smt3, Ubi4 and Med7, key proteins of nuclear transport, transcription, translation and protein degradation complexes, respectively. The 6 additional bridges appearing only after heat shock in the strongly linked subnetwork were Whi3, Rpn3, Rsp5, Cbk1, Hek2 and Srs2, key proteins of protein degradation, DNA repair, mRNA sequestration and metabolism, respectively: all essential processes for cell survival in stress.
In summary, a rather interesting, complex picture emerged on interactome changes of heat shocked yeast cells. On one hand, the interactome developed a decreased integrity apperaring at several hierarchical levels of the local to global topology. The most remarkable change of all these was the heat shock-induced partial uncoupling of interactome modules. On the other hand, the remaining inter-modular connections remained or became enforced by a few key proteins involved in cell survival.
The major findings of the current paper are the following: heat shock induces i.) an increase in the weighted diameter of yeast protein-protein interaction network (Table 1); ii.) subnetworks of strongest and weakest links as well as the modular structure show a more multifocal appearance upon heat shock (Figure 1 and Figure 2); iii.) modules became partially decoupled in heat shock (Figure 3); and finally, iv.) a few, selected, inter-modular proteins help the integration of the partially uncoupled interactome of heat shocked yeast cells (Figure 4, Figure 5 and Figure 6).
A minor part of our findings was rather obvious. As an example of this: it is more-less expected that many heat shock-induced proteins will have a larger community centrality, since they have an increased weight of their interactions (Figure S1 of Text S1), and therefore, may receive a larger influence of other interactome segments. However, the partial disintegration of the yeast interactome after heat shock is by far not an obvious consequence of heat shock-induced mRNA changes, but a highly non-trivial adaptation to stress at the system level. It is important to note, that this major finding, the partial disintegration of yeast interactome after heat shock, appeared at several levels on network topology. At the very local level, a significant decrease was observed in the weighted degrees upon heat shock (Table 1; Figure S3 of Text S1). At the mesoscopic level a remarkable and highly robust decrease of modular overlaps occurred (Figure 3). At the global scale, a close to 5-fold increase of the weighted network diameter was observed (Table 1.). All these changes point to the same direction and suggest that a more ‘sparing’ contact structure develops upon heat shock allowing a better isolation and discrimination of cellular functions. The heat shock-mediated isolation and discrimination of cellular functions is also reflected by the change in the structure of strongest links (cf. Figure 1A and Figure 1B), where a large number of disjunct network centres develop, and became connected by a few strong links after heat shock (Figure 1B), as opposed to a large density of strong links in a few centres in unstressed yeast cells (see Figure 1A, where the density is so large that it can not be readily visualized even in the magnified inset).
The observed findings were in a way indirect. Regretfully, no direct PPI network data exist for heat shocked cells, including yeast. Therefore, we had to calculate the yeast interactome weights after heat shock from mRNA data. As we noted earlier, this approach was justified by the finding that heat shock is a severe form of stress, where transcriptional and translational changes are better coupled [25]. Protein levels are also regulated by protein degradation. Though large-scale data on yeast protein half-lives exist [41], even these data cover only a part of the yeast genome, and their modification in heat shock is not known. Despite of these shortcomings of exact system level data in heat shock, the robustness of our major finding, the partial uncoupling of yeast interactome modules, suggests that the phenomenon we observed is a real, in vivo response of yeasts cells to heat shock.
The interactome modules of unstressed yeast cells defined in this paper correspond to the results of other modularization methods. When comparing our results with those obtained by the MCODE method [42] and of another method based on semantic similarity [43], the size of predicted complexes were different, but good functional matches could be identified. When we extended the comparison to methods detecting modules having a wide range of size, like the CNM method [44] or that of Mete et al. [45], besides some minor discrepances, nearly indentical modules were found having either a large size (like that of ribosomal assembly and maintenance) or a small size (like that of tRNA processing; data not shown). In a very interesting study Gavin et al. [14] defined core components and attachments of yeast protein complexes. Core components were constant parts, while attachments were more flexible, fluctuating parts of the protein complexes. Cores of several modules (see Table S2 of Text S1) were often highly similar to the core components Gavin et al. [14] (e.g. in case of the proteasome, mitochondrial translation or RNA polymerase complexes). Core proteins of the ribosome and carbohydrate metabolism were found to be in many attachment regions of Gavin et al. [14] (15 and 4 attachments as opposed to 0.2 and 0.8 cores on the average, respectively). This is in agreement with our current results showing that these proteins have an extremely high community centrality, i.e. accommodate a large influence of multiple interactome segments.
Our study provides the first detailed comparison of the interactome structure before and after heat shock. However, there were a few studies, which contained a part of this information directly, or indirectly. Valente and Cusick [16] mapped the modular structure of unstressed yeast cells, and (assuming that the structure is invariant) determined which modules are up- and downregulated under heat shock. They found several modules with similar functions to those of the unstressed cells detected in our study (e.g. ribosomes, proteasomes and complexes involved in cell cycle control, or the organization of the chromosome and cytoskeleton). The heat shock-induced changes were also similar, showing a high similarity of downregulated modules (e.g. those responsible for ribosomal function, or chromosome organization). The upregulated modules were partially consistent with our results (cell cycle control) with the exception of the proteasome and cytoskeleton organization complex. In these two exceptions we detected a central role of these two modules in the unstressed condition already, which made the detection of their further upregulation difficult. Another comparison arose from the study of Komurov and White [15], who identified static and dynamic modules. Very interestingly, modules that were found only in unstressed or heat shocked conditions in our study corresponded to their dynamic modules (regulation of intracellular pH, proteasome, ribosome biogenesis, trehalose biosynthesis). Wang and Chen [46] developed an integrated framework of gene expression profiles, genome-wide location data, protein-protein interactions and several databases to study the yeast stress response. Their study shows the system-level mechanism of the yeast stress response highlighting the major transcription factors of this process. The study complements ours describing stress-induced consequences at the systems level. The results of Wang and Chen [46] demonstrated a large degree of general similarity of various stress responses in yeast (among others showing that 136 out of 190 transcription factors are conserved in osmotic, oxidative and heat shock), which is in agreement with the similarity of interactome-level changes of network topology after various types of stresses we observed in yeast (Figure S5 of Text S1).
Our results may put the ‘stratus/cumulus debate’ [30], [47], [48] in the new contextual framework of cellular dynamics. Our findings showed that the unstressed yeast interactome resembles more to a stratus-type, whereas the heat shocked (stressed) interactome resembles more to a cumulus-type organization. This indicates that the stratus and cumulus interactome conformations may not be as antagonistic as thought before, and none of them may be a clear artifact. Our results suggest that both network conformations may occur in vivo, and may characterize different states of the organism. Regretfully no quantitative measures for this structural feature have been defined so far. This will be a subject of further interesting studies.
Our earlier surveys of the literature anticipated a stress-induced decrease in the number and weights of interactions, as well as the decoupling of network modules. Chaperones were hypothesized to play a major role in the coupling/decoupling processes, since they occupied a more central position during stress, and their occupation by damaged, misfolded proteins after heat shock led to a release of their former targets. This phenomenon was termed by us as ‘chaperone overload’ [39], [49]. Our recent results support these previous considerations. Moreover, the present findings considerably extend the earlier assumptions showing the details of the heat shock-induced partial disintegration of the yeast interactome.
What may be the reasons, which make a partial disintegration of the interactome an evolutionarily profitable response for yeast cells after heat shock? i.) The decreased number and weights of interactions may be regarded as parts of the energy saving mechanisms, which are crucial for survival. The specific decrease of inter-modular contacts may ‘slow down’ the information transfer of stressed cells, which is a further help to save energy. ii.) The increased weighted diameter and the partially decoupled modular structure of the interactome may localize harmful damages (e.g. free radicals, dysfunctional proteins), and thus may prevent the propagation of damage. iii.) Dissociation of modules may help the mediation of ‘intracellular conflicts’, e.g. opposing changes in protein abundance and dynamics in stress. iv.) The appearance of a more pronounced modular structure may allow a larger autonomy of the modules. This is beneficial, since more distinct functional units may work in a more specialized, more effective way, and at the same time may also explore a larger variety of different behavior, since in their exploratory behavior they are not restricted by other modules to the extent than before stress. The larger autonomy of modules increases both the efficiency and learning potential of the cell sparing additional energy.
The observed partial disintegration of the yeast interactome after heat shock is most probably only transient. The partial de-coupling of the interactome modules is presumably followed by a re-coupling after stress, which not only restores a part of the original, denser inter-modular connections, but may also build novel inter-modular contacts, giving a structural background to the adaptation of the novel situation [39], [40], [50], [51]. This brings a novel perspective to those proteins, which help to maintain the integration of the yeast interactome during heat shock, since some of these inter-modular proteins may play a role in the adaptive reconfiguration of PPI network as a response to the changed environment. The presence of 3 major chaperones among those 4 proteins, which increased their inter-modular overlap upon heat shock (Table 2), supports this assumption, since chaperones are well-known mediators of cellular adaptation in stress and during evolution [39], [49].
The decrease of modular overlap was similar in other stress conditions (e.g. in oxidative stress, reductive stress, osmotic stress, nutrient limitation; see Figure S5 of Text S1), although the heterogeneity of these conditions inhibited to create a coherent picture in every details. Prompted by the generality of stress-induced partial disintegration of the yeast protein-protein interaction network, and by the generality of the beneficial reasons behind these changes, we were interested to see, whether similar changes may occur in other complex systems. Bagrow et al. [52] showed that network failures of a model system cause the uncoupling of overlapping modules before the loss of global connectivity. A similarly modular, sequential disruption of (presumably inter-modular) links was observed, when single molecules of the giant protein, titin were pulled introducing a physical stress [53]. Bandyopadhyay et al. [54] showed that while protein complexes tend to be stable in response to DNA damage in a genetic network, genetic interactions between these complexes are reprogrammed. Similarly to the changes shown on Figure S4 of Text S1, the group of Uri Alon found that networks of organisms in variable environment are significantly more modular than networks that evolved under more constant conditions [33], [55]. These studies all revealed the stress-related dynamism of intermodular regions in other cellular contexts.
Looking at even broader analogies Tinker et al. [56] showed that food limitation causes a diversification and specialization of sea otters that greatly resembles to the changes of yeast interactome modules in stress. A similar increase of modularization (patchiness) was observed in increasingly arid environments suffering from a larger and larger drought stress [57]. A partial decoupling of social modules was also observed, when criminal networks faced increased prosecution [58]. A recent study detected a reorganization of brain network modules during the learning process [59]. As a far-fetched analogy, stress-induced psychological dissociation [60] may also be perceived as a partial decoupling of psychological modalities. The stress-induced uncoupling/recoupling cycle greatly resembles Dabrowski's psychological development theory of positive disintegration [61], as well as the Schumpeterian concept of “creative destruction” describing long-term socio-economic changes [62]. In agreement with this general picture, Brian Uzzi and co-workers [63] recently showed that brokers shift their link-structure of instant messaging from weak to strong ties under the initial phase of crisis-like events at the stock-exchange, which may reflect a partial de-coupling of weakly linked broker-network modules together with an increase of strong link-mediated intra-modular cohesion. Estrada et al. [64] proposed a model, where communicability and community structure of socio-economic networks are affected by external stress (e.g. by social agitation, or crisis). They showed that community overlaps diminished with the increase of stress. Increased modularity of the banking system may be a very efficient way to prevent the return and extension of the recent crisis in economy as pointed out recently by Haldane and May [65], and as applied by the Volcker Rule in the USA. These broad analogies are supported further by the previously proposed [31] generality of the two basic network conformations, the stratus- and cumulus-like network topology observed here before and after heat shock, respectively.
In summary, the major finding of our study was that heat shock i.) induces the increase in the weighted diameter of the yeast interactome; ii.) sets up multifocality in both subnetworks and modules of the yeast interactome, as well as iii.) contributes to the decoupling of the modules of the heat shocked yeast interactome. Parallel with these changes a few remaining inter-modular connections play an enhanced, prominent role in the residual integration of the yeast interactome. Our work may provide a model of a general, system-level adaptation mechanism to environmental changes.
The budding yeast (S. cerevisiae) PPI data were from the BioGRID dataset [24] (www.thebiogrid.com, 2.0.58 release), which is a freely accessible database of physical and genetic interactions. To avoid indirect interactions only the physical interactions of the database were used. These interactions (contained in the experimental system column of the database) included physical in vitro interactions such as biochemical activity-derived, co-crystal structure-related, far-Western, protein-peptide, protein-RNA, or reconstituted complex interactions, as well as physical in vivo (like) interactions, such as affinity capture mass spectrometry, affinity capture RNA, affinity capture Western, co-fractionation, co-localization, co-purification, fluorescence resonance energy transfer and two-hybrid interactions. The giant component of the obtained PPI network was used containing 5,223 nodes and 44,314 interactions. In the absence of reliable and large-scale weighted yeast protein-protein interaction data, network link weights were generated from mRNA microarray datasets as described later. We also analyzed the high-confidence PPI dataset of Ekman et al. [23], where the giant component of the network comprised 2,444 proteins and 6,271 interactions. These results were consistent with our presented findings (Figure S7A of Text S1), although the small scale of network and the nature of interactions (which were not restricted to physical interactions as our dataset), reduced the biological relevance of this latter analysis.
Yeast whole-genome mRNA expression datasets were from Holstege et al. [26] (called as the “Holstege-dataset”) as a reference dataset for the baseline, non-stressed yeast gene expression profile, and from Gasch et al. [19] (called as the “Gasch-dataset”) measuring relative expression profiles in various stress conditions. The Holstege-dataset contained data of 5,449, while the Gasch-dataset contained data of 6,152 yeast genes, respectively. From the Gasch-dataset we selected heat shock as the archetype of stress conditions. Besides being a widely examined form of stress, heat shock is considered as a “severe stress” by Halbeisen and Gerber [25], where a good correlation between translational and transcriptional changes have been found. We analyzed the ‘hs-1’ condition of the Gasch-dataset (15 minutes of 37°C heat shock), where broader time series were monitored than at ‘hs-2’ or other heat shock conditions (the stress condition names are the same as used by Gasch et al. [19]). We performed our analysis using longer durations of 37°C heat shock (40 and 80 minutes compared to that of the 15 minutes of the “hs-1” dataset, [19]). In line with the expectations, heat shock induced gene expression was less remarkable after 40 minutes and returned close to the baseline level after 80 minutes. Therefore we performed a detailed analysis only with the 15 minutes heat shock dataset. Importantly, our major finding, the decrease of modular overlaps after stress was robust, and persisted in all heat shock conditions tested. The decrease of modular overlap was similar in other stress conditions (e.g. in oxidative stress, reductive stress, osmotic stress, nutrient limitation, see Figure S5 of Text S1), although the specificity and heterogeneity of these conditions inhibited to create a coherent picture in every details.
Although logarithmic transformations are extensively applied in the literature, we used absolute expression values. The use of absolute expression values instead of logarithmic values was in part due to the technical difficulty that after the logarithmization step negative protein-protein interaction weights would also arose that could not be interpreted. Negative weights of the logarithmized mRNA data could be avoided applying a 1000-fold increase as a rescaling correction, which is appropriate all the more, since protein levels are roughly by this magnitude higher than the corresponding mRNA levels [66]. Using this methodology, we got similar major findings as those shown in the paper (Figure S7B of Text S1). However, due to the larger number of correction steps we did not pursue this approach in detail.
Weights of interactions in the PPI network were generated from the mRNA expression data in two steps. 1.) In the first step the baseline, non-stressed protein abundances were taken as the mRNA expression levels of the Holstege-dataset [26], then the baseline protein abundance values were multiplied by the relative mRNA changes of the Gasch-dataset [19], resulting in the approximated protein abundances after heat shock.
Since the Gasch-dataset contained only relative values, and therefore could not be used as a baseline-dataset, we had to use the Holstege-dataset to calculate the baseline weights of the PPI network. To check, whether our results are sensitive for baseline selection, we performed our analysis using another gene expression dataset, where time zero data were also provided [18]. This approach resulted in a similar decrease of modular overlaps (data not shown), showing that using two different datasets for mRNA abundances do not cause unexpected variability. Due to the greater ratio of missing data (∼14% in baseline data and ∼11% after heat shock) we did not prefer this dataset in detailed analyses. We also tried to use protein abundances instead of mRNA abundances for the unstressed condition [67], [68], but due to the large amount of missing data in these data sets (>50%) we have not pursued this approach further.
When using the mRNA changes as approximations of changes in protein levels, in agreement with Halbeisen and Gerber [25], we assumed that the mRNA expression data in heat shock correlate well with protein abundance. Missing expression data for proteins in the PPI network (436 nodes total in the baseline network, less than 9% in case of the Holstege-dataset, as well as 504 nodes total in the network after heat shock, less than 10% of the Gasch-dataset) were substituted by the median expression values (0.8 in case of the Holstege-dataset, and 0.9931 in case of the Gasch-dataset), where the median was selected instead of the mean, since the distributions also contained extreme values.
2.) In the second step link-weights of the PPI network were generated by averaging of the abundances of the two proteins linked. We also tried multiplication instead of averaging that gave very similar results and provided sufficiently robust data in case of the smaller, high-confidence PPI dataset of Ekman et al. [23] (see Figure S7A of Text S1). However, we rejected this approach in case of the BioGRID dataset, as in case of this much larger dataset it resulted in a ‘rougher’ community landscape with more extreme changes of community centralities than averaging, which has been generally used in calculation of our data.
The use of an unweighted baseline PPI network resulted in much less consistent data due to the large difference between the homogeneity of the unweighted baseline and the inhomogeneity of the weighted heat shocked PPI networks. The physical meaning of heat shock-induced changes in gene expression is encoded precisely by the changes of link weights at the network level. This assumption makes it understandable that an unweigthed network gave false positive results in important parts of the analysis. This has two major reasons. On one hand, community centrality values are largely affected by the density of interactions. Therefore, in an unweigthed network, proteins having a high link density in their neighborhood would result in high community centrality values independently from their expression level. On the other hand, the metrics used in the analysis (e.g. overlap as the effective number of modules) are sensitive measures of fine topological changes, therefore they were largely different in the unweighted, homogenous interactome as compared to the weighted, heterogeneous interactome.
In principle, ‘relative changes’ of mRNA expressions could also be used for comparison (where a, say, 4-fold increase in mRNA expression of a given gene can be split to a 2-fold decrease of its baseline abundance and a 2-fold increase of its abundance after stress corresponding to the abundances of the same protein in resting and stressed yeast cells, respectively). However the use of these ‘relative changes’ of mRNA expression resulted in a large variability of the baseline PPI network weights (Figure S8 of Text S1). The method using the average of protein abundance values as interaction weights, we described above, gave a reliable probabilistic model, since the more abundance the associated proteins had, the more possible they interacted, and the more weight of their PPI network link possessed. Moreover, by considering the baseline expression rates, we received a more exact description of the importance of proteins in the yeast cell in both baseline and stressed conditions.
Yeast PPI network modules were determined using the NodeLand influence function calculation algorithm with the ProportionalHill module membership assignment method of the ModuLand module determination method family described by the authors' lab recently [8]. During the post-processing of the module assignment no merging of primary modules was applied. The ModuLand method determines extensively overlapping network modules by assigning proteins to multiple modules, which reflects well the functional diversity of proteins. The ModuLand method constructs a community landscape, where the landscape height of a protein corresponds to a community centrality value showing the influence of the whole PPI network to the given protein, thus the importance of the appropriate protein in the whole yeast interactome. In fact, community centrality is a summarized value, where in the first step of the method (currently: the NodeLand influence function calculation algorithm) all increments of the influence of other proteins to the given protein are summed up. In the second step of the calculation process (currently: the ProportionalHill modules membership assignment method) proteins with locally high community centrality (corresponding to ‘hills’ of the community landscape, see the 3D image of Figure 2) form the core of a module of the interactome. Individual proteins are characterized by their membership assignment strength to all interactome modules. (Usually one or a few of the modules are the ones, where the protein belongs the most, while all the other modules contain the protein only marginally). With the ModuLand framework the functional annotation of modules becomes rather easy, since it can be derived from the functions of the ‘core proteins’ having the largest community centrality in the module. In the current work core proteins of a given module were determined as the 5 proteins having the maximal community centrality (the number of core proteins has been extended to 8 in some exceptional cases, where indicated). Comparison of the functions of proteins with lower community centralities than that of the core proteins did not change the consensus of functional annotation of modules ([8] and Table S1 of Text S1).
The effective degree of nodes and modules, as well as the effective number of modules were calculated as described earlier [8]. All effective numbers refer to a set of data, where the sum is not calculated as a discrete measure, but as a continuous measure taking into account the weighted values of the data. The effective numbers were calculated using the subsequent equation: , where data were in set V, V[i] was the value of data i, and . The dataset, V contained i.) in case of the effective degree of nodes the weights of the interactions of the given node as defined earlier; ii.) in case of the effective degree of modules the weights of the links of the given module to all neighboring modules as defined here later; and iii.) in case of the effective number of modules the module membership assignment strengths of the given node to all modules of the yeast interactome. The weight of the link between modules i and j was the sum of the node-wise calculated overlap values Oij(n): , where Oij(n) was proportional to the module membership assignment strengths Hi(n) and Hj(n), and was normalized to the community centrality as: , where c(n) was the community centrality of node n, and the factor 2 referred to that both directions between the modules have been taken into account.
For the functional categorization of yeast PPI network modules (see Table S1 of Text S1 and Table S2 of Text S1), the Gene Ontology (GO) term, biological process [69] (http://www.yeastgenome.org/cgi-bin/GO/goTermFinder.pl) of the core modular proteins (as defined above) were compared. A modular GO term was assigned, if the core proteins shared a significant (p<0.01) amount of their GO terms. GO terms of only the most central modules were identified, since they were supposed to have a relevant role in cellular functions. The threshold was applied by the community centrality values of the most central proteins of modules (where community centrality values were greater, than 500), and this resulted in 15 or 14 modules for the unstressed or heat shocked conditions, respectively. In those exceptional cases, when the 5 core modular proteins did not result in a meaningful functional assignment (in case of 5 modules representing 17% of the 29 modules total), we extended the core-set to 8 proteins. Only 2 modules (representing 7% of the 29 modules total) were found, where none of these definitions resulted in any common assignment.
For the statistical evaluation of data the non parametric statistical tests of the Mann-Whitney U test and the Wilcoxon paired test were applied using the R-statistical program (https://www.r-project.org) as described in the actual experiments. The hypergeometric test was performed as provided by the Gene Ontology Term Finder: http://www.yeastgenome.org/cgi-bin/GO/goTermFinder.pl.
|
10.1371/journal.pntd.0001610 | KSAC, a Defined Leishmania Antigen, plus Adjuvant Protects against the Virulence of L. major Transmitted by Its Natural Vector Phlebotomus duboscqi | Recombinant KSAC and L110f are promising Leishmania vaccine candidates. Both antigens formulated in stable emulsions (SE) with the natural TLR4 agonist MPL® and L110f with the synthetic TLR4 agonist GLA in SE protected BALB/c mice against L. major infection following needle challenge. Considering the virulence of vector-transmitted Leishmania infections, we vaccinated BALB/c mice with either KSAC+GLA-SE or L110f+GLA-SE to assess protection against L. major transmitted via its vector Phlebotomus duboscqi.
Mice receiving the KSAC or L110f vaccines were challenged by needle or L. major-infected sand flies. Weekly disease progression and terminal parasite loads were determined. Immunological responses to KSAC, L110f, or soluble Leishmania antigen (SLA) were assessed throughout vaccination, three and twelve weeks after immunization, and one week post-challenge.
Following sand fly challenge, KSAC-vaccinated mice were protected while L110f-vaccinated animals showed partial protection. Protection correlated with the ability of SLA to induce IFN-γ-producing CD4+CD62LlowCCR7low effector memory T cells pre- and post-sand fly challenge.
This study demonstrates the protective efficacy of KSAC+GLA-SE against sand fly challenge; the importance of vector-transmitted challenge in evaluating vaccine candidates against Leishmania infection; and the necessity of a rapid potent Th1 response against Leishmania to attain true protection.
| Leishmaniasis is a neglected disease caused by the Leishmania parasite and transmitted by the bite of an infective sand fly. Despite the importance of this disease there is no vaccine available for humans. Studies have shown that vector-transmitted infections are more virulent, promoting parasite establishment and abrogating protection observed against needle-injected parasites in vaccinated mice. KSAC and L110f, derived from Leishmania-based polyproteins, protected mice against the needle-injected parasites. Here, we tested the two molecules for their capacity to protect mice against cutaneous leishmaniasis transmitted by an infective sand fly. Our results show that KSAC, but not L110f, confers protection against Leishmania transmitted by sand fly bites where protection was correlated to a strong immune response to Leishmania antigens by memory T cells before and after sand fly transmission of the parasite. This is the first report of a Leishmania-based vaccine that confers protection against a virulent sand fly challenge. Our results support the importance of screening Leishmania vaccine candidates using infective sand flies before moving forward with the costly steps of vaccine development.
| Leishmaniasis is a neglected disease endemic in 98 countries with an estimated 350 million people at risk and an estimated burden of 2,357,000 disability-adjusted life years [1]. Visceral leishmaniasis is fatal if left untreated, and the morbidity and stigma caused by cutaneous leishmaniasis is significant [2]. Current treatment is dependent on long-term therapy with toxic drugs, most requiring parenteral administration and hospital supervision.
A vaccine against leishmaniasis is feasible because infection with certain species, including L. major, or exposure to live Leishmania (leishmanization) leads to a long-term protection in humans [3], [4],[5],[6],[7]. Unfortunately, there is no commercial vaccine available for humans despite the presence of an extensive list of vaccine candidates shown to be protective in various animal models [8]. With the exception of two vaccine candidates, a synthetic glycovaccine [9] and autoclaved L. major+CPG [10], all Leishmania vaccines tested to date were challenged with needle inoculation of the Leishmania parasite. L110f and KSAC, two fusion polyproteins, in various combinations with appropriate adjuvants were shown to confer strong protection against cutaneous and visceral leishmaniasis in mice following conventional needle challenge [11], [12]. None of these vaccines were challenged by infected sand fly bites, the natural route of transmission. For protection against L. major, the L110f and KSAC-containing vaccines were tested separately in susceptible BALB/c mice followed by an infected sand fly challenge.
Both susceptible and resistant mice strains have been used to study the immunology of leishmaniasis and the protective effect of potential Leishmania vaccine candidates [13], [14], [15]. It has been long established that protection from Leishmania parasites requires the induction of a Th1 immune response [16], [17], [18]. BALB/c mice produce a polarized Th2 type immune response against Leishmania spp. and are used extensively to test Leishmania antigens [19]. It has been hypothesized that protective antigen/adjuvant formulations in this model system are good vaccine candidates since they have to overcome the natural Th2 bias of this strain.
Recently, Peters et al. [20] demonstrated that transmission of Leishmania parasites by sand fly bites generates a specific innate immune response involving a sustained recruitment of neutrophils that promotes parasite establishment. Additionally, the authors demonstrated that vector transmission of Leishmania parasites can abolish protection observed in vaccinated mice following needle challenge [10].
In the current work, we use a natural sand fly challenge model in BALB/c mice to test the immunogenicity and protective efficacy of the two fusion proteins L110f and KSAC formulated with GLA-SE against L. major transmitted by the bite of its natural sand fly vector Phlebotomus duboscqi.
We used 6 to 8 week old female BALB/c mice (Charles River Laboratories Inc). Phlebotomus duboscqi sand flies, Mali strain, were reared at the LMVR, NIAID, NIH.
All animal experimental procedures were reviewed and approved by the National Institute of Allergy and Infectious Diseases Animal Care and Use Committee under animal protocol LMVR4E. The NIAID DIR Animal Care and Use program complies with The Guide for the Care and Use of Laboratory Animals and with the NIH OACU ARAC guidelines.
We used the L. major V1 (MHOM/IL/80/Friedlin) strain for all sand fly infections apart from that shown in supporting information (Figure S1). In Figure S1, we used the WR 2885 strain, a recent field isolate that originated in Iraq and was typed at the Walter Reed Army Institute of Research. Washed amastigotes were counted and added to the blood meal for sand fly infection or placed directly in culture for the generation of metacyclics for needle challenge.
KSAC and L110f recombinant proteins were prepared as previously described [11], [12], [21] and mixed at the time of injection with a stable emulsion (SE) formulation of the pure, synthetic hexa-acylated TLR4 agonist glucopyranosyl lipid A (GLA) [22]. More than one lot of each antigen was used in experiments with a residual endotoxin content ranging from 77 to 245 EU/mg protein for L110f and from <0.05 to 27 EU/mg protein for KSAC.
Mice were vaccinated subcutaneously (s.c.) in the base of tail, three times at three week intervals with 100 µl containing 10 µg of antigen (KSAC or L110f) formulated with 20 µg of GLA-SE or with GLA-SE alone.
Blood containing 3×106 L. major amastigotes/ml was used to artificially feed sand flies as previously described [23]. Sand flies were used for transmission 13–14 days post-Leishmania infection. Three weeks (early challenge) or 12 weeks (delayed challenge) after the last immunization, 10 infected sand flies were applied to a single mouse ear for 2 hours in the dark.
Vaccinated animals were injected intradermally in the right ear with 2×103 purified L. major metacyclics in 10 µl PBS using a 27-gauge needle.
The thickness of ear lesions was recorded on a weekly basis using a vernier caliper (Mitutoyo Corp.).
Parasite quantification was performed using JW11 and JW12 Leishmania-specific primers [24] as well as the 18S primers to amplify a housekeeping gene as previously described [25]. Expression levels were normalized to 18S DNA and corrected for the weight of the whole ear.
Subclass (IgG1 and IgG2a) responses were measured by ELISA using Immulon4-Thermo plates coated overnight at 4°C with L110f or KSAC (2 µg/ml). Diluted sera (1/100) were incubated for 1 hour at 37°C. After washing, plates were incubated with alkaline phosphatase-conjugated anti-mouse IgG1 or IgG2a antibodies (BD Biosciences, San Jose, CA) (1/1000). The plates were developed using alkaline phosphatase substrate (SIGMA). The reaction was recorded after 10 minutes at 405 nm.
Three weeks after the last immunization or one week post-challenge with infected sand fly bites, spleen cells were obtained as previously described [11] and stimulated with soluble Leishmania major antigen (SLA,100 µg/ml), KSAC (10 µg/ml) or L110f (10 µg/ml). Supernatants were collected 72 hours after incubation to evaluate cytokine production (IFN-γ, IL-10 and IL-4) by ELISA (BD Biosciences, San Diego, CA) according to the manufacturer's protocol.
Three weeks after the last immunization or one week post-challenge with infected sand fly bites, 2×106 splenocytes from individual mice were cultured in complete RPMI medium in flat-bottom 48-well plates with or without SLA (100 µg/mL), KSAC (20 µg/ml) or L110f (20 µg/ml) at 37°C in 5% CO2 for 18 h. Cells were incubated with Brefeldin A (BD Golgi Plug; BD Pharmingen) during the last 4 h of culture, washed with PBS, and blocked with anti-CD16/CD32 (BD Fc block, 2.4G2; BD Pharmingen) for 30 minutes at 4°C. Cells were stained with the fluorochrome-conjugated antibodies (BD Pharmingen and eBiosciences) PerCP-labeled anti-CD4 (RM4-5), APC-labeled anti-TCR-β (H57-597), PECy7-labeled anti-CD62L (MEL-14), and PE-labeled anti-CCR7 (4B12) for 30 minutes at 4°C, washed twice, fixed and permeabilized with Cytofix/Cytoperm Plus (BD Pharmingen), and stained with FITC-labeled anti-IFN-γ (XMG 1.2). A minimum of 100,000 cells were acquired using a FACS Calibur flow cytometer (BD Biosciences) and analyzed with the Flow Jo software (Tree Star, Inc., Oregon).
A two-tailed unpaired Student's t-test was used for statistical analysis using the GraphPad software (GraphPad Software Inc.). P values of 0.05 or less were considered significant.
Immunization of mice with KSAC+GLA-SE induced a robust antibody response following the first immunization and afterwards maintaining a positive IgG2a∶IgG1 ratio (Fig. 1 A, B). In contrast, L110f+GLA-SE induced a weaker overall antibody response that was biased towards IgG1 antibody production (Fig. 1A, B).
Mice vaccinated with KSAC+GLA-SE or L110f+GLA-SE controlled a needle-challenge infection for at least eight weeks, whereas mice receiving GLA-SE alone did not (P<0.001, Figure 2A). Animals immunized with either vaccine also controlled L. major infection post-sand fly challenge through 6 weeks, although the level of protection with the L110f-containing vaccine was reproducibly lower than that for the KSAC-containing vaccine (Figure 2C). KSAC+GLA-SE, but not L110f+GLA-SLE immunized mice displayed significant protection up to the final 8 week time point (Figure 2C). The parasite burden assessed eight weeks after challenge with either needle or infected sand flies supports the pathology data (Figure 2B, D). Following needle challenge, the parasite number was significantly decreased in KSAC+GLA-SE (P<0.01) and L110f+GLA-SE (P<0.05) compared to GLA-SE immunized mice (Figure 2B). Following sand fly-challenge, only KSAC+GLA-SE immunized mice showed a significant reduction of parasite number (P<0.01) compared to GLA-SE immunized mice (Figure 2D). The number of parasites in L110f+GLA-SE immunized mice were intermediate, but showed no significant difference from controls (Figure 2D).
KSAC+GLA-SE vaccination induced the production of KSAC-specific IFN-γ+ CD4+ T cells, and the relative size of this cell fraction was maintained after a challenge with L. major-infected sand flies (Fig. 3A). In contrast, CD4+ T cells of mice immunized with L110f+GLA-SE and stimulated with L110f produced IFN-γ only after the mice were challenged with infected flies (Figure 3B). Importantly, CD4+ T cells from KSAC+GLA-SE-immunized mice produced IFN-γ following stimulation with SLA, while cells from animals immunized with L110f+GLA-SE were non-responsive (Figure 3C). Before sand fly challenge the percentage of CD4+CD62LlowCCR7low effector memory T cells producing IFN-γ was greater in KSAC-immunized mice stimulated ex vivo with KSAC compared to GLA-SE alone or L110f-immunized mice stimulated with L110f (Figure 3D, KSAC and L110f panels). Of note, the percentage of L110f-specific effector memory IFN-γ+CD4+ T cells from L110f+GLA-SE -immunized mice was greater after sand fly challenge than either the KSAC+GLA-SE or GLA-SE control groups stimulated with the appropriate antigens. Nevertheless, SLA induced a greater percentage of IFN-γ+ effector memory T cells from KSAC+GLA-SE-immunized mice than from either L110f+GLA-SE or the control GLA-SE groups before and after sand fly challenge (Fig. 3D, SLA panel).
Mice vaccinated with KSAC+GLA-SE produced a high level of antigen-specific IFN-γ following ex vivo stimulation with SLA or KSAC both pre- and post-challenge with infected flies (Figure 4A). Prior to infected sand fly challenge, IFN-γ production was 23 ng/ml and 62 ng/ml following stimulation with SLA and KSAC, respectively (Figure 4A). This increased 8 and 3 fold when tested one week post-challenge, reaching 194 ng/ml and 212 ng/ml following stimulation with SLA and KSAC, respectively (Figure 4A). A distinctly different response was observed in L110f+GLA-SE-immunized mice where pre-challenge IFN-γ production was low to undetectable after stimulation with either SLA or L110f (Figure 4A), whereas IFN-γ was readily detectable after challenge with infected flies. Post-challenge, 57 ng/ml of IFN-γ and 222 ng/ml were produced in L110f-immunized mice in response to SLA and L110f, respectively (Figure 4A). Overall, antigen-stimulated splenocytes produced a low level of IL-10 compared to the relatively high concentration of IFN-γ produced (Figure 4A, B). The highest levels of IL-10 were produced by spleen cells of KSAC-immunized mice pre-challenge with infected flies (>3.3 ng/ml and 1.7 ng/ml with SLA and KSAC stimulation, respectively) compared to post-challenge levels (Figure 4A); post-challenge levels were reduced to <1 ng/ml following stimulation with either SLA or KSAC (Figure 4A). Similar to IFN-γ, IL-10 was not detectable in L110f+GLA-SE immunized mice pre-challenge and produced 1.2 and 2.4 ng/ml IL-10 after stimulation with SLA and L110f, respectively, after sand fly challenge (Figure 4A). The ratio of IFN-γ∶IL-10 highlight the dominance of IFN-γ over IL-10 in mice vaccinated with either KSAC+GLA-SE or L110f+GLA-SE and the lack of cytokine production in response to immunization with GLA-SE alone (Figure 4C).
Mice vaccinated with KSAC+GLA-SE maintained a positive KSAC-specific IgG2a∶IgG1 ratio for at least 12 weeks after the final vaccination (Fig. 5A). These mice were protected against a delayed challenge with L. major-infected sand flies, maintaining significantly smaller lesions (P<0.05 to <0.001) compared to mice vaccinated with GLA-SE alone (Figure 5B). In this delayed challenge, mice vaccinated with L110f+GLA-SE exhibited an almost equal IgG2a∶IgG1 ratio lower than that of KSAC+GLA-SE vaccinated mice (Figure 5A), possibly reflecting a slower and less dramatic shift to a Th1 response. These mice were only partially protected against L. major transmitted by infected sand flies (Figure 5B). The long-term protection conferred by vaccination with KSAC is reflected by the absence of disease pathology up to five weeks post-challenge compared to GLA-SE-vaccinated mice and to a lesser degree to the partially protected L110f+GLA-SE-vaccinated mice (Figure 5B panels).
To assess the relevance of the parasite strain used in challenge, we carried out a preliminary experiment using a recent field isolate of L. major (WR-2885), a strain kindly provided to us by Dr. Edgar Rowton, Walter Reed Army Institute of Research, Washington, D.C., to test the level of protection provided by KSAC or L110f vaccines. KSAC-immunized mice were protected while L110f-immuinized mice lost the partial protection displayed against the Friedlin V1 laboratory attenuated parasite strain (Figure S1).
The Leishmania-derived polyproteins L110f and KSAC have been extensively studied. L110f or its first generation antigen Leish-111f (equivalent to the clinical antigen LEISH-F1) and KSAC elicited protective immunity against L. major as well as L. infantum in rodent models of infection initiated by needle challenge [11], [12], [26], [27]. Both antigens are well-defined and used in formulation with MPL®-SE or GLA-SE, adjuvants suitable for human use.
Here, we vaccinated BALB/c mice using the same protocol used by Bertholet et al. [11] for L110f and Goto et al. [12] for KSAC with the exception of the exclusive use of a stable emulsion containing GLA, a synthetic glucopyranosyl lipid A molecule similar to the naturally derived MPL® [21]. The objective was to test whether vaccination of mice with either antigen is equally protective against challenge by L. major-infected P. duboscqi sand flies. This work is relevant to the value of these antigens as vaccines; particularly following findings by Peters et al. [20] demonstrating that parasite transmission by vector bite induces a specific innate immune response, which promotes the establishment of L. major infection. The authors went further to demonstrate that the virulence of vector-transmitted infections overcomes the protection observed against a needle challenge of vaccinated mice [10].
When BALB/c mice were challenged with infected sand flies, the KSAC+GLA-SE-vaccinated animals were protected from infection by both a laboratory-maintained parasite and a more virulent strain that was recently isolated from a soldier in Iraq (WR-2885). The partial protection observed in L110f-vaccinated mice following either needle or fly challenge with the Friedlin-V1 strain was abrogated when the mice were challenged with sand flies carrying the virulent WR-2885 strain. These data are consistent with the reported increase in virulence of a vector-transmitted infection and emphasize the need to test promising vaccines in vector-transmission models [10], [20]. Additionally, the data draw attention to the importance of the virulence of the parasite strain used in challenge experiments particularly for vaccine studies.
It is important to point out that the clinical forms of L111f and L110f (LEISH-F1 and LEISH-F2, respectively) delivered with MPL®-SE were safe and immunogenic in healthy subjects with and without histories of previous infection with L. donovani [28]. Additionally, LEISH-F1 had some therapeutic value in patients with mucosal and cutaneous leishmaniasis where it appeared to shorten time to cure when used with chemotherapy [29], [30]. Both Leish-111f and Leish-110f demonstrated therapeutic efficacy in dogs with canine leishmaniasis [31], [32]. Therefore, L110f should not be overlooked as a valuable vaccine in our fight against leishmaniasis.
Based on present and previous data [12] KSAC and GLA-SE used together show considerable promise as a preventive vaccine. Vaccinated mice were mostly pathology-free after challenge with either the Friedlin V1 or WR2885 L. major strains. Additionally, vaccinated mice were protected in a delayed challenge 12 weeks after the last vaccination using infected sand flies, indicative of the generation of long-lasting immunity. Protection was associated with a consistently positive IgG2a∶IgG1 ratio for KSAC. Interestingly, in L110f-immunized mice that were partially protected, this ratio fluctuated from negative to neutral at 62 days and 12 weeks post-vaccination, respectively. Such antibody fluctuation may reflect stabilization of antibody levels over time and further emphasizes the importance of testing the efficacy of the immune response to a vaccine in a delayed challenge. Of note, both KSAC and L110f generated a Th1-biased cell-mediated immunity. This was demonstrated by the predominant antigen-specific IFN-γ response (relative to the IL-10 response) of spleen cells from vaccinated mice one week post-challenge. However, L110f-vaccinated mice did not mount an immune response to SLA nor to pre-challenge stimulation of spleen cells with antigen. This finding is distinct from that of Bertholet et al. [11] where they demonstrated a balanced IgG2a/IgG1 response to L110f plus GLA-SE or MPL-SE, a sizable induction of CD4+CD44high IFN-γ+ cells, and good protection with both vaccines when challenged by needle. We cannot account with certainty for the apparent discrepancy between our immunogenicity results with L110f+GLA-SE and those of Bertholet et al. [11] and other reports using L111f/L110f+GLA-SE or +MPL-SE [26], [33], [34]. One difference between these studies and ours is the time chosen for pre-challenge analysis. In any case, the reduced protection we observed against the two L. major strains tested using L110f+GLA-SE correlate well with the relatively weak shift to a Th1 response that was observed after vaccination, but before parasite challenge. In contrast, splenocytes from KSAC+GLA-SE-vaccinated mice responded well to stimulation with antigen and, more importantly, to stimulation with SLA pre- and post- challenge by infected bites. KSAC-immunized mice generate a pool of effector memory CD4+IFN-γ+ T cells specific to KSAC that was efficiently stimulated with SLA both before and after a sand fly challenge. These results suggest that the rate and magnitude of the immune response are important for the generation of protection against a virulent sand fly-transmitted infection. Additionally, the observed differences in the protective effect of L110f and KSAC, both formulated with GLA-SE, against vector-challenge may be related to other factors such as antigenicity, accessibility or amount of the natural proteins making up these polypeptides.
Recent studies (S. Bertholet, personal communication) have shown that lower doses of adjuvant (GLA-SE) are more efficient at inducing long-lived CD4 memory responses, especially with L110f as an antigen (data not shown). This could indicate that optimal adjuvant doses vary for different antigens and might need to be titrated accordingly. The KSAC results reported here demonstrate that the synthetic TLR4 agonist GLA can be a powerful tool to direct a shift from Th2 to a Th1-type response, necessary to combat a vector-transmitted L. major infection.
In summary, the BALB/c mouse model was used in our experiments because it is especially susceptible to L. major infection as a result of its genetically determined Th2 immune response. We observed that immunization with KSAC in combination with the TLR4 agonist GLA in stable emulsion overcomes the Th2 bias of BALB/c mice, generating a robust, cell-mediated Th1 immune response in these mice. This immunological activation results in solid protection against vector-transmitted L. major infection, protection that is comparable to the one observed following needle challenge. As anticipated, immunization conditions (using L110f+GLA-SE) that produced a more modest immune response with a less dramatic shift from a Th2 to a Th1 response was less protective.
KSAC, a defined Leishmania-based vaccine candidate shows protection against a sand fly challenge, and this protection was produced in combination with the clinically viable adjuvant GLA-SE. With these encouraging results, more work is needed to test the protective nature of these vaccine components in more relevant models of cutaneous leishmaniasis.
|
10.1371/journal.pntd.0001258 | Prediction of Dengue Incidence Using Search Query Surveillance | The use of internet search data has been demonstrated to be effective at predicting influenza incidence. This approach may be more successful for dengue which has large variation in annual incidence and a more distinctive clinical presentation and mode of transmission.
We gathered freely-available dengue incidence data from Singapore (weekly incidence, 2004–2011) and Bangkok (monthly incidence, 2004–2011). Internet search data for the same period were downloaded from Google Insights for Search. Search terms were chosen to reflect three categories of dengue-related search: nomenclature, signs/symptoms, and treatment. We compared three models to predict incidence: a step-down linear regression, generalized boosted regression, and negative binomial regression. Logistic regression and Support Vector Machine (SVM) models were used to predict a binary outcome defined by whether dengue incidence exceeded a chosen threshold. Incidence prediction models were assessed using and Pearson correlation between predicted and observed dengue incidence. Logistic and SVM model performance were assessed by the area under the receiver operating characteristic curve. Models were validated using multiple cross-validation techniques.
The linear model selected by AIC step-down was found to be superior to other models considered. In Bangkok, the model has an , and a correlation of 0.869 between fitted and observed. In Singapore, the model has an , and a correlation of 0.931. In both Singapore and Bangkok, SVM models outperformed logistic regression in predicting periods of high incidence. The AUC for the SVM models using the 75th percentile cutoff is 0.906 in Singapore and 0.960 in Bangkok.
Internet search terms predict incidence and periods of large incidence of dengue with high accuracy and may prove useful in areas with underdeveloped surveillance systems. The methods presented here use freely available data and analysis tools and can be readily adapted to other settings.
| Improvements in surveillance, prediction of outbreaks and the monitoring of the epidemiology of dengue virus in countries with underdeveloped surveillance systems are of great importance to ministries of health and other public health decision makers who are often constrained by budget or man-power. Google Flu Trends has proven successful in providing an early warning system for outbreaks of influenza weeks before case data are reported. We believe that there is greater potential for this technique for dengue, as the incidence of this pathogen can vary by a factor of ten in some settings, making prediction all the more important in public health planning. In this paper, we demonstrate the utility of Google search terms in predicting dengue incidence in Singapore and Bangkok, Thailand using several regression techniques. Incidence data were provided by the Singapore Ministry of Health and the Thailand Bureau of Epidemiology. We find our models predict incident cases well (correlation greater than 0.8) and periods of high incidence equally well (AUC greater than 0.95). All data and analysis code used in our study are available free online and can be adapted to other settings.
| Google has reported success in using terms entered into its search engine (www.google.com) to predict trends in Influenza-Like Illness (ILI) cases one to two weeks ahead of the US CDC Morbidity and Mortality Weekly Report [1]. Several studies have reported similar results for influenza surveillance using Google search data, Yahoo search data, and internet advertising [2]–[8]. The Google research indicates that as the weekly incidence of influenza increases or decreases, the volume of certain internet search terms within the same geographical region change with a high level of correlation and predictability. Using a near real-time ability to collect search data (within 24 hours as opposed to one to two week lead time for US CDC reporting), the researchers were able to obtain information on the trend of ILI patterns in a more timely fashion than traditional surveillance.
Though the first efforts to use search terms from www.google.com have focused on influenza, this pathogen may be one of the more difficult to predict using internet searches. The presentation is non-specific to the pathogen and many searching behaviors that an ill person with influenza might exhibit overlap with searching by individuals afflicted by other pathogens. Pathogens exhibiting a distinct clinical presentation described by disease-specific terms that are widely used by the general population might exhibit the clearest correlation of search terms with disease incidence. Additionally, prediction of incidence is more important for pathogens that exhibit strong temporal variation. Dengue exhibits both of these characteristics: It exhibits a more distinct clinical presentation than influenza, giving rise to more disease-specific terms; and, dengue incidence in many locations exhibits large interannual variability with incidence varying by as much as a factor of ten from one year to the next [9].
The dengue virus is an arboviral illness of the family Flaviviridae and consists of 4 antigenically distinct serotypes. It is transmitted by the bite of an infected mosquito (Aedes aegypti poses the biggest threat to humans) and infection by one serotype does not confer life-long immunity to another serotype. The incubation period is typically 4–7 days and may present with undifferentiated fever, petechiae rash, nuchal headache, myalgia and arthralgia. The severe clinical manifestation, dengue hemorrhagic fever (DHF), is strongly associated with second infections and arises in around 3% of cases [10].
Prediction of outbreaks of dengue virus in countries with underdeveloped surveillance is of great importance to ministries of health and other public health decision makers who are often constrained by budget or man-power. The clinical presentation of dengue, although overlapping with other pathogens, is more specific to dengue than ILI is to influenza infection, and many of the search terms that individuals might search for when seeking information on dengue are specific to dengue (as opposed to terms such as ‘cold’). Thus, internet searches might exhibit stronger correlation with dengue incidence than influenza. Accurate predictions of dengue incidence might allow for more effective targeting of control measures such as vector control and preparation for surges in patients among hospitals, and may significantly increase the rapidity of dengue predictions in areas with less developed surveillance systems.
In Thailand, dengue has been a significant source of morbidity and mortality for over 70 years. DHF was first observed in Bangkok in 1949. The Thai Ministry of Public Health has conducted dengue surveillance since 1968. Incidence in Bangkok varies widely from year from 15,000 cases to over 175,000 cases annually. In Singapore, DHF was a significant cause of childhood mortality, in the 1960s, 1970s and 1980s, prompting vector control efforts that reduced the density of Aedes mosquito breeding sites and precipitated a decline in the incidence of DHF in the late 1980s and early 1990s [11]. However, from the late 1990s onwards, there has been a resurgence of dengue fever despite low levels of Aedes mosquito breeding, culminating in the largest observed epidemic in Singaporean history of over 14,000 cases in 2005. Although there is a trend towards an increase during the middle of the year, there is a wide variation of weekly incidence ranging from 32 to 713 cases during the 2004–2011 period.
The ability to accurately predict a rise in incidence would be a useful way to trigger a series of clinical interventions (deployment of medical teams, clearing of hospitial beds), and public health interventions (escalation in surveillance, public health education and pre-emptive source reduction measures) to reduce the transmission of dengue fever. Internet search term based surveillance could decrease delays associated with traditional surveillance systems and support under-developed systems.
Search data were downloaded from Google Insights for Search (http://www.google.com/insights/search/) on February 18th, 2011 for Singapore and March 2nd, 2011 for Bangkok. Relevant search terms for both were selected by brainstorming common words used in searching for dengue. We searched terms that include words in all three of the official languages in Singapore; English, Chinese Malay and Tamil. Terms for both Singapore and Bangkok were classified into 3 categories: nomenclature, signs/symptoms and treatment. The search terms for the “full models” are shown in Figure 1.
Google Insights for Search provides related searches that generate a significant volume of results. All relevant related search data were retrieved. Google Insights for Search ignores capitalization, but treats misspellings and different orderings (for example “symptoms flu” and “flu symptoms”) as distinct searches. However, the volume of search data for these are small and none were included in model testing.
Often, the Google Insight engine would only return data aggregated by month, because of uncertainty in weekly estimates in terms with low levels of search. For these terms a cubic spline was used to disaggregate the data to weekly responses (using R's spline()); negative values resulting from the spline were set to 0. The data were also regressed with the same model terms using the monthly aggregated data, and similar results were obtained (see below). Importantly, Google Insight returns a sample of the actual search volume, so exact replication of the estimates of the model covariates is impossible. To correct for seasonal variation and confounding by time we included both the month of the year (coded numerically as 1 for January, 2 for February, etc) and a numeric code indicating week and year of the current data point (given in R as the number of days since January 1st, 1970).
Epidemiologic surveillance data were obtained from the Singapore Ministry of Health website which conducts routine epidemiological data collection via the government polyclinics, public hospitals, clinical laboratories as well as via mandatory communicable disease reporting procedures [11]. Clinical and laboratory confirmed dengue fever cases have been reported to the Ministry of Health since 1977 and the data are aggregated by week. Thai monthly incidence data were gathered from the Thai Bureau of Epidemiology website. Since Google provides data on internet searches only since 2004, we only considered dengue incidence data from 2004. Incidence data for both Singapore and Bangkok are presented as the black lines in Figure 2.
We considered two outcomes, incident dengue cases and a binary outcome defined to be 1 during periods of high incidence and 0 otherwise. Multiple linear regression, negative binomial regression and generalized boosted regression (GBR) were used to model the weekly incidence of dengue fever using internet search terms [12]. A backwards and forwards step procedure was used to find the linear regression model that maximizes the Akaike information criterion (AIC). Negative binomial regression fit with the full set of search terms in each location was chosen over Poisson regression due to over-dispersion of the search term data. GBR models were fit using the gbm package in R [13].
Candidate models were trained using 2005–2010 data and used to predict 2011 incidence. Inclusion of 2004 data from Singapore reduced the predictive accuracy of the model. Because the predictions were not qualitatively different, and included a nearly overlapping set of covariates when 2004 was or was not included, we chose to optimize our predictions of incidence in later years by dropping 2004 from the models. To choose between the multiple linear regression, negative binomial regression and GBR models, we determined the model with the largest correlation between the 2010 prediction and lagged incidence. This model was then cross-validated to evaluate prediction performance. We used leave one out cross-validation and an expanding prediction window (both weekly and yearly, forward and backward) for search data dated between January 2005 and through December 2010, and evaluated the normalized root mean square error (NRMSE) of the predicted values of the left-out data from the observed incidence.
In addition to models predicting incidence, logistic regression and Support Vector Machine (SVM) models were used to predict periods of high incidence [12]. We built models for three different high incidence thresholds defined as the 50th, 75th and 90th percentiles of numbers of cases over the period 2005–2011. Model performance was assessed using the area under the receiver operating characteristic curve (AUC) for leave-one-out prediction.
All statistical analyses were conducted in R version 2.12.2 (R Core Development Team).
The AIC step-down model outperformed the GBR and negative binomial model for predicting numbers of incident cases and was chosen as optimal in both Singapore and Bangkok. The best fitting AIC step-down models have the predictor search terms shown in Figure 1. Table 1 shows the model diagnostics comparing the step-down and full models for Singapore and Bangkok. A multiple time series plot showing normalized dengue incidence, the results of the optimized model fits and the error between predicted and observed incidence is presented as Figure 2.
To assess the performance of the prediction on data that was not used to fit the model, we used multiple cross-validation techniques. We predicted incidence in 2010 in both locations using models fit to data from 2005–2010. Correlation between predictions of dengue in 2010 and observed dengue incidence for both Singapore and Bangkok are reported in Table 1.
We also assessed predictions for single and multiple observations that were left out of the data set used to fit the model. These results (reported in the Supporting Information S1) indicate a good fit of the step-down model relative to the full model. Additionally, the prediction errors are low in the leave-one-week-out case and the leave-52-weeks-out case. We also see poor performance of the negative binomial model relative to the other models.
For both Singapore and Bangkok, logistic regressions and SVM models were fit to predict the binary outcome of incidence above or below a threshold. Figure 3 summarizes the prediction of the SVM model in Singapore (a similar graph presenting the Bangkok SVM model is presented in the Supporting Information S1), and Table 2 presents the AUC and optimal sensitivities and specificities for the logistic and SVM models for each of the three cutoffs. We can see good prediction for the median and 75th percentile cutoffs.
We compared the performance of our model to a lag-1 autoregressive model using only dengue surveillance data from the last week (Singapore) or month (Bangkok) to predict the next observation. In Singapore, this model performs well, yielding a correlation between predictions and observed cases of 0.950. In Bangkok, the model performs much worse than models using search terms with a correlation of 0.766 (for comparison, the 8 search term model above has a correlation of 0.943). However, delays in compilation of these reports, especially in other locations could mean that these data would be unavailable for an autoregressive prediction model.
We have found that specific internet search terms are highly correlated with dengue incidence. Our best model for data from Singapore which included 16 terms showed a correlation of 0.931 with observed dengue incidence and an . The 8 term model for Bangkok performs equally well with a correlation of 0.869 and an . Out-of-sample predictions are predictably lower, but not significantly so. Our predictions of time periods with high dengue incidence are very accurate with sensitivities and specificities of 0.861–1.00 and 0.765–1.00 for multiple thresholds in each location. Together, these results demonstrate the viability of this data stream in supporting dengue surveillance.
Our model performed similarly to models built in other efforts to predict influenza incidence using internet search terms. Ginsberg et al. found a correlation of 0.90 for influenza incidence in the US using a model that included 45 search terms [1], and Polgreen et al. fit a series of models to influenza data in the United States and all had values of [4]. In out-of-sample prediction, our models performed slightly worse than the models of influenza produced by Ginsberg et al, which found a correlation of 0.97 (compared to 0.921 in Bangkok and 0.785 in Singapore). It should be noted that our model produces predictions for the entire year including high and low incidence seasons, whereas the models of Ginsberg produce predictions for only the influenza season. The accuracy of our predictions may be due to the clear clinical presentation of severe dengue. The larger interannual variability may also allow us to disentangle seasonal search behavior from dengue specific search behavior.
The search terms included in the models include nomenclature terms, terms describing signs and symptoms as well as treatment seeking. Interestingly, 11 of the 13 search terms that were found to be significant in our final model for Singapore were in English. This suggests that the typical language used for health seeking behavior in Singapore is English. In Bangkok, we also found that three of the seven significant terms are English.
We validated the candidate models using leave-one-observation-out, leave-one-year-out and forward and backward validation techniques. The model performance was fairly consistent across these different approaches. In our validations, we found one year with large incidence to be highly influential for the performance of our model (see Supporting Information S1). We expect that including future years with large incidence might further improve our results.
Singapore has an extremely well developed dengue surveillance system that makes reported cases available to policy makers and the general public with a delay of around one week. In a setting with this rapidity of reporting, it is challenging for an internet search term model to return results more quickly and with better performance than a model that uses only reported cases to predict future cases [14]. This point has been demonstrated elsewhere for predicting consumer behavior: predictive search term-based models perform better when used in conjunction with rich independent data sets [15]. Thus, in Singapore, this tool might best be used as a supplement to existing surveillance systems. However, in other settings, with less developed surveillance systems, an internet search term-based system may yield significant gains in the rapidity of predictions. In Thailand, there are significant delays in the reporting of cases from many areas of the country. Our model may give significant improvements in settings with significant delays. It is conceivable that some dengue-endemic settings in South and Southeast Asia may have significant internet use before surveillance systems are developed and thus an internet search term-based model may be a proxy for routine surveillance in these settings.
Caution must be used when generalizing our method to other settings. Even though we have chosen two settings that have very different rates of internet usage, both countries are of higher income than many of the countries in the region. However, it is reasonable to assume increasing internet penetration in the future. Individual models need to be developed for specific settings using local surveillance data and search terms. This effort shows that this approach may have promise in other settings.
There are several other limitations to our work. Internet searching behavior is susceptible to the impact of media reports as has been found for influenza systems [1], [16]. The rate of internet use and the rate of health information seeking in this setting may be changing over time and thus our parameters might need to shift over time to incorporate the impact of these changes. Although not affecting performance here, future outbreaks of other clinically similar diseases such as chikungunya may challenge the performance of our model for dengue. Finally, the Google Insight tool returns a sample of actual search data and limits the availability of search terms for which there are very few returns, often aggregating these terms to a large temporal discretization. This limits the utility of these terms for the purposes of prediction.
Search query surveillance is rapidly expanding into many areas of public health including the surveillance of noninfectious diseases and to influencing policy domains [17]–[21]. The current work demonstrates the utility of using search query surveillance to forecast the incidence of a tropical infectious disease. Additionally, and importantly, we have constructed forecasting models using freely available search query data from Google Insights and publicly available surveillance data from Singapore and Bangkok. In addition, we have developed these models using open source software from the R statistical project. Our approach can be readily adapted to other settings where other proprietary efforts can not be implemented. The approach may be an important tool in many dengue endemic settings in supporting the public health response to dengue.
|
10.1371/journal.pntd.0000846 | Coinfection with Different Trypanosoma cruzi Strains Interferes with the Host Immune Response to Infection | A century after the discovery of Trypanosoma cruzi in a child living in Lassance, Minas Gerais, Brazil in 1909, many uncertainties remain with respect to factors determining the pathogenesis of Chagas disease (CD). Herein, we simultaneously investigate the contribution of both host and parasite factors during acute phase of infection in BALB/c mice infected with the JG and/or CL Brener T. cruzi strains. JG single infected mice presented reduced parasitemia and heart parasitism, no mortality, levels of pro-inflammatory mediators (TNF-α, CCL2, IL-6 and IFN-γ) similar to those found among naïve animals and no clinical manifestations of disease. On the other hand, CL Brener single infected mice presented higher parasitemia and heart parasitism, as well as an increased systemic release of pro-inflammatory mediators and higher mortality probably due to a toxic shock-like systemic inflammatory response. Interestingly, coinfection with JG and CL Brener strains resulted in intermediate parasitemia, heart parasitism and mortality. This was accompanied by an increase in the systemic release of IL-10 with a parallel increase in the number of MAC-3+ and CD4+ T spleen cells expressing IL-10. Therefore, the endogenous production of IL-10 elicited by coinfection seems to be crucial to counterregulate the potentially lethal effects triggered by systemic release of pro-inflammatory mediators induced by CL Brener single infection. In conclusion, our results suggest that the composition of the infecting parasite population plays a role in the host response to T. cruzi in determining the severity of the disease in experimentally infected BALB/c mice. The combination of JG and CL Brener was able to trigger both protective inflammatory immunity and regulatory immune mechanisms that attenuate damage caused by inflammation and disease severity in BALB/c mice.
| Chagas disease, a life-long parasitic disease caused by the flagellate protozoan Trypanosoma cruzi, was discovered a century ago by the Brazilian physician Carlos Chagas, and remains one of the most neglected tropical diseases, affecting 13 million people in Latin America. Disease is characterized by distinct clinical courses, varying from asymptomatic to severe forms with damage to heart and/or gastrointestinal tract. The causes of the different clinical manifestations are not completely understood, but they certainly involve both parasite and host features. In this study, the authors analyzed immune response of BALB/c mice to infection with two different T. cruzi populations. One of them (JG) caused low parasitism and low levels of pro-inflammatory mediators associated with no clinical manifestation of the disease. The other (CL Brener) caused severe disease, high mortality and high levels of pro-inflammatory mediators. The coinfection, however, triggered singular regulatory immune mechanism(s) that attenuated damage caused by inflammation and disease severity that are typical of the single infection with CL Brener. As mixed infection is naturally found in patients in endemic areas, these results can explain, at least in part, the complexity of the immune responses and consequently the various clinical manifestations of the disease.
| Chagas disease (CD), a life-long complex illness caused by the protozoan parasite Trypanosoma cruzi, was firstly described by Carlos Chagas in 1909, but it is still acknowledged by the World Health Organization (WHO) as one of the most important neglected tropical diseases and as a significant public health problem in Central and South America [1]. T. cruzi is transmitted to humans and other susceptible hosts mainly through contact with the feces of infected blood-feeding triatomines, but alternative routes such as blood transfusion, organ transplant, vertical transmission (congenital) or ingestion of contaminated food (oral transmission) are presently more important in the current context of CD.
Despite one century of research, the most intriguing challenge to understanding the physiopathology of CD still lies in the complex host-parasite interrelationship. From the clinical point of view, T. cruzi infections progress in two phases. Patent parasitemia and parasitism in a wide variety of host cells characterize the acute phase of disease. This phase normally passes unnoticed because the signs and symptoms are similar to those of most common infections: fever, swollen lymph nodes, hepato- and/or splenomegaly. Sterile immunity is rarely achieved after T. cruzi infection, and most of the patients that survive the acute phase remain in a life-long asymptomatic state (indeterminate form) during the chronic phase of infection. However, a significant percentage of these patients (about 40%) develop deadly clinical forms of the disease up to 20 years after the first contact with the parasite, as a result of progressive tissue damage mainly involving the esophagus, colon and/or heart. On average, 5–10% of the T. cruzi infected individuals develop the digestive form of the disease and 30–40% develop cardiomyopathy (cardiac form), the most severe clinical manifestation of CD. The associated cardio-digestive form is observed in 2–3% of the patients [2].
The severity and prevalence of the different clinical forms of CD vary among different regions [2], but the cause of this clinical and epidemiological heterogeneity is a puzzling and yet unresolved question. Despite many uncertainties, it is more and more clear that the pathogenesis of CD is very complex and is a multifactorial trait influenced by several factors related to the parasite, the host and maybe also the environment [3]–[10].
Concerning to parasite related factors, there is extensive and well-characterized intraspecific genetic diversity in T. cruzi, which has been demonstrated by different biological, biochemical and molecular approaches [11]. The coexistence of mixed infections in vertebrate and invertebrate hosts has also been demonstrated in natural situations [12]–[14] and this certainly plays an important role in the context of the pathogenesis of CD. For instance, distinct parasite populations have been found in different tissues (blood, esophagus and heart) of the same chronically infected patients [14], [15], suggesting that specific tissue tropism of the parasite is one of the major factors determining the pathology of this illness. Similarly, T. cruzi genetic variability was shown to be an important factor influencing tissue tropism and pathogenesis in BALB/c mice double-infected with an artificial mixture of JG (T. cruzi II) and Col1.7G2 (T. cruzi I) monoclonal populations [3]. A clear difference in tissue tropism was observed after three months post-infection: the Co1.7G2 clone predominated over the JG strain in the rectum, diaphragm, esophagus, and blood, while a striking amount of the JG strain was observed in the heart muscles of coinfected mice. Intriguing results were also observed by Franco et al. (2003) in studying the effects of coinfection with two T. cruzi populations exhibiting opposing virulence and pathogenicity in Holtzman rats: the CL Brener (T. cruzi VI) clone, which induces severe and diffuse myocarditis with high mortality, and the JG strain, which causes moderate acute myocarditis with no mortality. Although less virulent when compared to CL Brener in single infections, the JG strain was the only parasite detected in the rat tissues at the end of the acute phase of the double infection, in contrast to the results observed in the single infection protocols [8].
Concerning to host related factors, it is also well accepted that genetic polymorphisms associated with the host's immune response have an essential role in determining the course of T. cruzi infection. In fact, there is evidence that changes in cytokine expression patterns during the course of infection play an important role in the disease outcome [7]. For instance, in vitro exposure to T. cruzi trypomastigotes induces higher expression of IL-10 in monocytes isolated from indeterminate patients relative to cardiac patients, suggesting an immunological imbalance among patients with the cardiac clinical form of CD [16]. The lower expression of IL-10 among cardiac patients was associated with occurrence of a polymorphism in the promoter region of the IL-10 gene [5]. Furthermore, associations of polymorphisms in the genes for BAT-1 and NFκB with the development of cardiomiopathy were also described for CD in the Brazilian population [10], [17]. Host genetic factors are also involved in determining parasite tissue tropism in experimental CD. Andrade et al. (2002) clearly demonstrated that the genetic background of mouse strains (BALB/c, DBA-2, C57BL/6, and Swiss) influences the differential tissue distribution of JG and Col1.7G2 populations in double-infected animals [4]. Subsequently, using congenic mice, Freitas et al. (2009) identified MHC-associated genes as those mainly involved in determining the differential tissue tropism of these two parasite populations [9].
In conclusion, there are many studies alternatively demonstrating the importance of parasite or host immune response factors influencing the pathogenesis of the CD, but the present work is probably the first one that simultaneously investigates the mechanism and the contribution of both parts. Herein, we assess the parasitemia, body weight evolution, survival rate, different hematological parameters, heart parasitism and histopathology, and heart differential tissue tropism. We also perform quantitative analyses of serum cytokines and nitric oxide as well as flow cytometry analyses of spleen cells during the acute phase of infection with the JG and/or CL Brener in BALB/c mice. We clearly demonstrate that coinfection with JG and CL Brener is able to trigger both protective inflammatory immunity and regulatory immune mechanisms that are capable of both attenuate damage caused by inflammation and disease severity induced by single infection with CL Brener in BALB/c mice.
All animals were handled in strict accordance with good animal practice as defined by the Internal Ethics Commitee in Animal Experimentation of the Universidade Federal de Minas Gerais (UFMG), Belo Horizonte (BH), Minas Gerais (MG), Brazil (CETEA/UFMG - Protocol no. 5/2007).
Six to eight-week-old inbred male BALB/c mice, bred and maintained in the animal breeding units at the Instituto de Ciências Biológicas (ICB/UFMG), or Centro de Pesquisas René Rachou/Fundação Oswaldo Cruz (CPqRR/FIOCRUZ), both in BH/MG, Brazil, were used. We used two different T. cruzi populations: the JG strain (T. cruzi II), which was isolated from a chagasic patient with mega esophagus, and the CL Brener clone (T. cruzi VI), which was obtained from CL strain isolated from a Triatoma infestans specimen. Parasite major lineages were identified as recently recommended by an expert committee [18]. Both T. cruzi populations were maintained by intraperitoneal (i.p.) inoculation of infective blood trypomastigotes in Swiss mice. Prior genetic characterization of T. cruzi populations used in this work (Table 1) was done by typing seven polymorphic microsatellite loci [19], [20], and the genes for 24Sα rDNA [21] and cytochrome oxidase subunit II (COII) [22]. For BALB/c mice infections, infective blood trypomastigotes were obtained from retroorbital plexus of JG or CL Brener infected Swiss mice. Both trypomastigote populations were counted and diluted in LIT medium. BALB/c mice were i.p. inoculated with 0.10 ml of a suspension containing 100 trypomastigotes of JG or CL Brener (single infection), or a mixture of 50 trypomastigotes of each (double infection). Age- and sex-matched non-infected BALB/c mice were used as controls. Experimental groups consisted of three, six or twelve mice. The experiments were repeated at least twice.
Parasitemia was assessed by counting the bloodstream form of parasites in 5.0 µl of tail vein blood of JG and/or CL Brener infected mice, on alternate days from the 5th day p.i. until the time point that the parasites became undetectable [23]. Data were expressed as number of trypomastigotes per milliliter of blood. Survival was determined by daily inspection post-infection (p.i.), and mice were weighed on alternate days to monitor the systemic repercussions during the course of infection.
At 7, 14 and 21 days post-infection, mice were bled from the axillary plexus under xylazine/ketamine anesthesia and peripheral blood (PB) was collected with anticoagulant for hematological analyses or without anticoagulant for serum cytokine and nitric oxide (NO) assays. The hematological parameters (leukocytes, red blood cells, platelets, hematocrit and hemoglobin) were determined using the ABX Micros ABC Vet automatic system (Horiba ABX diagnostics, Montpellier, France). Differential leukocyte counts were determined under the oil immersion objective (100×), using standard morphological criteria, in peripheral blood smears stained with May-Grünwald-Giemsa, and the absolute number of each leukocyte subtype per ml of PB was determined.
To compare the effects of JG and/or CL Brener infection on influx of inflammatory cells in the heart, we analyzed the intensity of myocarditis morphometrically. For this experiment, animals were euthanized by cervical displacement, and the hearts were removed and sliced transversally at 7, 14 and 21 days p.i. The apical half of each heart was washed in phosphate-buffered saline (PBS) and stored in absolute ethanol at 4°C for PCR assays; the heart bases were fixed in 4% phosphate-buffered formaldehyde and used for histopathology. After 24 hours of fixation, the tissues were paraffin-embedded, and three 5-µm thick, semi-consecutive sections were obtained and stained by hematoxylin-eosin (H&E). Heart inflammation was assessed in the left ventricle free wall. For quantitative analyses, ten fields from each of the three semi-consecutive sections were randomly captured with the 40× objective, corresponding to a total myocardium area of 234.376 µm2. Images were captured at a resolution of 1392×1040 pixels with a Cool SNAP-Pro cf Collor microcamera (Media Cybernetics, Bethesda, MD, EUA) and transferred to a computer using Image-Pro Express version 4.0 software for Windows (Media Cybernetics). After proper calibration, captured images were analyzed with KS300 software (Zeiss, Jena, Germany). The nucleus area from each cell presented in the analyzed fields was digitalized and automatically measured in µm2. The results were expressed by nucleus area/total area ratio. Heart parasitism was evaluated by counting the number of parasite nests in three semi-consecutive sections as visualized by light microscopy with a 40× objective. PCR was additionally performed in parallel samples.
Detection of parasites in heart tissue samples was performed by specific PCR amplification of a fragment of about 330 bp from variable regions of minicircle kinetoplast DNA (kDNA) molecules of T. cruzi, as previously described [3] with some modifications. Tissue samples, stored in absolute ethanol, were fragmented and submitted to alkaline lysis, as follows: the fragmented samples were boiled in the presence of 50 mM NaOH for 10 min, and after neutralization with 130 mM Tris-HCl pH 7, samples were 10-fold diluted in sterile Milli-Q water and used as the DNA template for PCR. Samples from uninfected BALB/c mice were used as a negative control. PCR was carried out in a final volume of 20 µl containing 1.5 mM MgCl2, Green GoTaq Reaction Buffer pH 8.5 (Promega, Madison, Wisconsin, USA), dNTPs at 250 µM, primers (S35: 5′-AAATAATGTACGGGKGAGATGCATGA-3′ and S36: 5′-GGTTCGATTGGGGTTGGTGTAATATA-3′) at 1.0 µM, 1.0 U of GoTaq DNA Polymerase (Promega) and 1.0 µl of 10-fold diluted alkaline lysis products. Amplification was performed in a PT100 thermocycler (MJ Research) using an initial denaturation step at 94°C for 5 min followed by 35 amplification cycles including an annealing step at 60°C, extension at 72°C and denaturation at 94°C, each for 1 min. At the end, the extension step was extended to 10 min. The PCR products were visualized on a 6% polyacrylamide gel using silver staining, as previously described (Santos et al., 1993). Differential tissue tropism of both T. cruzi populations was assessed by analyzing the LSSP-PCR profiles and one of the previously typed polymorphic microsatellite loci (TcAAAT6) in positive tissue samples from double-infected mice.
The relative proportions of JG and CL Brener in the positive heart tissue samples obtained from double-infected mice were estimated using the LSSP-PCR assay, as previously described [3] with some modifications. For this, kDNA amplicons were subjected to electrophoresis on an ethidium bromide stained 1.5% agarose gel (1.0% agarose, 0.5% low melting point agarose) at 100 V for 1 h 30 min. The DNA bands corresponding to the 330-bp amplicons from variable regions of T. cruzi kDNA minicircles were purified from the gel, diluted 10-fold in sterile Milli-Q water, and subjected to a second PCR assay at low stringency, using a single fluorescent primer. The PCR was carried out in a final volume of 10 µl containing 1.5 mM MgCl2, Colorless GoTaq Reaction Buffer pH 8.5 (Promega), dNTPs at 50 µM, fluorescent primer (S35G*: 5′-Fluorescein ATGTACGGGGGAGATGCATGA-3′) at 4.5 µM, 1.6 U of GoTaq DNA Polymerase (Promega) and 1.0 µl of a solution containing the ∼330-bp DNA fragments prepared as described above. Amplification was performed in a PT100 thermocycler (MJ Research) as follows: an initial denaturation step at 94°C for 5 min, followed by 40 amplification cycles with the annealing step at 30°C, extension at 72°C, and denaturation at 94°C, each for 1 min. The final extension step was extended to 10 min. To determine the DNA fragment sizes, the LSSP-PCR products were analyzed by 6% polyacrylamide gel electrophoresis under denaturing conditions (8 M urea) in an Automatic Laser Fluorescent (ALF) sequencer (GE Healthcare, Milwaukee, Wisconsin, USA) followed by data analysis using the Allelelocator software (GE Healthcare). Areas under specific peaks from JG and CL Brener curves were used to estimate the relative proportions of each population in reference to a standard curve, as previously described [3]. Briefly, genomic DNA samples from JG and CL Brener were mixed in different proportions (JG/CL Brener: 9/1 (lane 2), 3/1 (lane 3), 1/1 (lane 4), 1/3 (lane 5) and 1/9 (lane 6)), and subjected to PCR assays. Fluorescent products of PCR were loaded into a 6% polyacrylamide gel under denaturing conditions in an automated DNA sequencer. The proportions of the sum of areas under specific peaks (Figure 1A) of each population were used to construct a standard curve. The standard curve was obtained using GraphPad Prism 5.00 software (GraphPad Software, San Diego, California, USA) by point-to-point analysis without the choice of any specific model. We used 96 points calculated with the x values (relative proportion of JG/CL Brener) ranging from 0.0 to 0.9 for building the standard curve (data not shown).
The relative proportions of JG and CL Brener in the hearts of the double infected mice were also assayed by genotyping the TcAAAT6 microsatellite locus. To achieve that, a full nested PCR protocol was used, as previously described [20] with some modifications. Briefly, PCR was performed in a final volume of 15 µl containing 10 mM Tris-HCl pH 9.0, 50 mM KCl, 0.1% Triton X-100 (Buffer B, Promega), 2.5 mM MgCl2 (Promega), 0.5 U of Taq DNA Polymerase (Promega), dNTPs at 250 µM, primers (TcAAAT6ex-forward 5′-ACGCACTCTCTTTGTTAACAG-3′ and TcAAAT6ex-reverse 5′-CCGACAACGATGACAGCAAT-3′) at 0.3 µM and 1.0 µl of DNA template (10-fold diluted alkaline lysis products). Amplification was performed in a PT100 thermocycler (MJ Research) using the step-down protocol modified for amplification of T. cruzi DNA as follows: an initial denaturation step at 94°C for 5 min; annealing at 58°C for 30 s; extension at 72°C for 1 min and a denaturation step at 94°C for 30 s. After every five cycles, the annealing temperature was decreased by two degrees to 55, 53, 51 and finally 48°C. At this last temperature, the number of cycles was increased to 15, followed by a final extension step at 72°C for 10 min. A second round of amplification was performed in same conditions described above but with inner primers (TcAAAT6-forward 5′-FluoresceinGCCGTGTCCTAAAGAGCAAG-3′ and TcAAAT6-reverse 5′-GGTTTTAGGGCCTTTAGGTG-3′). For the second PCR round, 10% of the amplified products obtained in the first PCR round were used as the DNA template. The determination of allele sizes was performed as described above. Areas under specific peaks from JG and CL Brener were used to estimate the relative proportions of each population by reference to a standard curve (Figure 1B), as described above.
For cytokine analysis, serum samples were collected as previously described and stored at −20°C until used. Cytokines (IL-2, IL-4, IL-5, IL-6, IL-10, IL-12p70, IFN-γ, CCL2 and TNF-α) were measured with BD CBA Mouse Cytokine assay kits according to the manufacturer's specifications (BD Biosciences, CA, USA).
Serum nitric oxide (NO), an oxidation product of arginine by NO synthase, was measured as nitrite (NO2−), the stable product of reactive nitrogen intermediates, at 7, 14 and 21 days p.i. in samples collected as described above. Serum nitrite levels were assessed using the Griess reaction, after deproteination of samples with 1 M ZnCl2. Nitrite concentrations were determined by extrapolation from a standard curve constructed using various concentrations of sodium nitrite (NaNO2−), and the results were expressed in µM.
Spleen samples were collected at 7, 14 and 21 days p.i. in RPMI-1640 (GIBCO, Grand Island, NY, USA). Spleen cell suspensions were prepared as previously described [24] and kept on ice. Cells were counted and incubated for 12 h at 37°C in a 5% CO2 humidified incubator and re-incubated again for more 4 h in the presence of 10 µg/ml brefeldin A (BFA) (Sigma, St. Louis, MO, USA), in the same conditions. Cell samples were then treated with 2.0 mM ethylenediaminetetraacetic acid (EDTA) (Sigma, St. Louis, MO, USA) for 10 min at room temperature and washed once with FACS buffer (PBS with 0.5% of bovine serum albumin (BSA) pH 7.4 (Sigma, St. Louis, MO, USA). After washing, cells were incubated with undiluted rat anti-mouse (anti-CD4, anti-CD8, anti-CD49b or anti-MAC-3) or hamster anti-mouse (anti-CD69) monoclonal antibodies (mAbs) specific for different cell surface markers and labeled with fluorescein isothiocyanate (FITC), phycoerythrin (PE) or peridinin chlorophyll-alpha protein (PerCP), all purchased from BD Biosciences Pharmingen (San Diego, CA, USA). Cell suspensions were homogenized and incubated for 30 min at room temperature in the dark. Cell surface-labeled samples were treated with FACS Lysing/fix Solution (BD Pharmingen), immediately vortexed and incubated at room temperature for 3 min in the dark. After the lysis/fixation procedure, membrane-labeled spleen cells (except for the samples incubated with anti-CD69-PE) were permeabilized for 10 min with FACS permbuffer (FACS buffer with 0.5% saponin, Sigma, St. Louis, MO, USA), washed and resuspended in FACS buffer containing the following antibodies: anti-IL-10, anti-IL-12p70, anti-IFN-γ or anti-TNF-α (BD Biosciences Pharmingen, San Diego, CA, USA). After intracytoplasmic staining, cells were washed with FACS buffer and were fixed with FACS FIX Solution (10 g/L paraformaldehyde, 1% sodium cacodylate, 6.65 g/l sodium chloride, 0.01% sodium azide). Data acquisition was performed in a Becton-Dickinson FACScalibur flow cytometer (BD Pharmingen, San Diego, CA, USA) with CELLQuest software provided by the manufacturer. A total of 30,000 (for only surface labeling) or 50,000 (for intracellular cytokines) events per tube were acquired. Flow cytometry analyses were performed using CELLQuest software, and the absolute number of each spleen cell subtype per spleen was determined.
All statistical analyses were performed using GraphPad Prism 5.00 (GraphPad Software, San Diego, California, USA). The parameters studied, except survival analysis, were analyzed by One Way Analysis of Variance, and when differences between groups were verified, multiple comparisons were performed by the Student-Newman-Keuls' post-test. Survival analysis was carried out using the Kaplan-Meier method, and the significance of differences between groups was assessed using the logrank test. P-values of 0.05 or less were considered significant. The results were expressed as mean ± SEM.
BALB/c mice were infected with 100 trypomastigotes of JG or CL Brener (single infection) or coinfected with 100 trypomastigotes derived from a recently prepared mixture of both T. cruzi populations in a 1∶1 proportion via intraperitoneal route. The parasitemia levels, assessed from the 5th day p.i. to the time point that the parasites became undetectable, revealed that JG single infected mice presented lower parasitemia in relation to all other infected animal groups in spite of the day p.i. evaluated. In contrast, animals single-infected with CL Brener presented higher parasitemia, while mice coinfected with JG and CL Brener presented intermediate levels of parasitemia (Figure 2A). This behavior cannot be explained by the simple effect of the relative reduction in the CL Brener inoculum from 100 trypomastigotes (used in the single infections) to 50 trypomastigotes (used in the double infection), since mice single-infected with 50 trypomastigotes of CL Brener present similar parasitemia, symptoms and survival curve to those observed among mice single-infected with 100 trypomastigotes forms of CL Brener (data not shown).
Body weight loss was more significant among mice infected with CL Brener alone or in the presence or absence of JG in relation to other groups. This was especially noteworthy on the 21st day p.i., when body weights among CL Brener single infected (22.8±1.4 g) or coinfected (24.0±0.9 g) mice were significantly lower in relation to naïve (27.0±0.4 g) or JG single infected (29.0±0.5 g) animals. However, the body weights of all infected mice that overcame the acute phase of disease returned close to those of naïve mice during the course of infection (Figure 2B). Similar to the parasitemia and body weight loss levels, the mortality rate was null among naïve or single JG infected mice, higher among CL Brener single infected mice (75%) and intermediary among the coinfected mice (55%). However, despite slight differences observed in the mean survival time among CL Brener single infected (24±3 days p.i.) and coinfected (29±3 days p.i.) mice, the survival curves of animals infected with CL Brener alone or in the presence of JG were not significantly different (Figure 2C).
Peripheral blood samples were analyzed to assess the effects of JG and/or CL Brener infection on hematological parameters. There was a significant reduction in global leukocyte numbers on days 7 and 14 p.i. among CL Brener infected mice, in the presence or absence of JG. Infection with JG alone led to reduction only on day 7 p.i. (Figure 3A). Differential leukocyte counts also revealed variation among the groups of infected mice (Figure 3B–F). A significant reduction in lymphocyte counts was observed for all groups, but the magnitude of lymphopenia was more intense among CL Brener single infected mice (Figure 3B). Neutrophilia (Figure 3C) and bastonet neutrophilia (Figure 3D) were observed at day 21 p.i. among CL Brener infected animals. JG infected mice presented almost normal neutrophil and bastonet neutrophil counts, while coinfected animals presented intermediary counts (Figure 3C and 3D). Significant eosinopenia was observed in all infected mice on 7th and 14th days p.i., but returned to basal levels by 21 days p.i. (Figure 3E). Regarding monocyte counts, significant reduction was only observed on the 14th day p.i among animals infected with CL Brener in the presence or absence of JG (Figure 3F). Besides leukocyte amounts, the hemoglobin level, hematocrit and red blood cell concentration were also determined. A significant reduction in these parameters was only observed on the 21st day p.i. and only among animals infected with CL Brener in the presence or absence of JG (Table 2). Platelet counts were not significantly different among experimental groups during the course of infection (data not shown).
Parasite-induced cell destruction followed by focal inflammation usually correlates to tissue damage and heart malfunction. We evaluated heart inflammatory infiltrates and parasitism to assess the differential effects of infection with JG and/or CL Brener in heart tissue lesions at 7, 14 and 21 days p.i. As expected, JG and/or CL Brener infected mice presented typical heart histopathological alterations of the acute phase of infection, such as inflammatory infiltrates predominantly constituted by mononuclear cells, edema and some degree of degenerative changes of the myocardium (Figure 4). At 7 days p.i., the heart inflammatory response was more intense among JG infected mice in relation to animals infected with CL Brener or coinfected, and the inflammatory foci, when present, were small (data not shown). At the 14th day p.i., in all heart samples we noticed moderate inflammatory foci but we did not observe significant differences among infected mice (data not shown). Parasite nests were not visible yet. At the 21st day p.i., however, hearts from CL Brener infected mice presented more intense and diffuse inflammation in the ventricular and atrial walls when compared to JG infected mice, as well as when compared to coinfected animals (Figure 4A–D and 4F). The acute myocarditis induced by JG was predominantly focal and more restricted to the epicardial face of the myocardium (Figure 4B). Coinfected animals presented acute myocarditis of intermediary intensity (Figure 4D) when compared to JG or CL Brener single infected mice (Figure 4B, 4C and 4F). Furthermore, heart tissue parasitism, evaluated by counting the number of parasite nests in three HE-stained semi-conservative sections, was significantly higher among CL Brener single infected mice in comparison to animals infected with JG only or coinfected. This last group presented an intermediary level of heart parasitism at 21 days p.i. (Figure 4E). It is important to notice that the pattern of myocarditis induced by single infection with JG was rarely associated with tissue damage (see detail in Figure 4B). Meanwhile, the pattern induced by CL Brener was more often associated with cardiomyocyte degeneration associated to disrupted nests of parasites (see detail in Figure 4C) than that observed in hearts from coinfected mice, (see detail in Figure 4D).
As expected in the acute phase of experimental T. cruzi infection, parasite kDNA was detected by PCR in heart tissue samples from all infected mice on the 14th and 21st days p.i. However, only scarce amounts of parasite kDNA were detected in a small number of animals on the 7th day p.i. (data not shown). The differential tissue tropism of JG and CL Brener in heart tissue samples from coinfected animals was evaluated by analyzing both LSSP-PCR and polymorphic microsatellite locus profiles.
LSSP-PCR profiles of heart tissue samples obtained from coinfected mice revealed the presence of JG-specific amplicons (229, 233 and 258 bp) in 66% and 100% of samples collected on the 14th and 21st days p.i., respectively. The CL Brener-specific amplicon of 248 bp, in turn, was detected in 100% of samples collected on both the 14th and 21st days p.i. The relative amount of CL Brener/JG kDNA detected in the hearts analyzed varied from 76±9 to 77±8% in samples collected on the 14th and 21st days p.i., respectively.
Similar results were observed using TcAAAT6 microsatellite locus analysis. Heart tissue samples obtained from coinfected mice revealed the presence of JG-specific alleles (271 and 275 bp) in 16% and 50% of samples collected on the 14th and 21st days p.i., respectively. The CL Brener-specific allele (263 bp) was detected in 100% of samples collected on both the 14th and 21st days p.i. The relative amount of CL Brener/JG kDNA detected in the heart samples analyzed varied from 97±3 to 93±3% in samples collected on the 14th and 21st days p.i., respectively.
To assess whether differences in the outcome of infection with the JG and/or CL Brener were associated with particular patterns of cytokine response, we determined the levels of cytokines (IL-2, IL-4, IL-5, IL-6, IL-10, IL-12p70, IFN-γ, CCL2 and TNF-α) and nitrite, a more stable NO-derived metabolite, in serum samples collected at 7, 14 and 21 days p.i. A slight increase in serum levels of IL-2 (naïve mice: 1.70±0.13 and CL Brener infected: 2.18*±0.11, mean ± SEM, *P<0.05) and IL-5 (naïve mice: 2.70±0.45 and CL Brener infected mice: 5.08*±0.38, mean ± SEM, *P<0.05) was detected only among CL Brener infected mice in relation to naïve mice and only on the 14th day p.i. The biological significance of these small variations is unclear. Although measurable amounts of IL-4 and IL-12p70 were detected in serum samples from all experimental groups at all timepoints analyzed, no difference among groups was found (data not shown). Different patterns were observed, however, for all other measured cytokines. There were significant differences in serum levels of pro-inflammatory cytokines such as TNF-α, CCL2, IL-6 and IFN-γ at 14 and 21 days p.i. among CL Brener infected or coinfected mice in relation to naïve animals or JG single infected mice. It is noteworthy that IL-10 levels were maintained close to the basal level among infected animals on the 7th and 21st days p.i., while pro-inflammatory cytokine levels (TNF-α, CCL2, IL-6 and IFN-γ) presented a great variation among different experimental groups during the acute phase of infection (Figure 5A–E). Since cytokines act in a network of mutual interactions in vivo, ratios of pro-inflammatory cytokines (TNF-α, CCL2 and IFN-γ) to the immunoregulatory cytokine IL-10 were analyzed. The ratios of serum TNF-α, CCL2 or IFN-γ to serum IL-10 on the 7th, 14th and 21st days p.i. showed that animals infected with CL Brener or coinfected presented an increase in all TNF-α/, CCL2/ or IFN-γ/IL-10 ratios in at least one of three points analyzed in relation to both naïve mice and JG infected animals. However, coinfected animals had significant reductions in the ratio of TNF-α/IL-10 at 14 days, and of CCL2/IL-10 at 21 days p.i., suggesting a modulating role in the coinfection with JG and CL Brener in BALB/c mice (Figure 6A–C).
Regarding serum NO-derived metabolite levels, only NO2− was measured in the present work and no significant difference between experimental groups at 7, 14 and 21 days p.i. was found (data not shown).
Since the cytokines measured can be produced by more than one cell type, we evaluated next which cell type could be the source of the pro-inflammatory cytokine TNF-α and the immunoregulatory cytokine IL-10 in the spleen. These two cytokines were chosen because they represent opposite points on the inflammatory spectrum of immune response; the ratio between them revealed a significant difference between single infection with CL Brener and coinfection with JG and CL Brener. Analysis of TNF-α-producing cells showed that MAC-3+, NK, CD4+ and CD8+ T cells were important sources of this cytokine at days 14 and 21 p.i. for all infected mice (Figure 7A and 7B; 8A and 8B; 9). At day 21 p.i., the number of TNF-α-producing MAC-3+ cells was reduced in mice infected with CL Brener or coinfected (Figure 7A). However, the number of TNF-α-producing CD4+ and CD8+ T cells was augmented in coinfected and CL Brener infected mice, respectively (Figure 8A and 8B). Interestingly, the number of IL-10-producing macrophages and CD4+ T cells showed a significant increase in coinfected mice at day 14 p.i. (Figure 7C and 8C). In addition, no decrease in the number of these cell subpopulations was observed at any time point analyzed in coinfected mice.
The acute phase of CD is characterized by both high parasitemia and tissue parasitism, but most of the patients present few or no clinical symptoms in this phase of disease. Therefore, studies related to the early activation phase induced by natural T. cruzi infection in humans are scarce, and most information concerning parasite-associated features and host immunity related to T. cruzi infection is derived from studies using experimental models, in particular the murine model. Currently, we have a large amount of scientific information concerning the immune response during the early activation phase in animals acutely infected with T. cruzi, especially in BALB/c and C57BL/6 mice, which present different susceptibility to various intracellular pathogens, among them T. cruzi [25], [26]. However, most of these studies are either restricted to single-infected mice or are focused on analyses of few parasitological or immunological parameters. Co-existence of natural mixed infections among humans certainly plays an important role in the context of CD, and the complex interrelationships between host- and parasite-related factors might ultimately influence the outcome of infection with T. cruzi. Herein, we investigated the effects of the association of JG and CL Brener during the acute phase of infection in BALB/c mice by simultaneously analyzing different parasitological, histopathological and immunological parameters.
To better simulate natural infection conditions, inocula of 100 trypomastigotes of each parasite population (JG or CL Brener) or of a mixture of them (JG and CL Brener) were used. This inoculum is much smaller than those commonly used in the literature, which frequently reach 104 trypomastigotes or more per animal. Assessment of parasitemia and heart parasitism revealed great differences in parasite burden between JG and/or CL Brener infected mice. JG single infected mice presented lower parasitemia and heart parasitism compared to the CL Brener infection, which induced high parasitemia and heart parasitism, at least in the acute phase of infection. Animals coinfected with JG and CL Brener presented levels of parasitemia and parasitism at an intermediate level compared to those from JG or CL Brener single infected mice. The significant reduction in heart parasite nests observed in coinfected animals when compared to CL Brener infected ones correlates with a decrease in the number of inflammatory cells (measured by nucleus area), suggesting that coinfected animals had a less intense inflammatory reaction in the heart. It is plausible that the reduction of these two important parameters contributes to the trend observed in the mortality curve showing improved survival of coinfected mice.
Direct identification of parasite populations in heart tissue samples from double-infected mice revealed a relative predominance of CL Brener, varying from 50 to 100% in all analyzed tissues depending on the technique and on the time elapsed since infection considered. Although the percentage of JG detected was always lower than CL Brener, we observed a progressive increase in the presence of JG in the heart samples from double-infected animals throughout the acute phase of infection. The relative predominance of one of the parasite population in the hearts of the animals seems to correlate to the genetic aspects of the parasites and the hosts, rather than to the initial inoculum used, since similar results were observed using a mixture containing 50 or 100 parasites of each parasite population. In addition, previous studies have demonstrated that variation of one component of the parasite mixture or of the mouse genetic background, especially MHC-associated genes, can interfere in the relative predominance of a parasite population in different tissues, as well as in disease evolution [3], [4], [9], [27], [28].
The severity of disease induced by T. cruzi infection in BALB/c mice was measured by the assessment of body weight loss, heart tissue damage and mortality rate, and it corresponded to the parasite population involved. Absence of evident symptoms of disease, moderate acute myocarditis and null mortality were observed among JG single infected mice. On the other hand, CL Brener single infected mice presented gradual and progressive disease, characterized by anorexia, lethargy and cachexia, as well as severe acute myocarditis and a high mortality rate. Interestingly, coinfected animals presented symptoms similar to those presented by CL Brener infected mice, yet with lower magnitude. In addition, these animals presented less heart tissue damage, a reduced mortality rate and longer mean survival time compared to CL Brener single infected mice. It is interesting that Franco et al. (2003), working with the same T. cruzi populations but a different host, observed similar results [8].
Regardless of the experimental model studied, the inflammatory response triggered by infection or tissue damage involves the coordinated recruitment of blood components (plasma and leukocytes) to the site of infection or injury. The relative and absolute numbers of peripheral blood cells are critically regulated in physiological conditions, and disruptions in this physiological balance can be clinically detected in several disease states. In accordance with this, we found considerable variations in blood leukocyte counts among T. cruzi infected animals. The leukopenia observed among infected mice during the early phase of infection is probably caused by an intense recruitment of leukocytes to the inflammatory sites, and the return of total leukocyte counts close to basal levels at 21 days p.i. was mainly related to a significant increase in neutrophil counts. More importantly, we observed a severe and persistent lymphopenia among CL Brener single infected or coinfected animals at 14 and 21 days p.i., a condition that can be associated with both an immunosuppressive state and the high mortality rates observed among these animals during the course of infection. Marcondes et al. (2000) reported severe hematological alterations, characterized by pancytopenia and a low number of bone marrow blood cell precursors, in particular erythroblasts and megakaryoblasts, in mice infected with T. cruzi. Infection was accompanied by anemia, decrease in hematocrit and hemoglobin levels, as well as an exponential growth of parasites, and high mortality [29]. In the present study, we showed a significant anemia, with decreases in both number of red blood cells and hematocrit, as well as in hemoglobin levels among animals single infected with the CL Brener or coinfected at 21 days p.i., which may contribute to high mortality rates among these mice. The lifespan of murine red blood cells is from 1 to 2 months; anemia was detected earlier than this. Therefore, reduced lifespan of red blood cells should be considered as an additional factor that contributes to anemia, which can influence survival of T. cruzi infected hosts.
T. cruzi is capable of infecting a wide variety of host cells, but the persistence of this parasite in cardiac, skeletal and smooth muscle cells is, at least in part, a key aspect of both the chronic phase of the infection, as well as the outcome of disease. The first step to ensure T. cruzi survival and successful infection is to enter host cells. Several molecules present on host cells and on the parasite surface are essential for the process of cell invasion [30] and are capable of stimulating an innate immune response upon the first encounter [31]–[33]. These early interactions are critical for immediate control of parasitemia and parasitism, as well as for establishment of a cytokine-rich microenvironment that influences the generation and direction of the downstream adaptive immune response. With this in mind, we scrutinize the host response to T. cruzi infection through analysis of immunological parameters, such as serum cytokine and NO levels, as well as analysis of the expression profile of cytokines by several spleen cell subpopulations.
High levels of serum TNF-α during the course of disease are usually associated with toxemia symptoms (anorexia, lethargy and cachexia) and high mortality rates. TNF-α, also known as cachectin, is produced primarily by mononuclear phagocytes (monocytes and macrophages) and acts as a multipotent modulator of immune responses; it is also a potent endogenous pyrogen, a well-known mediator of cachexia, and a marker of sepsis. Due to its multiple functions in immunological activity, TNF-α plays a critical role in several conditions that involve systemic inflammatory responses, such as sepsis and toxic shock [34], [35]. In accordance with this, animals single-infected with JG presented serum TNF-α levels similar to those found in naïve mice, and no clear symptoms of disease. Hölscher et al. (2000) showed that the TNF-α neutralization not only attenuated disease progression, but also prolonged the survival of IL-10−/− mice infected with T. cruzi. Taking these findings together, it is reasonable to assume that TNF-α can be the direct mediator of mortality due to a toxic shock-like systemic inflammatory response observed among animals infected with CL Brener and, to a lesser extent, with both strains (coinfected).
At the same time, it is widely recognized that inflammatory responses have a critical role in protection against infection, though they may contribute to the pathology of it. Therefore, to avoid pathological side effects, the inflammatory reaction induced during immune responses must be tightly regulated. In tune with this, wild-type C57BL/6 mice infected with T. cruzi survived, but IL-10−/− mice with the same genetic background presented a high mortality rate, despite presenting low parasitemia levels and high systemic production of pro-inflammatory cytokines (IFN-γ, IL-12, and TNF-α) during the acute phase of infection [34], [36]. These findings show that IL-10, an anti-inflammatory cytokine, has a critical role in control of the immune response during experimental T. cruzi infection. In this study, we observed that although coinfected animals presented high levels of serum TNF-α during the acute phase of infection, the potential toxic effects of TNF-α were counterbalanced by the production of significant serum levels of IL-10. This resulted in a low and significant TNF-α/IL-10 ratio that may have contributed to the lower mortality rate and to the higher survival time observed among coinfected animals In fact, there are several reports on the immuno-modulatory role of IL-10 in infectious diseases including Chagas disease. In canine infection by T. cruzi, the development of chronic cardiomyopathy correlates with high levels of IFN-γ and TNF-α and low levels of IL-10 [37]. In human Chagas disease as well the presence of a polymorphic allele of IL-10 gene, which results in lower expression of this cytokine is associated with cardiomiopathy and a severe form of the disease [5]. Moreover, a study on cerebral malaria showed recently that coinfection of mice with non-lethal Plasmodium berghei XAT suppressed experimental cerebral malaria caused by infection with Plasmodium berghei ANKA. The modulatory effect of the coinfection was abolished in IL-10-deficient mice clearly showing the central role of IL-10 in inhibiting the inflammatory cytokines IFN-γ and TNF-α involved in brain damage [38].
In addition to the reduction in the ratio of TNF-α to IL-10, we also found a significant decrease in the CCL2/IL-10 ratio in serum samples from animals infected with CL Brener in the presence of JG. CCL2 (MCP-1) is another important pro-inflammatory mediator characterized as a monocyte-specific chemoattractant that also attracts NK cells and T lymphocytes. It is mainly produced by macrophages in response to a wide range of cytokines such as IL-6, TNF-α and IL-1β, but can upon stimulation also be produced by a variety of cells, such as fibroblasts and endothelial cells. CCL2 is secreted in the course of T. cruzi infection and participates in T. cruzi uptake and activation of trypanocidal activity in macrophages. Paiva et al. (2009) showed that mononuclear cells from T. cruzi-infected CCL2−/− mice (in contrast to WT mice) do not form heart focal infiltrates. In this case, the parasite burden is greater, and tissue infiltrates are composed of less-activated CD8 lymphocytes and macrophages, which are essential to control parasite growth [39]. In the present study, we found high levels of CCL2 among mice single-infected with CL Brener, which can explain, at least in part, the intense myocarditis characterized by inflammatory infiltrate (predominantly constituted of mononuclear cells) observed among these animals. Reduction in the CCL2/IL-10 ratio in the mice may have also contributed to controlling the inflammatory reaction in the heart and the improved survival of the coinfected mice.
Therefore, our results suggest that production of IL-10, a key element in the control of tissue damage triggered by exacerbated inflammatory response during the course of infection, elicited by coinfection with JG and CL Brener may have an important role in modulation of heart inflammation and survival. Flow cytometry analysis of spleen cell subpopulations producing IL-10 revealed that frequency of IL-10-producing MAC-3+ and CD4+ T cells were both elevated in coinfected mice when compared to single-infected ones.
In conclusion, our work reinforces that differential outcomes of T. cruzi infection can be influenced by the complexity of the infecting T. cruzi population and parasite load, as well as by factors related to regulation of acute inflammatory response that are essential for protection against infection, but may also contribute to pathology.
|
10.1371/journal.pgen.0030094 | Determinants of Cell- and Gene-Specific Transcriptional Regulation by the Glucocorticoid Receptor | The glucocorticoid receptor (GR) associates with glucocorticoid response elements (GREs) and regulates selective gene transcription in a cell-specific manner. Native GREs are typically thought to be composite elements that recruit GR as well as other regulatory factors into functional complexes. We assessed whether GR occupancy is commonly a limiting determinant of GRE function as well as the extent to which core GR binding sequences and GRE architecture are conserved at functional loci. We surveyed 100-kb regions surrounding each of 548 known or potentially glucocorticoid-responsive genes in A549 human lung cells for GR-occupied GREs. We found that GR was bound in A549 cells predominately near genes responsive to glucocorticoids in those cells and not at genes regulated by GR in other cells. The GREs were positionally conserved at each responsive gene but across the set of responsive genes were distributed equally upstream and downstream of the transcription start sites, with 63% of them >10 kb from those sites. Strikingly, although the core GR binding sequences across the set of GREs varied extensively around a consensus, the precise sequence at an individual GRE was conserved across four mammalian species. Similarly, sequences flanking the core GR binding sites also varied among GREs but were conserved at individual GREs. We conclude that GR occupancy is a primary determinant of glucocorticoid responsiveness in A549 cells and that core GR binding sequences as well as GRE architecture likely harbor gene-specific regulatory information.
| The glucocorticoid receptor (GR) regulates a myriad of physiological functions, such as cell differentiation and metabolism, achieved through modulating transcription in a cell- and gene-specific manner. However, the determinants that specify cell- and gene-specific GR transcriptional regulation are not well established. We describe three properties that contribute to this specificity: (1) GR occupancy at genomic glucocorticoid response elements (GREs) appears to be a primary determinant of glucocorticoid responsiveness; (2) the DNA sequences bound by GR vary widely around a consensus, but the precise sequences of individual GREs are highly conserved, suggesting a role for these sequences in gene-specific GR transcriptional regulation; and (3) native chromosomal GREs were generally found to be composite elements, comprised of multiple factor binding sites that were highly variable in composition, but as with the GR binding sequences, highly conserved at individual GREs. In addition, we discovered that most GREs were positioned far from their GR target genes and that they were equally distributed upstream and downstream of the target genes. These findings, which may be applicable to other regulatory factors, provide fundamental insights for understanding cell- and gene-specific transcriptional regulation.
| The great challenge of metazoan transcriptional regulation is to create specialized expression pathways that accommodate and define myriad contexts, i.e., different developmental, physiological, and environmental states in distinct organs, tissues, and cell types. This is achieved by a network of transcriptional regulatory factors, which receive and integrate signaling information and transduce that information by binding close to specific target genes to modulate their expression. For example, the glucocorticoid receptor (GR) associates selectively with corticosteroid ligands produced in the adrenal gland in response to neuroendocrine cues; the GR-hormone interaction promotes GR binding to genomic glucocorticoid response elements (GREs), in turn modulating the transcription of genes that affect cell differentiation, inflammatory responses, and metabolism [1,2]. Expression profile analyses have identified glucocorticoid responsive genes in different cell types [3,4], and it is striking that there is only modest overlap in glucocorticoid-regulated gene sets between two cell types. The mechanisms by which GR selectively regulates transcription in cell-specific contexts are not well established.
An intriguing feature of GREs and other metazoan response elements is that their positions relative to their target genes are not tightly constrained [5,6]. Although certain metazoan response elements have been described that operate from long range, most searches for such regulatory sequences have nevertheless focused for technical reasons on restricted zones just upstream of promoters, where prokaryotic and fungal elements reside. Thus, the GRE for interleukin-8 (IL8) is just upstream of the promoter [7], whereas the tyrosine aminotransferase GRE resides at −2.5 kb [8]. Recent, more systematic searches for response elements have revealed dramatic examples, such as an estrogen response element 144 kb upstream from the promoter of the NRIP gene [9], and an intragenic region 65 kb downstream from the Fkbp5 promoter that appears to serve as an androgen response element [10]. It has been suggested that long-range regulatory mechanisms are likely to facilitate and promote regulatory evolution [11]. However, it has not been determined whether the position of a response element relative to its target gene is functionally significant.
Evidence from numerous anecdotal, gene-specific studies indicates that native response elements are typically composite elements that encompass distinct sequence motifs recognized by two or more regulatory factors. In turn, the bound factors recruit non-DNA binding coregulatory factors, forming functional regulatory complexes that remodel chromatin and modify the activity of the transcription machinery. In this scheme, the structure and activity of the regulatory complex at a given response element would be specified by at least three determinants: the sequence motifs comprising the response element; the availability of those sequences for factor binding; and the availability and activity levels of regulatory factors present in the cell. For example, primary GREs, defined as those at which GR occupancy is required for glucocorticoid-responsive regulation, are a diverse family of elements that bind GR together with an array of additional factors defined by the above three determinants. Such composite response elements provide a powerful driving force for combinatorial regulation [2,12], vastly increasing the capacity of a single factor to assume multiple regulatory roles. Indeed, the mere presence of GR in a regulatory complex is not sufficient for glucocorticoid regulation [7]. It is not known, however, whether such “nonproductive” binding by GR is common, or if instead GR occupancy is a strong indicator of GRE function.
GR binds to a family of related sequences that defines a consensus motif: an imperfect palindrome of hexameric half sites separated by a three-bp spacer [13–15]. Within those 15-bp core GR binding sequences, a few positions are nearly invariant, whereas a substantial proportion can be altered with little effect on GR binding affinity [16]. However, the functional consequences of such “permitted” sequence variations are unknown. GR can mediate a range of regulatory processes within a single cell type, including activation and repression of specific genes [4,17]. These findings, together with the results of biochemical and structural studies, raise the possibility that the core GR binding sequences might themselves serve as distinct “GR ligands,” allosterically affecting GR structure to produce distinct GR functions [18,19]. Studies of other regulatory factors have led to similar conclusions [20,21]. If different core GR binding sequences indeed produce GRE-specific (and therefore target gene-specific) regulatory activities, we could expect that the core GR binding sequence associated with a given target gene would be strongly conserved through evolution, whereas the collection of core GR binding sequences across different genes would vary substantially. Analogously, if the architecture of composite GREs, i.e., the arrangements of additional sequence motifs surrounding the core GR binding site, are also important for gene-specific regulation, we would expect flanking sequences surrounding the core GR binding site to also be evolutionarily conserved in a GRE-specific manner but not across GREs within a single genome. Neither of these notions has been examined.
In the present work, we sought to define and characterize a set of GREs in A549 human alveolar epithelial cells. Thus, we determined in A549 cells the presence of GR at specific GREs close to genes that are steroid regulated across a range of cell types. We assessed whether the GR-occupied GREs were limited mainly to genes that are GR regulated in A549 and measured within and between species the conservation of GRE sequences, architecture, and genomic positions.
To assess the correlation of GR occupancy with glucocorticoid responsiveness, we examined GR binding at three classes of genes in A549 human lung carcinoma cells: first, genes regulated by GR in A549 cells; second, genes regulated by GR in U2OS human osteosarcoma cells but not in A549; third, genes regulated by GR or the androgen receptor (AR) in cells other than A549 or U2OS. The AR-responsive genes were of interest because AR is closely related to GR and shares similar DNA-binding specificity in vitro [14,16,22]. The first two classes of genes were identified in our lab using expression microarrays, whereas the third class was compiled from our own microarray data and from published reports of others [3,4,23,24]. Both positively and negatively regulated genes were included; together the three classes comprised 548 candidate GR target genes. By examining these genes for GR binding in A549 cells, we could determine if GR occupancy in vivo is restricted only at genomic sites of genes actually regulated by glucocorticoids in A549 cells; alternatively, GR might also bind at genes that are not under glucocorticoid control in A549, but are regulated by GR or AR in other cells.
To identify GR binding regions (GBRs), we used chromatin immunoprecipitation-microarray (ChIP-chip) to interrogate 100-kb genomic segments centered on the transcription start sites (TSSs) of our set of 548 genes. This ~55-Mb sample of the genome also included or impinged upon an additional 587 genes not previously reported to be regulated by GR; thus, we assessed GR occupancy in the vicinity of more than 1,000 genes. Immunoprecipitated chromatin samples from A549 cultures treated for one hour with the synthetic glucocorticoid dexamethasone (dex) (100 nM) or ethanol were hybridized onto the ChIP-chip tiling arrays. Independent biological replicates were hybridized onto two separate arrays, and GBRs were identified using the SignalMap detection program; we detected a 3.4% false positive rate for the GBRs found in both arrays as assessed by conventional ChIP and quantitative PCR (qPCR) analysis. Importantly, we did not detect GR occupancy at 22 regions that showed no GR binding in the arrays (unpublished data). The ChIP-chip experiments revealed a total of 73 GBRs adjacent to 61 genes (Table 1), which were validated by GR ChIP and qPCR analysis (Figure 1A). In addition to identifying GBRs previously detected by conventional ChIP, our experiments revealed novel GBRs in regions not searched in prior studies. For example, two known promoter proximal GBRs at SCNN1A [25] and SDPR [3] were confirmed in the ChIP-chip arrays as well as newly observed GBRs +3 kb and −20 kb from the SCNN1A and SDPR TSSs, respectively (Figure 1B).
Of the 73 A549 GBRs identified in the present study, 64 (88%) were associated with genes regulated by GR in those cells (Table 1). Although the remaining nine GBRs may be nonfunctional, they may mediate responses under different biological conditions. Notably, 27% of the genes that were glucocorticoid responsive specifically in A549 but not in U2OS cells were associated with a GBR, whereas only 1.9% of the genes responsive to glucocorticoids in U2OS but not in A549 contained A549 GBRs (Figure 2). Similarly, only 1.8% of the genes that were glucocorticoid or androgen responsive in other cells and only 0.3% of the genes that were sampled by the ChIP-chip arrays but were not steroid regulatory targets were associated with A549 GBRs (Figure 2; Table S1). Thus, GR occupancy in A549 cells is generally restricted to genes that are actually regulated by glucocorticoids in those cells; specifically, GR is rarely bound in A549 cells at genes responsive to glucocorticoids in other cells. We conclude that GR occupancy is a major determinant of glucocorticoid responsiveness in A549 cells at the genes assessed in this study.
To test whether the A549 GBRs can confer glucocorticoid-directed transcriptional responses, we cloned 500-bp DNA fragments encompassing the GBRs into luciferase reporter plasmids. Of the 20 GBRs randomly selected from the 73 GBRs identified in this study, 19 were dex responsive in A549 cells as assessed by reporter analysis (Figure 3A). We define primary GREs (denoted here simply as GREs) as genomic regions that are occupied in vivo by GR and confer glucocorticoid-regulated transcription in transfected reporters. Although the reporter analyses do not prove that the identified elements are functional in their native contexts (see Discussion), they establish that the 500-bp fragments tested harbor sufficient information for GR to regulate transcription. Thus, we shall refer to the GBRs henceforth as GREs.
We determined the positions of the A549 GREs relative to TSSs of their respective target genes (Figure 4A). For this analysis, the GREs were assigned to the nearest gene responsive to dex in A549 cells. Surprisingly, we found that 45% of the GREs were located downstream of the TSSs, suggesting that GR exhibits transcriptional regulation without a significant preference for regions upstream or downstream of TSSs (Table 1). Figure 4B summarizes the distribution of promoter proximal (within 5 kb from the TSS) and distal GREs (farther than 10 kb from the TSS). Strikingly, 63% of the GREs were distal, whereas only 31% of them were promoter proximal (Figure 4B). Mammalian response elements are commonly thought to reside upstream and proximal to their cognate promoters; thus, identification of GREs and response elements in general have mainly focused on these regions. Importantly, Figure 4B demonstrates that only a small fraction of the GREs (17%) identified in this study was positioned within these regions. These results indicate that GREs are just as likely to be located downstream of the TSSs and that the majority operate remotely from their target promoters, at least by linear DNA distance.
Our finding concerning GRE distribution is supported by two indirect analyses using nuclease sensitivity and sequence conservation. Sabo et al. found that DNAse I hypersensitive sites, indicative of chromatin-bound factors, are broadly distributed with a majority located >10 kb from the nearest TSS [26]. Furthermore, Dermitzakis et al. showed that conserved nongenic sequences (CNGs), ungapped 100-bp fragments with at least 70% identity between human and mouse that are presumed factor-binding regions, have no significant preference for promoter proximal regions [27,28]. As expected [29], we found that GR occupancy was correlated with DNAse I-hypersensitive cleavage at both promoter proximal (1.3, 1.5, 12.1, and 16.1) and distal GREs (2.4, 5.1, 6.1, 6.3, 7.3, and 20.2) (Figure 5A). In addition, by aligning the human GRE sequences with the corresponding regions in the mouse genome, we found that 23 of the GREs correspond to CNGs (Figure 5B). Moreover, GR occupancy and glucocorticoid responsiveness for several of these GREs/CNGs (6.4, 12.1, 5.1, 6.1, 6.2, 10.5, X.1, and X.2) were maintained in mouse cells (see Figure 6A, 6B). Thus, by testing the GREs identified in our study, we were able to provide direct support for the notion that DNAse I hypersensitive sites and CNGs serve as regulatory elements [26,28].
Native GREs, defined as naturally evolved genomic elements that confer glucocorticoid regulation on genes in their chromosomal contexts, are likely to be “composite elements,” made up of binding sites for GR together with multiple nonreceptor regulatory factors [2]. To assess whether we could detect such complex architecture, we used computational approaches (Bioprospector and MobyDick) to survey the 500-bp GRE-containing fragments for sequences related to known regulatory factor binding sites [30–32]. The most prominent motif found, present in 68% of the GRE sequences, was a series of imperfect palindromes similar to known core GR binding sites (Figure 3B). Potentially, GR may interact with the remaining 32% of GREs through other recognition motifs or through tethering to other factors [7]. Mutagenesis of computationally predicted core GR binding sites decreased or completely abolished dex stimulation for each of 13 randomly tested sites, validating this approach for identifying functional core GR binding sequences (Figure 3A). Some GREs, such as 6.2, 7.2, and 7.3, contained multiple GR binding sites; we found that reporters mutated at only one of those sites retained residual dex inducible activity. These experiments imply that most of the core GR binding sites identified in our computational analysis are functional.
In addition, we found that motifs similar to AP-1, ETS, SP1, C/EBP, and HNF4 binding sequences were enriched in the 500-bp GRE fragments (Figure 3B). For example, motifs resembling AP-1 and C/EBP binding sites were identified in the GRE of the IL8 gene. Importantly, the AP-1 binding site is known to be crucial for regulation of IL8 by the AP-1 factor [33]; similarly, C/EBPα enhances transcription of a reporter spanning the IL8 GRE region [34]. Thus, as with GR binding sequences, our computational analysis was capable of discovering functional nonreceptor binding sites. Detection of multiple factor binding sites within the GRE sequences is consistent with the hypothesis that native GREs are typically composite response elements that recruit heterotypic complexes for combinatorial control [2].
To estimate the extent of GRE conservation, we measured sequence identity in human and mouse across 4-kb regions centered on the core GR binding sites (see Figure 3C legend and Materials and Methods) averaged across 50-bp windows; a similar (albeit higher resolution) pattern was obtained with 15-bp windows (unpublished data). Strikingly, we found that flanking sequences roughly 1 kb surrounding the core GR binding sites were conserved relative to background (Figure 3C). This elevated evolutionary conservation implies that these segments are biologically functional, not only in reporter constructs (Figure 3A), but also in their native chromosomal contexts, further supporting the view that native GREs are composite elements.
We next sought to examine in detail the extent of sequence conservation of some of the individual core GR binding sequences and GREs that we had identified in our study. We chose a subset of 12 human GREs that are occupied by GR both in another species, mouse, and in another cell type, C3H10T1/2 mesenchymal cells (Figure 6A). Consistent with the correlation between GR occupancy and glucocorticoid responsiveness (Figure 2; Table 1), we confirmed that several of these genes (Fkbp5, Ddit4, Gilz, MT2A, and Sgk) were indeed dex inducible in the C3H10T1/2 cells (Figure 6B). These 12 GREs resided at very different locations relative to the TSSs of their human target genes (ranging from 0.1 kb to 86 kb) (Table S2); remarkably, however, each locus was approximately maintained in the mouse genome (Table S2). This finding suggests that the positions of individual GREs may be integral to their regulatory functions.
We then examined the extent of conservation of the 15-bp core GR binding sites within the GRE set defined above. As anticipated, the 12 core GR binding sites from the different human GREs differed substantially, with only five invariant positions across the 15-bp sequences (Figure 6C); for example, the binding sites of human GRE 5.1 and human GRE 10.3 match at only seven positions. In striking contrast, we found that the core GR binding site sequences within the individual GREs were highly conserved among human, mouse, dog, and rat (Figure 6C); for example, the core GR binding sequence at GRE 10.5 is identical in all four evolutionarily distant species.
Finally, we compared in human and mouse the patterns of conserved sequences flanking the core GR binding sites, which provide “architectural signatures” of individual GREs. We found that the patterns of sequence conservation differed dramatically among the different GREs (Figure 6D; Figure S3). For example, GRE X.1 contains conserved sequence elements at −900, −500, and +600bp, whereas GRE X.2 displays no conservation at those positions (Figure 6D). Although the functional significance of the conserved regions has yet to be tested (for example, we have not ruled out incidental overlaps with conserved noncoding expressed regions), the conserved regions are likely to correspond to regulatory or structural motifs. As predicted by these findings, pair-wise calculations of sequence identity of different human GREs (using a 15-bp window centered on the core GR binding sites) demonstrated that sequences flanking the core GR binding sites varied extensively among human GREs (Figure S4). Thus, the overall family of GREs is broadly divergent in sequence and organization, but each individual GRE retains a distinctive signature of conserved sequences, suggesting that each corresponds to a composite GRE that is functionally distinct.
We set out to examine the organization and function of genomic elements responsible for transcriptional regulation by GR. Our study yielded five conclusions: (1) GR occupancy at a GRE is generally a limiting determinant of glucocorticoid response in A549 cells; (2) the core GR binding sequences conform to a consensus that displays substantial GRE-to-GRE variation as anticipated, but the precise binding sequences at individual GREs are highly conserved through evolution; (3) GREs appear to be evenly distributed upstream and downstream of their target genes; (4) most GREs are positioned at locations remote from the TSSs of their target TSSs; and (5) native GREs are commonly composite elements, comprised of multiple factor binding sites, and they are individually conserved in position and architecture yet very different from each other. We shall consider the implications of these conclusions in turn.
We began by surveying more than 1,000 genes, with half of them candidates for steroid regulation, and a specific subset known to be GR-regulated in A549 cells. We found that GR occupancy of A549 GREs correlated strongly (nearly 90%) with genes that are glucocorticoid responsive in A549, suggesting that GR binding is generally a limiting determinant for response in these cells. In a small number of cases, we observed GR occupancy close to genes that were GR-unresponsive in A549 cells, but were steroid regulated in other cells [4] (E. C. Bolton and K. R. Yamamoto, unpublished results). This implies that GR occupancy at these genes likely reflects bona fide response element binding, but that GR binding is not a limiting factor for glucocorticoid regulation of this minority class of genes in A549 cells. Collectively, our data suggest that restriction of GR occupancy in A549 cells may be responsible for much of the cell-specific GR-mediated regulation in these cells. The mechanisms of occupancy restriction could be positive or negative mechanisms, such as accessory factors that stabilize GR binding, or chromatin packaging that precludes it. Although the strong correlation between GR occupancy and glucocorticoid responsiveness in A549 cells seems likely to hold in other cell types, it is conceivable that responsiveness may be determined differently in other cell types. Thus, it will be interesting to examine cell-specific GR regulation in other cells to complement the observations made in A549 cells. It is intriguing that one component, GR, within such varied and complex machineries would so strongly predominate as a determinant of transcriptional regulation in A549 cells. It will be interesting to examine regulatory complexes that mediate other types of responses (e.g., heat shock and DNA damage) to assess whether response element occupancy by a single factor in each class is a dominant determinant of responsiveness.
We examined sequence conservation of a set of GREs that are occupied by GR both in human lung epithelial cells and in mouse mesenchymal stem cells. We found that the 15-bp core GR binding sequences varied greatly among the different GREs (Figure 3B), whereas the sequences of the individual binding sites were nearly fully conserved across four mammalian species (Figure 6C). Crystallographic studies demonstrate that GR makes specific contacts with only four bases of the 15-bp core binding sequence [35], yet every position, including the “spacer” between the hexameric half sites, appears to be equivalently conserved. This indicates that the binding sequences serve functions in addition to merely localizing GR to specific genomic loci and instead may carry a regulatory code that affects GR function. Leung et al. reported similarly strong evolutionary conservation of individual κB binding sequences [36]. Indeed, Luecke and Yamamoto showed that GR directs distinct regulatory effects when tethered to NFκB at two κB response elements that differ by only one base pair [7]. Thus, one interpretation of our data findings is that factor binding sites may serve as allosteric effectors [19] in which individual binding sequences convey subtle conformational differences to specify distinct factor functions. Conceivably, this hypothesis might also explain why GR predominates as a limiting determinant of responsiveness, because factors that read allosteric regulatory codes might specify the rules for assembly of GRE-specific and thus gene-specific regulatory complexes.
To characterize the architecture of GREs, we took several approaches. In unbiased computational analyses, we identified enriched sequence motifs within 500-bp segments encompassing core GR binding sites. Sequence motifs resembling binding sites for GR, AP-1, ETS, SP1, C/EBP, and HNF4 were overrepresented relative to a background of unbound GR regions, consistent with the notion that native GREs are composite elements. For most of these GREs, the role of these factors in GR transcriptional regulation remains to be tested, but it is notable that ETS-1, SP1, and HNF4 have been shown at other genes to augment glucocorticoid responses [37–39]. Moreover, Phuc Le et al. [40] described motifs resembling AP1 and C/EBP binding sites within certain mouse GREs and showed that nearly half of the GREs predicted to encompass C/EBP binding sites did indeed bind C/EBPβ [40]. These findings further the view that our computational analysis can infer factors that potentially interact with GR at GREs. Using a similar approach, Carroll et al. [9] and Laganiere et al. [41] have interrogated estrogen response elements and identified FOXA1 as a factor playing an important role for both estrogen receptor binding and transcriptional activity. Thus, we anticipate that the factors that occupy the GR composite elements may interact physically, functionally, or both, thereby affecting binding as well as regulatory activity. Indeed, an averaged comparison of human and mouse sequences flanking core GR binding sites revealed that a region of approximately 1 kb was conserved above the background level (Figure 3C), suggesting that native composite GREs are extensive and typically may contain numerous factor binding sites. Interestingly, individual GREs displayed distinctive patterns of sequence conservation extending from the core GR binding sites (Figure 6D; Figure S3). These GRE signatures likely reflect conservation of various sequence motifs at different positions within each element, producing GRE-specific (and therefore gene-specific) architecture that likely creates distinct regulatory effects.
To investigate the distribution of regulatory elements relative to their target genes, we monitored GR occupancy across 100 kb regions centered on the TSSs of glucocorticoid responsive genes. We found that GREs were evenly distributed upstream and downstream of their target genes with the majority located >10 kb from their target promoters; other metazoan regulatory factors, such as estrogen receptor (ER) and STAT1, have similarly been reported to act from sites remote from their target genes [9,42–45]. In contrast to these factors, E2F1 was shown to mainly bind promoter proximal regions [42]; others have used computational approaches to infer factor binding sites close to promoters, but these have not been experimentally confirmed [46]. In parallel with our findings, Carroll et al. reported that only 4% of estrogen receptor ER binding regions was mapped within −800 bp to +200 bp from TSS of known genes from RefSeq [43]. Our data demonstrated that 9% of GBRs were positioned at this location. These studies together imply that steroid receptors, which include estrogen receptor and GR, in general regulate transcription from remote locations. Interestingly, we found that the positions of individual GREs were generally conserved across species (Table S2), implying that GRE position may be functionally important for target gene regulation. In any case, our findings differ dramatically from those in prokaryotes and fungi, where transcriptional regulatory elements are promoter proximal. It has been suggested that these two broad classes of regulatory mechanisms, so-called long range and short range, are mechanistically and evolutionarily related, and that long range control might facilitate regulatory evolution [11]. As predicted by that model, distal elements, far from target genes as measured by linear DNA distance, may operate in close proximity with their target promoters in 3-D space. For example, Carroll et al. detected an interaction between the NRIP-1 promoter and its distal estrogen response element [9]. It will be interesting to determine whether response element location (i.e., promoter proximal versus distal) is somehow related to mechanism or to physiological network.
Remote response element locations can complicate assignment of cognate target genes. An extreme example is olfactory receptor gene expression, which is governed by a regulatory element that can operate on target genes located on different chromosomes [47]. In this study, we assigned the GREs to the nearest RefSeq gene responsive to dex in A549 cells. In other contexts, these GREs may be nonfunctional or may operate on genes other than those assigned in A549 cells (Table 1). Clearly, unequivocal assignment of a GRE to a given target gene will require genetic manipulations not readily accessible in mammalian cells at present. It is encouraging, however, that GR occupancy of GREs correlated strongly with glucocorticoid responsiveness of adjacent genes, supporting the view that these are bona fide direct GR targets (Figure 2; Table 1). In fact, when these genes were subjected to Gene Ontology analysis, we found that they were enriched in cell growth and immune responses (unpublished data), two biological processes regulated by GR in A549 cells [48,49]. We found GR occupancy at genes up- and down-regulated in response to dex, consistent with GR serving either as activator or repressor in different contexts. At present, we cannot assess the significance of the finding that GR was detected at GREs adjacent to activated genes versus repressed genes at a 6:1 ratio in A549 cells; whether this difference reflects differences in GRE occupancy, epitope accessibility, crosslinking efficiency, or other variables has not been determined.
Genomic response elements orchestrate transcriptional networks to mediate cellular processes for single- and multicellular organisms. The present study advanced our understanding of the organization, evolution, and function of GREs and at the same time raised a series of interesting questions. Among the more intriguing: How is GR occupancy restricted to a small subset of potential GREs in a given cell context? What is driving the strong conservation of virtually every base pair within the core GR binding sequence at individual GREs? Addressing these and other questions raised in our study will contribute additional new insights about gene regulation by GR and by other regulatory factors.
A549 and C3H10T1/2 cells were grown in DMEM supplemented with 5% or 10% FBS, respectively, in a 5% carbon dioxide atmosphere. Before hormone treatment, media was replenished with DMEM containing charcoal stripped FBS, which depletes endogenous steroids. Plasmid PGL4.10 E4TATA (generously provided by Yuriy Shostak) was created by insertion of the E4TATA minimal promoter into pGL4.10 vector (Promega, http://www.promega.com). The 20 reporters tested (Figure 3A) represent randomly chosen GRE fragments. The QuikChange kit (Stratagene, http://www.stratagene.com) was used for reporter mutagenesis. The 13 core GR binding sites that were mutated in the reporters (Figure 3A) were also randomly chosen based on success of mutagenesis. GBR-containing DNA fragments (500 bp) were amplified by PCR and subcloned into pGL4.10 E4TATA using KpnI and XhoI sites (see Table S3 for primer sequences). A549 cells were grown in a 48-well plate and cotransfected with 19 ng of the reporter constructs, 10 ng pRL Luc (Promega), and 38 ng pCDNA3 hGR (human GR expression vector) using Lipofectamine 2000 (Invitrogen, http://www.invitrogen.com). After overnight transfection, cells were treated with hormone, harvested, and luciferase activity was measured as described for the dual luciferase reporter system (Promega) using a Tecan Ultra Evolution plate reader (Tecan, http://www.tecan.com).
ChIP assays were performed as described [7] with the following modifications. The chromatin samples were extracted once with phenol-chloroform and purified using a Qiaquick column as recommended by the manufacturer (Qiagen, http://www1.qiagen.com). The ligation-mediated PCR (LMPCR) process was adapted from Oberley et al. [50]. We used 3.5–20 ng of amplicon for real-time qPCR analysis, and data were normalized to Hsp70 (see Table S4 for primer sequence). Human and mouse DNA sequences were retrieved from University of California Santa Cruz (UCSC) Genome Browser (http://genome.ucsc.edu) NCBI Build 35, and qPCR primers were designed using Primer3 [51]. For the array, ~50 kb upstream and downstream regions were tiled with isothermal 50 mer oligos (spaced on average of every 54 bp apart) relative to the TSSs of the investigated target genes. Where 100-kb regions overlapped, the surrounding genomic region was tiled further bidirectionally. ChIP samples from (final concentration, 0.01% ethanol) or dex-treated A549 cells were labeled with Cy3 or Cy5, hybridized onto the arrays, and relative signal intensities were measured by NimbleGen (http://www.nimblegen.com). SignalMap was utilized to find peak enrichments with both window threshold detection (500-bp peak window size, 25% of Peak Threshold) and second derivative peak detection (500-bp peak window size, 20 bp smooth step, 25% peak threshold) (NimbleGen).
The RNA isolation, reverse transcription, and qPCR steps were performed as previously described [4]. Primers for cDNA amplification are displayed in Table S4.
The experiments were adapted from previous described protocol [52] with the following modifications. Briefly, nuclei from A549 cells treated with vehicle or dex for 1 h were treated with 6.25–200 units/ml of DNAse I (Qiagen) for 5 min at room temperature. The reaction was stopped and treated with Proteinase K for 1 h at 65 °C. The DNA samples were extracted once with 1:1 phenol-chloroform and further purified using MiniPrep columns (Qiagen). The samples were subjected to qPCR analysis to determine the relative amount of cleaved product (see Table S4 for primer sequences), which was converted to percent DNAse I cleavage.
For computational analysis of enriched motifs, all repeat-masked DNA sequences were downloaded from the UCSC genome browser (NCBI Human Build 35). BioProspector analysis was initially performed using nucleotide widths (w) 14 and 16 on GREs to identify GR binding sites and the top motifs were masked to identify other motifs [32]. For MobyDick analysis, both the human and the human/mouse aligned sequences were used as inputs to identify enriched motifs [30,31]. Similar motifs were clustered using CAST [53–55]. All p-values for enrichment were Bonferroni corrected to identify putative factor binding sites [55]. The top Bioprospector w14 position weight matrix (PWM) was used to score GREs for putative GR binding sites with a false positive rate of less than 10%. This upper bound was calculated from randomly sampling unbound GR regions (Figure S2).
We built position weight matrices (PWMs) of those motifs with p-values less than 0.05, which were used to measure similarity to known binding sites in TRANSFAC [56]. We measured the distance between the PWMs and those representing binding sites for known regulatory factors using relative entropy (Kullback-Liebler divergence) with a cutoff of less than 6.0 to associate motifs with putative regulatory factor binding sites. The known binding site matrices were obtained from TRANSFAC professional version 9.3.
The human–mouse conservation score was calculated as described [9] using a 50-mer window for 50 sequences containing a putative GR binding site based on our computational and experimental analysis (Figure S2 and Figure 3A). The conservation score was calculated as number of bp matches minus the number of bp deletions or insertions divided by the bp window size. We centered each alignment based on the highest scoring putative GR binding site in human and expanded equally on each side of the binding site to a total length of 4 kb. The background level was calculated by taking the average of all conservation scores across the 4-kb region. The human(hg)/mouse(mm) genome alignments were downloaded from Vista (http://pipeline.lbl.gov/cgi-bin/gateway2). |
10.1371/journal.pgen.1004121 | Differential Effects of Collagen Prolyl 3-Hydroxylation on Skeletal Tissues | Mutations in the genes encoding cartilage associated protein (CRTAP) and prolyl 3-hydroxylase 1 (P3H1 encoded by LEPRE1) were the first identified causes of recessive Osteogenesis Imperfecta (OI). These proteins, together with cyclophilin B (encoded by PPIB), form a complex that 3-hydroxylates a single proline residue on the α1(I) chain (Pro986) and has cis/trans isomerase (PPIase) activity essential for proper collagen folding. Recent data suggest that prolyl 3-hydroxylation of Pro986 is not required for the structural stability of collagen; however, the absence of this post-translational modification may disrupt protein-protein interactions integral for proper collagen folding and lead to collagen over-modification. P3H1 and CRTAP stabilize each other and absence of one results in degradation of the other. Hence, hypomorphic or loss of function mutations of either gene cause loss of the whole complex and its associated functions. The relative contribution of losing this complex's 3-hydroxylation versus PPIase and collagen chaperone activities to the phenotype of recessive OI is unknown. To distinguish between these functions, we generated knock-in mice carrying a single amino acid substitution in the catalytic site of P3h1 (Lepre1H662A). This substitution abolished P3h1 activity but retained ability to form a complex with Crtap and thus the collagen chaperone function. Knock-in mice showed absence of prolyl 3-hydroxylation at Pro986 of the α1(I) and α1(II) collagen chains but no significant over-modification at other collagen residues. They were normal in appearance, had no growth defects and normal cartilage growth plate histology but showed decreased trabecular bone mass. This new mouse model recapitulates elements of the bone phenotype of OI but not the cartilage and growth phenotypes caused by loss of the prolyl 3-hydroxylation complex. Our observations suggest differential tissue consequences due to selective inactivation of P3H1 hydroxylase activity versus complete ablation of the prolyl 3-hydroxylation complex.
| The prolyl 3-hydroxylase complex serves to hydroxylate a single residue in type I collagen and also serves as a collagen chaperone. The complex is comprised of prolyl 3-hydroxylase 1, cartilage associated protein, and cyclophilin B. Mutations have been identified in the genes encoding the complex members in patients with recessive Osteogenesis Imperfecta. Recent data suggest that prolyl 3-hydroxylation of collagen does not alter the stability of collagen but may rather mediate protein-protein interactions. Additionally, the collagen chaperoning function of the complex is an important rate limiting step in the modification of type I collagen. Irrespective of whether patients with mutations in the genes encoding the members of the prolyl 3-hydroxylase complex have hypomorphic or complete loss of function alleles, either circumstance leads to the loss of both functions of the prolyl 3-hydroxylase complex. Thus, it is unknown how collagen chaperoning and/or hydroxylation affect bone and cartilage homeostasis. In this study, we generated a mouse model lacking the prolyl 3-hydroxylation activity of the complex while maintaining the chaperoning function. We found that the hydroxylase mutant mice have normal cartilage and normal cortical bone but decreased trabecular bone, suggesting that there is a differential requirement for hydroxylation in different tissues.
| Although dominant mutations in the type I procollagen genes, COL1A1 and COL1A2, account for the majority of patients with Osteogenesis Imperfecta (OI) (#166200, #166210, #166220, #259420, #259440, #610682, #610915, #610967, #610968, #613848, #613849, #613982, #614856, #615066), the disorder can also be inherited in an autosomal recessive manner [1]. A mutation in cartilage associated protein (CRTAP) (*605497) was first identified by Morello et. al. in a class of patients with recessive OI [2]. CRTAP functions in a complex with prolyl 3-hydroxylase 1 (P3H1) (*610339) and cyclophilin B (CYPB) (*123841) to 3-hydroxylate a unique proline, Pro986, of the α1(I) chain and also to chaperone collagen trimer assembly [3]–[5]. Additionally, other clade A fibrillar collagens, such as collagen type II, are similarly hydroxylated. Subsequently, mutations in both leucine and proline enriched proteoglycan (LEPRE1), encoding P3H1, and peptidylprolyl isomerase b (PPIB), encoding CYPB, were identified in other patients with recessive forms of OI [6]–[13]. Mutations in additional genes have since been identified in recessive OI, revealing novel mechanisms of disease through alterations in post-translational collagen modification, trafficking and signaling.
Knockout mice have been created for each of the three genes that encode the prolyl 3-hydroxylase complex and these mice recapitulate the phenotype observed in recessive OI patients. Crtap−/− and Lepre1−/− mice display osteochondrodysplasia, severe low bone mass, kyphosis, rhizomelia and collagen fibrils with irregular diameter [2], [14]. Although the Ppib−/− mice do not have rhizomelia, they also have a low bone mass phenotype with kyphosis and wider collagen fibrils [15]. Together, these findings highlighted the importance of this complex in collagen modification and maintenance of bone mass.
Since null mutations in either LEPRE1 or CRTAP can result in recessive forms of Osteogenesis Imperfecta (OI) with almost identical features, P3H1 and CRTAP were hypothesized to stabilize each other [16]. Indeed, western blot and immunofluorescence studies demonstrated absence of both CRTAP and P3H1 in fibroblasts isolated from patients carrying either LEPRE1 or CRTAP mutations alone [16], [17]. These data supported the notion that CRTAP and P3H1 proteins stabilize each other in the endoplasmic reticulum.
The present study aimed to determine the relative contribution of the hydroxylation activity of the P3H1 complex to the pathogenesis of recessive OI. This important question was addressed in vitro and in vivo by introducing a point mutation in the catalytic domain of P3H1 that inactivated hydroxylase activity while preserving the protein secondary structure and the CRTAP/P3H1/CypB complex integrity.
To accomplish our goal of inactivating the hydroxylase function of P3H1, we employed an evolutionary trace algorithm to identify conserved residues that are essential for its enzymatic function. The four top ranking residues identified were the catalytic triad residues (HIS590, HIS662, and ASP592) and the 2-oxoglutarate binding residue (ARG672) (Figure 1A). The importance of these residues for hydroxylase activity was confirmed by earlier literature [18], [19]. Conversion of the catalytic triad histidines or aspartic acid to either alanine or glutamate abolished the enzymatic activity of Prolyl 4-hydroxylase activity in A. thaliana [20]. Similarly, conversion of the 2-oxoglutarate binding arginine to alanine also abolished the hydroxylase activity, suggesting that substituting the corresponding sites in P3H1 with an alanine could potentially inactivate its hydroxylase function [20]. Since alanine is a non-bulky, uncharged amino acid that can mimic the secondary structure of many other amino acids, we opted for this substitution rather than a glutamate substitution to preserve the structural integrity of P3H1.
Since the stability of the prolyl 3-hydroxylase complex is dependent on the interaction of P3H1 and CRTAP, it was important to verify that the mutation introduced into LEPRE1 did not disrupt its ability to form a stable complex with CRTAP and CYPB. To do this we first used an in vitro approach. We used immortalized patient fibroblasts carrying a loss of function mutation in LEPRE1 and tested whether the expression of 4 different LEPRE1 constructs containing alanine substitutions at H590, D592, H662, or R672 were able to restore the stability of CRTAP by immunofluorescence and immunoblot assays. We found that the mutant expression construct converting H662 to alanine (P3H1H662A) was the most effective at rescuing CRTAP compared with the others or the un-transduced LEPRE1 loss of function cells (figure 1B, C). These findings are consistent with P3H1H662A being able to interact with CRTAP and to restore it to the ER. Although P3H1H662A was not assayed for enzymatic activity, mutating the corresponding residue to an alanine in Prolyl 4-hydroxylase in A. thaliana resulted in complete inactivation of the hydroxylase activity, suggesting the P3H1H662A mutant is likely to be inactive [20].
We generated knock-in mice carrying the H662A mutation at the Lepre1 locus (Lepre1H662A/H662A) (figure S1). Importantly, we verified that CRTAP is also restored in vivo in the Lepre1H662A/H662A mice compared with wild-type littermates by confirming its presence by western blot using protein isolated from P1 calvaria (figure 2A). Since the Pro986 residue of α1(I), α1(II) and α2(V) procollagen chains is normally fully hydroxylated and complete loss of the P3H1 complex abolishes hydroxylation at these sites [2], [14], [15], [17], we analyzed the hydroxylation status of these residues in the Lepre1H662A/H662A mice to assess the in vivo enzymatic activity of P3H1H662A. Tandem mass spectrometry showed loss of prolyl 3-hydroxylation (3-Hyp) at Pro986 in the α1(I) chain and a residual 21% 3-Hyp in the α2(V) chain from bone (figure 2B, 3). Pro978 of the bone α2(V) chain remained minimally 4-hydroxylated, similar to the level observed in wild-type littermate bone (figure 3). This is in contrast to Crtap−/− mice in which 3-Hyp at Pro986 was missing but in effect replaced by 4-Hyp at Pro978 apparently as a consequence of over-modification in comparison to wild-type mice [17]. In cartilage, tandem mass spectrometry showed specifically a residual 9% 3-Hyp at Pro986 in the α1(II) chain of the Lepre1H662A/H662A mice (figure 2C), which is comparable to the 6% 3-Hyp observed in Crtap−/− mice (not shown). These findings are summarized in Table 1 together with the hydroxylation status of other partially occupied 3-Hyp sites in bone and cartilage collagens. The loss of 3-Hyp appears to be specific to Pro986 of types I, II and V collagens similar to findings in mouse models with complete loss of the P3H1 complex [2], [14], [15], [17].
Collagen cross-linking in bone was studied by determining the ratio of hydroxylysyl pyridinoline to lysyl pyridinoline (HP/LP). This ratio reflects the hydroxylation status of those triple-helical lysines at K87 and K930 in α1(I) and/or K87 and K933 in α2(I) that had participated in cross-link formation. In the Lepre1H662A/H662A mice, there was an increase in the HP/LP ratio compared to wild-type littermates indicating that there may have been an overall increase in hydroxylation at one or both of these sites. However, mass spectral analysis of linear (uncross-linked) sequences from the same helical cross-linking sites prepared by digestion either with bacterial collagenase or trypsin showed no significant differences between Lepre1H662A/H662A and wild-type bone collagen. Residue α1(I) K930 was 98% hydroxylated and non glycosylated in both genotypes and α1(I)K87 was 92% hydroxylated in wild-type and 93% in Lepre1H662A/H662A. So the difference in HP/LP ratio may be due to altered hydroxylation at the homologous sites in the α2(I) chain but we did not acquire informative peptides from the latter. Informative peptides from non cross-linking lysine sites also showed no significant differences between wild-type and Lepre1H662A/H662A bone. For example, α1(1) K174 was essentially all galactosyl Hyl in both genotypes and α2(1) K219 was 70% hydroxylated in wild-type and 77% in Lepre1H662A/H662A. Thus, no evidence of generalized over-modification was found from these site-specific mass spectral results.
The content of HP+LP in the bone collagen was not significantly altered (Table 2). Consistent with the latter observation, Lepre1H662A/H662A mice showed no significant difference in the ratio of telopeptide hydroxylysine to lysine in the extracted bone collagen α1(I) chains when compared to their wild-type littermates based on the mass spectral ratio of Hyl to Lys versions of the telopeptides not involved in cross-linking (Table 2). A caution here is that this estimate comes from a relatively minor fraction of the total matrix collagen.
At birth, the Lepre1H662A/H662A mice are indistinguishable from their wild-type littermates by gross physical appearance (data not shown). Radiographs at 3 months and 6 months of age showed normal skeletal patterning and no evidence of skeletal deformity such as kyphoscoliosis (figure 4A, not shown). Previous studies of the Crtap−/− and Lepre1−/− mice showed a disorganization of the growth plate causing smaller body size with shortening of the proximal long bone segments (rhizomelia) [2], [14]. To assess growth, the Lepre1H662A/H662A mice and their wild-type littermates were weighed weekly until 3 months of age. At all time points, the weight of Lepre1H662A/H662A mice was not statistically different from that of wild-type littermates (figure 4B). Similarly, there were no significant differences between the lengths of the femur, tibia, nor the femur/tibia ratio, thus excluding any rhizomelic defect in the Lepre1H662A/H662A mice [2], [14] (figure 4C).
Since the Lepre1−/− mice have a dysplasia of the growth plate that also affects the hypertrophic chondrocytes, we performed histology and specific staining of the hypertrophic zone with an antibody directed towards type X collagen in Lepre1H662A/H662A mice at P1 [14]. No defects were observed in the Lepre1H662A/H662A mice compared to their wild-type littermates (figure 5A, B; figure S2A, B). Additionally, we observed no difference in the width of the hypertrophic zone between the two genotypes (N = 8, p-value = NS) (figure 5C; figure S2C). Collectively, the normal growth curve, normal femur to tibia ratio and normal hypertrophic zone suggest that the cartilage in the long bones of Lepre1H662A/H662A mice is indeed normal despite of loss of Pro986 hydroxylation in type II collagen.
By assessing the femurs and spines at 3 months of age (n = 10, each genotype) by micro-computed tomography, we found that the Lepre1H662A/H662A mice had cortical bone mineral density and cortical thickness values comparable to wild-type littermates (figure 6). In addition, while the biomechanical analysis of femurs by 3-point bending test (n = 7, each genotype) demonstrated no differences in the extrinsic biomechanical properties (ultimate load, stiffness, energy to failure and post-yield displacement), the geometric value, cross-sectional moment of inertia, was increased and the elastic modulus, an intrinsic material property, was decreased in the Lepre1H662A/H662A mice compared to wild-type littermates (figure 6). Moreover, the Lepre1H662A/H662A mice have less trabecular bone when compared to wild-type littermates, which is quantified by decreased bone volume over tissue volume (BV/TV), decreased trabecular number (Tb.N), decreased trabecular thickness (Tb.Th), and increased trabecular separation (Tb.Sp) (figure 6). These findings are partly in contrast to what was described for Lepre1−/− mice, which are characterized by a decrease in both trabecular and cortical bone mineral density [14], and cortical stiffness and force to failure of femurs [14].
To further study the trabecular bone phenotype in the Lepre1H662A/H662A mice, bone histomorphometric analysis was conducted on 3-month spines (n = 9). A statistically significant decrease in bone volume over tissue volume and trabecular thickness confirmed the low bone mass phenotype (figure 7) [2]. However, no differences in the osteoblast number (N.Ob/BS), osteoclast surface over bone surface (Oc.S/BS), osteoid parameters (osteoid volume over bone volume and osteoid surface over bone surface) or bone formation rate were observed between the two genotypes (figure 7).
The morphology of collagen fibrils was then analyzed at the ultrastructural level in skin biopsies from Lepre1H662A/H662A mice. The electron micrographs showed collagen fibrils that were more homogeneous in diameter compared to wild-type controls (figure 8). This suggested that collagen trimers may not be efficiently assembled into higher order collagen fibrils as reflected by an increase in the proportion of smaller diameter collagen fibrils in the Lepre1H662A/H662A skin (N = 3, 150 collagen diameters measured per animal, p<0.05) (figure 8).
We then analyzed the collagen secretion rate, as measured by pulse-chase assays in dermal fibroblasts. The rate and amount of procollagen secreted from the Lepre1H662A/H662A fibroblasts were apparently similar to those secreted by wild-type fibroblasts (repeated 3 times) (figure 9A, B). These findings contrast with the delayed procollagen secretion observed in the Lepre1−/− mice and are apparently different from the increase in the collagen secretion rate observed in the Crtap−/− fibroblasts (figure 9A, B) [2], [14]. We also assessed whether there was collagen overmodification by steady-state analysis. The electrophoretic migration of type I collagen chains synthesized by wild-type and Lepre1H662A/H662A fibroblasts was similar and suggested normal post-translational modification (repeated 3 times) (figure 9C). These findings differ from the overmodification observed in collagen isolated from Crtap−/− fibroblasts [2]. These findings support a conclusion that the catalytically inactive P3H1H662A mutant protein can restore the chaperone activity and collagen assembly function of the P3H1 complex in the ER.
We generated and characterized a novel mouse model (Lepre1H662A/H662A) that harbors a knock-in mutation in the catalytic domain of P3H1. This mutation inactivated the 3-hydroxylase activity of P3H1 and caused loss of 3-Hyp at Pro986 in the collagen α1(I) chain of bone and a reduction to 9% 3-Hyp at Pro986 in the α1(II) chain of cartilage. Our findings are similar to what has been reported in the null mouse mutants for the components of the prolyl 3-hydroxylation complex [2], [14], [15]. Additionally, our data confirm the findings by Pokidysheva et. al. that the Pro986 site in the α1(I) chain of bone is exclusively hydroxylated by P3H1 [21]. The hydroxylation status of the A3 (Pro707) site of the α2(I) chain is similar between the wild-type and Lepre1H662A/H662A bone (although the wild-type percentage that we observed is lower than that observed by Pokidysheva et. al. and could be attributable to strain differences) and is in contrast to the reduction observed at this site in the Lepre1−/− bone [21]. The expression of Leprel1 (encoding P3H2) was found to be dramatically reduced in the bone of Lepre1−/− mice, possibly explaining the reduction in hydroxylation observed at the A3 site [21]. Pokidysheva et. al. also noted a slight increase in the expression level of Leprel2 (encoding P3H3), which could account for the residual hydroxylation observed in the Lepre1−/− mice [21]. Additionally, since the hydroxylation status of the A3 site isolated from bone of Lepre1H662A/H662A mice is similar to their wild-type littermates, it supports the notion that another P3H is responsible for hydroxylation at this site. The hydroxylation status of Pro986 in the α2(V) chain differed in our Lepre1H662A/H662A mice compared to Crtap−/− mice, i.e. 21% hydroxylation in the Lepre1H662A/H662A mice and <2% hydroxylation in the Crtap−/− mice [17].
Expression of Lepre1 (encoding P3H1) varies in cartilage, being highly expressed in the resting/proliferating chondrocytes and less so in pre-hypertrophic chondrocytes (not shown). With high expression of Leprel1 (encoding P3H2) in pre-hypertrophic chondrocytes and Leprel2 (encoding P3H3) in the resting/proliferating chondrocytes (not shown), it is possible that the 9% residual hydroxylation of Pro986 in α1(II) in cartilage could be a result of hydroxylation by P3H3 (or P3H2) analogous to the situation in bone with α2(V). Alternatively, it is possible that the histidine to alanine substitution at residue 662 may not completely inactivate the prolyl 3-hydroxylase function of P3H1 although we think this is less likely given published biochemical analysis of this residue and the lack of hydroxylation in bone [20].
The Lepre1H662A/H662A mice have reduced trabecular bone, but normal cortical bone and normal extrinsic cortical biomechanics. Although the geometric value, cross-sectional moment of inertia, is increased in the Lepre1H662A/H662A mice, this finding could be a compensatory mechanism to maintain similar extrinsic properties despite inferior intrinsic properties. Due to physiological higher remodeling rates in trabecular bone, a mild effect is more likely to manifest earlier in trabecular bone than in cortical bone, and this could account for the differential phenotype observed in our mice. The mild effect could also account for our histomorphometry data where we observe no difference in osteoid and bone formation parameters and is in contrast to the lower osteoid parameters and bone formation rate observed in the Crtap−/− mice [2]. Thus, due to a mild effect, we cannot detect a difference in the osteoid and the dynamic indices of bone formation.
Additionally, cartilage is normal in the Lepre1H662A/H662A mice suggesting that 3-Hyp modification in type II collagen is not required at least in the resting/proliferative zone. Table 3 compares the bone phenotypes exhibited by Crtap−/−, Ppib−/−, Leprel1−/− and Lepre1H662A/H662A mice. The pronounced growth defects and disorganized growth plates in the mice lacking a functional P3H1 complex may reflect a particular sensitivity of growth plate chondrocytes to ER stress caused by handling misfolded un-partnered subunits of the complex.
The effect of 3-hydroxyproline on collagen stability is not clear. Studies originally suggested the absence of a triple-helix structure for synthetic peptides containing 3-hydroxyproline at all Xaa positions [22], [23]. Recent studies indicate the presence of one or two 3-hydroxyprolines in the Xaa position does produce a triple helix with a consequent slight increase in stability [3], [24], [25]. These data suggest that although the absence of 3-hydroxyproline may not affect the stability of the triple-helix it may alter protein-protein interactions [24]. Since matrix-cell signaling is important in the development and maintenance of connective tissues, it is plausible that collagen post-translational modifications like prolyl 3-hydroxylation could specify protein-protein interactions between collagen and other ECM components. Furthermore, these interactions may be context-dependent differing in trabecular bone vs. cortical bone vs. cartilage. Candidates include ligand interaction sites mapped to the fibril in proximity to Pro986 and include fibronectin and α1β1/α2β1/α1β11 integrins [26].
Mutations that result in loss of the prolyl 3-hydroxylation complex can result in collagen overmodification [27]. This could also independently affect collagen fibril interaction sites, e.g., with small leucine rich proteoglycans (SLRPs) such as decorin. Such disruption or other signaling effects could explain the disorganization of the hypertrophic zone observed in the Lepre1−/− mice but not observed in the Lepre1H662A/H662A mice. Future work investigating the signaling defects present in bone and cartilage will be necessary to understand the chondrodysplasia present in patients carrying mutations in LEPRE1 or CRTAP.
Compared with other tissues, the extracellular matrix of bone is unique in the sense that it is able to mineralize [28]. As the dominant component of bone, it is likely that type I collagen plays an important role in the manner of mineralization of the extracellular matrix. It has been argued that the collagen of bone has evolved special features that equip it to constrain the growth internally of nanocrystal plates of hydroxyapatite [27]. Extrafibrillar non-collagenous proteins limit the amount of extrafibrillar crystal growth [27].
Evolutionarily, prolyl 4-hydroxylation increases the thermal stability of the triple helix through hydrogen bonding [29]. Although loss of prolyl 3-hydroxylation is a feature of recessively inherited OI, the evolutionary function of 3-hydroxyproline is still poorly characterized. One clue to the potential role of 3-hydroxyproline in collagen is through the identification of partially occupied 3-Hyp sites in type I and II collagen; these are D-periodically spaced and suggest the modification is involved in some aspect of collagen fibril assembly [30]. Peptide studies suggest that the 3-Hyp residues have selective affinity for one another [30]. Additionally, the 3-hydroxyl groups are outward pointing from the triple-helix, which provides evidence that short-range hydrogen bonding between collagen triple helices is likely to occur [30]. Although 3-Hyp appears early in collagen evolution, the 3-Hyp at Pro986 of the alpha I chain of type I collagen appeared much later [27], [31]. In fact, the presence of 3-Hyp at this residue coincides with the appearance of CRTAP and occurs just before the appearance of vertebrates and bone [32]. Taken together, with the potential role of 3-Hyp and the appearance of 3-Hyp at Pro986 just prior to the appearance of vertebrates and bone, we speculate that this unique modification could have functioned to equip collagen molecules for a polymeric architecture that allows organized hydroxyapatite nanocrystal growth within fibrils potentially explaining the lack of phenotype in cartilage vs. bone in our Lepre1H662A/H662A mice.
In addition to the loss of 3-Hyp at Pro986, we observed an increase in the HP/LP ratio in the bone of the Lepre1H662A/H662A mice compared to their wild-type littermates. In both dominant and recessive OI, there is an increase in the HP/LP ratio, supporting alterations in collagen crosslinking [33]. The increase in HP/LP ratio suggests a disturbance of the fibrillar architecture in bone and could result in the disorientation of nanocrystal plates of hydroxyapatite, but does not signify overmodification of the collagen. Since we observe no difference in collagen migration from the steady-state collagen analysis, we conclude that there is no gross overmodification in the collagen isolated from the Lepre1H662A/H662A mice. Additionally, we observe no difference in collagen secretion, providing indirect evidence that the P3H1 complex is able to form and bind to collagen, in contrast to the Lepre1−/− mice which have a delay in collagen secretion due to failure of complex formation [14].
Future work investigating the contribution of signaling vs. collagen cross-linking defects present in bone and cartilage will be necessary to understand the generalized connective tissue phenotype present in patients carrying mutations in LEPRE1 or CRTAP.
All research involving animals was conducted according to the relevant national and international guidelines. Veterinarians supervised animal care according to standard conditions of Baylor College of Medicine Center for Comparative Medicine, a program fully accredited by the Association for Assessment and Accreditation of Laboratory Animal Care International. All mouse work was approved by the Center for Comparative Medicine in conjunction with the Institutional Animal Care and Use Committee.
The Evolutionary Trace (ET) is an approach to identify molecular determinants of protein function and to target mutational analysis and protein engineering to the most relevant amino acids of a protein [34]–[38]. To help identify functionally important sites, ET builds a phylogenetic tree from a multiple sequence alignment and then scans the multiple sequence alignment for residue variations that correlate with major evolutionary divergences in the tree (Full residue invariance, or “conservation”, is a special case). The amino acids are ranked according to a score,where fiag is the frequency of the appearance at residue position i of the amino acid of type a within the group g of subalignments that appear at “distance” n from the tree root. is the number of sequences in the alignment.
In practice, the top-ranked ET residues cluster together spatially in the structure, a property that is used to assess the statistical significance of predictions.
Fibroblast cultures were grown in DMEM supplemented with 4 mM L-glutamine, 4500 mg/L glucose (HyClone), 10% FBS, 100units/mL penicillin, and 100 µg/mL streptomycin.
Viral vector and packaging plasmids were transfected into 293T cells to produce lentivirus, which was concentrated, and transduced into fibroblasts as previously described [39].
After fibroblasts were cultured to confluence in a 60 mm2 dish, cells were washed in 1× PBS. Subsequently, cells were scraped in 1 ml 1× PBS and transferred to a 1.5 ml Eppendorf tube. Cells were spun down at 14000 RPM for 5 minutes at room temperature. Samples were resuspended with 50 µl of 5%B-ME and Lamellae loading buffer (Biorad) prior to the separation of proteins on a polyacrylamide gel.
10 µg of protein was loaded into each well, run and transferred to a PVDF membrane using a standard wet transfer process. After blocking using 5% milk or 5% BSA, proteins were detected using the following antibodies directed against P3H1, CRTAP, and γ-Tubulin (sigma) diluted in appropriate buffer. The blot was washed 3 times, 10 minutes each with 1× TBS with 0.05% Tween20 and then incubated with the corresponding secondary antibody and washed again. Proteins were visualized using the Luminato Crescendo HRP substrate (Millipore) by incubating blot with the substrate for 2 minutes before exposing the blot to film and feeding it through the developer.
Fibroblasts were split into glass LAB-TEK 4-well chamber slides (Nunc), and 24 hours later were fixed with 4% paraformaldehyde, treated with 0.1% Triton X-100, blocked in 10% donkey serum and 1% BSA, and then sequentially incubated with 1∶250 dilution of CRTAP antisera or P3H1, 1∶500 donkey anti-rabbit secondary antibody conjugated to Alexa Flour 594 (Invitrogen), and mounted with Prolong Gold anti-fade reagent with DAPI (Invitrogen). The slides were visualized using a Zeiss fluorescence microscope.
Recombineering was utilized to generate the P3H1 knock-in mouse model using the method described by Pentao Liu [40]. Briefly, retrieval vector and targeting constructs were designed and cloned together in order to retrieve the section of Lepre1 for gene targeting and for introducing the alanine substitution into the gene locus. The linearized DNA construct was electroporated into AB2.2 embryonic stem (ES) cells and screened for positive recombination events by both Southern blot and sequencing. Germ-line transmission of the mutant allele into C57BL6 was obtained.
Lepre1H662A/H662A mice and wild-type littermates were sacrificed at 3 months of age. Spine, femurs and skin were dissected, fixed, paraffin embedded, and sectioned according to standard methods as previously described [2]. The Lepre1H662A/H662A mouse colony was maintained in a mixed 129Sv/ev-C57Black/6J genetic background and housed in the Baylor College of Medicine Animal Vivarium.
Standard protocols were followed for the following stains: Hematoxylin and Eosin. Immunofluorescence on mouse tissues was done as previously described [41]. Briefly, the paraffin sections were xylene treated, rehydrated, and heated for 20 minutes in a steamer for antigen retrieval. Subsequently they were incubated in blocking solution (3% normal Donkey serum, 0.1% BSA, 0.1% Triton X-100 in PBS), 1∶100 dilution of COLX antisera, 1∶600 donkey ant-rabbit secondary antibody conjugated to Alexa flour 594 (Invitrogen), and finally mounted with Prolong Gold anti-fade reagent with DAPI (Invitrogen). At the end of each described procedure, images were captured using a Zeiss Axioplan 2 microscope. Radiographs were obtained by Kubtec XPERT80 (Kubtec X-ray, Milford, CT). Routine histologic analysis of paraffin-embedded long bone and growth plates was done as per standard protocols. Histomorphometric analysis of static and dynamic parameters (using 25 mg/kg calcein injection) of bone resorption, formation, and volume was carried out according to standard procedure in 12 week-old (N = 6 each sex and genotype) [42] and analyzed using the Bioquant OsteoMetrics software system (BIOQUANT Image Analysis Corporation, Nashville, TN).
Spine and femur samples were placed into a 16 mm tube filled with 70% ethanol and scanned at 16 micron resolution using a ScanCo uct40 scanner (N = 10 each sex and genotype). Trabecular and cortical analysis was performed using the ScanCo software. The trabecular region of the L4 vertebrae was manually selected (contoured) every 5 slices, and then the remaining slices were morphed to enclose the region of interest, for a total of 100 slices. The region was thresholded at 210 with a gauss setting of 0. Cortical analysis was performed on 50 slices of the femoral midshaft, using the same thresholding.
Femurs were collected (N = 7, both genotypes) during tissue harvest, wrapped in gauze soaked in 1× PBS and stored at −20°C until ready for analysis. All testing was performed in a 3-point bending apparatus (Instron 5848) with a span of 7 mm as previously described [14], [43]. The posterior surface of the femur was placed on the lower supports and centered between the two supports. The displacement rate used for the analysis was 0.3425 mm/sec.
Stress was calculated using the following formula:where F is the load applied on the femur in N, L is the span length in mm, h is the specimen diameter in mm, and I is the cross-sectional moment of inertia in mm4. L was set at 7 mm for the three point bending jig. Both h and I were obtained by analyzing a midshaft micro-CT image using the slice geometry program in BoneJ, a plugin for bone image analysis in ImageJ. The program identified the bone in the grayscale micro-CT image by pixel brightness and then calculated its cross-sectional parameters once an accurate scale was assigned.
Strain was calculated using the following formula:where D is the actuator displacement in mm, h is the diameter in mm and L is the span length in mm.
To determine the Yield Point, a region was identified after the preload and before the maximum load on the Load-Displacement curve. This region was separated into 2 segments from which the fitted line of the segment with greatest slope was taken. Next, a 0.00876 mm offset was implemented on the line. The point of intersection between the offset line and the Load-Displacement curve was the Offset Yield Point. This yield point corresponded to a 0.2% offset strain, which is commonly chosen in the literature. The elastic region was identified as the region from the completion of the preload to the Yield Point. Post-Yield region was identified as the region from the Yield Point until the Failure Point, where the load dropped to zero. Elastic Displacement was the displacement during which specimen remained in elastic region. Post-Yield Displacement was the displacement during which specimen remained in the Post-Yield region. Total Displacement was calculated as the sum of Elastic Displacement and Post-Yield Displacement. Using a trapezoidal numerical integration method, Elastic Energy was calculated as the area under the elastic region of the Load-Displacement curve. Similarly, Post-Yield Energy was calculated as the area under the post-yield region of the Load-Displacement curve. Energy to Failure was the sum of Elastic Energy and Post-Yield Energy. Maximum Load was determined by finding the highest load value recorded by BLUEHILL, before the specimen failed. To calculate Stiffness, Least Square fit method was applied to the steepest segment of the elastic region of the Load-Displacement curve. Stiffness was the slope of least square fit line.
Ultimate Strength was determined by finding the highest Stress value before the specimen failed. Using a trapezoidal numerical integration method, Elastic Toughness was calculated as the area under the elastic region of the Stress-Strain curve. Similarly, Post-Yield Toughness was calculated as the area under the post-yield region of the Stress-Strain curve. Toughness to Failure was the sum of Elastic Toughness and Post-Yield Toughness. To calculate Elastic Modulus, Least Square fit method was applied to the steepest segment of the elastic region of the Stress-Strain curve. Elastic Modulus was the slope of least square fit line.
Freshly dissected tissues were fixed in 1.5% glutaraldehyde/1.5% paraformaldehyde with 0.05% tannic acid in 0.1 M Cacodylate buffer, pH 7.4 for 60 minutes on ice, rinsed in 0.1 M cacocylate overnight, then postfixed for 60 minutes in cacodylate buffered 1% OsO4, rinsed, then dehydrated in a graded ethanol series from 30–100%. The samples were washed in propylene oxide and embedded in Spurrs epoxy. Ultrathin sections were stained in Uranyl Acetate followed by Reynolds lead citrate and examined using a FEI Tecnai G2 TEM. Transmission electron microscopy was performed on skin of wild- type and Lepre1H662A/H662A mice (N = 3). The fibril diameter of ten fibrils in each of fifteen different areas per mouse was measured (N = 150 total measurements).
Pyridinoline cross-links (HP and LP) were quantified by HPLC after hydrolyzing demineralized bone in 6N HCl as described [44].
Types I and V collagens were prepared from minced bone decalcified at 4°C in 0.1M HCl overnight. Type I α-chains were extracted by heat denaturation (90°C) in SDS-PAGE sample buffer. Type V collagen was solubilized by pepsin in 3% acetic acid and selectively precipitated by 1.8 m NaCl [45]. Type II collagen was solubilized by CNBr digestion of rib cartilage in 70% formic acid [46]. Collagen α-chains and CNBr-peptides were resolved respectively on 6% and 12.5% SDS-PAGE gels [47].
Demineralized bone matrix was digested with bacterial collagenase as described [48]. Collagenase-generated peptides were separated by reversed-phase HPLC (C8, Brownlee Aquapore RP-300, 4.6 mm×25 cm) with a linear gradient of acetonitrile∶n-propanol (3∶1 v/v) in aqueous 0.1% (v/v) trifluoroacetic acid [49]. Individual fractions were analyzed by LC-MS.
Collagen α-chains or CB peptide bands were cut from SDS-PAGE gels and digested with trypsin in-gel [50]. Peptides were analyzed by electrospray LC/MS using an LCQ Deca XP ion-trap mass spectrometer (ThermoFinnigan) equipped with in-line liquid chromatography using a C8 capillary column (300 um×150 mm; Grace Vydac 208MS5.315) eluted at 4.5 ul min. The LC mobile phase consisted of buffer A (0.1% formic acid in MilliQ water) and buffer B (0.1% formic acid in 3∶1 acetonitrile∶n-propanol v/v). An electrospray ionization source (ESI) introduced the LC sample stream into the mass spectrometer with a spray voltage of 3 kV. Sequest search software (ThermoFinnigan) was used for peptide identification using the NCBI protein database. Large collagenous peptides not found by Sequest had to be identified manually by calculating the possible ms/ms ions and matching these to the actual ms/ms. Hydroxyproline and hydroxylysine calculations were done manually by scrolling or averaging the full scan over several minutes so that all of the post-translational variations of a given peptide appeared together in the full scan.
Mice were euthanized using isoflurane, the fur was scraped off the back of the mouse and a small section of skin was harvested. The skin was washed with 1× PBS and placed dermis down onto a well of a 6-well plate. 2 ml of DMEM complete was added to the well and the skin was cut into small pieces using a scalpel to allow fibroblasts to migrate from the skin.
We analyzed procollagen secretion by pulse-chase assay as previously described [51], [52]. Briefly, we plated 2.5×105 cells onto a 35 mm dish and let them grow overnight. The next evening, the medium was changed to DMEM complete with 50 µM ascorbate to induce collagen synthesis. After washing the cells 3 times with 1×PBS to remove FBS from the cells, the medium was replaced with serum free DMEM containing 50 µM ascorbate and 140 µCi of L-[2,3,4,5-3H] proline. Cells were labeled for 1 hour and then chased with fresh medium containing unlabeled proline. Both the cell layer and medium were harvested in 1× PBS containing 1× inhibitor at 20-minute intervals and the procollagen was precipitated with collagen carrier (Sigma) and absolute ethanol. Samples were electrophoresed on a 5% acrylamide gel containing urea under reducing conditions, dried, and imaged by exposing film to gel for 24 hours at −80°C and processed using a developer. Procollagen secretion over time was measured by comparing the amount of labeled procollagen present in the cell layer and medium at each time point. The pulse-chase assay was repeated three times to confirm the results.
We analyzed collagen modification by collagen steady-state analysis as previously described [51], [52]. Briefly, we plated 2.5×105 cells onto a 35 mm dish and let them grow overnight. The medium was changed to DMEM complete with 50 µM ascorbate to induce collagen synthesis. After 4 hours, the cells were washed 3 times with 1×PBS and the medium was replaced with serum free DMEM containing 50 µM ascorbate and 140 µCi of L-[2,3,4,5-3H] proline. Cells were labeled overnight. Both the cell layer and medium were harvested in 1× PBS containing 1× inhibitor and the procollagen was precipitated with collagen carrier (Sigma) and absolute ethanol. Collagens were obtained by overnight pepsin digestion (50 ug/ml) of procollagen samples. Samples were electrophoresed on a 5% acrylamide gel containing urea under reducing conditions for collagen samples, dried, and imaged by exposing film to gel for 24 hours at −80°C and processed using a developer. This assay was repeated three times to confirm the results.
Data are expressed as mean values ± standard deviation (SD). Statistical significance was computed using the Student's t test. A P value <0.05 was considered statistically significant.
|
10.1371/journal.pntd.0001396 | Examining the Relationship between Urogenital Schistosomiasis and HIV Infection | Urogenital schistosomiasis, caused by infection with Schistosoma haematobium, is widespread and causes substantial morbidity on the African continent. The infection has been suggested as an unrecognized risk factor for incident HIV infection. Current guidelines recommend preventive chemotherapy, using praziquantel as a public health tool, to avert morbidity due to schistosomiasis. In individuals of reproductive age, urogenital schistosomiasis remains highly prevalent and, likely, underdiagnosed. This comprehensive literature review was undertaken to examine the evidence for a cause-effect relationship between urogenital schistosomiasis and HIV/AIDS. The review aims to support discussions of urogenital schistosomiasis as a neglected yet urgent public health challenge.
We conducted a systematic search of the literature including online databases, clinical guidelines, and current medical textbooks. We describe plausible local and systemic mechanisms by which Schistosoma haematobium infection could increase the risk of HIV acquisition in both women and men. We also detail the effects of S. haematobium infection on the progression and transmissibility of HIV in co-infected individuals. We briefly summarize available evidence on the immunomodulatory effects of chronic schistosomiasis and the implications this might have for populations at high risk of both schistosomiasis and HIV.
Studies support the hypothesis that urogenital schistosomiasis in women and men constitutes a significant risk factor for HIV acquisition due both to local genital tract and global immunological effects. In those who become HIV-infected, schistosomal co-infection may accelerate HIV disease progression and facilitate viral transmission to sexual partners. Establishing effective prevention strategies using praziquantel, including better definition of treatment age, duration, and frequency of treatment for urogenital schistosomiasis, is an important public health priority. Our findings call attention to this pressing yet neglected public health issue and the potential added benefit of scaling up coverage of schistosomal treatment for populations in whom HIV infection is prevalent.
| Urogenital schistosomiasis is a parasitic infection caused by a worm, Schistosoma haematobium, which lives in the bloodstream of infected individuals. It affects at least 112 million people, mostly in sub-Saharan Africa, and has been suggested to be a risk factor for becoming infected with HIV. We reviewed publications in order to examine whether it seems likely that this parasitic infection could be a risk factor for HIV. Evidence from many types of studies supports the hypothesis that urogenital schistosomiasis does increase a person's risk of becoming infected with HIV. Studies also suggest that individuals who have both urogenital schistosomiasis and HIV have a more aggressive HIV infection and can more easily transmit HIV to their sexual partners. Praziquantel is an oral, nontoxic, inexpensive medication that is safe in pregnancy and is recommended for treatment of schistosomiasis. In areas where both infections co-exist, regular administration of praziquantel both to young girls and to sexually-active women may be an important approach to reducing HIV transmission. Our findings support the importance of making praziquantel more available to people who live in areas of the world where both urogenital schistosomiasis and HIV infection are widespread.
| An estimated 207 million people worldwide are infected with schistosomes [1], and 85% of these cases occur in Africa [1]–[3]. Schistosomiasis is a disease of poverty that arises in areas with poor sanitation where people come into contact with urine- or feces-contaminated water as part of their daily lives [4]. Individuals living in endemic countries are most commonly infected during childhood, and the prevalence peaks between the ages of 10 and 20 years [5], [6]. For those who are continually reinfected by contaminated water, schistosomiasis causes a chronic disease over decades. While the mortality caused by schistosomiasis is low, the morbidity is high, and includes anemia, stunted growth, and decreased ability to learn in children [1]. For these reasons, the World Health Organization (WHO) recommends annual treatment for school-aged children in areas of high endemicity [7].
Schistosoma haematobium causes more than half (at least 112 million) of worldwide schistosome infections [8]. Formerly known as urinary schistosomiasis, S. haematobium infection was recently renamed “urogenital schistosomiasis” in recognition that the disease affects both the urinary and genital tracts in up to 75% of infected individuals [9]. Adult S. haematobium worms inhabit the venules surrounding organs of the pelvis where they lay between 20 and 200 eggs daily [4]. These eggs subsequently penetrate the vessel wall and move towards the lumen of the bladder. An important proportion of the eggs become sequestered in the tissue of pelvic organs such as the urinary bladder, lower ureters, cervix, vagina, prostate gland, and seminal vesicles, where they cause chronic inflammation in the affected organs. This results in a number of symptoms and signs including pelvic pain, postcoital bleeding, and an altered cervical epithelium in women [10]–[11], and ejaculatory pain, hematospermia and leukocytospermia in men [12]–[13].
Epidemiologic mapping studies of HIV and S. haematobium in Africa depict a substantial overlap, in many regions, between areas in which S. haematobium is endemic and areas in which women have a high prevalence of HIV infection (Figure 1) [14]. In particular, HIV studies report an unexplained gender quotient disfavoring women over men [15]–[16]. In rural women whose limited access to clean water more often puts them at risk for schistosomiasis, HIV prevalence also peaks at younger ages than in urban women [17]. While some of this skewing has been attributed to social, behavioral, and cultural norms [17]–[20], this unexplained gender quotient also suggests that risk factors for HIV acquisition may be different between rural and urban populations [17].
Several cross-sectional studies have reported associations between urogenital schistosomiasis and HIV, but the infection still receives relatively little attention. Global Burden of Disease (e.g. disability-adjusted life years [DALY]) calculations treat schistosomiasis cases as a single sequel, leading others to argue that the estimates should be much higher [21]–[23]. Furthermore, DALY calculations have neither examined urogenital schistosomiasis as an entity separate from intestinal schistosomiasis nor considered it as a population-attributable risk factor for HIV transmission. Thus urogenital schistosomiasis remains a neglected disease, particularly in women and men of reproductive age.
This comprehensive literature review was undertaken to examine the evidence for a cause-effect relationship between urogenital schistosomiasis and HIV/AIDS. Our aim is to support discussions of urogenital schistosomiasis as an urgent public health challenge [24].
We conducted a broad review of the literature by performing a systematic search of online databases including PUBMED/MEDINE, EMBASE, POPLINE, GLOBAL HEALTH, and WEB OF KNOWLEDGE. We used search terms beginning with the text string ‘schistosom’ in all possible combinations with ‘HIV’, ‘HIV/AIDS’, and other related keywords including ‘urinary,’ ‘genital’, ‘gynecology’, and ‘adolescent.’ Case reports were excluded. We subsequently limited our search to articles published in the past 30 years which overlap with the HIV pandemic.
We also reviewed the most current editions of widely-used infectious diseases textbooks [6], [25]–[28] and WHO websites for relevant publications. We screened titles and abstracts for relevance, and subsequently reviewed the full texts of manuscripts that were potentially pertinent. Notable manuscripts and key learning points that emerged during our review are summarized in Tables 1 and 2.
Concurrent S. haematobium infection may also increase the ease with which HIV-positive women transmit HIV infection to their sexual partners. A recent meta-analysis found that a variety of genital tract infections were associated with HIV-1 viral shedding in the female genital tract [46]. This effect was most pronounced in conditions that resulted in the recruitment of high concentrations of leukocytes to the genital epithelium, including nonspecific cervicitis, genital ulcer diseases, Chlamydia trachomatis and Neisseria gonorrhea infections, and vulvovaginal candidiasis. The authors hypothesized that, because leukocytes typically harbor HIV, conditions that lead to higher genital tract leukocyte concentrations are those that most heighten the risk of sexual HIV transmission from women to men during sexual intercourse. Given the recruitment of leukocytes to the genital tract by S. haematobium infection [35], [36], it seems possible that, through this mechanism, women who are co-infected with HIV and S. haematobium may more easily transmit HIV to their sexual partners.
Genital schistosomiasis in men can involve several male reproductive organs. Since the penis is not affected by ova-induced lesions, male genital schistosomiasis is not believed to increase the risk of HIV acquisition through local effects [43] but rather through schistosomiasis-related immunomodulatory effects. Moreover, the infection could increase risk for HIV transmission by inciting inflammation in the male genital tract. Men with severe urogenital schistosomiasis have been found to have a higher prevalence of lymphocytes and eosinophils in seminal fluid than those without infection [47]. Infected men are also reported to have significantly higher levels of interleukin (IL)-4, IL-6, IL-10, and tumor necrosis factor-alpha in their semen. These cytokines may recruit more HIV-infected cells to the semen, upregulate viral replication, and increase the concentration of HIV virus in semen [47]. Six months after anti-schistosomal treatment, the concentrations of seminal lymphocytes and eosinophils were lower and the levels of cytokines were reduced. Another analysis of seminal fluid in S. haematobium-infected men demonstrated lower volumes of semen, higher levels of eosinophilic cationic protein (an established marker of inflammation and morbidity in urogenital schistosomiasis), and higher rates of sperm apoptosis, which lessened after praziquantel treatment [48].
With a mechanism similar to that discussed for women in the preceding section, it has been demonstrated that HIV-positive men with concomitant genital tract infections, such as urethritis, have higher concentrations of seminal HIV-1 RNA than those without dual infections [49]. The chronic inflammation and recruitment of lymphocytes and eosinophils to the male genital tract may increase the HIV-1 viral load in semen. In this manner, a female sexual partner of an HIV-positive male living in an S. haematobium-endemic area may have a doubly-amplified risk of HIV acquisition: her S. haematobium-infected partner's semen may contain disproportionately high concentrations of HIV RNA, and her own S. haematobium infection may increase the ease with which HIV can establish infection following exposure. Taken together, these data suggest that egg-induced inflammation in the male genital tract could be a risk factor for HIV transmission from men to women.
Schistosomal lesions are commoner in the vulva and the lower vagina before puberty, while in adult women they are more frequent in the cervix, uterus, ovaries and fallopian tubes [50]. These clinically-apparent lesions and the resulting compromise of the vaginal epithelium, therefore, are already present before a girl's first sexual intercourse. This is in contrast to lesions caused by STIs, which can develop only after sexual intercourse [32]. The presence of schistosomal lesions already in childhood makes it likely that schistosomal infection typically precedes HIV infection and that the temporal association reflects the fact that urogenital schistosomiasis is a risk factor for HIV acquisition rather than vice-versa [10].
Important risk factors for urinary tract morbidity in adulthood are cumulative intensity and duration of S. haematobium infection during early adolescence. Treatment of school-aged children can significantly reduce the cumulative lifetime egg burden as the intensity of infection is greatest during early teenage years [50]. Furthermore, treatment for schistosomiasis during childhood was significantly associated with the absence of cervical sandy patches and contact bleeding in adult women [51]. Thus treatment of S. haematobium infection before and during the teenage years may not only diminish genital schistosomiasis-associated morbidity in adulthood, but may simultaneously decrease the risk of HIV acquisition.
In addition to local and gender-specific effects of S. haematobium infection, schistosomiasis also appears to increase HIV susceptibility through chronic immune modulation. This topic has been studied far more extensively with regard to Schistosoma mansoni [52]. It has also recently been the subject of a comprehensive review [31] and for this reason will be summarized only briefly here.
Chronic schistosomiasis alters global immune function and in this manner may also increase susceptibility to HIV infection [38]. It preferentially stimulates the Th2-type immune response, with reciprocal down-regulation of the Th1-type cytotoxic responses [53] which are important in initial control of HIV infection. This is supported by work from Uganda, which demonstrated that HIV-positive patients with S. mansoni infection had decreased Gag-specific cytolytic CD8+ responses [54]. Moreover, CD4+ T-cells with a Th2 phenotype are more readily infected, and subsequently destroyed, by HIV-1 than are Th1 cells [55]. In individuals in Kenya with HIV and S. mansoni co-infection, Th2-type CD4+ T-cells were destroyed more quickly than in HIV-positive individuals without schistosomiasis [56].
Specifically, differences in cell surface receptors may lead to differences in HIV susceptibility between those with and without schistosomiasis. The chemokine receptors CCR5 and CXCR4 are co-receptors for HIV-1 and were found to be more dense on the CD4+ T-cell surfaces of individuals with active S. mansoni infection than on the CD4+ T-cells of individuals who had received prior anti-schistosomal treatment [38]. The levels of these co-receptors dropped in individuals who were studied pre- and post-praziquantel treatment [38]. This highlights the potential role that widespread anti-schistosomal treatment could play in reducing the progression and spread of HIV.
In addition to potentially increasing susceptibility to HIV infection, evidence suggests that S. haematobium infection may also speed progression of disease by raising plasma HIV RNA concentration (commonly known as “viral load”) in individuals who are co-infected. At the cellular level, the same CCR5 and CXCR4 chemokine receptors that are upregulated in schistosomiasis and facilitate HIV binding in initial infection may also promote cell-to-cell spread of HIV once infection is established [31], [57]. Multiple studies have shown that the plasma HIV RNA level is predictive of both HIV disease progression and risk of transmission of HIV to sexual partners [58], [59]. If the hypothesis is correct that schistosomiasis increases the HIV RNA levels in co-infected individuals, then treatment for schistosomiasis could delay the development of AIDS and decrease the spread of HIV in sub-Saharan Africa.
A recent randomized clinical trial conducted in Zimbabwe supports this hypothesis. Patients who were infected with both HIV and S. mansoni were randomized either to praziquantel treatment at enrollment or to praziquantel after three months [60]. Compared with the group in whom treatment was delayed, the early-treatment group experienced significantly smaller declines in CD4+ T-cell counts after three months (mean decline of 1.7 cells/ µL versus 35.2 cells/ µL in the delayed-treatment group) [61]. Notably, the HIV RNA levels in both groups of patients increased during the three months, but the mean increase in the early-treatment group (0.001log10 copies/mL) was significantly lower than in the delayed-treatment group (0.21log10 copies/mL).
Earlier non-randomized studies of HIV-positive patients who were treated for S. mansoni infections had found that HIV RNA levels remained stable or increased in patients regardless of treatment [62]–[65]. One notable study that reported significant HIV RNA level increases one month post-treatment noted corresponding increases in S. mansoni-specific Th2-type cytokine responses as well, though both of these reverted to pre-treatment levels by five months post-treatment [66]. Notably, none of the patients in these studies were receiving ART. In light of the findings of the randomized trial in Zimbabwe that did demonstrate a benefit with praziquantel treatment with regard to the viral load [60], it is plausible that the overall observed increases in HIV RNA levels reflect natural progression of untreated HIV infection. In this sense, while treatment for schistosomiasis is clearly not able to substitute for antiretroviral therapy, it may possibly be able to slow HIV disease progression [31].
In support of this hypothesis, two other randomized studies of HIV-infected patients co-infected with either Wucheria bancrofti [67] or soil-transmitted helminths [68] have explored the effects of treatment on parameters of HIV infection. Patients treated for lymphatic filariasis had significant decreases in their HIV RNA levels and insignificant increases in their CD4+ T-cell counts at 12 weeks as compared to pretreatment levels [67]. Patients with ascariasis who received albendazole experienced significantly higher CD4+ T-cell counts at 12 weeks and a trend towards lower HIV RNA levels [68].
Taken together, these studies of treatment in HIV and helminth co-infections support a positive effect of antiparasitic treatment on certain HIV infection parameters. While treatment for schistosomiasis in HIV-positive patients may not decrease HIV RNA levels, it may slow the increase of viral levels. It is also possible that praziquantel treatment may enhance immunocompetence by promoting an increase in CD4+ T-cell counts and an increased NK cell function [69]–[71].
Schistosoma haematobium infection is highly prevalent in sub-Saharan Africa. Increasing evidence supports that it is a plausible risk factor for HIV acquisition due both to its local genital tract effects in women, and to its chronic immunomodulatory effects in both men and women. It also could facilitate HIV transmission to the sexual partners of HIV-positive individuals with schistosomal co-infection, and could enhance HIV disease progression. Circumstantial, biological, immunological, and epidemiological evidence is strongly suggestive of a cause-effect relationship between S. haematobium and HIV infection.
Our review highlights the need for further innovative research, particularly appropriately-designed longitudinal studies which ultimately would be able to confirm the suggested causality of schistosomiasis in incident HIV infections. Such studies must carefully balance the ethical obligation to ensure treatment for study subjects while simultaneously managing to explore the cause-effect relationship between the two infections. This is by no means an easy task. Consideration should therefore be given to harnessing latent operational research opportunities that exist within the context of ongoing schistosomiasis control programs. These include studies such as exploring the effect of early, regular anti-schistosomal treatment of girls to prevent development of urogenital lesions in adolescence or testing for markers of active schistosomiasis in blood collected from HIV-positive women before their HIV-seroconversion.
Meanwhile, in schistosomiasis-endemic areas where coverage for preventive chemotherapy with praziquantel remains low, millions of individuals may be at higher risk for HIV infection. The presumptive causal association with HIV infection notwithstanding, urogenital schistosomiasis by itself leads to significant morbidity that can be lessened with inexpensive preventive chemotherapy. At an annual cost of about 40 cents per person [72]–[73], praziquantel stands as a powerful and economical public health intervention with the potential to prevent the development of urogenital lesions, prolong survival, and decrease new HIV infections on the African continent. In view of the plausible association between urogenital schistosomiasis and HIV transmission in areas where these infections are co-endemic, a salient effect on the health of millions of individuals could presumably be achieved if antischistosomal treatment and HIV prevention interventions were integrated. The WHO-recommended policy of early regular treatment of school-age children with praziquantel needs to be extended to adults and prioritized in national programs as a possible means of further preventing HIV infections in sub-Saharan Africa.
|
10.1371/journal.pcbi.1006187 | Bridging structure and function: A model of sequence learning and prediction in primary visual cortex | Recent experiments have demonstrated that visual cortex engages in spatio-temporal sequence learning and prediction. The cellular basis of this learning remains unclear, however. Here we present a spiking neural network model that explains a recent study on sequence learning in the primary visual cortex of rats. The model posits that the sequence learning and prediction abilities of cortical circuits result from the interaction of spike-timing dependent plasticity (STDP) and homeostatic plasticity mechanisms. It also reproduces changes in stimulus-evoked multi-unit activity during learning. Furthermore, it makes precise predictions regarding how training shapes network connectivity to establish its prediction ability. Finally, it predicts that the adapted connectivity gives rise to systematic changes in spontaneous network activity. Taken together, our model establishes a new conceptual bridge between the structure and function of cortical circuits in the context of sequence learning and prediction.
| A central goal of Neuroscience is to understand the relationship between the structure and function of brain networks. Of particular interest are the circuits of the neocortex, the seat of our highest cognitive abilities. Here we provide a new link between the structure and function of neocortical circuits in the context of sequence learning. We study a spiking neural network model that self-organizes its connectivity and activity via a combination of different plasticity mechanisms known to operate in cortical circuits. We use this model to explain various findings from a recent experimental study on sequence learning and prediction in rat visual cortex. Our model reproduces the changes in activity patterns as the animal learns the sequential pattern of visual stimulation. In addition, the model predicts what stimulation-induced structural changes underlie this sequence learning ability. Finally, the model also predicts how the adapted network structure influences spontaneous network activity when there is no visual stimulation. Hence, our model provides new insights about the relationship between structure and function of cortical circuits.
| The ability to predict the future is a fundamental challenge for cortical circuits. At the heart of prediction is the capacity to learn sequential patterns, i.e., the ability for sequence learning. Recent experiments have shown that even early sensory cortices such as rat primary visual cortex are capable of sequence learning [1]. Specifically, Xu et al. [1] have shown that the visual cortex can learn a repeated spatio-temporal stimulation pattern in the form of a light spot moving across a portion of the visual field. Intriguingly, when only the start location of the sequence is stimulated by the light spot after learning, the network will anticipate the continuation of the sequence as revealed by its spiking activity. While these results are remarkable, the cellular basis of this ability has remained elusive. A natural candidate for a cellular mechanism of sequence learning is spike-timing dependent plasticity (STDP) [2, 3]. Theoretical work has suggested that the inherent temporal asymmetry of STDP seems ideally suited for learning temporal sequences and storing them into the structure of cortical circuits [4–8]. At present, it is still unknown, however, exactly how such sequence memories are stored in real cortical circuits and how they become reflected in the structure of these circuits. Yet, a number of generic structural features of cortical circuits have been established in recent years. Among them is the lognormal-like distribution of synaptic efficacies between excitatory neurons [9–13], the distance-dependence of synaptic connection probabilities [12, 14], and an abundance of bidirectional connections between excitatory neurons [12, 15]. While recent theoretical studies have successfully modeled the origins of these generic structural features of cortical circuits, there is currently no unified model that explains both the emergence of structural features of cortical circuits and their sequence learning abilities. Here we present such a model and therefore establish a new conceptual bridge between the structure and function of cortical circuits.
Our model is a recurrently connected network of excitatory and inhibitory spiking neurons endowed with STDP, combined with a form of structural plasticity that creates new synapses at a low rate and destroys synapses whose efficacies have fallen below a threshold, as well as several homeostatic plasticity mechanisms. We used this network to model recent experiments on sequence learning in rat primary visual cortex [1]. The model successfully captured how multi-unit activity is changing during learning and explained these changes on the basis of STDP and the other plasticity mechanisms adapting the circuit during learning. It additionally demonstrated how homeostatic mechanisms prevent the runaway connection growth and unstable overlearning [16–19] that otherwise tends to occur from STDP alone. Furthermore, the model predicted that the changes to the network during learning also alter spontaneous activity patterns in systematic ways leading to an increased probability of spontaneous sequential activation. Finally, the model also captured the experimental finding that the training effect is only short-lasting. In sum, we present the first spiking neural network model that explains recent sequence learning data from rat primary visual cortex while also reproducing key structural features of cortical connectivity.
The network model we used is a member of the class of self-organizing recurrent neural network (SORN) models (see e.g. [6, 20–23]). Specifically, we used the leaky integrate-and-fire SORN (LIF-SORN) introduced by Miner and Triesch [22]. The LIF-SORN is a model of a small section of L5 of the rodent cortex. Here, we used a version of it consisting of NE = 1000 excitatory and NI = 0.2 × NE = 200 inhibitory leaky integrate-and-fire neurons with conductance based synapses and Gaussian membrane noise. The neurons are placed randomly on a 2500 μm × 1000 μm grid (Fig 1A) and their connectivity is distance-dependent, meaning a neuron is more likely to form a connection with a neuron nearby than with a remote neuron. While the weights of connections involving inhibitory neurons are fixed, recurrent excitatory connections are subject to a set of biologically motivated plasticity mechanisms. Exponential STDP with an asymmetric time window endows the network with the ability to learn correlations in external input. It is complemented by a form of structural plasticity (SP), which creates new and prunes weak connections. The network dynamics are stabilized by three additional plasticity mechanisms. First, synaptic normalization (SN) keeps the total incoming weight for each excitatory neuron constant. Second, intrinsic plasticity (IP) regulates the threshold potential of each excitatory neuron to counteract overly high or low firing rates. Third, short-term plasticity (STP) facilitates or impedes signal transmission along a specific connection based on the firing history of the presynaptic neuron.
Using this network model, we replicated the study by Xu et al. [1]. In this experiment, a multielectrode array was inserted in the primary visual cortex of rats and the receptive field of each channel was determined. The rats were then presented with a bright light spot, which was moved from a start point S ˜ to an end, or goal, point G ˜ along the distribution of receptive fields. The effect of this conditioning was assessed by measuring the responses to different kinds of cues of the full motion sequence. Xu et al. [1] investigated both awake and anesthetized rats, but unless noted otherwise, we compared our results to the results from awake animals. In the LIF-SORN, we modeled the movement of the light spot by sweeping a spot from x S ˜ = ( 375 μ m , 500 μ m ) T to x G ˜ = ( 2125 μ m , 500 μ m ) T. The amplitude of this spot at the position x n e of an excitatory neuron ne represents the rate r spot ( x n e , t ) of external Poissonian spike trains, which this neuron receives (Fig 1B). Furthermore, we approximated recording with a multielectrode array by introducing clusters of excitatory neurons. These clusters are located between x S ˜ and x G ˜ and are named A to H according to their distance to x S ˜ (Fig 1C). For the analysis of the sequence learning task, we only considered the activity of these clusters, which we defined to be the pooled spikes of the neurons part of a cluster. See the Materials and methods section for a more detailed description of our model.
To get an impression of the behavior of the LIF-SORN, we simulated the network for 500 s solely under the influence of the background noise, i.e., without external input. Thereby we also showed that it exhibits some basic properties of both the activity and connectivity in biological neural networks. We began by analyzing the spiking activity after an initial equilibration phase (Fig 2).
Both excitatory and inhibitory neurons seemed to exhibit unstructured firing (Fig 2A and 2B) with the excitatory neurons firing with frequencies distributed closely around 3 Hz due to the IP (Fig 2C) and inhibitory neurons firing with roughly twice this frequency (Fig 2D). The activity of the cortex in the absence of external stimuli is often assumed to lie in a regime of asynchronous irregular spiking. Synchrony refers, in this context, to the joint spiking of a set of neurons. Biological data shows that population level activity in the cortex is highly asynchronous [24, 25]. Mostly, the pairwise correlation coefficient is used to quantify synchrony (see e.g. [26]). The pairwise correlation coefficient between a neuron m and n is defined as
c m n = cov ( C m , C n ) Var ( C m ) Var ( C n ) , (1)
where Cm is the time series of spike counts of neuron m within successive time bins. Fig 2E shows the distribution of pairwise correlation coefficients of all disjoint pairs of excitatory neurons in the LIF-SORN for time bins of 20 ms duration. The pairwise correlation coefficients were closely distributed around zero, implying a low level of synchrony in the LIF-SORN. When varying the duration of the time bins, the mean of the distribution of pairwise correlation coefficients stayed close to zero while its width increased (decreased) with increasing (decreasing) duration of the time bins.
Regularity refers to the variability of the spiking of individual neurons. In the cortex, this spiking is highly irregular and can, apart from the refractory period, often be quite accurately described by a Poisson process [27], in which the interspike intervals follow an exponential distribution with a coefficient of variation equal to unity. We found that interspike intervals of excitatory neurons in the LIF-SORN were approximately exponentially distributed with a distortion caused by the refractory period (Fig 2F) and that the coefficients of variation were generally close to one (Fig 2G), indicating irregular spiking.
Next, we considered the structural properties of the LIF-SORN. The LIF-SORN was initialized without recurrent excitatory connections, but due to SP, these connections grew for about 200 s, as can be seen in Fig 3A. Afterwards the pruning rate of existing synapses approached the growth rate of new synapses and the network entered a stable phase in which the connection fraction of excitatory connections did not change anymore. The values of individual weights were, on the other hand, still fluctuating (Fig 3B). This constant change was also found in biological networks [13]. Additionally, the excitatory weights assumed an approximately lognormal-like distribution (Fig 3C) as observed in cortical circuits [9–13]. We also converted the connection weights in approximate amplitudes of the corresponding postsynaptic potentials (PSP). In excitatory neurons, the mean excitatory PSP (EPSP) amplitude was 0.72 mV and the mean inhibitory PSP (IPSP) amplitude was 0.96 mV and in inhibitory neurons, the mean EPSP amplitude was 0.74 mV and the mean IPSP amplitude was 0.94 mV. These values lie within the experimentally observed range [11, 12, 28]. See S1 Appendix for a description of how this conversion was done and figures of the distribution of PSP amplitudes and their ratios.
Taken together, the LIF-SORN displayed key features of both the activity and connectivity in cortical circuits. Besides the here mentioned structural properties, the LIF-SORN has also already been shown to reproduce more complex properties of cortical wiring, namely the overrepresentation of bidirectional connections and certain triangular graph motifs compared to a random network and various aspects of synaptic dynamics [22].
To investigate the sequence learning ability of the LIF-SORN, we employed a similar test paradigm as Xu et al. [1]. That means we trained the network with the moving spot as described above and tested it by stimulating the network with brief flashes of the spot at the start point x S ˜ = ( 375 μ m , 500 μ m ) T, the mid point x M ˜ = ( 1250 μ m , 500 μ m ) T and the end point x G ˜ = ( 2125 μ m , 500 μ m ) T. This testing was performed before and after training and the responses to each of the stimuli after training were compared to their counterpart before training.
Specifically, our simulation protocol started with a growth phase of 400 s duration, to initialize a network that exhibits key features of cortical circuits. It followed a test phase, during which one of the cues, i.e. a brief flash of the spot at the start point, mid point or end point, was presented once every two seconds. This first test phase lasted 100 s, leading to a total of 50 repetitions. After this test phase, the network was given a short relaxation phase of 10 s such that its activity got back to base level. Afterwards, the full motion sequence was shown to the network in the training phase, which lasted 200 s. The sequence was also presented once every two seconds leading to a total of 100 repetitions. After another relaxation phase of 10 s, the simulation ended with another test phase, during which the same cue as in the first test phase was shown to the network. This second test phase lasted 100 s leading to a total of 50 repetitions.
The purpose of the presentation of the test cue at the start point was to examine if the network learned the sequential structure of the motion sequence. Fig 4A shows the spike trains of the neurons that are part of one of the clusters A to H during training in response to one presentation of the full motion sequence as well as before and after training in response to one trial of cue presentation at the start point. While the spiking was clearly sequential during training, such sequential spiking was much less pronounced in response to the test cue before and after learning. Additionally, the spiking was quite variable over trials and different networks. Similar results were found by Xu et al. [1].
A common method to assess the sequential spiking in animal studies of sequence replay is to compute the cross-correlation between pairs of spike trains [1, 29]. We performed such an analysis in a similar way as Xu et al. [1] by pooling the spikes for each cluster and than computing the correlation between these spike trains for each trial and for all cluster combinations. Therefor, we only considered spikes within the window 0 ms–500 ms relative to stimulus onset to minimize the impact of spontaneous activity. Next, we pooled the cross-correlations according to the corresponding difference in cluster position. This was done for the test phases before and after training. We then also took the difference between the resulting cross-correlograms and finally normalized each of the three cross-correlograms to the range between 0 and 1. This was done independently for each of the sets of cross-correlations corresponding to a specific difference in cluster position. Fig 4B shows the thereby obtained cross-correlograms, which qualitatively resembled the cross-correlograms obtained from rats [1] in that the correlation function took on higher values for positive time delays compared to negative time delays even before learning—an observation that can be linked to the spread of activity from S ˜ towards G ˜—and in that this rightward slant enhanced due to training. This increase of the correlation function for positive time delays indicated that the network indeed learned about the sequential structure of the motion sequence.
To quantify the cue-triggered sequence recall, we again adapted the analysis used by Xu et al. [1]. That is to say we pooled the spikes of all neurons for each cluster and calculated their rate by convolving them with a Gaussian filter with width τrate = 50 ms. We only considered spikes within the window 0 ms–500 ms relative to stimulus onset to minimize the impact of spontaneous activity and defined the firing time of a cluster as the first peak of its rate curve. Then we computed for each test trial the Spearman correlation coefficient between the firing times of the clusters and their location on the S ˜ → G ˜ axis. The Spearman correlation coefficient between two variables is defined as the Pearson correlation coefficient between the rank values of those variables. Hence, it measured how much the replay order resembled the training order of clusters.
For the test phases with a cue at S ˜, we found a significant rightward shift of the correlation coefficient distribution after learning with a change in mean from 0.26 to 0.30 (P = 7.9 × 10−3; Kolmogorov-Smirnov test) as shown in Fig 5A. Thus, there was enhanced sequential spiking after training compared to before training as found by Xu et al. [1], who observed a change in mean from 0.21 to 0.29 (P = 1.5 × 10−3; Kolmogorov-Smirnov test; Fig 5A).
The purpose of the presentation of the test cue at the mid point was to avoid having a rightward bias even before training. For that case, Xu et al. [1] also found a significant rightward shift of the correlation coefficient distribution with a change in mean from −0.08 to −0.02 (P = 1.3 × 10−4; Kolmogorov-Smirnov test; Fig 5B). In the LIF-SORN, we observed only a very small rightward shift, which wasn’t significant, however (change in mean from 0.0 to 0.02; P = 0.31; Kolmogorov-Smirnov test; Fig 5B).
The purpose of the presentation of the test cue at the end point was to examine if the cue-triggered replay was specific to the direction of the motion sequence. Thereby, Xu et al. [1] computed the Spearman correlation coefficients between the firing times of the clusters and their location on the G ˜ → S ˜ axis and found no significant shift in the correlation coefficient distribution (change in mean from 0.20 to 0.20, P = 0.59; Kolmogorov-Smirnov test; Fig 5C), indicating that the direction of replay was indeed specific to the direction of the motion sequence used during training. In the LIF-SORN, we found a small but not significant leftward shift (change in mean from 0.22 to 0.20, P = 0.53; Kolmogorov-Smirnov test; Fig 5C). The small leftward shift can be explained by the weakening of connections between the clusters pointing in the opposite direction of the motion sequence due to STDP, since this decreased the correlation between the activity of one cluster and a cluster located further along the G ˜ → S ˜ axis. Although Xu et al. [1] did not observe even a small leftward shift, it may still be that STDP was also responsible for the training effect in rats as the weakening of connections pointing in the opposite direction of the motion sequence could have been too small to have an observable effect. Furthermore, Xu et al. [1] showed that blocking NMDA receptors lead to the disappearance of the training effect indicating that some form of NMDA-dependent plasticity was indeed responsible for the training effect.
To test whether the training effect was restricted to a small region of V1 or if it was also apparent elsewhere in V1, Xu et al. [1] performed an experiment where, during training, the motion sequence was shifted orthogonal to the long axis of the recorded distribution of the receptive fields, i.e. orthogonal to the S ˜ → G ˜ axis. They neither found a significant shift in the correlation coefficient distribution for S ˜-evoked nor for G ˜-evoked responses and concluded that the effect of learning was indeed specific to the location of the motion sequence. As in the animal study, the LIF-SORN exhibited for this scenario no significant shift in the correlation coefficient distribution for both S ˜-evoked and G ˜-evoked responses (S1 Fig).
To examine if training with a dynamic stimulus is actually necessary to achieve a more distinctive sequence replay, two different experiments using a static stimulus during training were performed by Xu et al. [1]. In the first one, this stimulus was a flashed bar spanning the region between S ˜ and G ˜. In the second one, the stimulus used during training was just a briefly flashed spot at S ˜. A significant shift in the correlation coefficient distribution was neither found for S ˜-evoked nor for G ˜-evoked responses. Again, the LIF-SORN also didn’t show significant training effects for these scenarios (S1 Fig).
The activity of the LIF-SORN is determined by the input and its connectivity. In this section, we show how the training with a moving spot modulated the connectivity through the plasticity mechanisms. Therefore, we analyzed the weight matrix of the recurrent excitatory connections before and after training with the full motion sequence from S ˜ to G ˜. This allowed us to connect a large part of the results of the previous sections with the change in connectivity.
We start by considering the weights between neurons part of one of the clusters A–H for one network instance. Fig 6A–6C show the connection weight matrix before and after training and their difference. Before training, we observed a structure with stronger weights distributed symmetrically around the diagonal. This reflected the distance dependency of the connectivity. After training, the symmetry was broken and the connections running in the direction of the moving spot used during training were strengthened while the connections in the opposite direction were weakened. To get a clearer picture of the weight change, we also determined the average connection weights between the different clusters A–H (Fig 6D). The connection weights between adjacent clusters in the forward direction were increasing due to training, while the opposite was true for the backward direction. The increase in connection weight was strongest for the A → B connections as these connections didn’t have as much SN-induced competition as the connections between adjacent clusters further along the S ˜ → G ˜ axis, since they had to compete with connections starting from other clusters located closer to S ˜.
This stripe-like connectivity was a result of STDP and caused the increase in sequential spiking when triggering the sequence with a cue at S ˜ and the decrease when triggering the sequence with a cue at G ˜.
Next, we examined if the training with a sequence had an impact on the spontaneous activity of the LIF-SORN. Therefor, we utilized the simulation protocol as described above with the difference that during testing no external input was used. Thus, only the noise drove the network during testing. As before, we determined the rate of each cluster by pooling the spikes of all neurons which were part of that cluster and convolving them with a Gaussian. However, we considered all spikes during the test phases and not only spikes within a window of 500 ms after stimulus presentation. Next we computed the times of all relative maxima of the firing rate for each cluster and ordered them. In the resulting sequence of firing times, we replaced each firing time with the corresponding cluster name A–H. Finally, we computed the transition probabilities between all clusters from this sequence. The transition probability from A to D, for example, was computed by dividing the number of times D was the cluster that fired directly after A by the total number of times A appeared. This was done for all combinations of clusters A–H and for the test phases before and after training.
Fig 7 shows the resulting transition probabilities before and after learning and their differences. Before and after learning, the transition probabilities from a cluster to itself were negligible and the transition probabilities between adjacent clusters as for example A to B or D to C were higher compared to the other transitions. This is not surprising as neurons close to one another strongly influenced each other due to the distance dependency of the connectivity. When considering the change in transition probabilities caused by the training, we observed that transitions between clusters in the S ˜ → G ˜ direction, which were separated by at most one other cluster, tended to be more likely while transitions between clusters in the opposite direction, which were separated by at most one other cluster, tended to be less likely. This finding was consistent with the weight change caused by the training (Fig 6D).
Thus, the characteristics of the sequence used during training were imprinted in the spontaneous activity of the LIF-SORN.
So far, we were only concerned with the order of the replayed sequence and did not pay attention to its speed. In this section, we examine the recall speed vrc after presentation of the test cue at S ˜ with varying velocities vspot of the motion sequence during training. We adopted the analysis of Xu et al. [1]. That is to say we considered only the test trials after learning for which the Spearman correlation coefficient was greater than 0.9. Then, we determined the recall speed for each of those trials by linear regression of the positions of the centers of each cluster as a function of their firing times. We found that the mean recall speed was independent of the training speed (Fig 8). It rather seemed to be determined by the network’s parameters.
Similar results were found by Xu et al. [1], i.e. they also observed no dependence of the recall speed on the speed used during training for anesthetized rats. Furthermore, spontaneous replay in cortex and hippocampus was found to be accelerated compared to training [29, 31]. All of these results suggest that only sequence order is learned and that the recall speed is primarily determined by the network’s dynamics and not the speed of the trained sequence. This observation on the level of local circuitry matches with the trivial fact that the recall of memories doesn’t happen with the speed with which they were experienced.
Xu et al. [1] also tested the persistence of the increase in sequential spiking caused by the training. To test this persistence in the LIF-SORN, we used a similar simulation protocol as before, i.e., training consisted of a moving spot shown along the S ˜ → G ˜ axis and testing of a briefly flashed spot at S ˜. The duration of the test phase after training was tripled.
We adapted the analysis of Xu et al. [1] in that we defined a match as a test trial whose Spearman correlation coefficient was above a threshold of 0.6 and computed the change in percentage of matches during the test phase after training compared to the test phase before training. This was done for different times after training. We found that the effect of training as measured by the change in percentage of matches decayed within approximately 5 min (Fig 9A). The training effect decayed within a similar time, namely within around 7 min, in rats (Fig 9A). Hence, the training effect was short-lasting in both rats and the LIF-SORN.
Again, we can link the results obtained from the LIF-SORN’s activity with its connectivity. During training, the forward weights between adjacent clusters were increasing for approximately 100 s and then stayed roughly constant until the training ended as a consequence of the synaptic normalization. In the following test phase, the values of the weights from one cluster to the next were decaying back to their initial values, resulting in the simultaneous decay of the training effect on the network’s activity (Fig 9B). This decay was caused by the interplay of STDP with the asynchronous, irregular network activity.
Establishing the relationship between structure and function of cortical circuits remains a major challenge. Here we have presented a spiking neural network model of sequence learning in primary visual cortex that establishes a new conceptual bridge between structural and functional changes in cortical circuits. The model posits that STDP is the cellular basis for the sequence learning abilities of visual cortex. The temporally asymmetric shape of the STDP window (pre-before-post firing leads to potentiation, post-before-pre firing leads to depression) allows the circuit to detect the spatio-temporal structure of the stimulation sequence and lay it down in the circuit structure. The homeostatic mechanisms prevent runaway weight growth, among other functions. Importantly, while doing so the model also explains the origin of key structural features of the connectivity between the population of excitatory neurons. Among them are the lognormal-like distribution of synaptic efficacies, the distance-dependence of synaptic connection probabilities and the abundance of bidirectional connections between excitatory neurons.
There are many studies that have addressed the functioning of STDP in feed-forward models (e.g. [32, 33]). In addition, several previous studies have successfully modeled elements of sequence learning with STDP in recurrent networks [5, 6, 34–37], and another set of studies has attempted to account for the development of structural features of cortical wiring [20, 22, 38, 39]. However, our model is the first to combine both. Thereby it offers the most advanced unified account of the relation between structure and function of cortical circuits in the context of sequence learning. Furthermore, it does so from a self-organizing, bottom-up perspective, a critical component missing in most other examples of artificial sequence learning in recurrent neural networks [40–42].
Equally important, however, the model makes precise and testable predictions regarding how the excitatory-to-excitatory connectivity changes during learning. Specifically, it predicts that synaptic connections that project “in the direction of” the stimulation sequence are strengthened, while the reverse connections are weakened. Furthermore, it makes the testable prediction that spontaneous activity after learning should reflect the altered connectivity such that it leads to an increased probability of sequential activation. Similarly, replay of activity patterns is a well-known and widely studied phenomenon in hippocampus [43], but has also been observed in the neocortex [29, 31]. Despite these contributions, our model also has several limitations.
First, while our network reproduced most of the experimentally observed results by Xu et al. [1], namely enhanced sequential spiking in response to a cue at the start point of the sequence (Figs 4B and 5A), independence of the recall speed on the training speed (Fig 8) and a short persistence of the training effect (Fig 9A), it did not show the experimentally observed significant rightward shift of the Spearman correlation coefficient distribution in response to a cue at the midpoint (Fig 5B) and it exhibited an, experimentally not observed, small leftward shift of the Spearman correlation coefficient distribution in response to a cue at the goal point (Fig 5C).
Second, most of the chosen neuron and network parameters were taken from studies on layer 5 of rodent cortex [11, 12, 14], while Xu et al. [1] recorded from both deep and superficial layers.
Third, synaptic plasticity in our model was restricted to the connections among excitatory neurons. As a consequence, inhibition is unspecific in our network. From a functional perspective, this shouldn’t make much of a difference for the simple sequence learning task we considered. For more complex situations, such as multiple disparate assemblies, multiple sequences or branching sequences, this may be different, however. Adding plasticity mechanisms to the other connection types would also make the model more realistic and may allow it to establish additional links between the structure and function of cortical circuits in the context of sequence learning. This will be an interesting topic for future work.
Additional limitations exist in the model as a function of computational practicality. These include network size and related subsampling effects, as well as more complex input structures, noise correlations, etc. Overcoming these limitations would also be an interesting topic for future investigation.
Finally, in both the experiments of Xu et al. [1] and our model the training effect persists for only a short time. Xu et al. [1] astutely noted that even such short term storage can be quite useful for perceptual inference [44, 45], as repeated experiences in the recent past are often a good predictor of similar experiences in the near future. It is also clear, however, from perceptual learning experiments that visual cortex can store information for long periods of time [46]. So how are new memories protected from being quickly forgotten? How can they be stabilized for weeks, months, and years? This is an important question for future work.
The LIF-SORN is a recurrent neural network model of a small section of L5 of the rodent cortex. It consists of noisy leaky integrate-and-fire neurons and utilizes several biologically motivated plasticity mechanisms to self-organize its structure and activity. It was introduced by Miner and Triesch [22] with the plasticity mechanisms being short-term plasticity (STP), spike-timing dependent plasticity (STDP), synaptic normalization (SN), structural plasticity (SP), and intrinsic plasticity (IP). Here, we employed a modified version in comparison to this study. The most major changes were the use of a conductance based model instead of an additive model of synaptic transmission to make the synaptic signaling more realistic, the adjustment of the SN mechanism to account for boundary effects and the enlargement of the network to a size that is similar to size of the cortical region considered by Xu et al. [1].
The parameter values in the LIF-SORN are set in accordance to experimental data from L5 of the cortex, although the timescales of SN, SP and IP are accelerated compared to biological findings in order to decrease the necessary simulation time. See [22] for a more detailed explanation of the selection of the individual values than the explanations given below. The network was simulated with the help of the Brian spiking neural network simulator [47] using a simulation timestep of Δtsim = 0.1 ms.
|
10.1371/journal.pntd.0001503 | Diversification of Schistosoma japonicum in Mainland China Revealed by Mitochondrial DNA | Schistosoma japonicum still causes severe parasitic disease in mainland China, but mainly in areas along the Yangtze River. However, the genetic diversity in populations of S. japonicum has not been well understood across its geographical distribution, and such data may provide insights into the epidemiology and possible control strategies for schistosomiasis.
In this study infected Oncomelania snails were collected from areas in the middle and lower (ML) reaches of the Yangtze River, including Hubei, Hunan, Anhui, Jiangxi and Jiangsu provinces, and in the upper reaches of the river, including Sichuan and Yunnan provinces in southwest (SW) China. The adult parasites obtained from experimentally infected mice using isolated cercariae were sequenced individually for several fragments of mitochondrial regions, including Cytb-ND4L-ND4, 16S-12S and ND1. Populations in the ML reaches exhibited a relatively high level of diversity in nucleotides and haplotypes, whereas a low level was observed for populations in the SW, using either each single fragment or the combined sequence of the three fragments. Pairwise analyses of F-statistics (Fst) revealed a significant genetic difference between populations in the ML reaches and those in the SW, with limited gene flow and no shared haplotypes in between. It is rather obvious that genetic diversity in the populations of S. japonicum was significantly correlated with the geographical distance, and the geographical separation/isolation was considered to be the major factor accounting for the observed difference between populations in the ML reaches and those in the SW in China.
S. japonicum in mainland China exhibits a high degree of genetic diversity, with a similar pattern of genetic diversity as observed in the intermediate host snails in the same region in China.
| Despite the existing threat of schistosomiasis in some rural areas along the Yangtze River, the genetic diversity of Schistosoma japonicum has not been investigated across its wide geographical distribution in China, and such information may provide insight into the disease epidemiology and the development of its control measures. In this study, the adult parasites, obtained through infecting mice with cercariae from snails of the genus Oncomelania collected from a wide range of localities in currently endemic areas of schistosomiasis in the middle and lower (ML) reaches of the Yangtze River, and in Sichuan and Yunnan provinces in the upper reaches of the river in southwest (SW) China, were sequenced individually for mitochondrial genes. In general, a relatively high degree of genetic variation was observed in populations in the ML reaches in terms of nucleotide and haplotype diversity, but a low level was observed in populations in the SW. The significant difference in genetic diversity as revealed by F-statistics, and the existence of no shared haplotypes, were observed between populations in the ML reaches and those in the SW, indicating the effect of geographical separation/isolation upon the schistosomes and probably the parasite-snail system in China.
| Schistosomiasis is one of the most neglected tropical diseases, with six species in the Schistosoma still infecting more than 200 million people in the world [1]–[3]. Schistosomiasis japonica is distributed in Indonesia, Philippines, and China. In mainland China, this parasitic disease is the most severe zoonosis infecting about 360,000 people and about 1% buffalo and/or cattle in endemic regions, particularly in lake/marshland and hilly areas of Hubei, Hunan, Anhui, Jiangxi and Jiangsu provinces and mountainous areas of Sichuan and Yunnan provinces [4]. Over the last 50 years, continuous efforts involving various measures, such as health education, snail control, community chemotherapy and environmental management have contributed significantly to the dramatic reduction in infection levels and epidemic areas of this parasitic disease in China, setting China as one of the most successful countries in control of schistosomiasis in the world [5]–[8]. However, recently available data have suggested that schistosomiasis has re-emerged over the last decade, probably as a severe threat once again to human health especially in rural areas of mainland China [9], [10]. The drastic pathogenesis, the number of reservoir hosts involved in epidemiology and the vast endemic areas of schistosomiasis japonica have inevitably resulted in a less investigated status for S. japonicum in respect with its genetic diversity, host immune response etc. when compared with other schistosomes [6], [11], [12].
The genus Oncomelania, which is the intermediate host of S. japonicum, was classified into different species and/or subspecies according to their morphology, biogeography and phylogeny [13], [14]. With the distinct diversity of snails in the genus Oncomelania which has been verified using various markers [14]–[16], the diversity of the parasite S. japonicum is of great interest for research from a co-evolutionary point of view. How diverse the digenean S. japonicum really is in such a large geographical range has not been well assessed especially in mainland China. An accurate measure of its population genetic diversity is certainly needed to clarify our understanding on the epidemiology of schistosomiasis [17], which may be also useful for implementing control measures, and for developing drugs or potential vaccines, as worms of different genetic backgrounds may respond differently to such treatments [18], [19].
In recent years, several molecular markers have been used to detect the variability of S. japonicum populations. Gasser et al. [20] found the variability among 7 geographical isolates across mainland China using the random amplified polymorphism DNA (RAPD) technique and suggested a potential strain complex for S. japonicum. Sorensen et al. [21] reported differences between S. japonicum populations from 6 localities in mainland China using NADH dehydrogenase subunit 1 (ND1) gene, but could not detect variability conclusively at the intrapopulation level. Bøgh et al. [22] did find 15 types of ND1 conformations and 23 types of cytochrome c oxidase subunit 1 (CO1) conformations in 9 populations from 7 provinces across mainland China by single-strand conformational polymorphism (SSCP). These results did in fact suggest the significant polymorphism among S. japonicum in mainland China, but provided very limited information relating to the population genetic diversity of this species. Upon the identification of polymorphic microsatellite loci, Shrivastava et al. [6] investigated the genetic variation of S. japonicum populations from 8 geographical locations in 7 endemic provinces across mainland China, and a high level of polymorphism was reported between and within populations. They considered that populations of S. japonicum in mainland China could be separated mainly into the populations in Sichuan and Yunnan provinces as being in southwest (SW) China and those in low-lying lake regions along the middle and lower (ML) reaches of Yangtze River. With three partial mitochondrial genes (cox3, nad4 and nad5) from 28 individual adult worms, Zhao et al. [23] reported recently that all parasites from SW China were grouped together, whereas those from the ML reaches of Yangtze River were not clustered together. However, the reports by Shrivastava et al. [6] and Zhao et al. [23] both contained limited specimens from relatively few localities, which may not represent the geographical distribution of this schistosome, and thus not its exact population genetic diversity, in mainland China. A comprehensive analysis is therefore needed using more molecular markers to examine more populations of S. japonicum from a wide range of its geographical distributions, especially in severely endemic areas along the ML reaches of Yangtze River in China.
In this study, mitochondrial DNA sequences including Cytb-ND4L-ND4, 16S-12S and ND1 were examined for S. japonicum collected from localities in seven provinces of China, where schistosomiasis is geographically endemic. The diversity in nucleotides and haplotypes was analyzed for different populations based on each of the three mitochondrial sequences and their combined sequences. Phylogenetic tree and parsimony network were constructed for observed haplotypes, and the genetic distance was examined against the geographical distance in order to understand the genetic diversity in populations of S. japonicum in mainland China.
The procedures involving animals were carried out in accordance with the guidelines of the Association for Assessment and Accreditation of Laboratory Animal Care International (AAALAC). The animal study protocol was approved by the Institutional Animal Care and Use Committee of Wuhan University.
The intermediate host, Oncomelania hupensis, from 18 localities of 7 schistosomiasis endemic provinces in mainland China, including Hubei, Hunan, Anhui, Jiangxi, Jiangsu provinces in the ML reaches of Yangtze River, and Sichuan and Yunnan provinces which are in the higher reaches of the river in SW China, but separated from the ML reaches by mountain ranges (Fig. 1 and Table 1), were collected and transported to laboratory from October 2005 to October 2006. After one month captivity, snails were washed and exposed individually in water for 3 h in a vial under light at 25°C to stimulate the emergence of cercariae for identifying the S. japonicum infection. Overall, snails from different localities had an infection rate ranging from 0.1% to 4%. To generate adult worms, the best source of DNA, 10 field-collected infected snails from each locality, with the exception of Zongyang in Anhui province (AHzy) and Pengze in Jiangxi province (JXpz) where only three and four infected snails were obtained respectively, were exposed to light for 4 hours to stimulate the emergence of cercariae. Five laboratory mice were infected percutaneously with 30 cercariae per mouse for each geographical locality. 6 weeks following the infection, adult worms were retrieved by perfusion from mesenteric veins using 0.9% NaCl, and worms from each mouse infected with cercariae were pooled together, and washed extensively in saline before being preserved in 95% ethanol at 4°C.
The total genomic DNA was extracted individually from both male and female schistosomes using a standard sodium dodecyl sulfate-proteinase K procedure [24]. Each worm was incubated and thawed in 200 µl extraction buffer containing 50 mM Tris-HCl, 50 mM EDTA, 100 mM NaCl, 1% SDS and 100 µg/ml proteinase K, at 56°C for 2 h with gentle mixing. DNA in solution was extracted using standard phenol/chloroform purification, followed by 3 M sodium acetate (pH 5.2) and ethanol precipitation. Pellets of DNA were washed in 70% ethanol, air-dried, and resuspended in 10 µl TE (pH 8.0).
For each adult worm, three fragments, i.e. Cytb-ND4L-ND4, ND1 and 16S-12S of the mitochondrial genome were sequenced. For the Cytb-ND4L-ND4 fragment, the forward primer ND4F (5′- TTGGGGGTTGTCATGCGGAGTA -3′) and the reverse primer ND4R (5′- CAAATACCCAATAGCAACGGAACAC -3′) were used based on available GenBank sequence AF215860. For the ND1 fragment, the forward primer ND1F (5′- TAGAGGGTTTGTTGGTTGTTTTG -3′) and the reverse primer ND1R (5′- ACCATACTTTCATACTACTGCC -3′) were used based on available GenBank sequence AF215860. For the 16–12S fragment, the forward primer 16S-12SF (5′- GATTATTTCTAGTTCCCGAATGG -3′) and the reverse primer 16–12SR (5′- TGTAACGCACAACAACCTATACC -3′) were used based on available GenBank sequence AF215860. The PCR protocols were 94°C for 3 min followed by 30 cycles of 94°C for 30 s, 58°C (for ND1) or 63°C (for Cytb-ND4L-ND4 and 16S-12S) for 30 s, and 72°C for 90 s and then a final elongation step at 72°C for 10 min. The amplified products were purified on 1.0% agarose gel stained with ethidium bromide, using the DNA gel extraction kit (Omega Bio-Tek). The purified PCR products were sequenced using ABI PRISM BigDye Terminators v3.0 Cycle Sequencing (ABI). The DNA sequences were deposited in the GenBank database under accession numbers FJ851893–FJ852573.
Sequences were aligned using ClustalX1.83 [25] at default settings followed by manual correction in SEAVIEW [26] for each molecular marker. DNAsp version 4.0 [27] was used to define the haplotype. The three parts, i.e. Cytb-ND4L-ND4, ND1 and 16S-12S, of mitochondrial data were also combined and aligned into a new combined mitochondrial data set, with this combined sequence named as combined mtDNA.
Nucleotide divergences within and between populations were calculated in Arlequin3.11 [28] and DNAsp. Genetic variation within different populations was estimated by calculating nucleotide diversity (π) and haplotype diversity (h) values. Selective neutrality was tested with Tajima's D [29] and Fu's F test [30]. The pairwise genetic difference was estimated for all populations by calculating Wright's F-statistics (Fst) based on gene flow (Nm). A Mantel g-test to compare the correlation between pairwise distance and geographical distance among localities was analyzed in Arlequin, with geographic distances (km) for the correlation analysis between geographical distance and genetic distance calculated using the great circle distance between localities.
The phylogenetic analysis for 96 haplotypes generated using combined mitochondrial DNA data was performed with Bayesian inference (BI), which was carried out with MrBayes 3.1 [31] under the best-fit substitution model. Analyses were run for 1×106 generations with random starting tree, and four Markov chains (with default heating values) sampled every 100 generations. Posterior probability values were estimated by generating a 50% majority rule consensus tree following the discard of first 3000 trees as part of a burn-in procedure. The HKY+I+G model was determined as the best-fit model of sequence evolution by using the hierarchical likelihood ratio tests implemented in Modeltest 3.7 [32]. The phylogenetic tree was rooted using Schistosoma mansoni as outgroup.
The genetic structure was phylogenetically evaluated by constructing unrooted parsimony network of haplotypes for combined mtDNA data sets using TCS version 1.21 [33].
The primary sequence data were obtained by amplifying and sequencing three partial regions of the mitochondrial genome, i.e. Cytb-ND4L-ND4 with 793–794 bp, ND1 with 767 bp, and 16S-12S with 1463–1466 bp. Measures of diversity of haplotypes and nucleotides within populations on the basis of the three mitochondrial regions are presented in Tables S1, S2 and S3, respectively. The highest values for the diversity were all observed for populations in the ML reaches, and the lowest all in populations from the SW (for details regarding each fragment, see Tables S1, S2 and S3). The pairwise genetic distance among all 18 populations showed a high degree of variation, as revealed respectively from the three different mitochondrial regions (for details, see Tables S4, S5 and S6).
A significant correlation was observed between geographical distance and genetic distance (pairwise Fst) for all 18 populations for Cytb-ND4L-ND4 (R = 0.642, P<0.001) and 16S-12S (R = 0.746, P<0.001), respectively, which indicates that genetic distance increased with the increase in geographical distance (Fig. 2a, b). No significant correlation was detected when ND1 was used, with the correlation coefficient R = 0.080 (P>0.05) (Fig. 2c). However, among 15 populations in the ML reaches, the value of the correlation coefficient decreased to 0.119 (P>0.05) and 0.061 (P>0.05) for Cytb-ND4L-ND4 and 16S-12S, respectively (Fig. 2d, e), implying that the genetic distance was not correlated with the geographical distance for populations in the ML reaches of Yangtze River.
Although some base substitutions were observed, selective neutrality of the observed nucleotide polymorphisms was suggested for S. japonicum, as indicated either by Tajima's D or Fu's F test (P>0.05) in each of the three regions.
As many studies have shown that longer genes contain generally more variable characters with proportionally more signals, and hence yield accurate phylogenetic estimates than shorter ones [34]–[36], the combined mitochondrial data sets were then deduced from 169 specimens by aligning combined Cytb-ND4L-ND4, ND1 and 16S-12S sequences (combined mtDNA), which had a range of 3024 to 3027 bp, resulted in 3028 characters, including gaps, and 166 variable sites (113 parsimony informative sites). A total of 96 mitochondrial haplotypes was observed (Table 1). Measures of haplotype and nucleotide diversity based on combined mtDNA are presented in Table 2. The highest values in the diversity of haplotype and nucleotide were all observed for populations in the ML reaches, and the lowest were all in populations from the SW, which is consistent with the findings from the three separate mitochondrial DNA sequences. 88 haplotypes were isolated from 143 specimens in five provinces along the ML reaches, with the mean haplotype and nucleotide diversity being 0.987±0.003 and 0.0036±0.0001, respectively. However, only 8 haplotypes were isolated from 26 specimens in the SW, with the haplotype and nucleotide diversity being 0.766±0.075 and 0.0017±0.0003, respectively. The Fst of all pairwise analyses varied from 0.482 to 0.870 between populations in the ML reaches and those in the SW (Table 3), showing highly significant difference (P<0.001). Among the 3 populations in the SW, the Fst between SCxc and two Yunnan populations (YNey and YNhq) showed highly significant differences (P<0.001), whereas no significant difference was observed between YNey and YNhq (P>0.05). Among the 15 populations in the ML reaches, the Fst varied from 0.014 to 0.807 (Table 3), with most of them being significantly different (P<0.05). When all specimens were classified into two populations according to whether they were from above or below the three Gorges region, i.e. population in the ML reaches of the Yangtze River and population in Sichuan and Yunnan provinces of the SW China, the value of genetic distance (Fst) and the gene flow (Nm) between them was 0.381 (P<0.001) and 0.410, respectively.
Significant correlation was also observed between geographical distance and genetic distance (pairwise Fst) among all 18 populations for combined mtDNA (R = 0.670, P<0.001), indicating that genetic distance increased with the increase in geographical distance (Fig. 2f). Among 15 populations in the ML reaches, the value of the correlation coefficient decreased to 0.077 (P>0.05) (Fig. 2g), implying that the genetic distance was not correlated with the geographical distance for populations in the ML reaches of Yangtze River.
As shown in the Bayesian phylogenetic tree (Fig. 3), two clades can be clearly separated. Clade A contains almost all haplotypes from all five provinces in the ML reaches of the Yangtze River. Although various divergence and some subclades were observed within this clade, support probabilities for each clade were generally very low. Haplotypes in the ML reaches were clustered in various subclades, and no obvious lineage was observed for haplotypes from different provinces along the ML reaches. However, subclades A1 and A2 include most haplotypes from Hubei, Hunan, Anhui, and Jiangxi provinces, and subclade A6 includes haplotypes from Hubei, Hunan, Anhui, and Jiangsu provinces. It is apparent that clade B can be separated into two distinct subclades, B1 and B2, with clade B1 having a high support probability and containing only haplotypes from Sichuan and Yunnan provinces in SW China, and B2 containing three haplotypes from three provinces in the ML reaches. Surprisingly, other trees (NJ, ML, MP; not shown), although inconsistent in their respects, all had such two clades containing haplotypes from SW China, and three from the ML reaches, despite a relatively low level of support probabilities.
The network constructed by statistical parsimony from 96 haplotypes on the basis of combined mtDNA sequences showed some characters as observed in the phylogenetic tree. The haplotype network was rather complicated, without any obvious lineages for those haplotypes from localities in the ML reaches (Fig. 4). However, all haplotypes from SW (from H26 to H33) were clustered together (Fig. 4), which corresponded exactly to clade B1 in Fig. 3, and this clade contained no haplotypes from the ML reaches of Yangtze River, but was related with a few haplotypes from the ML reaches (Fig. 4), as also indicated in clade B2 which formed, together with B1, into clade B (Fig. 3). A relatively large network containing haplotypes (from H71 to H93) from about 10 localities (Fig. 4) showed some similarity with clade A1 in Fig. 3, in composition of haplotypes. It is, however, impossible to detect any other patterns of haplotype networks, and impossible to find other geographical relationships or characteristic lineages in other network branches, which is largely consistent with the complex structure of clade A in Fig. 3.
The difference in genetic diversity of S. japonicum populations was demonstrated in samples collected from currently epidemic areas of schistosomiasis in mainland China, with the use of three mitochondrial fragments, Cytb-ND4L-ND4, ND1 and 16S-12S, respectively, and the combined sequences of these three fragments. The present study contains the mostly widespread and the largest number of S. japonicum populations in any attempts so far to examine the parasite genetic diversity in China. Overall, populations of S. japonicum in mainland China showed a relatively large degree of variation in terms of nucleotide and haplotype diversity. However, it is apparent that across the geographical distribution of schistosomiasis endemic areas in China, the genetic distance was correlated significantly with geographical distance when Cytb-ND4L-ND4, 16S-12S, and combined mtDNA were used, although non-significance was observed for ND1. It is even more obvious that as revealed through analyses of nucleotide and haplotype diversity, populations in Hubei, Hunan, Anhui, Jiangxi and Jiangsu provinces, namely in the ML reaches of Yangtze River showed a much larger degree of genetic variation than those in Sichuan and Yunnan provinces of the SW China in the upper reaches of the river, and no haplotypes were shared between populations in the ML reaches and those in the SW. Significant difference was also observed in genetic distance between populations in the ML reaches and populations in the SW, as revealed in pairwise analyses using individual and/or combined mitochondrial sequences.
Along the Yangtze River, are the endemic areas of schistosomiasis, and severe epidemic areas are mainly in the ML reaches [5]. However, in the Three Gorges area that is from Yichang going upwards to Yibin (Fig. 1), human schistosomiasis has never been reported [10]. It is quite obvious that the distribution of S. japonicum is geographically separated by the gorge area of the river. This apparent geographical separation may account for the observed difference in no-shared haplotypes, and in the genetic distance for S. japonicum between areas in the ML reaches of Yangtze River and areas in the SW China. When populations from the ML reaches and from the SW were further grouped separately, the Fst value (0.381) was greater than 0.25, a value which was considered to be ‘very great’ by Wright [37] for genetic differentiation between populations. It is therefore all indicated that a large level of genetic differentiation has evolutionarily occurred for S. japonicum, due to at least the geographical separation by the Three Gorges area and mountains. Phylogenetic analyses and haplotype network may support this conclusion, as parasites from Sichuan and Yunnan provinces in the SW were all closely clustered in the phylogenetic tree and the haplotype network. Using different molecular markers, other authors [6], [23] have also, to some extent, detected the genetic difference between S. japonicum populations in the SW and those in the flood plain of the ML reaches of the Yangtze River.
Despite the finding that the mean nucleotide and haplotype diversity of populations in the SW were rather low when compared with the same parameters in the ML reaches, the genetic distance had some significant difference between the population from Sichuan, SCxc, and the two populations from Yunnan, YNey and YNhq, as revealed by Fst of pairwise analyses using ND1, 16S-12S, and the combined mtDNA sequences, with the exception of Cytb-ND4L-ND4. Sichuan and Yunnan provinces are both distributed in Hengduan Mountains, and schistosomiasis was reported historically in various localities in these two provinces [38]. As various mountain ranges and rivers, as well as intermountain basins, are the general features in Hengduan Mountains [39], there must be some degree of geographical isolation in the distribution of S. japonicum in this region at a large geographical scale. However, only three populations were included in the present study and efforts to obtain more parasite samples have been unsuccessful, although the intermediate host snails were collected in a much wider range (unpublished data), due possibly to the continuous and extensive practices in either snail control or human chemotherapy in the two provinces. Thus, whether there is an effect of geographical isolation on populations of S. japonicum in this mountainous area will likely remain unknown, and whether the observed low level of genetic variation in these populations resulted from a recent bottleneck effect as a consequence of intensive control practices may also remain to be answered.
Ecological habitats were thought to affect population genetic diversity of S. japonicum in mainland China [40]. The mountainous habitats in Sichuan and Yunnan provinces may differ obviously from the habitats for the intermediate host in the ML reaches, in several aspects such as in hydrology, altitude and soil etc. [41], [42], but the difference should mostly be attributed to the geographical separation, rather than a simple impact from habitat difference. In the ML reaches of Yangtze River, it was impossible to clarify any patterns of haplotype clustering in relation to types of sample localities or to provinces, as haplotypes from a single locality were generally clustered in different clades. It can thus be speculated that S. japonicum might have experienced frequent gene flows in most populations in this region (Table 3). The localities for O. hupensis in the ML reaches have extensive physical connections through channels with the Yangtze River. With frequent occurrence of floods in the Yangtze River basin, especially in its ML reaches, snails in these habitats can be dispersed and subsequently deposited widely in various localities, and this naturally occurred instance was, in a previous research, proposed to explain the high genetic diversity of O. hupensis in the ML reaches [16]. It was further considered that this distinct genetic diversity in snail intermediate hosts may have strong implications in genetic diversity of schistosomes in mainland China [16], as demonstrated clearly in the present study. The accumulation of mixed sources of snails, especially infected snails can reconstitute the parasite population, leading to the existence of various haplotypes within a single population, and also to the limited degree of genetic distance between populations in the ML reaches as observed in the present study, which supports the speculation by Davis et al. [43] that floods may be the cause of the widespread mixing and dispersal of snails, leading to greater genetic diversity in O. hupensis populations along the Yangtze River plains compared with populations in SW China.
Surprisingly, the number of haplotypes, being 80 and 13 for the intermediate host snails in the ML reaches, and in Sichuan and Yunnan provinces [16], matches roughly, if not coincidently, with the number of haplotypes, 88 and 8, for S. japonicum in the ML reaches and in Sichuan and Yunnan provinces in this study, respectively. The intermediate host snails and the schistosome in China exhibit a lesser degree of genetic diversity in the SW, but a relatively larger degree in the ML reaches of the Yangtze River, as reported in a previous study on the intermediate host snails [16] and in this study. No shared haplotypes were observed either in the intermediate host snails or in the schistosomes between localities from the ML reaches and from the SW. Zhao et al. [44] recently reported that the intermediate host snails O. hupensis robertsoni in Sichuan and the snail O. hupensis hupensis in the ML reaches had a 10.3% genetic distance, strongly indicating that the two subspecies may differ at the species level. In a phylogenetic study on the Schistosomatidae, Lockyer et al. [45] considered that schistosomes in east Asia and their intermediate hosts in the Pomatiopsidae may be considered as the only co-evolutionary model between schistosomes and their intermediate host snails. Davis et al. [46] also speculated, as snail population forms have diverged genetically, so must their associated schistosomes or else become regionally extinct. However, it would be only possible to examine such relationship if the intermediate host snails and schistosomes are collected from a large geographical range in east Asia.
In a very small-scale area in Anhui province of China, Rudge et al. [40] detected strong genetic differentiation in S. japonicum between two types of habitats, lake/marshland region and hilly region, and suggested that contrasting host reservoirs may be associated with the genetic differentiation, with rodents and dogs being important infection reservoirs in hilly regions and bovines in lake/marshland regions. On the other hand, they found little or no parasite genetic differentiation among host species within most villages; but in another study, Wang et al. [47] reported that schistosomes were separated into two clades representing the parasites from different definitive hosts. It seems likely that S. japonicum has undergone genetic differentiation in a relatively small-scale area, as in a large geographical region reported in this study. In the above two studies, miracidia from definitive hosts were examined with microsatellite markers. In the present study, adult parasites were obtained through infecting mice with cercariae. As definitive host-based genetic variation in S. japonicum has been noticed [40], [47], the selection pressure through definitive host may need to be further investigated.
Unexpectedly, three haplotypes representing some schistosomes from three localities, each in Hubei, Hunan, Anhui provinces, were actually clustered together within another clade containing all haplotypes from Sichuan and Yunnan provinces. It is, however, at present impossible to explain this mixed cluster. As the movement of people has been frequent in China [48], the possible transmission through definitive host cannot be ruled out as a possible interpretation.
In conclusion, substantial genetic diversity was demonstrated in populations of S. japonicum in schistosomiasis endemic areas in mainland China. Overall, a significant correlation was observed between the genetic distance and the geographical distance among the populations. It is apparent that the populations from Sichuan and Yunnan provinces in SW China exhibited a relatively low level of genetic variation, and were genetically different from the populations in the ML reaches of the Yangtze River, which had a much more complicated genetic diversity. Such obvious genetic diversity should be taken into consideration in guiding any strategic control programmes and/or vaccine development/trials in the future.
|
10.1371/journal.pntd.0007514 | Sampling strategies for monitoring and evaluation of morbidity targets for soil-transmitted helminths | The current World Health Organization (WHO) target for the three major soil-transmitted helminth (STH) infections is to reduce prevalence of moderate-to-heavy infections to below 1% by 2020. In terms of monitoring and evaluation (M&E), the current WHO guidelines for control of STHs recommend evaluation of infection levels in school-age children (SAC) after five to six years of preventive chemotherapy (PC), using the standard Kato-Katz faecal smear. Here, we assess the predictive performance of various sampling designs for the evaluation of the morbidity target.
Using two mathematical models for STH transmission and control, we simulate how the number of villages and SAC sampled affect the ability of survey results in sentinel villages to predict the achievement of the morbidity target in PC implementation units (e.g. districts). As PC is stopped when the prevalence of infection in SAC in sentinel villages is less than 1%, we estimate the positive predictive value (PPV) of this indicator for meeting the morbidity target in the whole district. The PPV varies by species and PC strategy, and it is generally higher in areas with lower pre-control prevalence. Sampling a fixed number of SAC spread out over 10 instead of 5 sentinel villages may increase the PPV by up to 20 percentage points. If every SAC in a village is tested, a higher number of villages may increase the PPV by up to 80 percentage points. Increasing the proportion of SAC tested per village does not result in a relevant increase of PPV.
Although the WHO guidelines provide a combined strategy to control the three STH species, the efficacy of PC strategies clearly differs by species. There is added value in considering more villages within implementation units for M&E of morbidity targets, the extent varying by STH species. A better understanding of pre- and post-control local STH prevalence levels is essential for an adequate M&E strategy including the definition of morbidity targets at the appropriate geographical scale.
| Soil-transmitted helminths (STHs) affect approximately 1.5 billion people worldwide. The World Health Organization target for STHs is to achieve <1% prevalence of moderate-to-heavy infections among school-age children (SAC) through preventive chemotherapy (PC) with albendazole or mebendazole. For monitoring and evaluation (M&E) of STH infections, it is recommended to assess the infection levels in SAC after five to six years of treatment and to stop PC if the prevalence of infection is <1%. In this study, we use two mathematical models to assess the predictive performance of different sampling designs for the evaluation of the morbidity target. We find that the efficacy of PC strategies differs significantly by species and pre-control infection levels. Achieving 1% prevalence of infection in sentinel villages may still imply a prevalence of moderate-to-heavy infections >1% in the entire PC implementation unit. Evaluating the prevalence of any infection over a higher number of sentinel villages improves the accuracy in assessing the morbidity target at the implementation unit level, even when a lower proportion of SAC per village is tested. A better understanding of pre- and post-control local STH prevalence levels through large scale data collection is essential for the definition of an adequate M&E strategy for STH control.
| Approximately 1.5 billion people are infected with soil-transmitted helminths (STHs) worldwide [1]. The main STH species are Ascaris lumbricoides (roundworm), Trichuris trichiura, (whipworm) and hookworm (Ancylostoma duodenale and Necator americanus). Morbidity due to STHs is closely related to worm burden: chronic, moderate and high-intensity STH infections cause malnutrition, anaemia, stunted growth and impaired physical development in children [1]. The World Health Organization (WHO) global target for STHs is to eliminate morbidity in high-risk groups by 2020, defined as achieving less than 1% prevalence of moderate-to-heavy infections among pre-school-age children (preSAC, age 2–5), school-age children (SAC, age 5–14), and women of childbearing age [2]. Current WHO treatment guidelines recommend annual or semi-annual preventive chemotherapy (PC) using single-dose albendazole or mebendazole with a coverage of at least 75% of the aforementioned risk populations [1]. In practice, PC programmes are mostly school-based and therefore cover SAC and in some situations preSAC as well. PC is implemented semi-annually if the prevalence of any STH infection in SAC is higher than 50%, and annually if STH prevalence is between 20 and 50%. No PC is implemented if the pre-control STH prevalence is below 20%.
The WHO strategic plan for 2011–2020 [2] provides some recommendations on the monitoring and evaluation (M&E) of STH control programmes. However, they lack details on the recommended implementation of M&E and the interpretation of M&E results. Providing clear criteria to decide when to scale up/down or to stop PC would be very useful to guide programmes towards achieving the control goals [3]. Current recommendations [2] state that infection levels in SAC should be evaluated five to six years after starting annual or semi-annual PC. Based on the results, the decision is taken to either 1) continue PC as implemented initially (once or twice per year), 2) scale up PC frequency to three times per year, 3) scale down PC frequency to once every year or once every two years, or 4) stop PC altogether. The latter happens if the prevalence of infection (any intensity) in SAC is less than 1% [2]. All these criteria [4] are based on assessing the prevalence and intensity in sentinel sites selected proportionally to the number of SAC living in each ecological zone, using the Kato-Katz (KK) diagnostic method. Ecological zones are defined by the Food and Agriculture Organization of the United Nations (FAO) [5]. A clear indication on the number of sites and SAC to be sampled is lacking, as is the definition of standard indicators for M&E [2].
Previous work [6] has showed that there are practical difficulties in defining ecozones and in allocating different implementation units (e.g. districts consisting of a number of villages) to a single ecozone. Also, the impact of control programmes is likely non-homogeneous within ecological zones due to varying PC strategies in the different implementation units. In a comparison of survey designs for STH M&E in Kenya [6], using the recommended WHO approach (districts are aggregated according to ecological zone), appeared to be the least accurate sampling strategy.
A thorough understanding on how to assess the achievement of the prevalence thresholds to scale down or stop PC and how they relate to the achievement of the morbidity goal is currently lacking. In this study, we use two individual-based stochastic models developed independently by research groups at Imperial College London (ICL) and Erasmus Medical Center Rotterdam (Erasmus MC) to investigate how well different M&E strategies based on different sampling designs can detect whether the morbidity target of less than <1% prevalence of moderate-to-heavy intensity infections is achieved in a district after five to six years of PC.
Since the decision on whether to continue PC after the evaluation at five to six years is made by the programs at the level of PC implementation units, we construct ensembles of stochastic simulations that represent villages within districts, according to a realistic distribution of pre-control prevalence of infection. The models are then used to simulate the distribution of prevalence and intensity levels after five to six years of PC. Different post-control M&E strategies are then modelled in terms of different sampling schemes, i.e. different numbers of sentinel villages per district and different proportions of SAC sampled per sentinel village. Finally, we calculate the positive predictive value (PPV) of the indicators built using the prevalence of any infection in SAC (as measured by a single KK slide) and several threshold values to predict the achievement of <1% prevalence of moderate-to-heavy infection in all SAC in the entire district. When the threshold value is chosen to be 1%, this corresponds to assessing the probability that stopping PC in the implementation area correctly reflects the achievement of the morbidity target 5 years after the start of STH control.
The two models developed by Erasmus MC and ICL are individual-based stochastic transmission models, that allow the simulation of different STH transmission conditions and the impact of different control strategies. Both models are based on similar biological and demographic assumptions, the acquisition and death of worms are stochastic, with species-specific mean lifespans. In both models, exposure and contribution of the worms to the infective pool in the human habitat (i.e. practice of open defecation) are age-specific but differ in the functional forms and parametrisation. Based on the age pattern in hookworm infection levels, the Erasmus MC model assumes that the practice of defaecation increases with age up to age ten, and this pattern in open defaecation is then also applied to T. trichiura and A. lumbricoides. The ICL model assumes that age-dependent contribution is proportional to age-dependent exposure, and therefore it differs among the three species. Formal description of the Erasmus MC model has been published previously [7,8]. The individual-based ICL model has been presented in previous studies [9,10] and described in its deterministic version in earlier work [11]. Further details on specific assumptions, functional forms and parameter values can be found in S1 Table.
The simulation approach used in this work constructs ensembles of stochastic model realisations that represent villages within districts. The villages are independent units, and there is no exchange of individuals among villages. Each district is defined by a specific distribution of pre-control infection levels, characterised by a given mean and variance. To construct the districts, we first use the two mathematical models for the transmission of STHs developed by Erasmus MC and ICL to simulate a large pool of villages with stochastic transmission conditions, defined in terms of transmission rate for Erasmus MC model and basic reproduction number (i.e. R0, indicating the transmission intensity in a defined setting) for the ICL model. In both, the level of exposure heterogeneity is maintained fixed (values can be found in S1 Table). For each village, we simulate a pre-control prevalence of infection in SAC at baseline, measured using a single KK slide taken from all SAC in sentinel villages before the start of PC. Then, we assign a normalising weight to each village, based on the inverse of the density of its pre-control prevalence within the larger pool of village simulations, using a Gaussian kernel. The weights are used to repeatedly generate districts of 150 villages with a given desired distribution of pre-control prevalence of infection. Each village consists of approximately 500 simulated individuals. The choice of the population size and number of villages by district was informed by high resolution population count data generated within the WorldPop project [12,13] and by the implementation units shapefile for Sub-Saharan Africa available as part of the interactive mapping tool for control NTDmap [14].
We assume that the distribution of pre-control prevalences in a district follows a beta distribution with mean μ in the range between 0.2 and 0.4 (with 0.01 increments). To have these beta distributions represent a realistic level of geographical variation within a district at pre-control, we use sub-Saharan pixel-level prevalence predictions published in 2014 [15]. We aggregated these predictions over implementation units, and both mean and variance of implementation units were weighted by pixel-level population densities. Then a linear model was fitted to the log-transformed variance σ2 of the distribution of pre-control prevalences and its logit-transformed mean μ:
log(σ2)=τspecies+θspecieslogit(μ)
(1)
where τspecies and θspecies indicate the species-specific intercept and slope, respectively. The values for each species used here are reported in Table 1. Based on the available data, our analysis showed that for hookworm and T. trichiura the prevalence variation within a district increases as the mean prevalence μ increases. In contrast, the prevalence variation for A. lumbricoides does not depend on the mean prevalence and remains constant. Mean and variance were then used to obtain the shape parameters of a corresponding beta distribution employed for simulating implementation units.
PC strategy is decided at the district level: annual or semi-annual PC, targeting preSAC and SAC (ages 2–15) through schools, or treating the whole community (age 2 and above). For each village in a district we simulate the dynamics of STH infection levels using four treatment strategies (school-based annual PC, school-based semi-annual PC, community-wide annual PC and community-wide semi-annual PC). We only model treatment with albendazole, which is assumed to kill parasites with a probability of 99% (A. lumbricoides), 94% (hookworm), or 60% (T. trichiura), in line with measured egg reduction rates [16,17]. We assume a mean treatment coverage of 75% of the target population, with 95% of the villages having coverage between about 65% and 85%. This corresponds to a beta distribution with shape parameters α = 52.5 and β = 17.5 (mean 0.75, 95% CI (0.643–0.843)).
The flowchart in Fig 1 summarises all the different steps of the simulation methods. Further details on the sampling methodology, assumed drug efficacy and other parameters are described in S1 Table. For each generated district, we randomly select a sample of N sentinel villages (N = 2, 5, 10, 25, or 50) and evaluate the baseline prevalence in 100%, 50% or 25% of SAC in these villages before start of PC, using a single-slide KK per person. Five years after the start of PC, simulated single-slide KK measurements are taken again from a new sample of SAC to evaluate prevalence in the same sentinel villages. We extract model predictions for prevalence of infection in SAC in sentinel villages (the indicator) and the prevalence of moderate-to-heavy intensity infection in SAC in all villages as measured by a single-slide KK (the “true” outcome). We then calculate the PPV of the sentinel-village-level indicator for the district-level “true” outcome as the ratio A / B. The denominator B is the number of simulated districts with sentinel-village-level prevalences under a given threshold, and the numerator A is the number of simulated districts among B where the “true” prevalence of moderate-to-heavy infection is below 1% (i.e. achievement of the morbidity target). When the threshold value is chosen to be 1% and the predictor is the prevalence of any infection in the sentinel villages, the PPV corresponds to the proportion of districts interrupting PC, that have in fact reached the morbidity target 5 years after the start of STH control.
The distribution of pre- and post-control prevalences of STH infection in SAC in simulated districts is displayed in Fig 2, with prevalences based on KK testing of all SAC in the districts before the start of PC (2015) and after 5 years of PC (2020). Both Erasmus MC and ICL models adequately produce the predefined distribution of district level prevalences at baseline.
The two models qualitatively agree on the impact of PC on STH prevalence levels on any infection in SAC: community-wide PC vs. school-based PC and semi-annual vs. annual PC more strongly reduce the prevalence of any infection in SAC for all species. Because in the ICL models for T. trichiura and A. lumbricoides, adult humans are assumed to practice open defecation less frequently as they get older (i.e. age-dependent contribution to transmission is proportional to age-dependent exposure to transmission), school-based PC has a higher impact than predicted by the Erasmus MC model for these two species. Conversely, because in the Erasmus MC open defecation practices are assumed to be stable after the age of ten, adults contribute more to T. trichiura and A. lumbricoides transmission than in the ICL model, and hence, there is a larger additional benefit of implementing community-wide PC than predicted by the ICL model. S1 Fig compares baseline and post-control prevalences between the two models for district level means between 20 and 30% (first page) and between 30 and 40% (second page).
The histograms in Fig 3 show the post-control prevalence distribution of any (i.e. low, moderate and/or heavy) STH infection in SAC and the threshold value of 1% (vertical dashed line) used to make the decision of stopping PC. The plot further distinguishes between simulated districts that meet the morbidity target (prevalence of moderate-to-heavy infections <1%, turquoise) and those which do not meet the morbidity target (red). The bars to the left of the dashed lines (prevalence of any infection <1%) are entirely turquoise (as expected) because the morbidity target is always met in those districts. Considerable differences in these distributions are detectable depending on the PC target population, PC frequency, and the STH species (see S2 Fig for comparison stratified by baseline prevalences of 20–30%, first page, and 30–40%, second page).
Across all species, both models suggest that a school-based semi-annual PC is almost always a successful strategy for reaching the STH morbidity target (Fig 3) for the considered baseline prevalence levels. Community-wide PCs (both annual and semi-annual) resulted in the achievement of the morbidity target at district level for both mean district prevalence levels between 20 and 30% (S2 Fig, first page) and mean district prevalence levels between 30 and 40% (S2 Fig, second page).
According to our simulations, for hookworm intensive PC (school-based semi-annual PC and both the community-wide PC strategies) almost always meet the morbidity target in 2020, even if the prevalence of any infection in SAC is considerably higher than 1%. Five years of annual school-based PC is only sometimes sufficient to reach the morbidity target (54.8% vs. 71.3%, Erasmus MC and ICL model, respectively).
Furthermore, depending on STH species and PC strategy, the feasibility of reaching the morbidity target also depends on the pre-control prevalence levels (S2 Fig). Districts with higher endemicity (30–40%) have a lower chance to achieve a prevalence below 1% of moderate-to-heavy infections.
The prevalence of STH infection in sentinel villages after 5 to 6 years of PC as assessed by a single slide KK is used to determine whether PC should be scaled up/down or stopped altogether in an implementation unit. Table 2 shows the probabilities of scaling down or stopping PC prematurely based on two different sampling strategies: 2 sentinel villages per implementation unit where only 25% of SAC is tested for STH infection versus 50 sentinel villages with all SAC tested. Sampling more villages (and more SAC per village) reduces considerably the misclassification probabilities for treatment allocation at the district level.
Fig 4 shows the PPV for reaching the morbidity target for STH infection in a district given potential threshold values for the prevalence of infection (any intensity) in SAC in sentinel villages, comparing sampling of all SAC in 5 sentinel villages (solid lines) with sampling 50% of SAC in 10 sentinel villages (dashed lines). Only school-based annual PC is considered as the morbidity target was always met under more intensive PC strategies. In general, the PPV increases with lower threshold values, unless the morbidity target is never or always met in 2020, regardless of the prevalence of infection in sentinel villages (i.e. horizontal lines at either the bottom or top of the graph, respectively). Further, the PPV curve for districts with average pre-control prevalences in the range 20–30% (light blue) lies above the PPV curve for prevalence values between 30 and 40% (dark blue). This means that for districts with a lower pre-control prevalence (20–30%) higher threshold values can be used as an indicator for having met the morbidity target in 2020. In general, for a given pre-control prevalence level, the PPV curve associated with testing all SAC in 5 sentinel villages and PPV curve describing the sampling of 50% SAC in 10 sentinel villages coincide, suggesting that little additional information is provided by sampling the same total number of children from more villages. However, for threshold values close to 1%, the PPV curve based on sampling 10 villages and 50% SAC lies up to 20 percentage points higher than the curve for testing all SAC in 5 sentinel villages for high baseline prevalences (30–40%).
Using the prevalence of moderate-to-heavy infection measured in the sentinel villages as an indicator of meeting the morbidity target at district level (S3 Fig) shows a lower PPV but allows a smaller and consistent threshold across the three STH species.
Comparing sampling strategies that considers an increasing number of villages (2, 5, 10, 25 per implementation unit) with a fixed proportion of SAC tested per village (Fig 5) confirms the added value in terms of PPV, with a difference of up to 80 percentage points between the two extreme cases (2 villages vs. 25 villages testing all SAC). On the other hand, increasing the total population sampled by testing more SAC per village (S4 Fig) leads to a limited increase of the PPV for meeting the morbidity target.
We have investigated the performance of different M&E strategies in assessing whether the STH morbidity target is reached by 2020, using two stochastic individual-based transmission models developed by researchers at Erasmus MC and Imperial College London. Model predictions are in good agreement. In general, we found that the feasibility of meeting the morbidity target varies by species and baseline prevalence. For the range of prevalences considered here (20–40%) school-based PC with an annual frequency is not always sufficient to achieve <1% prevalence of moderate-to-heavy infections in SAC. Sampling strategies that involve few sentinel villages per implementation unit pose a considerable risk of prematurely scaling down/stopping PC. The positive predictive value (PPV) of prevalence of infection in SAC in sentinel villages for meeting the WHO morbidity target in a district is generally higher for areas with lower pre-control prevalence levels (lower R0 values) and/or past higher-intensity PC strategies. Sampling a fixed number of children spread out over 10 instead of 5 sentinel villages may increase the PPV by up to 20 percentage points. If the fraction of SAC sampled is fixed (e.g. 100%) increasing the number of sampled villages from 2 to 25 may result in a PPV up to 80 percentage points higher. On the other hand, for a random sample of 5 sentinel villages, testing of 100% instead of 50% or 25% of the SAC in those villages provides very little additional information. Furthermore, given that PC against STH infections is currently typically school-based, the WHO-recommended criterion of 1% prevalence of any infection in SAC as a marker of success is generally too optimistic for areas that have received 5 years of PC. The PPV for achieving the morbidity target is often well under 90% or even zero, except for the very low-endemic areas (low R0 values for the ICL model and low transmission rate for Erasmus MC) where the morbidity target is always met, regardless of the prevalence of any infection in SAC.
One of the novelties of this work is that we used the models to construct ensembles of stochastic simulations that represent villages within districts according to a defined distribution of pre-control prevalences of infection. The probability distributions of prevalence values represent local geographical variation in infection levels across villages in the same district. This description is important both because it is what is observed in large spatially structured epidemiological studies of STH infection in endemic regions and because the decision of whether to continue PC after the evaluation at 5 years is made by the programmes at the district level. In our analysis we based our assumptions about geographical variation within districts on an analysis of previous work by Pullan and colleagues [15] where a geostatistical model was used to obtain pixel-level estimates of STH prevalence in sub-Saharan Africa using all collated survey data until 2010. However, variation of prevalence of STH infection in specific localities at finer spatial scale may be different from what we assume here. Therefore, to produce model predictions for more specific situations it is important to gather more data and/or to collate existing data on the geographical variation in pre-control STH prevalences at clearly defined spatial scales. Two major Bill and Melinda Gates Foundation funded studies (the Tumikia study in Kenya [18] and the DeWorm3 study in India, Benin and Malawi [19]) will provide very detailed information on the heterogeneity in STH prevalence in large populations at village-level spatial scale in the near future. Given such information and the history of PC in an area, along with information or assumptions about the variation in local PC coverage, the simulations methods described in the paper can provide insight into the appropriateness of different sampling schemes for specific areas. For instance, for districts with higher variation in village-level prevalence of infection than in our analysis (e.g. because of higher variation in pre-control transmission conditions and/or patchy geographical coverage of PC), the benefit of sampling of additional sentinel villages would be higher than we predict here. A better understanding of the pre- and post-control variation in local STH prevalences will be essential to improve the definition of the WHO morbidity target and to define the appropriate geographical scale to consider in monitoring and evaluation progress towards the defined targets.
A current limitation of our approach is that we consider villages within a district as isolated units, such that no infections can be transferred between villages within a geographical area. Also, in our analysis, we assumed a roughly stable PC coverage (geographically), meaning that infection levels in all villages within the simulated district decline synchronously. This means that our model predictions are relatively optimistic particularly for areas where the geographical coverage of PC within an implementation unit is irregular such that spill-over of infection from untreated to treated villages becomes relevant, especially close to elimination. Such areas require more detailed and sophisticated modelling of human movement patterns to account for spill-over of infection between villages, which theoretically has a stabilising effect on infection levels and impedes achievement of control or elimination. Work is in progress on developing both stochastic frameworks of movement patterns and analytical studies of how such movements influence breakpoints in transmission.
Overall, we found that achieving the morbidity target is highly dependent on the dominant STH species, PC strategy and pre-control prevalence levels (transmission intensities in defined localities). The WHO treatment guidelines provide a combined strategy to control the three STH species. However, differences are evident in their responses to the different PC strategies and therefore also in the required action to meet the WHO morbidity goal. We would therefore advocate for an approach that differentiates among the three main species, such that the implemented policy is based on whatever dominant species present requires in terms of intervention and the associated M&E strategy. Within the current scope of WHO-recommended strategies (i.e. school-based deworming of SAC and preSAC) this means that five years of PC is too short a period to start considering evaluation of the morbidity target. To achieve the morbidity target within five years, community-wide treatment would need to be considered, and/or treatment with a combination of albendazole and ivermectin (not modelled here), which is more effective against T. trichiura in particular [20,21].
As previously pointed out by the STH Advisory Committee [22], survey methods currently endorsed by WHO to assess the prevalence of any STH infection are not designed to determine whether or not the morbidity goal in children has been achieved. Therefore, it may be advisable to evaluate achievement of the morbidity target using the prevalence of any infection in combination with the estimated proportion of moderate to heavy infections in the sentinel villages. The current recommended diagnostic is egg count through means of a KK, and therefore a direct quantification of moderate and heavy infection is possible. There would still be uncertainty due to the limited number of villages and people tested but the assessment of the real morbidity level in the district could be substantially improved and the definition of a threshold value to stop PC would be more straightforward. However, in terms of PPV, the use of prevalence of any infection in SAC as a criterion to stop PC remains the safest (S4 Fig).
The potential added value of more sensitive diagnostics such as multiple slide KK or quantitative (real-time) polymerase chain reaction (qPCR) is object of current investigation and will provide useful information for new WHO guidelines for the 2021–2030 period, also in terms of M&E sampling strategies to assess the morbidity target.
The current WHO treatment guidelines provide a combined strategy to control the three STH species. However, the efficacy of PC strategies clearly differs by species given differences in both drug efficacy and the age distribution of infection, with each having a different optimal strategy to meet the morbidity target. Meeting the criterion of 1% prevalence of any infection in district sentinel villages may still mean that the prevalence of moderate-to-heavy infections is higher than 1% in a considerable part of the district. Large-scale data collection through well designed M&E programmes that include information on mass drug administration coverage and individual longitudinal compliance to treatment, combined with further simulation studies, are required to further our understanding of how best to achieve the WHO STH control goals and how best to monitor and evaluate progress.
|
10.1371/journal.ppat.1001139 | Viral Replication Rate Regulates Clinical Outcome and CD8 T Cell Responses during Highly Pathogenic H5N1 Influenza Virus Infection in Mice | Since the first recorded infection of humans with H5N1 viruses of avian origin in 1997, sporadic human infections continue to occur with a staggering mortality rate of >60%. Although sustained human-to-human transmission has not occurred yet, there is a growing concern that these H5N1 viruses might acquire this trait and raise the specter of a pandemic. Despite progress in deciphering viral determinants of pathogenicity, we still lack crucial information on virus/immune system interactions pertaining to severe disease and high mortality associated with human H5N1 influenza virus infections. Using two human isolates of H5N1 viruses that differ in their pathogenicity in mice, we have defined mechanistic links among the rate of viral replication, mortality, CD8 T cell responses, and immunopathology. The extreme pathogenicity of H5N1 viruses was directly linked to the ability of the virus to replicate rapidly, and swiftly attain high steady-state titers in the lungs within 48 hours after infection. The remarkably high replication rate of the highly pathogenic H5N1 virus did not prevent the induction of IFN-β or activation of CD8 T cells, but the CD8 T cell response was ineffective in controlling viral replication in the lungs and CD8 T cell deficiency did not affect viral titers or mortality. Additionally, BIM deficiency ameliorated lung pathology and inhibited T cell apoptosis without affecting survival of mice. Therefore, rapidly replicating, highly lethal H5N1 viruses could simply outpace and overwhelm the adaptive immune responses, and kill the host by direct cytopathic effects. However, therapeutic suppression of early viral replication and the associated enhancement of CD8 T cell responses improved the survival of mice following a lethal H5N1 infection. These findings suggest that suppression of early H5N1 virus replication is key to the programming of an effective host response, which has implications in treatment of this infection in humans.
| Outbreaks of avian influenza (AI) viruses have continued in chickens in Southeast Asia, coupled with regular instances of direct bird to human transmission, with extremely high case fatality rates. The mechanisms underlying the disease pathogenesis and high mortality rate in humans are not well understood. In particular, we lack information on the development and/or failure of adaptive immune responses during AI infection. Our studies in mice have linked the pathogenicity of AI viruses to the virus' rate of replication in the lungs. Surprisingly, a strong T cell response was triggered by the infection, but virus-specific T cells were ineffective in controlling the rapidly replicating virus. The extremely high rate of AI virus replication likely outpaces and overwhelms the developing immune response. However, administration of anti-viral drugs, only early in the infection slowed viral replication, enhanced the number of effector CD8 T cells in the lung, and promoted survival and recovery from infection. These findings highlight the role of viral replication rate in pathogenesis and underscore the importance of controlling viral replication as an adjunct to immunotherapies in the treatment of this infection in humans.
| Severe outbreaks of highly pathogenic avian influenza (AI) H5N1 viruses in poultry continue to occur and are often coupled with reports of direct bird-to-human viral transmission. Between 2003 and 2009, 406 confirmed human cases of AI H5N1 were reported, with a fatality rate of >60% (http://www.who.int/csr/disease/avian_influenza/country/cases_table_2010_01_28/en/index.html). Although sustained human-to-human transmission has not yet occurred, there is increasing concern that these H5N1 AI viruses might acquire the ability to transmit efficiently between humans and cause a pandemic. The high virulence of H5N1 viruses in humans can be attributed to either a delay in development or the ineffectiveness of innate and/or adaptive immune mechanisms to control the infection in a timely fashion. However, little information exists on the dynamics of adaptive immune responses to H5N1 viruses during a primary infection, which constitutes a staggering gap in our understanding of the pathogenesis of lethal H5N1 infection in humans.
The adaptive immune response to seasonal influenza viruses has been extensively characterized using a murine model of intranasal (I/N) infection with mouse-adapted influenza viruses [1], [2], [3], [4], [5], [6]. Elicitation of a potent CD8 T cell response is of critical importance in resolving a primary influenza virus infection in mice [1], [3], [4], [7]. However, both CD8 T cells and antibodies might be required to clear highly pathogenic influenza viruses [8]. Mouse-adapted influenza viruses elicit robust CD8 T cell responses in the respiratory tract, which typically peak at day 10 after infection [5], [6]. Effector CD8 T cells control influenza virus replication by cytolytic mechanisms that require Fas and/or perforin [2]. In addition to their role in viral clearance, CD8 T cells are also implicated in mediating immune-mediated lung injury following influenza virus infection [9], [10], [11]. Pertaining to primary infection with H5N1 viruses, we do not yet know whether CD8 T cell responses are induced in the respiratory tract, or whether virus-specific CD8 T cells play a protective or immunopathologic role during a primary H5N1 infection. A high viral load is one of the hallmarks of a fatal H5N1 infection in humans [12], but the effect of high-level H5N1 virus replication on the emergence of CD8 T cell responses in the respiratory tract has not been studied.
In this study, using two human isolates of H5N1 viruses that differ in their pathogenicity in mice, we have systematically examined the following: 1) the relationship between the speed of H5N1 virus replication and viral pathogenicity on the dynamics of CD8 T cell responses, 2) whether ineffective control of H5N1 virus infection is related to the suppression of virus-specific CD8 T cell responses, 3) the effect of CD8 T cell deficiency on host survival, and 4) the effect of anti-viral therapy on CD8 T cell responses. Findings from these studies have provided novel insights into the virus/immune system interactions during an H5N1 infection from the standpoint of viral pathogenesis, immune control of viral replication, and immunopathology.
Unlike seasonal strains of influenza viruses, human isolates of H5N1 viruses readily replicate in other mammals, including mice, without prior adaptation and induce varying levels of pathogenicity [13]. Experimental infections of mice with the H5N1 viruses have led to the identification of viral determinants of pathogenicity [13], [14], [15], [16], [17]. Although high cleavability of hemagglutinin is essential to cause a lethal infection, a single amino acid residue in the PB2 protein controls the pathogenic potential of these AI viruses in mice [13]. An extremely low-dose infection of mice with the A/Hong Kong/483/97 (HK483) virus that has a Lys at position 627 of the PB2 protein induces a lethal infection (a dose of virus that kills 50% of infected mice (MLD50) of 1.8 plaque-forming units [PFU]), whereas the A/Hong Kong/486/97 (HK486) virus that has a Glu at position 627 of PB2 is less pathogenic (MLD50 of 7.6×103 PFU). To determine whether the two viruses differ in their rate of viral replication in vivo in the respiratory tract, we performed a detailed kinetic analysis of viral titers in the lungs of HK483- and HK486-infected mice (Figure 1). Although mice were infected with the same dose of both viruses and reached comparable maximum titers, viral growth kinetics in the lungs were dramatically different. The HK483 virus replicated at a remarkable pace within the first 24 hours, and the coefficient of expansion was calculated to be ∼4.3 log PFU/day; peak virus titers of ∼107 PFU/gram were attained in the lungs within 48 hours after infection. In striking contrast, the coefficient of expansion for the less pathogenic HK486 virus in the first 24 hours was only ∼1.3 log PFU/day, and peak titers in the lungs were not attained until 5 days after infection. Thus, the speed of early viral replication in the lungs might be a necessary and distinguishing trait of highly pathogenic AI viruses to rapidly reach high titers and potentially overwhelm the host immune responses.
The enhanced replication of the HK483 virus could be related to the virus's ability to evade the innate immune mechanism(s), especially the type I IFN pathway [18]. However, microarray analysis showed that the induction of IFN-β and interferon-stimulated genes is greater in the lungs of HK483-infected mice compared to HK486-infected mice at 48 hours after infection (Figure S1). To examine whether Type I IFNs play any role in controlling infection with HK483, we infected groups of wild type C57BL/6 (n = 5) and Type I IFN receptor-deficient (IFNRI−/−) mice (n = 5) with 18 PFU of HK483 virus. Upon infection with the HK483 virus, all wild-type mice survived at least until day 7 after infection, but 4 of 5 IFNRI−/− mice died by day 5 postinfection (PI), and the remaining IFNRI−/− mouse died on day 7 PI. These data suggested that the type I IFN pathway is induced and functional in HK483-infected mice, which is consistent with a recently published report [19].
To understand the relationship of viral pathogenicity and/or rapid viral replication rate to the development of adaptive immunity, we infected BALB/c mice I/N with the HK483 or HK486 virus and studied the evolution of virus-specific CD8 T cell responses in the respiratory tract. Sequence comparisons showed that the Kd-restricted CD8 T cell epitope NP147 of the PR8 virus was conserved in both HK483 and HK486 viruses. In mice infected with the less pathogenic HK486 virus, high numbers of CD8 T cells in the lung airways were not detectable until day 8 PI but rapidly accumulated within the next 24 hours (Figure 2A). The kinetics of the CD8 T cell response to HK486 were similar to those of mouse-adapted human influenza viruses [5], [6]. Surprisingly, in HK483-infected mice, virus-specific CD8 T cells were detectable earlier, at day 7 PI, and peak numbers were attained at day 8 PI. It is noteworthy that the peak numbers of NP147-specific CD8 T cells in HK483-infected mice attained at day 8 PI were lower, as compared to those in HK486-infected mice (day 9 PI). Additionally, CD8 T cells in HK483-infected mice appear to have initiated contraction between days 8 and 9 PI, when the number of CD8 T cells continued to increase in the respiratory airways of HK486-infected mice (Figure 2A). Similar contraction in the number of CD8 T cells was seen in the lungs of HK483-infected mice between days 8 and 9 PI in a separate experiment (data not shown).
In order to track the early events of CD8 T cell activation in the draining lymph nodes (DLNs), we adoptively transferred carboxyfluorescein succinimidyl ester (CFSE)-labeled influenza HA518-specific Clone 4 (CL-4) TCR transgenic CD8 T cells into congenic BALB/c mice [5], [6], which were subsequently infected I/N with 18 PFU of HK483 or HK486 virus. By day 5 PI, >90% of CL-4 CD8 T cells had divided several times in the DLNs of HK483-infected mice (Figure 2B), and a substantial fraction of these cells also exhibited markers of activation (LFA-1HI, CD43Hi, and CD62LLo) (Figure 2C). In contrast, in the DLNs of HK486-infected mice, only <50% of CL-4 CD8 T cells had proliferated by day 5 PI, and these cells did not upregulate expression of LFA-1 or CD43 until day 6 PI. The increased percentage of proliferated CL-4 CD8 T cells in the lymph nodes of HK483-infected mice was not linked to reduced trafficking of these cells out of the lymph nodes into the lungs because, the number of CL-4 CD8 T cells in the BAL of HK483-infected mice (2.9–4.1×103) were higher than in the BAL of HK486-infected mice (1.0–1.4×103) at day 5 PI. In addition to increased proliferation, a larger percentage of CL-4 CD8 T cells expressed granzyme B in HK483-infected mice at day 6 PI compared to those in HK486-infected mice (Figure 2C). Thus, CD8 T cells underwent accelerated activation in the DLNs of mice infected with the HK483 virus compared to those in HK486-infected mice. The early activation of virus-specific CD8 T cells in lymph nodes of HK483-infected mice corresponds with the faster replication kinetics of the virus in the respiratory tract.
The pathogenicity of influenza viruses in the experimental mouse model has been defined based on MLD50. Next, we determined whether the accelerated kinetics of CD8 T cell contraction is related to the clinical outcome of infection, i.e., lethality. Typically, varying the infecting dose alters the disease process and clinical outcome of influenza viruses, but this procedure is not feasible with the HK483 virus because of the extremely low MLD50 of 1.8 PFU. Therefore, we examined the effect of viral dose (based on MLD50) on the kinetics of CD8 T cell contraction by infecting BALB/c mice with 1 (4.6×103 PFU) or 10 MLD50 (4.6×104 PFU) of the HK486 virus. As controls, mice were infected with 10 MLD50 of the HK483 virus (18 PFU). As shown in Figure 3, premature contraction of total and NP147-specific CD8 T cells occurred in mice infected with 10 MLD50 of the HK483 or HK486 virus, but not in mice infected with 1 MLD50 of the HK486 virus. Thus, regardless of the H5N1 virus strain used, the pathogenicity of H5N1 viruses in mice (which is a function of infecting dose for the HK486 virus) regulated the dynamics of CD8 T cell contraction in the respiratory tract following infection with H5N1 viruses.
Next, we determined whether infection with the highly pathogenic H5N1 virus caused premature contraction of CD8 T cells by affecting cellular apoptosis. Following infection of BALB/c mice with 18 PFU of HK483 or HK486 viruses, apoptosis of CD8 T cells in the lung was assessed at days 7, 8, and 9 after infection by staining for active caspase 3. At all time points, the fraction of apoptotic CD8 T cells in the lungs of HK483-infected mice was two- to four-fold higher than in HK486-infected mice (Figures 4A and 4B). These findings suggested that infection with highly pathogenic AI viruses induces accelerated apoptosis and premature contraction of CD8 T cells in the respiratory tract.
Two distinct pathways of caspase-dependent cellular apoptosis have been described: the intrinsic and extrinsic pathways [20], [21]. The intrinsic apoptotic pathway is initiated following activation of the pro-apoptotic BH3-only proteins, such as Bcl-2-interacting mediator of death (BIM). On the other hand, interaction between death receptors and their ligands, such as Fas and Fas ligand, triggers the extrinsic pathway of cellular apoptosis. To determine which pathway of cellular apoptosis is triggered in CD8 T cells by the highly pathogenic AI virus, we infected wild-type C57BL/6 (+/+), Fas-mutant lpr/lpr (Fas KO), and BIM-deficient (BIM KO) mice with the HK483 virus. At day 8 PI, we quantified the number of apoptotic active caspase 3+ve CD8 T cells in the BAL of HK483-infected mice (Figure 4C). As expected, a substantial fraction of CD8 T cells was apoptotic in the lungs of C57BL/6 mice, and Fas deficiency did not significantly affect apoptosis of CD8 T cells induced by highly pathogenic HK483 infection. Notably, the percentage of apoptotic CD8 T cells was reduced by ∼90% in BIM KO mice compared to C57BL/6 or Fas KO mice. These data suggested that the apoptosis of CD8 T cells induced by highly pathogenic AI viruses is triggered by the intrinsic pathway of cellular apoptosis. As a consequence of reduced apoptosis in the absence of BIM activity, the numbers of PA224-specific CD8 T cells in the BAL of BIM KO mice (5.4×104) were higher than in +/+ mice (3.6×104).
Because BIM deficiency protected against CD8 T cell apoptosis, we next examined whether the loss of BIM would also improve survival of HK483-infected mice. Groups of +/+, Fas KO, and BIM KO mice were infected with the HK483 virus as above, and their survival was monitored daily. As shown in Figure 5A, a majority of +/+ mice infected with HK483 succumbed to infection by day 12 PI. Likewise, BIM KO and Fas KO mice also succumbed to HK483 infection, albeit with a slight delay. Neither BIM nor Fas deficiency significantly affected viral titers in the lungs (Figure S2). These data suggested that BIM deficiency-induced enhancement of virus-specific CD8 T cell responses is insufficient to enhance survival following a highly pathogenic H5N1 virus infection.
Next, we assessed whether BIM deficiency affected HK483 virus-induced cell damage in the lungs (Figure 5B). At day 8 after infection with the HK483 virus, the lung pathology in +/+ mice was characterized by extensive cellular necrosis and tissue disruption of medium-sized blood vessels and bronchioles, which was associated with infiltration of inflammatory cells composed mostly of neutrophils. Additionally, apoptotic cells were frequently observed in the lungs of +/+ mice. In striking contrast, in the lungs of BIM−/− mice, cellular necrosis was less pronounced and the tissue integrity was more intact, with less frequent apoptotic cells. Notably, the lungs of BIM−/− mice contained lymphocytic infiltrates in the connective tissues near medium-sized blood vessels and, more prominently, adjacent to the bronchioles (Figure 5B). Based on these findings, we infer that HK483-induced lung pathology is at least in part mediated by BIM-dependent mechanisms.
The neuraminidase inhibitor, oseltamivir phosphate, is an effective treatment for influenza A virus infection in humans if given early in the infection [22], [23], [24]. Treatment with oseltamivir reduces viral load and protects mice against a lethal H5N1 virus infection [25], [26]. It was of interest to determine whether a high rate of virus replication in HK483-infected mice, especially early in the infection could 1) lead to early activation and contraction of virus-specific CD8 T cells in the lung airways and 2) outpace and overwhelm the CD8 T cell response. Additionally, the effects of oseltamivir treatment on the adaptive immune response to H5N1 infection have not been examined. Therefore, we asked whether the reduction of virus replication by oseltamivir protected against the accelerated activation and contraction of CD8 T cells following infection of mice with the highly pathogenic HK483 virus. Mice that were infected with the HK483 virus were treated with graded doses of oseltamivir only early in the infection, and virus-specific CD8 T cells were quantified at days 7, 8, and 9 after infection. As expected, CD8 T cells in control vehicle-treated mice underwent contraction between days 8 and 9 PI (Figure 6A), but oseltamivir treatment at doses of 10 or 20 mg, but not 5 mg, mitigated contraction and led to a substantive increase in the number of virus-specific CD8 T cells in the lung airways of HK483-infected mice between days 8 and 9 PI. Notably, Figure 6B shows that oseltamivir treatment reduced viral titers, especially early in the course of the infection, regardless of the dose administered, but mouse survival was extended or increased only at doses of 10 and 20 mg, which suggested that suppression of early viral replication alone might be necessary but not be sufficient to enhance mouse survival. However, reduced viral titers coupled with enhanced CD8 T cell responses were associated with extended or improved survival.
Our studies showed that infection with the highly pathogenic HK483 virus elicited a readily detectable CD8 T cell response but failed to effectively control viral replication. Because there is evidence supporting a role for CD8 T cells in augmenting lung pathology following infection with seasonal influenza viruses [9], [10], [11], we questioned whether CD8 T cells contribute to the lethality induced by infection with highly pathogenic H5N1 viruses. Groups of +/+ and CD8-deficient (CD8 KO) mice were infected with graded doses of the HK483 virus, and mouse survival was monitored (Figure 7). As shown in Figures 7A and 7B, there was no difference in survival between HK483-infected +/+ and CD8 KO mice. These data suggested that the loss of a CD8 T cell response does not provide either a survival advantage or a disadvantage to mice infected with the highly pathogenic HK483 virus.
Until the AI epidemic of 1997, it was assumed that purely AI viruses could not cause a lethal disease in humans. However, since 1997, recurring instances of direct transmission of AI viruses from birds to humans have dismissed this assumption [14], [27], [28]. A unique feature of these H5N1 viruses is their ability to replicate to high levels in the lungs of several mammalian species, including humans, without adaptation [15], [29], [30]. Although a high viral load and hypercytokinemia are recognized hallmarks of fatal AI infections in humans [12], we still lack crucial information on the kinetics, magnitude, and nature of the adaptive immune response to these infections. In this study, we have examined the relationship of viral replication kinetics in the lungs and viral pathogenicity to the dynamics of virus-specific CD8 T cell responses to AI viruses in mice. We found that the extreme pathogenicity of H5N1 viruses is directly linked to the high viral replication rate and the consequent production of peak steady-state viral titers in the lungs within 48 hours after infection. Interestingly, we found that lethal H5N1 infection in mice stimulates a robust, virus-specific CD8 T cell response in the respiratory tract, but these CD8 T cells fail to control viral replication and undergo early contraction. The prevention of CD8 T cell contraction did not alter the survival of infected mice, but inhibition of neuraminidase activity and viral replication by therapeutic intervention mitigated the premature contraction of CD8 T cells and enhanced mouse survival following a lethal H5N1 infection. These findings suggest that the ability of H5N1 viruses to overwhelm and/or undercut the sustenance of the anti-viral CD8 T cell response and cause a lethal pulmonary infection is linked to a high viral replication rate, especially early in the infection. These findings further our understanding of the pathogenesis of H5N1 viruses, which should have implications on the development of novel therapies and prophylaxis for H5N1 infection in humans.
The infection of mice with mouse-adapted strains of influenza viruses elicits strong CD8 T cell responses in the respiratory tract, and there is ample evidence indicating an important role for CD8 T cells in the viral control of a primary influenza virus infection [3], [4], [7]. In contrast to a sublethal infection, inoculation of mice with high doses of mouse-adapted influenza virus leads to apoptosis of virus-specific CD8 T cells and lethal pulmonary injury [31]. Moreover, based on an analysis of gene expression in the lungs of mice infected with highly pathogenic H5N1 viruses, T cell activation might be impaired during an H5N1 virus infection [32]. However, we showed that the infection of mice with a highly pathogenic H5N1 virus elicits a readily detectable CD8 T cell response, which suggests that the initial events of T cell priming, including trafficking of dendritic cells to the DLN and antigen processing/presentation, are intact in H5N1 virus-infected mice. While virus-specific CD8 T cells continued to accumulate in the lung airways of mice until at least day 9 after infection with the less pathogenic HK486 virus, CD8 T cells in HK483-infected mice exhibited a decline after day 8 PI due to BIM-dependent apoptosis. The BIM-dependent intrinsic pathway of apoptosis of activated CD8 T cells appears to be unique to highly pathogenic H5N1 AI viruses because a high-dose infection of mice with mouse-adapted epidemic strains of the influenza virus induced CD8 T cell apoptosis that was dependent upon Fas/FasL interactions [31]. Highly pathogenic AI viruses are known to trigger hyperinduction of TNF-related apoptosis inducing ligand (TRAIL) in macrophages and cause T cell apoptosis in vitro [33]. Because TRAIL-induced apoptosis is BIM-dependent [34], [35], [36], we propose that apoptosis of activated CD8 T cells in HK483-infected mice might be triggered by interactions between macrophage-derived TRAIL and its receptors on CD8 T cells. It should be noted that our experiments did not test whether BIM triggered apoptosis of CD8 T cells by T cell intrinsic mechanisms. It is possible that reduced CD8 T cell apoptosis in BIM-deficient mice was an indirect effect, possibly linked to increased survival of dendritic cells [37]. Do differential direct infection of CD8 T cells by HK483 and HK486 viruses explain differences in CD8 T cell apoptosis? Studies of apoptosis in the lungs of mice infected with HK483 show that apoptotic cells are primarily localized to bronchial epithelial and subepithelial layers, and not to the cells with lymphocyte morphology [38]. Additionally, apoptotic HK483-infected cells are primarily found in the germinal centers of the spleen [38], where CD8 T cells are not typically present in significant numbers. Nevertheless, studies are warranted to assess whether HK483 but not HK486 induces apoptosis of T cells by direct infection. Tumpey et al have reported that the total number of CD8 T cells in the lungs and mediastinal lymph nodes of mice infected with 100 mouse infectious dose 50 (MID50) of HK483 was lower than those in HK486-infected mice at day 6 PI [38]. However, in our experiments, contraction in the number of CD8 T cells in the respiratory airways (Figure 2) or lungs (data not shown) did not occur until after 8 days after HK483 infection (dose of 18 PFU/mouse or 10 MLD50); the number of CD8 T cells in the BAL of HK483-infected mice was lower at day 9 PI, when compared to those in HK486-infected mice. The discrepancy in the kinetics of the CD8 T cell response between the two studies might be related to differences in experimental procedures including preparation of the virus stock, dose of virus used (100 MID50 versus 10 MLD50), infection procedures, and methods used for isolating mononuclear cells from the tissues.
Despite substantial expansion, virus-specific CD8 T cells were ineffective in controlling HK483 infection, and all mice succumbed within 10 days after infection. The inability of CD8 T cells to effectively control HK483 infection is not associated with functional impairment because virus-specific CD8 T cells in the lung airways contained high levels of granzyme (Figure 2) and also produced cytokines, such as IFN-γ, upon antigenic stimulation (Figure S3). Additionally, the impaired control of highly pathogenic H5N1 infection is not linked to premature apoptosis of CD8 T cells because protection of CD8 T cells against BIM-dependent apoptosis did not lead to effective viral control or enhanced mouse survival (Figure 5). Why is the CD8 T cell response unable to effectively control a lethal H5N1 infection? Recent work suggests that the effectiveness of a CD8 T cell response to successfully control viral replication depends upon the number and concentration of effector CD8 T cells in relationship to the number of virus-infected cells [39], [40]. Therefore, the inability of effector CD8 T cells to control the rapidly replicating HK483 virus might be explained by the large number of virus-infected cells, which leads to higher ratios of effector CD8 T cells to the number of virus-infected cells. The effector CD8 T cell response is perhaps neither fast nor large enough (even in BIM KO mice) to control viruses such as HK483 that are capable of rapid replication and dissemination. Immunotherapies to inflate the number of virus-specific CD8 T cells might be able to control infections with highly pathogenic H5N1 viruses.
The ratio of effector CD8 T cells to virus-infected cells in the tissues could be altered by increasing the number of effector CD8 T cells and/or by decreasing the number of virus-infected cells. Our studies show that oseltamivir treatment can achieve this objective. Oseltamivir therapy at certain doses not only suppressed H5N1 viral titers in the lungs but also enhanced the number of effector CD8 T cells in the lung airways, which in turn led to improved survival. The mechanism(s) underlying the enhancement in CD8 T cell responses by oseltamivir is purely conjecture at this point. One possibility is that oseltamivir reduces viral load, which in turn leads to inhibition of TRAIL induction and BIM-dependent apoptosis of effector CD8 T cells. A second theory is that the diminished viral load in oseltamivir-treated mice would be expected to reduce the amount of HA and HA-triggered cellular apoptosis [41]. A third theory is that CD8 T cell contraction is triggered by extrapulmonary dissemination of the HK483 virus, which elicits a systemic response, and oseltamivir treatment limits viral replication to the lungs. A fourth possibility is that reduced viral load induced by oseltamivir lowered/delayed antigenic stimulation of T cells by DCs, especially early in the infection, which in turn prevented accelerated activation and contraction of CD8 T cells in HK483-infected mice.
It should also be noted that oseltamivir treatment only affected virus titers early in the infection, and viral load in the lungs at the time of T cell contraction (8–9 days PI) was similar in the untreated group as well as in treated groups, regardless of the dose of oseltamivir. These data suggested that viral titers early in the infection might control the contraction kinetics of the anti-viral CD8 T cell response. It has been reported that the early inflammatory response triggered by an infecting organism programs the contraction of CD8 T cell responses [42]. Therefore, the hyperinflammatory response induced by high viral titers early in the H5N1 infection [43] could also be involved in accelerating the kinetics of CD8 T cell activation and contraction in the lungs. Consequently, lower HK483 viral titers induced by oseltamivir treatment would be expected to blunt the inflammatory response thereby delaying the onset of CD8 T cell contraction.
Interestingly, treatment of mice with 10 mg or 20 mg of oseltamivir reduced viral load in the lungs and modulated CD8 T cell responses to a largely similar extent. However, only treatment with 20 mg of oselatmivir led to substantial improvement in survival of HK483-infected mice. In addition to the well-characterized anti-viral effects, the increased survival of mice that received 20 mg of oseltamivir might be explained by at least two non-mutually exclusive mechanisms. First, only treatment with 20 mg or more of oseltamivir can restrict viral replication to the lungs and prevent viral dissemination into tissues like the brain, thereby averting a fatal infection. Second, oseltimivir at this dose might effectively attenuate the host inflammatory response and limit tissue damage by inhibiting pro-inflammatory responses of macrophages [44].
Although cytolytic influenza virus replication alone can cause significant cell death, CD8 T cells are implicated in accentuating tissue damage by immunopathologic mechanisms. We first showed that CD8 T cell deficiency had minimal effects on the survival of mice infected with the highly pathogenic HK483 virus. It is conceivable that in infections with highly lethal viruses, such as HK483, the extremely high rate of viral replication potentially outpaces the innate and adaptive immune responses, and overwhelming tissue damage caused by cytolysis of infected cells is sufficient to cause a lethal infection. A 100% mortality in +/+ mice and the delayed death in CD8 KO mice imply that viral replication is not controlled in +/+ mice, despite the development of a CD8 T cell response.
In summary, in this study, we have defined mechanistic links among the rate of viral replication, viral pathogenicity, the CD8 T cell response, and the clinical outcome of a lethal H5N1 infection in mice. These studies show that the extreme pathogenicity of H5N1 viruses is directly linked to the ability of virus to replicate rapidly and attain high steady-state viral titers in the lungs early in the infection and not due to the lack of a CD8 T cell response. Perhaps, the rapidly replicating virus simply overwhelms and outpaces the most potent CD8 T cell response. Therefore, restraining H5N1 virus replication to levels under a certain threshold early in the infection not only limits direct virus-induced cytopathicity but also allows the development of a CD8 T cell response that can now effectively clear the non-overwhelming infection. These findings have furthered our understanding of the pathogenesis of H5N1 infections and are expected to have significant implications on the development of effective therapies to treat H5N1 infection in humans.
6-week-old BALB/c, C57BL/6, BIM KO [45], Fas KO (lpr/lpr) [46], CD8 KO [47], and Clone-4 mice [48] were purchased from Jackson Laboratory (Bar Harbor, ME). The Type I IFNR−/− mice were provided by Dr. Murali-Krishna (University of Washington, Seattle, WA) [49]. All mice were used at 6–8 weeks of age according to the protocol approved by the University of Wisconsin School of Veterinary Medicine Institutional Animal Care and Use Committee (IACUC). The animal committee mandates that institutions and individuals using animals for research, teaching, and/or testing must acknowledge and accept both legal and ethical responsibility for the animals under their care, as specified in the Animal Welfare Act (AWA) and associated Animal Welfare Regulations (AWRs) and Public Health Service (PHS) Policy. Animal experimentation was done as per the PHS Policy on Humane Care and Use of Laboratory Animals as described in the Guide for the Care and Use of Laboratory Animals.
HK483 and HK486 viruses that were isolated from patients during the Hong Kong outbreak of 1997 were derived by reverse genetics and titered as described before [13]. Mice were infected I/N with different doses of HK483 or HK486 virus in a volume of 50 µl. Viral titers in tissues were quantified by a plaque assay using MDCK cells. All experiments with these H5N1 viruses were performed in a biosafety level 3 containment laboratory approved for such use by the CDC and United States Department of Agriculture.
Thy1.1/CL-4 CD8 T cells were labeled with CFSE and adoptively transferred into congenic Thy1.2/BALB/c mice by tail vein injection as described before [5]. Twenty-four hours after cell transfer, mice were infected I/N with the HK483 or the HK486 virus.
Kd/NP147 pentamers were purchased from Proimmune Inc. (FL USA). The Db/PA224 tetramers were kindly provided by the NIH Tetramer Facility (Emory University, Atlanta, GA). All antibodies were purchased from BD-Pharmingen unless stated otherwise. Mononuclear cells isolated from BAL or lymph nodes were stained with anti-CD8, anti-LFA-1, anti-CD62L, anti-CD43, and MHC tetramers/pentamers for 1 hr at 4C. For intracellular staining, cells were stained for cell surface molecules as above, and subsequently permeabilized and stained with anti-granzyme (Invitrogen) or anti-caspase 3 antibodies using the Cytofix/Cytoperm kit (BD-Pharmingen). Following staining, cells were fixed with 2% paraformaldehyde and analyzed using a FACSCalibur flow cytometer (Becton Dickinson). Flow cytometry data were analyzed using Flowjo software.
Mice were euthanized, and tissues were collected and fixed in 10% phosphate-buffered formalin. They were then dehydrated, embedded in paraffin, and cut into 5-µm-thick sections that were stained with standard hematoxylin-and-eosin.
Oseltamivir phosphate (Tamiflu, Roche Laboratories Inc., Basel, Switzerland) dissolved in 50% Ora-Plus Suspending agent (Paddock Laboratories, Inc., Minneapolis, MN, USA) in water and administered to mice once daily by oral gavage in a volume of 200 µL at −1 to 3 days relative to infection with HK483 virus.
|
10.1371/journal.pgen.1003839 | The Bacterial Response Regulator ArcA Uses a Diverse Binding Site Architecture to Regulate Carbon Oxidation Globally | Despite the importance of maintaining redox homeostasis for cellular viability, how cells control redox balance globally is poorly understood. Here we provide new mechanistic insight into how the balance between reduced and oxidized electron carriers is regulated at the level of gene expression by mapping the regulon of the response regulator ArcA from Escherichia coli, which responds to the quinone/quinol redox couple via its membrane-bound sensor kinase, ArcB. Our genome-wide analysis reveals that ArcA reprograms metabolism under anaerobic conditions such that carbon oxidation pathways that recycle redox carriers via respiration are transcriptionally repressed by ArcA. We propose that this strategy favors use of catabolic pathways that recycle redox carriers via fermentation akin to lactate production in mammalian cells. Unexpectedly, bioinformatic analysis of the sequences bound by ArcA in ChIP-seq revealed that most ArcA binding sites contain additional direct repeat elements beyond the two required for binding an ArcA dimer. DNase I footprinting assays suggest that non-canonical arrangements of cis-regulatory modules dictate both the length and concentration-sensitive occupancy of DNA sites. We propose that this plasticity in ArcA binding site architecture provides both an efficient means of encoding binding sites for ArcA, σ70-RNAP and perhaps other transcription factors within the same narrow sequence space and an effective mechanism for global control of carbon metabolism to maintain redox homeostasis.
| The cofactor NAD+ plays a central role in energy conservation pathways, shuttling electrons from the oxidation of growth substrates to respiratory or fermentative pathways. To sustain catabolism and cellular ATP demand, an appropriate balance between the reduced and oxidized forms of NAD+ must be maintained. Our genome-scale analysis of the transcription factor ArcA provides insight into how this process is transcriptionally regulated in E. coli in the absence of O2. We found that ArcA mediates a previously unrealized comprehensive transcriptional repression of genes encoding proteins associated with oxidation of non-fermentable carbon sources. Through the repression of these pathways, oxidized NAD+ is effectively preserved for fermentation pathways, facilitating energy conservation and preserving a balance between the oxidized and reduced forms of NAD+ in the absence of aerobic respiration. In addition, we found that the majority of ArcA binding sites contain additional sequence elements beyond that required for binding of an ArcA dimer, providing novel insight into how ArcA and other members of the largest class of two component system-response regulators (OmpR/PhoB family) may achieve global regulation of gene expression.
| Maintaining redox balance is a crucial function for cell survival. Alteration of the cellular redox environment has been shown to affect a broad range of biological processes including energy metabolism [1]–[3], protein folding [4], signaling and stress responses [5]–[9]. Despite this, we have only a superficial understanding of how cells control redox homeostasis at a global level. Since the cellular redox environment is a reflection of many different redox couples [10], some of which are linked together through enzymatic reactions, an improved understanding of this process requires knowledge of how the redox state of each couple is controlled. One such important redox couple is NADH/NAD+, which plays a central role in catabolic pathways, shuttling electrons between donor and acceptor molecules and allowing cells to convert energy from various reduced substrates into cellular ATP. To ensure that catabolism proceeds, a balance between the rates of oxidation and reduction of NAD+ must be maintained. Many diverse regulatory mechanisms have evolved amongst different organisms to control the redox state of the NADH/NAD+ couple [6], [11]–[14]. In this study we investigated transcriptional inputs into this process by mapping the regulon of the transcription factor ArcA in Escherichia coli.
The ArcAB two component system, comprised of the membrane bound sensor kinase, ArcB, and the response regulator, ArcA, coordinates changes in gene expression in response to changes in the respiratory and fermentative state of the cell [15], [16]. This system is maximally activated in E. coli under anaerobic fermentative conditions when NADH from central metabolism is recycled to NAD+ by formation of the end products succinate, ethanol and lactate. The DNA binding activity of ArcA is regulated through reversible phosphorylation by ArcB [17], whose kinase activity is governed by the redox states of the ubiquinone and menaquinone pools [18]–[20] that are linked to the NADH/NAD+ redox couple through respiration. In the absence of O2, decreased flux through the aerobic respiratory chain lowers the ratio of oxidized to reduced quinones, stimulating ArcB kinase activity and transphosphorylation of ArcA [19]. Additionally, fermentation products have been shown to enhance the rate of ArcB autophosphorylation [21] and there is a positive correlation between the rate of fermentation and the levels of phosphorylated ArcA (ArcA-P) [16]. Thus, enzymatic linkage of the NADH/NAD+ couple to the oxidation state of the quinone pool and the production of fermentation products provides a link between the redox state of the NADH/NAD+ couple and the activity of the ArcAB system. Indeed, artificial perturbation of the NADH/NAD+ ratio has been shown to alter ArcA activity [22].
Consistent with the role of the ArcAB system in redox regulation, the majority of known ArcA targets in E. coli are associated with aerobic respiratory metabolism. Under anaerobic conditions, ArcA-P directly represses the operons encoding enzymes of the TCA cycle (gltA, icdA, sdhCDAB-sucABCD, mdh, lpdA) [23]–[27], and for the β-oxidation of fatty acids (fadH, fadBA, fadL, fadE, fadD, fadIJ) [25], lactaldehyde (aldA)/lactate oxidation (lldPRD) [24], [28], and glycolate/glyoxylate oxidation (glcC, glcDEFGBA) [29]. In contrast, ArcA-P activates the expression of operons encoding three enzymes that are important for adapting to microaerobic or anaerobic environments [cytochrome bd oxidase (cydAB) [24], pyruvate formate lyase (focA-pflB) [30] and hydrogenase 1 (hya) [31]]. However, gene expression profiling analyses indicate that the ArcA regulon is more complex than originally expected, including genes encoding a wide variety of functions outside of redox metabolism [32], [33]. Salmon et al. [33] and Liu et al. [32] each identified >350 genes that were differentially expressed when arcA was deleted. However, there was only a minimal overlap between these datasets and it is unclear how many of these genes are direct vs. indirect targets of ArcA. Thus, although ArcA plays a prominent role in the anaerobic repression of genes that encode enzymes for aerobic respiratory metabolism, the full extent of the ArcA regulon remains unclear, preventing a comprehensive understanding of its physiological role.
Despite the identification of several ArcA binding regions by footprinting, the sequence determinants for ArcA DNA binding are also not well understood. This is in large part due to the unusually long length (30–60 plus bp) [23], [24], [26], [28]–[30] and degenerate nature of these sequences, which makes bioinformatic searches challenging. Nevertheless, a 15-bp site consisting of two tandem direct repeats has been proposed as the ArcA recognition site [34]. A similar motif has been derived for Shewanella oneidensis ArcA based on binding energy measurements for every possible permutation of a 15-bp site [35]. However, a 15-bp site is insufficient to explain the extended footprints, raising the question of whether additional sequence conservation beyond 15 bp is important for ArcA DNA binding and transcriptional regulation.
To determine the in vivo binding locations of ArcA in E. coli under anaerobic fermentative growth conditions, we utilized chromatin immunoprecipitation followed by sequencing (ChIP-seq) or hybridization to a microarray (ChIP-chip). Bioinformatic analyses of sequences corresponding to ArcA-enriched regions were used to predict individual ArcA binding sites and to search for a binding motif that could explain the large ArcA footprints. Novel ArcA binding site architectures were then validated by DNase I footprinting. Additionally, gene expression profiling was performed in arcA+ and ΔarcA backgrounds to determine the effect of ArcA DNA binding on gene expression. This combination of genome-wide approaches provided insight into the mechanism of ArcA DNA binding and transcriptional regulation. These results also allowed us to identify additional operons under direct ArcA control, thereby providing a more complete understanding of the physiological role of ArcA in E. coli.
We mapped 176 chromosomal ArcA binding regions (Table S1) across the genome of E. coli K-12 MG1655 during anaerobic fermentation of glucose using ChIP-chip and ChIP-seq (Figure 1). These sites include all but five of the 22 previously identified ArcA binding regions (uvrA/ssb [36], oriC [37], ptsG [38], rpoS [39] and sodA [40]; Figure 1); the absence of a binding region upstream of sodA is likely the result of Fur outcompeting ArcA from binding [40]. ArcA binding was also examined during aerobic respiration using ChIP-chip and as expected, revealed a pronounced decrease in site occupancy (Figure 1) except for a handful of peaks (e.g., ygjG and uxaB), which were not investigated further. As ArcA protein levels remained relatively constant between aerobic and anaerobic conditions (data not shown and [16]), the decrease in occupancy under aerobic conditions can be explained by decreased ArcA-P levels, resulting from the increase in the ratio of oxidized to reduced quinones [20].
Overall, there was good agreement between the ChIP-chip and ChIP-seq datasets (109 peaks in common). However, 15 regions identified by ChIP-chip were resolved into 32 binding regions (Table S2) using ChIP-seq and the CSDeconv peak deconvolution algorithm [41]. For example, compared to only one binding region resolved with ChIP-chip, three binding regions were identified upstream of cydA (Figure 2A) and two were identified within the divergent sdhC/gltA (Figure 2B) promoter region using ChIP-seq. Furthermore, the position of the peak calls with CSDeconv is consistent with the position of known ArcA binding sites mapped by DNase I footprinting within these promoters [24], [27] and 29 of these 32 regions contain a predicted ArcA binding site (Table S2). The correlation of footprinted sites and predicted sites with CSDeconv peak calls allowed us to establish that binding sites separated by as little as 76 bp (based on the CSDeconv-defined coordinate for each binding region) could be resolved. From this analysis, several novel closely spaced ArcA binding sites, e.g. three binding regions upstream of cyo and two binding regions upstream of nuo and pdhR-aceEF-lpdA, were identified. Thus, since ChIP-seq provided higher resolution identification of ArcA binding sites, this dataset was used for all other analyses.
DNase I footprinting experiments indicate that ArcA-P typically binds to long stretches of DNA (30–60+ bp) [23], [24], [26], [28]–[30]. However, the sequence determinants beyond a 15 bp direct repeat within these long stretches are not well understood. Using our high resolution binding regions, we searched for a common sequence recognition element [42], which identified a 18-bp sequence motif consisting of two direct repeat (DR) elements with a center to center (ctc) distance of 11 bp, close to the 10.5 bp per helical turn of B-form DNA, in nearly every (158 of 176) ArcA binding region (Figure 3A; Table S3). While this result extended the previously described ArcA box from 15 to 18 bp [34], we also found that many sites contained additional DR elements beyond the two DRs of the ArcA box. We then systematically searched the sequences surrounding each ArcA box with a 10-bp pair weight matrix (PWM), corresponding to a single DR element (Figure 3B), which revealed a diversity in the number and spacing of DR elements within ArcA binding sites. Although the largest class of binding sites contained just two DR elements at a ctc spacing of 11 bp (66), the majority of ArcA-binding sites (92) contain three to five DR elements predominantly at a ctc spacing of 11 bp (Figure 3C–D, Table S4).
To validate the bioinformatic predictions, DNase I footprinting was performed for a representative set of promoters. Since the OmpR/PhoB family of response regulators is expected to dimerize upon phosphorylation [43], we hypothesized that ArcA would bind as two adjacent dimers to sites with three consecutive DR elements (e.g., icdA and acs), three DR elements at which the distal DR is separated from DR2 by approximately two helical turns (22 bp; e.g., trxC), or four consecutive DR elements (e.g., astC) and in each case, protect a region the size of four DRs (∼44 bp). As anticipated, ArcA-P protected a 44 bp region at the astC promoter (Figure 4A) and a 48 bp region at the trxC promoter (Figure 4B). In contrast, ArcA-P only protected 33 bp and 37 bp regions, respectively, at the icdA and acs promoters, which encompassed the three consecutive DR elements (Figure 4C–D). The result for icdA is in agreement with previous footprinting data [23].
Our footprinting data also suggested that the spacing between DR2 and DR3 is likely important for ArcA-P binding, because ArcA-P did not protect a predicted DR3 element in which the ctc distance between DR2 and DR3 contained an extra bp (12-bp spacing; putP); protection corresponded to only DR1 and DR2 (Figure 4E). A potential explanation of this result is that the increased spacer distance disrupted potential protein-protein contacts between ArcA dimers. Additionally, our footprinting data identified 57 bp and 60 bp ArcA-P-binding regions, respectively, at the paaA and phoH promoters, which spanned from three consecutive predicted DRs to a distal DR element spaced nearly two full helical turns away (22 bp) (Figure 4F–G). As expected, no footprints were detected with unphosphorylated ArcA (data not shown).
Unexpectedly, the ArcA-P footprint at the dctA promoter extended 50 bp downstream of the predicted two DR site (Figure 4H), although this extended region was less well protected. A bioinformatic search revealed a second, but weaker two DR site at the downstream end of this protected region on the opposite DNA strand but no DR elements in the intervening 24 bp region, suggesting that protein-protein contacts may compensate for the absence of identifiable sequence elements at this site. Altogether, these results suggest that the length of the ArcA-P footprint reflects the location of the outermost DR elements within the binding site. In addition, these data reveal plasticity in the architecture among ArcA binding sites with anywhere from two to five DR elements of differing predicted strength present at any given site.
The footprinting results also revealed interesting features about ArcA-P DNA binding. At acs and astC, all DR elements were occupied at the same ArcA-P concentration, whereas at icdA, paaA, phoH, and trxC, occupation of DR3 or DR4 required a higher concentration of ArcA-P. The difference in concentration dependent occupancy of the DR elements at the icdA and acs promoters likely reflects the fact that DR3 of acs is a better match to the ArcA DR element PWM than DR3 of icdA (5 bits versus 3 bits). Furthermore, the transition from an unbound to bound state occurred over a narrow range in ArcA-P concentration, suggesting that ArcA-P binding to DR sites is cooperative, although the apparent degree of cooperativity also varied from site to site. Cooperative binding was particularly striking at the acs and astC promoters and for the three DR region at the phoH promoter, for which saturation occurred with less than a four-fold increase in ArcA-P levels. Finally, we also found that the average sequence conservation of DR elements in predicted binding sites with two, three and four equally spaced DR elements decreases with an increasing number of repeats (Figure S1).
DNase I hypersensitive sites were observed at six of the tested promoters, suggesting that ArcA-P binding to multiple DR sites also results in a bend or kink in the DNA. However, the locations of these hypersensitive sites differed from site to site. For example, a hypersensitive site was observed within the spacer region between the 22-bp spaced DR element and the other DR elements at the trxC, paaA and phoH promoters, whereas hypersensitive sites were observed within DR1 and DR2 at the icdA promoter (+8 and +19). In contrast, hypersensitive sites were located upstream and downstream of the footprinted regions at the acs and astC promoters, respectively. Thus, the binding site architecture appears not only to dictate the length of ArcA-P binding sites, but also to affect the concentration dependence of site occupancy and the DNA structure at target operons. These variations in ArcA-P binding likely have important implications for global transcriptional regulation.
To determine which ArcA binding regions exert an effect on transcription, genome-wide mRNA expression profiles for wild type (WT) and ΔarcA strains were examined. In total, 229 differentially expressed operons (Table S5) were identified, 85 of which were associated with one or more of 88 ArcA binding regions (Text S1) and, thus, are directly regulated by ArcA (Figure 5, Table S6). More than half of the operons that we found to be regulated directly by ArcA have not been previously reported (Table S6) but consistent with previous studies, ArcA acted predominantly as a transcriptional repressor (Figure 1).
Many intergenic ArcA binding regions (76) were associated with operons that did not show an ArcA dependent change in gene expression in our studies. However, previous studies indicated that 13 operons are regulated by ArcA but under different growth conditions (Table S7). For example, cydAB expression is activated by ArcA under microaerobic growth conditions, when FNR repression is relieved [62]. Furthermore, many binding regions (31) are associated with operons that are poorly expressed under our growth conditions in both the arcA+ and ΔarcA strains (e.g., paa operon; Table S8). Since ArcA is predominantly a repressor of transcription, we hypothesized that these promoters were repressed by a second transcription factor or require a transcriptional activator and, therefore, growth under inducing conditions would be required to see an effect of ArcA binding on the transcription of these operons.
To test this idea, we constructed a paaA promoter-lacZ fusion and measured β-galactosidase activity in WT and ΔarcA strains supplemented with phenylacetate (PA) because the paaABCDEFGHIJK operon is known to be repressed by PaaX in the absence of PA [63]. In the presence of PA, ArcA strongly repressed paaA-lacZ expression under anaerobic conditions (23 Miller units for WT), whereas repression was relieved in a strain lacking ArcA (404 Miller units) or under aerobic conditions (294 and 372 Miller units for WT and ΔarcA, respectively), indicating that ArcA prevents induction of the paa operon under anaerobic conditions even when PA is present. Examination of regulatory data in EcoCyc [47] indicated that 11 other poorly expressed operons also are associated with other annotated activators or repressors (Table S8) that may contribute to synergistic regulation with ArcA. Furthermore, ChIP-chip experiments for other transcriptional repressors indicated that under our growth conditions, 15 targets are also bound by Fur, H-NS, or both [Beauchene and Kiley, personal communication; [49]] (Table S8). Thus, repression by Fur and H-NS may mask effects of ArcA. Altogether, these results indicate that ArcA repression likely serves as a secondary layer of control at many of these operons, ensuring that induction does not occur under anaerobic conditions even when the specific inducer is encountered. Thus, the 85 operons that show a change in expression under fermentative growth with glucose represent just a subset of the complete ArcA direct regulon.
Of the 229 operons regulated by ArcA, 145 lacked ArcA binding in vivo and have not been shown previously to be directly regulated by ArcA. To assess whether an ArcA binding site was missed by our ChIP analyses at any of these operons, we searched the intergenic region upstream of each operon using a cutoff of 15 bits (representing the average sequence conservation of the ArcA sequence logo). An ArcA binding site was identified upstream of only seven operons (acnA, prpR, folE, yibF, yigI, dcuC/crcA), indicating that the remaining 135 operons are likely regulated through an indirect mechanism. Since ArcA directly regulates the expression of 17 transcription factors, a hierarchical mode of regulation could, in part, explain the differential expression of some of these operons. Although not all of these transcription factors are expected or known to be active under our growth conditions, differential expression of nine operons can likely be traced to one of these transcription factors (Figure 8). For example, the expression of the AppY dependent appCBA-yccB operon [52] is decreased when arcA is deleted, presumably because of the decrease in appY activation by ArcA. In addition, four target operons (folE, gpmA, dld and eco) of the ArcA-activated sRNA, FnrS were upregulated in the arcA mutant [57], [58]. Finally, although we did not identify an ArcA binding site upstream of arcZ, the downregulation of sdaC (the most strongly repressed target of the ArcZ sRNA in S. enterica [64]) in the absence of arcA is consistent with ArcA-dependent activation of arcZ [65].
Examination of EcoCyc (v15.5) [47] for annotated dehydrogenase enzymes (MultiFun term BC-1), indicated that ArcA either directly or indirectly regulates 37 out of 40 non-glycolytic dehydrogenase enzymes that are favored in the direction of reducing equivalent formation and are not involved in biosynthetic or detoxification functions (Table S9). The carbon oxidation pathways and transporters associated with the substrates of each repressed dehydrogenase are displayed in Figure 6 and the majority of these pathways feed into the TCA cycle for further carbon oxidation. The scope of this repression strongly suggests that a major function of ArcA is to repress all genes encoding enzymes that oxidize non-fermentable carbon compounds, thus preventing the formation of excess reducing equivalents (e.g., NADH, FADH2 and quinols) that cannot be readily re-oxidized in the absence of respiration. Nevertheless, despite the extensive upregulation of dehydrogenase enzymes, ArcA mutants have only a small increase in doubling time from 90 to 105 min (Figure S2A) and only a minor alteration in the distribution of fermentation end products (Figure S2B–C). Succinate and ethanol production were marginally increased and decreased by equivalent amounts in a ΔarcA strain, respectively, and lactate was not a major fermentation product (Figure S2C). This suggests that the NADH/NAD+ ratio was not likely perturbed in our ΔarcA strain in agreement with previous results [66], [67].
Although ArcA and FNR are known to mediate widespread changes in gene expression during the transition from aerobic to anaerobic conditions, the extent of the regulatory overlap between these factors has not been established. Previous gene expression studies have suggested that there may be a large overlap between the genes regulated by ArcA and FNR in both E. coli [33] and S. enterica [68]. However, comparison of our dataset with that determined recently for FNR using identical growth conditions, suggests that there is little direct coregulation (Figure S3). Of the 37 operons that showed both FNR and ArcA dependent changes in expression, only seven are directly regulated by both ArcA and FNR. Rather, differential expression may result from an indirect effect of a fnr deletion on ArcA-P levels, which has been previously suggested to explain the FNR-dependent effect on sdhC and lldP expression [69]. An additional 12 operons show both ArcA and FNR binding in vivo but are differentially expressed in only one dataset (e.g., focA-pflB, cydAB). This minimal overlap in the direct regulons of ArcA-P and FNR suggests that these regulators occupy distinct functional roles in anaerobic gene regulation; the ArcA regulon is largely centered around the repression of aerobic carbon oxidation pathways while FNR appears to function as a more general activator of anaerobic gene expression [49]. Some coregulated operons encode enzymes that direct carbon flow towards either oxidative or fermentative metabolism (e.g., pdhR-aceEF-lpdA, focA-pflB, yfiD) while others encode principal components of the respiratory chain (e.g., nou, ndh, cydA). However, coregulation of other operons (e.g. bssR, ompW, ompC, oppA, ygjG, msrB) by ArcA and FNR is surprising and the physiological implications of this coregulation are unknown.
By comparing ArcA binding in vivo with gene expression profiling data, we have greatly expanded the number of operons regulated by ArcA, leading to important insights into the physiological role, mechanism and sequence requirements for ArcA transcriptional regulation. Our analysis indicates that ArcA directly regulates the expression of nearly 100 operons and is predominantly a repressor of genes encoding proteins associated with carbon oxidation pathways. Furthermore, identification of binding sites upstream of many poorly expressed operons (e.g., paa) suggests that the direct regulon of ArcA could actually encompass as many as 150 operons. Additionally, our bioinformatic and DNase I footprinting analyses reveal a plasticity in the ArcA binding site architecture that likely has important implications for global regulation of carbon oxidation in E. coli.
Our finding that under anaerobic conditions, ArcA reprograms metabolism by either directly or indirectly repressing expression of nearly all pathways for carbon sources whose oxidation is coupled to aerobic respiration suggests a global mechanism for NAD+ sparing. This strategy would facilitate the preferential oxidation of the fermentable carbon source glucose and the sparing of NAD+ for glycolysis by recycling NADH to NAD+ via reductive formation of lactic acid, succinate and ethanol. Thus, ATP synthesis via substrate level phosphorylation is ensured and redox balance of NADH/NAD+ is maintained during anaerobic glucose fermentation. This function of ArcA exhibits parallels to carbon catabolite repression in that it is another mechanism for selective carbon source utilization in cells. Although carbon catabolite repression preferentially selects for glucose utilization over other sugars, ArcA reinforces glucose catabolism through the repression of non-glycolytic carbon oxidation pathways. By integrating signals from both respiratory and fermentative metabolism, which are both enzymatically linked to the NADH/NAD+ redox couple, the ArcAB two component system provides a means for E. coli to maintain the NADH/NAD+ ratio.
Despite the extensive upregulation of dehydrogenase enzymes in an arcA mutant, there was only a minor alteration in fermentation products. This result is in agreement with previous data, which also showed that the NADH/NAD+ ratio is not perturbed in strains lacking ArcA during fermentation [66], [67]. The ability of glucose fermenting cells to maintain redox balance in the absence of ArcA likely reflects thermodynamic and kinetic parameters that favor flux via glucose fermentation and the fact that although many dehydrogenases are upregulated, their substrates are not present preventing competition with glycolysis. Indeed, the activity of several dehydrogenases in cellular extracts was previously shown to be increased in an arcA mutant. However, the fact that the NADH/NAD+ ratio is altered in an arcB strain [70] may be explained by the additional roles of ArcB beyond regulating ArcA [39], [71].
Nevertheless, previous studies suggest that ArcA deficiencies may compromise growth more significantly under conditions that more closely parallel the natural habitats of E. coli. For example, an arcA mutant is defective in both survival during aerobic carbon starvation [72] and in colonization of the mouse intestine [73]. Increased NADH/NAD+ ratios have been observed in an arcA mutant during microaerobiosis [66], [67], which may contribute to the poor fitness of arcA mutants in the gut. Accordingly, it seems reasonable to conclude that this extensive repression of dehydrogenase enzymes by ArcA provides an evolutionary advantage for E. coli in its natural habitats where nutrient conditions are in flux and where many more growth substrates (i.e., both carbon sources and electron acceptors) could be encountered.
Surprisingly, very little in vitro data are available describing mechanisms of ArcA transcription regulation. Nevertheless, the location of the ArcA binding sites and the decrease in σ70 occupancy indicate that ArcA represses by occluding RNA polymerase binding like many repressors. However, the mechanism of activation is unlikely to occur through the direct recruitment of RNA Polymerase as observed with ArcA homologs OmpR [60], [61] and PhoB [59] since no conserved location or orientation of ArcA binding sites was evident. Rather, ArcA may increase transcription through an antirepression mechanism. In support of this notion, in vivo studies of hyaA [31], cydAB [74], appY [75] and yfiD [76] transcription suggest that ArcA activation occurs primarily through disruption of HNS (cydAB and appY), FNR (yfiD) or IscR (hyaA) binding. Furthermore, although the mechanism of ArcA activation of focA-pflB [77] and the PY promoter (from the conjugative resistance plasmid R1) [78] is unknown, DNA binding by ArcA alone appears insufficient for its transcriptional activation. In addition, binding of ArcA alone actually repressed transcription of ndh [79], despite the observation that ndh expression increased when arcA was deleted [32]. Although further in vitro experiments are necessary to investigate the activation mechanism, it seems plausible that ArcA functions solely by binding DNA and activates only indirectly when its binding interferes with the binding and repression by another transcriptional repressor.
The variation in the number, spacing, location and predicted strength of DR elements within the chromosomal ArcA binding regions suggests plasticity in the architecture of ArcA binding sites for either repressed or activated operons. Although the core of each site is an ArcA box containing two, 11-bp ctc spaced DR elements, the majority of binding sites contain an additional one to three DRs predominantly-spaced by approximately one or two turns of the helix of B-form DNA (11 bp or 22 bp ctc spacing). Multiple DR elements have also been observed for some promoters regulated by OmpR [80] and PhoB [59], [81], [82]. However, it is unclear how pervasive multiple repeat elements are for these regulators because the 41 genomic PhoB binding locations recently mapped by ChIP-chip were not searched for sequence elements beyond a single PhoB Box [83] and a conserved sequence motif was not identified within the majority of the 43 OmpR binding sites identified with ChIP-seq [84]. Although the three direct repeat binding site architecture represents a particularly novel finding for the OmpR/PhoB family of response regulators, at least one other example of a response regulator, ComA in B. subtilis, which binds three recognition elements (i.e., an inverted repeat and an additional half site) has been reported and all three elements were shown to be important for both DNA binding and transcriptional activation [85]. Whether the protection of only three DR elements by ArcA reflects binding by a dimer and monomer or two dimers, where the distal subunit is not bound sufficiently to protect sequences from DNase I cleavage, is not yet known.
Since the majority of ArcA binding sites overlap the σ70 promoter recognition elements, the plasticity of these cis-regulatory modules may provide an efficient means of encoding binding sites for ArcA, σ70-RNAP and perhaps other transcription factors within the same narrow sequence space. We propose that having binding sites with different architectures is also an effective mechanism for producing diverse transcriptional regulatory outputs. First, varying the number, strength or location of DR elements should modulate the extent of anaerobic repression. Second, embedding transcription factor binding sites within an ArcA binding site could either enhance or antagonize ArcA function. For example, the DR elements at the trxC, paaA and phoH promoters also overlap a binding site for a transcriptional activator (CRP for paaA [63], OxyR for trxC [45]) or a second promoter (P2 at phoH [86]), allowing additional regulatory control. Third, sites of varying affinities may also impact the sensitivity of promoters to the phosphorylation state of ArcA. For example, the different binding affinities of DR elements at the trxC, icdA, paaA and phoH promoters may allow the fine-tuning of expression in response to changing ArcA-P levels when O2 levels vary [16]. Fine tuning of ompF and ompC expression by OmpR has been observed in response to medium osmolarity due to the presence of multiple upstream OmpR boxes with different affinities [80]. Conversely, the highly cooperative mode of occupancy at the astC and acs promoters would likely render the expression of these operons exquisitely sensitive to changes in ArcA-P levels; thus, expression may more closely resemble an on-off switch. Ultimately, such flexibility in transcriptional regulatory outputs may be an important means for linking the redox sensing properties of the ArcAB two component system with the global optimization of carbon oxidation pathway levels. Further studies are underway to examine the contribution of different binding site architectures to both DNA binding and transcriptional regulation.
All strains were grown in MOPS minimal medium [87] with 0.2% glucose at 37°C and sparged with a gas mix of 95% N2 and 5% CO2 (anaerobic) or 70% N2, 5% CO2, and 25% O2 (aerobic). Cells were harvested during mid-log growth (OD600 of ∼0.3 on a Perkin Elmer Lambda 25 UV/Vis Spectrophotometer).
A paaA promoter-lacZ fusion was constructed as described previously [88] by amplifying the region from +15 to −194 relative to the translation start using primers flanked by XhoI or BamHI restriction sites. A TAA stop codon was incorporated after codon 5 to terminate translation from the Shine-Dalgarno sequence present in this region. The resulting PCR fragment was digested with XhoI and BamHI and directionally cloned into plasmid pPK7035. This lacZ promoter construct was then recombined into the chromosomal lac operon as previously described [88] to create the paaA promoter-lacZ fusion and then transduced using P1 vir into MG1655 and PK9416 (ΔarcA) to creating PK9959 and PK9960 (Table S10). For assays with paaA, 1 mM phenylacetic acid (Sigma Aldrich) was added to the minimal glucose media. To terminate cell growth and any further protein synthesis chloramphenicol (final concentration, 20 µg/ml) was added, and cells were placed on ice until assayed for β-galactosidase activity [89]. β-galactosidase values represent the average of at least three replicates.
arcA was amplified with primers which incorporated a NheI restriction site, a His6-tag and a Tev protease cleavage site (order listed in 5′-3′ direction) on the 5′ end of the gene and a XhoI site at the 3′ end. The NheI and XhoI digested fragments were cloned into plasmid pET 21-d to generate plasmid PK9431 for protein production. E. coli BL21(DE3), containing PK9431 was grown at 37°C until an OD600 of 0.5–0.6 was reached then 1 mM isopropyl-1-thio-β-D-galactopyranoside (IPTG) was added. After seven hours at 30°C, cells were harvested, suspended in 5 mM imidazole, 50 mM Tris-Cl, pH 8.3 and 0.3 M NaCl and lysed by sonication. His6-ArcA was isolated from cell lysates by passage over a Ni-NTA column pre-equilibrated with 5 mM imidazole, washing extensively with the same buffer followed by 50 mM imidazole, and then eluting with a linear gradient of 50–500 mM imidazole. Fractions containing the overexpressed His6-ArcA, determined by electrophoresis, were dialyzed against 50 mM Tris-Cl, pH 8.0 and 0.1 M NaCl and concentrated. Antibodies to ArcA were obtained from Harlan (Indianapolis, In), affinity purified prior to use and determined to be specific to ArcA by Western blot (data not shown). For DNase I footprinting, the His6 tag was removed from ArcA by overnight incubation with tobacco etch virus (TEV) protease at 4°C and passage over a Ni2+-agarose column (Qiagen). The protein concentration of ArcA (reported here as monomers) was determined with the Coomassie Plus protein assay reagent (Pierce), using bovine serum albumin as a standard.
ChIP was performed as previously described [90] using the affinity purified ArcA polyclonal antibodies. ChIP DNA along with corresponding input DNA were amplified by linker-mediated PCR and labeled with Cy3 or Cy5-random 9-mers then hybridized as previously described [49] to custom-made E. coli K-12 MG1655 tiled genome microarrays (Roche NimbleGen, Inc, Madison, WI). The hybridized microarrays were scanned using NimbleGen Hybridization System 4 and the PMT was adjusted as previously described [49]. Quantile normalization (“normalize.quantiles” in the R package VSN) [91] was used to obtain the same empirical distribution across the Cy3 and Cy5 channels and across biological replicate arrays to correct for dye intensity bias and to minimize microarray-to-microarray absolute intensity variations as previously described [92]. The log2 of the ratio of experimental signals (Cy5) to control signals (Cy3) was calculated. Regions of the genome enriched for occupancy by ArcA were identified using TAMALPAIS [93] L2 and L3 stringency levels (95th percentile/p<0.0001 and 98th percentile/p<0.05 of the log2 ratio for each chip, respectively) with the anaerobic fermentative ArcA data. Only enriched regions that were significant in both biological replicates were considered, resulting in the identification of 194 binding regions. Four false positives were eliminated from the data set by analyzing technical replicate ChIP-chip results from a strain lacking arcA (PK9416; Table S11). Fifty-three false positives were eliminated because we found that they resulted from ArcA co-immunoprecipitating with RNA polymerase at highly transcribed regions (Figure S4; Table S12; Text S1; Table S12) leaving 137 regions. The phosphorylation dependence of ArcA DNA binding at these sites was determined by performing a single biological replicate ArcA ChIP-chip experiment under aerobic conditions. For visualization, the anaerobic ArcA biological replicates were averaged then median smoothed using a 300 bp window using MochiView [94].
For ChIP-seq, enriched ChIP DNA from two additional biological replicates from anaerobic ArcA samples were submitted to the University of Wisconsin-Madison DNA Sequencing Facility for library construction and Illumina sequencing performed as previously described [49]. A total of 1,364,908 and 12,074,358 reads were obtained for the ChIP replicates. Greater than 90% and 80% of these reads, respectively, mapped uniquely to the K12 MG1655 genome (version U00096.2) using the software package SOAP release 2.20, allowing no more than two mismatches [95]. The CSDeconv algorithm [41] was then used to determine significantly enriched regions in high resolution using both ChIP-seq replicates and two anaerobic input samples [49] from the same sequencing run as the ArcA ChIP samples. Reads that mapped uniquely within the seven rRNA operon regions were eliminated to allow the algorithm to run more efficiently. CSDeconv was run with Matlab v7.11.0 (R2010b) using the following parameters: LLR = 21.75 and alpha = 800 for replicate one and LLR = 22 and alpha = 550 for replicate two. The find_enriched function was modified to account for differences in sequencing depth between the IP and Input samples. Correction factors of 2.98 (replicate 1) and 0.6579 (replicate 2), calculated by dividing the number of unique reads in the Input sample by the number of reads in the ChIP sample for replicates one and two, respectively, were multiplied by nip and the forward and reverse kernel density calculations for both the forward and reverse strands of the ChIP sample. FDRs of 0.0154 and 0.0156 for replicates one and two, respectively, were calculated by a sample swap (the number of peaks in the Input over the ChIP sample divided by the number of detections in the ChIP over the control sample). From 222 enriched regions generated from two independent ChIP-seq replicates, 146 ArcA-P binding regions (Table S1) were obtained using the same filtering criteria described for ChIP-chip (Table S12; Text S1). For visualization of the ChIP-seq data, the raw tag density at each position was calculated using QuEST version 2.0 [96] and normalized as tag density per million uniquely mapped reads.
The final list of 176 binding regions was obtained by searching binding regions that were found in only one ChIP-seq replicate (48) or were unique to ChIP-chip (28) with the ArcA box PWM (see below) using a cutoff of 10 bits as 99% of ArcA boxes in the alignment have an individual information content of 10 bits or greater. An ArcA binding site was identified in 30 of these binding regions (15 from ChIP-chip and 15 from ChIP-seq) which were, therefore, combined with the 146 regions found in both ChIP-seq replicates to produce the final list of 176 ArcA chromosomal binding regions (Table S1).
Based on the improved resolution of ChIP-seq, sequence corresponding to a 200 bp window around each of the 146 CSDeconv binding regions (averages of the two replicates) was searched for a common motif using MEME [42] with the parameters -mod zoops -nmotifs 1 -minw 18 -maxw 25. Using the alignment from MEME, a sequence logo was built using the Delila software package with the delila, encode, rseq, dalvec, and makelogo programs [97]. A PWM generated from this alignment was used to search the 146 binding regions with a cutoff of 9 bits as this represents the lowest scoring ArcA box included in the MEME alignment. Using the program localbest, only the best scoring ArcA box within a 200 bp region was retained due to several instances of overlapping ArcA-P boxes being identified (sites with three and four DR elements). The resulting 128 ArcA-P boxes were used to make the final sequence logo (Figure 3A). The delila program ri [97] was used to calculate the information content of individual sequences within the positions −3 and 14, which ranged from 9.1 to 21 bits (Table S3). A PWM derived from the conservation of bases between positions −3 and 14 in these 128 ArcA-P boxes, is referred to throughout the paper as the ArcA box PWM. No unique motif was identified within the 18 binding regions without a match to the ArcA box. The scan program [97] was used to search DNA sequences upstream of differentially expressed operons that were not enriched in ChIP using the ArcA box PWM. The E. coli K12 genome sequence [98] was obtained from GenBank (v. U00096.2) and a bit score cutoff of 15 bp bits was used as this represents the average information content of the ArcA box PWM. The localbest program was used to select the best scoring ArcA box within a 200 bp region in cases where two sites were predicted in close proximity.
To construct the 10 bp PWM corresponding to a single direct repeat element, positions −3 to 6 and 8 to 17 from the 128 sequences used to make the ArcA box sequence logo were aligned as they correspond to the nucleotides contacted by each PhoB monomer in the crystal structure of the C-terminus of PhoB bound to its PhoB box [99]. Due to the identical spacing between DR elements and the highly similar nucleotide compositions of the PhoB and ArcA boxes, this structure likely serves as a good model for the nucleotides contacted by each ArcA monomer. A bit score cutoff of 0, which represents the theoretical lowest limit of binding [97], was used to search a 100 bp region surrounding each identified ArcA box with the scan program to identify sites with additional repeat elements. Where displayed, sequence walkers were used to visualize matches to the ArcA-P binding site using the lister program [100].
An in-frame ΔarcA deletion strain was constructed by replacing the coding region of arcA (codons 2–238) with a CmR resistance cassette flanked by FLP recognition target (FRT) sites from plasmid pKD32 in strain BW25993/pKD46, as described previously [101] to generate PK7510. Transduction with P1 vir was used to move the arcA::cat allele into MG1655 to produce PK7514. The CmR cassette of PK7514 was removed by transforming this strain with pCP20-encoding FLP recombinase [101] then screening for loss of Cm, generating PK9416 (Table S10). The deletion was confirmed by sequencing.
RNA was isolated from triplicate MG1655 and ΔarcA (PK9416) strains using a hot-phenol method [102]. The RNA was reverse transcribed to cDNA, labeled with Cy3-random 9-mers and hybridized onto the Roche NimbleGen E. coli 4plex Expression Array Platform (4×72,000 probes, Catalog Number A6697-00-01) as previously described [49]. The expression data was normalized using Robust Multi-Array (RMA) [103] and statistical analysis was performed with Arraystar III software (DNASTAR). Transcripts exhibiting a statistically significant (moderated t-test p-value<0.05) change in expression greater than 2-fold were considered differentially expressed and grouped into operons using operon definitions in EcoCyc [47] if at least two of the genes in a particular operon exhibited differential expression.
Samples (2 ml) for end product analysis were collected during log phase, the transition to stationary phase and in stationary phase (Figure 7A). Cells were removed by passage through a 0.2 µm filter and the supernatant was stored at −80°C prior to analysis. For each sample, glucose, pyruvic acid, succinic acid, lactic acid, formic acid, acetic acid, and ethanol were separated by high-performance liquid chromatography (HPLC) and subsequently quantified as previously described [104].
Plasmids containing predicted ArcA-P binding sites were generated by PCR amplification of chromosomal DNA with primers flanked by XhoI or BamHI restriction sites and cloned into pPK7179 or pPK7035 (for the icdA promoter)(Table S10). The positions of the promoter fragments relative to the previously identified transcription start sites are as follows: for icdA [23], −216 to +65; for acs (P2) [105], −172 to +44; for phoH (P2) [86], −161 to +20; for paaA [106], −132 to +55; for astC [107], −166 to +62; for putP (P1) [108], −120 to +56; for trxC [45], −118 to +50 ; for dctA [109], −185 to +32. The icdA fragment contains two promoters: one whose expression is dependent on ArcA (P1) and a second promoter whose expression is dependent on FruR (P2) [23], [110]. To examine icdA expression from only P1 in future expression analyses, transcription from P2 was eliminated using the site-directed mutagenesis protocol described in [111] to mutate the −10 site from cattat to cggtga. DNA fragments were isolated from pPK7179 or pPK9476 (icdA) after digestion with XhoI and BamHI, radiolabelled at the 3′ BamHI end with [α-32P]-dGTP (PerkinElmer) and Sequenase Version 2.0 (USB Scientific), isolated from a non-denaturing 5% acrylamide gel and subsequently purified with elutip-d columns (Schleicher and Schuell). ArcA was phosphorylated by incubating with 50 mM disodium carbamyl phosphate (Sigma Aldrich) in 50 mM Tris, pH 7.9, 150 mM NaCl, and 10 mM MgCl2 for 1 h at 30°C [24] and immediately used in the binding assays. Footprinting assays were performed by incubating phosphorylated ArcA with labeled DNA (∼5 nM) for 10 min at 30°C in 40 mM Tris (pH 7.9), 30 mM KCl, 100 µg/ml BSA and 1 mM DTT followed by the addition of 2 µg/ml DNase I (Worthington) for 30 s. The DNase I reaction was terminated by the addition of sodium acetate and EDTA to final concentrations of 300 mM and 20 mM, respectively. The reaction mix was ethanol precipitated, resuspended in urea loading dye, heated for 60 s at 90°C, and loaded onto a 7 M urea, 8% polyacrylamide gel in 0.5× TBE buffer. An A+G ladder was made by formic acid modification of the radiolabeled DNA, followed by piperidine cleavage [112]. The reaction products were visualized by phosphorimaging.
All genome-wide data from this publication have been deposited in NCBI's Gene Expression Omnibus (GSE46415.
|
10.1371/journal.pmed.1002781 | Discovery and validation of a prognostic proteomic signature for tuberculosis progression: A prospective cohort study | A nonsputum blood test capable of predicting progression of healthy individuals to active tuberculosis (TB) before clinical symptoms manifest would allow targeted treatment to curb transmission. We aimed to develop a proteomic biomarker of risk of TB progression for ultimate translation into a point-of-care diagnostic.
Proteomic TB risk signatures were discovered in a longitudinal cohort of 6,363 Mycobacterium tuberculosis-infected, HIV-negative South African adolescents aged 12–18 years (68% female) who participated in the Adolescent Cohort Study (ACS) between July 6, 2005 and April 23, 2007, through either active (every 6 months) or passive follow-up over 2 years. Forty-six individuals developed microbiologically confirmed TB disease within 2 years of follow-up and were selected as progressors; 106 nonprogressors, who remained healthy, were matched to progressors. Over 3,000 human proteins were quantified in plasma with a highly multiplexed proteomic assay (SOMAscan). Three hundred sixty-one proteins of differential abundance between progressors and nonprogressors were identified. A 5-protein signature, TB Risk Model 5 (TRM5), was discovered in the ACS training set and verified by blind prediction in the ACS test set. Poor performance on samples 13–24 months before TB diagnosis motivated discovery of a second 3-protein signature, 3-protein pair-ratio (3PR) developed using an orthogonal strategy on the full ACS subcohort. Prognostic performance of both signatures was validated in an independent cohort of 1,948 HIV-negative household TB contacts from The Gambia (aged 15–60 years, 66% female), longitudinally followed up for 2 years between March 5, 2007 and October 21, 2010, sampled at baseline, month 6, and month 18. Amongst these contacts, 34 individuals progressed to microbiologically confirmed TB disease and were included as progressors, and 115 nonprogressors were included as controls. Prognostic performance of the TRM5 signature in the ACS training set was excellent within 6 months of TB diagnosis (area under the receiver operating characteristic curve [AUC] 0.96 [95% confidence interval, 0.93–0.99]) and 6–12 months (AUC 0.76 [0.65–0.87]) before TB diagnosis. TRM5 validated with an AUC of 0.66 (0.56–0.75) within 1 year of TB diagnosis in the Gambian validation cohort. The 3PR signature yielded an AUC of 0.89 (0.84–0.95) within 6 months of TB diagnosis and 0.72 (0.64–0.81) 7–12 months before TB diagnosis in the entire South African discovery cohort and validated with an AUC of 0.65 (0.55–0.75) within 1 year of TB diagnosis in the Gambian validation cohort. Signature validation may have been limited by a systematic shift in signal magnitudes generated by differences between the validation assay when compared to the discovery assay. Further validation, especially in cohorts from non-African countries, is necessary to determine how generalizable signature performance is.
Both proteomic TB risk signatures predicted progression to incident TB within a year of diagnosis. To our knowledge, these are the first validated prognostic proteomic signatures. Neither meet the minimum criteria as defined in the WHO Target Product Profile for a progression test. More work is required to develop such a test for practical identification of individuals for investigation of incipient, subclinical, or active TB disease for appropriate treatment and care.
| Tuberculosis (TB) is currently the leading cause of death by an infectious disease, yet diagnosis of TB is still hampered by poor tools that require a sputum sample.
An accurate, affordable, and easy-to-use diagnostic test would allow targeted antibiotic treatment before symptoms develop and the person becomes infectious, thus providing an opportunity to curb transmission and halt the global epidemic.
In this study, we sought to develop a blood test that can predict if a healthy individual is likely to progress to active TB disease before clinical symptoms manifest.
We analyzed plasma from healthy South African adolescents who were followed over 2 years. By comparing abundance of over 3,000 different plasma proteins from individuals who developed TB disease and others who remained healthy, we identified 2 biomarkers that comprised combinations of either 3 or 5 proteins and predicted onset of TB a year before traditional diagnosis was possible.
The protein biomarkers were validated for accuracy in an independent cohort of individuals from The Gambia.
To our knowledge, these are the first validated protein biomarkers with prognostic value for TB; however, neither meet the minimum performance criteria as set out by WHO for a TB progression test.
More work is required to improve the performance of such tests for practical identification of individuals for investigation of incipient, subclinical, or active TB disease.
| Global efforts to control the tuberculosis (TB) epidemic depend on new, more efficacious TB vaccines and drugs in addition to better diagnostic tests to accurately diagnose those with TB disease. Earlier identification of individuals during incipient or subclinical stages of TB disease progression holds great promise for targeted preventive therapy, which may provide a strategy to curb onward transmission of M. tuberculosis. Such a strategy requires prognostic tests that can accurately identify those at risk of TB disease before the onset of symptoms and further transmission. In 2017, 10 million cases of TB and 1.6 million deaths (more than any other infectious agent) were reported [1–3]. It is estimated that up to 40% of these TB cases are missed and thus not treated, highlighting the limitations of current diagnostic strategies and emphasizing the need for better, faster, and more tractable diagnostic tests [2].
In people with asymptomatic M. tuberculosis infection, the infecting organisms are primarily contained within lung granulomas and/or draining lymph nodes, making direct detection of the bacterium virtually impossible. However, host signals in the blood compartment, such as inflammatory markers, have been shown to reflect the host–pathogen interactions at the site of disease, which can be used to identify those who are progressing from M. tuberculosis infection to active TB disease. For example, we validated blood transcriptomic signatures of TB risk that identified those who progressed to active disease up to 18 months before TB diagnosis [4,5]. Although these RNA-based biomarkers show promise, measurement of plasma proteins is more amenable to development of point-of-care tests, as exemplified by lateral flow tests based on capillary blood collected by needle prick. Indeed, profound changes in abundance of many plasma proteins have been reported in TB patients, and we and others have described protein-based diagnostic TB signatures [6–9]. Further, by measuring kinetic changes in plasma proteins in TB progressors, we observed that proteins involved in inflammatory pathology, tissue repair, matrix-remodeling, elevated interferon responses, and activation of the complement pathway revealed stages of TB disease progression [10]. Similarly, Esmail and colleagues showed that HIV-infected individuals with subclinical TB had elevated plasma levels of immune complexes and blood signatures of complement activation [11].
In this study, we proposed to identify and validate parsimonious proteomic signatures of TB disease risk. We measured >3,000 proteins by multiplexed slow off-rate modified DNA aptamers (SOMAmers) in plasma from M. tuberculosis-infected progressors and nonprogressors and identified 2 proteomic signatures of TB progression, which were validated in an independent cohort.
Plasma samples were available for 37 progressors and 106 nonprogressors from the ACS and were primarily distributed between 1–18 months before TB diagnosis (Tables 1 and S1 and Fig 1A and S2 Text). Participants were randomly split into training and test sets for TRM5 signature discovery at a ratio of 2:1 (Fig 1A). Longitudinally collected samples from each participant were retained in each set and evaluated to ensure sufficient distribution of progressor samples in each 6-month time window approaching the diagnosis of TB disease.
Similarly, plasma samples from 34 progressors and 115 nonprogressors from the Gambian GC6–74 cohort were available for blind validation and distributed between 1–24 months before TB diagnosis [4,5] (Tables 1 and S2 and Fig 1B and S2 Text). A sample-by-sample hybridization normalization was first applied to control for differential hybridization of SOMAmers to the readout microarrays. An intraplate median signal normalization was then applied to control for bulk signal differences between samples. Finally, between-plate signal differences were corrected by calibrating each plate using replicate calibrator samples. For the GC6–74 samples, an additional 45 bridging samples were selected from the ACS cohort and were used to bring the distributions into alignment using a linear transformation.
Protein abundance data are presented in S4 and S6 Tables and S2 Text.
To identify host proteins with differential abundance, we compared all 197 nonprogressor plasma samples with 56 progressor samples from the ACS training set. One hundred thirty-five proteins were found to be different at a 1% Benjamini–Hochberg False Discovery Rate (bhFDR). Of these, 105 proteins were significantly more abundant and 30 proteins less abundant in progressors relative to nonprogressors (Fig 1C and S5 Table and S2 Text). The most differentially abundant protein between progressors and nonprogressors was Galactose-1-phosphate uridyl transferase 1 (GALT-1, log2 fold change = 0.112; P = 2.40 x 10−10; S5 Table and S2 Text), which is involved in galactose metabolism pathways, followed by Matrix Metalloproteinase 1 (MMP-1, log2 fold change = 0.680; P = 2.86 x 10−9), both of which were more abundant in progressors than nonprogressors. The protein found to be most abundant in progressors relative to nonprogressors was the acute-phase marker C-reactive protein (CRP, log2 fold change = 1.31; P = 1.17 x 10−5). The protein found at lowest levels in progressors relative to nonprogressors was Creatine Kinase type M/type B (CK-MB, log2 fold change = −0.528; P = 1.66 x 10−5).
Amongst all possible signatures with 1, 2, 3, 4, or 5 proteins, the signature with the highest AUC in cross-validation on the ACS training set was a 5-protein signature called TRM5, consisting of complement factor C9, insulin-like growth factor-binding protein 2 (IGFBP-2); B-cell antigen receptor complex–associated protein (CD79A), Matrix-Remodeling Associated 7 protein (MXRA-7), and neuronal cell-adhesion molecule (NrCAM). TRM5 signature scores were higher in progressors than nonprogressors, and the signature readily discriminated progressor from nonprogressor samples collected 1 to 180 days before TB diagnosis (AUC 0.961; 95% CI 0.931–0.99, Fig 2A and Table 2). Prognostic performance decreased for samples collected between 181 and 360 days before TB diagnosis, with an AUC of 0.761 (95% CI 0.648–0.874, Fig 2A and Table 2). The TRM5 signature did not significantly discriminate between progressor and nonprogressor samples collected more than 1 year before TB diagnosis (AUC 0.55; 95% CI 0.414–0.691, Fig 2A and Table 2).
To assess performance of the TRM5 signature on an unseen verification partition of the ACS progressors and nonprogressors, we applied it to blinded plasma samples from the ACS test set, comprising 13 progressors and 36 nonprogressors who were not included in the model discovery training set. The TRM5 signature discriminated progressor from nonprogressor samples spanning 1–720 days before TB with an AUC of 0.76 (95% CI 0.67–0.86, P < 0.001, Fig 2B), verifying the performance observed in the training set.
Our work on transcriptomic signatures showed that a 16-gene mRNA signature allowed significant discrimination between samples from progressors and nonprogressors at time points more than 12 months before TB diagnosis [4]. We therefore employed a different discovery approach that combined the ACS training and test sets to develop a signature that may provide better discrimination in samples collected more than a year before TB diagnosis. In this strategy, we also sought to make the signature as parsimonious as possible; we employed a pair-ratio strategy that incorporates a small ensemble of pairwise models, each comprising 1 protein with higher and 1 with lower abundance in progressors relative to nonprogressors. Using leave-one-out cross-validation, 3 proteins were selected, including C9 (higher in progressors than nonprogressors), CK-MB, and Complement C1q Tumor Necrosis Factor-Related Protein 3 (C1qTNF3/CTNFF3) (both lower in progressors than nonprogressors), which together formed the 3PR signature, an ensemble of 2 protein pairs (Fig 3A). Only 1 protein, C9, was common to the TRM5 and 3PR signatures. Proteins with differential abundance in the ACS training and test sets combined are in S7 Table and S2 Text.
Performance of the 3PR signature in the combined training plus test set was comparable to that of the TRM5 model (AUC 0.89, 95% CI 0.84–0.95) in samples between 1 and 180 days before TB diagnosis and in samples between 181 and 360 days before TB (AUC 0.72, 95% CI 0.64–0.81, Fig 3B). Notably, the 3PR signature also significantly discriminated between progressor and nonprogressor samples collected 361 to 720 days before TB, with an AUC of 0.71 (95% CI, 0.63–0.80). This enhanced performance at time points distal to TB may be due to a larger sample size of the discovery cohort used for the 3PR signature than that used for discovery of TRM5.
To validate the TRM5 and 3PR proteomic TB risk signatures in an independent cohort, we retrieved plasma samples from Gambian adult household contacts of TB cases who participated in the GC6–74 study [4,5] (S2 Table and S2 Text). Assignment of samples to progressor status, draw date, and participant were blinded. Raw fluorescence unit (RFU) signal levels in 45 ACS samples that were run on both the original SOMAscan discovery assay and the custom SOMAscan assay for bridging indicated a systematic intensity shift between signal levels. Despite the shift in mean signal intensity, most protein measurements generated with the ACS discovery array were well correlated with the original SOMAscan measurements, and the bulk intensity change was removed using the standard SOMAscan assay bridging procedure, which transforms the raw concentration ranges generated by the 45 ACS bridging samples on the validation array into the concentration ranges generated on the original discovery array. Fig 4 displays cumulative distribution functions of the TRM5 and 3PR analytes for the GC6–74 samples before and after assay bridging. A single progressor sample (of 61) and a single nonprogressor sample (of 193) failed SOMAscan operating procedure QC criteria—all other samples were deemed fit for analysis with the risk models. Distribution of progressor and nonprogressor signature scores in the ACS and GC6 cohorts was not different for the TRM5 model, although they were significantly different for the 3PR signature across discovery and validation assays (S4 Fig). Protein abundance data in the GC6 validation samples are in S6 Table and S2 Text.
Prognostic performance of both TRM5 and 3PR was determined on samples collected up to 2 years before the diagnosis of TB disease in the GC6–74 validation cohort (Fig 5). Both TRM5 and 3PR discriminated between Gambian progressors and nonprogressors within 1 year of TB diagnosis (TRM5: AUC 0.66 [95% CI 0.56–0.75]; 3PR: AUC 0.65 [0.55–0.75]).
Prognostic performance by both signatures was generally poor for samples collected from 1–2 years before diagnosis. When substratified into 6-month time windows before diagnosis of TB disease, performance of both models was, as anticipated, strongest most proximal to diagnosis (Table 2). The 3PR signature discriminated between progressor and nonprogressor samples collected 7–12 months before TB (AUC 0.67 [0.55–0.79], P = 0.019), and the TRM5 signature discriminated between progressor and nonprogressor samples 13–18 months before TB (AUC 0.75 [0.59–0.91], P = 0.0078). Neither signature showed significant performance for samples collected more than 18 months before TB diagnosis. After the bridge calibration procedure, only C9 and NrCAM were observed to have mean RFU values that were significantly different (Bonferroni P < 0.05) in the GC6–74 data set when explored in the ANOVA posthoc analysis for directionality of bias.
A target product profile for a test that predicts progression from TB infection to active disease, or an incipient TB test (ITT), was recently developed by FIND and WHO [21], which benchmarked the minimum sensitivity and specificity for such a test at ≥75% and ≥75%, respectively (optimal sensitivity and specificity were ≥90% and ≥90%). Neither TRM5 nor 3PR achieved these minimum criteria when tested for progression to incident TB diagnosed within a year of testing in the GC6 cohort; TRM5 achieved a sensitivity of 49% (95% CI 33%–65%) at a specificity of 75% (95% CI 68%–81%) and 3PR a sensitivity of 46% (95% CI 31%–63%) at a specificity of 75% (95% CI 68%–81%). By comparison, prognostic performance of CRP, the protein with the highest differential abundance between ACS progressors and nonprogressors, was promising in the combined ACS training and test sets in samples within 1 year of diagnosis (AUC 0.76; 95% CI 0.69–0.83) (S3A Fig). However, validation in the GC6–74 cohort was not statistically significant (AUC 0.62; 95% CI 0.49–0.74, P = 0.058). Despite this, CRP had a similar sensitivity of 41% (95 CI 22%–61%) at a specificity of 75% (95% CI 67%–82%) in the GC6–74 cohort (S3B and S3C Fig).
Using a well-characterized prospective longitudinal cohort of M. tuberculosis-infected South African adolescents, we discovered 2 prognostic protein signatures, TRM5 and 3PR, that successfully identified individuals at risk of incident TB disease risk within a year of the onset of disease symptoms. Validation of the prognostic performance of these signatures in an independent cohort of household contacts of TB patients from the Gambia represents a first step to an affordable and practical prognostic biomarker for TB.
While other proteomic biomarkers have been discovered with diagnostic potential for symptomatic TB disease [6–9], this outcome represents only one stage within the spectrum of M. tuberculosis infection. A biomarker with prognostic value that can identify asymptomatic individuals with incipient or subclinical disease would open the opportunity for early, targeted preventive treatment and the potential to curb M. tuberculosis transmission. A recent review of incipient or subclinical disease suggested that the number of individuals with these early stages of disease progression must be at least equivalent to the number of active TB cases: 10 million [22]. The only current tests that can identify those at risk of TB are interferon gamma release assays (IGRAs) or TSTs, which detect immunological sensitization to M. tuberculosis. These tests have low positive predictive value (PPV) for prognostic application [23,24], and the prevalence of TST+ or QFT+ people can be as high as 80% in countries endemic for TB. In fact, epidemiological models suggest that up to 23% of the global population may be infected with M. tuberculosis [25] and thus are at risk of disease progression, although a recent analysis has suggested that the proportion of individuals truly at risk of progression is likely smaller than the TST models suggest [26]. Regardless, these studies highlight the need for a prognostic test for incident TB that is more sensitive and specific than IGRAs and TSTs.
Neither TRM5 nor 3PR achieved the minimum criteria for an incipient TB test (ITT) set out by FIND and WHO [21], and it is clear that more work is needed to improve the performance of prognostic signatures based on proteins. The same was true of the prognostic performance of CRP. Notably, a recent diagnostic accuracy study conducted in 2 Ugandan HIV/AIDS clinics showed that point-of-care CRP screening of HIV-infected people with CD4 counts <351 cells per μL who were initiating antiretroviral therapy yielded 89% sensitivity and 72% specificity for culture confirmed TB [27]. The study supported use of CRP as a TB screening test to improve efficiency of case finding.
Nevertheless, our study reports, to the best of our knowledge, the first proteomic prognostic signature for TB and demonstrates feasibility of the approach. Prognostic transcriptomic signatures of TB risk have been developed using RNA sequencing [4,5], microarrays, in silico analysis of published data sets, as well as PCR-based methods [15,28]. While such transcriptomic signatures possess immense potential, their access to the market is hindered by high cost and the need to translate measurement of mRNA-based signatures to practical point-of-care devices for use in community healthcare or surveillance settings. A parsimonious proteomic signature could, in principal, be more amenable for adaptation to a portable and low-cost test, such as a lateral flow–based assay.
Interpretation of our results would benefit from verification with a different protein quantification technology, such as sandwich ELISA as proof-of-principle of antibody-based detection of proteins identified with SOMAmers, although commercial ELISA antibodies for detection of some of the proteins in the TRM5 and 3PR signatures at the appropriate biological range are limited. Ultimately, aptamer-based sandwich assays for analyte quantitation may be a viable alternative for point-of-care assays since aptamers can be manufactured reproducibly and do not require a cold chain. Translation to commercial methodologies would also allow easier uptake and external validation of these signatures in other populations and settings. This would also allow analysis of the effect on signature performance derived during the transition from the >3,000-plex SOMAscan discovery assay to the custom SOMAscan assay used for validation. We observed a systematic shift in signal magnitudes generated by the validation assay compared to the >3,000-plex discovery assay. Though the bridge calibration removed most of this artifact, there was still some residual shift in mean signal intensity for C9 and NrCAM, which may have contributed to the decrease in prognostic performance of TRM5 and 3PR in the GC6 validation cohort. Additionally, differences in disease epidemiology in the underlying populations, country of residence, strain of circulating M. tuberculosis, and/or the amount of heparin or other preanalytic processing variables in the plasma samples may also have contributed to a difference in performance between the ACS and GC6 cohorts. Regardless, our results showed that both proteomic signatures validated in the GC6 cohort and provide proof-of-principle that a prospective protein-based biomarker for incident TB is possible.
Our results of relative abundances of 2,872 plasma proteins in progressors and nonprogressors provide an opportunity to reflect on the biological pathways underlying progression from M. tuberculosis infection to active TB disease. We have previously shown that proteins associated with type I/II interferon responses (e.g., interferon gamma-inducible protein 10 [IP-10]) and complement cascade activation were elevated early during progression, up to 12 months before TB diagnosis, and are likely biomarkers of early incipient disease [10]. Elevated plasma proteins associated with myeloid inflammation, tissue repair, matrix remodeling, coagulation, and platelet activation were detected more proximal to TB diagnosis and suggestive of underlying pathology consistent with subclinical or active TB disease [10]. It was noteworthy that the methods employed to discover the TRM5 and 3PR signatures, which were completely agnostic to underlying biology, selected complement component C9 for inclusion in both proteomic signatures. This, along with the inclusion of C1qTNF3 in 3PR, further signifies the role of complement activation in TB disease progression, as shown by recent transcriptomic and proteomic studies [9–11]. C1qTNF3, which was less abundant in plasma from progressors than nonprogressors, has been shown to be inversely correlated with BMI and a proinflammatory obese state [29]. C1qTNF3 is a metabolic hormone with beneficial anti-inflammatory properties [30–32], and prior studies have found that obese individuals are at lower risk of incident TB [33] but greater risk of diabetes, which in itself is suggested as a TB risk factor [34]. The antidiabetes drug metformin, which has shown therapeutic potential in controling growth of M. tuberculosis [35], acts to increase C1qTNF3 levels [36]. Other studies have implicated low levels of C1qTNF3 in other inflammatory diseases such as rheumatoid arthritis [37], heart disease, lipid dysregulation, and apoptosis. Similarly, activation of the complement cascade in general and elevated C9 levels likely reflect the acute inflammatory responses and high type I interferon expression during TB disease progression [4,5,10,38]. The IGFBP-2 protein is implicated in growth and metabolism and was observed to increase during progressing infections [39], while plasma levels of insulin-like growth factor–binding proteins have been shown to change during TB treatment [40]. NrCAM is a member of the immunoglobulin superfamily and is important in cell adhesion and thought to be involved in immunity and pulmonary fibrosis [41,42]. While these inflammatory, immune activation, and tissue repair molecules provide some interpretation behind the biology of TB disease progression, the role of other differentially abundant proteins in the signatures, such as the dentin-associated ameloblastin (AMBN) and neuronal cell–associated NrCAM, are less clear and will require further investigation.
Our study had a number of limitations. Greater statistical power for signature discovery and validation would have been achieved with larger cohort sizes. It is critical that more progressor cohorts are assembled for future work on prognostic biomarkers for TB. In this light, the prospectively collected samples from the 76 progressors in both the ACS and GC6–74 cohorts—collected from 8,314 enrolled individuals—are of immense value. As such, the highly multiplexed SOMAscan assay was well suited for discovery, and the resulting data set is a valuable resource for the TB research community (S2 Text). The systematic shift in signal magnitudes generated by the validation assay compared to the discovery assay may be an important factor in the performance of TRM5 and 3PR in the validation cohort, as discussed above. New discovery using the entire ACS and GC6–74 data sets may allow discovery of a more universal signature, and it will be important to confirm the performance of these proteomic models on alternative platforms.
The performance of these signatures as diagnostic screening or triage tests should be further explored and compared with other protein-based diagnostic signatures [6–9], as such a signature with diagnostic utility would be an ideal tool for advancing the clinical care for TB. A next step is evaluation of the diagnostic performance in individuals with presumptive TB disease compared to those without confirmed TB but presenting with respiratory symptoms.
Successful validation of these proteomic signatures suggests that a simple proteomic test to predict progression to active TB disease is achievable. With further refinement and validation, the prospect of an affordable, point-of-care device to provide a tool to curb transmission is possible. While performance demonstrated here is not sufficient to meet minimal WHO guidelines for predicting progression of TB [1], the novelty of these prognostic signatures and the theoretical simplicity and robustness of a proteomic lateral flow test provides renewed hope in a prognostic marker for point-of-care.
|
10.1371/journal.pgen.1006827 | Arabidopsis RAD51, RAD51C and XRCC3 proteins form a complex and facilitate RAD51 localization on chromosomes for meiotic recombination | Meiotic recombination is required for proper homologous chromosome segregation in plants and other eukaryotes. The eukaryotic RAD51 gene family has seven ancient paralogs with important roles in mitotic and meiotic recombination. Mutations in mammalian RAD51 homologs RAD51C and XRCC3 lead to embryonic lethality. In the model plant Arabidopsis thaliana, RAD51C and XRCC3 homologs are not essential for vegetative development but are each required for somatic and meiotic recombination, but the mechanism of RAD51C and XRCC3 in meiotic recombination is unclear. The non-lethal Arabidopsis rad51c and xrcc3 null mutants provide an opportunity to study their meiotic functions. Here, we show that AtRAD51C and AtXRCC3 are components of the RAD51-dependent meiotic recombination pathway and required for normal AtRAD51 localization on meiotic chromosomes. In addition, AtRAD51C interacts with both AtRAD51 and AtXRCC3 in vitro and in vivo, suggesting that these proteins form a complex (es). Comparison of AtRAD51 foci in meiocytes from atrad51, atrad51c, and atxrcc3 single, double and triple heterozygous mutants further supports an interaction between AtRAD51C and AtXRCC3 that enhances AtRAD51 localization. Moreover, atrad51c-/+ atxrcc3-/+ double and atrad51-/+ atrad51c-/+ atxrcc3-/+ triple heterozygous mutants have defects in meiotic recombination, suggesting the role of the AtRAD51C-AtXRCC3 complex in meiotic recombination is in part AtRAD51-dependent. Together, our results support a model in which direct interactions between the RAD51C-XRCC3 complex and RAD51 facilitate RAD51 localization on meiotic chromosomes and RAD51-dependent meiotic recombination. Finally, we hypothesize that maintenance of RAD51 function facilitated by the RAD51C-XRCC3 complex could be highly conserved in eukaryotes.
| Meiotic recombination and sister chromatid cohesion are important for maintaining the association between homologous chromosomes and ensuring their accurate segregation. Meiotic recombination starts with a set of programmed DNA double-strand breaks (DSBs), catalyzed by the SPO11 endonuclease. Processing of DSB ends produces 3′ single-stranded DNA tails, which form nucleoprotein filaments with RAD51 and DMC1, homologs of the prokaryotic RecA protein. The eukaryotic RAD51 gene family has seven ancient paralogs, in addition to RAD51 and DMC1, the other five members in mammals form two complexes: RAD51B-RAD51C-RAD51D- XRCC2 (BCDX2) and RAD51C-XRCC3 (CX3). To date, the molecular mechanism of CX3 in animal meiosis remains largely unknown due to the essential roles of these two proteins in embryo development. In Arabidopsis, RAD51C and XRCC3 are required for meiosis and fertility, but their specific mechanisms are unclear. Here we present strong evidence that Arabidopsis RAD51 forms a protein complex with AtRAD51C-AtXRCC3 in vivo. Our data also support the previous hypothesis that CX3 promotes RAD51-denpendet meiotic recombination by affecting its localization on chromosomes. Given that the RAD51, RAD51C and XRCC3 proteins are highly conserved in plants and vertebrates, the mechanism we present here could be important for the regulation of meiotic recombination in both plants and vertebrate animals.
| Homologous recombination (HR) is important for repairing DNA damage and maintaining genomic stability. Meiotic HR and sister chromatid cohesion are required for maintaining physical associations between homologous chromosomes (homologs) and ensuring their accurate segregation. Meiotic HR is initiated by programmed DNA double-strand breaks (DSBs) that are catalyzed by SPO11, a topoisomerase-like protein [1]. The resulting DSB ends are processed by the MRE11- RAD50-NBS1 (MRN) protein complexes to generate 3′ single-stranded DNA (ssDNA) tails [2,3], which are subsequently protected by replication protein A (RPA) [4]. Functional homologs of the E. coli RecA protein, RAD51 and DMC1 [5,6] bind to the 3ʹ tails to form nucleoprotein filaments with the help of several proteins identified in multiple species, including Saccharomyces cerevisiae (Rad52 [7], Rad54 [8], Tid1/Rhd54 [9], Rad55-Rad57 [10], Swi5-Sfr1 [11] and PCSS complex [12]), Arabidopsis thaliana (RAD51C [13], XRCC3 [14], MND1-HOP2[15] and ATR/ATRIP [16]), and mammals (Mnd1-Hop2 [17] and Brca2-Dss1 [18]). The nucleoprotein filaments facilitate single-end invasion of a non-sister chromatid, resulting in the formation of a recombination intermediate called a D-loop, which is then processed to ultimately produce either crossovers (COs) or non-crossovers (NCOs) [19].
In vertebrate animals and plants, the RAD51 gene family is highly conserved with seven members: DMC1, RAD51, RAD51B, RAD51C, RAD51D, XRCC2 and XRCC3 [20–23], which share Walker A and Walker B motifs with over 37.5% similarity [24]. In mice, mutations in any of the paralogs, except DMC1, lead to embryonic lethality following spontaneous DNA damage or errors [25–29]. In the model plant Arabidopsis thaliana, all seven genes are dispensable for vegetative growth [13,14,24,30–33]. However, AtRAD51, AtRAD51C and AtXRCC3 are required for somatic and meiotic recombination, as well as plant fertility. Mutations in any of these three genes result in a meiotic chromosome fragmentation phenotype [13,14,24,30–32]. Moreover, AtDMC1 is specifically required for meiotic homolog pairing and recombination [34,35]. In contrast to atrad51, atrad51c and atxrcc3 mutants, atdmc1 mutants do not suffer meiotic chromosome fragmentation; instead their DSBs are thought to be repaired using sister chromatids as templates [34,35]. The three other paralogs, AtRAD51B, AtRAD51D and AtXRCC2, seem to be unnecessary for meiotic DSB repair, because the triple mutant has normal chromosome morphology and fertility [33]. Except for slight differences in synapsis, the chromosome morphology using light microscopy for DAPI-stained chromosomes and fertility phenotypes of atrad51c and atxrcc3 mutants are similar to those of atrad51, suggesting that their functions are related, but further analyses are needed to understand their mechanistic roles in meiotic DSB repair.
Biochemical studies in human cells demonstrate that RAD51 paralogs associate with one another in two distinct complexes: RAD51B-RAD51C-RAD51D-XRCC2 (BCDX2) and RAD51C-XRCC3 (CX3) [36,37]. The CX3 complex stabilizes RAD51 binding to ssDNA [36–39] in vitro, thus promoting single-end invasion. Moreover, RAD51C and XRCC3 also help mediate Holliday junction (HJ) resolution in vitro [40], suggesting a later role in meiotic recombination. A yeast two-hybrid assay demonstrated that the Arabidopsis RAD51 paralogs also interact with each other [41], supporting the idea that RAD51 paralogs function by formation of distinct protein complexes in both animals and plants. However, whether the RAD51 paralogs associate with each other in planta has not been tested.
In this study, we report that Arabidopsis homologs of RAD51, RAD51C and XRCC3 show highly similar meiotic chromosome morphological defects using immune-localization for key markers. We also provide evidence that AtRAD51C and AtXRCC3 are required for AtRAD51 localization on chromosomes. Both in vitro and in vivo data demonstrate that AtRAD51C interacts with AtRAD51 and AtXRCC3. Furthermore, observation of AtRAD51 foci in atrad51, atrad51c and atxrcc3 single, double and triple heterozygotes reveals that AtRAD51C and AtXRCC3 both are involved in AtRAD51 loading. Triple heterozygotes also experience non-homolog chromosome associations and have reduced CO frequencies. Together, these results demonstrate that AtRAD51C, AtXRCC3 and AtRAD51 form a complex in planta and are required for AtRAD51 loading on chromosomes.
Previous studies have found that AtRAD51, AtRAD51C and AtXRCC3 are required for meiotic DSB repair and plant fertility and mutation of individual genes cause indistinguishable chromosome entanglement and fragmentation phenotypes [13,14,31,32]. The similarity of the phenotypes suggests that these genes might function in the same genetic pathway or process. To test this hypothesis, we generated double mutants between atrad51-3 (SAIL_873_C08) [42], atrad51c (SALK_021960) [13], and atxrcc3 (SALK_045564) [14] and found that the chromosome morphologies of atrad51 atrad51c (48 cells), atrad51 atxrcc3 (65 cells), and atrad51c atxrcc3 (54 cells) double mutants showed no obvious differences compared with each of the single mutants (S1 Fig). The lack of an additive phenotype in the double mutants further supports the hypothesis that they act together in the same biological process.
To search for subtle chromosomal phenotypes that could discriminate between the three mutants, we used FISH with a centromere probe for atrad51 (82 cells); atrad51c (96 cells) and atxrcc3 (81cells) and a bacterial artificial chromosome (BAC-F19K16) probe that targets a telomere proximal region on chromosome 1 for atrad51 (31 cells); atrad51c (45 cells) and atxrcc3 (22 cells) [43]. Wild-type (WT) meiocytes had three to five centromere signals at pachytene, indicative of paired homologous centromeres in a cluster (Fig 1A). Although the three mutants had no typical pachytene chromosomes, they all displayed similar centromere clusters or numbers of signals at a stage similar to that of WT, suggesting that AtRAD51, AtRAD51C and AtXRCC3 are not required for early centromere pairing or clustering (Fig 1D, 1G and 1J). At diakinesis and metaphase I, WT meiocytes had five bivalents, each with two paired centromere signals corresponding to the associated homologs (Fig 1B). In contrast, the three mutants each had 10 centromere signals located on abnormally associated chromosomes (multivalents-with more than two chromosomes) (Fig 1E, 1H and 1K), indicating a failure to maintain homolog association, at least at the centromere regions. We next examined homolog pairing on the chromosome arms using the telomere-proximal BAC probe. Unlike the single focus observed on WT pachytene chromosomes, indicative of fully synapsed homologs, meiocytes from each of the three mutants showed two separate signals, indicating a failure to pair properly (Fig 1M–1P). We also performed ASY1 and ZYP1 immuno-localization in WT and mutants. No obvious difference of ASY1 signals at zygotene was found between WT and mutants (S2 Fig). However, unlike WT with linear ZYP1 distribution on pachytene chromosomes, ZYP1 was completely disappeared in rad51, while some punctate or discontinuous ZYP1 signals were observed in xrcc3 and rad51c (S2 Fig). Together, these results demonstrate that AtRAD51, AtRAD51C and AtXRCC3 are not required for recombination-independent centromere clustering, but are necessary for homolog pairing, consistent with previous findings obtained using FISH experiment [44]. The similarities of the mutant phenotypes further support the idea that they act in the same process.
Loading of RAD51 on ssDNA is aided by several proteins, including Rad52 [45], Rad55-57 (Rad51 paralogs) [46] and Sfr1-Swi5 [11] in yeast, the Brca2-Dss1 complex in mammalian cells [47], and also by AtBRCA2 in Arabidopsis [48]. The similarity of meiotic defects in Arabidopsis rad51, rad51c and xrcc3 mutants suggests the RAD51 paralogs RAD51C and XRCC3 may function in meiotic recombination by affecting RAD51 function. To test this hypothesis, we performed an immunofluorescence assay using an AtRAD51 antibody [49]. In Arabidopsis, formation of DSBs is thought to occur at leptotene [50]. At a similar stage, we found that WT plants had 187.7±24.5 AtRAD51 foci per meiocyte (n = 14), but the number of foci was greatly reduced in atrad51c (36.1±9.7, n = 17; P = 1.5E-13) and atxrcc3 (33.7±10.3, n = 34; P = 5.7E-13) mutant meiocytes (Fig 2A, 2C, 2D and 2Q). In contrast, a parallel experiment did not detect any AtRAD51 foci in atrad51 mutant meiocytes at zygotene (Fig 2B). A similar pattern was also observed using pachytene meiocytes (Fig 2E–2H). These results provide evidence that Arabidopsis RAD51C and XRCC3 are required for formation of wild type level of RAD51 foci on meiotic chromosomes. This is consistent with the previous findings for Rad51 paralogs in yeast [46]. Nevertheless, the reduction of AtRAD51 foci in atrad51c and atxrcc3 homozygous mutants does not preclude the possibility that normal level DSBs are formed in these mutants. To test whether DSB frequency is altered in atrad51c and atxrcc3 mutants, we examined the distribution of a DSB marker, phosphorylated histone H2AX (γ-H2AX) [51]. At zygotene, after DSBs have been formed, no significant differences in the number of γ-H2AX foci were detected between WT (189.3±26.5, n = 39), atrad51 (176.7±15.5, n = 19; P = 0.062), atrad51c (183.6±18.0, n = 18; P = 0.41) and atxrcc3 (178.3±13.5, n = 19; P = 0.097) mutants (Fig 2I–2L and 2R). In Arabidopsis, most meiotic DSBs are thought to be repaired during zygotene-pachytene. We found that γ-H2AX foci were obviously reduced in WT (56.9±15.2, n = 55) pachytene meiocytes compared those of atrad51 (132.1±15.4, n = 13; P = 1.5E-11), atrad51c (120.6±16.6, n = 14; P = 7.2E-11) and atxrcc3 (122.2±18.8, n = 18; P = 1.7E-11) mutants (Fig 2M–2P and 2S). The presence of normal numbers of zygotene γ-H2AX foci and reduced AtRAD51 foci suggests that AtRAD51C and AtXRCC3 are not required for meiotic DSB formation, but are necessary for AtRAD51 loading.
In yeast, the Rad51 paralogs Rad55 and Rad57 form a heterodimeric complex to stimulate RAD51 activity [10]. Vertebrate Rad51 paralogs interact with one another to form two distinct complexes: BCDX2 and CX3 [52]. Like vertebrates, Arabidopsis has seven RAD51 paralogs, and previous yeast two-hybrid assays have shown that XRCC3 interacts with both RAD51C and RAD51 [41]. However, whether these proteins interact in planta has not been investigated.
As an initial test for potential interactions we used a yeast two-hybrid assay (Y2H) and found that AtXRCC3 interacts with both AtRAD51 and AtRAD51C (Fig 3A), consistent with the previously identified interactions in Y2H system [41]. The interaction between AtRAD51C and AtXRCC3 was further supported by a pull-down assay using recombinant fusion protein of glutathione S-transferase (GST) with AtRAD51C and an AtXRCC3-His tag fusion protein (Fig 3B). In addition to the previously identified interactions, we also found that GST-AtRAD51 interacts with AtRAD51C-His (Fig 3B). To explore whether these associations also occurred in planta, we examined the interactions using bimolecular fluorescence complementation (BiFC) in tobacco (Nicotiana benthamiana) cells. Strong nuclear signals, indicating interaction, were observed for AtRAD51C with either AtRAD51 or AtXRCC3 (Fig 3C). These results provide the first direct evidence that plant RAD51 paralogs RAD51C and XRCC3 interact directly with RAD51 in vitro and in planta.
A recent study identified a weak atrad51 allele, atrad51-2 [42], with a T-DNA insertion in the 3′-untranslated region (UTR) that results in reduced AtRAD51 protein levels. This mutant had mild chromosome fragmentation and partial synapsis, as well as some bivalent formation with homologs and non-homologs [42]. In contrast, the atrad51-1 null mutant had severe chromosome fragmentation and formed multivalents during meiotic prophase I [31]. These findings suggest that reducing AtRAD51 level might be a strategy for investigating its meiotic function. Alternatively, analysis of double heterozygous mutants in genes encoding components of a complex can reveal phenotypic defects, even though the corresponding single heterozygotes are phenotypically normal [53,54]. We hypothesized that double/triple heterozygotes of atrad51, atrad51c and atxrcc3 might reduce, but not abolish, their interactions in a complex and reveal informative meiotic phenotypes
To test this hypothesis, we generated atrad51-/+, atrad51c-/+ and atxrcc3-/+ double and triple heterozygous mutants and compared their meiotic phenotypes with WT. Analysis of meiotic chromosome morphology after DAPI staining showed that atrad51-/+, atrad51c-/+ and atxrcc3-/+ single heterozygote meiocytes and atrad51-/+ atrad51c-/+ and atrad51-/+ atxrcc3-/+ double heterozygotes had similar phenotypes compared to WT (Fig 4A–4L; S3 Fig). In addition, meiocytes from atrad51c-/+ atxrcc3-/+ double heterozygotes had chromosome morphology similar to WT at pachytene (Fig 4M), but at diakinesis, WT formed five bivalents, whereas 32.8% (20 of 61, n = 61) of the atrad51-/+ atxrcc3-/+ double heterozygote meiocytes had non-homologous chromosome associations (Fig 4N). The cell appears to be able to resolve these associations since equal division of chromosomes was observed at anaphase I and II (Fig 4O and 4P). Meiocytes from atrad51-/+ atrad51c-/+ atxrcc3-/+ triple heterozygotes had a more severe non-homolog association phenotype (47.8% at diakinesis, 22 of 46, n = 46, Fig 4R) and had unequal chromosome segregation at metaphase II (16.7%, 2 of 12, n = 12, Fig 4T). No chromosome fragments were observed in the triple heterozygote, suggesting it is still capable of DSB repair. The results also suggest that RAD51C and XRCC3 are functionally more related to each other than either is to RAD51. Previous studies showed that T-DNA translocation can cause a similar pattern of chromosome association using light microscopy because the translocated chromosome can associated with two normal chromosomes [55,56]. We have verified the T-DNA insertion site by sequencing the junction with flanking genomic DNAs and the results indicated that these mutations are not associated with translocations.
To test whether meiotic DSB repair is delayed in the heterozygotes, we performed immunostaining experiments using a γH2AX antibody. As mentioned above, WT meiocytes had 189.3±26.5 (n = 39) and 56.9±15.2 (n = 55) γH2AX foci at zygotene and pachytene, respectively (Fig 5A, 5I, 5Q and 5R and Table 1). All single, double and triple heterozygotes had no obvious differences in the number of γH2AX foci at zygotene, but had significantly more foci at pachytene (Fig 5A–5R, S1 Table). Moreover, the double and triple heterozygotes had more foci at pachytene than the single heterozygotes. There are significantly fewer foci in atrad51-/+ (78.1±19.4, n = 17), atrad51c-/+ (83.3±10.8, n = 12) and atxrcc3-/+ (82.0±25.9, n = 24) (S1 Table) compared to the double mutants atrad51-/+ atrad51c-/+ (96.3±15.4, n = 30), atrad51-/+atxrcc3-/+ (100.4±14.8, n = 21) and atrad51c-/+atxrcc3-/+ (105.2±24.1, n = 15), which in turn have significantly fewer foci (S1 Table) than the triple atrad51-/+ atrad51c-/+ atxrcc3-/+ (113.3±14.8, n = 46) (Fig 5J–5P and 5R and Table 1). These data suggest that DSB formation is normal in the heterozygotes, but there is a defect in the progression of DSB repair, and that AtRAD51, AtRAD51C and AtXRCC3 function in this process.
Because AtRAD51C and AtXRCC3 are required for normal AtRAD51 localization, we next examined AtRAD51 localization in heterozygous mutant meiocytes. As described above, WT meiocytes have 187.7±24.5 (n = 14) AtRAD51 foci at zygotene and 51.2±14.0 (n = 65) foci at pachytene (see Fig 6A and 6I for examples and Table 1). In contrast, single, double and triple heterozygous mutant meiocytes have significantly fewer AtRAD51 foci at zygotene (p<0.05; Fig 6A–6H and 6Q). At pachytene, the three single mutant heterozygotes show no obvious differences in the number of AtRAD51 foci compared with WT (Fig 6I–6L and 6R), but the double and triple heterozygotes exhibited reduced AtRAD51 foci (p<0.05; Fig 6M–6P and 6R). These findings are consistent with the earlier results, suggesting that AtRAD51C and AtXRCC3 play related roles in AtRAD51 loading on chromosomes, likely in a protein complex.
As described earlier, the weak atrad51-2 allele is capable of forming bivalents and executing recombination [42]. Similarly, the heterozygous plants analyzed here also completed meiotic recombination to some extent and had partial fertility. To examine CO frequencies in comparison between the various genotypes, we counted the number of chiasmata, the physical manifestation of crossing-over, in WT and mutant meiocytes at both diplotene and metaphase I. On average, WT had 10.1±1.1 (n = 52) chiasmata per meiocyte and no obvious significant differences were observed in the single heterozygotes: atrad51-/+ with 9.6±0.7 (n = 20; P = 0.072) per meiocyte, atrad51c-/+ with 9.6±0.7 (n = 21; P = 0.052) per meiocyte and atxrcc3-/+ with 9.6±0.8 (n = 24; P = 0.066) per meiocyte. The atrad51-/+ atrad51c-/+, atrad51-/+ atxrcc3-/+ and atrad51c-/+ atxrcc3-/+ double heterozygotes showed a slight, but statistically significant, reduction of chiasmata with 8.4±1.2 (n = 14; P = 6.0E-05), 8.0±0.8 (n = 10; P = 2.9E-06) and 7.1±1.0 (n = 34; P = 1.2E-20) per meiocyte, respectively (Fig 7A–7D and 7K). The atrad51-/+ atrad51c-/+ atxrcc3-/+ triple heterozygous mutant also had a significant reduction, with only 6.9±1.0 (n = 15; P = 2.6E-10) chiasmata per meiocyte formed (Fig 7E and 7K). Furthermore, the chiasmata numbers per meiocyte of atrad51c-/+ atxrcc3-/+ double heterozygote (7.1; P values = 2.0E-03 and 1.1E-02, respectively) and the triple heterozygote (6.9, P values = 1.7E-03 and 8.8E-03, respectively) were significantly lower than those of the other two double heterozygotes.
Arabidopsis forms two types of COs: interference-sensitive Type I COs that require ZMM proteins like MSH4 and MLH1 [57–59], and interference-insensitive class II COs that are MUS81-dependent [60,61]. To assess the impact of RAD51 and its paralogs on Type I COs, we used an AtMLH1 antibody to visualize AtMLH1 foci, in WT, atrad51-/+ atrad51c-/+, atrad51-/+ atxrcc3-/+, atrad51c-/+ atxrcc3-/+and atrad51-/+ atrad51c-/+ atxrcc3-/+ meiocytes at diakinesis [59]. On average, WT meiocytes had 9.0±1.2 foci (n = 61, Fig 7F), whereas at similar stages, atrad51-/+ atrad51c-/+, atrad51-/+ atxrcc3-/+, atrad51c-/+ atxrcc3-/+ and atrad51-/+ atrad51c-/+ atxrcc3-/+ mutants had 7.9±1.4 (n = 40; P = 5.8E-05), 7.7±1.6 (n = 25; P = 5.4E-04), 6.4±1.3 (n = 39; P = 5.9E-16) and 5.9±1.0 (n = 16; P = 5.5E-12) foci, respectively (Fig 7G–7J and 7L). The reduction of AtMLH1 foci in the mutants is consistent with the observed reduction in chiasmata, and supports the idea that Type-I COs are reduced in the mutants.
Although the CO number was obviously reduced by ~30% in the atrad51-/+ atrad51c-/+ atxrcc3-/+ triple heterozygote, no univalents were observed, consistent with a mechanism that ensures at least one CO per chromosome [62]. If the COs were distributed among the 5 Arabidopsis bivalents randomly, they would follow the Poisson function P (k COs per bivalent) = (λke-λ)/k! where λ is the mean number of COs per bivalent. Using this function, from the analyses of 52 WT and 15 atrad51-/+ atrad51c-/+ atxrcc3-/+ triple heterozygote meiocytes, we would expect to find 36 and 19 univalents in WT and the triple mutant, respectively, but none were observed (Table 2).
To further quantify the remaining COs in atrad51-/+ atrad51c-/+ atxrcc3-/+, we used a flow cytometry-based assay that measures the segregation of transgenes encoding fluorescent marker proteins expressed using a pollen-specific LAT52 promoter (FTL markers) [63,64]. The number of viable pollen grains is dramatically reduced in atrad51-/+ atrad51c-/+ atxrcc3-/+, but it was still feasible to measure CO frequencies using this assay. We crossed atrad51-/+ atrad51c-/+ atxrcc3-/+ with line I2b, which carries two FTL markers (YFP and DsRed) on chromosome 2 (Fig 8A). Pollen grains which express both fluorescent proteins have not experienced a crossover between the markers, while those that express only one or the other have. The relative abundance of these two classes can be used to calculate the genetic distance between the two markers [65]. We scored 10,092 WT pollen grains and 15,460 pollen grains from the triple heterozygote (Fig 8B and 8C). The I2b map distance was 5.28±0.58 cM in WT and 2.87±0.33 cM in the triple heterozygote (Fig 8D–8G). The genetic distance between the two fluorescent markers was significantly reduced in atrad51-/+ atrad51c-/+ atxrcc3-/+ (Z score = 185.4, P value << 0.01) (Fig 8G), consistent with the reduction in chiasmata described above.
RAD51 family members are conserved across species, from yeast to humans [20]. The budding yeast S. cerevisiae has four RAD51 paralogs (Rad51, Dmc1, Rad55 and Rad57) [10], whereas humans have seven paralogs (RAD51, DMC1, RAD51B, RAD51C, RAD51D, XRCC2, XRCC3) [20]. In yeast, Rad55 interacts with Rad57 to form a stable heterodimer [10]. Similarly, in humans, two complexes are formed by the RAD51 paralogs: the BCDX2 and the CX3 complexes [36,37,39]. Moreover, a recent study in Caenorhabditis elegans showed that the RAD51 paralogs, RFS-1 and RIP-1, also exist as a heterodimer and interact with RAD51 [66].
Arabidopsis XRCC3 has been shown to interact with both RAD51 and RAD51C using a yeast two-hybrid assay [41]. We confirmed the yeast two-hybrid result (Fig 3A) and demonstrated that the AtRAD51C-AtXRCC3 interaction occurs in planta by using pull-down and BiFC assays (Fig 3B and 3C). It is noteworthy that both pull-down and BiFC assays support an interaction between AtRAD51C with AtRAD51 and AtXRCC3. Our results strongly support the idea that AtRAD51C is a central factor in complex formation, and is associated with AtRAD51 and AtXRCC3. These findings are consistent with previous results in human cells that show AtRAD51C associates with two protein complexes [36,37,39]. Our study is also the first time to show that RAD51 paralogs form a protein complex with RAD51 in plants, supporting the hypothesis that formation of RAD51-paralogs associated protein complexes is highly conserved across eukaryotes, including yeast, humans and plants.
Previous studies in yeast showed that RAD51 paralogs are unable to form filaments with ssDNA and do not have a direct role in homology search or single strand invasion [10,23]. Nevertheless, studies in different organisms have reported that RAD51 paralogs play important roles in promoting RAD51 function in both mitotic and meiotic HR [46,67–69]. For example, the yeast Rad55-Rad57 complex has a role in RAD51-dependent HR [10,46]. Similar roles have been found for the C. elegans heterodimer of RAD51 paralogs RFS-1/RIP-1 [66] and the human CX3 complex [36,37,39].
Due to the lack of direct biochemical data, the role of RAD51 paralogs in meiotic HR in planta is unclear. Studies in the monocot model plant, Oryza sativa (rice), showed that the RAD51 paralogs OsRAD51C and OsXRCC3 are required for meiotic DSB repair and mutations in either result in sterility, chromosome entanglement and fragmentation [70,71]. These results are consistent with similar findings in Arabidopsis [13,14,32]. Immunostaining showed that OsXRCC3 is required for OsRAD51C localization on chromosomes [70], suggesting the existence of a potential OsRAD51C-OsXRCC3 complex in rice. Additionally, the single-end processing proteins OsCOM1 and OsDMC1 no longer associate with DSB sites in rice osxrcc3, which suggests that OsXRCC3, and by extension OsRAD51C, might function upstream of OsRAD51 [70,72]. Nevertheless, the relationship between OsRAD51 and its paralogs OsRAD51C and OsXRCC3 remains unclear, because a RAD51 antibody is currently unavailable in rice. In the present study, we showed that Arabidopsis RAD51 foci were obviously reduced in atrad51c and atxrcc3 mutants, consistent to the discovery in rice. Together, these studies, in both rice and Arabidopsis, strengthen the idea that AtRAD51 depends on its paralogs for normal function and that this relationship is highly conserved in eukaryotes.
Previous studies showed that RAD51 paralogs have a later role in processing meiotic recombination intermediates [40]. Direct evidence to support the RAD51C-XRCC3 complex having a role in the later meiotic recombination process come from the observation that the RAD51C-XRCC3 complex is associated with HJ resolvase activity. Moreover, RAD51C- and XRCC3-defective hamster cells have reduced resolvase activity and HJ progression [40,73]. Similarly, the Arabidopsis RAD51 paralogs AtRAD51B and AtXRCC2 were also reported to affect meiotic recombination in terms of CO number [74]. However, mutations in these paralogs show an increase in meiotic recombination frequency [74], suggesting that they have roles in meiotic CO formation. In the present study, we found that atrad51c atxrcc3 double heterozygous mutant and the atrad51 atrad51c atxrcc3 triple heterozygous mutant have significantly fewer COs (Fig 7D, 7E and 7I–7L), compared with WT. Given that the reduced number of AtRAD51 foci observed in the double and triple heterozygous mutants, we propose that a diminished capacity to form wild type level of RAD51 foci results in fewer COs in the mutants. The previous finding further supports this idea that a weaker atrad51 allele had fewer chromosome fragments and some univalents, and also formed bivalents between homologs and non-homologs [42]. Therefore, we speculate that AtRAD51 could function in two manners, both dependent on the AtRAD51 paralogs AtRAD51C and AtXRCC3. Most AtRAD51 foci are required for DNA repair using either homologs or sister chromatids as templates without CO formation, while a small number of AtRAD51 foci might play a role in normal CO formation dependent also on AtDMC1. Therefore, the AtRAD51C-AtXRCC3 is critical for ensuring wild type number of AtRAD51 foci and COs and facilitating proper homolog recombination and association.
Based on our results and previous studies, we propose a model for how AtRAD51C and AtXRCC3 function in conjunction with AtRAD51 in meiotic HR (Fig 9). Meiotic recombination is initiated by programmed DSBs that are catalyzed by AtSPO11-1 and other proteins. The broken ends are further processed by the MRX protein complex to produce ssDNA tails [2,75–77]. In WT, interaction between the AtRAD51C-AtXRCC3 complex and AtRAD51 is proposed to alter the latter’s configuration and facilitates its binding with the ssDNA tails, thus resulting in single end invasion. Consequently, repair of the DSBs yields either COs or NCOs. In the heterozygous mutants, the reduced AtRAD51 level is likely insufficient for supporting the AtDMC1 function, consistent with previous studies in both Arabidopsis and yeast showing that normal DMC1 function in meiosis requires RAD51 [6,78]. Thus, with reduced amounts of RAD51 proteins, single end invasion is possibly more promiscuous and targets both homologous and non-homologous templates, resulting in multivalent formation. This aspect of the model is supported by the observation that the triple heterozygous mutant and the weak atrad51 mutant had non-homologous associations and reduced COs. In the homozygous mutants, when AtRAD51 is either completely absent or reduced below a threshold, most or all DSBs are unrepaired, leading to severe chromosome fragmentation and chromosome entanglements. Further investigations are needed to establish the precise AtRAD51 thresholds and how the AtRAD51 paralogs maintain the necessary level of AtRAD51 during the single-end invasion process.
In summary, meiotic DSB repair is essential for sexual reproduction in eukaryotes including budding yeast, animals and flowering plants. RAD51 paralogs facilitate the establishment of RAD51 at DSBs and mediate and single end invasion. These functions are also highly conserved in eukaryotes. We propose that facilitation of normal RAD51 function by its paralogs, such as RAD51C and XRCC3, may be a general mechanism for meiotic DSB repair.
The mutants atrad51-3 (SAIL_873_C08) [42], atrad51c (SALK_021960) [13], atxrcc3 (SALK_045564) [14] used in this study were shown previously to be null mutants in the Columbia (Col-0) background. atrad51-/+ atrad51c-/+ and atrad51-/- atrad51c-/- mutants were crossed by atrad51-/+ (male parent) and atrad51c-/+ (female parent), atrad51-/+ xrcc3-/+ and atrad51-/- xrcc3-/- mutants were crossed by atrad51-/+ (male parent) and xrcc3-/+ (female parent), atrad51c-/+ xrcc3-/+ and atrad51c-/- xrcc3-/- mutants were crossed by atrad51c-/+ (male parent) and xrcc3-/+ (female parent). Triple heterozygous mutants were crossed by atrad51c-/+ (male parent) and atrad51-/+ atxrcc3-/+ (female parent). Plants were grown at 21°C with 16 h light and 8 h dark. Mutant genotypes were confirmed by PCR using the primers described in S2 Table.
A minimum of 10 plants were characterized for each mutant. Chromosome spreads were stained with DAPI and centromere FISH, and immuno-localization experiments were carried out as described previously [79]. Rabbit polyclonal AtRAD51 and γ-H2AX antibodies were used at 1:200 fold dilutions and Alexa Fluor 488 Goat Anti-Rat IgG (H+L) secondary antibody (A-21428, Invitrogen, Carlsbad, CA, USA) was used at a 1:1000 fold dilution [80]. Chiasmata distribution statistics were performed following the protocol of Sanchez et al. [81]. BAC DNA extraction (F19K16) and probe labeling were described previously [43]. Images of chromosome spreads were obtained using an Axio Imager A2 microscope (Zeiss, Heidelberg, Germany) equipped with a digital camera (Canon, Tokyo, Japan), and processed using Photoshop CS (Adobe Systems, Mountain View, CA). Images were initially captured in black & white and, if necessary, globally false-colored post-capture for visual contrast. AtRAD51 and γ-H2AX foci in WT and mutant lines were counted and statistically analyzed using ImageTool version 3.0 software (University of Texas Health Science Center, San Antonio, USA).
In mutants that lacked synapsis, we distinguished zygotene from pachytene chromosomes by their relative condensation, with pachytene being more condensed than zygotene chromosomes.
To construct the vectors for yeast two-hybrid, pull-down and BiFC assays, full-length AtRAD51, AtRAD51C and AtXRCC3 cDNA were PCR-amplified using Phanta Super-Fidelity DNA polymerase (Vazyme Biotech Co., Ltd, China) and appropriate primers (S2 Table).
For the Y2H assay, full-length AtRAD51 and AtXRCC3 cDNA were purified and ligated into pGADT7 pGBKT7 by NdeI and BamHI double-enzyme digestion, and full-length AtRAD51C cDNA was purified and ligated into pGADT7 and pGBKT7 by NdeI and EcoRI double-enzyme digestion.
For the BiFC assay, full-length AtRAD51 and AtXRCC3 cDNA was purified and ligated into pXY103, pXY104, pXY105 and pXY106 by BamHI and SalI double-enzyme digestion, and full-length AtRAD51C cDNA was purified and ligated into pXY103, pXY104, pXY105 and pXY106 by XbaI and SalI double-enzyme digestion.
For the pull down assay, full-length AtRAD51 and AtXRCC3 cDNA was purified and ligated into pET32a and pGEX-6P-1 by BamHI and SalI double-enzyme digestion and full-length AtRAD51C cDNA was purified and ligated into pET32a and pGEX-6P-1 by EcoRI and SalI double-enzyme digestion. All constructs were verified by DNA sequencing.
Plasmid vectors were transformed into the Y2H gold yeast strain (pGBKT7 constructs) or the Y187 yeast strain (pGADT7 constructs) using the LiAc/PEG method. Transformants were mated on YPDA medium for 48 h, and selected on SD/–Trp–Leu plates for 36 h. Transformants were then selected on SD/–His–Ade–Trp–Leu with X-α-Gal and AbA plates to test for positive interactions [82].
AtRAD51, AtRAD51C and AtXRCC3 were expressed in E. coli using the pGEX6P-1 and pET32a plasmids. The tagged proteins were mixed and incubated for 2 h at 4°C, then pulled down by GST beads for 1 h at 4°C. The protein mixture was confirmed by western blotting with a GST antibody (AG768, Beyotime Co. Ltd, China) or a His-tag antibody (AH367, Beyotime Co. Ltd, China) at 1:100 dilutions, followed by application of an horseradish peroxidase (HRP) goat anti-mouse IgG (H+L) secondary antibody (A0216, Beyotime Co. Ltd, China) at a 1:2000 dilution.
BiFC plasmids (pXY103/104/105/106-RAD51, pXY103/104/105/106-RAD51C, pXY103/104/105/106-XRCC3 and pXY103/104/105/106) were transformed into Agrobacterium GV3101 cells. Transformants were harvested once the OD600 reached 2.0, and resuspended in MES/MgCl2/acetosyringone solution to a final OD600 of 1.0. Cell suspensions were mixed in 1:1 ratios of various combinations, and young Nicotiana benthamiana leaves were infiltrated. Leaves were excised and visualized using a LSM-710 confocal microscope (Zeiss) following 36 h incubation [83].
Open flowers from WT plants or atrad51-/+ atrad51c-/+ atxrcc3-/+plants that were hemizygous for the fluorescent-tagged line (FTL) interval I2b and either QRT+/+ or qrt-/+ were collected [64]. The flowers (50 or more) were mixed with 1 mL PBS buffer (10 mM CaCl2, 1 mM KCl, 2 mM MES, 5% w/v sucrose, pH 6.5) supplemented with 0.01% Triton X-100 in a 1.5-mL microcentrifuge tube. The mixture was vortexed at maximum speed for 2–3 min and the solution filtered through a 70-μm Falcon® cell strainer (352350, Corning Life Sciences, Tewksbury, MA, USA) at 450 ×g for 2 min at 4°C. The flow-through was resuspended in a fresh tube with 1 mL PBS buffer at 4°C. Flow cytometry analysis was performed using a Gallios flow cytometer (Beckman Coulter, Inc.). Statistical analysis was performed using Kaluza Analysis 1.3 software (Beckman Coulter, Inc.) using the two-color analysis methods described previously [65,84].
Excel 2016 (Microsoft, USA) was used to calculate the mean and standard error of the AtRAD51 foci, γ-H2AX foci, MLH1 foci and the chiasmata numbers of WT and mutants. Data was compared using Student’s t-tests and P values were reported as either exact values or Gaussian approximations.
|
10.1371/journal.ppat.1006909 | Insect tissue-specific vitellogenin facilitates transmission of plant virus | Insect vitellogenin (Vg) has been considered to be synthesized in the fat body. Here, we found that abundant Vg protein is synthesized in Laodelphax striatellus hemocytes as well. We also determined that only the hemocyte-produced Vg binds to Rice stripe virus (RSV) in vivo. Examination of the subunit composition of L. striatellus Vg (LsVg) revealed that LsVg was processed differently after its expression in different tissues. The LsVg subunit able to bind to RSV exist stably only in hemocytes, while fat body-produced LsVg lacks the RSV-interacting subunit. Nymph and male L. striatellus individuals also synthesize Vg but only in hemocytes, and the proteins co-localize with RSV. We observed that knockdown of LsVg transcripts by RNA interference decreased the RSV titer in the hemolymph, and thus interfered with systemic virus infection. Our results reveal the sex-independent expression and tissue-specific processing of LsVg and also unprecedentedly connect the function of this protein in mediating virus transmission to its particular molecular forms existing in tissues previously known as non-Vg producing.
| Rice stripe virus (RSV), which is completely dependent on Laodelphax striatellus for transmission between host plants, can also be vertically transmitted from the mother insect to its offsprings. Passing through the insect hemolymph is an essential step for both kinds of viral transmission. In this study, we found that RSV binds to L. striatellus vitellogenin (LsVg) in hemocytes. This tissue-specific LsVg-RSV interaction protects the virus for survival in the hemolymph and enhances both subsequent types of viral transport. More excitingly, we characterized novel and important properties of the insect form of Vg, an indispensable protein in almost all oviparous animals. In our study, we revealed for the first time that insect Vg transported into the ovary is also produced by tissues other than the fat body. Furthermore, we identified the tissue-specific molecular form of Vg responsible for its biological function in virus vertical transmission, uncovered non-female expression of Vg, and determined its function in virus horizontal transmission. These findings provide novel insights into plant-virus transmission within the vector, an important yet less explored part of the virus life cycle.
| Rice stripe disease is a serious problem during rice production, with epidemics occurring repeatedly in China, Japan and Korea [1–3]. Transmission of the causative pathogen, Rice stripe virus (RSV), is completely dependent on insect vectors, the most important of which is the small brown planthopper (SBPH; Laodelphax striatellus) [4]. RSV is transmitted by L. striatellus in a persistent-propagative manner [4]. The RSV filamentous ribonucleoprotein particles (RNPs) are ingested by L. striatellus individuals feeding on RSV-infected plants. Once inside the insect, the virus invades the midgut epithelium to establish infection; it then spreads within the gut and disseminates into the hemolymph. From the hemolymph, the virus further infects various L. striatellus tissues, including the salivary glands. RSV is then horizontally transmitted from the salivary glands into a healthy plant, and also invades the female ovaries, from where it is vertically transmitted to the offspring [5, 6]. Vertical transmission results in naturally existing RSV-infected L. striatellus, which presents a further challenge in disease control.
RSV RNPs contain four single-stranded RNAs, and the major nucleocapsid protein (CP) encoded by the ORF at the 5’ half of the viral complementary RNA3 [4]. Thus, CP is considered the key viral component for specifically interacting with the vector components and plays roles in RSV transmission. Recently we demonstrated that CP interacts with L. striatellus vitellogenin (Vg) in vitro, and that the virions co-localize with Vg in the insect germarium [5]. RSV accomplishes its vertical transmission by binding to Vg and via the uptake of Vg by developing oocytes [5]. In the present study, we focused on the molecular events of the Vg–RSV interaction prior to oocyte penetration. Detailed knowledge of virus transmission mechanisms is required for the design of novel disease control strategies.
Vgs are precursors of the major egg storage proteins in many oviparous animals [7, 8]. Insect Vg is usually synthesized extra-ovarially by the fat body. After processing and modification, also in the fat body, the protein is secreted into the hemolymph and taken up by oocytes via receptor-mediated endocytosis [9–12]. All the known Vgs have been reported to possess modifications; however, the extent of Vg modifications varies considerably. Vg undergoes co- and post-translational modifications, as well as proteolytic cleavage [7, 12–14]. The primary Vg precursor separates into Vg subunits upon proteolytic cleavage. All insect Vgs, excluding those of the honeybee suborder Apocrita, are cleaved in vivo at the tetra-residue motif R-X-R/K-R by subtilisin-like endoproteases [8, 15–18]. This conserved R-X-R/K-R motif is located near the N-terminus and is flanked by polyserine tracts (see review [7]). Cleavage at this motif gives rise to two subunits, one large (140–190 kDa) and one small (40–60 kDa) (see review [19]). In some insects, the Vg precursor contains additional RXXR motifs, and the large subunit is further cleaved into two medium-sized (~90–110 kDa) polypeptides [17, 20–22]. Vg subunits produced by proteolytic cleavage are usually assembled in tissues and secreted into the hemolymph as oligomeric proteins. However, only the small subunit is secreted into the hemolymph, while the large subunit is consumed in the fat body [23].
Vg was initially regarded as a female-specific protein; however, Vg synthesis, albeit in small quantity, has been shown to occur in males and sexually immature animals, indicating that the function of Vg extends beyond serving as an energy reserve for the nourishment of developing embryos [14, 24, 25]. In recent years, accumulating data have shown that Vg from fish plays a role in immune responses [26–32], either as a pattern recognition molecule to recognize bacteria, or as an opsonin to enhance macrophage phagocytosis [27, 33]. Moreover, Vg has been reported to directly kill bacteria via interaction with lipopolysaccharides and lipoteichoic acid present in bacterial cell walls [28], and to neutralize viruses by binding to and creating cross-links between virions [26]. A few studies on the immunological properties of Vg in species other than fish have also been reported. For example, Vg of the mosquito Anopheles gambiae Vg is able to interfere with the anti-plasmodium response by reducing the parasite-killing efficiency of the antiparasitic factor TEP1 [34]. Vg in honeybees also has immunological binding properties, and, moreover, is able to mediate trans-generational immunity by transporting microbe-derived molecules into developing eggs [35].
In a previous study, we have examined the interaction between L. striatellus Vg (LsVg) and the RSV CP and addressed the molecular interaction and its function in mediating RSV vertical transmission [5]. In the present investigation, we further explored molecular details of the RSV–Vg interaction occurring prior to ovary-uptake. We found that only the Vg produced by and processed in insect hemocytes can interact with RSV. This molecular form of Vg is also produced by non-female L. striatellus, in hemocytes only and facilitates RSV transmission.
Our previous study indicated that RSV RNPs entered the L. striatellus oocyte by binding to the Vg protein before reaching the germarium [5]. To ascertain in which tissue(s) the RSV–LsVg interaction occurs, we analyzed the expression levels of LsVg in various female L. striatellus tissues and carried out experiments to localize the LsVg protein. Tissues analyzed included the fat body, hemocytes, the midgut and salivary glands, all of which have been proposed previously to be involved in the transmission of persistent propagative viruses.
Quantitative real-time PCR (qRT-PCR) was performed to test the LsVg gene expression levels. Both the fat body and hemocytes produced abundant LsVg mRNA. By contrast, only a few of the salivary gland and midgut samples exhibited LsVg expression, all at low levels (Fig 1A). An immunofluorescence assay (IFA) was performed to visualize LsVg protein distribution. Using an LsVg-specific monoclonal antibody (designated as Ab47Km in this study) against the polypeptide fragment RNQQKTKSRSRRS [5], the LsVg protein was found to localize in four types of tissue at varying abundances (Fig 1B). Consistent with the observed mRNA levels, LsVg protein was abundantly synthesized in both the fat body and hemocytes (Fig 1B), but was only at the detectable levels in a few of the salivary gland and midgut samples.
Given that multiple L. striatellus tissues expressed LsVg and that both the fat body and hemocytes produced the protein at high levels, we next investigated whether LsVg from these tissues plays a role in mediating RSV transmission.
To determine the tissue or space in which the LsVg–RSV interaction occurred, IFAs were performed to detect the co-localization of LsVg and RSV in the insect tissues. By using antibodies against the RSV RNPs, we found that the virus was distributed in all four tested L. striatellus tissues (Fig 2). To our surprise, LsVg and RSV co-localized only in hemocytes. Even though LsVg was abundant in the fat body, the protein did not co-localize with RSV in that tissue. No co-localization was observed between RSV and the low levels of LsVg in the midgut or salivary glands either.
Previous studies have demonstrated that the Vg homologous protein is proteolytically cleaved into several subunits before being deposited in the eggs as vitellin (Vn) polypeptides [19]. Our earlier investigation of L. striatellus Vg also indicated that its N-terminal Vit-N domain does not interact with the RSV CP in vitro [5]. We thus hypothesized that LsVg is tissue-specifically processed and that the molecular form existing in the fat body is unable to interact with RSV. To test this hypothesis, we analyzed the LsVg cleavage profile and used subunit-specific anti-LsVg antibodies to investigate the subunit composition of LsVg in different tissues.
Insect Vgs are usually proteolytically cleaved prior to secretion into the hemolymph, whereas the Vg subunits, in some cases, are further processed in the ovaries [17, 19]. To reveal processing patterns of the LsVg protein in different tissues, we first analyzed the subunit composition of the L. striatellus vitellin (LsVn) and then used subunit-specific antibodies to determine the molecular form of LsVg in particular tissues.
LsVn was purified from female insects 3 days after molting, and the LsVn subunits were fractionated by SDS-PAGE. The purified LsVn resolved into four well-separated major bands with molecular sizes of approximately 178, 111, 67 and 42 kDa (Fig 3A). Mass spectrometry was performed to identify the LsVg-derived peptides of the four SDS-PAGE bands. Localization of the identified peptide hits revealed that the 42- and 178-kDa bands corresponded to the N- and C-parts of the LsVg protein, respectively. Peptide hits from the 67- and 111-kDa bands were included in the 178-kDa large subunit, where they clustered on its N- and C-parts, respectively. These results revealed the cleavage profile of LsVg. In particular, the precursor protein is first cleaved into small (42 kDa) and large (178 kDa) subunits, with the latter further cleaved into two medium-sized subunits (67 and 111 kDa). As insect Vgs are usually cleaved at the consensus tetra-residue motif RXXR by subtilisin-like endoproteases [15, 16], the LsVn subunit composition suggested the existence of two potential RXXR cleavage sites.
Inspection of the 2,045 amino acids of the LsVg protein sequence allowed us to identify seven RXXR motifs: RSRR ending at amino acid 445, RNNR at 496, RTAR at 719, RFQR at 844, RNNR at 1,043, RSGR at 2,020 and RCVR at 2,043. Based on ProPeptide cleavage site prediction (ProP 1.0 Server) [36], an above-threshold cleavage score, 0.873, was obtained for the RSRR motif flanked by conserved polyserine domains (Fig 3B). Cleavage after this motif resulted in two subunits with calculated molecular weights of 47 and 178 kDa. Because all the other RXXR motifs had cleavage scores below the prediction threshold, the second cleavage site was determined to be RNNR ending at amino acid 1,043; this inference, which was based on the molecular weight of the LsVn subunits (67 and 111 kDa) as well as on the mass-spectrometry results, was further verified experimentally (see below and Fig 3C). Cleavage at this motif divided the 178-kDa large subunit into two medium-sized subunits with molecular weights of 67 and 111 kDa (Fig 3B).
Peptide-based antibodies were produced for specific recognition of the corresponding LsVn subunits. Antibodies Ab42K and Ab111K were produced based on peptides within the mass-spectrometry-confirmed regions of the 42- and 111-kDa LsVn subunits (Fig 3B). Because multiple RXXR motifs were distributed within the calculated 67-kDa LsVn subunit, two antibodies, Ab67K1 and Ab67K2, were produced to recognize different regions of this fragment (Fig 3B). Western blotting confirmed the predicted RXXR cleavage motifs. As expected, antibody Ab42K recognized the 42-kDa band, both Ab67K1 and Ab67K2 recognized the 67- as well as the 178-kDa bands, and Ab111K exhibited immunoreactive signals to both 111- and 178-kDa bands (Fig 3C). The small LsVn subunit had a molecular weight (42 kDa) lower than the calculated value of 47 kDa, which suggests that further processing might occur in the ovaries. Given that no additional RXXR motif was present within the 47-kDa polypeptide, how this protein processing occurs is unclear.
According to the observed LsVg cleavage pattern, three subunit-specific antibodies (Ab42K, Ab67K2 and Ab111K) were used in the subsequent experiments to analyze the molecular forms of LsVg present in different tissues.
IFA was performed using subunit-specific antibodies to detect subunit distribution. Similar to the results obtained with antibody Ab47Km (Fig 1B), the use of the Ab42K antibody resulted in immunoreactive signals in both the fat body and hemocytes (Fig 4A). By contrast, Ab67K2 and Ab111K produced immunoreactive signals only in hemocytes (Fig 4A). These results indicated that only the N-terminus of LsVg exists in the fat body. Western blotting was performed to determine the molecular weights and the subunit composition of proteins in the fat body extract and hemolymph. Using the Ab42K antibody, a 60-kDa band was detected in both the fat body extract and the hemolymph (Fig 4B). Antibodies Ab67K2 and Ab111K both recognized a strong band of approximately 200 kDa in the hemolymph (Fig 4B) as well as a similar but much weaker band in the fat-body extract (Fig 4B). As no full-length LsVg protein was recognized by any of the three antibodies, we hypothesized that the LsVg protein, when synthesized, undergoes rapid cleavage to yield two subunits with molecular sizes of 60 and 200 kDa. While both subunits exist inside the hemocytes, only the small 60-kDa subunit exists in the fat body. The weak 200-kDa band detected by western blotting with LsVg C-terminal antibodies (Ab67K2 and Ab111K) in fat-body extracts may represent unconsumed proteins, or contamination from the hemolymph attached to fat-body tissues. The detected 60 and 200 kDa bands were larger than their calculated sizes (47 and 178 kDa); this increased molecular size may be due to post-translational modifications such as glycosylation, lipidation, phosphorylation or sulfation [12, 13]. These results also suggested that LsVg was cleaved at a single site (RSRR) prior to protein invasion of the ovaries. We then performed coimmunoprecipitation assay to confirm the in-vivo physical interaction between RSV and the LsVg subunits. The anti-RSV antibodies coimmunoprecipitated the LsVg large subunit, but not the small one, from the female crude extracts, confirmed the physical interaction occurred between RSV and the large LsVg subunit (S1 Fig).
We then performed dual immunofluorescence assay to analyze the distributional pattern of the two LsVg subunits within hemocytes. Hemocytes were probed with two different antibodies: Ab111Km antibody followed by staining with the fluorescence dye Alexa 488, or with Ab42K antibody stained with Alexa 568. The two types of fluorescence signals were merged in all LsVg-expressing hemocytes (Fig 4C), indicating the co-existence of N- and C-terminal LsVg subunits within individual hemocytes. These results, combined with the protein sizes revealed by western blotting, indicated that LsVg exists in hemocytes as a complex of large and small subunits.
To determine whether the full-length mRNA transcript or a truncated version is expressed in the fat body, we designed three pairs of PCR primers to amplify the different regions of LsVg. qRT-PCR analysis indicated that the three mRNA fragments were expressed at similar abundances (Fig 4D), suggesting the production of the full-length transcript. We also used RNA interference (RNAi) with double-stranded RNA of the large LsVg subunit to knockdown gene expression, and then measured the expression level of the small subunit. Western blotting with antibodies Ab42K and Ab67K2 revealed dramatically decreased LsVg protein levels following gene knockdown (Fig 4E), indicating a single transcript containing both the small and large subunits. We also cloned the full-length LsVg mRNA from both the hemocytes and fat body of female SBPHs. Comparison of the LsVg cDNA sequences from these tissues did not detect any variable splicing.
Taken together, our results indicated that the LsVg protein is expressed in both hemocytes and the fat body of L. striatellus. After synthesis, the protein is cleaved at the RSRR motif, resulting in large and small subunits. The two-subunit complex exists in hemocytes, whereas only the small subunit remains in the fat body. How and why the large LsVg subunit is depleted in the fat body remains unclear.
In addition to nourishing developing embryos, insect hemocytes and hemolymph play key roles in innate immunity. We thus investigated whether LsVg is also expressed by hemocytes of nymphal or male L. striatellus. According to a qRT-PCR analysis, both the nymphs and males expressed LsVg, and hemocytes were the only LsVg-expressing tissue (Fig 5A). An IFA using the three subunit-specific antibodies (Ab42K, Ab67K2 and Ab111K) further confirmed the qRT-PCR results. All antibodies reacted to the corresponding subunits in hemocytes (Fig 5B). A dual immunofluorescence experiment with Ab42K and Ab111Km was used to reveal co-localization of the large and small subunits (Fig 5C). When expressed in hemocytes, the LsVg protein co-localized with RSV (Fig 5B). To exclude the possibility that LsVg bound to various pathogens in a non-specific manner, we microinjected Escherichia coli into the L. striatellus hemolymph and looked for the localization of the LsVg protein with the phagocytosed bacteria. The bacteria did not co-localize with LsVg inside hemocytes (Fig 5D), indicating that the interaction between RSV and LsVg was a specific event.
Western blotting was performed to determine the subunit composition of LsVg in the male hemocytes. All the three antibodies Ab47K, Ab67K2 and Ab111K recognized a strong band of more than 200 kDa in the hemolymph (Fig 5E), indicating that the LsVg in male remained in its full-length molecular form.
To determine the role of LsVg in mediating RSV survival and transmission within the insects, we generated LsVg-deficient L. striatellus nymphs using RNAi with dsLsVg via microinjection. At day 6 following the onset of feeding, LsVg expression levels within hemocytes analyzed by IFA using antibody Ab111Km exhibited a dramatic reduction (Fig 6A). We then assessed whether RSV transmission was affected by RNAi-mediated LsVg deficiency. The RSV titers in various L. striatellus tissues were assessed by qRT-PCR. Compared with the control group injected with dsGFP, LsVg dsRNA-treated L. striatellus showed similar RSV titers in the midgut and fat body, whereas significantly lower virus titers were observed in the hemolymph and salivary glands (Fig 6B). These results indicate that LsVg functions in facilitating RSV survival in, and transmission through the hostile hemolymph environment.
To confirm the function of LsVg in facilitating RSV survival in the hemolymph environment, equal volumes of LsVg or GFP dsRNA were delivered into the hemocoel of RSV-free L. striatellus nymphs via microinjection. Insects were allowed to feed on RSV-free rice seedlings. At 72 h following the microinjection, when the LsVg expression levels had been dramatically reduced by dsLsVg treatment (Fig 6C), purified RSV RNPs were directly delivered into the insect hemocoel. At 1, 6, 12 and 24 h following virus delivery, RSV titers in the L. striatellus body were assessed by qRT-PCR. Compared with the control group treated with GFP dsRNA, LsVg dsRNA-treated L. striatellus showed significantly lower virus titers beginning 6 h post-injection (Fig 6D). These results further confirmed that the hemocyte-produced LsVg plays a role in facilitating RSV survival in the hostile hemolymph environment.
Immuno-blocking experiment with anti-Vg antibodies revealed similar results. Inoculation of anti-Vg antibody into hemolymoh before RSV delivery decreased RSV survival in the hemolymph, especially at the early stages (Fig 6E, 6h and 12 h); however, the difference was not significant at the later stage (Fig 6E, 24 h). It might be that most of the RSVs were inside the hemocytes at the late stage, so immuno-blocking did not affect the function of the intracellular Vg.
Plant viruses have been reported to achieve vertical transmission within insect vectors via the transovarial transportation system of the insect Vg protein [5, 37]. Traditionally, insect Vg transported into the ovaries has been thought to be synthesized in the fat body. In the current study, we demonstrated that L. striatellus hemocytes also synthesize abundant Vg protein and that only hemocyte-produced Vg interacts with RSV in vivo. By clarifying the subunit composition of LsVn and by using LsVn subunit-specific antibodies, we revealed that LsVg is synthesized and proteolytically cleaved into the N-terminal small subunit and the C-terminal large subunit in both the fat body and hemocytes. The large subunit produced in the fat body is further consumed, with the large subunit that remains in hemocytes capable of interacting with the RSV CP. Moreover, we showed that LsVg is also expressed in male and nymphal SBPHs and revealed the gender-independent function of LsVg. LsVg expressed in the hemocytes of non-female SBPHs can also interact with RSV in vivo, thus protecting the virus from the hostile hemolymph environment and facilitating its systemic infection.
This study identified hemocytes, a major component of the insect immune system, as a new Vg-producing tissue. To our knowledge, this is the first study to demonstrate that an insect Vg gene is expressed by hemocytes. We also addressed the function of the hemocytes-produced Vg protein in virus recognition and transmission. The insect Vg gene has been previously reported to be expressed in tissues other than the female fat body. Although the Vg protein was not functionally investigated in most of these studies, its action has been physiologically elucidated in some cases. For example, Apis mellifera Vg is synthesized in the hypopharyngeal glands and the adjacent head fat-body cells of functionally sterile honeybee workers, implying that Vg is used for brood food production [38–40]. In Bombus hypocrita, the Vg gene is expressed at various levels in different castes, including the queen, workers and drones, from pupal to adult stages [41]. Camponotus festinatus Vg has been identified in both the queen and workers. The concentration of Vg is higher in queenless workers than in queenright ones, and Vg is synthesized at low concentrations before adult eclosion [42]. In addition to tissue-specific expression, tissue-specific processing of the Vg protein has been reported. For example, Leucophaea maderae Vg is expressed in both the male and female fat bodies. In this species, Vgs produced in the two sexes are similar in terms of their native molecular weights, but differ in their cleavage profiles and polypeptide compositions [14, 43]. At present, little is known about the relationship between the function of Vg and its biochemical and structural properties. By addressing the functional relationship determined by the LsVg subunit/domain composition, we have confirmed the possibility that proteolytically processed Vg with different domains can mediate specific functions. To our knowledge, our investigation is one of the few studies to have classified the function of Vg according to its subunit or domain composition.
The mature Vg contains an N-terminal domain (Vitellogenin_N [Vit_N)], a middle-region domain of unknown function (DUF1943), and a von Willebrand factor type D (vWD) C-terminal domain [5, 33, 44–46]. Domain Vit_N is required for interaction with the Vg receptor (VgR) [9, 47], while domains DUF1943 and vWD play roles in pathogen recognition [48]. An example in vertebrate animals is that of Oreochromis aureus Vg, which interacts with VgR via a polypeptide fragment located in the Vit_N domain [9]. Arthropoda, Macrobrachium rosenbergii Vg also interacts with its receptor via a specific β-sheet region in the Vit_N lipoprotein domain [47]. In scallops, both the recombinant DUF1943 and vWD domains of the Patinopecten yessoensis Vg protein can interact with the lipopolysaccharides and lipoteichoic acid expressed on the bacterial cell wall [45]. In insects, the L. striatellus Vg protein has been well studied by in vitro experiments that have revealed strong, weak and non-existent interactions between the RSV CP and vWD, DUF1943 and Vit_N domains, respectively [5]. In this study, we verified that the fat body-produced N-terminal Vg subunit is included in the Vit_N domain sequence. Because the L. striatellus fat body-Vg lacks any microbe-binding domains, it plays no role in mediating RSV transmission. This subunit contains both a signal peptide for secretion and a recognition site for receptor binding, however, and is thus expected to be secreted and then taken up by oocytes.
Our analysis revealed the existence of an N-terminal LsVg fragment—in contrast to the complete protein—in abdominal fat body tissue. Three lines of evidence demonstrate that the N-terminal LsVg fragment is processed from the full-length LsVg protein in the fat body rather than being synthesized from a different gene. Firstly, the polyserine tracts flanking the R-X-R/K-R motif have been demonstrated to be cleaved in almost all Vg homologs (see review [7]). Second, qRT-PCR analysis detected similar mRNA abundance levels between N- and C-termini of the Vg gene (Fig 4D). Third, and most importantly, dsRNA of the C-terminal large subunit dramatically interfered with mRNA levels of the N-terminal small subunit (Fig 4E). The significance of the existence of this Vg fragment in the fat body remains unclear; however, this type of Vg processing has also been reported in honeybees [23]. Havukainen et al. have presented a structural model for the N-terminal Vg fragment that includs a conserved β-barrel-like shape, with a lipophilic cavity and two insect-specific loops, thus indicating a capacity for lipid transport.
In general, arthropod hemolymph is hostile to pathogens because it contains both circulating hemocytes and antimicrobial proteins [49–51]. Viruses acquired via the insect midgut can accordingly be transported to other tissues only after successful escape from the hostile hemolymph environment (e.g., through survival in plasma and evasion of hemocyte phagocytosis). Studies have revealed, however, that a successfully transmitted microbe can protect itself against the host immune system by exploiting host hemolymph proteins, thereby transforming the host hemolymph into a relative benign environment. Examples include Potato leafroll virus transmission by the green peach aphid Myzus persicae and Tomato yellow leaf curl virus transmission by the sweetpotato whitefly Bemisia tabaci. In these cases, the viruses gain protection and achieve successful transmission by binding to GroEL, a hemolymph protein produced by endosymbiotic bacteria of the insect [52–54]. During the transmission of West Nile Virus by Aedes aegypti, viruses can even establish infection in hemocytes by binding to the mosquito hemolymph protein that leads to the phagocytosis pathway [55]. In previous studies, Vg from fish has been revealed to have immunologic functions. By binding to the pathogen, the Vg protein can neutralize/kill the pathogen, either directly or in an indirectly fasion as opsonins mediate macrophage phagocytosis [26, 27, 33]. However, whether recognition by Vg of an arthropod vector would result in virus tolerance is not clear. In this study, the L. striatellus hemolymph was revealed to be a relatively benign environment for RSV. RSV tends to be cleared from the SBPH hemolymph slowly. When E. coli was delivered into the L. striatellus hemolymph through microinjection, the maximum level of phagocytosis was reached at 1.5–2 h post injection, and the bacterial numbers in the hemolymph (including in hemocytes) were dramatically decreased to <20% within 24 h (S2 Fig). When RSV suspension was delivered into the hemolymph, by contrast, maximum phagocytosis was reached at about 20 h post-injection (S2 Fig), and 50% of the virus was retained in the hemolymph (including the hemocytes) after 24 h (Fig 6D). By uncovering a positive correlation between the Vg presence and the RSV in-hemolymph survival (Fig 6), this study has provided clues regarding the contribution of the RSV-LsVg interaction to the protection of RSV in the hemolymph and the development of a benign L. striatellus hemolymph environment during RSV transmission. This vector-Vg-virus interaction may be a common molecular mechanism to facilitate the passage of the virus through the vector hemolymph.
In summary, we have revealed that an insect Vg protein undergoes tissue-specific processing, with the molecular form produced specifically by the hemocytes used by the virus to aid its survival in the hemolymph, thus facilitating virus transmission. Detailed analyses of the molecular interactions between the virus and its insect vector are required for the exploitation of novel virus control strategies that target specific points in the virus life cycle and interfere with virus transmission.
RSV-free and RSV-infected L. striatellus individuals used in this study were originally captured in Jiangsu Province, China, and were maintained in our laboratory. All plants used for L. striatellus rearing were grown inside a growth incubator at 25°C under a 16-h light/8-h dark photoperiod. To ensure a high offspring infection rate, viruliferous female imagoes were cultured separately, 15% of their corresponding offspring were tested for RSV infection through a dot-enzyme-linked immunosorbent assay using RSV-specific monoclonal antibodies provided by Dr Xueping Zhou (Institute of Biotechnology, Zhejiang University [56]). The RSV antibody was produced using the RSV RNPs as antigen. This virion-specific monoclonal antibody was used in all the in vivo experiments that were performed to determine RSV localization or co-localization with LsVg. An insect population with a predicted infection rate of 100% was used in the experiments.
For hemolymph isolation, the SBPHs were anesthetized at -20 centigrade for 3 min, and then the forelegs were severed at the coxa-trochanter joint by forceps. The hemolymph was expelled and drawn to the tip of clean forceps. Only clear droplets were collected to avoid contamination by fat body [55]. In droplets contaminated by fat body, transparent oil drops can be seen. The SBPHs were then dissected in pre-chilled PBS buffer. Insects were dissected from the abdomen, and the wound was rinsed gently two times with PBS buffer. Because the insect fat body compose of a meshwork of loose lobes suspended in the hemocoel and bathed in the insect hemolymph, it is difficult to dissect it out in its entirety and without contamination of hemolymph or other tissues. We collect most fat body without contamination from other tissues, and placed it in PBS buffer. Tissues including the midgut, salivary glands, ovaries of the female and testes of the male were washed twice in PBS to remove contaminating viruses or proteins from the hemolymph.
For quantitative analysis of LsVg expression in SBPH tissues, we dissected female, male or third-instar nymphal L. striatellus and collected the tissues according to the protocol described above. RNA was extracted from the tissues of individual insects. Reverse transcriptional PCR and SYBR-Green-based qPCR were performed according to the protocols provided by the manufacturer. Primer pairs used to amplify LsVg and LsVn / LsVg subunits were LsVg-QF / LsVg-QR, 47K-QF / 47K-QR, 67K-QF / 67K-QR and 111K-QF / 111K-QR (S1 Table). Viral RNA copies were measured by qRT-PCR using primer sequences pc3-F and pc3-R (S1 Table), which were designed and synthesized according to the nucleocapsid protein (Pc3 or CP) gene sequence (DQ333944). L. striatellus elongation factor 2 (ef2) was amplified as an internal control for the loading of cDNA isolated from different samples. Primers used for ef2 amplification were ef2-QF / ef2-QR (S1 Table). Water was used as a negative control.
Mouse anti-Vg monoclonal antibodies against Vg peptides RNQQKTKSRSRRS and RMQPLNKEEKQNVF were prepared by Abmart (Shanghai, China) as previously described [5], and were designated as Ab47Km and Ab111Km in this study. To prepare LsVn subunit-specific antibodies, the LsVn subunit-specific peptides KSRRNILPQSDSNQ, AQVDSDTKHMR, YKNPGEAPELR and RMQPLNKEEKQNVF were conjugated to mcKLH and injected into rabbits, and the corresponding antiserums were prepared by GenScript (Nanjing, China). The antibodies produced were designated as Ab42K, Ab67K1, Ab67K2 and Ab111K respectively.
Insects tissues were placed in PBS on silylated glass slides (Sigma cat. no. S4651; St. Louis, MO, USA) and allowed to dry. Tissues were then fixed in 4% paraformaldehyde at room temperature for 30 min. The slides were rinsed twice with PBS and then incubated in PBST/FBS (PBS containing 2% Tween 20 and 2% fetal bovine serum) for 30 min. To detect LsVg localization in different tissues, the slides were incubated with mouse anti-Vg monoclonal antibody Ab47Km (1:300 dilution in PBST/FBS) for 1 h and then Alexa 568-labeled goat anti-mouse antibody (1:200 dilution in PBST/FBS) for 1 h. The slides were rinsed three times with PBST, and the nucleoli were stained with TO-PRO-3 iodide (Invitrogen cat. no. T3605; Carlsbad, CA, USA) at room temperature for 3 min. The samples were examined using a Leica TCS SP8 confocal microscope.
To detect co-localization of LsVg with RSV in the SBPH tissues, slides were prepared according to the protocol described above. The anti-RSV monoclonal antibody was labeled with Alexa Fluor 488 according to the Alexa Fluor 488 Monoclonal Antibody Labeling kit (Invitrogen) instructions. The slides were sequentially incubated for 1 h each with antibody Ab47Km (1:300 diluted in PBST/FBS), Alexa 568-labeled goat anti-mouse antibody (1:200 diluted in PBST/FBS) and with Alexa 488-labeled anti-RSV monoclonal antibody and then stained with TO-PRO-3 iodide for 3 min.
To detect co-localization of LsVn subunits with RSV, slides were prepared as described above. The slides were incubated with mouse anti-RSV and rabbit anti-LsVn antibodies (Ab42K, Ab67K2 or Ab111K; 1:1000 dilution in PBST/FBS), followed by Alexa 488-labeled goat anti-mouse and Alexa 568-labeled goat anti-rabbit antibodies (1:200 dilution in PBST/FBS), and finally with TO-PRO-3 iodide for nucleolus staining. To detect co-localization of LsVg subunits in hemocytes, Ab42K and Ab111Km were used as the primary antibodies, and Alexa 568-labeled goat anti-rabbit and Alexa 488-labeled goat anti-mouse antibodies were used as the secondary antibodies. To detect co-localization of LsVg with E. coli in hemocytes, the GFP-expressing bacteria were suspended in sterile water at an OD600 of 1.0. Subsequently, 13.8 nl of the bacterial suspension was delivered into the hemocoel of third-instar nymphs; 1.5 h after microinjection, the insects were dissected and the hemolymph was collected. Slides for confocal microscopy were prepared as described above. The primary antibody for LsVg detection was Ab42K and staining was performed with Alexa 568.
Newly emerged female SBPHs were allowed to grow for 3 days before extraction of Vn. Two grams of the insects were ground in liquid nitrogen into a fine powder and incubated in 1 ml of 0.4 M NaCl solution for 20 min at 4°C. The suspension was centrifuged at 3,300×g for 10 min at 4°C to remove insect debris. The supernatant was centrifuged three times at 1,000×g for 10 min at 4°C to remove lipid on the surface of the supernatant. The sample was then treated three times as follows: after addition of 8 ml of ddH2O, the mixture was incubated overnight at 4°C followed by centrifugation (1,000×g, 20 min, 4°C) to precipitate the Vn protein. The precipitated protein was dissolved in 1 ml of 0.4 M NaCl solution and centrifuged again (3,300×g, 10 min, 4°C) to remove any undissolved precipitate. The final supernatant was applied to a size-exclusion Superdex 200 10/300 GL column (GE Healthcare, Piscataway, NJ, USA), and the fractions containing Vn of the highest concentration and purity were collected.
For mass spectroscopy analysis, the purified Vn protein was separated on a 10% SDS-PAGE gel (Bio-Rad Laboratories, Hercules, CA, USA) and stained with Coomassie Blue (Bio-Rad). The bands corresponding to the 178-, 111-, 67- and 42-kDa Vn subunits were excised and digested, and the peptides were subjected for liquid chromatography-tandem mass spectrometry analysis. Peptides were identified using an LTQ-Orbitrap XL with Easy nLC-1000 (Thermo Fisher Scientific), and proteomics data were analyzed using Proteome Discoverer 1.4 (Thermo Fisher Scientific).
To confirm the subunit composition of LsVn, the purified Vn protein was fractionated on a 10% SDS-PAGE gel (Bio-Rad) and processed for immunoblotting. The LsVn/LsVg subunit-specific antibodies Ab42K, Ab67K1, Ab67K2 or Ab111K (1:10,000 dilution in PBST/FBS) was used to probe the corresponding LsVn subunits. The bound antibodies were detected by using horseradish peroxidase-conjugated goat anti-rabbit secondary antibodies (Sigma), and the blots were developed using the enhanced chemiluminescence Western Blotting Detection System (GE Healthcare).
Western blotting was performed to measure the subunit composition and molecular sizes of LsVg in the fat body and hemocytes. Both fat body and hemolymph were dissected from the female insects (3 days after molting). Tissues from 50 insects were placed in 100 μl of PBS buffer and boiled in SDS-PAGE loading buffer. When the same amounts of total proteins were loaded to the gel, we found that LsVg was cleaved into two subunits, one small subunit of 60 kDa and one large of 200 kDa; however, the concentration of the two subunits in different tissues were very different (S3 Fig). It was difficult to determine whether the large subunit existed in the fat body or whether the small subunit was secreted from the fat body. We then adjusted the western blotting loading with same amount of the LsVg small subunit as a control, and compared the amounts of the large subunit to determine subunit distribution. We performed western blotting with Ab47Km to determine the amounts of the small subunit in fat body and hemolymoh. Then fat body and hemolymph protein samples with the same amount of the small subunit were applied to the SDS-PAGE, and probed with antibodies Ab42K, Ab67K2 or Ab111K.
Two DNA fragments, one specific to the coding sequences of the LsVg N-terminal small subunit and the other to the C-terminal large subunit, were PCR amplified and designated as VgN and VgC, respectively. Primer pairs used for the amplification of the VgN and VgC fragments were VgN-si-F/VgN-si-R and VgC-si-F/VgC-si-R, respectively (S1 Table). DsRNA was synthesized using a commercial kit (Ambion) and purified by phenol:chloroform extraction and isopropanol precipitation. Finally, 36.8 nl of dsRNA at 1 ng/nl was delivered into the insect hemocoel for gene silencing. GFP dsRNA, which was used as a negative control, was synthesized and microinjected following the same protocol.
To confirm that both LsVg subunits were expressed from the same transcript, dsRNA of VgC was delivered into the hemocoel of the fifth-instar nymphal L. striatellus individuals. The insects were cultured in new chambers with healthy rice seedlings until emergence of adults. Females were transferred to new chambers for an additional 48 h of culture. Then the insects were dissected, and the fat body protein extracts were prepared for Vg expression analysis. Western blotting was performed according to the procedure described above. The antibodies used for the detection of the fat body-expressing LsVg subunit were Ab42K and Ab67K2.
To determine the influence of LsVg on RSV horizontal transmission, dsRNA of VgC was delivered into the hemocoel of RSV-infected third-instar nymphs of L. striatellus. On day 6 of culture in new chambers, a subset of the insects was collected and the expression of Vg in their hemocytes was measured by confocal microscopy. The remaining insects were collected, and the RSV titers in various tissues were measured. Tissues, including the hemocytes, midgut, salivary glands and fat body, were collected as described above. Virus titers were determined by qRT-PCR using primer pair pc3-F / pc3-R (S1 Table).
To determine the effect of LsVg on RSV survival inside the hemolymph, dsRNA of VgC was delivered into the hemocoel of RSV-free third-instar nymphal L. striatellus individuals. After 72 h of culture in new chambers, some of the insects were collected, and the expression of Vg in their hemocytes was measured by confocal microscopy. Following successful knockdown of Vg expression, purified virus RNPs in PBS buffer were microinjected into the insect hemocoel. At 1, 6, 12 or 24 h after RSV microinjection, RNA was extracted from the whole insect body and virus titers were measured according to the protocol described above.
All graphing and statistical analyses were performed using Prism 6.0 software (GraphPad Software, CA, USA). Data were expressed as means ± standard deviation (SD). The significance of differences between groups was evaluated using Student’s t-test.
|
10.1371/journal.pgen.1007944 | Heme peroxidase HPX-2 protects Caenorhabditis elegans from pathogens | Heme-containing peroxidases are important components of innate immunity. Many of them functionally associate with NADPH oxidase (NOX)/dual oxidase (DUOX) enzymes by using the hydrogen peroxide they generate in downstream reactions. Caenorhabditis elegans encodes for several heme peroxidases, and in a previous study we identified the ShkT-containing peroxidase, SKPO-1, as necessary for pathogen resistance. Here, we demonstrated that another peroxidase, HPX-2 (Heme-PeroXidase 2), is required for resistance against some, but not all pathogens. Tissue specific RNA interference (RNAi) revealed that HPX-2 functionally localizes to the hypodermis of the worm. In congruence with this observation, hpx-2 mutant animals possessed a weaker cuticle structure, indicated by higher permeability to a DNA dye, but exhibited no obvious morphological defects. In addition, fluorescent labeling of HPX-2 revealed its expression in the pharynx, an organ in which BLI-3 is also present. Interestingly, loss of HPX-2 increased intestinal colonization of E. faecalis, suggesting its role in the pharynx may limit intestinal colonization. Moreover, disruption of a catalytic residue in the peroxidase domain of HPX-2 resulted in decreased survival on E. faecalis, indicating its peroxidase activity is required for pathogen resistance. Finally, RNA-seq analysis of an hpx-2 mutant revealed changes in genes encoding for cuticle structural components under the non-pathogenic conditions. Under pathogenic conditions, genes involved in infection response were differentially regulated to a greater degree, likely due to increased microbial burden. In conclusion, the characterization of the heme-peroxidase, HPX-2, revealed that it contributes to C. elegans pathogen resistance through a role in generating cuticle material in the hypodermis and pharynx.
| Reactive oxygen species (ROS) production by the host tissues is one of the first lines of defense when microbial infection occurs. ROS has been shown to be involved in multiple protective pathways in innate immunity. However, given the complexity of mammalian systems, the exact manner in which ROS are used for host defense remains incompletely understood. In this study, we use Caenorhabditis elegans as a simplified model system to decipher the protective functions of ROS in innate immunity. We describe a peroxidase, HPX-2, that protects C. elegans from multiple infectious microbes by strengthening barrier tissue. This finding brings insight into the mechanisms by which peroxidases utilizes ROS to contribute to innate immunity. With infectious diseases being one of the most important causes of morbidity and mortality around the world, understanding ROS production and its function in pathogen resistance will provide us with important information in developing new therapies against pathogens.
| When exposed to pathogen, production of Reactive Oxygen Species (ROS) is one of the first innate immune responses initiated by the host. ROS can exert immune-protective effects by acting directly as an antimicrobial, serving as a signal to activate downstream responses, and/or contributing to the generation of physical barriers (Reviewed by [1, 2]). There are two major groups of enzymes involved in this process–the ROS producing NADPH oxidases/Dual-oxidases (NOX/DUOX) and the ROS utilizing, heme-containing peroxidases (Reviewed by [1, 2]). In many cases, these enzymes are functionally, and sometimes physically linked to contribute to innate immunity against pathogen. The most canonical case studied in mammalian systems is that of myeloperoxidase (MPO), which resides in the phagolysosome of macrophages and neutrophils and utilizes the superoxide produced by NOX2 to generate the powerful oxidant HOCl during the oxidative burst that occurs as a result of microbial engulfment (Reviewed by [3]). Additionally, it has been shown that lactoperoxidase (LPO), coupled with DUOX2, is responsible for the generation of the antibiotic oxidant hypothiocyanite (OSCN-) from H2O2 and thiocyanate (SCN-) on mucosal surfaces [4–6]. These responses are not limited to animals; plants also employ peroxidases and NADPH oxidases as part of their antimicrobial arsenal (Reviewed in [7]).
In C. elegans, the one functional NOX/DUOX encoded by the genome is BLI-3, a dual oxidase that produces H2O2 when animals are exposed to pathogen [8–11]. Like all dual oxidases, BLI-3 consists of an NADPH oxidase domain and a heme peroxidase domain [10]. The production of H2O2 from the NADPH oxidase domain of BLI-3 is required for pathogen resistance and point mutants that affect this domain significantly increase pathogen susceptibility [8, 12]. In addition to its role during infection, BLI-3 is crucial for cuticle development. The peroxidase domain uses the H2O2 produced by the NADPH oxidase domain to cross-link collagen proteins [10]. Another heme-binding peroxidase, MLT-7, additionally contributes to this process [13]. Mutations in the peroxidase domains of BLI-3 and MLT-7 that impair activity result in a blistered phenotype. However, these mutations do not affect pathogen susceptibility [14].
Since peroxidases in other systems do play a role in innate immunity, we hypothesized that one or more of the peroxidases encoded in the C. elegans genome would affect pathogen resistance. Indeed, previous studies in our lab showed that a ShkT domain containing peroxidase, which we named SKPO-1, is required for resistance to E. faecalis [14]. A skpo-1 mutant was more susceptible to E. faecalis compared to the N2 reference strain. In addition to pathogen sensitivity, the skpo-1 mutant also had an incomplete penetrance of dumpy phenotype, suggesting a potential role in the process of cuticle bio-generation. Interestingly, SKPO-1 localized to the hypodermis of the worm, a tissue where BLI-3 also is present, indicating a potential physical interaction between the two [14].
Here, we characterized the immune-protective peroxidase, F09F3.5, which we named HPX-2, for Heme Peroxidase 2. Using a combination of RNA interference (RNAi) and CRISPR-Cas9 mediated mutation, we demonstrated that HPX-2 is important for protecting C. elegans against some, but not all, pathogens. Through fluorescent-tagging and tissue-specific RNAi, we demonstrated that HPX-2 is present in the pharynx and the hypodermis of the worms. Despite low expression of its gene at the organismal level, HPX-2 functions in these two tissues and contributes to cuticle integrity; hpx-2 mutants exhibited higher intestinal colonization by pathogens with thick cell walls and possessed weaker cuticles, more penetrant to dye. We also examined animals with a point mutation at the catalytic site of HPX-2 peroxidase domain and discovered that peroxidase activity is partly required for the protective function of HPX-2. Lastly, a transcriptomic analysis of the hpx-2 mutant revealed an altered immune response when exposed to E. faecalis. In summary, the results of these experiments demonstrated that HPX-2 is an immuno-protective peroxidase likely functioning in the development of cuticle barrier tissue associated with the hypodermis and pharynx.
Using RNAi, knock-down of hpx-2 resulted in increased sensitivity of C. elegans to E. faecalis infection (S1 Fig), as previously reported [14]. To further study the function of HPX-2, we generated a truncated mutant allele (hpx-2(dg047)) by CRISPR-Cas9 mediated gene editing. hpx-2(dg047) is missing most of the predicted peroxidase domain and is a predicted loss-of-function mutant. We additionally obtained a nonsense mutant allele, hpx-2(gk252521), from the Caenorhabditis Genetics Center that is predicted to produce a protein lacking the entire peroxidase domain (Fig 1A). The mutants were tested for general defects in fitness by measuring their longevity on live (S2A Fig) and heat-killed E. coli OP50 (S2B Fig). While both mutants showed slight longevity defects compared with the wild type reference strain, N2, on live E. coli, there was no significant difference on heat-killed E. coli. The minor longevity defect of the mutants on live E. coli OP50 might result from the slight pathogenicity of OP50 towards worms as they age [15]. Overall, these data suggest that the loss of HPX-2 does not dramatically affect the fitness of the worms.
We next tested the susceptibility of both hpx-2 mutants to E. faecalis. In agreement with the RNAi experiment, both mutants showed significantly decreased survival on this species of bacteria compared to N2 (Fig 1B). Furthermore, the difference disappeared when worms were exposed to E. faecalis inactivated by vancomycin treatment, suggesting live bacteria were required for killing, and again indicating that the animals do not have a general fitness defect (Fig 1C). To test if the pathogen susceptibility of the hpx-2 mutants is pathogen-specific, we exposed them to multiple human pathogens including the fungal pathogen Candida albicans SC5314 (Fig 1D), Gram-positive pathogens Staphylococcus aureus NCTC8325 (Fig 1E) and Corynebacterium diphtheriae NCTC12129 (Fig 1F), and finally Gram-negative pathogens Pseudomonas aeruginosa PA14 (Fig 1G), Escherichia coli O157:H7 Sakai (EHEC; Fig 1H), and Salmonella enterica SL1344 (Fig 1I). Interestingly, hpx-2 mutants showed susceptibility to all the microbial pathogens tested except for the Gram-negative bacteria, raising the possibility that differences in microbial cell wall structure might impact the phenotype observed with loss of hpx-2 (see Discussion).
To verify that HPX-2 is required for pathogen resistance, we complemented the mutant strains with DNA containing the hpx-2 gene and subjected the animals to E. faecalis exposure. When exposed to E. faecalis, the complemented strains showed a significant increase in pathogen tolerance compared to the parental strains, reaching a level of resistance similar to N2 (Fig 2A and 2B). These data collectively suggest that HPX-2 is required for resistance to many, but not all, pathogens capable of infecting C. elegans.
Previous studies showed that the only functional NOX/DUOX in C. elegans, BLI-3, is present in the pharynx, the hypodermis and the intestine of the worms [10, 12]. We postulated that heme peroxidases that might functionally interact with BLI-3 are likely to be in one or more of the tissues that harbor BLI-3. Indeed, we previously demonstrated that SKPO-1, another immune-protective peroxidase, localizes to the hypodermis [16]. To functionally localize HPX-2, we conducted tissue-specific RNAi to knock down hpx-2 in the intestine or hypodermis and measured the worms’ susceptibility to E. faecalis. We observed that intestinal RNAi did not affect pathogen susceptibility, whereas hypodermal RNAi resulted in a slight but significant decrease, suggesting that HPX-2 might be present in the hypodermis and contribute to pathogen resistance in this tissue (Fig 3A and 3B).
To directly visualize the expression of HPX-2, we created a partial translational fusion of HPX-2 to GFP. Specifically, the upstream sequence and the first two exons of the hpx-2 gene were cloned to generate a fusion with GFP, and the construct was injected into the worms (Fig 3C). The resulting extrachromosomal array was then integrated into the chromosome. We initially looked at the stable, transgenic worms at different development stages, but did not observe GFP expression in the majority. Occasionally (in about ~1% of the worms), we observed a green fluorescent stripe extending from the distal bulb of the pharynx to the anterior of the buccal cavity (Fig 3D and 3E). We postulated that the expression level might be too low to detect. In fact, expression data from Wormbase and our RNA-seq data (see below) indicate that the overall expression level of hpx-2 gene is extremely low at the whole worm level but increases significantly upon entry into dauer. When we subsequently analyzed dauer animals using confocal microscopy, we observed weak expression of hpx-2::GFP in the distal bulb of the pharynx in nearly all animals (Fig 3F). At higher magnification, the localization pattern was most consistent with expression in the gland cells (Fig 3G and 3H) [17]. Again, around 1% of the worms showed the striped expression pattern, resembling the structure called the “process” that extends from the dorsal g1 gland cell to the anterior of the buccal cavity (Fig 3E). Interestingly, the process functions in transporting excreted material from the gland cell to the pharyngeal lumen [18, 19]. Overall, these results indicate that hpx-2 is expressed in the hypodermis and pharynx.
A functional role in cuticle biogenesis was demonstrated for some of the previously studied heme peroxidases and loss-of-function mutations sometimes caused abnormal cuticle morphology. For example, MLT-7, was shown to function in conjunction with BLI-3 to cross-link the cuticular collagens during the molting process. Loss of MLT-7 resulted in a blistered phenotype due to incomplete cross-linking of these extracellular matrix proteins [13]. Another immune protective peroxidase previously studied by our lab, SKPO-1, has an incomplete penetrance of dumpy phenotype when mutated, again suggesting a role in cuticle generation [14].
Given the pathogen sensitivity phenotype that resulted from hypodermal-specific RNAi (Fig 3B), we postulated that HPX-2 might play a role in the generation and/or structural integrity of the cuticle. However, unlike the mlt-7 or skpo-1 mutants, the hpx-2 animals displayed normal morphology as observed under the dissecting microscope (S3A Fig). To test for more subtle defects, cuticle integrity was measured by testing how permeable the animals were to a DNA dye [20]. Specifically, we exposed the worms to the DNA staining agent Hoechst 33258 and scored how many exhibited nuclear staining of underlying tissue (Fig 4A). While about 15% of N2 animals showed evidence of nuclear staining, twice as many of the HPX-2 mutants displayed this phenotype, indicative of some loss of cuticle integrity (Fig 4B). The staining was observed at higher resolution to better classify which tissues were affected. Along the length of the worm, we sometimes observed enhanced staining of the hypodermal cell nuclei that underlie the cuticle on the outside of the worm’s body (Fig 4C). We counted animals that were specifically positive for this staining pattern and found that less than 1% of N2 animals were positive for dye penetrance along the body cuticle, but 2–5% of the hpx-2 mutants were, depending on the allele (Fig 4D). Staining of the hypodermal cell nuclei also occurs in the head region, but at higher resolution additional staining of the pharyngeal muscle tissue was observed, indicative of penetration of the cuticle that lines the pharyngeal lumen (Fig 4E). Approximately twice as much staining was apparent in the hpx-2 mutants compared to N2 (Fig 4F). These data are consistent with a weaker, more penetrant cuticle in both the hypodermal and pharyngeal cuticle.
Even though the loss of HPX-2 did not affect the low-resolution morphology of the worm, we observed that some of the strains generated by injection of higher amounts of the hpx-2 complementation construct displayed a slightly dumpy or partial roller phenotype. These animals were also more sensitive to E. faecalis than wild-type or hpx-2 mutant animals (S4A and S4B Fig). We hypothesized that overexpression of hpx-2 might be contributing to these phenotypes. The level of hpx-2 expression in the different strain backgrounds was measured, showing a correlation between higher mRNA levels and hyper-sensitivity to E. faecalis (S4C Fig). The results indicate that too much as well as too little HPX-2 cause cuticle defects and sensitivity to the pathogen.
Next, the functional relevance of the pharyngeal presence of HPX-2, was further examined. As mentioned, the apical surface of the pharyngeal lumen is lined with cuticle, which connects to the cuticle of the hypodermis. Specialized projections of the pharyngeal cuticle can form structures, such as the teeth-like formations found in the terminal bulb (grinder) that are thought to contribute to mechanical microbial disruption [18]. HPX-2 was found to be expressed in the gland cells of the pharynx (Fig 3D–3G), and the pharyngeal cuticle was more penetrant to Hoescht dye in the hpx-2 mutants (Fig 4E and 4F). The data is consistent with HPX-2 aiding cuticle development/remodeling of this organ following excretion into the pharyngeal lumen. We hypothesized that loss of hpx-2 weakens the pharynx, resulting in less disruption of ingested bacteria and increased intestinal colonization by pathogens. To test this hypothesis, we exposed L4 worms to a strain of E. faecalis OG1RF that constitutively expresses GFP and measured the level of intestinal colonization by both fluorescence microscopy and CFU plating. At Day 2 of infection, we observed a higher level of intestinal colonization in the hpx-2 mutant strains, indicated by higher fluorescent intensity and CFU counts (Fig 5A and 5B). In contrast, when we exposed worms to a P. aeruginosa PA14 strain expressing dsRed [21], we did not observe any significant differences between the N2 strain and the mutants in either fluorescent intensity or CFUs (Fig 5C and 5D), which is consistent with the observation that N2 and hpx-2 mutants are equally susceptible to P. aeruginosa (Fig 1G). Although the hpx-2 mutant might have impaired grinding capability, no obvious structural abnormalities of the pharynx were observed when examined by light and high-resolution transmission electron microscopy (TEM) (S5 Fig). TEM also show that the teeth-like structures found in the grinder are normally shaped (S5B Fig). In addition, the pumping rate and brood size of the worms were not significantly affected (S3B and S3C Fig). Altogether, these results suggest that HPX-2 functions in the pharynx to reduce intestinal colonization by some pathogens, possibility through structural reinforcement, hardening, of the pharyngeal cuticle, and not by changing morphological features such as the overall shape of the grinder and the grinder teeth.
Sequence comparison of HPX-2 to other peroxidases indicates that HPX-2 possesses all the conserved residues required for peroxidase catalytic activity, including the proximal histidine (H240), the catalytic Arginine (R372), and the distal histidine (H476) (reviewed by [22]). On the other hand, similar to BLI-3 and MLT-7, HPX-2 lacks the covalent heme-binding residues that are conserved in mammalian peroxidases [22]. However, this does not rule out the possibility of non-covalent heme binding since both human DUOX1 and 2 lack the conserved residues but still bind heme weakly [23, 24]. Thus, we hypothesize HPX-2 has peroxidase activity.
In previous work, a skpo-1 mutant was characterized as releasing more H2O2 than N2 when exposed to the pathogen E. faecalis, presumably because less H2O2 was being utilized [14]. In contrast, we found H2O2 levels generated by the hpx-2 (dg047) mutant were comparable to N2 following exposure to E. faecalis (S6 Fig). While this could be because HPX-2 possesses little peroxidase activity, it also could be due to its low expression level or a lack of activity under these conditions.
To further examine the possible contribution of HPX-2’s peroxidase activity to pathogen resistance, we analyzed a strain harboring a single amino acid mutation that changes an active site residue of the peroxidase domain. Specifically, the catalytic arginine residue was substituted to alanine using CRISPR-Cas9 mediated mutagenesis (S7 Fig). The catalytic arginine is conserved throughout the peroxidase–cyclooxygenase superfamily, and the substitution of the arginine to alanine abolishes peroxidase activity [25, 26]. The point mutant exhibited increased sensitivity to E. faecalis, however, the phenotype was not as strong as the deletion mutant (Fig 6B). These data suggest that the catalytic activity of the peroxidase active site of HPX-2 contributes to the pathogen resistance phenotype.
To better understand how the loss of HPX-2 affected the global gene profile of the worm, we conducted a transcriptome analysis using RNA sequencing. Specifically, we first examined the differences in gene expression of a hpx-2 mutant (hpx-2 (dg047)) compared to the N2 reference strain under both non-pathogenic (exposed to E. coli OP50) and pathogenic (exposed to E. faecalis OG1RF) conditions. Genes that were significantly altered in hpx-2 animals are plotted in Fig 7A. Under the non-pathogenic condition (exposed to E. coli OP50), there were a total of 69 genes that were significantly differentially expressed in the mutant, with 34 genes up-regulated and 35 down-regulated (S1 Table). 21 of the affected genes encode for structural constituents of cuticle, including 17 col genes, 3 dpy genes (dpy-3, dpy-4, and dpy-5), and 2 rol genes (rol-1 and rol-6). All of them are upregulated, suggesting a perturbed cuticle biogeneration process upon the loss of HPX-2 (S1 Table). We next analyzed gene expression in a hpx-2 mutant after 16 hours of exposure to E. faecalis OG1RF. We chose this early time point because the worms show limited damage and thus the indirect effects should be smaller than later during infection. Under these conditions, there were a total of 125 genes that were differentially expressed in the mutant (72 up-regulated and 53 down-regulated) with 106 of them unique to the pathogenic condition (S2 Table). Interestingly, genes involved in cuticle generation were again significantly altered. Compared to the 21 genes under the nonpathogenic conditions, a total of 41 cuticle genes were affected and up-regulated. In addition to the increased changes in cuticle structural genes, there was an enrichment of genes encoding proteins involved in defense response to Gram-positive bacteria (Fig 7B). These include 4 out of 10 lysozyme genes (lys-1, lys-2, lys-3, and lys-7), and f53a9.8, which is involved in defense response against Gram-positive bacteria and expressed in the intestine [27–29].
Although the transcriptome changes we detected were modest, both in the number of genes affected and the amplitude of the effect, several observations suggest these reflect the biological effect of HPX-2. First, hpx-2 gene expression is restricted to a few cells, and thus we would not expect large transcriptome effects that are averaged over the whole animal. Second, we performed five biological replicates of RNAseq and the differences were highly reproducible. Third, the affected genes are enriched for cuticle biogenesis, which correlates well with the known requirement for peroxidases in cuticle cross-linking and the increased permeability of the cuticle to dye that was observed (Fig 4). Fourth, although the overlap in genes that were significantly affected in the hpx-2 mutant under the two conditions is modest, in most cases genes that were significantly affected in one direction under the E. coli condition were affected in the same direction under the E. faecalis condition and vice versa, even if the effect reached significance only in one comparison (Fig 7C). Finally, we used qRT-PCR to validate ten genes that were changed in the hpx-2 mutant compared to N2 under pathogenic conditions. These qRT-PCR experiments were performed on RNA isolated from three biological replicates that were independent of the RNAseq RNA samples, yet there was a strong correlation between the results (Fig 7C; R2 = 0.92 between fold change by RNAseq and qRT-PCR). Therefore, we conclude that there are modest but significant effects on the transcriptome that implicate HPX-2 in cuticle biogenesis and defense against Gram-positive bacteria.
Next, we examined the gene profiles of animals exposed to E. faecalis as compared to those exposed to OP50 (Fig 7D, S3 and S4 Tables). In N2, a total of 4217 genes (2560 up-regulated and 1657 down-regulated) were significantly differentially expressed, many of which were consistent with previously published data (S3 Table) [30]. In the hpx-2 mutant, however, a larger number of genes were significantly differentially expressed: 8313 genes in total with 4255 of them up-regulated and 4058 down-regulated (S4 Table). To examine the relationship between the two sets of genes, genes that were significantly expressed under at least one condition were plotted (Fig 7D). Compared to the N2 strain, the hpx-2 mutant had similar, but stronger responses to the pathogen, contributing to the larger number of genes with significant changes in expression under the pathogenic condition. In summary, we conclude that the loss of hpx-2 results in a similar, but stronger response to infection. We speculate that the greater microbial load of E. faecalis observed in the hpx-2 mutant (Fig 5A) caused this observation of a more robust transcriptional response. In other words, the hpx-2 animals are essentially farther along in the infectious process than the N2 animals when they were assayed because they controlled the microbial load more poorly.
In this study, we demonstrate that the heme-containing peroxidase HPX-2 plays an important role in attenuating infection by a variety of pathogens. We postulate that HPX-2 functions in two tissues, the pharynx and the hypodermis, to protect worms from infection as modeled in Fig 8. In the pharynx, we favor a model in which HPX-2 contributes to the hardness/impermeability of the grinder, allowing better disruption of microbes with thick cells walls, and inhibiting microbial colonization of the intestine. In the hypodermis, we postulate that HPX-2 also contributes to the structural integrity of the cuticle, and changes to this barrier tissue perturb signaling pathways resulting in greater susceptibility. We discuss both these aspects of HPX-2 function in more detail below.
In the pharynx, the expression pattern of hpx-2::GFP resembles that of the gland cell reporter gene B02807::GFP [27, 31], with expression being observed in the process that extends from the terminal bulb to the posterior end of the buccal cavity. There are two groups of gland cells, three g1 cells and two g2 cells, located at the distal bulb of the pharynx. The gland cells are thought to secrete vesicles containing enzymes that function in the digestion of microbial food and aid in cuticle formation during the molting process [19]. Specifically, during the feeding process, small vesicles are transported from the g1 cells through the process to the secretory ducts to aid digestion. During molting, much larger vesicles are transported through the process and the enzymes they contain are thought to be involved in the tearing down and building up of the cuticle, a critical remodeling process [19].
HPX-2 has a signal peptide sequence and is predicted to be secreted. We postulate that HPX-2 is protective by virtue of it being released into the pharyngeal lumen by gland cells to aid in cuticle remodeling of the pharynx during development. A properly developed pharyngeal cuticle is likely necessary for effective disruption of microbes during the digestive process. Because Gram-positive bacteria and fungi like C. albicans have thicker, harder-to-disrupt cell walls, this model could explain why the hpx-2 mutants were colonized more efficiently by, and were more susceptible to, these pathogens compared to Gram-negative bacteria (Figs 1 and 5). The increased microbial load of E. faecalis in the hpx-2 mutant could also explain why the transcriptional response was stronger compared to N2 (Figs 5A, 5B and 7D). The cell envelope of Gram-negatives is comprised of two membranes, but only a thin layer of peptidoglycan, rendering these organisms easier to disrupt by mechanical means (Reviewed in [32]), and may explain why the hpx-2 mutants were not more susceptible to them. While we favor HPX-2 function ultimately affecting the mechanical efficiency of pharyngeal grinding, other explanations such as helping generate a microbicidal oxidant or effects on immune signaling remain formally possible.
In previous studies examining staged animals, significant levels of hpx-2 expression were only observed during dauer [33]. Using lines expressing an HPX-2::GFP fusion, we also detected expression during dauer in most animals examined (Fig 3E). However, we did not observe GFP expression in the vast majority of animals at other stages (L1-L4 and adult) and during pathogen exposure. The hpx-2 transcript was also not induced following pathogen exposure (S3 Table). Expression was observed only occasionally in the pharyngeal processes as shown in Fig 3D. Other genes encoding proteins involved in cuticle development like mlt-7 are expressed at intervals corresponding to the larval molts [13]. However, we did not observe consistent expression associated with molting using staged HPX-2::GFP animals. Based on all these observations, it appears that hpx-2 is expressed during dauer, a time of cuticle remodeling, but the timing of its role in non-dauer development remains unclear. It is possible that HPX-2 is subject to mechanisms of post-transcriptional regulation that would not necessarily be elucidated by these approaches focused on transcriptional regulation.
Peroxidases play important roles in cuticle bio-generation process and loss-of-function mutants frequently exhibit morphological defects. Mutations in MLT-7 or the peroxidase domain of BLI-3 result in severe blister phenotypes [13, 34]. skpo-1 mutants have a partially penetrant dumpy phenotype [14]. Loss of HPX-2 did not result in a morphological phenotype observable by light-microscopy or TEM. However, the increase in cuticle permeability, as evidenced by Hoechst staining, indicates some perturbation of the cuticle. Additionally, overexpression of hpx-2, by injection of higher concentrations of the transgene, resulted in some dpy and rol animals, again indicative of HPX-2 playing a role cuticle biogenesis in the hypodermis. Tissue-specific RNAi knock-down of hpx-2 in the hypodermis resulted in a weak susceptibility phenotype. Collectively, these data suggest that hpx-2 is expressed and plays an infection-protective role in the hypodermis, though expression of our HPX-2::GFP transgene was not detectable in this tissue.
How might a hypodermal cuticle impaired by lack of HPX-2 translate into susceptibility to pathogen, particularly pathogens that colonize and infect the intestine? First of all, it is important to note that not all cuticle defects alter sensitivity. For example, mlt-7 mutations and bli-3 peroxidase domain mutations result in severe blistered phenotypes, with no concomitant increase in pathogen susceptibility [8, 14]. Interestingly, a recent study showed that disruption of some, but not all, aspects of hypodermal cuticle structure triggers the activation of multiple stress response pathways [35]. Specifically, it was discovered that disruption of the annular furrows activates detoxification, osmolyte sensitivity, and most importantly, antimicrobial responses. In regards to HPX-2, we hypothesize that structural disturbance of the hypodermal cuticle perturbs signal transduction related to innate immune and/or stress responses. These changes in signaling in the hypodermis could affect the responses in other tissue by cell non-autonomous signaling events. However, a precise understanding of how cuticular localized HPX-2 impacts innate immunity will require more detailed knowledge of exactly how HPX-2 modulates cuticle structure.
In conclusion, we report the characterization of a heme peroxidase, HPX-2, that functions protectively against multiple pathogen infections in C. elegans. Our data suggest that HPX-2 contributes in a peroxidase-dependent manner by modulating cuticle structure, affecting barrier function and possibly signaling. Such functions are consistent with the role of peroxidases in other systems. For example, plant peroxidases function during infection in barrier tissue remodeling, specifically cell wall cross-linking and cell wall expansion [36]. Additionally, it has been shown in Drosophila that ROS contribute to barrier defense when utilized by peroxidases, but also serve as a signal to induce global production of antimicrobial peptides [37, 38]. This work highlights the importance of peroxidases in such roles extends to C. elegans.
C. elegans strains were grown and maintained as previously described [39]. The hpx-2 nonsense mutation strain VC20223 hpx-2 (gk252521) was obtained from the Caenorhabditis Genetic Center and was backcrossed with N2 Bristol six times. C. elegans strains, bacterial strains, and fungal strains used in this study are listed in Supplemental Material, S5 Table. For experiments requiring synchronized worms, L1 stage worms on starved plates were washed off, filtered through a 10μm filter (pluriSelect, pluriStrainer 10μm), harvested by centrifugation, transferred to seeded plates, and grown to the desired stage.
To generate the hpx-2(dg047) CRISPR knock-out strain, single guide (sg) sequences were designed by WU CRISPR (http://crispr.wustl.edu) using the un-spliced sequence of the hpx-2 gene. Four 18 bp sg sequences with WU score higher than 65 were selected and separately cloned into pJW1219 [40]. A mixture of four plasmid constructs was injected into N2 worms at a concentration of 10 ng/μl each, and with 10 ng/μl of pJW1219-dpy-10 as a co-CRISPR marker. Worms that displayed a roller or dumpy phenotype were then isolated for genotyping to test for insertions and/or deletions (INDEL) in the hpx-2 gene. Verified hpx-2 mutants were backcrossed with N2 to eliminate the dpy-10 mutation.
The HPX-2 catalytic residue mutant strain (HPX-2-R372A), was generated by Suny Biotech (http://www.sunybiotech.com/) using CRISPR-Cas9. The mutant strain PHX782 hpx-2(syb782) was verified by PCR sequencing and the primers are listed in S6 Table.
To generate the hpx-2 complemented strains, a 12 kb DNA fragment containing the hpx-2 gene and 5 kb of upstream and 4 kb of downstream sequences was amplified from the genomic DNA of N2 worms. The PCR product was then purified and injected into hpx-2(dg047) and hpx-2(gk252521) at 20 ng/μl concentration with 50 ng/μl of EcoRV-digested genomic DNA from N2 worms. pCFJ90 (pMyo2::mCherry) was used as a fluorescent co-injection marker at a final concentration of 2 ng/μl.
To visualize hpx-2 expression, the GF203 strain was generated by injecting the pPD95.75 (50 ng/μl) plasmid containing the 4 kb upstream of hpx-2 and the first two exons fused to gfp, with 20 ng/μl pRF4 (rol-6(su1006)) as a roller co-injection marker. The obtained strains with extrachromosomal arrays were then exposed to trimethylpsoralen (TMP) with UV irradiation and back crossed six times to generate stable integrated transgenic lines using a previously described protocol [41]. All the oligonucleotide primers used in this study are listed in S6 Table.
Killing assays and longevity assays were conducted as previously described, with some slight modifications [42–44]. All assays were done with 30 worms at the L4 stage on three replica plates (or 6-plate wells for the C. albicans assay) for a total of 90 animals and scored for survival over time. For plate preparation, E. faecalis OG1RF was grown in BHI media for 5 hours and seeded onto BHI agar plates with 50 μg/ml of gentamycin and grown overnight at 37°C. For the vancomycin inactivated killing assay, E. faecalis OG1RF was grown in BHI medium overnight and concentrated 10-folds prior to being seeded onto BHI agar plates with 50 μg/ml of gentamycin and 15 μg/ml of vancomycin and incubated for 5 hours at 37°C. P. aeruginosa PA14 was grown in LB broth overnight and seeded onto SK plates, incubated at 37°C for 24 hours, and then 25°C for another 24 hours. S. enterica SL1344 was grown in LB overnight and seeded onto SK plates and incubated at 37°C overnight. S. aureus NCTC8325, grown in TSB with 10 μg/ml nalidixic acid (Nal) overnight, was seeded to TSA+Nal plates and incubated at 37°C for 6 hours. C. diphtheriae NCTC13129 grown in BHI overnight was seeded onto BHI plates containing 25 μg/ml Nal and 50 μg/ml 5-fluoro-2-deoxyuridine (FuDR) and incubated at 37°C for 24 hours. C. albicans strain SC5314 was grown overnight in BHI media and spotted onto solid BHI plates. L4 stage worms were exposed to C. albicans for 4 hours and were collected, washed, and transferred to six-well plates containing 20% BHI and 80% M9W, and scored for survival daily [45].
To knock down hpx-2 gene expression, L1 to L4 stage larvae were exposed to E. coli HT115 containing the expression plasmid for double-stranded hpx-2 previously constructed [16]. L4 stage worms were then transferred to E. faecalis killing plates for the pathogen susceptibility assay.
To visualize HPX-2::GFP expression, worms with the roller phenotype indicative of the co-injection marker were picked and anesthetized with 25 mM tetramisole and mounted on 2% agarose pads. The worms were then visualized and imaged using Olympus FLUOVIEW FV3000 confocal microscopy equipped with Fluoview FV315-SW software. Z-stack images were acquired using a step size of 0.7 μm and processed using Olympus cellSens Dimension Desktop software.
To visualize intestinal colonization by E. faecalis and P. aeruginosa, C. elegans were exposed to these pathogens for 48 hours (E. faecalis) and 24 hours (P. aeruginosa) respectively on agar plates. They were then washed off the plates and washed four times with 1 ml of M9W. Following anesthesia with 25 mM tetramisole hydrochloride, the worms were mounted on 2% agarose pads for imaging, using Olympus FLUOVIEW FV3000 confocal microscopy equipped with Fluoview FV315-SW software.
To measure intestinal CFUs, animals were exposed to bacteria as described above and then washed off plates with M9W. They were then washed three times in 1 ml of M9W, then twice more with 1 ml of M9W containing 25 mM tetramisole hydrochloride. The worms were then treated with 500 μl M9W containing 25 mM tetramisole hydrochloride, 1 mg/ml of ampicillin, and 1 mg/ml kanamycin for 1 hour to kill all surface bacteria. The treated worms were washed twice more with 1 ml of M9W containing 25 mM tetramisole hydrochloride, transferred to an Eppendorf tube containing 100 μl of M9 (one worm per tube), and disrupted using a motorized pestle (Kontes cordless pestle (cat# K749540-0000) and pestles (cat# K749521-1590) for 1 min. The solution was then serially diluted in M9 and spotted on agar plates for CFU counting.
Hoechst staining of the worms was performed as previously described [46] with the following modifications. Specifically, worms were washed off of plates with M9W buffer, and then incubated in M9W containing 10 μg/ml Hoechst 33258 dye (Sigma) at room temperature for 20 minutes with gentle shaking, followed by three more washes with M9W before imaging. Images were acquired using Olympus FLUOVIEW FV3000 confocal microscopy equipped with Fluoview FV315-SW software. Z-stack images were acquired using a step size of 0.7 μm and processed using Olympus cellSens Dimension Desktop software. Hoechst-positive worms were scored based on staining of the cell nuclei, indicative of cuticle penetration.
L4 stage worms were collected and fixed in 3% glutaraldehyde overnight. Samples were then prepared and thin-sectioned for transmission electron microscopy as previously described [45]. Image acquisition was done using a JEOL 1400 transmission electron microscope at 60 kV, equipped with a 2K × 2K Gatan charge-coupled-device camera.
L4 animals were exposed for 16 hours to the condition of interest and total RNA was extracted using Trizol (Invitrogen) according to the manufacturer’s instructions. RNA samples were treated with Turbo DNA free kit (Applied Biosystems) to eliminate DNA contamination. qRT-PCR was performed as previously described [47]. The actin gene was used as an internal control. Primers used in qRT-PCR are listed in S7 Table.
L4 stage worms were exposed to E. faecalis or E. coli for 16 hours and total RNA was extracted for 4 biological replicates. Illumina Hiseq 4000 sequencer with 75 nt pair-ended read format was used to conduct the sequencing. The sequencing reads (ranging from 20 million to 37 million per biological replicate) were quality and adaptor trimmed and mapped to the reference genome (version WBcel235 downloaded from http://ensemblgenomes.org) using Tophat [48]. The expression level (RPKM) of annotated genes was measured using Cufflink [49], and differential expression analysis was conducted using Cuffcompare and Cuffdiff [49] using the gene annotation (Caenorhabditis_elegans.WBcel235.37.gff3 downloaded from http://ensemblgenomes.org). Gene Ontology analysis was carried out by using DAVID (the Database for Annotation, Visualization, and Integrated Discovery) 6.8 [50]. Gene enrichment with a Benjamini adjusted P < 0.05 is listed. Genes that were differentially regulated under different conditions are listed in S1–S4 Tables. The sequencing data are available from the GEO database under accession number GSE124372.
To measure H2O2 concentration of the worms, the Amplex red assay was performed as previously described [16] using the Amplex Red hydrogen peroxide/peroxidase kit (Invitrogen Molecular Probes, Eugene, OR) with the following modifications: L4 worms were exposed to a bacterial strain for 16 hours and transferred to 96 well plates with 30 worms in each well. A total of 80 mM diphenyleneiodinium chloride (DPI) (TCI, Tokyo) was added to some wells and allowed to incubate for 15 minutes prior to addition of Amplex Reagents. After 1 hour of incubation, fluorescence was measured at 540/590 nm excitation and emission, respectively.
To measure the pharyngeal pumping rate of the worms, contractions of the posterior pharyngeal bulb were observed and counted over a 10 second interval for 15 adult animals under a 20X magnification stereo microscope.
To measure the brood size of the worms, 9 L4 stage worms were singled on NGM plates, allowed to lay eggs and transferred to a new plate each day until no more eggs were produced. The offspring on the plates were counted to calculate brood size.
Survival, longevity assays, qRT-PCR, intestinal CFU, and Hoechst staining data were analyzed using GraphPad Prism version 7.0 (GraphPad Software, San Diego). Kaplan-Meier log rank analysis was used to compare longevity and survival curves. An unpaired Student’s t-test was used to determine the statistical significance of the intestinal CFU and Hoechst staining data. In all the experiments, P-values < 0.05 were considered to be significant and are noted in the Figs with one asterisk indicating P < 0.05, two indicating P < 0.01, three indicating P < 0.001 and four indicating P < 0.0001.
|
10.1371/journal.pgen.1007007 | CbtA toxin of Escherichia coli inhibits cell division and cell elongation via direct and independent interactions with FtsZ and MreB | The toxin components of toxin-antitoxin modules, found in bacterial plasmids, phages, and chromosomes, typically target a single macromolecule to interfere with an essential cellular process. An apparent exception is the chromosomally encoded toxin component of the E. coli CbtA/CbeA toxin-antitoxin module, which can inhibit both cell division and cell elongation. A small protein of only 124 amino acids, CbtA, was previously proposed to interact with both FtsZ, a tubulin homolog that is essential for cell division, and MreB, an actin homolog that is essential for cell elongation. However, whether or not the toxic effects of CbtA are due to direct interactions with these predicted targets is not known. Here, we genetically separate the effects of CbtA on cell elongation and cell division, showing that CbtA interacts directly and independently with FtsZ and MreB. Using complementary genetic approaches, we identify the functionally relevant target surfaces on FtsZ and MreB, revealing that in both cases, CbtA binds to surfaces involved in essential cytoskeletal filament architecture. We show further that each interaction contributes independently to CbtA-mediated toxicity and that disruption of both interactions is required to alleviate the observed toxicity. Although several other protein modulators are known to target FtsZ, the CbtA-interacting surface we identify represents a novel inhibitory target. Our findings establish CbtA as a dual function toxin that inhibits both cell division and cell elongation via direct and independent interactions with FtsZ and MreB.
| Bacterially encoded toxin-antitoxin systems, which consist of a small toxin protein that is co-produced with a neutralizing antitoxin, are a potential avenue for the identification of novel antibiotic targets. These toxins typically target essential cellular processes, causing growth arrest or cell death when unchecked by the antitoxin. Our study is focused on the CbtA toxin of E. coli, which was known to inhibit both bacterial cell division and also bacterial cell elongation (the process by which rod-shaped bacteria grow prior to cell division). We report that the effects of CbtA on cell division and cell elongation are genetically separable, and that they are due to direct and independent interactions with its targets FtsZ and MreB, essential cytoskeletal proteins that direct cell division and cell elongation, respectively. Our genetic analysis defines the functionally relevant target surfaces on FtsZ and MreB; in the case of FtsZ this surface represents a novel inhibitory target. As a dual-function toxin that independently targets two essential cytoskeletal elements, CbtA could guide the design of dual-function antibiotics whose ability to independently target more than one essential cellular process might impede the development of drug resistance, which is a growing public health threat.
| In E. coli, as in most other bacteria, cell shape is defined by the peptidoglycan sacculus [1], which is built by the coordinated efforts of two major protein complexes, the cell elongation complex and the cell division complex (reviewed in [2–4]). The cell elongation complex directs the insertion of new cell wall material into the E. coli lateral sidewall, causing a newly divided rod cell to increase in length (while maintaining a constant width). Once the elongated cell has approximately doubled its mass, the division complex (or divisome) builds a new septal wall at mid-cell, forming two equivalently sized rod-shaped daughter cells [2,5,6]. Functional disruption of either of these two complexes in E. coli results in striking cell morphological alterations. Cells that fail to divide form long filaments [7], whereas cells that are blocked for cell elongation lose their rod-shape and become spherical [8,9].
Peptidoglycan insertion by the cell division and cell elongation complexes is directed by a dedicated bacterial cytoskeletal element. Cell division is governed by the broadly conserved bacterial tubulin homolog and GTPase, FtsZ. FtsZ polymerizes into dynamic filaments that coalesce into a ring structure (referred to as the Z ring) at mid-cell. Once properly assembled at mid-cell, this Z ring serves as a scaffold for a large set of essential and non-essential protein components, resulting in formation of the mature division complex, which constructs the new septum (reviewed in [7,10]). Cell elongation in the majority of rod-shaped bacteria is mediated by the actin-homolog and ATPase, MreB [11–14]. MreB polymerizes to form antiparallel double filaments [15] that are peripherally associated with the inner leaflet of the cytoplasmic membrane [16]. In vivo fluorescence imaging studies have shown that MreB forms dynamic filament patches that move circumferentially along the long axis of the cell, directing the lateral incorporation of cell wall material [17–19].
The polymerization, assembly, and dynamics of these bacterial cytoskeletal elements are dictated by their inherent biochemical properties and further influenced by diverse modulatory proteins. FtsZ assembly is controlled by a complex set of positive and negative “house-keeping” regulators that spatiotemporally coordinate Z ring formation with the cell cycle [7,20–25]. FtsZ is also the target of several inhibitors that block its assembly in response to specific environmental cues. For example, in response to cellular DNA damage, the SOS inhibitor SulA blocks FtsZ assembly by sequestering FtsZ monomers [26–28]. Several exogenous inhibitors of FtsZ function, including phage-encoded proteins and small molecule inhibitors, have also been described in recent years [29–31].
The best-characterized MreB inhibitor is the small molecule antibiotic A22, which binds within the nucleotide-binding pocket of MreB and blocks double filament formation [15]. However, relatively few protein modulators of MreB function have been described [32–37] and the physiological relevance of their effects is unknown.
Among proteins that can alter cell shape, the CbtA (formerly known as YeeV) protein of E. coli is unusual in being able to inhibit both cell division and cell elongation. Previously proposed to target both FtsZ and MreB [32], CbtA is the toxin component of the prophage-encoded CbtA/CbeA chromosomal toxin-antitoxin system found in E. coli and other closely related species [38].Toxin-antitoxin systems are genetic modules that encode a small, stable toxin protein and a labile, cotranscribed antitoxin (reviewed in [39–41]). Capable of causing growth arrest or cell death, the toxins typically target essential cellular processes. Toxin-antitoxin systems are abundant in prokaryotic genomes [42] and have been implicated in the bacterial stress response [40,43,44].
Overexpression of the cbtA toxin gene in E. coli was shown by Tan et al. [32] to result in a cell growth defect and a loss of rod shape. Over the course of several hours, cells induced for cbtA expression formed swollen lemon-shaped cells with distinct poles; with prolonged induction, these lemon-shaped cells eventually lysed [32]. This morphology is reminiscent of the change in cell shape induced by a simultaneous block of cell division and cell elongation pathways in E. coli–specifically by the combined inhibition of FtsZ (via overexpression of sulA) and MreB (with A22 treatment) [45]. Consistent with the striking lemon-like morphology seen with cbtA overexpression, Tan et al. detected interactions between the CbtA toxin and both FtsZ and MreB in vivo (by yeast two-hybrid) and in vitro (by pull-down assay) [32]. Nonetheless, whether or not these interactions are directly responsible for the effects of CbtA overproduction on cell shape and cell growth has not been established; in particular, it is not known if CbtA, a small protein of only 124 amino acids, interacts independently with FtsZ and MreB to mediate its effects on cell shape and whether its interaction with these or other proteins is responsible for its toxic effects. Moreover, in light of evidence that in E. coli cell division and septum formation depend on an interaction between FtsZ and MreB [46], CbtA might conceivably exert its effects by interacting directly with only one or the other of these cytoskeletal elements [32].
Here, we genetically dissect the reported interactions of CbtA with FtsZ and MreB. Our analysis indicates that these interactions are direct and independent. We show further that both of these interactions are functionally relevant, contributing independently to CbtA-mediated toxicity and cell-shape perturbations. Our findings thus establish CbtA as a bona fide dual inhibitor of bacterial cell elongation and cell division. Moreover, by identifying the surface of each cytoskeletal element that is bound by CbtA, our findings describe new inhibitory surfaces that can be targeted to block cytoskeletal function.
Consistent with previous reports, we observed that overproduction of CbtA (as a His6-CbtA-GFP fusion protein) under the control of a hybrid T5-lac promoter [47] in E. coli resulted in a severe decrease in viability (Fig 1A). Furthermore, time-lapse microscopy confirmed that upon overproduction of His6-CbtA-GFP, cells failed to divide, adopting a morphology resembling swollen lemons, and eventually lysed (Fig 1B). Observation of GFP fluorescence in these cells revealed that the His6-CbtA-GFP fusion protein was distributed diffusely throughout the entire bloated cell (Fig 1C); a high background of diffuse cytoplasmic fluorescence even early after the induction of fusion protein synthesis obscured any possible co-localization with cytoskeletal elements at earlier time points. Importantly, we found that overproduction of untagged CbtA yielded an identical lemon-shape phenotype (Fig 1D). As the various images in Fig 1 illustrate, whereas essentially all the cells visualized manifested drastic morphological change, the individual lemon-shaped cells displayed striking heterogeneity. Many cells resembled smooth lemons, while others had pronounced tubular projections at one or both poles; bi-lobed lemon-shaped cells (such as the one shown in Fig 1C) were seen by time-lapse microscopy to form from pre-constricted cells (see S1A Fig). These morphologies are consistent with the varied cell shapes observed by Varma et al. upon combined FtsZ and MreB inhibition [45].
Tan et al. detected interaction between CbtA and its proposed cytoskeletal targets in a yeast two-hybrid system [32]. Similarly, we detected interaction between CbtA and both FtsZ and MreB in a bacterial two-hybrid system developed in our lab [48,49]. In this assay, contact between a protein domain (X) fused to the α subunit of E. coli RNA polymerase and a partner domain (Y) fused to the CI protein of bacteriophage λ (λCI) activates transcription of a lacZ reporter gene under the control of a test promoter bearing an upstream λCI-binding site (Fig 2A). In this case, we fused CbtA to λCI and either FtsZ or MreB to α. We detected an 18-fold increase in lacZ expression in the presence of the λCI-CbtA and α-FtsZ fusion proteins (Fig 2B) and a 3-fold increase in lacZ expression in the presence of the λCI-CbtA and α-MreB fusion proteins (Fig 2C). Whereas these results are consistent with the idea that CbtA can interact directly with both FtsZ and MreB, they do not exclude the possibility that chromosomally encoded FtsZ may serve as a protein bridge linking the fused CbtA and MreB moieties. Our genetic analysis below addresses this possibility.
To determine if CbtA can interact independently with FtsZ and MreB and to examine whether these interactions contribute directly to CbtA-mediated toxicity, we sought to identify mutations that specifically disrupt each of these interactions. We began by testing a CbtA variant that we had isolated on the basis of reduced toxicity (S1 Text and S1 Fig), which bore the substitution F65S. Although further analysis revealed that mutant CbtA-F65S was only slightly less toxic than wild-type CbtA when overproduced in E. coli (Fig 3A), bacterial two-hybrid analysis revealed that substitution F65S in the CbtA moiety of the λCI-CbtA fusion protein specifically disrupted its interaction with FtsZ (Fig 3B), without compromising its interaction with MreB (Fig 3C). Morphological observations were consistent with these results; upon overproduction of His6-CbtA-F65S-GFP to an intracellular level comparable to that of the wild-type protein (S1C Fig), cells adopted a sphere-like rather than a lemon-like morphology (Fig 3D). As seen by time-lapse microscopy (S1A Fig), over the course of a three-hour induction period, cells producing His6-CbtA-F65S-GFP lost their rod shape, becoming spheroidal. These spheroidal cells continued to divide for one to two generations, gradually increasing in diameter until they lysed, a phenotype that mirrors that observed upon depletion of MreB [8,9]. We conclude that CbtA’s ability to block cell division is due to a direct interaction with FtsZ. In addition, these findings suggest that CbtA-F65S is able to interact with MreB and mediate a block in cell elongation even in the absence of an interaction with FtsZ.
To further evaluate the proposition that CbtA’s ability to block cell elongation is due to a direct interaction with MreB, we sought to identify a CbtA variant with the opposite interaction profile: strong FtsZ interaction and abrogated MreB interaction. To do this, we used a two-hybrid-based screening strategy (see Materials and Methods). Specifically, we introduced random mutations into the gene fragment encoding the CbtA moiety of the λCI-CbtA fusion protein, transformed the mutagenized library into reporter strain cells containing the α-MreB fusion protein, and screened for clones with reduced expression of the lacZ reporter gene. λCI-CbtA mutants identified in this manner were then counter-screened to identify those that supported high levels of lacZ expression in the presence of the α-FtsZ fusion protein. Using this two-step screening procedure, we identified substitution R15C, which specifically disrupted the interaction of CbtA with MreB (Fig 3C), without compromising its interaction with FtsZ (Fig 3B). Consistent with these two-hybrid data, induction of His6-CbtA-R15C-GFP production was toxic and caused the cells to form filaments, rather than adopting the lemon shape observed when the wild-type protein was produced at comparable levels (Fig 3A, Fig 3D, S1C Fig). We conclude that CbtA’s ability to block cell elongation is due to a direct interaction with MreB, such that the CbtA-R15C variant, which still interacts with FtsZ, blocks cell division without blocking cell elongation. Together, the two-hybrid data and morphological phenotypes produced by the CbtA-F65S and CbtA-R15C variants demonstrate that the inhibitory functions of the CbtA toxin are independent and genetically separable.
Although cells overproducing either His6-CbtA- F65S-GFP or His6-CbtA-R15C-GFP were still inviable (Fig 3A), we found that overproduction of the His6-CbtA-R15C/F65S-GFP double mutant, which accumulated to comparable levels as the wild-type protein (S1C Fig), did not influence viability (Fig 3A). Furthermore, cells producing this variant maintained their rod shape, exhibiting only minor morphological perturbations (Fig 3D). These findings establish the functional relevance of both the CbtA-FtsZ interaction and the CbtA-MreB interaction, each of which contributes to the lemon-like morphology and the viability defect observed upon CbtA overproduction.
As an additional readout of the physiological perturbations caused by each of our CbtA variants, Z ring formation was monitored in cells producing the untagged mutant proteins. We first used a strain that constitutively produces a ZapA-GFP fusion from its native locus [50]. Because ZapA-GFP forms fluorescent ring structures that require proper assembly of the FtsZ ring, this fusion protein can serve as a proxy for FtsZ localization [50,51]. When the ZapA-GFP strain was transformed with an empty vector, fluorescent ZapA-GFP bands were observed at mid-cell in the majority of cells (S1D Fig). In contrast, after two hours of cbtA expression, ZapA-GFP exhibited patchy, cloud-like localization throughout the resulting lemon-shaped cells, suggesting disruption of Z ring formation (S1D Fig). When ZapA-GFP localization was observed in cells producing CbtA-R15C, the majority of cell filaments did not contain visible ring structures, and ZapA-GFP again formed cloud-like structures (S1D Fig). In cells overproducing the less toxic CbtA-R15C/F65S variant, rod shape was maintained, and ZapA-GFP rings were observed in most cells (S1D Fig). Importantly, overproduction of the untagged CbtA variants resulted in identical morphologies to those observed with the His6/GFP constructs (compare Fig 3D phase contrast images with those shown in S1D Fig).
A recent study reported similar ZapA-GFP cloud-like structures under conditions where FtsZ assembly and localization were disrupted, indicating that ZapA is able to localize in an FtsZ-independent manner [52]. To determine whether the patchy ZapA-GFP localization we saw upon overproduction of CbtA and CbtA-R15C was similarly occurring in an FtsZ-independent manner, we examined the localization of GFP-FtsZ (overproduced in a strain also containing wild-type endogenous ftsZ) in the presence of our untagged CbtA variants. Consistent with our ZapA-GFP data, we saw that after two hours of induction of CbtA or CbtA-R15C production, the majority of cells did not contain Z rings; however, unlike ZapA-GFP, GFP-FtsZ exhibited diffuse localization throughout the cell with no cloud-like structures observed (S1E Fig). Thus, it seems likely that the ZapA-GFP patches seen in S1D Fig are forming independently of FtsZ. GFP-FtsZ was found to localize to mid-cell ring structures in cells transformed with an empty vector and in cells producing the CbtA-F65S/R15C double mutant variant (S1E Fig). Taken together, the ZapA-GFP and GFP-FtsZ localization patterns suggest that wild-type CbtA and CbtA-R15C are able to disrupt FtsZ assembly and localization, blocking cell division, whereas the double mutant variant does not. The results of these analyses are consistent with the genetic evidence indicating that CbtA inhibits cell division and cell elongation via independent and separable interactions.
We next sought to identify the CbtA interaction sites on FtsZ and MreB required for CbtA to mediate its inhibitory effects on cell division and cell elongation. Tan et al. reported that removal of the last 66 residues of FtsZ eliminated the yeast two-hybrid interaction detected between CbtA and FtsZ as well as the interaction between MreB and FtsZ [32]. The last 66 residues of E. coli FtsZ include the conserved 15 amino acid C-terminal tail domain (CTT), which serves as a site of interaction for a variety of protein factors that regulate FtsZ assembly [53–62], raising the possibility that CbtA too binds the CTT. We sought to test this possibility using our bacterial two-hybrid assay.
As a positive control, we first tested the ability of the FtsZ membrane-anchoring protein ZipA (its cytoplasmic C-terminal domain) to interact with FtsZ. Structural data indicate that the C-terminal domain (CTD) of ZipA (ZipACTD) binds the FtsZ-CTT [57]. We detected a strong interaction (resulting in a 10-fold increase in lacZ expression) between FtsZ and the ZipACTD, and this interaction was compromised by removal of the C-terminal 66 residues of the FtsZ moiety (Fig 4A), consistent with the structural data [57] and previously reported yeast two-hybrid analysis [63]. However, surprisingly, we found that FtsZ-Δ66 maintained an interaction with CbtA comparable to that of the wild-type protein (Fig 4A), suggesting that the FtsZ-CTT is not necessary for the CbtA-FtsZ interaction. We were unable to detect an interaction between wild-type FtsZ and E. coli MreB in our two-hybrid system (S2A Fig) and thus could not determine whether or not this C-terminal truncation had any effect on that reported interaction.
Because, in our bacterial two-hybrid system, the last 66 residues of FtsZ did not appear to mediate the interaction with CbtA, we sought to identify substitutions in FtsZ that specifically disrupt its interaction with CbtA. To do this, we introduced random mutations into the gene fragment encoding the FtsZ moiety of the α-FtsZ fusion protein, introduced the mutagenized library into reporter strain cells containing the λCI-CbtA fusion protein and screened for colonies with reduced lacZ expression on appropriate indicator medium (see Materials and Methods). To identify FtsZ mutants specifically deficient for interaction with CbtA as opposed to generally destabilized variants, we performed a counter-screen based on the ability of FtsZ to interact with itself. Specifically, we detected a 4-fold increase in lacZ expression in reporter strain cells containing both the α-FtsZ fusion protein and a λCI-FtsZ fusion protein (Fig 4A); a similar interaction was previously reported in the context of both the yeast two-hybrid system [58] and an alternative bacterial two-hybrid system [46]. Thus, we screened for amino acid substitutions in the FtsZ moiety of the α-FtsZ fusion protein that reduced lacZ reporter gene expression in cells containing the λCI-CbtA fusion protein, but not in cells containing the λCI-FtsZ fusion protein. Among those amino acid substitutions that reduced lacZ expression by at least 60% in cells containing λCI-CbtA and by less than 25% in cells containing λCI-FtsZ, all localized to a small region encompassing residues 169–182 (Fig 4A). These amino acid substitutions did not compromise the interaction between FtsZ and the ZipACTD (Fig 4A).
FtsZ residues 168–182 make up a loop region connecting α-helices 6 and 7 (the H6/H7 loop) in the GTP-binding N-terminal domain of FtsZ. FtsZ protofilament crystal structures from several bacterial species show that the H6/H7 loop (shown in yellow in Fig 4B) lies at the longitudinal interface formed by stacked FtsZ monomers [64,66,67].
To evaluate whether or not the H6/H7 loop residues identified in our genetic screen are functionally important for the CbtA-FtsZ interaction, we sought to test the effect of CbtA overproduction in an E. coli strain bearing one of the H6/H7 loop mutations at the endogenous ftsZ locus. We found that the ftsZ-L169P allele was able to support growth when introduced into the chromosomal ftsZ locus (S2 Fig). Although this strain did not fully support cell division in fast-growth conditions (LB at 37°C)–we saw a subset of filamented cells and notable heterogeneity in cell length (S2C Fig)–this division defect was partially rescued by slower growth in LB at 30°C and fully rescued by growth in M9 minimal medium at 30°C (S2C Fig). Indeed, in minimal medium, we observed comparable cell lengths for the wild-type and ftsZ-L169P strains (S2D Fig). Western blot analysis with an FtsZ-recognizing antibody indicated that the FtsZ-L169P mutant protein accumulated within cells to levels comparable to that of the wild-type protein (S2E Fig).
We were therefore in a position to test whether or not the FtsZ L169P substitution specifically blocked the ability of CbtA to inhibit cell division. We found that when wild-type His6-CbtA-GFP was overproduced in the ftsZ-L169P strain, in M9 maltose at 30°C, cells lost their rod shape, but formed small spherical or sphere-like cells rather than lemon-shaped cells (Fig 4C), the expected phenotype for a defect in cell elongation. In particular, previous studies have shown that growth in minimal medium at low temperature can suppress the lethality of a cell elongation defect, such that the cells do not form large spheres and lyse, but instead are able to propagate as small spheres [9]. Overproduction of His6-CbtA-GFP in the wild-type background in these same growth conditions resulted in the formation of lemon-shaped cells, as expected (Fig 4C). We quantified these morphological observations by measuring cell roundness (width divided by length), confirming that His6-CbtA-GFP overproduction in the ftsZ-L169P strain caused a more pronounced increase in roundness than in the wild-type strain (Fig 4D). These findings are consistent with the two-hybrid data and provide strong support for the idea that H6/H7 loop region is functionally implicated in the CbtA-FtsZ interaction.
As a complementary approach, we developed a Bacillus subtilis heterologous system with which to evaluate the importance of residues in the FtsZ H6/H7 loop in enabling CbtA to interact functionally with FtsZ to inhibit cell division. Although the E. coli and B. subtilis FtsZ proteins share ~47% amino acid identity, comparison of the H6/H7 loop sequences reveals several non-conservative amino acid differences (Fig 5A). Thus, we surmised that if the H6/H7 loop mediates the interaction of CbtA with FtsZ, then the CbtA toxin should not interact with B. subtilis (Bsu) FtsZ. As shown in Fig 5B, CbtA was unable to interact with Bsu FtsZ by two-hybrid analysis; however, replacement of the B. subtilis H6/H7 loop with the E. coli loop (Bsu ftsZ (loopEco)) resulted in a strong interaction between Bsu FtsZ and CbtA.
To test whether or not CbtA could inhibit cell division in B. subtilis cells containing either wild-type FtsZ or an FtsZ chimera bearing the E. coli H6/H7 loop region, we constructed strains with either the wild-type or chimeric ftsZ (loopEco) (linked to spec) at the endogenous locus. These strains additionally harbored gfp, wild-type cbtA or cbtA-F65S (both alleles encode an N-terminal His6 tag preceding cbtA followed by a C-terminal GFP moiety) at the ycgO locus under the control of a strong inducible promoter (pHYPERSPANK). Overproduction of wild-type CbtA in the strain bearing the wild-type ftsZ allele had no effect on cell growth (Fig 5C and 5D) or any detectable effect on cell division (S3A Fig), consistent with our inability to detect an interaction between CbtA and Bsu FtsZ by two-hybrid analysis. The chimeric ftsZ (loopEco) allele itself caused a slight growth defect manifest as decreased colony size (Fig 5C) and decreased growth rate in liquid medium (Fig 5D); in addition, microscopic analysis of cells containing the chimeric ftsZ (loopEco) allele revealed a cell division defect (S3B and S3C Fig). However, CbtA overproduction in this strain caused a severe growth defect both on plates (Fig 5C) and in liquid (Fig 5D) and resulted in increased cell lysis (S3C Fig), but overproduction of CbtA-F65S to comparable levels (S3D Fig) did not (Fig 5C and 5D and S3C Fig). We conclude that residues in the H6/H7 loop region of FtsZ dictate whether or not CbtA can interact functionally with FtsZ.
We next sought to determine whether CbtA makes direct contact with the H6/H7 loop of FtsZ. To do this, we aimed to identify compensatory substitutions in CbtA that restored its interaction with specific FtsZ H6/H7 loop mutants, using our two-hybrid system to screen for such mutant-suppressor pairs. To facilitate this analysis, we first sought to identify a charge reversal substitution in the H6/H7 loop that disrupted the CbtA-FtsZ interaction. Having identified substitution D180N in our original screen for disruptive mutations, we tested the effect of a charge reversal substitution at the same position (D180K). We found that this charge reversal substitution almost completely eliminated the two-hybrid interaction between FtsZ and CbtA (Fig 6), making it a suitable starting point for seeking to identify compensatory substitutions in CbtA. Accordingly, we transformed reporter strain cells containing the α-FtsZ-D180K fusion protein with a mutagenized library of plasmids encoding the λCI-CbtA fusion protein (bearing random mutations in the cbtA moiety) and screened for clones with elevated expression of the lacZ reporter gene. These candidate suppressor mutants were then pooled and counter-screened to identify those that maintained a low level of lacZ expression in the presence of the wild-type α-FtsZ fusion protein, and thus to identify substitutions that enabled CbtA to interact with FtsZ-D180K but not wild-type FtsZ. With this two-step screening procedure, we identified CbtA substitution V48E, which partially restored the interaction between CbtA and FtsZ-D180K (resulting in a 9-fold increase in lacZ expression) (Fig 6). The effect of this substitution was allele-specific, as CbtA-V48E was unable to interact with wild-type FtsZ or other H6/H7 loop mutants identified in our original screen (L169P, S177P, D180N) (Fig 6). We conclude that CbtA interacts directly with the H6/H7 loop of FtsZ.
Next, in an attempt to identify the MreB surface targeted by CbtA, we sought to isolate MreB variants reduced for their interaction with CbtA in our bacterial two-hybrid system. Specifically, we introduced random mutations into the gene fragment encoding the MreB moiety of the α-MreB fusion protein, introduced the mutagenized library into reporter strain cells containing the λCI-CbtA fusion protein and screened for colonies with reduced lacZ expression. To identify MreB mutants that were specifically deficient for CbtA interaction, each candidate was counter-screened for interaction with the cytoplasmic N-terminal domain of RodZ (RodZNTD). RodZ is a component of the cell elongation complex, and its interaction with MreB has been established by bacterial two-hybrid and structural studies [68,69]. Using our two-hybrid system, we detected a 3 to 4-fold increase in lacZ expression in reporter strain cells containing both the α-MreB fusion protein and a λCI-RodZNTD (residues 2–84) fusion protein (Fig 7A and 7B). We confirmed the biological relevance of this interaction with the introduction of a charge reversal substitution (E319K in E. coli MreB, corresponding to E318K in C. crescentus MreB, shown in wheat in Fig 8) at the previously defined MreB-RodZ interface [69] (S4A Fig).
Using this two-step screening procedure, we identified four amino acid substitutions in MreB (I126V, V173A, E196G, and E262G) that reduced lacZ expression substantially in cells containing λCI-CbtA, but by less than 30% in cells containing λCI-RodZNTD (Fig 7A). We mapped these residues onto the C. crescentus (Cc) MreB double filament structure recently determined by the Lowe group [15] and found that they clustered at or near the interface formed by the paired protofilaments (Fig 8). This interface is formed by an interaction between the flat sides of the protofilaments, which pair in an antiparallel fashion. Stabilizing this antiparallel arrangement is an interaction between juxtaposed alpha helices (α-helix 3 in subdomain IA from one MreB subunit stacks onto α-helix 3 from the opposed subunit, with residue V121 (corresponding to Cc residue V118) playing a particularly important role; Fig 8). Two of the residues identified in our screen (E196 and E262, corresponding to Cc residues E193 and E261, respectively) are surface exposed on the flat side of the Cc MreB protofilament at the inter-protofilament interface. Although the other two residues (I126 and V173, corresponding to Cc I123 and Cc V170, respectively) are buried, they are located near the inter-protofilament interface and residue I126, in particular, lies within the critical dimerization helix (α3) (Fig 8) [15].
The Lowe group demonstrated that the interaction of MreB protofilaments via their flat sides is necessary for proper MreB function in E. coli [15]; thus, if CbtA is indeed interacting with the flat side of MreB, it may be blocking MreB function by preventing the formation of the essential double filament. To further evaluate the proposition that CbtA interacts with the flat side of MreB, we performed a targeted mutagenesis of other residues located on its flat side. Charge reversal substitutions were introduced at several positions (e.g. D192K) and non-conservative changes were made at additional positions (e.g. F84A). The ability of each MreB mutant (bearing a single amino acid substitution) to interact with both CbtA and RodZNTD was assessed in our two-hybrid system. S1 Table summarizes the two-hybrid interaction profiles of the complete set of MreB variants that was tested.
Among those MreB variants tested (as α-MreB fusion proteins), we identified three additional inter-protofilament interface mutants with altered CbtA-binding. The mutants α-MreB-F84A and α-MreB-D192K were unable to interact with λCI-CbtA, but maintained strong interaction with λCI-RodZNTD (Fig 7A). Conversely, the α-MreB-S269F variant was greatly increased in its ability to interact with λCI-CbtA, yielding a 25-30-fold increase in lacZ expression as compared to the highest empty vector control (Fig 7B, S4B Fig). This fold-change value is ~10 times higher than the 3-fold increase in lacZ expression consistently measured with wild-type α-MreB and λCI-CbtA. Importantly, the effect of the S269F substitution was specific to CbtA; α-MreB-S269F yielded a λCI-RodZNTD interaction profile identical to that of wild-type α-MreB across multiple induction levels (Fig 7B). Because MreB and CbtA are both known to interact with FtsZ [46], we considered the possibility that the S269F substitution might actually promote interaction between α-MreB and FtsZ; enhanced bridging of α-MreB-S269F and λCI-CbtA by endogenous FtsZ molecules could potentially lead to an apparent increase in the CbtA-MreB interaction. However, this explanation seems unlikely as α-MreB-S269F interacted similarly strongly with the λCI-CbtA-F65S variant, which is unable to interact with FtsZ (S4B Fig).
Single amino acid substitutions at various positions along the flat side of MreB (including several affecting residues that lie directly at the double protofilament interface) altered its interaction with CbtA in the context of our two-hybrid system. These data suggest that the MreB inter-protfilament interface may be the binding surface utilized by CbtA to inhibit cell elongation. To determine whether or not the MreB interface residues identified in our two-hybrid analyses are critical for the toxic block in cell elongation mediated by CbtA, we aimed to overproduce CbtA-F65S (whose toxicity derives exclusively from its ability to interact with MreB) in strains producing the various mutants as the sole source of endogenous MreB. Overproduction of CbtA-F65S results in a lethal (under rapid growth conditions) loss of rod shape, causing cells to become spherical and lyse. Accordingly, we predicted that in strains bearing MreB substitutions that disrupt the CbtA-MreB interaction, overproduction of CbtA-F65S would be less toxic and would not induce a spherical morphology. Additionally, we hypothesized that CbtA-F65S toxicity might be increased in a strain producing the “up” variant MreB-S269F, resulting in more severe growth defects and morphological perturbations. Importantly, this strategy required the use of MreB variants capable of supporting rod-shaped growth.
We tested the abilities of several of our isolated mutant alleles to complement the growth and morphological defects of an mreBCD depletion strain (FB30/pFB174) when expressed in an IPTG-dependent manner from a multi-copy plasmid along with operon partners mreC and mreD (Fig 9A) [9]. Only cells expressing mreB-E262G or mreB-S269F were able to support growth to a similar extent as the wild-type allele (Fig 9B); strains expressing these mutant alleles also maintained a rod shape comparable to that of the wild-type mreB-expressing strain (Fig 9C). We thus proceeded with these two alleles, testing the effect of overproducing CbtA-F65S in cells containing wild-type mreB, mreB-E262G or mreB-S269F as the sole source of MreB (see Materials and Methods). All three strains exhibited similar growth rates in liquid medium (S4C Fig), and Western blot analysis of these strains using an MreB antibody indicated that both mutant proteins were produced at levels comparable to that of the wild-type MreB protein (S4D Fig).
To test the effect of disruptive substitution E262G, we transformed our wild-type and mreB-E262G strains with either a plasmid producing CbtA-F65S (untagged) under the control of an arabinose-inducible promoter or an empty vector control and monitored cell growth and cell morphology in the presence of arabinose. Cells of both strains bearing the empty vector maintained rod shape in the presence of arabinose, and, as expected, cells containing wild-type MreB and the CbtA-F65S plasmid became spherical within two hours of arabinose addition (Fig 9D). In marked contrast, cells containing MreB-E262G did not become round, maintaining rod-like shape after two hours of arabinose addition (Fig 9D); CbtA-F65S-dependent growth inhibition was also reduced in the mreB-E262G strain (Fig 9E). Thus, MreB substitution E262G, which lies at the double protofilament interface and disrupted the two-hybrid interaction between MreB and CbtA, also interfered with the ability of CbtA-F65S in inhibit cell elongation.
To test the effect of substitution S269F, which strengthened the two-hybrid interaction between MreB and CbtA, we transformed our wild-type and mreB-S269F strains with either a plasmid producing CbtA-F65S (untagged) under the control of a tetracycline-inducible promoter or an empty vector control. We used a tetracycline-inducible system for these experiments to achieve a finer range of CbtA-F65S concentrations that might enable us to observe a wider range of growth and morphology phenotypes. Nonetheless, we did not observe any obvious morphological differences between these two strains at either 30°C or 37°C at multiple anhydrous-tetracycline (ATC) concentrations; both strains similarly transitioned from rod-shaped to spherical cells over the course of one hour (S4E Fig). However, we did see a more pronounced CbtA-F65S-dependent growth defect in the mreB-S269F strain (Fig 9F). In particular, expression of cbtA-F65S at a low ATC concentration (15 ng/mL) in the wild-type mreB strain caused a relatively modest growth defect on LB agar at 37°C, whereas expression of cbtA-F65S at the same ATC concentration in the mreB-S269F strain resulted in a 2-3-log decrease in plating efficiency (Fig 9F). Expression of the cbtA-R15C/F65S double mutant (recall that substitution R15C specifically disrupts the interaction of CbtA with MreB) had no effect on the growth of either strain, confirming that the increased toxicity of CbtA-F65S in the mreB-S269F background was dependent on the CbtA-MreB interaction (Fig 9F). Thus, MreB substitution S269F, which lies at the double protofilament interface and substantially strengthened the CbtA-MreB two-hybrid interaction, also sensitized MreB to the toxic cell elongation block mediated by CbtA-F65S. Taken together, our analyses of the MreB-E262G and MreB-S269F variants strongly suggest that the flat surface of MreB is critical for CbtA-dependent cell elongation inhibition and likely forms the inhibitory surface directly targeted by CbtA.
CbtA has two homologs in E. coli: the YkfI toxin of the YkfI/YafW toxin-antitoxin system, and the YpjF toxin of the YpjF/YfjZ toxin-antitoxin system [38]. The three toxins are encoded on different cryptic prophage elements within the E. coli genome, and have high amino acid sequence identity (58% identity between CbtA and YkfI, 62% identity between CbtA and YpjF, 78% identity between YpjF and YkfI) [38](S5A Fig). Overexpression of either ykfI or ypjF was previously shown to be toxic [38,70] and results in the formation of lemon-shaped cells [70]. Consistent with these previous results, we found that overexpression of his6-ypjF-gfp or his6-ykfI-gfp under the control of the hybrid pT5-lac promoter resulted in a decrease in viability (Fig 10A) and led to the formation of lemon-shaped cells (Fig 10B). We were also able to detect strong interactions between both toxins and FtsZ (Fig 10C), and between YpjF and MreB in our bacterial two-hybrid system (Fig 10D). Although YkfI is 78% identical to YpjF and blocks cell elongation when overproduced, we were unable to detect an interaction between YkfI and MreB in our bacterial two-hybrid system.
In order to determine whether YkfI and YpjF interact independently with FtsZ and MreB and utilize the same inhibitory surfaces as CbtA, we repeated many of the two-hybrid analyses described above. Residue F65 is conserved in all three toxins (S5A Fig), and as we saw with CbtA-F65S, overproduction of YpjF-F65S and YkfI-F65S yielded sphere-like rather than lemon-shaped cells (Fig 10B). Furthermore, we found that substitution F65S disrupted the two-hybrid interactions between YpjF and FtsZ and between YkfI and FtsZ (Fig 10C), but did not compromise the two-hybrid interaction between YpjF and MreB (Fig 10D). These analyses suggest that the FtsZ interaction determinants for all three toxins are conserved. Importantly, the morphology data also suggest that toxin interaction with FtsZ contributes to the striking lemon-shape phenotype. Interestingly, R15 is not a conserved residue; YpjF and YkfI both have a cysteine at this position (S5A Fig). Because substitution R15C decreased the interaction between CbtA and MreB, we wondered if the reverse substitution (C15R) in YpjF would increase its interaction with MreB, and in the case of YkfI, might allow for a detectable interaction. We found that substitution C15R had no effect in either case (S5B Fig). Thus, the genetic determinants within YpjF and YkfI that specify their abilities to inhibit cell elongation remain unknown.
To assess whether residues in the H6/H7 loop are necessary for the YpjF-FtsZ and YkfI-FtsZ interactions, we assayed the abilities of YpjF and YkfI to interact with wild-type Bsu FtsZ and our Bsu FtsZ chimera. We found that neither YkfI nor YpjF interacted with wild-type Bsu FtsZ, but both toxins interacted strongly with the Bsu FtsZ chimera containing the H6/H7 loop of E. coli (Bsu ftsZ (loopEco)) (Fig 10E). These findings provide strong support for the idea that all three homologous toxins interact directly with the H6/H7 loop of E. coli FtsZ. Similarly, as was seen with CbtA, single amino acid substitutions affecting residues on the flat side of MreB altered its interaction with YpjF. The E262G substitution disrupted the YpjF-MreB interaction, whereas the S269F substitution more than doubled the detected interaction (Fig 10F). Furthermore, we found that a strain harboring the mreB-E262G allele was less susceptible than the corresponding wild-type mreB strain to cell shape changes induced by YpjF-F65S (Fig 10G), suggesting that CbtA and YpjF both require residues lying at the double protofilament interface to inhibit the function of MreB. Taken together, these results strongly suggest that like CbtA, YpjF and YkfI act as dual inhibitors that block cell division and cell elongation in a genetically separable manner; furthermore, all three toxins appear to inhibit these processes by targeting the same surfaces of FtsZ and MreB.
We have shown that the chromosomally encoded toxin CbtA inhibits cell division and cell elongation via independent and genetically separable interactions with FtsZ and MreB. In particular, we identified amino acid substitutions in CbtA that specifically disrupt its interaction with FtsZ on the one hand (F65S), and MreB on the other (R15C), and we showed that both interactions contribute to CbtA toxicity. This genetic analysis enabled us to rule out the possibility that CbtA exerts its effect on either cell division or cell elongation indirectly, by interfering with a functionally relevant interaction between FtsZ and MreB [46]. Furthermore, we identified amino acid substitutions in FtsZ and MreB that disrupt their respective interactions with CbtA, and we constructed strains bearing one or the other type of substitution. This analysis defined the surfaces of FtsZ and MreB targeted by CbtA to inhibit cell division and cell elongation, respectively. Based on the location of these surfaces, we suggest that CbtA may block FtsZ polymerization and MreB double filament formation.
Our analysis of the interaction between CbtA and FtsZ uncovered the H6/H7 loop as a new target for inhibitors of FtsZ function. In particular, using our two-hybrid assay to screen for FtsZ mutations that specifically disrupted its interaction with CbtA, we found that the identified mutations mapped to the H6/H7 loop. We then showed that the identities of loop residues dictated whether or not CbtA could inhibit cell division both in E. coli and in B. subtilis. Finally, we identified an amino acid substitution in CbtA (V48E) that functioned as an allele-specific suppressor of a disruptive charge reversal substitution in the FtsZ H6/H7 loop (D180K), providing strong evidence for a direct physical interaction between CbtA and the H6/H7 loop. FtsZ subunits assemble as protofilaments by stacking vertically in a head-to-tail fashion, and the H6/H7 loop lies at the longitudinal interface formed by pairs of stacked subunits (see Fig 4B) [64,66,67]. We thus suggest that CbtA likely exerts its inhibitory effect on cell division by interfering with FtsZ protofilament formation. Additionally, residues in the H6/H7 loop have been implicated in FtsZ lateral interactions and bundling [71,72], raising the possibility that CbtA could also inhibit such higher order interactions of FtsZ. We note that the cognate antitoxin of CbtA, CbeA (previously YeeU), has been shown to interact with both FtsZ and MreB and stabilize protofilament bundling in vitro, suggesting that it neutralizes CbtA toxicity in vivo by stabilizing a higher order assembly of each of its targets, rather than by interacting directly with the toxin [33]. Interestingly, CbeA was also found to neutralize the toxicity of several other protein inhibitors of FtsZ function with distinct modes of action, suggesting that CbeA’s ability to stabilize a higher order assembly of FtsZ has a general protective effect [33].
Our findings do not support the previous proposal that CbtA interacts with the C-terminal region of FtsZ. Specifically, we found that removal of this C-terminal region (FtsZΔ66; Fig 4A) had no effect on the CbtA-FtsZ interaction, in contrast with previously reported results [32]. The FtsZ monomer consists of a globular core comprising two independently folded domains separated by α-helix H7; appended to this globular core is an unstructured linker region terminating in the highly conserved 15-residue C-terminal tail (CTT). Like their tubulin counterparts, the FtsZ N-terminal domain binds GTP, whereas the C-terminal domain contains the so-called synergy (T7) loop, which stimulates GTP hydrolysis in the context of an assembled protofilament by contacting the GTP-binding pocket of the next subunit. Specifically, the vertical stacking of FtsZ subunits enables the T7 loop of one subunit to insert into the GTP-binding pocket of the subunit just beneath [64] (see Fig 4B). Most previously characterized regulators of FtsZ assembly, including inhibitory factors such as SlmA [53,55] and the C-terminal domain of MinC [62], bind to the CTT, establishing it as an important hub of regulation [54,56–61]. Other protein regulators, such as SulA, the B. subtilis sporulation factor MciZ, and the N-terminal domain of MinC bind within the C-terminal domain (for example, in the vicinity of the T7 loop) and inhibit FtsZ assembly through diverse mechanisms [28,73–75]. CbtA and its homologs YpjF and YkfI provide the first example of protein inhibitors that target the FtsZ N-terminal domain, binding the H6/H7 loop.
The identification of an FtsZ inhibitor that binds to the H6/H7 loop could facilitate studies aimed at probing the polarity of protofilament assembly and disassembly. Both the GTP-binding pocket and the H6/H7 loop are located at the FtsZ plus end (defined by analogy with tubulin), whereas the T7 loop is located at the minus end [64,76]. Evidence from one study, in which plus-end and minus-end mutants were tested for their abilities to function as FtsZ cappers, suggested that FtsZ filaments assemble and disassemble with a polarity opposite that of microtubules, with FtsZ subunits being added primarily to the minus end and dissociating primarily from the plus end [76]. However, evidence from a more recent study, in which the sporulation-specific Z ring inhibitor MciZ was defined as a minus-end capper, was more consistent with designation of the plus end as the primary addition site [74]. CbtA, a plus-end binder, could provide a useful new tool for addressing this problem.
Our genetic analysis of the CbtA-MreB interaction identified amino acid substitutions that mapped to the flat side of the MreB monomer. Specifically, we identified six substitutions affecting flat-side residues (F84A, I126V, V173A, D192K, E196G, E262G) that disrupted and one (S269F) that strongly increased the two-hybrid interaction between CbtA and MreB. Among the affected residues, F84, E196, and E262 are surface-exposed, suggesting that they may contact CbtA directly; in addition, S269 is surface-exposed, suggesting that the mutant phenylalanine residue may form a new interaction at the CbtA-MreB interface. In the case of residue D192, which lies in a small pocket beneath residue E196, we speculate that the introduction of a positively charged lysine residue in its place may alter the position of surface-exposed E196, indirectly perturbing its interaction with CbtA. We found that cells containing either MreB-E262G or MreB-S269F (in the absence of wild-type MreB) grew as rods, enabling us to assess the effects of these substitutions on CbtA-dependent morphological changes, as well as CbtA-dependent growth inhibition. Our findings indicated that MreB substitution E262G mitigated the effects of CbtA on growth and cell shape, whereas MreB substitution S269F potentiated the toxic effect of CbtA (Fig 9). Taken together with the two-hybrid data, these findings support the idea that CbtA inhibits cell elongation by binding directly to the flat surface of MreB.
The discovery that pairs of MreB protofilaments associate in an antiparallel fashion along their flat sides to form a double filament that is required for MreB function in E. coli [15] leads us to propose that CbtA inhibits cell elongation by interfering with double filament formation. This proposed mechanism appears to be shared by other MreB inhibitors, as well. The small molecule inhibitor A22 (and its derivative MP265) was found to prevent double filament formation, evidently by displacing the main dimerization helix (α-helix 3 formed by Q120 to A133) that participates in essential inter-protofilament contacts (though the inhibitors also block nucleotide hydrolysis in the active site) [15]. Moreover, a recent study of B. subtilis sporulation factors YodL and YisK suggests that they can influence cell shape by targeting MreB and the MreB-like protein Mbl, respectively; specifically, Mbl substitution E250K (affecting the residue corresponding to E262 in E. coli MreB) was found to suppress the cell-shape defects caused by YisK [37]. Assuming that Mbl adopts a similar double-filament architecture as E. coli MreB, YisK may also disrupt double filament formation.
CbtA is the first example of a cytoskeletal inhibitor capable of independently targeting the cell division and cell elongation apparatuses. Given its small size (124 amino acids), CbtA’s ability to function as a dual inhibitor is particularly striking. It will be interesting to learn whether a single toxin molecule can interact simultaneously with FtsZ and MreB or whether inhibition of cell division and cell elongation depends on the combined action of subsets of molecules that interact with one or the other target. Whereas our work sheds light on the molecular basis for the effects of CbtA and its homologs YpjF and YkfI on cell shape and cell growth, it remains to be learned what roles these toxins might play in cellular physiology. A recent study reported that an E. coli strain deleted for all three toxin genes exhibited increased susceptibility to oxidative stress [70], raising the possibility that these toxin-antitoxin systems, like other chromosomally encoded toxin-antitoxin systems, contribute to the bacterial stress response. As these toxins are encoded on cryptic prophages, it is also interesting to consider what roles they might have played in the context of phage biology. A number of phage-encoded factors are known to block bacterial cell division, some of which (for example, the Kil peptide of bacteriophage λ) have been shown to target FtsZ [29]. It is of particular interest to note that the lytic phage T7 has recently been shown to encode separate inhibitors of FtsZ (Gp0.4) and MreB (Gp0.6) [30,36], one of which (Gp0.4) was shown to provide a growth advantage to the phage in dividing cells [30]. Kiro et al. [30] suggest that inhibiting cell division early after infection ensures that all the cell’s resources are available for phage replication by preventing daughter-cell escape. Whether or not the changes in cell size, cell shape and/or cell wall integrity that result from the combined inhibition of FtsZ and MreB have the potential to enhance phage production remains to be investigated.
FtsZ has been validated as a clinically relevant antibiotic target [31]. Our identification of its H6/H7 loop as a new epitope that can be exploited by FtsZ inhibitors may therefore have implications for the development of new antibiotics that target this essential protein. More speculatively, future structural studies of CbtA in complex with its partners could potentially inform the design of antibiotics that target both FtsZ and MreB (the latter also representing a potentially effective target for antibacterial agents). Such dual-function agents would be attractive due to the greater barrier towards the development of resistance.
A complete list of the bacterial strains used in this chapter is provided in S2 Table. Additionally, lists of the plasmids and oligonucleotides used in this chapter can be found in S3 and S4 Tables, respectively. NEB5-α F’Iq (New England Biolabs) was used as the cloning strain for all plasmid constructions outlined below. Two-hybrid studies were performed in FW102 OL2-62 [77] or BN30 [78]. Morphology observations were made primarily in strain BW27785 [79,80]. This strain also served as template for all colony PCRs. ZapA-GFP microscopy was performed using strain NP1 [50]. GFP-FtsZ microscopy was performed using strain TB28 HKHC488; this strain contains wild-type ftsZ at the endogenous locus and an IPTG-inducible allele of sfgfp-ftsZ (encoding FtsZ fused to superfolder GFP) integrated at the attHK site. E. coli strains were grown in LB (1% NaCl) broth at 37°C or 30°C, and on LB plates supplemented with appropriate antibiotics at the following concentrations (unless otherwise noted): carbenicillin (Carb), 100 μg/mL; chloramphenicol (Cm), 25 μg/mL; kanamycin (Kan), 50 μg/mL; spectinomycin (Spec), 50 μg/mL; streptomycin (Strep), 25 μg/mL; tetracycline (Tet), 5 μg/mL. Where noted, strains were grown in M9 minimal liquid medium or M9 agar (1 mM MgSO4) supplemented with either 0.4% maltose and 0.01% casamino acids, or 0.2% maltose and 0.2% casamino acids.
B. subtilis strains were grown at 37°C in LB (0.5% or 1% NaCl) broth without antibiotic or on LB plates supplemented with spectinomycin (100 μg/mL) or MLS (mixture of 1 μg/mL erythromycin and 25 μg/mL lincomycin). B. subtilis PY79 genomic DNA was used as template for all B. subtilis ftsZ constructs.
p3-37 is a derivative of the ASKA overexpression vector, pCA24N [47], encoding His6-YkfI-GFP under the control of the pT5-lac promoter. In this construct, his6-ykfI-gfp contains two SfiI sites flanking the ykfI sequence. Empty vector plasmid pMT136 encoding His6-GFP was made by cloning in a linker sequence composed of annealed oligonucleotides, oSG623 and oSG624, into SfiI-digested p3-37. This linker sequence contains ClaI and XbaI sites and encodes for the additional residues “IDAAASR” in between the SfiI sites in the His6-GFP sequence. To construct plasmids pMT138 (encoding His6-YpjF-GFP) and pMT139 (encoding His6-CbtA-GFP), colony PCR products generated using primer pair oSG639/oSG640 or oSG641/oSG642, respectively, were digested with AclI and XbaI and ligated into pMT136 digested with ClaI and XbaI. Plasmids pMT144 (encoding His6-YkfI-F65S-GFP) and pMT146 (encoding His6-CbtA-F65S-GFP) were generated by ligation of AclI/XbaI-digested overlap PCR products amplified with internal mutagenic primers (oSG663/oSG664 and oSG667/oSG668) and outside primers (oSG659/oSG660 and oSG641/oSG642) into ClaI/XbaI-digested pMT136 backbone. To construct plasmid pDH253, the cbtA-R15C allele was amplified from the two-hybrid construct pDH246 using primers oSG641 and oSG642, digested with AclI and XbaI, and ligated into pMT136 ClaI/XbaI backbone. Plasmid pDH262 (encoding His6-CbtA-R15C/F65S-GFP) was constructed in the same manner as pMT146 except using pDH253 as PCR template.
All bacterial two-hybrid α fusion constructs were cloned by restriction digest into the parent plasmid pBRα-β flap; all two-hybrid λCI fusion constructs were cloned by restriction digest into the parent plasmid pACλCI-β flap. Briefly, the parent plasmids were digested with NotI and BamHI to generate backbone. These backbones were ligated to relevant inserts generated by NotI/BamHI digestion of PCR products amplified using a NotI-containing forward primer and BamHI-containing reverse primer. Forward primers all contain an extra “A” base after the NotI site to maintain the reading frame. Reverse primers all encode a stop codon preceding the BamHI site. Mutant alleles (both point mutants and chimeric alleles) of ftsZ, mreB, or toxin genes were generated using internal mutagenic primers (see S4 Table for specific sequences). Oligonucleotides pBRα_F, pBRα_R, pACλCI_F, and pACλCI_R were used to sequence all two-hybrid constructs.
To generate plasmids pDH325, pDH326, pDH327, and pDH328 encoding untagged CbtA variants, the relevant cbtA allele was amplified from the appropriate pCA24N-derived construct described above using primers oDH446 and oDH447, digested with EcoRI/HindIII, and ligated into pSG360 (EcoRI/HindIII) backbone.
To construct cbtA-F65S and ypjF-F65S arabinose-inducible overexpression vectors (pDH212 and pDH289), alleles were amplified from pMT146 and pMT188 using primer pair oDH285/oDH286 or oDH380/381, respectively. PCR products were digested with NdeI/XbaI and ligated into the pBAD33 (NdeI/XbaI) backbone. To construct cbtA-F65S and cbtA-R15C/F65S tet-inducible overexpression vectors (pDH335 and pDH337), the EcoRI/HindIII inserts from plasmids pDH326 and pDH328 were ligated into pSG369 (EcoRI/HindIII) backbone.
pFB149 (plac-mreBCD lacIQ) contains an XbaI site upstream of mreB and a naturally occurring BamHI site within mreD that is unique on plasmid pFB149. To construct pFB149-derivatives for MreB mutant expression studies, mreB mutant alleles were generated by overlap PCR using pFB149 as template, outside primers oDH372/oDH369, and allele-specific internal mutagenic primers. PCR products were digested with XbaI/BamHI and ligated into the pFB149 (XbaI/BamHI) backbone. All pFB149- derivatives were verified by sequencing using primers oDH355, oDH369, oDH373, oDH374, and oDH375.
(The ftsZ-L169P allele was cloned into pCX41 (digested with HindIII/ClaI) in place of wild-type ftsZ by restriction digest (HindIII/ClaI) and ligation of overlap PCR products generated using wild-type pCX41 as template, internal mutagenic primers (oDH34_F and oDH35_R for L169P) and flanking primers oDH36_F, oDH37_R, which anneal within ftsA and lpxC, respectively. This generated plasmid pDH35, replication of which is controlled by a temperature-sensitive origin of replication. Plasmid is maintained at 30°C and lost at 42°C. Attempted integration of these mutant alleles into the endogenous chromosomal locus was performed essentially as described in [81]. Briefly, pDH35 was transformed into E. coli strain BW27785 in parallel with a pCX41 derivative encoding FtsZ-F268C. ftsZ-F268C is a known complementing allele and thus serves as a control for chromosomal integration. Transformants were plated on LB agar supplemented with Cm (10 μg/mL) and incubated overnight at 30°C. Several colonies were restreaked onto LB (Cm) and incubated at 42°C overnight in order to identify single crossover integrants. After an additional round of restreaking on LB (Cm) at the nonpermissive temperature, candidates were streaked onto LB (Cm) and incubated at 30°C overnight. Firing of the plasmid origin of replication on the chromosome causes a severe growth defect, and double crossover integrants that had looped out the plasmid were identified as healthy revertants within poorly growing streaks. These candidates were purified by restreaking, and were cured of plasmid by growth on LB (without Cm) at 42°C. The ftsZ locus was PCR amplified and sequenced (using sequencing primers generously provided by H. Cho and T. Bernhardt) from Cm-sensitive candidates in order to identify those in which allelic replacement occurred. All subsequent propagation of this strain was done at RT or 30°C to minimize growth defects. Multiple isolates of this strain exhibited identical phenotypes.
To construct strains DH118/pFB149, DH118/pDH278, and DH118/pDH332, plasmids pFB149, pDH278, and pDH332 were transformed into strain BW27785. To introduce the mreBCD::kanR deletion, a P1 lysate was grown on strain FB30/pFB174 and used to infect each recipient strain. Transductants were selected on M9 maltose plates (0.2% maltose, 0.2% casamino acids, 1 mM MgSO4) supplemented with 5 mM sodium citrate and 250 μM IPTG (for expression of mreBCD). Growth on minimal medium is known to suppress mreBCD defects and was used to prevent acquisition of suppressor mutations. Strains were checked for proper kanR insertion by colony PCR using primers oDH289 and oDH307. We note that in all DH118 strains, about 5% of cells failed to grow as rods, forming large spheres (as observed by microscopy); this is likely the result of plac-mreBCD plasmid loss. Growth of DH118 strains was monitored at 37°C over the course of 4 h. Four replicate M9 maltose overnight cultures were back diluted to a starting OD600 of 0.02 in 200 μL LB (Carb) supplemented with 250 μM IPTG in a 96-well microtitre plate. The plate was incubated shaking at 900 rpm in 90% humidity in a Multitron incubation shaker (Infors HT); OD600 readings were taken every 30 min with a microtitre plate reader (Molecular Devices).
B. subtilis strains were generated by directly transforming a PY79 derivative with either a linearized plasmid containing homology to the chromosomal locus where integration was desired or a PCR fragment containing chromosomal homology. In order to generate B. subtilis strains with gfp or various cbtA alleles integrated into the chromosome, plasmids pDH84 (pHYPERSPANK-his6-gfp), pDH85 (pHYPERSPANK-his6-cbtA-gfp), and pDH102 (pHYPERSPANK-his6-cbtA-F65S-gfp) were constructed. These plasmids were generated by restriction digest (HindIII/NheI) and ligation of PCR products amplified from pMT136, pMT139, or pMT146 using primers oDH108 and oDH116 into QER167 (generous gift of D. Rudner) HindIII/NheI digested backbone. These plasmids all contain homology to the ycgO locus flanking the insert. Plasmids were linearized by digestion with ScaI. DH84 (ycgO:: pHYPERSPANK-his6-gfp erm), DH85 (ycgO:: pHYPERSPANK-his6-cbtA-gfp erm), and DH104 (ycgO:: pHYPERSPANK -his6-cbtA-F65S-gfp erm) were generated by transformation of linearized plasmids pDH84, pDH85, and pDH102, respectively, into PY79 ycgO::spec. Transformants were selected on LB supplemented with MLS.
The wild-type ftsZ allele linked to a spec resistance cassette was assembled by Gibson assembly [82] of three PCR products with >20bp of overlapping homology: 1) part of the ftsA locus and the entire ftsZ locus amplified from PY79 genomic DNA using oligos oDH130 and oDH131, 2) amplification of spec from pDR111 using oDH132 and oDH133, and 3) 2 kb chromosomal sequence downstream of the ftsZ locus amplified from PY79 genomic DNA using oligos oDH134 and oDH135. This assembled PCR product was transformed directly into PY79 to generate strain DH98. Transformants were selected for on LB (Spec).
The chimeric ftsZ allele (containing E. coli ftsZ residues 169–182) linked to a spec resistance cassette, was assembled by Gibson assembly of three PCR products (all with at least 20 bp of overlapping homology: 1) 2 kb upstream of ftsZ amplified from PY79 genomic DNA using oligos ODH141 and ODH142, 2) the ftsZ chimeric allele amplified from pDH69 (Bsu ftsZ (loopEco)) using oligos oDH143 and oDH131 3) 2 kb downstream of ftsZ, including spec, amplified from DH98 genomic DNA using oDH132 and oDH144. This assembled PCR product was transformed directly into PY79 to generate strain DH99. Transformants were selected for on LB (Spec). The ftsZ loci from DH98 and DH99 were PCR amplified (using oDH167 and oDH168) and sequenced using oDH127, oDH172, and oDH173. oDH124 and oDH125, which anneal inside the ftsZ ORF were also used for PCR and sequencing.
Strains DH100, DH101, and DH105 were generated by direct transformation of DH98 genomic DNA into strains DH84, DH85, and DH104, respectively. Strains DH102, DH103, and DH106 were generated by direct transformation of DH99 genomic DNA into DH84, DH85, and DH104, respectively. Transformants were selected on LB (Spec) and patched on LB (MLS) to ensure the ycgO locus was unchanged. The ftsZ loci were re-sequenced after transformation.
cbtA and mreB gene fragments (located on pMT154 and pMT151, respectively) were mutagenized by error-prone PCR using Taq polymerase (Promega) and the outside primers pACλCI_F and pACλCI_R. The ftsZ gene fragment (located on pMT153) was amplified using Taq polymerase and the outside primers pBRα_F and pBRα_R. The mutagenized cbtA alleles were cloned into the pAC-λCI fusion vector; mreB and ftsZ alleles were cloned into the pBRα fusion vector.
To identify λCI-CbtA variants with a decreased ability to interact with α-MreB, the λCI-cbtA mutant library was transformed into a modified two-hybrid reporter strain (BN30) bearing pBRα-MreB (pMT151). Strain BN30 contains an F’ episome bearing a two-hybrid reporter with the λCI operator positioned at -42, 20 bp closer to the transcription start site than in the standard two hybrid strain FW102 OL2-62. This positioning allows for an additional stabilizing contact between λCI and region 4 of σ70 bound to the -35 promoter element and results in an elevated level of lacZ expression [78], which afforded us a better color range for blue-white screening than our standard reporter. Transformants were plated on LB (KanCarbCm) indicator medium containing IPTG (25 μM) and X-gal (40 μg/mL). Plates were incubated overnight at 30°C and refrigerated (4°C) for an additional 8–16 h. α-MreB mutants with decreased λCI-CbtA interaction were isolated under identical conditions. For each screen, several thousand colonies were screened to identify those exhibiting lower lacZ expression (white or light blue color) as compared to the dark blue colonies producing wild-type α-MreB and λCI-CbtA fusions. For each screen, candidate mutants were counter-screened to identify those that maintained the ability to interact with a second partner protein (FtsZ, in the case of CbtA, and the NTD of RodZ, in the case of MreB). Specifically, colonies containing prospective λCI-CbtA mutants were pooled into a single overnight culture, grown at 30°C; a pooled plasmid prep generated from this overnight culture was transformed into FW102 OL2-62/pBRα-FtsZ. Transformants were plated on LB (KanCarbCm) indicator medium supplemented with 5 μM IPTG, 40 μg/mL X-gal, and 250 μM TPEG (a competitive inhibitor of β- galactosidase; Gold Biotechnologies); dark blue candidates were selected and the pACλCI-CbtA plasmids were isolated and fusion gene sequenced. For colonies containing prospective α-MreB mutants, individual cultures of candidate clones were grown overnight at 30°C. Plasmids were prepped from these cultures, most likely generating a mixed prep of α and λCI plasmids in each case. Individual mixed preps were used to transform FW102 OL2- 62 cells containing either pACλCI-CbtA or pACλCI-RodZNTD. Transformants were selected on LB (CmCarbKan). β-galactosidase assays (see below) were performed to measure interaction between each α-MreB mutant and λCI-CbtA (at 100 μM IPTG) or λCI-RodZNTD (at 25 μM IPTG). pBRα-MreB plasmids were isolated from candidates that were down for λCI-CbtA interaction but maintained >60% of the wild-type λCI-RodZNTD interaction; the fusion genes were sequenced and the mutant plasmids re-tested by β-galactosidase assay.
To identify α-FtsZ mutants with a decreased ability to interact with λCI-CbtA, our α-ftsZ mutant library was transformed into FW102 OL2-62/pACλCI-CbtA; several thousand colonies were screened on medium containing 5 μM IPTG and X-gal (40 μg/mL) at 37°C. Colonies that were pale blue or white were selected and the plasmids were isolated and transformed into FW102 OL2-62 cells containing either pACλCI-CbtA or pACλCI-FtsZ. Transformants were selected on LB (CmCarbKan). β-galactosidase assays were performed to measure interaction between each α-FtsZ mutant and λCI-CbtA (at 100 μM IPTG) or λCI-FtsZ (at 100 μM IPTG). Those candidates that exhibited at least a 60% decrease in interaction with CbtA but maintained greater than 75% FtsZ-FtsZ self-interaction were sequenced and further assayed for their interaction with λCI-ZipACTD by β-galactosidase assay.
To identify CbtA variants with a restored ability to interact with FtsZ-D180K, the λCI-cbtA mutant library was transformed into FW102 OL2-62 cells pre-transformed with pBRα-FtsZ-D180K, and transformants were plated on LB (KanCmCarb) indicator medium supplemented with IPTG (5 μM), X-gal (40 μg/mL), and TPEG (250 μM) (Gold Biotechnology). Plates were incubated at 30°C overnight. Several thousand colonies were screened in order to identify those that exhibited increased lacZ expression as compared to the pale blue control colonies producing wild-type λCI-CbtA and α-FtsZ-D180K. Dark blue candidate colonies were pooled into a single overnight culture, grown at 30°C. In order to identify those candidates that specifically interact with α-FtsZ-D180K, a pooled plasmid prep generated from this overnight culture was transformed into FW102 OL2-62 cells containing pBRα-FtsZ. Transformants were plated on the same indicator medium as before; pale blue candidates were selected and the pACλCI-CbtA plasmids were isolated and the fusion genes sequenced. The interaction between λCI-CbtA-V48E and all relevant α-FtsZ fusions was assayed by β-galactosidase assay.
All β-galactosidase assays were performed in our standard two-hybrid reporter strain, FW102 OL2-62. FW102 OL2-62 cells were co-transformed with plasmids encoding the relevant α and λCI fusions. Cultures inoculated with transformants were grown in 1 mL LB (KanCmCarb) in deep-well 96-well plates at 37°C, 900 rpm, 90% humidity in a Multitron incubation shaker (Infors HT) overnight. Overnight cultures were back diluted 1:100 or 1:40 in LB (KanCmCarb) supplemented with the appropriate concentration of IPTG in sterile microtitre plates (total volume of 200 μL); subcultures were grown, shaking at 37°C until they reached mid-log phase (OD600 0.4–0.8). A 100 μL aliquot of subculture was lysed by addition of 10 μL PopCulture reagent (Novagen) supplemented with rlysozyme (400 mU/μL). LacZ levels were determined by β-galactosidase assay performed in microtitre plates with a microtitre plate reader (Molecular Devices), as described in [83]. All assays were done in triplicate and were repeated independently at least twice. All Miller Unit values shown are from a single representative experiment and represent averages of triplicate measurements. Fold-change values were calculated by normalizing to the highest relevant empty vector control.
For E. coli spot dilution assays, strain BW27785 (or the appropriate derivative) was transformed with the appropriate plasmid(s). Transformants were selected either on LB supplemented with appropriate antibiotic at 30°C or 37°C, or, for assays done with DH118 strains, on M9 maltose (0.2% maltose, 0.2% casamino acids, 1 mM MgSO4) plates supplemented with 250 μM IPTG at 30°C. Overnight cultures (grown in either LB or M9 maltose + IPTG) were back diluted in fresh medium + antibiotics to a starting OD600 of 0.03–0.05 and grown at the indicated temperature until cultures reached an OD600 of 0.5–1. Cultures were normalized by OD600 value and 1:10 serial dilutions were made in fresh LB or sterile phosphate buffered saline (PBS) in a microtitre plate. 5 μL of each culture were spotted on LB plates containing the appropriate antibiotics with or without the indicated level of inducer (IPTG, arabinose, or ATC). See figure legends for details on each experiment.
For B. subtilis spot dilution analysis (Fig 5C), relevant strains were streaked from glycerol stocks onto LB (Spec) agar and incubated at 37°C overnight followed by additional overnight incubation at RT. LB cultures were inoculated with single colonies, which were grown at 37°C until they reached an OD600 ~1. Cultures were normalized by OD600 value, and 1:10 serial dilutions were made in fresh LB in a microtitre plate. 5 μL of each dilution were spotted on LB agar supplemented with 100 μg/mL spectinomycin and LB agar supplemented with 100 μg/mL spectinomycin and 1 mM IPTG. Plates were incubated overnight at 37°C. Spot dilution analysis was done on both LB Miller (1% NaCl) and LB Lennox (0.5%) agar with identical results.
For B. subtilis growth curves, 5 mL LB cultures (no antibiotics) were inoculated with single colonies of relevant strains. Several dilutions of these cultures were made and grown with shaking at RT overnight. The next day, cultures that were in early to mid-log phase were back diluted to a starting OD600 of 0.01 in 5 mL fresh LB medium supplemented with 1 mM IPTG. Growth was monitored every 30–60 min by transferring 200 μL to a sterile microtitre plate and taking OD600 measurements on a plate reader (Molecular Devices). All strains were grown in triplicate, and growth curve experiments were repeated independently several times. Growth curve cultures were imaged at the indicated times using the same microscopy protocol as described below.
Cultures used for microscopy were handled as described in the corresponding figure legends. Briefly, BW27785, DH73, NP1, or TB28 attHKHC488 cells were transformed with the relevant plasmids and transformants were selected for on LB containing the appropriate antibiotic(s). FB30 or DH118 transformants were selected for on M9 maltose supplemented with the appropriate antibiotic(s) and inducer. For most experiments, all growth incubation steps were done at 30°C. LB or M9 overnight cultures were back diluted to a starting OD600 of 0.02–0.05, grown without induction for 1 h (cultures had reached an OD600 of ~0.1–0.2), and then induced for toxin expression with the addition of IPTG (50, 100, or 200 μM), arabinose (0.2%), or ATC (10, 15, or 25 ng/mL). Cultures were typically in an OD600 range of 0.4–1 at the time of imaging.
For all snapshot images, cells were mounted on 2% agarose pads containing PBS, and microscopic observation was performed using an Olympus BX61 microscope (objective UplanF1 100x). Images were captured with a monochrome CoolSnapHQ digital camera (Photometrics) using Metamorph software version 6.1 (Universal Imaging). Cropping and minimal adjustment was performed with ImageJ [65] or Adobe Photoshop. Cell roundness quantification was done manually in ImageJ with the ObjectJ plugin. Briefly, the length of the cell was measured along the long axis, and the width was measured as the axis roughly perpendicular to the long axis. Angle measurements were spot-checked to ensure the axes intersected at an angle close to 90°. Cell roundness data were compiled from three independent experiments; 200–300 cells of each strain from each independent experiment were measured.
For time-lapse imaging of CbtA-induced morphology changes, pMT136, pMT139, and pMT146 were individually transformed into BW27785 cells. Overnight cultures were back diluted to a starting OD600 of 0.05 and grown at 30°C for 1 h without induction. Cells were concentrated 5x, and 2 μL were spotted on the bottom of a glass-bottomed dish (Willco dish HBSt-5040; Willco Wells). A 2% agarose pad containing LB growth medium supplemented with 100 μM IPTG was placed on top of the culture aliquot. Cells were imaged on a Nikon Ti inverted microscope using Nikon Elements software, a Photometrics CoolSNAP HQ2 Interline CCD camera, and a Well Plate Holder stage (TI-SH-W; Nikon) equipped with a humid, temperature-controlled incubator (TC-MIS; Bioscience Tools). The objective was heated to ~30°C using a Bioptechs objective heater system. Images were acquired every 3 min for 3 h. Image analysis was performed in FIJI and ImageJ.
To compare levels of His6-CbtA-GFP, MreB, or FtsZ variants in E. coli cells, 2 mL of each culture grown as described in the relevant figure legends were pelleted and resuspended in various amounts of BugBuster lysis buffer (EMD) to normalize by OD600 value (OD600 of 1 = 100 μL lysis buffer). rlysozyme (3 kU; EMD) and Omnicleave (20U; Epicentre) were added to equal volumes of cell suspensions for a typical final concentration of 60U/μL and 0.4U/μL, respectively. Cells were lysed at room temperature for 30 min. Total protein concentration was measured by Bradford assay, and lysate volumes were adjusted using lysis buffer such that all samples contained equivalent amounts of protein. To compare levels of His6-CbtA-GFP and His6-CbtA-F65S- GFP in B. subtilis cells, Western blot analysis was performed on whole-cell lysates generated from growth curve cultures. Briefly, 2 mL of each B. subtilis culture were pelleted and resuspended in various amounts of lysis buffer (20 mM Tris-HCl pH 7.5, 50 mM EDTA, 100 mM NaCl) to normalize by OD600 value (OD600 of 1 = 100 μL lysis buffer). rlysozyme (30kU) and Omnicleave (20U) were added to cell suspensions, which were lysed for 30 min at 37°C.
All lysates were diluted 1:2 in 2x Laemmli buffer + BME (final concentration 1%) and boiled for 10 min; further dilutions were made in 1x Laemmli buffer + BME. Duplicate 10–20% Tris-glycine gels (Thermo Fisher) were run in MOPS-SDS buffer and transferred to nitrocellulose membranes using a wet transfer system (Life Technologies). Membranes were incubated with primary antibodies α-GFP 1:5,000 (Roche), α-RpoA 1:10,000 (Neoclone), α-FtsZ 1:10,000 (T. Bernhardt), α-MreB 1:5,000 (T. Bernhardt), α-SigA 1:5,000 (D. Rudner), or α-Spo0J 1:5,000 (D. Rudner) and HRP-conjugated α-mouse or α-rabbit secondary antibodies (Cell Signaling). Chemiluminescent signal was detected using ECL Plus reagent (GE Healthcare) on a ChemiDock XRS+ system (Bio-Rad).
|
10.1371/journal.pgen.1002247 | Increased RPA1 Gene Dosage Affects Genomic Stability Potentially Contributing to 17p13.3 Duplication Syndrome | A novel microduplication syndrome involving various-sized contiguous duplications in 17p13.3 has recently been described, suggesting that increased copy number of genes in 17p13.3, particularly PAFAH1B1, is associated with clinical features including facial dysmorphism, developmental delay, and autism spectrum disorder. We have previously shown that patient-derived cell lines from individuals with haploinsufficiency of RPA1, a gene within 17p13.3, exhibit an impaired ATR-dependent DNA damage response (DDR). Here, we show that cell lines from patients with duplications specifically incorporating RPA1 exhibit a different although characteristic spectrum of DDR defects including abnormal S phase distribution, attenuated DNA double strand break (DSB)-induced RAD51 chromatin retention, elevated genomic instability, and increased sensitivity to DNA damaging agents. Using controlled conditional over-expression of RPA1 in a human model cell system, we also see attenuated DSB-induced RAD51 chromatin retention. Furthermore, we find that transient over-expression of RPA1 can impact on homologous recombination (HR) pathways following DSB formation, favouring engagement in aberrant forms of recombination and repair. Our data identifies unanticipated defects in the DDR associated with duplications in 17p13.3 in humans involving modest RPA1 over-expression.
| The widespread use of genomic array technology has lead to the identification of a plethora of novel human genomic disorders. These complex conditions occur as a consequence of structural genomic alterations (deletions, amplifications, complex rearrangements). Understanding the specific consequences of such alterations on gene expression and unanticipated impacts on biochemical pathways represents an important challenge to help untangle the clinical basis of these conditions and ultimately aid in their management. Here, we demonstrate that individuals with specific duplications of 17p13.3 incorporating RPA1 exhibit modest over-expression of RPA1. Unexpectedly, this is associated with elevated levels of genomic instability and sensitivity to DNA damage. RPA1 is a component of the Replication Protein A heterotrimer, a complex that plays fundamental roles in DNA replication, repair, and recombination. Reduced RPA1 levels are associated with impaired DNA damage checkpoint activation, but the cellular impacts of over-expression of this subunit have not previously been described in the context of a genomic disorder. Using model cell and reporter systems, we show that modestly elevated levels of RPA1 can adversely impact on DNA double-strand break–induced homologous recombination resulting in elevated levels of chromosome fusions. This data highlights an unanticipated consequence of copy number variation on genomic stability.
| Variously sized contiguous deletions within 17p13.3-pter are associated with complex clinical features in humans including structural brain abnormalities (lissencephaly, agyria, microcephaly), growth retardation and developmental delay [1]. Multiple pathogenomic studies have identified haploinsufficiency of genes including PAFAH1B1 (LIS1) and YWHAE (14-3-3ε) as being particularly relevant in this context [2]–[5]. Previously, we have shown that patients with haploinsufficiency of RPA1 exhibit defective ATR-dependent DDR including failure of the G2-M cell cycle checkpoint suggesting RPA1 is sensitive to copy number variation [6]. Defective ATR-dependent G2-M arrest is associated with human conditions characterised by severe microcephaly (e.g. Seckel syndrome, Microcephalic primordial dwarfism type II, MCPH1-dependent Primary microcephaly, Nijmegen breakage syndrome) [7]. RPA1 (RPA1: RPA-70KD) encodes the largest subunit of the Replication Protein A complex, a heterotrimeric complex (RPA1-2-3: RPA-70KD-RPA-32KD-RPA14KD respectively) with single stranded DNA binding capability that appears to be involved in multiple DNA transactions. It functions to prevent unregulated nuclease digestion and/or hairpin formation as well as orchestrating the sequential assembly and disassembly of various DNA processing factors during DNA replication, repair and recombination [8]–[10]. With respect to the DDR, the DNA single stranded binding function of RPA1–3 plays a fundamental role in the recruitment of ATR to sites of DNA damage, for example stalled replication forks, via a direct interaction with ATR's binding partner, ATRIP [11]. Furthermore, through interactions with RAD51 and RAD52, RPA1–3 also plays an essential role in homology directed recombinational repair, likely facilitating RAD51 nucleofilament formation allowing strand invasion and homology searching [12]–[16].
Recently, distinct, variously sized, non-recurrent duplications within 17p13.3 have been identified in several individuals defining a novel genomic disorder. In two of these the duplication included RPA1 [17]. Consistent with other genomic disorders, the clinical duplication phenotype appears to be less severe compared to deletions within 17p13.3. Nevertheless, subtle over-expression of ‘normal’ genes within 17p13.3 is associated with profound clinical consequences [17]–[19]. Interestingly, over-expression of RPA1 has been implicated in genomic instability in other systems. For example, a quantitative over-expression screen in the budding yeast Saccharomyces cerevisiae found that over-expression of RFA1, the S. cerevisiae equivalent of mammalian RPA1, was associated with delayed cell cycle progression through G2-M, impaired chromosomal spindle attachment and activation of the DDR [20]. Furthermore, ectopic over-expression of individual RPA1–3 subunits in the human colorectal carcinoma cell line HCT116 promoted endoreduplication and aneuploidy [21]. Whether RPA1–3 over-expression functionally contributes to any cellular or clinical phenotype associated with genomic disorders has not been investigated. Since we had previously observed specific DDR-defects associated with reduced RPA1 expression in cell lines derived from individuals with variously sized contiguous deletions at 17p13.3-pter, we sought to determine if increased levels of RPA1 are associated with identical and/or related DDR-defects [6]. Herein, we show that cell lines derived from patients with 17p13.3 duplications that encompass RPA1 exhibit modest RPA1 over-expression, abnormal S phase distribution, attenuated DSB-induced RAD51 chromatin retention and enhanced sensitivity to killing by camptothecin, consistent with compromised homologous recombination (HR). Using various model and reporter systems we demonstrate that subtle over-expression of RPA1 is indeed associated with altered HR-mediated DNA double strand break repair.
Two of the 17p13.3 duplication cases recently described by Bi et al involve genomic duplication of RPA1, amongst other genes [17]. A schematic representation of the various CNVs in this region in several cell lines used in this study is shown in Figure 1A. The cell lines involving RPA1 duplication, BAB2668 and BAB2719, are shown in red (Figure 1A).
We examined RPA1 expression by western blotting following careful titration of whole cell extracts to ascertain the extent of over-expression at the protein level using EBV-transformed lymphoblastoid cells (LBLs) from both RPA1-duplication cases, compared to another LBL with a 17p13.3 duplication that does not involve RPA1 (BAB2721), as reported by Bi et al [17], and to an LBL (BAB2751; Case 6 [17]) exhibiting a novel 17p13.3 genomic deletion involving haploinsufficiency of RPA1 (Figure 1B). The left-hand panel of Figure 1B shows that RPA1 protein is modestly over-expressed in whole cells extracts from BAB2668 (Case 4 [17]) and BAB2719 (Case7 [17]) compared to BAB2721, an LBL from a patient who does not exhibit RPA1 duplication at the genomic level. LBLs from patient BAB2751 (Del; deleted for one copy of RPA1) associated with genomic haploinsufficiency of RPA1 show modestly reduced RPA1 expression at the protein level. Interestingly, modest over-expression of RPA2 was also evident in whole cell extracts from BAB2719 (Dup) and BAB2668 (Dup) (Figure 1B middle panels). This suggests that the 17p13.3 duplications involving RPA1 and resulting in RPA1 over-expression in these cells likely also results in elevated levels of the RPA complex since RPA2 levels appear modestly elevated in these cells (Figure 1B and 1C). Quantification of three separate experiments relative to MCM2 is shown in Figure 1C. Similar data from the other LBLs as described in Figure 1A is shown in Figure S1.
An important role of RPA1–3 in the ATR-dependent DDR is the recruitment of ATR-ATRIP to single stranded DNA (ssDNA) generated at the DNA damage site, thereby initiating ATR-dependent signalling [11], [22]. One aim of this process is the activation of cell cycle checkpoint arrest, particularly at the G2-M transition. Previously, we have shown that Miller-Dieker Syndrome (MDS) and severe Isolated Lissencephaly Sequence (ILS+) patient-derived LBLs with RPA1 haploinsufficiency fail to activate the ATR-dependent G2-M checkpoint [6]. Furthermore, we showed that this cellular phenotype was RPA1-dependent since it could be complemented by ectopic expression of RPA1 following transfection [6]. Interestingly, precedent exists whereby over-expression of a DDR-component is actually associated with a functional defect in the DDR [23]–[28]. Nevertheless, we did not observe a defective ATR-dependent G2-M cell cycle checkpoint arrest in LBLs derived from individuals with increased RPA1 levels associated with RPA1 duplication (Figure 2A). This was in contrast to LBLs exhibiting RPA1 haploinsufficiency with reduced RPA1 expression (BAB2751; Del; deleted for one copy of RPA1. Figure 2A and [6]). Therefore, modest over-expression of RPA1 in the context of 17p13.3 duplication is not associated with the same ATR-dependent DDR-defect as that of RPA1 haploinsufficiency.
Since ectopic over-expression of RPA1 has previously been shown to induce other forms of genomic instability including aneuploidy, we examined spindle assembly checkpoint (SAC) proficiency following prolonged exposure to the spindle poison nocodazole in RPA1-duplicated patient-derived LBLs [21]. Following 24 hrs treatment with 1.5 µM nocodazole, cells with a functional SAC exhibit an increased 4N population without any progression to >4N, as demonstrated by the propidium iodide staining flow cytometry profiles shown in Figure 2B (Unt; untreated. Noc; nocodazole treated). Quantification of the 4N population, with or without 24 hrs treatment with nocodazole, demonstrates that BAB2752 (WT; wild-type RPA1 copy number), BAB2668 (Dup; RPA1 duplication) and BAB2719 (Dup) all exhibit a similar arrest at 4N following nocodazole (Figure 2C). No increase in >4N was seen in either of the RPA1-duplicaiton containing LBLs, BAB2668 or BAB2719 (Figure 2B). Hence, we observed a normal nocodazole-induced arrest at mitosis with 4N DNA content suggestive of a proficient SAC in this context.
Ectopic over-expression of RPA1 is associated with endoreduplication in certain cell lines [21]. Since RPA1–3 complex is a fundamental component of normal DNA replication we examined S phase in one of our patient-derived LBLs with RPA1 duplication (BAB2668) using bromodeoxyuridine (BrdU) pulse-labelling-coupled two-dimensional flow cytometry (Figure 3A). No evidence for spontaneous endoreduplication was found (data not shown). Whilst we did not observe a difference in the overall amount of BrdU incorporated between patient-derived LBL BAB2668 (Dup; with RPA1 duplication) compared to those with normal (BAB2752;WT) or haploinsufficient RPA1 copy number (BAB2751;Del) (Figure 3A left hand graph), the distribution or pattern of BrdU labelling was specifically and reproducibly altered in BAB2668 LBLs with RPA1 duplication (Figure 3A middle flow cytometry panels). The BrdU positive cells within the boxed area represent those that have DNA content between 2N and 4N (mid S phase) but have not incorporated BrdU efficiently. These cells (i.e. mid S phase yet low BrdU incorporation) are approximately 3–4 fold more abundant in BAB2668 (Dup; RPA1 duplicated) compared to BAB2752 (WT; RPA1 copy no) or BAB2751 (Del; haploinsufficient RPA1) (Figure 3A right hand graph). This is suggestive of a stochastic problem in S phase progression or DNA replication in unperturbed asynchronously growing LBLs with RPA1 duplication.
We next examined the ability of our patient-derived LBLs to recover DNA replication following prolonged treatment with the DNA polymerase inhibitor aphidicolin (APH). APH treatment for 24 hrs efficiently reduced total BrdU incorporation as expected for all cell lines (Figure 3B upper panels). When this APH-induced DNA replication block was removed we found that RPA1-duplication associated LBL BAB2668 (Dup) failed to progress as efficiently as the control LBL BAB2752 (WT) through S phase, as judged by the distribution of BrdU positive cells in late S phase as indicated by the boxed area in Figure 3B (lower panels and graph). This is consistent with a constitutional problem in the ability to efficiently complete DNA replication, in this case following recovery from replicative stress, in these patient-derived cells. Interestingly, RPA1 haploinsufficiency (BAB2751) conferred a similar phenotype (Figure 3B).
Since the RPA1–3 complex is also an important functional component of HR, we sought to examine HR in our patient-derived LBLs. Furthermore, defective HR has previously been shown to result in impaired S phase progression, a phenotype suggested by our BrdU incorporation data (Figure 3) [29], [30]. Following IR treatment, we found modestly increased chromatin binding of RPA1, RPA2 and RAD51 in LBLs with normal RPA1 copy number (BAB2705; WT), in contrast to chromatin extracts from RPA1-duplicated BAB2668 LBLs (Dup) (Figure 4A and 4B). In fact, BAB2668 (Dup) exhibited increased endogenous levels of chromatin bound RPA1, RPA2 and RAD51, even in undamaged cells, in contrast to the WT LBLs, but this level did not change following IR (Figure 4A and 4B). Protein quantifications standardised to histone H2B loading and normalised to the un-irradiated BAB2705 (WT) for each of RPA1, RPA2 and RAD51 from three separate experiments are shown in Figure 4B. The attenuated IR-induced RAD51 chromatin retention observed in BAB2668 (Dup) indicates a potential problem in the ability to induce RAD51-dependent HR in these cells following IR-induced DSB formation.
Since the patient-derived LBLs with increased RPA1 copy number are also duplicated for other genes, it was important to examine whether indeed RPA1 is the protein conferring this phenotype or whether it is in fact the consequence of combined increased copy number of several genes. To demonstrate that this cellular phenotype was specifically associated with RPA1 over-expression, we constructed a conditional, isopropyl β-D-1-thiogalactopyranoside (IPTG)-inducible-RPA1 model system in the human glioblastoma line T98G based on the pTUNE vector (Origene). Interestingly, we found that transient ectopic over-expression of RPA1 from a high expression level CMV-promoter-containing pcDNA3.1 mammalian expression vector was consistently associated with overt toxicity (associated with detectable activated caspase 3; data not shown) in multiple commonly employed human tumour lines (e.g. HeLa, MG63, A549) suggesting that strong over-expression of RPA1 is not tolerated. We found significant leakiness associated with the T98G-RPA1 system (pTUNE-RPA1) as RPA1 and RPA2 levels appeared increased even in the absence of IPTG (Figure 4C). Nevertheless, IPTG treatment did induce a modest elevated expression of RPA1 in this context (Figure 4C Unt; untreated. IPTG; IPTG treated). Interestingly, this was also associated with elevated RPA2 expression (Figure 4C), similar to what was found in the patient-derived LBLs (Figure 1B and 1C).
We found that the modest IPTG-induced over-expression of RPA1 was associated with attenuated IR-induced chromatin retention of both RPA1 and RAD51 (Figure 4D). Attenuated IR-induced RAD51 chromatin retention is also a feature of the patient-derived LBLs (Figure 4A and 4B). Furthermore, and similar to LBLs exhibiting RPA1 duplication (BAB2668; Figure 4A), we observed more chromatin associated RPA1 even in undamaged cells following induction with IPTG (Figure 4D). Collectively these data suggest that modest over-expression of RPA1 is associated with attenuated RAD51 chromatin recruitment following IR treatment and that this cellular phenotype is observed in LBLs from patients with duplications in 17p13.3 involving RPA1 (Figure 4A).
To gain direct, independent insight into the consequences of subtle RPA1 over-expression on DSB-induced HR, we exploited I-Sce I restriction enzyme-induced HR using the established model DRneo reporter system in Chinese Hamster Ovary (CHO) cells following transient over-expression of human native untagged RPA1. Unlike other commonly used model HR reporter systems (e.g. DR-GFP), we opted to use this set-up specifically because the DRneo system enables the collective assessment of alternative forms of recombination alongside gene conversion (GC), such as single strand annealing (SSA) (Figure 5A) [31]–[33]. This is a heterologous system, although similar approaches have been used successfully before to study HR-mediated DSB-repair ([30] and references therein). Unfortunately, since the available RPA1 antibodies fail to cross-react with hamster RPA1, it was difficult to determine the precise extent of RPA1 over-expression following transient transfection. Nevertheless, transient expression of the human protein under these conditions appeared only modestly greater than endogenous RPA1 expression in T98G cells (Figure 5B). Interestingly, RPA1 expression in this context did not appear to grossly affect GC, although the limitations of such as heterologous system should be kept in mind (Figure 5C; white bars). Nevertheless, a reduced expression-induced RPA-dependent phenotype has been shown to give a different outcome using a similar system arguing against a simple dominant-negative effect here [16]. However, unexpectedly, we observed an approximately 2-fold increase in total levels of HR (i.e. all forms of GC+SSA; black bars) following I-Sce I-induced DSB formation (Figure 5C and 5D). This implies that RPA1 over-expression following I-Sce I-induced DSB results in increased forms of recombination such as SSA and/or GC with crossing-over which can be regarded as an aberrant or less favourable forms of recombination since they are associated with loss of genetic material [29]. Interestingly, RPA has recently been shown to be required for SSA in Xenopus [34]. Furthermore, increased RAD51 expression is also associated with increased genomic instability using a similar HR reporter system suggesting that over-expression of functional components of repair pathways likely to be involved in repairing such DSB's can adversely affect repair [35]. These data are consistent with attenuated IR-induced RAD51 chromatin recruitment observed in the RPA1-duplication associated patient-derived LBLs (Figure 4A) and the T98G-RPA1 system (pTUNE-RPA1; Figure 4D). Collectively, these results suggest that modest increased expression of RPA1 can influence HR sub-pathway choice.
Our findings with the model HR reporter system suggest that increased RPA expression was associated with increased recombination leading to the hypothesis that increased RPA expression could be associated with increased genome instability. Similarly, since RAD51 over-expression can also induce significant genomic instability [23]–[28], we examined mitotic spreads of our RPA1-over-expression model cell line (pTUNE-RPA1) for evidence of elevated genomic instability. Strikingly, significant levels of chromosome aberrations, fusions/derivatised chromosomes in particular, were observed in these cells (Figure 6A and 6B). Such abnormalities would be consistent with aberrant cross-over and/or ligation events. These aberrations were seen even without induction with IPTG which is a further indication of some inherent leakiness in this system and is consistent with elevated chromatin bound RPA1 seen here (Figure 4D). Despite the limitations of this artificial cell system, these data do demonstrate that subtle over-expression of RPA1 can induce significant levels of genomic instability, specifically elevated levels of derivatised chromosomes.
To examine whether the patient-derived LBLs with RPA1 over-expression exhibited a similar phenotype we examined mitotic spreads for chromosomal abnormalities in LBLs from BAB2719 (Dup; RPA1 duplication) compared to BAB2752 (WT; wild-type normal RPA1 copy number). We also observed elevated levels of chromosomal aberrations, specifically an over-representation of chromosomal fusions, in LBLs associated with RPA1 over-expression (Figure 7A and 7B). These aberrations were increased following IR-treatment further suggestive of an inability of these cells to properly repair DSBs (Figure 7B).
Defective and/or aberrant DNA damage-induced HR is also associated with DNA damaging-induced genomic instability. Consistent with this we found increased levels of hydroxyurea (HU)-induced micronuclei (MN) formation in LBLs associated with RPA1-duplication (Dup; BAB2668, BAB2719) indicative of increased DNA breakage following replication fork stalling in these cells (Figure 7C). Furthermore, compromised HR is also specifically associated with sensitivity to killing by topoisomerase I inhibitors such as camptothecin (CPT) [36]. Consistent with an underlying problem with HR associated with RPA1-duplication in our patient-derived LBLs, we also found that these lines were sensitive to apoptosis induction following CPT treatment, as judged by increased levels of sub-G1 cells by propidium iodide flow cytometry (Figure 7D). Interestingly, for both of these cellular phenotypes we observed a similar response in BAB2751 (RPA1 haploinsufficient) cells to that of lines over-expressing RPA1 (Dup; BAB2668 and BAB2719). This suggests that manipulation of RPA1 levels (increase or decrease) results in increased genomic instability following DNA damage.
In summary, we found that duplications involving RPA1 are associated with modest over-expression of RPA1 and also RPA2 at the protein level, impaired S phase distribution and spontaneously elevated levels of chromatin bound RPA1, RPA2 and RAD51, along with attenuated IR-induced RAD51 chromatin retention following DSB's suggestive of compromised HR. Using the DRneo model HR-reporter system we observed a hyper-recombinogenic phenotype consistent with a shift towards a less genomically preferable form of HR following modest RPA1 over-expression. We also found increased levels of complex rearrangements especially after DSB-induction in patient derived LBLs with RPA1 duplication. Furthermore, these patient derived cells exhibit other evidence of underlying problems in the DDR such as sensitivity to CPT and elevated HU-induced micronuclei formation.
Variously sized contiguous gene deletions at 17p13.3 are associated with severe neurodevelopmental phenotypes including microcephaly and neuronal migration deficits [1]. Recently, duplications within 17p13.3 have been identified in several patients exhibiting a milder though distinct phenotype that also incorporates aspects of autism spectrum disorder [17]–[19]. Much attention has focused on characterising the consequence of CNV of PAFAH1B1/LIS1 in this respect [17]–[19], [37]. Previously, we have shown that LBLs from some ILS+ individuals and from MDS patients, all of whom exhibit haploinsufficiency of RPA1, a gene telomeric to PAFAH1B1/LIS1, exhibit impaired ATR-dependent DDR [6]. Here, we find that for the reciprocal situation, that is, in LBLs from patients associated with duplication of RPA1, we observed a distinct DDR abnormality impacting upon HR.
The RPA1–3 complex is a fundamental functional component of many DNA processes involving the generation of single stranded DNA [8]. RPA1–3 complex is essential for several DNA repair pathways (e.g nucleotide excision repair, mismatch repair, base excision repair), for DNA replication and recombination events [12]–[16]. Therefore, a plausible assumption would be that a significant reduction in RPA expression/function results in embryonic lethality. Attempts to create knockout mice for RPA1 have not been reported. Nevertheless, mice bearing a semi-dominant heterozygous mis-sense mutation in Rpa1 (Rpa1L230P) exhibit gross genomic rearrangements and are highly cancer prone (Rpa1L230P homozygosity is cell lethal) [38]. Hence, precedent exists for altered RPA1, and likely, consequently RPA complex function, impacting on genomic stability at the organismal level. Furthermore, forced over-expression of RPA1 can cause genomic instability, at least in cancer cell lines [21].
There are several instances whereby over-expression of various DDR and/or cell cycle components disrupts or adversely affects the fundamental cellular processes/pathways in which they are functional components. For example, over-expression of CDC25A phosphatase is thought to be an important contributor to uncontrolled cell cycle progression from G2 into M frequently observed in certain malignancies [24], [25]. Over-expression of RAD51 and RAD52 has been found to reduce DSB-induced HR in mammalian cells [28]. Indeed, over-expression of separase or the SAC component MAD2 results in aneuploidy and malignancy in mice, consistent with defective SAC activity [26], [27]. Our findings suggest that a modest over-expression of RPA1 in LBLs derived from individuals with duplications in 17p13.3 involving RPA1 results in an abnormal distribution of cells in S phase, adversely impacts on HR and is associated with elevated chromosomal instability and sensitivity to DNA damaging agents.
RPA1 is thought to be of particular importance for RPA heterotrimeric function since it can bind DNA independently of the other subunits and contains the greatest surface area available to mediate protein-protein interaction [39]. Interestingly, we found that RPA2 also appeared to be over-expressed in our patient-derived LBLs associated with RPA1 duplication, potentially suggesting elevated levels of RPA complex in this context. This could have adverse implications for coordinating subsequent DNA processing pathways. For example, during HR, a ‘handover’ between RPA1–3 complex coated ssDNA and RAD51 must occur to allow RAD51 nucleofilament formation for strand invasion. As RPA1 can bind RAD51 directly, an excess of chromatin bound RPA complex could interfere with the timing, coordination and/or efficiency of this ‘handover’ (Figure 8). A direct consequence of this could be either uncontrolled elevated or reduced overall HR capacity and/or a preference for other forms of recombinational repair, aside from gene conversion (GC). Data from our patient-derived LBLs show spontaneously elevated RAD51 on chromatin but attenuated IR-induced recruitment. Furthermore, our data generated following transient RPA1 over-expression in the DRneo system indicates a hyper-recombinogenic phenotype (i.e. total HR increases whilst levels of GC without crossing-over remain fairly constant). Interestingly, RPA has recently been shown to be required for SSA, at least in Xenopus [34]. One possible interpretation of the DRneo system-derived data is a shift towards an elevated level of SSA and/or GC with crossing-over, both of which involved loss of genetic material. The elevated levels of derivatised chromosomes observed in mitotic spreads from the RPA1-duplication associated LBLs and in the pTUNE-RPA1 system cells are consistent with aberrant cross-over and/or ligation events. A hyper-recombinogenic phenotype can have serious consequences for genome stability. For example, elevated ‘mutagenic’ HR has been implicated as pathophysiological contributor to disease progression in haematological malignancies such as Chronic Myelogenous Leukaemia and Multiple Myeloma [40]–[42].
The complex clinical spectrum of 17p13.3 microduplication syndrome is undoubtedly a consequence of the combined increased copy number of several genes within 17p13.3, although some candidates may have greater impacts than others [17]–[19]. The specific pathological connection between increased RPA1 expression and the clinical features of those respective patients is unclear. Nevertheless, the cellular phenotypes described here, including impaired S phase and suboptimal HR, could adversely influence apparently unrelated biological pathways by affecting gene expression. For example, components of the DNA replication machinery have been shown to influence epigenetic control of gene silencing [43]. Furthermore, suboptimal/aberrant HR could conceivably ultimately alter the genomic architecture resulting in unanticipated cis and/trans effects on the expression of other genes. Interestingly, RPA1 has been implicated in such ‘allelic phasing’ together with TP53 with respect to carcinogenesis [44]. Congenitally elevated genomic instability is often associated with cancer predisposition, although this has not been noted in either 17p13.3 duplication syndrome patients associated with RPA1-duplication [17]. Obviously there are too few patients to make any definitive conclusions, although the cellular defects presented here may warrant consideration in this respect. Clearly, further work is required to untangle the clinical consequences of increased RPA1 expression.
In conclusion, we have found that LBLs derived from patients with duplications in 17p13.3 specifically incorporating RPA1 exhibit a modest over-expression of RPA1 and RPA2 which is associated with attenuated S phase transit, attenuated IR-induced RAD51 chromatin recruitment, elevated chromosomal instability, increased HU-induced MN formation and sensitivity to killing by CPT. All of these phenotypes are consistent with an inefficient HR pathway. Furthermore, using various model cell systems we showed that modest conditional over-expression of RPA1 alone impacts on IR-induced RAD51 chromatin retention and I-SceI-induced HR in a reporter construct, the latter phenotype indicative of a hyper-recombinogenic shift towards alternative forms of recombination coincident with elevated chromosomal fusions. Collectively, our findings highlight a novel association between impaired DDR and CNV resulting in copy number gain of RPA1.
EBV-transformed patient-derived lymphoblastoid cell lines (LBLs) were cultured in RPMI with 15% FCS, L-Gln and antibiotics (Pen-Strep) at 5% CO2. T98G glioblastoma cells were maintained in MEM supplemented with 10% FCS, pyruvate and non-essential amino acids. Chinese Hamster Ovary cell lines (CHOs) were cultured in 10% DMEM, L-Gln and antibiotics (Pen-Strep) at 5% CO2.
Urea extraction: Cells were lysed in 150 µl urea buffer (9 M urea, 50 mM Tris-HCl at pH 7.5 and 10 mM 2-mercaptoethanol), followed by 15 s sonication, 30% amplitude using a micro-tip (SIGMA-Aldrich). The supernatant was quantified by Bradford Assay.
Detergent lysis: Cell pellets were incubated for 1 hr on ice in buffer containing 50 mM Tris-HCl pH 7.5, 150 mM NaCl, 2 mM EDTA, 2 mM EGTA, 25 mM NaF, 25 mM β-glycerolphosphate, 0.1 mM Na-orthovanadate, 0.2% Triton X-100, 0.3% IGEPAL and protease inhibitor cocktail tablets as indicated by manufacturer (Roche). The supernatant was quantified by Bradford Assay.
Cells were harvested 24 hr after 10 Gy gamma irradiation. Gamma irradiation was performed using a 137Cs γ-ray source at a dose rate of 8 Gy/min. Cells were lysed in detergent lysis buffer (above) for 1 hr on ice followed by 15 min in high-salt IP buffer (an extra 500 mM NaCl added to the regular IP buffer). The cell pellet was re-suspended in urea buffer (see WCE above) and sonicated for 15 s. The supernatant was quantified by Bradford Assay.
Western blots were developed using ECL (Pierce) in a luminescent image analyser, Image Quant LAS 4000 (GE Healthcare). This analyser ensures all bands are in the linear range (during the developing any saturated bands are highlighted so that the exposure can be decreased). Image Quant TL 7.01 quantification software was used to quantify the band intensities.
Alternatively, following ECL, Western blots were developed using film and the scanned images quantified with Image J software.
Custom assembled pTUNE-RPA1 was obtained from Origene and stable T98G clones were obtained following transfection with MetafectenePro (Biontex Laboratories GmbH) and selection in G418 (1 mg/ml). For inductions, cells were treated with 500 µM IPTG for 3 hrs.
Anti-RPA1 (Ab-1 #NA13) and anti-RPA2 (Ab-2 #NA18) antibodies were from Calbiochem. Anti-RAD51 (H-92) was from Santa Cruz. Anti-H3 was from Cell Signaling and anti-H2B was from Millipore. Anti-BrdU-FITC conjugated antibody (347583) was from Becton Dickinson.
UV irradiation was carried out using a UV-C source (0.6 J/m2/s). Cells were irradiated with 5 J/m2 UV-C in PBS and immediately seeded into complete medium supplemented with 1.5 µM nocodazole for 24 hr. Cells were pelleted, swollen with 75 mM KCl for 10 min before fixing in Carnoy's solution (methanol: glacial acetic acid 3∶1), before counterstaining with 4′-6-Diamidino-2-phenylindole (DAPI). Cells were Cytospun (Shandon) onto poly-L-lysine coated slides and mounted with Vectashield (Vector Labs). Slides were scored using a Zeiss AxioPlan microscope.
Exponentially growing LBLs were treated with 1.5 µM nocodazole for 24 hrs then fixed in 70% ice cold ethanol prior to propidium iodide staining and analysis by flow cytometry.
LBLs were fixed in ice-cold 70% ethanol for 24 h and re-suspended in PBS containing 0.5% Tween-20, 10 µg/ml propidium iodide and 500 µg/ml RNase A. Data were collected using a Becton Dickinson FACS Calibur machine and were analysed with CellQuest software. For BrdU incorporation cells were labelled with 50 µM BrdU for 15 min. Incorporated BrdU was detected using FITC-conjugated anti-BrdU antibody (Becton-Dickson).
ERCC1.17 DRneo CHO cells were grown in 10% DMEM supplemented with 3 µg/ml Blasticidin-S and 0.05 mM hygromycin B [33]. Assay: 5×105 cells were seeded in 6 cm plates. Next day cells were co-transfected with 2 µg RPA1 and 2 µg I-SceI (or CMV control) using MetafectenePro, according to the manufacturer's protocol. 24 hr after transfection cells were put into selection. For the recombination frequencies, 5×104 cells per 10 cm plate were seeded with 1 mg/ml G418 and/or 0.5 mg/ml hygromycin-B. 103 cells were seeded to determine the cloning efficiency. Plates were incubated for 7 days, after which they were stained with methylene blue.
Exponentially growing LBLs were treated with 10 µM camptothecin (CPT) and incubated for 72 hrs, fixed (ice cold 70% ethanol), stained with propidium iodide and sub-G1 cells quantified by flow cytometry.
Cells were treated with 1 mM HU for 4 hrs before incubation for 24 hrs in 5 µg/ml cytochalasin B for bi-nucelate formation. Micronuclei were scored in bi-nucleated cells by immunofluorescence microscopy (Zeiss AxioPlan) following swelling in KCl (75 mM 10 mins), fixation (Carnoy's fix; 3∶1 methanol∶acetic acid. 10 mins) and staining with DAPI and acridine orange (2 µg/ml).
The pTUNE-RPA1 T98G cells, and wild type T98G cells, were induced with 500 µM IPTG for 3 hr before adding 0.2 µg/ml colcemid for 4 hr prior to harvesting. Cells were swollen (75 mM KCl 10 mins) and then fixed in Carnoy's fixative (10 mins) prior to being dropped onto slides from approx 50 cm above. The slides were air dried and Giemsa stained according to the manufacturer's (Sigma) protocol. Images were captured on a Zeiss AxioPlan microscope. Chromosomes spreads were scored blinded according to the following criteria; fusions between different chromosomes, breaks, branched structures and ‘other’ (a terminal fusion within a single chromosome). The results were represented as aberrations per 100 chromosomes, rather than per metaphase due to the aneuploid nature of T98G. LBLs were treated with 2 Gy ionising radiation (IR) and allowed to recover for 24 hrs. The IR treated (IR) and untreated control (Unt) cells were treated with 0.2 µg/ml colcemid for 4 hr prior to harvesting. Cells were swollen (75 mM KCl 10 mins), fixed (Carnoy's), Giemsa stained and analysed as above.
|
10.1371/journal.ppat.1003485 | Prion Replication Occurs in Endogenous Adult Neural Stem Cells and Alters Their Neuronal Fate: Involvement of Endogenous Neural Stem Cells in Prion Diseases | Prion diseases are irreversible progressive neurodegenerative diseases, leading to severe incapacity and death. They are characterized in the brain by prion amyloid deposits, vacuolisation, astrocytosis, neuronal degeneration, and by cognitive, behavioural and physical impairments. There is no treatment for these disorders and stem cell therapy therefore represents an interesting new approach. Gains could not only result from the cell transplantation, but also from the stimulation of endogenous neural stem cells (NSC) or by the combination of both approaches. However, the development of such strategies requires a detailed knowledge of the pathology, particularly concerning the status of the adult neurogenesis and endogenous NSC during the development of the disease. During the past decade, several studies have consistently shown that NSC reside in the adult mammalian central nervous system (CNS) and that adult neurogenesis occurs throughout the adulthood in the subventricular zone of the lateral ventricle or the Dentate Gyrus of the hippocampus. Adult NSC are believed to constitute a reservoir for neuronal replacement during normal cell turnover or after brain injury. However, the activation of this system does not fully compensate the neuronal loss that occurs during neurodegenerative diseases and could even contribute to the disease progression. We investigated here the status of these cells during the development of prion disorders. We were able to show that NSC accumulate and replicate prions. Importantly, this resulted in the alteration of their neuronal fate which then represents a new pathologic event that might underlie the rapid progression of the disease.
| Prion diseases are irreversible progressive neurodegenerative diseases, leading to severe incapacity and death. They are considered to be caused by an abnormally folded infectious protein named PrPSc. They are characterized in the brain by prion amyloid deposits, vacuolisation, astrocyte proliferation, neuronal degeneration, and by cognitive, behavioural and physical impairments. There is no treatment for these disorders. Transplantation of stem cells, or the stimulation of endogenous neural stem cells (NSC) in the adult brain therefore constitute new interesting and promising strategies. While our first interest was the development of a cell therapy approach, we rapidly realised that there were a lot of questions to address before investigating pre-clinical cell therapy assays. Some of them were: (i) what is the status of endogenous neural stem cells during the development of prion diseases and can they amplify prions, and (ii) can they still proliferate and give rise to new neurons. In this study we definitely demonstrate that PrPSc was not only able to replicate in adult neural stem cells derived from infected brains but also that this results in an impairment of the production of their neuronal derivatives.
| Prion diseases or transmissible spongiform encephalopathies (TSEs) are fatal neurodegenerative disorders, which include Creutzfeldt-Jakob disease in humans, scrapie in sheep and goats, and bovine spongiform encephalopathy in cattle. Their origin can be genetic, sporadic or infectious and there is currently no available treatment preventing the widespread neurodegeneration occurring in these disorders. TSEs are pathophysiologically characterized by the accumulation in the brain of a pathogenic abnormal isoform of a protein termed PrP scrapie (PrPSc) [1]. According to the prion hypothesis, the infectious isoform PrPSc can trigger the autocatalytic conversion of the neuronal host-encoded PrPC into PrPSc [2] through a poorly understood misfolding process [1], rendering the progression of the disease dependent upon PrP expression. Several studies have reported early, severe and selective loss of GABAergic interneurons in prion diseases [3], [4]. These specific changes in neuronal subset may underlie some of the clinical symptoms in prions. The diagnosis of these diseases is difficult and often leaves only a short therapeutic window after the appearance of the first clinical signs [5]. Although important efforts have been made to understand the physiopathogenesis of neurodegenerative disorders, Prion diseases are still incurable and new therapeutic approaches such as cell therapy need to be explored. As a matter of fact, the widespread existence of endogenous neural stem cells (NSC) in the adult brain [6], [7] offers hope that these endogenous cells may be harnessed to repair cellular damages caused by brain injuries. During the past decade, several studies have consistently shown that (i) NSC reside in the adult mammalian CNS and that (ii) adult neurogenesis occurs throughout the adulthood in the subventricular zone (SVZ) of the lateral ventricle (LV) or the Dentate Gyrus (DG) of the hippocampus (H). Accumulating evidences have clearly shown that a large number of newborn neurons can be generated from adult NSC, and integrate into pre-existing neural circuits [8]. Under physiological conditions, adult NSC follow a highly stereotypic differentiation path to generate neurons in the olfactory bulb and the DG. Adult neurogenesis is also highly sensitive to environmental cues, physiological stimuli and neuronal activity, suggesting that the tailored addition of new neurons might serve specific neuronal functions [9]. Endogenous NSC may also provide a cellular reservoir for replacement of cell lost during normal cell turnover but also after brain injury [10], [11]. In neurodegenerative affections, particularly those involving pathogenic protein misfolding, the field of adult neurogenesis only begins to be explored. The results are not always consistent between studies. For instance, hippocampal neurogenesis is increased in patients with AD [12], but it is decreased in some transgenic mouse models of AD [13]. Following brain injuries, adult neurogenesis can be increased and is even accompanied by a migration of neural precursors towards the injured area [13], [14]. However, the activation of this system does not fully compensate the neuronal loss that occurs during diseases and could even contribute to the disease progression [15].
In prion diseases, while it has been suggested that adult neurogenesis was increased [16], the role and the status of adult NSC are still obscure. Despite the fact that we were able to propagate prions in vitro in NSC from fetal [17] or adult origin, we did not know whether endogenous NSC also accumulated prion in vivo and the impact this would have. This question has been addressed in this study and we showed that endogenous NSC were not only infected by prions but also that their neuronal differentiation process was altered.
In order to investigate whether pathological prion protein (PrPSc) deposits were present in brain area containing adult neural stem cells and/or their derivative neuroblasts, we first performed immunohistochemical analyses of PrPSc, nestin and doublecortin in mice that have been intracerebrally infected with the ME7 prion strain. Nestin is a NSC marker and doublecortin (DCX) is a neuroblast and immature neurons marker. The PrPSc deposits were densely present in the DG and LV neuronal progenitor area as well as in the DCX neuroblasts area surrounding the LV in mouse infected brain at the endpoint of the disease (Figure 1A, B, C, E). The double immunohistochemical analysis of the nestin or DCX markers with PrPSc clearly confirmed that PrPSc deposits were present around nestin and DCX positive cells (Figure 1D, F, G, H).
We then aimed to determine whether endogenous NSC had been infected during the disease development. Adult NSC were therefore isolated from the hippocampus and the lateral ventricle of 10 mock and 10 ME7 prion inoculated mice, at 130 days post-infection (dpi). As expected, the cells cultivated under neurosphere free floating conditions were all, positive for the nestin NSC marker (Figure 2A). We showed after two subpassages (30 days after isolation) that only NSC derived from ME7 infected mice were positive for PrPSc accumulation as assessed by both immunofluorescence (Figure 2B) and western blot after Proteinase K digestion (Figure 2C). This PrPSc generation in adult NSC was shown to be stable since PrPSc could still be detected after 15 subpassages (Figure 2D).
A thorough control experiment was then designed to confirm that the isolated cells were endogenously infected before the derivation and not during the cell culture. It consisted in the derivation of adult NSC from actin-GFP mice in the presence of an equivalent amount of a 130 dpi ME7 infected brain tissue. To avoid non-actin-GFP cells to proliferate, and therefore obtain only actin-GFP-NSC, the infected tissue dissected in each neurogenesis area was successively frozen at −80°C and heated at +60°C (Figure 3A). This procedure kills all the cells in the extract from non actin-GFP mice. In this paradigm, the derived actin-GFP NSC were not positive for PrPSc (Figure 3B) indicating that, in our experimental conditions, the PrPSc particles present in a 130 days infected brain were not sufficient to infect NSC.
The differentiation potential of these different NSC isolated from mock and prion infected mice was then analysed. To avoid a massive anoikis which is a ROCK signalling induced apoptosis that occurs after neurosphere dissociation [18], neurospheres were gently trypsinized and put on polyornithine/laminine coated dishes for one passage before being seeded for differentiation studies. To induce neuronal and glial differentiation, NSC were placed in a differentiation medium during 5 days. In these conditions, the nestin markers completely disappeared (Figure 4A) and NSC gave rise to neuroblasts (Figure 4A), young neurons and astrocytes (Figure 4B). Counting analyses of the number of DCX positive cells and betaIII-Tubulin positive cells were performed and are presented in Figure 4 (Fig. 4A, B, C). DCX positive neuroblasts (5% and less than 1% for ME7 derived LV-NSC and H-NSC instead of 40% and 25% for non infected derived LV-NSC and H-NSC) and BetaIII-Tubulin positive new born neurons (23.5% and 18% for ME7 infected LV-NSC and H-NSC instead of 41% and 40% for non infected LV-NSC and H-NSC) were less numerous (p<0.05 and p<0.01 respectively, MannWhitney Test) when cells were infected, indicating a defect in neuronal differentiation. Inversely, the proportion of astrocytes was higher in the ME7 infected cells (Figure 3C).
In order to assess whether this PrPSc accumulation impaired the neuronal differentiation process itself, NSC cells were infected just when they began their differentiation [17]. Both non infected hippocampus and lateral ventricle derived NSC were plated on poly-L-ornithine/laminin coated dishes in minimal neural N2 medium [17]. They were exposed to ME7 prion homogenate at 0.05% (p/v) at the beginning of the differentiation process (simultaneously with the EGF/bFGF withdrawal from the medium). N2 medium was completely changed after 24 hours and cells were harvested at different time points to analyse PrPSc accumulation (Figure 5). PrPSc was able to replicate in both cellular types since PrPSc was detected at 6 dpi for cells derived from the hippocampus (Figure 5A) and 8 dpi for NSC derived from the lateral ventricles (Figure 5B). Infected KOPrP cells were used as negative control of the experiment in order to detect the remaining inocula (Figure 5C). These results demonstrated that adult NSC were also capable of replicating ME7 strain during differentiation. Since there was no detectable PrPSc at day 5 of differentiation we analysed the number of neuroblasts (DCX+ cells, Figure 6A) and young neurons (bIII-tubulin+ cells Figure 6B) after 10 days of differentiation. As observed for the cells that were isolated from prion infected brain, we obtained less DCX+ cells and less βIII-tubulin cells in the infected cells than in the non-infected cells. The difference between non infected and infected cells was also statistically significant for both DCX and βIII-tubulin marker (P value<0.01, Mann-Whitney Test).
The presence of PrPSc among lateral ventricle NSC and neuroblasts incited us to isolate NSC cells from the brain of prion infected mice. We then checked whether these cells were infected by prion or not and assess the impact this could have on their neuronal differentiation. Our results show for the first time that NSC present in prion infected mouse brain accumulate PrPSc, which leads to an alteration of their neuronal fate. Indeed, we isolated adult neural stem cells from neurogenic adult brain area (the dentate gyrus and the lateral wall of the lateral ventricles) of both prion infected and non infected mice. Importantly, we were able to keep these cells in culture for several subpassages while maintaining their prion replication. The neuronal differentiation of the infected cells was shown to be compromised since less neuroblasts and less newborn neurons were obtained when the cells were placed in a neuronal differentiation medium. This was also accompanied by an increase in the amount of astrocytes.
As neuronal differentiation impairment was also observed when the infection occurred at the beginning of the differentiation, it may suggest that some differentiation pathways are impaired when prion replicate in the cells. This is however not exclusive, with a possible additional impact of PrPSc on cell survival or on cell proliferation, but this remains to be checked in a further study. Indeed, the cellular prion protein has been shown to positively regulate neural precursor proliferation during developmental and adult mammalian neurogenesis [19], to enhance neural stem cell proliferation [20] and protect cells against oxidative stress [21], [22], [23]. A loss/alteration of PrP function could have modified the proliferative potential of the cells resulting in an inappropriate differentiation issue or apoptosis activation as it has been shown with Aβ peptides [15]. Moreover, it has been recently reported [24] that prion neurotoxicity dramatically depends on PrPC expression on established neurons that also appear more susceptible to various subtoxic stimuli such as reactive oxygen species and glutamate. This study suggests also that active prion replication in neurons sensitizes them to environmental stress regulated by neighbouring cells, including astrocytes [24]. Cross-talks between astrocytes and neurons derived from the NSC from infected brains therefore represent an additional mechanism we need to investigate in further studies.
In any case, during the development of the brain pathogenesis, the compromised neurogenesis observed may probably have taken place earlier than the onset of hallmark lesions or neuronal loss. We believe that this important alteration of neurogenesis, also suggested to a lesser extent in other proteinopathies [25], represents an essential mechanism that underlies the progression and the issue of prion diseases.
Though adult neurogenesis may be beneficial for regeneration of the nervous system, our results also suggest that this system is defective during the course of the disease. While it has already been shown that neural precursor proliferation was enhanced in prion infected mice [16] the authors did not check the accumulation of PrPSc in the cells of interest. Our study is in fact the first demonstration, in a relevant prion model, of the involvement of neural stem cell in the progression of the disease. Although it also remains to be assessed, adult endogenous neural progenitors could also constitute a “reservoir” of PrPSc amplification since they can proliferate while replicating PrPSc. Moreover, the fact that endogenous adult neurogenesis could be modified by the accumulation of the disease associated misfolded prion protein represents another great challenge. Inhibiting the misfolding of those pathogenic proteins would thus allow the endogenous neurogenesis to compensate injured neuronal system. Our observation regarding the status of neural stem cells during prion infection is also very important since neural stem cells graft approaches are thought to be future therapeutic strategies. This is illustrated in the case of prion diseases by one of our recent publications [26]. In this study the cell therapy approach we developed had a significant effect marked by an increase in both incubation (20.1%) and survival times (13.6%) in mice grafted before the appearance of the clinical signs. Indeed, the present results suggest that the issue of our preclinical trials would have been more successful if we had proposed a stem cell graft strategy combined with an anti-prion strategy preserving the grafted cells from prion infection and/or targeting the endogenous neural stem cells niches. This is therefore a critical issue in the search for disease-modifying therapies not only for prion disorders but also for other neurodegenerative diseases like Alzheimer or Parkinson diseases.
Five-week-old C57Bl/6J female mice were anesthetized via intraperitoneal route with 100 µg/g of ketamine (Imalgène, Merial, Lyon, France) and 5 µg/g of xylazine (Rompun, Bayer, Leverkusen, Germany). They were then intracranially inoculated with the ME7 prion strain (1%, 20 µl) and with 1% homogenate of healthy brain as control (mock). Mice were housed in an A3 facility. Transgenic mice expressing EGFP under beta-actin promoter kindly provided by Dr. M. Okabe were also used. KO-PrP mice were kindly provided by Dr C. Weissmann. All animal work has been conducted according to relevant national guidelines of the French Ethical Committee (decree 87–848) and European Community Directive 86/609/EEC regarding mice. Experiments were performed with the approval of the Regional Languedoc Roussillon Ethical Committee for Animal Experiments under the registration number CEEA-LR-1006. They were performed in the Biohazard prevention area (A3) (Biorad/Université MontpellierII).
We have used the NeuroCult Enzymatic Dissociation Kit for Adult Stem Cell (StemCell Technologies, Grenoble, France) to isolate adult NSC from ME7 infected and non infected mice. We first dissected the lateral ventricles and hippocampus from adult mouse brains 130 days after their inoculation. They were transferred into a 100 mm dish containing NeuroCult Tissue Collection Solution. The tissue was then chopped with a scalpel for 1 minute and suspended in NeuroCult Dissociation Solution for 7 minutes at 37°C. NeuroCult Inhibition Solution was added at a 1∶1 ratio v/v and the suspension was centrifuged at 700 rpm (100× g) for 7 minutes. Pellet was then resuspended in NeuroCult Resuspension Solution. The digested tissue was mechanically dissociated by pipetting up and down 10 times, and centrifuged at 700 rpm (100× g) for 7 minutes. This step was repeated two more times with a P200 micropipettor. The final pellet was resuspended in 1 mL of Complete NeuroCult NSC Proliferation Medium (Mouse) supplemented with 20 ng/mL of recombinant human Epidermal Growth Factor and 10 ng/mL recombinant human basic Fibroblast Growth Factor (PHG0311 and PHG0021, Gibco, LifeTechnologies, Saint-Aubin, France). The cell suspension was then filtered with a 40 µm cell strainer (StemCell Technologies, Grenoble, France) and counted using Trypan Blue. Adult cells were seeded at 3.5×103 cells/cm2 in 6-well tissue culture dishes (Nunc, VWR, Fontenay-sous-Bois, France).
Under proliferation conditions, adult NSC were cultivated in T-25 cm2 tissue culture flasks (Nunc, VWR, Fontenay-sous-Bois, France). Before differentiation induction, they were first mechanically dissociated and transferred into a Poly-L-Ornithine/laminin coated dish. At 80% of confluence, cells were seeded in Poly-L-Ornithine/laminin wells of 6-well culture dishes with a cell density of 2.5×105 cells/cm2. The day after, the NeuroCult NSC Proliferation Medium was replaced by the NeuroCult NSC Differentiation Medium (Stem Cell Technologies, Grenoble, France). This medium was replaced every 2 days and after 5 days of culture, cells were fixed with 4% paraformaldehyde for 15 minutes at room temperature.
Neurospheres were transferred onto Poly-L-Ornithine/laminin coated 6-well plates (Nunc) in N2 medium with a cell density of 2.5×105 cells/cm2. They were maintained on monolayer for several subpassages to adapt the cells to the new conditions. When the culture reached the 80–90% of density, 0.05% (p/v) of ME7 or healthy brain homogenate were added into the N2 medium. Cells were incubated in the presence of the inocula for 24 h. The culture was rinsed twice and fresh N2 medium was added. Media were replaced every 2 days. To monitor any remaining inocula, KOPrP NSC cells were used as control.
Cells were permeabilized with 0.1% triton X-100 in PBS during 3 minutes, washed with PBS-BSA 0.2% three times. Saturation was performed with PBS-0.2%BSA for 1 hour at room temperature. Cells were then incubated with the primary antibodies (Nestin (Chemicon, MerckMillipore, Billereca, USA), DCX (Abcam, Paris, France), GFAP (Dako, Trappes, France), and beta-III tubulin (TujI clone, Covance, Rueil Malmaison, France), 1∶500 in PBS-0.2%BSA) for 1 hour at 37°C. Cells were then washed with PBS-0.2%BSA and were incubated with the appropriate secondary antibodies (goat anti-rabbit-Alexa fluor 488 and goat anti-mouse-Alexa fluor 555 (Invitrogen LifeTechnologies, Saint-Aubin, France), 1∶7000) for 1 hour at room temperature. After sequential washes with PBS-0.2%BSA, nuclei were stained with Hoechst 33258 (Calbiochem, VWR, Fontenay-sous-Bois, France) for 5 minutes under agitation at room temperature and then rinsed with PBS and H2O. The slides were mounted using the FluorSave Reagent mounting medium (Calbiochem, VWR, Fontenay-sous-Bois, France). Photos were taken with a Leica DMRA2 microscope and ImageJ software was used to count the cells. For the statistical analysis we used the Mann-Whitney test. For each condition, images were acquired from 5 to 8 fields and the experiment was repeated three times. An average of 80 cells was counted in each field.
Cells were permeabilized with 0.5% triton X-100 in PBS during 5 minutes and washed with PBS-BSA 0.2% three times. PrPSc epitope retrieval was obtained using 3M guanidium thiocyanate/PBS during 5 minutes and washed with PBS-BSA 0.2% and PBS. The saturation was then obtained with PBS-0.2%BSA for 1 hour at room temperature. The cells were incubated with the primary antibody SAF61 (1∶300) in PBS-5% Milk over night at 4°C. The remaining steps were the same as described before [27].
PrPSc presence was checked by western blot analysis using the saf Mix anti-PrP cocktail (saf 60, saf 69 and saf 70 antibodies) as described elsewhere [27].
Mice were anesthetized as described above and then perfused with paraformaldehyde 4%. The brains of the mice were collected and placed in paraformaldehyde 4% for 24 hours at 4°C. They were then manually embedded in paraffin (Paraplast, Microm, Villefranche sur Saone, France) and cut in sections of 5 µm thick using a Leica microtome. The sections were collected on microscope slides without treatment (Starfrost, Microm, Villefranche sur Saone, France).
Tissues were dewaxed using Clearify solution (Microm, Villefranche sur Saone, France) and then rehydrated with decreasing degrees of ethanol washes.
Nestin and DCX immunohistochemistry: Sections were incubated in H2O2 0.5% for 20 minutes at room temperature and washed with H2O and PBS Epitope retrieval was performed by heathing the sections in 0.1 M EDTA. Sections were then saturated with PBS-0.1%BSA-0.1%Triton X-100 for 1 hour and then incubated overnight at 4°C with the pre-diluted anti-Nestin (Chemicon, MerckMillipore, Billereca, USA) primary antibody or anti-DCX primary antibody (1/300, Abcam, Paris, France). The secondary antibody used was a biotinylated goat anti-mouse or anti-rabbit antibody (Amersham, Velizy-Villacoublay, France) (1∶1000 in PBS-0.1%triton X-100). The avidin-peroxidase complex (Vectastain Elite kit, Vector laboratories, Clinisciences, Nanterre, France) was then added and then revealed with 3,3′-diaminobenzidine (DAB) (Vector laboratoriess, Clinisciences, Nanterre, France), according to the manufacturers' instructions.
PrPSc immunohistochemistry: PrPSc was analysed by immunohistochemistry using the Saf84 (0.5 µg/ml) anti-PrP antibody [28], [29]. SAF84 monoclonal antibody recognising the human 161–170 PrP sequence was kindly provided by Dr J. Grassi (CEA/SPI, Saclay, France). For PrPSc immunostaining, epitope retrieval consists in a treatment with formic acid (10 minutes) followed by an autoclaving treatments (121°C, 10 minutes). The secondary antibody used was a biotinylated goat anti-mouse antibody (Amersham, Velizy-Villacoublay, France) (1∶1000 in PBS-0.1%triton X-100). The avidin-peroxidase complex (Vectastain Elite kit, Vector laboratories, Clinisciences, Nanterre, France) was then added and then revealed with 3,3′-diaminobenzidine (DAB).
For the double immunostaining procedure, we proceeded as described in [30]. Briefly, we first performed the protocol described above for nestin or DCX immunostaining using EDTA epitope retrieval pretreatments. The slides were then revealed using DAB. The brown precipitate given by DAB resists to PrPSc specific pretreatments (formic acid and autoclave). Then the slides were treated according to the procedure described above for PrPSc immunostaining (formic acid and autoclave). PrPSc revelation was performed using histogreen kits giving a blue green coloration.
List of the accession numbers for genes and proteins mentioned in the text (UniProt):
PrP: P04925
Nestin: Q6P5H2
Doublecortin: O88809
Beta-III-tubulin: Q9ERD7
Glial fibrillary acidic protein: P03995
|
10.1371/journal.pgen.1008267 | Artificial selection on GmOLEO1 contributes to the increase in seed oil during soybean domestication | Increasing seed oil content is one of the most important breeding goals for soybean due to a high global demand for edible vegetable oil. However, genetic improvement of seed oil content has been difficult in soybean because of the complexity of oil metabolism. Determining the major variants and molecular mechanisms conferring oil accumulation is critical for substantial oil enhancement in soybean and other oilseed crops. In this study, we evaluated the seed oil contents of 219 diverse soybean accessions across six different environments and dissected the underlying mechanism using a high-resolution genome-wide association study (GWAS). An environmentally stable quantitative trait locus (QTL), GqOil20, significantly associated with oil content was identified, accounting for 23.70% of the total phenotypic variance of seed oil across multiple environments. Haplotype and expression analyses indicate that an oleosin protein-encoding gene (GmOLEO1), colocated with a leading single nucleotide polymorphism (SNP) from the GWAS, was significantly correlated with seed oil content. GmOLEO1 is predominantly expressed during seed maturation, and GmOLEO1 is localized to accumulated oil bodies (OBs) in maturing seeds. Overexpression of GmOLEO1 significantly enriched smaller OBs and increased seed oil content by 10.6% compared with those of control seeds. A time-course transcriptomics analysis between transgenic and control soybeans indicated that GmOLEO1 positively enhanced oil accumulation by affecting triacylglycerol metabolism. Our results also showed that strong artificial selection had occurred in the promoter region of GmOLEO1, which resulted in its high expression in cultivated soybean relative to wild soybean, leading to increased seed oil accumulation. The GmOLEO1 locus may serve as a direct target for both genetic engineering and selection for soybean oil improvement.
| Soybean seed oil is an important quality trait targeted during domestication and breeding. However, the molecular mechanism of soybean oil regulation is largely unknown due to its complex genetic architecture and environmental sensitivity. In this paper, we integrated GWAS across multiple environments, haplotype analysis, genetic transformation, and diversity analysis to study the genetic architecture of oil content and the underlying mechanism in soybean. This combined analysis enabled us to identify an environmentally stable QTL (GqOil20) and functionally verified that GmOLEO1 positively regulates total seed oil accumulation in soybean seeds. In addition, we found that GmOLEO1 showed a higher level of expression in cultivated soybean seeds than in wild soybean seeds, possibly as the result of the positive selection of the promoter, resulting in seed oil accumulation. Moreover, we identified an elite GmOLEO1 haplotype that correlated strongly with high oil content in soybean, holding great potential for assisting oil improvement in soybean breeding. Our study provided a new genetic resource for oil content improvement in soybean and other oilseed crops.
| Soybean (Glycine max (L.) Merr.) is an important food and oil crop. Soybean seeds accumulate large amounts of oil and protein and have been intensively targeted for human consumption during long-term domestication and cultivation. Given the high percentage of oil in soybean seeds, the demand for soybean oil production has increased dramatically due to the increasing demand for vegetable oils and expanded use of biodiesel, and the seed composition improvement is of particular interest in terms of increasing awareness of health issues around dietary fats [1]. However, oil accumulation in the seed is a complex metabolic process that is environmentally sensitive; thus, stably expressed oil-enhancing key genes that can be applied to soybean molecular breeding have rarely been reported, and the mechanism of the variance of oil content in soybean remains largely unknown.
In plants, accumulated oil in seeds is generally stored as triacylglycerols (TAGs). TAG synthesis is initiated from glucose in the cytosol, and the resulting products from glycolysis are transported into the plastid for fatty acid (FA) synthesis. The FAs are processed by a series of key enzymes to produce C16:0 and C18:0 acyl chains and desaturated products, such as C18:1. FA products are then exported to the endoplasmic reticulum (ER) to form TAGs via the acyl-CoA-dependent and acyl-CoA-independent pathways [2]. The resulting TAGs are present in subcellular spherical lipid droplets in various plant tissues; the lipid droplets stored in seeds are usually called oil bodies (OBs) and have been extensively investigated previously in studies of, for instance, the structure and composition of an OB and the essential role of OB-related proteins, such as oleosins, in OB formation, mobilization, and oil accumulation [3–10]. Previous studies have indicated that oleosins play conserved roles in OB formation in seeds in several oilseed plants [7–8]. Suppression of a soybean oleosin produces micro-OBs [10], while the effects of OBs on seed oil accumulation have rarely been reported in soybean. The soybean genome contains 13 putative oleosin-encoding genes [11], and if any of them are involved in oil accumulation remain unexploited.
By linkage and linkage disequilibrium mapping, over 300 quantitative trait loci (QTLs) associated with seed oil content have been identified across all 20 chromosomes in the soybean genome over the past decades (SoyBase, https://soybase.org). These studies have revealed the polygenic nature of oil regulation, and the majority of loci were found to have varying additive, epistatic or QTL×environment effects [12–15], implying that traditional breeding based on genetic crossing and phenotypic selection may be inadequate for oil improvement. Recent studies have shown that increased oil in soybean could be achieved by genetic engineering of transcription factors involved in oil accumulation [16–18] or a QTL gene controlling seed coat bloom [19]. However, QTLs directly related to seed oil accumulation in soybean have not been cloned; thus, the underlying mechanism has not been thoroughly elucidated to date. Therefore, identifying an environmentally stable major QTL regulating seed oil content is urgently needed to substantially enhance seed oil content and understand the underlying regulatory mechanism in soybean.
To reveal the genetic basis of seed oil content and elucidate how oil accumulation is regulated, we investigated the oil content variation in 219 diverse soybean genotypes across six different environments and conducted a high-density genome-wide association study (GWAS) using 201,994 genome-wide single nucleotide polymorphisms (SNP). In total, three QTLs were identified to be significantly associated with soybean oil content across at least two environments, with GqOil20 on chromosome 20 stably expressing across all six environments. We also found that an oleosin-encoding gene, GmOLEO1, in the GqOil20 linkage disequilibrium (LD) block was exclusively expressed in developing seeds and that its expression level was significantly correlated with oil content within selected genotypes. We subsequently verified that GmOLEO1 contributed to oil accumulation in soybean seeds by conducting a series of molecular assays. Our results reveal an environmentally stable QTL/gene controlling oil accumulation in soybean seeds, provide new insight into oil accumulation in soybean and offer new directions for breeding soybean varieties with enhanced seed oil content.
To identify the genetic variation in seed oil content, we measured the oil contents of 219 soybean genotypes with diverse genetic backgrounds across six different environments. Seed oil content exhibited large amounts of natural variation within the association panel in each environment and showed relative consistency across the six environments (S1A and S1B Fig, S1 Table). The mean oil content for the 219 accessions ranged from 18.10% to 18.97% across the six environments, and the observed maximum oil content reached 27.69% in Environment 1 (E1), which was approximately three times higher than the minimum value (9.64%) observed in E6 (S1A Fig, S1 Table). The distribution of oil content for the association panel in each environment was approximately normal (S1C Fig). Analysis of variance (ANOVA) indicated a significant difference (p < 0.001) in oil content among the genotypes, the oil content was significantly affected by environments (p < 0.001) (S1 Table), and the heritability was 0.64.
Because of the wide variation in seed oil content in the panel across the environments, we performed GWAS for the oil content in six environments (E1 to E6) and the best linear unbiased prediction (BLUP) using 201,994 genome-wide SNPs with a minor allele frequency (MAF) ≥ 0.05 in an effort to identify the genetic loci associated with soybean oil content. In total, 110 SNPs on three chromosomes (8, 12, and 20) were identified as significantly associated with oil content across at least two environments (S2 Fig, S2 Table). For the sake of simplicity, we empirically classified closely adjacent SNPs located within 5 Mb into one locus, as previously described [20]. The 110 SNPs were classified into three genomic loci, which were subsequently designated GqOil8, GqOil12, and GqOil20 (S2 Fig, S2 Table). Of these QTLs, the most significantly associated SNPs were identified in GqOil20, which was in physical proximity to oil-related QTLs identified in previous studies (S2 Table) [21–24]. Importantly, GqOil20 was consistently identified across all the environments and BLUP except E4 (Fig 1A), and it explained 13.4–24.4% of oil variation, representing the most stably expressed QTL for oil content in soybean.
It is known that oil content is a key domestication trait undergoing artificial selection [25], and the regulatory genes involved were likely selected during domestication. Thus, a comparison of the genetic diversity at the three loci between cultivated soybean (G. max) and wild soybean (G. soja), the progenitor of G. max, could be helpful in determining the most likely regions containing the oil-controlling gene(s). To this end, we calculated genetic differentiation (Fst) within the 140 kb regions upstream and downstream of each leading SNP per locus within a group containing this association panel (272 G. max accessions) and a panel of 122 G. soja accessions genotyped with the same microarray, as previously described [26]. After the comparison, we found that Fst showed variation, and the Fst across the entire group (G. max and G. soja) was lower than the average Fst in the association panel (G. max only) in two QTLs (GqOil8 and GqOil12). In contrast, most of the Fst values for GqOil20 were significantly higher in the G. max-G. soja group than in the association panel, suggesting that artificial selection might have occurred in this genomic region in relation to oil accumulation (S3 Fig), consistent with the fact that soybean oil content is a domestication trait [25]. In this regard, GqOil20 likely harbors a gene or genes that have important functions in the regulation of soybean oil accumulation. Thus, we next focused on GqOil20 to identify oil-related genes.
To identify the candidate gene, we analyzed the LD region harboring the leading SNPs using BLUP as a phenotype. GqOil20 contained a total of 33 significant SNPs located within a strong LD with an average r2 = 0.66 (Fig 1C). Of these genes within the LD according to the G. max Wm82.a2v1 reference genome (https://phytozome.jgi.doe.gov) (S2 Table), we found that a gene, Glyma.20G196600, encoding a putative oleosin protein, colocated with the significant SNP AX-93661332 (P = 4.98 × 10−10) (Fig 1A and 1B, S3 Table). Glyma.20G196600 is an ortholog of Arabidopsis AtOLE1 (AT4G25140), an oleosin-encoding gene with demonstrated roles in oil body formation [9], while other gene models in this block are annotated to be involved in defense responses (S3 Table). Thus, Glyma.20G196600 might be the candidate gene underlying GqOil20, and we designated it GmOLEO1 for further study.
To investigate whether GmOLEO1 underlies the domestication region GqOil20, we examined the expression patterns and sequence variations of GmOLEO1 alleles in 38 soybean accessions comprising 27 cultivated and 11 wild genotypes with significant differences in oil content between two subgroups (Fig 1D). Consistent with the observed high oil content in G. max relative to G. soja, GmOLEO1 showed significantly higher expression in cultivated soybeans than in wild soybeans (Fig 1D and 1E), indicating a correlation between the transcript abundance of GmOLEO1 and oil content.
Next, a 2.3-kb genomic region extending from -1,500 bp upstream of the start codon (ATG) to the 3’-untranslated region (UTR) of GmOLEO1 was sequenced and analyzed. Sequence analyses identified 12 nucleotide variants that divided the 38 germplasm into six haplotypes (Hap), which were clearly classified into two subgroups (cultivated and wild) by a phylogenetic tree (Fig 1H). Moreover, the six haplotypes represented six levels of seed oil content (Fig 1F and 1G), with Hap1 seeds containing the highest oil content. Of the 12 nucleotide variants, seven variants were found to be significantly associated with soybean oil content (Fig 1F), with four located at -442 (A/C/-, P = 5.22×10−4), -281 (G/C, P = 2.83×10−4), -237 (AAA/—, P = 1.26×10−3), and -167 (13-bp insertion/deletion, P = 3.13×10−3) being detected in the promoter region and three (PS26 = 5.36×10−3, PS265 = 0.02, PS393 = 0.04) occurring in the exon. Of these variants, those in the promoter region represent the most significant variation associated with the variation in seed oil, suggesting that resulting differences in the expression of GmOLEO1 among the haplotypes might account for the oil content variation.
To further determine whether variation in the promoter affected gene expression, we compared the transcriptional activity of the promoters of Hap1 and Hap5 (Hap1_pro and Hap5_pro) using a dual luciferase reporter gene assay. As shown in Fig 1I, Hap1_pro exhibited 3.58-fold higher activity than Hap5_pro, consistent with the observed higher expression of Hap1 than Hap5 (Fig 1I). These results suggest that GmOLEO1 is a strong candidate for GqOil20 and that expression level instead of exon variation is an important factor affecting seed oil content.
Given that GmOLEO1 was a strong candidate associated with seed oil content, we characterized its protein structure, phylogeny, and expression pattern. BLASTp showed that GmOLEO1 is an ortholog of Arabidopsis AtOLE1 (AT4G25140), an oleosin-like protein with demonstrated roles in OB formation [9] (Fig 2A and 2B). Similar to previously described OLE orthologs in other species, GmOLEO1 contains three conserved structural domains (Fig 2A) [5]. Two amphipathic domains are located at the N- and C-termini, respectively, and a hydrophobic domain is located at the center. In the central hydrophobic domain, GmOLEO1 also contains the conserved “proline knot” sequence (PX5SPX3P), which can form a loop including a hydrophobic hairpin that penetrates into the TAG matrix and two arms located on both sides of the knot (Fig 2A) [5]. This domain organization allows oleosins to be anchored on the surface of an OB, as illustrated in a previous study [3]. Phylogenetic analysis revealed that OLEO-like proteins from the Faboideae, Brassicaceae, and grass clades clustered separately, suggesting functional conservation within the clade and possible functional diversity between clades (Fig 2B). In addition, sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) analyses showed that GmOLEO1 has a low molecular mass (~16 KD) (Fig 2C), consistent with previous findings [9, 27]. These results indicated that the uncharacterized gene GmOLEO1 encodes a putative OB protein that might play roles associated with OB formation or oil accumulation in soybean.
To determine the temporal and spatial expression pattern of GmOLEO1, the expression levels of GmOLEO1 were examined in ten different tissues and two soybean varieties with different seed oil contents (H101, a high-oil variety; H112, a low-oil variety) (Fig 2D). Quantitative real-time PCR (qPCR) results showed that GmOLEO1 transcripts were undetectable in nonseed tissues, including the roots, stems, leaves, and flowers of both varieties, but its transcripts could be detected in developing seeds beginning at the seed-filling stage (Fig 2D). The abundance of GmOLEO1 transcripts in seeds increased with the number of days after flowering (DAF), with the highest expression level observed in developing seeds at 40 DAF, which was immediately before that seeds had completely matured (Fig 2D). Overall, the expression level of GmOLEO1 in the developing seeds of H101 was significantly greater than that in H112 seeds at all tested stages. These results indicated that GmOLEO1 functions specifically during seed maturation and that transcript abundance positively correlated with oil content (Fig 1G).
We next investigated whether GmOLEO1 was spatially related to OBs. We expressed a 35S::GmOLEO1-GFP (green fluorescent protein) construct in tobacco (Nicotiana benthamiana) leaf epidermal cells by agro-infiltration followed by staining with Nile Red, a lipophilic dye used to visualize OBs [4]. Confocal microscopy analysis revealed that GmOLEO1-linked GFP fluorescence and Nile Red fluorescence signal from OBs were colocalized in seed cells (Fig 2E), indicating that GmOLEO1 was localized to accumulated OBs.
Taken together, the results of haplotype analysis, diversity analysis, phylogenetic analysis, expression analysis, and subcellular localization supported the GWAS results and collectively indicated that GmOLEO1 was a strong candidate gene underlying GqOil20 associated with oil accumulation in soybean seeds.
To further demonstrate whether GmOLEO1 is functionally involved in oil accumulation in soybean seeds, we overexpressed GmOLEO1 in soybean using an improved cot-node transformation protocol [28]. Successful transformation was determined by detecting both the expression of the selective bar gene using the strip test and the presence of 35S::GmOLEO1 (Fig 3A) using polymerase chain reaction (PCR) analysis in T0 plant leaves (S4 Fig). Transgenic soybean lines were self-pollinated through three generations to obtain homozygous lines harboring 35S::GmOLEO1. Three independent homozygous transgenic lines (OE-9, OE-16, and OE-18) were selected and used for further analysis.
We first quantified the expression of GmOLEO1 in developing seeds (10, 20, 25, 30, and 40 DAF). As shown in Fig 3B, GmOLEO1 expression increased during seed development in both the OE lines and wild type (WT), while its expression in the OE seeds was significantly higher in OE than in WT at each stage of seed development. GmOLEO1 exhibited a sharp increase in its expression in the OE lines at 25 DAF, which was five days earlier than the increase in expression observed (30 DAF) in WT seeds. In agreement with the observed expression difference between the two soybean varieties described above (Fig 2D), expression of GmOLEO1 in both the OE lines and WT increased continuously as the seeds developed and reached its highest levels at 40 DAF. This gene expression result was further verified by comparative Western blot analysis of GmOLEO1 protein between WT and the OE lines using an antibody against GmOLEO1. A higher expression of GmOLEO1 was observed in the three OE lines than in the WT at 25 and 40 DAF (Fig 3C).
Compared with WT, mature seeds from OE lines had shinier surfaces with more yellowish colors and smaller sizes (Fig 4A). The oil contents of the seeds of the three OE lines were 22.35%, 21.91%, and 22.14%, respectively, which were all significantly higher (an absolute average increase of 2.12%, a relative increase of 10.6%, P = 4.6 × 10−6) than that in WT seeds (20.01%) (Fig 4F). Not surprisingly, the increase in oil content in the OE lines resulted in a significant decrease (P = 0.006) in protein content (Fig 4G). To further verify the oil increase in the OE seeds, we conducted a series of microscopy analyses of developing OE seeds (OE-9 and OE-18) at 25 DAF, where sharp increases in the expression of GmOLEO1 and oleosin were observed (Fig 4C–4F). Microscopy analyses of cross-sections from developing seeds stained with Oil Red O showed that OE seeds have markedly stronger Oil Red O staining than WT seeds, indicating that OE seeds contain a higher level of neutral lipid accumulation than WT seeds (Fig 4B). A further examination of the seed cells using an optical microscope (Nikon, Eclipse Ci, Japan) showed that more OBs were deposited in the two OE lines (OE-9 and OE-18) than in WT (Fig 4C), and a consistent result was found via staining with toluidine blue O (Fig 4D). These observations were further verified by a comparative analysis of Nile Red staining of accumulated oil between OE and WT seeds using a confocal microscope (Nikon, C2, Japan) (Fig 4E). These results visibly illustrated that OE plants overexpressing GmOLEO1 contained higher levels of oil accumulation in seed cells than WT.
In addition, we observed that overexpression of GmOLEO1 has pleiotropic effects on other agronomic traits. The phenotypic evaluation indicated that the overexpression resulted in a significant decrease in 100-seed weight in the OE lines compared with that in WT (Fig 4I). However, a significant increase (P < 0.01) in pod number per plant and a slight increase in plant height in the OE lines compared with WT lines (Fig 4H and 4J) were observed, which led to an increase (P = 0.017) in seed yield per OE plant compared with WT plants (Fig 4K, S4 Table). We also compared the seed germination between two lines. We found that seed germination and root growth were faster in the OE lines than in WT (Fig 4L). These results indicate that GmOLEO1 is involved in oil accumulation in soybean seeds with pleiotropic effects on yield-related traits, and no yield penalty was found in the current preliminary study.
In light of the role of GmOLEO1 in oil accumulation, we further measured and compared the fatty acids between WT and OE seeds to test whether GmOLEO1 affected FA composition (Fig 5). Compared with the WT, OE seeds contained a higher average total FA content of 12.7% (P = 3.2× 10−4). Further analysis of five important oil components (TAGs) indicated that two polyunsaturated oil components, linoleic acid (18:2) and linolenic acid (18:3), were significantly increased by 14.4% and 14.9% (P = 1.2× 10−4 and 9.7× 10−5, n = 3), respectively, in the OE seeds compared with WT, while no significant changes in the contents of palmitic acid (16:0), stearic acid (18:0) and oleic acid (18:1) were observed between the OE seeds and WT (Fig 5, S5 Table). This result indicates that the overexpression of GmOLEO1 also led to increased accumulation of polyunsaturated FAs.
Last, we compared the OBs of OE and WT seeds using transmission electron microscopy. At 25 and 40 DAF, the OBs of WT seeds showed typically spherical and ovoid structures and were distributed mostly between protein bodies at the periphery of the cells (Fig 6). In contrast, OE seed cells contained apparently smaller OBs than those of WT (Fig 6).
To better understand the molecular mechanism by which GmOLEO1 increased oil accumulation in soybean seeds, we compared the transcriptomes of OE and WT seeds at three seed development stages (20, 25 and 40 DAF) using RNA-seq analysis. In total, 796, 1238, and 1417 differentially expressed genes (DEGs) were identified by comparing OE with WT seeds at 20, 25, and 40 DAF, respectively (Fig 7A, S6–S8 Tables). The RNA-seq result was validated by qPCR analyses of 16 randomly selected genes (S9 Table, R2 = 0.84). We observed a trend toward increasing numbers of DEGs as DAF increased (Fig 7B). This increasing trend in the number of DEGs is consistent with the pattern of oil content increase in seeds as DAF increases (Fig 7B). This result indicated that overexpression of GmOLEO1 resulted in significant changes in the transcriptomes in the developing OE seeds, and the changes became more dramatic as the seeds developed. To understand the biological processes in which GmOLEO1 participates, we performed Gene Ontology (GO) enrichment analysis for these DEGs. In addition to the enrichment of GO terms associated with the regulatory pathways essential for plant growth and development, such as seed development and germination, amino acid and sucrose metabolism, and response to growth hormone, we found that GO terms associated with linoleic acid metabolism, fatty acid transport, lipid metabolism and storage were also significantly enriched for these DEGs (Fig 7D).
The RNA-seq results were further verified by the increased expression of several known genes participating in TAG biosynthesis in OE seeds as shown by qPCR, such as diacylglycerol acyltransferase (DGAT1) [18], wrinkled 1 (WRI1) [29], zinc-finger protein (GmZF351) [17], two Arabidopsis OLEO orthologs (AtOLE2 and AtOLE3) [30], and oil body associated protein 1 (OBAP1A) [31] (Fig 7C), indicating that the expression of these genes may be affected by GmOLEO1 overexpression. These results indicated that overexpression of GmOLEO1 promoted the expression of TAG biosynthesis-related genes and led to the enhancement of TAG biosynthesis.
Because higher expression of GmOLEO1 in cultivated soybean than in wild soybean was observed, we hypothesized that the variations in its promoter region were under selection. Statistical analyses were performed using a large population of 302 soybean accessions [24]. We first evaluated Fst for different comparisons, including wild soybean vs. cultivar, wild soybean vs. landrace, and cultivar vs. landrace. The results showed that the Fst between wild vs. cultivated soybean is considerably higher than that between cultivar vs. landrace, especially in the promoter region (Fig 8A and 8B). The nucleotide diversity (π) analysis showed that π was higher in wild soybean than cultivated soybean in the promoter region and the coding region (Fig 8A). Tajima's D in the promoter region was 2.00, 0.06 and -0.84 for wild, landrace and cultivar, respectively, while Tajima's D in the coding region was 1.22, -0.819, and 0.04 for wild, landrace and cultivar, respectively (Fig 8A), implying that positive selection had occurred in the promoter region. A phylogenetic analysis using variants in the promoter and coding regions identified three clusters, corresponding to wild soybean, landrace, and cultivar (Fig 8C). Taken together, these results indicated that the promoter region was subjected to artificial selection during domestication.
It is known that seed oil content has been subjected to artificial selection targeting higher oil content [25]. This finding was further validated by our study, in which a significant difference in seed oil content between cultivated and wild soybeans was observed (Fig 1D). Unlike other domestication traits in soybean, such as stem growth habit [32] and pod shattering [33], soybean oil content is highly complex; it is regulated by many genes of small effect and is easily influenced by various environmental factors [1]. Our GWAS study across multiple environments allowed us to identify a new environmentally stable QTL, GqOil20, and an underlying candidate gene, GmOLEO1 that is capable of increasing seed oil content in soybean. Notably, the GmOLEO1 locus was previously identified as a possible candidate for an eQTL associated with seed oil accumulation [34], and it is physically close to other oil-related QTLs previously identified by linkage mapping [21–23]. GmOLEO1 may have been identified in this study because the corresponding alleles were fixed with respect to oil variation during domestication. Given the complexity of oil metabolism, the observed phenotypic variation (23.7%) could be due to the combined effects of GmOLEO1 and other genes at this locus. Other oleosin genes, regardless of sequence variation, with increased expression at the gene or protein level [35–36] during seed filling/maturation may also substantially affect oil accumulation. Nevertheless, our study further functionally verified that human-selected GmOLEO1 might be involved in seed oil accumulation, possibly by indirectly affecting oil biosynthesis via efficient feedback.
Our study and previous studies have indicated that the improvement of seed oil content in soybean during domestication was achieved by artificial selection of multiple major genes, and some of those genes may not be directly involved in oil biosynthesis, such as B1 [37] and GmZF351 [17]. In contrast to previous studies that identified oil-related genes using a reverse genetic approach, GmOLEO1 was pinpointed in an artificially selected locus, GqOil20, using an integrated strategy of high-density genetic mapping and genomics. Haplotype and expression analyses of GmOLEO1 between cultivated and wild soybean in our study provided additional evidence of selection at the GmOLEO1 locus. This artificially imposed selection pressure on the expression of GmOLEO1 could be an important factor affecting the observed difference in oil accumulation in soybean, because the overexpression of Hap2 of GmOLEO1 resulted in enhanced oil accumulation in transgenic soybean (Figs 1G and 4F). Whether other genes in this block have functions associated with oil accumulation requires further determination.
Seed yield and quality represent two of the most important traits in soybean improvement. Breeding soybeans with high oil stability across environments while maintaining protein content and yield has been difficult due to the complex genetic architecture of oil regulation. In our study, GmOLEO1 was functionally identified as a candidate for the environmentally stable QTL GqOil20. Overexpression of GmOLEO1 significantly elevated oil content and the percentage of polyunsaturated FAs without detriment to the overall plant performance, especially yield, in our preliminary study, making GmOLEO1 a promising candidate gene for use in breeding high-oil soybeans with improved levels of healthy polyunsaturated FAs. Although Hap2 (Williams 82-type) is not the most favored haplotype for increasing oil accumulation, enhanced seed oil accumulation was observed in this study, indicating that the GmOLEO1 allele in Hap2 enhanced oil accumulation independent of the amino acid substitution (Ala265Pro). One possible reason for this finding is that the substitution, which does not change hydropathy, may not affect the secondary structure of oleosins in OBs. The strong correlation between Hap1 and high expression levels of GmOLEO1 alleles (Fig 1) suggests the importance of unique variation (Hap1) in the promoter region in enhancing the expression of GmOLEO1. The discovery of the molecular function DNA marker, Indel P237167, from the unique variation in the promoter of GmOLEO1 will facilitate marker-assisted selection (MAS) in soybean high-oil breeding programs.
The importance of oleosins in lipid accumulation and oil body formation in seed plants has been gradually recognized over the past three decades, and it has demonstrated an important role in the maintenance of OBs and preventing them from coalescence [9]. The Arabidopsis genome contains 17 oleosin genes [38], of which AtOLE1 has been reported to be involved in lipid biosynthesis [39]. In soybean, 13 putative oleosin genes were found in the G. max reference genome, while only GmOLEO1 colocalized with the associated SNP (AX-93661332) in GqOil20 in our study (S10 Table). The overexpression of GmOLEO1 showed consistent results, as observed in AtOLE1, revealing conserved functions between GmOLEO1 and AtOLE1 in increasing oil accumulation. Despite being rarely studied in other species, the high similarity in amino acid sequence and structural domains (Fig 2A and 2B) suggests that GmOLEO1-like proteins from the Faboideae, Brassicaceae, and grass clades might have a conserved function in determining OB size but lineage-specific roles [7]. For example, expression of GmOLEO1 correlated with oil content in our study, and OE seeds with increased oil content contained smaller OBs; conversely, the expression of oleosin genes was independent of oil content in maize, and a high-oil maize strain contained larger, more spherical OBs than did low-oil maize [40]. The conserved role of OLEOs from various plant species in enhancing oil accumulation suggests that GmOLEO1 orthologs have considerable potential for oil improvement in other oil-producing crops.
It has been demonstrated that oleosins have important functions in OB formation, stabilization, and transgenic addition of oleosin increased oil content in Arabidopsis, Brassica and yeast [10, 31, 41–42], but oleosins’ role in increasing oil accumulation in soybean seeds has rarely been reported previously. In addition to these potentially cross-species functions in determining the size of OBs and affecting oil accumulation, our study showed that increased oleosins resulted in apparent reductions in OB size and increases in OB number in seed cells (Fig 6) and increased seed oil contents (Fig 2), in agreement with a study in which suppression of OLEO1 resulted in larger OBs and reduced total lipid levels in seeds [9–10]. Oil accumulation was gradually regulated as the seeds matured, possibly due to a gradual increase in the expression of GmOLEO1, which was concomitant with the enhanced TAG metabolism during seed maturation identified by RNA-Seq (Fig 7). Thus, it is logical that a positive correlation between oil content and GmOLEO1 expression was observed in our study (Fig 1G). A similar correlation has also been observed in Brassica napus, where the expression levels of OLEO/oleosin in high-oil genotypes were considerably higher than those in low-oil seeds [41]. The higher expression levels of GmOLEO1 in high-oil soybean varieties might be attributed to the unique variation present in the promoter of Hap1. The presence of putative seed-maturation-related cis-elements (abscisic acid (ABA) response and seed regulation, Fig 1F) in the promoter region of GmOLEO1 may be responsible for its exclusive expression during seed maturation.
In addition to stabilizing the structures of lipid droplets (LDs), oleosins also serve other functions, including enzymatic and signaling roles. Some of these proteins are ubiquitous in cells with and without LDs, thus exerting broader functions in seeds and other organs [43]. In peanut, oleosin3 (OLE3) was shown to exhibit bifunctional activities and was phosphorylated by STYK (AhSTYK) to regulate MGAT and PLA2 activity; it could be involved in the biosynthesis and mobilization of TAGs during seed maturation and germination [44]. However, a recent report showed that the bifunctional enzymic motifs are present in only peanut oleosins and not in those of other plants [7]; thus, another possibility is that oil accumulation increases as a result of GmOLEO1 overexpression, which might lead to efficient feedback by producing smaller OBs [45]. The detailed mechanisms underlying the regulation of gene expression by GmOLE1 must be deciphered in future work.
The level of oleosin itself is regulated during seed development and germination. When seeds germinate, oleosin degradation occurs prior to OB degradation. A recent study revealed that the ubiquitin binding protein PUX10 and division cycle 48 homolog A (CDC48A) are core components of an LD-associated ERAD-like degradation machinery, which facilitates the dislocation of oleosins from LDs [46–47]. In our study, faster seed germination of OE lines might be associated with higher levels of some oleosin-degradation proteins (e.g., PUX10 and CDC48A) [46–47], but this hypothesis remains to be experimentally determined.
Based on the results of our preliminary study and a previous finding [9], we proposed that the biosynthesis of TAGs was enhanced in the OE lines, possibly because of the affected TAG metabolic pathway, as a result of increased expression of GmOLEO1/oleosins. Smaller OBs gradually accumulated as the newly produced TAGs reached the minimum size that could be completely covered by the increased number of oleosin proteins, given that oleosins serve as surfactant to prevent OBs from coalescence [10, 48]. Thus, increased oleosin production in OE seeds, which resulted in reduced size but increased turnover of OBs during seed maturation, could be a more efficient way to use the limited intracellular space than larger OBs, leading to increased total oil content in OE seeds.
The association panel for GWAS consisted of a diverse collection of 219 soybean accessions (including 195 landraces and 24 elite varieties) originating from 26 provinces across six different agroecological regions in China, ranging from latitudes 53 to 24°N and longitudes 134 to 97°E [49]. Field experiments were performed in the 2009, 2011, 2012, 2013 and 2014 growing seasons at four different geographic locations as previously described [50]. Briefly, soybean plants were examined under field conditions at the following experimental stations: Jiangpu Experimental Station of Nanjing Agricultural University (32.1°N 118.4°E), Nanjing, in 2009 (designated as Environment 1, E1); Maozhuang Experimental Station (34.8°N 113.6°E) of Henan Agricultural University Zhengzhou, in 2009 (E2) and 2011 (E3); the Fangcheng Experimental Farm (33.2°N 112.9°E) of Henan Agricultural University in 2012 (E4), and Yuanyang Experimental Station of Henan Academy of Agricultural Sciences, Zhengzhou, in 2013 (E5) and in 2014 (E6). A randomized block design was used for all field trials. In all environments, each accession was planted in a three-row plot, with each row 200 cm long and 50-cm row spacing.
Mature soybean seeds were harvested and air-dried, and fully filled seeds were used for oil content measurement. Measurement of soybean oil, protein, and FA components was conducted using a near infrared spectrophotometer (NIR) seed analyzer (DA7200, Perten Instruments, Huddinge, Sweden) as previously described [51]. This association panel was genotyped using the NJAU 355K SoySNP array as previously described [26], and a total of 292,035 high-quality SNPs were used for association mapping.
Phenotypic data for soybean seed oil across different environments were subjected to an ANOVA using the PROC GLM (general linear model) mixed model of SAS version 9.2 (SAS Institute, 2002). The linear statistical model includes the effects of genotype, environment and the environment × genotype interaction. The BLUP for each line was calculated with PROC MIXED in SAS (SAS Institute, 2002) and used as the phenotypic input for the subsequent GWAS. The violin plot was drawn using the R package vioplot [52]. The heritability of oil content was calculated using h2 = Vg/ (Vg+Ve), where Vg and Ve represent genetic and environmental variation, and each term was extracted from the ANOVA results.
GWAS was conducted using the compressed mixed linear model with TASSEL 5.0 [53, 54] using SNP with minor allele frequency greater than 0.05, and the threshold was determined with Bonferroni threshold of ≤ 4.95 × 10−6 (P = 1/n) [55], where n is the SNP number used in GWAS. The population structure and the relatedness were described previously [26]. The Manhattan plot was drown using the R package qqman [56]. The LD heat map was plotted using the LDheatmap R package [57].
Expression of the candidate gene was examined in different soybean tissues, including roots, shoots, leaves, flowers, pods, and developing seeds at different developmental stages (10, 20, 25, 30 and 40 days after flowering). Total RNA was isolated from the tissues using the RNAsimple Total RNA Kit (TaKaRa, Japan), and 1 μg of RNA was treated with 10 units of RNase-free DNase I (TaKaRa) prior to cDNA synthesis. The first strand of cDNA was synthesized using the SuperScript III First-Strand Synthesis System (Invitrogen, USA) following the manufacturer's instructions. Gene expression was determined using the Bio-Rad CFX96 Touch Real-Time PCR System (Bio-Rad, California, USA). The PCRs contained 5 μL of the first-strand cDNA, 0.5 μL of 10 μmol L−1 gene-specific primers (S11 Table), and 10 μL of Real-Time PCR SYBR Mix (PC3302; Aidlab). The PCR conditions were as follows: 94°C for 3 min and 40 cycles at 94°C for 15 s and 60°C for 15 s. The soybean tubulin gene (GenBank: AY907703.1) was amplified as an internal reference, and a negative control reaction was performed using water instead of cDNA. Three biological replicates per sample were used, and each reaction was performed in triplicate.
In the protoplast transient expression experiments, the dual luciferase assay vector pGreenII 0800-LUC was used to analyze the activity of the different promoters. This vector contains a firefly luciferase (LUC) reporter gene that can be driven by the target promoter and a Renilla luciferase (REN) reporter gene driven by 35S. The purified DNA fragment of the target promoter was fused with the LUC reporter gene in the vector digested with HindIII and SalI enzymes to construct the recombinant vector. The vector pGreenII 0800-LUC without promoter insertion before the LUC reporter gene was used as a control. The recombinant vector and the control were individually transformed into Arabidopsis protoplasts via PEG-calcium transfection. The isolation of Arabidopsis protoplasts and protoplast culture were performed according to standard protocols [58]. The ratio of LUC and REN activity (LUC/REN) was used to reflect the activity of the target promoter. The LUC/REN value was determined using the dual luciferase reporter assay system (Promega, USA).
The complete coding sequence of GmOLEO1 was amplified from the cDNA of Williams 82 by regular PCR using gene-specific primers (S11 Table). The PCR product was subcloned into the pMD-19 T vector (TaKaRa, Japan) for sequence verification. The verified GmOLEO1 sequence was then cloned into the dicotyledon expression vector pCAMBIA3300, which contains a selection marker gene, phosphinothricin acetyltransferase (bar), using the ClonExpress Entry One Step Cloning Kit. The resulting recombinant pCAMBIA3300-GmOLEO1 construct was transformed into Williams 82 via the Agrobacterium tumefaciens-mediated soybean cotyledon node transformation system as previously described [59]. Extraction of genomic DNA from the leaves of PPT-resistant plants and nontransformed plants was performed using the cetyltrimethylammonium bromide (CTAB) method [60]. Transformants were verified by leaf-painting assay with herbicide phosphinothricin (PPT), PCR analysis for the presence of introduced GmOLEO1 and bar (482 bp), and LibertyLink strip detection for the expression of the bar gene using the QuickStix Kit (EnviroLogix Inc., ME, USA) were considered positive transgenics for further analysis. For LibertyLink strip detection, a total of 100 mg leaf tissue was collected and ground completely in the bottom of a conically tapered 1.5 ml tube by pestle rotation, followed by adding 0.5 mL of extraction buffer and a strip into the tube. After ten minutes, strips containing only the control line were negative for PAT protein expression, while those with two lines (control line and test line) were positive for PAT protein expression [61].
The full-length GmOLEO1 cDNA was amplified and cloned into the pBWA (V)HS-osgfp vector to obtain the pBWA(V)HS-osgfp-35S::GmOLEO1-GFP construct under control of the cauliflower mosaic virus (CaMV) 35S promoter (Biorun Co., Ltd). The binary vector 35S::GmOLEO1-GFP was transiently coexpressed in the leaves of Nicotiana benthamiana via agro-infiltration. Then, the tobacco leaf epidermal cells agro-infiltrated with the GmOLEO1-GFP construct were stained with Nile Red, a lipophilic dye used to visualize OBs [4]. Fresh leaves were placed in a solution containing Nile Red stock (100 mg/mL dimethyl sulfoxide) diluted 100× with 1×PBS for 10 min and washed with PBS twice for 30 s each time. Fluorescence signals were detected using a confocal laser scanning microscope (Nikon C2-ER, Japan) 2–3 days after infiltration. GFP, mKate and Nile Red were excited at 488, 561 and 559 nm, and their emission was detected at 510 to 540, 580 to 620 and 570 to 670 nm, respectively. All of these fluorescence experiments were independently repeated at least three times.
Immunogenicity peptides of GmOLEO1 protein were predicted by bioinformatics analysis. The sequences of the peptides were as follows: MAELHYQQQHQYPHR and KDYGQQQISGVQAS. The peptides were commercially synthesized and purified (Wuhan GeneCreate Biological Engineering Co., Ltd, China). Two male Japanese White rabbits were used for the immune procedure. Next, a polyclonal antibody of GmOLEO1 protein was separated and purified for immunoblot analysis. Proteins of fresh soybean seed were extracted by Triton X-100 lysate (0.5%). Then, 30 μg of protein extracts mingling with 2× SDS-PAGE sample loading buffer (Solarbio, Beijing, China) were loaded and subjected to SDS-PAGE. Afterward, protein bands were transferred onto polyvinylidene fluoride (PVF) membranes (Solarbio, Beijing, China). The membranes were blocked with 5% skimmed milk powder solution for 2 h at room temperature, followed by incubation with a polyclonal antibody against GmOLEO1 diluted to 1:10000 in phosphate-buffered saline overnight at 4°C. Finally, the blot was detected with horseradish peroxidase (HRP)-conjugated goat-anti-rabbit secondary antibody (Santa Cruz Biotechnology, USA) for another 1 h. The protein bands were visualized using a chemiluminescence system (Pierce, Rockford, Illinois, USA).
Transcriptomes were compared between pooled OE seeds at 20, 25, and 40 DAF, respectively, with the WT seeds at the corresponding developmental stage. For each time point, two developing seeds from each of the three OE lines (OE-9, -16 and -18) were collected and pooled as one biological replicate, and three biological replicates were used per sample. Library construction was performed as previously described [62]. The library was sequenced with the Illumina HiSeq 2500 analyzer at Biomarker Technologies (Beijing, China), producing 200-bp paired-end reads. An average of 6.47 gigabases of clean data per sample was generated. Differential gene expression was determined using the DESeq R package [52]. A gene with an adjusted P < 0.05 and a fold change (FC) >1.5 were defined as DEGs. Enrichment analysis of Gene Ontology of biological pathways (GOBPs) was performed using the GOseq R packages [63] to compute P values that indicate the significance of each GOBP being represented by the genes. GOBPs with P < 0.01 were identified as enriched biological processes.
Fresh immature soybean seeds harvested at 25 and 40 DAF were fixed in FAA fixation solution for at least 24 h. The main experimental steps for Oil Red O staining are as follows: cutting the whole sample into small blocks, removing excess water with tissue paper, immersing the small tissue blocks in Oil Red O (Servicebio, G1016, Wuhan, China) solution and incubating at 37°C for 60 min. Excess staining solution was removed by rinsing with tap water. The stained tissue blocks were immersed in 75% ethanol for 30 min or until no fading occurred; then, they were preserved in 4% paraformaldehyde and kept in the dark. Photos were taken using a digital camera (Canon 7D). The fixed tissue samples were embedded with OCT compound (Sakura, Japan). Frozen sections (8–10 μm) were obtained with Cryostat Microtome (Thermo, CRYOSTAR NX50, USA) and mounted on a prechilled glass slide. The frozen sections were stained with 0.1% Nile Red (Servicebio, G1073, Wuhan, China) and Oil Red O (Servicebio, G1016, Wuhan, China). Image observation for Nile red staining was performed using a Nikon confocal scanning microscope (Nikon, C2, Japan). The excitation wavelength was 488 nm, the emission wavelengths were 593–654 nm, and the OBs were imaged at 800× magnification. Oil Red O staining was imaged using an optical microscope (Nikon, Eclipse Ci, Japan) at 800× magnification.
Tissues (1×3 mm3 in size) of developing soybean seeds were fixed in 2.5% glutaraldehyde buffered with 0.1 M phosphate buffer (pH 7.2) for 12 h. Postfixation was subsequently conducted in 1% osmic acid in 0.1 M phosphate buffer (pH 7.2) for 5 h. The blocks were then washed, dehydrated through an ethanol series of 30–100%, and embedded in EMbed 812 media. The samples were cut into 1 μm slices using an ultramicrotome (Leica UC7, Germany), stained with alkaline toluidine blue O solution (Servicebio, G1032, Wuhan, China), and then imaged (800×) using an optical microscope (Nikon, Eclipse Ci, Japan). For transmission electron microscopy (TEM), the samples were cut into 60 nm slices using an ultramicrotome (Leica UC7, Germany) and then separately stained with uranyl acetate and lead citrate for 15 min. The slice samples were photographed under a TEM (HT7700, Hitachi, Japan).
The 2.3-kb genomic region spanning from 1,500 bp upstream from the translation start codon (ATG) to the 3’-untranslated region (UTR) of GmOLEO1 was sequenced and analyzed. Haplotype analysis was performed by resequencing this region in 20 high-oil, 20 low-oil and 10 moderate-oil accessions. All primers (S11 Table) used in this study were designed using the Primer 3 online tool (http://frodo.wi.mit.edu/primer3/). All sequences were verified manually, and all observed polymorphisms were reverified by resequencing of another amplicon. All the verified sequences were aligned using ClustalX version 1.83 [64]. The polymorphism data were analyzed using DnaSP version 4.10 [65] to identify sequence variation. Prediction of cis-elements in the promoter region was carried out using the online web tool PlantCARE [66].
FA components in soybean seeds were analyzed as previously described [17]. Briefly, 10 mg fine powder of soybean seeds was used for FA isolation. FAs were extracted with 1 mL of extraction buffer (2.5% [v/v] H2SO4 in CH3OH) at 85°C for 1 h. The supernatant (500 μL) was mixed with 300 μL of hexane and 600 μL of 0.9% (w/v) NaCl. FA methyl esters were redissolved in 200 μL of ethyl acetate and analyzed immediately with a gas chromatography system (GC-2014; Shimadzu, Beijing, China). Peaks corresponding to each FA species were identified by comparison to a FA methyl ester analytical standard (Supelco, Poole, UK). Concentrations of FA species were normalized against the internal control heptadecanoic acid (Sigma-Aldrich, USA). Five biological replicates per line were analyzed in this experiment.
The seeds were surface-sterilized with chlorine gas for 4 h prior to germination in darkness in Petri dishes (90 mm in diameter) on two sheets of filter paper moistened with deionized water (15 seeds per Petri dish). Germination tests were carried out in an incubator (MGC-400B, YIHENG, Shanghai, China) equipped at 25°C with 75% humidity. The filter paper was replaced once a day, and germinated seeds with healthy roots were counted. Root length was measured using a ruler at 2, 3 and 7 days postgermination. Three replicates per treatment were performed.
The published whole genome sequencing data were used for gene diversity analysis [25]. VCFtools was used to estimate gene diversity (Fst, nucleotide diversity (π) and Tajima’s D) [67]. SNPRelate combined with APE was used to construct the phylogenetic tree [68, 69].
|
10.1371/journal.pbio.1000234 | Species-Specific Heterochromatin Prevents Mitotic Chromosome Segregation to Cause Hybrid Lethality in Drosophila | Postzygotic reproductive barriers such as sterility and lethality of hybrids are important for establishing and maintaining reproductive isolation between species. Identifying the causal loci and discerning how they interfere with the development of hybrids is essential for understanding how hybrid incompatibilities (HIs) evolve, but little is known about the mechanisms of how HI genes cause hybrid dysfunctions. A previously discovered Drosophila melanogaster locus called Zhr causes lethality in F1 daughters from crosses between Drosophila simulans females and D. melanogaster males. Zhr maps to a heterochromatic region of the D. melanogaster X that contains 359-bp satellite repeats, suggesting either that Zhr is a rare protein-coding gene embedded within heterochromatin, or is a locus consisting of the noncoding repetitive DNA that forms heterochromatin. The latter possibility raises the question of how heterochromatic DNA can induce lethality in hybrids. Here we show that hybrid females die because of widespread mitotic defects induced by lagging chromatin at the time during early embryogenesis when heterochromatin is first established. The lagging chromatin is confined solely to the paternally inherited D. melanogaster X chromatids, and consists predominantly of DNA from the 359-bp satellite block. We further found that a rearranged X chromosome carrying a deletion of the entire 359-bp satellite block segregated normally, while a translocation of the 359-bp satellite block to the Y chromosome resulted in defective Y segregation in males, strongly suggesting that the 359-bp satellite block specifically and directly inhibits chromatid separation. In hybrids produced from wild-type parents, the 359-bp satellite block was highly stretched and abnormally enriched with Topoisomerase II throughout mitosis. The 359-bp satellite block is not present in D. simulans, suggesting that lethality is caused by the absence or divergence of factors in the D. simulans maternal cytoplasm that are required for heterochromatin formation of this species-specific satellite block. These findings demonstrate how divergence of noncoding repetitive sequences between species can directly cause reproductive isolation by altering chromosome segregation.
| Speciation is most commonly understood to occur when two species can no longer reproduce with each other, and sterility and lethality of hybrids formed between different species are widely observed causes of such reproductive isolation. Several protein-coding genes have been previously discovered to cause hybrid sterility and lethality. We show here that first generation hybrid females in Drosophila die during early embryogenesis because of a failure in mitosis. However, we have discovered that this is not a general failure in mitosis, because only the paternally inherited X chromosome fails to segregate properly. Our analyses further demonstrate that this mitotic failure is caused by a large heterochromatic region of DNA (millions of base pairs) that contains many repetitive copies of short noncoding sequences that are normally transcriptionally quiescent. Interestingly, this block of heterochromatin is only found in the paternal species. We suggest that a failure of the maternal species to package this paternally inherited DNA region into heterochromatin leads to mitotic failure and hybrid lethality. If this is a general phenomenon it may explain other examples of hybrid lethality in which F1 females die but F1 males survive.
| A critical stage of speciation is the development of reproductive isolating mechanisms that prevent gene exchange between diverging populations. Hybrid sterility and lethality are major components of reproductive isolation. A key to understanding how these hybrid incompatibilities (HIs) evolve is discovering the causal genes and determining how they inhibit or perturb normal development. A number of HI genes have been identified, all of which are protein-coding. These genes are characterized by two distinct modes of evolution: either high rates of coding-sequence divergence that are consistent with adaptive evolution in many [1]–[3] but not all [4] cases, or structural changes such as in gene location [5] or gene silencing and loss following duplication [6],[7]. These cases suggest that rapid evolution of either protein-coding gene sequence or structure is a general principle underlying the evolution of HIs.
Are rapidly evolving protein-coding genes the only cause of HI? Noncoding repetitive sequences, including transposable elements (TEs) and satellite repeats, are major contributors to genome evolution in higher eukaryotes. These sequences comprise heterochromatin, chromosomal regions found primarily around the centromeres and telomeres that remain more condensed than gene-containing euchromatin through the cell cycle. Pericentric heterochromatin is known to play important roles in mitotic and meiotic chromosome segregation [8]–[10]. Heterochromatin may also be important for the transcriptional regulation of flanking sequences such as ribosomal DNA (rDNA) loci, since rDNA genes are often found in heterochromatic regions [11],[12]. Paradoxically, however, despite these apparently conserved functions in higher eukaryotes, heterochromatin can vary greatly in abundance and sequence composition even between closely related species [13]–[16]. These observations have led to speculation that divergence of repetitive noncoding sequences may also directly cause reproductive isolation between nascent species [13],[17]. However, to our knowledge no examples have been clearly demonstrated.
One hint that heterochromatin divergence may contribute to HI came from the discovery that the protein encoded by the Drosophila hybrid lethality gene Lhr localizes to pericentric heterochromatin. Lhr itself shows strong evidence of having diverged under the force of adaptive evolution, leading to the hypothesis that it may be co-evolving with heterochromatic sequences [18]. An additional possible link between HI and heterochromatin comes from the identification of the gene Prdm9 as causing hybrid male sterility between subspecies of mice, because the heterochromatic meiotic sex body is defective in both sterile hybrids and in Prdm9-mutant pure-species mice [19].
The sibling species D. melanogaster and D. simulans exhibit large differences in heterochromatin content [15] and strong reproductive isolation [20]. F1 hybrid females produced from D. simulans mothers and D. melanogaster fathers die as embryos [21]. This female-specific lethality is intriguing for several reasons. First, this lethality appears to have a different genetic basis than the F1 male lethality that occurs in the reciprocal cross [22]. While two major-effect genes causing this male lethality have been cloned [18],[23], nothing is known about the molecular basis of the female lethality. Second, this female-specific lethality is an exception to Haldane's rule, the observation that unisexual hybrid sterility or lethality typically affects the heterogametic (XY or ZW) sex rather than the homogametic sex (XX or ZZ) [24].
Third, a link between hybrid female lethality and heterochromatin was strongly suggested by studies of Sawamura and colleagues of the D. melanogaster Zygotic hybrid rescue (Zhr1) mutation, which suppresses lethality of these otherwise lethal hybrid females. Zhr1 was discovered on an X-Y translocation chromosome that is deleted for much of the X chromosome pericentric heterochromatin [25]. The deleted region is thought to consist primarily of satellite DNA composed of a tandemly repeated 359-bp long monomer [25]. We refer henceforth to the monomer unit as the 359-bp repeat, and the heterochromatic region of the D. melanogaster X chromosome as the 359-bp satellite block, and revisit in the Discussion the question of what specific DNA sequences within this block cause hybrid lethality. In the wild type this satellite DNA (also known as the 1.688 g/cm3 satellite) is estimated to form a multi-mega-bp block of heterochromatin [26]. Experiments showing that hybrid viability is sensitive to the dosage of a mini-chromosome containing part of the 359-bp satellite block led to the suggestion that repetitive sequences within the 359-bp satellite block are responsible for the hybrid lethal effect [27]. However, the mapping studies are consistent with the alternative possibility that the Zhr locus is a protein-coding gene embedded within this heterochromatic region. This is a plausible alternative, as an unexpected number of protein-coding genes have recently been found on Drosophila Y chromosomes, which otherwise contain mega-bp amounts of heterochromatic repeats [28].
If Zhr is not a protein-coding locus, then the possibility that an HI locus consists of noncoding, repetitive DNA raises important questions regarding how such sequences could kill hybrids. One possibility is that heterochromatic sequences such as those comprising the X-linked Zhr locus cause hybrid lethality by inducing in trans a global effect on chromatin structure or gene expression. Alternatively, the Zhr locus might operate in cis by affecting other adjacent, X-linked sequences such as the rDNA genes or the centromere. A third alternative is that the lethal effects are confined to this heterochromatic locus itself, such that an aberration in its structure somehow directly disrupts embryonic development.
Given the difficulties involved in the genetic manipulation of heterochromatic sequences, we addressed these questions by combining genetic and cytological approaches to determine when hybrid females die during development, to identify the cellular basis of the lethality, to investigate whether possible heterochromatic defects occur genome-wide or are confined to the Zhr locus, and to test whether such defects are suppressed in hybrid females carrying the Zhr1 rescue mutation and induced in hybrid males carrying a Zhr duplication. Our results strongly suggest that the Zhr locus directly causes hybrid lethality by inducing mitotic failure in early precellularized embryos, and that the underlying defect is a failure of the 359-bp satellite block to form or maintain a proper heterochromatic state. These results provide compelling evidence that noncoding heterochromatic DNA can directly cause HI and thus contribute to speciation.
To address the timing and nature of hybrid female lethality, we examined young (0–3 h) hybrid embryos produced from several different wild-type parental strains (Table 1). Normal embryonic development in Drosophila begins with a single diploid nucleus that gives rise to several thousand nuclei through 14 synchronous mitotic divisions in the large, single-celled blastula. During the first nine divisions, the nuclei migrate from the interior of the embryo to the cortex as they expand in number. Four additional nuclear divisions occur at the cortex before the formation of membrane furrows that transform the syncytial blastoderm into the cellular blastoderm. This process, termed cellularization, is followed by gastrulation (for a detailed review of early Drosophila embryogenesis see [29]). As expected, hybrid male embryos, which survive to adulthood [20], underwent normal nuclear divisions during the blastula stage and progressed into the gastrula stage (Figure 1A). Hybrid female embryos also had normal nuclear spacing and synchrony during the first nine mitotic divisions (Figure 1A). However, between mitotic divisions 10–13, hybrid female embryos exhibited large areas near the cortex devoid of nuclei and abnormal amounts of nuclei remained deep within the cytoplasm, indicating a high level of failed nuclear divisions (Figure 1A). The nuclei at the cortex were irregularly shaped and spaced (Figure 1A) and stained unevenly for the mitotic marker phospho-Histone-3 (PH3) (Figure 1B), demonstrating that these nuclei have asynchronous cell cycles.
We also observed lagging chromatin between the dividing chromosome sets during anaphase and telophase in hybrid female embryos (Figure 1C and 1D). Lagging chromatin was observed in all analyzed hybrid female embryos (n = 16), ranging from 40% (13/32) to 100% (11/11) aberrant anaphase spindles per embryo, which is consistent with the high hybrid female lethality (∼87%–100%) produced from these crosses (Table 1). It is likely that the lagging chromatin is the direct cause of the mitotic asynchrony and other nuclear defects in hybrid female embryos, an idea supported by studies showing that mutations causing chromosome bridges lead to similar mitotic defects in D. melanogaster embryos [30]–[32].
To determine whether the lagging chromatin in hybrid female embryos results from a general defect in chromosome segregation or is instead chromosome-specific, we performed fluorescent in situ hybridization (FISH) with probes that recognize distinct satellite sequences in the pericentric regions of different D. melanogaster and D. simulans chromosomes (Figure 2A). Probe signals for sequences on D. melanogaster Chromosomes 2 and 3 and the D. simulans X chromosome were found in condensed regions near the spindle poles and never within the lagging chromatin (n = 75/75 spindles from 13 embryos; Figure 2B), indicating normal segregation of these chromosomes.
We also analyzed the segregation of the D. melanogaster X chromosome in hybrid female embryos by using a probe for the 359-bp repeat. The 359-bp repeat probe labeled two abnormally stretched strands leading outward from the lagging chromatin toward opposite spindle poles (n = 56/100 spindles from nine embryos; Figure 2B). Stretched 359-bp repeat DNA was also observed in anaphase spindles in which there was no lagging chromatin (n = 37/100 spindles; Figure S1). Our mapping of the 359-bp repeat probe on chromosome spreads from larval brain tissue confirmed the presence of the major block of 359-bp satellite located on the D. melanogaster X, as well as several minor blocks of related satellites (353-bp, 356-bp, and 361-bp repeats) on the left arm of D. melanogaster Chromosome 3 (also see Figure S2) [33]. These smaller regions appeared unstretched and segregated normally in hybrid female embryos (Figure S3). A variant of the 359-bp repeat is also present in a small satellite block in the pericentric region of the D. simulans X chromosome (Figure S2) but does not cross-hybridize with the 359-bp repeat probe under our experimental conditions (Figures 2B and S2), presumably because of its high level of sequence divergence from the D. melanogaster repeats [34]. The lagging chromatin in hybrid female embryos, therefore, is derived solely from the D. melanogaster X chromosome. Moreover, this stretching effect likely results from partial or complete failure of the sister D. melanogaster X chromatids to separate during anaphase rather than from defective X chromatin condensation because the 359-bp satellite block appeared properly condensed during metaphase (Figure 2C).
We used FISH with additional probes to determine whether separation failure of the D. melanogaster X chromatids is confined to the 359-bp satellite block or occurs in other regions of this chromosome. Probe signals from a euchromatic region located at the distal end (cytogenetic location 1C3-4) of the major left arm and from the tandemly repeated rDNA genes (bobbed+ locus) in the distal pericentric heterochromatin appeared as unstretched and condensed foci (for euchromatic region, n = 28/28 spindles from six embryos; for rDNA locus, n = 13/13 spindles from eight embryos; Figure 3A). A discrete signal of the simple-repeat satellite AATAT, which spans a portion of the minor right arm and part of the centromere immediately adjacent to the large 359-bp satellite block [35], was present at each end of the stretched 359-bp satellite block near the spindle poles in a pattern similar to the centromeric regions of the other chromosomes (Figure 3A). Therefore, the centromeres of the sister D. melanogaster X chromatids are active and separate at anaphase. However, we also observed small amounts of AATAT DNA stretched across the spindle and in the lagging chromatin, similar to the 359-bp satellite block (n = 20/34 spindles from three embryos; Figures 3A and S4). These results demonstrate that the stretched DNA is confined to the proximal X pericentric heterochromatin containing 359-bp and AATAT satellites, suggesting that sequences in this region are responsible for separation failure of the D. melanogaster X chromatids.
To determine the particular causal region of the pericentric heterochromatin, we examined the segregation of the Zhr1 compound-XY chromosome (Figure 2A) in hybrid female embryos. Consistent with previous results [25], crosses between wild-type D. simulans females and D. melanogaster Zhr1 males resulted in full viability of F1 hybrid female adults (Table 1). Our analysis of larval brain chromosome spreads from the Zhr1 strain revealed that the compound-XY chromosome is completely devoid of the 359-bp satellite block but contains Y-derived AATAT repeats (Figure S2). In hybrid female embryos the Zhr1 compound-XY chromosome segregated normally, as indicated by the complete absence of lagging chromatin during anaphase (n = 68/68 spindles from six embryos; Figure 3B). Furthermore, these embryos advanced properly through subsequent developmental stages into adulthood.
We also analyzed hybrid male embryos whose Y chromosome carries a translocation of approximately half of the X-linked 359-bp satellite block to the Y long arm (see Figure 2A) [36]. This Zhr+ chromosome resulted in hybrid male lethality that was less severe than hybrid female lethality induced by the wild-type X chromosome (Table 1). A subset of these hybrid male embryos (4/14) exhibited mitotic asynchrony and lagging chromatin during anaphase and telophase (n = 34/45 spindles; Figure 3C), similar to but not as common as the defects described above in hybrid females. Moreover, FISH analysis showed that the chromatin bridges were comprised of Y-derived 359-bp repeat DNA in these hybrid males (Figure 3C). These results, together with our analyses of the Zhr1 chromosome, strongly suggest that sequences contained specifically within the 359-bp satellite block induce chromosomal segregation failure in hybrid embryos.
Segregation failure of the D. melanogaster X chromatids in hybrid females occurs between nuclear cycles 10–13, a period when embryonic development is primarily under control of maternally contributed RNA and proteins [37]. Our findings, therefore, suggest that the D. simulans maternal cytoplasm lacks factors that are compatible with and necessary for proper segregation of the D. melanogaster X-linked 359-bp satellite block. This hypothesis is consistent with the fact that hybrid females produced from the reciprocal cross, carrying the 359-bp satellite block and the D. melanogaster maternal cytotype, are fully viable [20]. We therefore investigated the localization patterns of D1 and Topoisomerase II (TopoII), two proteins known to associate with the 359-bp satellite block in D. melanogaster [38]–[40].
Previous studies showed that the protein D1 localizes to AT-rich heterochromatin, including the 359-bp and AATAT satellites, in larval mitotic tissues [39],[40]. Additionally, D1 was found to influence the localization of heterochromatin protein 1 (HP1) to the 359-bp satellite block [40]. On the basis of these results, it was suggested that D1 may be a structural heterochromatin component of these satellites. To determine if D1 plays a role in the defective structure of the 359-bp satellite block in hybrids, we analyzed the localization of D1 in wild-type D. melanogaster and hybrid embryos with an antibody raised against D. melanogaster D1 [39]. In Western blots, this antibody recognized a single band of approximately 60 kDa, the predicted size of D1 in both D. melanogaster and D. simulans (Figure S5). In D. melanogaster and hybrid embryos, D1 was present during anaphase at numerous sites near the spindle poles, which are likely the AT-rich satellites in the centric and pericentric regions (Figure 4A). However, in hybrid female embryos, we observed no D1 localized to the lagging chromatin containing the 359-bp DNA (Figure 4A). These observations suggested the possibility that D. simulans D1 fails to bind these sequences in hybrids.
To test this hypothesis, we expressed D. melanogaster and D. simulans D1 in D. melanogaster embryos using the GAL4-UAS system (see Materials and Methods). Transgenic D1 localized to pericentric regions that completely overlapped with endogenous D1 (Figure 4B). We performed immuno-FISH experiments to simultaneously visualize D. melanogaster or D. simulans D1 with several satellite sequences. Both orthologs exhibited identical binding patterns in young embryos (Figure 4C–4F). Contrary to the prominent localization of D1 to 359-bp DNA in larval mitotic cells (also see Figure S2) [40], we observed barely detectable levels of D1 on this satellite block (Figure 4C–4F). Instead, D1 localized primarily to AATAT satellite DNA (Figure 4G). We propose that the major foci of D1 detected in embryos in earlier studies [39] and presumed to correspond with the 359-bp satellite block actually represent the large regions of AATAT on Chromosome 4. Our results demonstrate that unlike in larval brain cells, D1 is not a major component of the 359-bp satellite block during early embryogenesis, and likely does not play a role in the 359-bp structural defects observed in hybrid female embryos.
We also analyzed the localization pattern of TopoII in hybrid female embryos. TopoII is the primary enzyme in Drosophila that decatenates newly replicated DNA strands and is also believed to be a structural component of condensed chromatin [41],[42]. In control D. melanogaster embryos, TopoII localized to 359-bp DNA during interphase and became more evenly distributed across the chromosomes through mitosis, with an occasional, slight enrichment on the 359-bp block at anaphase (Figures 5 and S6). However, in hybrid female embryos TopoII localized to the 359-bp satellite block during interphase but remained highly and consistently localized to this DNA through mitosis (Figures 5 and S6). We observed no TopoII foci during anaphase in hybrid male or D. simulans male or female embryos (Figure S7), in which the 359-bp satellite block is absent, further supporting the conclusion that abnormal TopoII persistence in hybrid female embryos occurs specifically on the 359-bp satellite block. This finding and the observed stretched and lagging 359-bp DNA together indicate the presence of a structural defect in this heterochromatin block that prevents chromatid separation.
We have shown that hybrid females produced from D. simulans mothers and D. melanogaster fathers die during early embryogenesis because of widespread mitotic defects induced by separation failure of the 359-bp satellite block on the paternal X chromatids.
Elegant genetic experiments by Sawamura and colleagues first suggested that hybrid female lethality is caused by a D. melanogaster heterochromatic locus Zhr [25],[36]. Genetic mapping localized Zhr to a pericentric region of the X chromosome containing the 359-bp satellite block. Because it is otherwise unprecedented for a heterochromatic locus to cause HI, this finding raised the key question of how Zhr kills wild-type female hybrids.
We suggest that our results strongly support the conclusion that the 359-bp satellite block directly and specifically causes hybrid lethality, as opposed to alternative possibilities outlined in the Introduction, including indirect effects on other genomic regions. First, we found that hybrid female embryos exhibit large chromatin bridges during anaphase and telophase of mitotic cycles 10–13 that are almost exclusively comprised of DNA from the 359-bp satellite block on the D. melanogaster X chromosome. While these bridges also included some flanking AATAT satellite, a large amount of this satellite is present on the Zhr1 chromosome, which segregates normally, arguing against the AATAT satellite being causal for lethality. The small amount of lagging AATAT DNA detected in hybrid female embryos may result from over-catenation and tangling of AATAT DNA with the 359-bp DNA due to mis-localized TopoII (see below) when the chromatin is uncondensed, and is thus likely a secondary effect. Second, the entire multi-mega-bp satellite block appears to be stretched across the metaphase plate, suggesting that hybrids suffer from a structural defect in this block. Third, concomitant with these chromatin bridges we observed mitotic asynchrony and other aberrations that have been found in D. melanogaster mutants that have chromatin bridges [30]–[32]. In these cases, the lagging chromatin prevents complete separation of the daughter chromosome sets, thus inhibiting further mitotic divisions. Fourth, we found that all of these mitotic defects are suppressed in the Zhr1 mutant, which lacks the 359-bp satellite block, and are induced on a Y chromosome that contains a translocation of the 359-bp satellite block and causes hybrid lethality in males, albeit with incomplete penetrance.
An important clue comes from our finding that TopoII localizes abnormally to the 359-bp satellite block during mitosis in hybrid female embryos. Both the DNA-decatenating and structural roles of TopoII are believed to be essential for normal chromatid separation [42]. These observations suggest several possible explanations for the hybrid phenotype. One possibility is that X chromatid separation failure results directly from incompatibility between D. simulans TopoII and the D. melanogaster 359-bp satellite block. TopoII is well conserved in the melanogaster subgroup (D. melanogaster and D. sechellia TopoII proteins are 95.6% identical based on analysis of the full-length D. melanogaster TopoII and the ∼98% of TopoII sequence available for D. sechellia; only ∼78% of D. simulans TopoII sequence has been assembled), arguing that TopoII is not a primary incompatibility factor. Nevertheless, future transgenic experiments will be important for testing this idea.
Alternatively, the persistence of TopoII may reflect a response to incomplete replication of the 359-bp satellite block as a result of incompatibilities with the D. simulans replication machinery. Extensive and unresolved tangling of daughter DNA strands would prevent separation of the D. melanogaster X chromatids at anaphase. Our observations suggest that the centromeres of the X chromatids are active and pulled toward the spindle poles, thus creating tension that results in stretching of the 359-bp satellite block. However, it is unlikely that an incompatibility with the D. simulans replication machinery is the primary cause because the first nine mitotic divisions occur normally, suggesting that replication during these divisions is normal.
A third possibility is that abnormal TopoII persistence may result from improper heterochromatin formation of the 359-bp satellite block. Chromatid separation failure in hybrid females occurs during mitotic cycles 10–13 when heterochromatin initially forms. This process involves visible changes in chromatin condensation and localization of HP1 to pericentric and telomeric regions, and precedes the major transition from maternal to zygotic gene expression [43],[44]. Our data thus argue that chromatin bridges and lethality result from a failure of heterochromatin formation at the 359-bp satellite block. Defective heterochromatin formation may lead to other effects such as improper replication and tangling of daughter DNA strands, ultimately causing failure of chromatid separation.
What DNA sequences are responsible for these Zhr lethal effects? Our data argue strongly against the possibility that Zhr corresponds to an unknown protein-coding gene embedded within the 359-bp satellite block. Such a hypothetical gene would have to have the highly unusual property of causing mis-segregation of the entire satellite block in which it happens to be located. Furthermore, there is unlikely to be sufficient time to transcribe such a gene to cause lethality since the mitotic defects occur during the early stages of embryogenesis when zygotic transcription is minimal [45].
Previous genetic studies by Sawamura and colleagues led them to propose that the Zhr hybrid lethal effect is caused by repetitive elements in the pericentric region of the D. melanogaster X [27],[36],[46]. By assaying a series of X pericentric deletions and duplications of different sizes they further concluded that the lethality is quantitative, and correlates with the amount of pericentric heterochromatin present. Several Zhr− stocks contained less 359-bp repeat DNA than a wild-type Zhr+ stock [46], a finding consistent with the possibility that a dosage threshold of the 359-bp repeat causes hybrid lethality. However, they excluded the 359-bp repeat (referred to as the 1.688 g/cm3 satellite) as causing hybrid lethality because two copies of two different mini-chromosomes containing 359-bp repeats did not induce hybrid lethality [46]. The authors inferred that the double dosage of these mini-chromosomes would contain more 359-bp repeats than a single dose of another mini-chromosome that did reduce viability, thus concluding that dosage of the 359-bp repeat does not correlate with hybrid lethality. We suggest two caveats to this conclusion. First, while Southern blots suggested that differences in the abundance of 359-bp repeats are present in the mini-chromosome stocks, quantitative methods were not used to estimate the abundance of 359-bp repeats that are present specifically on the mini-chromosomes. Second, increased dosage of 359-bp repeats may induce lethality only when present as a single block on a single chromosome, and not when dispersed over multiple chromosomes. Subsequent experiments, however, showed that a different mini-chromosome can induce lethality when in two doses [27]. Although the cause of the discrepancy between the two studies remains unclear, they were later interpreted to indicate that either the 359-bp repeat or other repetitive elements are causing hybrid lethality [47].
Our experiments do not allow us to rule out the possibility that other repetitive elements present in the 359-bp satellite block and also unique to the D. melanogaster X chromosome contribute to hybrid lethality. Various TEs are known to be interspersed within the 359-bp satellite block [48]–[50], however none are specific to the X chromosome and thus cannot account for the X chromosome-specific segregation defects we observed. In contrast, several lines of evidence argue that the 359-bp repeat is the primary contributor to the Zhr hybrid lethal effect. First, the 359-bp repeat is among the most highly abundant satellite repeats in the D. melanogaster genome [15]. And while there are scattered repeats along the D. melanogaster X chromosome [51], the vast majority are found in the proximal pericentric heterochromatin where Zhr maps. Second, the 359-bp satellite is essentially species-specific, being ∼50-fold more abundant in D. melanogaster than in D. simulans and highly diverged in primary sequence of its monomers between these species [15],[34]. This species-specificity makes it an attractive candidate in evolutionary models that can account for the nonreciprocal nature of the F1 female lethality in D. melanogaster/D. simulans hybrids (see below). Third, the entire 359-bp satellite block becomes stretched during mitosis in hybrids. If another unidentified repetitive element is causing this effect, it must be distributed evenly across the entire 359-bp satellite block and not on other chromosomes.
Our experiments are consistent with the idea that large amounts of the 359-bp repeat present in one block are required to induce chromosome segregation defects. First, the related 353-bp, 356-bp, and 361-bp repeats, located in much smaller amounts on D. melanogaster Chromosome 3, do not induce any observable mis-segregation in hybrids. This observation could mean that only the 359-bp monomer is capable of disrupting chromosome segregation, or, alternatively, that large amounts of this satellite class are required to cause lethality. Second, translocation of approximately half of the X-linked 359-bp satellite block to the Y chromosome resulted in lagging Y chromatin and hybrid male lethality that are proportionally less penetrant than the effects induced by the full-length 359-bp satellite block in hybrid females.
The multi-mega-bp size of the 359-bp satellite block precludes definitive genetic tests using transgenic methods. We suggest, however, that the available evidence strongly supports the hypothesis that the 359-bp repeat is the sequence element within the 359-bp satellite block that is the cause of the Zhr hybrid lethal effect.
The fact that hybrid females are lethal when produced from D. simulans mothers and D. melanogaster fathers but viable when produced from the reciprocal cross clearly demonstrates the involvement of a maternal effect in this incompatibility. Our results can explain this maternal effect as follows. First, we suggest that the 359-bp satellite block requires maternal factor(s) in order to be packaged as heterochromatin during normal embryonic development in D. melanogaster. Second, D. simulans does not require such factors because it does not contain the 359-bp satellite block. These factors are therefore diverged in or absent from D. simulans. Third, in F1 hybrids from D. simulans mothers, the paternally inherited D. melanogaster 359-bp block fails to be packaged properly as heterochromatin because the requisite maternal factors are missing or functionally diverged.
Our proposal that the heterochromatin structure of the 359-bp satellite block is defective in hybrid females provides several promising hypotheses to explain the molecular nature of this incompatibility and the underlying maternal component(s). Satellites and other repetitive DNA elements are normally packaged into heterochromatin with general heterochromatin factors such as HP1 [52],[53], and, in some cases, with repeat-class-specific proteins like D1 [40], GAGA [54], and Prod [55]. These findings suggest a model in which high divergence in both the primary sequence and the abundance of repeat elements leads to incompatibilities with DNA-binding proteins expressed in the hetero-specific maternal cytoplasm. We tested D1 as a candidate maternal incompatibility factor because of its specific association in larval tissues with AT-rich satellite DNA, including the 359-bp repeat, but found that D1 does not localize to the 359-bp satellite block during early embryogenesis. Additional studies will be required to identify new candidate proteins that associate with the 359-bp satellite block in embryos in order to further test this model.
Alternatively, hybrid female lethality may be due to a mechanism involving small RNAs. In the yeast Schizosaccharomyces pombe and other organisms, RNA interference pathways and small RNAs are required for heterochromatin formation [56]. Recent studies have identified 359-bp satellite-derived small RNAs in the maternal cytoplasm of D. melanogaster [57],[58], raising the possibility that they may be required for initial heterochromatin formation and epigenetic silencing of the 359-bp satellite block during early embryogenesis. We have proposed that hybrid female lethality occurs owing to the absence of 359-bp–derived small RNAs in the D. simulans maternal cytoplasm [59]. According to this model, hybrid females from the reciprocal cross are viable because these small RNAs are present in the D. melanogaster maternal cytoplasm.
Regardless of the mechanistic basis of the maternal effect, it remains interesting that only the 359-bp satellite block is aberrant in hybrids, even though other satellite DNAs show significant differences in abundance and location between D. melanogaster and D. simulans [15]. Larger, more complex satellite repeats such as the 359-bp repeat may be more prone to cause HI than simple-repeat satellites, which are also known to vary in abundance between Drosophila species [15], because of their greater repeat sequence variation and perhaps, more complex heterochromatic structure.
The incompatibility between the D. melanogaster X-linked 359-bp satellite block and the D. simulans maternal cytoplasm likely explains why the lethality from this cross violates Haldane's rule [24]. In this cross only hybrid females are lethal because they inherit the paternal (D. melanogaster) X chromosome carrying the 359-bp block, while viable hybrid males inherit the paternal Y. Although Haldane's rule is observed in many taxa, it is frequently violated in other Drosophila hybridizations that produce unisexual lethality, and in several of these cases, hybrids die during embryogenesis from a paternal X-linked locus [60]. We propose that paternally inherited X-linked heterochromatic repeats are strong candidates for causing hybrid female lethality in these interspecific crosses.
Much of repetitive DNA evolution is likely governed by neutral evolutionary processes [61]. However, variation in satellite DNAs can also be driven by their ability to mediate genetic conflicts such as segregation distortion [62]. In the D. melanogaster segregation distorter (SD) system, sperm bearing high-copy alleles of the 240-bp Responder (Rsp) satellite are targeted for destruction while sperm with low-copy alleles are immune to this effect [63], thus selecting strongly against high-copy alleles. Variation in satellite DNA abundance may also be influenced by meiotic drive in female meiosis [64]–[66]. Female meiosis is particularly prone to meiotic drive because only one of the four meiotic products becomes the maternal pronucleus of the egg. This situation creates an opportunity for competition among chromatids to gain access to the egg, with variation in centromeres being a prime candidate for mediating such antagonism among chromosomes. A recent example of this phenomenon was found in Mimulus, in which distinct centric or pericentric repeat alleles appear to confer a substantial chromosomal transmission advantage during female meiosis in conspecific crosses and a more extreme advantage in interspecific hybrids [67]. Meiotic drive and other types of genetic conflict may therefore be important for causing rapid evolution of repetitive sequences within species and fixed differences between closely related species. Our data demonstrate that as these interspecific differences accumulate, repetitive sequences can inhibit chromosome segregation in hybrids and thus directly cause reproductive isolation.
Strains used were: wild-type D. simulans C167.4 and NC48S [68], and white501 (SA32) (made by introgressing the white501 allele into an isofemale South African D. simulans strain that mates well with D. melanogaster, provided by C. Aquadro), and wild-type D. melanogaster Oregon R and Canton S. The hybrid rescuing Zhr1 chromosome (full genotype is XYS.YL.Df(1)Zhr) is described in [25] and the Zhr+ Y chromosome (full genotype of strain is Ts(1Lt;Ylt)Zhr/Dp(1;Y)y+) is described in [36]; the structures of both chromosomes are shown in Figure 2A. Crosses were conducted by combining 40–50 0–8-h-old virgin D. simulans females and 60–80 12–24-h-old D. melanogaster males. Flies were allowed to mate for 48 h in a 25°C incubator with a 12-h light/12-h dark cycle prior to embryo collection.
Embryos were collected on grape juice agar plates [69] over a 3-h period and dechorionated in 50% bleach. Immediately afterward, embryos were fixed for 10 min in 4% EM-grade paraformaldehyde (Electron Microscopy Sciences) and heptane (Sigma-Aldrich) and then devitellinized in 100% methanol (Sigma-Aldrich). Fixed embryos were hydrated by using a series of methanol∶1×PBTA buffer solutions (9∶1, 5∶5, 1∶9 by volume) and treated with RNaseA (Sigma-Aldrich) at 37°C for 2 h before carrying out FISH or immuno-staining.
The following sequences were used for FISH probes: (TTT-TCC-AAA-TTT-CGG-TCA-TCA-AAT-AAT-CAT) recognizing the 359-bp satellite block on the D. melanogaster X as well as minor variants on D. melanogaster Chromosome 3 [34]; (AAT-AC)6 recognizing a small block of this sequence on the D. melanogaster Y [70]; (AAT-AT)6 recognizing large amounts of this sequence on D. melanogaster and D. simulans Chromosome 4 as well as a small region on the D. melanogaster X [15],[70]; (AAG-AG)6 recognizing primarily a large block of this sequence on the D. melanogaster 2 and a small block on the D. simulans X [15]; (AAT-AAC-ATA-G)3 recognizing a single block of this sequence on D. melanogaster Chromosomes 2 and 3 [70]; and (AAT-AAA-C)4 recognizing a single region on the D. simulans Y (S. Maheshwari, personal communication). These sequences were chemically synthesized (MWG Biotech) and modified at the 5′ terminus with either fluorescein, Cy3, or Cy5 for fluorescent detection. The euchromatic FISH probe was made by random priming of BAC DNA (BACR19J01 from CHORI BACPAC Resources) using the BioPrime DNA Labeling System (Invitrogen), which incorporates amino-allyl-dUTP into amplified DNA products (protocol from A. Minoda). These products were sonicated into 100–150-bp fragments, cleaned with MiniElute columns (Qiagen), and conjugated with an Alexa546 fluorophore reactant group (Invitrogen). The labeled probe was ethanol-precipitated twice before hybridization. The D. simulans 360-bp probe was made by end-labeling a 360-bp PCR product with poly-amino-allyl-dUTP and Terminal Transferase (Roche). This product was amplified from D. simulans C167.4 genomic DNA using the following primers that recognize the D. simulans 360-family monomer repeat [34]: forward ACT-CCT-TCT-TGC-TCT-CTG-ACC-A and reverse CAT-TTT-GTA-CTC-CTT-ACA-ACC-AAT-ACT-A. The rDNA probe was made by using a 1,200-bp fragment of five tandem repeats of the rDNA IGS region cloned into pBluescript (construct was a gift from G. Bosco). This fragment was digested out of the plasmid using the restriction enzymes KpnI and SacI, digested into ∼150-bp fragments with AluI and MseI, and purified using the Qiaex II gel extraction kit (Qiagen). End-labeling was conducted as described above.
FISH was performed as described [70] with minor modifications. Overnight incubation of fixed tissues with probe was conducted in a thermocycler with a denaturation temperature of 92°C for 3 min and a hybridization temperature of 32°C overnight. This lower hybridization temperature (standard is 37°C) was important for detecting the 359-bp variant sequences on D. melanogaster Chromosome III and did not result in excessive nonspecific hybridization. Three additional 10-min washes in 50% formamide/2×SSCT were performed to maximize the removal of any nonspecifically bound probe. For immuno-FISH experiments, fixed embryos were hybridized with primary and secondary antibodies as described [69]. Embryos were subsequently fixed in 4% paraformaldehyde for 30 min, and washed 3× with PBTX and 3× with 2×SSCT. FISH was then conducted as described above. For immuno-cytological or immuno-FISH experiments, Rat anti-HA (Roche 3F10) and rabbit anti-D1 antibodies (a gift from E. Käs) [39] were used at 1∶100 and 1∶1,000, respectively. Rabbit anti-TopoII (a gift from T. Hsieh) [71] and mouse anti-PH3 (Santa Cruz Biotechnology) were used at 1∶1,000. Alexa555 anti-rabbit, Alexa555 anti-rat, and Alexa633 anti-mouse secondary antibodies were used at 1∶300 (Molecular Probes). Tissue preparation, FISH, and D1 immuno-staining of larval brain chromosomes were performed as described [39]. DNA was stained by using either OliGreen or TO-PRO-3 iodide (Molecular Probes) for embryos and Vectashield containing DAPI (Vector Laboratories) for brain tissues. All imaging was conducted at the Cornell Microscopy and Imaging Facility, using either a Leica DM IRB confocal microscope or an Olympus BX50 epifluorescent microscope. Confocal images were generated by using sequential collection of each wavelength to eliminate bleed-through of fluorophores and generated as maximum projections of multiple scans. Images were processed using Photoshop (Adobe, version 7.0). Contrast and brightness changes, when used, were applied globally across the image. Images shown in the figures were taken from either hybrid or pure species embryos produced from C.167.4 and/or Canton S strains, unless otherwise specified. However, cytological analyses were also performed on hybrid embryos produced from other parental lines shown in Table 1 for verification of the observed phenotypes.
0–3-h embryos from C167.4 and Canton S flies were collected as described above and washed in 1× PBS buffer. 50 µl embryos from each strain were lysed in an equal volume of 2× SDS Sample Buffer [72] and boiled for 5 min. Five µl of protein extracts were separated on a 10% polyacrylamide gel for 1 h at room temperature and 100 V. Proteins were transferred to a nitrocellulose membrane overnight at 4°C and 20 V, blocked with 5% powdered milk, and then blotted overnight at 4°C with anti-D1 serum (1/10,000 dilution). Membranes were then blotted with goat anti-rabbit HRP antibodies (1/5,000 dilution; Jackson) for 1 h at room temperature. HRP was detected using ECL Western blotting substrate (Pierce).
The complete D. melanogaster and D. simulans D1 coding sequences were PCR amplified from Canton S and C167.4 adult cDNA, respectively, by using the following primers: for D. melanogaster, forward CAC-CAT-GGA-GGA-AGT-TGC-GGT-AAA-GAA-G and reverse TTA-GGC-AGC-TAC-CGA-TTC-GG; for D. simulans, forward CAC-CAT-GGA-AGA-AGT-TGC-GGT-AAA-GAA-G and reverse TTA-GGC-AGC-TAC-CGA-TTC-GG. The resulting fragments were cloned into the pENTR/D-TOPO vector (Invitrogen). Positive clones were fully sequenced to confirm the absence of any errors. Each sequence was recombined into the pPHW plasmid downstream of the UAS transcriptional activation sequence and in frame with an N-terminal 3× HA peptide (Murphy collection; described at http://www.ciwemb.edu/labs/murphy/Gateway%20vectors.html). Additionally, the attB sequence was subcloned into this plasmid for site-specific integration into the D. melanogaster strain y1 w67c23; P{CaryP}attP2 [73]. The resulting transformants were crossed with the strain w; P{matα4-GAL-VP16}V37 (Bloomington Stock Center) for expression of the HA-tagged D1 in the early embryo.
|
10.1371/journal.ppat.1007962 | Convergent evolution in the mechanisms of ACBD3 recruitment to picornavirus replication sites | Enteroviruses, members of the family of picornaviruses, are the most common viral infectious agents in humans causing a broad spectrum of diseases ranging from mild respiratory illnesses to life-threatening infections. To efficiently replicate within the host cell, enteroviruses hijack several host factors, such as ACBD3. ACBD3 facilitates replication of various enterovirus species, however, structural determinants of ACBD3 recruitment to the viral replication sites are poorly understood. Here, we present a structural characterization of the interaction between ACBD3 and the non-structural 3A proteins of four representative enteroviruses (poliovirus, enterovirus A71, enterovirus D68, and rhinovirus B14). In addition, we describe the details of the 3A-3A interaction causing the assembly of the ACBD3-3A heterotetramers and the interaction between the ACBD3-3A complex and the lipid bilayer. Using structure-guided identification of the point mutations disrupting these interactions, we demonstrate their roles in the intracellular localization of these proteins, recruitment of downstream effectors of ACBD3, and facilitation of enterovirus replication. These structures uncovered a striking convergence in the mechanisms of how enteroviruses and kobuviruses, members of a distinct group of picornaviruses that also rely on ACBD3, recruit ACBD3 and its downstream effectors to the sites of viral replication.
| Enteroviruses are the most common viruses infecting humans. They cause a broad spectrum of diseases ranging from common cold to life-threatening diseases, such as poliomyelitis. To date, no effective antiviral therapy for enteroviruses has been approved yet. To ensure efficient replication, enteroviruses hijack several host factors, recruit them to the sites of virus replication, and use their physiological functions for their own purposes. Here, we characterize the complexes composed of the host protein ACBD3 and the ACBD3-binding viral proteins (called 3A) of four representative enteroviruses. Our study reveals the atomic details of these complexes and identifies the amino acid residues important for the interaction. We found out that the 3A proteins of enteroviruses bind to the same regions of ACBD3 as the 3A proteins of kobuviruses, a distinct group of viruses that also rely on ACBD3, but are oriented in the opposite directions. This observation reveals a striking case of convergent evolutionary pathways that have evolved to allow enteroviruses and kobuviruses (which are two distinct groups of the Picornaviridae family) to recruit a common host target, ACBD3, and its downstream effectors to the sites of viral replication.
| Enteroviruses are small RNA viruses that belong to the Enterovirus genus of the Picornaviridae family. They are non-enveloped positive-sense single-stranded RNA viruses with icosahedral capsids, currently consisting of 15 species. Seven enterovirus species (Enterovirus A-D and Rhinovirus A-C) contain human pathogens, such as polioviruses, numbered enteroviruses, echoviruses, coxsackieviruses, and rhinoviruses. They cause a variety of diseases ranging from common cold to acute hemorrhagic conjunctivitis, meningitis, myocarditis, encephalitis, or poliomyelitis [1]. The genome of the enteroviruses encodes the capsid proteins and seven non-structural proteins (named 2A-2C and 3A-3D). The latter carry out many essential processes including genome replication, polyprotein processing, host membrane reorganization, and manipulation of intracellular trafficking. To facilitate these functions, several host factors are recruited to the sites of enterovirus replication through direct or indirect interactions with viral proteins. For instance, the enterovirus non-structural 3A proteins directly bind to the Golgi-specific brefeldin A-resistant guanine nucleotide exchange factor-1 (GBF1) [2] and acyl-CoA-binding domain-containing protein-3 (ACBD3, also known as GCP60) [3].
ACBD3 is a Golgi resident protein involved in the maintenance of the Golgi structure [4] and regulation of intracellular trafficking between the endoplasmic reticulum and the Golgi [5]. ACBD3 is a multidomain protein composed of several domains connected by flexible linkers. Its central glutamine rich domain (Q domain) interacts with the lipid kinase phosphatidylinositol 4-kinase beta (PI4KB) and with the Rab GTPase-activating proteins TBC1D22A and TBC1D22B [6]. The interaction of ACBD3 and PI4KB causes membrane recruitment of PI4KB and enhances its enzymatic activity [7]. The C-terminal Golgi-dynamics domain (GOLD) of ACBD3 has been reported to interact with the Golgi integral protein giantin/golgin B1, which results in the Golgi localization of ACBD3 [5]. However, in enterovirus-infected cells, the ACBD3 GOLD domain interacts preferentially with viral non-structural 3A proteins, which causes re-localization of ACBD3 to the sites of virus replication [8].
The role of ACBD3 in enterovirus replication is not yet fully understood. It has been proposed that recruitment of ACBD3 to the sites of viral replication can lead to the indirect recruitment of its interactors and downstream effectors such as PI4KB, a well-known host factor essential for generation of PI4P-enriched membranes suitable for enterovirus replication [9, 10]. The 3A-ACBD3-PI4KB route represents one of the major described mechanisms of PI4KB recruitment to the sites of enterovirus replication [3, 11], although some other mechanisms employing the viral proteins 2BC [12] or 3CD [13] might be involved as well. Moreover, the formation of the 3A-ACBD3-PI4KB complex represents the major described mechanism of PI4KB recruitment to the replication sites of kobuviruses, members of a distinct group of picornaviruses [3, 14–16]. Previously, it has been suggested that PI4P directly recruits the viral RNA-dependent RNA polymerase (3Dpol) [9]. Further studies, however, revealed that the affinity of PI4P to 3Dpol is too weak to attract 3Dpol to target membranes by itself, suggesting that other factors may be involved [17]. Notably, PI4P gradients between various membranes can be used for the transport of other cellular lipids against their concentration gradient [18, 19]. The PI4P/cholesterol exchange machinery was implicated in replication of several enteroviruses [12, 20], suggesting that PI4P can be used by the viral machinery as a mediator to prepare membranes with a specific lipid composition suitable for viral replication.
ACBD3 is an important host factor of various enterovirus species [21], however, the structural determinants of its recruitment to the viral replication sites are poorly understood. To date, the structural information about any picornavirus 3A proteins is limited to a solution NMR structure of the uncomplexed poliovirus 3A protein [22] (pdb code 1NG7) and our previously published crystal structure of the aichivirus 3A protein in complex with the ACBD3 GOLD domain [23] (pdb code 5LZ3). Unfortunately, the latter cannot be used for homology modeling of the enterovirus 3A proteins, given the unrelated primary sequences of the enterovirus and kobuvirus 3A proteins, which indicates distinct mechanisms of hijacking ACBD3 by these two groups of viral pathogens.
In this study, we present a structural, biochemical, and biological characterization of the complexes composed of human ACBD3 and the 3A proteins of four representative enteroviruses. The crystal structures revealed the details of the ACBD3-3A interaction, the 3A-3A interaction causing the assembly of the ACBD3-3A heterotetramers, the interaction between the ACBD3-3A complex and the lipid bilayer, and the roles of these interactions in facilitation of enterovirus replication. The comparison of the structures of the ACBD3: enterovirus 3A complexes and the previously known structures of the ACBD3: kobuvirus 3A complexes [23] uncovered a striking convergence in the mechanisms of how the two distinct groups of picornaviruses recruit ACBD3 and its downstream effectors to the sites of virus replication.
For the structural characterization of the enterovirus 3A proteins in complex with the host ACBD3 GOLD domain, we selected 3A proteins of six human-infecting enteroviruses each representing different species as follows: enterovirus A71 (EVA71), coxsackievirus B3 (CVB3), poliovirus 1 (PV1), enterovirus D68 (EVD68), rhinovirus A2 (RVA2), and rhinovirus B14 (RVB14) (Fig 1a).
The recombinant cytoplasmic domains of all the 3A proteins were poorly soluble and tended to aggregate and precipitate at the required concentrations. Therefore, we used 3A proteins N-terminally fused to a GB1 solubility tag. For the crystallographic analysis of the complexes composed of the ACBD3 GOLD domain and the viral 3A proteins (hereafter referred to as GOLD: 3A complexes), the GB1-fused cytoplasmic domains of the 3A proteins were directly co-expressed with the ACBD3 GOLD domain in bacteria. The GOLD: 3A complexes were then purified, and the GB1 tag was cleaved off. The GOLD: 3A complexes exhibited better protein solubility than the uncomplexed 3A proteins, sufficient for the subsequent crystallographic analysis.
Of the six GOLD: enterovirus 3A complexes, only GOLD: 3A/EVD68 and GOLD: 3A/RVB14 formed crystals that diffracted to a resolution suitable for subsequent structure determination (i.e. 2.3 Å and 2.9 Å, respectively). Both structures were solved by molecular replacement using a previously published structure of the unliganded ACBD3 GOLD domain (accession number 5LZ1 [23]) as a search model (Fig 1b, Table 1). To improve the crystallization properties of the other four GOLD: 3A complexes, we used two different strategies. The first strategy was based on mutagenesis of selected surface-exposed hydrophobic residues of the 3A proteins to improve the solubility of the respective GOLD: 3A complexes and their capability to be crystallized at higher protein concentrations. This approach led to a successful crystallization of the GOLD: 3A/PV1 complex with an L24A point mutation within the PV1 3A protein. Its structure was then solved at a resolution of 2.8 Å (Fig 1b, Table 1). The second strategy took advantage of the fact that in all three solved GOLD: 3A structures the C terminus of the ACBD3 GOLD domain was located in the vicinity of the N terminus of the ordered part of the 3A protein. This allowed us to design GOLD-3A fusion proteins with the last residue of ACBD3 (R528ACBD3) fused through a short peptide linker (GSGSG) to the first predicted ordered residues of the respective 3A proteins (e.g. K153A/EVA71). This approach led to a successful crystallization of the GOLD-3A/EVA71 fusion protein and its structure solution at a resolution of 2.8 Å (Fig 1b, Table 1). The GOLD: 3A/CVB3 and GOLD: 3A/RVA2 complexes, however, failed to form diffracting crystals even after extensive optimization using both the mutagenesis and fusion-protein strategies.
The overall structures of all solved GOLD: 3A complexes are highly similar to each other. This suggests that neither the L24A mutation in the GOLD: 3A/PV1 complex nor the fusion-protein strategy used for the GOLD: 3A/EVA71 complex affected the overall fold of the complexes (Fig 1b). No electron density was observed for the N termini of the 3A proteins (approximately the first 15 residues) and we, therefore, assume that this region is intrinsically disordered. This part of the 3A proteins has been previously reported to be involved in the interaction with another host factor GBF1 [2] or it is largely absent (e.g. RVA2) (Fig 1a).
In order to determine the strength of the interaction between the ACBD3 GOLD domain and multiple enterovirus 3A proteins in vitro, we used microscale thermophoresis (Fig 1c). The dissociation constants of the GOLD: enterovirus 3A complexes ranged approximately from 1 μM (EVD68 and RVB14) to 15 μM (EVA71).
In summary, our experiments confirmed that the enterovirus 3A proteins interact with the host ACBD3 protein through the GOLD domain of ACBD3 and the cytoplasmic domains of the 3A proteins. These proteins interact directly with dissociation constants within the low micromolar range. Using several approaches, four GOLD: enterovirus 3A complexes were crystallized and their structures were solved. Taken together, these structures document a conserved mechanism how diverse enterovirus species recruit the host ACBD3 protein.
We performed an analysis of the GOLD: 3A interface to identify amino acid residues important for the ACBD3: 3A interaction, co-localization, stimulation of PI4KB recruitment, and facilitation of virus replication in human cells. For this analysis, we chose the GOLD: EVD68 3A complex because we resolved its structure at the highest resolution. Given the high similarity of the various GOLD: enterovirus 3A structures, we assume that the conclusions drawn from the ACBD3: EVD68 3A complex can be applied to the other ACBD3: enterovirus 3A complexes as well.
In the GOLD: EVD68 3A crystal structure, we could trace the polypeptide chain of the 3A protein from T163A to I583A. It contains four secondary elements: two alpha helices P193A-V293A (α13A, Fig 2a) and Q323A-K413A (α23A, Fig 2b), and two beta strands I443A-I463A (β13A, Fig 2c) and V533A-I583A (β23A, Fig 2d). All these segments contribute to the GOLD: 3A interaction mediated through multiple hydrophobic interactions and hydrogen bonds (Fig 2a–2d). The helices α13A and α23A bind to a mild cavity of the GOLD domain that is formed by four antiparallel beta strands of ACBD3. The strand β13A interacts with the strand K518ACBD3-R528ACBD3 of the ACBD3 GOLD domain, while the strand β23A binds to the strand V402ACBD3-P408ACBD3, both in the antiparallel orientation. The conformation of all these secondary elements is highly conserved among various GOLD: enterovirus 3A complexes. The lowest homology of the tertiary structures of these complexes within short linkers between the β13A and β23A strands of the 3A proteins corresponds to the lowest homology of the primary sequences of these proteins within this region (Fig 1a and 1b).
Calculations [24] of the changes of the interaction energies of various to-alanine mutants of these complexes based on their crystal structures uncovered that multiple amino acid residues of both 3A proteins and ACBD3 are involved in the interaction (S1 Fig). To evaluate the relative importance of various segments of the 3A protein on the complex formation, we designed the following EVD68 3A mutants: NLD (N23A/L26A/D30A), QRD (Q32A/R35A/D36A), IVH (I44A/V45A/H47A), and LVK (L52A/V54A/K56A) (Fig 3a; S1 Fig, panel a). For all the mutants, the ACBD3: 3A interaction was significantly attenuated both in the mammalian-two-hybrid assay (Fig 3b) and in the co-immunoprecipitation assay (Fig 3c), confirming that all four segments of the 3A protein are important for the ACBD3: 3A interaction. Nevertheless, some residual affinity of the 3A mutants to ACBD3 was still observed. All the 3A mutants co-localized with endogenous ACBD3 in the Golgi as did the wild-type 3A protein. The lipid kinase PI4KB, however, was recruited to the Golgi significantly more effectively in the cells expressing wild-type 3A compared to the cells expressing the 3A mutants (Fig 3d and 3e). Under physiological conditions, PI4KB cycles between the cytoplasm and Golgi, where it is recruited by a direct interaction with ACBD3 [7]. In enterovirus-infected cells, the viral 3A protein has been proposed to promote the ACBD3: PI4KB interaction [11]. Thus, considering that no direct interaction between the enterovirus 3A proteins and PI4KB has ever been observed, our data indicate that the stimulation of the ACBD3: PI4KB interaction by the 3A protein and the subsequent increase of the PI4KB recruitment to target membranes in infected cells depends on the ACBD3: 3A interaction. The Golgi-localized PI4P lipid was redistributed in the 3A-expressing cells possibly due to the Golgi disintegration caused by 3A overexpression, nevertheless, no significant change in the PI4P levels was observed in the wild-type 3A-expressing cells compared to the mock-transfected or mutant 3A-expressing cells (S2 Fig). Thus, a cooperation with some other viral proteins can be required to increase the PI4KB activity during viral infection.
To analyze the impact of these 3A mutations on enterovirus replication, we established a reporter subgenomic replicon assay for EVD68. To determine the background reporter expression directly from the transfected RNA, we used a viral polymerase-lacking mutant (Δ3Dpol). Unexpectedly, no significant replication of the wild-type replicon RNA compared to the Δ3Dpol mutant was observed in HeLa cells. However, screening of several human cell lines uncovered the U-87 MG glioblastoma cells and HaCaT keratinocytes in which the wild-type replicon RNA significantly replicated. For all analyzed mutants, the viral RNA replication was attenuated in both cell lines (Fig 3f; S3 Fig). We observed no replication of the NLD, QRD, and IVH mutants, and a significantly reduced replication of the LVK mutant. Notably, this mutant was the weakest ACBD3 interactor in both co-immunoprecipitation and mammalian-two-hybrid assays, indicating additional unknown important effects distinct from the strength of the ACBD3-3A interaction affecting virus replication. We tested whether this mutant gained resistance to the PI4KB inhibition, nevertheless, we found that this mutant was still sensitive to a highly specific PI4KB inhibitor (compound 10 in Mejdrova et al. [10]) (Fig 3g).
To address the effect of mutagenesis of selected residues within ACBD3, we designed the following ACBD3 mutants: WR (W375A/R377A), VTVRV (V403A/T404A/V405A/R406A/V407A), SYLF (S414A/Y415A/L416A/F417A), and RVYYT (R523A/V524A/Y525A/Y526A/T527A) (Fig 4a; S1 Fig, panels b-c). In the mammalian-two-hybrid assay (Fig 4b) and in the co-immunoprecipitation assay (Fig 4c), all these ACBD3 mutants displayed a significantly reduced ability to interact with the 3A protein. A weak yet significant effect was observed for the SYLF and RVYYT mutants, while a strong effect resulting in no detectable interaction in both assays was achieved for the WR and VTVRV mutants. Proper intracellular localization of these ACBD3 mutants was verified by their ectopic expression in ACBD3 knock-out cells derived from HeLa cells by CRISPR/Cas9 technology [21]. All these ACBD3 mutants co-localized with giantin, an integral Golgi protein, which has been proposed to directly recruit ACBD3 to the Golgi [5] (Fig 4d).
Finally, we tested the ability of these ACBD3 mutants to rescue enterovirus replication in ACBD3 knock-out cells. The ACBD3 F258A/Q259A mutant, which does not interact with the lipid kinase PI4KB and cannot rescue virus replication [21], was used as a control. The ACBD3 WR and VTVRV mutants failed to rescue virus replication as expected. However, virus replication was still sufficiently restored by the other tested ACBD3 mutants SYLF and RVYYT (Fig 4e). These data document that the remaining affinity of these ACBD3 mutants to the viral 3A protein is still sufficient to fully facilitate enterovirus replication. Surprisingly, the SYLF mutant supports viral replication significantly better than wild-type ACBD3. It is possible that this mutation affects some other ACBD3 properties, such as its ability to interact with some other (known or unknown) proteins involved in enterovirus replication, nevertheless, the exact mechanism of the enhanced enterovirus replication in the ACBD3 SYLF mutant-expressing cells remains unclear. Compared to the ACBD3 WR mutant, the single mutants W375A and R377A still could rescue virus replication (S4 Fig, panels a-b), indicating that both mutations at the ACBD3: 3A interface are required to sufficiently disrupt the ACBD3: 3A interaction to affect virus replication. Several other tested ACBD3 mutants (such as V403A/V405A/ V407A, Y415A/F417A, and R523A/Y525A/Y526A) displayed a reduced affinity to the enterovirus 3A protein and still were able to restore enterovirus replication (S4 Fig, panels c-d). Alternatively, virus replication can be inhibited by single mutations interfering with a proper intracellular localization of ACBD3 (through ACBD3 misfolding and/or loss of the interaction with giantin) as documented by the E419A mutant (S4 Fig, panels e-f).
In conclusion, our data document that the ACBD3: 3A interaction is essential for enterovirus replication. The viral replication, however, can be facilitated by weakly interacting ACBD3 mutants, provided that they are correctly folded and localized in the Golgi in non-infected cells.
The enterovirus 3A proteins have been proposed to form homodimers [22, 25]. Analysis of the crystal structures of the GOLD: 3A complexes revealed that the 3A proteins formed either one of the crystal-packing contacts (as in the case of EVD68 and PV1) or contacts with the second 3A molecule when two GOLD: 3A complexes per asymmetric unit were present (as in the case of EVA71 and RVB14). This putative dimerization interface is formed by the two central alpha helices of the 3A proteins, which are bent 180° to form a helical hairpin (Fig 5a). These helices are amphipathic with several hydrophobic residues oriented towards the hydrophobic residues of the other 3A monomer. Surprisingly, the C termini of the 3A proteins, which in the cellular environment are anchored to the membranes, are located on the opposite sides of the GOLD: 3A heterotetramers. Therefore, we were interested whether the plasticity and flexibility of the 3A dimerization interface together with the plasticity of the lipid bilayer allows to form the GOLD: 3A heterotetramers at the intracellular membranes.
To identify amino acid residues of the 3A proteins involved in the dimerization of the GOLD: 3A complexes, we calculated [24] the changes of the dimerization energies of various to-alanine mutants of these complexes based on the crystal structures (S5 Fig, panel a). The dimerization interface of the GOLD: 3A/EVD68 complex consists of the hydrophobic core formed by the residues L25, V29, V34, and Y37, and an additional intermolecular salt bridge between the residues D24 and K41 (Fig 5a). To analyze the dimerization of the GOLD: 3A complexes in more detail, we generated a mutant EVD68 3A protein (hereafter referred to as LVVY mutant) with the following four mutations at the putative dimerization interface: L25A, V29A, V34A, and Y37A. As expected, retention volumes of the recombinant wild-type 3A and its LVVY mutant in size exclusion chromatography were significantly shifted corresponding to the dimeric and monomeric state of the wild-type 3A and its LVVY mutant, respectively (Fig 5b).
At the request of a reviewer of our manuscript, we analyzed the dimerization of the L25V, V29Y, L25V/V34L, and V29Y/Y37V mutants (S6 Fig). Both L25V and V29Y mutations attenuated the 3A dimerization. The dimerization of the L25V mutant was restored by the V34L mutation, likely due to the compensation of weakening the L25-L25 interaction by strengthening the V34-V34 and V34-V29 interactions. On the other hand, the potential "rescue" Y37V mutation had a negative impact on the 3A dimerization, likely due to the attenuation of the Y37-L25 interaction and a loss of the hydrogen bond between Y37 and D24 (S6 Fig).
Next, we investigated the stoichiometry of the GOLD: EVD68 3A complexes. To ensure that the 3A protein is fully complexed with the ACBD3 GOLD domain and to avoid the formation of partial complexes with 1:2 stoichiometry, we designed a GOLD-EVD68 3A fusion protein using a similar approach as for the GOLD-EVA71 3A fusion protein used for the crystallographic analysis as described earlier. Taking advantage of the vicinity of the C terminus of the ACBD3 GOLD domain and the N terminus of the ordered part of the EVD68 3A protein, we connected the last residue of ACBD3 (R528ACBD3) through a short peptide linker (GSGSG) to the first ordered residue of the EVD68 3A protein (T163A/EVD68) (S5 Fig, panel b). Both GOLD-3A wild-type and LVVY mutant fusion proteins formed crystals, which diffracted to a resolution suitable for further structure determination (S5 Fig, panel c). The crystal structures of the GOLD: 3A complex consisting of two individual proteins, the GOLD-3A fusion protein, and its LVVY mutant were almost identical with low root-median-square deviations (S5 Fig, panel b), confirming that neither the fusion-protein approach nor the LVVY mutation affected the correct folding of these proteins.
Three lines of evidence document the dimeric state of the wild-type GOLD-3A fusion protein and the monomeric state of its LVVY mutant in vitro. First, the retention volumes of these proteins in size exclusion chromatography are significantly shifted (Fig 5c). Secondly, the small-angle X-ray scattering (SAXS) profiles of these proteins correspond to the calculated scattering curves of a dimer of the wild-type GOLD-3A fusion protein (Fig 5d; S7 Fig, panel a) and of a monomer of its LVVY mutant (Fig 5e). Thirdly, crystal contacts corresponding to the 3A dimerization interface are not preserved in the crystal structure of the GOLD-3A LVVY mutant (S7 Fig, panels b-c), indicating that this mutant cannot dimerize through this interface even at very high protein concentrations (of approximately 20 mM) present within the protein crystal.
Next, we investigated the stoichiometry of the GOLD: 3A complexes in cells. For this purpose, we ectopically co-expressed either wild-type GOLD-3A fusion protein or its LVVY mutant N-terminally fused to mAmetrine and mPlum fluorescent proteins in HeLa cells and determined the Förster resonance energy transfer (FRET) efficiency by flow cytometry. Owing to the crowding effect, the energy transfer was observed in the case of both proteins. Nevertheless, we observed a significant difference in FRET efficiency between the wild-type GOLD-3A fusion protein and its LVVY mutant (Fig 5f and 5g). These results confirm that the GOLD: 3A complexes are flexible enough to allow the formation of the heterotetramers consisting of two molecules of the viral 3A protein and two molecules of host ACBD3 even in cells at the respective intracellular membranes (Fig 5h).
Finally, we analyzed the impact of the LVVY mutation on enterovirus replication. Using a reporter subgenomic replicon assay for EVD68 established earlier, we found replication of this mutant significantly attenuated in both U-87 MG and HaCaT cells (Fig 5i; S3 Fig). These findings document that the intact dimerization interface of the viral 3A proteins is required for enterovirus replication.
In a previous study [23], we identified a novel ACBD3 membrane binding site (MBS) consisting of the residues R399, L514, W515, and R516. The hydrophobic residues L514 and W515 can be inserted directly into the lipid bilayer, while the positively charged residues R399 and R516 can interact with the negatively charged phospholipid head groups (Fig 6a). A vicinity of ACBD3 MBS and the expected position of the transmembrane domain of the enterovirus 3A protein within the ACBD3: 3A complexes suggests that ACBD3 MBS may be involved in the stabilization of the ACBD3: 3A complexes at the membrane as well.
To experimentally evaluate this hypothesis, we designed the following ACBD3 mutants with several point mutations within MBS: LWR514AAA and, to increase repulsion between ACBD3 MBS and the lipid bilayer, LWR514DDD. Then, we ectopically expressed wild type ACBD3 or its MBS mutants N-terminally fused to EGFP in HeLa ACBD3 knock-out cells. We found that wild-type ACBD3 co-localized with the Golgi marker giantin, while both ACBD3 MBS mutants LWR514AAA and LWR514DDD were mostly released to the cytoplasm, although minor yet significant fractions of their pools were still preserved at the Golgi (Fig 6b). Remarkably, when the ACBD3 MBS mutants were co-expressed with the enterovirus 3A protein, they were completely (LWR514AAA mutant) or partially (LWR514DDD mutant) re-localized back to the Golgi (Fig 6c and 6d). Thus, an intact MBS is required for ACBD3 recruitment to the Golgi by the action of giantin or other cellular factors, however, it is dispensable for ACBD3 stabilization at target membranes through its interaction with enterovirus 3A proteins.
Finally, we tested the capacity of wild-type ACBD3 and its MBS mutants to rescue virus replication in ACBD3 knock-out cells. Both ACBD3 wild type and the LWR514AAA mutant, but not the LWR514DDD and FQ258AA (used as a control [21]) mutants, effectively restored virus replication (Fig 6e). Thus, it seems that not ACBD3 MBS itself but rather the orientation of the ACBD3: 3A complex with respect to the membrane plays a role in facilitation of enterovirus replication. To-alanine mutations of ACBD3 MBS still allow the ACBD3: 3A complex at the membrane to adopt a conformation suitable for viral replication. On the contrary, to-aspartate mutations of ACBD3 MBS, which repel the negatively charged phospholipids of the lipid bilayer, result in an orientation of the ACBD3: 3A complex with respect to the membrane that does not support enterovirus replication.
In summary, ACBD3 MBS is not required for ACBD3 recruitment to target membranes by the action of the enterovirus 3A proteins, however, the proper conformation of the ACBD3: 3A complexes at the membrane mediated by ACBD3 MBS is essential for enterovirus replication.
Considering the commonness of enterovirus-mediated infections within human population, it is surprising that no antiviral therapy for enteroviruses has been approved yet. Targeting essential host factors instead of rapidly mutating viral enzymes represents a promising strategy. Several host factors essential for enterovirus replication are recruited to the sites of viral replication by a direct protein-protein interaction between the host factor and a viral protein. A detailed knowledge of the structures of such complexes can open up prospects for a structure-guided development of small chemical compounds targeting these interactions, yielding a novel class of antivirals to combat infections caused by these pathogens.
In this study, we present a series of crystal structures of complexes composed of the non-structural 3A proteins of four enterovirus species and the 3A-binding GOLD domain of the host factor ACBD3. Previously, the genetic inhibition of ACBD3 mediated by siRNA has yielded conflicting results on the importance of ACBD3 for virus replication [26, 27]. This conflict in the literature has been recently addressed using CRISPR/Cas9-generated ACBD3 knock-out cells, in which enterovirus replication was severely impeded [8, 21]. This confirmed that ACBD3 is, indeed, an essential host factor for enterovirus replication. However, it seems that a very low concentration of ACBD3 within the cells is still fully capable of facilitating enterovirus replication. This hypothesis is in agreement with our observations that enterovirus replication in ACBD3 knock-out cells can be restored by several ACBD3 mutants with a very low affinity to the viral 3A proteins even at the detection limit of conventional methods assessing the protein-protein interactions, such as protein co-immunoprecipitation.
Among picornaviruses, the interaction between the viral 3A protein and host ACBD3 is not unique for enteroviruses. ACBD3 has been proposed to interact also with the 3A proteins of kobuvirus (e.g. aichivirus), hepatovirus, salivirus (klassevirus), and parechovirus, but not with those of cardiovirus (e.g. Saffold virus) or aphthovirus (foot-and-mouth disease virus, FMDV) [6]. To our best knowledge, all picornaviruses sensitive to PI4KB specific inhibitors (such as enteroviruses and kobuviruses) are able to hijack ACBD3, arguing for ACBD3 as a main mediator of PI4KB recruitment by these viruses. Notably, hepatovirus replicates independently of PI4KB [28], indicating either functionally irrelevant interaction with ACBD3 or another, PI4KB-independent, role of ACBD3 in hepatovirus replication. Picornaviruses that cannot hijack ACBD3 through their 3A proteins are either PI4P-independent (such as FMDV [29]) or their replication depends on another PI4P-producing lipid kinase PI4KA (e.g. cardioviruses [30]).
During the past decades, multiple enterovirus mutants resistant to the inhibitors of PI4KB and the oxysterol binding protein (OSBP), which acts downstream of PI4KB, were isolated and characterized. Most of the resistance-conferring mutations were localized to the 3A-encoding regions of these viruses, e.g. PV1 N45Y, R54W, N57D, A70T, and A71S, CVB3 V45A, I54F, and H57Y, and RVB14 E30D/V/Q, I42V, and M54I (S8 Fig) [31–34]. Although many of the PI4KB/OSBP-inhibition resistance-conferring mutations are localized within the ACBD3-interacting regions of the 3A proteins, they seem unlikely to act through modulation of the ACBD3-3A interaction. On the other hand, it is possible that the mutations clustered within the β2 strands of the 3A proteins, such as CVB3 H57Y or PV1 R54W, can act through modulation of the interaction of the ACBD3-3A complex (or the uncomplexed 3A protein) with the membrane. The loss of the positive charge of the mutated residues can possibly compensate for the loss of the negative charge of the PI4P head groups upon PI4KB inhibition. Nevertheless, at least the mechanism of action of the mutations located distally with respect to the membrane, mostly clustered within the β1 strands of the 3A proteins, remains unclear.
Apart from the crystal structures of the enterovirus 3A: GOLD complexes, to date only the structures of the kobuvirus 3A: GOLD complexes are known [23]. The enterovirus (e.g. poliovirus) and kobuvirus (e.g. aichivirus) 3A proteins share a common overall architecture, i.e. a similar size of approximately 10 kDa, large N-terminal soluble cytoplasmic domains followed by hydrophobic membrane-anchoring regions and small C-terminal domains (Fig 7a). Despite this common architecture, primary and predicted secondary structures of the enterovirus and kobuvirus 3A proteins are unrelated and cannot be aligned. Furthermore, positions of the ACBD3 binding regions of the enterovirus and kobuvirus 3A proteins are distinct. The ACBD3 binding region of the enterovirus 3A proteins forms the C-terminal segments of the cytoplasmic domain (and is preceded by the N-terminal GBF1 binding region), while the ACBD3 binding region of the kobuvirus 3A proteins is located at the N terminus (and a GBF1 binding region is completely missing). Superposition of the crystal structures of the enterovirus and kobuvirus 3A: GOLD complexes reveals that the enterovirus and kobuvirus 3A proteins bind to the same regions of the ACBD3 GOLD domain, nevertheless, the polypeptide chains of the enterovirus and kobuvirus 3A proteins have opposite polarities (Fig 7b and 7c). For instance, the poliovirus 3A strand β23A/PV1 binds in the antiparallel orientation to the strand V402ACBD3-P408ACBD3 of ACBD3, while the aichivirus 3A strand β13A/AiV1 binds at the same position to the same strand of ACBD3, but in the parallel orientation (Fig 7b and 7c). The reverse orientation of the enterovirus 3A proteins compared to the kobuvirus 3A proteins may be caused by the specific need of the enterovirus 3A proteins to bind GBF1.
A notable difference between the enterovirus and kobuvirus 3A proteins is represented in the way they are anchored to the membrane. In addition to the C-terminal hydrophobic membrane binding regions, kobuvirus 3A proteins are membrane-anchored by the myristoylated N-terminal glycines, which is very unusual among picornaviruses [3]. Among enteroviruses, the N-terminal myristoylation is not present. To gain more insight into the membrane binding mode of the enterovirus 3A: GOLD protein complex, we performed all-atom molecular dynamics simulation of this complex at the membrane and compared it with our previously published [23] simulation of the kobuvirus 3A: GOLD protein complex at the surface of the lipid bilayer (Fig 7d). These simulations uncovered a similar conformation of the ACBD3 GOLD domain recruited to the membrane by the poliovirus or aichivirus 3A protein, including the insertion of the ACBD3 membrane binding site residues into the lipid bilayer. The position of the N-terminal myristoylation of the aichivirus 3A protein functionally substitutes the position of the C-terminal transmembrane domain of the poliovirus 3A protein, whereas the C-terminal transmembrane domain of the aichivirus 3A protein has no equivalent in the case of the poliovirus 3A protein.
In summary, our findings reveal structural details of how two groups of viral pathogens, enteroviruses and kobuviruses, developed a similar mechanism of hijacking the same host factor (ACBD3) and its downstream effectors (such as PI4KB). These viruses use their 3A proteins with a common architecture yet totally unrelated primary sequences to bind to the same regions of the host ACBD3 protein in the opposite orientations, representing a striking case of convergence in picornavirus evolution. Our results are in agreement with a pioneering work by Greninger and colleagues [6, 35], which has forecast such convergent evolutionary strategies of kobuviruses and enteroviruses based on extensive mutagenesis of the ACBD3-3A interface. There are still several other picornavirus genera proposed to recruit ACBD3 through the 3A: GOLD interaction (such as salivirus, hepatovirus, or parechovirus [6]) whose 3A: GOLD complexes remain structurally unexplored. Structural details of their ACBD3 recruitment uncovering whether they utilize the mechanism described for enteroviruses, kobuviruses, or another mechanism distinct from those two, remain to be further elucidated.
For expression in E. coli, full-length human ACBD3 and various enterovirus 3A proteins and their deletion mutants were cloned into pRSFD vector (Novagen) with an N-terminal 6xHis tag followed by a GB1 solubility tag and a TEV protease cleavage site using PCR and restriction cloning. For bacterial expression of the EGFP-fusion proteins, the EGFP encoding sequence was inserted between the TEV cleavage site and the target gene encoding regions. For expression of the EGFP-fusion proteins in human cells, target genes encoding regions were recloned into pEGFP-C1 vector (Clontech) with an N-terminal EGFP tag. For expression of the GST-, mAmetrine-, and mPlum-fusion proteins in human cells, the EGFP encoding region was replaced by GST or corresponding fluorescent protein encoding sequence by PCR and restriction cloning. The pRib-EVD68/mCherry plasmid for viral subgenomic replicon assays was generated by subcloning of the EVD68 cDNA of a prototypical Fermon strain under T7 promoter and replacing the capsid proteins-encoding region with the mCherry fluorescent protein-encoding gene by Gibson assembly. Mutations were generated using the Q5 Site-Directed Mutagenesis Kit (New England BioLabs). All DNA constructs were verified by sequencing. The pEGFP-ACBD3 and pBIND-ACBD3 plasmids were kindly provided by Carolyn Machamer and Jun Sasaki, respectively. The mAmetrine-C1 and mPlum-C1 plasmids were gifts from Robert Campbell and Michael Davidson (Addgene plasmids #54660 [36] and #54839 [37]). The pEGFP-GalT plasmid was a gift from Jennifer Lippincott-Schwartz (Addgene plasmid #11929 [38]).
All recombinant proteins used in this study were bacterially expressed as fusion proteins with an N-terminal 6xhistidine (His6) tag followed by a GB1 solubility tag and a TEV protease cleavage site. For the crystallographic analysis of the GOLD: 3A complexes, the N-terminally His6-GB1-TEV site-fused cytoplasmic domains of the 3A proteins were directly co-expressed with the untagged ACBD3 GOLD domain. The proteins were expressed in E. coli BL21 DE3 NiCo cells (New England Biolabs) using the autoinduction ZY medium. Bacterial cells were harvested and lysed in the lysis buffer (50 mM Tris pH 8, 300 mM NaCl, 3 mM β-mercaptoethanol, 30 mM imidazole, 10% glycerol), the lysate was incubated with the HisPur Ni-NTA Superflow agarose (Thermo Fisher Scientific), and the bound proteins were extensively washed with the wash buffer (50 mM Tris pH 8, 300 mM NaCl, 1 mM β-mercaptoethanol, 20 mM imidazole). The protein was eluted with the elution buffer (50 mM Tris pH 8, 200 mM NaCl, 3 mM β-mercaptoethanol, 300 mM imidazole).
For the biochemical analysis of the 3A proteins by microscale thermophoresis or SAXS, the N-terminal His6-GB1 tags were preserved uncleaved to increase protein solubility and to avoid aggregation at required concentrations. For the crystallographic analysis of the GOLD: 3A complexes and for the biochemical analysis of the GOLD-3A fusion proteins by SAXS, the N-terminal His6-GB1 tags were removed with home-made TEV protease. Next, the proteins were purified using the size exclusion chromatography at HiLoad 16/60 Superdex 75 or Superdex 200 prep grade columns (GE Healthcare) in the storage buffer (10 mM Tris pH 8, 200 mM NaCl, 3 mM β-mercaptoethanol). In addition, the GOLD: 3A complexes used for the crystallographic analysis were further purified by reverse immobilized metal affinity chromatography using the HisTrap HP column (GE Healthcare), while the EGFP-fused ACBD3 GOLD domain used for microscale thermophoresis was further purified using the ion exchange chromatography at a MonoQ 10/100 GL column (GE Healthcare) and then dialyzed back into the storage buffer. The molecular weight and purity of all proteins was verified by SDS-PAGE (S9 Fig) and Matrix-Assisted Laser Desorption/Ionisation (MALDI). Purified proteins were concentrated to 1–10 mg/ml, aliquoted, flash frozen in the liquid nitrogen, and stored at -80 °C until needed.
Crystals grew at 291 K in sitting drops by the vapor diffusion method. They were cryoprotected, flash frozen in liquid nitrogen, and analyzed. Measurements were carried out at the MX14.1 beamline of the synchrotron BESSY II at Helmholtz-Zentrum Berlin [39]. The crystallographic datasets were collected from single frozen crystals. Data were integrated and scaled using XDS [40] and XDSAPP [41]. Structures were solved by molecular replacement using the uncomplexed ACBD3 GOLD domain structure (pdb code 5LZ1) as a search model. The initial models were obtained with Phaser [42] from the Phenix package [43]. The models were further improved using automatic model building with Buccaneer [44] from the CCP4 suite [45], automatic model refinement with Phenix.refine [46] from the Phenix package [43], and manual model building with Coot [47]. Statistics for data collection and processing, structure solution and refinement are summarized in Table 1. Structural figures were generated with PyMol [48]. The atomic coordinates and structural factors were deposited in the Protein Data Bank (www.pdb.org).
MST measurements were carried out using the Monolith NT.115 instrument (NanoTemper Technologies) according to the manufacturer's instructions. The Monolith NT.115 standard treated capillaries were loaded with a mixture of a recombinant EGFP-fused protein at a constant concentration of 150 nM in the MST buffer (30 mM Tris pH 7.4, 150 mM NaCl, 3 mM β-mercaptoethanol) and its binding partner in the appropriate series of concentrations. The thermophoretic motion of the fluorescent protein and its temperature-dependent changes of fluorescence were analyzed with the Monolith NT Analysis Software.
Human cervical-carcinoma cells HeLa (American Type Culture Collection / ATCC), embryonic kidney cells HEK293T (ATCC), and keratinocytes HaCaT (AddexBio) were maintained in Dulbecco's modified Eagle's medium (Sigma) supplemented with 10% fetal calf serum (Gibco). Human glioblastoma cells U-87 MG (ATCC) were maintained in Minimum Essential Medium Eagle (Sigma) supplemented with 10% fetal calf serum (Gibco), GlutaMAX Supplement (Thermo Fisher Scientific), and non-essential amino acids (Biowest). HeLa cells were transfected using Lipofectamine2000 reagent (Thermo Fisher Scientific) or X-tremeGENE HP DNA Transfection reagent (Sigma/Roche) according to manufacturer's instructions. Transfections of HEK293T cells were performed using polyethylenimine (Sigma) or Fugene6 (Promega).
HEK293T cells were transfected with the appropriate mutants of the EGFP-fused EVD68 3A protein and GST-fused ACBD3. The next day, cells were harvested, washed twice with phosphate-buffered saline (PBS) and lysed in the ice-cold lysis buffer (20 mM Tris pH 7.4, 100 mM NaCl, 50 mM NaF, 10 mM EDTA, 10% glycerol, 1% NP-40), supplemented with protease inhibitors (Complete protease inhibitor cocktail, Sigma/Roche). After solubilizing for 15 min on ice, the lysate was pre-cleared by centrifugation at 16,000g for 15 min. The resulting supernatant was incubated with sepharose beads coupled either to GFP nanobody (GFP-Trap, ChromoTek) or glutathione (Protino Glutathione Agarose, Macherey-Nagel) for 1h at 4 °C. After three washes with 10 volumes of the lysis buffer, the bound proteins were directly eluted with the Laemmli sample buffer, subjected to SDS-PAGE, and analyzed by immunoblotting. The whole cell lysates and eluted proteins were stained with mouse monoclonal antibodies to ACBD3 (Santa Cruz Biotechnology, sc-101277) and GFP (Santa Cruz Biotechnology, sc-9996). The images were acquired using the LI-COR Odyssey Infrared Imaging System.
HeLa cells were co-transfected with plasmids encoding target proteins fused to the FRET pair of fluorescent proteins mAmetrine ("donor") and mPlum ("acceptor"). The next day, cells were harvested, washed twice with PBS, and analyzed by flow cytometry using BD LSR Fortessa (BD Biosciences). The donor and acceptor fluorescence as well as the energy transfer was determined using the optical configurations as follows: mAmetrine—405 nm laser—525/50 nm bandpass filter; mPlum—561 nm laser—670/30 nm bandpass filter; FRET—405 nm laser—655/8 nm bandpass filter. Acquired data were analyzed with the FlowJo software. The acquired fluorescence intensities were compensated and the same gate corresponding to the live transfected cells with the approximately 1:1 donor:acceptor ratio was applied to all samples. The acquired events were binned on the basis of the acceptor fluorescence intensity, and the average FRET fluorescence intensities of each bin were plotted against the respective acceptor fluorescence intensity.
HEK293T cells grown in 96-well plates were co-transfected with 50 ng of each pACT, pBIND, and pG5Luc plasmids using Fugene6 (Promega). At 24 hours post transfection, the cells were lysed, and both firefly and Renilla luciferase activities were measured using the Dual-Luciferase assay kit (Promega) and Centro LB 960 luminometer (Berthold Technologies) according to the manufacturer's instructions. The firefly luciferase activity was normalized to the Renilla luciferase activity (used as an internal control of the transfection efficiency) and then to the activity determined in cells co-expressing wild-type ACBD3 and 3A (which was set to 100%).
HeLa cells grown on coverslips in 24-well plates were transfected with 400 ng of the plasmid DNA using Lipofectamine2000 (Thermo Fisher Scientific). At 16 hours post transfection, the cells were fixed with 4% paraformaldehyde for 15 min at room temperature, permeabilized with 0.1% Triton X-100 in PBS for 5 min, and immunostained with the appropriate primary and secondary antibodies diluted in 2% normal goat serum in PBS. Sources of the antibodies were as follows: anti-ACBD3 (Sigma, WH0064746M1), anti-PI4KB (Merck, 06–578), anti-GM130 (BD Biosciences, 610822), anti-giantin (Enzo Life Science, ALX-804-600-C100), anti-myc (Thermo Fisher Scientific, PA1-981), anti-PI4P (Echelon, Z-P004), and goat-anti-mouse and goat-anti-rabbit secondary antibodies conjugated to AlexaFluor 488, 596, or 647 (Molecular Probes). Nuclei were stained with DAPI. Coverslips were mounted with FluorSave (Calbiochem), and confocal imaging was performed with a Leica SpeII confocal microscope.
The pRib-EVD68/mCherry wild-type and mutant plasmids were linearized by cleavage with SalI-HF (Thermo Fisher Scientific) and purified using the mini spin columns (Epoch Life Science). Viral subgenomic replicon RNA was generated with TranscriptAid T7 High Yield Transcription Kit (Thermo Fisher Scientific) and purified using the RNeasy mini spin columns (Qiagen). For replicon assays, U-87 MG or HaCaT cells grown in 12-well plates were transfected with T7-amplified RNA using the TransIT mRNA transfection kit (Mirus Bio). At 12 hours post transfection, the reporter mCherry fluorescence was determined by flow cytometry using BD LSR Fortessa (BD Biosciences) and the following optical configuration: 561 nm laser, 670/30 nm bandpass filter. Acquired data were analyzed with the FlowJo software. The level of RNA replication was expressed as a transfection efficiency-normalized percentage of cells with the mCherry signal above the threshold determined using the viral polymerase-lacking mutant Δ3Dpol.
To test the effect of PI4KB inhibition on virus replication, a PI4KB-specific inhibitor (compound 10 from [10] kindly provided by Radim Nencka) was added to the medium at a final concentration of 1 μM 30 min prior transfection of the viral subgenomic replicon RNA.
Wild-type or ACBD3 knock-out HeLa cells grown in 96-well plates were transfected with plasmids encoding wild-type or mutant ACBD3 or another Golgi-resident protein GalT as a control. At 24 hours post transfection, the cells were infected with the Renilla luciferase-expressing CVB3 virus (RLucCVB3) [49]. At 8 hours post infection, the intracellular Renilla luciferase activity was determined using the Renilla luciferase assay system (Promega) and a Centro LB 960 luminometer (Berthold Technologies). The Renilla luciferase activity was normalized to the activity determined in wild-type ACBD3-expressing cells (which was set to 100%).
Proteins were dialyzed against the SAXS buffer (30 mM Tris pH 7.4, 150 mM NaCl, 1 mM TCEP) and concentrated as follows: c1 = 1.03 mg/ml, c2 = 1.4 mg/ml and c3 = 1.54 mg/ml for the wild-type GOLD-3A fusion protein, and c1 = 0.96 mg/ml, c2 = 2.36 mg/ml and c3 = 3.26 mg/ml for the GOLD-3A LVVY mutant. The SAXS data were collected using the beamlines BioSAXS Beamline BM29 (ESRF, Grenoble) and EMBL SAXS beamline P12 (Petra III DESY, Hamburg) that are both equipped with the 2M Pilatus detector (Dectris). The three datasets overlay after rescaling, indicating no protein aggregation in the samples. To structurally interpret the SAXS data, we incorporated the missing loop (D437-K473) into the structures of the wild-type GOLD-3A dimer and GOLD-3A LVVY mutant monomer. Next, we performed the coarse-grained molecular simulations [50] in which only the conformations of the D437-K473 loop were sampled while the crystallized portion of the protein was kept rigid, yielding 10,000 structural models of both proteins. For all structural models, we computed the SAXS intensity profiles using the previously developed algorithms [51], compared them individually to the experimental SAXS data, and selected the models best fitting the SAXS data collected on the samples with the highest protein concentrations. The best models fit the SAXS data with χ = 1.5 for the wild-type GOLD-3A dimer and χ = 1.4 for the GOLD-3A LVVY mutant monomer.
The intrinsically disordered region of the ACBD3 GOLD domain, which was missing in the crystal structure (D437-K473), was modeled as described previously [23]. The C-terminal segment of the poliovirus 3A protein (N57-Q87) was modeled as a loop (N57-M60) followed by a transmembrane alpha helix (T61-F83) and a short C-terminal tail (A84-Q87). This segment was positioned in a planar segment of a lipid bilayer using the PPM server [52].
MD simulations of the ACBD3 GOLD domain in complex with the poliovirus 3A protein in the environment of a lipid bilayer were performed following the procedures recently used to study the ACBD3 GOLD domain in complex with the aichivirus 3A protein [23]. The initial system for MD simulations was prepared using VMD version 1.9.2 [53]. Namely, a POPC bilayer segment with the lateral dimensions of 10 nm by 10 nm was formed using the Membrane Plugin version 1.1 in VMD. The GOLD: 3A complex was placed on top of the resulting lipid patch. The lipids overlapping with the transmembrane alpha helix of the 3A protein were removed. The system was solvated using the Solvate Plugin version 1.5 in VMD. Sodium and chloride ions were added to neutralize the simulated system and to reach a physiological concentration of 150 mM. The MD simulations were performed using the NAMD package [54] version 2.9. The CHARMM22 force field [55, 56] with the CMAP correction [57] and the TIP3P water model were used. The simulations were carried out in the NPT ensemble. Temperature was kept at 298K through a Langevin thermostat with a damping coefficient of 1/ps. Pressure was maintained at 1 atm using the Langevin piston Nose-Hoover method with a damping timescale of 50 fs and an oscillation timescale of 100 fs. Short-range non-bonded interactions were cut off smoothly between 1 and 1.2 nm. Long-range electrostatic interactions were computed using the particle-mash Ewald method with a grid spacing of 0.1 nm. Simulations were performed with an integration time step of 2 fs. After initial energy minimization with a conjugate gradient method, a 10 ns simulation was performed with constraints on the protein backbone atoms in order to equilibrate the lipids, ions and water molecules. Namely, a harmonic potential with the spring constant of 5 kcal/(mol Å2) was applied to all backbone atoms of the GOLD: 3A complex. After the equilibration, the system was simulated with no constrains for 200 ns. The resulting MD trajectory was visualized and analyzed using VMD.
To localize the PI4KB/OSBP-inhibition resistance-conferring mutations within the structure of the GOLD: CVB3 3A complex, a homology model of this complex was generated by the I-TASSER server [58] using the crystal structure of the GOLD: EVD68 3A complex as a template.
In the graphs, data are presented as mean values ± standard errors of the means (SEMs) based on three independent experiments, unless stated otherwise. For statistical analyses, two-tailed two-sample Student's t-tests were used. P-values below 0.05 were considered significant.
The crystal structures of the ACBD3 GOLD domain in complex with the 3A proteins from EVD68, RVB14, and PV1 (L24A mutant), and the ACBD3 GOLD domain fused to the 3A proteins from EVA71, EVD68 (wild type), and EVD68 (LVVY mutant) from this publication have been submitted to the Protein Data Bank (www.pdb.org) and assigned the identifiers 6HLN, 6HLT, 6HLV, 6HLW, 6HM8, and 6HMV, respectively.
|
10.1371/journal.pgen.1000038 | Fine Mapping of the Psoriasis Susceptibility Locus PSORS1 Supports HLA-C as the Susceptibility Gene in the Han Chinese Population | PSORS1 (psoriasis susceptibility gene 1) is a major susceptibility locus for psoriasis. Several fine-mapping studies have highlighted a 300-kb candidate region of PSORS1 where multiple biologically plausible candidate genes were suggested. The most recent study has indicated HLA-Cw6 as the primary PSORS1 risk allele within the candidate region in a Caucasian population. In this study, a family-based association analysis of the PSORS1 locus was performed by analyzing 10 polymorphic microsatellite markers from the PSORS1 region as well as HLA-B, HLA-C and CDSN loci in 163 Chinese families of psoriasis. Five marker loci show strong evidence (P<10−3), and one marker locus shows weak evidence (P = 0.04) for association. The haplotype cluster analysis showed that all the risk haplotypes are Cw6 positive and share a 369-kb region of homologous marker alleles which carries all the risk alleles, including HLA-Cw6 and CDSN*TTC, identified in this study. The recombinant haplotype analysis of the HLA-Cw6 and CDSN*TTC alleles in 228 Chinese families showed that the HLA-Cw6−/CDSN*TTC+ recombinant haplotype is clearly not associated with risk for psoriasis (T∶NT = 29:57, p = 0.0025) in a Chinese population, suggesting that the CDSN*TTC allele itself does not confer risk without the presence of the HLA-Cw6 allele. The further exclusion analysis of the non-risk HLA-Cw6−/CDSN*TTC+ recombinant haplotypes with common recombination breakpoints has allowed us to refine the location of PSORS1 to a small candidate region. Finally, we performed a conditional linkage analysis and showed that the HLA-Cw6 is a major risk allele but does not explain the full linkage evidence of the PSORS1 locus in a Chinese population. By performing a series of family-based association analyses of haplotypes as well as an exclusion analysis of recombinant haplotypes, we were able to refine the PSORS1 gene to a small critical region where HLA-C is a strong candidate to be the PSORS1 susceptibility gene.
| Psoriasis is a common skin disease with strong genetic risk. The analysis of psoriatic families with multiple affected individuals has identified several genomic regions that are linked (showing linkage evidence) to the development of psoriasis. Of them, the region on 6p21.3 (PSORS1) is a well-confirmed major risk locus. The identification of the disease risk gene within the PSOR1 locus, however, has been difficult, largely due to the fact that several genes show evidence for association with the disease development and the evidences are highly correlated and hard to be separated from each other. In this study, we performed a fine mapping study of the PSORS1 locus in Chinese families with psoriasis. By analyzing recombinant haplotypes that carry different risk-associated genetic variants within different genes, we were able to separate the genetic effects observed within multiple genes and identify strong supporting evidence for HLA-C to be the primary risk gene of the PSORS1 locus. We have further demonstrated that the genetic variation within the HLA-C gene does not explain the full linkage evidence at the PSORS1 locus, suggesting that there might be other risk genes and/or alleles within the region. Our findings have improved our understanding about the genetic complexity of the PSORS1 locus.
| Psoriasis (OMIM*177900) is a common chronic inflammatory skin disorder affecting approximately 2–5% of Caucasian population [1] and 0.123% of the Chinese population [2]. It has long been widely accepted that psoriasis is a complex disease involving multiple genetic and environmental factors. Past genome-wide linkage analyses have identified nine susceptibility loci (designated PSORS1–9) and an additional 13 suggestive linkage loci for psoriasis. Among them, the PSORS1 locus on 6p21.3 is a well-confirmed major susceptibility locus for psoriasis, which accounts for about 30% to 50% of the genetic contribution to the disease [3],[4],[5].
Many fine-mapping studies have been performed to refine the localization of the PSORS1 gene. By using HLA types as markers, Schmitt-Egenolf et al [6],[7] shown that familial psoriasis is associated with the class I end of the EH57.1 haplotype. By performing a linkage disequilibrium (LD) mapping in the Caucasian population, Balendran et al [8] suggested a 285-kb critical region for PSORS1 between the makers tn821 and HLA-C. Oka et al [9] further mapped the PSORS1 gene to a 111 kb interval telomeric to HLA-C through an association analysis in the Japanese population. Similarly, Orru et al's study also highlighted a 70-kb critical region around the CDSN gene that is not recombinant with PSORS1 by identical-by-descent (IBD) haplotyping analysis in a Sardinian population [10]. A 150-kb region telometic to HLA-C was also shown to be associated with psoriasis in a Jewish population [11]. More recently, Lench et al [12] performed a SNP-based association analysis and found strong association to a 46-kb interval telomeric to HLA-C in both Caucasian and Japanese populations. Helm et al [13] also performed a comprehensive case/control and family-based association study using SNPs and located the PSORS1 to a haplotype block harboring HLA-C and distinct from CDSN and HCR. Nair et al [14] carried out a comprehensive analysis of the MHC region and narrowed the candidate interval for PSORS1 to a 224-kb region in an American Caucasian population. Taken together, although the results are not totally consistent with each other, these fine-mapping studies generally suggested a critical region of about 300 kb for PSORS1, containing HLA-C and at least 10 other genes.
As the PSORS1 locus contains the major histocompatibility complex (MHC) that is known to be involved in many autoimmune disorders, MHC genes have been intensively evaluated as the candidates for the PSORS1 susceptibility gene. HLA-Cw6 has long been known to be associated with susceptibility to psoriasis [15] and was suggested to be a marker allele in LD with the PSORS1 susceptibility allele because of the lack of its functional role in psoriasis [16],[17],[18]. However, a recent study indicated that there are distinct differential expression patterns of HLA-C in psoriasis and eczema, suggesting a functional role of HLA-C in psoriasis-related immune response rather than general inflammation [19]. In addition, several other genes within the PSORS1 locus, including MICA [20], CDSN [21],[22],[23], HCR [24],[25],[26],[27],[28], and PSORS1C3 [29],[30], were shown to be expressed in skin cells and are, therefore, plausible candidates for the PSORS1 gene. Other genes like SEEK1 and SPR1 genes were also suggested as the candidates for PSORS1 [31]. Through a systematic screening of densely distributed SNPs within a 150-kb region around HLA-C by family-based analysis, Veal et al [32] identified the strongest association evidence at two SNPs (n. 7 and n. 9) that are located 4 and 7 kb centromeric to HLA-C, respectively. Using the haplotype sharing statistic HSS, Foerster et al [33] suggested the localization of PSORS1 to a small region telomeric of HLA-C and indicated that an endogenous retroviral dUTPase should be considered a candidate for the PSORS1 gene.
Identification of the PSORS1 susceptibility gene has been difficult due to the existence of extensive LD and multiple biologically plausible candidate genes within the PSORS1 region [34]. Most of the genes within the 300-kb critical region of PSORS1 have been shown to be strongly associated with psoriasis [24],[29],[31],[35],[36]. However, the risk alleles of these genes are inseparable from each other due to strong linkage disequilibrium among them. Single marker and haplotype-based association analyses have therefore not been able to determine the primary susceptibility allele. Functional analysis of the known genes within this critical region has also failed to shed light on this problem, because many of these genes have been shown or suggested to be involved in the development of skin tissue and/or the pathogenesis of psoriasis. To overcome this difficulty, Elder and his colleagues genetically dissected the PSORS1 locus by directly sequencing risk-associated haplotypes and subsequently analyzing recombinant haplotypes that can separate the risk-associated alleles of the several genes [14]. They demonstrated that of the eleven genes within the 300-kb critical region of PSORS1, only HLA-C and CDSN carry nonsynonymous alleles (protein alleles) that are unique to the risk haplotypes of the PSORS1 locus, making them the more likely candidates to be the PSORS1 gene. By analyzing the recombinant haplotypes that carry only either the HLA-C or CDSN risk allele, they further demonstrated that HLA-C is more likely to be the PSORS1 susceptibility gene than the CDSN gene.
In this study, we performed the first fine mapping analysis of the PSORS1 locus in Chinese population. The PSORS1 locus was confirmed in Chinese population by our previous genome-wide scan analysis [37]. Our previous case-control study had also shown association of certain class I and II MHC alleles with psoriasis [38]. However, no fine mapping analyses of PSORS1 have been done in Chinese population. By performing both population- and family-based association analysis of haplotypes as well as an exclusion analysis of recombinant haplotypes, we were able to refine the PSORS1 gene to a small critical region where HLA-C is a strong candidate of the PSORS1 susceptibility gene.
A total of 228 Han Chinese psoriasis families, including 61 families used in our previous genome-wide linkage study, as well as additional 192 sporadic cases and 192 healthy controls were recruited in this study. All samples were recruited from the Dermatology Department at No. 1 Hospital of the Anhui Medical University by using an ascertainment procedure described previously [37]. The structure and clinical characteristics of all the 228 pedigrees are summarized in Table 1 and Supplementary Table 1. All the samples were recruited with informed content. The study was approved by the Ethics Committee of Anhui Medical University and conducted according to the Declaration of Helsinki Principles.
10 microsatellite markers (D6S1660, D6S1691, M6S187, C2_4_5, C2_4_4, C1_3_2, C1_2_6, M6S172, D6S273 and D6S1645) were genotyped in 163 families as well as the 192 sporadic cases and 192 healthy controls. The primer information of the 10 microsatellite markers was obtained from the UniSTS Database (http://www.ncbi.nlm.nih.gov/). Marker order and distances were obtained from the National Center for Biotechnology Information (NCBI) database (the build 36.2). HLA-B antigen alleles were also genotyped in the same 163 pedigrees by using DNA-based methods (PCR-SSP) described elsewhere [39].
HLA-C and CDSN were genotyped in the 163 families and additional 65 affected sib pairs. HLA-C alleles were determined by genotyping seven SNPs within HLA-C using the method developed by Nair et al [14]. The seven SNPs are located in exons 2 and 3 of the HLA-C gene at positions 213, 218, 341, 361, 387, 459, and 540 (NM_002117.4). HLA-Cw6 genotype can be definitely distinguished from all other alleles by testing the observed genotypes for the seven HLA-C SNPs against those produced by all possible combinations of alleles in the IMGT/HLA database [14]. The constructed HLA-C haplotypes were retained if they were the only possible outcome given the genotypes of that person and other family members or if the only other choices involved HLA-C haplotypes known to be very rare in the study population. The CDSN*TTC allele (defined as CDSN*632T-1249T-1256C) was determined by genotyping three SNPs within CDSN. The three polymorphisms within CDSN gene are located at positions 632, 1249, and 1256 of the coding sequence (NM_001264).
Genomic DNA was extracted from peripheral blood leukocytes by using standard procedures [40]. Microsatellite markers were genotyped by using a procedure described previously [41]. Briefly, PCR amplifications were performed in a thermocycler 9700TM (Perkin Elmer, Foster City, CA) and PCR products were pooled together and analyzed on an automated ABI Prism 3730 DNA sequencer (Applied Biosystems, Foster City, CA, USA). GeneMapper 4.0 software (Applied Biosystems) was used for determining the fragment sizes of alleles that were further reviewed by two persons independently. All SNPs were genotyped by using the Snap-Shot method of the Applied Biosystems. Pedstats [42] was used to check for non-Mendelian inheritance of alleles and Hardy–Weinberg equilibrium in controls. Unlikely genotypes were further identified by using Merlin (version 1.0.1) [43] and corrected by reviewing the raw genotyping results or re-genotyping if necessary.
9-marker haplotypes were constructed in the 163 families by using the genotypes of 7 microsatellite markers as well as the bi-allelic genotypes of HLA-Cw6 and the quadric-allelic (TTT, TTC, TGC and CTC) genotypes of CDSN locus. The haplotype construction was done by using Merlin. Phase ambiguities in the most-likely Merlin haplotypes were then resolved by PHASE (version 2.1.1) whenever the confidence of the phase call was at least 99%. The 9-marker founder haplotypes from the 163 families were then clustered by using an average-distance agglomerative hierarchical method [18] and SPSS 11.5. For haplotypes to be clustered, more than 90% alleles of the haplotypes were required to be typed and of known phase. The criteria for assigning haplotypes to a cluster are ≥80% homogeneity of marker alleles and a minimum number of five founders. Rare haplotypes that are different from a haplotype cluster by only a single repeat (addition or deletion) were merged into the haplotype cluster. The haplotypes that can not be clustered using the above criteria were lumped into a single cluster. The genotypes for HLA-B antigen alleles were assigned to the haplotype clusters by the inspection of informative pedigrees.
Single-marker case-control association analysis was performed using the two-tailed chi-squared statistic to compare allele frequency differences. Correction for multiple testing was done by calculating the false discovery rate (FDR) using Benjamini and Hochberg's method [44]. The odds ratio (OR) was calculated by 2×2 contingency tables.
A family-based transmission/disequilibrium test (TDT) was performed by using either single locus genotypes or the haplotype clusters. For all the pedigrees with at least one heterozygous parent where allele or haplotype transmission could be determined, a single affected child was chosen randomly for performing the TDT test. Inferred genotypes and haplotypes of founder individuals without DNA samples were not used in the TDT test due to the potential bias [45].
Family-based conditional association analysis was performed by using the WHAP program version 2.09 [46]. Single trio was randomly selected from each of the 163 families. The 7 microsatellites were treated as multi-allelic locus by using the –usat command. We first evaluated the association of every marker with disease phenotype after controlling for the genotypes of the surrounding loci and then the omnibus association evidence within the region after dropping the HLA-Cw6 or CDSN*TTC risk allele from the null model.
The haplotypes of the HLA-Cw6 and CDSN*TTC alleles were constructed in the 228 families by using Merlin and the biallelic genotypes (+/+, +/− and −/−) of the two loci. The recombinant haplotypes that carry either the HLA-Cw6 or CDSN*TTC risk allele were identified and tested for association with psoriasis in the same 228 families. Subsequently, the recombination breakpoints of the recombinant haplotypes were mapped in the 163 families where 10 microsatellite markers (in addition to the Cw6 and CDSN*TTC alleles) were genotyped. The recombination breakpoints were mapped to the last marker on both sides of the CDSN locus that bears a risk allele, to ensure that these recombinant haplotypes fully retained the portion of the candidate interval being tested for exclusion. TDT test was subsequently performed on the recombinant haplotypes sharing the same breakpoints to assess association between the different portions of the PSORS1 candidate region and psoriasis. Due to the low frequency of the HLA-Cw6+/CDSN*TTC− recombinant haplotypes, only the HLA-Cw6−/CDSN*TTC+ recombinant haplotypes were used in the analysis of identifying the maximum portion of the HLA-Cw6+/CDSN*TTC+ risk haplotype that is retained by the non-risk HLA-Cw6−/CDSN*TTC+ recombinant haplotypes.
To evaluate the contribution of the HLA-Cw6 allele to the linkage evidence observed at the PSORS1 locus, we performed the Linkage and Association Modeling in Pedigrees (LAMP) analysis [47] in the 163 families using 7 microsatellite markers surrounding the maximum LOD score peak identified in our previous genome-wide analysis as well as the HLA-C and CDSN loci. LAMP linkage test, direct association test and indirect association test have been performed to get the linkage signal and to assess whether there are other variants that can explain the linkage signal.
To follow up the linkage evidence at the HLA region (PSORS1) identified by our previous genome-wide analysis [37], we performed a family-based association analysis by genotyping 10 microsatellite markers from the PSORS1 locus in 163 Chinese pedigrees (including the 61 families used in our genome-wide linkage analysis). As expected, the family-based TDT test revealed strong supporting evidence for our initial linkage finding (Fig 1). Of the 10 microsatellite markers analyzed, 5 markers (M6S187, C2_4_5, C2_4_4, C1_3_2 and M6S172) show strong evidence for association with psoriasis (TDT p<10−3), and one marker (C1_2_6) yields weak evidence for association (p = 0.0437). The results from the family-based TDT test are consistent with the results from the association analysis of the same 10 microsatellite markers in 192 unrelated cases and 192 controls (Supplementary Table 2), except at D6S273 locus where the association with psoriasis is suggested by the case-control analysis but not supported by the family-based analysis.
Association between the HLA-Cw6 and CDSN*TTC+ alleles and psoriasis was also investigated by TDT test in the same 163 families. The HLA-Cw6 allele was determined by genotyping the seven coding SNPs that can uniquely define the HLA-Cw6 allele, and the CDSN*TTC allele was determined by directly genotyping the three relevant SNPs (see Methods and Materials). The seven SNPs of the HLA-C locus yielded 9 haplotypes, and the three SNPs of the CDSN gene yielded 4 haplotypes. The haplotypes of each gene were tested for association with psoriasis by using TDT test, and the result is summarized in Table 2. As expected, only the haplotypes carrying either HLA-Cw6 (HLA-C haplotype 11) or CDSN*TTC (CDSN haplotype 2) allele are positively associated with psoriasis. Both alleles are strongly associated with psoriasis, although HLA-Cw6 allele seems to be more strongly associated with psoriasis than CDSN*TTC allele (85.6% vs. 73.9% transmission; 2.0×10−19 vs. 1.7×10−8 TDT p value) (Table 2).
We then performed a haplotype-based TDT analysis in the 163 families. The haplotypes were constructed using the genotypes of the 7 microsatellite markers showing single-locus association evidence as well as the HLA-B, HLA-Cw6, CDSN*TTC loci. After removing 31 inferred haplotypes, only 699 founder haplotypes from the 163 families were used in the following haplotype clustering and TDT test.
As described in Materials and Methods, the 699 founder haplotypes were grouped into 29 haplotype clusters where the last cluster (cluster 29) holds all the haplotypes that can not be clustered otherwise. All the haplotype clusters, except the cluster 29, were tested for association with psoriasis by using family-based TDT test, and the results are summarized in Table 3. The family-based association analysis of the haplotype clusters was also performed by using FBAT test, and the results are very consistent to the TDT results (data not shown). Of the 28 clusters tested, 4 haplotype clusters (cluster 4, 5, 6, and 7) are associated with risk. The four risk haplotypes share a 369-kb region of homologous marker alleles between HLA-B and M6S187 where the HLA-Cw6 and CDSN*TTC risk alleles as well as the risk-associated alleles of the 5 microsatellite markers are all present on these four risk haplotypes. The four risk haplotypes seem to be derived from two known ancient risk haplotypes (HLA-Cw6-B57 and HLA-Cw6-B13) [36], and the 369-kb interval is the minimum fragment of the ancestral haplotypes retained by all four risk haplotypes. The clusters 24 and 27 also show moderate evidence for protective effect. However, because the moderate evidence becomes insignificant after correction for multiple testing and the two clusters do not show any allele sharing, the evidence for the cluster 24 and 27 is likely to be false positive.
Hence, the haplotype analysis determined a 369-kb critical region for the PSORS1. However, due to the strong linkage disequilibrium within this region, association evidences observed at the HLA-Cw6 and CDSN*TTC alleles as well as other 5 marker alleles are not independent from each other, and the haplotype-based analysis can not discriminate which alleles is the primary cause of association with psoriasis observed within the PSORS1 locus.
To evaluate which of the risk alleles within the 369-kb critical region is likely to be the primary cause of association, we employed the strategy of recombinant haplotype analysis where different risk allele or alleles are evaluated for association separately. First, we evaluated which of the HLA-Cw6 or CDSN*TTC allele is more likely to be the primary PSORS1 risk allele, by analyzing the recombinant haplotypes that have separated the HLA-Cw6 and CDSN*TTC risk alleles in the 163 families as well as additional 65 affected sib pairs. After discarding 101 (10.2%) phase unknown haplotypes, the 889 founder haplotypes of the HLA-Cw6 and CDSN*TTC alleles from the 228 pedigrees were analyzed by TDT test (Table 4). Of the 889 founder haplotypes, HLA-Cw6+/CDSN*TTC+ and HLA-Cw6−/CDSN*TTC− haplotypes are both common with 32.6% and 52.6% frequency, respectively. As expected, the HLA-Cw6+/CDSN*TTC+ haplotype is strongly associated with risk for psoriasis (T∶NT = 129:34, p = 9.9×10−14), whereas the HLA-Cw6−/CDSN*TTC− haplotype is clearly not associated (T∶NT = 50:117, p = 2.2×10−7). The HLA-Cw6−/CDSN*TTC+ recombinant haplotype is also common (12.9%) and is surely not associated with risk for psoriasis (T∶NT = 29:57, p = 0.0025). The HLA-Cw6+/CDSN*TTC− recombinant haplotype is rare (1.8%) so that its association with risk can not be determined (Table 4). This result strongly indicates that the CDSN*TTC allele itself does not confer any risk without the presence of the HLA-Cw6 allele and further suggests that the PSORS1 gene is likely to be located within the HLA-C side of the 369 critical region.
We also performed a conditional association analysis using the WHAP program. A total of 90 trios were selected randomly from the 163 families. As shown in the table 5, only HLA-Cw6 shows significant association evidence (empirical p-value from 1000 permutations = 9.9×10−4) after controlling for the genotypes of the surrounding loci (including the CDSN*TTC allele), whereas the CDSN*TTC allele no long shows significant association evidence after controlling for the genotypes of the surrounding loci (including the HLA-Cw6 allele). Furthermore, the omnibus association test of the region becomes insignificant after controlling the effect of HLA-Cw6 alleles (p = 0.14), whereas the same omnibus association test remains significant after controlling the effect of CDSN*TTC allele (p = 0.03). Being consistent with the results from the recombinant haplotype analysis, the conditional association analysis provides further supporting evidence that HLA-C is more likely than CDSN to be the candidate of the PSORS1 gene or the PSORS1 gene is more likely to be located nearby HLA-C than CDSN within the 369-kb critical region.
To further refine the location of the PSORS1 gene in Chinese population, we then performed an exclusion analysis within the 369-kb critical region (between HLA-B and M6S187), as demonstrated in Nair et al's study [14]. We first mapped the breakpoints of ancestral recombination between the HLA-Cw6 and CDSN*TTC alleles in the 163 families. By analyzing the genotypes of 7 markers (HLA-Cw6, CDSN*TTC and 5 microsatellite markers) within the 369-kb region, 105 recombinant haplotypes between HLA-Cw6 and CDSN*TTC loci were identified (among the 699 founder chromosomes) in the 163 families, including 7 HLA-Cw6+/CDSN*TTC− haplotypes and 98 HLA-Cw6−/CDSN*TTC+ haplotypes (Table 6). Of the observed 105 recombinant haplotypes, 32 (30.5%) of the recombination breakpoints are mapped to the 48-kb region between HLA-C and M6S172, and 67 (63.8%) are mapped to the 92-kb region between M6S172 and C1_3_2. Only 6 (5.7%) recombination breakpoints are mapped to the 7-kb interval between C1_3_2 and CDSN. Hence, the recombination events between HLA-C and CDSN loci appear to be unevenly distributed and are largely limited to the 140-kb interval between HLA-C and C1_3_2 in the Chinese population. We then performed an exclusion analysis within the 369-kb critical region by identifying the maximum portion of the HLA-Cw6+/CDSN*TTC+ risk haplotype that is retained by the non-risk HLA-Cw6−/CDSN*TTC+ recombinant haplotypes. As shown in Table 7, the HLA-Cw6−/CDSN*TTC+ recombinant haplotypes that carry the C1_3_2-C2_4_5 portion of the HLA-Cw6+/CDSN*TTC+ risk haplotype are clearly not associated with risk to psoriasis (T∶NT = 28:47, p = 0.028). It is also true for the HLA-Cw6−/CDSN*TTC+ recombinant haplotypes that carry the C1_2_6-C2_4_5 portion of the HLA-Cw6+/CDSN*TTC+ risk haplotype (T∶NT = 27:45, p = 0.034). For the HLA-Cw6−/CDSN*TTC+ recombinant haplotypes that carry the M6S172-CDSN portion of the HLA-Cw6+/CDSN*TTC+ risk haplotype, our data suggested an under-transmission of the recombinant haplotypes (T∶NT = 11:20, p = 0.11), but the evidence is not statistically significant. Therefore, our results provided significant evidence for supporting the exclusion of the C1_2_6-C2_4_5 interval from the 369-kb critical region of the PSORS1 locus. Our result also provides suggestive evidence to further exclude the M6172-C1_2_6 interval, but statistical evidence for this exclusion is moderate and needs to be confirmed by further study.
Finally, we evaluated the contribution of the HLA-Cw6 risk allele to the linkage evidence observed at the PSORS1 locus. We performed a LAMP analysis in the 163 families by using the HLA-Cw6 and CDSN*TTC alleles as well as the 7 microsatellite markers. As expected, the LAMP analysis revealed highly significant evidences for both linkage (LOD = 5.19, p = 2.6×10−5) and association (LOD = 45.88, p = 7.2×10−48), and both the evidences are mapped to the HLA-Cw6 locus. After controlling for the effect of the HLA-Cw6 allele, the linkage evidence is dramatically reduced, but still significant (LOD = 2.25, p = 0.005). This suggests that the Cw6 allele is a major contributor to the linkage evidence observed at the PSORS1 locus, but it cannot explain the full linkage evidence.
We have done the first fine mapping analysis of the PSORS1 locus in Chinese population. Our single-locus association analysis provides strong supporting evidence for the PSORS1 locus in a Chinese population and reveals a similar association pattern across the PSORS1 region to that demonstrated in various populations. Our haplotype-based association analysis indicates that all the risk haplotypes within the PSORS1 locus are HLA-Cw6 positive and likely derived from two known ancient risk haplotypes (HLA-Cw6-B57 and HLA-Cw6-B13) in the Chinese population, consistent with what has been found by previous studies in other populations. The HLA-Cw7-B58 haplotype, which was found to be positively associated with psoriasis in a Sardinian population [10], was not present in the Chinese population. B13-Cw1 and B13-Cw3 haplotypes are present in our Chinese samples, but the sample size of these two haplotypes was too small to test for association with psoriasis. Allele comparison among all the risk haplotypes has identified a 369-kb critical region for PSORS1 where all the risk-associated alleles of the 5 marker loci as well as HLA-C and CDSN loci are carried by the identified risk haplotypes, reflecting the extensive LD across the PSORS1 region in the Chinese population as reported in other populations. The 369-kb critical region identified in Chinese population is very similar to the 300-kb critical region identified in a western population by similar methods [14].
Within the critical region, HLA-C and CDSN are the two leading candidates for the PSORS1 gene both with strong supporting evidence. HLA-Cw6 allele is the most significant marker for psoriasis risk prediction and has been indicated to be involved in the psoriasis-related immune response [19]. CDSN is the second extensively investigated candidate and codes corneodesmosin involved in keratinocyte cohesion and desquamation [48]. To evaluate which of the two genes is more likely to be the candidate, we performed the association analysis of the recombinant haplotypes that carry only one of the two risk alleles of the two genes. The frequencies of these two risk alleles in Chinese population are 39% and 47%, respectively, which are slightly higher than their frequencies in Caucasian population (21% and 23%, respectively) [14]. The frequency of the recombinant haplotypes of the HLA-Cw6 and CDSN*TTC alleles is also higher in Chinese psoriatic families (14.7%) than in Caucasian psoriatic families (2.7%) [14]. The majority of the recombinant haplotypes are HLA-Cw6−/CDSN*TTC+ haplotypes (12.9%) that are clearly not associated with disease phenotype, suggesting that without the HLA-Cw6 allele, CDSN*TTC does not confer any risk. This result is further supported by our conditional association analysis where only the HLA-Cw6 allele shows independent effect from the genotypes of the surrounding marker loci including the CDSN*TTC allele. Therefore, our results have clearly excluded CDSN as the PSORS1 susceptibility gene, which is consistent with the conclusion from Nair et al's study in Caucasian population [14].
We then attempted to further narrow down the location of the PSORS1 gene within the 369-kb critical region by mapping the recombination breakpoints and performing an exclusion analysis, as illustrated in Nair et al's study [14]. The common region of the recombination breakpoints between HLA-C and CDSN loci in Chinese population (HLA-C to C1_3_2) is only partially overlapped with the one observed in Caucasian population (M6S169 to CDSN) where over 90% of recombination breakpoints were mapped [14]. In Caucasian population, most of the recombination breakpoints between HLA-C and CDSN loci are located within the region telomeric to C1_2_6 (M6S111) (89.5%), whereas in Chinese population, most of the recombination breakpoints are located within the region centromic to C1_2_6 (about 86.7%). Our results can exclude the C1_2_6-C2_4_5 interval from the 369-kb critical region of PSORS1 with confidence. For the HLA-Cw6−/CDSN*TTC+ recombinant haplotypes that carry the M6S172-C2_4_5 portion of the HLA-Cw6+/CDSN*TTC+ risk haplotype, our result also indicates an under-transmission of the haplotypes (T∶NT = 11:20, p = 0.11), but the evidence is not statistically significant. Therefore, we can narrow down the PSORS1 locus to the 172-kb region between HLA-B and C1_2_6 and potentially to the interval between HLA-B and M6S172, if the exclusion of the M6S172-C2_4_5 interval can be confirmed by further study in Chinese population. The more centromeric location of the recombination breakpoints in Chinese population (than in Caucasian population) allows us to narrow down the location of the PSORS1 gene into a smaller region than Nair et al's study [14].
Within our conservatively mapped critical region of the PSORS1 between HLA-B and C1_2_6, there are only two known genes HLA-C and HCG27. It has been confirmed by many studies, including our present study, that the HLA-Cw6 allele is a strong susceptibility allele for psoriasis. But, there is limited evidence for HCG27 to be the primary susceptibility gene for psoriasis. Direct sequence analysis of two risk haplotypes and eight non-risk haplotypes by Nair et al [14] in Caucasian population indicated that HCG27 carried no alleles unique to the risk haplotypes. Besides these two known genes, there are a few ESTs as well as a few conserved non-coding sequences within the critical region. However, there is no supporting evidence for them to be functional as either coding or regulatory sequences. Searching in the UCSC Genome Browser (May 2006 version) also failed to identify any evidence of known microRNAs within this critical region. Therefore, our recombination mapping analysis has provided strong supporting evidence for HLA-C to be the primary susceptibility gene of the PSORS1 locus.
HLA-Cw6 is a major contributor to the linkage and association evidence observed at the PSORS1 locus, but may not be the only risk allele within the PSORS1 locus. The importance of the HLA-Cw6 allele is demonstrated by the fact that after controlling for the effect of the HLA-Cw6 allele, we could no longer identify significant association evidence at the PSORS1 locus by the family-based test, and the supporting evidence for linkage to the PSORS1 locus has also been dramatically reduced. However, the HLA-Cw6 allele cannot fully explain all the linkage evidence, because the linkage evidence for the PSORS1 locus is still significant after controlling for the effect of the HLA-Cw6 allele. This result can be explained by two possibilities. The HLA-Cw6 is not the only risk allele, and there are additional risk allele(s) within the region. Alternatively, the HLA-Cw6 is only a marker allele and therefore indirectly associated with disease phenotype through LD with an unidentified causal allele. Although we can detect residual linkage evidence after controlling the effect of the HLA-Cw6 allele, we could not identify independent association evidence from the HLA-Cw6 allele in our family-based association study. This could be because our conditional family-based association analysis does not have sufficient power for detecting the residual association evidence due to the moderate number of informative families for family-based association test. It is very interesting to see that the HLA-Cw6 can only partially explain the observed linkage evidence at the PSORS1 locus, but the result needs to be confirmed by more studies.
Our fine mapping result is only partially consistent with the ones from previous studies. The minimal candidate region of PSORS1 mapped by our study overlaps the regions mapped in an American Caucasian population [14] and a Jewish population [11], but is different from the ones mapped by other studies in Caucasian, Japanese and Sardinian psoriatic populations [9],[10],[12] (Fig 2). This difference could be due to the existence of different risk haplotypes in other populations. For example, The HLA-Cw7-B58 recombinant was found to be positively associated with psoriasis in Sardinian samples but not present in our Chinese samples. The difference could also be due to the employment of different methodologies for fine mapping analysis. Another explanation is that there might be different susceptibility alleles in different populations. Giving the fact that psoriasis is a chronic inflammatory skin disorder and that many genes within the PSORS1 region are known to be involved in autoimmune response and skin development, it is not implausible that there might be more than one gene within this region to play a role in the pathogenesis of psoriasis. It is therefore possible that different ethnic populations might carry different susceptibility alleles from different genes and that the results from the fine mapping studies done in different populations are not always consistent with each other due to the genetic heterogeneity of the PSORS1 locus. Further studies will be needed to confirm this hypothesis. |
10.1371/journal.pntd.0000369 | The Importance of pH in Regulating the Function of the Fasciola hepatica Cathepsin L1 Cysteine Protease | The helminth parasite Fasciola hepatica secretes cathepsin L cysteine proteases to invade its host, migrate through tissues and digest haemoglobin, its main source of amino acids. Here we investigated the importance of pH in regulating the activity and functions of the major cathepsin L protease FheCL1. The slightly acidic pH of the parasite gut facilitates the auto-catalytic activation of FheCL1 from its inactive proFheCL1 zymogen; this process was ∼40-fold faster at pH 4.5 than at pH 7.0. Active mature FheCL1 is very stable at acidic and neutral conditions (the enzyme retained ∼45% activity when incubated at 37°C and pH 4.5 for 10 days) and displayed a broad pH range for activity peptide substrates and the protein ovalbumin, peaking between pH 5.5 and pH 7.0. This pH profile likely reflects the need for FheCL1 to function both in the parasite gut and in the host tissues. FheCL1, however, could not cleave its natural substrate Hb in the pH range pH 5.5 and pH 7.0; digestion occurred only at pH≤4.5, which coincided with pH-induced dissociation of the Hb tetramer. Our studies indicate that the acidic pH of the parasite relaxes the Hb structure, making it susceptible to proteolysis by FheCL1. This process is enhanced by glutathione (GSH), the main reducing agent contained in red blood cells. Using mass spectrometry, we show that FheCL1 can degrade Hb to small peptides, predominantly of 4–14 residues, but cannot release free amino acids. Therefore, we suggest that Hb degradation is not completed in the gut lumen but that the resulting peptides are absorbed by the gut epithelial cells for further processing by intracellular di- and amino-peptidases to free amino acids that are distributed through the parasite tissue for protein anabolism.
| Fasciola hepatica is a helminth parasite that causes liver fluke disease (fasciolosis) in domestic animals (sheep and cattle) and humans worldwide. Cathepsin L cysteine proteases (FheCL) are secreted by the parasite to invade its host, migrate through tissues and to degrade host haemoglobin (Hb), a major source of nutrient to the parasite. FheCL1 is a very stable protease and active over a broad pH range (3.0–9.0), making it very suitable for functions both inside and outside the parasite. The slightly acidic pH of the parasite gut not only regulates the autocatalytic activation of the proFheCL1 zymogen to an active FheCL1 protease but also induces relaxation of the Hb structure, making it more susceptible to proteolysis. The action of FheCL1, which is enhanced by glutathione (GSH), the major reducing agent found in red blood cells, degrades Hb to small peptides (predominantly 4–14 residues) that can be absorbed by the gut epithelial cells. Further processing within these cells by exopeptidases provides the necessary amino acids required for protein anabolism by the parasite.
| Fasciolosis is a disease caused by helminths of the genus Fasciola. F. hepatica is found in temperate climates whereas F. gigantica is predominant in tropical regions. However, the distribution of the two species overlap in the Asia-Pacific region where hybrid forms have been isolated [1],[2]. Fasciolosis is of major global economic importance as it infects primary production livestock of humans, especially sheep, cattle and water buffalo. Moreover, epidemiological surveys carried out over the last 15 years have uncovered fasciolosis as a significant human zoonosis. To-date, high prevalence of human infection has been reported in South America (Ecuador, Peru and Bolivia), Vietnam, Thailand, Egypt and Iran [2],[3]. Animals and humans become infected by ingesting vegetation contaminated with infective larvae which emerge from cysts and migrate through the intestinal wall and liver tissue causing extensive tissue damage and haemorrhaging as they burrow and feed. The parasites then enter the bile ducts where they mature and produce eggs that are carried into the environment with the bile juices [1].
The success of F. hepatica as a parasite is related its ability to infect and complete its lifecycle in wide range of mammalian hosts. Besides domestic ruminants and humans these include a large number of relevant reservoir hosts, such as deer, rabbits, hares, rats and mice [2]. Over the last few centuries European colonisation accelerated the distribution of the disease by introducing infected animals into many countries [2],[3]. Most remarkably, within this relatively brief time period the parasite has adapted to local host species such as camelids in Africa, llamas and alpaca in South America and kangaroo in Australia [2].
Fasciola parasites infect and survive in their hosts by secreting cathepsin cysteine proteases. RNAi-mediated knock-down of cysteine protease activity of infective larvae was shown to prevent their ability to migrate through the host intestinal wall [4]. Also, blocking the function of these enzymes using anti-cysteine protease inhibitors or by vaccination with purified enzymes protects animals from infection [5],[6]. The primary function of the F. hepatica cathepsin L proteases is in the digestion of host haemglobin (Hb), the main source of nutrient for the parasite. This takes place within the lumen of the parasite gut, which is believed to be slightly acidic at around pH 5.5 [7],[8]. Adult parasites draw blood with a muscular pharynx through punctures they make in the wall of the bile duct and use it to supply the amino acids needed for the massive production of eggs [7]. Due to the blind-ended nature of the adult parasite gut it must be emptied regularly by regurgitation (approximately every three hours) and refilled with fresh blood [9]. This process is also important for the extrusion of the cathepsin L proteases into the host tissues where they are involved in additional pivotal functions to parasitism including penetration of the host's tissues, cleavage of host immunoglobulins and suppression of immune cell proliferation [3],[6]. These extracorporeal functions of the parasite enzymes are performed, therefore, in an environment of neutral physiological pH.
The functions of proteases are not only related to their physico-biochemical properties but also to their cellular/tissue location and physiological environment (particularly pH). Since F. hepatica cathepsin L proteases are required to function both inside and outside the parasite we considered it important to investigate the regulatory influence of pH on the autocatalytic processing and activation of the inactive zymogen, and on the structural stability and hydrolytic activity of the major secreted enzyme cathepsin L1 (FheCL1). We found that this enzyme was most rapidly activated at low pH but, once activated, was stable and functional over a broad pH range with an optimal hydrolytic activity at pH 6.2. While FheCL1 readily cleaved peptide and protein (ovalbumin) substrates at neutral pH, Hb was resistant to cleavage at this pH. The degradation of Hb required acid-induced structural changes that made it susceptible to FheCL1 cleavage. Degradation was enhanced by the presence of small thiol agents, such as glutathione and cysteine, which activate FheCL1 and are present in physiologically-relevant concentrations in red blood cells and plasma. Our experiments suggest that Hb is digested by the parasite in a microenvironment, likely between the lamellae of the gut epithelial cells, at a pH of approximately 4.5. Under low pH and reducing conditions FheCL1 is capable of generating small peptides but not free amino acids. We propose that these peptides are absorbed by the gut epithelial cells of the parasite where further processing takes place by intracellular dipeptidases [10] and aminopeptidases [11] to release amino acids that are distributed to the parasites tissues and used for protein anabolism.
Z-Phe-Arg-NHMec was obtained from Bachem (St. Helens, UK). E-64, DTT, l-cysteine, GSH (reduced glutathione), EDTA and ovalbumin were obtained from Sigma-Aldrich (Sydney, Australia). Prestained molecular mass markers were obtained from Invitrogen (Victoria, Australia).
Expression, production and purification of recombinant wild-type proFheCL1 and variant proFheCL1Gly25 (procathepsin L) in the yeast Pichia pastoris have been described elsewhere [12],[13]. The variant proFheCL1Gly25 is an inactive zymogen since the active site Cys was replaced by a Gly. Auto-activation of the active wildtype proFheCL1 was carried out by incubating 0.2 mg/ml enzyme at 37°C in 100 mM sodium acetate buffer, pH 4.5, containing 1 mM DTT and 1 mM EDTA. Aliquots (15 µl) were removed at time intervals and added to tubes containing 1 µl of 1 mM E-64 to stop the reaction. Proteolytic cleavage of the prosegment was visualised by 15% SDS-PAGE.
Auto-activation was also monitored in the presence of the fluorogenic substrate Z-Phe-ArgNHMec by measuring the release of fluorescence over time using a KC4 Bio-Tek micro-plate reader in 96-well fluorescent plates. proFheCL1 (5 nM) was incubated in 100 mM buffer pH 4.0–pH 7.0 in the presence of 2 µM Z-Phe-Arg-NHMec. Final linear rates of substrate hydrolysis at each pH were measured with FheCL1 after auto-catalysis was completed.
Fluorescence assays measuring activity of mature FheCL1 was carried out in 96-well plates using a KC4 Synergy HT micro-plate reader (Bio-Tek Instutments Inc., Vermont, USA). Assays were carried out with a final substrate concentration of 0.5 µM in a volume of 200 µl. When [S]<KM the initial rate is proportional to kcat/Km. Assays contained 0.14 nM cathepsin L1 in the following buffers: 100 mM formate (pH 3.24–4.0), 100 mM sodium acetate (pH 4.0–5.5), 100 mM sodium phosphate (pH 5.5–8.0), and 100 mM sodium borate (pH 8.0–10.0). The assay also contained final concentrations of 1 mM DTT and 1.0 mM EDTA. The data were fitted to the equation:
Stability of FheCL1 was investigated by incubating 0.1 mg/ml enzyme in 100 mM buffer (pH 2.5–pH 9.0) at 37°C. Enzyme activity towards 5 µM Z-Phe-Arg-NHMec in 100 mM sodium acetate buffer, pH 5.5 and containing 1 mM DTT was measured at time intervals over a 10-day period.
Human red blood cells were washed three times by resuspending 0.25 ml of whole blood in 5 ml PBS and centrifugation at 5000 rpm. The supernatant with the buffy coat was removed each time. After the final wash, the cells were lysed to release haemoglobin (Hb) by adding 1 ml ice-cold distilled H2O for 10 min and then the suspension was centrifuged at 15000 rpm to remove insoluble material [14]. To remove any free amino acids or low molecular mass material Hb was dialysed twice against 1.5 L phosphate-buffered saline (PBS), pH 7.3, for 3 h using a dialysis membrane with a 3000 Da molecular mass cut-off (Sigma Chemical Co., Sydney, Australia). Hb was quantified using an extinction coefficient of 125 000 M−1 cm−1 at 414 nm [15] and was in good agreement with the total protein in lysates measured by the Lowry method [16] using BSA as standard.
Spectrophotometry was carried out in in 96-well plates in a KC4 Synergy HT micro-plate reader. Hb was diluted to a final concentration of 5 µM into 100 mM buffer and denaturation was recorded for one hour by monitoring the decrease in absorbance at 414 nm [15]. Buffers used were 100 mM sodium acetate buffer, pH 3.5–pH 5.5 and 100 mM sodium phosphate buffer, pH 6.0–pH 7.0. Absorption spectra were recorded after one hour for each sample from 600 nm–300 nm. Hb denaturation was also monitored for 90 minutes at pH 4.5 and pH 7.0 in the presence of 1 mM GSH, with or without FheCL1.
Stock FheproCL1Gly25 in PBS was dialysed into 50 mM sodium phosphate buffer, pH 7.5 or 50 mM sodium acetate buffer, pH 4.0, to remove any NaCl that could interfere with the CD spectrum. CD spectra of 5.3 µM proFheCL1Gly25 (∼0.2 mg/ml) were recorded over the wavelength range 195–250 nm, in steps of 0.5 nm and speed of 20 nm/min using a Jasco720 spectropolarimeter in quartz cuvettes with a 0.1 cm pathlength. Spectra were the average of three scans and were buffer baseline corrected.
Hb (1.8 nmoles) and ovalbumin (1.2 nmoles) were incubated with FheCL1 (0.18 nmoles) in 0.1 M buffers, pH 3.5–8.0 and containing 1 mM DTT. Control experiments contained no enzyme. The buffers used were 100 mM sodium acetate (pH 3.5–5.5) and 100 mM sodium phosphate (pH 5.5–8.0). The reactions were stopped after 30 min by adding 1 µl 1 mM E-64 to the tube and aliquots were analysed by 15% SDS-PAGE under reducing conditions. Gels were stained with a 0.1% w/v solution of Coomassie Brilliant Blue R-250 in 40% methanol/10% acetic acid [17].
Hb (1.8 nmoles) was digested with purified recombinant FheCL1 (0.9 nmoles) in 0.1 M sodium acetate buffer (pH 4.0) containing 1 mM GSH and 1 mM EDTA for 0, 10, 20, 30, 45, 60, 75, 90, 120 and 180 minutes at 37°C. 10 µl aliquots of the digests were analysed using NuPage Novex 4–12% Bis-Tris gels (Invitrogen) according to the manufacturer's instructions. Gels were stained with Colloidal Coomassie Blue G250 (Sigma).
Hb digests were spun at 13,000 rpm for 15 min to remove particulates and were concentrated to a final volume of 15 µl using a Concentrator 5301 (Eppendorf). Peptides were analysed by nanoLC-ESI-MS/MS using a Tempo nanoLC system (Applied Biosystems) with a C18 column (Vydac) coupled to a QSTAR Elite QqTOF mass spectrometer running in IDA mode (Applied Biosystems). Peak list files generated by the Protein Pilot v1.0 software (Applied Biosystems) were exported to local MASCOT (Matrix Science) and PEAKs (Bioinformatics Solutions Inc.) search engines for protein database searching. MS/MS data was used to search 3239079 entries in the MSDB (20060809) database using MASCOT whereas PEAKs software was used to search a custom-made database containing only human Hb-alpha and Hb-beta sequences. The peptide mass tolerance was set at 0.1 Da, oxidation of methionine residues was set as a variable protein modification and the “no enzyme” function was selected. For MASCOT searches, matches with a MOWSE score >70 were considered to be significant [18],[19] and matched peptides achieving a score >60% were accepted during PEAKs searches. The matching peptides were then mapped onto the primary amino acid sequences of human Hb-alpha and Hb-beta to identify FheCL1 cleavages sites.
For matrix-assisted laser desorption ionisation time-of-flight mass spectrometry (MALDI-TOF MS) the Hb digest was desalted and concentrated by zip-tip (Millipore Perfect Pure C18) and spotted using 1 µL matrix (α-cyano-4-hydroxycinnamic acid, 4 mg/mL in 70% v/v acetonitrile, 0.06% v/v TFA, 1 mM ammonium citrate) onto a target plate, and allowed to air dry (Australian Proteome Analysis Facility, Macquarie University, Sydney.). The sample was then analysed using a 4700 Proteomics System TOF mass spectrometer (Applied Biosystems, USA) operated in reflectron mode in the mass range of 100 m/z to 400 m/z. Spectra were analysed manually and externally calibrated using ACTH (fragment 18–37), neurotensin, angiotensin I, bradykinin to give a mass accuracy 50 ppm or less.
The zymogen of the F. hepatica cathepsin L1, proFheCL1 (Mr ∼38 kDa) is auto-catalytically processed at pH 4.5 by inter-molecular cleavage and removal of the prosegment to release a fully mature and active enzyme (Mr ∼25 kDa) (Figure 1A). Analysis of the in vitro auto-activation process by 4–20% SDS-PAGE shows that a band corresponding to the processed ∼25 kDa mature enzyme is observed within 5 minutes and that full removal of the prosegment from the zymogen occurs between two and three hours. Peptides representing products of the cleaved prosegment are observed below the 10 kDa molecular size marker (Figure 1A).
The rate of formation of an active mature enzyme from the inactive zymogen (5 nM) was monitored between pH 4.0–7.0 by performing the autocatalytic reaction in the presence of the fluorogenic substrate Z-Phe-Arg-NHMec and calculating the rate of hydrolysis (Figure 1B). The rate of hydrolysis of Z-Phe-Arg-NHMec, and hence the rate of activation from proFheCL1 to FheCL1, was linear over this pH range; however, hydrolysis at pH 4.0 was ∼40-fold greater than at pH 7.0 indicating that autocatalytic activation occurs much more rapidly in an acidic environment (Figure 1B).
The relationship between the activity of the fully processed mature FheCL1 and pH was examined by determining the kcat/Km against Z-Phe-Arg-NHMec at various pH values in the range 2–10. The results show that the enzyme has the capacity to cleave substrates over a wide pH range (pH 3.0–9.0). Maximal activity was observed between pH 5.5–7.0 with a peak at pH 6.2 (Figure 1C).
We recently described the production of a catalytically inactive proFheCL1 zymogen by replacing the active site Cys25 with a Gly [12],[13]. This variant, proFheCL1Gly25, was correctly folded but inactive and, therefore, unable to autocatalytically activate even at low pH. In the present study, we used this proFheCL1Gly25 variant to examine the stability of the zymogen at various pH values by subjecting it to analysis by circular dichroism (CD) in various solutions buffered in the pH range 4.0–7.5 (Figure 2A). No significant difference in the far-UV CD spectra of proFheCL1 was observed showing that no conformational shifts occur in the secondary structure over this pH range. These data indicate that proFheCL1 would remain stable during the auto-catalytic activation process to FheCL1, even at pH 4.0.
To investigate the susceptibility of the mature activated enzyme to pH denaturation, mature FheCL1 was incubated for at various time-periods at 37°C in buffers over the pH range 2.5–9.0 and then assayed for activity towards Z-Phe-Arg-NHMec in the presence of 1 mM DTT (Figure 2B). The enzyme exhibited optimal stability at pH 4.5; even following a 10-day incubation period the enzyme retained ∼45% activity at pH 4.5 and ∼5% activity at pH 3.0 demonstrating that FheCL1 is very stable in a moderately acidic environment. When incubated at pH 2.5, enzyme activity was not completely lost until day three.
The activity of cysteine proteases is enhanced in the presence of small thiol molecules that reduce the active site cysteine. Dithiothreitol (DTT) is typically included in reactions carried out in the laboratory, but since this is not a physiological-relevant thiol we investigated whether reduced glutathione (GSH) and cysteine could activate the mature FheCL1 at concentrations found in blood (GSH is found predominantly in red blood cells at concentrations of approximately 1.2 mM, while cysteine is found in plasma at 0.23 mM, [20]). To do this FheCL1 was incubated for 5 minutes at pH 4.5 in a range of concentrations of dithiothreitol (DTT), GSH and L-cysteine. Substrate (Z-Phe-Arg-NHMec) was then added and endopeptidase activity determined by monitoring release of -NHMec with time. We found that in the presence of DTT, GSH and L-cysteine FheCL1 exhibited similar activation curves with maximal enzyme activity observed in the presence of each reducing agent at a concentration of ∼0.1 to 1.0 mM (Figure 3).
Since Hb is a major physiological substrate for FheCL1 we examined the pH dependence for its hydrolysis by FheCL1 and compared this to the hydrolysis of ovalbumin. Firstly Hb was incubated alone in solutions buffered at various pHs in the range 3.5 to 8.0 for one hour and then analysed by SDS-PAGE. We observed that in the pH range 5.0–8.0 the molecule migrated as a major band at ∼15 kDa representing the Hb-alpha and Hb-beta monomers and a minor band at ∼30 kDa representing the alpha-beta dimers. However, at lower pH values the intensity of the band at ∼30 kDa increased and new bands ≥50 kDa were observed most likely due to aggregation of Hb (with incubation times of greater than one hour precipitation of the Hb was observed in the acidic pH solutions) (Figure 4A).
When Hb (50 µg) was incubated in the presence of FheCL1 (1 µg) no difference was observed in the migration pattern within the pH range 5.0 to 8.0 (compare Figure 4B with 4A). At pH 4.5, however, addition of FheCL1 caused the bands at ∼15 kDa and ∼30 kDa to disappear and smearing in the respective lanes indicated the presence of low molecular mass products due to Hb digestion (Figure 4A and B). These Hb bands underwent greater degradation by FheCL1 in reactions carried out pH 4.0 and 3.5 (Figure 4A and B).
The above results indicate that FheCL1 cannot digest Hb at pH≥5, whereas digestion is efficient in acidic conditions of pH≤4.5. The lack of digestion at pH≥5 is not due to the inability of FheCL1 to function in this pH range as the studies above showed that the enzyme could cleave peptide substrates optimally between pH 5.5 and 7.0. To support this observation we analysed the digestion of the protein ovalbumin by FheCL1 over the pH range 3.5 to 8.0. Ovalbumin incubated in various buffered solutions at pH values from 3.5 to 8.0 migrates in SDS-PAGE as a single band at ∼45 kDa (Figure 4C). When ovalbumin was incubated with FheCL1 a series of digestive products (<45 kDa) were produced (Figure 4D). SDS-PAGE clearly shows that degradation was optimal between pH 5.0 and 7.0 (Figure 4D), which is in agreement with the optimal activity of the enzyme determined against the peptide substrate (Figure 1C).
To investigate the effect of pH on the structure of Hb we obtained absorption spectra of the molecule in various buffered solutions. The absorption spectrum of Hb at physiological pH is characterised by a large Soret peak 414 nm due to the bound heme moiety; disruption of the Hb conformation causes shifts in this peak (Figure 5A). No alteration in the Soret peak was observed between pH 7.0 and pH 5.5, but the height of the peak began to decrease at pH 4.5. When Hb was exposed to pH 4.0 and pH 3.5 the Soret peak completely disappeared (Figure 5A) indicating that structural changes are occurring in the Hb molecule such that it can no longer bind the heme moiety [15].
The denaturation of Hb at low pH was shown to be a time-dependent process (Figure 5B). Progress curves obtained by monitoring the decrease in absorbance at 414 nm clearly show that Hb is stable at pH 7.0 and 5.5 but that partial denaturation occurs at pH 4.5. Hb denaturation was complete at pH 4.0 and 3.5 within one hour.
To determine if the rate of Hb denaturation made it more susceptible to FheCL1 degradation we mixed Hb at pH 7.0 and pH 4.5 in the presence and absence of 5 µM FheCL1 and 1 mM GSH and monitored denaturation at 414 nm for 1 hour (Figure 4C). The results show that while FheCL1 had no effect on Hb denaturation at pH 7.0, the rate of Hb denaturation/digestion was significantly increased at pH 4.5 in the presence of the protease. Thus the Hb molecule at physiological pH is resistant to proteolysis by FheCL1 but at pH 4.5 alterations in its structure take place that make it susceptible to hydrolysis, which is consistent with our SDS-PAGE analysis described above (Figure 4). Finally, the reducing agent GSH alone, at a concentration of 1 mM, had no effect on Hb denaturation at pH 4.5 or pH 7.0 (Figure 5C).
To examine the process of Hb degradation by FheCL1 Hb was mixed with the protease at pH 4.0 for 120 minutes at 37°C. Reactions were stopped at several time points by addition of E-64 (an irreversible inhibitor of cysteine proteases) and the degradation products were analysed by SDS-PAGE (Figure 6A). The bands representing the 15 kDa Hb monomers and 30 kDa Hb dimers were gradually degraded to smaller protein bands in the molecular size region of 3–10 kDa within the first 10–20 minutes of the reaction and completely degraded between 60 and 120 min. It is noteworthy that during this digestive process the FheCL1 (∼25 kDa) was not degraded (Figure 6A) supporting our earlier data showing that the enzyme is very stable under acid conditions (Figure 2).
To identify the cleavage sites for FheCL1 within Hb, the 10 min and 120 min reaction aliquots were analysed by mass spectrometry. The peptides were then mapped onto the primary amino acid sequences of human Hb-alpha and Hb-beta to identify FheCL1 cleavage sites. Within 10 mins FheCL1 cleaved Hb-alpha at 47 sites and Hb-beta at 52 sites while at 120 min additional cleavage sites, totalling 83 sites in Hb-alpha and 89 sites in Hb-beta were observed (Figure 6B). Examination of the cleavage map presented in Figure 6B shows that within a 10 min time-frame FheCL1 could generate small peptides of 4–8 amino acids from Hb. The map also indicates that these would conceivably be further degraded to release dipeptides and free amino acids after 120 mins. The 120 min digests were, therefore, analysed by LC-MS/MS to determine the masses and sequence identities of the resulting hydrolytic products. This analysis revealed that FheCL1 had degraded Hb into peptides ranging from 3–26 amino acids in length (Figure 7) but not dipeptides or free amino acids. The average length of the released peptides (from both the Hb alpha and beta chains) was 10 amino acids with 13- and 12-residue peptides occurring most frequently in the digested Hb alpha and beta chains, respectively. Accordingly, FheCL1 must not cleave all Hb molecules in the same manner and, thus, the cleavage map shown in Figure 6B represents a composite of cleavage sites.
To verify that free amino acids and/or small peptides (di- or tri-peptides) were not end-products of the proteolysis the digests were also analysed by MALDI-TOF MS (specifically using the mass range 100 m/z to 400 m/z). Only 12 mass ions were detected within this range the masses of which five could be mapped to di- or tri-peptides present in either the Hb-alpha or -beta chains. Importantly, ion masses corresponding to free amino acids were not observed.
Residues present at the P2 position from the scissile bond interact with the S2 subsite of the active site of papain-like cysteine proteases and determine the efficiency by which the bond is cleaved [21]. Therefore, we examined the frequency of each amino acid in the P2 site of the proteolytic cleavage site identified in aliquots of the 10 min Hb digest described above (Figure 8). Consistent with our previously published studies using fluorogenic peptide substrates and positional-scanning of synthetic combinatorial libraries [13] FheCL1 preferentially cleaved bonds where the P2 position was occupied with hydrophobic residues; this preference followed the order Leu>Val>Ala>Phe, and was observed for the digestion of both Hb-alpha and Hb-beta (Figure 8). Due to the promiscuity of the FheCL1 for peptide bonds no obvious trend for P2 preference could be discerned in digests taken at 75–120 minutes (data not shown). Finally, in support of other studies using synthetic combinatorial libraries [13] the P1 position could be occupied by many amino acids but most preferentially Leu.
Proteomic analysis of proteins secreted from adult F. hepatica parasites in situ within the bile ducts [22] and in culture [22],[23] showed that >80% of the secreted proteins are cathepsin L cysteine proteases. Furthermore, no other class of endopeptidase or exopeptidase was identified in these secretions demonstrating the exclusive reliance of the mature parasites on cathepsin Ls [22],[23]. The cathepsin L proteases are synthesised within epithelial cells lining the parasite gut; these cells have both a secretory and absorptive function and spread extended lamellae into the gut lumen [24]. We have shown that cathepsin L zymogens are concentrated and stored in numerous secretory vesicles that lie at the apex or luminal side of these cells ready for secretion into the gut [7],[24]. By the time the enzymes are secreted outside the parasite they have undergone complete processing to mature enzymes by removal of the prosegment portion, which informed us that the activation process takes place within the gut lumen [12]. The gut lumen of F. hepatica, like that of other trematodes such as the schistosomes, is believed to be maintained at a slightly acidic pH, approximately 5.5 [8],[24],[25]. In the present study we have demonstrated using recombinant pro-cathepsin L that auto-catalytic processing and activation can take place at neutral pH but that this occurs far more rapidly at lower pH values (activation at pH 4.0 was 40-times faster than at pH 7.0). Circular dichroism studies showed that the zymogen does not undergo any significant conformational alteration in the pH range 4.0 to 7.0, and like the lysosomal cathepsin Ls of mammals is stable under acid conditions [26]. Enzymatic studies demonstrated that the mature activated enzyme is also very stable under the pH conditions it would experience in the parasite gut. We can conclude, therefore, that the slightly acidic conditions of the parasite gut are very suitable for the autocatalytic activation and digestive function of the cathepsin L proteases.
The primary function of the cathepsin Ls in the parasite gut is to digest host macromolecules and tissues to usable products. Haemoglobin (Hb) is the principle source of amino acids for protein anabolism by the parasite and our present studies demonstrate that FheCL1 can efficiently degrade this substrate in an acidic environment, pH≤4.5. Surprisingly, however, the cathepsin L protease could not cleave Hb at pHs≥5.0, despite the fact that it has optimal activity towards small-peptide and protein (ovalbumin) substrates between pH 5.5 and pH 7.0. These observations revealed the importance of low pH in regulating the structure of Hb and its susceptibility to proteolysis. By monitoring the Soret peak of Hb over a range of pH values we examined the conformational changes that are induced in the molecule. The Hb molecule retained it structure and bound heme in the pH range 5.5 to 7.0 but partial loss of heme-binding was observed when the pH was reduced to 4.5. Hb underwent full denaturation after 1 hour at pH 3.0–4.0, which in solution could be observed by precipitation of Hb in the reaction tubes. Thus, the susceptibility of Hb to FheCL1 proteolysis, as revealed by our SDS-PAGE analysis of digestion reactions, correlated with the pH whereby Hb becomes denatured.
A recent study of acid-induced unfolding of Hb monitored by ESI-mass spectrometry proposed the following model for Hb denaturation:where subscripts “h” and “a” refer to holo- and apo-forms (i.e. heme and non-heme forms of Hb, respectively) [27]. This model indicates that the release of heme from the Hb molecule accompanies the separation and unfolding of the α and β subunits. The final steps in the denaturation scheme, from αβ dimers to heme-bound monomers and then to unfolded non-heme-binding monomers occurred at ∼pH 4.4 and ∼pH 4.0, respectively. In our present study we showed that the addition of FheCL1 to Hb increased the rate by which Hb lost bound heme at pH 4.5 and confirmed that partial denaturation of Hb at this pH was sufficient to relax the structure of the molecule and make it susceptible to proteolysis. Our results are consistent with a much earlier study by Kimura et al. [28] who showed that pH-induced denaturation of Hb increased its susceptibility to trypsin digestion. In conclusion, our studies underscore the importance of the low pH of the parasite gut lumen for denaturing ingested Hb to facilitate its proteolytic hydrolysis. This process is not unlike the denaturation of proteins for hydrolysis in the acid human stomach.
Determining the precise pH of the gut lumen presents a practical hurdle. As mentioned above the pH of the gut lumen in F. hepatica has been suggested to be ∼pH 5.5 [8],[25] while that of the related trematode Schistosoma mansoni has been estimated to be pH 5.0–6.0 by Senft [29], pH 6.0–6.4 by Chappell and Dresden [30] and 6.84 by Sajid et al. [31]. These were not direct measurements of the intraluminal pH but were generally obtained by measuring media into which parasites had extruded their gut contents. Our data showing that FheCL1 could not digest Hb at pHs≥5.0 is biochemical evidence suggesting that the site of proteolytic activity within the gut must be lower than pH 5.0. Electron micrographs of the gut lumen of both F. hepatica [8],[25] and S. mansoni [32] often visualise Hb as a dense precipitate, representing presumably denatured protein, in the vicinity of the gut lamellae. Halton's [8],[25] interpretation of micrographs of the gut structure was that digestion in F. hepatica takes place between the lamellae of the secretory epithelial cells. Derived from these studies it was suggested that the pH in this local microenvironment is maintained at a more acidic pH than the gut lumen per se [24]. In support of this suggestion Delcroix et al. [33] found sequestered compartments between lamellae of the schistosome gut with pH as low as 3.9. Their observation explains why the schistosome gut aspartic protease, SmCD, whose activity is confined to the range of pH 2.5–4.6 [34] could participate in Hb digestion. Similarly it explains the role of the schistosome cathepsin L cysteine protease, SmCL1, which could efficiently cleave Hb only in the pH range 4.0–4.5 [14].
FheCL1 and other papain-like cysteine proteases are activated in the presence of low molecular mass thiols such as cysteine or DTT [35],[36]. Although these compounds are routinely used to activate cysteine proteases as part of in vitro activity assays, they are not considered physiologically relevant reducing agents. GSH is the most abundant intracellular reducing agent and its concentration inside red blood cells is particularly high, estimated to be 1.192 mM by Mills and Lang [20] and ∼3.2 mM by Chappell et al. [37]. Here we found that GSH effectively enhances FheCL1 activity towards small synthetic substrates with an optimum at ∼0.1 to 1.0 mM GSH, and accelerates the digestion of Hb by FheCL1. A concentration of ∼0.1 to 1.0 mM GSH could conceivably be reached in the parasite gut following lysis of ingested red blood cells notwithstanding variations in the size of the blood meal and dilution in the parasite gut.
We used mass spectrometry to identify the cleavage sites of FheCL1 within Hb and to determine the size of the peptide products generated by its complete digestion. FheCL1 digested Hb at 83 cleavage sites in Hb-alpha and 89 sites in Hb-beta that resulted in short peptides of at least 4–14 amino acids, with some appearance of tripeptides. Residues in the P2 position are known to influence the efficiency of all papain-like cysteine proteases, and we found that those residues in Hb that were most susceptible to cleavage by FheCL1 were invariably a hydrophobic residue, and in the order Leu>Val>Ala>Phe. These results are consistent with our earlier studies using fluorogenic peptides and peptide libraries that showed FheCL1 to have a more restricted S2 active site compared to human cathepsin L and most readily accommodates hydrophobic P2 residues, particularly Leu [13]. It is pertinent to note that the amino acids Leu, Val, Ala, Phe make up approximately 42% of the Hb molecule and, therefore, we would propose that FheCL1 has been specificity adapted to degrade this substrate. However, our studies also show that cleavage by FheCL1 does not generate free amino acids and, by extension, suggests that Hb degradation is not completed within the parasite gut but that small peptides are taken up by the gut epithelial cells during their absorptive phase for further processing within cells [24]. The enzymes involved in this process likely include a dipeptidylpeptidase [10] and an aminopeptidase [11] that function at neutral pH and have been located by immunofluorescence microscopy within the cytosol of the epithelial cells.
Our studies on F. hepatica point to a digestive machinery that requires proteases of only one mechanistic class i.e. cathepsin cysteine proteases. However, F. hepatica secretes different forms of these proteases with overlapping specificity that may complement each other [23]. Nevertheless, the mechanism of gut digestion appears to differ markedly from other helminths so far studied. Dalton et al. [38] were first to propose that schistosomes exploit a cascade involving aspartic and cysteine (cathepsin B, L1 and L2) proteases within their gut lumen to achieve the complete degradation of Hb. The more recent studies by Delcroix et al. [33], which exploited selective protease inhibitors and RNA interference (RNAi) to explore the mechanism of Hb digestion in schistosomes, supports the role of a network or combination of cysteine proteases, aspartic protease and an asparaginyl endopeptidase. However, Correnti et al. [39] showed that while knockdown of cathepsin B expression in schistosomes by RNAi retarded parasite growth it did not prevent Hb digestion in the parasite gut. On the other hand, Morales et al. [40], also using RNAi, demonstrated that the cathepsin D aspartic protease is essential to survival of schistosomes through its pivotal role in Hb digestion.
A multi-enzyme cascade involving cysteine and aspartic proteases is also necessary for Hb digestion in canine hookworm Ancylostoma caninum [41] and aspartic and cysteine proteases in the nematode Ostertagia ostertagi have also been shown to have activity against Hb [42]. Although several proteases appear to be involved in Hb digestion in these helminths it is still not clear whether digestion is regulated in an ordered manner, each enzyme working sequentially, or whether all proteases work simultaneously and in a random manner. Dalton et al. [7],[24] suggested that the activity of each protease within the gut was regulated by pH, and therefore as the bloodmeal (pH 7.0) was drawn into the gut the pH slowly decreased (perhaps by proton pumps in the epithelial cells), each enzyme would come into play when its appropriate pH range for activity was reached; thus in schistosomes cathepsin B (optimum pH 4.0–6.0) would be activated before cathepsin L (optimum pH 4.0–4.5), which would be followed by aspartic proteases (optimum pH 2.9–4.0).
The cathepsin L proteases of F. hepatica also participate in functions outside the parasite gut; these include liver tissue degradation, cleavage of host antibodies and suppression of host immune cell function (see [6]). The blind-ended gut of the parasite is emptied every 3 hours, thus depositing the cathepsin L proteases outside [9]. The extracorporeal roles of the proteases are performed at pH values that are between two and three pH units higher than the microenvironment at which the proteases function in the parasite gut. Our studies showing that FheCL1 are active and highly stable at neutral pH points to a specific adaptation of these molecules to carry out functions over a wide pH range. It is interesting to note that the pH optimum of the FheCL1, pH 6.2, is approximately mid-point between the pH values at which it works inside and outside the parasite. In contrast, lysosomal cathepsin Ls of mammals are active only at pHs values of approximately 4.5, in keeping with the environment in which they function, and are inherently unstable at neutral pH so that cellular damage due to leakage from the lysosome is avoided [43].
To conclude, the helminth parasite F. hepatica secretes cathepsin L proteases that are specifically adapted to be functional at pHs at which they perform essential roles in this parasite's biology. The low pH of the parasite gut is important in regulating the activity of these proteases by providing a milieu whereby the proteases readily autocatalytically activate from inactive zymogens secreted by the surrounding epithelial cells, and by facilitating the denaturation of the protein substrates on which the proteases act. The mature cathepsin L proteases are extremely stable at this pH and their hydrolytic activity is greatly enhanced by GSH, most likely derived from ingested host red blood cells. FheCL1 is specifically designed to cleave peptide bonds with N-terminal hydrophobic residues which are most common in Hb with the goal to provide small peptides that can be absorbed by the gut epithelial cells for further processing to amino acids within cells before distribution to parasite tissues via amino acid transporters [24]. However, following completion of the digestive process in the gut lumen unwanted material is extruded which delivers the proteases to the outside where they can perform their additional extracorporeal roles at physiological pH conditions in which they are also highly active and stable.
|
10.1371/journal.pntd.0001268 | Chagasic Thymic Atrophy Does Not Affect Negative Selection but Results in the Export of Activated CD4+CD8+ T Cells in Severe Forms of Human Disease | Extrathymic CD4+CD8+ double-positive (DP) T cells are increased in some pathophysiological conditions, including infectious diseases. In the murine model of Chagas disease, it has been shown that the protozoan parasite Trypanosoma cruzi is able to target the thymus and induce alterations of the thymic microenvironment and the lymphoid compartment. In the acute phase, this results in a severe atrophy of the organ and early release of DP cells into the periphery. To date, the effect of the changes promoted by the parasite infection on thymic central tolerance has remained elusive. Herein we show that the intrathymic key elements that are necessary to promote the negative selection of thymocytes undergoing maturation during the thymopoiesis remains functional during the acute chagasic thymic atrophy. Intrathymic expression of the autoimmune regulator factor (Aire) and tissue-restricted antigen (TRA) genes is normal. In addition, the expression of the proapoptotic Bim protein in thymocytes was not changed, revealing that the parasite infection-induced thymus atrophy has no effect on these marker genes necessary to promote clonal deletion of T cells. In a chicken egg ovalbumin (OVA)-specific T-cell receptor (TCR) transgenic system, the administration of OVA peptide into infected mice with thymic atrophy promoted OVA-specific thymocyte apoptosis, further indicating normal negative selection process during the infection. Yet, although the intrathymic checkpoints necessary for thymic negative selection are present in the acute phase of Chagas disease, we found that the DP cells released into the periphery acquire an activated phenotype similar to what is described for activated effector or memory single-positive T cells. Most interestingly, we also demonstrate that increased percentages of peripheral blood subset of DP cells exhibiting an activated HLA-DR+ phenotype are associated with severe cardiac forms of human chronic Chagas disease. These cells may contribute to the immunopathological events seen in the Chagas disease.
| The thymus is a primary lymphoid organ that plays an important role on the development of the immune system and maturation of the T cell repertoire. During the normal life span, this organ undergoes involution during the aging and also in the presence of a wide variety of infectious diseases. It has been shown that the protozoan parasite Trypanosoma cruzi is able to target the thymus and induce alterations of the thymic microenvironment. In the acute phase, this results in a severe atrophy of the organ and early release of immature double-positive (DP) T cells into the periphery. The effect of the changes promoted by the parasite infection on thymic central tolerance has remained not clear. The present study shows that the intrathymic key elements that promote the negative selection of thymocytes during the thymopoiesis remains functional in the acute chagasic thymic atrophy. However, we found that the DP cells released into the periphery acquire an activated phenotype and its high frequency in the peripheral blood are associated with severe cardiac forms of human chronic Chagas disease.
| Chagas disease is caused by the flagellate protozoan Trypanosoma cruzi (T. cruzi) and represents an important public health burden in the Americas. After infection, the initial acute phase of the disease progresses to an asymptomatic indeterminate period with virtually undetectable parasitemia. The patients have a strong humoral and cellular anti-T. cruzi responses, resulting in non-sterile control of the parasite. Up to several years after the initial infection, approximately 20 to 30% of all infected individuals develop a chronic inflammatory disease primarily affecting the heart [1]. The pathogenesis of Chagas disease is controversial and distinct hypotheses have been considered, including autoimmune manifestations and parasite-driven tissue damage [2]. In any case, it is accepted that the events occurring during the acute phase of T. cruzi infection are determinant for the pathological features to be settled later, during the chronic phase of the disease [3].
In experimental models of Chagas disease, several alterations in lymphoid organs were observed, including the thymus where the parasite has been detected [1]. Previous studies have revealed a severe thymic atrophy in acutely infected animals mainly due to apoptotic depletion of CD4+CD8+ double-positive (DP) thymocytes in the cortical area of the thymic lobules [4], [5]. Thymocyte depletion parallels T. cruzi-induced alterations of the thymic microenvironment, comprising phenotypic and functional changes in the thymic epithelial cell (TEC) network, with enhancement in the deposition of cell migration-related molecules such the extracellular matrix (ECM) proteins, fibronectin and laminin, as well as the chemokines CXCL12 and CCL21 [6]–[8]. These changes are likely related to the abnormal release of DP cells from the thymus into the periphery, resulting in more than 15 fold increase in DP cell numbers in subcutaneous lymph nodes. In this vein, it is noteworthy that DP cells seen in peripheral lymphoid organs express high densities of ECM and chemokine receptors [9]. Among these abnormally released DP cells in the periphery, we found lymphocytes expressing potentially autoreactive TCRs, which are normally deleted in the thymus of uninfected mice. This suggests that during the infection, immature T lymphocytes escape from the thymic central tolerance process and migrate to the lymph nodes where they eventually differentiate into mature CD4+ or CD8+ cells [9], [10].
During lymphocyte development, TCR gene assembly occurs by combinatorial linking of gene segments. The nature of this process gives rise to lymphocytes that can recognize self-antigens and have the potential to induce autoimmunity [11]. In the thymus, potentially autoreactive lymphocytes are induced to undergo apoptotic cell death during the negative selection process, which involves the elimination of potentially autoreactive T cell clones bearing high affinity TCR that recognize autoantigens presented by TEC. Intrathymic expression of tissue-restricted antigens (TRAs), activated by the autoimmune regulator (Aire) protein in TECs, has an important role in the induction of negative selection of thymocytes [12]. The Aire protein has several domains that are characteristic to transcriptional activators and has been reported to interact with a common transcriptional regulator, the CREB binding protein (CBP) [13], [14]. In turn, the apoptosis mechanism in thymocytes is dependent on the activation of the pro-apoptotic Bcl-2 family member Bim. The TCR ligation upregulates the BH3-only protein Bim expression and promotes interaction of Bim with Bcl-XL, inhibiting its survival function [15]–[17]. Loss of either Aire or Bim expression results in altered thymic deletion of autoreactive lymphocytes and has been implicated in autoimmune manifestations [18]–[21].
Herein we attempted to clarify whether the changes in the thymic microenvironment, as a result of parasite infection, play a role on the modulation of the negative selection of thymocytes, a key element for the control of autoimmunity. To this end, we analyzed the intrathymic expression of Aire and TRAs as well as the expression of pro-apoptotic Bim protein during the course of the thymic T. cruzi infection. Using the ovalbumin specific DO11.10 TCR transgenic system, we further investigated whether the transgenic thymocytes are hypersensitive to stimulation with the specific OVA peptide in mice undergoing T. cruzi promoted-thymic atrophy. In addition, we analyzed the activation profile of the DP cell subset that is prematurely released to the periphery during the course of the infection. Our study reveals that regardless of thymic changes promoted by the T. cruzi infection, the negative selection is still functional during the acute infection of the parasite. However, we show that in contrast to the physiological condition, the DP cells released into the periphery during the course of the infection acquire an activated phenotype similar to what is described for activated single-positive T cells. Furthermore, we show that the presence activated DP cells in the periphery is correlated with the development of severe clinical form in chronic human Chagas disease.
The study was approved by the Research Ethics Committee of National University of Rosario, (protocol UNR-CD 2854/2008) and Fiocruz (protocol CEUA-LW8/10). Protocols for animal and human studies were approved by the Institutional Ethical Committees in accordance with international guidelines. All animal experimentation was performed in accordance with the terms of the Brazilian guidelines for the animal welfare regulations. All individuals provided written informed consent.
Healthy volunteers and T. cruzi chronic chagasic patients were recruited from Chagas Unit, Hospital Provincial del Centenario de Rosario, UNR, Argentina. Subjects using any medication thought to affect immune functions were excluded from the study. The diagnosis was based on standard serological test, including indirect immunofluorescence and haemaglutination assay. All chronic infected patients and noninfected individuals participated in serological tests to confirm the diagnosis for T. cruzi infection, with ages ranging from 30 to 64 years. The seropositive cases included fifteen cardiac chronic-infected patients presenting dilated cardiomyopathy diagnosed based in a detailed clinical examination, including electrocardiography (ECG) and chest X-ray. Additionally, we included fifteen chagasic patients without any cardiac alterations detected, being diagnosed as in the indeterminate form of the disease. Fifteen sex and age matched-controls were also included.
Male BALB/c mice, aged 4–8 weeks, were obtained from the Oswaldo Cruz Foundation animal facilities. The OVA-TCR transgenic mice (BALB/c background, clone D011.10, that recognize the 323–339 peptide fragment of OVA in the context of I-Ad), were bred under standard conditions. Acute T. cruzi infection was performed by inoculating the mice intraperitoneally with 102 blood-derived trypomastigote forms of the Tulahuén strain. After 2 weeks of acute infection (at the peak of parasitemia), animals were bled and sacrificed, and the organs to be studied were removed. Blood parasites were counted using Neubauer's chambers.
Thymuses from normal and infected animals (after 2 weeks of acute infection) were fixed in neutral buffered formaldehyde or Bouin's solution and processed for paraffin wax sectioning. The 3–5 µm thick paraffin sections were stained with hematoxylin-eosin, impregnated with ammoniacal silver according to Weil–Davenport method or stained with Periodic acid-Schiff for demonstration of metallophilic macrophages, as described [45].
Immunoperoxidase assays were performed using standard techniques on 4% formaldehyde-fixed frozen sections or chamber slides. Endogenous peroxidase was blocked with 3% H2O2. Non-specific protein binding was blocked by incubation with 5% normal swine serum. Sections were sequentially incubated with rabbit anti-mouse Aire polyclonal antibody for 60 min at 37°C, horseradish peroxidase (HRP) labeled swine anti-rabbit polyclonal antibody for 60 min at 37°C, with intervening phosphate buffer solution (PBS) washes. After a complete wash in PBS, the slides were developed in 0.05% freshly prepared 3,3′-diaminobenzedine solution with 0.03% hydrogen peroxide for 5 min, and then counterstained with hematoxylin, dehydrated, air-dried, and permanently mounted. For frozen sections, the optimum dilution of each antibody was first determined by serial dilutions on normal mouse thymus sections. Normal swine serum was used to substitute for the primary antibody to serve as a negative control. The stained cells were observed under a Bio-Rad MRC 1024ES regular microscope equipped with a Zeiss Diaphot inverted microscope (Jena, Germany). Rabbit anti-mouse Aire polyclonal antibody and HRP labeled swine anti-rabbit polyclonal antibody were purchased from Santa Cruz Biotechnology Inc. (California, USA).
Thymocytes were lysed at a concentration of 3×108 cells/ml in 0.2% Nonidet P-40 lysis buffer. Lysates were incubated on ice for 15 min, centrifuged at 16,000 x g for 5 min at 4°C, and pre-cleared with protein A/G beads. Bim was immunoprecipitated from lysates with 3 µg of monoclonal rat anti-Bim clone 3C5 Ab, gently provided by Professor Andreas Strasser (The Walter and Eliza Hall Institute of Medical Research, Australia). Control lysates were incubated with 3 µg of normal rabbit IgG (Santa Cruz Biotechnology). The lysate mixtures with antibodies were gently rocked at 4°C for 1 h. Protein A/G beads (50 µl of a 50% (v/v) slurry) were added to each sample and rocked at 4°C for an additional hour. After incubation, the beads were washed with lysis buffer three times, then washed with PBS and transferred to fresh tubes. The bound proteins were eluted with nonreducing 4x SDS sample buffer by boiling for 5 min and subjected to Western blot analysis.
RNA was isolated using TRIzol (Invitrogen, Life Technologies) and reverse-transcribed to cDNA using the SuperScript TM III Reverse Transcriptase (Invitrogen, Life Technologies). Real-time PCR was performed with the ABI Prism 7900HT Fast Real-Time PCR System instrument (Applied Biosystems) using the qPCR SYBR Green Core Kit (Eurogentec) according to the manufacturer's instructions, except that 2 mM MgCl2 concentration was used. The amplification program included an initial denaturation step at 95°C for 10 min, followed by denaturation at 95°C for 15 s, and annealing and extension at 60°C for 1 min, for 45 cycles. SYBR Green fluorescence was measured after each extension step, and the specificity of amplification was evaluated by melting curve analysis. Primers used to amplify specific gene products from murine cDNA were K5 sense, 5′-agtcaacatctccgtcgtcac-3′; K5 antisense, 5′-gggactgcctaaaagaagcag-3′; Aire 11/12, 5′-ccccgccggggaccaatctc-3′; Aire 12/13, 5′-agtcgtcccctaccttggcaagc-3′; Ins2 sense, 5′-gacccacaagtggcacaac-3′; Ins2 antisense, 5′-tctacaatgccacgcttctg-3′, Mup1 sense, 5′-tctgtgacgtatgatggattcaa-3′; Mup1 antisense, 5′-tctggttctcggccatagag-3′; Spt1 sense, 5′-aacttctggaactgctgattctg-3′; Spt1 antisense, 5′-gaggcctcattagcagtgttg-3′; IFN-γ sense, gaaagcctzgaaagtctgaataact; IFN-γ antisense, atcagcagcgactccttttccgct; GAPDH sense, gccgcctggagaaacctcccaagt; GAPDH antisense, tattcaagagagtagggagggctc. The relative gene expression levels were calculated using the comparative Ct method (according to Applied Biosystems), where Ct represents the threshold cycle. Every sample was run in three parallel reactions.
Heparinized whole blood was collected and diluted 1/1 with PBS before separated by density centrifugation on Ficoll-Hypaque (Sigma) for 30 min at 2000 rpm. All antibodies were purchased from BD Biosciences (CA, USA). For human T cell phenotyping, individual samples contained 1.106 living cells were pre-blocked with human AB serum for 15 min and stained at 4°C for 30 min simultaneously with four colors using the following antibodies for FACS analysis: APC-labeled anti-CD3, APCCy7-labeled anti-CD4, PECy7-labeled anti-CD8 and FICT-labeled anti HLA-DR antibodies (BD/PharMingen). For murine T cell phenotyping experiments, thymuses and subcutaneous lymph nodes (pool of inguinal, axilary and brachial lymph nodes) were removed from infected and control animals. The organs were minced, washed and resuspended in PBS-FCS 5% for subsequent evaluation of cellularity, which was followed by triple or quadruple immunofluorescence staining, as previously described [8]. Cells were then fixed and analyzed by flow cytometry in a FACSCalibur flow cytometer. Analyses were done after recording 25,000–50,000 events for each sample, using a CELLQuest software (Becton Dickinson). Lymphocytes were gated based on forward and side scatter parameters, so as to avoid larger leukocytes such as macrophages and granulocytes.
CD4+CD8+ T cells were obtained from lymphoid tissues of normal and infected animals in the peak of parasitemia (2 weeks post-infection). The total number of CD4+CD8+ T cells was determined by FACS analysis. The purification of the DP subset was obtained by FACS cell sorting using anti CD3-FITC, anti CD4-PERCP, and anti CD8-PE. After FACS cell sorting, the total RNA from DP subset was extracted using standard Trizol extraction. mRNA transcripts for mouse IFN-γ were amplified using specific primers in quantitative RT-PCR as described [46].
The mouse mastocytoma cell line, P-815, was labeled with CFSE based on the manufacturer's instructions (Molecular Probes) and used as target cells. The labeled cells were adjusted to 5×104 cells/mL, and 100 µL/well was plated in 96-well microtiter plates. CD4+CD8+ T cells obtained from lymphoid tissues of infected animals or normal single-positive CD4+ or CD8+ T cells as controls were added at different effector-target ratios. Plates were incubated in a humidified atmosphere of 5% CO2 and 37°C in the presence of anti-CD3 mAb at a concentration of 20 µg/ml for TCR triggered-redirected cytotoxicity activity. After 6 hours, the wells were harvested to allow quantitative analysis of the cell populations by FACS analysis. To stain for dead cells, propidium iodide (1 µg/mL) was added, and samples were mixed properly and directly analyzed by flow cytometry. The absolute number of surviving cells was determined at each time point by calculation of the acquisition of CFSE-lived cells at a fixed period of time. The absolute number of cells at the moment of T- cell addition (t = 0) is the number of target cells added to the wells (5,000). The percentage of survival was calculated as follows: % survival = [absolute no. viable CFSE+ target cells (t = x)]/[absolute no. viable CFSE+ target cells (t = 0)] x 100.
Results were expressed as mean ± standard error, unless otherwise indicated. Parametric and non parametric tests were used, depending on the characteristics of variables (normal distribution or not, etc). Comparisons among groups were made by parametric analysis of variance (k groups >2) followed by Student's t-test (k groups = 2) or non parametric Kruskall-Wallis test (k groups >2) followed by the U Mann-Whitney test (k groups = 2). Differences between control versus infected groups were considered statistically significant when P≤0.05.
The acute infection experimental model of T. cruzi induces a severe thymic atrophy that is evident at 10 days post-infection after the intraperitoneal injection of BALB/c mice with 102 Tulahuén trypomastigote parasites. In this model, the parasitemia progresses with a peak at day 14, and 20 days after infection, when the animals start to succumb. The thymuses of normal BALB/c mice show a well-developed cortex and medulla with a clear cortico-medullary boundary, delicate capsular and septal tissue and medullary blood vessels of usual appearance (Fig. 1, upper left panel). In the acute model of Chagas disease, the thymuses exhibit progressive atrophy of the cortex and medulla with a blurred cortico-medullary boundary region. In addition, rough and broadened capsular and septal connective tissue, and more numerous and dilated medullary blood vessels can be observed during the course of infection (Fig. 1, upper right panel).
As the macrophages represent a parasite reservoir in the thymus [22], we evaluated whether the thymic changes observed during the infection could compromise the viability of macrophages and their localization in the thymic microenvironment. The metallophilic macrophages correspond to a special type subpopulation of mature resident macrophages present with prominent dendrites in the thymus. They are distinguished by a high non specific esterase activity and can be stained with silver impregnation [23], and have an important role in the process of thymocyte maturation, being able to present antigens by both MHC class I and II molecules [23], [24]. Accordingly, they are particularly active in the process of negative selection, which is in keeping with their strategic localization within thymic parenchyma, forming a distinct row of cells at the cortico-medullary junction of the thymic lobules [23].
Interestingly, in contrast to normal thymus (Fig. 1, lower left panel), we found that in the infected thymus, numerous metallophilic macrophages are dispersed not only in the cortico-medullary area but also throughout the cortical compartment, appearing as large cells with abundant cytoplasm (Fig. 1, lower right panel). Nevertheless, in infected thymuses we noted the presence of these cells at the cortico-medullary junction where the clonal deletion is believed to occur. As the metallophilic macrophages are mostly participating in the uptake of dying thymocytes, but not in induction of apoptosis that is mediated by DC or medullary epithelial cells, it suggests that their altered location in cortical areas is not necessarily affecting the thymic clonal deletion.
Previous studies have demonstrated that TRAs are expressed in the thymus and this expression is needed for the deletion of self-reactive T cells. An important molecule in regulation of TRA expression by medullary TEC (mTEC) is Aire [12]. Several studies have shown the Aire protein to be predominantly expressed in TEC and suggest it has a role in regulation of immune tolerance by modulating the TRA expression in the thymus [25]. Therefore, we explored whether the acute atrophy of the thymus induced by the T. cruzi infection could modulate the expression of TRAs and Aire in the thymus. For this purpose, we collected the thymuses at day 15 post-infection when the atrophy is severe and compared the expression of Aire and TRAs in infected and normal thymus by real-time PCR.
To study the expression of self-antigens we chose three TRAs (Ins1, Mup1 and Spt1) that are downregulated in Aire deficient mice and follow the same expression pattern as the Aire gene. All three genes have highly selective, tissue-specific expression. For example, Ins1 encodes for insulin expressed by pancreatic beta cells, Spt1 is expressed in salivary and lacrimal glands, and Mup1 is expressed in the liver and also in salivary, lacrimal and mammary glands [25]. The results were normalized to the expression level of keratin 5 (k5) mRNA, which is specifically expressed in medullary thymic epithelial cells [26].
The real-time PCR analysis showed similar levels of Aire expression in both normal and infected thymus (Fig. 2a). To more accurately assess the protein expression of Aire within the thymus, in situ immunohistochemical staining with anti-Aire polyclonal antibody was performed in thymus sections. Despite the atrophy observed during the acute infection, we did observe Aire+ cells in thymuses at day 15 post-infection as compared to physiological condition (Fig. 2b). This profile was followed by the expression of TRA genes, showing that the expression of ectopic genes was unaffected by T. cruzi infection as compared to the control mice. Collectively these data show that both Aire and TRAs are normally expressed in the thymus of T. cruzi infected mice.
In order to evaluate the checkpoints of the intrathymic thymocyte negative selection, we next assessed the proapoptotic Bim expression levels in DP thymocytes during T. cruzi-induced thymus atrophy. To analyze Bim expression levels, thymuses were collected from normal and infected mice at the indicated time points, and the viable cells were counted by trypan blue exclusion (Fig. 3a), and the FACS sorted DP cells were analyzed for Bim expression by Western blotting with anti-Bim antibodies. The Bim protein expression was readily detectable in thymocytes obtained from uninfected control mice as seen in Figure 3b. From day 10 to 16 post-infection, we found that the expression levels of Bim in infected mice along the course of thymus atrophy were similar to control animals (Fig. 3b-c), thus indicating that there is no major change in expression level of proapoptotic Bim protein.
As the expression of Aire/TRA genes and the pro-apoptotic Bim protein were not changed in the thymus undergoing atrophy upon T. cruzi infection, we next wanted to confirm that the negative selection from infected thymus is functional. To address this question, we took advantage of the OVA-transgenic system in which the administration of the cognate antigen promotes the TCR-induced killing of semi-mature thymocytes in vivo [27]–[29]. We found that injection of OT II TCR transgenic non-infected mice with the ovalbumin 323–339 cognate peptide (three times every 24 h) significantly decreased the number of DP thymocytes at 3 days after the first injection, which is consistent with the results described earlier [28], [29]. Deletion of immature CD4+CD8+ DP cells was assessed by counting total thymocyte numbers and by determining the proportions of cell subsets by flow cytometry staining with antibodies to CD4 and CD8 cell-surface markers.
We then evaluated whether the depletion of DP cells by co-administration of the cognate antigen is affected in the presence of thymus atrophy during T. cruzi infection. Similarly to non-infected BALB/c OT II TCR transgenic mice, we observed a drastic reduction of the intrathymic DP cell population in T. cruzi infected transgenic mice at day 12 post-infection (Fig. 4). Next, we investigated whether there is a significant change in numbers of DP cells caused by the thymic negative selection of thymocytes undergoing differentiation upon infection. As shown in Figure 3, the deletion of DP cells from infected mice was increased to the levels of control mice upon administration of OVA 323–339 peptides. These results suggest that the changes promoted in the thymus by acute infection of T. cruzi do not affect negative selection of DP cells undergoing antigen-driven thymic differentiation.
Since all our findings suggested that negative selection in T. cruzi infected thymus was functionally normal, we next focused on the phenotypic characterization of the DP cells that are abnormally released from the thymus upon T. cruzi infection. In this respect, we determined whether these cells exhibit an activated profile similar to effector and memory single positive T cells. Using multiparameter FACS analysis we assessed the expression of the major cell surface markers that are known to undergo changes after in vivo activation of T cells. In parallel, we phenotyped thymocytes from T. cruzi infected or normal mice. We found that DP cells from T. cruzi infected thymuses, but not from controls, express high levels of CD44 and CD69. These levels were consistent with those previously described for activated CD4+ and CD8+ T cells (Fig. 5a).
Moreover, analysis of thymocytes for the expression of L-selectin (CD62L), an up-regulated marker of recent thymic emigrant and mature T cell subsets, indicated that DP cells obtained from acutely chagasic animals express high levels of this molecule, as compared to corresponding controls (Figs. 5a–b). In fact, our data show that CD62L expression levels of intrathymic DP cells obtained from chagasic animals (but not from normal mice) are comparable to the normal single-positive CD4+ and CD8+ T cells that mature and are exported from the thymus (Fig. 5b). When comparing the activation profile of DP cells obtained from peripheral lymphoid tissues at day 15 after T. cruzi infection, the surface markers CD44 and CD69 were highly expressed as compared to naïve CD4+ and CD8+ T cells (Fig. 5c). We also determined that the DP cells released from thymus during the organ atrophy showed a characteristic profile of CD44 and CD69 expression similar to single-positive T cells differentiated upon infection with T. cruzi (Fig. 5d).
To further compare the differentiation of DP cells with activated cells, we determined the changes in level of IFN-γ transcripts after T. cruzi infection. Total RNA was isolated from DP cells sorted by FACS from peripheral lymph nodes 15 days after infection (Fig. 6a). The positive controls were obtained from activated CD4+ and CD8+ T cells purified from peripheral lymphoid tissues at day 15 after T. cruzi infection by FACS cell sorting based on the expression of high levels of the CD44 activation marker. As negative control, we purified naive CD4+ and CD8+ T cells with a CD44low phenotype from normal mice. Based on the real-time RT-PCR, a significant induction of IFN-γ was detected in DP cells from T. cruzi infected mice, at comparable levels of CD44high activated CD4+ and CD8+ T cells. As expected, naïve CD44low single-positive CD4+ and CD8+ T cells obtained from uninfected control mice did not exhibit IFN-γ mRNA induction (Fig. 6b).
To further study the differentiation status of DP cells, we tested whether this particular T cell subset had cytotoxic activity. To address this issue, DP cells were purified by cell sorting from chagasic peripheral lymph nodes at day 15 post-infection and varying numbers of sorted DP cells were incubated with P815 target cells. The effector:target (E:T) ratio was normalized based upon the frequency of T cells in order to compare the cytolytic activity on a per cell basis. Anti-CD3 mAb-redirected cytotoxic activity was measured because specific antigens for DP cells are unknown at present. In contrast to naïve CD4+ and CD8+ T cells, DP cells showed significant cytotoxic activity (Fig. 6c), which was specific for the anti-CD3 mAb signal, since P815 cell incubated in the absence of the antibody did not induce any cytotoxic activity (data not shown).
We further compared the cytotoxic activity of DP cells and activated CD8+ T cells obtained from peripheral lymphoid tissues at day 15 after T. cruzi infection. In these experiments, activated CD8+ T cells from chagasic peripheral lymph nodes were sorted by FACS based on the expression of high levels of the CD44 activation marker. As control, we purified naive CD8+ T cells with a CD44low phenotype from normal mice. Our data show that the level of cytotoxic activity from DP T cells were reduced as compared to activated CD44highCD8 T cells but significantly higher than naïve CD44lowCD8 T cells (Fig. 6d).
In the light of evidence that the DP cells released into the periphery acquire an activated phenotype similar to what is described for activated effector or memory single-positive T cells, we next evaluated a possible relationship between the presence of these cells and the development of severe clinical forms of human Chagas disease. For this purpose, we first examined the frequency of peripheral blood DP cell subset in cross-sectional studies of chronic chagasic patients at the indeterminate or cardiac clinical forms of Chagas disease. The results show higher percentage of DP cells from peripheral blood of cardiac chagasic patients as compared to noninfected individuals. However, this statistically significant increase was not seen in the percentages of DP cells from peripheral blood of patients at the indeterminate stage of chonic disease, when compared to control individuals (Fig. 7a).
Moreover, we showed a statistically significant increase in the activation profile of extrathymic activated DP cells in cardiac chagasic patients to assess the expression of activation marker HLA-DR indicated that peripheral blood subset of DP cells obtained from chronic cardiac chagasic patients (but not from patients at the indeterminate form of the disease) highly express significant levels of this molecule, as compared to normal individuals (Fig. 7b). Furthermore, the increase of extrathymic HLA-DR+ DP cell population in patients having the cardiac form of human Chagas disease was also correlated with the augmentation of the frequencies of activated HLA-DR+ single-positive CD4+ T cell (Fig. 7c) and CD8+ T cell subsets (Fig. 7d) obtained from chronically infected chagasic patients. CD8+ T cells also exhibited an activated phenotype regarding to the HLA-DR+ expression in the peripheral blood of patients at the indeterminate form of the disease (Fig. 7d).
Additionally, phenotype FACS analysis of the expression marker VLA-4 indicated that the peripheral blood subset of DP cells from both indeterminate and cardiac chronic patients express highed levels of this molecule as compared to normal individuals (Fig. 7e). Interestingly, we showed that there is no modulation on the expression of this marker in single-positive CD4+ cell (Fig. 7f) and CD8+ T cell (Fig. 7g) subsets obtained from chronic cardiac patients. Taken together, these results indicate that increased percentages of circulating DP cells are associated with severe clinical form of chronic human Chagas disease. While the percentages of these cells are not significantly increased in chronic patients at the indeterminate form, as the disease progresses to a cardiac clinical form, the peripheral blood subset of DP cells acquire a full activated HLA-DRhighVLA-4high profile.
Several pathogens, including T. cruzi, cause thymic atrophy [30]. Although the precise mechanisms underlying this phenomenon are not completely elucidated, it is most likely that this event is linked to a particular pathogen-host relation established during infection. Based on the studies developed with the T. cruzi model, it has been shown that the inflammatory syndrome mediated by TNF-α during the acute phase of infection induces the activation of hypothalamus-pituitary-adrenal (HPA) axis with the consequent release of corticosterone. In its turn, the glucocorticoid rise is likely associated with profound effects on the changes observed in the thymuses of T. cruzi infected mice, including the lymphoid and non-lymphoid compartments, as well as upon the disease outcome [31], [32]. Nevertheless, it should be pointed out that other host-derived molecules can be involved in the T. cruzi-induced thymocyte depletion. This is the case for galectin-3: thymic atrophy is not seen in T. cruzi-infected galectin-3 knockout mice [33]. Moreover, the parasite-derived trans-sialidase is involved in the generation of intrathymic T cell death [5], [34]. In a second vein, the thymic microenvironmental changes seen after acute infection comprise the enhanced expression of extracellular matrix ligands and receptors, which correlates with a higher fibronectin-driven migration of DP thymocytes, and the abnormal raise of immature DP cells in lymph nodes [8], [9].
Nevertheless, it was not determined if the changes of the thymic microenvironment seen following T. cruzi infection, would also lead to an altered intrathymic negative selection of the T cell repertoire. There is an emerging consensus stating that clonal deletion normally occurs late in thymocyte development at the double-positive to single-positive transition, and is believed to take place at the cortico-medullary junction of the thymic lobules [35]. In this regard, it has been shown that the thymic metallophilic macrophages restrictedly located at the cortico-medullary junction are related to the process of thymocyte maturation [23]. These cells show a number of features that differentiate them from both cortical macrophages and medullary interdigitating cells. They possess extremely developed specialized endocytic compartments for the processing and presentation of antigens by MHC class II molecules [24]. Here, we demonstrate that, although the alterations observed in the cortex and medullary compartment reveal the severe atrophy of the organ during the acute infection, the changes promoted by the infection in the thymic architecture changed the topological distribution of the metallophilic macrophages. Indeed, these cells are also found in the cortico-medullary region but in higher numbers and are spread along the cortex, as compared to controls. The distribution of these cells in the thymic compartments may represent a functional shift for the clonal deletion from the cortico-medullary junction to the cortex during the acute phase of the infection.
It is largely established that interactions between TEC and thymocytes control the development of the thymic microenvironment and T cell development. Previous studies have shown that the disruption of normal thymic architecture is known to affect the expression pattern of autoantigens by TEC and functionality of thymus [36]-[38]. Thymic medullary atrophy and decreased expression of Aire and TRAs have been reported in mouse models deficient in several genes involved in the NFκB pathway, such as TRAF6, NIK, RelB or p52, suggesting an important role of this pathway in the development of thymic medulla [12], [39]. We showed that the expression of Aire and highly selective tissue restricted antigens was readily detectable in whole thymus by real-time PCR analysis from infected mice, rather similar to controls. These data suggest that the expression of peripheral antigens in the infected thymuses is sufficient to modulate the tolerance induction by the negative selection process.
As the acute phase of infection progresses, the thymic atrophy becomes evident, as is the increase in numbers of apoptotic intrathymic DP cells, compared to their respective normal counterparts [4], [5]. Although this phenomenon may be a consequence of the changes observed in the organ, our data show that along the DP depletion there is sustained expression of Bim, an pro-apoptotic factor essential for thymocyte negative selection. Finally, by using an OTII TCR transgenic system, we were able to demonstrate that the administration of the cognate OVA peptide in the acutely infected mice undergoing thymic atrophy can induce TCR-stimulation-induced apoptosis of semi-mature thymocytes. Collectively, these data point out that negative selection operates normally during infection-promoted thymic atrophy, since the DP cells can be negatively-selected in the infected thymus by antigen-induced depletion. This supports previous work showing that intrathymic mature single-positive CD4+ or CD8+ T cells do not bear forbidden TCR genes as compared with their DP counterparts undergoing intrathymic differentiation [10].
Although the intrathymic checkpoints necessary to avoid the maturation of T cells expressing a forbidden T cell receptor repertoire are present in the acute phase of murine Chagas disease, it has been shown that DP cells are early released from infected thymus to the periphery [7], [9]. Exit of DP cells from thymus during both acute and chronic T. cruzi infection [10], may be favored by the upregulation of CD62L (L-selectin), which controls lymphocyte homing to lymph nodes. Since DP cells are progressively accumulated in peripheral lymphoid organs of T. cruzi acutely infected animals, we determined whether those cells exhibit an activated profile similar to effector/memory single positive T cells. The existence of this unconventional and rare (<5%) lymphocyte population in the periphery was explained as a premature release of DP cells from the thymus into the periphery, where their maturation into functionally competent single-positive cells continues [10]. There is, however, considerable evidence of an increased frequency of peripheral CD4+CD8+ T cells during viral infections and during acute T. cruzi infection. For example, in human immunodeficiency virus or Epstein-Barr virus infections, the percentage of DP cells can increase to 20% of all circulating lymphocytes [40], [41]. This fluctuation is also present in the secondary lymph nodes as we demonstrated in the experimental model of Chagas disease, in which DP cell subset increases up to 16 times in subcutaneous lymph nodes [10].
Interestingly, despite the expansion of peripheral DP lymphocytes in the experimental model of Chagas disease, these cells develop an activated phenotype, upregulating the activation markers CD44 and CD69, tightly linked to the differentiation status of T cells. In addition, we showed that highly purified (>98%) DP cell populations from infected mice obtained after cell sorting, produced high levels of IFN-γ mRNA. Furthermore, similar to previous studies showing high cytotoxic activity and effector memory phenotype of extrathymic DP cells in cynomolgus monkeys [42] and in a chimpanzee with experimental hepatitis C virus infection [43], our results indicate that the DP cells purified from peripheral lymphoid tissues of chagasic animals show cytotoxic activity as compared to naïve single-positive CD4+ or CD8+ T cells. Most interestingly, we found that patients having the cardiac form of human Chagas disease presented higher percentages of peripheral blood HLA-DR+ DP T cells as compared to non-infected individuals. These data suggest that this T cell subset might be associated with the development of the cardiac clinical form of the disease, probably as activated cells.
All together, our results indicate a role of the DP T cell subset in the inflammatory process caused by parasite-driven immune responses. It is possible that co-expression of CD4 and CD8 molecules on the T cell membrane would enhance the avidity of the TCR to its target cell. In this way, the concomitant trigger of CD4 and CD8 co-receptors through MHC-antigen complex would reduce the threshold of the antigen signal and decrease the requirements for the costimulatory signals to generate T cell activation [44]. This cascade of events would favor the DP T cell activation in the presence of low antigen concentrations. This scenario could promote a quick activation of the peripheral DP subsets upon the onset of infection where the antigens derived from the parasite are limiting. It is thus plausible to conceive that these cells may play a relevant role in modulating the adaptive immune responses via cytokine secretion. Accordingly, the cytokine profile secreted by DP T cells may drive the function of dendritic cells during the adaptive immune responses, thereby providing a link between innate and adaptive immunity. In addition, DP cells may have regulatory roles in host immunity. As DP cells are theoretically recognizing both class I and class II MHC complexes, it is also possible that these cells may eliminate activated antigen presenting cells, which would favor the parasite during of infection.
In conclusion, our data indicate that the key intrathymic checkpoints necessary to promote negative selection process of thymocytes are effective during acute chagasic thymic atrophy. Although the negative selection process is still functional in the acute phase, DP cells are prematurely released to the periphery. During the course of experimental infection, these peripheral DP cells acquire an activated phenotype similar to what is described for activated and memory single-positive T cells with high IFN-γ production, CD44+CD69+ expression and cytotoxic activity. This is particularly important, as we show for the first time that the appearance of full activated peripheral DP cells correlate with the cardiac clinical form of human chronic Chagas disease. Most likely, the presence of peripheral, mature and activated DP lymphocytes challenges the perception of the T cell populations involved in adaptive immune responses during the T. cruzi infection and suggests that DP cells promptly participate in the immune response in T. cruzi infection. Therefore, although the function of this DP T cell population remains to be defined in vivo, the presence of peripheral activated DP cells with potentially autoreactive TCR may contribute to the imunopathological events found in both murine and human Chagas disease. Correlations between the changes in the levels of DP T cell subsets and the extent of myocardial lesion during the evolution of cardiac disease could identify a clinical marker of disease progression and may help in the design of alternative therapies for the control of chronic morbidity of chagasic patients.
|
10.1371/journal.pntd.0003348 | Genome Update of the Dimorphic Human Pathogenic Fungi Causing Paracoccidioidomycosis | Paracoccidiodomycosis (PCM) is a clinically important fungal disease that can acquire serious systemic forms and is caused by the thermodimorphic fungal Paracoccidioides spp. PCM is a tropical disease that is endemic in Latin America, where up to ten million people are infected; 80% of reported cases occur in Brazil, followed by Colombia and Venezuela. To enable genomic studies and to better characterize the pathogenesis of this dimorphic fungus, two reference strains of P. brasiliensis (Pb03, Pb18) and one strain of P. lutzii (Pb01) were sequenced [1]. While the initial draft assemblies were accurate in large scale structure and had high overall base quality, the sequences had frequent small scale defects such as poor quality stretches, unknown bases (N's), and artifactual deletions or nucleotide duplications, all of which caused larger scale errors in predicted gene structures. Since assembly consensus errors can now be addressed using next generation sequencing (NGS) in combination with recent methods allowing systematic assembly improvement, we re-sequenced the three reference strains of Paracoccidioides spp. using Illumina technology. We utilized the high sequencing depth to re-evaluate and improve the original assemblies generated from Sanger sequence reads, and obtained more complete and accurate reference assemblies. The new assemblies led to improved transcript predictions for the vast majority of genes of these reference strains, and often substantially corrected gene structures. These include several genes that are central to virulence or expressed during the pathogenic yeast stage in Paracoccidioides and other fungi, such as HSP90, RYP1-3, BAD1, catalase B, alpha-1,3-glucan synthase and the beta glucan synthase target gene FKS1. The improvement and validation of these reference sequences will now allow more accurate genome-based analyses. To our knowledge, this is one of the first reports of a fully automated and quality-assessed upgrade of a genome assembly and annotation for a non-model fungus.
| The fungal genus Paracoccidioides is the causal agent of paracoccidioidomycosis (PCM), a neglected tropical disease that is endemic in several countries of South America. Paracoccidioides is a pathogenic dimorphic fungus that is capable of converting to a virulent yeast form after inhalation by the host. Therefore the molecular biology of the switch to the yeast phase is of particular interest for understanding the virulence of this and other human pathogenic fungi, and ultimately for reducing the morbidity and mortality caused by such fungal infections. We here present the strategy and methods we used to update and improve accuracy of three reference genome sequences of Paracoccidioides spp. utilizing state-of-the-art Illumina re-sequencing, assembly improvement, re-annotation, and quality assessment. The resulting improved genome resource should be of wide use not solely for advancing research on the genetics and molecular biology of Paracoccidioides and the closely related pathogenic species Histoplasma and Blastomyces, but also for fungal diagnostics based on sequencing or molecular assays, characterizing rapidly changing proteins that may be involved in virulence, SNP-based population analyses and other tasks that require high sequence accuracy. The genome update and underlying strategy and methods also serve as a proof of principle that could encourage similar improvements of other draft genomes.
| Paracoccidioides spp. is a thermally dimorphic pathogenic fungus that causes paracoccidioidomycosis (PCM), a neglected health-threatening human systemic mycosis endemic to Latin America where up to ten million people are infected. Disease can progress slowly, with roughly five new cases of disease per million infected individuals per year, with a male to female ratio of 13 to 1. About 80% of PCM cases occur in Brazil, followed by Colombia and Venezuela [2].
Within the Paracoccidioides genus, the three characterized phylogenetic lineages of P. brasiliensis (PS2, PS3, S1) and the one characterized lineage of P. lutzii (Pb01-like) can infect humans, and these groups can vary in virulence and induce different immune responses by the host [3], [4]. To better understand the pathogenesis and to enable genomics-based studies, the genomes of Paracoccidioides spp. were sequenced, analyzed and made publicly available in 2011 [1]. The Broad Institute of MIT and Harvard in partnership with the Paracoccidioides research community selected three reference isolates for sequencing and genomic analysis; assembly size for these strains varied between 29.1 and 32.9 Mb, and between 7,875 and 9,132 genes were identified in each strain [1]. These included two strains of P. brasiliensis (Pb18 representing the S1 lineage and Pb03 representing the PS2 lineage) and one strain of P. lutzii (Pb01) [1].
These sequenced isolates are extensively referenced in molecular biology and experimental mycology laboratories working with Paracoccidioides spp. and also other pathogenic fungi, including those working with yeast phase specific genes expressed during host infection. These sequences also serve as a reference to analyze high-throughput data increasingly generated by genomic, metagenomic, transcriptomic and proteomic approaches. Additionally, accurate sequences are critical for evolutionary analyses, e.g., to identify positively selected genes, as well as to provide new targets for the design of diagnostic assays.
The P. brasiliensis Pb18 and Pb03 strains and the P. lutzii Pb01 strain were sequenced using the sequencing technology and computational methods available at the time, which produced high quality draft assemblies. However, the assemblies included a large number of gaps and uncertain or low quality nucleotides in the final consensus sequences. Also, the annotation pipelines flagged only the most extreme annotation errors for curation and did not address the larger number of smaller scale errors in the gene models and underlying sequence [5]. Correction of such errors requires re-evaluation of the assembly consensus sequence and associated annotation.
Assembly errors that could not be detected in previous data and passed standard quality control criteria at that time can now be corrected using next generation sequencing (NGS) for systematic assembly improvement. These include errors in gene-containing regions of the original genomic assembly affected by poor quality sequence or ambiguities, which can cause incorrect gene structure predictions. Since predicted genes of reference genomes are now frequently used for homology-based inference or confirmation of gene structures in closely related species, errors in the reference sequence may be propagated to other genomes [6], [7]. Therefore, systematic improvement of a genome assembly and annotation can impact not just the understanding of that particular species, but also that of other related species for which it is used as a reference for comparison.
Here, we present an update of the three Paracoccidioides reference genome sequences achieved using Illumina re-sequencing to correct assembly errors and document the improvements obtained. The improved and updated reference genome assemblies and annotations of this important human fungal pathogen now allow more accurate SNP analyses, genome-wide evolutionary (e.g., selection) analyses that depend on high-quality sequences, phylogenetic footprinting studies of regulatory regions, or primer and probe design for diagnostic assays.
Three reference isolates of Paracoccidioides spp. (Pb01, Pb03 and Pb18), representing two species, were previously sequenced. The isolate of P. lutzii (Pb01) was a clinical isolate originating from an acute form of paracoccidioidomycosis (PCM) in an adult male. The two P. brasiliensis isolates were from individuals presenting chronic PCM; Pb03 represents the PS2 phylogenetic group and Pb18 the S1 group [1], [4]. In a partnership between the Broad Institute and the Paracoccidioides research community, these genomes were previously sequenced using multiple whole genome shotgun libraries constructed from genomic DNA for each strain; paired-end sequences were generated for each with Sanger technology and assembled using Arachne [1] (assembly v1; S1 Table).
The reference strains Pb01 (previously sequenced DNA sample) and Pb03 and Pb18 (newly extracted DNA samples) were re-sequenced using Illumina technology. For library construction, 100 ng of genomic DNA was sheared to ∼250 bp using a Covaris LE instrument and prepared for sequencing as previously described [8]. A library for each of the three samples was used to generate 101 base paired-end reads on the Illumina HiSeq2000 platform, producing an average genome coverage of 165X.
To improve the genome sequence of Paracoccidioides spp. strains Pb01, Pb03 and Pb18, Illumina paired-end reads were aligned to the draft reference assemblies (assembly v1) using BWA version 0.5.9 with default settings [9]. The assembly consensus sequence was re-evaluated by providing these alignments as input to the automated assembly improvement program Pilon (version 1.4, default parameters, www.broadinstitute.org/software/pilon/). Pilon uses the Illumina read alignments for multiple classes of assembly correction. First, Pilon scans the read alignments for positions where the sequencing data disagree with the input genome (assembly v1) and corrects small errors such as single nucleotide differences and small insertion/deletion events. Second, Pilon looks for coverage and alignment discrepancies to identify potential mis-assemblies and larger variants. Finally, Pilon uses reads anchored adjacent to discrepant regions and gaps in the input genome to reassemble the region, attempting to fill in the true sequence including large insertions. As output, Pilon provides the sequence of this improved genome assembly (assembly v2; Fig. 1) along with files summarizing the changes and quality measures used in the assessment.
Protein-coding genes were predicted in the improved assemblies (assembly v2) using a combination of gene models from the prediction programs Augustus [10], Genemark-ES [11], GlimmerHMM [12], Genewise [13], and Snap [14], as well as automated revision based on EST data (e.g., from [15]) and manual gene revision of flagged calls. The predicted gene sets were then provided as input to EvidenceModeler (EVM) [16] to obtain the best consensus model for a given locus. The consistency of the gene models was evaluated by examining alignments of protein orthology groups identified using OrthoMCL [17]. EVMLite was used to rescue orphan genes not captured in EVM; only those genes with additional evidence such as overlap to Genewise or non-repeat HMMER3 PFAM domains were rescued, as well as non-redundant genes overlapping the OrthoMCL genes in clusters containing 2 or more genomes. Lastly spurious gene models matching repetitive or low-complexity sequences were removed.
For each Paracoccidioides genome, we compared the original annotation (v1) with the updated annotation (v2) to evaluate the changes in the new gene sets. To precisely characterize the types of changes across the v1 and v2 annotations, we first mapped the corresponding gene between the two assemblies. The v1 and v2 assemblies were aligned using nucmer [18], and the alignment coordinates were used to assign gene correspondence between the initial annotation v1 and the new annotation v2. This mapping also allowed us to preserve locus numbers in the updated gene set. Each annotated gene was assigned a locus number, keeping where appropriate the previous locus number of the form PAAG_##### (Pb01), PABG_##### (Pb03) or PADG_##### (Pb18), which serves as a unique identifier within each genome and across assemblies. New genes, merged genes, and genes with large structure and sequence changes in transcripts were assigned new and unique locus numbers following the last locus number of annotation v1. Locus numbers of deleted genes do not appear in the final gene sets.
To evaluate whether the changes in gene sequence and structure produced a more accurate gene set, the gene sequences of annotation v1 and of the updated annotation v2 were compared via sequence similarity and orthology analysis. To evaluate the consistency of gene structures for orthologs as well as their conservation between species, OrthoMCL version 1.4 with a Markov inflation index of 1.5 and a maximum e-value of 1e-5 was used to identify orthologous clusters across the six total protein sets corresponding to annotations v1 and v2 of each Paracoccidioides strain. For each cluster group representing putative orthologs, we compared the maximum length difference among the three Paracoccidioides genes in annotation v2 to that in the annotation v1.
To compare the functional content of the v1 and v2 gene sets, we evaluated both protein domain families (PFAM) and pathway information (KEGG). Using HMMER3 [19], we mapped v27 of the PFAM domain database [20] to both the v1 and v2 gene sets. KEGG domains [21] from release 65 were also mapped to both gene sets using BLAST.
To evaluate changes in the gene structure, the corresponding transcripts from annotation v1 and v2 were identified as described above. We also aligned gene sets v1 and v2 using BLASTn [22] version 2.2.28+ with default parameters, using an in house Perl script to determine the types of modification for each gene, which included changes in gene length, gene coverage and percent nucleotide identity. We manually checked a random gene sample of each type of change (up to 10 genes) from both gene correspondence and BLAST analyses to verify that changes in gene set v2 were actually gene improvements. To evaluate changes in the coding regions of genes of high interest to the community, we selected known specific yeast-phase genes or virulence factors of Paracoccidioides spp., as well as other genes that are generally considered relevant for research on Paracoccidioides or related dimorphic pathogens, for manual review. The sequences of these genes' coding regions were aligned at the protein level with CLUSTALW [23] version 2.1, using both the v1 and v2 annotations.
The coverage of Core Eukaryotic Genes defined by CEGMA [24] was evaluated using the CoreAlyze tool (http://sourceforge.net/projects/corealyze/) to summarize results for all the v1 and v2 gene sets. BLASTp version 2.2.28+ was run with default settings using protein sets from annotations v1 and v2 as the database, with Saccharomyces cerevisiae and Schizosaccharomyces pombe CEGMA proteins as the query. We also included the protein gene sets of two close relatives of Paracoccidioides, the dimorphic fungal pathogens Blastomyces dermatitidis and Histoplasma capsulatum. In order to obtain a detailed picture of the changes where gene annotations were modified but not completely overridden, we compared protein sequences between the two versions, excluding proteins that were added or deleted from the final gene set v2, as well as proteins for which the new annotation was for a completely different transcript at the same locus. For a hit to be counted, the protein needed to match a protein in the reference set with at least 75% identity for the v1 and v2 annotations. This percent identity cutoff was determined empirically to eliminate spurious low similarity alignments. The percent identity and the bit score between the query protein and each version of the Paracoccidioides annotations (v1 and v2) were compared.
The strains of the genomes of Paracoccidioides spp. previously sequenced [1], Pb18 and Pb03 and Pb01, were re-sequenced using Illumina 101 bp paired-end reads. This sequencing generated 93.6 million reads for Pb18 with an average coverage of 198X, 124.2 million reads for Pb03 with an average coverage of 150X and 110.0 million reads for Pb01 with an average coverage of 148X. This high coverage sequence data was then used to refine the consensus sequence of the original assembly by assessing differences between the new sequence and the previous assemblies. This can target a wide range of improvements, including correcting base calls, resolving ambiguous bases and closing gaps within scaffolds.
Fig. 1 shows a simplified overview of the workflow of genome improvement. The new Illumina data were used to systematically improve the three Paracoccidioides spp. assemblies using Pilon (http://www.broadinstitute.org/software/pilon/). Pilon bases its improvement calls on an alignment of the reference genome and the sequenced reads. The aligned bases and depth at each sequenced position provides evidence for the reference base or for an alternative; where changes are supported they can result in single base differences, insertion or deletion of single bases or larger regions, identification of collapsed regions and more complex changes and gap filling based on local reassembly. Reads of each of the genomes of Pb18, Pb03 and Pb01 were aligned to the corresponding reference assembly using BWA [9] and the resulting bam file was used as input for Pilon.
In each of the Paracoccidioides assemblies, Pilon identified and fixed base errors in the consensus sequence. The statistical improvements for the assemblies v2 of Paracoccidioides spp. are summarized in Table 1. The most frequent class of changes was single base substitutions, identified as single nucleotide polymorphisms (SNPs) between the assembly and reads. Between 3,018 and 3,290 single base errors were corrected in each assembly. Small insertions and deletions were also incorporated into each assembly. The major classes of changes can be attributed to bases added and removed in reassembly fixes, collapsed bases in the new assembly and the closing of gaps (Table 1). Regions of misassembly identified and fixed by Pilon resulted in bases added or removed but no new gaps opened. Across all the assemblies, 20% of all initial gaps were closed by Pilon; the number of gaps closed were 113, 56 and 212, for Pb18, Pb03 and Pb01, respectively. Overall, the assembly improvement process led to an increase of contig N50 for all strains. About ∼99% of low quality nucleotides in assemblies v1 were well supported or fixed with high flag coverage in assemblies v2.
Overall, the P. lutzii Pb01 genome assembly was most substantially improved, based on comparing statistics for all v1 and v2 assemblies (S1 Table). The contig N50 for Pb01 v2 increased by 29.1 kb; more bases were added and removed after re-assembly fixes and more gaps were closed than in the other two genomes. The genome size and number of scaffolds of Pb18 and Pb03 were essentially unchanged. The Pb01 genome size decreased slightly from 32.94 to 32.93 Mb; the updated assembly contains one scaffold fewer, as two scaffolds were merged by closing the gap between them. The number of contigs was reduced in all three strains, which considering the increase in the contig N50 indicates that the assemblies v2 for Pb18, Pb03 and Pb01 were less fragmented. All these changes indicate that the genome assemblies v2 after Pilon improvements were more contiguous, contained more bases with high quality, and had fewer gaps and errors.
The gene annotations of the reference strains Pb18, Pb03 and Pb01 were updated using a pipeline to transfer and revise gene structures (Methods). The implemented annotation pipeline was an updated and improved version of the previous protocol used to annotate Paracoccidioides spp. assemblies v1. The current pipeline includes an updated set of gene prediction programs, including the EVM caller used to select the best call for each locus. Databases used for training these gene prediction programs are also more comprehensive, with more sequences available from the dimorphic fungi group for comparison. Also, the databases used for homology inference and functional annotation were updated since the previous annotation. In addition, we identified orthologs to evaluate the gene calls for consistency (see below). The incorporation of these new methods and data improved the evidence supporting gene prediction.
The updated gene sets are more consistent across the three Paracoccidioides genomes (S1 Table). The total gene count for the two P. brasiliensis genomes now only differs by 37 in v2 whereas the v1 gene counts differed by 866; overall the update removed 351 genes from Pb18 and added 552 genes to Pb03. P. lutzii (Pb01) also has a more similar gene count, due to 306 fewer genes in the v2 compared to v1. A more detailed view of the gene structure changes by major categories is provided in Table 2; these statistics were calculated by mapping the transcripts from the previous annotation to the corresponding locus on assembly v2. Notably, this analysis helped recover a large number of genes missed by the original annotation in each genome; the total genes newly added to a region was 840 in Pb18, 933 in Pb03 and 936 in Pb01. In addition, dubious genes were removed from each genome; the number of genes no longer present at the same locus was 1187 in Pb18, 490 in Pb03 and 1265 in Pb01. Other changes include extending or truncating transcripts, merging or splitting transcripts, changes to splice sites, and changes to UTRs (Table 2). Only 23% of genes in the v2 annotations were unchanged from v1; the primary transcripts were identical for 1816 genes in Pb18, 2599 in genes Pb03 and 1581 in genes Pb01. Genes with any type of change in their coding sequences represent a smaller subset, in total 5734 (68%) in Pb18, 4895 (58%) in Pb03 and 6309 (71%) in Pb01.
Both sequence addition (gap filling and local reassembly) and small changes in the genome assemblies (single-nucleotide substitutions or insertion/deletion events (indels)), contributed to the improvement of the gene annotation in the update v2. Two examples of how indel correction fixed gene structures are shown in Fig. 2. In the first case (left panel), an extra C was inserted at a polyC tract in PABG_00129 of Pb03; correction of this position resulted in extending the coding DNA sequence (CDS) of this gene by 423 bases. In the second case (right panel), an A was deleted at a polyA tract in PABG_00790; correction of this position also corrected the reading frame, allowing for removal of a false intron that was needed to step over a stop codon and extension of the CDS of this gene by 252 bases. While these are small changes to the underlying assembly, both have had larger impact on correcting these gene structures.
The annotation improvements were also analyzed by comparing the alignments of orthologs for all three Paracoccidioides genomes, identified by OrthoMCL (Methods). For orthologs identified either from the v1 or v2 assemblies, maximum and minimum gene length was computed for each ortholog cluster. In comparing these gene lengths (Fig. 3A), the v1 gene annotations (red points in scatterplot) exhibited a higher variation among Pb18, Pb03 and Pb01 orthologs compared to annotation v2 (blue points). The positions that are closer to the diagonal correspond to smaller differences in gene length between orthologs; as expected for an improved annotation, the v2 points are closer to the diagonal than v1. These differences between maximum and minimum length of the genes within each orthologous cluster group were also plotted on a logarithmic scale (Fig. 3B), based on sorting cluster differences from smallest to largest. The v2 annotation differences (blue curve) were lower and well separated from the v1 annotation (red curve), providing additional support of the increased length concordance in the v2 annotation.
Further analysis of gene conservation also supported the greater consistency among the Paracoccidioides spp. genomes in the v2 annotation. The number of genes found in all three genomes increased, whereas the number of unique genes specific to only one genome decreased; this has produced a more uniform set of protein coding genes (Figure S2). The improved structural annotation also led to improvements in functional annotation. The v2 annotation had more genes with assigned protein domain families (PFAM) and pathway information (KEGG), using the same version of these databases for the v1 and v2 gene sets (Figure S2). This supports the higher functional content of the revised gene sets, despite the lower total gene counts in two of the genomes.
We also manually reviewed and curated the predicted structures of a number of protein-coding genes that are of importance to the Paracoccidioides research community, including well-characterized yeast-phase specific genes and other virulence factors. This introduces changes to the transcript sequence of 27 of these genes (Table 3). The improvements to the assemblies resulted in updated transcript predictions for the vast majority of genes of the three reference strains, with substantially corrected gene structures for several virulence-associated or yeast-phase specific genes of central importance in Paracoccidioides or other dimorphic fungi, including HSP90 [25], PbGP43 [26], PbP27 [27], RYP1-3 [28], BAD1 [29], catalase B, alpha 1,3 glucan synthase and the beta glucan synthase target gene FKS1 [30], [31]. An extreme example is the HSP90 gene, where corrections were made to the sequence of each of the three Paracoccidioides genomes (Fig. 4). This example illustrates the annotation errors in v1 of all Paracoccidioides reference strains that were fixed in v2 after Pilon improvement and re-annotation. In this case one or more single-nucleotide errors, unknown single nucleotides (N's), and/or single nucleotides that were erroneously reported as absent or duplicated by Sanger sequencing resulted in radically different predicted gene structures (intron/exon and/or gene boundary errors). This is shown in detail for a cluster of errors present at the end of HSP90 in Pb03 (Fig. 4B), which included alteration of the proper stop codon, resulting in premature truncation of this gene. Another example of coding sequence updates to multiple genomes is shown for FKS1, where different regions of the Pb03 and Pb18 proteins were restored in the updated assemblies and annotations (Figure S1).
The improvements in the annotation v2 were also analyzed for completeness by comparing to a set of highly conserved fungal genes defined by CEGMA and to protein sets of related dimorphic human pathogenic fungi. Genes in the v2 annotation showed a higher coverage of both the CEGMA and related dimorphic fungal data sets in comparison with annotation v1, suggesting these v2 genes are more complete (Figure S3). Furthermore, we examined the level of conservation to other fungi, by analyzing the difference of the BLASTp score between the v1 and v2 protein sets compared to those of B. dermatitidis, H. capsulatum and the CEGMA genes of S. cerevisiae and S. pombe. We observed that in all cases the v2 annotation had more hits greater than the minimum-similarity cutoff, and that the vast majority of genes of the v2 annotation had higher BLAST score values than their counterparts from v1 annotation (Figure S4).
The initial draft genomes of three isolates of Paracoccidioides (P. brasiliensis isolates Pb03 and Pb18, and P. lutzii isolate Pb01) served as the first complete genome references for this fungal species [1]. Although these assemblies were obtained using the best technology available at that time, they included gaps and low quality sequence in genic and intergenic regions, which in turn resulted in a number of suboptimal gene structures, coding sequences and predicted protein sequences. This work has revised these reference genomes, providing the Paracoccidioides community with more complete and accurate sequences; this provides a more accurate foundation for future genome-based, molecular biological or genetic research on paracoccidioidomycosis and the fungal strains that cause it. The strategy we have followed will more widely be useful also for other groups wishing to update fungal and other microbial genomes in future.
The updating of a reference genome, in particular of the underlying assembly and annotation, can be thought of as a largely computational form of deep sequence curation. The success of the update we present here shows that next-generation sequencing (NGS) together with publicly available software tools can markedly enhance the quality of a eukaryotic genome resource. Indeed, the availability of affordable NGS sequencing opportunities makes such endeavors accessible to small bioinformatics groups. The massively computer-assisted component of such an update, which can include tabular and graphical views for monitoring improvements and performing quality control, can be complemented by choosing and following a few ‘guide genes’ to evaluate the process. This focused analysis provides tangible examples of how the update affected predicted properties of important genes, such as gene structure or encoded proteins.
The accuracy of a genome sequence and associated annotations are critically important for many types of analysis; therefore validating and improving the accuracy of the sequence and annotation can have wide impact, especially for methods highly sensitive to sequence errors. One example involves examining a genome sequence for evidence of genes and genic regions likely to be under positive selection. Such genes and genic regions, which are believed to be relatively rare in many eukaryotic genomes (see, e.g., [32]), are sometimes associated in pathogenic organisms with virulence or rapid adaptation to host conditions, including resistance to defense by the host or avoidance of the host immune system. An example of such adaptation has been found for surface proteins of diverse pathogens [32] and in fungi of the proline-rich antigen gene in Coccidioides spp. [33]. Positive selection can also occur in response to antimicrobial drugs, as in chloroquine resistance in Plasmodium falciparum [34]. Candidate regions under positive selection are commonly identified as sections of coding regions having unusually high rates of nonsynonymous (amino-acid changing) substitutions. Precisely because such regions are quite rare, a coding region of low sequence quality having several sequencing errors could be categorized as a region under positive selection, and if there are several such regions in a genome, an automated genome-wide screen will report a high percentage of false positives. Conversely, assembly and annotation improvements such as we describe here can effectively evaluate and fix such regions of a genome so that even error-sensitive evolutionary analyses become realistic.
Similar considerations also apply to analyses that have more obvious clinical relevance. For example, improving a DNA sequence's accuracy can bring it closer to being ‘clinical grade’ or ‘diagnostic grade’. Indeed, identification of a clinical sample of a human pathogenic fungus isolated in a hospital using sequence comparison requires certainty that any nucleotide differences (e.g., resulting from single nucleotide polymorphisms/SNPs) observed between the sequenced sample and trusted reference strain(s) or isolate(s) are not simply errors in the reference. For fungi encountered in clinical contexts, only one or a few traditionally used loci are typically represented by reliable reference sequences, which are often from the ribosomal DNA, or from one or two protein-coding genes known beforehand to be diagnostically informative. Reference diagnostics, as well as diagnostic PCR assays (e.g., primer/probe design in real-time PCR assays), depend on such regions that have been reliably characterized at the molecular level in a fair number of related species or strains that could be present in clinical settings. Whole-genome gene sets offer, however, new perspectives; if their sequence quality is high, one could then systematically and exhaustively screen alignments of the full gene sets for diagnostically promising genes and genic regions that are likely to be informative for the identification task at hand. Such genome-wide screens should be able to identify new, candidate target loci, and molecular assays could then be developed for them and validated. Genome sequences also allow for metagenomic or metatranscriptomic analysis, where reference genomes enable identification of the pool random sequence from the population of microbes in a sample. Such wide applications will be better powered by efforts such as this to improve the set of reference genomes that form a fundamental basis of comparison and analysis.
By re-sequencing three reference strains of Paracoccidioides spp., using deep sequencing depth of Illumina paired-end reads, we have been able to substantially improve the assemblies and annotations for this important human fungal pathogen. Here we have presented the updated and improved annotated genome sequences, which constitute new references that can be used in diverse future molecular projects by those working in the field of medical mycology. Since the process leading to the new sequences is largely automated using publicly available programs and the NGS technology used is cost-effective, the success of our strategy represents a proof of concept that may stimulate similar updates of other genomes in future.
|
10.1371/journal.pgen.1003957 | Oct4 Is Required ∼E7.5 for Proliferation in the Primitive Streak | Oct4 is a widely recognized pluripotency factor as it maintains Embryonic Stem (ES) cells in a pluripotent state, and, in vivo, prevents the inner cell mass (ICM) in murine embryos from differentiating into trophectoderm. However, its function in somatic tissue after this developmental stage is not well characterized. Using a tamoxifen-inducible Cre recombinase and floxed alleles of Oct4, we investigated the effect of depleting Oct4 in mouse embryos between the pre-streak and headfold stages, ∼E6.0–E8.0, when Oct4 is found in dynamic patterns throughout the embryonic compartment of the mouse egg cylinder. We found that depletion of Oct4 ∼E7.5 resulted in a severe phenotype, comprised of craniorachischisis, random heart tube orientation, failed turning, defective somitogenesis and posterior truncation. Unlike in ES cells, depletion of the pluripotency factors Sox2 and Oct4 after E7.0 does not phenocopy, suggesting that ∼E7.5 Oct4 is required within a network that is altered relative to the pluripotency network. Oct4 is not required in extraembryonic tissue for these processes, but is required to maintain cell viability in the embryo and normal proliferation within the primitive streak. Impaired expansion of the primitive streak occurs coincident with Oct4 depletion ∼E7.5 and precedes deficient convergent extension which contributes to several aspects of the phenotype.
| Embryogenesis is an intricate process requiring that division, differentiation and position of cells are coordinated. During mammalian development early pluripotent populations are canalized or restricted in potency during embryogenesis. Due to considerable interest in how this fundamental state of pluripotency is maintained, and the requirement of the transcription factor Oct4 to maintain pluripotency, Oct4 has been intensively studied in culture. However, it is not clear what role Oct4 has during lineage specification of pluripotent cells. Oct4 removal during lineage specification indicates that it is required in the primitive streak of mouse embryos to maintain proliferation. The consequences of Oct4 removal diverge from the consequences of removing another factor required for pluripotency between preimplantation development and early cell fate specification suggesting that the network Oct4 acts within is altered between these stages.
| Oct4 is a homeodomain-containing transcription factor (TF) of the POU family required for pluripotency in ES cells and preimplantation embryos [1]. It has been extensively characterized in ES cells, and established as a hub of the signaling network that maintains pluripotency [2]–[5]. Embryonically, Oct4 is present in the developing zygote and down-regulated somatically between E7.0 and E9.0 depending on the cell type (see Supplementary (S) Figure (Fig.) S1 and S2 for detail) [6], [7]. After E9.0 of murine development Oct4 is restricted to the germline, persisting until maturation of type A to type B spermatogonia in the male germline, in contrast to the female gametic lineage where it is depleted during meiosis (E14–16) before up-regulation as oocytes mature within primordial follicles [6], [8]–[10]. Several regulators of Oct4 have been established in vivo. Oct4 is maintained through the early stages of embryonic development by intercellular Nodal acting in part through Smad2 [11], [12]. Conversely, Cdx2 mediates repression of Oct4 in trophectoderm of the early blastocyst, while both Eomes and Gcnf mediate repression in the embryo after implantation [13], [14].
Oct4 buffers the ICM against differentiation into trophectoderm (the embryonic contribution to the placenta), but the proposal that Pou5f1 (gene symbol for Oct4) emergence relates to evolution of the mammalian placenta [15] is not supported given that Pou5f1 evolved before the origin of amniotes [16]. It is unknown whether Oct4 has a conserved role, or any post-implantation function in murine somatic development. Pluripotent somatic cells persist until E7.5–8.5 based on teratogenesis experiments [17], [18] and ∼E8.0 based on epiblast stem cell (EpiSC) derivation [19], suggesting that Oct4 might continue to maintain pluripotency during this window of development. in vitro studies have also inferred many roles for Oct4 between the pre-streak and headfold stages, ∼E6.0–E8.0, including regulating neural versus mesendoderm differentiation [20], [21] as well as promoting cardiomyocyte [22] and neuronal differentiation [23]. However aside from maintaining the viability of primordial germ cells (PGCs), Oct4's role in post-implantation development has not been characterized in vivo [1], [2], [24], [25].
The extent of Oct4's function at the molecular level is also unclear. Physical interactions suggest Oct4 may have roles in chromatin modification, regulation of transcription, DNA replication and DNA repair as well as post-transcriptional modification, ubiquitination, and various other functions [2]–[4], [26], [27]. Oct4 both activates and represses transcription [28]. It binds thousands of sites in the ES cell genome, often co-occupying these sites with Sox2, Nanog, Smad1 and Stat3 [5]. The majority of genes occupied by several of these transcription factors (TFs) are active in ES cells, but their binding does not ensure expression [5].
Since Oct4 protein normally persists in somatic cells until ∼E7.0–E9.0 but Pou5f1 null embryos arrest at E3.5, we asked what role Oct4 had later in murine development, using a conditional system to deplete it ∼E7.5. We show that Oct4 depletion ∼E7.5 results in craniorachischisis, random heart tube orientation, failed turning, defective somitogenesis as well as posterior truncation. The phenotype is not the result of a general delay in development, nor does it result from a failure in the pluripotency network present in the ICM. Depletion of Sox2, another core member of the pluripotency network in an overlapping window of development does not phenocopy Oct4 depletion. Instead, Oct4 is required until ∼E7.5 to maintain cell viability in the embryo and proliferation in the primitive streak. In its absence, convergent extension is disrupted leading to several morphogenetic defects.
We used a conditional mutant of Oct4 to study its role after E3.5 when it is essential for development. We used floxed Pou5f1 alleles (Oct4f) [25] and a tamoxifen inducible recombinase (CreERT2) that is ubiquitously expressed from the ROSA locus [29]. To establish the window of development during which embryos are sensitive to Oct4 depletion, we staggered the initial dose of tamoxifen with respect to embryonic maturity and administered a second supplementary dose 12 hrs later to enhance overall recombination efficiency. Oct4f/f;CreERT2+/− embryos administered tamoxifen ∼E8.0 and ∼E8.5 before analysis ∼E9.5 did not have a phenotype (Table S1, row A (S1A), while tamoxifen administration ∼E7.5 and ∼E8.0 before analysis ∼E9.5 resulted in a partially penetrant phenotype (Fig. S3; Table S1B). Unlike tamoxifen administration beginning ∼E7.5 or ∼E8.0, all Oct4f/f; CreERT2+/− embryos induced ∼E6.0 and ∼E6.5 before analysis ∼E9.5 were amorphous, lacking structures aside from what resembled anterior neural head folds (Fig. S4; Table S1C). Tamoxifen administration ∼E7.0 and ∼E7.5 also led to a fully penetrant phenotype ∼E9.5 (Table S1D).
E9.5 embryos administered tamoxifen ∼E7.0 and ∼E7.5 failed to turn, had severe posterior truncations, randomly oriented heart tubes, craniorachischisis (open neural tube along its entire length) as well as impaired somitogenesis (Fig. 1A–C). Such animals are referred to as Oct4COND MUT in the remainder of this report. The phenotype is not a consequence of tamoxifen administration, leaky recombinase activity prior to tamoxifen administration, or associated with recombination of a single Pou5f1 allele: no Oct4f/f embryos induced ∼E7.0, no uninduced Oct4f/f;CreERT2+/− embryos, nor any Oct4+/f;CreERT2+/− embryos induced ∼E7.0 had phenotypes ∼E9.5 (Table S1E–G). Reducing the quantity of tamoxifen per dose administered ∼E7.0 or failure to administer the second dose ∼E7.5 led to incomplete penetrance of the Oct4COND MUT phenotype (Table S1H–J): 80%, 40% and 0% of embryos ∼E9.5 exhibited the Oct4COND MUT phenotype when a single full, half, and quarter tamoxifen dose was administered ∼E7.0 (Table S1H–J). This suggests reduced recombination with these lower tamoxifen doses. Collectively, these data support Oct4 depletion causing the Oct4COND MUT phenotype.
To determine the time course of Oct4 depletion with this system, we compared Oct4 transcript and protein abundance between Oct4f/f and Oct4f/f;CreERT2+/− littermates administered tamoxifen ∼E7.0. A single dose of tamoxifen was used to avoid a compound effect from a second dose. Relative Oct4 transcript abundance (Oct4f/f;CreERT2+/−/Oct4f/f;CreERT2−/− littermates) was significantly different 12 hrs after tamoxifen administration (ATA) (Fig. 1D; Table S1K; F5,13 = 15.48, p<0.05 1-way ANOVA, *p<0.05, **p<0.01 Bonferroni posttest). The fraction of cells in which Oct4 was detectable by immunohistochemistry was lower 20 hrs ATA, which is ∼E7.5 (Fig. 1E, Fig. S5A–D; Table S1L; F3,10 = 12, p<0.05 1-way ANOVA, **p<0.01 Bonferroni posttest). A distinct primary antibody indicated that Oct4 protein was undetectable 24 hrs ATA in Oct4f/f; CreERT2+/− embryos (Fig. 1F,G; Table S1L). Since penetrance of the phenotype is complete when tamoxifen administration begins ∼E7.0, partial when tamoxifen administration begins ∼E7.5, and the fraction of cells with detectable Oct4 protein reduced ∼20 hrs ATA (following administration ∼E7.0), these data indicate that Oct4 is required until ∼E7.5.
Oct4 depletion does not cause a global delay in development. Administering tamoxifen ∼E7.0 and ∼E7.5 to avoid partial penetrance, Oct4f/f;CreERT2+/− embryos were recovered in a ratio of 1∶1 with Oct4f/f littermates until E9.5, but less frequently at E11.5 (Fig. 2A; Table S1M–O). Features disrupted in Oct4COND MUT remained arrested in the mutants that persisted beyond E9.5 (Fig. 2B,C), indicating that the Oct4COND MUT phenotype is not a global delay in development but disruption of select features. Indentation of the otic cup occurred and the branchial arches formed in Oct4COND MUT, events that normally occur by E9.0. Forelimb buds also protruded in Oct4COND MUT as they normally do by E9.5. Conversely, the neural tube normally closes rostrally between E8–9 and caudally by E9–10 (we refer to caudal and rostral neural tube closure with respect to closure point 1 at the hindbrain cervical boundary throughout; see Figure 2D) [30], turning normally occurs by ∼9.0 and posterior extension normally reaches 21–29 somites by E9.5 in WT embryos. These events always failed at E9.5 when Pou5f1 excision was induced ∼E7.0 (Fig. 1A–C; Table S1D; 26.5 versus 4.6 somites in Oct4f/f versus Oct4f/f;CreERT2+/− littermates). Additionally, heart tube orientation was randomized, 38.6% of Oct4f/f;CreERT2+/− had situs inversus while the orientation of 6.8% was ambiguous (Table S1P; p>0.05 Chi-square test). The neuroepithelium of Oct4COND MUT embryos was also thicker in regions, particularly in the distal portion of the embryo (Fig. S6A–C; Table S1D; F1,287 = 94.95, p<0.05 2-way ANOVA, ***p<0.001 Bonferroni posttest). These data indicate that Oct4 is required for posterior extension, turning, heart tube orientation and neural tube closure (NTC).
Partial phenotype penetrance following tamoxifen administration ∼E7.5 was used to assess whether the cause of disrupted features in Oct4COND MUT embryos were related. Coincidence of features in litters with incomplete phenotype penetrance suggests related causation of the coincident features. Craniorachischisis and posterior truncation coincided in all 23 of the 36 embryos analyzed (Fig. S2; Table S1B; p = 1.64E-10, hypergeometric test). Conversely 2 turning defects in the 9 embryos where rostral NTC failed suggests independence of these processes, although the small number of embryos limits statistical power in this case (Fig. S3; Table S1B; p = 0.72, hypergeometric test). These data suggest independent requirements for Oct4 in closure at closure point 1/posterior extension and rostral NTC.
Craniorachischisis occurs when closure at closure point 1 fails (see Figure 2D). Convergent extension elongates the embryo in the anterior-posterior axis during gastrulation and neurulation, bringing the neural folds into opposition prior to adhesion at closure point 1. Failed convergent extension results in broad midlines and enlarged notochord diameter as both narrow during convergent extension. Oct4COND MUT embryos exhibit broad neural plates (Fig. 2H–J; Table S1D; F2,22 = 17.42, p<0.05 2-way ANOVA, **p<0.01 Bonferroni posttest) and enlarged notochord diameter (Fig. S6D–F; Table S1D; p<0.05, two-tailed student t-test). Concordance between posterior truncation and craniorachischisis, broadened neural plates, and broader notochords are consistent with deficient convergent extension.
NTC rostral and caudal to closure point 1 occur by different mechanisms. Unlike the spinal region where expansion of paraxial mesoderm is not required for elevation and subsequent NTC, cranial NTC is initiated by expansion of underlying mesenchyme [30]. Mesenchyme density, including cranial mesenchyme, was reduced in Oct4COND MUT (Fig. 2E–G; Table S1D; F1,13 = 54.59, p<0.05 2-way ANOVA, *p<0.05, ***p<0.001 Bonferroni posttest). Hence expansion of cranial mesenchyme that is required for cranial NTC is deficient in the absence of Oct4.
A requirement for Oct4 in extraembryonic tissue offers one possible explanation for the Oct4COND MUT phenotype: ∼E7.5 Oct4 is present in extraembryonic mesoderm, allantoic angioblasts as well as extraembryonic endoderm which promotes proliferation and organization of the primitive streak [6], [31]. To test this possibility, Oct4+/+ Red fluorescent protein positive (RFP+) ES cells were aggregated with tetraploid Oct4f/f;Z/EG+/−;CreERT2+/− embryos, where ES cells contribute to the embryo, and tetraploid cells generate trophectoderm and visceral endoderm [32]. In this scheme, tamoxifen administration will selectively remove of Oct4 from the tetraploid extraembryonic lineages. Tetraploid Oct4f/f;Z/EG+/−;CreERT2+/− embryos induced ∼E6.5 and ∼E7.0 supported development of WT ES-derived embryos to E9.5 (Fig. 3A–C,E; Table S1Q). Embryos were dosed on this relatively early schedule to avoid false negatives that might result from altered timing of development associated with transferring embryos to pseudopregnant mothers. In practice transferred embryos synchronize with the maternal uterine environment [33], suggesting false negatives for this reason are unlikely. Normal embryonic development after excision of Pou5f1 in trophectoderm and visceral endoderm suggests Oct4 is required in embryonic tissue.
To identify non-autonomous effects of Oct4 depletion, we tested whether lineage-specific removal of Oct4 affected development of other tissues. Since Oct4 is present in the primitive streak, neuroepithelium and portions of mesoderm ∼E7.5 as well as mosaically in definitive endoderm (Fig. S1 and S2), a primary effect in one of these lineages might non-autonomously cause other aspects of the Oct4COND MUT phenotype [6]. To test this possibility, Oct4 was removed in the neuroepithelium using Sox1-Cre, which is expressed and catalytically active from ∼E7.5 [34]; in definitive endoderm using tamoxifen-inducible Foxa2mcm, which is expressed ∼E6.25 [35]; as well as in embryonic mesoderm using Brachyury (Bry)-Cre, which is expressed and catalytically active from ∼E6.25 [36].
Excision of Pou5f1 by lineage-specific recombinases (Bry-Cre, Sox1-Cre or Foxa2mcm) did not result in a phenotype or impact embryonic viability at E9.5. Oct4f/f; Z/EG+/; lineage-specific Cre+/− embryos should reveal aspects of the Oct4COND MUT phenotype related to requirements for Oct4 within their respective expression domains or cause the embryo to resorb by E9.5 if development is more severely impacted than in Oct4COND MUT embryos. Recombination at the lacZ/enhanced GFP (Z/EG) locus yields GFP expression, so the Z/EG allele was incorporated to gauge recombination efficiency [37]. Based on the parental genotypes used in the cross (Table S1R–T), a genotypic ratio where Oct4f/f; Z/EG+/−; lineage-specific Cre+/− embryos comprise ¼ of the progeny is expected if this genotype, where lineage-specific excision of Pou5f1 occurs, does not impact viability. Such embryos with no phenotype comprised ¼ of each litter (Table S1R–T). To test whether the lineage-specific recombinases yielded false negative results due to infrequent biallelic excision, we assessed the development of embryos where one Pou5f1 allele was removed prior to recombinase expression. Even with this sensitized approach, Oct4Δ/f; Z/EG+/−; lineage-specific Cre+/− embryos with no phenotype comprised ¼ of the progeny at E9.5. This genotypic ratio indicates that excision of Pou5f1 by these lineage-specific recombinases did not impact viability (Table S1U,V).
Since false-negatives may arise due to low recombination efficiency in this scheme, we used the GFP expression resulting from recombination at the Z/EG locus in Oct4f/+; Z/EG+/−; lineage-specific Cre+/− embryos as a proxy for recombination efficiency. By E9.0 Sox1-Cre and Bry1-Cre induced >95% and >51% recombination within their respective domains (Fig. S7A–C; Table S1W–Y), while Foxa2mcm yielded <5% (data not shown). However, prior to E8.0 when embryos are sensitive to Oct4 depletion, Sox1-Cre and Bry-Cre also yielded <5% recombination (Fig. S7C; Table S1Z,AA) [30]. Notably, the distribution of Oct4Δ/f; Z/EG+/−; Bry-Cre+/− cells did not appear altered ∼E9.5 (Fig. S7D,E), suggesting that any effect Oct4 has on cell fate either coincides with lineage specification or precedes it.
To investigate how recombination frequency influences phenotype penetrance in embryos where Pou5f1 is removed by lineage-specific recombinases, we generated diploid chimeras by aggregating WT and Oct4f/f;HisGFP+/−;CreERT2+/− morulas. The ubiquitously expressed fusion protein ‘HisGFP,’ which is comprised of histone H2B and eGFP was used to mark transgenic cells [38]. Following tamoxifen administration ∼E6.5 and ∼E7.0, we recovered 16 chimeras where contribution by Oct4f/f;HisGFP+/−;CreERT2+/− morulas ranged from 20–60% (Table S1AB). 11 of these 16 embryos had no phenotype, while the remaining 5 chimeras had rostral NTC deficits (Fig. 3D,E). This indicates that Oct4+/+ cells rescue the developmental deficiencies caused by Oct4−/− cells in mosaic embryos. Since efficient depletion of Oct4 is required for the Oct4COND MUT phenotype, the inefficient recombination of Bry-Cre, Sox1-Cre and Foxa2mcm during the window of development in which embryos are sensitive to Oct4 depletion does not resolve whether Oct4 is ubiquitously required ∼E7.5, required only in unspecified progenitors, or necessary in a subset of specified lineages, such as in specified Oct4+Bry+ mesoderm.
Since this data suggested that differences in the kinetics of Pou5f1 excision with lineage-specific recombinases and CreERT2 (when tamoxifen is administered ∼E7.0) are responsible for the absence and presence of phenotypes following Pou5f1 excision, we tested whether expansion of specified lineages was affected in Oct4COND MUT embryos. Lineage-specified Bry+ and Sox2+ cells were present 48 hrs ATA in Oct4f/f;CreERT2+/− embryos (Fig. S8A,B; Table S1AC). We quantified the fraction of phosphorylated Histone H3 (PH3)+ cells in specified lineages. The PH3+ fraction of neural or mesoderm cells (Oct4f/f;CreERT2+/− versus Oct4f/f) was the same (Fig. S8C, Table S1AC). The data indicate that expansion of these specified lineages is not impacted by Oct4 depletion.
To test whether disruption of the pluripotency network causes the Oct4COND MUT phenotype, we removed Sox2 using the same conditional approach [39]. Sox2 is a core component of the pluripotency network that complexes with Oct4, co-occupies many genomic sites (Oct4/Sox2) and is required for maintenance of Pou5f1 expression in ES cells. ES cells differentiate into trophectoderm when Sox2 is removed [40], however the ability of Oct4 over-expression to rescue pluripotency in these cells suggests that the critical role of Sox2 in pluripotency is to maintain Pou5f1 expression [40]. Sox2 null embryos lack epithelial cells typical of the epiblast and have a later extraembryonic defect which does not permit development past E7.5 [41]. Following tamoxifen administration ∼E6.5 and ∼E7.0 to Sox2f/f;CreERT2+/− embryos [39], hydrocephalus was evident in 11/20 Sox2f/f;CreERT2+/− and 2/20 others had kinked neural tubes ∼E9.5 (Fig. 4A–C; Table S1AD). Thus Sox2 removal did not phenocopy Oct4 depletion ∼E7.5. These data do not rule out partial compensation for loss of Sox2 by redundant factors, however between E7.0–E8.0 Oct4 and Sox2 only overlap spatially in anterior neuroepithelium (compare Figure S1, S2 and S9) [6], [41]. The distinct phenotypes produced by depletion of Sox2 and Pou5f1 indicate that at least part of their functions do not overlap ∼E7.0–E8.0, in contrast to ES cells.
Oct4 is reported to bind 784–4234 genomic loci in ES cells depending on the methodology used to map binding sites [5], [42], [43]. To determine which targets might be contributing to the Oct4COND MUT phenotype, we measured gene expression changes that occurred coincident with Oct4 depletion (∼E7.5) and thereafter (∼E8.0 and ∼E8.5). Oct4f/f;CreERT2+/− embryos were separated from Oct4f/f littermates by genotyping extraembryonic tissue, and differential expression assessed within litters with ≥3 CreERT2+/− and ≥3 CreERT2−/− embryos (Table S1AE). RNA was extracted 24, 36 and 48 hrs ATA, when Oct4 transcript abundance in CreERT2+/− embryos is <5% CreERT2−/− littermates (Fig. S5A–D). 754 unique genes were differentially expressed (p<0.01) at one or more of these three timepoints.
To determine whether the differential expression following Oct4 depletion was a direct consequence of Oct4 loss at its genomic targets, we assessed whether Oct4's direct targets were enriched amongst up- or down-regulated genes as Oct4 both activates and represses transcription [28]. Systematic mapping of TF targets in early embryos is currently prohibitive [44], so a genome-wide binding map of Oct4 in ES cells was used [5]. This particular genomic binding map, which is based on ChIP-seq data, was used because it offers more complete genomic coverage than target maps based on ChIP-chip data, and also contained the most extensive set of other TF binding maps for additional analysis (alternatives include: [42], [43]).
Enrichment of TF binding targets from ES cells amongst differentially expressed genes after ∼E7.5 requires that binding sites be conserved between these stages. Oct4 binding sites from ES cells were enriched amongst up-regulated genes (Fig. 5B), supporting conservation of the binding sites between ES and ∼E7.5–E8.5 embryos. Oct4 binding targets were also enriched when alternative datasets were analyzed. For comparison, with the aggregate of differentially expressed genes (24, 36 and 48 hrs ATA), enrichment using hypergeometric tests were: p = 3.45E-11 [43], p = 2.13E-08 [5], and p = 7.36E-4 [42]. This suggest that the expression changes at these sites were a direct consequence of Oct4-mediated transcriptional regulation being removed after ∼E7.5.
Oct4 targets whose transcription is regulated by Oct4 in ES cells were differentially expressed coincident with Oct4 depletion ∼E7.5. Lefty1 and Klf2 that are activated by Oct4 in ES cells decreased [45], [46], while Xist was notable among the most up-regulated genes following Oct4 depletion as it is repressed by Oct4 in ES cells [41]. An unbalanced male∶female ratio in the intra-litter comparisons, rather than Oct4 depletion, might explain the increase in Xist transcript abundance since embryos were not sexed in the microarray, however Quantitative (Q)-PCR on independent balanced comparisons confirmed that the increase related to Oct4 depletion. An intra-litter comparisons to match developmental stage, and inter-litter comparisons to reduce biological variance associated with comparing a small number of embryos both supported Oct4-mediated repression of Xist ∼E7.5: Xist was 3.20 times more abundant in the intra-litter comparison, and 2.85±0.76 s.e.m. more abundant in the inter-litter comparison of Oct4f/f;CreERT2+/−/Oct4f/f 24 hrs ATA (Table S1AF).
Enrichment for genomic targets of Oct4 is expected with this approach, but transcriptional activators of Oct4 and proteins that physically interact with it were also differentially expressed. Ligands that maintain Oct4 such as Nodal and Wnt3a [11], [47] exhibit decreased transcript abundance coincident with Oct4 depletion ∼E7.5, while transcriptional activators of Oct4 such as Sp1 [48] and Ago2 [49] exhibited increased transcript abundance, perhaps due to a feedback loop. Proteins that physically interact with Oct4 were also enriched amongst the genes up-regulated following Oct4 depletion (see Table S2 for cofactor identities; p = 1.99E-08 24 hrs ATA, p = 1.64E-05 36 hrs ATA, p = 5.55E-07 48 hrs ATA enrichment using hypergeometric tests). Interestingly, we found considerable enrichment for Oct4 within genomic regulatory elements of these physical cofactors (p = 5.34E-07 for 24,36 and 48 hrs ATA collectively using a hypergeometric test). This suggests that ∼E7.5 Oct4 directly represses expression of a subset of the genes it physically interacts with in ES cells and that its absence triggers positive indirect feedback of the expression of others. Collectively, these data suggest that several regulatory relationships of Oct4 are maintained between preimplantation development and ∼E7.5–8.5.
To test whether signaling networks other than direct targets of Oct4 might contribute to the Oct4COND MUT phenotype, we determined the transcriptional response that target sets bound by TFs other than Oct4 had to Oct4 depletion. The binding maps of 12 other TFs, and combination of several with Oct4, were assessed for enrichment amongst the genes differentially expressed after Oct4 depletion (Fig. 5A) [5]. Targets of c-Myc and Smad1 were enriched amongst genes up-regulated after Oct4 depletion [5]. Unlike c-Myc, which does not cluster at binding sites with Oct4 in the genome, Oct4 facilitates the binding of Smad1 such that they overlap at a subset of sites [5]. However up-regulation of Smad1 targets after Oct4 depletion occurred at sites Smad1 occupies independent of Oct4, indicating that enrichment of up-regulated Smad1 targets is not due to direct relief of Oct4-mediated repression at sites that the two co-occupy [5]. The enrichment of Smad1 targets amongst up-regulated genes that are not co-occupied by Oct4 are: p = 6.14E-06 24 hr ATA, p = 4.55E-03 36 hr ATA, p = 3.53E-09 48 hr ATA (hypergeometric test). Like Oct4, Smad1 has been implicated in both activation and repression of target genes [50], consistent with a separate subset of Smad1 targets are de-repressed 24 hrs ATA. These data suggest that the absence of Oct4 yields a transcriptional environment conducive to target activation by c-Myc and Smad1. Conversely, enrichment of co-occupied Oct4/Sox2 target sites amongst down-regulated genes (Fig. 5C) suggests that Oct4 participates in transcriptional activation of these ∼E7.5 and after. Since conditional removal of Sox2 and Pou5f1 do not phenocopy (compare Figure 1A to 4A), Sox2 is either not essential for activation of these sites, which is consistent with data from ES cells [40], or down-regulation of these targets does not contribute to the Oct4COND MUT phenotype.
Oct4 binds thousands of sites in the genome, and it is unlikely that disruption of a single target gene causes the Oct4COND MUT phenotype. To relate molecular changes resulting from Oct4 depletion with the Oct4COND MUT phenotype, we determined which signaling pathways were disrupted coincident with Oct4 depletion and prior to the onset of the phenotype. Unsupervised clustering was used to assess the function of differentially expressed genes collectively. To discern primary effects of Oct4 depletion, we sub-setted for genes that are direct targets of Oct4 based on the ES binding maps [5], clustered these (Fig. 6A; Table S1AE), and then compared the clusters to global changes (Fig. 6B; Table S1AE). 3 of the 4 pathways showing the strongest enrichment in the set of direct targets also showed significant enrichment in the global set. Coordinate regulation of additional genes that are not targets of Oct4 within the same pathways as those directly regulated by Oct4, suggests amplification of the direct effects (Fig. 6A,B; Table S1AE). QPCR on independent biological samples confirmed a subset of changes from the global expression analysis (Fig. 6C, Table S1AG), supporting the reproducibility of the differential expression.
Differential expression was then considered in relation to the Oct4COND MUT phenotype. The expression profiling suggested that decreased TGF-β signaling and increased nuclear import of NF-κB were primary effects as they occurred within hours of Oct4 depletion (24 hrs ATA) amongst direct targets of Oct4, while decreased Notch signaling and increased protein translation are other candidates that occurred later (Fig. 5A).
The node is required to coordinate left-right asymmetry, specification of definitive endoderm and somitogenesis [51]. Given these roles in development, we considered the possibility that Oct4 was required in node formation a candidate that might explain the situs inversus, defective somitogenesis and the posterior truncation (via either endoderm specification or defective somitogenesis) observed in Oct4COND MUT embryos. Gene expression changes following Oct4 depletion also suggested the possibility of node malformation: decreased Dll1 contributed to the ‘Notch signaling’ enrichment in the microarray and was confirmed by QPCR in separate litters (Fig. 6C; Table S1AG). Decreased Dll1 following Oct4 depletion is relevant because loss of Dll1 was previously shown to disrupt node formation and cause defects in left/right asymmetry [52]. While these data were suggestive of a candidate mechanism underlying the Oct4COND MUT phenotype, the presence and appropriate localization of the node marker Chordin both 24 hrs ATA (Fig. 7A,B, Table S1AC) and 36 hrs ATA (Fig. S10A,B; Table S1AC) suggests that initial node specification occurs in Oct4COND MUT [53]. The disruption of left-right asymmetry is likely downstream of node specification, as transcript abundance of laterality specifiers that are asymmetrically distributed by the node during development is altered: Nodal, Dll1, Lefty1 and Lefty2 are decreased while Hand1 and Hand2 are increased. These data do not support the Oct4COND MUT phenotype being caused by a failure in Notch-mediated node specification.
Contraction of actin-myosin microfilaments contributes to the morphogenetic processes of turning and convergent extension. A decrease in ‘actin filaments’ (p = 1.88E-07) following Oct4 depletion (Fig. 6B; Table S1AE) suggests that actin networks are affected by Oct4 depletion. The distribution of actin appeared altered 24 hrs ATA with phalloidin staining (Fig. S10C,D; Table S1AC). Indeed the distribution of actin in Oct4f/f;CreERT2+/− embryos suggests that adhesion between anterior and posterior neuroepithelium in the distal portion of the embryo may contribute to thicker neuroepithelium in this regions and impaired embryonic morphogenesis.
TGF-β signaling has also been implicated in several processes disrupted in Oct4COND MUT embryos: expansion of primitive streak [54], patterning derivatives of the anterior primitive streak [55], establishment of definitive endoderm [56], maturation of the node [57] and left/right asymmetry establishment [58], [59]. Unsupervised clustering indicates that Oct4 directly maintains TGF-β signaling (Fig. 6A). TGF-β signaling through Smad2 competes with Smad1 for the co-activator Smad4 [60], so up-regulation of Smad1 targets following Oct4 depletion may involve an increase in Smad1, expansion of the domain of activated phosphorylated-Smad1 (p-Smad1), or diminished competition from TGF-β-Smad2. Increased transcript abundance of Smad1 was confirmed by Q-PCR (Fig. 5C; Table S1AG). The p-Smad1 domain also appears altered 24 hrs ATA (Fig. 7C,D; Table S1AC). Variance in p-Smad1 introduced by differences in embryonic stage and ‘batch effects’ during detection prohibited making a statistically meaningful quantitative comparison of protein abundance between stage-matched Oct4f/f; CreERT2+/− and Oct4f/f embryos. Quantitative comparison with high-content image analysis software did suggest a difference in p-Smad1 abundance related to Oct4 depletion (Fig. S11), but this approach would require a considerable increase in sample size to test significance. These data suggest a direct effect of Oct4 depletion on diminished TGF-β signaling.
Presence of Oct4 in the primitive streak ∼E7.5 (Fig. S1), impaired axial extension in Oct4COND MUT embryos and differential expression of TGF-β signaling that is essential for expansion of primitive streak [54] suggested an effect on its expansion. An effect on the primitive streak and consequently its derivatives might have broad relevance: cranial mesenchyme supports NTC, while mesendoderm facilitates posterior extension, somitogenesis and turning. The frequency of cells undergoing apoptosis (Caspase-3+) in the Oct4COND MUT was increased (Fig. 7I; Table S1AC), suggesting that diminished cell viability might contribute to the phenotype. Notably, the distribution of apoptotic cells throughout the embryo, including regions where Oct4 is not expressed, suggests that some apoptosis may be a secondary defect. Conversely, fewer cells proliferated indicated by phosphorylated histone H3 positive (PH3+) in the primitive streak of embryos 24 hrs ATA (Fig. 7G,H,J; Table S1AC). To confirm the localization of these effects, we divided embryos into three segments (proximal anterior, distal and proximal posterior) and quantified the abundance of transcripts regulating apoptosis and proliferation. To obtain sufficient material for comparison, CreERT2+/−;Oct4f/f samples 24 hrs ATA were compared to CreERT2+/−;Oct4f/f stage-matched samples from separate litters. While there was no difference in the transcript abundance of apoptosis regulators Bax and Bcl2, a negative regulator of proliferation, Cdkn1c, which exhibited increased transcript abundance in the differential expression analysis was selectively increased in the posterior third of embryos coincident with the loss of Oct4 (Fig. 7K; Table S1AH). These data indicate that ubiquitous Oct4 depletion leads to increased apoptosis and deficient proliferation in the primitive streak.
∼E7.5, Oct4 is still present in the primitive streak, posterior visceral endoderm, several mesoderm derivatives, neuroepithelium as well as extraembryonic endoderm and mesoderm (Fig. S1) [6]. Proliferation of the primitive streak decreases and apoptosis increases within the embryo coincident with Oct4 depletion ∼E7.5, and by ∼E9.5 several morphogenetic processes are disrupted: turning, posterior extension, laterality and NTC all are affected, demonstrating that Oct4 is required for somatic development after implantation.
Reduced proliferation in the primitive streak coincident with Oct4 depletion suggests that Oct4 might maintain potency ∼E7.5 as it does in the ICM [1]. EpiSC-derivation and teratoma assays support the persistence of pluripotent somatic cells ∼E8.0, while lineage tracing indicates the presence of neuro-mesodermal progenitors ∼E8.0 [61]. However excision of pluripotency factors Sox2 and Oct4 ∼E7.0 do not phenocopy as their depletion in ES cells do [1], [40], indicating that the pluripotency network is altered between the ICM and ∼E7.5. Differences in localization contribute: at the latest stage embryos are sensitive to Oct4 depletion and a proliferation deficit is evident in the primitive streak of Oct4COND MUT embryos (∼E7.5), Sox2 transcript is limited to the chorion and anterior neuroectoderm (Fig. S9) [41]. Neural-specific Sox2 excision results in enlarged lateral ventricles ∼E19.5 due to decreased proliferation of neural stem and progenitor cells [62], suggesting that hydrocephalus in Sox2COND MUT embryos may result from insufficient expansion/thickening of the neuroepithelium. This might render the neuroepithelium more elastic and distended as a result of the positive fluid pressure in the neural lumen [63], or precede the collapse or kinking of neural tubes that infrequently occurred. The differing phenotypes following depletion ∼E7.5 indicate that Sox2 is not required for Pou5f1 transcription or as a cofactor in the processes disrupted in Oct4COND MUT embryos.
Oct4 promotes mesoderm as opposed to neural fate during ES differentiation [20], as does XlPou91 (the paralog in X. laevis) in response to FGF [64], [65], suggesting that Oct4 depletion might divert mesoderm to neural tissue. Decreased expression of Tbx6 [66] and Wnt3a [67] whose loss is associated with diversion to ectopic neural tubes from paraxial mesoderm following Oct4 depletion is consistent with this possibility, as is thicker neuroepithelium of Oct4COND MUT embryos near closure point 1. However this differential expression may not reflect altered specification per se, but altered proportions of the embryo associated with defective axial extension. Similarly, neuroepithelial thickening unrelated to cell fate divergence is common amongst mutants with NTC defects such that this is not a reliable indicator of fate changes [30]. Finally, the distribution of Oct4Δ/f; Z/EG+/−; Bry-Cre+/− cells did not appear altered. This suggests that any effect Oct4 has on cell fate either coincides with lineage specification or precedes it.
An alternative to an effect on cell fate specification is that Oct4 promotes expansion of unspecified progenitors by driving the cell cycle. Reduced mesenchyme density, decreased proliferation in the primitive streak, increased Trp53 (p53) expression and increased Cdkn1c expression in the Oct4COND MUT embryonic posterior all indicate that expansion of posterior progenitors is disrupted when Oct4 is depleted. The G1/S transition is effectively absent from ES cells, and binding of Oct4 to micro-RNAs that suppress inhibitors of the G1/S transition [68] may promote its bypass and limit the window for lineage-specific chromatin remodeling. Indeed, genes regulating ‘chromatin modification’ are up-regulated 24 hrs ATA coincident with reduced proliferation in the primitive streak (cluster 1–295: p = 2.1E-04 and cluster 613–908: p = 3.7E-04 using hypergeometric tests). Finally, c-Myc activates G1/S checkpoint complexes [69], [70], suggesting that c-Myc may be required to promote G1/S transition when the G1/S checkpoint is established coincident with Oct4 depletion.
Morrison and Brickman proposed that the evolutionarily conserved role of Oct4 might be facilitating expansion of progenitor populations during and after gastrulation based on work with paralogs: Pou2 in D. rerio and XlPou91 in X. laevis [64]. These D. rerio Pou2 mutants [71] and X. laevis embryos treated with morpholinos against XlPou91 share posterior truncations [64]. Since Pou5f1 arose by duplication of Pou2 [64], these data support a conserved role for Oct4 in posterior extension, which in mice includes maintaining proliferation in the primitive streak.
All procedures were approved by the University of Toronto Animal Care Committee in accordance with the Canadian Council on Animal Care. Foremost, both euthanasia and surgery were minimized. When performed, stress was minimized to the greatest extent possible before rapid depressive action on the CNS during euthanasia. Minimally invasive surgeries were performed under anesthetic to achieve complete depression of feedback from the PNS and analgesic used for recovery. For staging, embryos were assumed to be 0.5 days post coitum at 1pm on the day a vaginal plug was found. This is 12 hrs after the midpoint of the 14 hr light/10 hr dark cycle we used, where the lights were shut off every night at 8 pm and came on every morning at 6 am. Given the relevance of staging to this set of experiments, it is important to note that use of vaginal plugs –as opposed to direct observation of conception– is accompanied by ±7 hrs of variability in embryonic staging and is inferred from the midpoint of the dark period in the light/dark cycle. Embryos were dissected in Dulbecco's PBS (Gibco) and immediately placed in either liquid nitrogen (for microarrays and QPCR analysis) or in 4% paraformaldehyde (for sectioning and immunohistochemistry). Dissections for embryonic stages that are whole numbers (e.g. E8.0 or E9.0) were performed between 9 and 11 pm, while those occurring 12 hrs apart from whole days post coitum (e.g. E9.5 or E10.5) were performed between 12 and 2 pm. For the experiments assessing the timeframe of Oct4 depletion (Fig. 1D–G, S5A–D), tamoxifen was administered at 9 pm±30 min, and dissections performed the indicated number of hours ATA, e.g. dissections for the time-point 3 hrs ATA were done at midnight (12 am). The following stocks were used in the study: CD1 (Charles River), Oct4f/f [25], lacZ/eGFP (Z/EG) [37], B6.Cg-Tg(Hist1H2BB/Egfp)1Pa/J (Histone H2B/eGFP fusion ‘HisGFP’) [38], Bry-Cre [36], Sox1-Cre [34], Foxa2tm2.1(cre/Esr1*)Moon/J [35], Sox2f [39], CreERT2 [29]. Individual embryos or the associated extraembryonic tissues were genotyped as originally described.
Because a variety of experimental permutations were used in this project, the details of each permutation, including the mouse strains, genotypic ratios, tamoxifen administration regimen and other relevant features are provided on a separate row in Table S1 (the relevant row is noted as the experiment is described where ‘S1, row A’ is ‘S1A’).
Tamoxifen was administered according to the protocol optimized following CreERT2 development [29]. 99 mg of tamoxifen (Sigma) was dissolved by sonication in a solution of 100 ul of ethanol (Sigma) and 1 ml of peanut seed oil (Sigma) [29]. The solution was kept in a ∼50°C water bath during preparation and prior to administration to avoid precipitation. 50 µl doses of this solution were administered to pregnant mothers by oral gavage using a 250 µl gastight #1725 syringe (Hamilton) [29]. Because of the uncertainty associated with staging embryos with vaginal plugs (±7 hrs), the time-point(s) indicated for tamoxifen administration are approximations, and listed as such (∼) within the text to reflect this uncertainty. In practice, tamoxifen was given at 9pm±30 min (∼E6.0, ∼E7.0 or ∼E8.0) or 9 am±30 min (∼E6.5, ∼E7.5 or ∼E8.5). The time-point(s) when tamoxifen was administered for each experimental permutation are listed in Table S1 as well as in the figure captions.
The density of mesenchyme, frequency of apoptosis and proliferation, relative abundance of transcripts (other than Oct4), distance between neural folds and thickness of neuroepithelium were compared using 2-way ANOVAs. Depletion of Oct4 protein and transcript were compared with 1-way ANOVAs. F-values from the embryonic genotype's contribution (Oct4f/f versus Oct4f/f;CreERT2+/−) to variation are indicated except for Figure 7K and S6C where the intra-embryo segment contribution is reported (e.g. difference between segments in the same embryo). Binding enrichment amongst differentially expressed genes and common causality of disrupted features in partially penetrant Oct4COND MUT embryos was assessed using hypergeometric tests. The thickness of notochords was compared using a two-tailed t-test. A threshold of p<0.05 was used for each test (ANOVA, hypergeometric and t-test). Please see the Supplementary Methods (‘Text S1, page 1’) for detail on how measurements of Oct4 protein depletion, mesenchyme density, neuroepithelium thickness, notochord thickness, distance between the neural folds, and the fraction of Ph3+, Caspase-3+ and Oct4+ cells were taken (‘Basic Measurements’).
Images in Figure 1F,G; Figure 2 F,G,I,J; Figure 3A–D; Fig 7A–J; Figure S6A,B,D,E; Figure S7A,B,D,E; Figure S8A,B and Figure S10A–F were taken with a Zeiss Axio Observer, images of Figure S5A–D were taken with an Olympus Fluoview 1000, images of Figure 2 B,C and Figure 4A–C were taken with an Olympus SZ61, and images of Figure 1A–C; Figure 3SA,B and Figure S4A were taken with a Leica MZ16 FA stereomicroscope. Contrast of the images in Figure 3D, 4A and 4C was enhanced with Adobe Photoshop v12.
Oct4 staining was performed as described previously [6]. For all other immunohistochemistry, embryos were fixed in 4% PFA overnight at 4°C, sectioned at a thickness of 10 µm and primary antibodies applied overnight at 4°C at the following concentrations: Oct-3/4 1∶200 (C-10 Santa Cruz), Chordin 1∶100 (R & D Systems), p-Smad1 1∶400 (Cell Signaling), Caspase-3 1∶500 (Promega), Ph3 1∶500 (Cell Signaling), Bry 1∶50 (R & D Systems), Sox2 1∶50 (R & D Systems). An antigen retrieval step of boiling the sample in 10 mM Sodium Citrate Buffer, pH 6.0 for 15 min was used for Oct-3/4 (C-10 immunofluorescent) and Chordin staining. Phalloidin staining (Alexa Fluor, Life Technologies) was performed according to the manufacturer's instructions. Hematoxylin and Eosin (Sigma) staining was performed according to the manufacturer's instructions.
Different litters from those used in the microarray analysis were used to confirm changes in gene expression by QPCR. Please see Supplementary Methods (‘Text S1, page 2) for assay details.
Chimeras were produced as outlined in [72], and contribution was assessed by semi-quantitative PCR. Please see Supplementary Methods (‘Text S1,’ page 2) for details.
RNA was extracted with Trizol according to the manufacturer's instructions (Invitrogen) and sent to the UHN Microarray Centre (Toronto, ON, Canada) for fluor-labeling (protocol GE2 v5.7), microarray hybridization, and array scanning. Please see Supplementary Methods (‘Text S1,’ page 4) for additional detail and analysis methodology.
Please see ‘Text S1.’
Please see ‘Text S1.’
Please see ‘Text S1.’
Please see ‘Text S1.’
Please see ‘Text S1.’
Please see ‘Text S1.’
|
10.1371/journal.pntd.0006992 | Basophils are dispensable for the establishment of protective adaptive immunity against primary and challenge infection with the intestinal helminth parasite Strongyloides ratti | Infections with helminth parasites are controlled by a concerted action of innate and adaptive effector cells in the frame of a type 2 immune response. Basophils are innate effector cells that may also contribute to the initiation and amplification of adaptive immune responses. Here, we use constitutively basophil-deficient Mcpt8-Cre mice to analyze the impact of basophils during initiation and execution of the protective type 2 responses to both, a primary infection and a challenge infection of immune mice with the helminth parasite Strongyloides ratti. Basophil numbers expanded during parasite infection in blood and mesenteric lymph nodes. Basophil deficiency significantly elevated intestinal parasite numbers and fecal release of eggs and larvae during a primary infection. However, basophils were neither required for the initiation of a S. ratti-specific cellular and humoral type 2 immune response nor for the efficient protection against a challenge infection. Production of Th2 cytokines, IgG1 and IgE as well as mast cell activation were not reduced in basophil-deficient Mcpt8-Cre mice compared to basophil-competent Mcpt8-WT littermates. In addition, a challenge infection of immune basophil-deficient and WT mice resulted in a comparable reduction of tissue migrating larvae, parasites in the intestine and fecal release of eggs and L1 compared to mice infected for the first time. We have shown previously that S. ratti infection induced expansion of Foxp3+ regulatory T cells that interfered with efficient parasite expulsion. Here we show that depletion of regulatory T cells reduced intestinal parasite burden also in absence of basophils. Thus basophils were not targeted specifically by S. ratti-mediated immune evasive mechanisms. Our collective data rather suggests that basophils are non-redundant innate effector cells during murine Strongyloides infections that contribute to the early control of intestinal parasite burden.
| Helminths are large multicellular parasites that infect approximately every third person. Infections are controlled by a concerted action of innate and adaptive immune responses. Basophils and mast cells are innate effector cells with overlapping functions that have recently been implicated in the initiation and promotion of adaptive immune responses. Here, we analyze the function of basophils during murine infection with Strongyloides ratti, an intestinal parasite with tissue migrating stages. Using mice constitutively lacking basophils, we show that their most important function is the early control of intestinal parasite burden during a primary infection. Basophils are not needed for efficient killing of tissue migrating S. ratti larvae, the timely termination of infection or the initiation and execution of protective immunity against S. ratti challenge infections in immune mice. Since basophils were shown to be important for the immunity against different helminth parasites in intestine or tissue (Heligmosomoides polygyrus and Nippostrongylus brasiliensis), their function has to be evaluated for every parasite species individually.
| Approximately one third of the human population is infected with parasitic helminths. Strongyloides ratti is a rodent-specific helminth parasite that can be used as a model parasite to study the immune response against intestinal helminth infections with tissue migrating stages in the mouse system [1]. Infective third stage larvae (L3i) actively penetrate the skin of the mammalian host, migrate within 2 days via the tissue and partially the lung towards the nasofrontal region of the host. They are swallowed by day 2–3 post infection (p.i.) and reach their final destination, the small intestine, where they molt twice to become parasitic female adults that live embedded in the mucosa of the intestine. The females reproduce via parthenogenesis by day 5–6 p.i. and release eggs and already hatched first stage larvae (L1) with the feces into the environment. Immune competent mice terminate the infection within a month and remain semi-resistant to subsequent infections [2, 3]. We have previously shown that the early expulsion of parasitic adults from the small intestine is predominantly mediated by mucosal mast cells [4]. Thereby, mast cells represented terminal effector cells that were not involved in killing of tissue-migrating L3 but were indispensable for the final expulsion of S. ratti during a primary infection. However, the generation of the adaptive immune response that mediates partial protection against a challenge infection of immune mice was readily established in the absence of mast cells.
Basophils are a rare population of late effector cells that are phenotypically and functionally related to mast cells. They arise from a common precursor and share a set of effector mediators that are stored in vesicles ready to be released [5]. In contrast to mast cells that differentiate and reside in the tissue, basophils leave the bone marrow as mature cells, are predominantly found in the blood stream and spleen and have a relatively short lifespan of about 60 h under steady-state conditions [6]. After exposure to stimuli such as allergens or parasitic helminth infections, basophils can be recruited to the site of inflammation. We and others have shown that, although clearly contributing to the early expulsion of Strongyloides adults from the intestine, basophils play only a minor role in the final control and termination of infection [4, 7]. Also intestinal parasite burden during a primary infection with Nippostrongylus brasiliensis, Heligmosomoides polygyrus, Trichinella spiralis and Litomosoides sigmodontis were not affected by basophil deficiency or depletion [8–11]. However, several lines of evidence suggest a contribution of basophils in the generation and execution of the protective immune response during a secondary infection with N. brasiliensis and H. polygyrus [6, 9, 10, 12, 13]. Thereby, possible functions of basophils in the polarization of Th2 immune responses, the early production of type 2 polarizing cytokines and antigen presentation are being debated [14, 15].
Here, we analyze the role of basophils in the protective immune response to a S. ratti challenge infection in immunized mice. Although basophils expanded during the course of infection, a protective adaptive immune response was readily established in the absence of basophils. Cytokine and antibody responses and subsequent mast cell activation were unchanged in basophil-deficient and basophil-competent mice. Efficient killing of migrating larvae as well as rapid expulsion of intestinal parasites during a challenge infection of immune mice was not impaired by basophil deficiency, thus ruling out a non-redundant role for basophils in the initiation or execution of a protective memory response to S. ratti infection.
Animal experiments were conducted in agreement with the German animal protection law and experimental protocols were approved by Federal Health Authorities of the State of Hamburg. BALB/c Mcpt8-Cre mice [10] and DEREG mice [16] have been described previously and were bred heterozygously. Wistar rats were obtained from Janvier Labs (Le Genest-Saint-Isle, France). Heterozygous BALB/c DEREG mice were intercrossed with heterozygous BALB/c Mcpt8-Cre mice in the animal facilities of the Bernhard Nocht Institute for Tropical Medicine to provide littermates for the experiments. All mice were bred in house and kept in individually ventilated cages under specific pathogen-free conditions. For all experiments, male and female mice were used at 7 to 10 weeks of age.
The S. ratti cycle was maintained in Wistar rats and infections were performed by s.c. infection of 2000 L3i in the hind footpad of mice [2]. Mice were vaccinated with 2000 irradiated L3i (160 Gy) 4 weeks before challenge infection with 2000 viable L3i as described [4]. For Treg depletion, groups of mice received 0.5 μg DT (Merck, Darmstadt, Germany) dissolved in PBS (pH 7.4) i.p. on three consecutive days, starting one day prior to S. ratti infection. Treg depletion was controlled by analysis of peripheral blood samples for GFP, Foxp3, and CD4 expression at day 1 p.i. Parasite burden in the intestine and quantification of the S. ratti 28S RNA-coding DNA in the feces of infected mice was performed as described [17, 18].
For surface staining, cells were stained for 30 min on ice in the dark with FITC-labeled antibodies against CD4 (clone: RM4-5), CD8 (clone: 53–6.7) and CD19 cells (clone: 1D3), PerCP Cy5.5-labelled anti-mouse CD11b (clone: M1/70), PE-labelled anti-mouse IgE (clone: RME-1), Brilliant Violet 421-labelled anti-mouse CD117 (c-Kit; clone: 2B8) and PE Cy7-labelled anti-mouse CD49b (clone: DX5). Ab were purchased from BioLegend or Affymetrix eBioscience. For intracellular staining cells were permeabilized with 250 μl fixation/permeabilization buffer for 30 min at 4°C, washed with permeabilization buffer and stained with PE- or Alexa Fluor 700-labeled anti-Foxp3 (clone FJK-16s, eBiosciences, SanDiego, USA). Samples were analyzed on a LSRII Flow Cytometer (Becton Dickinson) using FlowJo software (TreeStar).
Mice were sacrificed either naïve or on day 6 p.i. A total of 2×105 spleen and mesenteric lymph node (mLN) cells were cultured in 3–5 replicates 96-well round-bottom plates in RPMI 1640 medium supplemented with 10% FCS, 20 mM HEPES, L-glutamine (2 mM), and gentamicin (50 μg/mL) at 37°C and 5% CO2. The cells were stimulated for 72 h with S. ratti antigen lysate (20 μg/mL) or anti-mouse CD3 (145-2C11, 1 μg/mL) or with medium only. The supernatant was harvested for analysis of cytokine production by ELISA. IL-3, IL-4, IL-10 and IL-13 in culture supernatants were measured using DuoSet ELISA development kits (R&D Systems, Wiesbaden, Germany), according to the manufacturer’s instructions. IL-9 detection was performed by coating with 2 μg/mL anti-IL-9 Ab (BD, Heidelberg, Germany) overnight at 4°C. Plates were blocked with 10% FCS/0.05%Tween/PBS for 2 h at RT. Samples and recombinant IL-9 standard (Peprotech, Hamburg, Germany) were incubated overnight and detection was performed with an anti-IL-9-biotin AB (BD, Heidelberg, Germany) for 1 h at RT and subsequent Streptavidin-HRP incubation for 20 min before development with 100 μL/well tetramethylbenzidine 0.1 mg/ml, 0.003% H2O2 in 100 mM NaH2PO4 (pH 5.5). The reaction was stopped after 10 min by adding 25 μL of 2 M H2SO4.
Blood was collected from day 14 infected mice and Strongyloides-specific Ig in the serum was quantified by ELISA as described [2]. Briefly, 50 μL/well S. ratti Ag lysate (2.5 μg/mL) in PBS was coated overnight at 4°C on Microlon ELISA plates (Greiner, Frickenhausen, Germany). Plates were washed four times with PBS 0.05% Tween 20 and blocked by incubation with PBS 1% BSA for 2 h at RT. Serial dilutions of sera in PBS 0.1% BSA were incubated in duplicate, adding 50 μL/well overnight at 4°C. Plates were washed five times, and Strongyloides-specific Ig was detected by incubation with 50 μL/well horseradish peroxidase (HRP)-conjugated anti-mouse IgM, anti-mouse IgG1 or anti-mouse IgG2b (Zymed Karlsruhe, Germany) for 1 h at RT. Plates were washed five times and developed by incubation with 100 μL/well tetramethylbenzidine 0.1 mg/ml, 0.003% H2O2 in 100 mM NaH2PO4 (pH 5.5) for 2.5 min. Reaction was stopped by addition of 25 μL/well 2 M H2SO4, and OD at 450 nm (OD450) was measured. The titer was defined as the highest dilution of serum that led to an OD450 above the doubled background. Background was always below 0.15 OD450. Concentration of IgE was quantified using the IgE ELISA kit (BD, Heidelberg Germany) according to the manufacturers recommendations.
For analysis mouse mast cell protease-1 (mMCPT-1), blood was collected from infected mice at the indicated time points and allowed to coagulate for 1 h at room temperature (RT). Serum was collected after centrifugation at 10,000× g for 10 min at RT and mMCPT-1was detected using the mMCPT-1 ELISA Ready-SET-Go kit (eBioscience, San Diego, USA) according to the manufacturers recommendations.
All data were assessed for normality. Groups were compared by using Mann Whitney-U test (non-parametric comparison of two groups) or Kruskal-Wallis test corrected with Dunn’s multiple comparisons test (non-parametric multiple comparisons) or 2 Way ANOVA with Bonferroni post test (parametric comparison of two groups over time), using GraphPad Prism software (San Diego). P values of ≤0.05 were considered to indicate statistical significance.
To analyze a potential role of basophils during S. ratti infection we first monitored their expansion in the blood and mLN by flow cytometry (Fig 1). Basophil expansion in the skin was not quantified in this study. Basophils were identified as lineage negative (i.e. CD19-, CD4-, CD8-), CD11b negative, and c-kit negative cells that were positive for IgE and CD49b [6] (Fig 1A). Blood basophil numbers remained at baseline levels at day 3 p.i., a time point that marks the end of the L3 tissue migration phase and at day 7 p.i. that is the peak of parasite burden in the intestine. Blood basophil numbers increased one week after the peak of intestinal parasite burden, at day 14 p.i., returned to naïve levels at the resolution of infection by day 35 p.i. (Fig 1B and 1C). In the mLN, basophils increased in frequency and numbers already at day 7 p.i., the time point of maximal worm burden, and decreased again to show a non-significant trend towards elevation at day 14 p.i followed by contraction to baseline level at clearance of infection at day 35 p.i. (Fig 1D). Thus, S. ratti infection resulted in expansion of basophils both systemically in the blood and locally in the mLN with kinetics that correlated to the presence of parasitic S. ratti adults in the intestine.
To analyze the impact of the expanding basophil population on the generation of the anti-helminth immune response, we compared the T and B cell responses in constitutively basophil-deficient Mcpt8-Cre mice and basophil-competent Mcpt8-WT mice (Fig 2).
Thereby the absence of basophils in basophil-deficient Mcpt8-Cre during the entire course of S. ratti infection was controlled by quantification of basophils as CD19, CD4, CD8, CD11b, c-kit negative and IgE, CD49b positive cells in the peripheral blood of S. ratti-infected Mcpt8-Cre and Mcpt8-WT littermates (S1 Fig).
Ex vivo re-stimulation of mLN cells derived from day 6 S. ratti infected mice with crude S. ratti antigen (Fig 2A) or anti-CD3 mAb (Fig 2B) elicited secretion of Th2 associated cytokines such as IL-13, IL-4 and IL-5 in basophil-deficient and basophil-competent mice to the same extent. Production of IL-9 and IL-3, cytokines that are central in immunity to Strongyloides [19, 20], were unchanged by basophil deficiency. Finally, the production of IL-10 that may, according to context and parasite species, contribute to protection or immune evasion [21] was not modulated by basophil deficiency. The observed cytokine production was infection-specific as mLN cells derived from naïve Mcpt8-WT and Mcpt8-Cre mice did not respond to S. ratti antigen-specific stimulation. Naïve mLN cells produced very low amounts of IL-13 and IL-10 and failed to produce IL-4, IL-5, IL-9 or IL-3 in response to polyclonal anti-CD3-mediated stimulation reflecting the expected absence of immune activation or polarization (Fig 2B). However, naïve mLN produced IL-2 in response to anti-CD3 stimulation, thus proving their general viability (Fig 2C). Also spleen cells derived from day 6 S. ratti-infected Mcpt8-WT and Mcpt8-Cre mice displayed comparable cytokine production to either CD3-engagement or S. ratti antigen-specific stimulation while unstimulated spleen cells did not secrete cytokines (S2 Fig). Absence of basophils did not impair the antibody response at day 14 p.i. (Fig 2D). The serum concentration of S. ratti-specific IgG2b was unchanged in basophil deficient Mcpt8-Cre mice. Antigen-specific IgE was not detectable in complete or in IgG1-depleted sera of S. ratti-infected mice (S3 Fig) but polyclonal IgE concentrations were not reduced in basophil deficient mice. Interestingly, serum concentrations of S. ratti-specific IgM and IgG1, the isotypes that predominantly mediate opsonization and subsequent clearance of migrating L3 in the tissues [22, 23], were even increased by trend (IgM) or statistically significant (IgG1) in basophil-deficient Mcpt8-Cre mice compared to their basophil-competent Mcpt8-WT littermates (Fig 2D).
Final expulsion of S. ratti adults from the small intestine depends on activated mucosal mast cells [4]. We measured mast cell activation by quantification of mouse mast cell protease 1 (mMCPT-1) that is specifically released by degranulating mucosal mast cells [24]. The concentration of mMCPT-1 in the serum was unchanged in basophil-deficient Mcpt8-Cre mice and their basophil-competent littermates from day 3 to day 14 p.i (Fig 2E), indicating comparable mast cell activation in the presence and absence of basophils.
In summary, we did not record a dominant impact of basophils on the initiation of helminth-specific Th2 and Th9 response.
Immune competent mice terminate Strongyloides infection within a month and remain semi-resistant to a secondary infection [22, 25]. Thereby already migrating L3 in the tissues of immune mice are potently attacked and killed and only few parasitic adults can be detected in the small intestine of immune mice. We have shown previously that vaccination with irradiated S. ratti L3i that cannot molt to parasitic adults confers the same immunity to a challenge infection as a resolved patent primary infection [4]. To evaluate the impact of basophils on protection of immune mice during challenge infection, basophil-deficient and basophil-competent mice were vaccinated with irradiated S. ratti L3 and 4 weeks later challenge infected with viable L3. Migrating L3 numbers in the tissue (head and lung) were recorded on day 2 post re-infection and parasitic adults were counted at day 6 post re-infection in the small intestine and compared to age- and gender-matched non-vaccinated mice that were infected for the first time (Fig 3). To additionally monitor the kinetic of infection we measured S. ratti-derived DNA in the feces as a rough indicator for release of eggs and first stage L1 [2]. Regarding the primary infection, basophil deficiency did not change the number of L3 in the head or lung (Fig 3A) but elevated intestinal parasite burden (Fig 3B) and fecal output of S. ratti-derived DNA (Fig 3C) as we had shown before [4]. Although intestinal parasite burden and L1 release were elevated in basophil-deficient mice during a primary infection, the kinetic of infection termination was not changed as all mice cleared the infection by day 28 p.i. (Fig 3C). Both, vaccinated basophil-deficient Mcpt8-Cre mice and vaccinated basophil-competent Mcpt8-WT mice showed drastically reduced L3 numbers in the head during challenge infection compared to their naïve counterparts that were S. ratti infected for the first time (Fig 3A). Thus, protection against the tissue migrating L3 was established in the absence of basophils. Furthermore, almost no parasitic adults were detectable in day 6 challenge infected mice compared to mice that were infected for the first time irrespective of the presence or absence of basophils (Fig 3B). The reduced parasite burden in the small intestine during challenge infection was reflected by a reduced release of S. ratti DNA in the feces of vaccinated Mcpt8-WT and Mcpt8-Cre mice from day 6 to day 14 post re-infection (Fig 3C). The intact immunity in Mcpt8-Cre mice was not caused by a repopulation of basophils during primary infection or challenge infection (S1 Fig).
In summary, these results show that basophils contribute to the control of intestinal parasite burden during a primary infection but are dispensable for the protection during challenge infection of immune mice regarding both, the efficient control of tissue migrating L3 and the expulsion of parasitic adults from the small intestine.
The combined results of this study rule out a central role for basophils in the generation and execution of the adaptive anti-helminth response during a secondary infection. However, within this study we reproduced our earlier results showing a non-redundant contribution of basophils, next to mast cells, in the expulsion of S. ratti parasitic adults from the small intestine during a primary infection ([4] and Fig 3). We have shown previously that mast cell activation and subsequent parasite expulsion was suppressed by Foxp3+ regulatory T cells (Treg) that expanded upon S. ratti infection [18, 26]. Since mast cells and basophils are innate effector cells with overlapping functions that both contribute to the early control of intestinal S. ratti burden, we asked, whether Treg would also directly interfere with basophil function during S. ratti infection. To this end we crossed basophil-deficient Mcpt8-Cre mice to Depletion of Treg (DEREG) mice. DEREG mice are transgenic for a bacterial artificial chromosome driving the expression of a fusion protein consisting of the diphtheria toxin receptor (DTR) and enhanced green fluorescent protein (eGFP) under the control of the Foxp3 promoter [16]. The application of diphtheria toxin (DT) leads to a rapid and transient depletion of Foxp3+ regulatory T cells in mice that are heterozygous for the DEREG allele. Already the heterozygous expression of the Cre recombinase under the control of the Mcpt8 promoter in Mcpt8-Cre mice is sufficient to induce constitutive and complete depletion of basophils. Therefore the F1 of Mcpt8-Cre and DEREG mice yielded four different genotypes and phenotypes: basophil-competent Mcpt8-WT mice and basophil-deficient Mcpt8-Cre mice that were both DEREG negative and thus not susceptible to DT-mediated Treg depletion as well as basophil-competent Mcpt8-WT DEREG mice and basophil-deficient Mcpt8-Cre DEREG mice that were both susceptible to DT-mediated Treg depletion. All groups were treated with DT one day before S. ratti infection and the frequency of Treg in the peripheral blood was measured day 1 p.i. to control efficient Treg depletion in DEREG positive groups compared to DEREG negative littermates (Fig 4A). Parasite burden in the small intestine was counted at day 6 p.i. (Fig 4B). Depletion of Treg in basophil-competent BALB/c DEREG mice reduced parasite burden in the intestine as we have shown before [18, 26], demonstrating that Foxp3+ Treg interfered with efficient parasite expulsion. Basophil deficiency elevated the parasite burden in the presence of normal Treg frequency as observed before (Fig 3B). However, depletion of Treg in basophil-deficient mice still led to a significant reduction of this initially higher intestinal parasite burden. Thus, Treg interfered with efficient expulsion of S. ratti independently of the presence and thus independent of the function of basophils.
Analyzing the role of basophils during intestinal helminth infection of mice we demonstrate that basophils expanded during infection with S. ratti and specifically contributed to the early control of intestinal parasite burden. Basophils were not involved in the control of S. ratti tissue migrating larvae during primary infection and were dispensable for the generation of a protective Th2/9 immune response during S. ratti infection. Parasite-specific cytokine and Ab production as well as mast cell activation were not reduced in S. ratti-infected Mcpt8-WT and Mcpt8-Cre mice. Moreover, numbers of migrating larvae in the tissue, numbers of parasitic adults in the intestine and fecal release of eggs and L1 in a secondary infection were not affected by basophil deficiency.
In line with our results, IgG1 and IgE production, eosinophilia and Th2 expansion were not affected by constitutive basophil deficiency in Mcpt8-Cre mice during primary H. polygyrus and N. brasiliensis infection [9, 10]. Basophils were also not required for the initiation of a Th2 response to Schistosoma mansoni in constitutive basophil-deficient Basoph8 x Rosa-DTa mice [27] or Mcpt8-Cre mice [28]. Moreover, Litomosoides sigmodontis infection, a model for human filarial infections [29] elicited comparable cellular and humoral immune responses and was cleared with similar kinetics in basophil-deficient Mcpt8-Cre mice and Mcpt8-WT mice [11].
By contrast, depletion of basophils by injection of diphtheria toxin to BaS-DTR mice during T. spiralis infection led to an impaired production of type 2 associated cytokines such as IL-5, IL-4 IL-13 [8]. However, as the diminished Th2 cytokine production had no impact on intestinal parasite burden in the basophil-depleted T. spiralis-infected mice, the clinical relevance of the observed alteration in cytokine production appeared to be limited. In this context, also we cannot formally rule out a contribution of basophils to the amplification of the anti-S. ratti Th2/9 response after day 6 p.i. (i.e. the time point analyzed here) that had no impact on the kinetics of parasite clearance or parasite burden during a challenge infection.
One hallmark of effective initiation of an adaptive immune response to parasite infection is the establishment of a potent immune memory that is characterized by a more efficient protection against a secondary infection in many models of parasite infection. Our results show that the established anti-S. ratti Th2/9 immune response in the absence of basophils was readily translated into protection against a secondary challenge infection where hardly any adult parasite was detected in the small intestine. In line with this finding, other mouse models for basophil deficiency, basophil depleted MasTRECk and Mcpt8DTR mice, were protected against a secondary infection with the closely related helminth S. venezuelensis [7].
By contrast, absence of basophils clearly impaired the efficient control of parasite burden during a secondary infection with either N. brasiliensis or H. polygyrus, although via different mechanisms [9]. Using mixed bone marrow chimeras, the authors showed that specifically the absence of basophil-derived IL-4 and IL-13 resulted in reduced Th2 cell expansion and increased intestinal parasite burden during secondary H. polygyrus infection [9]. Interestingly Th2 cell expansion and IgE response were intact during secondary N. brasiliensis infection in basophil-deficient mice [9, 10]. Here, basophils were important as effector cells mediating efficient eradication of migrating N. brasiliensis larvae already in the skin [12]. Comparable to S. ratti, N. brasiliensis larvae penetrate the skin and migrate within 3–4 days via the lung and mouth to the small intestine. Depletion of basophils by DT injection into Mcpt8DTR mice before a secondary N. brasiliensis infection abolished larval retention in the skin and caused a higher parasite burden in the lung. Thereby, larval trapping in the skin depended on IgE and FcεRI-mediated activation of basophils [12]. A similar role for-IgE activated basophils was demonstrated in mediating resistance to a secondary tick infection [30].
Although immunity to secondary S. ratti infections is also characterized by efficient killing of tissue migrating larvae [3], attack and trapping of Strongyloides larvae predominantly depends on eosinophils and neutrophils that are activated by complement and IgG [22, 31–35]. In line with these reports we did not record changed numbers of tissue migrating larvae in basophil-deficient mice neither during primary nor challenge infection, suggesting that basophils are not as central for trapping of tissue migrating Strongyloides larvae as for trapping of N. brasiliensis larvae [12].
In summary, our results show that the dominant function of basophils during S. ratti infection in mice is the promotion of early intestinal parasite expulsion that is mediated by both basophils and mast cells [4, 7]. Although basophil-deficient mice displayed a higher intestinal parasite burden at day 6 p.i., clearance of infection was executed with WT kinetics within 4 weeks showing that lack of basophils as intestinal effector cells can be compensated by other cells. In sharp contrast, deficiency of mast cells resulted in prolonged infection to more than 20 weeks pointing out the non-redundant function of IL-9 activated mast cells as intestinal effectors during S. ratti infection [4, 20].
Although S. ratti infection is terminated after 4 weeks by immune competent mice, we have shown before that, in order to establish intestinal infection and to survive this 4 weeks, S. ratti actively down-modulates the immune response that promotes its expulsion [18, 26, 36]. S. ratti infection induces expansion of Treg and their depletion drastically reduced intestinal parasite burden in BALB/c mice [18, 26]. Since additional mast cell deficiency abrogated the beneficial effect of Treg depletion, the expanding Treg population directly interfered with mast cell mediated parasite expulsion [18]. In the current study, we show that basophil-deficiency did not abrogate the beneficial effect of Treg depletion. Although the basophil-deficiency alone elevated intestinal S. ratti parasite burden compared to basophil-competent mice, additional Treg depletion significantly reduced the intestinal parasite burden in the absence of basophils. Thus, basophil function in the intestine is not specifically suppressed by S. ratti-mediated immune evasive mechanism, emphasizing their minor role in the efficient control of this particular parasite.
|
10.1371/journal.ppat.1006630 | IFN-γ extends the immune functions of Guanylate Binding Proteins to inflammasome-independent antibacterial activities during Francisella novicida infection | Guanylate binding proteins (GBPs) are interferon-inducible proteins involved in the cell-intrinsic immunity against numerous intracellular pathogens. The molecular mechanisms underlying the potent antibacterial activity of GBPs are still unclear. GBPs have been functionally linked to the NLRP3, the AIM2 and the caspase-11 inflammasomes. Two opposing models are currently proposed to explain the GBPs-inflammasome link: i) GBPs would target intracellular bacteria or bacteria-containing vacuoles to increase cytosolic PAMPs release ii) GBPs would directly facilitate inflammasome complex assembly. Using Francisella novicida infection, we investigated the functional interactions between GBPs and the inflammasome. GBPs, induced in a type I IFN-dependent manner, are required for the F. novicida-mediated AIM2-inflammasome pathway. Here, we demonstrate that GBPs action is not restricted to the AIM2 inflammasome, but controls in a hierarchical manner the activation of different inflammasomes complexes and apoptotic caspases. IFN-γ induces a quantitative switch in GBPs levels and redirects pyroptotic and apoptotic pathways under the control of GBPs. Furthermore, upon IFN-γ priming, F. novicida-infected macrophages restrict cytosolic bacterial replication in a GBP-dependent and inflammasome-independent manner. Finally, in a mouse model of tularemia, we demonstrate that the inflammasome and the GBPs are two key immune pathways functioning largely independently to control F. novicida infection. Altogether, our results indicate that GBPs are the master effectors of IFN-γ-mediated responses against F. novicida to control antibacterial immune responses in inflammasome-dependent and independent manners.
| The cell-intrinsic immunity is defined as the mechanisms allowing a host cell infected by an intracellular pathogen to mount effective immune mechanisms to detect and eliminate pathogens without any help from other immune cells. In infected macrophages, the Guanylate Binding Proteins (GBPs) are immune proteins, induced at low levels in a cell autonomous manner by endogenous type I IFN or at high levels following IFN-γ production by innate and adaptive lymphocytes. The antibacterial activity of GBPs has been recently tightly linked to the inflammasomes. Inflammasomes are innate immune complexes leading to inflammatory caspases activation and death of the infected cell. Francisella novicida, a bacterium replicating in the macrophage cytosol, is closely related to F. tularensis, the agent of tularemia and is used as a model to study cytosolic immunity. GBPs contribute to F. novicida lysis within the host cytosol leading to DNA release and AIM2 inflammasome activation. In addition to their regulation of the AIM2 inflammasome, we identified that GBPs also control several other pyroptotic and apoptotic pathways activated in a hierarchical manner. Furthermore, we demonstrate that IFN-γ priming extends GBPs anti-microbial responses from the inflammasome-dependent control of cell death to an inflammasome-independent control of cytosolic bacterial replication. Our results, validated in a mouse model of tularemia, thus segregate the antimicrobial activities of inflammasomes and GBPs as well as highlight GBPs as the master effectors of IFN-γ-mediated cytosolic immunity.
| Intracellular pathogens have evolved sophisticated mechanisms to invade and replicate within host cells. In parallel, multi-cellular organisms have evolved multiple mechanisms allowing a host cell to detect microbial infection and to mount an effective antimicrobial response. Key actors of the host cell intrinsic immunity include the Guanylate Binding Proteins (GBPs)[1–3]. GBPs constitute a family of interferon-inducible dynamin-like GTPases [4,5]. 11 GBPs are encoded by the murine genome in two clusters on chromosomes 3 and 5 [6,7]. The antimicrobial functions of GBPs are still poorly understood. One key mechanism of GBPs' potent antimicrobial activity resides in their ability to target and disrupt pathogen-containing vacuoles. Indeed, chromosome 3-encoded GBPs (Gbpchr3) are required to disrupt Toxoplasma gondii parasitophorous membrane and to control the parasite replication in mice [8,9]. Similarly, Gbpchr3 are required to lyse the Salmonella-containing vacuole leading to the release of this bacterium into the host cytosol and to the subsequent activation of the caspase-11 non-canonical inflammasome [10].
Cooperation between GBPs and the NLRP3, the AIM2 and the non-canonical caspase-11 inflammasome complexes have emerged recently as central to the innate immune responses against intracellular bacteria [10–14]. However, the functional molecular links between GBPs and the inflammasomes remain unclear. GBP5 was described to bind NLRP3 and directly promote NLRP3-dependent inflammasome assembly [14]. This finding was later challenged by several groups [10,15]. Chromosome 3-encoded GBPs are key host factors to trigger the non-canonical caspase-11 inflammasome in macrophages infected with various Gram-negative bacteria [16] including Salmonella typhimurium [10], Legionella pneumophila [12] and Chlamydia trachomatis [11]. Three different models have been proposed to explain the role of GBPs in promoting non-canonical inflammasome activation. Similarly to what have been demonstrated for IFN-γ-inducible Immunity-Related GTPases (IRG) in their antimicrobial role against Toxoplasma gondii [17], GBPs might disrupt the Salmonella-containing vacuole leading to the release of this bacterium and its associated LPS into the host cytosol [10]. Alternatively, GBPs might orchestrate the recruitment of IRGB10 onto cytosolic bacteria to liberate LPS for sensing by caspase-11 [16]. Finally, as GBPs can promote caspase-11 activation without any detectable recruitment around Chlamydia muridarum inclusions and upon Legionella LPS transfection into the host cytosol, Coers and colleagues suggested that GBPs might directly facilitate caspase-11 activation [11,12].
The link between GBPs and the AIM2 inflammasome is less controversial and has been mostly studied by our group and others in macrophages infected with Francisella novicida [13,15,18]. F. novicida is a close relative of F. tularensis, the agent of tularemia. The virulence of Francisella strains is linked to their ability to rapidly lyse the phagosome, escape into the host cytosol [19,20] and replicate within this compartment. This process is dependent on a cluster of genes in the Francisella-pathogenicity island [21,22], which encodes an atypical type VI secretion system [23]. GBP2 and GBP5 are recruited onto cytosolic F. novicida and are required to lyse bacteria and release the bacterial genomic DNA into the host cytosol. DNA in the host cytosol is then recognized by AIM2 [13,15,24–26]. Recently, GBPs were demonstrated to be required for IRGB10 recruitment onto cytosolic F. novicida to mediate cytosolic bacterial killing, DNA release and AIM2 inflammasome activation [16].
Chromosome 3-encoded GBPs and the AIM2 inflammasome are both equally required to resist F. novicida infection in vivo [13,15]. Yet, whether GBPs might have anti-F. novicida functions dependent on other inflammasomes or inflammasome-independent antibacterial responses remain unclear [10,16]. In this work, we demonstrate that GBPs control F. novicida-mediated host cell death in a hierarchical manner implicating at least 3 different canonical and non-canonical inflammasome complexes as well as apoptotic caspases 8, 9 and 3. Furthermore, we demonstrate that upon IFN-γ treatment, GBPs control F. novicida replication independently of the canonical and non-canonical inflammasome pathways and independently of macrophage cell death. IFN-γ-, GBPs-mediated inhibition of intracellular bacterial growth was also effective during F. tularensis spp. holarctica Live Vaccine Strain infection, while IFN-γ was inefficient to block the replication of the highly virulent F. tularensis SCHU S4 strain. Finally, we demonstrate in vivo that IFN-γ-mediated host protection against F. novicida is largely GBPs-dependent and inflammasome-independent. Our work thus positions the GBPs as the master effectors of the IFN-γ-mediated anti-F. novicida responses.
We and others have previously reported that Gbpchr3-deficient and inflammasome-deficient mice (Aim2-/-, Asc-/- and Casp1/Casp11-/- mice) were highly susceptible to F. novicida infection [13,15,24,25,27,28]. In vitro studies have demonstrated that GBPs act upstream of the AIM2 inflammasome [13,15,16] suggesting that the Gbpchr3 deficiency should phenocopy deficiencies in the AIM2 inflammasome. In a survival experiment, Gbpchr3-KO and Asc-/- were almost as susceptible to F. novicida infection (Fig 1A) although we consistently noticed that Gbpchr3KO mice died slightly faster than Asc-/- mice. Similarly, Gbpchr3-KO and Asc-/- mice displayed very high bacterial burden both in the spleen and in the liver at day 2 post-infection (PI) with an average of 40-fold more bacteria than WT mice (Fig 1B). Surprisingly, at 48 h PI, IFN-γ level in the serum of Gbpchr3-deficient mice reached WT levels while in agreement with previous work [28], IFN-γ level in the serum of Asc-/- mice was strongly decreased (Fig 1C). We and others have previously reported that early (16 h PI [13], 24 h PI [16]) IL-18 production in vivo is GBPs-dependent, a finding which we reproduced here (Fig 1D). However, at later time points (48 h PI), IL-18 levels were similar in Gbpchr3-KO and in WT mice (Fig 1E). As expected, IL-18 levels in infected Asc-/- mice were not statistically different from the levels observed in uninfected mice. These results indicated that inflammasome activation was only delayed in vivo in Gbpchr3-deficient mice. The increase in IL-18 levels observed over 48h in infected Gbpchr3-KO mice likely explained the high IFN-γ serum level observed in these mice. The high susceptibility of Gbpchr3-KO mice to F. novicida infection despite high levels of circulating IFN-γ is remarkable since IFN-γ is considered to be one of the most important cytokine to fight Francisella infection [29–31]. This conundrum led us to hypothesize that GBPs might be the main IFN-γ effector in F. novicida-infected mice. Furthermore, this result indicates that while the overall susceptibility of Gbpchr3-deficient mice and Asc-/- mice to F. novicida infection are similar, the in vivo antibacterial mechanisms of GBPs are, at least partially, independent of the inflammasome.
The action of GBPs against F. novicida has been mostly studied in unprimed macrophages [13,15,16]. Under these conditions, GBPs induction relies on the endogenous recognition of nucleic acids by the cGAS pathway, secretion of type I IFN, signaling through the type I IFN receptor (IFNAR1) and activation of the IRF-1 pathway [15,24,32]. Importantly, there was a drastic quantitative shift in GBP2 and GBP5 transcripts levels upon priming with IFN-γ compared with F. novicida-mediated endogenous induction or to induction following IFN-β priming (S1A Fig). Indeed while GBP2 transcript levels increased by a factor of 15 upon F. novicida infection, IFN-γ priming of infected macrophages, led to a 325-fold increase in GBP2 transcript levels relative to its level in uninfected macrophages. As previously reported, this very strong induction likely results from synergistic NF-κB and IFN signaling [14,33,34]. Indeed, we obtained comparable levels of GBP2 induction when BMDMs were primed with both IFN-γ and Pam3CSK4, a TLR2 agonist (S1A Fig). ProIL-1β transcript levels were not impacted by IFN-γ priming (S1C Fig) indicating that this synergy was specific for GBPs induction. Similar results were obtained while investigating GBP5 transcript levels (S1B Fig) or while monitoring GBP2 and GBP5 protein levels (S1D Fig). This quantitative shift in GBPs levels upon IFN-γ treatment led us to investigate the impact of IFN-γ on the F. novicida-mediated cell death.
In F. novicida-infected bone marrow-derived macrophages (BMDMs), the only reported cell death pathways are dependent on the AIM2/ASC complex [24,25,27,35]. By monitoring, in real time, propidium iodide influx and fluorescence over 24 h in unprimed macrophages deficient for various inflammasome components, we observed substantial differences in the kinetics of propidium iodide between various knock-out macrophages suggesting that several cell death pathways were engaged following F. novicida infection (Fig 2A). These differences in the kinetics of propidium iodide incorporation/ fluorescence were demonstrated to be statistically significant by calculating the area under the curve corresponding to each kinetics (Fig 2B). As previously described [13,15,24,25], in the absence of IFN-γ priming, F. novicida-infected BMDMs died in an AIM2-dependent manner. Indeed, at MOI 10, propidium iodide incorporation/fluorescence sharply increased around 6 h post-infection in WT macrophages while Aim2-/- BMDMs presented cell death kinetics delayed by more than 7 h compared with that of WT macrophages. Interestingly, incorporation/fluorescence increase of propidium iodide was significantly further delayed in Gbpchr3-KO BMDMs compared to Aim2-/- BMDMs suggesting that GBPs control both AIM2-dependent and -independent cell death pathways. Accordingly, by monitoring macrophage cell death in real time and in single cells using time-lapse microscopy (Fig 2E and 2F; S3 Fig), we clearly observed that the number of propidium iodide-positive cells increased significantly later in Gbpchr3-KO BMDMs compared to Aim2-/- BMDMs. This difference in cell death kinetics was exacerbated in the presence of IFN-γ priming (Fig 2C, 2D and 2F). Finally, to exclude any bias associated with propidium iodide incorporation, we used a luminescent cell viability assay based on the quantitation of ATP, a signature of metabolically active cells (CellTiter-Glo; Fig 2G). This assay confirmed that Gbpchr3-KO BMDMs survived longer than Aim2-/- BMDMs upon F. novicida infection. Altogether, our data strongly suggest that GBPs control both AIM2-dependent and -independent cell death/survival pathways.
In F. novicida-infected murine macrophages, the only inflammasomes described so far are dependent on AIM2, while our results (Fig 2A–2G) clearly demonstrate that F. novicida-mediated cell death can proceed independently of AIM2. We thus used Aim2-/- macrophages to unravel other cell death pathways controlled by GBPs. Using siRNA, we first knock-downed various inflammasome NLRs (S4A Fig) and observed a contribution of NLRP3 in F. novicida-mediated cell death (S4B Fig). To confirm this result, we generated Aim2-/-/Nlrp3-/- mice and compared their BMDMs response with that of BMDMs single knock-out for Aim2 (Fig 3A and 3B). Aim2-/-/Nlrp3-/- BMDMs displayed a kinetics of propidium iodide incorporation/fluorescence slower than that of Aim2-/- BMDMs, indicating that in the absence of AIM2, NLRP3 controls F. novicida-mediated cell death. Aim2-/-/Nlrp3-/- BMDMs phenocopied Asc-/- BMDMs indicating that activation of the canonical inflammasome pathways in F. novicida-infected BMDMs is exclusively dependent on AIM2 and NLRP3.
To assess whether the non-canonical caspase-11 inflammasome could be involved in the AIM2-independent detection of F. novicida, we used siRNA against caspase-11. We observed a consistent and significant delay in Aim2-/- macrophage death upon treatment with a caspase-11 siRNA (Fig 3C and 3D and S4B Fig) indicating that while F. novicida is largely able to escape caspase-11 detection ([36], Fig 3E, S4C and S4D Fig), caspase-11 may contribute to macrophage cell death at late time points of the infection in the absence of AIM2. The contribution of caspase-11 in the immune response of Aim2-/- macrophages was further strengthened by investigating IL-1β release. Indeed, caspase-11 expression knock-down strongly decreased the late secretion of IL-1β observed in Aim2-/- macrophages (Fig 3F). Nlrp3 expression knock-down or knock-out also consistently decreased IL-1β release in Aim2-/- macrophages (Fig 3F and 3G). This decrease may be partly due to its involvement downstream of caspase-11 [37].
Altogether, our data demonstrate the involvement of at least two sequential cell death pathways (mediated by the sensors AIM2, NLRP3 and caspase-11) elicited in response to F. novicida infection. The extensive survival of Gbpchr3-KO BMDMs strongly suggests that GBPs contribute to activation of these three inflammasome complexes during F. novicida infection. Of note, at 24 h post-infection in absence of IFN-γ, propidium iodide incorporation/ fluorescence was similar in Asc-/- and Gbpchr3-KO BMDMs. The late cell death occurring in Gbpchr3-KO BMDMs was associated with IL-1β release while as expected no IL-1β was observed in the supernatant of Asc-/- BMDMs (S5 Fig). This result suggests that while chromosome 3-encoded GBPs are instrumental in promoting fast inflammasome activation upon F. novicida infection, they are not strictly required to trigger inflammasome activation likely explaining the bi-phasic dependence of IL-18 serum level on GBPs (Fig 1D and 1E). Interestingly, analysis of macrophages deficient for both Asc and Gbpchr3 demonstrated a strong delay in propidium iodide incorporation/fluorescence increase compared to that of Asc-/- and Gbpchr3-KO BMDMs (S6A and S6C Fig) providing genetic evidence that GBPs can act independently of the canonical inflammasomes.
In addition to uncovering AIM2-independent cell death pathways, real time cell death analysis of F. novicida-infected BMDMs (Figs 2A–2F and 3A–3D and S3 and S4 Figs) revealed both known and unsuspected cell death pathways. In agreement with previously reported LDH-release quantifications [13], IFN-γ could complement Ifnar1-/- BMDMs inability to rapidly undergo cell death upon F. novicida infection (Fig 2). Furthermore, as previously described [35,38], Casp1/Casp11-/- BMDMs died significantly faster than Aim2-/- BMDMs (Fig 2A–2G, S3 Fig). Apoptosis is associated with cell retraction while pyroptosis is associated with cell swelling [39]. We took advantage of wheat germ agglutinin (which binds cell membrane glycoproteins) staining intensity to quantify cell retraction and of well-characterized stimuli triggering pyroptosis (LPS + nigericin [40]) and apoptosis (gliotoxin [41]) to validate this cell retraction quantification (Fig 4A). F. novicida-infected Casp1/Casp11-/- BMDMs cell death proceeded via a major cell retraction (S1 Movie, Fig 4B and 4C), a morphological feature of apoptosis. The ability of AIM2/ASC complex to recruit caspase-8 and trigger apoptosis in Casp1/Casp11-/- BMDMs [35,38] may explain the kinetics of cell death observed in Casp1/Casp11-/- BMDMs. Indeed, at 10 h post-infection, we observed processing of the apoptotic caspases-8, 9 and 3 in infected Casp1/Casp11-/- BMDMs. As expected and as previously reported [35], we did not observe such apoptotic caspases processing in WT pyroptotic macrophages (Fig 4D). Furthermore, a strong DEVDase activity suggestive of active caspase-3 (or caspase-7) was observed in Casp1/Casp11-/- BMDMs but not in WT macrophages (Fig 4E and 4F). Interestingly, the kinetics of propidium iodide incorporation/fluorescence in infected Gbpchr3-KO BMDMs were even slower than the corresponding kinetics of Casp1/Casp11-/- BMDMs (Fig 2A–2D). This difference strongly suggests that GBPs not only control pyroptosis, but also has the ability to control apoptotic pathways.
This conclusion was further strengthened by comparing cell death kinetics, morphological and molecular features of IFN-γ-primed Asc-/- and Gbpchr3-KO BMDMs. Priming with IFN-γ accelerated the cell death kinetics in the different macrophages with one notable exception (Fig 2C–2F). Indeed, we did not detect any substantial propidium iodide incorporation/fluorescence increase in IFN-γ-primed Gbpchr3-KO macrophages while IFN-γ-primed Asc-/- BMDMs died although with delayed kinetics compared with that of WT macrophages (Fig 2C–2F, S3B Fig). The death of Asc-/- BMDMs was confirmed by addition of triton (TX100) at the end of the 24 h kinetics. Indeed, TX100 treatment did not further increase propidium iodide incorporation/fluorescence in Asc-/- BMDMs (S2B Fig). In contrast, TX100 addition in Gbpchr3-KO BMDMs led to a strong increase in propidium iodide fluorescence indicating that the plasma membrane of most Gbpchr3-KO BMDMs was still intact (and not permeable to propidium iodide) at 24 h PI. The relatively lower intensity of propidium iodide fluorescence in Asc-/- BMDMs compared to WT BMDMs might be related to intrinsic differences in propidium iodide incorporation/fluorescence upon incorporation in apoptotic vs. pyroptotic nuclei. Cell death of IFN-γ-primed Asc-/- BMDMs was associated with cell retraction (S2 Movie and Fig 4C) and was reminiscent of the morphological cell death observed in Casp1/Casp11-/- BMDMs although it progressed with a ≈6 h delay compared to the latter BMDMs. Accordingly, apoptotic caspase-8/9/3 processing and DEVDase activity were clearly detectable in Asc-/- BMDMs infected for 16 h corroborating the morphological features and demonstrating that at late time points of infection, Asc-/- BMDMs die by apoptosis.
The extensive survival of Gbpchr3-KO BMDMs upon IFN-γ priming was confirmed using CellTiter-Glo assay (Fig 2G). Furthermore, we observed a limited number of cell death events in infected Gbpchr3-KO BMDMs using time-lapse microscopy (Fig 2F, S3 Fig and S2 Movie) suggesting that a limited GBPs-independent cell death pathway occurs at late time points of infection. The cell survival effect of IFN-γ on Gbpchr3-KO BMDMs (compare Figs 2A and 2B with 2C and 2D; S3A with S3B and S6A with S6B, 2F and 2G, S3C and S6C) correlates with a decrease in IL-1β release (S5 Fig). This result suggests that in infected Gbpchr3-KO BMDMs, IFN-γ might inhibit inflammasome activation possibly via nitrosylation, an IFN-inducible mechanism previously reported to inhibit inflammasome activation [42,43]. While the three different cell death/survival assays showed subtle differences [44], these assays converge to indicate that upon IFN-γ priming, Gbpchr3-KO BMDMs survive better than Casp1/Casp11-/- and Asc-/- BMDMs suggesting that GBPs control both pyroptotic and apoptotic pathways. Importantly, the GBPs-mediated control of apoptosis was confirmed by generating mice doubly deficient for Asc and Gbpchr3. Macrophages deficient for both Asc and Gbpchr3 failed to demonstrate caspase-8/9 and 3 maturation and DEVDase activity in contrast to Asc-/- macrophages (S6D and S6E Fig). Altogether, our results demonstrate that GBPs act as major cell death regulators by controlling numerous pyroptotic and apoptotic cell death pathways.
The high IFN-γ levels observed in Gbpchr3-KO mice in the mouse model of tularemia coupled to the inability of Gbpchr3-KO mice to control F. novicida burden (Fig 1) led us to investigate if GBPs contribute to the IFN-γ-mediated growth restriction observed in vitro [45]. As previously described [13,15,24,35], we observed a robust F. novicida replication in unprimed macrophages (Fig 5A corresponding to the Raw data presented in S7 Fig), which was partially controlled in a GBP-dependent manner by the inflammasome. IFN-γ priming led to F. novicida killing as visualized by a net decrease in the recovered intracellular colony forming units (Fig 5A) as soon as 6 h PI. Bacterial killing was highly dependent on GBPs. Indeed, F. novicida was not killed in IFN-γ-primed Gbpchr3-KO macrophages, but robustly replicated by a 40-fold over 12 h of infection. We did not observe any substantial effect of IFN-γ on phagosomal rupture (S8 Fig) suggesting that IFN-γ does not act by restricting F. novicida access to its cytosolic niche. This finding is consistent with a previous study demonstrating a direct activity of IFN-γ on cytosolic F. tularensis [45]. Furthermore, the IFN-γ-mediated bacterial growth inhibition was independent of the NADPH oxidase and of the IFN-γ-inducible NO synthase (iNOS also known as NOS2), two immune effectors described to act downstream of GBPs [46] (S9C–S9E Fig). In WT macrophages, the GBPs-mediated bacterial killing was difficult to segregate from the antibacterial effects of host cell death due to the very rapid inflammasome activation (see Fig 2B). Yet, bacterial killing was also observed in IFN-γ-primed Asc-/- macrophages infected for 6 h at a MOI of 1 in the absence of substantial pyroptotic and apoptotic cell death. Similarly, IFN-γ-mediated blockage in bacterial replication was observed in Casp1/Casp11-/- and Aim2-/- macrophages (S9A Fig) suggesting that intracellular bacterial killing is independent of canonical and non-canonical inflammasomes. To assess whether IFN-γ-mediated inhibition of bacterial growth was dependent or independent of cell death we took advantage of flow cytometry using GFP-expressing F. novicida and propidium iodide to exclude dead cells. In the absence of IFN-γ priming, replication was observed in ≈30% of live Asc-/- and ≈50% Gbpchr3-KO macrophages (as determined by the number of propidium iodide- GFP+ cells). In the presence of IFN-γ, this number dropped to less than 4% in Asc-/- macrophages indicating a robust bacterial growth restriction in these cells. In contrast, a large number of Gbpchr3-KO macrophages (>40%) sustained F. novicida replication despite IFN-γ priming (Fig 5B). These results suggest that this anti-bacterial GBP-dependent mechanism proceeds independently of cell death at least in Asc-/- BMDMs.
To quantify F. novicida replication in a large number of macrophages, we analyzed infected macrophages using high-resolution microscopy in Flow (ImagestreamX). This single cell quantification technique further demonstrated that in the absence of Gbpchr3, IFN-γ was almost ineffective to control bacterial replication. Indeed, in the presence of IFN-γ, there was a 100-fold increase in the number of Gbpchr3-KO macrophages permissive for bacterial replication (containing more than 100 bacteria) compared with that of Asc-/- macrophages (Fig 5C, S9F Fig). Representative immunofluorescence images of this striking phenotype are presented in Fig 5D. While this difference was exacerbated in the presence of IFN-γ and as previously noticed using Aim2-/- macrophages [13], we observed that in unprimed macrophages, GBPs also controlled in an ASC-independent manner the bacterial burden.
GBP5-associated F. novicida loose their GFP expression, which has been associated with a loss of bacterial viability [15]. We thus quantified GFP intensity in single bacteria in propidium iodide-negative BMDMs as a surrogate marker of bacterial viability/metabolic activity (Fig 5E). In WT macrophages, IFN-γ treatment led to a significant reduction in the average GFP intensity (-10.2%) of intracellular F. novicida with a large increase (30%, n = 640) in the number of bacteria expressing low GFP levels (as defined by a GFP intensity <80% of the average intensity of bacteria in unprimed WT macrophages). Similarly, IFN-γ priming significantly reduced the average GFP intensity of bacteria in Asc-/- BMDMs (-8.4%) with a large increase (+19%, n = 843) in the number of low GFP-expressing bacteria. Surprisingly, IFN-γ priming had a paradoxical effect on bacterial GFP expression in Gbpchr3-KO BMDMs with an increase in the average GFP intensity (+20.6%) and a strong decrease in the number of low GFP-expressing bacteria (-30%, n = 658). While this paradoxical increase (also visible in Fig 5D) remains to be understood, this experiment demonstrates that IFN-γ-induced GBPs in live macrophages affect the metabolic activity of bacteria. The decrease GFP expression observed in the presence of IFN-γ and GBPs is likely due to GBP-mediated antibacterial activity although we were unable to directly quantify bacteriolysis in single cells. Importantly, the antibacterial activity of IFN-γ against F. tularensis live vaccine strain (LVS) was also fully dependent on Gbpchr3 extending our results to an attenuated strain of the F. tularensis species (Fig 5F). While an IFN-γ-dependent GBPs antibacterial activity was clearly observed upon LVS infection (Fig 5F; 23x fold reduction in WT BMDMs upon IFN-γ treatment), LVS sustained a substantial replication (≈100x) in IFN-γ-primed BMDMs. More strikingly, we were unable to observe a robust IFN-γ-mediated growth restriction of F. tularensis SCHU S4 in BMDMs (S10 Fig). These results suggest that F. tularensis strains have evolved mechanisms to avoid, at least partially, IFN-γ-mediated GBPs-dependent antibacterial activity.
The synergistic roles of GBP2 and 5 in promoting the antibacterial effect of IFN-γ could be demonstrated in J774.1 macrophage-like cells using CRISPR/cas9 (S11 Fig). This result rules out that the extensive replication observed in Gbpchr3-KO BMDMs despite the priming with IFN-γ could be due to a non-specific defect of Gbpchr3-KO BMDMs associated with their large genomic deletion. This is well in line with control experiments previously performed on Gbpchr3-KO mice demonstrating normal induction of IFN-inducible genes and normal susceptibility/resistance to certain pathogens including L. monocytogenes [8,16]. Altogether, these findings demonstrate that, in vitro, in infected macrophages, the anti-bacterial function of IFN-γ, a cytokine known to induce a large number of antibacterial effectors relies almost exclusively on Gbpchr3.
GBPs have been tightly linked to inflammasome complexes [10–16]. Yet our in vitro data clearly demonstrate that GBPs have potent inflammasome-independent antimicrobial functions. Particularly, in vitro, GBPs are the main IFN-γ antimicrobial effectors. In contrast, in the presence of IFN-γ, the inflammasome complex seems largely facultative to control F. novicida replication. ASC is required in vivo to induce IFN-γ via IL-18 release [27,28,47]. To further explore these potential differences in vivo, we administered rIFN-γ to F. novicida-infected mice. Early rIFN-γ administration allowed WT mice to survive F. novicida infection (Fig 6A). rIFN-γ administration clearly extended ASC-/- mice survival (Fig 6B) and strongly delayed Asc-/- mice weight loss (S12 Fig), while it had a very moderate (although statistically significant) effect on Gbpchr3-KO mice survival and weight loss (Fig 6C and 6D, S12 Fig). To assess the functional links in vivo between IFN-γ, GBPs or the inflammasomes and their impact on bacterial replication, we analyzed the bacterial burden in the spleen and the liver at 48h PI following rIFN-γ injection at day 0 and day 1 PI. rIFN-γ was highly efficient to control the bacterial burden in both WT (Fig 6E) and Asc-/- mice (Fig 6F). In contrast, upon IFN-γ injection in Gbpchr3-deficient mice, there was no statistical reduction in the bacterial burden in the liver and the spleen (4-fold reduction in the average splenic burden in Gbpchr3-KO mice versus a 40-fold reduction in Asc-/- mice) (Fig 6F). These results demonstrate that, as observed in infected macrophages, the antibacterial action of IFN-γ is mostly mediated by GBPs in vivo and is largely independent of the canonical inflammasomes.
With the recent discoveries of the tight links between the inflammasome complexes and the GBPs, it became unclear whether most antimicrobial functions of GBPs in infected macrophages were mediated by various inflammasome complexes or were independent of the inflammasomes [48]. Indeed, Gbpchr3- and inflammasome-deficient mice are similarly highly susceptible to F. novicida infection. One of the current models to explain the link between GBPs and the inflammasomes positions GBPs as the molecular platform promoting the inflammasome supramolecular complex [2,3,48,49]. This model is strongly supported by evolution since GBPs from jawed fish display inflammasome-related CARD domains [14]. During F. novicida infection, we and other have previously demonstrated that GBPs are required to trigger AIM2 inflammasome activation while they are dispensable upon direct delivery of DNA into the host cytosol [13,15]. GBPs, in cooperation with IRGB10, were demonstrated to participate in F. novicida lysis into the host cytosol [13,15,16] suggesting that GBPs mostly act to release bacterial DNA into the host cytosol. The polymeric nature of DNA might alleviate the requirement for host factors to promote AIM2 inflammasome activation. Indeed, cytosolic dsDNA could provide the scaffold for AIM2 oligomerization and subsequent inflammasome activation [50,51].
In this work, we demonstrate that GBPs are required not only for the AIM2 canonical inflammasome activation but for most of the programmed cell death pathways that can take place in F. novicida-infected macrophages. In vitro, in F. novicida-infected WT murine BMDMs, so far the only described cell death pathway was the AIM2 inflammasome. Yet, using various knock-out macrophages, we have demonstrated here that several canonical and non-canonical inflammasome complexes can be active during F. novicida infection. These alternative pathways were revealed thanks to the use of Aim2-/- BMDMs and are masked in WT macrophages. Interestingly, Harton and colleagues recently demonstrated that F. tularensis strains, in contrast to F. novicida, elicit NLRP3 inflammasome activation in BMDMs [52]. This result suggests that F. tularensis has evolved to escape AIM2 inflammasome activation while the delayed NLRP3-dependent sensing is conserved in response to various Francisella species. The identification of a caspase-11-dependent role in mediating Aim2-/- BMDMs death and IL-1β release in the absence of AIM2 was unexpected. Indeed, the direct delivery of Francisella LPS into BMDMs cytosol does not activate caspase-11 due to its underacylated structure ([36]). It is still unclear whether the delayed caspase-11-dependent activation observed in Aim2-/- BMDMs is due to sensing of F. novicida LPS or of another endogenous or bacterial ligand. The large number of inflammasome complexes activated upon F. novicida infection is reminiscent of what have been observed during infection with other intracellular bacteria [11,53,54], although F. novicida infection of murine macrophages is somewhat unique due to its high dependence on AIM2 [18]. Remarkably, IFN-γ-induced GBPs are required to trigger all these pathways suggesting that they either act upstream of the inflammasomes or that they have conserved mechanisms to facilitate activation of several (AIM2, NLRP3, caspase-11) inflammasome sensors. Such facilitation of the activation of multiple inflammasomes by GBPs has been previously observed upon Chlamydia infection [11]. Yet, GBPs-mediated control of Chlamydia-mediated cell death was only partial in contrast to what we observed upon F. novicida infection. Importantly, Asc-/- (and Casp1/Casp11-/-) BMDMs died by apoptosis as demonstrated by morphological and molecular analyses (Fig 4, S1 and S2 Movies)[35] indicating that GBPs control both pyroptotic and apoptotic pathways. Indeed, processing of caspase-8, 9 and 3 and DEVDase activity were clearly visible in Asc-/- BMDMs but absent in Asc-/-Gbpchr3-KO BMDMs (S6D and S6E Fig). The GBP-dependent pathway leading to apoptotic caspase activation in Asc-/- BMDMs is still unclear. Antibiotic-mediated bacteriolysis of another cytosolic pathogen (Shigella flexneri) triggers massive caspase-9-dependent apoptosis of epithelial cells [55]. Based on the antibacterial role of GBPs ([13,15,16], this work), we speculate that GBPs-mediated action releases a PAMPs that directly or indirectly triggers apoptosis in the absence of the inflammasome adaptor ASC. The concurrent maturation of both caspase-8 and caspase-9 suggests activation of both extrinsic and intrinsic (e.g. mitochondrial) apoptotic pathways. This dual activation may be due to the cross-activation of the mitochondrial intrinsic apoptosis pathway following cleavage of Bid by caspase-8 [56]. Yet, we cannot exclude a direct GBPs-dependent induction of mitochondrial dysfunction such as the one occurring in S. flexneri-infected cells of non-myeloid lineages [57]. IRGs can lyse T. gondii vacuole leading to parasite permeabilisation and a caspase-1-independent necrotic death in mouse embryonic fibroblats [58]. Due to the diversity of cell death pathways controlled by GBPs and IRGs, we favor the hypothesis that these IFN-induced GTPases act as PAMPs-shedders to release/uncover various microbial cell death-activating ligands. Future studies are needed to establish the pathways linking GBPs to the different cell death pathways.
IFN-γ is the most potent cytokine against intracellular bacteria due to its ability to induce hundreds of genes promoting host defense [1]. Remarkably, our data indicate that the antimicrobial action of IFN-γ against F. novicida and F. tularensis Live Vaccine Strain is almost exclusively dependent on GBPs. While it was previously known that IFN-γ could restrict cytosolic Francisella growth independently of cell death, reactive oxygen or nitrogen species, autophagy and IDO-mediated tryptophan degradation [45], the mechanisms responsible for this growth restriction were unknown. The IFN-γ-mediated F. novicida growth restriction is independent of caspase-1 and caspase-11 (S3 Fig). It thus differs from the recently described mechanisms responsible for growth inhibition of cytosolic Salmonella [59]. Interestingly, the highly virulent F. tularensis SCHU S4 largely escaped IFN-γ-mediated antibacterial activity in BMDMs (S10 Fig). It remains unclear whether this escape is the result of an active process or due to the failure of innate immune sensors to detect/recognize cytosolic F. tularensis SCHU S4.
In a concurrent work, Kanneganti and colleagues identified IRGB10 as an IFN-inducible GTPase recruited onto cytosolic F. novicida and required to lyse the bacterium and trigger AIM2 inflammasome activation [16]. IRGB10 recruitment is abolished in Gbpchr3-KO macrophages indicating that GBPs and IRGB10 may act together and that the latter protein is likely involved in IFN-γ-mediated growth restriction in murine macrophages. In contrast to GBPs [6], IRGs (with the exception of the constitutively expressed IRGM) are absent in humans [60]. Yet, IFN-γ priming efficiently restricts cytosolic Francisella growth in human macrophages ([45], S13 Fig) indicating that IRGs are facultative for the IFN-γ-mediated antimicrobial role. While our results identify that GBPs are required for IFN-γ-mediated killing of cytosolic bacteria, other host factors are likely involved upstream of GBPs to facilitate GBPs targeting onto cytosolic bacteria. The molecular mechanisms sustaining IFN-γ-dependent GBPs-mediated antibacterial activity remains to be understood.
All animal experiments were reviewed and approved by the animal ethics committee (CECCAPP, Lyon) of the University of Lyon, France under the protocol number #ENS_2012_061, #ENS_2014_017 and #ENS_2017_002 and in strict accordance with the European regulations (#2010/63/UE from 2010/09/22) and the French laws ("Décret n 2013–118 du 1er février 2013 relatif à la protection des animaux utilisés à des fins scientifiques" and "Arrêté ministériel du 1er février 2013 relatif à l'évaluation éthique et à l'autorisation des projets impliquant l'utilisation d'animaux dans des procédures expérimentales").
Gbpchr3-KO, Nos2–/–, Cybb–/–, Casp1–/–/Casp11–/–(a.k.a caspase-1 knockout), Asc–/–, Aim2–/–, Nlrp3–/–mice, all in the C57BL/6, have been previously described [8,24]. Double knock-out mice (Nlrp3-/-Aim2-/- and Asc-/- Gbpchr3-KO) were generated in the framework of this project. The presence of the functional C57BL/6 caspase-11 was verified by PCR amplification of exon 7 boundaries followed by sequencing [37]. Mice were bred at the PBES (Lyon, France).
Age- and sex-matched animals (6–10 weeks old) were infected subcutaneously with 5x103 or 5x104 or 4x105 CFU of F. novicida in 100 μl PBS (as indicated in the figure legends). When applicable, 106 U/ml of rIFN-γ was injected intraperitonealy in 100 μl PBS. Blood was collected by retro-orbital bleeding at 16 h post-infection or intra-cardiac puncture at 48 h post-infection. Animals were sacrificed at the indicated time point post-infection. Mice were examined twice daily for signs of severe infection and euthanized as soon as they displayed signs of irreversible morbidity or as soon as weight loss exceeded 20%.
F. novicida strain U112, its isogenic ΔFPI mutant [61] and F. tularensis subspecies holarctica Live Vaccine Strain (LVS) were used. When applicable, strains were transformed with pKK219-GFP [62].
Preparation and culture of BMDMs were performed as previously described [63]. BMDMs were differentiated in DMEM medium (Invitrogen) with 10% v/v FCS (Thermo Fisher Scientific), 10% MCSF (L929 cell supernatant), 10 mM HEPES (Invitrogen), 5% Sodium pyruvate. 1 day before infection, macrophages were seeded into 6-, 24-, or 96-well plates at a density of 1.25x106, 2.5x105, or 5x104 per well. When applicable macrophages were pre-stimulated with 100ng/ml Pam3CSK4, LPS O111:B4 (InvivoGen) or 100u/ml mIFN-β or mIFN-γ (immunotools). For infections with F. novicida, bacteria were grown overnight in TSB supplemented with 0.1% (w/v) cysteine at 37°C with aeration. The bacteria were added to the macrophages at the indicated MOI. The plates were centrifuged for 15 min at 1500 g and placed at 37°C for 60 min. Cells were washed and fresh medium containing 10 μg.ml-1 gentamycin (Invitrogen) was added. For LVS, cells were infected for 2 h at an MOI of 0.4, washed and incubated in the presence of gentamicin at 5 μg.ml-1.
For F. novicida intracellular replication assay, macrophages were lysed with 1% (w/v) saponin (Sigma) in water for 5 min. Dilution, plating on TSA supplemented with 0.1% (w/v) cysteine and counting was performed using the easySpirale Dilute (Interscience). For LVS replication assay, cells were lysed in PBS with 0.1% deoxycholate, serially diluted in PBS and plated on modified GC-agar base plates.
Gene expression knockdown was done using GenMute (SignaGen laboratories) and siRNA pools (siGenome, Dharmacon). Briefly, wild-type BMDMs were seeded into 24-, or 96-well plates at a density of 1.5x105 or 3x104 per well. siRNA complexes were prepared at 25 nM in GenMute Buffer according to the manufacturer’s instructions for forward knockdowns. siRNA complexes were mixed with BMDMs medium and added onto the cells. BMDMs were infected with F. novicida at an MOI of 10:1 after 48 h of knockdown and analyzed for inflammasome activation as outlined below. siRNA pools included: Aim2 (M-044968-01), Caspase-11 (that is, Casp4) (M-042432-01), Mefv (M-048693-00), Nlrp3 (M-053455-01), Caspase-1 (M-048913-01) and NT (non-targeting) pool 2 (D-001206-14).
IL-1β and IL-18 were measured by ELISA (R&D systems and platinum ebioscience, respectively). Cell viability was determined by the CellTiter-Glo Luminescent Cell Viability Assay (Promega). Global cell death kinetics was monitored in BMDMs by assessing in real time incorporation of propidium iodide (used at 5 μg/ml) through measurement of fluorescence emission at 635 nm every 15 min on a microplate reader (Tecan, see. S2A Fig for the sensitivity of the technique). When indicated triton X100 (Sigma) at 1% (v/v) was added at the end of the kinetics to further control cell death/viability. Area under the curve were computed using Prism software (GraphPad) to obtain a single quantitative readout of the full kinetics as recently described [64]. Gliotoxin (Enzo Pharma) and Nigericin (Sigma) were used at 5 μM, the latter after a 3 h priming with LPS at 100 ng/ml. Single cell death kinetics was determined using an automated time-lapse video microscope (Arrayscan high-content system, Thermo Fisher Scientific). Image analyses of four sparse fields per well were performed using the HCS studio analysis software. Wheat-germ agglutinin (WGA)-labelled BMDMs were used. Individual dead cells were detected and numerated based on the propidium iodide fluorescence staining and normalized to the total number of cells numerated through the vital Hoechst staining at time 0. WGA intensity in propidium iodide positive cells was calculated to quantify cell retraction using HCS studio software. CO2-independent medium (ThermoFisher Scientific) was used for all cytokines dosage and cell death kinetics.
Following BMDMs infection, protein extracts were obtained by lysing cells in the following buffer (10 mM Hepes/KOH, 2mM EDTA, 0.1% CHAPS, 250 mM sucrose, 5mM dithiothreitol). Samples were clarified by centrifugation at 4°C, 13 000g for 15 minutes. Protein concentration was determined using Bradford method (Bio-Rad). Fluorimetric analysis of caspase-3/7 activity was performed as previously described [35] by incubating protein extracts (4 μg/sample) with Ac-DEVD-AFC (Enzo pharma, ALX-260-037) at 40 μM final concentration. Fluorescent reading over 3 h was performed on a fluorimeter (Tecan).
Blotting was done as described before using 15 to 20 μg of protein sample per lane depending on the antibodies [13]. Antibodies used were rabbit anti-GBP2 and rabbit anti-GBP5 (1:1,000; 11854-1-AP/13220-1-AP; Proteintech), anti-caspase-8 (1:2,000; Enzo pharma; ALX-804-447), anti-caspase-9 (1:2,000; MBL; M054), anti-caspase-3 and anti-cleaved caspase-3 (1:1,000; Cell signaling Technologies; #9662 and #9661, respectively). Cell lysates were probed with anti-β-actin antibody (Sigma) at 1:2,000.
Macrophages were seeded on glass coverslips and infected as described above. At the desired time, cells were washed 3 times with PBS and fixed with 4% paraformaldehyde for 15 min at 37°C. Following fixation, coverslips were washed and the fixative was quenched with 0.1 M glycine for 10 min at room temperature. Coverslips were stained with primary antibodies at 4°C for 16 h, washed with PBS, incubated for 1 h with appropriate secondary antibodies at room temperature (1:500, AlexaFluor, Invitrogen), washed with PBS and mounted on glass slides with Vectashield containing 6-diamidino-2-phenylindole (DAPI) (Vector Labs). Antibodies used were chicken anti-Francisella (1:1000, a gift from D. Monack). Coverslips were imaged on a Zeiss LSM710.
Quantification of GFP in GFP-expressing F. novicida was performed using an automated process in ImageJ (NIH, USA). The threshold was adjusted using the moments-preserving thresholding method with the dark background option. Binary watershed process was used to separate individual bacteria. Fluorescence intensity quantification was restricted to individual particle of 0.2 to 2 μm2 with a circularity comprised between 0.5 and 1.
For assessment of bacterial replication by flow cytometry, macrophages seeded onto non-tissue culture-treated plates were infected as described above with GFP-expressing F. novicida strains. At desired time, cells were lifted with trypsin and immediately analyzed by Flow cytometry on a Canto 2 cytometer (BD biosciences). Dead cells were excluded based on staining with propidium iodide.
For the microscopy in flow experiments, macrophages infected with GFP-expressing bacteria were fixed in PFA 4% and analyzed on ImageStream X mark II (Amnis, EMD-Millipore) using the Inspire software with the Extended depth of field (EDF) function activated to increase the spot counts accuracy. Images of single cells were analyzed with the Ideas Software (Amnis, EMD-Millipore) as previously described [13] to quantify the number of bacteria per cell.
Statistical data analysis was done using Prism 5.0a (GraphPad Software, Inc.). To evaluate the differences between three selected groups or more (cell death, cytokine release, FACS, CFU and immunofluorescence-based counts) one-way ANOVA analysis was performed with Tukey's correction for multiple analysis. Komogorov-Smirnov test was used to compare the cell distribution as determined by Imagestream and the distribution of GFP intensity in single bacteria. P values were adjusted for multiple comparisons with the Bonferroni correction approach. Animal experiments were evaluated using Kruskal-Wallis analysis with Dunn's correction except when only two groups were present in the analysis, Mann-Whitney analysis was performed. Survival experiment was analyzed thanks to log-rank Cox-Mantel test. In figures NS indicates ‘not significant’, P values are given according to the following nomenclature: *P<0.05; **P<0.01; ***P<0.001.
|
10.1371/journal.pcbi.1006321 | Profiling cellular morphodynamics by spatiotemporal spectrum decomposition | Cellular morphology and associated morphodynamics are widely used for qualitative and quantitative assessments of cell state. Here we implement a framework to profile cellular morphodynamics based on an adaptive decomposition of local cell boundary motion into instantaneous frequency spectra defined by the Hilbert-Huang transform (HHT). Our approach revealed that spontaneously migrating cells with approximately homogeneous molecular makeup show remarkably consistent instantaneous frequency distributions, though they have markedly heterogeneous mobility. Distinctions in cell edge motion between these cells are captured predominantly by differences in the magnitude of the frequencies. We found that acute photo-inhibition of Vav2 guanine exchange factor, an activator of the Rho family of signaling proteins coordinating cell motility, produces significant shifts in the frequency distribution, but does not affect frequency magnitude. We therefore concluded that the frequency spectrum encodes the wiring of the molecular circuitry that regulates cell boundary movements, whereas the magnitude captures the activation level of the circuitry. We also used HHT spectra as multi-scale spatiotemporal features in statistical region merging to identify subcellular regions of distinct motion behavior. In line with our conclusion that different HHT spectra relate to different signaling regimes, we found that subcellular regions with different morphodynamics indeed exhibit distinct Rac1 activities. This algorithm thus can serve as an accurate and sensitive classifier of cellular morphodynamics to pinpoint spatial and temporal boundaries between signaling regimes.
| Many studies in cell biology employ global shape descriptors to probe mechanisms of cell morphogenesis. Here, we implement a framework in this paper to profile cellular morphodynamics very locally. We employ the Hilbert-Huang transform (HHT) to extract along the entire cell edge spectra of instantaneous edge motion frequency and magnitude and use them to classify overall cell behavior as well as subcellular edge sectors of distinct dynamics. We find in fibroblast-like COS7 cells that the marked heterogeneity in mobility of an unstimulated population is fully captured by differences in the magnitude spectra, while the frequency spectra are conserved between cells. Using optogenetics to acutely inhibit morphogenetic signaling pathways we find that these molecular shifts are reflected by changes in the frequency spectra but not in the magnitude spectra. After clustering cell edge sectors with distinct morphodynamics we observe in cells expressing a Rac1 activity biosensor that the sectors with different frequency spectra associate with different signaling intensity and dynamics. Together, these observations let us conclude that the frequency spectrum encodes the wiring of the molecular circuitry that regulates edge movements, whereas the magnitude captures the activation level of the circuitry.
| Cell morphology and morphodynamics are used to phenotype the state of a cell throughout various processes, including differentiation, proliferation, migration and apoptosis[1–5]. Moreover, numerous signaling pathways converge onto cytoskeleton architecture that determines morphological variation among cells. Therefore, parameters of cell morphology and morphodynamics can also serve as indicators of signaling states[6, 7]. Indeed, analysis of cellular morphology and morphodynamics has been applied, for example, in cancer cell screens[8], drug development[9–11], cell transformation characterization[12] and cell cycle analysis[13, 14].
A number of strategies have been developed to elucidate the physical cause and signaling regulation of cell morphology. Quantification of cell edge movements using physical and mathematical models revealed different modes of motility associated with actin-based spreading[6, 15–25], myosin-related contraction[26] and transverse wave propagation[22, 27–31]. Moreover, shape descriptors have been used for statistical classification of cell morphological patterns[32–38]. However, these studies generally applied a global parameterization of cell morphological changes, such as degree of polarization, cell area change and migration rate, and did not consider the local and dynamic behavior of the cell edge. This has been in part due to the significant complexities in robustly tracking cell edge motion at the subcellular scale. We[39] and others[40] have developed the necessary image analysis framework to track complex cell boundary movements in time-lapse cell image sequences. Densely sampled protrusion and retraction velocities were compiled in space-time maps that offer an opportunity to identify distinct cell morphodynamic states as well as to unveil putative functional links to underlying cytoskeleton dynamics[41–44] and signaling activities[45–47]. Nonetheless, a systematic classification of the spatiotemporal patterns captured by these maps has yet to be performed.
Here, we implement a framework based on the Hilbert-Huang Transform (HHT) to decompose the spatiotemporal signal into instantaneous frequencies and amplitudes. Applied to a population of spontaneously migrating fibroblast-like Cos7 cells, we show that the frequencies encode information on the wiring topology of pathways involved in the regulation of morphodynamics, whereas the amplitudes reflect pathway activation levels. We then validate these results by acute manipulation of the wiring topology of a pathway using optogenetics[38]. We also show that the decomposition into temporally and spatially localized frequency spectra offers an opportunity to identify time windows and cell edge sectors with distinct morphodynamic signatures. This permits determination with subcellular resolution of switches between morphodynamic states that are associated with particular signaling motifs.
We hypothesized that subcellular morphodynamic profiling would be highly informative regarding the states of signaling pathways that regulate cytoskeleton and adhesion dynamics at the cell periphery. To test this, we first imaged unstimulated Cos7 monkey kidney fibroblast-like cells. These cells often exhibit a robust spontaneous migration, and because of their tight adhesion to the substrate, are ideal for high-resolution live cell imaging. We tracked the motion of virtual fiduciaries on the cell boundary by identifying the outline of the cell edge in each frame of a time-lapse sequence, and mapping the outlines of consecutive frames subject to minimizing the overall displacement and strain that are associated with the deforming cell shape[39, 48] (Fig 1A, see Methods and S1 Fig for details on the mapping strategy). We subsequently sampled time series of local protrusion (positive velocities) and retraction (negative velocities) by averaging the motion within edge sectors of ~10 pixel (i.e. ~3 μm) width each. This spacing in sampling corresponds to the half-width-at-half-maximum (HWHM) of the spatial autocorrelation of the edge motion[49]. Velocity time series along the cell boundary were then compiled sector-by-sector into the rows of a matrix referred to as a protrusion activity map[39] (Fig 1B). Accordingly, a matrix column represents the velocity variation over all edge sectors in a particular time point. For the particular cell displayed in Fig 1A, the boundary region encompassing sectors 12–38 prominently protrudes for the first 15 min of the movie, interspersed with short periods of retraction. After 15 min the region splits into two protrusive subregions. The boundary region encompassing sectors 40–54 retracts for the first 10 min before converting into a relatively quiescent zone (see also Video 1). These examples show that the velocity time series is nonstationary. Accordingly, edge motion analysis must be temporally localized.
To analyze edge velocity time series, we adopted the Hilbert–Huang transform (HHT)[50–55]. The HHT relies on an empirical mode decomposition (EMD), which divides the submitted time series into a finite and generally small number of component signals, referred to as intrinsic mode functions (IMFs). The set of IMFs forms a complete and nearly orthogonal basis for the original signal satisfying the following two conditions: i) The number of local extrema and the number of zero-crossings either is equal to each other or at most differs by one. ii) The mean value of the upper envelope defined by the local maxima and the lower envelope defined by the local minima is equal to zero. Under these conditions, the Hilbert Transform is guaranteed to converge to an unbiased estimate of the instantaneous frequency spectrum of the IMF[50, 54].
The EMD procedure involves iterative application of the following steps: i) Identifying all local extrema in the original target time series X(t). ii) Connecting all local maxima by a cubic spline to generate the upper envelope; similarly, connecting all local minima by a cubic spline to generate the lower envelope. iii) Computing the mean m1(t) of upper and lower envelopes and subtracting it from the target time series to generate a reduced series h1(t).
If h1(t) satisfies the aforementioned conditions of an IMF, it is the first IMF component c1(t). Usually this is not the case. Instead h1(1)(t) = h1(t) is considered the new target time series, and the above procedure is repeated k-1 times,
h1(1)(t)−m1(2)(t)=h1(2)(t)⁞h1(k−1)(t)−m1(k)(t)=h1(k)(t)
until h1(k)(t) satisfies the conditions of an IMF. This is the first IMF component c1(t).
The residual signal r1(t) is then defined as
r1(t)=X(t)−c1(t)
and used in a next iteration as the initial target time series. Usually, the decomposition is terminated after n iterations, subject to the condition that the residual signal is either a constant, or a monotonic function, or a function with only one maximum and one minimum, from which no more IMF can be generated. However, in our application IMF sets were compared between experiments. Therefore, it was necessary to fix the number of iterations such that the majority of decomposed data fulfilled the above defined termination criterion. Irrespective of the termination rule, the EMD generates n IMF components c1(t), …, cn(t) and a residual signal rn(t) that satisfy
X(t)=∑i=1nci(t)+rn(t)
where rn(t) either fulfills the above termination criterion or its variance is less than, e.g., 5% of that of the original target time series X(t).
Application of the Hilbert Transform to a particular IMF produces an instantaneous frequency spectrum at each time point t.
H[ci(t)]=1π∫−∞∞ci(τ)t−τdτ
(1)
F(t)=12π⋅ddt(arctan(H[ci(t)]ci(t)))
(2)
A(t)=ci2(t)+H2[ci(t)]
(3)
where i = 1, …, n. The instantaneous frequency spectrum is the temporal derivative of the phase change in the IMF signal ci(t), which is defined by the inverse tangent function of the quotient between the Hilbert Transform of the original signal ci(t) (see Eq (1)) and the original signal ci(t) (see Eq (2)). The corresponding instantaneous amplitude spectrum is the root of the square sum of the original signal ci(t) and its Hilbert Transform (see Eq (3)).
We show an example of the decomposition of the velocity time series at a specific sector (Fig 1C) in Fig 1D. By definition of the EMD procedure, higher order IMFs (Fig 1D) tend to contain lower frequencies and lower amplitudes. However, in our data the instantaneous frequency and amplitude spectra at a specific sector overlapped between IMFs (Fig 1E and 1F). We computed the frequency and amplitude spectra for all IMFs in all edge sectors, which generated at each time point for each cell boundary sector six temporal frequency and amplitude values. Moreover, we repeated the HHT computation for all columns of the protrusion activity map to capture the instantaneous spatial frequency and amplitude spectra. As with the time domain, we restricted the EMD to six spatial IMFs, which generated at each cell boundary sector for each time point another six spatial frequency and amplitude values. We chose the number (six) of IMFs empirically and found it works well to capture the variation of complex cellular morphodynamics.
To illustrate the meaning of the EMD and to better interpret the related spectral decomposition outcomes, we reconstructed six movies and associated activity maps that visualize the cell edge motion captured by the six IMFs (Fig 1D, S2 Fig and and Video 2). For a particular IMF at a particular cell edge sector we extracted time point by time point the velocity magnitude and integrated the values into a displacement time series (Fig 1C). After computing the displacements for all sectors in one time point we plotted the virtual cell edge and repeated the procedure for all time points to generate a movie associated with the IMF. Each of the six movies starts with the true cell edge image at the first time point. Video 2 clearly indicates the distinct levels of motion persistence and magnitude captured by the six IMF signals. For example, IMF1 captured the protrusion signal with highest frequency and greatest magnitude, which yields rapid and jerky changes in cell shape. In contrast, IMF6 captured only subtle long-range position changes of the cell edge with almost no shape change associated. Hence, the instantaneous frequencies extracted from these different IMF orders represent, on average, different length scales and ranges of persistence in the protrusion-retraction cycles of a cell.
We first applied the spectral decomposition to the edge movements of spontaneously protruding Cos7 fibroblasts. These cells exhibited a wide range of cell shapes and morphodynamics at a basal level of activity. For example, some cells showed persistent polarity and protruded/retracted over large parts of their peripheries (top panel in Fig 2A and Video1). Other cells showed an unpolarized morphology with only small oscillatory edge movements along the entire periphery (lower panel in Fig 2A and Video 3). For the two cells illustrated in Fig 2A, we extracted histograms of instantaneous frequencies from each of the six IMFs (Fig 2B). Despite the vast differences in cell shape and motion, the two sets of histograms appeared strikingly similar. For both active and quiescent cell, the central frequencies of IMFs decreased exponentially (Fig 2C). Comparison of cumulative distribution functions (CDFs) using Kolmogorov–Smirnov (K-S) test statistics confirmed that the frequency spectra of the two cells were statistically indistinguishable (Fig 2D and S3 Fig). In contrast, the K-S test statistics of the instantaneous amplitudes were different (Fig 2E and S3 Fig). This observation also held for 48 spontaneously protruding Cos7 cells (Fig 2F and 2G). The instantaneous frequency distributions of cells with comparable molecular makeup and similar levels of stimulation were conserved regardless of morphological and morphodynamic differences. In contrast, morphological and morphodynamic differences manifested themselves in significant variations of the amplitude spectra. The more different the velocities of two cells were, the larger the difference between their instantaneous amplitude spectra (Fig 2G)). Of note, the small differences between instantaneous frequency spectra were independent of the cell order (Fig 2F). Those analyses indicate the orthogonality between instantaneous amplitude and frequency spectra in capturing cell morphodynamic behaviors.
The conservation of instantaneous frequency distribution in molecularly similar, spontaneously migrating Cos7 cells led us to ask whether induced shifts in morphogenetic signaling greater than the basal level of variation in a control cell population would systematically alter the frequency components. To address this, we employed a recently introduced optogenetic construct that allows acute and reversible inhibition of the guanosine exchange factor (GEF) Vav2[38]. We have previously shown that Vav2 acts as a core element of a signaling resonator that controls the oscillatory protrusion and retraction of cells[56]. To capture the morphodynamic response to Vav2 inhibition, we filmed cells for 6 minutes without light-activation of the inhibitor construct, followed by 12 min of pulsed blue-light inhibition, and then another 12 min in the dark to examine the recovery of Vav2 activation levels (Fig 3A). Activation pulses of three different lengths were examined: 1000 msec, 100 msec, and 1 msec (Fig 3A).
Based on the comparison of instantaneous frequency distributions, photo-inhibition with pulse lengths of 1000 msec and 100 msec changed the spectra. We did not observe any evident change with a pulse length of 1 msec (see Fig 3B). Importantly, the shifts were limited to the first three IMFs, which covered frequencies in a range 0.006–0.035 Hz (S4 Fig). Frequencies below this band were unaffected. Overall, inhibition of Vav2 signaling yielded lower frequencies, suggesting that this signal is implicated in pathways that promote fast exploratory protrusion and retraction cycles. Strikingly, scatter plots of frequency versus amplitude indicate that Vav2 inhibition has no effect on the amplitude (Fig 3C), i.e. the speed of the protrusion-retraction cycles. Those experiments suggest a separation of pathways that set the pace of the protrusion machinery from pathways that define the power of this same machinery.
In the experiments described thus far, we used instantaneous frequency and amplitude as morphodynamic signatures reflecting the state of an entire cell with sufficient sensitivity. We then asked if those signatures could also be applied to distinguish the potentially transient signaling states of subcellular regions. We first evaluated the spectral signatures of a migrating Cos7 cell with obvious polarity (Fig 4A and 4B, upper panels). The subcellular region indicated by the solid green box represents the actively protruding cell front, whereas the region indicated by dashed green box represents the retracting/quiescent cell rear. We separately applied the HHT to the time series encompassed by these two regions, extracted the instantaneous frequency distributions and conducted the K-S test to obtain K-S statistics of all six IMFs. For comparison, we repeated this analysis on two randomly selected subcellular regions of a quiescent Cos7 cell (Fig 4A and 4B, lower panels). For all IMFs, the K-S statistics comparing the front to back dynamics in a polarized cell was greater than the K-S statistics comparing two randomly selected regions of a quiescent cell (Fig 4C). The former K-S statistics also systematically exceeded the average K-S statistics quantifying cell-to-cell variability in the population of control cells (Fig 4C, black dash line) analyzed in Fig 2. However, they did not exceed the level of K-S statistics that were related to the morphodynamic shifts induced by acute Vav2 inhibition (Fig 4C, red dash line). This suggests that the signaling changes we experimentally introduced were stronger than the differences in signaling programs between front and back of a polarized cell.
To further test the postulation that spectral signatures of cell edge motion could distinguish the signaling states of subcellular regions, we compiled the instantaneous temporal and spatial frequency and amplitude spectra in a feature vector at each time point and each location and performed statistical region merging (SRM)[57] to identify regions of the cell edge with distinct motion regimens. Specifically, we formulated two 12-dimensional vectors in each sector s at each time point t (Eq (4)). The vector contains components 1 to 6 of the instantaneous temporal frequencies at time point t computed from the sector’s six IMFs along the time axis. Components 7 to 12 contain the instantaneous spatial frequencies in sector s computed from the six IMFs capturing the cell edge undulations at time point t along the space axis. The vector represented in Eq (5) captures the instantaneous amplitudes in the same fashion.
The feature vector ϕ(s,t) in each sector s at each time point t is then composed of amplitude-weighted instantaneous temporal and spatial frequencies (Eq (6)). The amplitude weights are normalized by Amax,t(s), which denotes the maximum amplitude for a specific sector along the time axis and by Amax,s(t), denoting the maximum amplitude at a specific time point along the spatial axis. We chose quadratic amplitudes because they reflect the instantaneous relative energy consumed by a particular IMF in the temporal and spatial domain. In summary, the feature vector captures the instantaneous spectral properties that characterize the local morphodynamic activity of a particular sector at a particular time point.
We exploited the feature vector to identify in the protrusion activity map regions of homogeneous morphodynamics, i.e. regions of the cell edge that move over a specific time period under the same regimen. To define such regions we applied the SRM algorithm[57]. For a multi-dimensional feature vector, this algorithm merges two regions R1 and R2 if the difference in every feature component between the two regions is less than a threshold (Eq (7)). The threshold penalizes regions of very large area and includes a user-controlled merging delicacy parameter Q (Eq (8)). |Rj| denotes the size of a region, and |Rj|max is an estimate for the largest region clustered in the map. Throughout this work, we set the value of |Rj|max to 256. Nt is the number of time frames in the cell imaging, and Ns is the number of sectors or windows along the cell periphery.
The merging started with the feature vectors in individual edge sectors and time points and iteratively grew regions with sufficient similarity in morphodynamics until none of two regions in the protrusion activity map fulfilled the merging criteria. We showed the response of SRM to different levels of merging delicacy in Fig 5A. At Q = 0, only two protrusion regimens were differentiated, while at Q = 8 the activity map was decomposed into a high number of regimens that spanned very few sectors and lasted for only a few time points. To determine an optimal value for Q, we computed the ratio of intra- vs inter-region variance as a function of Q (Fig 5B). Beyond Q = 3 the fraction of explained variance increased only marginally, indicating that this level of granularity captures the spectrum of relevant morphodynamic regimens.
To demonstrate how critical the combination of instantaneous frequency and amplitude is for the formulation of a distinguishing feature vector, we compared the SRM results of the full feature vector using the combined instantaneous frequencies and amplitudes (Fig 5A) versus the results from using the instantaneous amplitudes only (Fig 5C). We also computed SRM results using the weighted instantaneous frequency of IMF1 only (Fig 5D). It is evident that the combined frequency and amplitude features accounting for all IMFs captures much finer spatiotemporal patterns. Thus, this feature vector is effective and suitable for SRM clustering.
We applied SRM to two cells with distinct initial morphodynamics (Fig 6A, and Videos 6–7). Both cells were perturbed for 12 min by photoinhibition of Vav2 activity and then released for another 12 min. The first cell displayed a clear polarity with a morphodynamically active front between sector 20 and 50 and a more quiescent back. The difference in this activity is easily perceived in the protrusion activity map (Fig 6B, top) and, as with the cell presented in Fig 4, described by clearly separated motion regimens, where the active front breaks into two regimens (red and orange) with slightly different morphodynamic feature values (Fig 6B, bottom). The remainder of the cell edge was described by a single regimen with significantly lower feature values, reflecting the relative quiescence of this cell region. During Vav2 photoinhibition the active front was abrogated and largely merged with the more quiescent regimen. Interestingly, after release from the inhibition the higher activity regimens were restored, yet around the entire cell perimeter. Hence, while the cell regained full morphodynamic activity, it lost polarity. The second cell was less active overall and showed weaker polarity. The effects of Vav2 photo-inhibition were much harder to perceive in the protrusion activity map (Fig 6C, top), yet the region merging unveiled a clear demarcation of motion regimens before and during photo-inhibition (Fig 6C, bottom). After release from inhibition, the cell restored for short time intervals and along the entire perimeter the regimens of the more active zone before inhibition. Together, these experiments highlight the sensitivity of the instantaneous spectral decomposition to outline the spatial and temporal boundaries of distinct morphodynamic activity patterns.
Based on our finding that acute switches in Vav2 activity cause acute shifts in the instantaneous frequency spectra of cell edge motion (Fig 3), we hypothesized that the different motion regimens identified by SRM analysis could be associated with peripheral cell areas of distinct signaling activity. To test this hypothesis, we employed a Förster resonance energy transfer (FRET) biosensor probing the activity of the GTPase Rac1 in Cos7 cells, which is one of the targets of Vav2 (Fig 7A, and Video 8) and a key regulator of cytoskeleton processes implicated in cell protrusion activity. Like the construction of the protrusion activity map, we sampled the Rac1 activity locally in probing windows. Each window corresponded one-to-one with a 3 μm–wide sector for protrusion measurements and had a window depth of 3 μm. The average activity values per probing window for one time point were then pasted into the column of a matrix and the procedure repeated over all time points to generate a Rac1 activity map (Fig 7B). Next, we spectrally decomposed the protrusion activity map (Fig 7C) and performed SRM analysis on those spectral features of cell protrusion dynamics to identify distinct motion regimens (Fig 7D). Using Q = 3 we found four distinct regions, each with a different average level of Rac1 activity (Fig 7E). It should be noted that the motion regimens are transient in space and time. We visualized this behavior in a movie where the probing windows are color labeled in correspondence with their association to a particular motion regimen (Video 9; Fig 7F displays selected snapshots at certain time points).
In previous work[45], we demonstrated that cycles of edge protrusion and retraction corresponded to cycles of Rac1 activity. Dependent on the distance from the edge, the motion and signaling cycles had distinct time lags and also showed different levels of correlation, which is a measure of their mutual association. While in the past we manually or semi-manually selected regions along the edge boundary suited for correlation analysis, we wondered whether the boundary regions identified by SRM would now in a more objective manner indicate differences in the magnitude and time lag of the correlation. We performed this region-based correlation analysis in a layer of probing windows at the edge and a second layer of windows shifted into the cell interior by 3 μm, thus covering a band 3–6 μm from the cell edge. Fig 7G–7J display the correlation functions for individual sectors (blue) and their average (red) for the four identified motion regimens in the first and second layers. Both regimens 1 and 2 displayed correlation functions with significant positive lobes for negative time lags and negative lobes for positive time lags in the first layer (Fig 7G and 7H). Consistent with our previous analyses of Rac1 activation in protrusion-retraction cycles[45], this meant that in these regions Rac1 activity was delayed by ~50–60 sec relative to cell protrusion, whereas Rac1 activity was minimal ~30–40 sec prior to protrusion events. Neither regimen 3 nor 4 displayed a significant correlation function in the first or second layer, indicating overall weaker Rac1 signaling in these regions, and especially a weaker coupling between edge motion and Rac1 activity (Fig 7I and 7J). None of the four identified motion regimens displayed significant correlations between Rac1 activity in the second layer and edge motion. This is also consistent with our previous data[45], which showed a decay of spatially finer sampled correlation values to insignificant values at 4.5 μm and longer distances. The correlation functions in the first layer for regimens 1 and 2, however, showed a remarkable difference in the widths of the significant lobes. Regimen 1 had nearly symmetric lobes with a full width at half maximum (FWHM) of 40 sec, whereas regimen 2 had skewed lobes with a FWHM of 75 sec, We note that in previous analyses of correlations between molecular and cell protrusion activities such differences were obscured by the need for averaging over multiple sectors. It is tempting to speculate that the differences in signaling dynamics identified between regimen 1 and 2 are associated with different molecular programs driving Rac1 activity. With the presented SRM analysis of motion regimens, we now have the tool to systematically probe subcellular signaling activities that may even be transient in space and time, and relate them to cell morphogenesis and other cell functions.
In this work we implemented a framework for profiling cellular morphodynamics using spectral decomposition, instantaneous frequency analysis, and unsupervised clustering. First, we extracted the local dynamics of cell edge motion from time-lapse live cell image sequences by sampling protrusion and retraction velocities in discretized sectors of ~3 μm width along the cell periphery. Then, we conducted in every sector HHT-based spectral decomposition of the sampled velocity time series. The HHT resulted in several intrinsic mode functions, here fixed to six, each of which was transformed into instantaneous frequency and amplitude distributions. Hence, unlike a static spectral decomposition such as a Fourier Transform, the HHT-based decomposition captures variations in the oscillatory behavior between different time points and thereby allows detection of switches in the spectrum.
A critical question to address in our development was how much the uncertainty of mapping the displacements of a cell edge between consecutive frames affects the spectral decomposition analysis. Given the previously published mapping algorithm (see Methods), we performed a worst-case error assessment for the displacement data and then simulated how such an error level projects into the distribution of instantaneous frequencies (S5 and S6 Figs). Specifically, our algorithm for computing edge displacements at the level of single pixel guarantees topological consistency among the virtual edge markers between consecutive frames, i.e. protrusion vectors are never allowed to cross each other (S1 Fig). Hence, the maximal mapping error corresponds to the length difference Δdp of a vector that originates from a virtual edge marker at time t and targets one of the neighboring virtual marker positions at time t+1 (S5A and S5B Fig). We then defined a relative mapping error rate (S5B Fig), which for all marker points and an entire movie was approximately uniformly distributed between 0 and 8%, with an average mapping error rate of ~4% (S5C Fig). To assess the effect of such errors on the spectral analysis we randomly perturbed an actual protrusion activity map (S6A Fig) with error rates spanning from 1% to 100% (S6B–S6F Fig) and computed the K-S statistics between the frequency distributions of original and perturbed protrusion time series (S6G Fig). The simulations showed that mapping error rates of less than 10% generate deviations from the ground truth with K-S statistics less than the threshold value ~0.08 associated with the average variation of protrusion dynamics in an molecularly homogeneous cell population (Fig 2F). Hence, we were assured that the 4% level of errors from the cell edge tracking per se does not significantly contribute to conclusions from spectral analysis of morphodynamic behaviors.
Biologically, the most striking finding in this first study with HHT-based protrusion analysis is the distinct, nearly orthogonal meaning of instantaneous frequency and instantaneous amplitude spectra in terms of protrusion regulation. While the amplitude spectra report the speed of cell edge motion, the frequency spectra report how protrusion and retraction cycles are regulated. This is consistent with previous reports from our and other labs that have shown high sensitivity of measurements of protrusion persistence to perturbation of regulatory signals, whereas measurements of protrusion speed were largely unaffected by these same manipulations[44, 46, 58–60]. Cell protrusion requires on the one hand a process that stalls retraction and initiates forward edge motion. On the other hand, it requires a process that stabilizes and eventually reinforces the forward motion against increasing mechanical resistance by the environment and/or the stretched cell plasma membrane[42]. While initiation is stimulated by cell external signals or occurs spontaneously, as is the case for all data analyzed in this work, persistent edge advancement depends on the coordinated engagement of signaling pathways that converge on the activation of a series of nucleators and modulators for actin filament assembly after protrusion onset[61]. Many of these pathways are regulated by feedbacks, which integrate environmental and cell-intrinsic mechanical and chemical cues. Accordingly, dynamic or permanent changes in environmental cues, or in the pathways that process them, primarily affect the protrusion persistence. In contrast, the protrusion speed is less sensitive to the coordination of pathway engagement but more on the overall level of engagement. Moreover, the maximal velocity is reached in the early phases of protrusion, before the pathways critical for reinforcement are engaged[42, 61]. This temporal separation of molecular processes that affect protrusion speed from processes that affect protrusion reinforcement explains also the orthogonality between maximal velocity and persistence measurements. To demonstrate the orthogonality of speed-related amplitude spectra and regulation-related frequency spectra we employed a recently developed optogenetic toolkit to instantaneously deactivate and reactivate a specific node in one of the regulatory pathways. We chose the GEF Vav2, which is one of several activators of the Rac1 GTPase[38, 56] implicated in the assembly of actin filaments required for lamellipodia-driven cell protrusion[62]. Given the redundancy of Vav2 with other Rac1 GEFs, we suspected that inactivation of Vav2 would cause a rewiring of the regulatory circuitry without completely shutting down the regulation process. Indeed, we found a light-dose dependent shift in the frequency spectra. The amplitude spectra, on the other hand, were remarkably stable across the range of applied Vav2 inhibition. This underscores the orthogonality between amplitude and frequency of spontaneous protrusion-retraction cycles, allowing us to distinguish molecular processes implicated in setting the activation level of the protrusion machinery from processes that control protrusion regulation.
HHT-based profiles now introduce a refined framework to describe the state of the regulatory circuitry. Compared to the analysis of protrusion persistence, which requires time integration over multiple cycles, the decomposition of the motion signal into instantaneous frequencies allows distinction of spatially and temporally localized states of the circuitry. We therefore thought that the HHT-based spectra would allow us to define a multidimensional feature vector to distinguish edge sectors with transiently consistent morphodynamic behavior. We also hypothesized that these sectors would correspond to significantly different regulatory signaling activities, i.e. be associated with transient ‘signaling microdomains’[63]. We tested this by separately analyzing both the activation levels of Rac1 as well as the temporal correlation between Rac1 activity and cell edge motion in edge sectors belonging to distinct morphodynamic regimens. The temporal correlation is a surrogate for the coupling of Rac1 signaling with motion, i.e. how much the pathways downstream of Rac1 activation contribute to the morphodynamics analyzed by HHT-based profiling. Indeed, we found that cell edge regions with different morphodynamic behavior displayed different Rac1 activation patterns. This demonstrates that the profiling framework not only detects differences between cells with different molecular makeups, but also provides the means to identify subcellular regions with distinct signaling activities. A critical future application of this capacity will be in the analysis of signal transduction pathways implicated in the regulation of cell shape and migration. In the past, we have manually selected sectors obeying qualitative criteria of cell edge dynamics to perform image fluctuation-based analyses of signaling activities[44, 45, 47], or have included all of the cell edge independent of their dynamics. In both cases, the spatiotemporal averaging necessary to extract meaningful information from signaling fluctuations was executed over the boundaries of signaling microdomains, resulting in less sensitivity and bias. Moving forward, HHT-based morphodynamic profiling and spatiotemporal clustering of similar profiles will be used to automatically identify sub-cellular regions of homogeneous signaling activities with high resolution in both time and space. A second future application of HHT-based frequency decomposition and spatiotemporal clustering will be in the time series analysis of activity biosensor fluctuations per se. While the presented work focused on domain definitions only of motion fluctuations, the same framework could also be applied to fluctuations in signal activation throughout the entire cells. This will potentially enable the complete mapping of sub-cellular signaling regimes, and in combination with perturbations of signaling nodes, the identification of sub-cellularly distributed functions of signals, which is currently experimentally inaccessible.
In sum, HHT-based profiling and clustering will have numerous powerful applications in the quantitative analysis of cell behavior, from classifying whole-cell migration states to focusing on subcellular regulatory microdomains. The software to enable these types of analyses is accessible under the Github link: https://github.com/DanuserLab/MorphodynamicProfiling and https://github.com/DanuserLab/Windowing-Protrusion.
To identify the cell-edge location in the examples presented here, automatic thresholding was combined with morphological post-processing. Thresholds were automatically selected by fitting a smoothing spline to the image intensity histogram and by finding the first local minimum after the lowest-intensity maximum, thus selecting a threshold, which separates the low-intensity histogram mode corresponding to the background from the higher-intensity peak(s) associated with cellular fluorescence. In cases where this automatic approach failed, thresholds were manually selected. Images were pre-filtered with a Gaussian approximation of the point spread function prior to binarization by thresholding to minimize the effects of image noise. To ensure that the resulting segmentation contained only a single connected component corresponding to the cell, the thresholding was followed by automated morphological post-processing including hole-filling for small intracellular areas of low intensity, a closure operation to fill small gaps in low-intensity edge regions, and only the largest remaining connected component was retained to remove small background spots.
Cell edge velocities were derived from pixel-by-pixel matching of cell contours between consecutive time points, as described in ref. [64] and reproduced here for completeness. In summary, a B-form spline was fitted to the edge pixel positions of the segmented cell area, with nodes corresponding to each edge pixel (S1A Fig). The spline representations of two consecutive frames were then divided into segments between their intersections. To map a correspondence between the edge splines on consecutive frames, the following objective function was iteratively minimized:
(o^1,…,o^n)=argmin(o^1,…,o^n){∑i=1n[x(t+1,oi)−x(t,pi)]2+ω∑i=2n[oi(t+1)−oi−1(t+1)pi(t)−pi−1(t)]2SUMASUMB}
(M.1)
withthetopologicalconstraintse1=o1<o2<…<on=en
(M.2)
The variable n denotes the number of nodes, which in the absence of down-sampling (see below) is equal to the number of edge pixels in that segment. p1,2,…nt are the parameters of the spline at time t defining equally spaced edge nodes x(t,pi), one at each edge pixel. The goal of Eqs M.1 and M.2 is to identify n spline parameters o1,2,…nt+1 in between the intersection points e1 and en that define non-equally spaced nodes x(t+1,oi) at t+1 such that the overall displacement (SUMA) and strain, i.e. changes in spacing of nodes (SUMB) is minimized. M.2 imposes the constraint to the minimization that displacement vectors must not cross. The two sums in M.1 have different physical units. To balance them correctly we introduce a factor ω as follows:
ω=w*(SUMASUMB)iteration=1=w*{∑i=1n[x(t+1,oi)−x(t,pi)]2∑i=2n[oi(t+1)−oi−1(t+1)pi(t)−pi−1(t)]2}iteration=1
(M.3)
The factor ω is calculated only in the first iteration of the minimization, as the unit conversion by the ratio SUMA/SUMB changes insubstantially thereafter. The parameter w is a free user-control that allows the definition of the trade-off between minimal edge displacement and minimal lateral strain (S1B–S1E Fig). For w = 1 these two competing criteria have equal weight. The global solution of the edge mapping is fairly insensitive to the value of w. However, adjustments may be useful to track particularly rugged features of the cell edge, or vice versa, to oppress the mapping of spiky edge features.
The minimization of Eq M.1 can be computationally costly when the number of edge pixels in a segment exceeds 100. To circumvent this problem, we introduce a control parameter 10<Nmax<100. When the number of edge pixels in a segment is greater than Nmax, we downsample the number of nodes to Nmax, calculate the boundary displacement for this number of nodes, and then up-sample to the original number of edge pixels by interpolation. This control parameter therefore not only allows the flexibility of trading computational speed for accuracy, but allows the method to be applied to cells of any size imaged at any resolution. Once the boundary displacements are identified, the projections of these displacements onto the boundary normal vector are calculated to obtain a signed local measurement of the instantaneous normal edge velocity. The nodes are reset with every time step (S1F Fig). Accordingly, to compute a continuous path for a virtual edge marker throughout an entire movie it is necessary to interpolate marker positions for each time point (not applied in this study).0020
Our software supports two methods for sampling window creation. The first is a discrete pixel space method, which is faster and ensures windowing of the entire segmented area. The second is a sub-pixel method, which allows more flexibility and precision, but which excludes some segmented areas that do not meet strict criteria. The second method is also more computationally intensive. In both methods the intracellular frame of reference used to create the image sampling windows is based on the Euclidean distance transform D of the cell edge[65].
The discrete pixel space sampling window generation method creates sampling windows using both the discrete distance transform D and the nearest-neighbor transform or feature transform F:
di=D(ui,x1,2,…n)
(M.4)
fi=F(ui,x1,2,…n)
(M.5)
where, for the ith pixel ui in the segmented cell, fi is the index of the closest pixel on the cell boundary x, and di is the distance to this pixel and therefore the shortest distance to the cell boundary. We then also calculate the associated distance along the cell boundary for each pixel,
li=L(ui)=∑k=2fi|xk−xk−1|
(M.6)
The location of the origin of the sampling windows, x1, is determined by the user. A given sampling window can then be defined as:
Wm,p={u1,2,…,I|ui∈Ω∧ bm<di≤bm+1∧ sp<li≤sp+1}
(M.7)
where Ω is the segmented cell area, b1,2,…M are the user-selected distances from the cell edge, and s1,2,…P are the user-selected distances along the cell edge. That is, a particular window Wm,p is defined as all pixels with distances between bm and bm+1 from the cell edge, and for which the distance from the origin along the closest cell edge is between sp and sp+1. Note that in discrete pixel space it is non-trivial to define a distance measure L at a contour other than the cell boundary x. This is because with near convex cell edges the nearest feature transform, fi at positions inwards from the cell edge will not contain indexing which represents all of the pixels on the cell boundary. Therefore, the discrete windowing approach in our software package does not currently support specification of a ‘master contour’ other than the cell edge.
In the sub-pixel windowing method, the cell interior is subdivided with respect to distance from the edge by defining isocontours (or level sets) C of the Euclidean distance transform D at distance isovalues specified by the user:
Cm={u1,‥,un|D(u1,‥,un)=bm}
(M.8)
Where Cm is the mth isocontour at the distance value bm. Isocontour coordinates are refined to sub-pixel precision by linear interpolation of the original distance transform, which is calculated on the discrete pixel grid. The subdivisions of the cell interior into window slices are defined by first defining ‘slice’ start positions σ along a user-specified ‘master-contour’ Cμ:
σ={Cμ,1,‥,Cμ,n|Lμ(Cμ)∈s1,2,…P}
(M.9)
Where s1,2,…P are again the user-selected distances along the master contour μ and
Lμ(Cμ,i)=∑k=2i|Cμ,k−Cμ,k−1|
(M.10)
is the distance along the master contour from the origin Cμ,1. The position of this origin can be set by the user as well, e.g. to mark the back or front of the cell. Note that this choice has no influence on the actual geometry of the sampling windows. The ‘slice’ curves S used to subdivide the cell from these slice start positions are then determined by a maximal-gradient ascent of the Euclidean distance transform:
Sp,i=Sp,i−1+∇D(Sp,i−1)
(M.11)
with
Sp,1=σp
(M.12)
and the local gradients are again estimated via linear interpolation. The geometry of an individual sampling window Wm,p is then defined as the area enclosed by the two isocontours Cm and Cm+1 and the slice curves Sp and Sp+1. This ensures that the image area within each sampling window occupies a specific range of distances from the cell edge, and that the closest region of the cell edge is the one delineated by the intersection of gradient ascent polygons and the cell edge. Regions of the cell interior that do not meet these criteria are excluded from the windowing. This includes regions spanning ridges in the distance transform, which are therefore equally proximal to two disconnected regions of the cell edge, and regions near image borders, where the association with the cell edge is indeterminate.
In an image time-lapse sequence, the position of the sampling windows in each frame can be determined in several ways: The algorithm described above can be applied to each frame using constant isocontour and gradient ascent curve spacing, and only the location of the origin varies with time. The location of this origin is propagated between subsequent frames either by using the closest edge displacement vector or by finding the closest point on each subsequent cell edge to the original user-selected origin. In either of these cases the number of sampling window slices can vary with respect to time if the length of the cell edge changes. Alternatively, the number of sampling window slices can be held constant, allowing the width of each to vary as the length of the cell edge changes. Finally, it is also possible to allow each gradient ascent start-point to follow the edge-displacement vectors adjacent to it. This propagation method will maintain the number of window sampling slices, but will allow each slice to expand or contract as the associated region of the cell edge protrudes or retracts. This last method tends to generate the most stable window configurations. Irrespective of the propagation method chosen, each sampling window band will always maintain its distance from the cell edge. For all analyses in Fig 7 we used the setting with constant number of window sampling slices.
Once the sampling windows are generated for each image of a dataset, the associated image signals, image-derived data and edge velocities can be sampled. For image sampling, a variety of statistics are calculated (mean, standard deviation, maximum etc.) for each pixel whose center lies within a given sampling window, yielding sample statistics in the activity matrix for each sampling window at each time point. For sampling of edge velocities, statistics are calculated for the displacement vectors associated with a particular cell edge sector. For example, instantaneous cell edge velocities are calculated for each edge sector as the average of the projections of the displacement vectors onto the associated edge normal divided by the time interval between frames. Because the sample sizes per edge sector may vary with the local cell edge geometry and motion, the number of edge displacement vectors contributing to each sample is also quantified.
Cos7 cells were maintained in DMEM growth medium supplemented with 10% (vol/vol) FBS at 37°C and 10% CO2. The transient transfection of Cos7 cells were performed using Fugene 6 transfection reagent (Promega) under the guidelines of the manufacturer. The YFP-PI(WT)-Vav2(DPZ) (Addgene #86974) and mCherry-lifeAct expressing plasmids were used to photo-inhibit Vav2 and label actin[38]. For 48 spontaneously migrating cells, a plasmid expressing mCherry-lifeAct was used. To monitor Rac1 activity, cells were transfected with dual chain Rac1 biosensor Rac1FLARE.dc1g[45] that has a dTurquoise and YPet fluorescent protein pair. To obtain a fixed ratio of two chains, two consecutive 2A viral peptide sequences from porcine teschovirus-1 (P2A) and Thosea asigna virus (T2A) were inserted between two chains, leading to cleavage of the two chains during translation.
For live cell imaging, cells were plated on sterilized coverslips coated with 5 μg/mL of fibronectin (Sigma) and incubated in DMEM growth medium supplemented with 10% (vol/vol) FBS at 37°C. On the day of imaging, cell medium was replaced with L15 imaging medium (Invitrogen) supplemented with 5% (vol/vol) FBS. The coverslips with cells were placed in an open heated chamber (Warner Instruments) and live cell imaging was performed with an Olympus IX-81 inverted epifluorescence microscope equipped with an Olympus 40x UPlan FLN1.25 N/A silicon oil objective and a Flash 4 sCMOS camera (Hamamatsu) with temperature control (BC-100 20/20 Technology). For excitation, a 100 Watt mercury arc lamp with a 3% ND filter and a 510–520 (YFP) nm or 565–595 (mCherry) nm band-pass filter was employed. A 1% ND filter and a 426–446 nm band-pass filter were used with a 100 Watt mercury arc lamp (~ 1 nW/μm2 of power density at λ = 488 nm, measured at the specimen plane) for blue light pulse illumination. For emission ratio imaging of Rac1, CFP: (ex)FF-434/17, (em)FF-482/35; FRET: (ex)FF-434/17, (em)FF-550/49; YFP: (ex) FF-510/10, (em)FF-550/49 filters (Semrock) were used. Images of control cells and cells expressing the PI-Vav2, and cells expressing Rac1 biosensor were taken every 10 and 5 sec, respectively.
|
10.1371/journal.pgen.1007246 | Morphogenetic defects underlie Superior Coloboma, a newly identified closure disorder of the dorsal eye | The eye primordium arises as a lateral outgrowth of the forebrain, with a transient fissure on the inferior side of the optic cup providing an entry point for developing blood vessels. Incomplete closure of the inferior ocular fissure results in coloboma, a disease characterized by gaps in the inferior eye and recognized as a significant cause of pediatric blindness. Here, we identify eight patients with defects in tissues of the superior eye, a congenital disorder that we term superior coloboma. The embryonic origin of superior coloboma could not be explained by conventional models of eye development, leading us to reanalyze morphogenesis of the dorsal eye. Our studies revealed the presence of the superior ocular sulcus (SOS), a transient division of the dorsal eye conserved across fish, chick, and mouse. Exome sequencing of superior coloboma patients identified rare variants in a Bone Morphogenetic Protein (Bmp) receptor (BMPR1A) and T-box transcription factor (TBX2). Consistent with this, we find sulcus closure defects in zebrafish lacking Bmp signaling or Tbx2b. In addition, loss of dorsal ocular Bmp is rescued by concomitant suppression of the ventral-specific Hedgehog pathway, arguing that sulcus closure is dependent on dorsal-ventral eye patterning cues. The superior ocular sulcus acts as a conduit for blood vessels, with altered sulcus closure resulting in inappropriate connections between the hyaloid and superficial vascular systems. Together, our findings explain the existence of superior coloboma, a congenital ocular anomaly resulting from aberrant morphogenesis of a developmental structure.
| Ocular coloboma is a disease characterized by gaps in the lower portion of the eye and can affect the iris, lens, or retina, and cause loss of vision. Coloboma arises from incomplete closure of a transient fissure on the underside of the developing eye. Therefore, our identification of patients with similar tissue defects, but restricted to the superior half of eye, was surprising. Here, we describe an ocular developmental structure, the superior ocular sulcus, as a potential origin for the congenital disorder superior coloboma. Formation and closure of the sulcus are directed by dorsal-ventral eye patterning, and altered patterning interferes with the role of the sulcus as a pathway for blood vessel growth onto the eye.
| Aberrant ocular morphogenesis during embryonic development frequently results in reduced visual acuity or blindness. Morphological development of the eye begins with evagination of retinal precursors from the forebrain to produce bilateral optic vesicles and subsequent invagination of the associated ectoderm to create the lens [1,2]. Each optic vesicle reorganizes into a bilayered optic cup, with the distal (lens-facing) layer forming the presumptive neural retina and the proximal layer forming the retinal pigmented epithelium (RPE). To provide an entry point for vasculature and an exit pathway for axons of the optic nerve, a transient inferior (choroid) fissure forms along the ventral/inferior side of the optic cup and stalk. In cases where the inferior fissure fails to close, gaps remain within tissues of the eye (iris, retina, choroid and/or occasionally lens) [3,4]. This congenital anomaly, referred to as ocular coloboma, is estimated to occur in 1 out of 4–5,000 live births and cause 3–11% of pediatric blindness [4,5]. Ocular coloboma has a complex causality encompassing mutations in over 20 genes [5,6]. Although both clinically and genetically heterogeneous, coloboma predominantly affects the inferior aspect of the eye.
The posterior segment of the developing eye receives two vascular supplies [7]. The transient hyaloid vasculature is a plexus between the retina and lens, and is connected to the hyaloid artery, which enters the eye via the inferior fissure. A second circulatory system, the choroidal vasculature, grows over the surface of the optic cup to nourish the RPE and the light-sensing photoreceptor cells in the outer retina. Although development of the choroidal vessels is poorly understood, zebrafish studies demonstrated that the complex choroidal vascular plexus is preceded by a simple set of pioneer vessels [8,9]. To form this so-called superficial vascular system (distinct from the superficial retinal vessels and also known as the ciliary vasculature), three radial vessels grow over the optic cup and anastomose to create an annular vessel encircling the lens. The highly stereotypical formation of the superficial vessels suggests precise developmental regulation, but the mechanisms that guide their growth are currently unknown.
In the context of studying a large cohort of patients with ocular coloboma, we identified five local patients with a novel ocular anomaly characterized by gaps in tissues of the superior eye. Although it is logical that such an anomaly represents another fissure disorder, common models of vertebrate eye development do not feature a division in the embryonic dorsal/superior eye. However, a careful examination of zebrafish, chick, and mouse eye development did reveal a transient groove, or sulcus, bisecting the dorsal optic cup. Moreover, we utilized patient exome sequencing and zebrafish models to define the importance of dorsal-ventral patterning in morphogenesis of this ocular sulcus. Functionally, the superior ocular sulcus serves as a conduit for the advancing first vessel of the superficial vasculature, and we note profound errors in vascular growth and connectivity in embryos with abnormal sulci.
Over a six-year period (2007–2012), we identified five local patients with superior ocular defects affecting the iris, lens, retina, optic nerve and/or sclera (Fig 1 and S1 Table); notably, these were unassociated with a family history of such anomalies. On the basis of apparent similarity to coloboma (gaps in inferior/ventral ocular tissue), yet inverse orientation, we propose the term superior coloboma to describe this disorder. The first patient, with tuberous sclerosis attributable to a rare TSC2 (c.C5026T; p.R1676W) mutation, exhibited a prominent unilateral iris coloboma situated at 12 o'clock. Bilateral disease was present in a single patient (#2), and involved both iris and lens (Fig 1, images 2 and 3). Two of the five patients were diagnosed in infancy, and for one (#4), examination under anesthesia was required to fully characterize pathology. As is evident from Fig 1, the diversity of tissue involvement in superior colobomata recapitulates that present in inferior colobomata. We subsequently received, from pediatric ophthalmologists at US and UK tertiary referral centers, clinical data on three further patients with superior colobomata. These cases extended the range of associated phenotypes to include additional structural ocular malformations (microphthalmia, or small eye; #8). All eight patients in our cohort had profoundly reduced visual acuity, precluding normal stereopsis.
To identify candidate genetic variants carried by superior coloboma patients, exome sequencing was performed on the initial five probands (S2 Table). Identified variants were prioritized by comparison to SNP databases (frequency <1%), in silico prediction algorithms (Mutation Taster >0.95) and expression within the developing eye or previously identified connections to coloboma (see methods). We focused our efforts on understanding genetic alterations in the single patient with bilateral superior coloboma (#2, Table 1). In particular, we noticed that patient #2 carries compound heterozygous variants in the Retinoic Acid (RA) synthesis gene CYP1B1 [10](S1A Fig) as well as a rare (dbSNP: 1 in 60,706; NHLBI and 1000 Genomes: 0 in 14,000) missense variant in Bone Morphogenetic Protein Receptor 1A (BMPR1A, S1B and S1C Fig). As RA and BMPs are morphogens with essential roles in eye development, including regulation of inferior fissure closure [11–17], we hypothesized that the identified mutations contributed to the patient’s ocular disorders. In order to examine how disruption of eye patterning genes could lead to superior coloboma, we next turned to animal models and conducted an in depth analysis of dorsal eye morphogenesis.
Inferior coloboma arises from failed closure of the choroid fissure located in the ventral eye. Given the comparable phenotype despite opposite orientation seen in superior coloboma patients, we hypothesized a similar etiology. Although the standard model of eye development describes an uninterrupted dorsal retina, two older studies of fish eye development identified a groove present in this space [18,19]. To determine if such a structure exists broadly across vertebrates and whether it is a Laminin-lined space, we chose to revisit the study of dorsal eye morphogenesis in fish, chick and mouse. Using multiple microscopy methods, we identified a transient groove/sulcus in the dorsal zebrafish eye [dorsal in fish and superior in human are equivalent; for consistency with superior coloboma, we describe this structure as the superior ocular sulcus (SOS)] (Fig 2A–2D). The sulcus is visible by stereoscope but more obvious in compound or confocal observations of live embryos (Fig 2A), and most easily discernible from 21–25 hpf. When imaged under an electron microscope, the SOS can be seen to transect the distal portion of the dorsal retina (Fig 2B), while single confocal optical slices reveal the SOS as a distinct space (Fig 2C) lined by basal lamina (Fig 2D).
To ascertain whether a similar structure exists in chick, we examined tissue sections immunostained for Laminin and counterstained with DAPI. At stage HH16, we observed the presence of a Laminin-lined division in the distal portion of the chick dorsal optic cup (n = 6/8 eyes; Fig 2E, S2 Fig). For evidence of a comparable structure in mammals, we next examined mice and found a Laminin-lined separation across the inferior portion of the dorsal optic cup at embryonic day 10.5 (Fig 2F). A collaborator also shared older SEM studies of newt (Taricha tarosa) development, which similarly demonstrate the presence of a division across the dorsal embryonic eye (S3 Fig, personal communication, A. Jacobson). Thus, we present clear evidence for the existence of an evolutionarily conserved, Laminin-lined sulcus in the dorsal optic cup of multiple vertebrate species.
The inferior fissure temporarily bisects the ventral retina prior to closing through progressive fusion of the nasal and temporal margins of the ventral optic cup [3]. The SOS similarly extends across the dorsal zebrafish retina (Fig 2A–2D) to partially separate the nasal and temporal retinal lobes, and is also present only transiently. To determine the mechanism of SOS closure, we followed ocular morphogenesis over time. The SOS arises soon after optic cup formation (19–20 hpf) as a distinct and narrow structure (S1 Video and Fig 2A and 2B). Notably, formation of the sulcus occurs at a time when the developing retinal pigmented epithelium is spreading around the optic cup, but is not associated with significant cell movement or apoptosis in the forming dorsal retina (S2 Video). Unexpectedly, the edges of the narrow SOS do not migrate toward one another and fuse, but instead the SOS transitions at 22–24 hpf to a shallow and wide trough and gradually disappears after 26 hpf (S3–S5 Videos and S4 Fig). Both phases are visible in representative scanning electron microscopy images (Fig 2B). As we observed the transition from narrow to wide, and never detected an epithelial fusion event, it is logical to propose that the sulcus closes via cell rearrangement or shape modification, mechanisms distinct from the epithelial fusion that occurs within the choroid fissure.
CYP1B1 mutations cause ocular malformation and are a major cause of congenital and adult glaucoma [20]. Patient #2 carries one of the known disease-causing alleles (R368H) while the second allele is a truncation (A287Pfs6), and so both alleles are expected to be pathogenic. Retinoic acid can be synthesized through both the Cyp1b1 and Aldh pathways [10,21], and mRNA encoding both types of RA synthesis enzyme is expressed in the dorsal zebrafish eye (S5A Fig)[12,22]. In order to test whether Cyp1b1 is necessary for SOS closure, we used TALEN mutagenesis to create zebrafish carrying a 13 bp frameshift deletion within the P450 domain, resulting in an early stop codon and a truncated protein. Surprisingly, the zebrafish cyp1b1 mutants did not display defects in sulcus closure, even when the Aldh pathway was additionally inhibited (S5B Fig)[23]. Given the lack of a phenotype with reduced RA signaling, we next investigated the BMPR1A variant and Bmp-dependent regulation of SOS closure.
Bmp ligands (Gdf6/Bmp13 and Bmp 2, 4, and 7) pattern the eye at the time of SOS closure [11–14,24,25] and the identified BMPR1A patient variant alters a highly conserved residue in the kinase domain (p.Arg471His, S1 Fig); therefore, we tested whether reduced Bmp receptor activity affects closure of the SOS. The small molecule DMH1 is an inhibitor of type IA BMP receptors, with robust and specific activity in zebrafish [26,27]. Embryos were treated with DMH1 either just after gastrulation or just prior to optic cup invagination (10 and 18 hpf, respectively) and evaluated for SOS presence at 28 hpf, a time point when the sulcus is no longer visible in wildtype embryos. Exposure to DMH1 prevented SOS closure in a dose-dependent manner (Fig 3A and 3B), establishing that Bmp signaling regulates sulcus morphogenesis.
We next used a zebrafish overexpression assay to evaluate whether the patient variant disrupts BMPR1A function. As injection of one-cell stage embryos with wildtype human BMPR1A mRNA failed to elicit alterations to dorsal-ventral axis specification, we used site-directed mutagenesis to introduce a Q233D mutation previously shown to render BMPR1A constitutively active [28]. Injection of mRNA encoding the constitutively active BMPR1A receptor (caBMPR1A) efficiently induced ventralization of whole zebrafish embryos, while caBMPR1A carrying the patient variant (R471H-caBMPR1A) showed mildly reduced activity (Fig 3C and S6 Fig). The patient variant therefore does not completely inactivate the protein, but this assay does suggest that it could be a hypomorphic allele and may have been one of multiple factors contributing to the development of superior coloboma. Overall, our data support a role for Bmp signaling in regulating SOS closure.
Within the zebrafish eye, Bmpr1a mediates signaling from the Gdf6a (Growth Differentiation Factor 6a, Bmp13) ligand [29] and absence of Gdf6a results in almost complete loss of dorsal (superior) ocular genes, expansion of ventral (inferior) genes, and a small eye phenotype [12,13,30]. Knockdown of Gdf6a signaling in wildtype embryos by injection of antisense morpholino oligonucleotides caused a highly penetrant SOS closure defect, very similar to that seen with DMH1 exposure (Fig 4A–4C). Recapitulation of the persistent sulcus phenotype in both homozygous [12,13] and a subset of heterozygous gdf6a embryos (Fig 4D and 4E) shows that SOS closure is sensitive to the precise level of Bmp signaling. A lack of Gdf6a also affected formation of the SOS, as seen by the deeper sulcus in a representative SEM image (Fig 4D, bottom right panel) and in animations showing the surface morphology of the dorsal eye in 22 hpf wildtype (S6 Video), gdf6a heterozygous (S7 Video) and gdf6a homozygous (S8 Video) embryos. While the sulcus eventually closes in most Gdf6a-deficient embryos, two adult gdf6a-/- fish displayed superior colobomata (Fig 4F), demonstrating that an early closure defect can lead to the disease phenotype.
There are diverse outputs of Gdf6a signaling, regulating cellular functions such as apoptosis, cell proliferation, and dorsal-ventral retinal patterning [12,13,17,31]. Because proliferative defects are visible after sulcus closure and apoptotic cells are not concentrated near the SOS [31], we reasoned that dorsal-ventral retinal patterning is the Gdf6a function most essential for SOS closure. During development, dorsal ocular Bmp signaling is balanced by midline Sonic Hedgehog (Shh) activity [16,32], and gdf6a-/- mutants exhibit an expansion of the Shh downstream gene vax2 into the dorsal retina [13]. We therefore tested whether increased Shh signaling in Bmp-deficient embryos underlies the persistent SOS phenotype. Indeed, treatment of gdf6a-/- and gdf6a+/- embryos with the Shh inhibitor cyclopamine significantly rescued the delayed closure phenotype (Fig 5A and 5B). Cyclopamine treatment also partially rescued patterning in the dorsal retina, as it restored the tbx5 expression domain in gdf6a heterozygotes (Fig 5C and 5D). These data support the idea that SOS closure is dependent on proper pattern formation within the developing retina and that sulcus morphogenesis is regulated by a balance of ventral Shh and dorsal Bmp signaling pathways.
Transcriptome analyses of Gdf6a-depleted retinas have highlighted critical regulators of dorsal-ventral patterning within the zebrafish eye [31]. Using this dataset, we interrogated the superior coloboma patient exome data, and identified a variant in TBX2 (p.Pro329His). Zebrafish tbx2b is expressed in the dorsal eye in a Gdf6a- and BMP-dependent manner (Fig 6A)[13]. To analyze the function of zebrafish tbx2b in regulating sulcus morphogenesis, we compared dorsal eye morphology between wild type embryos and tbx2bfby (from beyond) mutants [33]. We note a statistically significant increase in the proportion of embryos displaying an open SOS in tbx2bfby mutants compared to wildtype embryos at 28 hpf (Fig 6B and 6C). Such experimental results support a model in which dorsal-ventral patterning within the embryonic eye provides essential cues for morphogenesis of the SOS.
The inferior fissure demarcates the boundary between nasal and temporal retinal lobes and allows for ingrowth of blood vessels into the developing eye, both of which are also logical functions for the SOS. Alignment of the SOS with naso-temporal markers was examined in gdf6a+/- embryos because of their easy-to-visualize sulcus and undisturbed nasal-temporal patterning. In situ hybridization with probes for foxg1a (nasal retina) and foxd1 (temporal retina) demonstrates that the expression boundaries align with the position of the sulcus. Although the SOS lies at the division between nasal and temporal retina (Fig 7A), its significance in separating retinal domains or, conversely, the role of nasal-temporal patterning in establishing the location of the sulcus remain to be tested.
Vascular inputs to the developing zebrafish eye include both the hyaloid artery that extends through the inferior fissure to form a plexus behind the lens, and the superficial vasculature that grows over the eye and encircles the lens [8,9]. The two systems are connected ventrally by the hyaloid vein. We hypothesized that the SOS forms a channel for the dorsal radial vessel (DRV, the first vessel of the superficial vasculature) as it grows over the dorsal retina and toward the lens. Indeed, SEM imaging shows a vessel extending into the SOS, and both DIC and confocal time-lapse imaging demonstrate that the nascent DRV grows through the sulcus (S3 and S9 Videos and Fig 7B–7D).
If the SOS functions to direct the DRV toward the lens, then altered sulcus morphology and dynamics would be expected to modify vascular development. Since our data demonstrate that Bmp signaling regulates SOS closure, we therefore evaluated development of the superficial vasculature in embryos lacking Gdf6a. The DRV does form in gdf6a-/- mutants and extends through the abnormally deep SOS; however, compared to control embryos, the DRV is of reduced caliber and unbranched at 26 hpf [gdf6a-/-: 0±0 branch points (n = 11) vs. siblings: 1.3±1.0 branch points (n = 24)] (Fig 8A, 8C and 8D). The gdf6a mutants always form a single DRV, compared to approximately half of control embryos where two DRV converge in the SOS (see control 41 hpf embryo in Fig 8A, Fig 8D). Moreover, instead of its normal course around the lens, the DRV in gdf6a-/- embryos projects deeply and ectopically travels dorsal to the lens to connect with the hyaloid vasculature (Fig 8A, 8B and 8E). The DRV subsequently degenerates in most gdf6a mutants, but the ectopic vessel remains as a dorsal connection between the superficial annular vessel and the hyaloid plexus (Fig 8A, 8E and 8F). Imaging of Tg(rx3:GFP;kdrl:mCherry) embryos revealed that the deep sulcus in gdf6a-/- mutants creates a notable divot in the optic cup immediately dorsal to the lens (Fig 9A). In all cases (n = 8), the forming ectopic vessels grew directly into this space between the dorsal edge of the lens and the retina. Given the defects observed for the DRV in Bmp-deficient embryos, we conclude that dorsal retinal patterning is necessary for superficial vascular pathfinding.
Patterning of the ventral retina is regulated by Shh [16,32], and our earlier data suggest a balance between Bmp and Shh signaling impacts SOS morphogenesis. In contrast to Bmp loss, cyclopamine inhibition of Shh signaling in wild type embryos resulted in a shallow SOS that closes early (Fig 9B), and an increased proportion of embryos with multiple DRVs spread across the dorsal retina (Fig 9B and S7 Fig). A similar change in growth of the DRV was noted previously in embryos where the Shh receptor Smoothened is non-functional [34]. In summary, disrupted dorsal-ventral patterning of the retina leads to profound alteration of the superficial vasculature.
Aberrant vasculature in gdf6a-/- mutants or cyclopamine-treated embryos could result either from a direct role of the morphogens in regulating vascular pathfinding or from altered SOS dynamics. To determine whether the SOS itself directly influences growth of the superficial vasculature, we prevented the Gdf6a-dependent sulcus defects by manipulating Hedgehog signaling. Indeed, cyclopamine treatment of gdf6a-/- mutants both rescues SOS closure defects and precludes ectopic connection with the hyaloid vasculature (Fig 5 and S7 Fig). Similarly, loss of Gdf6 rescues the DRV overgrowth phenotype observed in cyclopamine-treated embryos (S7 Fig). Therefore, the data support a model in which proper SOS formation and closure are necessary for DRV pathfinding.
In this manuscript, we classify superior coloboma as a separate disease with a developmental origin distinct from, but comparable to, inferior coloboma. Eight patients display gaps in tissues of the superior eye, including retina, lens, and iris. We demonstrated the existence of a transient dorsal groove in vertebrate eye development that is conserved amongst fish, chick, newt and mouse. Failure to close the superior ocular sulcus can result in adult zebrafish displaying a phenotype that resembles superior coloboma. Furthermore, it supports the evolutionary conservation of the SOS amongst vertebrates, an evolutionary distance of some 450 million years.
There are rare reports in the scientific literature of patients with “atypical” coloboma [35–39], ocular anomalies contrasting with the position of the known inferior embryonic fissure. The vast majority of such cases (macular coloboma, aniridia, or nasally/temporally oriented iris coloboma) are unlikely to arise from defects of sulcus closure. However, at least two of the described atypical coloboma patients display iris colobomata with a superior orientation [38,39]. Although the embryonic mechanism was originally considered anomalous, our identification of the SOS provides a likely explanation for the unusual coloboma identified in these two patients.
Exome sequencing of our superior coloboma patients identified rare variants in the genes encoding the type 1 BMP receptor and transcription factor T-box 2. In the absence of multigenerational pedigrees of affected patients, we are unable to causally link such variants to the incidence of disease. However, the connection between Bmp signaling and inferior fissure morphogenesis is well established. Indeed, variants in GDF6 (BMP13), BMP4, and SMOC1 are linked to inferior coloboma and microphthalmia [5,6,40–42]. Furthermore, zebrafish, Xenopus, chick, and mouse studies have demonstrated a key role for Bmp signaling in optic cup morphogenesis, apoptosis, proliferation, and dorsal-ventral eye patterning [11–13,40,41,43–45]. Consistently, abrogating Bmp signaling either by DMH1 treatment or loss of Gdf6a results in profound SOS closure defects. Beyond the gdf6a homozygous mutant phenotype, we also detected a partially penetrant sulcus closure defect in the otherwise morphologically normal gdf6a heterozygotes, arguing that the sulcus is particularly sensitive to the levels of Bmp signaling. Further, loss of Tbx2b function in zebrafish fby mutants leads to comparable aberrations in SOS morphogenesis. Such data, taken together with the detrimental nature of the patient BMPR1A variant, support a model whereby Bmp signaling modulates SOS closure via regulation of target genes such as tbx2.
Research on ocular Bmp signaling defines roles in regulating eye precursor cell number, apoptosis, proliferation, and dorsal-ventral gene expression [12,13,17,31,41,44,46,47]. However, apoptotic cell populations are not localized to the SOS, and proliferative defects are present only after SOS closure [31,47]. Furthermore, we note that gdf6a+/- heterozygotes display aberrant sulcus closure, yet lack apoptotic or proliferative defects. In contrast, gdf6a+/- heterozygotes display detectable alterations to dorsal-ventral gene expression, providing a correlation between patterning and SOS closure defects. To further test the role of dorsal-ventral patterning in sulcus dynamics, we asked whether rescue of the patterning defects in gdf6a-/- mutants would also promote SOS closure. Given the expansion of inferior markers into the superior retina of gdf6a-/- mutants [12,13], and the rescue of SOS defects with Shh inhibition, we conclude that the aberrant closure of the SOS in Gdf6- and Tbx2-depleted embryos is linked to dorsal-ventral patterning defects of the vertebrate eye.
The identification of a patient with two variants in CYP1B1 prompted us to carefully examine retinoid signaling in SOS closure. A role in ocular morphogenesis is well established for the retinoid signaling pathway, with mutations in the RA synthesis gene ALDH1A3 known to cause inferior coloboma [48]. Furthermore, RA regulates proliferation and migration of periocular mesenchyme (POM), a neural crest- and mesoderm-derived cell population that modulates inferior fissure closure. The inability of extensive zebrafish experiments to reveal a role for cyp1b1 in SOS closure, even in the context of Gdf6a deficiency (S8 Fig) may reflect the greater complexity of the family of retinoid synthesis enzymes in humans and their distinct expression patterns compared to zebrafish. Given the proximity of RA signaling to the SOS and known roles for RA in regulating morphogenesis in other systems, it remains plausible that RA signaling contributes to the causality of human superior coloboma.
Eye morphogenesis and patterning are dependent on multiple signaling pathways, in addition to Bmp and RA. For example, overexpression of the Wnt inhibitor Dkk1 results in loss of dorsal ocular gene expression [49], and mutation of the Wnt receptor FZD5 (thought to function as a receptor for both canonical and non-canonical Wnts) causes inferior coloboma [50]. In examining the prioritized list of rare variants identified in superior coloboma patients, we note rare variants in NKD1, CELSR2, FZD4, SCRIB, and WNT9B (components of canonical or non-canonical Wnt pathways). The rare TSC2 (Tuberous Sclerosis Complex 2/Tuberin) variant in patient #1 plausibly implicates other cellular mechanisms in the induction of superior coloboma. TSC2 complexes with TSC1 to regulate the mTOR signaling pathway [51], and loss of either gene leads to unregulated cell growth and proliferation.
The rare incidence of superior coloboma argues that the disorder is unlikely caused by simple, single-gene inheritance. Rather, a model incorporating multi-gene inheritance or incomplete penetrance is more plausible. Seven of the eight patients with superior coloboma in the current study display unilateral disease, also a common characteristic of inferior coloboma [52]. The highly penetrant defects found in zebrafish gdf6a mutant larvae, which only infrequently result in an adult superior coloboma phenotype (Fig 4F), are consistent with an impressive ability of the developing eye to recover from embryonic defects. However, the absence of an obvious coloboma does not preclude abnormal SOS morphogenesis generating subtler abnormalities, such as vascular misrouting. Although defining the relative contribution of heritability and environment is challenging, other disorders offer potential insight. Characterized by appreciable globe enlargement, high myopia represents an ocular disorder with substantial genetic and environmental components, where unilateral cases account for up to one third of the total [53]. Anisometropia represents a second example of an asymmetric developmental ocular phenotype [54], and the pattern apparent in the current cases (Fig 1) corresponds with such examples.
The parallels with the inferior ocular fissure, which provides a passageway for the hyaloid vasculature [7,9,55], are strong. The close coordination between the development of both structures is highlighted by the ability of the dilated hyaloid vein in zebrafish lmo2 mutants to disrupt fissure closure and cause inferior coloboma [56]. The tight association between the superficial vasculature’s DRV and the SOS provides convincing evidence that the SOS serves a similar retinal vascular guidance function. While developing blood vessels follow guidance cues in the same manner as growing axons [57], our data argues that the physical landscape of a tissue can also direct angiogenesis. First, the DRV in wildtype embryos travels directly through the SOS to reach the lens, whereas only a thin and unbranched DRV grows through the particularly deep sulci of gdf6a-/- mutants. Second, the shallow or absent SOS in cyclopamine-treated embryos correlates with the appearance of multiple DRVs spread across the dorsal retina. Finally, the divot above the lens in gdf6a-/- mutants aligns with the position of the ectopic connection between hyaloid and superficial vasculature. Taken together, these data support a model in which the SOS provides a path for directing and restraining DRV growth (Fig 10).
Here, we have characterized a previously unrecognized developmental structure with a significant disease connection. Further studies will be needed to discern the exact mechanisms of sulcus formation and resolution, and to more deeply analyze the causes of superior coloboma.
All experiments were conducted in accordance with the guidelines provided by the institutional IACUC. The research on chick and mouse was performed at the University of Texas and approved by their institutional AC committee (#2015–00089). The research on zebrafish was performed at the University of Alberta and approved by ACUC, Biosciences (#0082) and ACUC, Faculty of Medicine (#1476). Anesthesia and euthanasia were performed with MS-222.
Whole exome sequencing was performed on genomic DNA from each proband (#1 - #5) as part of FORGE Canada Consortium at the McGill University and Génome Québec Innovation Centre. Exome target enrichment was performed using the Agilent SureSelect 50Mb (V3) All Exon Kit and sequencing was performed on the Illumina HiSeq 2000, multiplexing three samples per lane. The mean coverage of coding sequence regions, after accounting for duplicate reads was greater than 70x. WES data was analyzed by performing alignment with BWA, duplicate read removal with Picard, local indel realignment with GATK, variant calling with SAMtools, and annotation with Annovar and custom scripts [58]. Subsequently exome sequencing was repeated commercially (Beijing Genomics Institute). In parallel, array CGH was performed to identify any causative copy number variations (CNV) using an Affymetrix cytoscan HD array that comprises approximately 1,800,000 CNV and 700,000 genotyping probes. Within patients #1–5, we identified 783, 843, 942, 708, and 721 rare (<1%) non-synonymous and stop-gain/loss variants, respectively. By filtering such variants using Mutation Taster (score >0.95), patients #1–5 contain 163, 155, 139, 112, and 148 higher probability variants, respectively. Subsequent prioritization included literature searches associating genes with ocular function and zfin.org examination of in situ expression patterns within the developing eye at 18–24 hpf, yielding a restricted subset of high priority variants in each proband (S1 Table).
Zebrafish were cared for according to standard protocols, and embryos grown in embryo media at 28.5°C and staged appropriately. Zebrafish embryos grown past 24 hours post fertilization (hpf) were treated with 0.004% 1-phenyl 2-thiourea (PTU; Sigma-Aldrich, St. Louis, MO) to prevent pigment formation. The AB strain of wild-type (WT) fish, Tg(rx3:GFP), Tg(kdrl:eGFP)la116, and Tg(kdrl:mCherry)ci5 transgenic lines [59–61], and cyp1b1 (see below), gdf6as327 [13], and tbx2bfby [33] mutant lines were used. The gdf6as327 mutation encodes a S55X truncation producing a 54 amino acid peptide lacking the mature domain. The tbx2bfby mutation is a point mutation resulting in a T>A substitution, resulting in a premature stop codon within the T-box sequence. In situ hybridization (ISH) was performed as described previously [12] with embryos fixed overnight at 4°C in 4% paraformaldehyde (PFA) and permeabilized by incubation in 10 μg/mL Proteinase K for 20 minutes for 28 hpf embryos and 1 min for 5.3 hpf (50% epiboloy) embryos. Following in situ hybridization, eyes from 28 hpf embryos were removed and mounted under a coverslip in 70% glycerol, and then photographed using a Zeiss Axioimager Z1 compound microscope and an Axiocam HR digital camera (Carl Zeiss Microscopy, LLC). 5.3 hpf embryos were imaged using an Olympus stereoscope and a Qimaging micropublisher camera.
For the chick studies, fertilized Leghorn eggs (Texas A&M, Bryan, TX) were incubated at 38°C in a humidified forced-draft incubator. Chick embryos were staged according to Hamburger and Hamilton [62] and Swiss Webster mice were collected at E10.5
Immunohistochemistry was performed as previously described [63]. Chick embryos were stained with antibodies against Laminin-1 (#3HL1; Developmental Hybridoma Studies Bank, Iowa; Conc: 1:250), whereas for mouse Laminin alpha 1 stains, we utilized a different antibody (Sigma L9393). Alexa-Fluor conjugated Goat anti-Rabbit IgG (#411008; Life Technologies,: Conc: 1:250) was used for fluorescent detection [64]. Antibodies used in the current study were validated for use in chicks in previous studies [64,65]. DAPI staining was used for detecting nuclei. Confocal images were obtained with an Olympus IX51 spinning disc microscope and data analyses carried out with Slidebook Pro (3I, CO). Images are presented as single 0.5–0.8 μm thick optical sections. The position in the dorsal-ventral plane is based on the acquisition of multiple serial sections and respective alignment to those sections (just ventral) that contain lens tissue.
A gdf6a Morpholino (5’-GCAATACAAACCTTTTCCCTTGTCC-3’) was used to block splicing at the exon 1–intron 1 boundary of the gdf6a pre-mRNA transcript [41]. 5–10 ng were injected into one-cell stage wildtype zebrafish embryos.
TALEN mutagenesis constructs targeting the Cyp1b1 cytochrome P450 domain (nts 924–977) were created by Golden Gate cloning [66]. The target region was TTCGGGGCCAGTCAAGACACtctgtctacagctCTCCAGTGGATCATCCTGCTA, with the spacer region shown in small letters. 100 pg of RNA for each TAL construct were injected into one-cell stage zebrafish embryos and the offspring were screened for mutations by HRM. The 13 bp deletion causes a frameshift at aa317 (p.Cys317SerfsX23), followed by a stop codon at aa340. The wildtype protein is 526aa.
The offspring of gdf6a heterozygous incrosses were genotyped by high resolution melt (HRM) temperature analysis performed on genomic DNA extracted in 10 μL of 50 mM NaOH (95°C, 10 minutes) and neutralized with 2 μL Tris-HCl, pH 8.0. PCR was performed using primers optimized for HRM (GCGTTTGATGGACAAAGGTC; CCGGGTCCTTAAAATCATCC) and Qiagen Master Mix on a Qiagen Rotor Gene Q qPCR machine (Qiagen). Conditions for amplification were 1 cycle at 95°C for 5 min, 40 cycles of 95°C for 10 seconds, 55°C for 30 seconds, followed by HRM ramp from 70–90°C, 0.1°C per step. Fish carrying the cyp1b1 mutation were HRM genotyped using primers CCATCTCAGATATTTTCGGGG and GTTATTTACCTGACAAGTAGCAG and a 52°C annealing temperature for amplification. Results were analyzed via Qiagen software v2.02 (Qiagen) and variants initially confirmed by Sanger sequencing. tbx2bfby mutants were genotyped by PCR followed by MseI restriction digest. Genomic DNA was extracted as above and diluted 10X for use as template. PCR was performed with Ex Taq DNA Polymerase (TaKaRa Bio Inc.) using the following primers: Forward-TGTGACGAGCACTAATGTCTTCCTC; Reverse-GCAAAAAGCATCGCAGAACG. Conditions for amplification were 1 cycle at 94°C for 2 min, 40 cycles of 94°C for 15 seconds, 58°C for 15 seconds, and 72°C for 20 seconds, followed by 1 cycle at 72°C for 3 min. The PCR products were then digested with MseI (NEB) for two hours and analyzed via gel electrophoresis using a 3% agarose gel.
For Laminin staining, whole embryos fixed in 4% PFA were permeabilized with ice-cold acetone (7 min), washed in water (5 min), and PBS with 0.5% Tween-20 (PBST, 5 min). Embryos were then treated for 5 min with 10 μg/mL Proteinase K in PBST, washed four times in PBST, blocked for 1 hour at room temperature in 5% normal goat serum and 2% BSA, and then incubated overnight at 4°C in block plus rabbit anti-Laminin primary antibody (1/100, L-9393, Sigma-Aldrich). After washing, embryos were incubated at room temperature for two hours in goat anti-rabbit Alexa Fluor 488 or 568 secondary antibody (1/1000, Molecular Probes). All embryos were washed 4 times for 10 minutes in PBST after primary and secondary antibody incubations. Eyes were removed and mounted on slides in 70% glycerol, or were imaged as whole embryos mounted in 1% low melting temperature agarose.
Embryos at 22 hpf from AB (wildtype) and gdf6a+/- incrosses were fixed overnight in 2.5% Glutaraldehyde; 2% Paraformaldehyde. After washing in 0.1M phosphate buffer, embryos were gradually dehydrated in ethanol, transferred to Hexamethyldisilazane (HMDS; Electron Microscopy Sciences) and left to dry overnight. Embryos were then mounted on SEM stubs, sputter coated with Au/Pd using a Hummer 6.2 Sputter Coater (Anatech), and imaged on a XL30 scanning electron microscope (FEI) operating at 20 kV.
The full coding domain of human BMPR1A mRNA (Ensembl transcript ID: ENST00000372037) was cloned into pCS2+. Using site-directed mutagenesis, the Q233D mutation was created to make caBMPR1a [28] and the patient mutation, G1412A, was subsequently added to create R471H-caBMPR1a, with sequences confirmed by Sanger sequencing. The constructs, which were identical except for the patient mutation, were linearized (NotI, NEB) and mRNA generated using the SP6 mMessage mMachine kit (Ambion). mRNA was purified using YM-50 Microcon columns (Amicon, Millipore) and concentration determined through spectrophotometry. The mRNA, which was newly synthesized for each round of injections, was diluted with DEPC-treated water, and embryos were injected with 25 pg of RNA at the single-cell stage. Both injections and analysis were performed in a blinded fashion.
Primers were validated prior to the experiment, as previously described [67]. Endogenous control primers (elongation factor 1a; ef1a), previously used [67], were chosen from the Universal Probe Library Assay Design Centre for Zebrafish (Roche). Primer sequences for BMPR1A are as follows: F-CGTGTTCAAGGACAGAATCTGG; R-AAAGGCAAGGTATCCTCTGGTG. RNA was isolated from 40 embryos per group at the 256-cell stage using RNAqueous (Ambion), first-strand cDNA synthesis was performed using AffinityScript qPCR cDNA Synthesis (Agilent) and qRT-PCR was performed as described previously [67].
For drug treatments, embryos were transferred at 10 hpf or 18 hpf into 35 mm dishes containing 5 mL embryo media plus 0.05–0.2 μM DMH1 (Sigma-Aldrich), 1–5 μM N,N-diethylaminobenzaldehyde (DEAB; Sigma-Aldrich), 0–10 μM cyclopamine (Sigma-Aldrich), or an equivalent volume of vehicle (DMSO or ethanol). For each experiment, 15 embryos were added to each dish and two dishes were used for each treatment (total of 30 embryos). For analysis of SOS presence, embryos were grown at 28.5°C to 28 hpf, treatment conditions were blinded and the embryos were analyzed under a Zeiss Discovery V8 80x stereoscope for the presence of the SOS. Alternatively, live or fixed embryos at 22–54 hpf were mounted in 1% low melting temperature agarose for imaging.
Two-factor analysis was done by Students t test. Multivariable analysis was performed by two-tailed, one or two-factor ANOVA with Tukey posthoc test. *P<0.05, **P<0.01, ***P<0.001.
Fixed or anaesthetized live transgenic embryos were mounted laterally in a 35mm dish in 1% low-melting temperature agarose and imaged using a Zeiss W Plan-Apochromat 20x/1.0 water immersion objective and a Zeiss LSM700 laser scanning unit mounted on a Zeiss Axioimager Z1 compound microscope. Z-stacks were made by taking optical slices at intervals of 2–3 μm for a total of ~60 μM, and maximum projections or surface projections were created from the resulting stacks using either ZEN (Carl Zeiss) or Imaris (Bitplane) software.
All DIC images were taken on an Axiocam HR digital camera mounted on a Zeiss Axioimager Z1 compound microscope. Photos and videos were annotated, assembled and processed for brightness and contrast in Adobe Photoshop software.
FORGE Canada Consortium: Finding of Rare Disease Genes in Canada; Steering Committee: Kym Boycott (leader; University of Ottawa), Jan Friedman (co-lead; University of British Columbia), Jacques Michaud (co-lead; Université de Montréal), Francois Bernier (University of Calgary), Michael Brudno (University of Toronto), Bridget Fernandez (Memorial University), Bartha Knoppers (McGill University), Mark Samuels (Université de Montréal), Steve Scherer (University of Toronto).
|
10.1371/journal.pcbi.1003009 | Origin and Evolution of Protein Fold Designs Inferred from Phylogenomic Analysis of CATH Domain Structures in Proteomes | The spatial arrangements of secondary structures in proteins, irrespective of their connectivity, depict the overall shape and organization of protein domains. These features have been used in the CATH and SCOP classifications to hierarchically partition fold space and define the architectural make up of proteins. Here we use phylogenomic methods and a census of CATH structures in hundreds of genomes to study the origin and diversification of protein architectures (A) and their associated topologies (T) and superfamilies (H). Phylogenies that describe the evolution of domain structures and proteomes were reconstructed from the structural census and used to generate timelines of domain discovery. Phylogenies of CATH domains at T and H levels of structural abstraction and associated chronologies revealed patterns of reductive evolution, the early rise of Archaea, three epochs in the evolution of the protein world, and patterns of structural sharing between superkingdoms. Phylogenies of proteomes confirmed the early appearance of Archaea. While these findings are in agreement with previous phylogenomic studies based on the SCOP classification, phylogenies unveiled sharing patterns between Archaea and Eukarya that are recent and can explain the canonical bacterial rooting typically recovered from sequence analysis. Phylogenies of CATH domains at A level uncovered general patterns of architectural origin and diversification. The tree of A structures showed that ancient structural designs such as the 3-layer (αβα) sandwich (3.40) or the orthogonal bundle (1.10) are comparatively simpler in their makeup and are involved in basic cellular functions. In contrast, modern structural designs such as prisms, propellers, 2-solenoid, super-roll, clam, trefoil and box are not widely distributed and were probably adopted to perform specialized functions. Our timelines therefore uncover a universal tendency towards protein structural complexity that is remarkable.
| Proteins are vital and central macromolecular players necessary for the functioning of the cell. The redundant and highly conserved structural makeup of proteins reflects their ability to act as genomic repositories of evolutionary history. These structures are fundamental subjects for the study of molecular evolution. Structural biologists have demonstrated the existence of a wide array of compact 3-dimensional fold structures, the protein domains. Their classification resulted in hierarchical taxonomies that describe protein fold space, most notable SCOP, CATH and FSSP. Studies have shown that certain types of protein shapes are more abundant than others and this uneven distribution implicates processes by which new shapes are discovered. Our evolutionary genomic research explores the evolution of protein domains at the deeper levels of classification. However, we have not embarked in a systematic study of the origin and evolution of general structural designs. These designs include topologies such as sandwiches, bundles, barrels, prisms, solenoids, and propellers. The appearance and diversification of general structural designs and their confirmation from published literature defines a unique chronology of structural innovation. The study also uncovers a recent trend of architectural sharing between Archaea and Eukarya and benchmarks the phylogenomic analysis of CATH domains with SCOP domains.
| The polypeptide chains of proteins generally fold into highly ordered and well-packed three-dimensional (3D) atomic structures [1]. These protein folds represent spatial arrangements of more or less wound helices (generally α-helices) and extended chain segments (β-strands) that are separated by flexible loops and relatively rigid regions in the form of turns and coils. Helices are stabilized by local main-chain (backbone) hydrogen bonding interactions. In turn, β-strands establish main-chain interactions with other strand elements that are distant. Parallel and antiparallel arrangements of β-strands form β-sheets, which often curve to form open and closed barrel structures. Folds are generally defined by the composition, architecture and topology of their core ‘helix’ and ‘sheet’ secondary structure elements [2]. The satisfaction of the hydrogen bonding potential of main-chains gives rise to regular secondary and super secondary structural elements in globular proteins. Analysis of protein folds indicates that those that occur frequently tend to adopt regular architectures, such as the αβ Rossmann folds, α/β-barrels, β-sandwiches, and bundles [3]. Main-chain hydrogen bonding is also important for the formation of complex turns and coils that link α-helices and β-strands.
Protein domains are compact, recurrent, and independent folding units of protein structure that sometime combine with other domains to form multi-domain proteins. They are considered evolutionary units and are the basis for several protein structure classification schemes. Two of them, CATH and SCOP, are accepted as gold standards and share a number of common features [4]. SCOP [5] is a largely manual collection of protein structural domains that aims to provide a detailed and comprehensive description of the structural and evolutionary relationships of proteins with known structures. In contrast, CATH [6] uses a combination of automated and manual techniques, which include computational algorithms, empirical and statistical evidence, literature review and expert analysis. Both classifications are hierarchical but dissect 3D structure differently, focusing more on either evolutionary or structural considerations [4]. SCOP unifies domain structures that are evolutionarily related at sequence level (>30% pairwise residue identities) and are unambiguously linked to specific molecular functions into fold families (FFs), FFs with common structures and functions with a common evolutionary origin into fold superfamilies (FSFs), FSFs with similarly arranged and topologically connected secondary structures (not always evolutionarily related) into folds (Fs), and finally Fs that share a general type of structure into classes. CATH unifies domain structures hierarchically (bottom-up) into sequence families (SFs; analogous to FFs), homology superfamilies (Hs; analogous to FSFs), topologies (Ts; analogous to Fs), architectures (As), and protein classes [6] (see also Figure 1 for comparisons of SCOP and CATH levels of structural abstractions). Multi-linkage clustering groups domains into SFs based on sequence similarity. SFs with structures that are thought to share common ancestry and can be described as homologous are grouped into Hs. H structures sharing patterns of overall shape and connectivity of secondary structures are grouped into Ts. T structures that share and overall shape of the domain structure according to the orientations of the secondary structures but ignoring their connectivity are unified into As. Finally, A general shapes are grouped into four protein structural classes, mainly-alpha, mainly-beta, alpha-beta and few secondary structures [6].
Protein structures are evolutionarily conserved and capable of preserving an accurate record of genomic history [1], [7]. They represent ‘relics’ of molecular evolution [2] and express the greatest levels of redundancy and reuse that exist in molecular biology [8]. Many studies have been conducted to unfold the evolution and diversification of protein domain structures and proteomes of extant organisms [1], [9]–[11]. Structural phylogenies describing the evolutionary relationship of SCOP F, FSF and FF domains were built by data-mining the census of structures in hundreds of genomes [12]–[15]. Timelines of F, FSF and FF appearance were derived from the phylogenetic trees and revealed the existence of three epochs in protein evolution, ‘architectural diversification’, ‘superkingdom specification’ and ‘organismal diversification’. A universal core of domain structures that is central for cell function was the first to unfold in the timelines during the architectural diversification epoch. During the superkingdom specification epoch, patterns of reductive evolution in the domain repertoire consistently segregated the archaeal lineage from the ancient community of organisms and established a first organismal divide. Finally, the appearance of eukaryotic and archaeal signature domains marked the start of the organismal diversification epoch and the rise of domain structures specific to proteome lineages. Finally, trees of proteomes (i.e. trees of life) placed Archaea at the root and confirmed this organismal supergroup represents the most ancient superkingdom of life [7], [16].
While we have studied how F, FSF and FF domains appeared and distributed in the world of organisms, we have not embarked in a systematic study of the origin and evolution of general structural designs. Here we study how these designs evolve in trees of domain structures, this time focusing on the CATH classification. The appearance and diversification of general protein structural designs at A-level (e.g., sandwiches, bundles, barrels, solenoids, propellers) and published literature define in this study a unique chronology of structural innovation. Structural phylogenies of domains at T and H levels of structural abstraction uncover global evolutionary patterns of structural distribution in the world of organisms. The study benchmarks previous phylogenetic analysis of SCOP-defined domains and again reveals the early origin of the archaeal superkingdom. Congruent patterns of diversification derived from protein structure provide remarkable support to the ancient history of the cellular world, and trees of life confirm the primordial evolutionary patterns.
Domain structures are unevenly distributed in the world of proteins and proteomes [1]. They distribute differently in superkingdoms Archaea (A), Bacteria (B) and Eukarya (E) and can be pooled into seven taxonomical groups depending on whether they are unique to a superkingdom (A, B and E) or are shared by two (AB, AE and BE) or three superkingdoms (ABE). The taxonomical groups can be visualized in a simple Venn diagram (Figure 2). Bias in the relative number of domains structures corresponding to each taxonomical group persists regardless of the classification used (CATH or SCOP) or the level of structural abstraction of the classification scheme (Figure 2). This bias cannot be attributed to non-vertical patterns of inheritance (e.g. the effect of horizontal transfer) since research groups have confirmed independently that convergent evolution is relatively rare (∼2–12%) at these high levels of structural conservation (e.g., [17], [18]).
Distribution biases among taxonomical groups show some striking features. First and as expected, higher taxonomical levels show higher levels of structural sharing between superkingdoms (especially ABE) than lower taxonomical levels, confirming the contention that they are evolutionarily more conserved. Second, the highest level of structural abstraction (CATH A) does not contain a single superkingdom-specific taxonomic group, suggesting that these groups represent sets of structures that are late evolutionary additions to the protein repertoire. Finally, ABE and BE domain structures are consistently the dominant taxonomic groups at all hierarchical levels, from FF to A. This final observation suggests they represent the most ancestral and common taxonomical groups. The most parsimonious corollary of these evolutionary patterns of domain distribution is that the ancient BE taxonomical group must arise by loss of archaeal-specific domain structures, suggesting Archaea is the most ancient superkingdom. As we will now show, this suggestion can be confirmed by phylogenomic reconstruction.
We generated phylogenomic trees describing the phylogenetic relationship of 38 A, 1,152 T and 2,221 H domain structures (Figures 3 and 4). Tree distribution profiles and metrics of skewness suggest significant cladistic support (P<0.01). The trees were well resolved. However, internal branches for trees of Hs and Ts were poorly supported by bootstrap analysis, an expected outcome with trees of this size. Chronologies of evolutionary appearance [7] of CATH domain structures were derived directly from the phylogenomic reconstructions. The relative age of domains (nd) was measured on the trees as a relative distance in nodes from the hypothetical ancestor of domains at the base of the trees, and used to build the timelines. Since our method produces rooted trees that are highly unbalanced and reject the Yule and random speciation models [19] and since molecular speciation in our trees has clock-like behavior and does not depend on changes in domain abundance [20], nd was considered a good and most-parsimonious proxy for time. To study how domain structures distribute in proteomes, we calculated a distribution index (f), the number of species that use each structure given on a relative 0–1 scale. The f index was plotted along the timelines of domain structures, i.e. against nd (Figure 5). Three As (ndA = 0–0.068), fifteen Ts (ndT = 0–0.061) and fifteen Hs (ndH = 0–0.049) were present in all proteomes examined (f = 1) and were the most ancient in the timeline. A list of the fifteen Hs is given in Table S1. The f of As decreased with increasing age. The f of Ts and Hs decreased with their increasing age until f approached zero at ndT = 0.55 and at ndH = 0.55, respectively. We term these ages “crystallization points” of the T and H structural chronologies, borrowing the idea of a phase transition from physics. At these time points, a steady decrease in f results in a large number of structures being specific to a small number of organisms. After crystallization, an opposite trend takes place, in which Ts and Hs increase their representation in genomes. In contrast, the architectural chronology that describes the appearance of As remained unaffected by the crystallization event since the losing trend of As started at ndA = 0.56–0.60 but rarely reached zero (Figure 5).
To uncover hidden patterns of organism diversification in our dataset, we divided structures according to their distribution in superkingdoms and constructed three separate structural chronologies for the genomes of each superkingdom at A, T and H levels of structural abstraction (Figure 5). Taxonomical groups of domain structures were identified in the time plots with different colors. We previously observed that a superkingdom must ‘lose’ a significant number of SCOP structures before the evolutionary appearance of the first superkingdom-specific ‘signature’ structure [7]. In our study, this loser trend of domain structures was also observed for the CATH annotated genomes in each superkingdom. This observation strengthens our claim of reductive evolution in protein domains of the lineages that emerge from the cellular urancestor (the last universal common ancestor; LUCA) that we find is functionally complex [11]. The loser trend of SCOP and CATH structures reveals the primordial birth of Archaea followed by the birth of Bacteria and Eukarya. For example, the complete loss of Hs first starts in Archaea (ndH = 0.176) with the membrane-bound lytic murein transglycosylase D (chain A) H domain (3.10.350.10). Its appearance is congruent with the loss of the first SCOP FSF in Archaea (ndFSF = 0.174), the LysM domain (d.7.1), observed in previous studies [7]. Both domain definitions are very much similar in how they describe functions in the cell. Analysis of domain distribution in Archaea shows that the vast majority of ancient Ts and Hs that were lost in proteomes were present in all superkingdoms (ABE; colored grey). These were followed by AB (orange), A (wine) and few AE (red) structures, most of which started to appear after the crystallization point and during the superkingdom specification and organismal diversification epochs [7]. Clear decreases in structural representation (f-value) also occurred in Bacteria and Eukarya, but involved fewer and younger structures. Analysis of domain distribution in Bacteria shows that AB and B structures (dark yellow) started to increase representation after the crystallization point, leading towards their diversification and specification. Similarly, the eukaryotic chronology showed that comparatively younger architectures [e.g. BE (blue) and E (green)] increased their popularity among the eukaryal lineages. The appearance and distribution of the seven taxonomical groups of T and H structures was unfolded in the timelines using boxplots describing the range of ndT and ndH values and measures of central tendency for each group (Figure 6). Only domains shared by the three superkingdoms (ABE) span the entire chronology, from the origin of proteins (nd = 0) to the present (nd = 1). These structures represent instantiations of the domain content of the urancestor but their late appearance may also indicate events of horizontal transfer between lineages. Boxplots for BE, AE and AB explain relationships among superkingdoms over time. The BE boxplot is the most ancient of the three, suggesting Archaea diversified early by reductive evolution. The A, B and E boxplots reflect the history of ‘signature’ structures that are unique to individual superkingdoms. These signatures appear first in Bacteria and then concurrently in Archaea and Eukarya, an observation that is congruent with timelines derived from SCOP domains [7]. Despite its early specification, Archaea tends to acquire Archaea-specific structures very late in evolution and their number is limited when compared to Bacteria and Eukarya. This may stem from very strong adaptive pressures that were historically imposed by lifestyle. Archaea are very simple organisms that usually live in harsh and extreme environments [21]. We believe their extremophilic lifestyles impose constraints on their molecular make up that: (i) limit the possibility of acquiring new structures, and (ii) induce a constant selective pressure to maintain a minimal structural set necessary for survival. We therefore propose that Archaea maintained a minimal set of structures while losing structures by strong reductive evolution. We note that signature As exhibit very low f values, suggesting these molecular designs were acquired as adaptations to new environments and lifestyles. The appearance of structures shared by only two superkingdoms was also revealing. For example, the AE boxplot's upper whisker approached ndH = 1, implying a recent relationship between Archaea and Eukarya. Comparatively, the nd values for SCOP FSFs for the AE taxonomical group was ndFSF = 0.85, supporting the late appearance of the interaction [7]. Note that a sister relationship between Archaea and Eukarya is usually used to claim the canonical bacterial rooting of the tree of life [22], but that in our studies this relationship is only supported by domain structures that are quite derived. It is also noteworthy that the early loser trend in the BE taxonomic group, made explicit by smooth decreases in f-values in the timeline, occurs in the absence of signature domain structures specific to superkingdoms (Figure 5). This weakens other evolutionary scenarios of superkingdom origin, including chimerism mediated by massive horizontal gene transfer (endosymbiosis or fusion) processes, and the possibility that phylogenetic signal of these events (e.g. those between Bacteria and Eukarya) would make Archaea appear artificially ancient in phylogenomic reconstructions (see below).
We previously reconstructed trees of proteomes from a genomic census of SCOP domains and made inferences about the rooting of the tree of life [7], [11], [16]. We found trees of proteomes reconstructed from ancient domain structures were rooted paraphyletically in Archaea while trees reconstructed using derived structures exhibited the canonical rooting with Bacteria emerging at their base. We also revealed how parasitic and symbiotic lifestyles can complicate phylogenetic interpretation [7], [16]. The proteomes of organisms that are parasitic or that establish symbiotic relationships with other organisms have frequently experienced reductive evolution, discarding enzymatic and cellular machineries in exchange for resources from their hosts. Since their inclusion can lead to incorrect phylogenetic trees, we excluded proteomes from all but 295 free-living (FL) organisms and reconstructed rooted trees that most parsimoniously describe their evolution. The FL set included 41 archaeal, 189 bacterial, and 65 eukaryotic organisms. The tree of FL proteomes reconstructed from a census of H domain structures supported the trichotomy of the superkingdoms (Figure 7). The number of bacterial proteomes was however overrepresented in the FL-tree and could cause long-branch attraction during phylogenetic reconstruction possibly leading to incorrect deep phylogenetic relationships. Since taxon sampling can also affect phylogenomic inference [23], we randomly sampled equal numbers of proteomes per superkingdom (a maximum of 41) and generated replicated trees of proteomes. Reconstruction of equally sampled FL proteomes improved tree resolution and bootstrap support values of deep branches. More importantly, the trees consistently showed a paraphyletic rooting in Archaea and the derived placement of monophyletic Bacteria and Eukarya (Figure 7). We also reconstructed trees of FL proteomes from three subsets of phylogenetic characters: ancient H structures common to all superkingdoms corresponding to the architectural diversification epoch (ndH<0.176), H structures of intermediate ancestry corresponding to the superkingdom specification epoch (0.176<ndH<0.55) and H structures that are derived and reflect the organismal diversification epoch (0.55<ndH). The proteome tree reconstructed from the most ancient H structures was rooted paraphyletically in Archaea, reflecting their early segregation through the minimalist strategy. Reconstructions from H structures of intermediate ancestry produced trees with three clades corresponding to the three superkingdoms that were rooted in Archaea. Finally, reconstructions from H structures that were derived yielded the canonical tree of life rooted in Bacteria. It is noteworthy that the rooting of these trees reflects the early appearance of Bacteria-specific domain structures (Figure 7, see trees reconstructed using most ancient, ancient and younger characters sets). We note the split of Archaea in three groups in the tree reconstructed from ancient H structures. We believe this anomaly stems from using subsets of characters in phylogenomic reconstructions and from the existence of a ‘modern effect’ [11] imposed by relatively recent changes in abundance of domain structures belonging to the ABE taxonomic group. Both factors impoverished phylogenetic signal and obscured deep phylogenetic relationships. The modern effect is an embodiment of recent evolutionary processes affecting ancient repertoires, the effects of which must be identified and removed when reconstructing the set of domain structures present in the urancestor [11].
The structural chronology, especially at H level, unveils a relatively recent (perhaps ongoing) sharing of protein architectures between archaeal and eukaryal genomes. The timeline reveals that while AE domain structures appeared for the first time when Archaea and Eukarya acquired their superkingdom-specific signature structures, the vast majority of them appeared quite late in evolution (e.g., Figure 6D). This was unanticipated. This finding inspired us to resolve the phylogenetic contribution of each structural character set in the tree of proteomes. Interestingly, characters that are shared by archaeal and eukaryal genomes exhibited high retention index (RI) values (Figure 8), indicating that the sharing pattern did not result from annotation artifacts. The RI measures the amount of synapomorphy (features that are shared and derived) expected from a data set that is retained as synapomorphy on a cladogram. Boxplots of structural character sets shared by the seven taxonomical groups were also plotted (Figure 8). Since low RI values indicate high levels of homoplasy (i.e. non-vertical phylogenetic signal), the low values of bacterial signature structures confirm the high incidence of horizontal gene transfer that exists in the bacterial superkingdom. In turn, the relatively high RI levels of the common ABE group is surprising. Most members of the group include very ancient structures (Figure 8), many of which were part of the urancestor. High RI levels in this taxonomical group challenge the common assumption that horizontal transfer was rampant during early life [22].
These RI boxplots are powerful enough to explain the relationships of superkingdoms in our tree of proteomes. The AE boxplot is the only one exhibiting very high RI values. In turn, bacteria-specific characters had the most dispersed RI boxplot. Hence, archaeal and eukaryotic lineages share good signal characters that are very recent and are widely present; their high f values indicate for example their presence in most of archaeal and eukaryotic proteomes (Figure 5C). More than 30 years ago, Woese and Fox [24] defined the existence of three ‘aboriginal’ lines of descent – superkingdoms Archaea, Bacteria and Eukarya. The microbial Archaea and Bacteria lines were conceptualized as ‘urkingdoms’ of deep origin that were qualitatively different from the eukaryotic kingdoms. This prompted reconstructions of a tripartite tree of life and later proposals of the early rise of Bacteria with rooting determined using paralogous gene couples (e.g., EF-Tu/EFG). This classical (canonical) tree topology induces sister lineages corresponding to Archaea and Eukarya and an exclusive common ancestor of both. Many archaeal components involved in informational systems (e.g. translation, replication and transcription) and transmission of genetic information show a higher sequence similarity with their eukaryotic homologue than their bacterial homologue [25], [26]. For instance, more than 30 ribosomal proteins are shared between the Archaea and Eukarya that are not present in Bacteria [27]. Moreover, Archaea and Eukarya also share a similar base excision repair system that is different than the system in bacteria [28]. If the phylogenetic signal in the sequence of these RNA and protein molecules adequately depicts history, these findings would explain the evolutionary link between Archaea and Eukarya and the topology of the canonical tree of life that emerges in some phylogenetic studies from their close relationship. However, many genes do not share the archaeal and eukaryal link and the canonical root must be considered contentious. Remarkably, the tree of proteomes reconstructed using the modern structural character set in our experiments (Figure 7, epoch III or younger character sets) is the only tree with the canonical topology that places the root branch in Bacteria. This topology mostly results from protein domain structures of very recent origin that are shared between Archaea and Eukarya. We contend that these very recent domains retain good phylogenetic signal, especially in their sequences, and will be less affected by processes of mutation saturation. Consequently, the close evolutionary relationship of Archaea and Eukarya in trees of life derived from analyses of these sequences [22], [24] can be considered an artifact of the focus on sequence.
Current trees of life built for example from sequence concatenation, such as those in refs. [29], [30], include genes encoding for multidomain proteins (e.g. aminoacyl-tRNA synthetases). Some of these domains are of recent origin and may fall within the derived domain set we have analyzed. We claim that strong phylogenetic signal in the sequence of these domains likely drives the reconstructed topologies. Instead, weak phylogenetic signal embedded in the sequences of older and universal domains is swamped by the recent archaeo-eukaryotic signal that is in part responsible for the canonical tree. Our focus on CATH domain structure (not gene sequence) can dissect the differential contribution of old and recent protein domains that belong to the proteome-encoding gene repertoire. A similar focus on deep phylogenetic signal in RNA structure has also shown the basal placement of Archaea in phylogenetic reconstructions from tRNA, RNase P RNA and 5S rRNA [31]–[35], including analysis of paralogy in tRNA [35]. For example, a timeline of accretion of helical RNA substructures of RNase P complexes showed the most ancient substructures were universal and harbored the core catalytic activities of the endonuclease [34]. However, the first substructures that were lost were specific to Archaea and this episode occurred before molecules were accessorized with superkingdom-specific substructures. The early origin of Archaea was also shown in trees that describe the structural evolution of RNase P RNA, which placed archaeal molecules at its base. These results obtained by studying the evolution of RNA structure clearly parallel the evolutionary patterns of CATH domain accumulation of this study. Clearly, deep phylogenetic signal in protein and RNA structure is free from the limitations of gene sequence and associated non-vertical patterns arising from horizontal gene transfer but more importantly from domain rearrangement and can therefore reveal historical patterns without bias [36]. Here we show the importance of considering the age heterogeneity of a biological repertoire, in this case the proteome, when making phylogenetic statements.
The architectural chronology of As is evolutionarily more conserved than chronologies of Hs and Ts (Figure 5A). The timeline shows that As are widely shared and are refractory to loss in genomic lineages. In fact, very few As are lost in superkingdoms (4 in Archaea, and one each in Bacteria and Eukarya) and are thus very old and popular in the world of organisms. The 3-layer (αβα) sandwich (3.40) is the most abundant and ancient of all proteins. The orthogonal bundle (1.10) and the α/β-complex (3.90) are equally abundant and are the second and third most ancient architectures. Remarkably, the phylogenomic tree of As shows that comparatively simpler shape structural designs are more favored than complex designs and in general are more ancient, appearing at the base of the tree. Architectural complexity was here evaluated on empirical grounds by focusing on the topology and regularity of spatial arrangements of secondary structures in a structural design. For example, the most ancient 3.40 and 1.10 architectures involve simple arrangements of secondary structure that can be very diverse in different structural variants while more recent shape designs are spatially more convoluted and regular (Figures 3). As time progresses the complexity in architectural make up of structural designs also increases (Figure 3). The few As that are lost in superkingdoms are quite complex and as expected their appearance is quite derived. The first loss occurred in Eukarya (ndA = 0.76) with the very complex Clam architecture, and then in Archaea and Bacteria. We note that Archaea loses four As quite late and in a row, showing that the pervasive reductive trends of Archaea described above extend almost to the present. This also reflects the conservative nature of extremophilic Archaea, which are not in need of modern structural designs. Bacteria loses the most recent A structural design, Box (2.80), at ndA = 1, which is shared by both archaeal and eukaryal genomes. Box is involved in nucleotide excision repair, a molecular function that has a unique place in cellular defense because of its wide substrate range and its ability to virtually remove all base lesions from a genome. Ögrünç et al. [28] reported a similar base excision repair system used in Archaea and Eukarya and argued that a different set of proteins are employed by the bacterial nucleotide repair system. Interestingly, the f index for Box in Archaea (f = 1) and Eukarya (f = 0.96) again indicates a recent sharing of structural designs between archaeal and eukaryal organisms. Architectures constitute the second highest level of structural abstraction in CATH, and because of their high conservation it is difficult to clearly delimit the three epochs of the protein world. In contrast, our results indicate CATH H and SCOP FSF are the most suitable levels to uncover the evolution of domain structures in genomes. These levels of abstraction are structurally and evolutionarily conserved. They preserve deep phylogenetic signatures and are variable enough to dissect evolutionary history of proteomes and molecular functions.
To obtain a detailed view of architectural discovery and usage over time, we grouped As into 10 major structural designs: sandwiches, bundles, barrels, prisms, horseshoes, rolls, solenoids, propellers, complexes and other (a category with structural designs that could not be clearly grouped into the main categories) (Table 1 and Figure 9). We found that most sandwiches, bundles, barrels, complexes and rolls have high f values (f∼1) and rather simple structural designs (Figure 9). In turn, structural designs such as propellers, horseshoes, solenoids (2 Solenoid, 2.150), prisms, trefoil and box, have low f values (f = 0.85–0.10) and are very complex. Under the assumption that widespread and abundant designs are old, complex folds appear to have evolved later than simpler folds. We also mapped the appearance of T and H structures harboring individual A designs, plotting ndH and ndT values for Hs and Ts belonging to each of the 38 known As (Figure 10).
The structural makeup of the most ancient 3-layer (αβα) sandwich (3.40) architecture (Figure 3) represents the central theme of the most ancient SCOP FFs [37]. These structures consist of repeating α-β-α supersecondary units, such that the outer layer of the structure is composed of helices packing against a central core of parallel β-sheets. Many enzymes, including most of those involved in glycolysis, are α/β layered proteins and are cytosolic [38]. These α/β structures harbor repeats of the α-β-α arrangement (e.g., the α-β-α-β-α sequence). The β-strands are parallel and hydrogen bonded to each other, while the α-helices are all parallel to each other but are antiparallel to the strands. Thus the helices pack against the sheet forming a sandwich-like structure. We note that the β-α-β-α-β (αβα) subunit, often present in nucleotide-binding proteins, represents the Rossmann structural motif found in proteins that bind nucleotides, especially the cofactor NAD(H) [39].
The orthogonal bundle (1.10) and α-β-complex (3.90) appear immediately after the 3-layer (αβα) sandwich (3.40) design. The orthogonal bundle consists of a 3–4 α-helix bundle and is found in a number of different proteins, most of which associate with membranes. Due to physical constraints imposed by the lipid bilayer of membranes the list of possible membrane protein structures is limited to either bundles [40], [41] or barrels [42], [43]. In many cases the α-helices are part of a single polypeptide chain and are connected to each other by three loops. In the 4-helix bundle proteins the interfaces between the helices consist mostly of hydrophobic residues while polar side chains on the exposed surfaces interact with the aqueous environment. A number of cytokines consist of 4-helix bundles, such as interleukin-2, interleukin-4, human growth hormones, and the granulocyte-macrophage colony-stimulating factor (GM-CSF) [38] and DNA binding proteins (e.g., transcription factors, repressors proteins) [44]. CATH has grouped the complex shaped structures into the ‘complex’ bin, until alternative assignment methods are developed. The α/β-complex architecture groups together all those designs that include significant α and β secondary structural elements in a mixed fashion. Examples of α/β-complex proteins include bacterial and mammalian pancreatic ribonucleases [45], Zn metallo-proteases and DNA topoisomerases [46]. Two kinds of barrel structures are the most ancient and abundant in the protein world, the α/β-barrel (3.20) and the β-barrel (2.40) [6], and both appeared at about the same time (ndA = 0.13). The α/β-barrel is composed of eight α-helices and parallel β-strands that alternate along the peptide backbone. The α/β-TIM barrel is the most prominent example of α/β-barrel and is widely present in enzymes of central metabolism [47]. A β-barrel is a large β-sheet that twists and coils to form a closed structure in which the first strand is hydrogen bonded to the last. β-strands in β-barrels are typically arranged in an antiparallel fashion. Barrel structures are commonly found in porins and other proteins that span cell membranes and in proteins that bind hydrophobic ligands in the barrel center, such as lipocalins [48]. The roll is a complex nonlocal structure in which 3–4 pairs of antiparallel β-sheets, only one of which is adjacent in sequence, are ‘wrapped’ in 3D space to form a barrel shape [49]. Rolls appear for the first time at ndA = 0.3.
A number of distinct and more complex architectures appear later on in the chronology, including solenoids, horseshoes, prisms, propellers and trefoils. Solenoid proteins, with their arrays of repeating motifs, tend to have elongated structures that contrast with the majority of globular proteins whose polypeptide chains follow more complex trajectories [50]. These are constructed from tandem structural repeats arranged in superhelical fashion, a feature that is important for many cellular processes [51]. Solenoid proteins constructed from HEAT repeats [52] and armadillo repeats [53], [54] constitute the principal transport receptors. A key structural property that differentiates solenoid proteins from other structured proteins is the lack of contacts between distal regions of protein sequence (sequence-distal contacts). For this reason, solenoid proteins are often more flexible than other structured proteins and this flexibility is an important feature of their specific functions [50]. The solenoid structure appears for the first time at ndA = 0.46. The α-horseshoe protein appears at ndA = 0.4, is a super helical structure made up of a number of three α-helical orthogonal bundle repeats. The α-β horseshoe appeared at ndA = 0.56, consists of several α/β-repeating units [55]. The structure of the ribonuclease inhibitor, a cytosolic protein that binds strongly to any ribonuclease that may leak into the cytosol, takes the concept of the repeating α/β unit to the extreme [55]. The structure is made of a 17-stranded parallel β-sheet curved into an open horseshoe shape, with 16 α-helices packed against the outer surface. Prisms are similar to solenoids in geometry but completely different in connectivity. A more self-contained β-sheet forms each face of a triangular prism. They appear late at ndA = 0.86. The trefoils consist of an unusual β-sheet formed by six β hairpins arranged with three fold symmetry into ‘Y’ like structures [56] and are also quite derived (ndA = 1).
The most ancient and popular architecture, the 3-layer (αβα) sandwich (3.40), harbors the most ancient and abundant topology, the Rossmann fold (3.40.50) and the most ancient and abundant superfamily, the P-loop containing nucleotide triphosphate hydrolases (3.40.50.300). Despite differences of topology and ranking within databases [57], this H structure of CATH is analogous to the “P-loop containing nucleotide triphosphate hydrolase” FSF (c.37.1) of SCOP [4], since both have Rossmann fold topology and also agree on their keyword definitions. A careful analysis of CATH and SCOP structures phylogenies show that the ancient domains structures at T (3.40.50) and H (3.40.50.300) levels are in global agreement with timelines of F (c.37) and FSF (c.37.1) structures [7]. Despite differences in domain definitions of tertiary structure in CATH and SCOP, the remarkable conservation of evolutionary signal indicates both classification systems effectively preserve evolutionary information in protein structure and uncover global patterns of origin and diversification that are for the most part congruent.
We note that levels of structural abstraction above H and FSF unify domains that may not be necessarily homologous. In other words, T and A in CATH and F in SCOP may show episodes of structural convergence. This could complicate evolutionary interpretations. The fact that the same evolutionary patterns observed using H domain structures in this study (and FSF structures in previous studies; reviewed in [1]) could be recovered at higher levels of the structural hierarchy is encouraging and suggests that the influence of convergent processes at these higher levels is limited and that the classifications do in general a good job in capturing true evolutionary information.
In this study we follow the history of protein fold structures and proteomes in the tripartite world of organisms. Instead of generating trees of life from protein sequence with standard methods, we use a genomic structural census and robust cladistics methods to build trees of domain structures and proteomes. Structural phylogenies describing the evolution of CATH domains at A, T and H levels of structural abstraction revealed patterns of reductive evolution and the three epochs in the evolution of the protein world that were previously proposed [7]. Structural diversification patterns match those observed in the analysis of SCOP domain structures [7], [16], [58]. Reconstruction of phylogenomic trees of proteomes describing the evolution of lineages confirms Archaea is the most ancient superkingdom.
Provided assumptions of our phylogenomic method are considered valid, six major findings summarize novel results and take advantage of the ability of CATH to better describe topological features of protein structure:
We note that these conclusions entrust CATH with the ability to properly apportion domain structures in fold space and are only valid if assumptions of character argumentation are valid.
Our trees of domain structures define timelines that trace back the history of innovation, diversification and distribution of protein structural designs. Our finding that protein architectures tend to become more complex in evolution is very significant. In a previous study, analysis of β-barrel structures revealed that the curl and stagger and complexity of the connectivity of supersecondary structures increases in evolution [12]. The very early appearance of multilayered sandwich structures is also compatible with the finding that the most ancestral folds share a common architecture of interleaved β-sheets and α-helices [12]. An even more recent study shows that 36 out of the 54 most ancient FFs harbor α/β/α-layered sandwich structures [37]. The very early appearance of the P-loop hydrolase motif in the first FF, the ABC transporters, was associated with a built-in lateral bundle, which resembles the trans-membrane domains of transporter proteins. This suggests that first proteins contained sandwich and bundle structures and were associated with the membranes of primordial cells. Remarkably, P-loop hydrolase folds and bundles make up important membrane complexes, such as ion channels and transporters. Their very early origin highlights a crucial links between the origin of proteins and the origin of cells.
Phylogenomic trees describing the evolution of domain structures and proteomes were reconstructed using a census of domain abundance in proteomes using PAUP* version 4.0b10 [59]. Figure S1 presents a flowchart of the methodology. CATH annotations for the proteomes of 492 fully sequenced genomes (42 Archaea, 360 Bacteria and 90 Eukarya) were retrieved from Gene3D [60]. We used CATH version 3.3 and its corresponding Gene3D assignments. Table S2 lists the organisms studied and Table S3 lists the subset that is free-living and was used in phylogenomic analyses. Gene3D is a repository of manually curated HMM predictions with a false positive prediction rate of only 0.2–0.6%. As with SUPERFAMILY [9], [61], a repository of SCOP domain predictions, proteomes deposited in Gene3D were searched against HMM libraries using the iterative Sequence Alignment and Modeling System (SAM) method. Data matrices of genomic abundance (G) of domains at A, T and H levels were assembled for phylogenetic analysis. Empirically, G values represent numbers of multiple occurrences of an A, T and H domain in a genome, ranging from 0 to thousands and resembling morphometric data with large variances. Because existing phylogenetic programs can process only tens of phylogenetic character states depending on user's CPU performance, the space of G values in the matrix was reduced using a standard gap-coding technique with the following formula:in which denote either an A, T or H domain structure, a genome, and the abundance of in . indicate maximum values for all genomes. The round function normalizes G values on a 0–20 scale (). These values define character states, which are encoded as linearly ordered multistate phylogenetic characters using an alphanumeric format of numbers 0–9 and letters A–K that is compatible with PAUP*. A ‘by hand’ generic example of data normalization and encoding is shown in Protocol S1. The actual raw matrix describing the A-level domain census is shown in Dataset S1 as an example. Transposition of the data matrix (switching characters and taxa) allowed reconstruction of trees of either proteomes or domain structures. Trees of A, T and H domains were built by polarizing states from ‘K’ to ‘0’ using the ANCSTATES command in PAUP*, with ‘K’ being ancestral. Trees of proteomes were built by polarizing character states from ‘0’ to ‘K’, with ‘0’ being ancestral. The trees were rooted without invoking outgroup taxa using the Lundberg method, which positions the most ancient proteomes and domain structures at the base of their corresponding trees. Assumptions of character argumentation have been discussed in previous publications [1], [7], [12], [15]. Our model of structural evolution (‘K’ to ‘0’ polarization) considers that the abundance of individual domain structures increases progressively in nature, even when expanding domain levels suffer loss in individual lineages or are selectively constrained during evolution (we consider that character state transformation is reversible). Consequently, ancient structures are more abundant and widely present than younger ones. In contrast, our model of proteome evolution (‘0’ to ‘K’ polarization) assumes proteomes have built their structural repertoires progressively, increasing both the diversity and abundance of their structural make up.
While character argumentation considers that domain structures that appear early in evolution are prominent in genomes and that their numbers increase in steps corresponding to the addition or removal of a homologous gene in a family, the model is agnostic about how changes occur. For example, duplications of domains with simple structural motifs that occur in multiples may involve the entire array, and if these tandem duplicates confer selective advantage, they can be retained in the lineage and can distribute throughout proteome lineages. This is the case for example with proteins that contain tandem repeats of several domains from a same family that are common in Eukarya [62]. While this mechanism of domain gain in not accounted by the model, evolutionary statements relate to domain taxa and their definitions, which generally consider domains as structural and evolutionary modular units.
Phylogenomic trees were reconstructed using the maximum parsimony (MP) optimality criterion in PAUP* with 1,000 replicates of random taxon addition, tree bisection reconnection (TBR) branch swapping, and maxtrees unrestricted. Phylogenetic confidence was evaluated by the nonparametric bootstrap method with 1,000 replicates (resampling size matches the number of the genomes sampled; TBR; maxtrees, unrestricted). The degree of phylogenetic signal for taxa was measured using the skewness (g1) test with a tree length distribution obtained from 1,000 random trees.
Since trees of domain structures are rooted and are highly unbalanced, we unfolded the relative age of protein domains directly for each phylogeny as a distance in nodes (node distance, nd) from the hypothetical ancestral architecture at the base of the trees in a relative 0–1 scale. nd was calculated by counting the number of internal nodes along a lineage from the root to a terminal node (a leaf) of the tree on a relative 0–1 scale with the following formula:where a represents a target leaf node (either an A, T or H domain), r is a hypothetical root node, and m is a leaf node that has the largest possible number of internal nodes from node r. Consequently, the nd value of the most ancestral taxon is 0, whereas that of the most recent one is 1. Node distance can be a good measure of age given a rooted tree because the emergence of protein domains (i.e., taxa) is displayed by their ability to diverge (cladogenesis or molecular speciation) rather than by the amount of character state change that exists in branches of the tree (branch lengths).
In this study we have not compared phylogenies recovered using different versions of CATH. However, our experience with SCOP definitions over the years has shown that tree topologies do not change significantly and that evolutionary inferences stand despite biases in the databases [8] and addition of new domain structures to the known repertoire of proteins [1]. We note that the atomic structures of most protein folds have been acquired (∼1,200 out of 1,500 expected)[63]. Consequently, new domain structures are by definition either rare in genomes or intrinsically difficult to recover. Since important evolutionary patterns obtained using CATH definitions match those derived from SCOP, we do not expect CATH updates will change the central conclusions of our study.
|
10.1371/journal.pntd.0002064 | T Cell Hypo-Responsiveness against Leishmania major in MAP Kinase Phosphatase (MKP) 2 Deficient C57BL/6 Mice Does Not Alter the Healer Disease Phenotype | We have recently demonstrated that MAP kinase phosphatase 2 (MKP-2) deficient C57BL/6 mice, unlike their wild-type counterparts, are unable to control infection with the protozoan parasite Leishmania mexicana. Increased susceptibility was associated with elevated Arginase-1 levels and reduced iNOS activity in macrophages as well as a diminished TH1 response. By contrast, in the present study footpad infection of MKP-2−/− mice with L. major resulted in a healing response as measured by lesion size and parasite numbers similar to infected MKP-2+/+ mice. Analysis of immune responses following infection demonstrated a reduced TH1 response in MKP-2−/− mice with lower parasite specific serum IgG2b levels, a lower frequency of IFN-γ and TNF-α producing CD4+ and CD8+ T cells and lower antigen stimulated spleen cell IFN-γ production than their wild-type counterparts. However, infected MKP-2−/− mice also had similarly reduced levels of antigen induced spleen and lymph node cell IL-4 production compared with MKP-2+/+ mice as well as reduced levels of parasite-specific IgG1 in the serum, indicating a general T cell hypo-responsiveness. Consequently the overall TH1/TH2 balance was unaltered in MKP-2−/− compared with wild-type mice. Although non-stimulated MKP-2−/− macrophages were more permissive to L. major growth than macrophages from MKP-2+/+ mice, reflecting their reduced iNOS and increased Arginase-1 expression, LPS/IFN-γ activation was equally effective at controlling parasite growth in MKP-2−/− and MKP-2+/+ macrophages. Consequently, in the absence of any switch in the TH1/TH2 balance in MKP-2−/− mice, no significant change in disease phenotype was observed.
| Leishmania species are parasites that are of extensive public health importance in the tropics and subtropics. Within humans the parasites are intracellular and reside particularly within macrophages. Classical activation of macrophages by Interferon-γ (IFN-γ) induces the enzyme nitric oxide synthase (iNOS) and parasites are killed via the production of nitric oxide (NO) from the substrate L-arginine. Alternative activation by Interleukin-4 (IL-4) results in Arginase-1 expression, which depletes L-arginine and facilitates parasite growth. We have recently shown that MAP Kinase Phosphatase-2 (MKP-2) suppresses macrophage Arginase-1 and that C57BL/6 mice with a deletion of this gene are subsequently extremely susceptible to New World cutaneous leishmaniasis caused by Leishmania mexicana. Surprisingly, MKP-2 deficient mice have been shown here to be resistant to Old World cutaneous leishmaniasis caused by L. major. We show that during infection with L. major, enhanced Arginase-1 in MKP-2 deficient mice serves to induce a generalized T cell hypo-responsiveness so that IFN-γ and IL-4 levels are equally suppressed compared with intact mice. In addition, unlike L. mexicana, classically activated MKP-2 deficient macrophages were able to control L. major growth equally as well as MKP-2 intact macrophages, highlighting a fundamental difference in the control of these two species.
| Leishmania species are protozoan parasites that are transmitted by infected female sandflies and cause a wide spectrum of diseases ranging from self-healing cutaneous lesions to fatal systemic disease. After entering their vertebrate host, promastigotes are taken up initially by neutrophils and ultimately macrophages and dendritic cells, where they turn rapidly into amastigotes and survive within parasitophorous vacuoles [1]. Resistance against cutaneous infection with Leishmania (L.) major typically requires the presence of an antigen-specific type 1 immune response comprising of IFN-γ/TNF-α/IL-2 producing CD4+ T [2], [3] cells but also CD8+ T cells have been shown to play an important role in parasite clearance [4], [5]. Subsequently, activated T cells migrate to the site of infection where they release IFN-γ and TNF-α which in turn upregulate inducible nitric oxide synthase (iNOS) in infected macrophages, enabling nitric oxide (NO) mediated killing of the intracellular parasites [6], [7]. Susceptibility, on the other hand, has been associated with a failure to produce a type-1 response, which may be a consequence of IL-10 production from Fc-γ mediated macrophage uptake of amastigotes [8], or from natural and or type-1 regulatory T cells [9]–[11], or regulatory B cells [12], or an elevated TH2 response and the excessive production of IL-4 by CD4+ T cells [2], [13], or indeed a combination of all these factors (reviewed by [14]). IL-4 in particular has been shown to promote alternative macrophage activation including increased expression of Arginase-1 [15], suppression of iNOS [16] and increased growth of L. major.
Mitogen-activated protein kinase phosphatase 2 (MKP-2) is a dual-specific nuclear phosphatase (DUSP) and is associated with the MAPK signalling pathway, where it has been shown to dephosphorylate and thereby inactivate protein kinases ERK, JNK but not p38 in vitro [17], [18]. MKP-2 could therefore have significant effects on L. major infection as these parasites have a well recognized ability to subvert the development of TH1 responses partly via effects upon MAP kinase signalling. Studies using L. major metacyclic promastigotes indicated that the parasite via lipophosphoglycan (LPG) differentially regulated IL-12 as well as NO production by targeting ERK and p38 MAPK, respectively [19], [20]. In order to better understand the role of MKP-2 in immune functions we, and others, have recently created MKP-2 deficient mice on a C57BL/6 background [21]–[24]. Thus regulatory roles specific for MKP-2 have been demonstrated in the inflammatory response associated with sepsis [22], cell cycle progression and apoptosis [23] and infection [21]. Furthermore, MKP-2−/− macrophages have severe ablation of LPS or IFN-γ -induced iNOS expression and nitric oxide release and enhanced basal expression of Arginase-1. Given that Arginase-1 competes with iNOS for their common substrate L-arginine, it suggested that MKP-2 could have a regulatory function significant in immune responses involving intracellular pathogens [25]. Indeed, following infection with the intracellular parasite L. mexicana we demonstrated that it was changes in Arginase-1 and iNOS rather than changes in kinase mediated signalling that dictated the subsequent in vivo response to MKP-2 deletion [21]. MKP-2−/− mice displayed increased lesion size and parasite burden, and a significantly modified TH1/TH2 bias compared with wild-type counterparts [21]. This was related to a significant down-regulation of specific TH1 activity in MKP-2 deficient animals. However, there was no intrinsic defect in MKP-2−/− T cell function as measured by anti-CD3 induced IFN-γ production. Rather, MKP-2−/− bone marrow-derived macrophages, as a consequence of increased Arginase-1 expression, were found to be inherently more susceptible to infection with L. mexicana than MKP-2+/+ derived macrophages.
The immune-regulatory mechanisms controlling L. mexicana and L. major differ significantly and while most mouse strains heal following infection with L. major the vast majority develop chronic infections following infection with L. mexicana (reviewed by [26]). Thus, while C57BL/6 mice can control L. mexicana lesion growth they do not cure and MKP-2 deficiency results in progressive disease following L. mexicana infection [21]. C57BL/6 mice on the other hand are more resistant to L. major and lesions heal following infection. Given the well documented importance of NO killing in controlling L. major infections we addressed the question as to whether MKP-2 deficiency would make C57BL/6 mice more susceptible to this parasite. To our surprise no distinct disease phenotype was noted in MKP-2−/− mice infected with this parasite. Although type-1 responses in infected MKP-2−/− mice were down-regulated, type-2 responses were equally suppressed and the TH1/TH2 balance remained unaffected. Furthermore although MKP-2−/− naïve macrophages were more permissive host cells for L. major than wild-type macrophages, if classically activated they were equally effective at controlling parasite growth.
Female DUSP-4 (MKP-2) wild-type and deficient mice were generated as previously described and bred onto C57BL/6 background [21] and female BALB/c mice were bred and maintained under specific pathogen-free conditions in the animal facilities of the Strathclyde Institute for Pharmacy and Biomedical Sciences at the University of Strathclyde. Animals were used at 6–9 weeks of age and were age matched within each experiment. All animal experiments adhered to the UK Animals (Scientific Procedures) Act 1986 and were conducted under Project Licenses to RP (PPL60/3439 “genetic models of cancer and inflammation”) and JA (PPL60/3929 “mechanism of control of parasite infection”) granted by the UK Home Office and with local ethical approval.
Leishmania major (IR75) promastigotes were grown in HOMEM (Gibco) supplemented with 10% FCS (Biosera, East Sussex, UK) until late stationary phase. In order to enable direct comparison with previous studies on the role of MKP-2 during L. mexicana infections [21] and to Arginase-1/L. major studies [27], [28] relevant to the work presented here we used the high dose inoculation model. Mice were given 2×106 L. major subcutaneously into the left hind foot pad and lesion development was monitored as the difference in thickness between infected and uninfected foot pads using a dial gauge calliper. For determination of parasite burden, mice were sacrificed by cerebral dislocation and foot pads, popliteal lymph nodes and spleens were removed and homogenized. Single cell suspensions were adjusted to equal volumes and subjected to limiting dilution assay as described elsewhere [29].
Single cell suspensions from spleens and lymph nodes of infected mice were prepared and erythrocytes were lysed using Erythrocyte lysis buffer pH 7.3 (160 mM NH4Cl, 10 mM KHCO3 and 100 µM EDTA). Cells were resuspended in RPMI containing 10% FCS, L-Glutamine and Penicillin/Streptomycin and 2×106 splenocytes were restimulated either on dendritic cells pulsed with 5 µg L. major antigen over night, PMA (10 ng/ml)/Ionomycin (500 ng/ml), ConA (10 µg/ml) or medium alone for 6 h (intracellular staining) or 48–72 h (ELISA) in 24-well plates. For intracellular staining, cytokine release was inhibited by the addition of BrefeldinA (10 µg/ml) at 3 h into re-stimulation. All reagents were supplied by Sigma-Aldrich (St. Louis, USA).
CD4+ T cells from spleens and lymph nodes of infected mice were isolated on LS magnetic columns (Miltenyi Biotec GmbH, Bergisch Gladbach, Germany) using the negative selection kit (Miltenyi) following manufacturer's instructions.
Mouse serum was analysed individually for antigen-specific IgG1 or IgG2b using single sided ELISA as previously described [30]. Resulting titres were expressed as reciprocal values of the half-maximal absorption at 450 nm using a Spectromax 190 plate reader.
Cytokines from T cell supernatants were determined with Sandwich-ELISA using cytokine specific capture and detection antibodies (BD Bioscience) and recombinant cytokines for standard curves (R&D Systems, Minneapolis, USA). IL-4 was determined using the mouse IL-4 Quantikine kit (R&D Systems) following manufacturer's instructions. To allow detection of IL-4, 5 µg/ml IL-4 receptor blocking antibody (anti-mouse CD124, BD Pharmingen, USA) was added to the cultures.
Cells were harvested and passed through a nitex mesh to remove clumps. For isotype control, 5×105 cells of medium-, antigen- and PMA/Ionomycin-restimulated splenocytes were pooled for each individual. After blocking unspecific binding with 10% mouse serum and Fc receptor blocking antibodies (anti-mouse CD16/32, eBioscience, UK), cell surface was stained with conjugated antibodies for CD3e (PerCP), CD4 (APC-H7), both BD Pharmingen (USA) and CD8b (Alexa Fluor 488, eBioscience UK) for 45 min at 4°C. Cells were fixed for 15 min and permeabilized using the Fix and Perm kit (Invitrogen, Paisley, UK). Intracellular staining was carried out simultaneously with permeabilisation for 1 h at room temperature using conjugated antibodies for IFN-γ (Allophycocyanin (APC)) and TNF-α (Phycoerythrin (PE)) or their respective isotypes anti-rat IgG1 (APC) and anti-rat IgG1 (PE), all eBioscience. After washing steps, cells were resuspended in 200 µl PBS and subsequently run and analysed on the FACS Canto flow cytometer (BD Bioscience) using FACS Diva software.
Bone marrow-derived macrophages (BMDM) and dendritic cells were derived from tibia and femur of 6 to 8 week old mice. Bone marrow was flushed, and in order to obtain macrophages, cells were resuspended in DMEM containing 10% foetal calf serum (FCS), 30% L929-conditioned medium, 2 mM L-Glutamine (Gibco, Invitrogen, Paisley, UK), 1% Penicillin/Streptomycin (Gibco) and seeded into Petri dishes. After 10 days at 37°C, adherent macrophages were harvested with cold/warm PBS, washed and resuspended in complete RPMI (10% FCS, 2 mM L-Glutamine, 1% Penicillin/Streptomycin). Dendritic cells were generated from bone marrow precursors by culturing in RPMI containing 10% FCS, 2 mM L-Glutamine, 1% Penicillin/Streptomycin and 2.5–10% GM-CSF conditioned medium (X63). Non-adherent cells were harvested at day 7, washed and resuspended in complete RPMI.
BMDMs were seeded onto cover slips at 2×105 cells/500 µl in 24-well plates and left to adhere at 37°C over night. For infection L. major promastigotes (IR75) were added at a multiplicity of infection (MOI) of 5 and plates were briefly spun at 300×g to allow close proximity of parasite and macrophages. After 1 h at 34°C, supernatants were removed and macrophages were washed in PBS to remove any free parasites. Fresh complete RPMI supplemented with or without 100 ng/ml LPS, 100 U/ml IFN-γ or 100 U/ml IL-4 was added and cells were incubated at 34°C for different periods of time.
To determine number of parasites and infection rates, macrophages were fixed in methanol and stained with Giemsa solution. Cover slips were mounted onto glass slides and intracellular parasites were counted in a total of 200 macrophages across each cover slip using a bright field microscope.
Arginase-1 expression was determined from whole cell lysates of 1×106 L. major-infected BMDM. SDS-PAGE analysis and detection was carried out as described elsewhere [21].
Statistical analysis was performed using GraphPad Prism Program (Version 4.0, GraphPad Software, San Diego, California). P values below or equal to 0.05 were considered significant.
MKP-2 deficient and MKP-2 wild-type C57BL/6 mice were given 2×106 late-stationary phase L. major promastigotes subcutaneously into the left hind footpad and lesion development was monitored over 12–13 weeks. Surprisingly, given that MKP-2−/− mice were previously found to be more susceptible to L. mexicana than their wild-type counterparts [21], no difference in lesion growth between MKP-2−/− and MKP-2+/+ mice was detected throughout the course of infection with L. major. Both MKP-2−/− and MKP-2+/+ mice developed lesions which healed spontaneously after several weeks and in a manner typical for L. major resistant C57BL/6 mice (Figure 1A). At two time points, the peak of infection (Figure 1B) and after onset of healing (Figure 1C), five mice were sacrificed and the parasite burdens of infected footpads, popliteal draining lymph nodes and spleens were determined. Consistent with the lesion development, parasite numbers were high at the peak of infection and consequently dropped with the onset of healing. We could not observe a difference in phenotype between wild-type and MKP-2−/− mice at either peak lesion growth or after healing in any of the tissues examined. This observation was made in several independent experiments.
Our previous studies demonstrated that the susceptibility of MKP-2 deficient mice is to a significant degree due to MKP-2−/− macrophages being inherently more susceptible to the growth of L. mexicana than wild-type macrophages a dichotomy that was maintained even following classical activation [21]. We therefore examined the infectivity of L. major metacyclic promastigotes for MKP-2−/− and MKP-2+/+ naïve, innately, classically or alternatively activated bone marrow derived macrophages (Figure 2 A, B and C). Over several experiments we found that infected non-stimulated MKP-2−/− BMDM cultures maintained significantly increased parasite burdens and increased frequency of infected macrophages than cultures using their wild-type counterparts (Figure 2A). Our previous studies with L. mexicana had demonstrated the increased susceptibility of non-stimulated MKP-2−/− macrophages to infection to be related to elevated Arginase-1 levels compared with MKP-2+/+ macrophages and similarly enhanced levels of Arginase-1 were found in MKP-2−/− macrophages following infection with L. major promastigotes (Figure 2E). Innate macrophage activation with 100 ng/ml LPS greatly decreased the number of intracellular parasites as well as the frequency of infected cells in both MKP-2−/− and MKP-2+/+ macrophages (Figure 2A). However, despite activation with LPS, MKP-2 deficient macrophages still presented higher parasite survival than their wild-type counterparts. As nitric oxide (NO) is essential for killing of intracellular parasites, we determined the production of NO by measuring nitrite levels in the supernatants of infected macrophages 24 and 48 h after activation with LPS. Consistent with our previous observations with L. mexicana, MKP-2−/− macrophages showed a reduced NO production at both time points in response to LPS activation (Figure 2D). However, whereas MKP-2−/− macrophages classically activated with LPS and IFN-γ failed to reduce L. mexicana parasite burdens to the level of similarly treated MKP-2+/+ macrophages [21] such treatment sufficed following infection with L. major promastigotes to ablate any differences in susceptibility between MKP-2−/− and MKP-2+/+ host cells (Figure 2B) and corresponded to LPS and IFN-γ treatment ablating the deficiency in NO production in MKP-2−/− macrophages infected with L. major (data not shown). The addition of recombinant IL-4 to macrophages infected with L. major doubled the numbers of intracellular parasites in both wild-type and MKP-2−/− macrophages (Figure 2C), highlighting the disease-promoting attributes of IL-4 and alternative macrophage activation in L. major infections characterized by elevated Arginase-1 and reduced iNOS expression [16], [31].
As no discernible change in the healing disease phenotype was observed between MKP-2−/− and MKP-2+/+ mice infected with L. major, and given that classically activated macrophages from MKP-2−/− mice were equally as effective as their wild-type counterparts at killing parasites we compared the type-1 response generated in MKP-2+/+ and MKP-2−/− following infection. After re-stimulation with soluble L. major antigen-pulsed DCs, flow cytometric analysis of CD4+ T cells from the spleens of infected animals demonstrated that MKP-2−/− mice had a significantly lower frequency of IFN-γ single producing and IFN-γ/TNF-α double producing CD4+ and CD8+ T cells (P<0.05) than their wild-type counterparts (Figure 3A, B and C). Similarly significantly reduced IFN-γ production (P<0.05) was measured in the supernatants of MKP-2−/− derived T cell cultures after stimulation with L. major antigen-pulsed DC, PMA/Ionomycin or ConA (Figure 3D). Finally we analysed the serum of L. major infected mice for the presence of antigen-specific IgG2b, a surrogate marker for TH1 responses, as IFN-γ is essential for IgG class switch to 2a and 2b [32], [33]. At both stages of infection, MKP-2 deficient mice showed lower IgG2b titres, being statistical significant (P<0.05) at day 42 post-infection (Figure 3E). Overall, our data suggest that MKP-2 deficiency results in a diminished TH1 response during infection with L. major, similar to our previous observations using L. mexicana infections [21].
In stark contrast to experimental infection with L. mexicana, MKP-2 deficient mice did not show an increased susceptibility to infections with L. major compared with wild-type mice. This was despite a clearly diminished TH1 response in MKP-2−/− compared with MKP-2+/+ mice following L. major infection (Figure 3). However, MKP-2 is a generalized negative regulator of Arginase-1in tissues rich in phagocytes (Figure S1) and it is well documented that Arginase-1 can induce hypo-responsiveness in a pan-T cell fashion [27], [28] [34]. We therefore examined whether the lack of a parasite-specific TH1 response could be counterbalanced by concomitant impairment of a TH2 response, leaving the relative TH1/TH2 balance undisturbed. Over a number of experiments we found MKP-2−/− mice infected with L. major produced less IgG1 but not significantly less than their similarly infected wild-type counterparts (Figure 4A). To directly measure the parasite induced production of IL-4, we infected MKP-2 wild-type and deficient mice as well as susceptible BALB/c mice with L. major promastigotes. At the peak of infection, mice were sacrificed, spleens and draining lymph nodes removed and analyzed for the production of IL-4. Whole spleen and lymph node cells were re-stimulated with soluble L. major antigen-pulsed DC, PMA/Ionomycin or ConA for 72 h in the presence of IL-4 receptor blocking antibody to prevent immediate uptake of freshly produced IL-4 in an auto- or paracrine fashion. As expected, susceptible BALB/c mice produced high levels of IL-4 compared with mice from the C57BL/6 background (Figure 4B). However, while antigen induced IL-4 production was reduced in MKP-2−/− splenocytes and draining lymph node cells (Figure 4C and D) compared with wild-type equivalent cell populations this was not clearly significant because of small sample sizes. For a more precise examination of the TH2 response we therefore isolated CD4+ T cells from spleen suspensions by negative selection and re-stimulated these as described above. The levels of IL-4 produced by CD4+ T cells in response to L. major infection were indeed clearly and significantly reduced in MKP-2 deficient mice when compared with wild-type mice (Figure 4E), thus confirming a reduced TH2 response in infected MKP-2−/− mice.
Previous studies have suggested that it is not the quantity of IFN-γ or IL-4 production that is important in determining protective immunity to L. major but rather the overall TH1/TH2 balance [35], [36]. Consequently we calculated the relative TH1/TH2 balance between L. major infected MKP-2−/− and MKP-2+/+ mice. In the first instance we calculated the ratio of IgG2b to IgG1 for each individual mouse, as this is a strong indicator of TH1/TH2 balance [37]–[39]. However no difference in IgG2b/IgG1 levels between L. major infected MKP-2−/− and MKP-2+/+ mice were observed throughout infection (Figure 5A). Not surprisingly, the ratio increased after onset of healing toward an IgG2b bias and thus a TH1 response in both MKP-2−/− and MKP-2+/+ mice. We further calculated the percentage reduction in the levels of type-1 and type-2 cytokine responses induced in MKP-2−/− compared with wild-type mice following infection. MKP-2−/− splenocytes from L. major infected mice produced, upon restimulation with L. major antigen-pulsed DCs, 32.6% less IFN-γ (Figure 5B) than MKP-2+/+ splenocytes (20.79±5.53 (SD) and 30.87±5.39 (SD) ng/ml, respectively). Under the same experimental conditions a 37.8% reduction in IL-4 production (Figure 5B) was generated by MKP-2−/− splenocytes compared with MKP-2+/+ splenocytes (35.46±9.32 (SD) pg/ml compared with 56.99±12.13 (SD) pg/ml, respectively). Moreover, when comparing the percentage of IFN-γ/TNF-α double producing CD4+ T cells, as measured by flow cytometry, we found a 28.7% reduction (Figure 5B) in infected MKP-2−/− compared with MKP-2+/+ mice. Thus MKP-2 deficiency results in a T cell hypo-responsiveness following L. major infection, which effects both the protective TH1 and the disease-exacerbating TH2 response to a similar degree (Figure 5C). Consequently, MKP-2−/− mice have a healing phenotype similar to their wild-type counterparts.
In a previous study we identified a major role for MKP-2 in protecting C57BL/6 mice from infection with L. mexicana as MKP-2 deficiency resulted in uncontrolled lesion growth with massively increased parasite burdens [21]. Increased susceptibility was associated with MKP-2−/− macrophages being inherently more susceptible than wild-type macrophages to parasite infection as a result of increased Arginase-1 expression and also reduced NO production [21]. In addition, while no specific direct defect in T cell function could be attributed to MKP-2 deficiency, a diminished parasite-specific TH1 and an enhanced TH2 response developed in MKP-2−/− mice infected with L. mexicana [21]. Surprisingly therefore, in the present study, no differences whatsoever in the normal healing phenotype were observed between MKP-2−/− and MKP-2+/+ mice infected with L. major. This was despite firstly, naïve and innately activated MKP-2−/− macrophages being more permissive to L. major infection than wild-type macrophages which, as with L. mexicana infections, was associated with increased Arginase-1 and reduced iNOS activities, and secondly, a clearly reduced parasite-specific TH1 response in infected MKP-2−/− mice.
Infection of MKP-2−/− mice and their host cells with L. major differed to that of infection with L. mexicana in two significant ways that undoubtedly had profound effects on disease outcome. Firstly, in contrast to infection with L. mexicana, infection of MKP-2−/− mice with L. major resulted not only in a reduced type-I response, but also in an equally reduced TH2 response compared with wild-type mice. Consequently there was no discernible difference in the TH1/TH2 bias between MKP-2−/− and MKP-2+/+ mice infected with this parasite. It is well established in C57BL/6 mice, that once parasite peptide reactive-CD4+ and CD8+ T cell populations reach the proper balance in draining lymph nodes and the sites of infection, there is rapid healing, and immunity is maintained by a persistent small amastigote population in equilibrium with both effector and regulatory T cell populations [10], [40]. Furthermore studies in mice have demonstrated that the absolute amount of IFN-γ generated following infection with L. major did not correlate with protection or cure [35], and rather, it was the balance in the TH1/TH2 cytokine profile that was important in determining disease outcome. Secondly, classically activated MKP-2−/− macrophages were equally as effective as MKP-2+/+ macrophages in controlling the growth of L. major although not, as previously demonstrated, L. mexicana [21]. Therefore, as the TH1/TH2 balance remained unaltered, albeit equally diminished, in L. major infected MKP-2−/− compared with MKP-2+/+ mice, the comparative polarisation of classical macrophage activation would remain the same. Consequently healing takes place in MKP-2−/− mice infected with L. major in a similar manner to infected resistant wild-type mice.
Among the intriguing questions regarding the differential outcome of L. major and L. mexicana infections in MKP-2 deficient mice is why TH2 responses are maintained following L. mexicana infection [21] but down-regulated following L. major infection compared with their infected wild-type counterparts? It has been demonstrated previously that during L. major infections high local Arginase-1 levels at the site of infection mediate L-arginine depletion, which results in impaired local CD4+ (and CD8+) T cell function, particularly IFN-γ production but also to a lesser extent IL-4 and IL-10 [28], [41]. Investigations carried out by Kropf et al. [34] using Arginase-1 expressing placenta cells and Jurkat T cells showed that Arginase-1-mediated T cell hypo-responsiveness is a consequence of the down-regulation of the CD3ζ chain, a crucial signal transducing component of the TCR [34] and significantly CD3ζ has recently been found to be down-regulated on CD4+ and CD8+ T cells along with increased Arginase-1 activity in the lesions of patients infected with L. aethiopica [42]. Depletion of L-arginine, as a result of the generalised elevated Arginase-1 levels in MKP-2−/− mice would explain the general T cell hypo-responsiveness observed in these mice following infection with L. major. However, why should the TH2 response be maintained if not enhanced following infection of MKP-2−/− mice with L. mexicana [21]? Studies to date clearly implicate the multiple CPB isoenzymes as important species-specific virulence factors for the L. mexicana complex and provide one possible explanation as to why these parasites, unlike L. major (reviewed by [43]) tend to induce chronic lesions in the majority of mouse strains such as the C57BL/6 strain used in this study. Not only are L. mexicana CPBs potent inhibitors of TH1 responses [44] as a consequence of disrupting macrophage signalling pathways [45], but they have also been shown to be potent inducers of IL-4 production and TH2 responses [46]. Consequently C57BL/6 mice infected with L. mexicana CPB null mutants, unlike infection with wild-type parasites, develop a healing response with reduced IL-4 production and TH2 responses along with elevated TH1 responses [47]. Thus, as a result of their highly expressed TH2 promoting and TH1 inhibiting CPBs infection of MKP-2−/− mice with L. mexicana, unlike infection with L. major, is able to rescue and promote the TH2 component of the parasite specific T cell response which is less subject to Arginase-1 induced hypo-responsiveness than TH1 responses [27].
Alternative activation of both MKP-2−/− and MKP-2+/+ macrophages resulted in increased susceptibility to L. major. This would support the view that the TH1/TH2 balance is important and that IL-4 and IFN-γ activities regulate each other not just at the T cell level but also at the level of the macrophage by modulation of iNOS and Arginase-1 expression. In agreement IL-4Rα signalling via macrophages/neutrophils has been shown to promote early lesion growth in L. major infected BALB/c mice and macrophage/neutrophil specific (LysMcreIL-4Rα−/lox) IL-4Rα−/− mice display delayed lesion growth [16]. The control of L. major early in infection in LysMcreIL-4Rα−/lox mice has been identified as being due to enhanced macrophage microbicidal NO mediated activity in the absence of alternative macrophage activation. Paradoxically we failed to identify any disease promoting contributory role for IL-4Rα signalling via macrophages/neutrophils during L. mexicana infection [48]. In agreement in the present studies we also failed to identify a disease promoting role for IL-4 in either MKP-2−/− or MKP-2+/+ macrophages infected with L. mexicana (Figure S2). Thus the interaction of L. major and L. mexicana with their host macrophages must be significantly different. What may be critical in this regard is that L. amazonensis parasites, which belong to the “mexicana” complex of parasites, have been shown to be more resistant to macrophage-mediated control than L. major requiring higher levels of NO to induce killing [49], [50]. The evidence would suggest that macrophage killing of “mexicana” complex parasites unlike L. major requires NO and additionally, superoxide [51]. Furthermore, recent evidence indicates that, unlike L. major, there is in fact enhanced replication of the amastigote stage of L. amazonensis in IFN-γ-stimulated murine macrophages despite higher NO production [52], reportedly due to the induction of a novel L-arginine pathway independent of iNOS or host Arginase-1 [53]. Induction of Arginase-1 by L. amazonensis has also been shown to enhance replication of the amastigote stage of the parasite [52], [53] while inhibition studies have shown that the enhanced susceptibility of MKP-2−/− macrophages for L. mexicana is associated with enhanced Arginase-1 expression [21]. Thus given the relatively higher sensitivity of L. major to NO matched with the increased importance of Arginase-1 to infection with L. mexicana it is perhaps not surprising that classical macrophage activation ablates MKP-2 deficiency mediated differences in infectivity with the former but not the latter parasite.
Overall the present study confirms our previous observation that MKP-2 is a major factor in determining the immune response against intracellular parasites and potentially the outcome of infection. Naïve MKP-2 deficient macrophages are inherently more susceptible to L. major than wild-type macrophages and following infection there is a generalised T cell hypo-responsiveness. However, despite these apparent deficiencies the disease phenotype of MKP-2−/− mice following L. major infection does not differ from wild-type mice. Our results suggest that this is a consequence of the TH1/TH2 balance remaining unaltered in MKP-2−/− mice infected with L. major and that classical macrophage activation suffices to ablate the innate permissiveness of MKP-2−/− macrophages to this species.
|
10.1371/journal.pcbi.1002880 | Metallochaperones Regulate Intracellular Copper Levels | Copper (Cu) is an important enzyme co-factor that is also extremely toxic at high intracellular concentrations, making active efflux mechanisms essential for preventing Cu accumulation. Here, we have investigated the mechanistic role of metallochaperones in regulating Cu efflux. We have constructed a computational model of Cu trafficking and efflux based on systems analysis of the Cu stress response of Halobacterium salinarum. We have validated several model predictions via assays of transcriptional dynamics and intracellular Cu levels, discovering a completely novel function for metallochaperones. We demonstrate that in addition to trafficking Cu ions, metallochaperones also function as buffers to modulate the transcriptional responsiveness and efficacy of Cu efflux. This buffering function of metallochaperones ultimately sets the upper limit for intracellular Cu levels and provides a mechanistic explanation for previously observed Cu metallochaperone mutation phenotypes.
| Copper (Cu) toxicity is a problem of medical, agricultural, and environmental significance. Cu toxicity severely inhibits growth of plant roots significantly affecting their morphology; Cu overload also accounts for some of the most common metal-metabolism abnormalities and neuropsychiatric problems including Wilson's and Menkes diseases. There is a large body of literature on how Cu enters and exits the cell; the kinetic and structural details of Cu translocation between trafficking, sensing, metabolic, and pumping proteins; and phenotypes associated with defects in metalloregulatory and efflux functions. Although the role of metallochaperones in Cu-cytotoxicity has been poorly studied, it has been observed that in animals deletion of metallochaperones results in elevated intracellular Cu levels along with overexpression of the P1-type ATPase efflux pump, ultimately causing malformation with high mortality. These observations are mechanistically explained by a predictive model of the Cu circuit in Halobacterium salinarum, which serves as an excellent model system for Cu trafficking and regulation in organisms with multiple chaperones. Constructed through iterative modeling and experimentation, this model accurately recapitulates known dynamical properties of the Cu circuit and predicts that intracellular Cu-buffering emerges as a consequence of the interplay of paralogous metallochaperones that traffic and allocate Cu to distinct targets.
| Copper (Cu) is an essential trace element in nearly all biological systems [1] but highly cytotoxic when in excess [2], [3]. Biological systems possess sophisticated trafficking systems that include ion importers to control Cu entry [4]; metallochaperones for shuttling intracellular ions to and from targets[5], [6]; efflux pumps that export excess Cu [7], [8]; and metalloregulators that sense internal abundance and modulate expression of all trafficking proteins [9]. There is a large body of literature on how Cu enters and exits the cell [10], [11]; the kinetic and structural details of Cu translocation between trafficking, sensing, metabolic, and pumping proteins [12], [13], [14]; and phenotypes associated with defects in metalloregulatory and efflux functions [15].
Defects in Cu efflux pumps, for example, are associated with neurodegenerative disorders such as Menkes and Wilson's syndromes, and Alzheimer's disease [7], [16], [17], [18], [19]. Although we do not fully understand the mechanistic role of metallochaperones in any Cu-related disease, they are universally present [20], [21], [22], [23], [24], [25], [26] and known to be important in Cu-trafficking and preventing cellular damage [27], [28]. Furthermore, deletion of the Cu metallochaperone Atox1 in immortalized human cell lines resulted in elevated intracellular Cu levels along with overexpression of Cu transporting ATP7A [19], [29]. Mice harboring metallochaperone deletions are malformed with high mortality [30]. Combined, these data suggest that beyond their known Cu trafficking role, metallochaperones also influence the activity of metalloregulators and intracellular Cu levels via an uncharacterized mechanism.
In this study we present an integrated experimental and computational analysis of transcriptional regulation of Cu efflux in Halobacterium salinarum NRC-1 and the role that metallochaperones play therein. In brief, H. salinarum possesses a post-transcriptionally regulated Cu trafficking system that was characterized by systems analysis of cellular response to growth sub-inhibitory levels of extracellular Cu. The main components of the Cu efflux network in H. salinarum include a Cu sensing metalloregulator (VNG1179C) that regulates transcription of a P1-type ATPase efflux pump (VNG0700G), and two metallochaperones (VNG0702H and VNG2581H) with Cu binding sites that are highly conserved across all domains of life (Figure S1). We integrated the functional interplay between these components into a system of ordinary differential equations with iterative refinement of model parameters to fit experimental data of Cu response dynamics. Using this model we explored the consequence of deleting or overexpressing metallochaperones on activity of VNG1179C (assayed directly by measuring VNG0700G transcript levels, or indirectly by measuring fluorescence in live cells transformed with a GFP reporter tagged to the promoter of VNG0700G) and ultimately on intracellular Cu levels (measured with ICP-MS). The three rounds of iterative experimentation and computation has revealed that each of the two metallochaperones in H. salinarum have distinct functions, and that their interactions with other components of Cu efflux acts as a buffer, setting the upper threshold of homeostatic intracellular Cu. Altering the absolute abundance of metallochaperones significantly affects sensitivity of the metalloregulator to Cu levels, efficacy of Cu efflux by VNG0700G, and ultimately results in higher level of intracellular Cu.
Glycerol stocks of H. salinarum with requisite gene knockouts were revived on solid CM (NaCl–250 g/l, MgSO4•7H2O–20 g/l, Na·Citrate–3 g/l, KCl-2 g/l, and peptone 10 g/l) agar (1.8% w/v) plates incubated at 37°C for 1–2 wks. Single colonies were selected from plates and placed into liquid CM media (typically 50 ml in a 125 ml Erlenmeyer flask unless otherwise noted) and grown to an optical density (measured at 600 nm) (OD600) of 0.6–0.8 to create stock cultures. Prior to experiments, stock cultures were split into replicates and diluted in fresh medium to a starting OD600 of 0.05. Cu addition experiments were initiated once replicate cultures reached OD600 of 0.4–0.6. Strains carrying GFP expression vectors required medium supplemented with 200 ng/ul mevinolin.
Mutant strains harboring deletions of VNG1179C, VNG0702H, and VNG2581H were created via a two step in-frame deletion method previously described [31], [32] using a Δura3 strain as the parent background.
Strains studied were grown in liquid cultures 250 ml in volume (500 ml flask) at 37°C to accommodate 4 ml samples taken at specified timepoints. Total RNA was purified from cell pellet lysates, stained, and hybridized to a spotted microarray using a previously reported dye-flip protocol [31]. Transcriptome expression data presented herein is original to this work.
Metal free basal salts solution (NaCl–250 g/l, MgSO4•7H2O–20 g/l, Na·Citrate–3 g/l, KCl-2 g/l) and MilliQ water was created by overnight treatment with 5 g/l Chelex. For metal free basal salts solution, metal pure MgSO4 (<0.001% metal impurity) was added after Chelex treatment to avoid saturating the ion binding capacity of the resin. Prior to experiments, all glassware and sample collection tubes were washed twice in 2% nitric acid to strip any trace metals. For ICP-MS analysis, 10 ml samples were retrieved from cultures pre- and post- copper addition into 50 ml conical tubes. Cells were pelleted by centrifugation and washed three times in basal salts solution. After the final wash, cells were lysed in MilliQ water and sonicated for 10 min and stored at 4°C for analysis. Samples were analyzed by ICP-MS using minor modifications to previously published protocols [33].
GFP timeseries experiments were conducted using 50 ml cultures seeded with clonal populations of cells from solid medium colonies. Once cultures reached an OD600 of 0.4–0.6, CuSO4•5H2O was spiked in to a final concentration of 0.85 mM. At selected timepoints, 500 ul samples were taken from cultures and pelleted. Cells were then fixed by resuspending in 1 ml of basal salts solution with 0.25% (w/v) paraformaldehyde and incubated at room temp for 10 min. Fixative was removed by pelleting cells and resuspending in basal salts solution and samples were stored at 4°C.
GFP Cu dose experiments were conducted in 96 deep-well plates (1.6 ml culture per well). Cultures were started in 500 ml flasks and distributed to wells at a nominal OD600 of 0.05. Plates were sealed with BreathEasy film and incubated at 37°C in a shaking incubator set to 200 rpm. Once the average OD600 of the plate reached 0.4–0.6, a pre- Cu addition sample was taken and processed as below. Afterwards, CuSO4•5H2O was spiked in and the plate was incubated for 300 min at 37°C, shaking at 200 rpm, at which time an endpoint sample was taken. For each sample, 150 ul of culture was removed from each well and transferred to a v-bottomed 96-well plate prealiquoted with basal salts solution with 0.5% (w/v) paraformaldehyde. Cells were pelleted in a tabletop centrifuge at 2800G for 10 min at room temp. Supernatant fixative was removed and cells were resuspended in basal salts solution. Plates were then sealed with aluminum foil tape stored at 4°C until analysis.
Flow cytometry analysis was performed within 2–3days of sample collection using a BD InFlux cell sorter fitted with a 100 um flow nozzle and primed with 100 g/l NaCl in MilliQ water as sheath. Prior to injection, each sample was spiked with 1 um yellow/green fluorescent beads to a final concentration of 1×107 ml−1 for use as an internal reference. 100,000 total events were collected and population gating was done via FlowJo. Gated cell populations were exported to ASCII files and further analyzed using the statistical analysis environment, R (http://www.r-project.org).
In R, fluorescence concentrations were calculated by dividing each event's GFP signal by its corresponding forward scatter value. Population concentration means resulting from this analysis are presented in the results.
When stressed with excess Cu, activation of efflux mechanisms protects the cell by restoring intracellular Cu to homeostatic levels. To determine the temporal dynamics of transcription of all genes involved in Cu efflux, we performed a global time course survey of transcript level changes in cells stressed with 0.85 mM CuSO4, a previously determined growth sub-inhibitory Cu concentration [31]. To rid the cell of excess Cu, we discovered that H. salinarum relies primarily on transcriptional activation of three genes: a P1-type ATPase efflux pump yvgX (VNG0700G), and two HMA-domain containing metallochaperones VNG0702H and VNG2581H. The temporal transcript changes for each of these genes demonstrated pulsed induction following Cu exposure (Figure 1). These genes were also previously determined to be regulated by VNG1179C [31], a Lrp family transcription factor with a TRASH domain for metal sensing [36]. The expression of this regulator did not change under Cu stress suggesting that it is post-translationally activated – e.g. upon binding Cu.
The pulse-like transcriptional response of yvgX was expected due to its direct role in relieving the cell of excess intracellular Cu. Conversely, the pulse in metallochaperone expression was unexpected. Absent information on transcriptional dynamics of metallochaperones, we had initially assumed that they were constitutively expressed at low levels. However, the regulated transcription of these genes in direct response to excess Cu suggested that the metallochaperones might have an important role in tuning the transcriptional dynamics of the Cu efflux network. Notably, negative feedback linking activity of YvgX to repression of VNG1179C must exist to conserve cellular resources. The mechanisms that enable such feedback are most likely indirect as YvgX is membrane bound and VNG1179C is cytoplasmic and bound to DNA. Our hypothesis was that feedback must occur via metallochaperones because of their ability to interact directly with Cu ions, VNG1179C, and YvgX.
Based on existing mechanistic understanding from diverse organisms [11], [37], [38], our prior work [31], and the dynamics of the transcriptional response investigated in this study, we developed a computational model for the transcriptional regulation of Cu efflux (Figure 2A). A system of ordinary differential equations that describe transcriptional, translational, and post-translational regulatory events, this model (Model 0) makes two important assumptions based on known biology. First, intracellular Cu is rarely unchelated or “free” [25], but instead is readily bound by metallothioneins [9], [13], [39], glutathione [40], [41], and other Cu binding proteins [2]. We modeled this Cu sequestration capacity with a quota element (Q) whose demand is fulfilled prior to activation of Cu efflux. Second, we assumed that the two metallochaperones VNG0702H and VNG2581H were functionally indistinguishable given the high level of similarity in both their expression profiles and Cu binding motifs. Therefore, metallochaperones were simulated as a single species with two-fold higher copy number than other elements.
We tested this model by performing simulations of the Cu response to a step-increase in extracellular Cu to a growth sub-inhibitory level (0.85 mM). The model assumes that a single cell only reacts to a shell of surrounding volume equivalent to 3× the cell volume. A concentration of 0.85 mM of dissolved copper therein is equivalent to ∼5×105 molecules. This value is held constant throughout simulations, representing non-changing copper supply in the bulk growth medium and presenting a boundary condition for copper at the cell membrane.
All simulations were run until 300 min to give sufficient time for restoration of regulated Cu homeostasis in normal (wild-type or wt) cells. The simulations accurately recapitulated known dynamics of transcriptional induction of yvgX and metallochaperones with transcript and protein levels peaking at ∼18 and 200 min, respectively (Figures 2B and 2C). Based on these encouraging results we proceeded to explore the functional and mechanistic role of metallochaperones in regulation of Cu efflux.
First, we investigated whether changes in abundance of metallochaperones had any consequence on expression of yvgX and, ultimately, on intracellular Cu concentration. The model predicted that the metallochaperones had to be within an optimal range of 100–1000 molecules per cell to produce 100–200 copies of YvgX for maintaining low levels of intracellular Cu (Figure 2D). Interestingly, increasing or decreasing the concentration of metallochaperones outside this optimal range had significantly different effects on steady state levels of YvgX and intracellular Cu. Lowering the abundance of metallochaperones below 100 molecules per cell resulted in increased levels of both intracellular Cu and YvgX. In contrast, increasing the metallochaperones to above 1000 molecules per cell resulted in Cu accumulation with undetectable YvgX levels.
Next, we investigated the consequence of increasing or decreasing metallochaperone abundance on sensitivity of the Cu efflux response over a wide range of extracellular Cu concentrations. The model predicted significant differences in the threshold concentrations of Cu that were necessary to activate expression of YvgX in presence of low, normal, and high abundance of metallochaperone (Figure 2E). Depletion of metallochaperones was predicted to significantly increase the sensitivity of the response, with steady state YvgX levels rising to 1000 copies per cell in the presence of micromolar quantities of external Cu. In contrast, overexpression of metallochaperones was predicted to repress YvgX expression over almost the entire range of extracellular Cu. Thus, the model predicted that metallochaperones tune responsiveness of the metalloregulator, modulate the absolute abundance of the efflux pump, and, ultimately, set the homeostatic level of Cu.
In Model 0 we made the assumption that both metallochaperones had identical functions and, therefore, provided equivalent Cu trafficking and buffering capacity. Alternatively, there could be differences in the specific roles of each metallochaperone. Eukaryotes make use of several Cu-specific metallochaperones: ATOX1, CCS1, and COX17. ATOX1, the eukaryotic ortholog of metallochaperones in H. salinarum, trafficks Cu to P1-type ATPases ATP7A and ATP7B [29] while CCS1 delivers Cu to Cu/Zn dismutases and COX17 delivers Cu to cytochrome oxidases [27]. This raises the possibility that there might be similar distinction in Cu trafficking by the two metallochaperones in H. salinarum.
To investigate if this was indeed the case, we revised the model to include both VNG0702H and VNG2581H as independent elements. We first incorporated subtle differences in trafficking functions (Model 1.1, Figure 4A). Specifically, both metallochaperones were capable of directly binding intracellular Cu, distributing Cu to Q, activating VNG1179C by allocating excess Cu to the metalloregulator, and trafficking excess Cu to YvgX. However, only one metallochaperone was ideally suited for each task, while the other was 10-fold less efficient, resulting in “strong” and “weak” interactions, respectively.
In a revised version of this new model, we incorporated distinct functions for each metallochaperone (Model 1.2) (Figure 4B). In this model, Cu trafficking by VNG0702H is restricted to YvgX, while VNG2581H is responsible for Cu allocation to all remaining targets including VNG1179C and Q. Importantly, VNG2581H was modeled as the only metallochaperone that actively binds free intracellular Cu ions, thus production of Cu-bound VNG0702H requires a hand-off of Cu from VNG2581H, an inter-metallochaperone interaction that has been previously observed in eukaryotes [27]. Thus, Models 1.1 and 1.2 represent alternate extremes for functions of the two metallochaperones. Eliminating one of the two models would ascertain whether the two metallochaperones have interchangeable (Model 1.1) or distinct (Model 1.2) functions in intracellular Cu trafficking and buffering.
The two models made significantly different predictions of intracellular Cu levels at steady state under Cu stress. Model 1.1 predicted increased Cu levels in all single and double chaperone mutant backgrounds (overexpression and deletion), except Δ0702Δ2581, in which intracellular Cu was predicted to be at wt levels (Figure 4C). In contrast, Model 1.2 predicted increased Cu levels in all mutants except when only VNG0702H is overexpressed (Figure 4D).
We have demonstrated that, similar to eukaryotes, the two chaperones in H. salinarum have distinct roles in Cu trafficking. Remarkably, by simultaneously modeling these distinct functions and their interplay we were able to explain why mutations that either increase or decrease the abundance of metallochaperones result in elevated intracellular Cu levels.
In absence of metallochaperones, trafficking of Cu is completely disrupted. When subjected to Cu stress, intracellular Cu concentration increases and eventually overcomes diffusional limitations to activate VNG1179C and increase transcription of yvgX. Despite its increased levels, YvgX is unable to receive Cu efficiently to perform its efflux function. Consequently, Cu levels are perpetually increased and VNG1179C remains locked in an activated state to constitutively drive the expression of yvgX. Importantly, it is the disrupted trafficking of Cu to YvgX that is ultimately responsible for similar consequences in both the single and double metallochaperone deletion mutants.
At the other end of the spectrum, when abundance of metallochaperones is increased, activation and deactivation of VNG1179C proceeds normally, and as a result we do not observe significant differences in transcriptional dynamics of yvgX. However, intracellular Cu level rises because of the increased number of Cu-binding sites in the overexpressed metallochaperones. Importantly, we observe this only with overexpression of VNG2581H; as increased VNG0702H expression also increases Cu efflux by trafficking to YvgX.
Thus, the interplay between metallochaperones with distinct trafficking roles is critical for modulating transcriptional responsiveness and efficacy of Cu efflux. We have demonstrated that this system of interactions among metallochaperones and their targets sets an upper threshold for intracellular Cu levels. As a result, biological systems are under stringent selection pressure to maintain a fine balance in the activity of metallochaperones and their abundance. Changes to either can significantly affect responsiveness of the metalloregulator to modulate transcriptional dynamics of the efflux pump, and, ultimately, alter the homeostatic intracellular level of Cu.
In conclusion, while mathematical modeling of Cu trafficking has been performed previously [45], what is unique about this study is that it incorporated three iterations of experimentation and computation to refine model architecture and parameters (Figure 5). The modeling incorporated actual experimental measurements, recapitulated known dynamics, and predicted new dynamics, properties, and functions that were experimentally validated. Similarly, the experimental validations included microarray analysis to assay global transcriptional dynamics of the Cu response, GFP-based reporter assays to measure high resolution transcriptional dynamics of the Cu efflux pump, and ICP-MS measurements of intracellular Cu levels. This iterative computation and experimentation strongly supports a novel buffering role for metallochaperones to mechanistically explain the cause for elevated intracellular Cu levels and overexpression of the ATP7A efflux pump in cell lines harboring ATOX1 mutations [19]. Ultimately, we have presented a quantitative model that explicitly demonstrates the role of metallochaperones in regulating intracellular Cu, a contribution that is novel to the field of metal biology. Indeed, additional iterations of experimentation and computation are necessary to further refine this model and reveal new insights.
|
10.1371/journal.pgen.1003095 | Genes Contributing to Pain Sensitivity in the Normal Population: An Exome Sequencing Study | Sensitivity to pain varies considerably between individuals and is known to be heritable. Increased sensitivity to experimental pain is a risk factor for developing chronic pain, a common and debilitating but poorly understood symptom. To understand mechanisms underlying pain sensitivity and to search for rare gene variants (MAF<5%) influencing pain sensitivity, we explored the genetic variation in individuals' responses to experimental pain. Quantitative sensory testing to heat pain was performed in 2,500 volunteers from TwinsUK (TUK): exome sequencing to a depth of 70× was carried out on DNA from singletons at the high and low ends of the heat pain sensitivity distribution in two separate subsamples. Thus in TUK1, 101 pain-sensitive and 102 pain-insensitive were examined, while in TUK2 there were 114 and 96 individuals respectively. A combination of methods was used to test the association between rare variants and pain sensitivity, and the function of the genes identified was explored using network analysis. Using causal reasoning analysis on the genes with different patterns of SNVs by pain sensitivity status, we observed a significant enrichment of variants in genes of the angiotensin pathway (Bonferroni corrected p = 3.8×10−4). This pathway is already implicated in animal models and human studies of pain, supporting the notion that it may provide fruitful new targets in pain management. The approach of sequencing extreme exome variation in normal individuals has provided important insights into gene networks mediating pain sensitivity in humans and will be applicable to other common complex traits.
| Chronic widespread pain is a complex clinical problem. Identification of underlying genetic factors would shed light on the biology of pain and offer targets for novel therapies. We aimed to identify rare genetic variants in the normal population associated with pain sensation by performing exome sequencing on individuals who were more or less sensitive to heat pain. While we did not identify any single variants having large effect, we did observe major group differences between the sensitive and insensitive individuals. Network analysis suggested a role for the angiotensin pathway, which previous work in animal models has suggested is important in pain mediation. Our results cast light on the genetic factors underlying normal pain sensation in humans and the utility of exome analyses. It suggests that further exploration of the angiotensin pathway may reveal novel targets for the treatment of pain.
| Chronic pain has a prevalence of nearly 20% in Europe [1] and similar estimates are reported for North America. The symptom is poorly controlled by existing therapies and the resulting personal and socio-economic burden is considerable. While many analgesic drugs are available, the vast majority of analgesic prescriptions are drawn from two classes of drug, opiates and nonsteroidal anti-inflammatory-like drugs, and have either limited efficacy or significant side effects. There is, therefore, a considerable need to develop novel analgesic treatments. The use of human genetics for identification of intrinsic factors that contribute to chronic pain states is attractive for several reasons. Chronic pain conditions as well as experimentally induced pain have been shown to have a considerable genetic component [2]. Twin studies have shown observed heritabilities of about 50% for different pain traits [3]. The manifestation of pain in response to experimental stimuli such as skin heating, or to clinical pathologies such as joint degeneration, is known to vary markedly. It is clear that a range of factors, including personality, expectation and mental state modulate the expression of chronic pain and these features are themselves genetically mediated. Modelling in twins, however, suggests that there are two separate predisposing genetic factors [4] including variants that modulate sensitivity to pain, as well as those mediating anxiety and depression. A number of approaches to pain sensitivity genetics have been adopted including the examination of rare (monogenic) syndromes of pain insensitivity (reviewed in [5]) and candidate genes identified from transcriptional profiling in animal models [6]. Candidate gene studies in humans with chronic pain have been unconvincing, and confirmed candidate gene associations are still lacking (reviewed in [4] and [7]). The aim of the present study was to examine the influence of genetic variation, particularly rare variants having minor allele frequency <5%, on pain sensitivity in normal human volunteers. Two hypotheses were tested; that a single rare variant having large effect influences pain sensitivity and that the burden of variation would differ between sensitive and insensitive individuals.
Attempts to standardise and quantify pain sensibility in humans have led to the introduction of standardised thermal, mechanical or chemical stimuli that activate the nociceptive (pain signalling) system. Such quantitative sensory testing (QST) has been used to show that an individual's sensitivity to experimental pain predicts risk of developing chronic pain after surgical interventions such as hernia repair [8] and arthroscopy [9]. That pre-operative pain sensitivity is a major risk factor for chronic post-operative pain suggests that exploration of genetic variation underlying experimental pain might be a useful approach. The pain stimulus, its site of application and methods of rating have all been standardised - unlike spontaneous pain in a disease state. A further benefit is that the genetic influence on pain sensitivity is studied, rather than its influences on disease and disease progression. In the present study, we sought to determine whether rare variants associate with extremes of pain sensitivity in healthy volunteers. Using heat as the stimulus for QST in a large sample of healthy twin volunteers (www.twinsuk.ac.uk) we observed the normal variation in pain sensitivity using two objective tests, the heat pain threshold (HPT) and the heat pain suprathreshold (HPST). From a study population of >2500 individuals having QST, we compared approximately 200 individuals categorised as having high and low sensitivity to HPST (approximately 100 from each; TUK1 set) then repeated the process in a further 200 individuals (TUK2 set).
Our initial analysis sought to identify genes harbouring single nucleotide variants (SNVs) in either pain sensitive or insensitive subjects, with a focus on non-synonymous exonic and nonsense mutations. A large number of methods have been proposed for such an analysis [40]–[44]. We employed a battery of such tests including both old and new techniques, as well as tests examining a range of hypotheses; a difference between pain groups (sensitive vs. insensitive) in the proportion of subjects harbouring rare variants; a difference in abundance of rare variants, weighted by function; and a multivariate difference in variant patterns between the two groups, allowing simultaneous excess in either pain group for any single rare variant within a gene.
We found no single rare variant to have a statistically significant association to heat sensitivity, after multiple testing correction. The strongest signal was found for GZMM, a serine protease from immune cell granules. However, our network analysis identified up to 30 genes harbouring rare SNVs as belonging to the Angiotensin II pathway, which has previously been linked to the pain phenotype in a number of settings.
Singleton females were drawn from same sex twin pairs included in the sample so that gender- and relatedness bias were removed; after quality control all subjects were of north European descent. Complete data were available on 413 singleton subjects: TUK1 comprised 203 and TUK2 210 individuals. Analysis was performed in stages: TUK1, TUK2, combined TUK1 and TUK2 and pathway analysis. Details of the study participants are shown in Table 1. Based on the sample size required for exome sequencing and on the distributions obtained for HPT and HPST, insensitivity to heat pain was defined as HPST≥49.2°C, and sensitivity as HPST≤45.5°C. An individual designated insensitive/sensitive on HPST was included only if their HPT measure was higher/lower than the median HPT (46.6°C). The distributions of the TUK2 set were somewhat shifted, with median HPT = 46.0°C. Insensitivity to pain in TUK2 was defined as HPST> = 48.9°C while sensitivity was defined as HPST< = 45.4°C, with subjects required to have HPT above/below 46.0°C, respectively. Description of the exome sequencing findings in TUK1 and TUK2 groups is shown in Table S1. Details of the SNVs identified in the 2 datasets are shown in Table 2. The TUK2 set identified more variants of all types (except partial codons, which were extremely infrequent), which likely reflected the different exome capture arrays used and was consistent with greater coverage captured for the TUK2 set. However, the HapMap samples (n = 3) duplicated in the TUK1 and TUK2 exome sequencing showed no significant difference between number of SNVs called by the two platforms on commonly captured regions (by paired t-test, p = 0.24). The relative frequencies of the variants identified in the 2 datasets were compared to those recorded in dbSNP (http://www.ncbi.nlm.nih.gov/projects/SNP/) (Figure S1). Unsurprisingly, the majority of novel SNVs identified were rare, with estimated minor allele frequency (MAF)<0.005.
Results of comparison between the 21 tests for analysing rare variant association were represented using a heat map (correlation matrix, Table S2). As expected, methods having similar underlying assumptions provided highly correlated results and show “hot” on the heat map. A pair of tests was selected from each category/correlation block based on the correlation matrix and their QQ plots (Figure 1) giving 6 gene-centric variant burden tests employed in the final analysis. Using these 6 methods we identified genes containing variants associated with pain sensitivity, shown in Table 3 ranked by strength of evidence. Of the 20,038 exonic gene regions tested, 17,129 (from 14,109 unique genes) gave consistently non-missing p-values across the 6 selected variant burden tests. The p-value considered significant under Bonferroni correction that would apply for a single set of tests based on this number of genes was p<3.0e-06: no variants passed this threshold so none could be considered unequivocally associated with pain sensitivity. The gene GZMM was the most highly associated with heat pain sensitivity, p = 6.86e-05 in the combined TUK1 and TUK2 analysis. Variants identified in GZMM are shown in Figure 2. For SNV A95T there were 12 alleles (1×2+10) in the heat insensitive vs 1 in heat sensitive (p = 0.005, by Fisher's exact test) in TUK1. While in TUK2 we found 17 alleles in the heat insensitive vs 4 alleles in the heat sensitive (p = 0.0016). Individuals insensitive to heat pain manifested rare variants more frequently than the sensitive, across GZMM (Figure 2). Finally, the distribution of variants differed between the pain insensitive and sensitive groups, with the pain insensitive showing a relative enrichment of rare variants (Figure S2).
The 2nd lowest p-value among 6 gene-centric variant burden tests was used as a cut-off to prioritise genes for pathway analysis (see Methods, statistical analysis). After merging TUK1 and TUK2 datasets, we identified 138 unique genes harbouring a rare variant with a 2nd lowest p-value<0.01. First we examined the functional annotations of these 138 genes using the online functional annotation tool DAVID (http://david.abcc.ncifcrf.gov/) [10]. Nine high level GO terms were nominally significantly enriched in the gene list eg. “plasma membrane” and “intracellular signalling cascade”. None reached significance after multiple testing correction or offered obvious insights into mechanisms of altered pain sensitivity (results not shown). We applied causal reasoning to our data [11], which uses a large curated database of directed regulatory molecular interactions to identify the most plausible upstream regulators of a gene set. Of the 138 genes 86 were present in our database of causal interactions, from which we identified 4 nominally significant regulatory networks (Table 4). One of the regulatory networks, angiotensin II (Figure 3), was highly enriched for a pain signal with 12 out of 204 genes in the network also in the set of 86 genes with a nominal genetic burden. This yields an odds ratio of 7.6, an enrichment p = 3.4×10−7 and a correctness p = 1.2×10−8. Since 1108 pathways were tested, this adjusts to enrichment p = 3.8×10−4 and correctness p = 1.4×10−5 under multiple test correction.
We also investigated whether the genes identified were known to interact physically with proteins playing a role in pain. For this we used the BioGrid database of protein-protein interactions. Notable connections included the binding of synaptotagmin-9 (SYT9), a membrane trafficking protein activated by calcium, to TRPV1, the capsaicin receptor, which plays a key role in thermal nociception [12]. The extracellular matrix glycoprotein laminin B1 chain (LAMB1) interacts with the voltage dependent calcium channel Cav2.1 (CACNA1A) [13]. The receptor activity modifying protein 3 (RAMP3) binds to the calcitonin receptor (CALCRL), for transport to the membrane. Here the calcitonin receptor recognises the calcitonin gene related peptide (CGRP), a hormone proposed to contribute to pain transmission and inflammation [14]. Finally, the sodium-hydrogen exchanger regulatory factor 1 (SLC9A3R1), binds the beta-2-adrenergic receptor (ADRB2) [15], nitric oxide synthase 2 (NOS2) [16], membrane metallo-endopeptidase (MME) [17] and the opioid receptor kappa 1 (OPRK1) [18].
Patients with chronic pain have increased sensitivity to noxious stimuli such as heat and pressure compared to controls [19] as well as to non-noxious stimuli such as sound [20]. These observations support the notion that the processing of external stimuli is heightened or exaggerated in chronic pain states. Thus, people harbouring gene variants associated with greater sensitivity to heat pain stimulus are thought to be at increased risk of developing chronic widespread pain. The premise of this work was that understanding better the genetic influence on normal pain processing would shed light on the biological pathways underlying the pathology of chronic pain. In this project we adopted novel methods - biotechnological and statistical - to identify rare sequence variation contributing to pain sensitivity in normal individuals. The advent of high throughput genotyping technologies has helped to unravel the aetiology of many complex diseases and quantitative traits. In particular, genome-wide association (GWA) studies have uncovered many common variants associated with quantitative phenotypes. However, GWA is underpowered to detect association of rare variants, and the common variants identified so far explain only a fraction of the trait heritability. As whole-genome sequencing has become more cost-efficient it is now feasible to examine the effect of rare variants. The hypothesis that multiple rare variants explain a proportion of the missing heritability is gaining more attention [21].
Rare variants with moderate to high penetrance have been associated with a number of extreme phenotypes (summarised in [22]). For quantitative phenotypes, sampling and comparing the extremes of traits has become an accepted strategy for identifying disease-causing rare variants in exome sequencing [23]. In this novel exome project of pain perception in normal individuals, no genetic variants of large effect were identified. Considering that the statistical power after applying stringent multiple test correction was limited, we can't exclude moderate or small contributions by individual SNVs to the experimental pain phenotype. Indeed, we have noted a differential distribution of rare variants between the pain sensitive and insensitive subjects (Figure S2), which suggests enrichment of multiple SNVs of small effect at the extremes of the normal distribution. This study also provides proof of principle of the utility of the exome sequencing method.
Such an approach has been used successfully in the, albeit more limited, setting of sequencing ion channel genes in epilepsy [24]. The authors highlighted the need for cell and network analysis to optimise information obtained from such a study. A variety of statistical methods have been developed for analysis of association of rare variants with complex traits, but there remains a paucity of data regarding the genetic architecture underlying complex traits such as pain perception. For this reason we elected to use a variety of tests based on different underlying assumptions so that no rare variant associated with pain perception would be missed.
GZMM was the only gene classified as having “very high” evidence of association to thermal nociception (Table 3 and Figure 2: see Methods: statistical analysis for classification definitions). It encodes granzyme M, one of the serine proteases produced and stored in the granules of immune cells such as lymphocytes and natural killer cells [25]. While we could not find reports of association with pain in the literature, granzymes are known to play an important role in apoptosis [26] and in the initiation of inflammation: elevated levels have been detected in rheumatoid synovial fluid [27] and granzyme B expression increased in lesional atopic dermatitis skin [28]. In the “high” evidence category, the enzyme encoded by the seventh gene, DDAH1, plays a role in nitric oxide generation by regulating cellular methylarginine concentrations, which in turn inhibit nitric oxide synthase. Although both anti-nociceptive and pro-nociceptive roles of NO have been reported, overproduction of NO - together with free radicals - contribute to central sensitisation and the pathogenesis of abnormal pain states via association with NMDA receptor mediated signalling events. In support of this, circulating NO has been shown to be elevated in chronic widespread pain patients [29]. The links between pain and other genes listed in Table 3 (such as CCNJL and TBK1) are tenuous at present.
To explore further the interplay between the SNV-containing genes identified we applied causal reasoning, an algorithm using directed molecular relationships between biological entities to identify up-stream regulators of a set of input genes [30]. We identified 4 regulatory networks that were nominally significant, one of which (angiotensin II) remained significant after correction for multiple testing (correctness p = 1.4×10−5, enrichment p = 3.8×10−4). Angiotensin II is a peptide hormone involved in the control of blood pressure. This network connected 12 of our identified genes into a causal network (Figure 3). Angiotensin II has been already been implicated in central pain: it has been shown to facilitate pain-related behaviours in experimental animals [31] including responses to thermal stimuli similar to those employed in the current studies. The mechanism appears to be via the modulation of descending brainstem pathways. Blocking the receptors for angiotensin II (so called AT1 receptors) reverses some pain-related behaviours in models of chronic pain, suggesting a role for endogenous angiotensin II. For example, AT-1 receptor antagonist telmisartan has been shown to abrogate pain in the sciatic nerve constriction model in rats [32]. The data from several small clinical studies in humans have been conflicting [33], [34] but a recent phase II clinical trial of a AT2 receptor antagonist (AT2 receptors are expressed by primary afferent nociceptors) found a significant improvement in the pain of a group of patients with post-herpetic neuralgia (http://www.spinifexpharma.com.au/DRUG-DISCOVERY.html).
Our causal reasoning analysis allowed for only one interaction upstream of the genes in our dataset to be included. However, allowing two interactions increased the number of genes from this study that may be causally linked to angiotensin II to 30 genes. Angiotensin II can also be causally linked to known pain relevant processes. For example, PTGS2, the gene encoding cyclooxygenase 2 (COX-2, the target of the non-steroidal anti-inflammatory drugs) is regulated by angiotensin II [35]. COX-2 produces prostaglandin E2 (PGE2), which is released in damaged or inflamed tissues and binds to nociceptive nerve terminals via PGE2 receptors (so called EP receptors), leading to cAMP production. This leads to post-translational modification of several target proteins within nerve terminals that regulate nociceptor excitability, including voltage-gated sodium channels [36]. The current study using novel exome sequencing methods supports the notion that the angiotensin II pathway is important in pain regulation in man and suggests that genetic variation in the pathway may influence sensitivity to heat pain, at least in the Northern European population.
A third form of analysis examined the target genes in a network of all human protein-protein interactions from the BioGRID database. We asked if any of the proteins encoded by the genes identified in this study were known to interact directly with proteins having a role in pain. We found known physical interactions with several pain-relevant proteins including ion channels (TRPV1 and Cav2.1), the CGRP receptor and the kappa opioid receptor. It is clear therefore that although we did not identify any genes immediately associated with nociception, several play key roles in processes linked to the reception and transduction of pain signals by their physical and biochemical interactions with important pain mediating complexes.
This study highlights the potential of using a combination of sophisticated analytical methods to identify associations underlying rare variants in quantitative traits. While the predicted effect sizes are relatively small and require large samples, we have made progress in understanding the genetic architecture underlying heat pain sensitivity. Despite recent advances in both DNA sequencing technology and the statistical methods to analyse such complex datasets, the identification and follow-up of associations of individual gene variants remains a challenge. Our results lend weight to the notion that angiotensin II plays in important role in signal transduction in pain and this pathway merits further biological investigation.
Ethics committee approval was obtained from Guy's and St Thomas' Hospital research ethics committee. All subjects were volunteer singleton members of female monozygotic (MZ) and dizygotic (DZ) twins from the TwinsUK register of King's College London [37]. Thus we did not perform a classical twin study and did not need to adjust for relatedness. QST was performed according to standard methods (see Supporting information) in which measures of heat pain threshold (HPT) and heat pain suprathreshold (HPST) were made.
HPST score was selected as the primary metric because reproducibility was greater (intra-class correlation coefficients, HPST = 0.59 (0.51, 0.68); HPT = 0.34 (0.23, 0.46)). HPST was also found to have greater heritability (HPST h2 = 0.44; HPT h2 = 0.29). The two phenotypes were correlated (r = 0.64). To select subjects who were relatively pain sensitive/insensitive for exome sequencing, the following protocol was adopted: a subject was included only if their HPT score was in the same half of the distribution as the HPST and, in the case of MZ twin pairs, the co-twin also resided in the same HPST tail. For DZ twins, the entire pair was excluded if they fell into opposite tails; if both were in the same tail, the more extreme twin was selected. In no case were two members from a twin pair selected. In addition, three samples provided by HapMap were analysed twice – in TUK1 and TUK2 – to enable comparison of the methods. Additional detail is provided in Text S1.
DNA extracted from whole blood was sent to BGI for exome sequencing [38]. The qualified genomic DNA sample was randomly fragmented by Covaris technology with resultant library fragments 250–300 bp. Adapters were ligated to both ends of the fragments. Extracted DNA was amplified by ligation-mediated PCR (LM-PCR), purified and hybridized to the NimbleGen human exome arrays for enrichment; non-hybridized fragments were then washed out. The target enrichment of the TUK1 samples were performed using hybridization to the NimbleGen 2.1 M array, while the shotgun libraries of the TUK2 samples were enriched using NimbleGen EZ v2 library. The captured LM-PCR products were subjected to quantitative PCR to estimate the magnitude of enrichment. Each captured library was then loaded on Illumina platforms and high-throughput sequencing was performed on each library. The BGI used Illumina GAIIx for sequencing of the TUK1 samples and a Hiseq2000 platform for TUK2 samples. Raw image files were processed by Illumina base-calling software v1.6 (and v1.7), and the sequences of each individual were generated as 75 bp (and 90 bp) paired-end reads for TUK1 (and TUK2) sets respectively. The fastq files were generated from the raw data after removing the adapters and low quality reads.
Both datasets were mapped to the NCBI Human Reference (GRCh37; hg19) using BWA v0.5.5 (v0.5.9). We considered the default parameter –q 15 for read clipping, and a maximum insert size of 600 bp for proper pairing of the short reads. The alignment files for each lane were sorted and indexed by SAMtools [39] before constructing the library-level bam files. We also tried to improve the accuracy of the base quality scores by running a recalibration stage using Genome Analysis Toolkit (GATK) v1.0.5777 [40]. On average 5% of each library was contaminated with duplicate fragments, which were removed before variant calling. An extra step of local re-alignment was applied only to the TUK2 data to improve the sensitivity and specificity of mismatches near indel sites.
For quality control (QC) of the TUK1 data, we studied the histogram of depth distribution, the distribution of inferred insert sizes in the bam files, the GC content distribution for reads mapped to forward and reverse strands, the depth of coverage as a function of percentile of unique sequences ordered by GC content, and the fraction of each chromosome covered by the exomes. The distribution of per-base sequencing depth for each sample was evaluated as was the cumulative depth distributions in target regions, and sequencing depth and coverage of the target region per chromosome. The TUK2 dataset had a slightly higher depth of coverage over the capture target region (CTR), with average 71× depth (compared to 69× for TUK1), whereas average coverage of the CTR was 97.5% for TUK2 (compared to 96.5% in TUK1). In the TUK1 panel we discarded and re-sequenced a few lanes, which showed very low target coverage; hence requiring all the exomes to cover more than 70% of the CTR by at least 20× in both datasets. We observed that although the mean depth was comparable, the fraction of CTR covered at a given depth was generally lower in TUK2 set, e.g. CTR coverage at ≥20× was 80.3% for TUK2 compared to 89.1% for TUK1. This alludes to the greater coverage uniformity of the 2.1 M array compared with that of the solution-based EZ sequence capture.
For TUK1, we ran SAMtools v0.1.8 ‘pileup’ while limiting maximum depth for indels to 500. Then we filtered the SNVs (with ‘varFilter’) with SNV and indel Phred-scale quality scores less than 20, and minimum and maximum depth at 8 and 300 respectively. The GATK v1.0.5777 was run for TUK1 using default values and a minimum confidence threshold 30 and minimum read mapping quality at 10. We subsequently filtered the GATK SNVs by keeping only those with alternative allele quality score ≥ 20 and depth within [8,300] interval. For TUK2, we ran SAMtools v0.1.16 ‘mpileup’ together with ‘bcftools’ using default parameters, but requiring the SNV quality score and depth interval to satisfy the same criteria of the TUK1 set (i.e. QUAL≥20 and 300≥DP≥8). The GATK calling for TUK2 data followed the same procedure as for TUK1. We further filtered all the variants outside the capture target region. Overlapping results SAMtools and GATK were extracted. The discordance (about 5%) was largely attributed to unique calls, however we observed a small fraction (less than 1%) of SNVs called by both algorithms were assigned mismatching genotypes (homozygous non-reference vs heterozygous). We ran the GATK on the coordinates of the overlap to determine the non-variant genotypes hence adjusting the missing rates. The single-sample variant files were then merged (using ‘merge-vcf’) into two large variant call files (VCF) each containing the entire sample variants. Table S1 compares the SNV statistics for TUK1 and TUK2 samples.
We evaluated the genotype concordance between the exome and pre-existing GWAS datasets. We observed greater agreement between GWAS and TUK2 (average 99.8%) than GWAS and TUK1 (99.3% concordance) (Table S1). Three samples were identified as highly discordant with GWAS (52%, 54% and 51% rates). A multi-dimensional clustering analysis of these three exomes together with the entire GWAS dataset for 5,654 twins, confirmed that they were true outliers so were excluded from statistical analysis. Duplicate samples in TUK1 allowed estimation of genotype error rate. Out of ∼35 M bases on the 2.1 M array which had been genotyped, 295 and 374 sites were discordant between duplicates. This sets a type 1 error rate for genotyping of approximately 1.0e-05, or 0.001%.
The wide variety of methods to analyse rare variants generally fall into three broad categories: “collapsing” methods, which test for differences in rare variant accumulation; “carrier-based” tests, which test for differences in the number of subjects carrying a certain class of variant (usually at least partially based on frequency thresholds); and “multivariate” tests, which test for differences in variant patterns, and is further subdivided into kernel-based and regression-based methods. Using several tests from each category we ran 21 different gene-centric variant burden tests on the TUK1 set and the results correlated (and displayed as a “heat” map, Table S2). A pair of tests was selected from each category/correlation block based on the correlation matrix and the QQ plots (Figure 1). The six statistical methods selected for this project were:-
A primary list of genes harbouring rare variants was drawn up based on combining the p-values from TUK1 and TUK2 sets using Fisher's method. To identify signals from genes with concordant variant patterns across TUK1 and TUK2 datasets, the top genes from the merged raw TUK1 and TUK2 datasets were also considered as relevant signals. This combination did not comprise the primary list because the TUK1 and TUK2 sequencing were performed on different capture platforms: some regions did not overlap between the two. Further details are provided in Supporting information.
In addition to the issue of combining TUK1 and TUK2 was the challenge of combining and sorting the results of the 6 gene-centric variant burden tests which were relatively new and not well understood. Because the 6 tests comprised 3 pairs of similar test methods (Table S2) we considered that a result was not robust if it was significant for only 1 test category. Significance in more than 1 category added confidence that a result was less likely to be a false signal. To prioritize genes that were either significant in more than one category or consistently significant for both tests within a pair, we prioritized genes based on the 2nd lowest p-value from the 6 selected tests. This approach also ensured that the top gene list could not be dominated by anomalies from a single test. Significant results using the 2nd lowest p-value were obtained in two ways: from combining the TUK1 and TUK2 p-values via Fisher's formula, and by merging the datasets (Table 3). A gene was classified as “High” evidence if its 2nd lowest p-value achieved p<0.00044 (the p-value such that replication would achieve a genome-wide significant meta-analytic p-value), and “Very High” if this occurred with the combined dataset being more significant than the combination of the p-values across the two halves. “Medium” priority was given any gene which achieved p<0.001 for its 2nd lowest p-value in either the merged dataset or the combination of the p-values across the two halves.
After removing genes showing an opposite direction of effect and after merging the datasets, we identified 138 unique genes having a 2nd lowest p-value<0.01. These were considered for more detailed analysis. We looked first for enriched Gene Ontology categories within these genes using DAVID [10] with an EASE p-value<0.05. Then we undertook causal reasoning [11] which uses a large curated database of directed regulatory molecular interactions to identify the most plausible upstream regulators of a gene set. Consequently it allows the recapitulation of regulatory networks/pathways associated with genes of interest. The method offers two measures of statistical significance. The enrichment p-value corresponds to a standard gene set enrichment test on the set of downstream genes, whereas the correctness p-value takes the direction of regulation into account. For the latter, each associated gene was considered as a down-regulated transcript in the causal reasoning network ie. assumed loss-of-function mutations. As a background set for the significance calculations we considered the intersection of the set of all genes covered in either the TUK1 or TUK2 study and all transcripts in our causal reasoning database. This set consists of 9275 genes. A regulatory hypothesis was considered nominally significant with a p-value<0.05 and significant at a 0.05 level after application of the Bonferroni correction for multiple testing. As we are considering 1108 potential upstream regulators in the underlying database, a Bonferroni corrected p of 0.05 corresponds to a nominal p-value of 4.5×10−5. Finally, we searched for direct physical interactions between proteins identified in this study and proteins known to have a role in pain using protein interaction data from the BioGrid database [46].
|
10.1371/journal.ppat.1007126 | Defects in intracellular trafficking of fungal cell wall synthases lead to aberrant host immune recognition | The human fungal pathogen, Cryptococcus neoformans, dramatically alters its cell wall, both in size and composition, upon entering the host. This cell wall remodeling is essential for host immune avoidance by this pathogen. In a genetic screen for mutants with changes in their cell wall, we identified a novel protein, Mar1, that controls cell wall organization and immune evasion. Through phenotypic studies of a loss-of-function strain, we have demonstrated that the mar1Δ mutant has an aberrant cell surface and a defect in polysaccharide capsule attachment, resulting in attenuated virulence. Furthermore, the mar1Δ mutant displays increased staining for exposed cell wall chitin and chitosan when the cells are grown in host-like tissue culture conditions. However, HPLC analysis of whole cell walls and RT-PCR analysis of cell wall synthase genes demonstrated that this increased chitin exposure is likely due to decreased levels of glucans and mannans in the outer cell wall layers. We observed that the Mar1 protein differentially localizes to cellular membranes in a condition dependent manner, and we have further shown that the mar1Δ mutant displays defects in intracellular trafficking, resulting in a mislocalization of the β-glucan synthase catalytic subunit, Fks1. These cell surface changes influence the host-pathogen interaction, resulting in increased macrophage activation to microbial challenge in vitro. We established that several host innate immune signaling proteins are required for the observed macrophage activation, including the Card9 and MyD88 adaptor proteins, as well as the Dectin-1 and TLR2 pattern recognition receptors. These studies explore novel mechanisms by which a microbial pathogen regulates its cell surface in response to the host, as well as how dysregulation of this adaptive response leads to defective immune avoidance.
| Disease causing microorganisms have adapted many strategies to avoid host immune detection in order to facilitate their survival. Pathogenic fungi alter their cell surface in order to mask immune stimulatory epitopes in their cell wall. We have identified a novel protein involved in this host-induced cell wall remodeling process in Cryptococcus neoformans. We have created a mutant strain with a targeted deletion of this gene and demonstrated that this protein is involved in intracellular trafficking of the β-(1,3)-glucan synthase enzyme and is differentially localized in response to host-like conditions. In the absence of this protein, the cell wall has decreased levels of β-glucan and increased exposure of chitooligomers. Importantly, improper cell wall maintenance by this mutant leads to a hyper-immunostimulatory fungal cell surface that induces increased macrophage activation. We show that this activation is dependent on several pattern recognition receptors previously demonstrated to be important for fungal pathogenesis.
| The microbial surface is the first point of contact for interactions with the innate immune system, representing the site at which an infected host might recognize a microbe as a potential pathogen. This recognition is achieved through host pattern recognition receptors (PRRs) that distinguish specific pathogen-associated molecular patterns (PAMPs) on microbial surfaces, directing downstream signaling events that ultimately lead to the initiation of an immune response. The fungal cell wall is a dynamic structure composed of a complex matrix of polysaccharides including α- and β-glucans, mannoproteins (mannans), and chitin/chitosan. These fungal specific components have been shown by many groups to be recognized by host PRRs including Toll-like receptors (TLRs) and C-type lectin receptors (CLRs) [1,2]. Several fungi have developed strategies to mask their surfaces from immune detection. Examples include Histoplasma capsulatum cell wall α-(1,3)-glucan and Aspergillus fumigatus conidial RodA hydrophobin, which both serve to block exposure of the more immunogenic β-glucan molecule [3,4]. The fungal cell surface is also responsive to different environments, including various micro-environments within the infected host. For example, Candida albicans differentially exposes β-glucan in response to diverse host niches, drug treatments, and growth conditions, resulting in varying degrees of Dectin-1-mediated host responses [5,6].
The opportunistic human fungal pathogen Cryptococcus neoformans continues to be a significant health threat for immune compromised populations, particularly those with HIV/AIDS, among whom it causes over 175,000 deaths per year [7]. This ubiquitous fungus colonizes the lungs after inhalation from the environment. It can then disseminate to the central nervous system in immunocompromised individuals, where it causes life-threatening meningoencephalitis [8]. C. neoformans has developed several adaptations to avoid immune detection and to direct the host immune response in its favor. The polysaccharide capsule on the cell surface shields immunostimulatory cell wall components from host recognition. Additionally, secreted capsular material actively represses various immune responses [9]. C. neoformans cells can also grow to massive sizes, forming so-called titan cells in the setting of infection. These giant and hyper-encapsulated cells are unable to be engulfed by host immune cells, but instead they drive a non-protective immune response to C. neoformans leading to pathogen persistence [10–12].
Both capsule production and titan cell formation are induced in the host environment and involve significant cell wall remodeling. Previous work has shown that capsule polysaccharide likely attaches to the cell surface through interaction with α-(1,3)-glucan in the cell wall [13]. Titan cell walls are also thicker and more chitin-rich then normal sized cells [12]. In addition to these macromolecular cell surface changes, our lab has shown that C. neoformans actively remodels its cell wall in response to host pH signals in order to avoid immune detection [14–16]. We have recently demonstrated that aberrant chitooligomer exposure leads to a detrimental immune response [16]. Although few investigators have performed detailed cell wall analyses in C. neoformans, this structure appears to contain significantly more chitin and chitosan than that of other pathogenic species [17]. There is increasing evidence to support the concept of chitin as an immune modulatory molecule, and recent studies have highlighted the complexity of chitin recognition, indicating that the source and size of chitin molecules can differentially direct immune responses [18–20].
C. neoformans cell wall regulation, particularly in response to the host environment, remains incompletely defined. Our lab has previously identified cell surface properties that drive immune detection and through a previously published screen for cell wall regulators, we have identified a novel protein involved in this process [14,21,22]. In this study we explore the mechanism by which this protein, Mar1, regulates cell wall remodeling, and the implications of an aberrant cell wall architecture on host immune detection. We report that in the absence of Mar1, cells display a defect in capsule attachment, as well as increased exposure of cell wall chitin and chitosan. This concurs with our observation of decreased levels of glucans and mannans in the cell wall, decreased expression of α- and β-glucan synthases, and mislocalization of the β-glucan synthase, Fks1. While canonical secretion is intact in the mar1Δ mutant strain, general intracellular trafficking appears altered. We also report the localization of the Mar1 protein itself to cellular membranes and demonstrate differential localization in host-like tissue culture conditions. The implications of this cell wall remodeling defect include increased recognition and activation by innate immune cells and attenuated virulence. We further show that this innate immune recognition is dependent on the Card9 and MyD88 adaptor proteins and the cell surface receptors Dectin-1 and TLR2. In addition to highlighting intracellular processes involved in cell wall remodeling, these studies underscore the importance of proper cell wall regulation in host immune detection and broaden our understanding of the recognition of individual fungal cell wall components.
The cell wall is the interface between microbial pathogens and host immune cells. To identify genes required for proper fungal cell wall homeostasis, we performed a random mutagenesis screen in the human fungal pathogen Cryptococcus neoformans using Agrobacterium tumefaciens-mediated transformation (AMT). Mutagenized strains were screened for phenotypes corresponding to cell wall changes, including growth impairment and dry colony morphology in alkaline conditions (pH 8) and sensitivity to elevated salt concentrations (1.5 M NaCl). We hypothesized that a subset of these cell wall mutants would also display alterations in the host-pathogen interaction. Preliminary results from this screen were previously reported [22].
From this screen, we identified a mutant displaying dry colony morphology and slight sensitivity to alkaline pH as well as decreased growth on elevated salt concentrations. This strain had a mutation in a previously uncharacterized gene (CNAG_06695), which we have named MAR1 (macrophage activating regulator of cell wall-1). The encoded protein is predicted to contain two transmembrane domains, which comprise a larger domain of unknown function (DUF4112) (Fig 1A). However, Mar1 has no annotated predicted functions based on its sequence, and it shares no significant sequence homology with proteins from other species, including other basidiomycetes, with the exception of highly related Cryptococcus species (C. deneoformans, C. deuterogattii, and C. gattii). We confirmed the phenotypes of this mutant by independently disrupting the entire MAR1 gene in the wild type background (Fig 1B). We also complemented all phenotypes by reintroduction of the wild type allele into the mar1Δ mutant.
We assessed the sensitivity of the mar1Δ mutant to common cell wall stressors. These agents included calcofluor white (CFW, binds and blocks chitin assembly), Congo red (inhibits assembly of cell wall polymers, especially chitin), caffeine (affects signal transduction and general cell wall integrity), and SDS (cell membrane stressor) [23–26]. When incubated in the presence of cell wall stressors, growth of the mar1Δ strain was severely inhibited by Congo red (0.5%) and SDS (0.06%) compared to the WT strain (Fig 1C). In contrast, the mar1Δ mutant only displayed a slight decrease in colony size on CFW (1 mg/ml), and growth was comparable to WT when incubated in the presence of caffeine (1 mg/ml).
Given the enhanced susceptibilities of the mar1Δ mutant to cell wall perturbing agents, we considered that the Mar1 protein might be involved in other cell-signaling pathways that regulate or respond to defects in cell wall integrity, including (1) the Rim/alkaline response pathway (2) the PKA/cAMP pathway, (3) the PKC/cell wall integrity (CWI) pathway, (4) the calcineurin pathway, and (5) the high-osmolality-glycerol (HOG) pathway. Therefore, we compared the mar1Δ mutant to strains with mutations in these other pathways (representative mutants used for each pathway: (1) rim101Δ; (2) pka1Δ; (3) mpk1Δ; (4) cna1Δ; (5) hog1Δ) to specifically test patterns of susceptibility to common cell wall stressors (Fig 1C). While sensitivity to Congo red and SDS is observed in mutants of the PKC/CWI and calcineurin pathways, these strains are also sensitive to caffeine. In a manner distinct from the SDS susceptibility of the mar1Δ strain, Rim and PKA/cAMP pathway mutants grew more robustly on SDS. Hog pathway strains were exclusively sensitive to SDS and no other stressors tested.
Additionally, we tested these strains for phenotypes associated with virulence including melanin production and the ability to grow at high temperature. The mar1Δ mutant displayed a modest growth defect when incubated at 37°C, however melanin production was comparable to WT (Fig 1C). These phenotypes contrasted sharply with calcineurin pathway mutants, which display defective thermotolerance, and PKA pathway mutants, that have defective melanin production [27,28]. Due to the modest temperature sensitivity of the mar1Δ strain, we also tested the effect of the Hsp90 inhibitor, radicicol, on mar1Δ growth at 30°C and 37°C. We observed identical MICs for WT and mar1Δ cells (12.5 μM at 30°C and 1.56 μM at 37°C). Together these data indicate that MAR1 is required for normal cell wall integrity under certain cell wall stress conditions. However, this particular combination of sensitivities does not precisely mimic that of mutants in any of the previously studied cell wall responsive pathways.
The C. neoformans cell wall serves as the site of attachment for the polysaccharides that comprise the cell surface capsule [13]. Therefore, some C. neoformans strains with defects in cell wall organization and structure also display defective encapsulation. We incubated the mar1Δ strain in capsule-inducing tissue culture medium and assessed capsule microscopically by India ink staining. Compared to WT, the mar1Δ mutant displayed a marked reduction in surface capsule (Fig 2A). Capsular polysaccharide is synthesized in the cytoplasm, and then secreted, where it binds to the cell surface [29]. To differentiate between a capsule biosynthesis and capsule attachment defect, we assayed the relative amount of secreted capsule in the culture supernatant using a previously described immuno-blotting technique [21,30]. In brief, we used the mAb18B7 monoclonal antibody directed against the main capsule component GXM (glucuronoxylomannan) to probe for secreted capsule polysaccharide. By this method, we observed WT-levels of this capsular polysaccharide secreted by the mar1Δ mutant (Fig 2B). However, the electrophoretic mobility of this polysaccharide appears to differ slightly from the WT and complemented strains. These results indicate that mar1Δ has no defect in total capsule polysaccharide production, but that the defect in surface encapsulation is likely due to a defect in capsule attachment, and perhaps polysaccharide composition/structure. This pattern of altered capsule attachment is distinct from PKA pathway mutants that display impaired capsule synthesis [21]; however, capsule attachment defects have been observed in other strains with cell wall defects [13,21].
To more fully explore the cell wall changes of the mar1Δ mutant, we performed microscopy with stains and antibodies specific for various cell wall components. After the mar1Δ mutant was incubated in host-like tissue-culture medium (TC), we observed a striking increase in its staining by FITC-conjugated wheat germ agglutinin (WGA), a lectin that recognizes exposed chitin and chitooligomers (Fig 2C) [16,31]. This pattern of increased WGA staining does not occur when the cells are incubated in rich medium (YPD), indicating this chitooligomer-exposure phenotype is induced by host-like TC conditions (Fig 2C and 2D). These initial observations were quantified, demonstrating a significant increase in the average fluorescence of WGA-stained mar1Δ cells incubated in TC medium compared to WT cells or mar1Δ cells incubated in YPD (Fig 2D).
Whereas chitooligomer exposure is well approximated by quantifying cell surface binding of the large WGA molecule, total cell wall chitooligomer content is better assessed using calcofluor white (CFW), a smaller fluorescent molecule that more easily penetrates into the cell surface and binds chitin and other chitooligomers. In contrast to the marked shift in WGA staining, we only observed a slight increase in CFW staining of mar1Δ cells compared to WT, and only when the cells were incubated in TC medium (Fig 2C and 2D).
We also measured levels of chitosan, the de-acetylated form of chitin, by Eosin Y (EY) staining (Fig 2E) [32]. Similar to staining for chitin, we did not observe a major change in fluorescence when cells were incubated in YPD (Fig 2E); however, we observed enhanced staining of mar1Δ cells incubated in TC medium compared to WT cells. Together these cell wall staining results show that in host-like tissue culture conditions the mar1Δ strain has increased exposure of chitooligomers, including both chitin and chitosan.
To better understand how host-like tissue culture conditions induce cell wall changes in the mar1Δ mutant, we sought to better define what component of this condition is responsible. To determine if the mar1Δ cell wall changes were dependent on temperature, we incubated cells in YPD or TC media at 30°C and 37°C, and measured staining of exposed chitooligomers by WGA (Fig 3A and 3B, S1A Fig). We observed a small, but not significant, increase in staining of cells incubated in YPD at 37°C. By contrast, cells incubated in TC regardless of temperature displayed significant and equivalent increases in staining. These data indicate that increased temperature is not the major driver of the mar1Δ cell wall changes.
We next tested the role for pH in inducing cell wall changes. Standard YPD medium has a pH of 5–6, while TC medium is buffered to pH 7.4. To test if increased pH could induce the mar1Δ mutant cell wall changes, we buffered YPD to pH 7.4 and measured staining by WGA (Fig 3C and 3D, S1A Fig). We observed a small, but significant increase in chitooligomer staining of mar1Δ cells, suggesting that increased pH is involved in TC-induced cell wall changes. Interestingly, when we buffered YPD to pH 8, we observed increased WGA staining of both WT and mar1Δ cells (S1B Fig).
While higher pH induced increased WGA staining, it did not fully recapitulate the level of staining observed in TC medium. We next tried to suppress the mar1Δ cell wall changes by supplementing TC medium with common nutrients found in rich medium. We observed that when we incubated mar1Δ cells in TC medium supplemented with 2% glucose, the intensity of WGA staining was partially suppressed (Fig 3C and 3D). We did not observe any changes in WT staining or any suppression in TC medium supplemented with 1x complete amino acids (S1A and S1B Fig). Together these data indicate that a combination of increased pH and decreased glucose availability in TC medium induces increased chitooligomer exposure of the mar1Δ mutant.
We used several independent methods to further quantify the relative levels of cell wall carbohydrates in our strains. First, we used an enzymatic method [17] to quantify total chitin and chitosan levels in the mar1Δ and WT cell walls after incubation in TC medium, the condition in which we observe differences in staining intensity. Compared to the WT, the mar1Δ mutant displayed no increase in total chitin or chitosan using this biochemical assay (Fig 2F).
We also quantified our initial staining observations using flow cytometry, demonstrating a striking increase in the mean fluorescence intensity (MFI) of WGA-stained mar1Δ cells compared to WT (Fig 4A). Similar to what we saw by staining, we observed a modest increase in the MFI of CFW-stained mar1Δ cells (Fig 4A). Interestingly, we observed two peaks of EY-stained mar1Δ cells, one corresponding to the mean fluorescence intensity (MFI) of WT stained cells, and a second shifted to a higher MFI (Fig 4A). This suggests a non-homogeneous pattern of chitosan exposure after incubation in host-like conditions in mar1Δ cells. We were unable to accurately assess the relative levels of other cell wall carbohydrates such as α-glucan, β-glucan, and mannoproteins using staining and microscopy methods adapted from other fungal species. In all cases, the staining signal of WT cells was similar to unstained controls (Fig 4A). Therefore, we utilized biochemical methods to quantify cell wall carbohydrate composition. We extracted total cell wall material from our strains and used high-performance liquid chromatography (HPLC) to quantify the levels of glucosamine (chitin and chitosan together), glucose (glucans), and mannose (mannosylated proteins) in these extracts [16,33]. Similar to our enzymatic measurements of chitin and chitosan, we observed no significant difference in the level of glucosamine between the WT and mar1Δ strains (Fig 4B). However, by HPLC, we measured a significant decrease in the levels of glucose and mannose in the mar1Δ mutant compared to the WT (Fig 4B). Together, these results suggest that Mar1 is required to maintain normal levels of the outer glucan and mannan cell wall layers. Accordingly, in the absence of functional Mar1, the resulting changes in outer cell wall carbohydrate abundance result in increased exposure, but not total levels, of the inner cell wall chitooligomers, chitin and chitosan.
To determine the site of regulation for these cell wall changes, we performed quantitative real time PCR analysis of selected cell wall synthesis genes in the WT and mar1Δ strains. Interestingly, as the WT strain transitions from a rich medium to tissue culture medium, there is a marked increase in the expression of genes encoding α- and β-glucans (Fig 4C). In contrast, there is no significant transcriptional change in most chitin synthase and chitin deacetylase genes during this environmental transition, except for the CHS4 chitin synthase. Consistent with our chitooligomer cell wall staining and quantification data, the expression of these chitin synthase and chitin deacetylase genes was not significantly different between WT and mar1Δ in tissue culture medium. By contrast, the genes encoding the major glucan synthases, including the α-glucan synthase AGS1 and the β-1,3-glucan synthase FKS1, displayed statistically significantly decreased expression in tissue culture medium in mar1Δ compared to WT. These data indicate that Mar1 is required for the transcriptional induction of glucan cell wall genes, including AGS1 and FKS1, that are typically upregulated during host-like conditions.
We chose to more fully explore the relationship of Mar1 function with the Rim and CWI signaling pathways given our observation that the mutant strains associated with each of these pathways most closely shared cell wall-related phenotypes with the mar1Δ strain. C. neoformans strains defective in Rim pathway signaling are unable to grow well at alkaline pH or in the presence of elevated salt concentrations, similar to the mar1Δ strain. This signaling pathway is responsible for activating the Rim101 transcription factor, which in turn regulates the expression of many genes required for cell wall integrity and organization during growth in host-like conditions. Additionally, Rim pathway mutants have increased chitooligomer exposure [16].
The MAR1 gene was not identified as a Rim101 target in previously published comparative transcriptional profiling experiments [14]. However, direct analysis of the raw data from these experiments (NCBI GEO database accession number GSE43189) indicated that the MAR1 locus was not included in the statistical analyses due to its relatively low transcript abundance. To more definitively determine whether MAR1 is a downstream target of the Rim101 transcription factor, we first performed quantitative real time PCR to measure the expression of MAR1 in the WT background in YPD and TC media. We observed that MAR1 had significantly induced expression in TC medium compared to YPD (p = 0.0017) (S2A Fig). To next determine if the expression of MAR1 is regulated in a Rim101-dependent manner, we analyzed updated transcriptional profiling of the rim101Δ strain in TC medium recently carried out in our laboratory (NCBI GEO database accession number GSE110723). Expression of the MAR1 transcript in the rim101Δ mutant versus WT was only modestly reduced (log2-fold change = 0.772, p = 0.04), suggesting that MAR1 is likely not a major target of Rim101.
We also assessed whether Mar1 might function in the Rim signaling pathway upstream of Rim101. Rim pathway activation occurs in a pH-dependent manner, culminating in the cleavage of the Rim101 transcription factor and its translocation to the nucleus [34]. Rim101 cleavage and nuclear localization are disrupted in mutants of upstream Rim pathway activators [35]. In contrast, Rim101 cleavage/activation is intact in the mar1Δ background (S2B Fig). Together these data indicate that Mar1 likely operates in cellular processes independent of the C. neoformans Rim pathway, despite phenotypic similarities of these mutant strains.
The PKC/CWI pathway is responsible for the activation of the Mpk1 MAP kinase protein, which in turn coordinates enhanced cell wall stress resistance. Accordingly, mutants in this pathway display defective phosphorylation of Mpk1. We therefore assessed Mpk1 phosphorylation in response to cell wall stress in the WT and mar1Δ mutant strains. Western blots of total cell lysates from these strains were analyzed using an antibody directed against phosphorylated (activated) Mpk1 [36,37]. After 3.5 hours of incubation in YPD, WT and mar1Δ cells had a modest level of phosphorylated Mpk1 (S3A Fig). After incubation in TC medium, both WT and mar1Δ displayed enhanced Mpk1 phosphorylation (S3A Fig). These results demonstrate that Mar1 is not required for CWI integrity pathway activation under host-like tissue culture conditions as measured by Mpk1 phosphorylation. To further test this, we constructed a mar1Δ mpk1Δ double mutant and analyzed chitooligomer exposure by WGA staining. In agreement with the mar1Δ mutation inducing distinct cell wall changes, the mar1Δ mpk1Δ double mutant displayed more WGA staining then either mpk1Δ or mar1Δ single mutants (S3B Fig). The patterns of Rim and CWI signaling pathway activation in the mar1Δ strain suggest that the Mar1 protein regulates cell wall integrity in a novel manner.
To better understand the function of Mar1, we generated an N-terminally tagged GFP-Mar1 fusion protein and analyzed its localization by microscopy. We expressed this fusion protein under the constitutively active HIS3 promoter in the mar1Δ mutant strain background. This fusion protein was functional, displaying partial suppression of mar1Δ dry colony morphology on pH 8 plates. After overnight incubation in YPD medium, GFP-Mar1 localized throughout the cell to small punctate structures on endomembranes and at the cell surface (Fig 5A). 3D-projection of Z-stacked images indicated that many of the observed puncta were located on the cell surface (Fig 5B). Compared to what was observed in YPD-incubated cells, the number of GFP-Mar1 puncta was decreased after incubation in TC medium (Fig 5A and 5B). These puncta also appeared larger and more globular in nature, and overall intracellular endomembrane staining was less intense after incubation in TC medium.
To further characterize the dynamics of GFP-Mar1 protein localization, we analyzed protein fluorescence over time after shifting cells to TC medium (Fig 5C). After 2 and 4 hours in TC medium, GFP-Mar1 localizes to small punctate structures as well as to endomembranes, similar to what we observed for cells incubated in YPD overnight. After 6 hours, some GFP-Mar1 puncta begin to appear globular, more similar to what we observed for cells incubated in TC overnight. By 8 hours, the majority of GFP-Mar1 puncta appear globular. Interestingly, when we measured mar1Δ chitooligomer exposure over time in TC medium, we observed a similar time course of changes (Fig 5D). Increased WGA staining of mar1Δ cells can be observed beginning around 6 hours after incubation in TC medium, similar to when globular GFP-Mar1 puncta are seen.
The dynamic localization of GFP-Mar1 led us to next investigate intracellular trafficking processes in the mar1Δ mutant strain. To determine if these processes were impaired, we used the lipophilic dye FM4-64 to assess rates of endocytosis in YPD and TC media (Fig 6A). Medium-sized, bright endocytic vesicles can be seen after 30 minutes of staining in WT and mar1Δ cells incubated in YPD. Endocytic vesicles can also be observed in WT cells incubated in TC medium; However, by contrast, less well-defined, tubular structures are observed in mar1Δ cells incubated in TC medium.
To further assess intracellular trafficking, we measured acid and alkaline phosphatase activity in cell supernatants. In S. cerevisiae, acid phosphatase is secreted through the canonical secretory pathway, whereas alkaline phosphatase is trafficked to the vacuolar membrane through the alternative ALP pathway [38]. Both enzymes are induced under low phosphate conditions and can be assayed by measuring the colorimetric hydrolysis of the para-Nitrophenylphosphate (pNPP) substrate. Over a time-course of incubation with the pNPP substrate in acidic medium, phosphate starved (induced) WT and mar1Δ cells showed increased phosphatase activity compared to phosphate replete (non-induced) cells (Fig 6B). A similar induced increase in phosphatase activity was observed for WT cells in alkaline conditions; however, there was a minimal difference in phosphatase activity between phosphate starved and phosphate replete mar1Δ cells in alkaline conditions, suggesting a defect in secreted alkaline phosphatase activity (Fig 6C). Together, these data are consistent with defects in endocytic and vesicular trafficking in the mar1Δ mutant, but not a defect in classical secretion.
To further explore the mar1Δ mutant defect in glucan abundance, we constructed a C-terminally tagged Fks1-GFP fusion protein. We transformed this allele into the WT H99 background and selected those transformants in which the allele replaced the endogenous locus for further analysis. Previous work has indicated that FKS1 is an essential gene in C. neoformans [39]. Therefore, the ability to replace the WT FKS1 allele with the FKS1-GFP allele suggests that this fluorescent fusion protein is functional. We subsequently mutated the MAR1 gene in the strain expressing Fks1-GFP, and we analyzed the localization of Fks1-GFP in two independent mar1Δ mutants. After incubation in YPD medium, we observed by fluorescent microscopy that Fks1-GFP localizes to punctate structures on cellular membranes in both the WT and mar1Δ backgrounds (S4A Fig). After incubation in TC medium, Fks1-GFP localization is more heterogeneous among the cell population. It is still present in puncta throughout the cell in the majority of cells, but it is enriched at the plasma membrane of some cells in the WT background, predominately mother cells (Fig 6D). By contrast, Fks1-GFP localization is more homogeneous among cells in the mar1Δ background, with reduced plasma membrane puncta and very few cells displaying the uniform plasma membrane localization observed in the WT (Fig 6D and S4B Fig). These data suggest that Mar1 is required for proper trafficking and localization of Fks1 to the cell membrane in TC medium.
To better visualize the cell wall changes of the mar1Δ mutant, we examined our strains by transmission electron microscopy (TEM) after incubation in either YPD or TC media. In rich YPD medium, both WT and mar1Δ cells had thin, organized cell walls with a well-defined lamellar appearance (Fig 7A). The mar1Δ cell walls displayed a trend to be slightly thinner than WT cell walls in this condition, but this difference was not statistically significant (p = 0.3431). In TC medium, the cell walls of WT cells remained compact and well-ordered (Fig 7B). In contrast, mar1Δ cell walls were less compact and well-organized after incubation in TC medium and significantly thicker than WT or mar1Δ cell walls incubated in YPD (p = 0.0011 and 0.0007 respectively). Several cells appeared to have layers of cell wall material sloughing away from the cell periphery, a phenotype that was not observed for WT cells (Fig 7B). Accordingly, a significant amount of debris was also observed in the space between cells in the mar1Δ TC samples, likely representing degraded cell wall material, or perhaps changes in chemical cross-linking resulting in altered cell wall integrity during sample processing.
In addition to the notable TC-induced cell wall changes in the mar1Δ strain, the TEM images also suggested alterations in vesicular trafficking. In both WT and mar1Δ strains, there were numerous membrane-bound vesicles. These structures have been suggested to be secretory vesicles transporting cargo toward the cell surface [40]. In TC medium, both the WT and mar1Δ strains demonstrated a relative increase in the number of these vesicles carrying electron-dense material localized near the cell surface. In mar1Δ cells, an increase in large electron-lucent vesicles was also observed (Fig 7C). In some cells, multiple large “empty” vesicles can be seen near the cell periphery. In S. cerevisiae, the accumulation of similar vesicles has been observed for several secretory mutants [41,42]. Similarly, the sav1Δ secretory mutant in C. neoformans accumulates post-Golgi secondary vesicles [43]. Along with the altered FM4-64 staining and alkaline phosphatase secretion described, these images support a defect in intracellular trafficking in this mutant.
Macrophages and dendritic cells are likely the first immune cells that C. neoformans contacts within infected lungs. To determine if the changes in the mar1Δ cell wall would affect this interaction, we quantified the production of tumor necrosis factor-alpha (TNF-α) after co-culturing WT and mar1Δ strains with primary bone marrow-derived macrophages (BMMs) and bone marrow-derived dendritic cells (BMDCs). We have previously used this pro-inflammatory cytokine as a marker of macrophage activation in vitro, observing that C. neoformans strains with increased chitooligomer exposure often induce more TNF-α production than WT [14,16]. Accordingly, we observed that mar1Δ cells induced significantly more TNF-α production from both BMMs and BMDCs, compared to the WT or reconstituted strains (Fig 8A and 8B). To determine if this is due to an active cellular process, we tested the macrophage response after co-culturing with heat-killed mar1Δ cells. Similar to live cells, we found that heat-killed mar1Δ cells also induce increased TNF-α production by BMMs, demonstrating that mar1Δ macrophage stimulation is not dependent on cell viability (Fig 8A).
Due to our observation that the mar1Δ cell wall changes are only induced when the cells are incubated in host-mimicking tissue culture conditions, we tested both YPD- and tissue culture media-incubated cells in our BMM assays. Consistent with the host-induced changes in the mar1Δ cell wall, only cells pre-incubated in TC medium prior to co-culture induced increased TNF-α production from BMMs (Fig 8C). This was the case for both live and heat-killed cells. We also examined whether the cell wall itself was sufficient to elicit a response from BMMs. Compared to WT and reconstituted strain controls, we observed that isolated cell walls from the mar1Δ strain pre-incubated in TC medium induced more TNF-α production by BMMs (Fig 8D). Together these data suggest that macrophage activation by this strain is dependent on specific cell wall changes that are induced by host-mimicking tissue culture conditions.
The polysaccharide capsule can serve to shield the immunogenic cell surface of C. neoformans from immune recognition. Additionally, capsule polysaccharide itself can be immunosuppressive [9]. We have observed that the mar1Δ mutant does not properly attach capsule to its cell surface, but it seems to secrete the polysaccharide similarly to WT (Fig 2A and 2B). Therefore, we wanted to determine if the capsule-deficient phenotype of mar1Δ is contributing to its ability to stimulate macrophages. To test this, we generated a mar1Δ cap59Δ double mutant; Cap59 is involved in capsule biosynthesis, and the cap59Δ mutant does not produce any detectable capsule polysaccharide [14,44]. We used the cap59Δ single mutant as our control strain, and tested macrophage activation of the mar1Δ cap59Δ double mutant compared to cap59Δ alone. We observed that the mar1Δ cap59Δ double mutant induced significantly more TNF-α than the single cap59Δ mutant (S5 Fig). This indicates that the cell surface of mar1Δ has a role in activating macrophages that is independent of this strain’s capsule defect and separate from any cell surface changes that the cap59Δ mutant exhibits as a result of its inability to produce capsular polysaccharide.
We used a murine inhalation model of cryptococcosis to assess the role of Mar1 in virulence [45]. We intranasally inoculated C57BL/6 mice with WT, mar1Δ, or mar1Δ + MAR1 complemented cells and monitored mice over the course of 40 days for clinical endpoints predictive of mortality. Mice infected with the WT or mar1Δ + MAR1 complemented strains exhibited a median survival time of 18 days (Fig 9A). In contrast, mice infected with the mar1Δ mutant had a median survival time of 28 days. We also measured fungal burden in the lungs at early time points after infection. As early as days 1 and 4 after infection, the number of mar1Δ mutant cells were significantly decreased in the lungs of infected mice compared to WT cells (Fig 9B). However, half of the mar1Δ-infected mice eventually succumbed to infection, suggesting there is not complete clearance of these cells.
It has been previously documented that mice from different genetic backgrounds display varying levels of sensitivity to cryptococcal infection [46,47]. C57BL/6 mice predominate towards protective Th1 type responses, while BALB/c mice have a non-protective Th2 type bias [48]. Therefore, we also tested the susceptibility of BALB/c mice to mar1Δ infection. We intranasally inoculated BALB/c mice as above and monitored mice over the course of a 40-day infection. In contrast to C57BL/6 mice, all of the mar1Δ-infected BALB/c mice survived the course of the experiment (Fig 9C). At the end of the experiment, we measured fungal burden in the lungs and brain post-mortem and observed that mar1Δ is not completely cleared from all mice, despite mouse survival (Fig 9D).
The initial innate interaction between fungi and the host begins with pathogen recognition by pattern recognition receptors (PRRs) on innate immune cells. Members of the C-type lectin receptor (CLR) and Toll-like receptor (TLR) families have been implicated in recognizing fungal pathogen-associated molecular patterns (PAMPs), including fungal cell wall components. Many of the CLRs signal through the adaptor protein, Card9, to downstream cellular pathways to activate pro-inflammatory cytokines. Similarly, several of the TLRs signal through the adaptor protein, MyD88, to activate downstream cellular responses.
To determine if members of the CLR or TLR families are responsible for recognizing the mar1Δ cell surface, we examined the roles of the Card9 and MyD88 adaptor proteins in macrophage activation by these strains. We co-cultured our strains with BMMs isolated from Card9-/- and MyD88-/- mice and compared the production of TNF-α with BMMs isolated from WT mice. Consistent with our results above, the mar1Δ mutant induced significantly more TNF-α production from WT BMMs than isogenic WT C. neoformans strains. By contrast, the Card9-/- BMMs had a significant, but incomplete, reduction in their response to the mar1Δ mutant, indicating a role for both Card9-dependent and–independent receptors (Fig 10A). Strikingly, the exaggerated TNF-α response to the mar1Δ strain was completely absent in MyD88-/- BMMs, suggesting a significant role for TLRs in recognizing the mar1Δ cell surface (Fig 10B). Combined, these data demonstrate that both CLR and TLR family receptors are likely involved in sensing and responding to the mar1Δ strain.
Based on the importance of both Card9 and MyD88 in sensing and responding to mar1Δ cells, we next focused on specific PRRs that have been previously implicated in fungal cell wall recognition, in particular chitin and chitooligomers. The C-type lectin receptor, Dectin-1 has been studied extensively for its role in recognizing the fungal cell surface, in particular β-glucan in the cell wall [49]. It has also been implicated in sensing fungal chitin and leading to a pro-inflammatory response, including the production of TNF-α, from innate immune cells [19,20]. To test the role of Dectin-1 in recognizing mar1Δ cells, we co-cultured our strains with Dectin-1-/- BMMs and measured the production of TNF-α after 6 hours (Fig 10C). Compared to WT BMMs, the response to mar1Δ by Dectin-1-/- BMMs was significantly decreased, indicating that this C-type lectin receptor is involved in sensing mar1Δ.
Due to the marked decrease in the mar1Δ response by MyD88-/- BMMs, we used BMMs isolated from TLR2/4-/- mice to focus on the extracellular TLRs that have been previously implicated in fungal PAMP recognition [50–52]. We co-cultured our strains with TLR2/4-/- BMMs and measured the TNF-α response after 6 hours compared to WT BMMs. Similar to what we observed for MyD88-/- BMMs, TLR2/4-/- BMMs showed significantly reduced responses to mar1Δ, indicating a role for TLR2 and/or TLR4 in sensing this fungal cell surface (Fig 10D).
To test the individual role of TLR4, we utilized C3H/HeJ mice that have a null mutation in the TLR4 gene [53]. These mice have been used in several studies to assess the role of TLR4 in the response to a variety of microbial pathogens or pathogen products including bacteria (Mycobacterium [54], Bordetella [55], Rickettsia [56], Neisseria [57]) and fungi (Coccidioides [58], Aspergillus [59]). We isolated BMMs from C3H/HeJ mice and C3H/HeOuJ control mice and assessed their TNF-α responses to mar1Δ. We found a modest increase in the response to mar1Δ between C3H/HeJ TLR4 mutant BMMs and C3H/HeOuJ control mice (Fig 10E). In control experiments, both mutant and control BMMs responded similarly to a control ligand, zymosan; however, as expected, C3H/HeJ BMMs did not respond to the canonical TLR4 ligand, LPS, while C3H/HeOuJ BMMs responded normally (S6 Fig). These results suggest that the response to mar1Δ is TLR4-independent.
In previous work, TLR2 has been described to have a role in recognizing fungal chitin, leading to a pro-inflammatory response and the production of TNF-α [18–20]. To test if TLR2 is responsible for recognizing the mar1Δ surface, we carried out our co-culture assays using TLR2-/- BMMs. The TNF-α response after 6 hours was significantly reduced in the TLR2-/- BMMs compared to WT BMMs (Fig 10F). These data suggest that the decrease in response to mar1Δ in the TLR2/4-/- BMMs was due to the defect in TLR2. Together these results support a model in which the TLR2 and Dectin-1 PRRs are both involved in sensing and responding to the mar1Δ cell surface.
The fungal cell wall is a dynamic structure that is constantly being remodeled. Fungal pathogens carefully regulate their cell wall in the context of the host in order to adapt to this environment, as well as to avoid detection by the immune system. Here we have identified a novel cell wall remodeling protein in C. neoformans, Mar1, and demonstrated that this protein has a role in intracellular trafficking that results in the proper localization of the cell wall β-(1,3)-glucan synthase, Fks1, as well as the continuous reorganization of the fungal cell surface in host-like conditions. The immunological consequence of this cell wall mis-regulation in the mar1Δ mutant is an increased activation of macrophages and dendritic cells as well as attenuated virulence in vivo.
Our lab and others have demonstrated that the cell wall changes in size and in composition in response to the host environment [14–16,60]. Here, in WT cells, we have observed an increase in cell wall thickness in TC medium by TEM. This occurs in the setting of the well characterized formation and expansion of polysaccharide capsule. We have also measured increased expression of many of the known cell wall synthase genes in response to TC medium, and we have demonstrated that Fks1 is enriched at the plasma membrane in a specific population of WT cells after incubation in TC medium. By contrast, we observed a thickened, but less structurally sound cell wall in mar1Δ cells incubated in TC medium. Despite this observed increase in cell wall thickness, we showed that the mar1Δ strain has decreased total levels of glucans and mannans, decreased expression of glucan synthase genes, and decreased Fks1 protein at the plasma membrane in TC-incubated cells. We also demonstrate that mar1Δ cells do not properly attach capsule to their cell surface. Together these data suggest a model in which the mar1Δ strain has decreased total levels of outer cell wall components (glucans and mannans), leading to the observed capsule attachment defect and increased exposure of the inner cell wall components chitin and chitosan (Fig 11).
Our model demonstrates a prominent role for Mar1 in the cell wall response to TC medium, in particular through the regulation of cell wall synthase expression and localization. Previous studies in model fungi have shown differential localization of cell wall synthases in response to different conditions. The S. cerevisiae chitin synthase, Chs3, is maintained under normal conditions in internal compartments called “chitosomes”, where it is cycled between early endosomes and the trans-Golgi network in a clathrin adaptor protein (AP) complex dependent manner [61]. However, under stress conditions, Chs3 is trafficked to the plasma membrane in order to synthesize increased chitin [62]. Similar clathrin AP complex-dependent trafficking has been described for the Schizosaccharomyces pombe β-(1,3)-glucan synthase, Bgs1. Mutations in the conserved AP-1 adaptor complex protein, Sip1, and the clathrin light chain protein, Clc1, lead to defects in Bgs1 localization/delivery to the plasma membrane [63,64].
Here we have observed that Fks1 plasma membrane localization is enriched in specific WT cell populations upon shifting cells to tissue culture medium, suggesting a similar stress-induced cycling of this protein to its functional location at the cell surface. Mar1 could be impacting the cycling of Fks1 between internal compartments and the plasma membrane in multiple ways. For example, Mar1 may be required for exocytic movement towards the plasma membrane, and therefore in the absence of this protein, exocytosis and/or secretion of Fks1 is defective. However, the fact that capsule, melanin, and acid phosphatase secretion remain intact in this strain would suggest that Mar1 is not impacting generalized exocytosis or secretion. Another way in which Mar1 could be impacting Fks1 trafficking is if it were required for protein maintenance on the PM, for example, by preventing excessive endocytosis or recycling. Our data showing altered endocytosis of FM4-64 in the mar1Δ strain would support this model.
Another alternative mechanism by which Mar1 might impact Fks1 localization and ultimately cell wall remodeling could be through a role in cell surface signaling. The Mar1 protein has two transmembrane domains and our microscopy data indicates that it localizes to cellular membranes at the cell surface. Therefore, it is feasible to hypothesize that Mar1 is serving a signaling role, perhaps as a cell surface receptor, propagating the cue to traffic cell wall synthase proteins from internal stores to the PM under stress conditions. This model for Mar1 function, unlike the previous, would also explain the lack of transcriptional response of the FKS1 and AGS1 genes in the mar1Δ background. If the mutant cells are unable to sense specific stress conditions, it is plausible that they would not induce transcription of the enzymes required to respond to that stress.
The differential localization of Mar1 in YPD and TC media is also consistent with Mar1 serving a sensor/receptor role. We observed many Mar1 puncta, particularly on the cell surface, but also on intracellular membranes resembling the endoplasmic reticulum when the cells were incubated in YPD medium. By contrast, when shifted to TC medium, we observed a decrease in the number of puncta, as well as changes in their size/shape. It is possible that this decrease in puncta represents the endocytosis of Mar1 from the membrane to transmit a signal downstream, or perhaps internalization for degradation. Interestingly, we observed these changes in Mar1-localization at similar time points as increases in cell wall chitooligomer staining in the mar1Δ mutant; these observations suggest a correlation between Mar1 localization and the resulting cell wall response.
A role for Mar1 as a sensor implies a particularly intriguing model, as few cell surface stress sensors have been identified by sequence homology in C. neoformans. Interestingly, the putative Rra1 pH sensor was recently identified in C. neoformans; while appearing structurally and functionally similar to that of the pH sensors in ascomycete fungi, it was found to share no significant sequence similarity [35]. While Mar1 does not share sequence similarity with any known sensor proteins, its structural features and pattern of dynamic localization suggest that it could be serving as a functional homologue of a cell stress sensor.
The importance of cell wall remodeling, in particular as it relates to cell wall masking, has been well-described in other fungi. The α-glucan layers of H. capsulatum and A. fumigatus have both been implicated in shielding β-glucan in these fungi. β-glucanases that degrade exposed β-glucan have also been elucidated in H. capsulatum [65]. In C. albicans, several factors have been implicated in β-glucan exposure including morphotype switching, antifungal exposure, phospholipid production, and carbon source [5,6,66]. For example, Ballou, et al. recently demonstrated that lactate exposure elicits increased β-glucan masking in C. albicans and other pathogenic Candida species, leading to reduced recognition by immune cells [6]. Furthermore, cell wall remodeling has been shown to drive immune function even after ingestion by host phagocytes. O’Meara, et al. showed that C. albicans actively remodels its cell surface upon phagocytosis in order to induce inflammatory cell death pathways in macrophages to aide in fungal cell escape [15].
C. neoformans is particularly adept at hiding from the immune system. Importantly, the polysaccharide capsule serves to effectively shield potentially immunostimulatory molecules from host detection [29,67]. However, there are several examples of the immune consequences of improper cell wall organization in C. neoformans. The loss of α-(1,3)-glucan results in a highly disorganized cell wall, with increased chitin and chitosan content and redistributed β-glucan, rendering cells avirulent in mouse models of infection [68]. Chitosan-deficient strains of C. neoformans are also unable to cause disease in mice, inducing a protective proinflammatory host immune response that leads to their rapid clearance [69–71]. Imprecise regulation of the cell wall has also been implicated in excessive immune responses related to improper C. neoformans cell wall exposure, in particular chitooligomers (chitin and/or chitosan). Our laboratory has previously studied the importance of the Rim101 alkaline pH transcription factor in regulating cell wall remodeling [14–16]. The rim101Δ mutant exposes increased cell wall chitooligomers, which results in an excessive and nonprotective immune response in vivo [16]. Likewise, Wiesner, et al. demonstrated that strains with increased chitin abundance induced unfavorable immune outcomes and exacerbated disease [12].
Here, we propose that both the exposure and total levels of cell wall components are important determinants of immune recognition of fungi. Based on our HPLC data, our model suggests that mar1Δ has less β-glucan in its cell wall (Fig 11). Therefore, one might predict that immune signaling would be less active with reduced immunostimulatory β-(1,3)-glucan to sense. However, the mar1Δ cell wall has more exposed chitooligomers than WT. As noted above, several studies have highlighted the importance of chitin and chitin-derived structures in C. neoformans immune recognition [12,16]. Additionally, cells in which chitooligomer exposure was blocked by WGA exhibited reduced association with murine macrophages [72]. While a single chitin receptor has not been identified to date, several PRRs have been implicated in different aspects of chitin recognition. Here we have demonstrated that multiple PRRs are also involved in the recognition of the mar1Δ cell wall. Macrophage activation by mar1Δ was partially dependent upon Card9, the adaptor protein required for most C-type lectin receptor signaling, and entirely dependent on MyD88, the adaptor protein required for signaling through many Toll-like receptors (TLRs). Accordingly, we demonstrated a requirement for Dectin-1 and TLR2 in the activation of macrophages in response to the mar1Δ mutant strain. There are several possibilities as to why more than one PRR might be involved in this process. In addition to their roles individually, PRRs can function together to induce downstream immune signaling pathways. In particular, several groups have demonstrated a collaborative role between Dectin-1 and TLR2 in detecting fungal epitopes [73–75].
Wagener and colleagues proposed a model in which chitin recognition occurs in different stages during C. albicans interaction with immune cells [20]. During early interaction, chitin on the surface of this fungus is recognized by Dectin-1 and TLR2, resulting in the secretion of proinflammatory TNF-α. Over time this leads to the secretion of chitinases and a decrease in pathogen load. Later in this interaction, digested fragments of chitin are phagocytosed by immune cells and recognized by intracellular TLR9 and NOD2 in a mannose receptor (MR)-dependent manner, resulting in the secretion of IL-10. This latter state occurs during the resolution of infection. In agreement with this model, other groups have demonstrated that intermediate-sized chitin stimulated macrophage production of TNF-α that is dependent on TLR2, Dectin-1, and MR. In contrast, small chitin fragments induced IL-10 production [19]. Given the importance of chitooligomer size and structure in the immunostimulatory process, future work on the mechanism of chitin and chitosan synthesis and degradation will elucidate its role in fungal immune recognition and evasion. Our data demonstrating that intact mar1Δ cells induce increased TNF-α from macrophages in a TLR-2 and Dectin-1 dependent manner support these emerging models of the centrality of chitooligomer exposure in fungal stimulation of the host immune system.
It is still possible, that other factors are contributing to the mar1Δ macrophage activation phenotype. While we demonstrated decreased overall levels of glucans and mannans in mar1Δ cell wall extracts, this does not necessarily exclude these components from playing a role. We did not observe a significant signal from Fc-Dectin-1 labeled cells above baseline, however we did see a very modest increase in mar1Δ cells stained with the α-glucan antibody, MOPC-104E, and the mannoprotein-binding lectin, Concanavalin A (Fig 4A). While α-glucans generally serve to mask other cell wall components, Cryptococcal mannoproteins are well-known to illicit immune responses [76]. We have also shown that mar1Δ cells lack a polysaccharide capsule, and further demonstrated that this is due to a capsule attachment defect, rather than a biosynthesis or secretion defect. Polysaccharide capsule serves to shield the cell surface from immune detection, and free capsule has been shown to have immunomodulatory effects [9,77]. Using a mar1Δ cap59Δ double mutant, in which capsule biosynthesis is inhibited, we showed that the mar1Δ cell wall has an impact on macrophage activation that is independent of the cap59Δ single mutant strain, suggesting this response is not due simply to the lack of capsule in the mar1Δ background. We also determined that heat killed mar1Δ cells, in which capsule polysaccharide is not being actively shed, are still immunostimulatory.
Based on our data, we suggest multiple ways in which this capsule defect may arise. First, it has been previously shown that capsule attachment to the cell wall requires Ags1 expression and α-glucan [13]. We demonstrated that the mar1Δ mutant has decreased glucans in its cell wall and severely decreased Ags1 expression. We also observed that while mar1Δ sheds a similar amount of capsule polysaccharide into the media as WT cells, it migrated at a different rate. This could suggest that the major capsule polysaccharide component, GXM, is modified in some way as to inhibit its ability to attach to the cell wall.
Additional yet unidentified PRRs may also be involved in this interaction. In addition to further interrogating the role of mannoproteins in the mar1Δ response, future work will examine the importance of the mannose receptor in this interaction. The role of TLR9 and NOD2 in this interaction is also intriguing. We did not explore these intracellular receptors in this study, however TLR9 and NOD2 have been previously implicated in recognizing fungal chitin that has been digested and phagocytosed by mammalian macrophages [20]. This recognition leads to the production of the anti-inflammatory cytokine, IL-10, and is thought to lead to the resolution of the pro-inflammatory immune response to intact fungal chitin [20]. In earlier studies, it was observed that, in fact, TLR9-/- macrophages co-incubated with C. albicans and S. cerevisiae had increased TNF-α production, suggesting a role for TLR9 in modulating the pro-inflammatory response to these fungi [78]. TLR9 deficiency has also been associated with worsened immune outcomes in C. neoformans [79,80].
In summary, we have identified a novel cell wall regulatory protein, Mar1. This protein, while not apparent in any canonical cell wall regulatory pathway, is required for proper cell wall organization in host-like conditions. The Mar1 protein localizes to discrete puncta on cellular membranes, with an apparent reduction in puncta upon shifting cells to host-like conditions. In the absence of Mar1, transcription of the α- and β-glucan synthases is not induced, leading to a decrease in these cell wall components and an increase in the exposure of the chitooligomers, chitin and chitosan. We propose that the role of Mar1 in cell wall integrity is in orchestrating the proper response to host-like conditions, occurring in part at the level of intracellular trafficking and results in the mis-localization of the β-(1,3)-glucan synthase, Fks1. Here we have also shown that this dysregulated cell wall manipulates the host-pathogen interaction, leading to increased macrophage activation that is dependent on multiple pattern recognition receptors. Finally, Mar1 is required for full virulence in two mouse models of systemic cryptococcosis.
Cryptococcus neoformans strains used in this study are listed in Table 1. Unless otherwise noted, all strains were generated in the C. neoformans var. grubii H99 background and maintained on YPD agar plates (2% yeast extract, 1% peptone, 2% dextrose). Strains created by crosses were co-incubated on MS mating media [81], followed by spore isolation by microdissection. Recombinant spores were identified by PCR and selectable marker resistance. To assess cell wall associated phenotypes, NaCl (1.5 M) and Congo red (0.5%) were added to YPD medium prior to autoclaving; caffeine (1 mg/ml) and calcofluor white (1 mg/ml) were filter sterilized and added to YPD after autoclaving. Alkaline pH plates were made by adding 150 mM HEPES buffer to YPD and adjusting the pH (pH 8.15) with NaOH prior to autoclaving. To induce and visualize capsule, strains were incubated in CO2-independent tissue culture medium (TC, Gibco) for 72 hours with shaking at 37°C, followed by staining with India Ink.
For cell wall staining, cell wall isolation, protein localization microscopy, FM4-64 staining, and in vitro co-culture experiments, overnight YPD cultures were diluted 1:10 in YPD liquid medium (at 30°C) or in TC medium (at 37°C) for 16–18 hours with shaking (150 rpm), unless otherwise noted. These methods were described previously [16].
All primers used in this study are listed in Table 2. All plasmids used in this study are listed in Table 3. The mar1ΔT-DNA strain was generated by Agrobacterium tumefaciens-mediated transformation (AMT) as described previously [22]. C. neoformans targeted gene deletion strains were generated by homologous recombination to replace the entire open reading frame (ORF) with a dominant selectable marker. The deletion cassettes were created using overlap PCR as described previously [86,87] using the primers indicated in Table 2. Deletion cassettes were introduced into the indicated background strain by biolistic transformation [88]. All deletion strains were confirmed by PCR and Southern blot; primers used to design probes for Southern blot can be found in Table 2.
The mar1Δ deletion strain (MAK1) was constructed by replacing the MAR1 ORF with the nourseothricin (NAT) cassette [90]. The mar1Δ + MAR1 complemented strain (MAK11) was constructed by co-transformation of the WT MAR1 allele with the pJAF neomycin (NEO) resistance vector into the MAK1 background.
The GFP-Mar1 strain (SKE106) was constructed by transforming pSKE26 into the mar1Δ (MAK1) background. The pSKE26 plasmid contains an N-terminally GFP-tagged Mar1 protein under the control of the histone H3 promoter. A fragment consisting of the MAR1 open reading frame and ~ 500 bp of the 3’ UTR/terminator sequence was amplified from H99 genomic DNA using primer pair AA4894/AA4895. Using InFusion cloning (Clontech), this fragment was cloned in frame into the pCN50 backbone at a BamHI site at the end of GFP. Transformants were screened by wet colony morphology on pH 8 and GFP-Mar1 fusion was confirmed by PCR and western blot.
The mar1Δ + eGFP-RIM101 (MAK8) strain was constructed by crossing MAK1 with KS208. Recombinant spores were screened by epifluorescent microscopy and confirmed by PCR.
The MPK1-4FLAG-NEO strains (SKE94 and SKE96) were generated by transforming pSKE19 into the WT or mar1Δ (MAK1) background. The MPK1-4FLAG-NEO tagging construct was designed such that a C-terminal 4xFLAG epitope tag would homologously recombine into the 3’ end of the MPK1 locus. The pSKE19 plasmid was generated by In-Fusion cloning (Clontech) the following fragments into the pUC19 backbone: (1) ~500 bp of the 3’ end of MPK1 ORF, amplified from H99 genomic DNA using primer pair AA4829/AA4830; (2) 4xFLAG linked to the HOG1 terminator, amplified from pSG27 [91] using primer pair AA4831/AA4832; (3) NEO resistance cassette amplified from pJAF using primer pair AA4264/AA4598; (4) ~ 1 kb 3’ MPK1 flank amplified from H99 genomic DNA using primer pair AA4599/AA4833. Transformants were screened by PCR and confirmed by Southern blot.
The mar1Δ mpk1Δ (SKE87) strain was constructed by replacing the MAR1 ORF with the NEO cassette in the mpk1Δ deletion background (KK3). Transformants were screened by dry colony morphology on pH 8 and PCR.
The FKS1-GFP-NEO strain (KMP13) was constructed by transforming pKP6 into the WT background. The FKS1-GFP-NEO tagging construct was designed to facilitate homologous recombination at the FKS1 locus. The pKP6 plasmid was created by In-Fusion cloning the following fragments into the pUC19 backbone: (1) ~ 1 kb of the 3’ end of the FKS1 ORF, amplified from H99 genomic DNA using primer pair AA4545/AA4362; (2) GFP amplified from pCN19 (Price 2008) using primer pair AA4364/AA4400; (3) FKS1 terminator (464 bp) amplified from H99 genomic DNA using primer pair AA4553/AA4426; (4) NEO resistance cassette amplified from pJAF using primer pair AA4264/AA1668; (5) 1 kb 3’ FKS1 flank amplified from H99 genomic DNA using primer pair AA4433/AA4546. Transformants were screened by PCR and epifluorescent microscopy and integration into the locus was confirmed by PCR. The mar1Δ FKS1-GFP-NEO strains (CLT1 and CLT2) were generated by replacing the MAR1 ORF with the NAT cassette in the KMP13 background. Transformants were screened by dry colony morphology on pH 8 and confirmed by PCR and Southern blot.
The mar1Δ cap59Δ (SKE60) strain was created by crossing MAK1 with CBN377. Recombinant spores were screened and confirmed by PCR.
The relative amount of capsule shedding in the cell supernatant was assayed as previously described [21,30]. Briefly, capsule induced cultures (incubated as described above) were incubated at 70°C for 15 minutes to denature enzymes, after which the cells were pelleted and the supernatant was sterile filtered. This conditioned medium was then run on a 0.6% agarose gel for 15 hours at 25 V, followed by transfer to a positively charged nylon membrane using Southern blotting methods. The membrane was air dried overnight, followed by blocking with 5% skim milk in Tris-Buffered Saline-Tween-20 (TBST). To detect capsule polysaccharide, blots were incubated with a mouse monoclonal anti-GXM antibody, MAb18B7 (1 μg/ml) [92,93] for 1 hour, washed 3x with TBST, and incubated with an anti-mouse horseradish peroxidase-conjugated secondary antibody (Jackson ImmunoResearch) for 1 hour. Blots were washed 3x with TBST and capsule polysaccharide was detected by enhanced chemiluminescence (ECL Prime Western blotting detection reagent; GE Healthcare).
Prior to all cell wall staining, cells were pelleted and washed 1-2x with phosphate buffered saline (PBS). For quantification by microscopy, stained cells were imaged on a Zeiss Axio Imager.A1 fluorescence microscope equipped with an AxioCam MRm digital camera (60X objective). The same exposure time was used to image all strains with the same stains. The mean gray value (sum of gray values for all the pixels in a cell divided by the number of pixels that make up the cell) of at least 100 cells was calculated using ImageJ/Fiji [94,95]. Results are reported as mean fluorescence values +/- standard error of the means.
For flow cytometry, cells were fixed with 3.7% formaldehyde for 5 minutes at room temperature, followed by washing 2x with PBS. Eosin Y stained samples were fixed with 10 mM sodium azide for 10 minutes at room temperature, followed by washing 2x with PBS. A total of 107 cells/ml were stained and 106 cells/ml were submitted to the Duke Cancer Institute Flow Cytometry Shared Resource for analysis using a BD FACSCanto II flow cytometer. Data was analyzed using FlowJo v10.1 software (FlowJo, LLC). Relevant events were gated in the forward scatter/side scatter (FSC/SSC) plots and then represented as histograms with mean fluorescence intensity (MFI) on the x-axis and cell counts on the y-axis. Unstained cells and cells incubated with secondary antibodies alone were used as negative controls.
To visualize chitin, cells were stained with 100 μg/ml FITC-conjugated wheat germ agglutinin (WGA; Molecular Probes) for 35 minutes in the dark, followed by 25 μg/ml calcofluor white (CFW) for 10 minutes. Prior to analysis, cells were washed 2x and resuspended in PBS. For microscopy, WGA was imaged using a GFP filter and CFW was imaged using a DAPI filter. For flow cytometry, WGA cells were analyzed using a 488 nm laser and CFW cells were analyzed using a 405 nm laser.
To visualize chitosan, sodium azide fixed cells were washed 2x with McIlvaine’s buffer (0.2 M Na2HPO4, 0.1 M citric acid, pH 6.0), followed by staining with 300 μg/ml Eosin Y (EY) in McIlvaine’s buffer for 5 minutes at room temperature. Prior to analysis, cells were then washed 2x and resuspended in McIlvaine’s buffer. For microscopy cells were visualized using a GFP filter. For flow cytometry cells were analyzed using a 488 nm laser.
The MOPC-104E antibody (Sigma) was used to visualize α-glucan, as described previously [3,14]. Briefly cells were incubated with 400 ng/ml MOPC-104E primary antibody for 1 hour, washed 2x with PBS, and incubated with 4 μg/ml anti-mouse AlexaFluor 488 secondary antibody (Jackson ImmunoResearch) for 30 minutes in the dark. Cells were washed 2x and resuspended in PBS prior to analysis. For flow cytometry cells were analyzed using a 488 nm laser.
An Fc-Dectin-1 fusion protein was used to visualize β-glucan (a gift from Gordon Brown, University of Aberdeen) [96–99]. Cells were resuspended in FACS block (0.5% BSA, 5% HI-rabbit serum, 5 mM EDTA, 2 mM NaAzide in PBS) for 10 minutes, followed by incubation with 5 μg/ml Fc-Dectin-1 protein for 40 minutes on ice. Cells were washed 3x with PBS and resuspended in 3.75 μg/ml anti-human AlexaFluor 488 secondary antibody (Jackson ImmunoResearch) in FACS wash (0.5% BSA, 5 mM EDTA, 2 mM NaAzide in PBS) for 30 minutes on ice. Cells were washed 2x and resuspended in PBS prior to analysis. For flow cytometry cells were analyzed using a 488 nm laser.
Concanavalin A conjugated to AlexFluor 488 (ConA; Molecular Probes) was used to visualize mannoproteins. Cells were resuspended in 50 μg/ml ConA for 1 hour, then washed 2x and resuspended in PBS prior to analysis. For flow cytometry cells were analyzed using a 488 nm laser.
Differential interference microscopy (DIC) and fluorescent images were visualized with a Zeiss Axio Imager.A1 fluorescence microscope (60X or 100X objectives). Images were taken with an AxioCam MRm digital camera with ZEN Pro software (Zeiss). High-resolution fluorescent images were taken using a DeltaVision Elite deconvolution microscope equipped with a CoolSnap HQ2 high-resolution charge-coupled-device (CCD) camera. Images were processed using softWoRx software (GE). Images taken on both microscopes were additionally analyzed using ImageJ/Fiji software [94,95].
Chitin and chitosan levels were quantified using a modified MBTH (3-methyl-benzothiazolinone hydrazine hydrochloride) method as previously described [16]. Cell wall isolation and high performance anion-exchange chromatography with pulsed amperometric detection (HPAEC-PAD) were performed as previously described [16,33].
Cells from an overnight YPD culture were washed 1x with water, diluted to 107 cells/ml in YPD or TC medium in duplicate and incubated for 1.5 hours at 30°C (YPD) or 37°C (TC). Cultures were spun down and flash frozen on dry ice, followed by lyophilization. RNA was extracted using the RNeasy Plant Mini Kit (Qiagen) with optional on-column DNAse digestion. cDNA for real time-PCR (RT-PCR) was prepared using the AffinityScript cDNA synthesis kit (Agilent) with oligo(dT) primers. For RT-PCR, cDNA was diluted 1:3 in RNase-free water, added to IQ SYBR Green Supermix (Bio-Rad) per protocol instructions, and analyzed on an iCycler iQ Real-Time PCR Detection System (Bio-Rad). GPD1 was used as an internal control, and negative control samples without reverse transcriptase were included. All RT-PCR primers are listed in Table 2.
To assess Rim101 processing, overnight cultures were diluted to an optical density of 1 in 25 ml YPD pH 4 and pH 8. Cells were incubated for 1 hour, washed 1x with water, flash frozen on dry ice, and stored at– 80°C until cell harvesting.
For cell wall integrity pathway activation analysis, overnight YPD cultures were diluted to an optical density of 0.8 in 25 ml YPD in duplicate and incubated for 2.5 hours at 30°C. Cultures were then spun down and duplicates were resuspended in YPD or TC medium and incubated at 30°C (YPD) or 37°C (TC). After 3.5 hours, samples were taken, washed 1x with water, flash frozen on dry ice, and stored at -80°C until cell harvesting.
For protein extraction, cells were lysed by bead beating and the lysate was collected in 1.4 ml NP40 lysis buffer (6 mM Na2HPO4, 4 mM NaH2PO4, 1% Nonidet P-40, 150 mM NaCl, 2 mM EDTA, 1x protease inhibitors [Complete mini, EDTA-free; Roche], 1x phosphatase inhibitors [Phos-Stop; Roche], and 1 mM phenylmethylsulfonyl fluoride [PMSF]), as described previously [100]. For Western blot analysis, samples were normalized by BCA assay (Thermo Scientific), diluted in 4x lithium dodecyl sulfate (LDS) loading sample buffer, and boiled for 5 minutes. Normalized protein was loaded on a NuPage 4–12% Bis-Tris gel (Invitrogen) and western blots were performed as previously described [100]. To detect Gfp-Rim101, blots were incubated with an anti-GFP primary antibody (1/5,000 dilution; Roche) and an anti-mouse peroxidase-conjugated secondary antibody (1/25,000 dilution; Jackson Labs). To detect phosphorylated Mpk1, blots were incubated with a phospho-p44/42 MAPK (Thr202/Tyr204) rabbit polyclonal primary antibody (1/2,500 dilution; 4370 Cell Signaling Technology) and an anti-rabbit peroxidase-conjugated secondary antibody (1/50,000 dilution; Jackson Labs). Proteins were detected by enhanced chemiluminescence (ECL Prime Western blotting detection reagent; GE Healthcare).
Cells from an overnight YPD (30°C) or TC (37°C) culture were normalized and stained with the lipophilic dye, FM4-64 (1/1000 dilution in the indicated medium, Molecular Probes) for 30 minutes shaking, after which the cells were pelleted and YPD or TC medium was refreshed for an additional 30 minutes at with shaking. For microscopy, cells were pelleted, washed 2x with PBS, and resuspended in PBS. FM4-64 was visualized using a Texas Red filter. The same exposure time was used for all images.
Cells were incubated overnight in phosphate replete minimal medium (15 mM dextrose, 10 mM MgSO4, 13 mM glycine, 3 μM thymine, 0.4% KH2PO4). Cultures were then diluted to an OD of 0.9 in phosphate replete or phosphate deficient minimal medium (15 mM dextrose, 10 mM MgSO4, 13 mM glycine, 3 μM thymine, 0.4% KCl) and washed 1x in respective minimal media. 100 μl aliquots were plated in triplicate in a 96-well plate and incubated for 3 hours at 30°C with shaking at 150 rpm. Para-Nitrophenylphoshate (pNPP) solutions were prepared by dissolving a 5 mg pNPP substrate tablet (Thermo) in 50 mM sodium acetate (pH 5.2) for acid phosphatase testing or 1x diethanolamine substrate buffer (Thermo) for alkaline phosphatase testing. 100 μl of pNPP solution was added to each well and plates were incubated for 3 hours at 37°C with shaking at 150 rpm. Phosphatase activity was measured at an absorbance of 410 nm and adjusted for cell density, as determined by absorbance at 600 nm, over a time course of 3 hours.
Overnight YPD and TC cultures were diluted to an OD of 0.5 and recovered in the same media for 4–6 hours. Cells were prepared as described previously [14,68,101]. Briefly, cells were spun down and washed 1x in pre-fixation mix (0.1 M sorbitol, 1 mM MgCl2, 1 mM CaCl2, 2% gluteraldehyde in 0.1 M PIPES, pH 6.8), followed by fixing in fresh pre-fixation mix overnight at 4°C. The next day, cells were washed 3x for 10-minute intervals in water. Next the cells were washed 3x in 2% KMnO4, and post-fixed for 45 minutes at room temperature in fresh 2% KMnO4. The cells were then washed repeatedly in water until no purple color was visible, and partially dehydrated for 10-minute intervals in increasing concentrations of ethanol (30%, 50%, 70%). Partially dehydrated samples were submitted to the Duke Shared Materials Instrumentation Facility (SMIF) for further processing, embedding, and sectioning as follows: Samples were rinsed thoroughly in PBS and post-fixed in 1% osmium tetroxide for 1 hour at room temperature. Samples were then stained with 0.5% uranyl acetate for 1 hour, further dehydrated in a series of graded ethanol (30%, 50%, 70%, 90%, 100%) and infiltrated overnight in resin. Samples were then embedded in resin and cured in a 55°C oven for 48 hours. The cured samples were thin sectioned with an ultramicrotome to approximately 60–90 nm. Thin-sections were mounted on copper grids and stained with uranyl acetate and lead citrate to enhance contrast. Grids were examined and digital images were taken on the FEI Tecnai G2 Twin transmission electron microscope with an Eagle digital camera.
Bone marrow cells derived from female C57BL/6 mice purchased from Jackson Laboratories were used as WT controls for all experiments, unless otherwise noted. MyD88-/- and TLR2/4-/- bone marrow cells were a generous gift from Marcel Wüthrich at the University of Wisconsin-Madison. Bone marrow cells from Card9-/- mice were provided by Floyd Wormley. Dectin1-/- mice were a generous gift from Mari Shinohara at Duke University. C3H/HeOuJ TLR4 mutant mice (Stock no. 000659), C3H/HeJ control mice (Stock no. 000635), and TLR2-/- mice (Stock no. 004650, [102]) were purchased from Jackson Laboratories.
Murine bone marrow cells were isolated and prepared as previously described [16,103,104]. Briefly, femurs and tibias were isolated from mice and each bone was flushed with 5 to 10 ml cold PBS using a 27½ gauge needle. Red blood cells were lysed in 1x RBC lysis buffer (0.15 M NH4Cl, 1 mM NaHCO3, pH 7.4) and cells were resuspended in 1x Dulbecco’s modified Eagle’s medium (DMEM; + 4.5 g/L D-Glucose, + L-Glutamine, +110 mg/L sodium pyruvate) with 1 U/ml pencillin/streptomycin. Bone marrow cells were cryopreserved in 90% FBS/10% endotoxin-free DMSO at a concentration of 1 x 107 cells/ml and later thawed for use as previously described [103].
Fresh or frozen bone marrow cells were used to generate bone marrow derived macrophages (BMMs) or bone marrow derived dendritic cells (BMDCs). BMMs were differentiated in BMM medium (1x Dulbecco’s modified Eagle’s medium [DMEM; + 4.5 g/L D-Glucose, + L-Glutamine, +110 mg/L sodium pyruvate], 10% fetal bovine serum [FBS; non-heat inactivated], 1 U/ml penicillin/streptomycin) with 3 ng/ml recombinant mouse GM-CSF (rGM-CSF; R&D Systems or BioLegend) at a concentration of 2.5 x 105 cells/ml in 150 x 15 mm petri plates at 37°C with 5% CO2. The media was refreshed after 3–4 days and the cells were harvested on day 7 as previously described [103]. BMMs were counted (by hemocytometer, with Trypan blue to discount dead cells), plated in BMM medium in 96-well plates at a concentration of 5 x 104 cells/well, and incubated at 37°C with 5% CO2 overnight prior to fungal co-culture experiments.
BMDCs were differentiated in BMDC medium (1x RPMI, 10% FBS [non-heat inactivated], 1 U/ml penicillin/streptomycin, 1x beta-mercaptoethanol) with 20 ng/ml rGM-CSF at a concentration of 5 x 106 cells/ml in 20 ml in 150 x 15 mm petri plates at 37°C with 5% CO2. After 3 days an additional 20 ml of BMDC medium with 20 ng/ml rGM-CSF was added to plates. After 6 days, 20 ml of culture supernatant was collected, centrifuged, resuspended in fresh BMDC medium with rGM-CSF and returned to the culture plate. BMDCs were harvested on day 10 as described previously [105] and BMDCs were counted, plated in BMDC medium in 96-well plates at a concentration of 5 x 104 cells/well, and incubated at 37°C with 5% CO2 overnight prior to fungal co-culture experiments.
BMM and BMDC co-cultures with C. neoformans were performed as described previously [16]. Briefly, C. neoformans cells were washed 2x with PBS, counted, and added to BMM or BMDC containing 96-well plates at a concentration of 5 x 105 fungal cells per well (10:1 C. neoformans cells:BMMs/BMDCs). Isolated cell wall material was added at a concentration of 10 mg/ml. Co-cultures were incubated for the indicated amount of time at 37°C with 5% CO2. Supernatants were collected and stored at -80°C until analysis. Secreted cytokines (TNF-α) were quantified in supernatants by enzyme-linked immunosorbent assay (ELISA; BioLegend). Data are represented as the average TNF-α values (pg/ml) for biological replicates; each fungal strain was tested a minimum of 3 times. BMM/BMDC only control wells, in which fresh media was added in lieu of fungi are included as negative controls. Ultrapure lipopolysaccharide (List Biolabs) and zymosan from S. cerevisiae (Sigma) were diluted to the indicated concentrations in BMM medium and used as positive controls.
As described previously [16], the cap59Δ mutation causes cell aggregation that makes quantification by hemocytometer inaccurate. As a result, these strains were normalized to 2 mg wet cell pellet/ml of medium, which was used previously for other mutants with similar mass/cell ratios and approximates the milligram-per-milliliter concentration used for standard co-culture experiments [16].
We used the murine inhalation model of Cryptococcosis to assess virulence [45]. For each strain, 9–10 female C57BL/6 mice and 9–10 male and female BALB/c mice were used. Mice were anesthetized by isoflurane inhalation and intranasally inoculated with 1 x 105 fungal cells of the following strains: WT (H99), mar1Δ (MAK1), and mar1Δ + MAR1 (MAK11). Mice were monitored over the course of 40 days and sacrificed based on clinical endpoints that predict mortality. The statistical significance of difference between survival curves of mice infected with different strains was determined by log-rank test with Bonferroni correction (GraphPad Prism).
An additional 9–10 mice per strain were intranasally inoculated as described above for organ burden. Mice (5 per time point) were sacrificed on days 1 and 4 post inoculation. From each mouse 1 lung was harvested, weighed, and homogenized in cold PBS. Colony forming units (CFU) were calculated by quantitative culture and are represented as CFU/gram of tissue. For BALB/c post-mortem CFU analysis, lungs and brains were harvested from mice and homogenized in 1 ml PBS. Viable cells were calculated by quantitative culture and are represented as CFU/ml.
All animal experiments were performed in compliance with guidelines at Duke University, the University of Texas at San Antonio, and the American Veterinary Medical Association. All mice were anesthetized by isoflurane inhalation. Mice were sacrificed by CO2 with approved secondary methods of ensuring animal death. The Duke University Institutional Animal Care and Use Committee reviewed and approved the protocol (A138-17-06) used for animal experimentation in these studies. The specific projects were reviewed for congruence with this protocol, and approval was granted on 6/29/2015. Duke University maintains an animal program that is registered with the United States Department of Agriculture (Animal Welfare Act, Customer Number: 863), assured through the National Institutes of Health/Public Health Service (Assurance Number D16-00123 (A3195-01)), and accredited with AAALAC International (Accreditation Number: 363).
|
10.1371/journal.pgen.1003800 | Mitochondrial Transcription Terminator Family Members mTTF and mTerf5 Have Opposing Roles in Coordination of mtDNA Synthesis | All genomes require a system for avoidance or handling of collisions between the machineries of DNA replication and transcription. We have investigated the roles in this process of the mTERF (mitochondrial transcription termination factor) family members mTTF and mTerf5 in Drosophila melanogaster. The two mTTF binding sites in Drosophila mtDNA, which also bind mTerf5, were found to coincide with major sites of replication pausing. RNAi-mediated knockdown of either factor resulted in mtDNA depletion and developmental arrest. mTTF knockdown decreased site-specific replication pausing, but led to an increase in replication stalling and fork regression in broad zones around each mTTF binding site. Lagging-strand DNA synthesis was impaired, with extended RNA/DNA hybrid segments seen in replication intermediates. This was accompanied by the accumulation of recombination intermediates and nicked/broken mtDNA species. Conversely, mTerf5 knockdown led to enhanced replication pausing at mTTF binding sites, a decrease in fragile replication intermediates containing single-stranded segments, and the disappearance of species containing segments of RNA/DNA hybrid. These findings indicate an essential and previously undescribed role for proteins of the mTERF family in the integration of transcription and DNA replication, preventing unregulated collisions and facilitating productive interactions between the two machineries that are inferred to be essential for completion of lagging-strand DNA synthesis.
| All genomes require a system for preventing collisions between the machineries of DNA replication and transcription. We have investigated the roles in this process of two proteins of the mTERF (mitochondrial transcription termination factor) family in Drosophila. These factors, mTTF and mTerf5, share common binding sites in the mitochondrial genome, which we found to coincide with sites of replication pausing. Knockdown of either factor by RNA interference resulted in mtDNA depletion and developmental arrest. mTTF knockdown decreased site-specific replication pausing, but led to an increase in random stalling and regression of replication forks, with impaired synthesis of the lagging strand. This we attribute to random collisions with the transcriptional machinery. Conversely, mTerf5 knockdown led to enhanced replication pausing at mTTF binding sites. These findings indicate an essential and previously undescribed role for proteins of the mTERF family in the integration of transcription and DNA replication, preventing unregulated collisions and facilitating productive interactions between the two machineries that are inferred to be essential for completion of lagging-strand DNA synthesis.
| The mitochondrial genome and its expression are essential to maintain oxidative phosphorylation (OXPHOS), a central metabolic process in higher eukaryotes. OXPHOS failure during development leads to developmental arrest in a diverse range of metazoans, including both insects [1], [2] and vertebrates. In the mouse, for instance, ablation of genes required for mitochondrial DNA (mtDNA) maintenance results in lethality at embryonic day 8–9 [3]–[5]. OXPHOS dysfunction also underlies many pathological states in humans [6]. Elucidation of the mechanisms of faithful mitochondrial genome maintenance and expression is therefore of both developmental and medical relevance [6].
In metazoans, mtDNA replication has been most extensively studied in mammals, where several competing models have been proposed. The strand-displacement model [7], originally based on imaging and end-mapping studies (see also [8]–[10]), contrasts with the evidence from two-dimensional neutral agarose gel electrophoresis (2DNAGE) analyses [11]–[16], supporting various types of strand-coupled replication. In the strand-displacement model, leading-strand synthesis initiates in the major non-coding region (NCR), at a site designated as the origin of heavy-strand synthesis (OH) [12], [13]. It then proceeds two-thirds of the way around the circle until reaching the site designated as the origin of light-strand synthesis (OL). Synthesis of the two strands on this model is asynchronous, but continuous on both strands, i.e. without a need for Okazaki fragments.
2DNAGE was developed almost three decades ago, to separate and characterize branched from linear DNA [17]. It has been widely used to analyze replication intermediates, starting in 1987 with the yeast 2 µ plasmid [18], and subsequently in hundreds of other publications. The method is considered definitive for inferring replication mode and origins, termination sites, fork barriers and molecular recombination (for review see [19]–[23]). 2DNAGE has indicated the existence of two classes of strand-coupled replication intermediate in mammalian mtDNA, which have been suggested to reflect alternate modes of replication that may operate in parallel. In the unidirectional RITOLS mode (RNA Incorporation Throughout the Lagging Strand), a provisional lagging-strand, consisting of RNA segments derived from processed transcripts, is established as the replication fork proceeds [14]. This RNA is then replaced by DNA in a subsequent maturation step, in which lagging-strand DNA synthesis is initiated at one or more preferred sites, including OL. RITOLS shares many features with the strand-displacement model, the only major difference being that the latter postulates that the parental strand displaced during heavy-strand replication remains single-stranded until the light-strand initiates. The second type of replication intermediate detected by 2DNAGE is composed fully of dsDNA, whose structure implies bidirectional initiation of replication across a wider origin zone, stretching beyond the NCR. However, termination at OH means that this mode of replication is also effectively unidirectional [11], [16].
Mitochondrial DNA replication in Drosophila melanogaster, based both on early TEM [24], [25] and more recent 2DNAGE analyses [26], also involves two replication modes. The majority of replication intermediates are composed entirely of dsDNA, with no evidence of extensive RNA incorporation. Their structure implies unidirectional strand-coupled DNA synthesis, commencing in the NCR, with an initiation site as inferred previously by end-mapping [27]. A minority of replicating molecules retain a region of single-strandedness encompassing the rRNA gene locus just downstream of the origin, indicative of delayed lagging-strand completion in this limited part of the genome. Also of note was the inference of specific replication pause regions through which the replication fork travels more slowly, based on the pronounced accumulation of replication intermediates containing fork structures therein. The two major pause regions of the mitochondrial genome [26] correspond approximately with zones of convergence of oppositely transcribed blocs of genes (Fig. 1A).
The coding region of metazoan mtDNAs shows a highly compact organization, with little or no non-coding sequences between genes. Typically, genes are encoded on both strands, a type of organization that unavoidably risks encounters between the transcription and replication machineries, which compete for the same template. As in other genetic systems, these processes should be subject to regulation, in order to minimize and resolve potential conflicts, including both co-directional and anti-directional collisions between the two molecular machineries. Defects in this collision regulation have been shown to cause abortive DNA synthesis, mutagenesis and genomic instability in a wide range of organisms [28]–[33]. In E. coli, for example, transcription termination is essential for the maintenance of genome integrity [34], by minimizing the generation of double-strand breaks arising from replication-fork collapse. A recent report has documented the importance of a machinery to regulate replication pausing caused by collisions with transcription complexes [35].
The mitochondrial transcription termination factor (mTERF) family comprises a set of mitochondrial DNA-binding proteins with diverse, documented roles in mitochondrial gene expression [36], [37]. The key structural feature of these proteins is the presence of multiple TERF motifs (I–IX), which have been shown, at least in the case of human MTERF1 and MTERF3, to form left-handed helical repeats that create a superhelical DNA-binding domain [38], [39]. mTERF family members have been implicated in the regulation of transcriptional initiation [4], [40], [41] as well as attenuation [40], [42], [43], and have also been shown to participate in mitoribosome assembly and translation [44]–[47]. In the mouse, Mterf3 and Mterf4 are essential genes [4], [45], while Mterf2 is not [41]. Human MTERF1 terminates transcription bidirectionally in vitro at its major binding site downstream of the rRNA genes [48]–[50], but manipulation of its activity in cultured cells or knockout mice has rather modest effects on transcript levels [43], [51], whose physiological significance, if any, is unknown.
Four proteins of this family have been identified in Drosophila, of which the best characterized is mTTF (CG18124). mTTF binds two sequence elements in Drosophila mtDNA [42], each located at the junctions of convergently transcribed blocs of genes (see Fig. 1A). Its binding facilitates transcriptional termination bidirectionally in vitro and is required for transcriptional attenuation in vivo [52], [53]. The amount and activity of mTTF therefore influences the steady-state levels of mitochondrial RNAs whose coding sequences lie between the mTTF binding sites and the putative promoters [52]. Knockdown of an insect-specific paralog of mTTF, mTerf5 (CG7175), was found to have opposite effects on transcript levels to knockdown of mTTF, despite the fact that mTerf5 binds to the same sites in mtDNA in an mTTF-dependent manner [54].
As DNA binding proteins with an established role in the regulation of transcription, mTERF family members are strong candidates for mediating conflicts between the replisome and transcription complexes. Moreover, MTERF proteins may have multiple roles in mtDNA metabolism, considering that alterations in the levels of MTERF1 or its homologs MTERF2 (MTERFD3) and MTERF3 (MTERFD1) were reported to modulate the levels of paused replication intermediates in cultured human cells [55], [56]. The sea urchin mTERF protein mtDBP has been demonstrated in vitro to have contrahelicase activity [57]. This feature, commonly seen in replication termination proteins, is shared also by the mammalian nuclear rDNA transcription terminator TTF-1, which has been suggested to regulate entry of the replication machinery into an actively transcribed region [58]. The possible correspondence of the mTTF binding sites in D. melanogaster mtDNA with the regions of replication pausing identified in our earlier study suggests that mTERF family proteins could be considered as candidates for a similar role.
To test the possible involvement of mTTF and mTerf5 in mtDNA replication, we investigated their effects on mtDNA metabolism after manipulation of their expression by RNAi, both in cultured cells and in vivo. Here we show that both factors are required for normal mtDNA topology and maintenance. Lack of either (or both) resulted in developmental arrest at L3 larval stage. mTTF knockdown led to the accumulation of nicks, dsDNA breaks and recombination junctions. 2DNAGE demonstrated stalled and reversed replication forks over broad zones surrounding the mTTF binding sites, and an accumulation of aberrant replication intermediates with extended segments of RNA/DNA hybrid, indicating a failure to complete lagging-strand DNA synthesis. Knockdown of mTerf5 had an opposite effect on mtDNA replication intermediates, bringing about an increase in replication pause strength when compared to wild-type, a decrease in fragile replication intermediates containing single-stranded segments, and the disappearance of species with even the short segments of RNA/DNA hybrid that we were able to detect in wild-type cells.
Because of their opposing but essential roles in mtDNA expression and synthesis, we propose that the balance of these two mTERF family members facilitates the orderly and productive passage of oppositely moving replication and transcription complexes, preventing collisions that would otherwise result in abortive replication and loss of genome integrity.
Replication pauses are revealed as discrete spots on arcs of replication intermediates resolved by 2DNAGE [17], [59]. The two major replication pause regions of D. melanogaster mtDNA were previously localized to approximately 1/3 and 2/3 of genome length from the replication origin, located in the NCR [26]. In order to map these pause sites more precisely, we conducted 2DNAGE on overlapping short restriction fragments, in a size range considered optimal for resolution on the standard two-dimensional gel system, i.e. 3–5 kb [60]. These analyses revealed the pause sites as the expected discrete spots (Figs. 1, S1, red arrows), lying on standard Y-arcs which are characteristic of non-origin fragments through which a replication fork passes unidirectionally (see [19]–[23], [60] for full explanations of the signals seen on 2DNAGE). Within the ∼50 bp resolution of the method, and based on multiple digests (Fig. S1), each pause maps precisely to one of the two binding sites for mTTF in the genome, namely at the ND1/tRNASer(UCN) gene boundary (here designated bs1) and the tRNAPhe/tRNAGlu gene boundary (bs2; see Fig. S1C for explanation of mapping). The HindIII fragment beyond bs2, encompassing the remainder of the coding region, did not reveal any discrete pause signals. However, an enhanced signal relative to that seen in the adjacent ClaI fragments was evident at the start of the Y arc in this fragment (blue arrow), suggestive of a more diffuse replication slow-zone at the origin-proximal end of this fragment, consistent with previous data [26]. Treatment with the single strand-specific nuclease S1 had no effect on the migration of replication intermediates in any of the fragments tested, consistent with the previous inference that DNA replication in these regions is fully strand coupled [26].
The coincidence of replication pauses with the previously mapped binding sites for mTTF suggests a role for this protein in mtDNA maintenance. To investigate this we used dsRNA-based RNA interference to knock down mTTF in S2 cells. A ∼70% decrease in mTTF mRNA levels (Fig. S1A) resulted in altered mitochondrial transcript levels consistent with the previous report by Roberti et al. [52]. Depending on their location within the transcription map, transcripts were either upregulated (e.g. cytochrome b mRNA), downregulated (e.g. 16S rRNA and COX2 mRNA), or little altered (e.g. ND5 mRNA) (Fig. 2A). Furthermore, in untreated cells, transcript levels of a given strand were observed to decrease markedly as the mTTF binding sites are successively traversed within the transcription unit (Fig. S1B), consistent with the proposed role of mTTF as a transcriptional attenuator (although this may also be influenced by differential RNA stability).
Next we analyzed mtDNA levels in cells knocked down for mTTF, using three different methods: qRT-PCR (Fig. 2B), PicoGreen staining of mtDNA nucleoids (Fig. 2C) and Southern blotting of both digested (Fig. S1C, D) and undigested mtDNA (Fig. 2D). qRT-PCR indicated that mtDNA levels fell to approximately 20% of control levels following 4–5 days of mTTF knockdown (Fig. 2B), whereas mtDNA levels were unchanged in untreated cells or cells treated with an inert dsRNA against GFP. The intensity of PicoGreen staining after 5 d of mTTF knockdown (Fig. 2C) was also greatly diminished compared with control cells treated with the dsRNA against GFP and similar to the effect of dsRNA treatment targeted against genes with well-established roles in mtDNA synthesis, such as tamas (encoding the catalytic subunit of DNA polymerase gamma) or CG5924 (the Drosophila homologue of the mammalian mitochondrial helicase Twinkle). The relative amount of intact mtDNA detected by Southern blot was also diminished by mTTF knockdown (Fig. 2D), with a progressive disappearance of the supercoiled form relative to genome-length linear molecules (Fig. 2D). The total amount of mtDNA detectable by Southern hybridization after digestion with a restriction enzyme was also diminished (Fig. S2C, S2D). Analysis of full-length mtDNA by 2DNAGE revealed a relative increase both of recombination structures and broken replication replication intermediates (Fig. 2E: see [26] for a full explanation of the arcs revelaed by 2DNAGE of Drosophila mtDNA digested with restriction enzymes curring once in the genome). Recombination structures linking two whole copies of the genome following such linearization are most easily revealed in the Bsp1407I digest, where the characteristic X-arc that they form (see [19]–[23], [60]) is well resolved from termination and dY structures. Their accumulation was most prominent after 3 d of knockdown of mTTF (Fig. 2E, red arrows). Broken replication intermediates, arising from scission of one branch at or near the origin, migrate on or close to a standard Y-arc, instead of a bubble, double-Y or eyebrow arc (see Figs. 2E, 2F, S3). They are normally found only at a low-level in control cells, but are generated in material from control cells by treatment with S1 nuclease, which cuts the region that remains single-stranded in some replicating molecules, extending from the replication origin across the rRNA gene locus (see Fig. 6 of [26], panels from which are reproduced here in Fig. S3 for comparison). After mTTF knockdown, these broken intermediates were much more abundant, and further treatment with S1 nuclease had no effect (Fig. 2F, red arrows). The characterstic eyebrow arc seen in the Bsp1407I digest, resulting from non-digestion in the partially single-stranded region, was already absent, consistent with systematic strand-breakage in this region following mTTF knockdown (Fig. 2E, blue arrows).
Roberti et al. [52] earlier found no significant effect on mtDNA levels from 3 days of mTTF knockdown, but using a different dsRNA. To clarify this inconsistency and exclude possible off-target effects, we repeated the experiment using either the same dsRNA as Roberti et al. [52] or our own custom-designed dsRNA. mtDNA levels were decreased by ∼80% at day 5 in both cases, although the dsRNA used by Roberti et al. [52] produced its effects more slowly, with only a 15% drop in mtDNA levels at day 3 (Fig. S2E).
To investigate the effects of mTTF knockdown on mtDNA maintenance in the whole organism, we expressed a (hairpin) dsRNA transgene targeted on mTTF, using the ubiquitous and constitutive daughterless-GAL4 (da-GAL4) driver. We confirmed that the parental RNAi line (itself homozygous viable) produced normal numbers of progeny with a wild-type phenotype when mated to flies not expressing da-GAL4. We also confirmed that RNA interference in vivo produced ∼90% knockdown of mTTF at the mRNA level at larval stage (Fig. S4A), which was seen also at the protein level (Fig. S4B). mTTF knockdown larvae gained weight more slowly than wild-type larvae of the same genetic background (Fig. 3A). More than 90% of individuals failed to develop beyond the L3 larval stage although larval weight exceeded the critical range for developmental progression (Fig. 3A, [61]]. None of the few aberrant pupae progressed to the late pupal stages. The persistent larval stage lasted approximately 30 days, during which the larvae became progressively inactive and then died.
Mitochondrial RNA levels were altered in a similar manner as in mTTF knockdown cells, e.g. COX2 mRNA was decreased, whereas cytochrome b mRNA was elevated (Fig. 3B). Mitochondrial DNA copy number failed to increase as typically occurs during wild-type development, remaining at 40% of the wild-type level 3 days after hatching (Fig. 3A). During the persistent larval stage, mtDNA copy number steadily declined to approximately 25% of the maximum level observed in wild type L3 larvae, 25 days after hatching.
A similar accumulation of broken replication intermediates was observed in mTTF knockdown larvae as in S2 cells, e.g. as revealed by NdeI digestion (Fig. 3D, red arrows; compare with Figs. 2F, S3). The control strain (w1118 ; da-GAL4/+) displayed an identical pattern of replication intermediates to that described previously for the Oregon-R wild-type strain (Fig. 4 of [26]], ruling out any confounding effect of genetic background.
The observed drop in mtDNA copy number and topological changes following mTTF knockdown prompted us to characterize mtDNA replication intermediates in more detail, in cells and larvae knocked down for mTTF. In each of the two ClaI fragments that contained the mTTF binding sites, the discrete spots corresponding to replication pauses were observed to fade out and spread over a wider region of the Y-arc, during 4 days of mTTF dsRNA treatment of S2 cells (Fig. 4A, red arrows). After 4 days of treatment, the proportion of this novel material migrating along the Y-arc, relative to the unit-length fragment, was significantly increased compared to day zero for both bs1 and bs2. Concomitantly we observed a transient increase in cruciform DNA species, particularly a subclass of Holliday junction-like molecules (Fig. 4A, blue arrows). This is consistent with the increase in recombinational forms linking two full copies of the genome seen after 2–3 days of RNAi following digestion with restriction enzymes that cut once in the genome (Fig. 2E, red arrrows). Spreading of the pauses (Fig. S5, red arrows), with accumulation of recombination junctions (Fig. S5, blue arrows) was seen in mTTF knockdown larvae, although to a lesser extent than in mTTF knockdown cells.
Stalled replication forks have a tendency to regress along the template and, under some conditions, can adopt a “chickenfoot” structure around a Holliday-like junction [62], [63]; (see Figs. 4B and S6). If such fork reversal is relatively limited, the species formed would still migrate close to a classical Y-arc. However, they should become sensitive to nucleases targeting Holliday junctions [64]. To test this, we treated mtDNA with the bacterial cruciform-cutting enzyme RusA. This removed substantially more material from the region of the Y-arc in cells knocked down for mTTF compared to control cells (Fig. 4B, red arrows; see also Fig. S7 for side-by-side comparisons of gels at equivalent exposures). This supports the idea that mTTF knockdown resulted in the accumulation of regressed replication forks containing Holliday-like junctions, which may be considered a signature of replication stalling. Note the decrease of the recombination structures (blue arrows in Fig. 4B) migrating on the X-arc, following RusA treatment, thus confirming the functionality of the enzyme in this experiment.
Our findings are consistent with the idea that bound mTTF provides a natural barrier to fork progression, avoiding unregulated replication stalling that might arise, for example, from collisions of the replication and transcription machineries. Since mTTF is already known to promote transcriptional termination, we reconsidered the issue of the role of RNA in Drosophila mtDNA replication. Previous 2DNAGE analyses indicated that mtDNA replication intermediates in D. melanogaster were fully double-stranded [16], except for the rRNA locus, which exhibited single-strandedness in a minority of molecules. Restriction sites across the remainder of the genome were completely digestible, indicating that extensive regions of RNA/DNA hybirid, such as seen in RITOLS replication intermediates in vertebrates [12], [13], [65], were absent, although the presence of limited patches of RNA/DNA hybrid could not be excluded. We investigated the issue further by treating mtDNA, after restriction digestion, with RNase H, which digests regions of RNA hybridized to DNA, thus modifying the migration pattern of RITOLS-type replication intermediates. This analysis revealed a prominent, novel arc, migrating just below the Y-arc (Fig. 5, red arrows), whose trajectory is consistent with the presence of one or more short segments of ssDNA, arising from the enzymatic removal of RNA from some replicating molecules. Other species detectable by 2DNAGE were essentially unaffected by RNase H treatment, indicating that the novel arc arose from material previously not resolved on this gel system, which is consistent with the clear increase in signal seen after RNase H treatment (Fig. S8).
The nature and trajectory of the novel arc released by RNase H treatment differed markedly after knockdown of mTTF. The forms migrating just below the standard Y-arc (Fig. 5, red arrows), were replaced by a much shallower sub-Y arc, extending beyond the limit of the fragment analysed (Fig. 5, blue arrows). Its trajectory is consistent with much more extensive ssDNA regions (i.e. much longer segments of RNA/DNA hybrid prior to RNase H treatment) than in the replication intermediates that formed the sub-Y arc generated by RNase H treatment in untreated cells.
To test whether the mTTF partner protein mTerf5 antagonizes the effects of mTTF on replication as well as on transcription, we investigated the effect of mTerf5 knockdown on mtDNA copy number in S2 cells (Fig. 6A). We observed a substantial depletion of mtDNA to a similar extent (∼70%), and with similar kinetics, as mTTF knockdown, although there was no cross-reaction between the two dsRNAs used (Fig. S9). Simultaneous knockdown of both factors in S2 cells produced a small initial increase in mtDNA copy number, followed by a gradual decline to the same low level as produced by knockdown of either factor alone, after 5 days of treatment. In the developing fly, mTerf5 knockdown using each of three independently isolated RNAi lines driven by da-GAL4, produced the same phenotype as mTTF knockdown, i.e. a persistent larval stage with failure of pupariation. The congruent phenotype effectively excludes off-target effects as an explanation. Simultaneous knockdown of both factors in the developing fly also yielded this phenotype.
Despite the fact that mTerf5 knockdown produced similar effects on mtDNA copy number and development as mTTF knockdown, 2DNAGE analysis of mtDNA from mTerf5 knockdown cells revealed different effects on the pattern of replication intermediates. We observed enhanced replication pausing at both mTTF binding sites (Fig. 6C: for comparison based on gels of equivalent exposure, see Fig. S10). The broken intermediates that accumulated when mTTF was knocked down were absent (compare Fig. 6B with Fig. 2F, shown alongside in Fig. S3B), and treatment with S1 nuclease failed to release such intermediates in comparable amounts as in control cells (Fig. 6B, Fig. S3B). Treatment with RNase H had no discernible effect (compare Fig. 6C with the corresponding digests of Fig. 5). mTerf5 knockdown thus had opposite effects on replication intermediates as mTTF knockdown, appearing to enhance replication pausing at specific sites and shifting the balance of mtDNA replication intermedaites towards those composed fully of dsDNA, as opposed to those with patches of RNA/DNA hybrid or single-strandedness.
mTTF and mTerf5 were previously shown, using RNAi, to have reciprocal effects on transcription. Here we investigated their roles in mtDNA maintenance, using a similar strategy. Both factors were essential for mtDNA copy number maintenance, but had opposing effects on mtDNA replication. These findings allow us to propose a model whereby these factors co-operate to facilitate the productive interaction between oppositely moving replication and transcription complexes on the same template, thus contributing to the maintenance of genomic fidelity.
Contrary to a previous report [52], our data demonstrate that mTTF is required in vivo to maintain mtDNA levels. The different findings are attributable to the kinetics of action of the dsRNAs used in the two studies. The effects on transcription were broadly similar [36]: the minor differences are most likely due to early changes in mtDNA levels compounding those on RNA. The apparent drop in steady-state transcript levels as the mTTF sites are successively traversed reflects the organization of the mitochondrial genome, but makes no obvious sense for the equimolar supply of polypeptides belonging to any given OXPHOS complex. The transcription termination activity of mTTF might therefore serve primarily a different role, such as in DNA replication, with effects on mitochondrial transcripts being accommodated (post-) translationally.
The developmental arrest at larval L3 stage caused by deficiency of mTTF or mTerf5 is a phenotype shared by knockdown of many genes for mitochondrial functions, including those encoding mitochondrial transcription factor 2 (mtTFB2), single-strand binding protein (mtSSB) and CCDC56, required for the assembly of cytochrome c oxidase [66]–[68]. Whether it is a direct result of OXPHOS deficiency or of deranged developmental signaling remains to be determined.
Although we previously found no evidence for RNA-containing mtDNA replication intermediates in Drosophila [26], finer scale analysis indicated the presence of short patches of RNA scattered around the mitochondrial genome, based on the prominent sub-Y arcs seen on 2DNAGE after treatment with RNase H. Standard Y-arcs, which were already present before RNase H treatment, also remained after the treatment. There was a clear and reproducible increase in total signal in the resolving portion of 2D gels following treatment with RNase H. Logically, this material must have been released by the specific action of the nuclease, from high molecular-weight complexes or tangles previously unable to enter the gel, This is supported by similar observations on human heart mtDNA [69], much of which remained trapped in the well upon 2DNAGE, unless treated with suitable nucleases and/or topoisomerases to disrupt tangles visualized also by electron microscopy. We infer that mtDNA replication intermediates in Drosophila must consist of two classes, as in vertebrates. One class appears to be composed entirely of dsDNA, and is represented by the standard Y-arcs seen both before and after RNase H treatment. The second class, akin to the RITOLS intermediates seen in vertebrates [12], [13], contains tracts of RNA/DNA hybrid, except that here such tracts must be very short, so that RNase H generates a novel sub-Y arc which migrates close to the trajectory of the standard Y-arc. Short segments of RNA hybridized to replicating DNA may be covalently joined to longer transcripts, forming the complex tangles unable to enter gels unless released by RNase H treatment. The Y-like structure of the products, and the fact that they were created, not destroyed by RNase H, indicates that they are not simple intermediates of transcription, DNA repair or recombination.
These observations raise the issue of whether transcription and DNA replication can occur simultaneously on the same template and, if so, whether this association is obligatory. The existence of a population of mtDNA molecules only able to enter agarose gels after treatment with a ribonuclease strongly suggests that these are molecules engaged in active transcription. After RNA removal, they migrate along 2DNAGE arcs expected for an iterative set of replication intermediates, strongly supporting the idea that replication and transcription can take place on the same template. Those replication intermediates that can be resolved on 2D gels without ribonuclease treatment may represent a distinct subset of replicating molecules, in which transcription is prevented. Resolving these issues will require the development of novel methods for metabolic labeling and analysis of replication and transcription intermediates.
Knockdown of mTTF or mTerf5 produced specific and reciprocal effects on mtDNA synthesis. Lack of mTTF caused random stalling and fork regression, whilst decreasing those molecules specifically paused at the binding site itself. RusA treatment confirmed the presence of Holliday-like junctions, a signature of fork reversal associated with stalling due to replication impediments [63], and proposed as a necessary intermediate in replication repair [70]. The logical explanation for replication stalling in the present case is random collisions with the transcriptional machinery, as observed in other systems [29], [30], [71], such as the rDNA locus in yeast. The formation of Holliday-like chicken-foot structures at stalls of this type has not been reported previously, but our observation of an increase in X-form species containing recombination junctions after 2–3 days of mTTF knockdown suggests that stalling creates substrates for a recombinational repair and/or restart machinery. The observed mtDNA depletion and shift in topology indicates that such processes are unable to support the completion of replication sufficiently to maintain mtDNA copy number. The concomitant accumulation of broken replication intermediates, akin to those that can be created in material from unmanipulated cells by S1 nuclease treatment, indicates that the ssDNA region in the rRNA locus was systematically broken, suggesting that it was more pervasive or persistent than in control cells. Finally, a novel class of putative replication intermediates was observed to accumulate, inferred to contain more extensive RNA segments, based on the generation of shallower sub-Y arcs by RNase H. These replaced the forms with only short RNA segments, that were seen in control cells. Conversely, mTerf5 knockdown produced opposite effects, namely enhanced pausing at the mTTF binding sites, a decrease in replication intermediates broken at the rRNA locus and disappearance of the RNA-containing species. Thus, whereas mTTF is required for physiological pausing, mTerf5 allows paused replication to resume. Additional enzymatic treatments, as well as the use of in gel-digestion [72], heat denaturation prior to second dimension electrophoresis [73], [74], and electron microscopy, will be needed to reveal the detailed structural differences between replicating molecules paused naturally by mTTF/mTerf5, and those arising from unregulated or persistent stalling in their absence.
Some of the effects of mTTF knockdown on mtDNA replication could be indirect, e.g. resulting from altered transcript levels. However, a failure of replication due to primer insufficiency would lead to the progressive disappearance of shorter replication bubbles, rather than the accumulation of broken termination intermediates. Evidence of a role for preformed transcripts in RITOLS replication of mammalian mtDNA, via the bootlace mechanism [13], [14], suggests the possibility that mTTF deficiency might impair DNA replication by distorting the relative abundances of different processed transcripts that must be incorporated during fork progression. However, this would not explain the accumulation of random collision products. Thus, we favor a more direct role for mTTF in DNA replication.
The effects of mTTF and mTerf5 knockdown imply that RNA incorporation, replication-fork pausing and lagging-strand synthesis are related phenomena. RNA incorporated via the bootlace mechanism is one possible source of primers for the synthesis of lagging-strand DNA, although other mechanisms of lagging-strand initiation are consistent with RITOLS [9]. Our data suggest that proteins belonging to the mTERF family are crucial factors in execution and/or regulation of such a process, at least in Drosophila, as illustrated in Fig. 7. The proposed model postulates that the balance of mTTF and mTerf5 nurses the productive interaction of replication and transcription machineries moving in opposite directions, and that replication pausing is vital for ensuring the incorporation of RNA transcripts into replication intermediates at the replication fork (Fig. 7). Capture of a new bootlace, resulting from the arrival of a transcription complex that undergoes termination, is also proposed to be essential for the priming of lagging-strand DNA synthesis, not only at the immediate site of mTTF/mTerf5 binding, but also further downstream, as the replication fork progresses.
The prevention and/or regulation of collisions between the transcription and replication machineries is indispensable for all genetic systems [30], [75], to avoid knotting of the daughter molecules [76], the generation of recombinogenic ends [77] and other types of genomic instability [29]. Proteins that perform this function are well documented in other systems, for example in the rDNA of both fungi [78], [79] and mammalian cells [80], although these proteins (Fob1 in S. cerevisiae, Reb1 in S. pombe and TTF1 in mammals) are unrelated to the mTERF family and to each other. Thus, there is both a precedent and a rationale for mTTF and mTerf5 to integrate transcription and DNA replication.
However, the many cases of mitochondrial proteins having multiple functions, e.g. Ilv5, Aco1 and RNase P [81]–[83], means that it cannot be excluded that mTTF and mTerf5 regulate transcriptional and replication independently. The two proteins may also be considered as an example of an antagonistic pair that together control a specific process, a type of regulation widespread in biological systems. An intriguing parallel is provided by the related helicases Rrm3 and Pif1, which exert opposing effects on DNA synthesis at the replication fork barrier of Saccharomyces cerevisiae rDNA [84]. Unlike mTTF and mTerf5, Rrm3 and Pif1 do not bind DNA sequence-specifically, but can recognize and process unusual DNA structures in G+C-rich regions [85]. Although mTTF and mTerf5 are not themselves helicases, they may recruit an antagonistic helicase pair that act in a similar manner to Rrm3 and Pif1, or may confer alternate properties on a single helicase.
S2 cells [86] were cultured in Schneider's Medium (Sigma-Aldrich) at 25°C. Cells were passaged every 3–4 d at a density of 0.5×106 cells/ml. Standard Drosophila strains, plus the mTTF RNAi line 101656 and mTerf5 RNAi lines 2899, 2900 and 107227 from the Vienna Drosophila RNAi Center (VDRC), were cultured as described previously [87].
Gene-specific dsRNAs were synthesized from templates created from S2 cell cDNA by a nested PCR strategy, which introduced the T7 promoter sequence on both sides of each final amplicon (see Table S1 for primer sequences). S2 cells were transfected with 4 µg of each dsRNA added to 0.5 ml of culture medium, and grown for the times indicated in figures and legends. Where transfections were to be continued for >3 d, cells were passaged every 3 d and fresh dsRNA was added. For visualization of nucleoids, dsRNA against Tfam was added for the final 2 d, as described in SI, Nucleoids were detected by fluorescence microscopy, after staining with Quant-iT™ PicoGreen (Invitrogen).
Nucleic acids for mtDNA copy-number analysis, 2DNAGE and Q-RT-PCR were isolated from S2 cells, adult flies, larvae or purified mitochondria thereof using variants of standard methods. In general, 2DNAGE used total nucleic acids isolated from sucrose density gradient-purified mitochondria (see SI).
Q-RT-PCR to measure RNA levels was performed essentially as described earlier [52], using cDNA prepared by random priming or, where indicated, by gene- and strand-specific primers as detailed in Table S1. Assays always included three or more independent biological replicate samples, with normalization to the transcript of nuclear gene RpL32. Relative quantitation of mtDNA content was performed similarly, using total DNA as template, plus primers for mitochondrial 16S rDNA (Table S1), also with normalization to RpL32.
Standard one-dimensional electrophoresis used 0.6% agarose gels in TBE buffer. 2DNAGE and blot-hybridization were conducted essentially as described earlier [13], using slightly different conditions for resolving large and small DNA fragments (see SI). For details of restriction digests and treatment with DNA modifying enzymes see SI. Radioactive probes for specific fragments of Drosophila mtDNA were generated by PCR, with [α-32P]-dCTP (Perkin-Elmer, 3000 Ci/mmol) in the reaction mix (see Table S1 for primers).
For further details, see Text S1.
|
10.1371/journal.pgen.1005281 | A Transposable Element within the Non-canonical Telomerase RNA of Arabidopsis thaliana Modulates Telomerase in Response to DNA Damage | Long noncoding RNAs (lncRNAs) have emerged as critical factors in many biological processes, but little is known about how their regulatory functions evolved. One of the best-studied lncRNAs is TER, the essential RNA template for telomerase reverse transcriptase. We previously showed that Arabidopsis thaliana harbors three TER isoforms: TER1, TER2 and TER2S. TER1 serves as a canonical telomere template, while TER2 is a novel negative regulator of telomerase activity, induced in response to double-strand breaks (DSBs). TER2 contains a 529 nt intervening sequence that is removed along with 36 nt at the RNA 3’ terminus to generate TER2S, an RNA of unknown function. Here we investigate how A. thaliana TER2 acquired its regulatory function. Using data from the 1,001 Arabidopsis genomes project, we report that the intervening sequence within TER2 is derived from a transposable element termed DSB responsive element (DRE). DRE is found in the TER2 loci of most but not all A. thaliana accessions. By analyzing accessions with (TER2) and without DRE (TER2Δ) we demonstrate that this element is responsible for many of the unique properties of TER2, including its enhanced binding to TERT and telomerase inhibitory function. We show that DRE destabilizes TER2, and further that TER2 induction by DNA damage reflects increased RNA stability and not increased transcription. DRE-mediated changes in TER2 stability thus provide a rapid and sensitive switch to fine-tune telomerase enzyme activity. Altogether, our data shows that invasion of the TER2 locus by a small transposon converted this lncRNA into a DNA damage sensor that modulates telomerase enzyme activity in response to genome assault.
| Telomerase is a highly regulated enzyme whose activity is essential for long-term cellular proliferation. In the presence of DNA double-strand breaks (DSBs), telomerase activity must be curtailed to promote faithful DNA repair. We previously showed that the flowering plant Arabidopsis thaliana rapidly down-regulates telomerase in response to DSBs, and further that this mode of regulation is dependent on TER2, a non-canonical telomerase RNA subunit. Here we demonstrate that the unique regulatory properties of TER2 are conveyed by a transposable element (TE) embedded in the TER2 gene. A comparison of A. thaliana accessions with and without the TE revealed that the element increases the binding affinity of TER2 for the telomerase catalytic subunit TERT relative to the canonical telomerase RNA subunit. The TE also increases TER2 turnover. In response to DSBs, TER2 is induced and accumulates in TERT containing complexes in vivo. Thus, invasion of a TE endows TER2 with a DNA damage sensor to rapidly and reversibly modulate enzyme activity in response to genotoxic stress. These findings provide an example of how exaptation of a TE altered the function of a long noncoding RNA. In this case, a duplicated gene (TER2) was used as the platform, and the TE as the tool to engineer a novel mode of telomerase regulation.
| The discovery of long noncoding RNA (lncRNA) has challenged the prevailing paradigm of protein-mediated regulation of gene expression and cell behavior. lncRNAs play essential roles in epigenetic regulation, stem cell biology and signal transduction and are emerging as key targets in human disease [1–3]. Unlike small regulatory RNAs (e.g. miRNAs, siRNAs), lncRNAs are not subjected to purifying selection, and as a consequence they are very poorly conserved, tending to emerge quickly and evolve swiftly [4]. Although transcriptome analyses have uncovered a vast array of lncRNAs, just a tiny fraction of these have an assigned biological function, and fewer still an ascribed molecular mechanism. Little is known about the evolutionary pathways via which lncRNAs gain new functions.
The telomerase RNA subunit TER is a lncRNA and an integral component of the telomerase enzyme. TER functions as template to direct the synthesis of telomeric DNA by the telomerase reverse transcriptase TERT. Telomerase continually synthesizes telomeric DNA in stem and germline cells to avert cellular senescence. Conversely, in cells with limited proliferation programs telomerase activity is repressed, an outcome in vertebrates that may have evolved to avert tumorigenesis [5,6]. Mechanisms of telomerase regulation are varied and complex, and include modulation of telomerase localization, recruitment to the telomere and enzymology at the chromosome terminus [7]. Within the telomerase ribonucleoprotein itself, the major target of enzyme regulation is TERT. However, TER is also implicated in telomerase control. In addition, different isoforms of core telomerase components influence telomerase behavior [8,9].
In conjunction with modulating telomerase action at natural chromosome ends, the enzyme must also be restrained from acting at sites of DNA double-strand breaks (DSBs). Barbara McClintock coined the term “chromosome healing” to describe the acquisition of telomeres on broken chromosomes in maize [10]. Although de novo telomere formation (DNTF) protects the terminus from subsequent repair activities, it leads to loss of the centromere distal chromosome fragment. Thus, DSBs must be sheltered from telomerase action to prevent gross chromosomal rearrangements and loss of heterozygosity. Multiple pathways evolved to prevent the establishment of telomeres at DSBs in yeast [11]. For example, phosphorylation of the Cdc13 telomere binding protein decreases its affinity for DSBs [12]. In addition, the Pif1 helicase is activated by DSBs, resulting in removal of telomerase from DNA [13]. Less is known about how DNTF is repressed in multicellular eukaryotes. In mammals, DSBs trigger TERT phosphorylation leading to decreased telomerase activity [14]. In addition, ionizing radiation causes transient sequestration of TERT in the nucleolus [15]. In Arabidopsis thaliana, a non-canonical TER represses telomerase activity in response to DSBs [16].
TER ranges in size from 150 nt in Tetrahymena to >2 kb in certain fungi, and while the nucleotide sequence is highly variable across species, core secondary and tertiary structures are conserved and essential for TER interaction with TERT and for telomerase catalysis [17–21]. TER is transcriptionally regulated in mammals [22], but the transcript is highly stable with a half-life of several days [23]. Recent data show that that 3’ terminus of Schizosaccharomyces pombe TER is generated by an additional RNA processing step termed slicing, which involves only the first step in mRNA splicing [24,25]. Conventional introns have not been associated with TER.
Arabidopsis thaliana is unusual in that it harbors two TER genes, TER1 (784 nt) and TER2 (748 nt) [26]. Within TER1 and TER2, there are two regions of high similarity spanning ~219 nt termed conserved region 1 (CR1) and conserved region 2 (CR2). In TER2, CR1 and CR2 are separated by a 529 nt intervening sequence. An additional unique 36 nts lie at the 3’ end of the TER2 CR2 termed 3’R. The intervening sequence and 3’R are removed in vivo to create a truncated isoform called TER2S [16]. Sequences flanking the intervening sequence do not adhere to consensus splice donor and acceptor sites, suggesting that removal of this element may not proceed via conventional mRNA splicing.
Although the function of TER2S is unclear, TER1 and TER2 play opposing roles in the control of telomerase enzyme activity. TER1 serves as the canonical telomere repeat template necessary for telomere length maintenance in vivo [26]. Plants deficient in TER1 exhibit progressive telomere shortening, and mutations in the TER1 template alter the telomere repeat sequence in vivo. In contrast, TER2 does not direct telomere repeat incorporation in vivo. Instead, this RNA negatively regulates TER1-mediated enzyme activity. Telomerase activity is elevated in plants lacking TER2, while in plants over-expressing TER2, telomerase activity is decreased and telomeres shorten [16].
TER2 is regulated by DNA damage. Under standard growth conditions, the steady state levels of TER1 and TER2S are similar, and 10-20-fold higher than TER2 [16]. However, in response to DSBs, TER2 is rapidly induced and becomes the predominant TER isoform. The increase in TER2 is coincident with a reduction in telomerase activity. Indeed telomerase inhibition is dependent on TER2: ter2 mutants do not down-regulate telomerase in response to DNA damage [16]. Telomerase repression is not elicited by replication stress or telomere dysfunction, indicating that TER2-mediated telomerase regulation is specific for DSBs and thus may play a role in repressing DNTF. While the mechanism of TER2-mediated telomerase inhibition is not known, TERT has a higher affinity for TER2 than for TER1 or TER2S, and preferentially assembles into TER2 containing RNP complexes in vivo. Therefore, TER2 may serve as a molecular sponge to sequester TERT in a non-functional RNP in response to DSBs [16].
TER is evolving rapidly in Arabidopsis and its relatives. Analysis of sixteen closely related species within the Brassicaceae lineage revealed that these species contain a single locus that bears similarity to the 3’ end of TER1 and the 5’ end of TER2 from A. thaliana [27]. Remarkably, several of these TER-like loci lack a template domain altogether, indicating that a functional TER must be encoded elsewhere in the genome. The intervening sequence associated with A. thaliana TER2 is missing from the TER-like genes of other Brassicaceae. Thus, the appearance of TER2 and its intervening sequence represent recent events likely generated during a massive genome rearrangement that occurred on the branch leading to A. thaliana [28].
In this study we employ a comparative genomics approach to investigate the regulatory function of TER2. Using data acquired from the 1,001 Arabidopsis genomes project, we show that the intervening sequence in TER2 has the characteristics of a solo long terminal repeat (LTR) from a Copia-like retrotransposon. The element is associated with most, but not all of the TER2 loci. We report that the unique regulatory functions of TER2, including its responsiveness to DSBs, are derived from this transposable element. Consequently, invasion of the TER2 locus by a transposon transformed this lncRNA into a highly sensitive DNA damage sensor that modulates telomerase enzyme activity.
Since a clear TER2 ortholog could not be discerned in other members of the Brassicaceae, we analyzed genomic sequence data for different A. thaliana accessions, natural strains of A. thaliana collected from the wild. A. thaliana diverged from its closest relative 10 million years ago [29]. It is estimated that Col-0 and Ler-0, the two best studied A. thaliana accessions, are approximately 200,000 years divergent from one another [30]. We retrieved TER1 and TER2 loci from 853 accessions compiled by the 1001 Arabidopsis genomes project (http://signal.salk.edu/atg1001) and analyzed them for variation against Col-0, the A. thaliana reference genome where a regulatory function for TER2 was first described [16]. The TER1 locus is highly conserved, including the 5’ and 3’ regions flanking CR1 and CR2 (Fig 1A), which lie upstream of the RAD52 coding region or within a predicted intron [27,31]. TER1 exhibits 92% identity across the sequenced accessions, but a few polymorphisms are scattered across the RNA (Fig 1A and 1B, S1A Fig). The most notable variations lie within the TER1 template domain (S2A Fig). A transition of A to C occurred three times while a T-A transversion appeared in 44/853 accessions. In neither instance are the two variations found within the same TER1 gene. Because the A. thaliana TER template is 11 nt in length and encodes one and a half copies of the telomere repeat, these TER1 RNAs retain the potential to direct synthesis of TTTAGGG repeats. More intriguing is the C to T mutation in the middle of the template in Bela-1 (S2A Fig). Whether this variation reflects a sequencing error or indicates that an alternative TER1 locus is present in this accession is unknown.
Like TER1 much of TER2 is strongly conserved. CR1 retains high percent identity among the accessions (92%) (Fig 1C). CR2 and the 3’R are also very well conserved with complete conservation in >60% of the accessions analyzed (S3 Fig). Conservation of 3’R was unanticipated since this segment of TER2 is eliminated in the production of TER2S (Fig 1A). Nevertheless, the high degree of conservation in CR1, CR2 and 3’R argues that these regions are important for TER2 function.
Although the intervening sequence within TER2 is completely conserved in more than 60% of the accessions, striking sequence divergence was observed in many of the other accessions. Two islands of conservation with ≥ 50% identity were identified within the intervening sequence, one corresponding to 63 nt and a second of 123 nt (S2B Fig). Hyper-variable sequences flank these regions within the 65 accessions bearing an incomplete intervening sequence. To verify the TER2 sequencing data, we performed PCR genotyping on a sampling of accessions predicted to harbor an intact intervening sequence (Col-0, Ws-2), a partial intervening sequence (Aa-0, Ang-0, Co-1 and Ei-2) or no intervening sequence (Ler-0). PCR primers were positioned within CR1 and 3’R (S4A Fig). A 784 bp PCR product is expected for accessions bearing an intact intervening sequence, a 255 nt product for accessions completely lacking the intervening sequence, and an intermediate size product for accessions with a partial intervening sequence. Products of the expected sizes were obtained for loci predicted to contain an intact or no intervening sequence, but for all TER2 loci predicted to contain a partial intervening sequence, the genotyping results indicated that this element was completely absent (S4B Fig). Genotyping repeated with siblings from accessions predicted to contain a partial intervening sequence gave the same result (S4C Fig). Genotyping was performed on several additional accessions reported to contain a partial intervening sequence (S1 Table). In all cases, the intervening sequence was absent. Finally, PCR products were sequenced from TER1 and TER2 reactions, with TER1 polymorphisms serving as a control to ensure that seed stocks were as expected (S4B and S4D Fig). The sequencing results confirmed the PCR genotyping data. For all partial intervening sequence accessions tested, there was complete loss of this element. The sequencing data also revealed a substantial deletion (~20 bp) within CR2 in two accessions (S4D Fig).
The simplest explanation for these genotyping results is that the TER2 locus was mis-annotated in some of the A. thaliana accessions. However, we cannot exclude the possibility that the intervening sequence within TER2 is extremely labile and between the time the genome sequencing was performed and our acquisition of seeds, partially deleted elements were completely eliminated.
For reasons discussed below, we named the intervening sequence within TER2 DSB responsive element (DRE). BLAST analyses against the A. thaliana genome using DRE as a query returned two hits, one on the left arm of chromosome 3 (adjacent to At3G30120) bearing 94.6% identity to DRETER2 termed DRE3L, and another on the right arm of chromosome 3 (adjacent to At3G50120) showing 63.4% identity called DRE3R (Fig 2A). Both DRE3L and DRE3R are found within intergenic regions and display a number of single-nucleotide polymorphisms among A. thaliana accessions (S5A Fig).
BLAST was performed to determine if the DRE is present in other species within the Brassicaceae family. Arabidopsis lyrata, A. thaliana’s closest relative, contains 32 copies of DRE dispersed throughout the genome (Fig 2B). A significant fraction of these elements exhibit a high degree of similarity within the 5’ 200nt of DRETER2, and are associated with open reading frames encoding typical retrotransposon proteins (S6 Fig). Three DREs were also detected in Capsella rubella, four in Brassica rapa, and ten in Eutrema salsugineum (Fig 2B). The presence of multiple copies of DRE in A. thaliana and its relatives suggests that it is a transposable element (TE). Consistent with this conclusion, sequences at the 5’ and 3’ borders of DRETER2 contain a 5 nt tandem inverted repeat of TGTTG/ACAAC (Fig 2C, brown bar). The tandem inverted repeat at the 5’ and 3’ boundaries of DRETER2 and DRE3L are highly conserved across the A. thaliana accessions and are present at the boundaries of DREs detected in other species (S6 Fig). In addition, a target site duplication of TCGTC is present at the 3’ end of CR1 and the 5’ end of CR2 of TER2 (Fig 2C, green bar). Tandem site duplications flank all three DREs in A. thaliana, ranging in length from 5 nt for DRETER2 and DRE3L to 18nt for DRE3R (Fig 2C, green bar). The tandem site duplication sequence varies, consistent with the hypothesis that these insertions represent unique TE insertion events rather than gene duplications. The small size of DRE and its association with tandem inverted repeats and target site duplications suggest that DRE is derived from a solo LTR of the abundant Copia family. Based on synteny mapping with Arabidopsis lyrata we confirmed that all three Copia-like solo LTRs in A. thaliana (TER2DRE, DRE3L, and DRE3R) are unique insertion events and are of approximately the same age (S7 and S8 Figs). Since the large majority of A. thaliana accessions apparently harbor an intact DRE within the TER2 locus, it is likely that the element was inserted soon after the TER duplication and was subsequently lost in a small subset of accessions.
The presence of two distinct TER2 alleles in A. thaliana provided us with an opportunity to study the functional impact of DRE. We previously showed that two RNA transcripts are derived from the Col-0 TER2 locus: the primary TER2 transcript and a processed isoform, TER2S, in which DRETER2 is removed along with 3’R [16,26]. In the Ler-0 accession, the TER2 locus lacks DRE, and thus the primary transcript is predicted to be TER2Δ. To assay for TER2Δ, RT-PCR was performed on RNA from Ler-0 seedlings using primers directed at CR1 and 3’R, which is unique to TER2 (Fig 3A and 3B). A product of the expected size was generated, indicating that a Ler-0 transcript containing CR1, CR2 and 3’R is present. Sequence analysis confirmed this conclusion. Notably, the CR1/CR2 junction in Ler-0 TER2Δ is distinct from Col-0 TER2S [26] as it contains only a single 5’ TCGTC 3’ motif instead of the two found in Col-0 (Fig 3B bottom, underlined sequence). Although a faint signal for TER2 was observed in Col-0 using our PCR conditions, TER2Δ was not (Fig 3A), suggesting that TER2Δ is either a transient processing intermediate, or is not generated during the conversion of TER2 to TER2S.
Col-0 TER2 is a poorly expressed transcript (Fig 3A) and is substantially less abundant than TER1 or TER2S [16]. To assess the relative abundance of Ler-0 TER2Δ, we performed qPCR (Fig 3C). The steady state level of TER1 was similar in Ler-0 and Col-0. However, Ler-0 TER2Δ was approximately 6–8 fold more abundant than Ler-0 TER1. By comparison, Col-0 TER2 was 15–20 fold less abundant than Col-0 TER1 (Fig 3C). Thus, Col-0 TER2 and Ler-0 TER2Δ are differentially regulated in vivo.
In Col-0, TER2 but not TER1 or TER2S is rapidly induced by DSBs [16]. Therefore, we asked if regulation is confined to TER2 by examining TER2 and TER2Δ in other A. thaliana accessions (Fig 4A). Seven day-old Ler-0 and Col-0 seedlings were treated with 20μM zeocin and qPCR was performed. In control reactions, BRCA1 mRNA was induced in both accessions after 2 hours and peaked at 4 hours, confirming that a DNA damage response was elicited (Fig 4B). As expected, the level of TER1 was unchanged in Ler-0 and Col-0 following zeocin treatment (S9 Fig). In addition, Col-0 TER2 increased 2.5 fold after 2 hours in zeocin relative to untreated seedlings (Fig 4C). In marked contrast, there was no significant change in TER2Δ over the 4 hour zeocin treatment (Fig 4C).
To test if DSB-mediated regulation of TER2 is a peculiarity of the Col-0 accession, we examined TER2/TER2Δ transcripts in two additional accessions: Ws-2, which contains DRETER2 and Co-1, which lacks it (Fig 4D). Consistent with the findings in Ler-0 and Col-0, there was no change in Co-1 TER2Δ, while Ws-2 TER2 was induced (Fig 4D). We conclude that the effect of DSBs on TER2 varies across A. thaliana accessions, and correlates with the presence of DRETER2.
We next asked if transcripts were derived from the other two DRE-like sequences in Col-0, and if so whether they responded to DSBs. Semi-quantitative RT-PCR was performed with primers specific for DRE3L and DRE3R on seedlings in the presence or absence of zeocin (S5B Fig). DRE3L transcripts could not be detected under either condition. However, transcripts from DRE3R were observed in the presence of zeocin (S5B Fig), indicating that a DNA damage-sensing element resides within DRETER2 as well as DRE3R
We previously showed that Col-0 TERT displays a hierarchy of binding favoring TER2 > TER1 >> TER2S both in vitro and in vivo [16]. The molecular basis for the enhanced affinity of TERT for TER2 is known. Since DRETER2 and the 3’R are unique to TER2, it seems likely that one of these elements influences TERT binding. To investigate this possibility, we examined the relative affinity of TERT for TER2Δ. Col-0 and Ler-0 seedlings were subjected to immunoprecipitation with TERT antibody followed by qPCR (Fig 4E). We set the ratio of TER2 to TER1 in the Col-0 TERT IP to 1, and then assessed the change in TERT-bound TER2 following zeocin treatment. The relative abundance of TER2 containing TERT complexes increased ~ 7-fold in response to DSBs (Fig 4E). Since the input level of TER2 increased by only 2.5-fold under these conditions (Fig 4C), the data raise the interesting possibility that other DNA damage-induced factors promote TER2 assembly with TERT. In marked contrast to TER2, we found that TER2Δ is not a preferred binding partner for TERT in vivo, and further zeocin treatment did not change the relative abundance of TER2Δ containing TERT complexes (Fig 4E). These results argue that the increased affinity of TERT for TER2 in Col-0 reflects the presence of DRETER2 and not 3’R.
Since Col-0 plants lacking TER2 do not down-regulate telomerase activity in response to DSBs [16], we asked if DSB-induced telomerase regulation is dependent on DRETER2 by comparing the level of telomerase activity in Ler-0 and Col-0 in the presence of zeocin. As expected application of quantitative telomere repeat amplification protocol (qTRAP) to Col-0 seedlings treated with zeocin for 2 or 3 hours showed reduced telomerase activity (70% decrease) compared to untreated seedlings (Fig 4F and S10 Fig). Although there was an alleviation of the inhibitory effect after 3–4 hours of treatment, enzyme activity was maintained at 50% of untreated level. In contrast, under the same treatment regime, telomerase activity was unaltered in Ler-0 (Fig 4F). Similar results were obtained with Ws-2 (plus DRETER2) and Co-1 (minus DRETER2) accessions, respectively (Fig 4G). These findings imply that DRE is necessary for DSB-induced telomerase repression.
To further assess the role of DRE in telomerase regulation, we generated two transgenic A. thaliana lines. First we asked if the presence of TER2 was sufficient to alter the level of telomerase activity in Ler-0 by expressing TER2 from its native promoter in this accession. In one of the transformants, the steady state level of transgenic TER2 was higher (2.5 fold) than the basal level of endogenous TER2 in wild type Col-0 (Fig 5A). qTRAP revealed a small, but statistically significant decrease in telomerase activity in the transformant (Fig 5B), indicating that Ler-0 telomerase can be down-regulated by Col-0 TER2. Next we asked if over-expression of TER2Δ altered telomerase activity in Col-0. TER2Δ expression was driven by the powerful CaMV promoter in wild type Col-0. As expected, there were no change in TER1 or TER2, but the steady state level of transgenic TER2Δ was ~8-fold higher than endogenous TER2Δ in wild type Ler-0. However, qTRAP showed no change in telomerase activity relative to untransformed Col-0 controls (Fig 5A and 5B). We conclude that the regulation of telomerase by TER2 is dependent on DRETER2.
The rapid induction of Col-0 TER2 in response to DSBs could occur through increased TER2 transcription or increased RNA stability. Because the sequences upstream of all TER2 genes are highly conserved, we considered the former possibility less likely. Indeed, when TER2 transcription was monitored in seedlings expressing a fused GUS reporter to a TER2 or TER2Δ promoter in Col-0 and Ler-0, respectively, approximately the same level of GUS staining was observed in the presence or absence of zeocin (S11 Fig). Hence, TER2 induction in response to DNA damage is not caused by increased transcription.
We assessed TER2 stability using six day-old seedlings treated with the transcription elongation inhibitor cordycepin. TER1 and TER2 RNA levels assessed by qPCR showed that Col-0 and Ler-0 TER1 have similar half-lives, t1/2 = 75 and 84 min, respectively (Fig 6A). The stability of TER2Δ was even greater with t1/2 = 244 min (Fig 6B). TER2, on the other hand, had a much shorter half-life than either TER2Δ or TER1: TER2 t1/2 = 13 min (Fig 6B). Thus, TER2 is an intrinsically unstable transcript.
To test if DSBs reduce TER2 turnover, Col-0 seedlings were treated with cordycepin to pause transcription and then zeocin was added after 90 min to produce DSBs. Although there was a slight change in the abundance of TER1 and BRCA1 mRNA in the presence of zeocin, this change was not statistically significant (Fig 6C and 6D). In contrast, TER2 abundance declined sharply over the 3.5 hour time course, but immediately after the introduction of zeocin, TER2 was stabilized (Fig 6E). These data implicate DRETER2 as the causal factor in destabilizing TER2 and in turn negatively regulating telomerase activity during bouts of DNA damage.
When the insertion of a TE within or adjacent to a gene leads to a change in gene function the process is termed “exaptation” [32]. Exaptation can alter gene regulation through myriad different mechanisms. A prominent example in plants is the insertion of multiple TEs adjacent to teosinte branched1 (tb1), which gave rise to domesticated maize [33]. One of the TEs disrupts a regulatory region of tb1, leading to increased expression and enhanced apical dominance. In vertebrates, exaptation of TEs is more prevalent at lncRNA loci than in protein-coding genes [34]. Approximately 41% of vertebrate lncRNA sequence is derived from TEs [35,36], leading Johnson and Guigo to propose that TEs can behave as pre-formed functional RNA domains, and further that exaptation of TEs is a major driving force in lncRNA evolution [36]. A recent systematic survey in vertebrates catalogued multiple instances of TEs altering lncRNA promoters, splice sites, and polyadenylation sites [37]. LncRNAs can also acquire novel interaction partners as a direct result of TE exaptation [32]. For instance, TEs within XIST facilitate interaction with a host of protein complexes including PRC2 and splicing factor ASF2 [38].
Here we show that invasion of a small TE (DRE) into the A. thaliana TER2 locus profoundly altered the function of this lncRNA (Fig 7). This exaptation event is not fixed, as the TER2 genes in 9% of the 853 accessions examined lack DRE. Insertion and subsequent loss of TEs is not uncommon in Arabidopsis. Some 80% of the annotated TEs in A. thaliana were lost in one or more accessions [39]. In the 200,000 years since Col-0 and Ler-0 diverged, at least 200 TEs have been active, and the unique insertions/deletions between the two accessions have biological implications [30]. One illustrative example of TE exaptation occurred at the Flowering Locus C (FLC) in Ler-0. Insertion of a Mutator-like transposon in this accession decreased FLC transcription, causing early flowering [40]. In this study we exploited the natural genetic heterogeneity within the TER2 locus, and discovered that many of the unique functions ascribed to this lncRNA derive from DRE.
First, DRE destabilizes TER2. A survey of ~800 lncRNAs in mouse revealed that only a small fraction are unstable, defined as RNAs with a half-life of less than 60 minutes [41]. By this criterion, TER2 is a highly unstable transcript with a half-life of only 13 minutes (Fig 7). TER1 (t1/2 = 80 min) and TER2Δ (t1/2 = 240 min), on the other hand, are categorized as stable RNAs. Unstable lncRNAs, like their unstable mRNA counterparts, are typically associated with regulatory functions, while stable RNAs are thought to serve housekeeping roles [42]. With Col-0 A. thaliana TER1 and TER2, this paradigm also holds.
A second key observation is that the instability of TER2 arising from DRE is reversed in response to DNA damage (Fig 7). The abundance of TER2, but not TER1 or TER2Δ is elevated in response to DSBs, and this change is largely, if not entirely, dependent on RNA stabilization rather than new transcription. Exaptation of a TE is known to endow host genes with the capacity to respond to environmental cues. For example, a cold-sensitive TE was inserted into the promoter of Ruby, a transcription factor that regulates flesh color in Citrus sinensis (blood orange). Cold activates the transposon, which in turn activates Ruby and downstream anthocyanin production [43]. In the case of TER2, DRE imparts DNA damage sensitivity, which increases TER2 abundance. How TER2 is regulated in response to DSBs is unknown. One possibility is that DRE carries binding sites for one or more interaction partners responsive to DNA damage, which then stabilize TER2. RNA binding proteins can play a significant role in the DNA damage response by regulating specific target genes post-transcriptionally [44]. TER2 turnover might be controlled through the small RNA regulatory pathway. A 24 nt RNA is associated with DRETER2 [45]. This finding is particularly intriguing given the recent discovery that small RNAs modulate the response to DSBs in both vertebrates and Arabidopsis [46]. Finally, it is possible that DNA damage blocks the RNA processing steps (e.g. splicing) that lead to production of TER2S (Fig 7). Splicing machinery has emerged as a target of the DDR [47].
The third key observation from this work is that DRE increases the affinity of TER2 for TERT (Fig 7), and correlates with the down-regulation of telomerase activity. DRE could modify TER2 structure in a manner that enhances its inherit affinity for TERT. Alternatively, DRE may make independent contacts with TERT to increase TER2 affinity. Intriguingly, zeocin treatment causes an even greater enrichment of TER2 containing TERT complexes than expected based on the fold induction of TER2, suggesting that a TER2 associated factor that is also responsive to DNA damage might drive the assembly of TER2-TERT RNPs.
Altogether, our data are consistent with a model in which exaptation of a TE into the A. thaliana TER2 locus gave rise to a new mode of telomerase regulation. Specifically, we propose that the DRE converted TER2 into a DNA damage sensor that controls telomerase enzyme activity through sequestration of TERT. Furthermore, because this regulatory pathway is regulated by changes in RNA stability, it is both rapidly responsive and reversible, allowing the A. thaliana accessions that carry DRE to fine-tune telomerase activity when the plant is under genome assault. These discoveries provide a fresh perspective on the role of TE exaptation in shaping lncRNA function and evolution.
For experiments with seedlings, seeds from different accessions (Col-0, Ler-0, Ws-2, etc) were sterilized in 50% bleach with 0.1% Triton X-100 and then stored in 4°C for 2–4 days. Liquid Murashige and Skoog (MS) medium were used for germination and growing [16]. After transferring cold-treated seeds to MS, plants were grown at 22°C under long day light condition for ~7 days. The Col-0 TER2 gene including 3kb upstream sequence and 300bp downstream sequence was cloned in the pMDC99 vector for transformation in the Ler-0 background. Hygromycin MS plates were used for selection. For Col-0 transformation, TER2Δ together with 300 bp downstream flanking region was cloned into the pBA002 vector with 35S promoter. BASTA MS plates were used for the selection.
Sequences corresponding to TER2 (Genbank accession number: HQ401285.1) were obtained using the genome browser at http://signal.salk.edu/atg1001. The search query AT5G24660 was used to pinpoint the region of interest, and all available tracks (accessions) were selected. Two sequences were removed from our analysis. Hov 3–2 was removed because it was the only accession with two deletions in the 5’ end, corresponding to 20 nt from the 5’ start of TER2, and a 100 nt deletion starting at nucleotide #101. The template region was not disturbed in this accession, possibly indicating a functional TER2 is generated. The Tottarp-2 accession was removed because the sequence corresponding to our search region did not contain sequences corresponding to TER2, most importantly, a template region.
Sequences were trimmed in MEGA5, and then analyzed using Geneious v6.0 (Biomatters). Sequence conservation and alignments were performed using Geneious. DRE-like sequences were obtained by BLAST searches of the A. thaliana (www.arabidopsis.org), A. lyrata, Capsella rubella, Brassica rapa, and Thellungiella halophila genomes accessed via www.phytozome.net v9.1 [48,49].
A. thaliana seedlings (5–7 day old) were transferred to fresh MS liquid medium with 20 μM zeocin (Invitrogen) as described [16]. Seedlings were kept in the dark with gentle agitation for 1, 2 or 4 h. Multiple seedlings were combined and flash frozen in liquid nitrogen for RNA extraction or protein extraction for TRAP. The combined sample was treated as a single biological replicate.
DNA samples were prepared from the leaves of different accessions. Both TER1 and TER2 loci were used for genotyping. PCR samples were resolved in 1% agrose and gel purified and sequenced. RNA was extracted from seedlings using the Direct-zol RNA MiniPrep kit (Zymo Research, Epigenetics) according to the manufacturer’s instructions. 1 μg total RNA was used for preparing cDNA. For RT-PCR, cDNA was synthesized by SuperscriptIII Reverse Transcriptase (Invitrogen) using random primers. For qRT-PCR, reverse transcription was performed using the Superscript cDNA master mix (Quanta), according to the manufacturer’s instructions. 1:5 diluted cDNA was used for qPCR. qPCR was performed on a Bio-Rad CFX-1000 using the following primers: qTER2Δ F: 5’-AGAACGTTGACGGCTAAAGG-3’; qTER2Δ R: 5’- TGTGGCATAAGGCAAACTGA-3’; TER2, BRCA1, TER1 and GAPDH primers are used as described before [16]. Data were analyzed using Bio-Rad’s CFX manager software. ΔΔCT values were obtained by comparing against GAPDH levels.
qTRAP assays were performed as described [50]. Data were normalized against untreated Col-0. For immunoprecipitation, TERT antibody [50] was conjugated with Dynabeads Protein A (Invitrogen) then incubated with protein extracts in 4°C. RNA was recovered from the IP sample using phenol/chloroform followed by ethanol precipitation [16]. qPCR was performed on TER1 and TER2/TER2Δ. The ΔCT value was used to determine the relative level of TER2 or TER2Δ against TER1.
5–7 day old seedlings were treated with cordycepin (100 ng/μl as a working concentration) for 2 h before RNA extraction. RNA was analyzed by qPCR normalized to eIF-4a [51]. RNA abundance was converted to the decreased level relative to untreated. RNA half-life was determined by the absolute value of inverse of the slope of the equation plotted by untreated and treated data. For the combined cordycepin/zeocin experiment, seedlings were pre-incubated with cordycepin for 1.5 h followed by zeocin and the incubation was continued for 2 h. RNA extraction and qPCR were used to determine RNA abundance. RNA half-life was determined by plotting RNA abundance versus time as described in [51].
3 kb of sequence upstream of the TER2 5’ terminus was cloned in a GUS reporter vector pMDC163. The construct was transformed into A. thaliana Col-0 and Ler-0 as described [52]. After selection in hygromycin, transformants seedlings were treated with zeocin for 2 h and then subjected to GUS histochemical staining as described [53].
|
10.1371/journal.pbio.1001133 | The Glial Regenerative Response to Central Nervous System Injury Is Enabled by Pros-Notch and Pros-NFκB Feedback | Organisms are structurally robust, as cells accommodate changes preserving structural integrity and function. The molecular mechanisms underlying structural robustness and plasticity are poorly understood, but can be investigated by probing how cells respond to injury. Injury to the CNS induces proliferation of enwrapping glia, leading to axonal re-enwrapment and partial functional recovery. This glial regenerative response is found across species, and may reflect a common underlying genetic mechanism. Here, we show that injury to the Drosophila larval CNS induces glial proliferation, and we uncover a gene network controlling this response. It consists of the mutual maintenance between the cell cycle inhibitor Prospero (Pros) and the cell cycle activators Notch and NFκB. Together they maintain glia in the brink of dividing, they enable glial proliferation following injury, and subsequently they exert negative feedback on cell division restoring cell cycle arrest. Pros also promotes glial differentiation, resolving vacuolization, enabling debris clearance and axonal enwrapment. Disruption of this gene network prevents repair and induces tumourigenesis. Using wound area measurements across genotypes and time-lapse recordings we show that when glial proliferation and glial differentiation are abolished, both the size of the glial wound and neuropile vacuolization increase. When glial proliferation and differentiation are enabled, glial wound size decreases and injury-induced apoptosis and vacuolization are prevented. The uncovered gene network promotes regeneration of the glial lesion and neuropile repair. In the unharmed animal, it is most likely a homeostatic mechanism for structural robustness. This gene network may be of relevance to mammalian glia to promote repair upon CNS injury or disease.
| The process of tissue regeneration has long been studied as a route to understanding what promotes structural robustness of cellular networks in animals. In the central nervous system (CNS), neurons and glia interact throughout adult life and during learning, at the same time accommodating functional changes while preserving the structural integrity necessary for function. The mechanisms that confer this combination of structural robustness and functional plasticity in the CNS are unknown, but they may be shared with the cellular responses to injury, which also require structural changes while retaining function. The glial cells that enwrap axons respond to injury by dividing and re-enwrapping them, leading to partial recovery of function. Here, we use Drosophila genetics to uncover a gene network underlying this glial regenerative response. This gene network enables glia to divide upon injury, prevent uncontrolled proliferation, and differentiate. We find that the network also has homeostatic properties: two cell-cycle activators (Notch and NFκB) promote the expression of a cell cycle inhibitor (Pros), providing negative feedback on cell division. Pros is also essential for glial differentiation, enabling the clearance of cellular debris and axonal enwrapment, and priming glia for further responses. By removing these genes or adding them in excess, we can shift the response to injury from prevention to promotion of lesion repair. This gene network is thus a homeostatic mechanism for structural robustness. Our findings from Drosophila may also help manipulation of glia to repair the damaged human CNS.
| The structure of organisms is robust. Cells accommodate changes in their environment during development and throughout life by adjusting cell number and cell morphology to preserve overall organismal integrity. In the central nervous system (CNS), adjustments are carried out in interacting populations of neurons and glial cells, from development to learning, ultimately enabling function. Injury and regeneration experiments and theoretical models have long been used to uncover the cellular and molecular mechanisms of how cells “sense” and maintain normal organ structure [1]. The premise is that shared mechanisms may underlie normal structural homeostasis and plasticity, and cellular responses to injury. Understanding such mechanisms is one of the frontiers in biology. It will also lead to a greater understanding of regeneration and repair of relevance to the treatment of human injury and disease. The fruitfly Drosophila is an ideal model organism for discovering gene networks, and has been successfully used to investigate cellular responses to CNS injury [2]–[6]. Here, we use Drosophila to uncover a gene network that controls the glial regenerative response to injury and promotes robustness in the normal CNS.
Previous experiments had revealed two important findings about glial responses to injury in Drosophila. Firstly, enwrapping glial cells become phagocytic upon injury clearing cellular debris. This phagocytic function requires the corpse engulfment receptor Draper, which is specifically expressed in enwrapping glia, and whose function involves Simu, Src42A, and Shark [4],[6]–[9]. Secondly, stabbing injury in the adult head [5] and neuronal ablation in the embryonic Ventral Nerve Cord (VNC) [10] induce the proliferation of glial cells (including the enwrapping Longitudinal Glia). Similar findings had long before been observed in the cockroach [11]–[13]: surgical lesioning and chemical ablation of enwrapping glial cells induced cell transformation leading to the phagocytosis of cellular debris, and most remarkably glial proliferation. This restored glial numbers, enwrapment, and normal electrophysiological function.
Insect glia may be evolutionarily distant from mammalian glia, but it can be insightful to compare them. Injury induces distinct responses in mammalian CNS glial cells. Astrocytes normally maintain ionic homeostasis, provide nutrients, and participate in synapses. Microglia are the immune cells of the brain, and are normally in a resting state. Upon injury, microglia and astrocytes phagocytose degenerating axons and other cellular debris, and they can also form a glial scar that inhibits axonal regeneration [14]. Ensheathing glia (oligodendrocytes) normally myelinate axons for saltatory conduction in vertebrates, maintain ionic homeostasis, and provide metabolic and trophic support to axons [15],[16]. Oligodendrocyte progenitor cells (OPCs) respond to injury by dividing, and resulting oligodendrocytes remyelinate [17]–[19]. This latter response is regenerative, leading to spontaneous re-enwrapment of axons and partial functional recovery, for instance of locomotion [20]–[22]. Conditions such as spinal cord injury, stroke, and multiple sclerosis induce the proliferation of OPCs resulting in spontaneous remyelination of CNS axons, and underlie the “remission” phases of multiple sclerosis [18],[23]–[28].
Thus, the regenerative response of ensheathing glia occurs across the animals, from insects [11] to fish [23] and humans [20]. In cockroach, fruitflies, and vertebrates, ensheathing glia proliferate upon injury, and both in insects and mammals this response can lead to limited remyelination and some recovery of function. This reveals that there is an endogenous tendency of the CNS to repair itself. Its manifestation across species may reflect a common underlying gene network. If understood, it could be harnessed to stimulate CNS repair. Here, we search for a Glial Regenerative Response (GRR) gene network that can promote repair after injury and confer structural robustness in the normal animal.
The following factors are promising candidates to belong to this gene network. The Drosophila TNF super-family member Eiger triggers the proliferation of adult brain glia upon injury in fruitflies [5]. TNFα also triggers the proliferation of mammalian oligodendrocytes progenitors through its receptor TNFR2 upon injury [29]. While in other contexts TNFR2 is thought to function by activating NFκB [30] (which can promote the cell cycle), whether this is the case for CNS glial cells and whether this activates glial proliferation are unknown.
Notch maintains the undifferentiated and stem cell state in many contexts [31]. In Drosophila, Notch maintains the mitotic potential of embryonic ensheathing glia in interaction with the Jagged1 homologue, Serrate, from axons [10],[32]. Similarly, in vertebrates, Notch1 maintains the oligodendrocyte progenitor state by interacting with its ligand Jagged1 present in axons [33]. However, the functions of Notch in the glial regenerative response remain unsolved. Notch1 is present in adult NG2+ OPCs, and it is upregulated upon injury and during regeneration, but conditional Notch1 knock-out in OPCs does not prevent the regenerative response [34],[35]. Notch1 can inhibit the differentiation of progenitors into myelinating oligodendrocytes, preventing repair [33],[36], but presence of Notch1 signaling in these cells does not prevent the regenerative response either [34],[35]. Finding out how to control Notch1 function, to enable glial proliferation, and to subsequently promote ensheathing glial differentiation is thus a key issue.
The transcription factor Prospero (Pros) interacts with Notch in ensheathing glia in Drosophila embryos [10], but how Pros and Notch affect each other's function is not understood. Pros appears to have opposite functions in neuroblasts and in glia. In neuroblasts Pros is a tumour suppressor, as mutations in pros result in over-proliferation [37]–[40]. Instead in glia both Pros and Notch are necessary for proliferation, and mutations in pros do not result in glial hyperplasia [10],[32]. Unravelling the relationship between Notch and Pros may hold the key to understanding how glial proliferation and differentiation are regulated.
Here, we uncover the Glial Regenerative Response gene network in Drosophila. For this, we establish a new CNS injury paradigm in the larval VNC. We use the larva because it is more accessible than the adult while it has locomotion, senses, learning, and memory, enabling the investigation of repair in the context of a fully functional CNS. We show that the ensheathing Interface Glia of the VNC respond to injury by phagocytosing and clearing cellular debris and by dividing. We reveal a gene network that controls the balance of glial proliferation and differentiation, and it is comprised of two feedback loops: one involving Pros and Notch, and a second involving Pros and Dorsal/NFκB, connected via Eiger/TNF and Wgn/TNFR signaling. By manipulating this gene network we could shift from exacerbating the damage to promoting repair of the damaged neuropile. The uncovered gene network is a homeostatic mechanism for structural robustness and plasticity.
To test how Drosophila larval glia respond to injury, dissected ventral nerve cords (VNCs) were stabbed with a fine needle and cultured (Figure 1A). Stabbing was applied dorsally into the neuropile, which comprises the bundles of CNS axons and Interface Glia (IG) [41]. To obtain an overview of the injury response, we performed a time-lapse analysis. Axons were visualised with GFP driven by the protein trap line G9, and glia with repoGAL4>DsRed (Figure 1B, Video S1 and Video S2). The wound initially expanded, and vacuoles formed within the neuropile. However, after 6 h of culture, glial processes invaded the wound, the axonal and glial wound began to shrink, and by 22 h the wound could heal considerably. This suggested that there is a natural mechanism that can promote repair. Here (Figure 1C), we (1) characterised the glial responses to injury, (2) investigated the gene network controlling the regenerative potential of glia, and (3) tested whether altering the functions of this gene network could promote glial regeneration and axonal repair.
To characterise the glial responses to injury, glial membranes were visualised with repoGAL4>mCD8-GFP and Interface Glia (IG) with anti-Glutamine Synthetase 2 (GS2) (Figure 2A,B and Figure S1). This revealed the stabbing wound dorsally and an indentation ventrally (Figure S1, Video S3 and Video S4). Although stabbing damaged to some extent surface and cortex glia (unpublished data), it affected most prominently the IG, causing GS2+ glial loss (Figure 2A,B and Figure S1) and GS2+ glial debris at 6 h after injury (Figure 2B). We used the alrmGAL4 driver, which is restricted to the Pros+ IG (Figure S2A), to visualize IG nuclei with HistoneYFP, and this showed that some IG were lost through apoptosis (Figure 2C, intact VNCs had 0 cleaved-Caspase-3+ YFP+ IG n = 10 VNCs versus stabbed VNCs with an average of 1.5 cleaved-Caspase-3+ YFP+ in 57% of the VNCs n = 14). In remaining IG, injury provoked an increase in the size and complexity of cytoplasmic projections (seen with the membrane reporter mCD8GFP, Figure 2D). As observed in time-lapse, injury led to vacuolization of the neuropile, as holes formed within the axonal bundles (Figure 1B and Figure 2E). IG projections enveloped these vacuoles (Figure 2E).
To further characterise these aspects, we used Transmission Electron Microscopy (TEM). In wild-type wandering larvae, the IG nuclei were located outside the neuropile and their cytoplasms enwrapped the entire neuropile (Figure 3A). IG processes projected into the neuropile, where they could enwrap smaller axonal bundles (Figure 3B) and individual axons (Figure 3C). As seen with confocal microscopy, TEM confirmed that injury caused glial loss with breakdown of neuropile enwrapment after 6 h (Figure 3D), and vacuoles formed within the neuropile (Figure 3D). Some Interface Glia degenerated via necrosis as seen by swelling of mitochondria (Figure 3E). Remaining IG expanded their cytoplasmic projections both around and within the neuropile (Figure 3F–J), which was never observed in intact specimens. IG processes lined vacuoles (Figure 3F,G), phagocytosed axonal fragments and other cellular debris, as revealed by phagosomes and multilamellar bodies within the glial processes (Figure 3H–J). IG processes frequently wrapped around isolated axons that could be degenerating (Figure 3K–N). These data show that upon injury IG phagocytose cellular debris, presumably clearing the lesion. Altogether, these data demonstrate that larval IG enwrap CNS axons, and that they are damaged by, and respond to, injury.
Next, we investigate if stabbing induced IG proliferation. Based on their location, the IG are classified into dorsal (dIG), lateral (lIG), and ventral (vIG) IG (Figure 4A). They are identified by co-localisation of the pan-glial marker Repo and the transcription factor Pros in nuclei, surrounded by the cytoplasmic marker Ebony (Figure S2B,C). IG do not normally divide [42], but are arrested in G1 and have mitotic potential (Text S1 and Figure S3). To examine cell proliferation, we used PCNA-GFP, a reporter with E2F binding sites that reveals GFP expression when cells go through S-phase (Figure 4B,C,D) [43]. PCNA-GFP+ Pros+ Ebony+ IG were rarely seen in non-stabbed controls, but stabbing increased their frequency at the lesion site (Figure 4B) and throughout the neuropile (Figure 4C). We did not find any Ebony-negative IG with PCNA-GFP (unpublished data), suggesting that Ebony+ Pros+ IG are the only IG that divide in response to injury. Normally there is one Ebony+ vIG per hemisegment, but the number of Ebony+ vIG adjacent to the wound increased significantly in stabbed larvae (Figure 4E,F). Altogether these data show that stabbing causes a local increase in proliferation of Pros+ Ebony+ IG at the lesion site.
A BrdU pulse experiment (a commonly used method to visualise proliferating cells) also revealed an overall increase in the number of dividing IG upon stabbing, to 50% of Ebony+ IG being also BrdU+ (p<0.05), comprising local IG at the lesions site (Figure S4) and at some distance along the VNC. This suggested that neuronal damage may also affect glial cells at a distance from the original lesion, which is explained as axons extend along the whole length of the VNC. Stabbing may also affect other glial classes than IG. To take these facts into account, we purposely developed DeadEasy Glia software to automatically count in vivo all Repo+ glial cells (Figure 4G and Figure S5). After 22 h culture, stabbed VNCs had more glial cells than non-stabbed controls (Figure 4G). This effect was abolished when a cell cycle inhibitor—constitutively active Retinoblastoma protein factor (Rbf280/Rb)—was expressed in glia (Figure 4G). This demonstrates that the increase in glial number upon stabbing was due to the induction of glial proliferation. These data show that stabbing the larval VNC causes an increase in glial proliferation and a consequent increase in glial cell number. Although we cannot rule out that other glial cells might also divide, our data demonstrate that this response involves the IG.
We next asked what genes might control the proliferative glial response to injury. Notch and Pros regulate the mitotic potential of embryonic glia [10]; thus, we wondered if they might be involved. prosvoila1/prosS044116 hypomorphic mutants specifically affected larvae, since embryogenesis proceeded normally but the levels of Pros dropped in IG by the third instar larval stage (Figure S6A,B). In prosvoila1/prosS044116 VNCs, Ebony was downregulated, meaning that Ebony is a downstream target of Pros (Figure S6B), but there were no major developmental defects as Repo and GS2 expression were normal (Figure S6C,D). Expression of the Notch antagonist numb with repoGAL4 to knockdown Notch specifically in glia did not cause general developmental defects either (see below). However, the glial proliferative response to injury was significantly reduced both upon the glial over-expression of numb (Figure 5A) and in prosvoila1/prosS044116 mutant larvae (Figure 5A). In particular, IG number decreased upon stabbing in prosS044116 mutant larvae (Figure S7 p<0.01). These data show that Notch and Pros are required for the glial proliferative response.
Drosophila Egr/TNF is required for glial proliferation in response to injury in the adult brain [5], but how it implements this is unknown. In mammals, TNF can induce cell proliferation via the activation of NFkB [30], but whether it does in glial progenitors upon CNS injury is unknown. Thus we asked whether in our injury paradigm, IG proliferation required Egr/TNF, Wengen (Wgn)/TNFR, and Drosophila NFκB, Dorsal. egr/TNF and wgn/TNFR are expressed in the VNC (Figure S8A–D) and Dorsal/NFkB is distributed preferentially in Ebony+ pros-lacZ+ IG (Figure S8E). There were no major developmental defects in the VNC of dorsalH/dorsal1 or egr1/egr3 mutant larvae (Figure S9). However, the glial proliferative response was abolished in stabbed egr1 (unpublished data), wgne00637/Df(1)Exel7463, and dorsal1/dorsalH mutant larval VNCs (Figure 5A). These data suggest that Wgn/TNFR, its ligand Egr/TNF, and Dorsal/NFκB are required for the glial proliferative response to injury.
To verify this, we asked whether stabbing resulted in the activation of Dorsal/NFκB in glia. In its inactive form, Dorsal/NFκB is cytoplasmic, and upon signaling it is translocated to the nucleus to function as a transcription factor. We found intense nuclear distribution of Dorsal/NFκB in IG upon injury in wild-type (Figure 5C p<0.05), but not in egr1 mutants (Figure 5C). This shows that stabbing induces the activation of Dorsal/NFκB in IG, which depends on Egr/TNF (Figure 5B).
Therefore, we sought to find out how might Pros, Notch, Eiger/TNF, and Dorsal/NFκB implement their functions in the glial proliferative response to injury.
To investigate the function of Pros in glial proliferation, we generated prosJ013 null mutant MARCM clones in larval glia. The number of IG in prosJ013 mutant clones (1–8 cells per clone in 8 clones generated in n = 786 VNCs) did not differ from the number of IG in wild-type clones (1–9 cells per clone in 15 clones in n = 1,254 VNCs). Furthermore, in wandering larvae (120 h AEL) the number of glial cells in prosvoila1/prosS044116 mutants was indistinguishable from wild-type (Figure 6A,D). These data demonstrate that Pros does not affect the extent of glial proliferation in the normal, non-stabbed larva. However, loss of pros function affected the timing of glial cell division (Figure S10A,B). In younger (96 h AEL) prosvoila1/prosS044116 mutant larvae, there were more glial cells than in wild-type (Figure S10A), implying that the excess glial cells arose from faster (but not more) cell divisions. Cell division is speeded up by shortening the G1 phase, for instance with the up-regulation of CycE. Consistently, Pros activates the expression of the CycE repressor Dacapo (the p21/p27 homologue) in glia (Figure S10C p<0.05). To further test if Pros can halt larval glial proliferation, we over-expressed pros in larval glia using tubGAL80ts;repoGAL4. This resulted in early larval lethality, and escapers had decreased glial number compared to controls (Figure 6G), showing that Pros inhibits glial proliferation. Altogether, our data show that Pros functions as a repressor of cycE in glia and it inhibits cell cycle progression by keeping glia arrested in G1 (Figures 6B).
If Pros inhibits cell division, why isn't there glial hyperplasia in the mutants? And why can't pros mutant glia proliferate upon injury? To solve this conundrum, we wondered if Pros might interact with Notch. Notch signalling is present in larval IG, its ligand Serrate is in axons (Figure S11A), and Notch maintains the expression of pros. Pros is also required for Notch signalling (Figure S11B–G), like in embryonic LG [10]. Thus, Notch and Pros maintain each other in IG. In other contexts, Notch promotes cell division by regulating the G1/S transition (Figure S10H) [43],[44]. We found that constitutive activation of Notch signalling—expressing the Notch Intra Cellular Domain (NotchICD) —in all glia increased both total glial number (Figure 6A,Ai,D) and the number of Ebony+ IG (Figure 6E). Activation of Notch restricted to the IG only also resulted in an increase in IG cell number (Figure 6F). Consistently, transient activation of Notch signalling induced PNCA-GFP expression (Figure S10D,E) and BrdU incorporation (Figure S10F,G) specifically in IG. These data show that Notch can promote glial cell division.
So if Notch signalling is normally activated in IG, why don't they divide in the intact larva? Our data show that Pros and Notch have antagonistic functions in the control of glial proliferation. Since they also maintain each other, a “tug of war” between Notch and Pros is likely to keep IG in cell cycle arrest. To test this, we asked whether cell cycle arrest could be evaded by interfering with this feedback loop. Over-expression of NotchICD in glia resulted in the up-regulation of Pros (Figure S11E), which would repress cycE expression. When we expressed cycE together with NotchICD, this increased glial number and expanded VNC size (Figure 6A,D). Over-expressing NotchICD in glia in prosS044116 mutant larvae further increased glial number, causing a tumourous expansion of the VNC (Figure 6A,Ai,B). The increase in abdominal VNC size was not due to a non-autonomous effect on neuroblast proliferation or increased neuronal number (Text S2 and Figure S12 and Figure S13), but to increased glial divisions (Figure 6C). Altogether, our data show that Notch promotes cell cycle progression in glia while Pros inhibits it, and positive feedback between Notch and Pros counterbalances the effects of each other, maintaining glial cells on the brink of dividing (Figure 6B and Figure S10H). Interfering with this feedback loop has dramatic consequences in glial number and VNC size.
To find out whether Notch and Pros influence IG differentiation, we visualised IG morphology using alrm>mCD8GFP upon loss or gain of function for each of these genes. To knockdown Notch function only in larvae, we used a temperature sensitive allele of Notch—Notchts1. In Notchts1 mutant larvae, IG filopodia and lamellipodia are thinner than in wild-type controls (Figure 7A). Conversely, over-expression of NotchICD in glia results in larger and rounder glial cells (Figure 7A). In hypomorphic prosS044116 mutant larvae, IG hardly developed cytoplasmic projections (Figure 7B). Conversely, over-expression of pros in glia induced more elaborate IG projections (Figure 7B). These findings show that Notch and Pros have opposite effects on glial differentiation.
To further test how loss of pros function affects IG differentiation, we analysed MARCM clones of prosJ013 null mutant IG: glial morphology was aberrant, with dramatic loss of glial projections compared to wild-type (Figure 7C,D). The glial differentiation markers Ebony and GS2 were also influenced by Pros. Ebony is a glial enzyme involved in neurotransmitter recycling [45]–[48], and it was down-regulated in prosvoila1/prosS044116 mutants (Figure 7E and Figure S6B). GS2 is an enzyme involved in Glutamate recycling normally restricted to enwrapping glia [49]. Larval over-expression of pros with tubGAL80ts; repoGAL4 induced its ectopic expression in non-enwrapping glia (Figure 7F). Over-expression of NotchICD did not induce Pros, Ebony, or GS2 expression in non-enwrapping glia (Figure S14). Thus, GS2 and Ebony are directly regulated by Pros but not by Notch. Altogether, these findings demonstrate that Pros controls IG differentiation.
We have shown above that the glial proliferative response to injury is abolished in Egr/TNF, Wgn/TNFR, and Dorsal/NFκB mutants. The number of Repo+ glia, as well as the expression of GS2 and Ebony, were normal in egr1 and dlH/dl1 mutant wandering larvae (Figure 8A,C and Figure S9), meaning that the glial functions of Egr/TNF and Dorsal/NFκB are dormant in the normal, non-stabbed larva.
To investigate if activation of Dorsal/NFkB could promote glial proliferation, we over-expressed dTRAF2 in all glia. The Drosophila TRAF6 homologue dTRAF2 binds Wgn/TNFR and induces the nuclear translocation of NFκB homologues [50]–[52]. When dTRAF2 was expressed in glia, the number of glial cells increased (Figure 8A,C). This effect was rescued by expressing dTRAF2 in a dorsalH/dorsal1 mutant background (Figure 8C), showing that the effect of dTRAF2 is mediated by Dorsal. Activation of Dorsal/NFκB by expressing dTRAF2 in glia resulted in an increased number of Ebony+ IG (p<0.01). Temporal over-expression of dTRAF2 also induced BrdU incorporation in Ebony+ IG, showing that it activated mitosis cell-autonomously (Figure S15). These data show that activation of Dorsal by dTRAF2 promotes glial proliferation. Glial number and VNC size increased further upon glial expression of dTRAF2 in pros mutants (UASdTraf2;repoGAL4 prosS044116/prosS044116 Figure 8A,C), indicating that Pros antagonizes the proliferative function of Dorsal/NFkB (Figure 8E).
Our data show that Pros-Notch feedback keeps glia on the brink of dividing, and upon injury Egr/TNF signalling via dTRAF2 activates Dorsal/NFκB, tipping the balance towards cell division (Figure 8E). Thus we asked whether these two genetic mechanisms are linked. We found that in prosvoila1/prosS044116 mutant larvae Dorsal is decreased from IG (Figure 8B), suggesting that Pros is required for dorsal/NFκB expression. Since the glial response to injury critically depends on Dorsal/NFkB, this means that the ability of IG to respond to injury is regulated by Pros (Figure 8E).
The glial regenerative response is constrained, as it induces glial proliferation but not tumours, indicating that cell cycle arrest is restored in daughter cells. Tumorous-like over-growth was induced by expressing dTRAF2 in glia in pros mutants (Figure 8A,C), as was the case when expressing NotchICD in pros mutants (Figure 6A,Ai,D). This suggested that Dorsal/NFκB might activate pros expression restoring arrest. To test this, we used hypomorphic prosS044116 mutant larvae that still produce Pros at low levels in a few glial cells. Expression of dTRAF2 in glia in prosS044116 mutant larvae resulted in the up-regulation of Ebony and Pros (Figure 8D). These data show that Dorsal/NFκB activates pros expression in glia. Since Pros inhibits cell cycle progression whereas Dorsal/NFκB promotes it, the “tug of war” between Pros and Dorsal/NFkB is likely to restore G1 arrest in the daughter cells (Figure 8E).
Thus, we have shown that a gene network involving Notch, Pros, TNF, and NFκB controls the balance between glial proliferation, arrest, and differentiation.
To test whether manipulating this gene network was regenerative to enwrapping glia, we examined the glial wound. Stabbing disrupted the GS2+ Ebony+ glial mesh in the neuropile, and the area devoid of these markers was measured (Figure 9A). In egr1; prosvoila1/prosS04416 double mutant larvae, in which the proliferative glial response and glial differentiation were both affected, the glial wound increased significantly compared to controls (Figure 9C). In larvae expressing NotchICD in glia, resulting in over-proliferation, the glial wound was consistently significantly smaller than in controls (Figure 9A,D). This indicates that either Notch itself or increased glial number is regenerative. We showed above that over-expression of NotchICD in glia also induced pros expression, and that Pros promoted glial differentiation. Thus we asked whether the regenerative function of Notch relied on Pros. When stabbing was carried out in larvae that over-expressed NotchICD in glia but were also mutant for pros (repoGAL4 prosS044116/UASNotchICDprosS044116), wound size increased significantly (Figure 9D). Since glial cells proliferated in excess in this genotype (Figure 6A,D), this means that glial proliferation alone is not sufficient for repair and glial differentiation is also required.
To test what consequence the uncovered gene network might have on neuropile repair, we examined cell death levels. Apoptotic cells were visualized with anti-cleaved-Caspase3 antibodies, and counted in vivo automatically using purposely adapted DeadEasy Caspase software [53]. Injury increased the extent of apoptosis over non-injured controls (Figure 9B,E). Expression of NotchICD in glia did not rescue baseline apoptosis, but it rescued injury-induced apoptosis (Figure 9B,E). This suggests that either NotchICD itself or the resulting increase in glial cell number is protective upon injury.
To test the effect of these genes in enwrapment, we used TEM. Over-expression of NotchICD in glia dramatically increased glial projections and axonal enwrapment (Figure 9F). However, enwrapment was reduced when NotchICD was over-expressed in pros mutant larvae (Figure 9F), indicating that Pros is required for enwrapment. Altogether, these data show that the glial response is regenerative, that both glial proliferation and differentiation are necessary for glial regeneration, and that Notch and Pros play central roles.
To test what effects the glial regenerative response (GRR) may have on the axonal bundles, we carried out time-lapse recordings of stabbed larval VNCs, with glial cells labeled with repoGAL4> or alrmGAL4>UAS-DsRed, and all axons labeled with the GFP-protein-trap line G9. In wild-type larvae the neuropile wound first increased in size, and numerous vacuoles formed, consistently with TEM and confocal microscopy data from fixed samples (Figure 10A,G,H and Video S1, refer also to Figure 2E and Figure 3D,F,G). Subsequently, the vacuoles might disappear and the wound might shrink (Figure 10B,F,G,H and Video S2). In Notchts1 mutant larvae, wound size in the axonal neuropile was considerably larger, had greater vacuolization than controls, and did not decrease over time (Figure 10C,G and Video S5). Similarly, when glial proliferation was prevented by over-expressing pros in IG, wound size and vacuolization were also more extensive than in controls (Figure 10D,H and Video S6). Conversely, when glial cell proliferation was increased by over-expressing NotchICD, wound enlargement and vacuolization were considerably constrained, wound size decreased over time, and even repaired (Figure 10E,H and Video S7). In both wild-type and upon over-expression of NotchICD there was a correlation between repair and presence of DsRed+ glial processes within or around the vacuoles and in areas of axonal damage (Figure 10F and Video S2, Video S7). Together with the TEM data (Figure 3D–N), the time-lapse data suggest that upon injury, glial processes engulf the vacuoles, and phagocytose axonal fragments and other cellular debris, contributing to repair. Altogether, our data show that Notch and Pros control glial proliferation and differentiation required for glial regeneration and debris clearance, and this enables neuropile repair.
We present here a regenerative response of Interface Glial cells to stabbing injury in the Drosophila CNS, we have identified an underlying gene network, and we show that it can promote repair. This gene network controls the balance between glial proliferation, arrest, and differentiation, and it promotes repair by “hitch-hiking” on a developmental mechanism for structural robustness.
Stabbing injury in normal larval VNCs caused an initial loss of IG, wound expansion, and neuropile vacuolization. Ensheathing glia extended large processes within the neuropile, phagocytosed axonal fragments and cellular debris and dissolved vacuoles, some remaining glial cells divided, and neuropile integrity could be restored. This natural mechanism was enhanced by activating Notch signalling in glia in the presence of Pros. Together, NotchICD and Pros prevented wound enlargement and vacuolization, they prevented injury induced apoptosis, increased ensheathing glial number, and promoted glial regeneration and axonal neuropile repair (Figure 11A). Remarkably, the stabbing injury wound could be completely repaired in these larvae. This was achieved through the balance of glial proliferation and differentiation under the control of Notch and Pros (Figure 11B). NotchICD promotes glial proliferation and Pros promotes their differentiation. In the normal intact larva, the balance between NotchICD and Pros keeps IG in the brink of dividing. Pros also promotes the expression of cytoplasmic NFκB and of the glial differentiation factors Ebony and GS2. Upon injury NFκB shuttles to the nucleus increasing the relative levels of cell cycle activators, and glia divide. NFκB and NotchICD activate Pros expression, and as Pros levels rise, Pros halts further glial cell division and promotes glial differentiation. Pros also promotes Notch expression, thus restoring the original balance. We have shown that interfering with the functions of these genes prevents repair (Figure 11C). When both glial proliferation and differentiation were inhibited as in egr-pros- double mutant larvae, the glial wound enlarged. When glial proliferation was abolished in Notchts mutants or upon over-expression of pros in glia, the glial and neuropile wounds enlarged and vacuolisation increased. Conversely, increasing glial proliferation by activating Notch signaling promoted glial regeneration. The regenerative effect of Notch not only relied on the increase in glial cell number, but also on pros. Glial regeneration was prevented if NotchICD was expressed in a pros mutant background. Pros promotes IG differentiation, increasing the complexity of cytoplasmic processes and promoting axonal enwrapment. In the absence of Pros glia have fewer filopodia and lamellipodia, and downregulate the glial differentiation marker Ebony—an enzyme involved in the recycling of neurotransmitters [45]–[48]. Conversely, upon over-expression of pros, glia have more processes and up-regulate the expression of the enwrapping glial marker GS2—and enzyme involved in the recycling of Glutamate [49]. The control of glial differentiation by Pros is conceivably required for glia to phagocytose and clear cellular debris, restore neurotransmitter homeostasis, and re-enwrap the neuropile and axons. Thus, the uncovered gene network is regenerative and Pros is the critical link in the control of glial proliferation and differentiation that enables repair.
The identified gene network enables the regenerative response to injury in the following way. (1) In the normal larva it maintains glial cells arrested with mitotic potential, enabling them to respond to injury (Figure 11D). Generally, a cell that has exited the cell cycle cannot divide again. Pros and Notch together prevent cell cycle exit and maintain glia in a proliferative yet arrested state. That is, the IG can divide, but do not normally do so. This state is achieved as Pros and Notch maintain each other but have antagonistic functions on the cell cycle. Pros prevents cell cycle progression by repressing cycE, and Notch promotes the G1/S transition. Their mutual maintenance counterbalances their effects on the cell cycle, and maintains glial cells on the brink of dividing. (2) The gene network enables IG to respond to injury by proliferating at the lesion site (Figure 11E). This is achieved as Pros regulates the expression of dorsal/NFκB. Dorsal/NFκB is a transcription factor located in the cytoplasm in its inactive form. Upon injury, the pro-inflammatory cytokine Egr/TNF via its receptor Wgn/TNFR induces the translocation of Dorsal/NFκB to the nucleus, where it promotes cell cycle progression. This breaks the balance of the Pros-Notch loop pushing glia to divide. (3) The gene network restores cell cycle arrest, preventing uncontrolled proliferation (Figure 11F). Dorsal/NFκB activates the expression of pros, which inhibits cell cycle progression. Presumably as the total input from cell cycle activators (Notch and NFκB) and inhibitor (Pros) balances out, it restores cell cycle arrest. The antagonistic function of Pros versus Notch and Dorsal/NFκB, and their mutual dependence, restricts cell proliferation. Overgrowth is induced when negative feedback breaks down, upon activation of Notch or Dorsal/NFκB in the absence of Pros. Thus the GRR gene network prevents tumourous overgrowth. (4) The gene network controls glial differentiation, which critically depends on Pros. We have shown that in the absence of Pros, IG have reduced processes, and wound size and vacuolization enlarge. This would indicate that Pros may be required for the phagocytic response of glia and lesion clearance. We have also shown that Pros is required for axonal enwrapment, an end point of repair. Thus the gene network is a homeostatic cycle by which injury triggers a response in glia that not only repairs the wound but also primes the restored glia to respond to further injury or smaller changes. This would suggest that there must be a mechanistic link between Pros and the corpse engulfment pathway of Simu, Draper, Src42A, and Shark [4],[6]–[9]. While our data show that the IG underlie the regenerative response, we cannot rule out that other glial classes might also be involved. For instance, cortex glia have also been reported to be phagocytic and activate Draper [8]. It will be interesting to explore these uncovered research avenues in the future.
Drosophila Pros and the mammalian homologue Prox1 have different effects on cell proliferation in different cell lineages [10],[37]–[40],[54]–[56]. In Drosophila ganglion mother cells, Pros promotes cell cycle exit by repressing cycE expression, and is a tumour suppressor [37]–[40]. However, in Drosophila embryonic glia Pros enables cell division [10], but how this might occur remained unexplained. Our findings demonstrate that Pros functions as a repressor in cycE also in glia. However, in IG loss of pros does not result in hyperplasia because as Pros maintains Notch signaling, loss of pros leads to loss of Notch signaling, consequently reducing cell cycle activation. Although glial cell division initially occurs faster in pros mutants as the G1 phase shortens upon the de-repression of cycE, cell division soon stops due to the loss of Notch. The diverse outcomes of Pros function depend on cell type specific gene networks.
We have shown that in normal larvae the IG do not divide and that loss of function mutations in the gene network genes can result in normal glial cell number and distribution of glial markers. It would appear that these glial gene functions are uncovered upon injury. However, since in the wild CNS injury most likely results in fruitfly death, this raises the intriguing question of what might this gene network be for in the non-injured fruitfly. The fact that the glial regenerative response is also found in other insects, fish, and humans might imply an underlying common genetic mechanism, but why should it be?
Our findings suggest that the GRR is a homeostatic mechanism that promotes structural robustness in the non-injured animal. Homeostasis is grounded in two features of the feedback loops. Firstly, both feedback loops result in negative feedback on cell proliferation. Injury initially causes cell loss, and glial proliferation and differentiation are subsequently induced followed by cell cycle arrest, restoring cell number and enwrapment. This is achieved as the cell cycle activators (Notch and NFκB) induce the expression of a cell cycle inhibitor (Pros). Thus, although glial cells divide initially, as more activator protein is produced, more inhibitor is produced too, restoring normal cell number but halting further cell division (Figure 11C). This homeostatic control enables glial proliferation for repair while preventing excess, which would result in tumours.
Secondly, the two feedback loops limit the amounts of cell cycle regulators. The mutual maintenance between Pros and Notch, and Pros and NFκB would result in their levels forever rising. Instead, the two positive feedback loops are constrained, in different ways. Pros-Notch feedback is established spatially through interactions between enwrapping glia and the axons that express the Notch ligand Ser. Pros is a transcription factor and can directly activate the expression of Notch. However, Notch only functions as a transcription factor after it has been cleaved at the membrane as NotchICD, which then translocates to the nucleus. Thus positive feedback only takes place when Notch contacts Ser in neighbouring axons. If contacts with Ser are saturated, NotchICD is not processed and cannot activate Pros. In this way the Pros-Notch loop is stabilized relative to the amount of Ser in axon-glia contacts.
Pros-NFκB feedback is constrained through time, in response to changes in the cellular environment. Although NFκB is a transcription factor, it can only activate Pros expression when it translocates to the nucleus. In the uninjured glia, NFκB is trapped in the cytoplasm and cannot activate pros. In this way, Pros-NFκB positive feedback is frozen in time and released only in an injury event.
Following injury, the homeostatic feedback loops restore initial conditions. Injury is likely to compromise contact between axons and glia, perhaps causing an initial drop in Notch signalling in glia, which would consequently down-regulate Pros levels. But injury also triggers the nuclear shuttling of NFκB, inducing cell proliferation and pros expression. As Pros levels rise, it halts further cell division and promotes glial differentiation and the expression of Notch. As glial differentiation restores neuron-glia interactions, this activates Notch signalling in glia, re-establishing the Pros-Notch loop, cell cycle arrest and priming glia for future responses to injury, thus closing the cycle.
In normal, non-injured larvae, the IG divide sporadically, and these divisions may represent homeostatic adjustments in glial number in response to cellular changes due to genetic variability or exogenous influences (e.g., changes in temperature). This would help clear debris and neurotransmitters, maintain axonal enwrapment and ionic homeostasis, modulate axonal growth and fasciculation, and provide trophic support to neurons. It would maintain enwrapment of the neuropile preserving architecture. This adjustment could result from fluctuations in axon-glia interactions that altered the relative levels of NotchICD and Pros. We propose that the normal function of the GRR gene network is the homeostatic regulation of glial proliferation and differentiation to provide structural robustness. The glial regenerative response to injury may “hitch-hike” on this developmental mechanism to restore structural integrity. This would explain why a gene network underlies these events and how it emerged in the course of evolution—as a mechanism that confers robustness, reproducibility, and reliability to CNS structure.
We have shown that IG can enwrap the neuropile, axonal bundles, and individual axons. Similar glia in insects have been compared to mammalian oligodendrocytes [57],[58]. However, IG do not form nodes of Ranvier, thus resembling non-myelinating enwrapping glia, such as Remak glia in the peripheral nervous system (PNS) and olfactory ensheathing glia of the CNS [15]. Drosophila IG also express molecules involved in neurotransmitter recycling such as Ebony and Glutamate synthetase (GS2). In vertebrates, neurotransmitter reuptake is mostly carried out by astrocytes in the CNS and Schwann cells in the PNS [59]. NG2+ OPCs and oligodendrocytes also express molecules involved in glutamate recycling, including glutamate synthetase, at non-synaptic sites [60]–[62]. Like microglia and astrocytes in mammals, and Drosophila ensheathing glia in other contexts [63]–[65], Drosophila IG are phagocytic and engulf axonal and other debris [66]. And like Schwann cells, Drosophila Pros+ IG can function as differentiated cells but can also divide. Thus, Drosophila IG cells are neuropile glia that behave like, and carry out multiple functions attributed to, distinct glial types in mammals.
The injury progression we observed—wound expansion, followed by debris clearance, glial regeneration, and neuropile repair—reproduces that documented for insect [12],[13] and mammalian CNS injury [67]. The molecular mechanisms underlying debris clearance have been little explored in mammals [68]. Although astrocytes proliferate to some extent upon injury [18],[19], NG2+ OPCs are the prominent cell type to divide [18],[19],[69]. Replenishing ensheathing glia promotes axonal regrowth and insulation, and protects against axonal degeneration and neuronal death. Thus, transplantation of enwrapping glial progenitors is a relevant strategy for the therapeutic treatment of spinal cord injury and demyelinating diseases [20],[22]. A critical aim has been to identify a gene interacting with Notch1 that will enable differentiation of progenitors into enwrapping oligodendrocytes. Our Drosophila findings have revealed shared functions of Notch between insect IG and NG2+ OPCs: as with IG, Notch1 maintains the mitotic state of OPCs [33]; it is present in OPCs that divide upon injury [34],[35]; and it prevents oligodendrocyte differentiation [70]. We show here that Pros antagonizes Notch function and it has a critical role inducing glial differentiation in fruitflies. The vertebrate homologue Prox1 can promote cell cycle exit and differentiation [54],[71] and antagonize Notch1 function in mammalian neural stem cells [72]. If a gene network similar to that uncovered here operates in human glial progenitors, its manipulation may facilitate CNS repair. The GRR gene network should also bring insights into the understanding of glioma, as Notch, NFκB, and cycE are hyper-activated in human gliomas [73],[74].
Finally, like in Drosophila, the response to spinal cord injury by mammalian OPCs recapitulates developmental events [69]. The physiological function of enwrapping glial plasticity in the adult may be to promote re-enwrapment following focal loss and to modulate myelination also during learning [15],[75]. Similarly, the clearance of axonal degeneration by glia during circuit remodeling shares mechanisms with that after injury [4],[66]. Accordingly, the GRR gene network may be a common, homeostatic mechanism for structural robustness and plasticity.
Supporting Information is linked to the online version of the article. Supporting Information comprises Figures S1 to S15, Table S1 with statiscal analyses, Videos S1 to S7 and TextS1 and Text S2 with data, and Text S3 with detailed methods.
Details on methods can be found in Text S3. A summary is given below.
Conventional genetics was used to generate lines of flies bearing multiple mutations, drivers, and other genetic tools. For the description of genotypes and crosses, please see Text S3.
To drive gene expression in larval stages only, thus enabling normal embryogenesis, we used (1) the temperature sensitive GAL4 repressor GAL80ts driven by the general tubulin promoter in tubGAL80ts flies (Bloomington) [76]; GAL80ts represses GAL4 at 18°C but not at 30°C, so larval GAL4 expression is controlled by shifting larvae from 18°C to 30°C at the required time. (2) hsGAL4 (gift of S. Brogna) flies, where GAL4 is switched on after heat-shock at 37°C. (3) Notchts1. For details, see Text S3.
Eggs were collected for 6 h and kept at 18°C until heat-shock was applied to the larvae. Mosaic Analysis with a Repressible Cell Marker (MARCM) clones were generated as described [77]; for genotypes, see Text S3.
VNCs were dissected from 96 h AEL old larvae (unless otherwise indicated elsewhere) in Shields and Sang M3 insect culture media (Sigma). The VNCs were stabbed from the dorsal side with a fine tungsten needle (Fine Scientific Tools), of 0.5 mm diameter at the base and 1 µm diameter at the tip. Culture of dissected injured or uninjured VNCs was done according to [78] with the indicated adaptations. Each brain was cultured separately in a well (24-well plate) containing 500 µl culture medium with 7.5% fetal bovine serum (Sigma), 1% Penicillin, and streptomycin (Sigma) for 18 to 22 h at 25°C. Control VNCs were dissected and cultured in the same way, without stabbing injury. Following culture, VNCs were fixed and stained as normal.
Immunohistochemistry, in situ hybridisations to mRNA and BrdU incorporation experiments of larval VNCs were done following standard procedures. Samples were mounted either in Vectashield with the nuclear dye DAPI (Vector Laboratories) or in 80% Glycerol PBS after staining nuclei with Daunomycin 5 µg/ml (Sigma) or Hoechst33342 5 µg/ml (Sigma). For BrdU detection, the VNCs were treated with 2 M HCl for 20 min at room temperature after immunolabeling for other proteins. Samples were mounted either in Vectashield with DAPI (Vector Laboratories) or in 80% Glycerol PBS after staining nuclei with Daunomycin 5 µg/ml (Sigma) or Hoechst33342 5 µg/ml (Sigma). For antibodies used, details on plasmids, and other, see Text S3.
Bright field, laser scanning confocal and transmission electron microscopy, and image processing were done following standard procedures (more details in Text S3).
Time-lapse confocal laser scanning throughout the entire neuropile was done visualizing axons with G9 (gift of W. Chia) [79], an protein-trap line with GFP in all CNS axons, and all glia (except midline glia) with repoGAL4/UASDsRed. Experimental genotypes were generated by crossing G9; repoGAL4/TM6B flies to the following flies: (1) UASDsRed S197Y (gift of K. Ito) [80], (2) UASDsRed S197Y;UASNotchICDmyc, (3) Notchts1 UASDsRed S197Y/FM7(sn+)actGFP, and (4) by crossing G9;alrmGAL4 (alrmGAL4 is a gift of Marc Freeman) [64] to UASDsRed S197Y; UASpros. The stabbed VNCs were placed dorsal side down in a 35 mm glass based dish (Iwaki) treated with PolyLysine (Sigma). Time-lapse scans (xyzt scan) were carried out using a Leica SP2-AOBS confocal inverted microscope with a temperature controlled chamber set to 25°C, or to 30°C for Notchts1 and its control experiments, 1- to 2-h intervals per Z-scan (8 time points in average), 20× lens with 4 times zoom, and 1 µm interval between optical slices. The obtained images were processed in ImageJ using plugins Turboreg and Stackreg to correct accidental sample movement. The Videos were arranged using ImageJ and Adobe Photoshop.
To count automatically the number of all larval glia stained with anti-Repo and acquired as confocal microscopy images, we purposely wrote DeadEasy Larval Glia software in Java as an ImageJ plug-in. Confocal serial sections were obtained with the BioRad Radiance 2000 confocal microscope as images of x,y = 0.5665 µm/pixel and z = 1 µm/pixel dimensions. The region of interest (ROI) was defined as the region starting from immediately posterior to the Ebony-positive ventral IG of the last thoracic segment to the posterior tip of the VNC, hereby referred to as abdomen. Peripheral nerves exiting the VNC were excluded from the ROI. DeadEasy Larval Glia identifies the stained cells first in 2-D based on shape (circular or elliptical) through each confocal slice and then in 3-D throughout the stack, based on minimum and maximum volume and minimum pixel intensity. DeadEasy creates a stack of processed images, where the identified objects reproduce those of the Repo glia in the raw images, enabling easy comparisons and validation. Each cell can be uniquely identified, as placing the mouse over each cell highlights a number, with which it is possible to check if, for example, two adjacent cells are counted as one. The programme was validated using n = 997 cells out of 3 different stacks of images. The mathematical algorithm for DeadEasy Larval Glia will be published elsewhere (Forero, Kato, and Hidalgo, in preparation).
To count the number of dying cells (labeled with active Caspase3-positive) throughout the VNC, we adapted the programme DeadEasy Caspase [81] to work on larvae. Given the strong variation in background intensity in larval samples, the outlier thresholding method originally used for embryos did not provide good results under the new conditions. Entropy thresholding was used instead, which provided better results. However, some other labeled tissues could not be rejected automatically anymore, given that they did not have a particular shape or size which would have allowed one to differentiate them from the apoptotic cells. Poor signal-noise ratio due to thickness of larval VNCs also resulted in false negatives/positives in deeper slices. Therefore, we manually corrected the error by deleting false positive and adding back false negatives for the final counting, employing the ImageJ DeadEasy Manual macro (described below).
Anti-Ebony stained IG were counted manually with support of the ImageJ DeadEasy Manual macro, which we purposely developed to speed up manual counting and eliminate error. With this macro, Ebony positive cells in a confocal stack of images are labelled manually with a digital colour, and DeadEasy Manual software automatically counts the colour labels.
Wound area was measured on longitudinal confocal microscopy images of the neuropile, using the ROI manager in ImageJ, for the glial wound on anti-Ebony and anti-GS2 stained VNCs, and for the axonal wound and vacuoles in time-lapse on G9-GFP-expressing VNC. The largest outline of the wound throughout the stack of images (i.e., a neuropile) was set as the ROI and measured in µm2. The kinetics of the area affected by the wound and vacuoles were obtained by normalising the size of affected area at each timepoint to the size of the area at 0–1 h after stabbing.
For statistical tests applied to each experiment and p values, please see Text S3 and Table S1.
|
10.1371/journal.pgen.1004022 | Meta-Analysis Identifies Gene-by-Environment Interactions as Demonstrated in a Study of 4,965 Mice | Identifying environmentally-specific genetic effects is a key challenge in understanding the structure of complex traits. Model organisms play a crucial role in the identification of such gene-by-environment interactions, as a result of the unique ability to observe genetically similar individuals across multiple distinct environments. Many model organism studies examine the same traits but under varying environmental conditions. For example, knock-out or diet-controlled studies are often used to examine cholesterol in mice. These studies, when examined in aggregate, provide an opportunity to identify genomic loci exhibiting environmentally-dependent effects. However, the straightforward application of traditional methodologies to aggregate separate studies suffers from several problems. First, environmental conditions are often variable and do not fit the standard univariate model for interactions. Additionally, applying a multivariate model results in increased degrees of freedom and low statistical power. In this paper, we jointly analyze multiple studies with varying environmental conditions using a meta-analytic approach based on a random effects model to identify loci involved in gene-by-environment interactions. Our approach is motivated by the observation that methods for discovering gene-by-environment interactions are closely related to random effects models for meta-analysis. We show that interactions can be interpreted as heterogeneity and can be detected without utilizing the traditional uni- or multi-variate approaches for discovery of gene-by-environment interactions. We apply our new method to combine 17 mouse studies containing in aggregate 4,965 distinct animals. We identify 26 significant loci involved in High-density lipoprotein (HDL) cholesterol, many of which are consistent with previous findings. Several of these loci show significant evidence of involvement in gene-by-environment interactions. An additional advantage of our meta-analysis approach is that our combined study has significantly higher power and improved resolution compared to any single study thus explaining the large number of loci discovered in the combined study.
| Identifying gene-by-environment interactions is important for understand the architecture of a complex trait. Discovering gene-by-environment interaction requires the observation of the same phenotype in individuals under different environments. Model organism studies are often conducted under different environments. These studies provide an unprecedented opportunity for researchers to identify the gene-by-environment interactions. A difference in the effect size of a genetic variant between two studies conducted in different environments may suggest the presence of a gene-by-environment interaction. In this paper, we propose to employ a random-effect-based meta-analysis approach to identify gene-by-environment interaction, which assumes different or heterogeneous effect sizes between studies. Our approach is motivated by the observation that methods for discovering gene-by-environment interactions are closely related to random effects models for meta-analysis. We show that interactions can be interpreted as heterogeneity and can be detected without utilizing the traditional approaches for discovery of gene-by-environment interactions, which treats the gene-by-environment interactions as covariates in the analysis. We provide a intuitive way to visualize the results of the meta-analysis at a locus which allows us to obtain the biological insights of gene-by-environment interactions. We demonstrate our method by searching for gene-by-environment interactions by combining 17 mouse genetic studies totaling 4,965 distinct animals.
| Identifying environmentally specific genetic effects is a key challenge in understanding the structure of complex traits. In humans, gene-by-environment (GxE) interactions have been widely discussed [1]–[12] yet only a few have been replicated. One reason for this discrepancy is the inability to accurately control for environmental conditions in humans as well as the inability to observe the same individuals in multiple distinct environments. Model organisms do not share such difficulties and for this reason can play a crucial role in the identification of gene-by-environment interactions. For example, in many mouse genetic studies the same traits are examined under different environmental conditions. Specifically, knock-out or diet-controlled mice are often utilized in the study of cholesterol levels. The availability of these studies presents a unique opportunity to identify genomic loci involved in gene-by-environment interactions as well as those loci involved in the trait independent of the environment.
In order to utilize genetic studies in model organisms to identify gene-by-environment interactions, one needs to directly compare the effects of genetic variations in studies conducted under different conditions. This practice is complicated for a number of reasons, when combining more than two studies. First, environmental conditions are often variable across studies and do not fit to the standard univariate model for interactions. For example, in one study, cholesterol may be examined under different diet conditions (eg. low fat and high fat) and then in another study cholesterol is examined using gene knockouts. In this case, it is not straightforward to analyze these studies in aggregate using a single variable to represent the environmental condition. Applying a multivariate model, one in which the environment is represented using multiple environmental variables, results in increased degrees of freedom and low statistical power. Second, model organisms such as the mouse exhibit a large degree of population structure. Population structure is well-known for causing false positives and spurious associations [13], [14] in association analysis and can be expected to complicate the ability to combine separate studies.
In this paper, we propose a random-effects based meta-analytic approach to combine multiple studies conducted under varying environmental conditions and show that this approach can be used to identify both genomic loci involved in gene-by-environment interactions as well as those loci involved in the trait independent of the environment. By making the connection between gene-by-environment interactions and random effects model meta-analysis, we show that interactions can be interpreted as heterogeneity and detected without requiring uni- or multi-variate models. We also define an approach for correcting population structure in the random effects model meta-analysis, extending the methods developed for fixed effects model meta-analysis [15]. We show that this method enables the analyses of large scale meta-analyses with dozens of heterogeneous studies and leads to dramatic increases in power. We demonstrate that insights regarding gene-by-environment interactions are obtained by examining the differences in effect sizes among studies facilitated by the recently developed m-value statistic [16], which allows us to distinguish between studies having an effect and studies not having an effect at a given locus.
We applied our approach, which we refer to as Meta-GxE, to combine 17 mouse High-density lipoprotein (HDL) studies containing 4,965 distinct animals. To our knowledge, this is the largest mouse genome-wide association study conducted to date. The environmental factors of the 17 studies vary greatly and include various diet conditions, knock-outs, different ages and mutant animals. By applying our method, we have identified 26 significant loci. Consistent with the experience of meta-analysis in human studies, our combined study finds many loci which were not discovered in any of the individual studies. Among the 26, 24 loci have been previously implicated in having an effect on HDL cholesterol or closely related lipid levels in the blood, while 2 loci are novel findings. In addition, our study provides insights into genetic effects on several disease loci and their relationship between environment and sex. For example, we identified 3 loci (Chr10:21399819, Chr19:3319089, ChrX:151384614), where female mice show a more significant effect on HDL phenotypes than male mice. We also identified 7 loci (Chr1:171199523, Chr8:46903188, Chr8:64150094, Chr8:84073148, Chr10:90146088, Chr11:69906552, Chr15:21194226) where male mice show a more significant effect on HDL than female mice. In addition, many of the loci show strong gene-by-environment interactions. Using additional information describing the studies and our predictions of which studies do and do not contain an effect, we gain insights into the interaction. For example, locus on chromosome 8 (Chr8:84073148) shows a strong sex by mutation-driven LDL level interaction, which affects HDL cholesterol levels.
Part of the reason for our success in identifying a large number of loci is that our study combined multiple mouse genetic studies many of which use very different mapping strategies. Over the past few years, many new strategies have been proposed beyond the traditional F2 cross [17] which include the hybrid mouse diversity panel (HMDP) [18], [19], heterogeneous outbred stocks [20], commercially available outbred mice [21], and the collaborative cross [22]. In our current study, we are combining several HMDP studies with several F2 cross studies and benefit from the statistical power and resolution advantages of this combination [15]. The methodology presented here can serve as a roadmap for both performing and planning large scale meta-analysis combining the advantages of many different mapping strategies. Meta-GxE is publicly available at http://genetics.cs.ucla.edu/metagxe/.
The Meta-GxE strategy uses a meta-analytic approach to identify gene-by-environment interactions by combining studies that collect the same phenotype under different conditions. Our method consists of four steps. First, we apply a random effects model meta-analysis (RE) to identify loci associated with a trait considering all of the studies together. The RE method explicitly models the fact that loci may have different effects in different studies due to gene-by-environment interactions. Second, we apply a heterogeneity test to identify loci with significant gene-by-environment interactions. Third, we compute the m-value of each study to identify in which studies a given variant has an effect and in which it does not. Forth, we visualize the result through a forest plot and PM-plot to understand the underlying nature of gene-by-environment interactions.
We illustrate our methodology by examining a well-known region on mouse chromosome 1 harboring the Apoa2 gene, which is known to be strongly associated with HDL cholesterol [23]. Figure 1 shows the results of applying our method to this locus. We first compute the effect size and its standard deviation for each of the 17 studies. These results are shown as a forest plot in Figure 1 (a). Second we compute the P-value for each individual study also shown in Figure 1 (a). If we were to follow traditional methodology and evaluate each study separately, we would declare an effect present in a study if the P-value exceeds a predefined genome-wide significance threshold (P ). In this case, we would only identify the locus as associated in a single study, HMDP-chow(M) (P = ). On the other hand, in our approach, we combine all studies to compute a single P-value for each locus taking into account heterogeneity between studies. This approach leads to increased power over the simple approach considering each study separately. The combined meta P-value for the Apoa2 locus is very significant (), which is consistent with the fact that the largest individual study only has 749 animals compared to 4,965 in our combined study.
In order to evaluate how significantly different the effect sizes of the locus are between studies, we apply a heterogeneity test. The statistical test is based on Cochran's Q test [24], [25], which is a non-parametric test for testing if studies have the same effect or not. In this locus, the effect sizes are clearly different and not surprisingly the P-value of the heterogeneity test is significant (). This provides strong statistical evidence of a gene-by-environment interaction at the locus. Below we more formally describe how heterogeneity in effect size at a given locus can be interpreted as gene-by-environment interaction.
If a variant is significant in the meta-analytic testing procedure, then this implies that the variant has an effect on the phenotype in one or more studies. Examining in which subset of the studies an effect is present and comparing to the environmental conditions of the studies can provide clues to the nature of gene-by-environment interactions at the locus. However, the presence of the effect may not be reflected in the study-specific P-value due to a lack of statistical power. Therefore, it is difficult to distinguish only by a P-value if an effect is absent in a particular study due to a gene-by-environment interaction at the locus or a lack of power. In order to identify which studies have effects, we utilize a statistic called the m-value [16], which estimates the posterior probability of an effect being present in a study given the observations from all other studies. We visualize the results through a PM-plot, in which P-values are simultaneously visualized with the m-values at each tested locus. These plots allow us to identify in which studies a given variant has an effect and in which it does not. M-values for a given variant have the following interpretation: a study with a small m-value() is predicted not to be affected by the variant, while a study with a large m-value() is predicted to be affected by the variant.
The PM-plot for the Apoa2 locus is shown in Figure 1 (b). If we only look at the separate study P-values (y-axis), we can conclude that this locus only has an effect in HMDP-chow(M). However, if we look at m-value (x-axis), then we find 8 studies (HMDPxB-ath(M), HMDPxB-ath(F), HMDP-chow(M), HMDP-fat(M), HMDP-fat(F), BxD-db-5(M), BxH-apoe(M), BxH-apoe(F)), where we predict that the variation has an effect, while in 3 studies (BxD-db-12(F), BxD-db-5(F), BxH-wt(M)) we predict there is no effect. The predictions for the remaining 6 studies are ambiguous.
From Figure 1, we observe that differences in effect sizes among the studies are remarkably consistent when considering the environmental factors of each study as described in Table 1. For example, when comparing study 1–4, the effect size of the locus decreases in both the male and female HMDPxB studies in the chow diet (chow study) relative to the fat diet (ath study). Thus we can see that when the mice have Leiden/CETP transgene, which cause high total cholesterol level and high LDL cholesterol level, effect size of this locus on HDL cholesterol level in blood is affected by the fat level of diet. Similarly, when comparing study 12–15, the knockout of the Apoe gene affects the effect sizes for both male and female BxH crosses. However, in the BxD cross (study 8–11), where each animal is homozygous for a mutation causing a deficiency of the leptin receptor, the effect of the locus is very strong in the young male animals, while as animals get older and become fatter, the effect becomes weaker. However in the case of female mice, the effect of the locus is nearly absent at both 5 and 12 weeks of age. Thus we can see that sex plays an important role in affecting HDL when the leptin receptor activity is deficient. We note that there are many genes in this locus and the genetic mechanism of interactions may involve genes other than Apoa2. Despite this caveat, the results of Meta-GxE at this locus provides insights into the nature of GxE and can provide a starting point for further investigation.
We note that an alternate explanation for differences in effect sizes between studies is the presence of gene-by-gene interactions and differences in the genetic backgrounds of the studies. While this is a possible explanation for differences in effect sizes between the different crosses and the HMDP studies, in Figure 1, we see many differences in effect sizes among studies with the same genetic background. Thus gene-by-gene interactions can only partially explain the differences in observed effect sizes.
Gene-by-environment interactions, random effects meta-analysis and heterogeneity testing are closely related. Suppose we have studies each conducted under different environmental conditions. We define the following linear model, where is the observed phenotype for study , is the phenotype mean for study , is the genetic effect on the phenotype for study , is the genotype, and is the residual error.(1)
Since each environment is different, the effect size is partially determined by environmentally-specific factors and partially determined by factors common to all studies. Given that we can decompose the effect into environment-independent and environment-dependent factors. Then we define the following linear model, where is the environment-independent genetic effect and is the environment-dependent genetic effect for study .(2)
In order to test for the presence of an effect shared across environments, we test the null hypothesis and to test for the presence of a gene-by-environment interaction, we test the hypothesis that .
In the random effects meta-analysis, we assume that the effect size is sampled from a normal distribution with mean and variance , denoted . Under this assumption, we test the null hypothesis and , in order to obtain a study-wide P-value. Additionally, we perform a heterogeneity test to test the null hypothesis versus the alternative hypothesis . We posit that by conducting hypotheses tests in the meta-analysis framework, we are simultaneously testing for the presence of environmentally-independent and environmentally-specific effects and that by applying heterogeneity testing we are testing for only environmentally-specific effects.
Consider that in the meta-analysis framework is analogous to and the variation () around is analogous to variation among s. In the random effects meta analysis testing framework we are testing if and . This is equivalent to testing both environmentally-independent () and environmentally-dependent () effects simultaneously. In heterogeneity testing, we test the null hypothesis versus the alternative hypothesis . When the environmentally-dependent effect () is 0 it means that and thus . When , we expect that will vary around , so that we do not expect that . Since the variation () of around is analogous to the variable , heterogeneity testing in the meta-analysis framework is approximately equivalent to testing for environmentally-specific effects.
The presence of heterogeneity in the effect size at causal genetic loci due to gene-by-environment interactions is naturally expected in mouse genetic studies when combining studies with varying environmental conditions. One extreme example comes from a knock-out experiment. If the knocked-out gene is causal for a particular trait, then we can expect that the gene would have no effect on a knock-out mouse, while the gene would have an effect on the wild type mouse. This is a binary form of heterogeneity. In a less extreme form of heterogeneity, the effect of a given gene may be affected by an environmental factor which varies in different mice – ranging from small effects to large effects.
To see the relationship between significance of the association and gene-by-environment interactions, we compute and compare this P-value for each SNP from the 17 studies using the random effects meta-analysis to a measure of heterogeneity. Heterogeneity can be assessed by statistic, which describes the percentage of variation across studies that is due to heterogeneity rather than chance [26].
Figure 2 compares statistic with the meta-analysis P-value for each SNP. In this figure, we see that is uniformly distributed for the non-significant SNPs (blue dots), while it is right skewed for significant SNPs (red dots), indicating that more significant SNPs have a greater potential for exhibiting heterogeneity in effect. Since heterogeneity in this case can be interpreted as representing gene-by-environment interactions, as heterogeneity is induced by differences in the environment, we see that the presence of a GxE interaction confers higher power to detect an association.
The power to identify both gene-by-environment and main effects in a meta-analysis of mouse studies depends on both the main effect size and the amount of heterogeneity. We performed simulations using the genotypes of the 17 mouse studies analyzed in this paper. We simulated a range of main effect (mean effect) sizes and a range of gene-by-environment effects. We are simulating the realistic scenario in which we do not know exactly the set of covariates which are responsible for the gene-by-environment effects. We simulated gene-by-environment effects by drawing the effect in each study from a distribution with a mean given by the main effect size and a variance controlling the magnitude of gene-by-environment interactions. If this variance is small, then all of the studies have close to the same effect size and there are few gene-by-environment effects. If the variance is high, then there are strong gene-by-environment effects. Figure 3 shows the results of our simulations. 1000 simulated phenotypes were generated for each mean and variance pair. Statistical power is estimated by computing the proportion of the datasets in which a simulated effect is detected. We observe that the power is high for a wide range of main effect sizes and gene-by-environment effect sizes which is explained by the large sample size of the study. We also observe that even for small main effects, if there are strong gene-by-environment effects, we can still identify the locus. This is because in this case a subset of the studies will have strong effect sizes due to gene-by-environment effects.
Our approach is not the only way to analyze a meta-analysis study. We compare the power to two other meta-analytic approaches. The first is the traditional meta-analysis strategy which uses a fixed effects model (FE) in which all of the effect sizes across studies are assumed to be the same. We utilize an extension of the fixed effects model which corrects for population structure [15]. A second alternate strategy is to simply apply the heterogeneity test (HE), which in our framework is only applied to loci first identified using random effects meta-analysis. The HE test follows the intuition that loci with high heterogeneity will harbor gene-by-environment interactions. For the purposes of the comparison we refer to Meta-GxE as the random effects (RE) model.
The level of gene-by-environment interactions can be simulated by changing both the environment-dependent and environment-independent effect simultaneously, when simulating the phenotype. Figure 4 (a)–(c) shows the power of the three approaches (RE, FE, HE) respectively when we vary the mean and variance of the effect size distribution we sampled from. In this simulation study, mean effect represents shared effect and variance of the effect size represents interaction effect. As expected, RE has high power in cases where the shared effect or the interaction effect is large. FE has high power when the shared effect is large and the HE test has high power when the interaction effect is large. Figure 4 (d) shows the heatmap which is colored with the color of highest powered approach. FE is most powerful at the top-left region, HE is most powerful at the bottom-right region, while RE is most powerful for a majority of the simulations. In the Text S1, we show through simulations that our methodology outperforms the alternative fixed effects and heterogeneity testing approaches when the effect is present in a subset of the studies, which is another possible interaction model we can assume. We also show in the Text S1 that our approach is more powerful than the traditional uni- or multi-variate gene-by-environment association approach which assumes knowledge of the covariates involved in gene-by-environment interactions. For the traditional uni- or multi-variate approach, required knowledge includes kinds of variable (e.g. sex, age, gene knockouts) and encoding of the variables (e.g. binary values, continuous values). In the Text S1, we also show the our proposed approach controls the false positive rate.?
We applied Meta-GxE to 17 mouse genetic studies conducted under various environmental conditions where each study measured HDL cholesterol. Table 1 summarizes each study. More details are provided in the Materials and Methods section and in Text S1. We analyzed all 17 studies together and we also analyzed the 9 male and 8 female studies separately. Some significant associations are shared and some associations are specific to males and females.
The Manhattan plots in Figure 5 show the meta-GxE result when applied to the 17 studies, 9 male only studies and 8 female only studies. Table 2 summarizes 26 significant peaks (P) showing the P-values obtained by applying meta-GxE to the male only studies (9 studies), the female only studies (8 studies) and the male+female studies (17 studies). For each significant locus, we computed m-values, interpreted as the posterior probability of having an effect on the phenotype and report the number of studies with an effect (E), the number of studies with ambiguous effect size (A) and the number of studies without an effect (N). We also report the number of individual studies where the locus was significant (P). As seen in the table, many of the loci were not significant in any of the individual studies and would not have been discovered without combining the studies. We note that we use a more stringent genome wide threshold of P than was used in the original studies. The Genes in Region and Gene Refs columns contain the gene names near the locus previously known to affect HDL cholesterol level or closely related lipid level in the blood and associated literature citations.
Among the 26 loci that we identified by applying Meta-GxE, 24 loci are near the genes (mostly genes are located within 1MB of the peak) known to affect HDL or closely related lipid level in the blood, while 3 loci are novel.
For example, we identified 3 loci (Chr10:21399819, Chr19:3319089, ChrX:151384614) female mice show a more significant effect on HDL phenotypes than male mice. We also identified 7 loci (Chr1:171199523, Chr8:46903188, Chr8:64150094, Chr8:84073148, Chr10:90146088, Chr11:69906552, Chr15:21194226) where male mice show a more significant effect on HDL than female mice.
Interestingly, we observed that in 3 loci (Chr10:21399819, Chr19:3319089, ChrX:151384614), female mice are more highly affected, while in 7 loci (Chr1:171199523, Chr8:46903188, Chr8:64150094, Chr8:84073148, Chr10:90146088, Chr11:69906552, Chr15:21194226) male mice are more highly affected. Among 26 loci, many show a significant heterogeneity in effect sizes between the 17 studies, which we interpret as gene-by-environment interactions.
One interesting example showing strong gene-by-environment interaction is a locus in Chr8:84073148. This locus is located near the gene , which is known to affect the abnormal lipid levels in blood [27]. Figure 6 shows the forest plot and PM-plot for this locus. If we look at the forest plot of the locus in Figure 6, we can easily see that there are two groups: 12 studies with an effect (red dots) and 5 studies with an ambiguous prediction of the existence of an effect (green dots). Interestingly, the log odds ratios of effect size for the 12 studies with an effect is about the same (around 0.2). The common characteristic in 4 of the 5 studies (HMDPxB-chow(F), HMDPxB-ath(F), BXH-apoe(F), CXB-ldlr(F)) is that they are female mice with high LDL levels in the blood. In addition, in all 4 cases, these high LDL levels are caused by mutant genes. Mice in HMDPxB-chow and HMDPxB-ath studies have transgenes for both Apoe Leiden and for human Cholesterol Ester Transfer Protein (CETP), while mice in the BXH-apoe and CXB-ldlr studies carried knockouts of the genes for Apoe and LDL receptor, respectively. This is a strong evidence that there is an interaction between sex×mutation-driven LDL levels through this locus (Chr8:84073148) when affecting HDL levels in mice.
Figures S5, S6, S7, S8, S9, S10, S11, S12, S13, S14, S15, S16, S17, S18, S19, S20, S21, S22, S23, S24, S25, S26, S27, S28, S29, S30 show the forest plots and PM-plots for each locus, which show information such as effect sizes, the direction of the effect, which study has an effect and which study does not have an effect for each of 17 studies at the given locus.
In this paper, we present a new meta-analysis approach for discovering gene-by-environment interactions that can be applied to a large number of heterogeneous studies each conducted in different environments and with animals from different genetic backgrounds. We show the practical utility of the proposed method by applying it to 17 mouse HDL studies containing 4,965 mice, and we successfully identify many known loci involved in HDL. Consistent with the results of meta-analysis in human studies, our combined study finds many loci which were not discovered in any of the individual studies.
A point of emphasis is that in our study design, in each of the combined studies, all of the individuals in the study are subject to only a single environment. This is distinct from other approaches for discovery of gene-by-environment interactions using meta-analysis such as those described in [28]. In these approaches, in each of the combined studies, the individuals in the study are subject to multiple environments and information on each individual's environment is collected. Gene-by-environment statistics are then computed in each study and then combined in the meta-analysis. In our study design, we compute main effect sizes for each SNP and then look for variants where the effect sizes are different suggesting the presence of a gene-by-environment interaction.
In our meta-analysis approach, we assume that we do not have any prior knowledge of the effect size in any particular study. However one might incorporate prior knowledge of the specific environmental effects. In some cases, one might know that some of the studies have similar effect sizes as compared to others. Or the prior knowledge might suggest that one specific study needs to be eliminated in the meta analysis. If we utilize such prior knowledge, we may be able to achieve even higher statistical power.
In this paper we have addressed how to perform meta-analysis when the studies have different genetic structures, building off the results of our previous study [15]. While in this paper we combine 7 HMDP studies with 10 genetic crosses, the approach in principle can be used to combine any variety of study types. Recently, several strategies for mouse genome-wide association mapping have been proposed [29] [17]. These include HMDP [18], collaborative cross [30] and outbredstock [21] [17]. The approach presented here can be utilized to combine these different kinds of studies and is a roadmap for integrating the results of different strategies for mouse GWAS.
Although we have focused on explaining heterogeneity by gene-by-environment interaction, it is possible that the differences in effect sizes can be caused by gene-by-gene interactions on different genetic backgrounds, where the interacting variants differ in frequency in the different studies. While gene-by-gene interactions certainly contribute to locus heterogeneity, we predict that, in combining studies with similar genetic structures, locus heterogeneity more likely arises from gene-by-environment interactions. In any case, determining whether or not these heterogeneous loci are environment-driven or interaction-driven is an important and interesting direction for future study.
In the model organism studies for which we can control the environment, the standard study design for testing gene-by-environment interactions is to combine multiple cohorts whose environments are known. The environmental value that we vary is typically a quantitative measure that we can model with a single random variable. Thus, the standard univariate linear model can be appliedwhere is vector of phenotype measurements from individuals, is the phenotype mean, is the main environmental effect mean, is environmental status vector, is the genetic effect, is genotype vector, is GxE interactions effect, denotes the dot-product between two vectors, and is the residual error, which follows normal distribution. In this model, vector is a vector of indicators which describes the environmental status of each individual. study. For example, Suppose the environmental condition of one study is wildtype and that of another is gene knockout. In this case, the environmental condition of wildtype is described as 0 and that of knockout is described as 1. In order to test if there are interactions, we test the null hypothesis versus the alternative hypothesis . Another possible testing strategy is to test the interactions effect together with the genetic effect, that is, the null hypothesis versus the alternative hypothesis . This strategy is powerful in detecting loci exhibiting both the genetic effects and the interactions effects.
For more complicated scenarios where the different environments can not be modeled with a single variable, a straightforward extension of the standard univariate interactions model is the multivariate model. Suppose that there are k different possible environments and the information on the environments of each individual are captured by a matrix D which has k columns where each column corresponds to one environment. Then, the standard multivariate interactions model will be(3) is the column of the D matrix, is the environment specific mean, denotes the phenotype measurements, denotes the genotypes, denotes the fixed genetic effect, denotes GxE interactions effect of environmental variable and, and denotes the residual error. Then the testing will be between the null hypothesis versus the alternative hypothesis . The test statistic will bewhere is the z-score corresponding to . follow under the null. Similarly to the univariate model, if we want to test the interactions effect together with genetic effect, we add the z-score corresponding to into the statistic, in which case the statistic will follow .
Before we describe the relationship between gene-by-environment interactions and meta-analysis, we first describe the standard fixed effects and random effects meta-analysis in details.
Here we explain more about the relationship between gene-by-environment interactions and meta-analysis based on the explanation in Results section. If we do not consider the interactions, it has been already known that the fixed effects model meta-analysis is approximately equivalent to the linear model of combined cohorts [35]. That is, the fixed effects model equation (5) gives approximately equivalent results to the combined linear model(8)where is the combined genotype vector from all cohorts, is a matrix that includes indicator columns which identify which individual is in each cohort, is the column of matrix A, and is the cohort specific mean. The two methods are approximately equivalent because they both test the fixed mean effect ( in equation (8) and in equation (5)). The subtle difference between the two models is that in equation (8), we assume the error follows a single normal distribution (e.g. ), whereas in equation (5), the variance of the distributions may differ between studies (e.g. for each ). In other words, under the constant error variance assumption (), the two models become equivalent and in equation (8) equals in equation (5),
Similarly, by considering interactions, we extend this argument to show the relationship between gene-by-environment interactions and meta-analysis. We consider the relationship between equation (3) and equation (4). For simplicity of the notation, we consider the case where the matrix D is defined in such a way that each individual is only in one environment such that the D matrix is equivalent to the matrix A described above. If we assume the constant error variance assumption, we establish the following relationship,where the left hand side is the coefficient of the genotype of study from the meta-analysis equation (4) and the right hand side is the same coefficient of (the study 's part within the combined genotype matrix ) from the equation (3).
Suppose that there are no interactions (null hypothesis of interaction testing). Then, for each study . Thus, the effect size of meta-analysis is equivalent to , the genetic effects that are invariant across studies. Therefore, (null hypothesis of heterogeneity testing). On the other hand, suppose that (null hypothesis of heterogeneity testing). Naturally, for all studies (null hypothesis of interaction testing). This shows that the null hypothesis of the interactions test in the model (3) and the null hypothesis of the heterogeneity test in meta-analysis are equivalent. As a result, we can utilize meta-analytic heterogeneity testing to detect interactions.
Using reasoning, it is straightforward to show that we can utilize the random effects model meta-analysis method to detect the mean effect and the interaction effect at the same time, which can be powerful for identifying loci bearing both kinds of effects.
Model organism such as the mouse are well-known to exhibit population structure or cryptic relatedness [36], [37], where genetic similarities between individuals both inhibit the ability to find true associations and cause the appearance of a large number of false or spurious associations. Mixed effects models are often used in order to correct this problem [38]–[42]. Methods employing a mixed effects correction account for the genetic similarity between individuals with the introduction of a random variable into the traditional linear model.(9)
In the model in equation (9), the random variable represents the vector of genetic contributions to the phenotype for individuals in population . This random variable is assumed to follow a normal distribution with , where is the kinship coefficient matrix for population . With this assumption, the total variance of is given by . A z-score statistic is derived for the test by noting the distribution of the estimate of . In order to avoid complicated notation, we introduce a more basic matrix form of the model in equation (9), shown in equation (10).(10)
In equation (10), is a matrix with the first column being a vector of 1 s representing the global mean and the second vector is the vector and is a coefficient vector containing the mean and genotype effect . We note that this form also easily extends to models with multiple covariates. The maximum likelihood estimate for in population is given by which follows a normal distribution with a mean equal to the true and variance . The estimates of the effect size and standard error of the () are then given in equation (11) and equation (12), where is a vector used to select the appropriate entry in the vector .(11)(12)
When we test gene-by-environment interactions with meta analysis approaches, one important step is correcting for population structure. This can be achieved by correcting for population structure within each study first as described above. For example, consider the random effects model meta-analysis method that we primarily focus on. We employ population structure control, using (11) and (12). Then the likelihood ratio test statistic will be(13)where and .
After identifying loci exhibiting interaction effects, we employ the meta-analysis interpretation framework that we recently developed. The m-value [16] is the posterior probability that the effect exists in each study. Suppose we have number of studies we want to combine. Let be the vector of estimated effect sizes and be the vector of estimated variance of effect sizes. We assume that the effect size follows the normal distribution.(14)(15)We assume that the prior for the effect size is(16)A possible choice for in GWASs is 0.2 for small effect and 0.4 for large effect [43]. We also denote be a random variable whose value is 1 if a study has an effect and 0 otherwise. We also denote as a vector of for studies. Since has binary values, can be possible configurations. Let be a vector containing all the possible these configurations. We define m-value as the probability , which is the probability of study having an effect given the estimated effect sizes. We can compute this probability using the Bayes' theorem in the following way.(17)where is a subset of whose elements' value is 1. Now we need to compute and . can be computed as(18)where denotes the number of 1's in c and B denotes the beta function and we set and as 1 [16]. The probability given configuration , , can be computed as(19)(20)(21)where where is the indices of 0 in and is the indices of 1 in , denotes the probability density function of the normal distribution with mean and variance . is the inverse variance or precision and is a scaling factor.(22)All summations appeared for computing , and are with respect to .
The m-values have the following interpretations: small m-values(0.1) represent a study that is predicted to not have an effect, large m-values(0.9) represent a study that is predicted to have an effect, otherwise it is ambiguous to make a prediction. It was previously reported that m-values can accurately distinguish studies having an effect from the studies not having an effect [16]. For interpreting and understanding the result of the meta-analysis, it is informative to look at the P-value and m-value at the same time. We propose to apply the PM-plot framework [16], which plots the P-values and m-values of each study together in two dimensions. Figure 1 (b) shows one example of a PM-plot. In this example, studies with an m-value less than are interpreted as studies not having an effect while studies with an m-value greater than are interpreted as studies having an effect. For studies with an m-value between and , we cannot make a decision. One reason that studies are ambiguous () is that they are underpowered due to small sample size. If the sample size increases, the study can be drawn to either the left or the right side.
|
10.1371/journal.pntd.0003774 | Exposure to an Indoor Cooking Fire and Risk of Trachoma in Children of Kongwa, Tanzania | Elimination of blinding trachoma by 2020 can only be achieved if affected areas have effective control programs in place before the target date. Identifying risk factors for active disease that are amenable to intervention is important to successfully design such programs. Previous studies have linked sleeping by a cooking fire to trachoma in children, but not fully explored the mechanism and risks. We propose to determine the risk for active trachoma in children with exposure to cooking fires by severity of trachoma, adjusting for other known risk factors.
Complete census of 52 communities in Kongwa, Tanzania, was conducted to collect basic household characteristics and demographic information on each family member. Information on exposure to indoor cooking fires while the mother was cooking and while sleeping for each child was collected. 6656 randomly selected children ages 1-9yrs were invited to a survey where both eyelids were graded for follicular (TF) and intense trachoma (TI) using the WHO simplified grading scheme. Ocular swab were taken to assess the presence of Chlamydia trachomatis.
5240 (79%) of the invited children participated in the study. Overall prevalence for trachoma was 6·1%. Odds for trachoma and increased severity were higher in children sleeping without ventilation and a cooking fire in their room (TF OR = 1·81, 1·00–3·27 and TI OR 4·06, 1·96–8·42). Children with TF or TI who were exposed were more likely to have infection than children with TF or TI who were not exposed. There was no increased risk with exposure to a cooking fire while the mother was cooking.
In addition to known risk factors for trachoma, sleeping by an indoor cooking fire in a room without ventilation was associated with active trachoma and appears to substantially increase the risk of intense inflammation.
| Trachoma remains the leading preventable cause of infectious blindness in the world. Identifying risk factors for active disease that are amenable to intervention is key to successfully designing effective control programs to eliminate blinding trachoma. Association between cooking fire and eye inflammation makes biological sense, and multiple other studies, but conducted in adults, have been reported. This is the first detailed study of the risk of trachoma or infection with Chlamydia trachomatis in children associated with a broader exposure to cooking fires. We were able to identify an important risk factor: a strong relationship between cooking fire exposure while sleeping and active trachoma that appears to substantially increase the risk of intense inflammation, which may play a role in perpetuating the disease in the community reservoir, the children. Hence a higher risk for transmission and re-emergence in communities seeking elimination. This finding may suggest modifications to current behavioral risks that were not considered before and may significantly impact the progression of the disease.
| Trachoma remains the leading infectious cause of preventable blindness in the world, and a significant public health problem in endemic areas.[1] Elimination of trachoma by 2020 can only be achieved if all affected areas have effective control programs in place at a prudential period before the target date. The World Health Organization (WHO) advocates the SAFE strategy to control blinding trachoma (Surgery, Antibiotics, Facial cleanliness, and Environmental change) but identifying risk factors for trachoma that are amenable to intervention at family, or community level is important for designing successful control programs.[2]
Children are the main reservoir of the disease, so by controlling the risk factors and rate of infection in this population, we can decrease the community pools and risk of transmission, hence a great step in elimination of endemic trachoma. Known risk factors for active trachoma include young age, poor water access, unclean faces, and other household characteristics that are markers of poor socioeconomic status.[3] We had previously also found an association between children sleeping by a cooking fire and increased odds of trachoma; however the assessment of exposure was a simple question.[4] This association makes biological sense as multiple other studies, but conducted in adults, have associated pollution from cooking fires with eye irritation and alteration of ocular immunity. [5–14] However, no detailed study of the risk of trachoma or infection with Chlamydia trachomatis in children, associated with a broader exposure to sleeping rooms and rooms with a cooking fire has been carried out. We propose to determine the risk in children ages 1 to 9 years of active trachoma, both follicular diseases and intense trachoma, with a detailed assessment of exposure to indoor cooking fire.
The research complied with the tenets of the Declaration of Helsinki and all guardians gave written informed consent for study procedures. Research was conducted with approval from the Johns Hopkins institutional review board and the national institute of medical research of Tanzania.
This cross-sectional study was conducted in 52 communities in Kongwa district of central Tanzania. These communities underwent mass drug administration for the previous five years, and at the time of the survey were at least one year from having had treatment with azithromycin. The school curriculum stresses face washing as one of its hygiene components but no other hygiene campaign has being carried out.
A complete census of households in each community was carried out and a random sample of 128 children from each community between ages 1–9 years was selected for the trachoma survey. A total of 5240 children from 4311 households were surveyed. Details are described below.
The census was carried out prior to the survey by a trained census team that collected demographic information for all household members and household information including type of roof of the house, presence of latrine, distance to water source, level of education of the household leader, number of children in the house, age and sex of the children. A specific cooking fire questionnaire was done to collect information about the type, use and location of cooking stoves/fires in and around the house, at different times of the year. The characteristics of the room where each resident child slept was directly observed and recorded. We observed the presence or not of a cooking fire, and observed whether the room had ventilation defined as present if either of the following were present: the room had windows that allowed air in or vents at the top of the walls (at least 3 inches of space between the roof and the wall where sky could be seen).
Examination of each everted eyelid was performed by trained trachoma grader using a 2·5X loupe. The trachoma grader was trained by a GTMP certified grader (HM), and had to have a kappa of >.6 against the trachoma certified grader. Trachoma was assessed in both eyes using the WHO simplified grading scheme, which assesses the presence or absence of follicular trachoma (TF), severe intense trachoma (TI), conjunctival scarring (TS), trichiasis (TT), and corneal opacity (CO). For this study, the relevant signs are TF and TI, signs of active trachoma.[15,16] For our analyses, we defined active trachoma as the presence TF or TI, alone or together, and Intense trachoma as TI alone, or with TF in at leat one eye.
For quality control purposes photographs of the right upper eyelid of a random 20% sample of children examined (using a Nikon D-40 camera with a 105mm f/2·8D AF Macro lens). These images were used to monitor the consistency of grading. Inter-observer agreement between the grader and the master grader (SW) was kappa 0·72 (95 CI: 0·62–0·82).
All children had ocular swabs taken from the left eye for determination of infection, using strict protocols to avoid field contamination. In each village a 5% sample of children had “air swabs” taken to check for field contamination. Samples were stored at KTP in a refrigerator and shipped within 30 days to the Johns Hopkins International Chlamydia Laboratory to be analyzed using the APTIMA ACT commercial test for C. trachomatis (Gen-Probe Inc., San Diego CA). Lab personnel were masked to the identified study and “air” swabs. None of the “air” swabs were positive.
For sleeping in a room with a cooking fire, we created an exposure index as follows: the lowest exposure was sleeping in a room without a cooking fire and with ventilation, the next lowest was sleeping in a room without a cooking fire but with no ventilation, the higher category was sleeping in a room with a cooking fire but the room had ventilation, and the highest exposure was sleeping in a room with a cooking fire and the room had no ventilation. However, few households were at the high extreme only 3% of children slept in a room with a cooking fire, regardless of ventilation. Therefore, we used three levels to reflect exposure: sleeping in a room with a cooking fire, sleeping in a room with no cooking fire and no ventilation, and sleeping in a room with no cooking fire and with ventilation.
Results were adjusted for age. Exposure while sleeping was assessed with logistic regression models to determine the associations between presence of TF and TI and exposure, adjusting for other risk factors. In the multivariate model, a backward elimination procedure was used to construct a parsimonious model that included only factors with significance level ≤ 0·5. Exposure during the time a mother cooked was also assessed. We used stratified analyses to evaluate the effect of exposure to a cooking fire on risk of infection in children with trachoma.
Random effects models including a random intercept for the community were used to account for the correlation of trachoma within residents of the same community. Less than 10% of the children belonged to the same household; in a sensitivity analysis we determined that adjusting for this level of clustering had no effects on our results. All analyses were conducted in SAS (version 9.2, SAS Institute Inc., Cary, NC, USA).
A total of 6656 children were randomly selected, and 1416 (21%) did not participate primarily because they were absent from the village the day of the exam; 5240 children from 4311 households were examined. Demographic characteristics of the participating children were slightly different than children who were non-participants (Table 1). Non- participants were more likely to be older, male, and live in houses where the water source was farther and the head of household had no former education. However, the two groups were similar in terms of the characteristics of the room where they slept. Non-participants tended to live in houses where the room with the cooking fire had ventilation. The majority, 99%, of households had an open fire stove with predominately wood or charcoal fuel.
The overall prevalence of active trachoma was 6·1%. Three year olds had the highest prevalence (9·5%). TI was present in all age groups and in 38% of the active trachoma cases, shown in Fig 1A. The overall C. trachomatis infection prevalence for these groups was 3.8% and prevalence by age is shown in Fig 1B.
A greater odds of active trachoma and severe trachoma was seen in children ages 1–5 years compared to the 6–9 year olds. After adjusting for age, a positive association was seen between active trachoma and severe trachoma and several indicators of socioeconomic status. Living in a household with no latrine, living farther from the water source, and in a house with a mud roof were all associated with active trachoma and severe trachoma (Table 2). Children whose head of household had no formal education were also significantly more likely to have active and severe trachoma. No increased risk was observed for children who spent time in the room with the cooking fire during cooking. However, we observed a dose response increase in odds of active trachoma, and especially severe trachoma, in children according to the proximity to the cooking fire and the degree of ventilation (Table 2). Compared to the lowest risk category, children who slept in a room without the cooking fire but no ventilation had an increased odds of trachoma, and children who slept in a room with a cooking fire had a 2·4 fold increased odds of active trachoma and a 5·5 fold increased odds of TI. in children who slept in a room with a cooking fire.
After multiple adjustments for age and the other variables, only longer distance to water remained a significant association with trachoma and severe trachoma. Living in a house with a mud roof, or no latrine, was no longer significant (Table 3). Children who slept in a room with a cooking fire were 1·8 times more likely to have active trachoma and 4·1 times as likely to have severe trachoma as children who slept in a room with ventilation and without a cooking fire.
Only the younger children were exposed to the cooking fire during the day while their mothers cooked, resulting in exposure confounded by age. We restricted the analyses to children age 0–5 years and examined the relationship between active and severe trachoma with exposure while cooking, according to the characteristics of the room (Table 4). Although the prevalence of active trachoma was highest in children who were exposed in a room with no ventilation, the test for trend was not significant.
We examined the infection rates in children according to trachoma status and type of sleeping room. We had few infections, but the trend for increasing infection in children with trachoma who slept in unventilated rooms or slept with a cooking fire was observed (Table 5).
Our study found a strong relationship between cooking fire exposure while sleeping and active trachoma, especially severe trachoma in children.
There is good biologic plausibility why trachoma may be higher in children exposed to indoor air pollution (IAP). In addition to increased tearing and irritation, which may result in auto-re-infection, IAP appears to have a direct effect on the immune system. In women who were users of biomass fuels, an increased TH2 response was described with an increase in the Treg cells, CD4+ and CD25+, a subset that can inhibit effector T cell response.[9] In this case, much of the regulatory activity is exerted by IL-10 and TGFβ, perpetuating the TH2 response and leading to chronic inflammation, and less clearance of infection.[7,8] Our data supports this finding, as we noted that infection rates were higher with TF in children who were exposed to a cooking fire, suggesting a prolongation of infection. Another study in children described the effects in T cell immunity from exposure to ambient polycyclic aromatic hydrocarbons, which are compounds produced from combustion of organic matter such as wood or coal. Epigenetic modifications associated with impaired immunity suggesting an increased TH2 response was observed as well.[17] For clearance of acute trachoma, an adequate CD4+ response of the Th1 phenotype appears to be necessary, and the Th1 cytokine gamma interferon assists in infection clearance. So not only might IAP lead to effects that delay clearance of infection, but might also increase risk for trachoma sequelae.
Previous studies of exposure to a cooking fire while cooking in women in non-trachoma areas reported an increasing eye irritation, conjunctival damage and tearing of the eyes.[5,6,18] Similar ocular effects were reported in children exposed to wildfire smoke exposure in California Ellegard and Diaz describe in their studies an increase in tear production or “tears while cooking” (TWC) in women and the direct association with indoor air pollution (IAP). The study also found a strong relationship between ocular symptoms in women who cooked inside a room, compared to women who cooked outside, suggesting a role for ventilation.[6] We did not find a statistically significant association of trachoma in children who were in the cooking fire room during cooking, although the greatest prevalence of trachoma in children age five years and under was in those in a cooking room with no ventilation.
In addition, the type of stove used also has been shown to play a role in eye symptoms in women. Smoke free stove versus traditional or open fire stoves have been assessed in relationship with the amount of IAP in the house and symptoms that the women experienced.[5,6] Women were found to have increased eye symptoms when exposed to open fire stoves especially those who use wood or charcoal as fuel. In our study population 99% of the women cooked using an open fire stove with predominately wood or charcoal, adding to the burden of indoor air pollution.
It is possible that the children who sleep in a room with cooking fire or rooms that are not ventilated belong to poorer families with less access to water and general sanitation. We tried to account for these socioeconomic factors in our study and found a strong, independent effect of sleeping in a room with a cooking fire.
Interestingly, we found that 3% of the children slept in a room with a cooking fire, compared to the results from the same district in 1986, where 60% of the children slept in a room with a cooking fire.[4] In 1986, the estimated prevalence of trachoma in children ages 1–7 years was 60%, and at that time, most of the children slept on animal skins in the room with a cooking fire. Since 1986, many changes have occurred, with most households now building a separate cooking fire area from the sleeping room, and using beds for sleeping. Trachoma has declined in this district, most markedly in the last eight years in connection with trachoma control measures, to an overall estimate last year of 12%. This change is encouraging, suggesting some environmental and socioeconomic improvement in these communities.
A difference between age groups was noted as well, where younger children had more TF and TI than older children. This age difference in clinical signs has previously been described.[19–27] and may reflect decreased exposure to re-infection as children reach school age. Of note, the older aged children did not spend time during the day in the room with the cooking fire, and we had to confine our analyses of exposure during cooking to the younger aged children.
There were some significant differences observed between the participating and non-participating children, the latter being older and more likely to be male, and to live in a house farther from a water source. However, the groups were similar in terms of exposure to cooking fire while sleeping. Since those who are older are less likely to have trachoma, but those who live far from water are more likely to have trachoma, the effects of these differences are uncertain. In any case, there was no difference in exposure to cooking fire, suggesting an absence of bias due to differential participation.
In addition, these communities are also undergoing mass drug administration. The last MDA was more than one year ago for these villages, but the low rates of trachoma, 6%, likely reflect at least in part the high compliance with this program. We were underpowered to detect a significant difference in infection rates in trachoma in children who slept in a room with a cooking fire and those who did not, although the trend was observed. The fact that Taylor et al in 1986 found a similar risk for sleeping next to a cooking fire in these communities prior to any intervention argues for an effect of exposure on trachoma even when disease rates are low or in the presence of a program with MDA.[4]
In conclusion we confirmed that those children who sleep in a room with a cooking fire have an increased risk of trachoma independently from other risk factors for disease. The prevalence of exposure to sleeping next to a cooking fire in this population has declined over time, and programs to encourage continued decline are warranted. Further studies on the impact of exposure to indoor air pollution and the risk of trachoma sequelae should be done.
|
10.1371/journal.pntd.0007460 | Zebra skin odor repels the savannah tsetse fly, Glossina pallidipes (Diptera: Glossinidae) | African trypanosomosis, primarily transmitted by tsetse flies, remains a serious public health and economic challenge in sub-Saharan Africa. Interventions employing natural repellents from non-preferred hosts of tsetse flies represent a promising management approach. Although zebras have been identified as non-preferred hosts of tsetse flies, the basis for this repellency is poorly understood. We hypothesized that zebra skin odors contribute to their avoidance by tsetse flies.
We evaluated the effect of crude zebra skin odors on catches of wild savannah tsetse flies (Glossina pallidipes Austen, 1903) using unbaited Ngu traps compared to the traps baited with two known tsetse fly management chemicals; a repellent blend derived from waterbuck odor, WRC (comprising geranylacetone, guaiacol, pentanoic acid and δ-octalactone), and an attractant comprising cow urine and acetone, in a series of Latin square-designed experiments. Coupled gas chromatography-electroantennographic detection (GC/EAD) and GC-mass spectrometry (GC/MS) analyses of zebra skin odors identified seven electrophysiologically-active components; 6-methyl-5-hepten-2-one, acetophenone, geranylacetone, heptanal, octanal, nonanal and decanal, which were tested in blends and singly for repellency to tsetse flies when combined with Ngu traps baited with cow urine and acetone in field trials. The crude zebra skin odors and a seven-component blend of the EAD-active components, formulated in their natural ratio of occurrence in zebra skin odor, significantly reduced catches of G. pallidipesby 66.7% and 48.9% respectively, and compared favorably with the repellency of WRC (58.1%– 59.2%). Repellency of the seven-component blend was attributed to the presence of the three ketones 6-methyl-5-hepten-2-one, acetophenone and geranylacetone, which when in a blend caused a 62.7% reduction in trap catch of G. pallidipes.
Our findings reveal fundamental insights into tsetse fly ecology and the allomonal effect of zebra skin odor, and potential integration of the three-component ketone blend into the management toolkit for tsetse and African trypanosomosis control.
| The use of repellents from non-preferred hosts represents an innovative approach to control animal African trypanosomosis by limiting contact between tsetse flies and livestock. Although zebras are non-preferred hosts, the possible chemical basis of their avoidance by tsetse flies is unknown. We hypothesized that certain chemical components of zebra skin odor play a role in their avoidance. We tested this hypothesis by screening crude zebra skin odor for repellency, identifying the chemicals detected by tsetse flies (G. pallidipes) in the skin odor matrix of zebra, and establishing the specific chemicals eliciting repellency in field trials. We identified a three-component blend as contributing to the repellency of the crude zebra skin odor. Repellency of the crude zebra skin odor and the three-component repellent blend was comparable to the known tsetse repellent, WRC (waterbuck repellent compounds), derived from waterbuck skin odor. Our study shows that odors play a role in the avoidance behavior of tsetse flies to zebras. The three-component repellent blend identified in zebra skin odor provides a new effective tool in the management of tsetse flies and trypanosomosis.
| Tsetse flies (Glossina sp.) feed exclusively on blood and are the sole cyclical vectors of the trypanosome parasites that cause African trypanosomosis, a neglected tropical disease [1–3]. Two forms of the disease exist in sub-Saharan Africa, both of which are major constraints to development: Human African Trypanosomosis (HAT) or sleeping sickness, affecting humans, and Animal African Trypanosomosis (AAT) or nagana, affecting livestock, especially cattle [3,4]. While successes have been achieved in recent times to control HAT, the livestock disease is still a huge burden [2] with devastating economic consequences [2,3,5]. Particularly, nagana discourages sustainable agriculture and accounts for at least 3 million cattle deaths annually [4], leading to USD 4.75 billion annual losses in crop and livestock production [3,6]. There is, therefore, an urgent need for enhanced and concerted efforts towards its control.
Control of AAT by chemotherapy has not been sustainable due to widespread and increasing resistance of trypanosomes to existing drugs [2,7], toxicity, relatively high cost and the use of counterfeit or sub-standard drugs in some areas [7,8]. Chemotherapy has also been jeopardized by the presence of wildlife trypanosome reservoirs which effectively maintains the transmission cycle that includes livestock [9,10]. Furthermore, there are no vaccines against any trypanosome pathogen, owing to their complex mechanism of antigenic variation [2,11]. Finally, the use of trypanotolerant cattle is not effective because these breeds are limited in geographical distribution and can lose trypanotolerance when under heavy tsetse densities [6]. Given these challenges, tsetse control efforts constitute a cornerstone in disease suppression and eradication efforts.
Several vector control methods are available for disease management [1,3]. However, strategies exploiting visual and chemical preferences of tsetse flies in traps and targets are the most cost-effective and are particularly promising [1,3]. Adult tsetse flies use a combination of odor and visual cues for host location, the latter being more important for landing [12,13]. By exploiting this host-seeking behavior, blue and black-coloured traps and targets combined with host attractants have been developed to suppress tsetse population by over 90% [3,13]. Here, the blue color of the traps and targets attracts tsetse flies, the black panel triggers landing responses, and the odor cues lure them beyond the vicinity of the trap [13]. For instance, in the Ngu tsetse trap, a blue and black paneled trap for Glossina pallidipes patented by icipe [14,15], cow urine and acetone are combined with the visual trap as odor cues to enhance trap attractiveness [14]. However, the application of traps and targets is limited to small defined areas [3] and offers no protection to freely-grazing livestock. More versatile vector control strategies addressing these loopholes are warranted for an effective animal African trypanosomosis control.
Innovative mobile tools amenable to African nomadic pastoralists to protect their livestock from tsetse bite and trypanosomosis infection has led to the development of the tsetse repellent technology [14,16]. Certain vertebrates such as waterbuck, zebra, wildebeest and impala are abundant in tsetse habitat but not preferred for blood meals [17–20]. This phenomenon has been strongly linked to allomones emanating from the skin of the non-preferred vertebrate hosts [21] and has been exploited in repellent technology. For instance, potent tsetse repellents have been identified from waterbuck by elucidating the odor basis of the avoidance of this bovid by tsetse flies [22,23]. These have further been harnessed into the development of a four-component tsetse repellent blend comprising geranylacetone, guaiacol, pentanoic acid and δ-octalactone (herein referred to as waterbuck repellent compounds; WRC). This blend is now in use as a tsetse repellent collar to protect livestock [16]. Besides its excellent efficacy, the tsetse repellent technology has been reported to be cost effective compared to available trypanocides [16].
Investigating other non-preferred vertebrate hosts could reveal cheaper, fewer-component repellent blend for tsetse flies with the same or superior potency as the known WRC [16]. Zebras constitute important non-preferred hosts of tsetse flies [17–19]. Previous studies have argued that the striped coats of zebra might play a role in their avoidance by tsetse flies [24,25]. However, the allomonal basis for this avoidance is not understood. Therefore, we tested the hypothesis that much like the waterbuck, zebra skin volatiles contribute to this avoidance behavior of the tsetse fly. Using field trials and chemical analyses, we show that zebra skin volatiles contribute to repellency in the savannah tsetse fly species (G. pallidipes) and we identify the compounds responsible.
This study was carried out between October 2016 and August 2018. For field evaluations, skin odors of zebra were collected in Nguruman (Kajiado County, Kenya) (Fig 1). Similar zebra odor sampling was conducted in Serengeti (Mara County, Tanzania) but only used for laboratory analysis. Nguruman maintains an abundant community of savannah tsetse flies (G. pallidipes Austen) and a forest species (G. longipennis) [22,26] whereas, in Serengeti, two savannah tsetse fly species (G. pallidipes and G. swynnertoni) are present [17]. Both areas have similar vegetation cover, ranging from open savannah grasslands to dense woodlands [17,26] suitable for the savannah species of tsetse flies [27] which was the focus of the present study. The choice of the study sites was informed by the co-existence of both zebra and tsetse flies in the same natural habitat. In addition to zebra, both areas have diverse community of wildlife species such as giraffe, wildebeest, elephant, antelope, buffalo, warthog, gazelle, waterbuck, bushbuck and non-human primates.
Adult plains zebras (Eguus quagga) were accessed in situ in Nguruman and Serengeti with the help of the Kenyan Wildlife Service (KWS) and the Tanzanian Wildlife Research Institute (TAWIRI) personnel, respectively. Zebras were immobilized and anaesthetized with 0.015mg/ kg body weight opiate etorphine hydrochloride (Captivon 98, Wildlife Pharmaceuticals Ltd, South Africa) in combination with 0.20mg/kg body weight azaperone (Kyron Laboratories, South Africa) [28]. All zebras were positioned on lateral recumbency to prevent respiratory complications and facilitate efficient working and sampling. Skin odors were collected from fourteen plains zebras including seven males and seven females by rubbing Soxhlet-extracted (with dichloromethane) and oven-dried cotton materials (23 cm x 23 cm, Lux Premium, Bidhannagar, West Bengal, India) on the belly and upper parts of the front legs where most tsetse flies are known to feed [29] for 10–12 min [30]. The zebra anaesthesia was reversed with intravenous injection of opioid antagonist diprenorphine (Activon, Wildlife Pharmaceuticals Ltd, South Africa) at a dosage rate of 0.045mg/kg body weight [28]. All animals recovered well with no complications after antagonist administration. To minimize contaminations with human odor and zebra fecal droppings, latex-gloved hands were used to collect skin odors and the rear legs, anal region and genitals of the animals were carefully avoided [31]. Collected odor samples were immediately wrapped in at least four layers of aluminum foil, placed in separate ziplock plastic bags and kept in a cool box underlaid with dry ice for use in field trapping experiments and laboratory analyses.
The Ngu tsetse trap (100% polyester, Vestergaard Frandsen, Lausanne, Switzerland) was used throughout this study. The traps were placed about 200 m apart and the odor dispensers were placed at the base of each trap approximately 30 cm downwind according to the standard procedures adopted for trapping tsetse flies [12,22].
We monitored trap catches of tsetse flies (G. pallidipes) in a series of randomized 3 x 3 Latin square designed experiments [12,22] with treatments comprising i) crude zebra skin odor, ii) WRC, a known tsetse repellent as negative control and iii) a combination of cow urine and acetone, known attractants, as positive control. Each treatment was replicated 30 times. Five pieces of cotton materials with crude skin odors of zebra were placed next to the Ngu tsetse trap in cylindrical canisters (diameter 9.5 cm, height 22.5 cm) fabricated from a stainless-steel wire mesh (1 mm x 1 mm x 1 mm sq mesh holes). WRC sachets (2 sachets of 4.5 ml per trap) were made of polyethylene (0.15 mm thickness, 50 cm2 surface area) folded and double-sealed into a tetrahedron shape [22]. Trap catches were recorded daily (between 1100 and 1300 hr, after the morning activity period which is usually between 0900 and 1100 hr) [32] to ensure captures of G. pallidipes both during their morning and afternoon activity periods. Daily trap catches were sorted by species and sex. The treatments were rotated daily according to the Latin square design and fresh zebra skin odors were used for each 24 h trapping period.
Aliquot samples of zebra skin odors were trapped on site on to adsorbent filters (Carbopak B 3.5” with 30 mg ± 5 mg, Sigma Scientific, Gainesville, Florida, USA) in a volatile entrainment system using a battery-powered field pump (assembled at the USDA/CMAVE, Gainesville, Florida, USA) [31]. Headspace odors from the cotton materials containing freshly collected zebra skin odor were trapped on to the adsorbent filter for 12 hr by the field pump preset to supply charcoal-filtered clean air at a flow rate of 348 ml/min. Adsorbent filters containing trapped odors were tightly sealed with Teflon tape, wrapped in aluminum foil in a cool box underlaid with dry ice and later transported to the laboratory in icipe, Duduville campus. Trapped skin odors were eluted from each adsorbent filter with 200 μl dichloromethane (≥ 99.9% Sigma-Aldrich, St Louis, Missouri, USA) and concentrated to 100 μl under a very gentle stream of charcoal-filtered nitrogen. Eluted skin odor extracts were kept at -80 °C until use.
Gas chromatography-electroantennographic detection (GC/EAD) analysis was carried out on a HP 5890 GC model fitted with a non-polar capillary column HP-1 stationary phase (30 m x 0.2 mm x 0.2 μm film thickness) and a flame ionization detector (280 °C) operated on a splitless injector mode (220 °C). Three three-day-old adult females G. pallidipes, reared in the laboratory under 12L:12D photoperiod, 25 ± 2 °C and 70 ± 5% relative humidity [33], were used. Individual insect that was fed two days earlier was immobilized on ice and the whole insect was mounted on a microscope stage between two capillary glass tubes filled with Ringer solution [34,35]. The tip of the insect antenna was gently inserted into the capillary tube on the recording electrode and to complete the circuit, the capillary tube at the reference electrode, grounded by an Ag-AgCl wire, was inserted to the base of the antenna. The column oven was held at 35 °C for 5 min, thereafter programmed to increase at 10 °C/min to 280 °C and maintained at this temperature for 10 min. An aliquot (2 μl) of the crude zebra skin odor extract was injected into the GC and components were separated on the column under the temperature programming mode with nitrogen as a carrier gas at 1.2 ml/min flow rate. Upon exit, nitrogen make-up gas was added to the column effluent, split 1:1 and delivered to the FID and the antenna of the mounted tsetse fly, via a stainless-steel delivery tube (5 mm ID), for simultaneous detection by the FID and EAD. The recordings were later analyzed with GC/EAD 2000 software (Syntech, Hilversum, the Netherlands). Commercially purchased authentic standards of identified EAD-active compounds were also analyzed under similar conditions as the crude odor samples.
Samples (1 μl) were analyzed on an Agilent GC/MS (Agilent technologies 7890A series). The GC was fitted with an autosampler, a split-splitless injection port (200 °C), an HP-5 Agilent fused silica capillary column (30 m length x 0.2 mm id x 0.22 μm film thickness), an Agilent technologies 5975C EIMS (electron energy 69.922 eV) triple axis mass selective detector (MSD), and an Agilent ChemStation data system [31]. The injection port ran on a splitless mode and the column oven ran on a temperature programmed mode as in the EAD but with helium as the carrier gas (1.2 ml/min flow rate, 8.8271 psi head pressure). After tentative identification of the EAD-active compounds in zebra skin odor extract, a final validation of identities was conducted using commercial standards (1 μl in 4000 μl dicholoromethane) made in into blends (aldehydes, ketones) in the EAD, and by comparing their GC retention indices and MS fragmentation patterns. Further, an external quantification of the identified EAD-active compounds was achieved using calibration curves prepared for each class of compounds over five different known concentrations within their expected range in the zebra odor extract. Nonanal was selected for the aldehydes, geranylacetone for the aliphatic ketones and acetophenone for the benzenoid ketone. These calibration curves were obtained using the peak area of each selected compound over three replicate runs for each concentration. Using the resulting linear equation from the calibration curves, the concentrations (ng/μl) and their natural ratio in the zebra odor were estimated.
The chemicals used included: 6-methyl-5-hepten-2-one (Aldrich, 99%), acetophenone (Sigma-Aldrich, ≥ 99%); geranylacetone (Aldrich, 65% geranylacetone and 35% nerylacetone), heptanal, octanal, nonanal and decanal (Aldrich, 95%).
To evaluate their effects on trap catches of G. pallidipes, seven identified EAD-active compounds—heptanal, 6-methyl-5-hepten-2-one, octanal, acetophenone, nonanal, decanal and geranylacetone—were tested in blends and singly in a series of Latin square-designed field experiments [12,22]. This field evaluation was carried out in three experiments described below. The compounds were used neat and to prevent oxidation, 10% antioxidant (2,6-di-tert-butyl-4-methylphenol, Aldrich, Gillingham—Dorset, UK) was added to each aldehyde before use [31]. Blends of compounds were constituted to mimic their natural ratio of occurrence in zebra skin odors (S1 Table) and dispensed from polyethylene sachets (0.15 mm thickness, 50 cm2 surface area) folded and double-sealed into a tetrahedron shape as used for WRC. Two 4.5 ml sachets of individual compounds and blends, pre-informed as the optimum repellent doses in initial trials with one, two and three sachets, were used per trap, except for the seven-component blend (Blend Z) in which three sachets gave the optimum repellency (S2 Table).
We evaluated the effect of a seven-component blend of the identified EAD-active compounds on field catches of G. pallidipes in Ngu traps combined with attractants (cow urine and acetone) in a 3 x 3 randomized Latin square designed experiments replicated 10 times. The treatments evaluated were: (i) attractant-baited (cow urine and acetone) Ngu trap (positive control), (ii) attractant-baited trap with WRC (negative control), and (iii) attractant-baited trap with Blend Z (7-component blend of EAD-active compounds mimicking natural zebra skin odor).
Here, we assessed the relative contribution of each EAD-active compound to the observed repellency in field trials. We used two sachets each containing 4.5 ml of the individual compounds, predetermined as optimum repellent dose based on trap catches as described earlier (S2 Table). For the aldehydes (heptanal, octanal, nonanal, decanal), a 6 x 6 Latin square design was used, and each treatment tested in 12 replicate trials comprising: (i) attractant-baited (cow urine and acetone) Ngu trap (positive control); (ii) attractant-baited trap with WRC (negative control); (iii-vi) attractant-baited trap with each of the four aldehydes, separately. For the ketones (6-methyl-5-hepten-2-one, acetophenone, geranylacetone), a 5 x 5 Latin square design was used and each treatment, listed as follows, had 10 replicates: (i) attractant-baited (cow urine and acetone) Ngu trap (positive control); (ii) attractant-baited trap with WRC (negative control); (iii-v) attractant-baited trap with each of the three ketones, singly.
In this experiment, blends of all seven EAD-active compounds, all four aldehydes and all three ketones were compared using a 5 x 5 Latin square with components tested as follows in 10 replicate trials: (i) attractant-baited (cow urine and acetone) Ngu trap (positive control); (ii) attractant-baited trap with WRC (negative control); (iii) attractant-baited trap with blend A (blend of four aldehydes); (iv) attractant-baited trap with blend K (blend of three ketones); (v) attractant-baited trap with blend Z (blend of all seven EAD-active compounds in zebra skin odor).
Additionally, we explored the possibility of having a 2-component tsetse repellent blend of ketones (using acetophenone and geranylacetone, which showed higher repellency than 6-methyl-5-hepten-2-one when tested individually) instead of a 3-component blend. The 3-component repellent blend was also compared with trap alone (unbaited Ngu trap). The experiment followed a 5 x 5 Latin square design having the following components evaluated in 20 replicates each: (i) attractant-baited (cow urine and acetone) Ngu trap (positive control); (ii) trap alone; (iii) attractant-baited trap with WRC (negative control); (iv) attractant-baited trap with Blend K (three-component blend of ketones); (v) attractant-baited trap with 2C Blend K (two-component blend of ketones—acetophenone and geranylacetone). All the blends tested above (blend A, blend K, blend Z, 2C blend K) were constituted to mimic the natural ratios of their occurrence in zebra skin odor (S1 Table). The dose used for each blend (number of sachets) was the optimum repellent dose determined in preliminary trials (S2 Table).
Daily trap catches of G. pallidipes for each treatment were analyzed in R (version R i386 3.2.3) using a generalized linear model with negative binomial error structure with abundance as response variable and treatment, day and site as predictor variables. Mean catches and standard error for each treatment were calculated. Means were separated using ‘lsmeans ()’ function, embedded in Least-Squares Means (lsmeans) R package, with Tukey adjustment for post hoc comparison. Further, the percentage reduction in catches for each treatment compared to the control was calculated using the mean catches. The higher the catch reduction, the more the treatment is avoided by savannah tsetse flies (G. pallidipes) and the better it is as a repellent for the fly. For all statistical tests α was set at 0.05.
The use of zebra in this study was approved by the KWS (permit number: KWS/BRM/5001) the TAWIRI, the Tanzania National Park (TANAPA) and the Commission for Science and Technology (COSTECH) (COSTECH permit number: 2016-223-NA-2016-96). In addition, consent was sought from community elders in Nguruman before traps were set in the forest.
A total of 326 G. pallidipes were caught (82.2% female, 17.8% male). Crude skin odors of zebra significantly reduced trap catch of G. pallidipes (66.7% catch reduction, IRR = 0.33, p < 0.001) when compared to a baited trap (cow urine and acetone, positive control). However, its repellent effect was not significantly different from WRC (58.1% catch reduction, IRR = 0.42, p < 0.001) (Fig 2). In addition to G. pallidipes, G. longipennis (a forest species of tsetse flies) was trapped. Fifty-one G. longipennis were caught in the baited trap and only four were caught in the trap containing zebra odor, translating to a 92.2% catch reduction (IRR = 0.08, p < 0.001). Similarly, ten G. longipennis were caught in the trap containing WRC (catch reduction = 80.4%, IRR = 0.20, p < 0.001).
In GC/EAD analysis, the antennae of three individual female G. pallidipes consistently detected seven components which were present in small amounts in the extract (Fig 3). These components were identified as the aldehydes heptanal, octanal, nonanal, decanal and ketones 6-methyl-5-hepten-2-one, geranylacetone, and acetophenone (Table 1). The identities of the EAD-active components were further confirmed using commercially-purchased standards, however, the antennal response to some of the synthetic compounds is reduced as compared to components detected in the zebra extract (S1 Fig). Notably, the response of the olfactory sensory neuron (OSN) was higher for the initial stimulations but this response declined with stimulations from the latter eluting ketones (S1 Fig).
In the experiment evaluating the seven-component blend of all EAD-active compounds on field catches in baited (cow urine and acetone) traps, a total of 597 G. pallidipes were collected (62.0% female, 38.0% male). As observed for crude zebra skin odor, the 7-component blend resulted in significant reduction (p < 0.001) in the catch of G. pallidipes [catch index = 0.42, 95% confidence interval, CI (0.28–0.61)] compared to the baited trap alone (positive control). The reduction in catches was recorded for both males and females of G. pallidipes which were not significantly different from that exhibited by the WRC (Fig 4).
In the experiment examining the effects of the individual compounds, a total of 2,693 G. pallidipes were caught (67.7% female, 32.3% male). The individual aldehydes (heptanal, octanal, nonanal, decanal) had no significant effect on G. pallidipes catches in baited traps (positive control). However, for the individual ketones, acetophenone (CI = 0.57) and geranylacetone (CI = 0.57), demonstrated significant reduction in catches (Table 2).
A total of 996 G. pallidipes were caught (62.6% female, 37.4% male) in the experiment comparing the identified compounds grouped by class. Baited Ngu traps combined with the four-component blend of the aldehydes (blend A) did not significantly impact catch reduction of G. pallidipes compared with the baited trap alone (positive control). On the other hand, baited traps combined with the three-component blend of ketones (blend K) significantly reduced the catch of G. pallidipes (Catch index 0.39, p < 0.001), which compared favorably with the trap catch recorded for the seven-component blend and WRC (Table 3).
An additional experiment which compared the 3-component blend of ketones above with a 2-component blend (acetophenone and geranylacetone) with baited trap (cow urine and acetone) as positive control, recorded 1,658 G. pallidipes (72.4% female, 27.6% male). The two-component blend (2C blend K) reduced field catches of G. pallidipes (IRR 0.59) and the performance was, again, not significantly different from WRC (IRR 0.71). However, blend K (the 3-component ketone blend) performed better than the 2-component blend (IRR 0.42) in repelling G. pallidipes. Catches of G. pallidipes in baited traps, combined with either “blend K”, “2C blend K” or “WRC”, were not significantly different from the catches in the trap alone (Table 4).
Exploiting odors of non-preferred vertebrate hosts of tsetse flies for repellents represents a sustainable innovative strategy for the control of both Human and Animal African Trypanosomosis [16]. Here, we established the presence of potent repellents in zebra skin odor for male and female savannah tsetse fly G. pallidipes. This observed repellency was maintained regardless of gender biases in field captures of G. pallidipes which may relate to differences in their population due to seasonal variations. This is particularly important since both male and female tsetse flies feed exclusively on blood and can transmit African Trypanosomosis [3]. The observed repellency of crude skin odor of zebra and a 7-component blend of the identified EAD-active compounds simulating their natural ratio of occurrence in zebra skin odor strongly supported our study hypothesis. Further, we formulated and identified a 3-component blend of ketones repellent to G. pallidipes. Subject to further field evaluation for performance across different seasons and ecologies, this blend could be used to protect livestock, particularly cattle, from tsetse bites and trypanosome infections.
We found that the crude skin odor of zebra was effective in reducing field trap catches of savannah tsetse flies, like the known tsetse repellent, WRC. Previous research has shown that zebras, although present in tsetse habitat, are usually avoided by these flies [17–19]. A previously proposed hypothesis for this avoidance suggests that the polarization effects of the striped pelage of zebras is the driving component of the observed avoidance [24]. However, the stripes of this ungulate is visible to tsetse flies only at a distance of about 5–10 m and beyond this distance, zebras appear uniformly grey to these flies [36]. This indicates that fitness benefits conferred by the stripes on zebra skin against tsetse flies are only within the boundaries of this proximity. Also, tsetse flies utilize odor cues in locating suitable hosts and discriminating potentially unsuitable vertebrates [21]. In this case, the odor cues might complement visual cues in conferring fitness benefits beyond the vicinity of zebras. Our study shows that odor cues from zebra skin contribute to the avoidance behavior of tsetse flies to this equid. This finding is not characteristic of zebras alone. Skin odors of non-preferred vertebrates of hematophagous arthropods are known to contain potent repellents. For example, beagle dogs are avoided by the brown dog tick [37]. Similarly, chickens are avoided by Anopheles mosquitoes [38], and waterbucks are avoided by tsetse flies [23]. The skin odors of each of the aforementioned contain repellents for these blood feeding arthropods [23,37,38] and are being exploited for their potential to control these flies. Our study does not rule out the importance of visual cues in the avoidance behavior of tsetse flies to zebras but shows the contribution of chemical cues from the skin odor. Future studies can focus on exploiting possible synergy between visual and odor cues in this avoidance behavior and its possible integration as a novel strategy in protecting livestock hosts from tsetse bites. More research can also be conducted to include other sources of odors including zebra breath, urine and dung.
Our current study identified key EAD-active aldehyde and ketone components of zebra skin odor extract, which were present at different concentrations and showed varying magnitudes in G. pallidipes antennal responses. The magnitude of antennal responses depends on several factors such as the nature and concentration of the stimulus, number and strength of previous stimulations, insect species and the quality of the antennae [34]. Furthermore, the distribution of the olfactory sensilla for some compounds are localized [39], and antennal response could be affected by the mounting technique. This could contribute to the lower response to heptanal compared to the latter aldehyde stimulations. Also, adaptation due to repeated stimulation of the olfactory sensory neurons (OSNs) as the individual compounds elute from the GC column is possible. As shown both in insect and mammalian olfactory sensory neurons, repeated and longer stimulation with high concentration induce OSN adaptation [40–42]. This adaptation due to repeated stimulations and higher concentrations could explain the reduction in responses to the latter-eluting compounds. For example, as noted for the response to geranylacetone when the commercially-purchased ketones were used as stimulus compared to the natural zebra skin odor extract. The reduced EAD response for subsequent GC stimulation could also enable us to hypothesis that the ketones most probably are detected by the same receptor [43]. However, electroantennographic detection by an insect does not necessarily translate to behavioral responses and such data are better used as a qualitative indicator of antennal responses [35]. For instance, in the moth Manduca sexta, odor detection by females did not translate to increased behavioral response of the females compared to males [44]. Therefore, a detailed behavioral assay with the identified EAD-active compounds is required, as conducted in our study.
The identifed EAD-active ketones (geranylacetone, acetophenone and 6-methyl-5-hepten-2-one) and the aldehydes (heptanal, octanal, nonanal and decanal) are constituents of skin odors of certain vertebrates, and different blood feeding insects will show variant behavioral responses to these odors [22,33,38,45–47]. Geranylacetone is present in waterbuck skin odor [33] and a key contributor to the repellency of WRC [22]. This compound has also been shown to repel effectively several mosquito disease vectors [46]. It could be prudent to investigate the probable utilization of geranylacetone as biomarkers for potentially unsuitable vertebrates for feeding to tsetse and other hematophages. Also, the repellence of acetophenone to tsetse flies has previously been established [47]. The ketone 6-methyl-5-hepten-2-one is a signature component of human odor [48], and has been shown to have either attractive or repellent effects on mosquitoes, depending on the dose and formulation [45,46,49]. In our study, 6-methyl-5-hepten-2-one alone did not impact trap reduction of tsetse flies. However, when combined with acetophenone and geranylacetone, it contributed to the repellency of the 3-component blend by increasing the catch reduction by about 50%. We recorded a more significant repellency when the three ketones were formulated into a blend, than as individual compounds, thus suggesting a combined activity of these chemicals. Also, 6-methyl-5-hepten-2-one could have some combined effects on lactic acid which is present in human skin odor, and was previouly shown to be the major chemical responsible for the avoidance of humans by tsetse flies [50]. The identified EAD-active aldehydes (heptanal, octanal, nonanal, decanal) are commonly found in skin odors of vertebrates, and are attractive to blood feeding insects [31,33,38]. For example, they are present in skin odors of preferred (buffalo, cattle) and non-preferred (waterbuck) vertebrate hosts of tsetse [33]. In previous laboratory experiments, these aldehydes were suggested as attractants for tsetse flies [23]. It was therefore not surprising that the aldehydes had no significant effect on field catches of tsetse flies in cow urine and acetone-baited Ngu traps when used either alone or in blends.
The empirical application of tsetse repellents in disease control has recently been demonstrated [16]. Collars containing repellent chemicals worn by cattle can be used to protect them from tsetse bites and trypanosomosis infection either alone, as these livestock are grazed in tsetse infested areas, or in a push-pull strategy, in which the repellent collars serve as push and a attractant-baited trap/target serve as a dead-end pull. This approach will not only reduce disease incidence and trypanocide use, but also reduce tsetse population over time. This is particularly advantageous as the tsetse repellent technology becomes more efficient as the disease incidence and tsetse population reduce [51]. Repellents for tsetse flies can also be successfully applied in areas where trypanotolerant cattle are in use, and in the control of HAT because of the characteristic low infection rates of Trypanosoma brucei species [51]. Key strengths of this approach over the existing ones are its mobility and ease of use which have triggered Kenyan livestock farmers’ interest in embracing the technology in tsetse and trypanosomosis control [16,51]. Like the existing tsetse repellent, WRC, the newly identified repellent blend also contains geranylacetone, but the other components are different. However, unlike the WRC which contains a ketone, an alcohol, an acid and a lactone, the repellent blend identified in our study consists of only ketones and are more likely to maintain their integrity for longer periods of time, although this needs to be established in future studies. More specifically, the blend is made up of three components and may offer a cheaper, yet potent, alternative to WRC. Also, the new repellent blend could have synergistic effect if combined with WRC though this needs to be determined in further studies. This repellent technology could be combined with other tsetse and African trypanosomosis control strategies for their integrated management.
We investigated the chemical basis of the interactions between zebras and tsetse flies and established that odor contributes to the avoidance behavior of tsetse flies to zebras. Our results show that, like the known tsetse repellent WRC, ketones present in zebra skin odor are largely responsible for the tsetse fly avoidance behavior. The three-component ketone blend can be incorporated into the tsetse repellent technology to protect livestock, particularly cattle, from tsetse bites and trypanosome infections. We recommend multi-site field testing of this newly identified tsetse repellent blend and evaluation of their effect on disease incidence in livestock in tsetse-infested areas. Finally, we recommend future studies to explore a possible interaction between odor and vision in the tsetse-zebra interactions and the comparison of skin odor profiles of zebra with other equids that are relatively more attractive to tsetse flies.
|
10.1371/journal.ppat.1005728 | Preventing Vaccine-Derived Poliovirus Emergence during the Polio Endgame | Reversion and spread of vaccine-derived poliovirus (VDPV) to cause outbreaks of poliomyelitis is a rare outcome resulting from immunisation with the live-attenuated oral poliovirus vaccines (OPVs). Global withdrawal of all three OPV serotypes is therefore a key objective of the polio endgame strategic plan, starting with serotype 2 (OPV2) in April 2016. Supplementary immunisation activities (SIAs) with trivalent OPV (tOPV) in advance of this date could mitigate the risks of OPV2 withdrawal by increasing serotype-2 immunity, but may also create new serotype-2 VDPV (VDPV2). Here, we examine the risk factors for VDPV2 emergence and implications for the strategy of tOPV SIAs prior to OPV2 withdrawal. We first developed mathematical models of VDPV2 emergence and spread. We found that in settings with low routine immunisation coverage, the implementation of a single SIA increases the risk of VDPV2 emergence. If routine coverage is 20%, at least 3 SIAs are needed to bring that risk close to zero, and if SIA coverage is low or there are persistently “missed” groups, the risk remains high despite the implementation of multiple SIAs. We then analysed data from Nigeria on the 29 VDPV2 emergences that occurred during 2004−2014. Districts reporting the first case of poliomyelitis associated with a VDPV2 emergence were compared to districts with no VDPV2 emergence in the same 6-month period using conditional logistic regression. In agreement with the model results, the odds of VDPV2 emergence decreased with higher routine immunisation coverage (odds ratio 0.67 for a 10% absolute increase in coverage [95% confidence interval 0.55−0.82]). We also found that the probability of a VDPV2 emergence resulting in poliomyelitis in >1 child was significantly higher in districts with low serotype-2 population immunity. Our results support a strategy of focused tOPV SIAs before OPV2 withdrawal in areas at risk of VDPV2 emergence and in sufficient number to raise population immunity above the threshold permitting VDPV2 circulation. A failure to implement this risk-based approach could mean these SIAs actually increase the risk of VDPV2 emergence and spread.
| Global, coordinated withdrawal of serotype-2 OPV (OPV2) is planned for April 2016 and will mark a major milestone for the Global Polio Eradication Initiative (GPEI). Because OPV2 withdrawal will leave cohorts of young children susceptible to serotype-2 poliovirus, minimising the risk of new serotype-2 vaccine-derived poliovirus (VDPV2) emergences before and after OPV2 withdrawal is crucial to avoid large outbreaks. Supplementary immunisation activities (SIAs) with trivalent OPV (tOPV) could raise serotype-2 immunity in advance of OPV2 withdrawal, but may also create new VDPV2. To guide the GPEI strategy we examined the risks and benefits of implementing tOPV SIAs using mathematical models and analysis of data on the 29 independent VDPV2 emergences in Nigeria during 2004–2014. We found that in settings with low routine immunisation coverage, the implementation of a small number of tOPV SIAs could in fact increase the probability of VDPV2 emergence. This probability is greater if SIA coverage is poor or if there are persistently unvaccinated groups within the population. A strategy of tOPV SIA in sufficient number and with high coverage to achieve high population immunity in geographically-focused, at-risk areas is needed to reduce the global risk of VDPV2 emergence after OPV2 withdrawal.
| Global and synchronous withdrawal of all live-attenuated oral poliovirus vaccines (OPV) is one of the major objectives of the global Polio Eradication & Endgame Strategic Plan 2013–2018 [1] and part of the global transition from OPV to inactivated poliovirus vaccine (IPV). Serotype 2 will be the first to be removed, with a planned date of April 2016. This means that trivalent OPV (tOPV) will be replaced by bivalent OPV (bOPV, containing Sabin virus types 1 and 3) in routine immunisation programmes, and tOPV will no longer be used in supplementary immunisation activities (SIAs). Furthermore, all OPV-using countries are recommended to introduce at least one dose of IPV in their routine immunisation programmes before the switch from tOPV to bOPV [2].
OPV use needs to be stopped because of its genetic instability. Attenuated vaccine (Sabin) polioviruses lose key genetic determinants of attenuation through mutation and/or recombination with other enterovirus serotypes during replication in the human gut [3]. In countries using OPV, approximately 1 child per 900,000 first OPV doses is estimated to develop vaccine-associated paralytic poliomyelitis (VAPP) [4]. The relative contributions of viral evolution (loss of key attenuating sites), immune function of the vaccine recipient and chance in the aetiology of VAPP are unclear. More significantly for the Global Polio Eradication Initiative (GPEI), vaccine polioviruses may spread from the recipient to his or her contacts, in rare cases leading to an outbreak of a vaccine-derived poliovirus (VDPV). VDPVs are defined as OPV-related isolates whose ~900-nucleotide sequence encoding the major capsid protein VP1 differs from that of the parental strain by >1% for serotypes 1 and 3, and >0.6% for serotype 2 [5]. VDPVs are classified into three categories: circulating VDPVs (cVDPVs), when there is evidence of person-to-person transmission; immunodeficiency-associated VDPVs (iVDPVs), shed by individuals with primary immunodeficiencies who have prolonged, sometimes chronic, virus excretion; and, ambiguous VDPVs (aVDPVs), which are isolates that cannot be classified as cVDPV or iVDPV despite thorough investigation [6,7]. Until July 2015, the definition of cVDPV required that genetically linked VDPVs were isolated from at least two AFP cases, but the GPEI now considers even single individual or environmental sample isolates to be cVDPV if their genetic features indicate prolonged circulation [7].
cVDPVs have transmission dynamics similar to wild polioviruses [8]. Since 2006, more than 680 acute flaccid paralysis (AFP) cases due to VDPVs have been reported worldwide [9], underlining the importance of VDPVs for the polio eradication endgame. Strikingly, >97% of those cases have been associated with serotype 2 [9], whose wild counterpart was last detected in 1999 [10]. The burden of serotype 2 VDPV (VDPV2) and the eradication of serotype 2 wild poliovirus (WPV2) in 1999 are the main motivations for the global withdrawal of serotype 2 OPV (OPV2) planned for April 2016.
Polioviruses spread where levels of immunity in the population are low and where environmental conditions such as sanitation and crowding facilitate virus transmission. As such, the detection of cVDPVs has historically been associated with poor population immunity [3,8,11–15]. However, the initial appearance of a VDPV in a population depends on different factors and the relationship with population immunity may be more complex. For example, the number of people infected with Sabin poliovirus, the duration of excretion among those infected, the extent of secondary transmission and the prevalence of other enteroviruses may all be important in determining the probability of VDPV emergence.
Worldwide OPV2 withdrawal will put the 155 countries currently using tOPV in their routine immunisation programmes at risk of outbreaks of VDPV2 given the associated increase in the number of children susceptible to that type. SIAs with tOPV prior to OPV2 withdrawal would increase population immunity to serotype 2 and have been proposed as a strategy to mitigate the risk of VDPV2 emergence and spread [16]. However, infrequent or poor-coverage SIAs could lead to limited immunity and potentially an adverse increase in risk resulting from poliovirus shedding and seeding of new VDPV. A better understanding of the factors associated with the risk of VDPV emergence and subsequent spread will help the GPEI to define a clear strategy on the number, timing and geographic extent of any tOPV SIA that minimises the risk of VDPV2 emergence at the time of and immediately after OPV2 withdrawal. Defining such a strategy is one of the priorities of the polio eradication program.
In this article, we first present mathematical models that describe the relationship between the coverage of routine and supplementary immunisation activities, and the probability of VDPV emergence and subsequent spread. To test the conclusions from the mathematical models, we then identified risk factors associated with past VDPV2 emergences in Nigeria. For this aim, we carried out a case-control analysis of those districts reporting the first case of poliomyelitis associated with each of the 29 independent VDPV2 emergences in Nigeria during 2004−2014 compared with districts without emergences. We also used logistic regression to identify the risk factors associated with the probability that a VDPV2 emergence resulted in >1 case of poliomyelitis. We finish by discussing the implications of our findings for the tOPV SIA strategy to reduce the risk of VDPV2 emergence during and post OPV2 withdrawal.
The number of people infected with Sabin polioviruses is primarily determined by the number of doses of OPV administered during routine and supplementary immunisation activities and the level of population immunity. As the amount of OPV administered increases from zero, the number of Sabin-infected individuals will initially increase, but at some point further increases in OPV administration are likely to result in a decrease in the number of individuals infected because of the associated increase in the level of population immunity. This implies a trade-off in the levels of OPV use that will favour VDPV emergence. We developed two mathematical models to study this trade-off: an analytical model that includes only SIAs, and a more complex model that includes both routine immunisation and SIAs, which must be solved through numerical simulation. We used these models to investigate the risks and benefits of carrying out preventive campaigns with tOPV as a strategy to maximise population immunity to serotype 2 prior to OPV2 withdrawal.
We explored the probability of a VDPV outbreak for different numbers of supplementary campaigns in a scenario without routine immunisation and considering a population completely susceptible (Fig 2). Using the analytical model, if the individuals reached at each campaign are randomly chosen (assuming the same coverage at each campaign), the risk of a VDPV outbreak is maximised at low and intermediate levels of SIA coverage (Fig 2A). The exact location of the peak in risk depends on the number of SIAs, rapidly shifting to lower values of SIA coverage for increasing number of SIAs. In particular, for a single SIA with 100% coverage, the probability of a VDPV outbreak is around 70%, which is explained by the relatively small proportion of children that will be protected after the campaign, due to the limited (~50%) immunogenicity of OPV. As expected, the size of any resulting outbreak is also significantly smaller for increasing number of SIAs (Fig 2D). Although the absolute risk of VDPV emergence depends on the assumed probability of reversion of Sabin poliovirus to a VDPV (ρ) and the assumed population size (N) via σ = ρN, the location of the peak in risk does not change unless the value of σ is so low or high as to make VDPV emergence impossible or inevitable respectively. We provide a sensitivity analysis of the probability of VDPV outbreak to the value of σ in Fig B in S1 Text. In particular, when σ becomes sufficiently large, the probability of a VDPV outbreak becomes a stepwise function of SIA coverage (Fig B in S1 Text).
If the same individuals are reached at each campaign, thus leaving a “missed” group that is only immunised through secondary spread of Sabin virus, the risk of a VDPV outbreak is maximised at intermediate levels of SIA coverage. More importantly, increasing the number of SIAs above 4 barely reduces the risk, which becomes zero only above 70% coverage after 4 or more SIAs (Fig 2B). In other words, there is a threshold in SIA coverage under which the risk does not decrease despite increasing the number of campaigns. The existence of this threshold can be shown analytically (Section A.2.3 in S1 Text), and for both random and fixed coverage, an expression for the minimum SIA vaccine coverage required to have zero probability of outbreak can be found (Section A.2.3 in S1 Text). A sensitivity analysis of the minimum SIA coverage required for zero probability of a VDPV outbreak to a broad range of values of the reproduction number of Sabin virus and the reproduction number of VDPVs is shown in Figs D and E in S1 Text. As expected, the minimum SIA coverage to bring the probability of a VDPV outbreak to zero increases for increasing values of the reproduction number of VDPVs, however, it slightly decreases for increasing values of the reproduction number of Sabin virus, because of the associated increase in the number of individuals who will be immunised by secondary spread of OPV from vaccinees.
The risk of observing an outbreak of VDPV obtained with the more complex model in the absence of routine immunisation coverage displays a shape similar to that obtained with the analytical model, although stochastic extinction results in a lower risk at low values of SIA coverage (Fig 2C).
The stochastic SIR model allows the study of the risk of VDPV2 outbreak in the context of OPV2 withdrawal. Including reasonable levels of routine immunisation coverage results in a significant reduction in the risk of VDPV2 outbreaks during the 6 months that follow OPV2 withdrawal (Fig 3A and 3B). This risk becomes almost negligible when routine coverage is high (Fig 3B), but for low levels of routine immunisation coverage, multiple tOPV SIAs preceding OPV2 withdrawal are needed to avoid seeding new VDPV2 (Fig 3C and 3D). Notably, in the context of low routine immunisation coverage, a single SIA seems to highly increase the risk irrespective of SIA coverage, and at least 3 campaigns at high coverage are needed to bring that risk close to zero (Fig 3C and 3D). If campaign coverage is only intermediate and there is a persistently “missed” group (i.e. SIAs reach the same individuals at each round), the risk of VDPV2 outbreak remains high even after 4 or 5 SIAs (Fig 3D).
A total of 29 independent VDPV2 emergence events were identified in Nigeria during the study period, of which 7 resulted in more than one case of poliomyelitis (Fig 4A, Table C in S1 Text). This resulted in 28 cases in the case-control analysis, since two emergences took place in the same district during the same 6-month period (Maiduguri, Borno state, between April and September 2006). The 28 cases were matched to 560 controls.
The number of tOPV SIAs in the previous 6 months varied between 0 and 5 (Fig H in S1 Text). Important changes over time occurred, due to a progressive and rapid removal of tOPV from SIA since 2006, which was replaced by bOPV, mOPV1 and mOPV3 [29]. tOPV was re-introduced in SIAs since mid-2009. These changes over time were also reflected in estimated serotype-2 immunity among children 0–2 years old, which reached very low levels in 2008 and 2009, and increased again from 2010 onwards (Fig G in S1 Text). In general, serotype-2 population immunity was higher in Southern districts. Routine immunisation coverage was also higher in the South and increased during the study period (Fig 4B).
The annual number of births per district was highly variable, ranging from 42 children (Bakassi, Cross River state) to 57,710 children (Alimosho, Lagos state), with a median of 7,542 (Fig I in S1 Text). Population density was also highly variable, ranging between an average of 9.37 (Teungo, Adamawa state) and 55,450 people per km2 (Ajeromi-Ifelodun, Lagos state), with a median of 218.70 (Fig J in S1 Text). The mean number of household members per district remained nearly constant over the study period, ranging between 3.26 and 6.31, and displayed a North-South gradient (Fig Q in S1 Text).
In the univariable analyses, a number of variables were associated with cases of VDPV2 emergence: (i) geographic region (North vs. South), (ii) serotype-2 population immunity, (iii) routine immunisation coverage, (iv) number of tOPV campaigns in the previous 6 months, (v) number of months since the last tOPV campaign, (vi) number of births, and (vii) number of household members (Table 1). Districts in the North had an increased risk of VDPV2 emergence compared to the South (odds ratio 5.52, [95% confidence interval 1.89−16.16]). This association may well reflect the existence of a North-South gradient in Nigeria for many demographic, social and economic variables [30]. The number of births was also associated with cases of VDPV2 emergence as it is a proxy for the size of the population exposed to OPV. Interestingly, among the variables related to OPV use, increased population immunity, routine immunisation coverage and the number of months since the last tOPV SIA were associated with a reduced odds of VDPV2 emergence. However, the number of campaigns in the previous 6 months was associated with an increase in the odds of VDPV2 emergence (Table 1).
The best multivariable model (lowest AIC, 146.63) retained two variables statistically significantly associated with cases of VDPV2 emergence: routine immunisation coverage and the annual number of births (Table 1). In this model, an absolute increase of 10% in routine immunisation coverage was estimated to reduce the odds of VDPV2 emergence by 31%. Adding the number of tOPV SIAs in the previous 6 months to the best model gave a very similar AIC (148.08), but the variable was not statistically significant (odds ratio 1.54, [95% confidence interval 0.47−5.07]).
The seven VDPV2 emergences that established circulating lineages (>1 case of poliomyelitis) occurred in districts with low to middle serotype-2 population immunity (<55%) and low routine immunisation coverage (<20%) (Fig 5). A univariable logistic regression analysis found that the probability of an emergent VDPV2 to establish a circulating lineage decreased for higher serotype-2 population immunity (p = 0.051). The other variables did not show a statistically significant association with the probability of a VDPV2 to establish a circulating lineage.
This study presents an analysis of the risk factors associated with the emergence and spread of VDPV, and provides a basis for strategic decisions about the optimal extent and number of mass campaigns with OPV in advance of OPV withdrawal.
First, using two simple mathematical models, we describe a trade-off between OPV use and the risk of VDPV emergence. Our findings indicate that immunity provided through routine immunisation counterbalances well the risk of VDPV emergence and spread. However, we found that in settings where routine immunisation coverage or the baseline level of population immunity is low, a small number of SIA campaigns could increase the risk of VDPV emergence compared to no campaigns. This is partly due to the low immunogenicity of OPV that makes necessary a certain number of SIA rounds to increase population immunity to levels that counterbalance the risk of seeding new VDPV through those campaigns. For example, our model predicted that in a setting with just 20% routine coverage with three tOPV doses (e.g. many districts in northern Nigeria), a single OPV SIA increased the risk of VDPV outbreak irrespective of SIA coverage (Fig 3). To bring population immunity to levels that completely counterbalanced the risk of VDPV outbreak, at least three rounds of supplementary campaigns at 80% coverage were needed. If only intermediate levels of campaign coverage were attained and SIAs persistently reached the same population leaving a persistently “missed” group, the risk of VDPV emergence remained high even when a high number of campaigns were implemented (Fig 3D). The existence of this threshold in SIA coverage under which the risk cannot decrease despite an increasing number of SIA illustrates how groups of unvaccinated children may hamper the efforts to minimise the risk of VDPV after OPV withdrawal.
Second, we identified risk factors associated with VDPV2 emergence and subsequent spread in Nigeria using epidemiologic, virologic and demographic data for 2004−2014. In both univariable and multivariable analyses, districts reporting the first case of poliomyelitis associated with a given VDPV2 emergence were more likely to have low routine immunisation coverage and a higher number of births. These districts were also more likely to have had a higher number of tOPV SIA in the previous 6 months, although this association was not statistically significant in the final multivariable model. This may be a result of the small number of observations, or a confounding between routine immunisation and tOPV campaigns, which are used to fill the immunity gaps. Finally, we also found that VDPV2 emergences were more likely to establish a circulating lineage and thus be responsible for more than one AFP case when they emerged in districts with low serotype-2 population immunity. Interestingly, the VDPV2 emergences that established a circulating lineage (7/29) occurred in districts with estimated serotype-2 population immunity <55%.
The statistical analyses of data on VDPV2 emergence in Nigeria are consistent with our transmission model results in suggesting that tOPV SIAs can in some settings increase the risk of VDPV emergence. Past experience has also highlighted how limited use of OPV (Sabin or other attenuated strains) either during small clinical trials (e.g. Poland [31,32]) or vaccination programmes (e.g. Byelorussia, former USSR [33]) can lead to widespread circulation of VDPV and outbreaks of poliomyelitis [31–33]. In Nigeria, the association that we found could be explained either by an insufficient number of campaigns, low coverage of those campaigns (settings with a higher number of campaigns may have poorer campaign coverage) or the existence of persistently “missed” populations, leading to insufficient levels of population immunity to avoid seeding new VDPV. Introduced in 2009 to monitor the quality of SIAs, lot quality assurance sampling (LQAS) showed that SIA coverage in 65% of districts in Nigeria did not reach 60% by the end of 2009 [34], suggesting that possibly only intermediate levels of SIA coverage were reached during the first half of the study period. Promisingly, significant improvements in SIA coverage have been reported since [34].
There are several limitations to our analyses. Firstly, the results of the case-control analyses were limited by the small number of emergence events in Nigeria, resulting in wide confidence intervals for some odds ratios. However, we chose Nigeria because it has experienced the greatest number of recorded VDPV2 emergences, with each VDPV2 undergoing detailed genetic sequencing and molecular epidemiological analysis [23]. Secondly, we were only able to examine risk factors associated with the first reported case of poliomyelitis caused by a VDPV2, rather than the initial emergence of the VDPV2. These VDPV2 isolates had between 6 and 17 nucleotide substitutions in the VP1 coding region, corresponding to an estimated average time of circulation since the initiating tOPV dose of around 9 months [23], thus leaving the possibility that the district where the first AFP case associated to a given emergence was detected did not correspond to the district where the initiating tOPV dose was administered. Thirdly, although our simple mathematical model is mechanistic–describing OPV transmission and VDPV emergence–it does not attempt to capture the detailed genetic changes that result in reversion of Sabin poliovirus to a VDPV with transmissibility and virulence equivalent to wild-type virus [8]. Attenuating mutations in the Sabin polioviruses have been identified, but the process of reversion and the significance of genetic changes for poliovirus transmission are not well understood [3]. Instead, we chose to capture the process of reversion by a simple probabilistic process and derive results that are likely to be robust to the details of genetic reversion. Fourth, although we considered models of immunisation that included persistently “missed” populations, we did not explicitly consider geographic heterogeneity in coverage and risk. Areas with poor routine immunisation coverage are also often challenging places to deliver vaccine during mass campaigns. These heterogeneities in coverage are therefore likely to increase the risk of VDPV2 emergence associated with tOPV SIAs, and it may be advisable to increase the number of campaigns in areas with heterogeneous coverage to account for this risk, as has been necessary during the eradication of wild-type polioviruses. Finally, other variables that could play a role in VDPV emergence such as hygiene behaviour, sanitation or the prevalence of other enterovirus serotypes (which could act as partners for recombination with Sabin viruses) were not included in the analysis of data from Nigeria, because they were not available at the district level.
Our transmission models do not account for the introduction of 1 or more doses of IPV into routine immunisation, which is a pre-requisite for OPV2 withdrawal [1]. Most countries currently using OPV only will introduce a single IPV dose at 14 or 16 weeks. This is likely to protect about 50% of vaccine recipients against poliomyelitis [35]. However, it is unclear what impact this vaccine will have on poliovirus transmission, which may well be limited as a result of the poor mucosal protection induced by this vaccine [36]. In particular, children born post OPV2 cessation (thus non-exposed to live serotype-2 virus) may not benefit from the boost in mucosal immunity induced by IPV [37,38]. Therefore, the impact of IPV on VDPV2 emergence and transmission may be limited [39], leaving populations at risk of silent transmission of poliovirus. Environmental surveillance will play a major role during the polio endgame, in particular to detect highly divergent Sabin-2 viruses that could be silently circulating and help to determine the extent of possible SIAs with monovalent OPV2 to control the spread of those and avoid new VDPV2 outbreaks.
Taken together, our results have important implications for efforts to prevent the emergence and subsequent spread of VDPVs during the polio endgame. First, they highlight the importance of enhancing routine immunisation coverage, which is already one of the main objectives of the Polio Eradication & Endgame Strategic Plan 2013–2018 [1]. Second, they may help to define a strategy for the use of tOPV in SIAs preceding the withdrawal of serotype 2 OPV that minimises the risk of VDPV2 post OPV2 withdrawal. In settings where population immunity is low and cVDPV2 currently absent, our findings and past experience suggest that 1 or 2 tOPV SIAs could increase the (small) risk of VDPV2 emergence compared to doing nothing. This risk is enhanced where tOPV immunogenicity is low, SIA coverage poor or there is a persistently “missed” group. Therefore, if tOPV SIAs are implemented preceding OPV2 withdrawal then they should be of sufficient number and high coverage to achieve high serotype-2 population immunity. In the context of limited resources, these SIAs should be targeted to countries considered at high risk of VDPV2 emergence according to the WHO risk assessment system. Finally, given the recently demonstrated advantage of IPV compared with OPV in terms of the boost to humoral and intestinal immunity in previously OPV-immunised children [37,38], IPV SIAs in high-risk areas may also be considered as part of the strategy to minimise the risk of VDPV2 emergence during OPV2 withdrawal.
|
10.1371/journal.ppat.1005522 | Candida glabrata Binding to Candida albicans Hyphae Enables Its Development in Oropharyngeal Candidiasis | Pathogenic mechanisms of Candida glabrata in oral candidiasis, especially because of its inability to form hyphae, are understudied. Since both Candida albicans and C. glabrata are frequently co-isolated in oropharyngeal candidiasis (OPC), we examined their co-adhesion in vitro and observed adhesion of C. glabrata only to C. albicans hyphae microscopically. Mice were infected sublingually with C. albicans or C. glabrata individually, or with both species concurrently, to study their ability to cause OPC. Infection with C. glabrata alone resulted in negligible infection of tongues; however, colonization by C. glabrata was increased by co-infection or a pre-established infection with C. albicans. Furthermore, C. glabrata required C. albicans for colonization of tongues, since decreasing C. albicans burden with fluconazole also reduced C. glabrata. C. albicans hyphal wall adhesins Als1 and Als3 were important for in vitro adhesion of C. glabrata and to establish OPC. C. glabrata cell wall protein coding genes EPA8, EPA19, AWP2, AWP7, and CAGL0F00181 were implicated in mediating adhesion to C. albicans hyphae and remarkably, their expression was induced by incubation with germinated C. albicans. Thus, we found a near essential requirement for the presence of C. albicans for both initial colonization and establishment of OPC infection by C. glabrata.
| Understanding how Candida glabrata is able to establish oral mucosal infections is particularly important since many C. glabrata strains are innately resistant to azole antifungal drugs used in treating mucosal and disseminated infections. The epidemiology of C. glabrata oral infections shows that C. glabrata is very often present as a co-infection with Candida albicans. Here we suggest a mechanism to explain this clinical finding. We show that C. glabrata is unable to colonize the oral mucosa in a murine oral infection model. However, prior or co-colonization by C. albicans allows C. glabrata to colonize and persist in the oral cavity. Mechanistically, we show that C. glabrata binds specifically to C. albicans hyphae, mediated by hyphally expressed ALS adhesins in C. albicans and cell surface proteins in C. glabrata that are transcriptionally up-regulated in the presence of C. abicans. In this sense, C. glabrata is a piggy-back fungus that relies upon binding to C. albicans hyphae for oral colonization. This finding has implications for treatment of oral candidiasis and may shed light on colonization mechanisms of other non-hyphae producing fungi.
| Oropharyngeal candidiasis (OPC) is an opportunistic mucosal infection caused by Candida species [1,2]. Candida albicans and Candida glabrata are the first and second major etiological agents of OPC, respectively [3]. Although other Candida species, including C. parapsilosis, C. tropicalis, and C. krusei, may be isolated as the sole species from oral infection sites, single species infection by C. glabrata alone is rare [4,5]. C. glabrata is most frequently co-isolated along with C. albicans in mixed species oral infections [4,6,7]. Oral infections involving C. glabrata have increased by 17% over the past several years [7], and are particularly common in cancer patients, denture-wearers, or following prolonged use of broad spectrum antibiotics, steriods or following head and neck radiation therapy [3]. These infections were often associated with multiple Candida species [3,4]. Oral infections with mixed C. albicans and C. glabrata were found to be more severe and difficult to treat [5] since many C. glabrata strains are innately resistant to azole antifungal agents used in treating mucosal infections. Prophylactic use of azole antifungal drugs has been implicated as a major cause for the increase in non-C. albicans fungemia [8]. Fungemia caused by C. glabrata has high mortality especially in adult patients in intensive care units [9], and although fluconazole prophylaxis has reduced the incidence of invasive candidiasis in high-risk neonates and immunosuppressed patients, there has been little effect on the overall incidence of C. glabrata candidiasis. Given the frequency of C. glabrata and C. albicans co-infection, it is imperative to understand the mechanisms deployed by C. glabrata in co- infections with C. albicans.
C. albicans is a diploid, polymorphic fungus that exists in yeast, hyphal, and psuedohyphal forms [10]. C. albicans hyphae express numerous proteins that enhance virulence by adhering to host cells or damaging host tissue [11]. C. albicans hyphae are known to penetrate epithelial surfaces, damage endothelial cells, and aid in systemic infection by colonizing different organs such as kidneys, spleen and brain [10,12]. Als (Agglutinin Like Sequence proteins), Hwp1p (Hyphal wall protein), and Eap1 (Enhanced Adherence to Polystyrene) are well-characterized C. albicans hyphal wall adhesins that mediate C. albicans interaction with host epithelial, endothelial and host tissue proteins [13–15]. C. albicans adhesins contribute not only to its ability to adhere and colonize multiple types of host tissues, but also serve as binding moieties for other microbes such as Streptococcus gordonii, Pseudomonas aeruginosa, and Staphylococcus aureus [16–19]. It is therefore possible that one or more C. albicans hyphal-specific adhesins may play a role in C. glabrata interaction as well.
In terms of host tissue invasion, C. albicans has a fitness advantage over C. glabrata in terms of its ability to switch between yeast to hyphal forms. By contrast, C. glabrata virulence must be independent of its morphology, since it lacks the ability to form true hyphae. However, C. glabrata is likely to express specific adhesins in order to establish colonization [20,21]. Phylogenetic analysis of the C. glabrata genome showed 66 putative cell wall proteins, of which only a few have been well characterized in terms of host cell adhesion [13]. Cell wall protein families known to be involved in adhesion to endothelial and epithelial cells include the EPA (Epithelial cell adhesin), AED (Adherence to endothelial cells), and PWP (PA14 domain containing Wall Protein) proteins [13]. C. glabrata Epa1, 6, and 7 adhesins bind to both endothelial and epithelial host cells [22,23], while Pwp7p and Aed1p are known to interact with endothelial cells [13]. Deletion of these Epa1 adhesins attenuated virulence in a murine model of disseminated candidiasis [22,23]. The role of C. glabrata adhesins, beyond their ability to mediate adherence to host tissues, is understudied. We hypothesize that one or more of these adhesins may promote interspecies interaction with C. albicans during mixed species OPC.
Co-adhesion is the basis for both single and multispecies colonization in the host [24]. Co-adhesion in bacteria is well studied and it has been established that the expression of multiple bacterial adhesins drive interspecies oral bacterial colonization [24,25]. Although mixed infections of C. glabrata and C. albicans occur frequently, the mechanism of co-adhesion and interspecies colonization is not well understood [4]. In our study, in spite of C. glabrata encoding several cell wall adhesins known to bind host epithelial and endothelial cells, we documented poor colonization in our murine OPC model. We hypothesized that co- infection or prior infection with C. albicans may facilitate C. glabrata infection. Here we characterize the co-colonization of C. glabrata and C. albicans in a murine model of OPC, and explore the role of cell wall proteins from both species in mediating cell-cell interaction and co-colonization.
We initially performed an in vitro biofilm assays to test whether C. albicans and C. glabrata have any cooperative growth effects. Two strains of C. glabrata (BG2 wild type, WT) and a GFP-expressing strain CgVSY55 (ura3Δ::hph ScPGKp-yEGFP-URA3-CEN-ARS) derived from a CgDSY562 WT [26] and two strains of C. albicans (CAI4 WT with URA replaced, URA+) or CAF2-yCherry strain [27] were used in biofilm experiments. In a static plate assay, C. albicans CAI4 and CgBG2 each formed single species biofilms with similar robustness. However, when grown together as a dual species biofilm, the total dry weight was significantly (P<0.001) higher compared to single species (Fig 1A). Fluorescent quantitation of co-culture of C. albicans CAF2-yCherry with C. glabrata CgVSY55 under static biofilm growth showed enhanced growth of both species occurred compared with single species (Fig 1B). In contrast, under dynamic flow conditions, C. glabrata (CgVSY55) alone was unable to form biofilms within the flow chamber, while C. albicans formed abundant biofilms. However, when both species were co-cultured under dynamic flow conditions, C. glabrata CgVSY55 cells (green) were found associated with C. albicans (red) nascent biofilm regions, and were concentrated along C. albicans hyphae (Fig 1C arrows).
To further examine how these two Candida species might be interacting, we examined their association directly by fluorescence microscopy. C. albicans cells were grown in YNB + 1.25% glucose (for yeast phase cells) or in YNB + 1.25% N-acetyl glucosamine at 37°C (to induce hyphal cells) for 3 h. C. albicans cells were then incubated with C. glabrata cells at 1:1 ratio for 60 min. C. glabrata cells did not adhere with C. albicans yeast cells (Fig 2A, upper left); however they showed strong adhesion along the length of germinated C. albicans hyphae (Fig 2A left). Scanning Electron Microscopy (SEM) further illustrated this interaction showing that C. glabrata cells adhered along the entire length of C. albicans hyphae (Fig 2A right). We observed that C. glabrata cells formed rows of adherent cells along the length of hyphae, but did not adhere to other C. glabrata cells. Next, we quantified adhesion as defined by the number of C. glabrata cells adhering to 10 μm length of C. albicans hyphae in seven different strains of C. glabrata. Among the C. glabrata strains examined, CgDSY56 (the parent strain of CgVSY55) had significantly (P<0.0001) higher adherence (6.4 ± 0.2 cells / 10 μm hyphae, high adherence strain) when compared to other strains tested. CgBG2 and Cg960032 (4.2 ± 0.2 cells / 10 μm hyphae) showed medium adherence; and Cg931010, Cg932474, Cg148042, and Cg90030 showed low adherence (3.0 ± 0.2cells / 10 μm hyphae) (Fig 2B).
Yeast to hyphae transition in C. albicans induces expression of hyphal-specific proteins as well as altering mannans and glucans levels in the hyphal cell wall [28]. To identify whether binding between C. albicans and C. glabrata was mediated by cell wall carbohydrates, we performed blocking experiments with C. albicans using concanavalin A (which binds cell wall mannans) and an antibody to β,1–3 glucan at concentrations that we previously showed provided good cell coverage [29]. C. albicans hyphae were treated with concanavalin A or with β,1–3 glucan Ab for 30 min, washed, then incubated with C. glabrata; however C. glabrata adhesion to C. albicans hyphae was unchanged, suggesting that C. albicans adhesion is not mediated by binding to C. albicans mannose or β,1–3 glucan. This is consistent with the fact that we did not detect C. glabrata binding to other C. glabrata cells since the C. glabrata cell wall contains both mannan and β,1–3 glucan.
We hypothesized that C. glabrata might bind C. albicans cell wall proteins directly. To test candidate C. albicans hyphal wall adhesins required for C. glabrata adhesion, we performed co-adhesion assays with ALS1 and ALS3 deficient C. albicans (Fig 3). We found that C. glabrata had significantly decreased adherence to hyphae of a C. albicans als3Δ/Δ mutant (72.3% reduction), an als1Δ/Δ mutant (28.8% reduction), and an als1/als3Δ/Δ double mutant (86% reduction). ALS1 and ALS3 complementation strains showed restoration of adherence to levels closer to that of wild type strain (Fig 3A). To further validate the role of C. albicans Als1 and Als3, we performed a quantitative adherence assay by direct microscopic observation using S. cerevisiae strains expressing C. albicans ALS1 and ALS3 with a GFP-tagged C. glabrata strain. Both Als1 and Als3 expressing S. cerevisiae strains showed significantly higher binding (Binding Index = 52.0 ± 3.0 and 58.2 ± 1.4, respectively) compared to S. cerevisiae expressing an empty vector (Binding Index = 19.7 ± 2.3) (Fig 3B).
To determine whether our observed binding between C. albicans and C. glabrata has relevance in vivo, we examined the ability of C. glabrata to establish infection in our murine model of OPC. Since C. glabrata has not been used before in OPC infection models, we began with a single species oral infection of C57BL/6 mice with C. glabrata alone as we have previously done with C. albicans (Fig 4A). In this model, sublingual infection with a C. albicans inoculum of 1 X 106 cells/ml typically produces clinical symptoms and white tongue plaques 4–5 days post infection, and recovery of 1 X 107 CFU / gm tongue tissue at 5 days post infection. Surprisingly, in no infection experiments using C. glabrata did we observe the typical appearance of white tongue plaques indicative of clinical infection. We tried varying immunosuppressive agents (triampicinolone acetonide, cyclophosphamide), mouse strains (BALB/c, IL17RAk/o) and used inocula size of C. glabrata ranging from 1 X 107 to 1 X 1010 cells/ml. In all cases, infection with C. glabrata alone resulted in no clinical appearance of disease or weight loss in animals. Consistent with this lack of disease, the recoverable C. glabrata CFUs from the tongue were extremely low (4–7 X 102 CFU/g of tongue tissue).
Since our in vitro biofilm and adhesion assays showed enhanced adhesion and growth of C. glabrata when mixed with C. albicans, we next attempted a mixed infection with C. glabrata (1X109 cells/ml) either as a co-infection with C. albicans (5 X 107 cells/ml); or as a delayed infection with C. glabrata 24 or 48 h after infection of C. albicans (Fig 4). C. glabrata CFU were significantly (P<0.02) increased by ten-fold (3 X 103 CFU/g of tongue tissue) when mice were co-infected with C. albicans (Fig 4A). Co-infection with C. glabrata did not alter C. albicans infection levels (1.2 X 107 CFU/g of tongue tissue) compared with C. albicans infection alone (Fig 4A). However, delaying C. glabrata infection for 24 or 48 h after establishment of C. albicans infection further increased C. glabrata oral infection by a further 10-fold (3.5–4.5 X1 04 CFU/g of tongue tissue, P<0.0001) compared to C. glabrata single species infection (Fig 4B). Mean animal weights did not change upon C. glabrata infection only (Fig 4C). However, mice lost weight more rapidly following a mixed infection compared with infection by C. albicans alone, so that mice in the mixed infection group had to be sacrified one day sooner due to total weight loss compared with mice infected with C. albicans only (Fig 4C). Thus our data show that levels of oral infection of C. glabrata were signifcantly increased by an established C. albicans oral infection and the rate of weight loss was increased upon dual species infection.
Next, we examined tongues of mixed-infected mice histologically to determine whether C. glabrata alters C. albicans invasive properties and to identify the localization of C. glabrata infection within the mucosal epithelium. For these experiments we infected mice with fluorescent-tagged strains of C. albicans (CAF2-yCherry) on day 0 and C. glabrata (CgVSY55) on day 2; and collected tongue tissues on day 5. Tongues were sectioned and stained with either PAS to visualize fungal-tissue architecture or cryo-sectioned for visualization of yeast cell localization by fluorescence microscopy. Tongues from mice with mixed infection showed robust fungal plaque formation as well as extensive C. albicans hyphal penetration of the superficial epithelium (Fig 5A, boxed region) as well as invasion into some regions of the underlying epithelium and lamina propria (Fig 5A, arrows). Closer inspection of these regions showed widespread C. albicans hyphae; and in some areas yeast cells were observed both adherent to hyphae and as unattached cells that were likely to be C. glabrata (Fig 5B and 5C, arrows). Fluorescent imaging of these regions confirmed that the majority of tissue invasion was with C. albicans hyphae (Fig 5D, boxed region, red), however C. glabrata cells (green) were also observed within these tissues both associated with C. albicans hyphae as well as being unconnected and separate within the epithelium (Fig 5E and 5F, arrows). In contrast, mono-species C. glabrata infection resulted on only very small superficial plaques that were localized on the surface mucosa without any invasion. Thus, infection of oral epithelium with C. albicans and the presence of its hyphae were permissive for infection and tissue invasion by C. glabrata.
To further confirm the requirement of C. albicans for C. glabrata for initial infection, we treated mice with fluconazole (Flu) after establishing mixed infection using Flu sensitive (CaFluS) or Flu resistant C. albicans (CaFluR) strains and Flu resistant C. glabrata (CgFluR) (Fig 6). Mice were treated with Flu for four days after an oral mixed infection was already established for four days. As expected, Flu treatment did not alter infection levels of either species in a mixed infection with CgFluR and CaFluR strains. However, for a mixed infection with C. glabrata CgFluR and C. albicans CaFluS strains, Flu treatment resulted in significant (by two logs, P<0.001) reduction of both C. glabrata and C. albicans. Flu treated animals infected with CaFluR strains in a mixed infection lost significantly (P<0.05) more weight (21.2 ± 0.2%) than mice infected with C. albicans CaFluS strains (18.9 ± 0.3%). Although we could not determine the co-locallization of C. albicans and C. glabrata histologically due to lack of fluorescent markers in Flu resistant strains, examination of tongues confirmed the reduction in superficial epithelial fungal burden and invasion upon Flu treatment (Fig 6). Thus, C. glabrata infection levels were proportional to those of C. albicans, showing that C. glabrata requires the presence of C. albicans for early infection in vivo.
Since our in vitro data showed that C. albicans Als1 and Als3 adhesins were important for C. glabrata adherence, we next examined their role in mixed C. glabrata-C. albicans oral infection in vivo. A 48 h delayed infection of C. glabrata following infection with C. albicans wild type or Als adhesin deficient strains was performed (Fig 7). C. albicans als1Δ/Δ and als3Δ/Δ mutants were able to establish infection at the same levels as WT cells. However, C. glabrata tongue CFUs were significantly (P<0.05) decreased (2.8 X 104 CFU/g) following infection with the C. albicans als1Δ/Δ mutant; and were even further reduced (6.6 X 103 CFU/g, P< 0.001) following infection by C. albicans als3Δ/Δ. Infection of C. glabrata with C. albicans Als1 and Als3 complemented strains showed restoration of C. glabrata colonization to levels similar to those observed with the wildtype C. albicans strain (Fig 7). No differences in animal weights between the groups was found since levels of infection by C. albicans were similar between groups.
To identify adhesion partners on C. glabrata, we screened 44 S. cerevisiae strains expressing C. glabrata cell wall proteins and identified five strains expressing CgEpa8, CgEpa19, CgAwp2, CgAwp7 or ORF CAGL0F00181 that were most adhesive (2–5 cells/10 μm C. albicans hyphae) (Fig 8A). Most other tested strains, including the S. cerevisiae parental strain, had no adhesion to C. albicans hyphae. Next, we examined comparative transcription levels of these five candidate genes in C. glabrata strains which have high adherence (CgDSY562), medium adherence (CgBG2), and low adherance (Cg90030) in vitro to C. albicans. We used C. glabrata EPA1 and EPA6 genes as a negative control since S. cerevisiae expressing C. glabrata Epa1 and Epa6 did not bind to C. albicans hyphae, although they are highly expressed major adhesins in C. glabrata. To confirm that these strains also had differential binding to C. albicans during infection, we compared infection levels in a mixed infection in OPC, and found that indeed, the low and high adherence strains had a significant (P<0.01) difference in infection levels (Fig 8B). Then, transcriptional levels of these candidate genes were measured by qPCR before and after incubation with germinated C. albicans. Although CgEPA8 and CgAWP7 were most highly expressed in the high adhesion strain compared to the lower adhesion strains, we did not find significant differences in basal expression levels among the three other candidate genes among the C. glabrata strains. However, transcriptional levels of four genes (CgEPA8, CgEPA19, CgAWP2, and ORF CAGL0F00181) were increased significantly by 6–7 fold, while CgAWP7 was increased by 2-fold in the high adherence strain (CgDSY562) upon incubation with C. albicans hyphae. This induction was less for CgEPA19, CgAWP2, and CgCAGL0F00181 in the intermediate adherent strain, while the low adhesion Cg90030 strain had the least induction by C. albicans for all five genes (Fig 8C). Expression of CgEPA6 and CgEPA1 genes, which serve as controls since they do not mediate adherence to C. albicans, were both modestly down-regulated in the presence of C. albicans. Taken together, these results show that C. glabrata cell wall genes EPA8, EPA19, AWP2, AWP7 and CAGL0F0018 are upregulated by C. albicans and may promote a dual species oral infection.
Although clinical studies have shown that C. albicans and C. glabrata are common partners co-isolated from oral infections, C. glabrata alone rarely causes oral infection. This work identifies for the first time that C. glabrata adherence to C. albicans hyphae is the basis for this partnership and that it is mediated by specific adhesins on both species. Previous in vitro studies found that C. glabrata alone was unable to colonize or invade reconstituted human vaginal epithelium (RHVE) [30] or reconstituted human oral epithelium (RHOE) [31]. Mixed infections using both C. glabrata and C. albicans increased tissue damage in RHOE [32] and were permissive for infection in RHVE [33] and in vivo in tongues of immunosuppressed mice [34], although others found no difference in host damage or inflammation in co-infected human oral epithelial [35]. These and our own studies are in agreement that C. glabrata alone is non-invasive in respect to oral-esophageal mucosal epithelium, in contrast to its ability to penetrate gastric epithelium [34]. The basis for this difference in tissue tropism is unknown, although it is possible that differences in the gut environment induces differential expression of C. glabrata adhesins.
We found that two major fungal hyphal wall adhesins Als3 and Als1 contribute to binding C. glabrata in vitro and to establish oral infection in vivo. C. albicans Als3 appears to make the major contribution towards binding with C. glabrata, with Als1 having a secondary role. Consistent with this, loss of Als1 on its own does not strongly reduce adherence to C. glabrata. However, in strains deleted for ALS3, additional loss of ALS1 further reduced adherence by an additional two-fold. Als3 is a well known multifunctional surface protein, however we have identified an additional novel function of this adhesin in binding C. glabrata. Since Als 3 proteins are very abundant on C. albicans hyphae, and we only find 2–6 C. glabrata cells per hyphae, we expect that substantial numbers of Als proteins would still be available on hyphae to carry out other functions in the context of oral infection. It is also possible that Als3 might have a similar role in binding other non-hyphal forming Candida species such as C. krusei that are frequently co-isolated along with C. albicans in OPC. C. albicans Als3 seems to be promiscuous in its binding partners since Als3 proteins have been shown to bind the oral bacteria S. gordonii through its SspB cell surface protein in a mixed species biofilm [36] and to S. aureus during polymicrobial biofilm growth [37]. Hence Als3 may to be an excellent target for disruption of mixed species and inter-kingdom biofilms.
The C. glabrata Epa family consists of at least 20 GPI-anchored surface exposed adhesins whose expression of individual members is strain dependent [38]. Epa proteins recognize host glycans, and C. glabrata Epa1 is the best characterized member that is involved in adhesion to mammalian epithelium. Epa1 preferentially recognizes Gal β1–3 glycans, and variations of its adhesion domain conferred promiscuity of ligand binding [39]. Recently, Epa binding domains were functionally classified according to their ligand binding profiles, and interestingly our identified adhesins C. glabrata Epa8 and Epa19 were found to be very closely related and within the functional class III of Epa ligands that have weak binding to epithelial cells [40]. Thus, we speculate that some Class III Epa adhesins may have ligand functions with other cell types including C. albicans.
Another similarlity among the C. glabrata adhesins we identified (EPA8, EPA19, AWP2, AWP7 and CAGL0F00181) is that their expression levels were all induced by incubation with C. albicans hyphae (Fig 8C). In contrast, C. glabrata EPA1 and EPA6 (both Class I ligands with high binding to epithelial and endothelial [41] cells, and highly expressed in log phase cells [23]), were not up-regulated following incubation with C. albicans hyphae. In agreement with our findings, no increase in expression levels of C. glabrata EPA1, EPA6 or EPA7 was found following co-infection with C. albicans in RHVE cells [30]. Based on our results, we propose a role of these C. glabrata CWPs (EPA8, EPA19, AWP2, AWP7, and CAGL0F00181) in interspecies binding and further suggest that C. glabrata is able to transcriptionally regulate selected genes needed for its colonization and survival in a host. It is known that many C. glabrata EPA genes are transcriptional silenced. Since EPA1 and EPA6 (both of which are strongly silenced) are not up-regulated by co-culture with C. albicans (Fig 8), this suggests that the transcriptional regulation of C. glabrata EPA8, EPA19, AWP2, AWP7, and CAGL0F00181 is not through general antagonism of sub-telomeric silencing [42]. How C. glabrata regulates these genes in the presence of C. albicans remains to be determined.
C. glabrata alone was not competent to cause infection in our OPC model. Our data further suggest that while C. glabrata colonizes oral mucosa poorly (even in an immunosuppressed host), it has evolved to exploit the presence of hyphae-producing C. albicans to establish colonization and invasion of oral epithelium; and its presence enhanced the severity of OPC as measured by rate of weight loss of animals. Furthermore, co-infections treated with Fluconazole reduced levels of C. glabrata concomitantly with C. albicans over four days, showing its dependence upon the presence of C. albicans in early infections. However, our results show that C. glabrata is found both together and apart from C. albicans hyphae in tissues, suggesting that once it gains a foothold in oral epithelium by binding C. albicans hyphae, it can survive alone in mucosal tissues, albeit at low levels. These C. glabrata cells existing independently in oral mucosa may be a colonization reservoir for dissemination if the oral epithelium is breached by trauma, chemotherapy or other factors. In this regard, our preliminary experiments showed that mice with mixed C. glabrata and C. albicans oral infections had significantly higher stomach colonization of both species, suggesting that gut colonization might serve as such a reservoir. Also, these reservoirs may become clinically significant following long-term azole therapy providing an environment in which drug resistant C. glabrata could emerge.
Our data suggests a model whereby oral tissues that are inherently resistant to infection by C. glabrata, are colonized by piggybacking with C. albicans to establish a foothold of tissue infection. Of interest, and the subject of ongoing studies in our lab, is the role of oral and gut reservoirs of C. glabrata in subequent colonization of other tissues that have a naturally higher tropism for infection by C. glabrata, as well as their role in subsequent dessimination.
All Candida and S. cerevisiae strains used are listed in Table 1 and Table in S1 Table. C. albicans cells were maintained in yeast extract/peptone/dextrose (YPD; Difco) medium with the addition of uridine (50 mg/ml; Sigma) when required and stored as -80°C. S. cerevisiae containing pADH or pADH-ALS3 were maintained on synthetic medium lacking uracil (CSM-glu) (0.077% CSM-ura, 0.67% yeast nitrogen base [Difco], 1.25% glucose, and 2.5% agar). S cerevisiae strains expressing N-terminal domains of C. glabrata Cell Wall Proteins (CWP) were made as described [42], and are in preparation for publication elsewhere). The ORFs whose domains mediate adherence to Candida hyphae are shown in S1 Table.
C. albicans cells were cultured overnight in YPD broth, diluted to an OD600 = 0.3 in pre-warmed YNB medium supplemented with 1.25% GlcNAc, and incubated for 3 h at 37°C with gentle shaking to induce germination. C. glabrata or S. cerevisiae strains were grown similarly except in YNB + 1.25% of glucose. Cells were collected by centrifugation (100 X g), washed once in PBS, and then re-suspended in PBS. Germination of C. albicans cells was confirmed by microscopic observation. C. albicans cells were then incubated with C. glabrata cells at a 1:1 ratio for 60 min. Blocking experiments described previously [29], were carried out using washed C. albicans cells incubated with concanavalin A (100 ug/ml; mannan binding lectin, Sigma) or β,1–3 glucan Ab (10 μg/ml, Biosupplies) for 30 min (concentrations that gave high coverage of cells as determined by FACScan), then washed in PBS before assay. For adhesion assays of S. cerevisiae strains expressing C. albicans Als1 and Als3 adhesins, S. cerevisiae cells or an S. cerevisiae empty vector (control) were incubated with CgVSY55 for 1 h at 37°C (at cell ratio 1:1), then a Binding Index was calculated as the number of C. glabrata cells bound to S. cerevisiae cells divided by (number of bound C. glabrata cells plus unbound C. glabrata cells plus unbound S. cerevisiae cells) X 100 per field. At least 10 separate fields were used to obtain averages.
Each Candida strain was grown overnight to OD600~2.0, washed twice in Phosphate Buffered Saline (PBS), re-suspended in YNB without uridine, and 1 ml cells (1 X106 cells/ml) were added to polystyrene wells. For mixed species biofilms, 500 μl of each species (5 X 105 cells /ml) for a total of 1 ml was added to the well. After incubation for 3 h to allow adhesion, non-adherent cells were gently removed by aspiration and 1 ml of fresh media was added. Biofilms were grown for 24 h at 37°C on an orbital shaker and biofilm dry weight was measured as previously described [43]. For fluorescence biofilm assays, single and dual species biofilms were grown on 96 well microtiter plates using a yCherry expressing strain of C. albicans and a GFP expressing C. glabrata strain. Fluorescent counts were recorded at 37°C using a Bio-Tek multifunction plate reader and analyzed using Gen5 software. Alternatively, we examined non-static dual species biofilms grown under flow conditions. For these experiments, YPD media containing the C. albicans WT strain CAF2 cells expressing the fluorescent protein mCherry and the C. glabrata WT strain VSY55 expressing GFP (both at 1 × 106 cells/ml) were circulated through a μ-Slide I 0.8 Luer family ibiTreat flow chamber (ibidi, Martinsried, Germany) for 2 h at 37°C and a shear force at the coverslip surface of 0.8 dynes/cm2. Images were obtained using a Zeiss LSM 510 confocal microscope, and analyzed using ZEN imaging software (Zeiss, Göttingen, Germany). Flow was maintained during image acquisition.
Overnight cultures of C. albicans were diluted to an OD600 = 0.3 in pre-warmed YNB medium supplemented with 1.25% GlcNAc and incubated for 3 h at 37°C to induce germination, or diluted in YNB medium supplemented with 1.25% glucose at room temperature for yeast cells. C. glabrata CgDSY562, CgBG2, and Cg90030 overnight cultures were grown similarly using YNB + 1.25% of glucose. Cells were collected by centrifugation (100 X g), and re-suspended in PBS. C. glabrata cells were then incubated with germinated or yeast form C. albicans at a 1:1 ratio for 30 min. Total RNA was isolated from C. glabrata, C. glabrata and C. albicans 1 X 107 cells using an RNeasy minikit (Qiagen). Reverse transcription (RT) was performed using SuperScript III reverse transcriptase, and oligo(dT)20 primer (Invitrogen). cDNA was purified (Geneflow PCR purification kit) and quantified with a NanoDrop spectrophotometer (NanoDrop Technologies). Quantitative RT-PCR (qRT-PCR) was performed in triplicate for CgACT1, CgAWP2,7, CgEPA1,6,8,19 and CgCAGL0F00181 using gene-specific primers (Table in S2 Table). C. albicans cDNA was used as a negative control in all experiments to verify specifity of amplification. Genes were normalized to CgACT1 in each respective strain and condition as decribed previously [44].
Fluorescent microscopy was done using a yCherry expressing strain of C. albicans [45] and the GFP expressing C. glabrata (VSY55: ura3Δ::hph ScPGKp-yEGFP-URA3-CEN-ARS) derived from a C. glabrata DSY562 clinical isolate [46]. Scanning electron microscope observations were carried out on C. albicans hyphae and C. glabrata cells. C. albicans cells were grown in YNB for yeast phase cells or in YNB + 1.25% GlcNAc at 37°C (to induce hyphae) for 3 h. C. albicans cells were then incubated with C. glabrata cells at 1:1 ratio for 30 min. Cells were incubated on a concavalin A (100ug/ml; Sigma) coated glass slide for 1 h at RT. Cells were washed twice with PBS, fixed with 2% glutaraldehyde (Sigma) for 30 min at 4°C, then washed twice with distilled water. Samples were dehydrated in 30%, 50%, 70%, 85%, and 95% ethanol for 15 min each and 100% ethanol twice for 15 min each. Samples were exchanged into 100% hexamethyldisilazane (HMDS) and allowed to dry in a hood before visualization. SEM observation was done under the following analytical condition: L = SE1 and EHT = 2.5 kV to study the binding of C. glabrata on C. albicans cells with Hitachi SU70 FESEM operating at 2.0 keV.
C. albicans murine OPC model [47,48] was used for infection with C. glabrata. Mice (BALB/c, C57BL/6, and IL17RAk/o) were immuno-suppressed with cortisone acetate (150–250 mg/kg), triampicinolone acetonide (100–150 mg/kg) or cyclophosphamide (100–150 mg/kg) one day before infection with C. glabrata (1 X 107 to 1 X 109 cells/ml). For mixed infections, mice (female C57BL/6, 4–6 weeks old) were immunosuppressed with cortisone acetate 225 mg/kg (Sigma) on day -1, +1, and +3, and then infected with C. albicans (5 X 107 cells/ml) on day 0; or infected with C. glabrata, (1 X 109 cells/ml) on day 2 after pre-establishing C. albicans infection on day 0. C. albicans and C. glabrata colonies from tongue tissues were differentiated on CHROMagar media. On the fifth or sixth day after infection, mice were euthanized by cervical dislocation under anesthesia (ketamine/xylazine); tongue tissues were excised and hemi-sectioned along the long axis with a scalpel. One half was weighed and homogenized for quantification of fungi, and the other half was processed for histopathological analysis. Tongue hemi-sections were fixed in 10% buffered-formalin for 24 h, paraffin embedded, and then cut into 5μm sections for Periodic Acid-Schiff (PAS) staining as we previously described [49]. For histological co-localization experiments, animals were infected with C. albicans yCherry and the GFP expressing C. glabrata (VSY55) strains as described above. For these experiments, tongue hemi sections were fixed in 4% (w/v) paraformaldehyde (PFA) for 24 h, incubated in 30% sucrose for 3 days, snap frozen in OCT compound (Tissue-Tek, Sakura, Torrance, CA) with liquid nitrogen, and cut into 8μm cryosections.
For Fluconazole (Flu) treatment studies, ten mice were used for each group (drug treatment and controls using combinations of Flu resistant and sensitive strains of C. albicans and Flu resistant C. glabrata shown in Table 1). Sensitivities of each strain to Flu was verified using MIC assays. Immunosuppression was induced on days −1, +1 and +3 post-infection. Mice were infected sublingually with C. albicans (5 X 107 cells/ml) on day 0, and C. glabrata (1 X 109 cells/ml) on day 2, and were sacrificed on day +7. Mice received daily intraperitoneal injections of 100 mg/kg Fluconazole that was initiated 48 h after C. glabrata infection and continued through post-infection day 7.
Statistical analyses were performed using GraphPad Prism software version 5.0 (GraphPad Software, San Diego, CA, USA) using unpaired Student's t-tests. Differences of P<0.05 were considered significant. All experiments were performed at least thrice.
This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. This protocol was approved by the University of Buffalo Institutional Animal Care and Use Committee (Project Number: ORB06042Y).
|
10.1371/journal.pntd.0001153 | Predictors of Visceral Leishmaniasis Relapse in HIV-Infected Patients: A Systematic Review | Visceral leishmaniasis (VL) is a common complication in AIDS patients living in Leishmania-endemic areas. Although antiretroviral therapy has changed the clinical course of HIV infection and its associated illnesses, the prevention of VL relapses remains a challenge for the care of HIV and Leishmania co-infected patients. This work is a systematic review of previous studies that have described predictors of VL relapse in HIV-infected patients.
We searched the electronic databases of MEDLINE, LILACS, and the Cochrane Central Register of Controlled Trials. Studies were selected if they included HIV-infected individuals with a VL diagnosis and patient follow-up after the leishmaniasis treatment with an analysis of the clearly defined outcome of prediction of relapse.
Eighteen out 178 studies satisfied the specified inclusion criteria. Most patients were males between 30 and 40 years of age, and HIV transmission was primarily via intravenous drug use. Previous VL episodes were identified as risk factors for relapse in 3 studies. Two studies found that baseline CD4+ T cell count above 100 cells/mL was associated with a decreased relapse rate. The observation of an increase in CD4+ T cells at patient follow-up was associated with protection from relapse in 5 of 7 studies. Meta-analysis of all studies assessing secondary prophylaxis showed significant reduction of VL relapse rate following prophylaxis. None of the five observational studies evaluating the impact of highly active antiretroviral therapy use found a reduction in the risk of VL relapse upon patient follow-up.
Some predictors of VL relapse could be identified: a) the absence of an increase in CD4+ cells at follow-up; b) lack of secondary prophylaxis; and c) previous history of VL relapse. CD4+ counts below 100 cells/mL at the time of primary VL diagnosis may also be a predictive factor for VL relapse.
| Visceral leishmaniasis (VL) is the most serious form of an insect-transmitted parasitic disease prevalent in 70 countries. The disease is caused by species of the L. donovani complex found in different geographical regions. These parasites have substantially different clinical, drug susceptibility and epidemiological characteristics. According to data from the World Health Organization, the areas where HIV-Leishmania co-infection is distributed are extensive. HIV infection increases the risk of developing VL, reduces the likelihood of a therapeutic response, and greatly increases the probability of relapse. A better understanding of the factors promoting relapses is essential; therefore we performed a systematic review of articles involving all articles assessing the predictors of VL relapse in HIV-infected individuals older than 14 years of age. Out of 178 relevant articles, 18 met the inclusion criteria and in total, data from 1017 patients were analyzed. We identified previous episodes of VL relapse, CD4+ lymphocyte count fewer than 100 cells/mL at VL diagnosis, and the absence of an increase in CD4+ counts at follow-up as major factors associated with VL relapse. Knowledge of relapse predictors can help to identify patients with different degrees of risk, facilitate and direct prophylaxis choices, and aid in patient counseling.
| Visceral leishmaniasis (VL) and human immunodeficiency virus (HIV) co-infection has emerged as a serious disease pattern [1], [2]. HIV infection increases the risk of developing VL by 100 to 2,320 times in endemic areas [3], [4] and, on the other hand, VL promotes the clinical progression of HIV disease and the development of AIDS-defining conditions [5]. Both infections switch the predominantly cellular immunological response from Th1 to Th2 through complex cytokine mediated mechanisms leading to a synergistic detrimental effect on the cellular immune response [6], [7], [8]. Other important findings related to HIV-Leishmania co-infection is a reduction in therapeutic response and high rate of relapse, which is the clinical deterioration after clinical improvement, observed in 25–61% of patients [9], [10], [11], [12]. Although the term recurrence has also been used as synonym for relapse, recurrence applies to the finding of a parasite repeatedly. It is important to emphasize that neither of these two terms distinguishes parasitological persistence from re-infection.
The poor therapeutic outcome, the high rate of relapse, the poliparasitic nature of VL in HIV-infected persons, as well as the atypical manifestations of the disease and the impaired access to health-care resources make HIV-infected individuals prone to enlarge the number of human reservoirs [13]. This concern is of utmost importance in Asia, where HIV and Leishmania co-infections are increasingly being reported in countries that are also facing parasite resistance to antimonial drugs [14].
Recent changes in the epidemiological patterns of HIV and Leishmania infections are likely to lead to a greater degree of overlap and greater risk of co-infection and they justify increased alertness. From a global epidemiologic viewpoint, two parallel trends are alarming: the ruralization of the HIV pandemic and the urbanization and spread of VL [1], [15]. World Health Organization (WHO) [16] reports that the public health impact of leishmaniasis worldwide has been grossly underestimated for many years because notification was compulsory in only 32 of the 88 countries where 350 million people were at risk. The reported global incidence of co-infection is likely underestimated either because VL has not been included in the list of AIDS related opportunistic infection in all endemic areas.
Before the widespread use of antiretroviral therapy, such co-infection was common in Europe [5]. The co-infection is now becoming proportionately more prominent in areas with poor access to antiretrovirals, such as Africa. In areas where it is available, highly active antiretroviral therapy (HAART) has changed the course of the HIV/AIDS epidemic and the outcome of associated opportunistic infections. However, evidence of relapse rate reduction in patients using HAART is conflicting [17]. This work is a systematic review of studies describing the predictors of VL relapse in HIV-infected patients.
This review was conducted on all papers published before July, 31, 2010. To ensure scientific rigour, the Preferred Reporting of Systematic Reviews and Meta-Analysis (PRISMA) guidelines [18] were used for systematic data synthesis. Studies were identified by a Medline/PubMed search using a combination of terms that has been shown to maximize sensitivity [19]. The search terms used are shown in Figure 1. The LILACS and Cochrane databases were used for literature review using a Boolean combination of search terms. Additional reports were located using a manual search of references from retrieved papers. Two independent reviewers (GFC and MRS) initially checked the lists of titles and abstracts identified by this search to determine whether an article contained relevant data. Studies were considered eligible if they were presented in an original article, examined HIV-infected individuals over 14 years of age with a VL diagnosis, included follow-up after the leishmaniasis treatment and clearly analyzed predictors of relapse.
There were no restrictions on the publication language, date of publication, use of secondary prophylaxis, or duration of follow-up in the study. We excluded studies evaluating fewer than ten cases and studies evaluating mixed populations of HIV-infected and uninfected subjects unless separated results for HIV patients were identified. The selected articles were read in full to confirm eligibility.
Data were extracted directly from the full-length articles into structured tables containing all of the descriptive variables and relevant outcomes. The following information was extracted: country and period of enrollment; sample size; clinical characteristics of the included patients; study design; the number of excluded patients if specified; statistical analyses utilized; duration of follow-up and number of subjects lost to follow-up; outcome of interest; prognostic variables assessed in each study and quality of the regression model [20], [21], [22]. When data were available tests required for completion of the tables were performed. To summarize the results regarding secondary prophylaxis, the software Comprehensive Meta-Analysis Version 2.2.048 was used.
Our selection process is illustrated in Figure 2. Of 178 articles, 136 were excluded because they did not meet the inclusion criteria following reading of titles and/or abstracts. Twenty more articles were excluded after reading the entire article: six analyzed less than ten patients [23], [24], [25], [26], [27], [28], [29], one was a review [30], and thirteen did not evaluate the risk on relapse of different predictors [3], [28], [31], [32], [33], [34], [35], [36], [37], [38], [39], [40], [41]. Four studies [42], [43], [44], [45] were excluded because they included cases published elsewhere [10], [46], [47]. Thus, 18 studies (Table S1) satisfied the specified inclusion and exclusion criteria and constituted the basis of this investigation.
Table S1 summarizes the characteristics of the 1017 patients encompassed by the 18 included studies. The year of study publication ranged from 1989 to 2008. The design of 8 of the studies examined was prospective. Fourteen studies were reported in Spain, two in Italy, and one in Ethiopia and one in France. Eight studies had an enrollment period exclusively after 1996, when HAART became available. Twelve papers stated the proportions of patients receiving HAART involving two nucleoside reverse transcriptase inhibitors and one or two protease inhibitors or non-nucleosides reverse transcriptase inhibitors at VL diagnosis or at relapse or both.
A large proportion of the patients in these studies (87.7%) were male and most were young adults; the median or mean ages reported varied from 27 to 37 years (Table S2). In the 14 studies in which patients' presumed transmission route was known, 72.3% (420/581) of the infections were likely due to intravenous drug use. The median CD4+ T lymphocyte count ranged from 11 to 82 cells/mL. Most patients had an AIDS-defining condition [48] at the time of VL diagnosis (332/572, 58% of patients).
In the majority of the studies, the diagnosis of VL was established by direct demonstration of amastigotes (by cytological study of Wright stains) or by the observation of promastigote growth in samples cultured in specific media. In one study [49], the VL diagnosis was supported either by positive results from Leishmania-specific PCR (polymerase chain reaction) of peripheral blood or bone marrow samples. Three studies [47], [50], [51] also included patients diagnosed by serologic tests (direct agglutination, indirect immunofluorescence or rK-39 dipsticks).
The drug used in the treatment of the primary episode of VL was reported for 89% of the treated patients. Of this total, 73.4% of cases (733 patients) were treated with pentavalent antimonial drugs, 12.4% with amphotericin B deoxycholate (124 patients), and 2.1% (21 patients) received amphotericin in lipid formulations. A minority of patients (1.2%) received pentamidine isethionate and three papers included patients treated with miltefosine [47] or unconventional regimens such as a combination of allopurinol with an azole compound [50], [52]. A test of cure (staining with Giemsa stain and parasite culture or PCR) at the end of treatment was carried out in 8 of 18 studies. In most of these studies, this control was performed for patients whose clinical response was uncertain. Secondary prophylaxis for leishmaniasis was reported in eleven studies.
Three studies explored the impact of mono or dual antiretroviral therapy at VL diagnosis [47] or during the follow-up [50], [53] on relapse. Only one [47] of these studies demonstrated a reduction in relapse rate compared with patients who did not undergo retroviral therapy. Similarly, only one [49] of four studies [10], [49], [51], [54] that followed patients on HAART at VL diagnosis reported a reduction in relapse rate. HAART use on follow-up has also been studied in relation to risk of relapse and none of the five [9], [51], [52], [54], [55] studies showed reduction on VL relapse rate.
Two studies [52], [54] that evaluated VL prophylaxis without specifying the drug used noted a significant reduction in relapse. In a report of ten cases, Bossolasco et al. [55] showed that the relapse rate in patients groups with and without prophylaxis were 60% and 100%, respectively, but this difference did not reach statistical significance. Three studies evaluated specific prophylactic regimens (antimony compounds [46], [50] and liposomal amphotericin [50]) and demonstrated reduction on VL relapse. Although the confidence intervals did not reach statistical significance, another author [56] concluded that lipid-complexed amphotericin prophylaxis also reduced the relapse rate. Finally, Laguna et al. [57] showed a trend towards (p = 0,08) a reduction in VL relapse rate following treatment with pentamidine prophylaxis. A meta-analysis of results from all studies evaluating the impact of secondary prophylaxis is shown in Figure 3. This analysis could consistently demonstrate that secondary prophylaxis reduces VL relapse rate.
CD4+ lymphocyte count at VL diagnosis and follow-up has been studied in relation to risk of relapse. Nine articles [10], [11], [12], [46], [50], [51], [52], [55], [58] compared CD4+ lymphocyte cell counts at VL diagnosis between relapsing and non-relapsing patients as a continuous variable. Neither of these studies showed significant differences between these two groups. On the other hand, two studies [47], [49] that compared relapse rate between patients with CD4+ count at VL diagnosis as a dichotomic variable (above and below than 100 cell/mL) noted that the arms with higher CD4+ counts had lower relapse rate. Similarly, an increase in CD4+ lymphocyte count at follow-up was protective against VL relapse in 5 of 7 studies [10], [11], [49], [55], [58]. In another study [12], univariate analyses of CD4+ counts at follow-up revealed a trend towards a reduction in relapse (p = 0.09).
Other variables explored in relation to relapse are shown in Table S3. Factors such as age, route of HIV transmission, history of intravenous drug use, HIV viral load at VL diagnosis, various clinical findings, specific anti-Leishmania treatments given, time from VL diagnosis to the introduction of protease inhibitor therapy, HAART compliance, the presence of an AIDS-defining disease before VL diagnosis and the presence of serum anti-Leishmania antibodies were not substantially different between relapsing and non-relapsing patients. Tuberculosis co-infection [47], hepatitis C virus co-infection [49] and an incomplete course of VL treatment [52] were evaluated in multivariate analysis and showed a statistically significant association of these conditions with the occurrence of relapse. Previous VL episodes were identified as risk factors for relapse in 3 studies, two of which were multivariate analyses.
The statistical quality and the presentation of methods and results in many studies were poor. In nine studies, the Kaplan-Meier method was used in a univariate survival analysis to analyzed VL relapse. Three prospective studies and two retrospective cohort studies employed Cox regressions for multivariate analysis of independent predictors. One study randomized patients to compare prophylaxis (liposomal amphotericin versus no treatment) and performed multivariate analysis to compare relapse rates by logistic regression, including some predictors as covariates. None of these six studies mentioned collinearity assessment (i.e., a high degree of correlation between 2 predictive variables) or developed a risk score for relapse based on their multivariable results. Also, none of the multivariate analyses reported a goodness-of-fit test of their models. Other studies analyzed isolated relapse predictors by univariate association tests in series of prospective or retrospective cases or in intervention studies.
The present study is the first systematic review of predictors of VL relapse in HIV-infected patients. Our main conclusions are that VL relapse in HIV-infected patients receiving HAART is high and that secondary prophylaxis provides some protective effect but does not completely prevent the occurrence of relapse. We found that patients who did not relapse showed significantly higher CD4+ count at follow-up than patients with relapsing course. Our analysis also suggests that CD4+ count greater than 100 cell/mL at VL diagnosis reduces the odds of relapse. Unlike other opportunistic infections there are some reports of VL relapse in patients with a CD4+ count greater than 200 cell/mL in Ethiopia, and rarely in Europe [9]. This evidence shows that factors other than a CD4+ cell increase are involved in VL control. A threshold for safely discontinuing of secondary prophylaxis has not been established because of these uncertainties.
Most cases reported showed severe reductions in T cells. It could indicate that VL affects HIV-1 patients who exhibit a significant disturbance of cellular immunity; however, VL by itself may reduce CD4+ lymphocyte counts [59]. On the other hand, a CD4+ count greater than 100 cell/mL at VL diagnosis is a potential protective factor against relapse, although the analysis of this beneficial effect may be complicated by the immunosuppression of many the patients included in the studies. When analyzing the CD4+ count range and number of patients with CD4+ counts of greater than 100 cell/mL in the two studies [47], [49] demonstrating an association between higher baseline CD4+ counts and reduced VL relapse, it is possible to speculate that studies that did not demonstrate an influence of CD4+ cells had few patients with CD4+ counts of greater than 100 cell/mL. Studies using animal models reported that CD4+ cells are responsible for the initial control of parasite proliferation and dissemination [60]. Thus, a low initial CD4+ count might allow a wide dissemination of the parasite throughout the phagocytic mononuclear system at the beginning of infection, increasing the number of sites that could harbor quiescent parasites (so-called “sanctuaries”) [61].
Relapses of VL are suggested to occur mainly in individuals with poor responses to antiretroviral treatment who have no improvement in CD4+ counts [11], [12], [58], [62], with a few exceptions [9], [47]. The evolution of patients who acquire VL and thereafter show a significant increase in CD4+ counts while on HAART is currently receiving attention [47][50][51][52]. It has already been established that the outcome of VL is not influenced by humoral immunity but appears to be regulated by CD4+ T helper cells with different patterns of cytokine activity [63]. Protective immunity can be attributed to T helper (Th)-1 cells, whereas Th-2 cell responses produce IL-4 and IL-10 and facilitate the intracellular survival of the parasite [64]. It might be expected that highly active antiretroviral drug combinations would favor an immunological shift from type 2 to type 1 cytokines in HIV-infected individuals. However, increased CD4+ values in peripheral blood and lymphoid tissues as a result of antiretroviral therapy may have negligible effects on cytokine production during the first 24 weeks [65]. In addition, patients on HAART show an initial increase in the CD4+ memory subset, whereas naive CD4+ cells consistently increase only after 1 year [66].
It is known that HIV patients who are receiving HAART have fewer opportunistic infections and recent data shows that there has been a decline in the incidence of VL after the introduction of HAART [41], [54], [67], [68], [69]. HAART seems to be insufficient to prevent VL relapse. Studies in patients receiving HAART showed a relapse rate similar to other studies performed in the pre-HAART era. Only one [49] observational study noted a reduction in the relapse rate among patients on HAART at VL diagnosis. None of the studies reported a statistically significant difference in VL relapse between patients receiving and not receiving HAART on follow-up. These disappointing results so far disagree with a statistically significant association between improvement of CD4+ count at follow-up and reduction of VL relapse. They may be due to the small sample sizes of the studies performed, poor patient adherence to antiviral therapy or insufficient immune response. One possibility to be explored in the future is the role of cytokines [70] influencing the control of VL independently of the CD4+ lymphocyte. The heterogeneity of zymodemes that exhibit different degrees of virulence or parasite burden could contribute to the differences observed in the host immune response and clinical evolution [9]. HAART increases CD4+ count thus influencing the control of VL, but may not be enough in this complex scenario created by the co-infection HIV and Leishmania. Fernandéz-Cotarelo et al. [54] and others [41] have shown a decrease in the number of new episodes of VL in HIV-infected patients receiving HAART but also a tendency toward VL relapse. According to these authors the high rates of relapse could be explained by the increased patient survival that results from effective antiretroviral therapy.
Previous episodes of VL were strongly associated with relapse. Also in agreement with the immune-inflammatory theory, it was hypothesized that the enhancement of the Th-2 response following one early relapse could complicate or prevent the later control of Leishmania infection [54].
Secondary prophylaxis seemed to only partially protect against relapse. Some of studies that observed a reduction in VL relapse following the use of secondary prophylaxis had few patients on HAART, which may not reflect the current reality. Data analysis suggests that the small sample sizes and heterogeneity of regimens used make the results less robust. Nevertheless, the evaluation of these studies through meta-analysis indicates a clear benefit of secondary prophylaxis in reducing VL relapse. Based on six studies whose data were available, the average relapse rate in patients not receiving secondary prophylaxis was 67%, while in the secondary prophylaxis arm, the relapse rate was 31%. Given this difference, the estimated total sample size needed for a study with 80% power would be 70 patients. Three out of the six studies examining secondary prophylaxis were not able to demonstrate statistical significance, possibly because of small sample sizes. It is important to emphasize that despite the heterogeneity of prophylaxis regimens used; statistical results are positively homogeneous in meta-analysis.
Thresholds for safe discontinuation of secondary prophylaxis for Spanish patients have been suggested to be CD4+ counts of 200 [71] and 350 cells/mL [11]. Differently of the European experience, one Ethiopian study [47] has shown that many patients suffering relapse (11 from 39 cases) had a CD4+ count above 200 cells/mL before relapse. These data may suggest that L. donovani, the predominant causative agent of VL in east Africa and south Asia, is a more virulent and anthroponotic species than L. infantum. Another plausible explanation for this difference may be the influence of other variables that can affect the host immune response such as nutritional status and the presence of other infections and co-morbidities.
It has been postulated that the maintenance of an undetectable viral load protects against the development of VL [17] and that a high viral load could predict a weak response to antiparasitic treatment [12] although there are contradictory reports on this point [54], [72]. None of the papers reviewed here linked HIV load by PCR at VL diagnosis with relapse. On the other hand HIV load by PCR at follow-up was statistically related to relapse in one [58] of four studies that evaluated this variable in a univariate analysis. These observations support the idea that a sustained immunological response is more important than a virological response to cure VL in HIV-infected patients.
It is important to note that a wide range of therapeutic drugs were utilized for the treatment of VL in the studies we have reviewed. There was no notable difference in the relapse rate with regard to specific VL treatment used (all analyzed in univariate analysis); however only four studies explored this association and most of them included a limited number of patients and only two [11], [73] involved randomly assigned patients. Few comparative clinical studies have been conducted of the efficacy of treatment for HIV–VL co-infection outside the Mediterranean area. In some instances [74], [75], the development of drug resistance could contribute to therapeutic failure and the relapsing course observed in HIV-infected patients. These observations do not allow us to refute the influence of anti-parasite treatment on relapse outcome.
Although we have made an extensive review, our analysis includes studies with different definitions of cure and different lengths of follow-up. Cure is seldom defined parasitologically in these studies and the difference between treatment failure and relapse is arbitrary in some studies. It is possible that some episodes of relapse in the group of patients in which parasitological cure were not documented by bone marrow examination were treatment failures rather than relapses. Moreover, re-infection was not distinguished from relapse in any paper. There is a high degree of heterogeneity in the evaluated populations as shown by the wide range of reported mortality (6.5% to 83.8%), treatment failure (0 to 47.6%) and relapse rates (20% to 70%). These studies included patients with different degrees of immunosuppression, and different treatment and prophylaxis regimens. Also, there are differences in the study designs, the types of statistical methods used and the prognostic variables included in analysis. These variations may have resulted in patient selection bias or low statistical power, thus hampering a meta-analysis of all studied predictors of relapse. In spite of these limitations, we believe that the meta-analysis results of secondary prophylaxis are consistent, considering the available evidence. In addition, the quality of published reports was heterogeneous and usually poor. Despite these limitations, this review may assist clinicians in making decisions and may also help in the design of future studies.
The results of this systematic review suggest there are identifiable predictive factors of VL relapse, such as previous episodes of VL relapse and lack of recovery of CD4+ lymphocyte numbers after primary visceral leishmaniasis. HAART did not produce the anticipated decrease in the incidence of VL relapses and more data is needed in order to better assess the evolution of VL in the HAART era. In contrast, secondary prophylaxis was shown to be protective against relapse. CD4+ count below 100 cells/mL at the time of VL primary diagnosis is a potential predictor of relapse.
Based on these observations, a high-risk population might be identified and such patients might then be eligible for secondary prophylaxis. Strong surveillance will certainly contribute to improved data quality for decision-makers in this complex scenario. Randomized trials to compare the efficacy of different drugs and their role either in treatment or in prophylaxis are required.
|
10.1371/journal.pgen.1003441 | Identification of Rtl1, a Retrotransposon-Derived Imprinted Gene, as a Novel Driver of Hepatocarcinogenesis | We previously utilized a Sleeping Beauty (SB) transposon mutagenesis screen to discover novel drivers of HCC. This approach identified recurrent mutations within the Dlk1-Dio3 imprinted domain, indicating that alteration of one or more elements within the domain provides a selective advantage to cells during the process of hepatocarcinogenesis. For the current study, we performed transcriptome and small RNA sequencing to profile gene expression in SB–induced HCCs in an attempt to clarify the genetic element(s) contributing to tumorigenesis. We identified strong induction of Retrotransposon-like 1 (Rtl1) expression as the only consistent alteration detected in all SB–induced tumors with Dlk1-Dio3 integrations, suggesting that Rtl1 activation serves as a driver of HCC. While previous studies have identified correlations between disrupted expression of multiple Dlk1-Dio3 domain members and HCC, we show here that direct modulation of a single domain member, Rtl1, can promote hepatocarcinogenesis in vivo. Overexpression of Rtl1 in the livers of adult mice using a hydrodynamic gene delivery technique resulted in highly penetrant (86%) tumor formation. Additionally, we detected overexpression of RTL1 in 30% of analyzed human HCC samples, indicating the potential relevance of this locus as a therapeutic target for patients. The Rtl1 locus is evolutionarily derived from the domestication of a retrotransposon. In addition to identifying Rtl1 as a novel driver of HCC, our study represents one of the first direct in vivo demonstrations of a role for such a co-opted genetic element in promoting carcinogenesis.
| HCC is the third deadliest cancer worldwide, largely due to a lack of effective treatment options. Therapeutic approaches targeted at the molecular mechanisms underlying tumor formation and progression have shown great efficacy for treating other tumor types. Unfortunately, however, much remains to be learned about the molecular pathogenesis of HCC. There is an urgent need to identify and characterize genetic alterations that drive HCC in order to facilitate the development of more effective targeted therapeutics for patients. Here, we present data showing that recurrent mutations identified in a mouse model of HCC result in overexpression of the Rtl1 gene. We have validated Rtl1 as a driver of HCC by demonstrating that its overexpression in mouse liver causes tumor formation. We also detected overexpression of this gene in a significant proportion of human HCC samples, suggesting that it may be a relevant therapeutic target for patients with this disease.
| Hepatocellular carcinoma (HCC) is the third leading cause of cancer-related deaths worldwide [1]. In contrast to the downward trends in incidence observed for most cancer types, that of HCC continues to rise, particularly in the United States [2]. This is due in part to increases in obesity and hepatitis C viral infection, both of which have been implicated in HCC pathogenesis. Treatment options for patients are limited, particularly for those with advanced disease, and the five-year survival rate remains low at ∼10%.
A major goal of HCC research is to develop therapies targeted at the molecular mechanisms underlying tumor development and progression. This type of approach is expected to be much more efficacious, increasing survival rates for HCC patients. Consistent with this idea, treatment with sorafenib, a multi-kinase inhibitor, has shown survival benefits for late-stage patients [3] – a rare achievement in HCC treatment. Nevertheless, sorafenib treatment is only able to extend median survival by three months, underlying the need for improved targeted therapies. Unfortunately, the molecular drivers of HCC remain poorly characterized, precluding the development of such therapeutics. Large-scale sequencing efforts currently being undertaken by The Cancer Genome Atlas (TCGA) project will likely characterize the recurrent genetic alterations present in human liver tumors and may identify novel therapeutic targets. However, it is becoming increasingly clear that human tumors are incredibly complex, and identifying molecular drivers of carcinogenesis among the larger number of background events has proven difficult. Comparative analysis of the information gained from human tumor profiling with data from animal models provides an improved ability to distinguish driver events contributing to human disease.
The Sleeping Beauty (SB) transposon mutagenesis system has proven useful for identifying drivers of tumorigenesis in a wide variety of tissue types [4]. We have previously used SB mutagenesis to generate mice that developed HCC [5]. Subsequent genetic analysis of SB-induced liver tumors identified the Dlk1-Dio3 imprinted domain as a common target of transposon-induced mutations. This highly complex domain contains genes encoding protein-coding transcripts, long non-coding RNAs (lncRNAs), microRNAs (miRNAs), and small nucleolar RNAs (snoRNAs). Expression of domain members is regulated in an allele-specific manner and depends on epigenetic modifications established in the germline [6]. Regulation of this expression pattern is maintained, at least in part, by multiple differentially methylated regions (DMRs) throughout the domain that are methylated on the paternally inherited allele. Maintenance of imprinting is critical for normal function, as evidenced by the fact that uniparental disomy (UPD) for either parental allele leads to severe and widespread developmental defects in both mouse models [7] and human patients [8].
A link between the Dlk1-Dio3 domain and HCC has previously been identified. Interestingly, it has been reported that adeno-associated viral (AAV) vector integration within the same region of the domain as the SB transposon integrations in our model is associated with HCC development in mice [9], [10]. AAV integrations were found to alter expression of several domain members, preventing elucidation of a clear molecular mechanism of tumorigenesis. Other studies have also identified correlation between disrupted expression from the Dlk1-Dio3 domain and HCC [11]–[15], often with several domain members showing aberrant expression. The majority of these studies are correlative in nature, and no attempt is made to validate tumorigenic function of domain members through direct modulation of gene expression.
Here we describe a series of experiments that initially utilized deep-sequencing analyses to obtain detailed gene expression profiles of the SB-induced HCCs. This approach revealed that transposon integration within the Dlk1-Dio3 domain has variable effects on expression of several elements throughout the imprinted domain, but uniformly drives dramatic overexpression of Retrotransposon-like 1 (Rtl1). Validation experiments demonstrate that hepatic overexpression of Rtl1 promotes tumorigenesis in vivo. Additionally, we find that RTL1 is aberrantly expressed in ∼30% of human HCC samples, suggesting that it may be a relevant therapeutic target.
Rtl1 is a poorly characterized gene that encodes a predicted transmembrane protein with aspartic protease activity. Interestingly, this locus is derived from domestication of a sushi-ichi-related retrotransposon [16] and is unique to placental mammals [17]. This study identifies Rtl1 as a novel oncogene involved in hepatocarcinogenesis and suggests that its expression may be used as a prognostic indicator and/or targeted therapeutically to improve outcome for patients with HCC. It also represents one of the first direct in vivo demonstrations of a role for a co-opted genetic element in driving carcinogenesis.
We previously reported the identification of a 33 kilobase region of the imprinted Dlk1-Dio3 domain as a common target of transposon insertion in an SB-induced model of HCC [5] (Figure 1A). Given the domain's complexity and previous studies demonstrating altered expression of multiple domain members in response to insertion of exogenous DNA [9], [10], [18], we used both transcriptome and miRNA sequencing approaches to obtain expression profiles of eight SB-induced HCCs with Dlk1-Dio3 integrations and six normal livers for comparison (Figure 1B–1C, Figures S1 and S2, Tables S1 and S2). Expression of Dlk1-Dio3 domain miRNAs was low to undetectable in normal liver. Similar results were detected for three of eight tumors, while the remaining five tumors displayed activated expression of several imprinted miRNAs. Thus, transposon insertion in the Dlk1-Dio3 domain does not consistently alter miRNA expression. Interestingly, tumor samples with elevated expression of imprinted miRNAs also showed enhanced expression of Meg3 and Rian, suggesting a possible transposon-mediated loss of imprinting effect. Dramatic activation of expression from the locus encoding Rtl1 and Rtl1 antisense (Rtl1as) was observed in all eight SB-induced HCCs, while no significant expression was detected in normal liver. Notably, elevated expression from this locus is the only event that was consistently observed in all SB-induced HCCs with Dlk1-Dio3 integrations (Figure 1B–1C). Because transcription can occur on either strand at this locus [19], strand-specific RT-PCR was performed to determine whether the observed increase resulted from expression of Rtl1, Rtl1as, or a combination of both transcripts. As shown in Figure 2A, reads from the locus encoding Rtl1 and Rtl1as detected in HCCs were derived primarily from transcription of the protein-coding sense strand (i.e. Rtl1). The lack of detectable Rtl1 in normal liver suggests that transposon integration results in activation of a normally transcriptionally silent allele.
As we previously reported, SB transposon integration sites in HCC samples clustered near the 5′ end of Rian within the Dlk1-Dio3 domain [5]. Our initial characterization of transposon integrations was performed using ligation-mediated (LM)-PCR followed by pyrosequencing. It has been shown that this approach yields suboptimal sequencing depth for confident identification of clonal insertion sites [20]. To ensure adequate sequence coverage, the SB-induced HCCs were re-sequenced for the current study using the Illumina platform. Surprisingly, while integrations near the 5′ end of Rian were still found to be the most common event, a transposon orientation bias was revealed that had not previously been evident. For many of the tumors, multiple transposon integrations were identified in this region, and for each of the tumors at least one of these integrations was in the same orientation as Rtl1 (Figure 1A).
To validate the significance of transposon integrations upstream of Rtl1 in SB-induced HCCs, insertion sites from a larger set of tumors, as well as some normal livers (Rogers et al., in press), were sequenced using the Illumina platform. A quantitative analysis of all transposon integrations in the Dlk1-Dio3 domain for these samples is provided in Figure S3. Consistent with recent studies demonstrating minimal insertion bias for SB transposon integration [21], [22], background insertion sites identified in normal liver and subclonal insertions in HCC samples did not show any evidence for preferential integration within the Dlk1-Dio3 domain. In contrast, clonal sites identified in tumors were highly enriched upstream of Rtl1, suggesting positive selection for insertions in this region during the process of tumorigenesis. This analysis further confirmed that transposon integrations in the same transcriptional orientation as Rtl1 are preferentially detected specifically in HCCs. Based on these results, we hypothesized that the high levels of Rtl1 observed in tumors were driven directly by transposons integrated upstream. Amplification of transposon/Rtl1 fusion products from cDNA confirmed transposon-driven Rtl1 overexpression for each of the tumors harboring integrations in this region (Figure 2B). Two different sizes of fusion products were detected, representing direct splicing of the T2/Onc3 transposon into Rtl1 (smaller product) or inclusion of a cryptic upstream exon (larger product). Importantly, both fusion products encode the full Rtl1 open reading frame and are thus predicted to drive overexpression of functional Rtl1 protein.
Two additional Sleeping Beauty screens have been reported in which liver tumors were generated and characterized [23], [24]. Neither of these studies identified the Dlk1-Dio3 domain as a common site of integration. Both screens utilized T2/Onc mice as the source of mutagenic transposons. This transposon is similar in structure to that of the T2/Onc3 strain used in our study, but a distinct promoter is included within the transposon. T2/Onc transposons contain the murine stem cell virus (MSCV) 5′ long-terminal repeat (LTR) promoter, while T2/Onc3 transposons contain the cytomegalovirus (CMV) enhancer/chicken β-actin (CAG) promoter. Differences in promoter activities likely affect the profile of mutations that are selected for in tumors resulting from SB mutagenesis. We suspect that the MSCV promoter may be too weak to overcome the influence of imprinting within the Dlk1-Dio3 domain to drive sufficient hepatic Rtl1 expression to provide cells with a selective advantage and promote tumorigenesis. The CAG promoter, which has a much higher activity in epithelial cells like hepatocytes, may be better able to drive Rtl1 overexpression when integrated upstream, resulting in frequent selection of cells with such mutations in tumors. Consistent with this idea, insertional mutations upstream of Rtl1 have been linked to liver tumor development in two independent studies that utilized viral vectors containing promoters with high activity in hepatocytes [9], [15].
Our RNA profiling analyses and fusion transcript detection led us to conclude that the primary tumor-driving event under positive selection in SB-induced HCCs is activation of Rtl1. While we cannot exclude the possibility that other domain members play a role independently and/or cooperatively with Rtl1, in our model it seems to be the dominant driver of hepatocarcinogenesis. It should be noted that other models of HCC have been described in which altered expression of maternal Dlk1-Dio3 domain members is observed in the absence of Rtl1 activation [25], suggesting that distinct roles may exist for both paternal and maternal components of the domain in different subtypes of HCC.
To study the effects of Rtl1 overexpression on hepatocyte growth and morphology in vitro, we stably overexpressed it in the murine hepatocyte cell line TIB-73. Importantly, this cell line is non-tumorigenic and lacks endogenous expression of Rtl1. Based on the predicted protein structure of Rtl1, which contains an extracellular protease domain, we hypothesized that its effects may be mediated via cleavage of a substrate within the extracellular matrix (ECM). To test this hypothesis, TIB-73 cells expressing either Rtl1 or an empty vector were embedded in a matrix of Matrigel, plated in 24-well plates, and cultured in serum-free medium. Two weeks after plating, cells expressing Rtl1 had grown to form dozens of cyst-like colonies composed of several cells (Figure 3B, 3D). In contrast, cells lacking Rtl1 expression formed less than one colony per well on average, and colonies that did form were much denser and smaller (Figure 3A, 3C). These results demonstrate that Rtl1 expression promotes growth of hepatocytes in the presence of ECM in the context of physiologically relevant levels of growth factors, and they are consistent with our hypothesis that Rtl1 acts by cleaving an ECM component. ECM is an important aspect of the tumor microenvironment, particularly in the liver. The process of liver fibrosis, which involves ECM remodeling and expansion, is strongly linked to HCC, with nearly 90% of cases developing in this context [26]. One mechanism by which fibrosis may contribute to the development of HCC is through sequestration of growth factors in the newly remodeled ECM [27]. According to this model, subsequent release of growth factors through protease-mediated cleavage of ECM components promotes proliferation of adjacent hepatocytes. Our results suggest that Rtl1 may contribute to hepatocarcinogenesis via this mechanism.
We next sought to determine if Rtl1 overexpression is sufficient to promote hepatocarcinogenesis in vivo. Mice with stable hepatic expression of Rtl1 were generated by hydrodynamic tail vein injection of transposon-based expression constructs [28] into Fah-deficient male mice expressing SB transposase [24]. Selective repopulation of the liver was achieved through inclusion of a separate Fah expression vector that allowed stably transfected cells to survive withdrawal of NTBC [29], an event that triggers the death of Fah-null hepatocytes. Mice were euthanized nine months post-injection to assess liver tumorigenesis. Of fourteen mice injected with Rtl1 overexpression constructs, twelve (86%) developed liver tumors, with an average of 2.9 tumors per mouse (Table 1, Figure 4). In another experimental condition, a third construct encoding a short hairpin directed against Trp53 was additionally included. Loss of p53 function is one of the most commonly observed molecular abnormalities in human HCC, occurring in ∼30% of cases and making this a relevant context in which to validate putative oncogenes. Of twelve mice injected with all three transposon constructs, ten (83%) developed liver tumors, with an average of 4.3 tumors per mouse. Six of the mice from this cohort were sacrificed at time points earlier than nine months. When considering only those mice that were aged for nine months to allow direct comparison between the two experimental groups, five of six (83%) mice with p53 knockdown in addition to Rtl1 overexpression developed liver tumors, with an average of 6.7 tumors per mouse. This is significantly higher (p = 0.027) than the number of tumors per mouse developed with Rtl1 overexpression alone. Knockdown of p53 in tumors was assessed by western blot (Figure S4A). Although efficiency was somewhat variable, the majority of tumors showed significant knockdown.
It has been shown that following liver repopulation, the Fah mouse model is predisposed to tumor formation in the absence of any additional transgene [30], [31]. The tumors that develop in this context uniformly lack expression of Fah. We assessed expression of both Rtl1 and Fah by RT-PCR in fourteen tumors developed following hydrodynamic injection (Figure S4B). Of these fourteen tumors, eleven were found to express both genes. This result suggests that while a small subset of our tumors are likely background events developed independently of Rtl1 expression due to the model's predisposition, the majority of tumors were induced directly by overexpression of Rtl1. Further evidence for the tumorigenic activity of Rtl1 in vivo comes from a recently published study showing that liver tumors develop in mice following hepatic lentiviral delivery [15].
In order to determine the prevalence of RTL1 activation in human disease, RT-PCR was performed on a collection of thirty-three human HCC RNA samples, along with matched benign adjacent liver tissue (Figure 5A, Figure S5). A lack of significant expression was observed for all but one of the benign liver samples. In contrast, significant activation of RTL1 was detected in 30% (10/33) of analyzed tumors. To assess RTL1 expression in another set of human HCCs, we utilized RNASeq data available through The Cancer Genome Atlas (TCGA) consortium. Consistent with our initial analysis, RTL1 expression was found to be significantly activated in 30% (10/33) of analyzed tumors (Figure 5B). Low-level expression was detected in two of the adjacent benign tissue samples for which sequence data was available. It should be noted that four of the tumor samples included in the TCGA dataset overlap with the initial set of 33 samples analyzed by RT-PCR. No expression of RTL1 was detected in these four samples by either analysis. A notable gender disparity is observed in human HCC, wherein men are around three times more likely to develop the disease than women [1]. We analyzed our human expression data to determine if RTL1 overexpression was associated with tumors from one gender or the other, but failed to detect evidence of any bias. Based on the combined set of human samples that we analyzed, RTL1 was found to be overexpressed in samples from 12/38 males (32%) and 8/24 females (33%).
Unfortunately, there is very little existing data on the expression of RTL1 in disease states, including cancer. Most expression analyses utilize commercially available microarray platforms, the vast majority of which lack probes for RTL1. While multiple studies have identified correlative links between disrupted expression of other Dlk1-Dio3 domain members and HCC [9]–[14], expression of RTL1 has not typically been assessed. This may be due in part to the fact that RTL1 is a single exon gene, preventing straightforward design of primers that specifically amplify from cDNA and not genomic DNA. Notably, we have utilized a method for RTL1 expression analysis that adds a unique sequence tag during reverse-transcription [32], thus allowing specific amplification from cDNA and eliminating background amplification from genomic DNA.
In the setting of spontaneous hepatocarcinogenesis in humans, RTL1 activation may occur as a result of loss of imprinting (LOI) within the Dlk1-Dio3 domain. Epigenetic abnormalities are known to play a large role in driving tumor development and progression, in part through induction of LOI [33]. A direct causal role for LOI in cancer was demonstrated by Holm et al., who showed that chimeric mice created using embryonic stem cells lacking imprinting-specific DNA methylation develop multiple tumor types with nearly complete penetrance [34]. The most common tumor type observed was HCC, suggesting that LOI in the liver confers a strong predisposition to cancer. While expression from the Dlk1-Dio3 domain was not examined in the study, the results we present here suggest that hepatic activation of Rtl1 may be a driving factor in the HCCs that were developed. Interestingly, Wang et al. reported loss of methylation within the Rtl1 locus in mouse HCCs resulting from AAV integration [10], although effects on Rtl1 expression were not determined. To assess whether or not Rtl1 overexpression is associated specifically with altered expression of other imprinted genes in our SB-induced HCCs, analysis of variance (ANOVA) was conducted on the whole transcriptome to identify genes with differential expression between Rtl1-overexpressing tumors and normal liver. Following Bonferroni correction, 3 of 125 imprinted genes and 474 of 20,707 non-imprinted genes were identified as having significantly different expression between the two sample sets. By Fisher's exact test, these proportions are not significantly different (p = 0.760). This analysis shows that activation of Rtl1 does not correlate specifically with altered expression of other imprinted genes in our tumors.
Next we sought to determine if Rtl1-induced HCCs in mice resemble a specific subtype of human HCC. An integrative meta-analysis of human HCC gene expression profiles has identified three major expression subtypes called S1, S2, and S3 [35]. Transcriptome sequencing data from the mouse HCCs overexpressing Rtl1 was used to determine the extent to which these SB-induced tumors resemble human HCC. Expression levels of genes defining the S1, S2, and S3 subclasses of human HCC were assessed for each of the SB-induced tumors and normal liver samples. Unsupervised clustering of samples based on expression of constituent genes was performed individually for each subclass. The results show that the SB-induced tumors resemble human HCCs within the S1 subclass (Figure 6). This was further supported by Gene Set Enrichment Analysis (GSEA) [36], [37] that showed a statistically significant association (p = 0.039) between Rtl1-induced HCCs and the S1 expression class. Immunohistochemistry was performed to validate protein expression of two S1 subclass genes in SB-induced HCC (Figure S6). This subclass of human HCC is associated with poor to moderate cellular differentiation, activation of the WNT signaling pathway, and early tumor recurrence.
Rtl1 is a poorly characterized gene that encodes a predicted transmembrane protein with aspartic protease activity. Knockout studies in mice have demonstrated a role in the placental feto-maternal interface [38], but functional studies in other tissues are lacking. Experiments to determine the necessity of Rtl1's protease domain for its ability to promote tumorigenesis and to identify targets of its activity will help to clarify the oncogenic mechanism. If required, RTL1's protease activity represents a promising target for therapeutic intervention in HCC patients. Pepstatin is a naturally occurring bacterial peptide that demonstrates broad potential to inhibit aspartic proteases [39]. Additionally, more specific inhibitors have successfully been developed that target the activity of other aspartic proteases, including renin [40] and HIV-1 protease [41]. It is also possible that RTL1 expression could be a useful biomarker for HCC. Based on the human samples that we analyzed, its expression appears to be highly tumor-specific. Although low-level expression was detected in three non-tumor liver samples, all of the benign samples came from HCC patients and are therefore unlikely to be representative of truly normal liver.
In this study we identify Rtl1, a co-opted imprinted gene, as a novel driver of hepatocarcinogenesis. Mutations resulting in its overexpression were highly selected for in liver tumors developed using a forward genetic screen. While several correlative results linking the Dlk1-Dio3 domain to HCC development have been reported, our study provides direct evidence that modulation of a domain member in vitro and in vivo promotes a tumorigenic phenotype. We show here that overexpression of Rtl1 in cultured hepatocytes results in an increased growth ability in extracellular matrix. We also show that overexpression via hydrodynamic gene delivery results in highly penetrant liver tumor formation in mice. Additionally, a subset of human HCCs displays overexpression of RTL1, suggesting it may be a relevant therapeutic target for patients.
SB-induced mouse HCCs used in this study were generated as previously described [5]. All tumors used in this study came from male mice and were collected using procedures approved and monitored by the Institutional Animal Care and Use Committees at the National Cancer Institute-Frederick and the University of Minnesota.
Paired tumor and benign liver tissues were obtained from 33 patients undergoing resections for HCC at Mayo Clinic between 1987 and 2003, snap-frozen in liquid nitrogen, and stored at −80°C. The Mayo Clinic Institutional Review Board approved the study.
DNA from SB-induced tumors was prepared for sequencing of transposon integration sites as previously described [20].
Stable cell lines were generated by delivery of piggyBac transposon constructs encoding either Rtl1 or an empty vector into TIB-73 (ATCC: BNL CL.2) cultured mouse hepatocytes. 24-well plates were coated with a thin layer of Matrigel basement membrane mix (BD Biosciences) and allowed to set up for 30 minutes at 37°C. For each stable cell line, cells were trypsinized and washed with PBS before resuspension of 5,000 cells in additional Matrigel. The resuspended cells were plated on top of the thin layer of basement membrane mix and allowed to set up, followed by addition of serum-free, low-glucose DMEM (Life Technologies). Images were taken two weeks after plating.
Hydrodynamic tail vein injection into Fah-deficient male mice expressing SB11 transposase was performed as previously described [24]. A plasmid expressing Rtl1 from the human PGK promoter and flanked by SB transposon inverted repeat/direct repeats (IR/DRs) was generated by amplifying the open reading frame of Rtl1 from C57Bl/6J mouse genomic DNA and subcloning it into pT2/PGK-pA. This plasmid was co-injected with PT2/PGK-FAHIL, a plasmid containing an SB IR/DR-flanked expression cassette for Fah and firefly luciferase. Some mice were additionally injected with pT2/shp53, a plasmid containing an SB IR/DR-flanked expression cassette for a short-hairpin RNA directed against Trp53 [29], [45].
Total protein was collected from liver tumor samples by homogenization in RIPA lysis buffer. Samples were boiled for five minutes in a reducing buffer and SDS-PAGE was performed. Proteins were transferred to nitrocellulose membranes for blotting. Primary antibodies used were anti-p53 (Cell Signaling Technology #2524), anti-GFP (Clontech #632380), and anti-β-tubulin (Sigma-Aldrich #T4026).
GSEA [36], [37] was performed using default parameters. Analyzed gene sets were comprised of all the genes defining human HCC subclasses S1, S2, and S3 [35] for which mouse orthologs have been annotated.
Formalin-fixed, paraffin-embedded liver samples were sectioned to a thickness of 4 µm and baked onto glass slides. Samples were de-paraffinized, rehydrated, and treated with citrate antigen unmasking solution (Vector Laboratories). Endogenous peroxidase activity was blocked by treatment with a 3% solution of hydrogen peroxide for fifteen minutes. The anti-rabbit ImmPRESS reagent kit (Vector Laboratories) was used for immunolabeling with primary antibodies anti-Fyb (Abgent #AJ1306a) and anti-Ier3 (Abgent #AP11790a). Both primary antibodies were diluted 1∶100 and incubated with samples for one hour at room temperature. The ImmPACT DAB kit (Vector Laboratories) was used for detection. Sections were counterstained with hematoxylin QS (Vector Laboratories) and mounted in Permount (Fisher Scientific) for light microscopy.
|
10.1371/journal.pbio.2001627 | Convergent evolution of SWS2 opsin facilitates adaptive radiation of threespine stickleback into different light environments | Repeated adaptation to a new environment often leads to convergent phenotypic changes whose underlying genetic mechanisms are rarely known. Here, we study adaptation of color vision in threespine stickleback during the repeated postglacial colonization of clearwater and blackwater lakes in the Haida Gwaii archipelago. We use whole genomes from 16 clearwater and 12 blackwater populations, and a selection experiment, in which stickleback were transplanted from a blackwater lake into an uninhabited clearwater pond and resampled after 19 y to test for selection on cone opsin genes. Patterns of haplotype homozygosity, genetic diversity, site frequency spectra, and allele-frequency change support a selective sweep centered on the adjacent blue- and red-light sensitive opsins SWS2 and LWS. The haplotype under selection carries seven amino acid changes in SWS2, including two changes known to cause a red-shift in light absorption, and is favored in blackwater lakes but disfavored in the clearwater habitat of the transplant population. Remarkably, the same red-shifting amino acid changes occurred after the duplication of SWS2 198 million years ago, in the ancestor of most spiny-rayed fish. Two distantly related fish species, bluefin killifish and black bream, express these old paralogs divergently in black- and clearwater habitats, while sticklebacks lost one paralog. Our study thus shows that convergent adaptation to the same environment can involve the same genetic changes on very different evolutionary time scales by reevolving lost mutations and reusing them repeatedly from standing genetic variation.
| When organisms colonize a new environment in replicate, natural selection often leads to similar phenotypic adaptations. Such “convergent evolution” is known from both distant relatives, e.g., sea cows and whales adapting to an aquatic life, and from multiple populations within a species, but the causing genetic changes are rarely known. Here, we studied how a fish, the threespine stickleback, repeatedly adapted its color vision to living in red light–dominated blackwater lakes. Using multiple natural populations and a 19-y evolution experiment, we found selection on a blue light–sensitive visual pigment gene. One allele of this gene with a red-shifted light sensitivity facilitated repeated blackwater colonization. Surprisingly, two amino acid changes responsible for the red-shift have independently occurred 198 million years earlier, after the gene was duplicated in the ancestor of all spiny-rayed fish and modified into blue- and red-shifted gene copies. While other fish species today use these two gene copies to adapt to clear- and blackwater, stickleback have lost a copy and reevolved these mutations on different alleles of the same gene causing convergent adaptation to these habitats. Thus, we conclude that the same genetic changes can be responsible for convergent evolution on very different time scales.
| Successful colonization of a new habitat requires adaptation to a multitude of different selection pressures. When similar habitats are colonized in replicate by different populations or species, phenotypic adaptation is often convergent [1], and this is most striking in adaptive radiations in multiple lakes or on several islands [2]. Whether a similar phenotypic adaptation is caused by selection on variants present in a shared ancestor due to recurrent mutation or due to changes in different genes, however, is still poorly understood [3, 4]. Only recently, genetic and population genomic studies have started to unravel the evolutionary mechanisms of phenotypic convergence [5–12].
An adaptive radiation with replicate habitat colonization is found among threespine stickleback (Gasterosteus aculeatus) inhabiting the Haida Gwaii archipelago, British Columbia, Canada. Since the retreat of the ice sheets 12,000 years ago, and likely before that [13–15], marine stickleback have colonized hundreds of freshwater habitats independently in different watersheds and adapted in predictable ways to highly divergent “ecological theatres” [16]. One major predictor of natural selection in the Haida Gwaii stickleback radiation is the spectrum of visible light [16]. Most Haida Gwaii lakes are either oligotrophic clearwater lakes featuring full-spectrum light to blue-shifted light with increasing depth, or they are dystrophic blackwater lakes, stained by dissolved tannins leading to a red-shifted light spectrum [16–18]. Blackwater lakes are extreme, almost “nocturnal” visual environments, as both downwelling short-wavelength light and almost all up- or sidewelling light is absorbed, leaving only downwelling red light in a small cone above the focal animal.
Evolutionary adaptation to blackwater lakes in Haida Gwaii stickleback had consequences for multiple traits: stickleback have evolved larger body sizes and reduced lateral plates, both maximizing burst velocity and agility to escape from a predator on short reaction distance, and the former increasing postcapture resistance to predators [16]. Not only natural selection by predators, but also sexual selection interacts with light spectra: blackwater stickleback males have replaced red with black nuptial throat color, which maximizes contrast to the blue eye and against the background via reversed counter-shading [17]. And both traits, the black throat and blue eyes, are preferred by females choosing their mates [19]. Also, color vision was adapted to the blackwater light spectrum: double cones in the retina of stickleback from blackwater systems express only red light–sensitive photopigments instead of one red and one green light–sensitive photopigment, increasing the stickleback’s visual sensitivity to dominant red light [18]. Expression differences are heritable and replicated between independently colonized blackwater lakes [18], but the genetic mechanisms underlying these differences are still unknown.
Here, we study the evolutionary history of color vision genes during adaptation to blackwater environments. To perceive color, vertebrates use a combination of membrane-bound photosensitive proteins, called visual opsins, that are expressed in cone cells in the retina (i.e., cone opsins) and have peak sensitivities at different wavelengths [20]. Vertebrates have evolved large opsin repertoires via gene duplication and divergence [21, 22]. Previous research showed that opsins have evolved in response to the visual environment by sequence or expression divergence (i.e., “spectral tuning,” [18, 23–30]). Adaptation of color vision has been found both between [31–34] and within species [35–37], sometimes without gene duplication [38] or without sequence divergence [39]. Remarkably, spectral tuning of opsins has led to convergent adaptation by recurrent mutations, leading to the same amino acid sequences in distantly related species, families, orders, or phyla [31, 40–43]. These observations together with extensive biochemical and mutagenesis study of opsin proteins have led to the identification of several “key site” substitutions [44–46], from which genotypes the light absorption phenotype can be predicted. Both functional experiments and the evolutionary history of opsins thus show that there are many different, functionally tractable “molecular roads” to color vision adaptation.
Most ray-finned fish possess large repertoires of eight or more cone opsin genes, originating from a combination of ancient and recent lineage-specific gene duplication events, facilitating adaptation to a diversity of visual environments in aquatic systems [21, 22]. However, threespine stickleback have only four cone opsins: the UV sensitive SWS1, a single blue-sensitive SWS2, a single green-sensitive RH2, and a red-sensitive LWS [18, 22]. This is an impoverished repertoire compared to most other fish species, and when compared to relatives among spiny-rayed fish, two SWS2 paralogs and one RH2 paralog have been lost [22, 47].
Although we know of parallel and heritable expression differences between blackwater and clearwater habitats [18] and between marine and freshwater habitats [36], the targets of selection in the genome are unknown and it is unclear whether spectral tuning of amino acids is involved in adaptation to divergent freshwater habitats [6, 36]. Here, we assess whether and which cone opsin genes have experienced recent selection using two types of evidence. First, we use whole genomes from one oceanic and 27 freshwater fish from across the Haida Gwaii adaptive radiation and from coastal British Columbia, including 15 clearwater and 12 blackwater populations derived from 18 watersheds independently colonized by marine ancestors over the last 12,000 y [13]. Second, we use whole genomes from a selection experiment in which 100 adult stickleback from a blackwater lake were transferred to a barren clearwater pond, from which 11 individuals were resampled after 19 y [48]. Then, we screen cone opsin genes for amino acid variation with predictable effects on color vision and test whether selection on such coding changes or noncoding variation has facilitated the colonization of blackwater habitats. Finally, we compare the molecular mechanisms of adaptation to blackwater habitat among Haida Gwaii threespine stickleback to other blackwater-inhabiting fish species and therein uncover convergent evolution on vastly different time scales.
We sequenced the genomes of 58 threespine stickleback from 25 freshwater populations on Haida Gwaii, two freshwater sites from coastal British Columbia, and one marine site (Table 1), resulting in a dataset of 7,888,602 high-quality SNPs with transition to transversion ratio (Ts/Tv) 1.26 (see materials and methods). We split this dataset into an “adaptive radiation” partition, containing single individuals from each natural population on Haida Gwaii, two coastal British Columbia populations, and one mid-Pacific marine population (n = 28 individuals, 6,526,842 SNPs, Ts/Tv = 1.31), and a “selection experiment” partition, containing 12 individuals from the blackwater Mayer Lake source population and 11 individuals from the clearwater Roadside Pond transplant population (n = 23 individuals, 4,180,622 SNPs, Ts/Tv = 1.26). We scanned the adaptive radiation dataset for evidence of selective sweeps at the four cone opsin genes, using the haplotype-based statistics iHS and H12 [49, 50] and their variation across the genome to identify outlier regions. We identified a prominent outlier region for both iHS and H12 metrics in Haida Gwaii sticklebacks, suggesting a selective sweep centered on a genomic region containing both the blue- and red-sensitive opsin genes, SWS2 and LWS (Fig 1). In contrast, H12 and iHS around the green- and UV-sensitive opsins RH2 and SWS1 were not significantly different from the genome-wide expectation, although both statistics are elevated around RH2, which is in the vicinity of a more heterogeneous genomic background due to reduced recombination in this region. The selective sweep signature centered on SWS2 and LWS is caused by haplotypes with a long run of reference alleles, uninterrupted by recombination and thus leading to an extended haplotype homozygosity (EHH) across the 28 populations for the reference “sweep haplotype” (Fig 2). The same EHH signature for the sweep haplotype was found in the selection experiment within the blackwater population Mayer Lake. However, 13 generations after the transfer of Mayer Lake fish into the clearwater habitat of Roadside Pond, the alternate haplotype shows a stronger EHH signature (Fig 2), indicative of the alternate haplotype quickly rising to high frequency in the selection experiment.
Patterns of nucleotide diversity, differentiation, site frequency spectra, and allele-frequency change across the genome from the selection experiment data support the presence of a selective sweep signature in the region containing the two opsins SWS2 and LWS and two tightly linked genes, HCFC1A and ENSGACG00000022160 (Figs 3 and 4, S1 Fig). In the blackwater source population Mayer Lake, nucleotide diversity is significantly reduced compared to the genome-wide expectation, and Tajima’s D is strongly negative, as expected under a selective sweep (Fig 3). The transplant into a clearwater habitat, however, has led to an increase in frequency of the alternate haplotype (Fig 2) and therefore to significant differentiation in this region between the two populations, as measured by FST, ranking this region among the highest 0.1% differentiated regions in the genome (Fig 3). While nucleotide diversity and Tajima’s D on LWS are still reduced in the source population Mayer Lake from the initial selective sweep, the rising frequency of the alternate haplotype has led to a positive Tajima’s D centered on SWS2. These high Tajima’s D values are among the top 0.1% outliers even against the positively shifted genome-wide distribution of Tajima’s D, which was caused by the bottleneck experienced during the population transplant [48]. This positive Tajima’s D is a consequence of both reduced diversity due to the first sweep shared with Mayer Lake (cf. Fig 3) and a rapid increase of the alternate haplotype in the transplant population to a similar frequency as the haplotype favored in Mayer Lake (Fig 2). A transient phase of a “reverse selective sweep” for the alternate haplotype, associated with the shift in light regime in the selection experiment, or the combined effect of a past selective sweep and a bottleneck may have caused this pattern. The fact that this genomic region is among the top 0.1% FST outlier windows and that linked low-frequency alleles associated with SWS2 and LWS are among the top 1% allele-frequency changes in the genome (Fig 4) support selection for the alternate haplotype in the transplant population. Note that linkage disequilibrium is not increased beyond the region containing the two opsins SWS2 and LWS and the two linked genes, perhaps due to high recombination in this chromosomal segment, making it unlikely that selection on genes further up- or downstream was involved (S1 Fig., [51, 52]).
We identified segregating amino acid polymorphisms in all four cone opsins and genes linked to the selective sweep around SWS2 and LWS in both the adaptive radiation and selection experiment datasets (Fig 5). The blue-sensitive SWS2 opsin contains the highest number of amino acid polymorphisms, with seven alternative amino acid residues occurring at high frequency and in nearly perfect linkage (Fig 5), in contrast to only four synonymous substitutions. The sweep haplotype identified in the adaptive radiation dataset carries the same alleles as the reference genome for all seven amino acid polymorphism in SWS2, leading to a nearly perfect association of the SWS2 polymorphisms with the sweep haplotype (χ2 tests, all SWS2 Bonferroni-corrected p < 0.05, Fig 2), while no association is found between the LWS polymorphism and the sweep haplotype. Segregating amino acid polymorphisms in SWS2 are both numerous and occur at high frequency in the adaptive radiation SNP dataset, leading to a high mean pairwise ratio of non-synonymous to synonymous substitutions (dN/dS) estimate of 1.08 between the 56 haplotypes in this dataset, an exceptional value when compared to other functional protein-coding genes in the threespine stickleback genome (Fig 6A).
Thanks to earlier mutagenesis and protein structure studies, we can predict functional consequences for four SWS2, one LWS, and two RH2 amino acid polymorphisms. Three SWS2 polymorphisms are at opsin key sites 96, 97, and 109 and cause shifts in the peak light absorption of SWS2 [53] and the rod opsin RH1 [54]. Notably, the sweep haplotype reference alleles at SWS2-specific key sites, S97 and A109, lead to a “red-shift” in the absorption spectrum, i.e., a peak absorbance at a longer wavelength, while the alternate haplotype alleles at sites C97 and G109, lead to a “blue-shift,” a shorter wavelength absorption maximum at SWS2 [53]. Furthermore, these two key sites and three other polymorphic amino acids in SWS2 (site 40) and RH2 (sites 179 and 203) face the “retinal binding pocket” of the opsin proteins, in which a functional effect is likely [55]. In SWS1 and LWS, no polymorphism falls into a key or retinal binding pocket site, but the single LWS substitution (A217T) replaces a hydrophobic with a hydrophilic residue and may thus have functional consequences. While this polymorphism is not associated with the sweep haplotype in the adaptive radiation dataset (Fig 5), it has increased in frequency linked with the blue-shifted SWS2 haplotype in the selection experiment alongside noncoding SNPs around HCFC1A (Figs 4 and 5).
Population genomic patterns and functional predictions suggest that the amino acid polymorphisms in SWS2 are the most likely target of the selective sweep among Haida Gwaii stickleback and again in the selection experiment, while noncoding variation in both opsins and the two linked genes, the latter lacking coding variation, may have hitchhiked on the selected haplotype. Notably, the same haplotype with the same red-shifted SWS2 key sites is found across the Haida Gwaii adaptive radiation: nearly all populations show identical SWS2 coding sequence haplotypes, either the same blue-shifted or red-shifted haplotype (Fig 7). Selection thus favored the same red-shifted SWS2 cone opsin sequence across multiple Haida Gwaii populations and the alternate and widespread blue-shifted SWS2 haplotype in the selection experiment (Figs 2, 5 and 7).
We asked whether the selective sweep for a red-shifted SWS2 cone opsin was associated with the colonization of blackwater. For this, we tested for an association between genetic variation at the four cone opsins and three predictors and covariates of visual environment: light transmission at 400 nm (T400), lake area, and lake depth. Light transmission at 400 nm is a predictor of light intensity and spectrum, with lower transmission indicating a red-shifted light spectrum in Haida Gwaii lakes [18]. Lake area is a strong predictor of between lake differences in predation regime and might capture the interaction of visual environment and predation landscape influencing selection on color vision [16]. The third covariate, lake depth, is a predictor for variation in light spectra found within a single lake, with more divergent light spectra and thus potential for disruptive selection on color vision in deeper lakes [16, 18]. Genetic variation at cone opsins SWS2, SWS1, and LWS was significantly associated with T400 but not with lake depth or area (Table 2, Fig 6C). When we tested more specifically for an association with blackwater (T400 ≤ 74%), we found only genetic variation at SWS2 and LWS—sweep haplotype variation—to be significantly associated with blackwater (Table 2, Fig 6C). Correlation with blackwater was strongest for four coding and seven noncoding SNPs in SWS2, one noncoding SNP also being an upstream regulatory SNP for LWS (single SNP χ2, p < 0.01). In addition, 28 noncoding SNPs around the UV-sensitive SWS1 opsin showed such a strong correlation, likely due to a high frequency of certain haplotypes in clearwater habitat (Fig 6C). In conclusion, the selective sweep on a single haplotype carrying a red-shifted SWS2 cone opsin coding sequence and linked noncoding variation in SWS2 and LWS may be associated with successful repeated colonization of blackwater habitat across the Haida Gwaii adaptive radiation.
The selection experiment supports a role of the sweep haplotype and associated coding variation in SWS2 during adaptation to different light regimes: The blackwater-associated SWS2 haplotype with the red-shifted SWS2 allele occurs at high frequency in the blackwater population, Mayer Lake, where a selective sweep signature is persistent (Figs 2–5). In contrast, the frequency of the alternate haplotype has rapidly increased after 13 generations in the clearwater habitat, with blue-shifting SWS2 key site substitutions rising from 13% to 40% frequency, with significant population differentiation, and with positive Tajima’s D centered on SWS2 (Figs 4 and 5). Shifts in allele frequency at SWS2 key site substitutions and linked regulatory sites ranks them among the top 5% and 1%, respectively, for allele-frequency change across the genome in the selection experiment (Fig 4), making the blue-shifted SWS2 haplotype a genome-wide outlier and thus a likely target of reversed selection due to habitat shift. Under a pure selection model, an allele-frequency shift of 27% over approximately 13 generations would correspond to an evolutionary change of 1.12 haldanes and to a selection coefficient of 0.28 (see materials and methods). The experimental transfer of a blackwater population to a clearwater habitat thus lead to evolution in the expected direction, given the habitat association across the adaptive radiation: a change from red- to blue-shifted SWS2 allele after the transfer into clearwater habitat.
The blue light–sensitive SWS2 gene has undergone two duplication and divergence cycles [47], of which the first, in the ancestor of spiny-rayed fish, has led to a blue-shifted SWS2B paralog and a red-shifted SWS2A paralog [22, 26, 53, 56–58]. The single SWS2 gene copy of threespine stickleback is derived from an SWS2A paralog, while the other paralogs have been lost in the stickleback lineage [22, 47]. Strikingly, the two SWS2 key site amino acid polymorphisms found in our study are also key sites that have led to red- and blue-shifts in the SWS2A and SWS2B paralogs, respectively (Fig 6B, [47, 53]): at these two key sites, the sweep haplotype in threespine stickleback shows the same amino acids as the ancestral red-shifted SWS2A paralog, and the alternate haplotype shows the same amino acids as the blue-shifted SWS2B paralog lost in the stickleback lineage (Fig 8). Also, at three of the remaining five SWS2 substitutions segregating among threespine stickleback (sites 33, 150 and 169), the amino acids on the sweep haplotype are the same as in SWS2A paralogs of medaka (Oryzias latipes) and bluefin killifish (Lucania goodei), while some amino acids on the nonsweep haplotype are identical to their SWS2B paralogs.
We tested whether the similarity of the red- and blue-shifted stickleback SWS2 haplotypes with each of the ancestral paralogs was due to shared ancestry, gene conversion, or convergent evolution. First, we reconstructed the evolutionary origin of the two threespine stickleback SWS2 haplotypes to ask whether both threespine stickleback haplotypes are derived from an SWS2A ancestor. For this, we embedded the two most common SWS2 threespine stickleback haplotypes (nsweep = 74, nnon-sweep = 23 of 116 haplotypes from all 58 sequenced individuals, Fig 7) into a phylogeny with an orthologous SWS2 sequence from blackspotted stickleback (Gasterosteus wheatlandi) and both SWS2A and SWS2B paralogs from shorthorn sculpin (Myoxocephalus scorpius), medaka, and bluefin killifish. We chose these taxa because their SWS2 paralogs have not been excessively affected by gene conversion [47]. The phylogeny confirmed that both threespine stickleback SWS2 haplotypes and blackspotted stickleback SWS2 were derived from an SWS2A paralog, in line with findings of earlier studies on the origin of the stickleback SWS2 [22, 47], ruling out shared ancestry as an explanation for the similarity between SWS2 paralogs and stickleback haplotypes (Fig 6B).
Second, we asked whether gene conversion between SWS2A and SWS2B paralogs in the lineage leading to threespine stickleback may have contributed to the similarity between SWS2 paralogs and the threespine stickleback SWS2 haplotypes. For this, we computed divergence at synonymous sites (dS) between the threespine stickleback haplotypes and both paralogs of the shorthorn sculpin in sliding windows across the gene (Fig 6D). The divergence distribution shows that the SWS2B paralog is more divergent from both stickleback haplotypes than the SWS2A paralog almost throughout the whole gene, inconsistent with ancestral gene conversion. A single region of reduced synonymous divergence in the last third of the protein suggests ancient gene conversion but does not overlap with the amino acid polymorphisms among Haida Gwaii stickleback. This gene conversion signal has been ascribed to the shorthorn sculpin lineage and not the threespine stickleback lineage in a larger phylogenetic analysis [47]. Ancestral gene conversion thus cannot explain key site similarity between segregating stickleback haplotypes and divergent ancestral paralogs. Instead, new, recurrent mutations in threespine stickleback must have led to convergent amino acid changes with the ancestral paralogs.
We tested whether the SWS2 sweep haplotype in threespine stickleback has accumulated amino acid–changing mutations at an extraordinarily rapid rate (“positive selection”). Branch-specific dN/dS estimates in the SWS2 phylogeny show an elevated dN/dS ratio of 1.64 on the branch leading to the selective sweep haplotype (Fig 6B), indicative of accelerated amino acid substitution and positive selection. A branch-site model test for positive selection on protein sequence, however, could not distinguish between positive selection on this branch and a null model without positive selection (ΔLRT = 0.65, p = 0.13). As the terminal threespine stickleback branches contain only a few substitutions (N * dN = 5.2 and S * dS = 1.0 on the sweep haplotype branch and N * dN = 1.9 and S * dS = 1.0 on the nonsweep haplotype branch), the low divergence of the two haplotypes has likely limited our power to distinguish the two models using the branch-site test, which is most powerful for divergent sequences from interspecific comparisons [59, 60].
Our study reveals that threespine stickleback have adapted wavelength sensitivity through selection at the SWS2 locus. A single red-shifted SWS2 allele has been favored across the adaptive radiation and blackwater lakes, most of which were colonized in replicate from marine ancestors and are almost exclusively inhabited by individuals with this red-shifted SWS2 allele. The evolution of a red-shifted SWS2 opsin thus likely facilitated the colonization of blackwater lakes and subsequent establishment in this extreme habitat. Tannin-stained blackwater is characterized by a red-shifted light spectrum with reduced transmission of short wavelengths [18]. A blue-sensitive opsin spectrally tuned to a longer wavelength will thus increase an individuals’ ability to detect any residual short wavelength light in blackwater. Such an adaptation mechanism is supported by genomic signatures of selection, by functional effect predictions and by genotype-environment associations across the adaptive radiation and in the selection experiment. Also, previously observed phenotypic differences [18] support this mechanism: cones expressing SWS2 in blackwater stickleback from some of these and other populations had an absorption spectrum red-shifted by ~10 nm [18], a stronger shift than is explainable by alternate chromophore use. The combined results from our study and Flamarique et al. [18] thus show that threespine stickleback adapted visual perception to blackwater habitats by spectral tuning of SWS2 key sites, causing a higher sensitivity to the remaining short wavelength light, and increased expression of LWS in double cones, maximizing the sensitivity to background light.
Remarkably, the molecular mechanism underlying this recent adaptation in threespine stickleback recapitulates the duplication and divergence of SWS2 around 198 million years ago in the spiny-rayed fish ancestor [22, 53]. Amino acid replacements at SWS2 key sites are identical and thus convergent between the threespine stickleback SWS2 alleles and the SWS2A and SWS2B paralogs, respectively [18, 47, 53]. Such convergent spectral tuning at key sites of cone opsins has been found previously but exclusively at larger evolutionary timescales, such as between damselfish species [40], between butterflies and vertebrates [41], butterflies and bees [42], humans and poeciliid fish [31], or across the animal kingdom [43]. Convergent spectral tuning between such vastly different time scales—on one side, a microevolutionary, intraspecific level and on the other side, a 198 million-y-old duplication-divergence process—has not yet been shown to our knowledge.
Not only the mechanism of spectral tuning at SWS2 is convergent, but also the environmental context: two other fish species inhabiting both tannin-stained blackwater and clearwater habitats, bluefin killifish and black bream Acanthopagrus butcheri, show expression divergence between the red-shifted SWS2A and blue-shifted SWS2B paralogs [24, 26, 37, 61]. Bluefin killifish populations occur either in blackwater or clearwater habitats [24, 37], and black bream live in clearwater as juveniles and migrate to blackwater habitats where they spend their adult life [26, 61]. In both species, the red-shifted SWS2A cone abundance is higher in blackwater habitats and the blue-shifted SWS2B cone abundance is higher in clearwater habitats, which is due to reduced SWS2B expression and an increased SWS2A expression relative to SWS2B, respectively [24, 61]. While the two key amino acids at sites 97 and 109 each are convergent between bluefin killifish paralogs and stickleback alleles living in either blackwater or clearwater (Fig 8, [37, 47]), black bream has substituted these with other amino acids but still shows the same function for the two paralogs (red- and blue-shift) [26] and thus ecological, phenotypic, and functional convergence.
Adaptation to blackwater environment via SWS2 spectral tuning and therein improved visual capacities can have a multitude of consequences for survival and reproduction. The light environment in blackwater lakes is limited to downwelling, red-shifted light and thus visual detection of predators and prey is much reduced, leading to short action and reaction distances. Any improved detection of prey or predators, for example, via increased sensitivity to color contrast at residual short wavelengths, would be favored by natural selection. Bluefin killifish from blackwater environments indeed showed increased color contrast attention towards blue objects in blackwater [62]. Increased color contrast attention might also be favored by sexual selection: the “blue morph” in bluefin killifish is more abundant in blackwater, and blue males are preferred by individuals raised in stained water [63, 64]. Similarly, stickleback inhabiting blackwater systems have lost red nuptial throat color and instead show black throats, contrasting with the background and with blue eyes, and these two traits are under sexual selection by choosy females [17, 19]. Spectral tuning of blue-sensitive SWS2 may thus be under both natural and sexual selection in threespine stickleback and other blackwater-adapted fish species.
Our selection experiment confirmed that the segregating SWS2 alleles are favored by selection in different light environments: after only 13 generations in a clearwater habitat, the red-shifted SWS2 allele associated with the blackwater sweep haplotype decreased in frequency while the alternate blue-shifted allele swept to high frequency, indicating a reverse, ongoing sweep in clearwater habitats. The direction of change is in line with both functional predictions and genotype–environment association across the adaptive radiation. The evolutionary change of 1.12 haldanes estimated from this allele-frequency shift is much larger than the change in feeding morphology or predator defense morphology traits, which show a mean change of 0.22 haldanes over 12 generations [48]. This could arise from inherent differences between genotype- and phenotype-based estimates, such as the increased variation due to a complex genetic basis and environmental effects in phenotypic estimates [65]. Change in allele frequency at SWS2 by 27% is comparable to the strongest relative changes in trait means: gill raker length was reduced by 43%, lateral plate three height by 18%, and lateral plate two frequency and dorsal spine length by 16% [48]. Selection on color vision thus led to similarly rapid or slightly faster evolutionary change compared to other, feeding and predator defense–related traits; and genetic and phenotypic change was in the direction predicted from independently evolved populations across the adaptive radiation, recapitulating the same habitat contrast.
Repeated use of the same red-shifted SWS2 haplotype during replicated adaptation to blackwater lakes in Haida Gwaii suggest that these adaptive mutations have been present as standing genetic variation in the marine population prior to colonization. Indeed, the marine individual in our dataset is heterozygous for all SWS2 amino acid polymorphisms, confirming the presence of both red-shifted sweep and blue-shifted nonsweep haplotypes in a marine population (Fig 2). Also, freshwater populations outside Haida Gwaii, such as one individual from mainland British Columbia in our study (Table 1, Figs 5 and 7) and the reference genome, a female freshwater stickleback from Alaska, carry the same red-shifted SWS2 haplotype, while another mainland individual carries the blue-shifted SWS2 haplotype. Maintenance of two spectrally tuned opsin alleles as standing genetic variation might be a “microevolutionary rescue” solution to the loss of multiple functionally divergent SWS2 paralogs in the lineage leading to threespine stickleback, which could explain convergent evolution with the ancestral SWS2 paralogs. To maintain such divergent SWS2 alleles, more complex selection scenarios than selection in blackwater might be necessary, including disruptive or fluctuating selection or selection for a red-shifted allele in other freshwater habitats than blackwater lakes. Visual spectra in freshwater habitats rapidly change with depth, dissolved organic particles, and other biophysical properties, making more complex scenarios likely. Further study of cone opsin variation in additional freshwater and marine populations, taking ecological knowledge of visual environments into account, may provide better insight into the maintenance of variation and repeatability of adaptation to light spectra.
By combining population genomic data, functional genomic analysis and a selection experiment, we have uncovered the genetic mechanisms underlying repeated adaptation of color vision to divergent visual environments in threespine stickleback. This mechanism of adaptation is convergent at the molecular, functional, and ecological level with other fish species that have used 198 million-y-old paralogs to adapt to similar blackwater environments. Convergent evolution at the same gene happened thus at vastly different timescales, involving two mechanisms: repeated de novo mutation, leading to convergent amino acid changes, and the reuse of standing genetic variation for repeated adaptation in an adaptive radiation. Our study thus supports the emerging view that mechanisms underlying adaptive evolution are often highly repeatable and likely predictable [4] and that evolutionary tinkering with the same, constrained toolset can lead to convergent adaptation, both within species and between distantly related groups.
Stickleback collection followed guidelines for scientific fish collection in British Columbia, Canada, under Ministry of Environment permits SM09-51584 and SM10-62059. Collections in Naikoon Provincial Park and Drizzle Lake Ecological Reserve were carried out under park use permits: 103171, 103172, 104795, and 104796.
Among the more than 100 stickleback populations previously studied from the Haida Gwaii archipelago [16], a subset of 25 populations was chosen to comprise the full range of biophysical attributes of the freshwater habitats on Haida Gwaii, including water spectra, lake area, bathymetry, and predation regime. Stickleback from these 25 freshwater sites, two freshwater sites in coastal British Columbia [66], and one marine site were collected between 1993 and 2012, using minnow traps or recovery from salmon stomachs (marine sample, Table 1, for coordinates, see [16, 66]; coordinates BKW2: 53.375089°N, −130.177378°W). We selected one to four individuals per population for whole-genome resequencing. From the selection experiment [48], we chose 12 fish from the source population Mayer Lake and 11 from the population introduced into Roadside Pond (equivalent to “Mayer Pond” in [48]), the latter sampled in 2012, 19 y or approximately 13 generations after the release of 100 Mayer Lake fish, assuming a generation time of 1.5 y being intermediate between Mayer Lake (2 y generation time) and Roadside Pond (1 y generation time). In total, 58 individuals, 56 females and two males, were resequenced to 6x depth using paired-end Illumina reads as described in [6] at the Broad Institute.
We aligned reads to the Broad S1 reference [6] using BWA v0.5.9 [67] with parameters -q 5 -l 32 -k 2 -o 1 and recalibrated base qualities using the GATK v1.4 tools CountCovariates and TableRecalibration [68], with read group, quality score, cycle, and dinucleotide covariates in the recalibration model. This resulted in 2,992,040,331 aligned and recalibrated reads. Variants were called using GATK’s UnifiedGenotyper for each chromosome separately and all 58 individuals combined, with default parameters for SNP and indel calling, respectively. We removed variants with quality normalized by depth ≥ 2, read position rank sum test value ≥ −20, and allele-specific strand bias ≤ 200, using GATK’s VariantFiltration from the dataset. We recalibrated variants using the GATK’s VariantRecalibrator and ApplyRecalibration with a VQSR-LOD cutoff of 98.5%. Filtered and recalibrated variants were lifted over to an improved ordering of scaffolds in the reference stickleback genome [51] using Picard v2.2.1 (http://broadinstitute.github.io/picard). Also, we realigned base quality recalibrated reads to this improved reference using samtools v1.3 [69] and BWA v0.7.12 with the same alignment parameters as above, in order to enable read-backed phasing and read-based genotype likelihood–based analyses (see below), resulting in 2,935,821,595 aligned reads, which have been deposited on the NCBI short read archive under accession SRP100209.
We obtained a set of high-quality SNPs by removing all variants failing variant recalibration, variants with quality < 45 and with a mean depth > 9.51 (= average mean depth plus 1.5 times the interquartile range of the mean depth distribution), variants with less than four reads of each allele, variants with more than two alleles, and indels, using bcftools v1.3.1 [69]. This dataset was partitioned by chromosome, and males (individuals in populations Banks 70, Laurel) were removed from the sex chromosome XIX partition. SNPs were further split into an “adaptive radiation” and “selection experiment” SNPs partition. The “adaptive radiation” SNP partition contained one randomly picked individual from each of the 28 populations except the transplant population Roadside Pond (Table 1) in order to perform further analyses with equal sample size for all natural populations. The “selection experiment” SNP partition contained all 12 and 11 individuals from Mayer Lake and Roadside Pond, respectively. In both adaptive radiation and selection experiment SNP datasets, genotypes with less than four reads and sites with more than 50% missing genotypes were removed using vcftools v0.1.15 [70], resulting in 15.3% and 16.2% missing genotypes, respectively. Both the adaptive radiation and selection experiment SNP datasets were phased and missing genotypes imputed with the read-backed phasing algorithm implemented in SHAPEIT v2.r790 [71]. Phase-informative reads covered 9.3% of all heterozygote genotypes and 32.7% of all graph segments.
We scanned the genomic regions containing the four cone opsins for signatures of selective sweeps by using variation across the whole genome to identify outlier regions. We computed two haplotype-based statistics, integrated haplotype score iHS and H12 [49, 50], for the phased adaptive radiation SNPs. These statistics have been developed to detect signatures of incomplete hard and soft selective sweeps, based on extended haplotype homozygosity around an allele under selection compared to its alternate allele (iHS, [49]) or based on the haplotype frequency spectrum expected under a selective sweep (H12, [50]). Applied to the adaptive radiation dataset, these statistics will capture selective sweeps shared by multiple members of the adaptive radiation. iHS for each SNP in the genome was computed in selscan v1.1.0b [72] with default parameters and standardized in 5% allele frequency bins. In addition, we calculated the percentage of absolute iHS values > 2 in nonoverlapping 10 kb windows with more than 10 iHS estimates [49]. H12 was computed in bins spanning 81 SNPs using scripts published alongside the definition of H12 [50]. We used an outlier approach to identify significant iHS and H12 regions. We identified the top 0.1% genome-wide outliers for SNP-iHS, window-iHS, and H12 in recombination rate bins (<0.5, 0.5–2, 2–3.5, 3.5–5, >5 cM/Mb) because of the sensitivity of these statistics to variation in recombination rate. Local recombination rates were estimated from the “FTC x LITC”-cross recombination map published in [51] with a cubic splines smoothing approach described in [73]. For a significant outlier region indicating a selective sweep centered on opsins SWS2 and LWS, we used the top H12 estimates to identify the haplotype under selection or “sweep haplotype.” We visualized haplotype structure around a selective sweep in both adaptive radiation and selection experiment datasets using the extended haplotype homozygosity (EHH) statistic [74] calculated in selscan with default parameters.
We further traced the evolution of the selective sweep region around SWS2 and LWS in the selection experiment. For this, we computed population differentiation (FST) between Mayer Lake and Roadside Pond as well as nucleotide diversity (π) and Tajima’s D (TD) in each population across the genome and linkage disequilibrium (r2) in the selective sweep region for the unphased selection experiment SNPs. We first estimated the folded two-dimensional site-frequency spectrum (2D-SFS) from genotype likelihoods at all sites, from aligned reads with mapping-quality ≥ 17 and bases with quality ≥ 17 using angsd v0.911 [75, 76]. Using this 2D-SFS, we computed π and TD in 10 kb nonoverlapping as well as 10 kb wide, 2 kb step sliding windows across the genome for each population using angsd [76]. Then we estimated population allele frequencies with angsd and used them with the 2D-SFS to compute FST in 10 kb nonoverlapping as well as 10 kb wide, 2 kb step sliding windows across the genome using realSFS from the angsd software package [76, 77]. As for iHS and H12, we identified the top 0.1% outliers among nonoverlapping windows, based on the genome-wide distribution of π and TD in recombination rate bins. We computed linkage disequilibrium (r2) across the selective sweep region for SNPs with minor allele frequency (MAF) ≥ 5% and maximum 20% missing data in both populations using vcftools.
We identified synonymous- and nonsynonymous variation in the coding sequence of the four cone opsins SWS1, SWS2, RH2, and LWS in both the adaptive radiation and selection experiment SNPs. For the adaptive radiation SNPs, we tested whether coding variation at cone opsins was associated with the sweep haplotype identified above by using both alleles at each nonsynonymous SNP and chi-square tests with Bonferroni-corrected p-values. We also estimated the mean ratio of pairwise sequence divergence at synonymous and nonsynonymous sites (mean pairwise dN/dS) for all pairs of haplotypes in the adaptive radiation dataset using PAML v4.8 [78] and following the approach by Yang and Nielsen [79]. This statistic measures the relative frequency of segregating amino acid polymorphism to silent mutations [80]. We assessed whether any of the four cone opsins showed an unusually high frequency of amino acid changes by computing the distribution of mean pairwise dN/dS for all functional amino acid–coding genes on assembled chromosomes in our dataset (n = 17,846 genes).
We tested whether genetic variation at the four cone opsins was associated with variation in light spectrum, using three environmental proxies of light spectrum, percent light transmission at 400 nm (T400), lake depth in meters, and log-transformed lake area in square meters. We assigned SNPs to up- and downstream regulatory regions, introns, exons, and 3′/5′-untranslated regions of each cone opsin gene using SnpEff v4.2 [81] and combined all SNP alleles per gene into a single multidimensional scaling (MDS) coordinate in R v3.3.1 [82]. We used the MDS coordinate as a response variable in a general linear model with three predictor variables: T400, lake depth, and log-transformed lake area. To test more specifically for an association with blackwater environment, we repeated the general linear model analysis with a categorical light transmission variable “clearwater” for lakes with T400 > 74% and “blackwater” with T400 ≤ 74%, following [16]. Significance of effects was determined after Bonferroni-adjustment for multiple testing. We qualitatively assessed which SNPs are most strongly correlated with blackwater habitat from single-SNP chi-square tests for each gene-associated SNP.
For the selection experiment populations Mayer Lake and Roadside Pond, we estimated allele frequencies based on genotype likelihoods with angsd v0.911 [75] using raw-aligned reads of mapping quality > 17 for sites with quality > 17 and the GATK genotype likelihood model [68]. We computed allele-frequency changes for all variable sites in the genome and the empirical quantiles for absolute allele-frequency changes at SNPs surrounding SWS2 in Mayer Lake–based MAF bins of width 0.05. We also computed evolutionary change in haldanes at SWS2 key sites following equation 1 in [65], with raw allele frequency mean, standard deviations, and a generation time of 12.7 as input. Furthermore, we estimated the expected selection coefficient under a pure selection model, following equation 3.2 in [83], assuming incomplete dominance h = 0.5 and using a per generation allele-frequency change by dividing the observed allele-frequency change by 12.7 generations.
We reconstructed the evolutionary history of the cone opsin associated with a selective sweep, SWS2, using a Bayesian phylogenetic approach implemented in MrBayes v3.2.6 [84] and the same evolutionary model and run parameters as in [47]. For phylogenetic reconstruction, we used the two most common threespine stickleback SWS2 haplotypes, one associated and the other not associated with the sweep haplotype; an SWS2 sequence of blackspotted stickleback ([85], SRA-accession DRR013347); and both SWS2A and SWS2B paralogs from shorthorn sculpin ([47], genbank accession KM978046.1), medaka ([56], genbank accessions AB223056.1, AB223057.1), and bluefin killifish ([53], genbank accessions AY296737.2, AY296736.1). To get the blackspotted stickleback SWS2 sequence, we aligned whole-genome sequence data with a subset of the threespine stickleback reference sequence spanning the SWS2 coding sequence and 1 kb sequence up- and downstream of the start and stop codon, respectively, using bowtie2 v2.2.3 [86] with parameters -N 1 and -L 20, called genotypes using the GATK v3.5 tool UnifiedGenotyper [68] with default parameters and genotype likelihood model “BOTH,” converted the variants to FASTA format using the GATK-tool FastaAlternateReferenceMaker, and reintroduced missing sites using bedtools v2.25.0 [87]. We then aligned threespine stickleback and blackspotted stickleback SWS2 with the SWS2A and SWS2B paralogs of the other species manually in BioEdit v7.2.5 [88] and clipped the alignment to codons present in all sequences. We created the same alignment of codons for all 116 phased threespine stickleback haplotypes in our dataset to compute a neighbor-joining network using standard parameters in PopART v1.7 (http://popart.otago.ac.nz).
We tested whether selection on the threespine stickleback SWS2 haplotype lead to a rapid accumulation of amino acid–changing mutations compared to synonymous mutations (“positive selection”) on this haplotype. We used the phylogeny from above, estimated branch-specific dN, dS, and dN/dS and performed a branch-site test for positive selection with a subset of the phylogeny containing only stickleback SWS2 and other species’ SWS2A paralogs [60, 89], using PAML v4.8 [78] and the threespine stickleback’s sweep-associated haplotype as the foreground branch. Following [47], we also tested for gene conversion between SWS2A and SWS2B paralogs in the threespine stickleback ancestor by calculating dS between threespine stickleback SWS2 haplotypes and shorthorn sculpin SWS2A and SWS2B paralogs in 30 bp sliding windows with 1 bp step size in DnaSP v.5.10.01 [90]. After excluding gene conversion, we assessed whether amino acid substitutions at opsin key sites in the stickleback SWS2 haplotypes paralleled the divergence of ancestral SWS2A and SWS2B paralogs. We also added a partial coding sequence of black bream SWS2A and SWS2B paralogs ([26], genbank accessions DQ354580.1, DQ354581.1) to assess potential molecular and functional convergence with two other species inhabiting both clear- and blackwater (black bream, bluefin killifish).
|
10.1371/journal.pgen.1007217 | Genome-wide association across Saccharomyces cerevisiae strains reveals substantial variation in underlying gene requirements for toxin tolerance | Cellulosic plant biomass is a promising sustainable resource for generating alternative biofuels and biochemicals with microbial factories. But a remaining bottleneck is engineering microbes that are tolerant of toxins generated during biomass processing, because mechanisms of toxin defense are only beginning to emerge. Here, we exploited natural diversity in 165 Saccharomyces cerevisiae strains isolated from diverse geographical and ecological niches, to identify mechanisms of hydrolysate-toxin tolerance. We performed genome-wide association (GWA) analysis to identify genetic variants underlying toxin tolerance, and gene knockouts and allele-swap experiments to validate the involvement of implicated genes. In the process of this work, we uncovered a surprising difference in genetic architecture depending on strain background: in all but one case, knockout of implicated genes had a significant effect on toxin tolerance in one strain, but no significant effect in another strain. In fact, whether or not the gene was involved in tolerance in each strain background had a bigger contribution to strain-specific variation than allelic differences. Our results suggest a major difference in the underlying network of causal genes in different strains, suggesting that mechanisms of hydrolysate tolerance are very dependent on the genetic background. These results could have significant implications for interpreting GWA results and raise important considerations for engineering strategies for industrial strain improvement.
| Understanding the genetic architecture of complex traits is important for elucidating the genotype-phenotype relationship. Many studies have sought genetic variants that underlie phenotypic variation across individuals, both to implicate causal variants and to inform on architecture. Here we used genome-wide association analysis to identify genes and processes involved in tolerance of toxins found in plant-biomass hydrolysate, an important substrate for sustainable biofuel production. We found substantial variation in whether or not individual genes were important for tolerance across genetic backgrounds. Whether or not a gene was important in a given strain background explained more variation than the alleleic differences in the gene. These results suggest substantial variation in gene contributions, and perhaps underlying mechanisms, of toxin tolerance.
| The increased interest in renewable energy has focused attention on non-food plant biomass for the production of biofuels and biochemicals [1]. Lignocellulosic plant material contains significant amounts of sugars that can be extracted through a variety of chemical pretreatments and used for microbial production of alcohols and other important molecules [2–5]. However, there are major challenges to making biofuel production from plant biomass economically viable [6]. One significant hurdle with regards to microbial fermentation is the presence of toxic compounds in the processed plant material, or hydrolysate, including weak acids, furans and phenolics released or generated by the pretreatment process [7–10]. The concentrations and composition of these inhibitors vary for different pretreatment methods and depend on the plant feedstocks [7, 9, 11]. These toxins decrease cell productivity by generating reactive oxygen species, damaging DNA, proteins, cell membranes [12–14], and inhibiting important physiological processes, including enzymes required for fermentation [15], de novo nucleotide biosynthesis [16], and translation [17]. Despite knowledge of these targets, much remains to be learned about how the complete suite of hydrolysate toxins (HTs) acts synergistically to inhibit cells. Furthermore, how the effects of HTs are compounded by other industrial stresses such as high osmolarity, thermal stress, and end-product toxicity remains murky.
Engineering strains with improved tolerance to industrial stresses including those in the plant hydrolysate is of the utmost importance for making biofuels competitive with fuels already in the market [6]. A goal in industrial strain engineering is to improve lignocellulosic stress tolerance, often through directed engineering. Many approaches have been utilized to identify genes and processes correlated with increased stress tolerance, including transcriptomic profiling of cells responding to industrial stresses [18–21], genetic mapping in pairs of strains with divergent phenotypes [22–25], and directed evolution to compare strains selected for stress tolerance with starting strains [26–29]. However, in many cases the genes identified from such studies do not have the intended effect when engineered into different genetic backgrounds [30–33]. One reason is that there are likely to be substantial epistatic interactions between the genes identified in one strain and the genetic background from which it was identified [34]. A better understanding of how tolerance mechanisms vary across genetic backgrounds is an important consideration in industrial engineering.
Exploring variation in HT tolerance across strain background could also reveal additional defense mechanisms. The majority of functional studies in Saccharomyces cerevisiae are carried out in a small number of laboratory strains that do not represent the rich diversity found in this species [35, 36]. The exploration of natural diversity in S. cerevisiae has revealed a wide range of genotypic and phenotypic variability within the species [36–40]. In some cases, trait variation is correlated with genetic lineage [36, 41–43], indicating a strong influence of population history. At least 6 defined lineages have been identified in the species, including strains from Malaysia, West Africa, North America, Europe/vineyards, and Asia [41] as well as recently identified populations from China [38, 44]. In addition to genetic variation, phenotypic variation has cataloged natural differences across strains, in transcript abundance [37, 45, 46], protein abundance [47–49], metabolism [50–52], and growth in various environments [32, 36, 37, 42, 52–54]. Thus, S. cerevisiae as a species presents a rich resource for dissecting how genetic variation contributes to phenotypic differences. In several cases this perspective has benefited industry in producing novel strains by combining genetic backgrounds or mapping the genetic basis for trait differences [25, 55–59].
We used genome-wide association (GWA) in S. cerevisiae strains responding to synthetic hydrolysate (SynH), both to identify new genes and processes important for HT tolerance and to explore the extent to which genetic background influences mechanism. We tested 20 genes associated with HT tolerance and swapped alleles across strains to validate several allele-specific effects. However, in the process of allele exchange we discovered striking differences in gene contributions to the phenotype: out of 14 gene knockouts tested in two strains with opposing phenotypes, 8 (57%) had a statistically significant effect on HT tolerance in one of the backgrounds but little to no significant effect in the other background. In most of these cases, the specific allele had little observable contribution to the phenotype. Thus, although GWA successfully implicated new genes and processes involved in HT tolerance, the causal variation in the tested strains is not at the level of the allele but rather whether or not the gene’s function is important for the phenotype in that background. This raises important implications for considering natural variation in functional networks to explain phenotypic variation.
We obtained 165 Saccharomyces cerevisiae strains, representing a range of geographical and ecological niches, that have high quality whole genome sequencing reads (coverage ~30X), coming from published sequencing projects across the yeast community [39, 42, 52, 60] (S1 Table). We identified 486,302 high quality SNPs (see Methods). 68% of them had a minor allele frequency less than 5%. Nucleotide variation compared to the well-studied S288c-derived reference strain varied from as low as 0.08% for the closely related W303 lab strain and as high as 0.72% for the bakery strain YS4 (S1 Table). The majority of strains were largely homozygous (in some cases due to strain manipulation by sequencing projects); however, we identified 21 strains with >20% heterozygous sites. Most of these were from natural environments (11 strains) but they also included clinical samples (5 strains), baking strains (3 strains), a sugar cane fermenter (1) and a laboratory strain (FL100, which was scored as 98% heterozygous and may have mated with another strain in its recent history (S1 Table)).
Sixty-three percent of the variants were present in coding regions (S2 Table), which is lower than random expectation (since 75% of the yeast genome is coding) and consistent with purifying selection acting on most gene sequences. Indeed, coding variants predicted to have high impact, such as SNPs that introduce a stop codon, eliminate the start codon, or introduce a defect in the splicing region, were very rare (0.004% of genic SNPs)–a third of these were in dubious ORFs (22%) or genes of unknown function (8%) [61] that are likely nonfunctional and under relaxed constraint. However, 54 genes with debilitating polymorphisms are reportedly essential in the S288 background; nearly half of these polymorphisms are present in at least 3 strains and in some cases are lineage specific (S3 Table). Tolerance of these polymorphisms could arise through duplication of a functional gene copy [62], but could also arise due to evolved epistatic effects as has been previously reported [63], highlighting the complexity behind genetic networks and the role of genetic variation in determining their regulation.
Principal component analysis of the genomic data recapitulated the known lineages represented in the collection, including the European/wine, Asian/sake, North American (NA), Malaysian, West African (WA), and mosaic groups [36, 41, 42, 64] (S1 Table). Our analysis split the West African population into three subgroups not previously defined (Fig 1A). Construction of a simple neighbor-joining (NJ) tree broadly confirmed the population groups present in the 165-strain collection (Fig 1B).
We scored variation in lignocellulosic hydrolysate tolerance in several ways. Strains that are sensitive to hydrolysate grow slower and consume less sugars over time [65], thus we measured final cell density and percent of glucose consumed after 24 hours to represent SynH tolerance. Growth and glucose consumption were significantly correlated (R2 = 0.79), although there was some disagreement for particular strains (including flocculant strains) (S1 Table). We also determined tolerance to HTs specifically, to distinguish stress inflicted by HTs from effects of the base medium that has unusual nutrient composition and high osmolarity due to sugar concentration. To do this, we calculated the relative percent-glucose consumed and final OD600 in media with (SynH) and without HT toxins (SynH–HTs, see Methods) (S1 Table). Tolerance to SynH base medium without toxins (SynH–HT) and SynH with the toxins was only partly correlated (R2 = 0.48) (S1 Fig), suggesting that there are separable mechanisms of growing in base medium and surviving the toxins.
There is wide variation in tolerance to lignocellulosic hydrolysate that partly correlates with populations (Fig 2, S1 Fig). North American and Malaysian strains displayed the highest tolerance to SynH. As expected, phenotypic variation within each population was related to genetic variation, e.g. West African strains in Population 5 showed low genetic and phenotypic variation while mosaic strains with genetic admixture showed the widest range of phenotypes.
We used GWA to map the genetic basis for the differences in SynH tolerance, for each of the four phenotypes introduced above. The population signatures in S. cerevisiae are problematic for GWA, since the strong correlations between phenotype and ancestry obscure the identification of causal polymorphisms [66, 67]. To overcome this, we incorporated a large number of mosaic strains in the analysis and used a mixed-linear model to account for strain relationships, as implemented in the program GAPIT [68] (see Methods). We used as input SNPs that were present in at least 3 strains, eliminating 42% of SNPs in the dataset (see Methods). Of the remaining SNPs, 45% have a minor allele frequency of less than 5%; only those with an allele frequency >2% were used for GWA. GWA identified loci whose variation correlated with phenotypic variation. None of the GWA-implicated loci passed the stringent Bonferroni p-value correction based on the number of effective tests (see Methods), which is not uncommon for GWA at this scale [42, 69, 70]. We therefore used a somewhat arbitrary p-value cutoff of 1e-04 and performed additional filtering to minimize false positive associations (see Methods).
The combined analysis yielded 76 SNPs that met our p-value threshold (S4 Table, S2 Fig). Thirty-eight of these SNPs, linked to 33 genes, passed additional filtering (See Methods, Table 1). Of these, 17 SNPs are associated with growth in SynH, while 23 SNPs are associated with tolerance to HTs specifically (Table 1). Eight of the SNPs are intergenic and 20 are located within genes, with 13 of those predicted to change the coding sequence. Although we would expect that SNPs linked to HT tolerance should be identified in both sets of analyses, only 2 SNPs were significantly associated with both SynH and HT tolerance. This almost certainly highlights limited statistical power with the small set of strains used here. For most SNPs, the allele associated with tolerance was more frequent in our strain collection (Fig 3A), but for some it was the allele associated with sensitivity that was nearly fixed. We carried out additional GWA filtering to ensure that results were not driven by population structure (see Methods), since we note that many of sensitive alleles were prominent in the Asian population (S3 Fig). As expected for a largely additive trait, there was a significant linear correlation between the number of deleterious alleles a strain harbored and its tolerance to hydrolysate (R2 = 0.48, p = 2.2e-16, Fig 3B).
Interestingly, the genes associated with the 38 implicated SNPs capture functionally related processes, suggesting mechanistic underpinnings of hydrolysate tolerance. Lignocellulosic hydrolysate contains a large number of toxins that affect multiple cellular functions and can target energy stores, membrane fluidity, protein and DNA integrity, and other processes [10, 65]. Our analysis implicated several genes involved in redox reactions (ADH4, ALD3), protein folding or modification (CYM1, UBP5, UFD2, AOS1), ergosterol or fatty acid synthesis (ERG12, HMG2, NSG2), DNA metabolism and repair (REV1, DAT1, MCM5, SHE1), mRNA transcription and export (LEU3, SIR3, ELF1, RIM20, MEX67), mitochondrial function (MNE1, MAS1), and flocculation (FLO1, FLO10). Several of these processes were already known to be associated with hydrolysate stress, including flocculation [71], ubiquitin-dependent processes that may be linked to protein folding challenges [13, 72, 73], and sterol biosynthesis which affects tolerance to multiple stresses present in this media [32, 74, 75]. Nearly a third of these genes were identified as differentially expressed in our previous study comparing strain responses to SynH and rich medium [32], although this was not enriched above what is expected by chance. Thus, although gene expression differences can be informative in suggesting affected cellular processes, many of the genes implicated by GWA cannot be predicted by expression differences, especially SNPs that affect function without altering gene expression. Additional genes identified here belong to functional groups previously identified in our differential expression analysis, such as amino-acid and NAD biosynthesis.
We sought to confirm the importance of the GWA-implicated genes in SynH tolerance, first through gene-knockout analysis and then with allelic replacement in two different strains backgrounds. We began by knocking out 19 of the implicated genes in the tolerant North American strain, YPS128. Of these, 37% (7/19) of the knockout mutants had a significant phenotype when grown on SynH: four displayed decreased SynH tolerance, while 3 showed increased performance (Fig 4A). We note that 4 of the 7 knockouts had a mild phenotypic effect in standard growth medium (that was generally exacerbated in SynH), while 3 of these had a phenotype only in response to SynH (S4 Fig). The most significant knockouts decreasing tolerance in the YPS128 strain included the transcription factor LEU3, ribosomal protein RPL21B, protein phosphatase subunit SAP190, and to a milder extend the mitotic spindle protein SHE1. None of these genes has been directly implicated in tolerance to hydrolysate in previous studies.
The effect of deleting LEU3, which encodes the leucine-responsive transcription factor, was intriguing, since our prior work reported that amino-acid biosynthesis genes are induced specifically in response to HTs [32]. To confirm that this response was due to the toxicity found in the media and not due to amino acid shortage in SynH, we compared growth in synthetic complete (SC) medium, which has similar levels of branched-chain amino acids compared to SynH. The LEU3 knockout strain grew as well as the wild type in SC, but it grew to 54% lower final density in SynH–HT medium and 79% lower density in SynH medium with the toxins added (Fig 5A). The defect was not fully complemented by supplementing synthetic hydrolysate with 10X the normal amino acid mix (Fig 5B), indicating that amino acid shortage in the medium is unlikely to fully explain the growth defect.
The most striking phenotypic improvement was caused by deletion MNE1, encoding a splicing factor for the cytochrome c oxidase-encoding COX1 mRNA [76]. Aerobically, the mutant grew to roughly similar cell densities but consumed 44.7% more glucose and generated 64% more ethanol than the wild type, generating significantly more ethanol per cell (S5 Fig). A logical hypothesis is that this mutant has a defect in respiration and thus relies more on glycolysis to generate ATP and ethanol than wild-type cells [76]. Under this hypothesis, the effect of the mutation should be normalized when cells are grown anaerobically because both the mutant and wild type must rely on fermentation. However, under anaerobic conditions the mutant grew significantly better than the wild type (Fig 6A), consumed 70% more glucose (Fig 6B), and produced 63% more ethanol after 24-hour growth (Fig 6C). Thus, a simple defect in respiration is unlikely to explain the result, suggesting that Mne1 may have a separable role relating anaerobic toxin tolerance and/or metabolism.
We next knocked out 16 genes in the sensitive strain YJM1444, with the intention of allelic exchange (Fig 4B). We were unable to recover knockouts for some of the genes tested in YPS128, but of those we acquired 14 overlapped the YPS128 knockouts, and two (REV2 and HMG2) that we were unable to knock out in the tolerant strain were added. Remarkably, knockouts had strikingly different effects between the two genetic backgrounds–while three of the gene deletions affected hydrolysate tolerance in YJM1444, there was no overlap with the gene deletions causing a statistically significant effect in YPS128 (although some mild effects may be below our statistical power to detect). The three knockouts specific to YJM1444 improved SynH tolerance and included two genes involved in sterol biosynthesis (NSG2 and HMG2) and one involved in flocculation (Fig 4B). In fact, deletion of FLO1 dramatically reduced the flocculation phenotype of YJM1444 and resulted in >236% increased glucose consumption in SynH. This single mutation converted YJM1444 tolerance to the level of SynH tolerance seen in YPS128 (S6 Fig). To test that this phenotypic effect was directly caused by the FLO1 allele, we deleted its paralog FLO5, which caused neither a change in flocculation nor increased glucose consumption of the culture (S7 Fig).
There appeared to be subtle, but not significant, effects of the MNE1 deletion in YJM1444 and we wondered if the was obscured by flocculation. Therefore, we measured glucose consumption in high-rpm shake flasks that disrupt flocculation. Indeed, MNE1 deletion had a significant benefit under these conditions; however, the magnitude of the effect was more subtle than MNE1 deletion in YPS128 (S8A Fig). We also tested this deletion in an industrial strain, Ethanol Red (E. Red). Deletion of MNE1 in a haploid spore derived from E. Red produced a minor, reproducible benefit although it was not statistically significant (S8A Fig). Nonetheless, these results show that MNE1 plays a role in SynH tolerant, albeit to different levels, in three different strain backgrounds.
We tested allelic effects in two ways. First, we introduced a plasmid-borne copy of the tolerant allele or sensitive allele (S5 Table) into YPS128 lacking the native gene, and measured percent final glucose consumption in SynH (S9 Fig) in synthetic complete medium (required to allow drug-based plasmid selection) with HTs (Fig 7A). The assay was fairly noisy, nonetheless, there was a clear effect for the FLO1 allele, which caused YPS128 to become flocculant and dramatically decreased growth in the SC with HTs. We did not observe other allele-dependent effects that overcame the variability of the assay, including for the genes whose knockout produced a defect in YPS128. Second, we performed reciprocal hemizygosity analysis for six genes, including three genes that whose deletion produced differential effects in YPS128 and YJM1444. We crossed the YPS128 and YJM1444 backgrounds such that the resulting diploid was hemizigous for either the tolerant or sensitive allele (Fig 7B). In this case, none of the six genes had an allele-specific effect–surprisingly, this included FLO1 for which there was clear allelic impact in the haploid backgrounds. We realized a unique phenotype in the YPS128-YJM1444 hybrid: whereas the strain is heterozygous for the functional FLO1 allele, the hybrid lost much of the flocculence of the YJM1444 strain (S8B Fig). FLO1 expression is known to be repressed in some diploid strains [77]. Thus, simply mating the strains in effect created a new genetic background that changed the allelic impact of the gene.
We wondered if this effect explained the lack of allele-specific phenotypes for other implicated genes. We therefore created homozygous deletions in the diploid hybrid for six genes whose deletion had strain-specific impacts in the haploids (Fig 7C). Two of the knockouts (leu3Δ and sap190Δ) produced a defect in the hybrid, similar to the effect seen in YPS128. Homozygous deletion of MNE1 produced a unique growth defect in 24-well plates that was not seen in the haploids or the hemizigous diploids. This appeared to be due to increased flocculation in the hybrid diploid; growth in shake flasks to disrupt flocculation resulted in a mild but statistically insignificant benefit to the hybrid when grown in flasks, similar to that seen for YJM1444. In contrast, deletion of RIM20 or FLO1 had no effect under these growth conditions–this explains the lack of allele specific effect, because the genes are no longer important in this background and under these growth conditions.
Mating YJM128 and YJM1444 created a new background that surpassed performance of YPS128 (Fig 7D). We wondered if hybridization could benefit other strains as well. We mated industrial strain E. Red crossed to YJM1444 and YPS128. E. Red and YJM1444 were both scored as sensitive and perform similarly in SynH (Fig 7D). However, the hybrid had a striking jump in SynH tolerance, exceeding the tolerance of YPS128. This benefit may be in part because the new diploid background changes the flocculation phenotype. On the other hand, YJM1444 and E. Red harbor alternate alleles at 71% of the SNPs implicated by GWA, raising the possibility that complementation of recessive alleles could also contribute to the strain improvement (see Discussion).
Engineering strains for tolerance to lignocellulosic hydrolysate has proven difficult due to the complex stress responses required to deal with the combinatory effects of toxins, high osmolarity, and end products such as alcohols and other chemicals. Even when the cellular targets of stressors are known, the mechanisms for increasing tolerance are not always clear. We leveraged phenotypic and genetic variation to implicate new mechanisms of hydrolysate tolerance, by finding correlations between phenotypic and genetic differences among a collection of Saccharomyces cerevisiae strains, which allowed us to implicate specific genes and alleles involved in hydrolysate tolerance. The results indicate several important points relevant to engineering improved hydrolysate tolerance and genetic architecture of tolerance more broadly.
Perhaps the most striking result is the level to which gene involvement varies across the strains in our study. We expected that swapping alleles of implicated SNPs should contribute to variation in the phenotype. Most alleles did not detectably affect tolerance, although it is likely that they may have a minor contribution below our limit of detection. Indeed, strains that harbor more deleterious alleles are significantly more sensitive to SynH (Fig 3B), as expected for an additive trait. But at the same time, we uncovered significant variation in whether the underlying gene was involved in the phenotype. Among the genes that we were able to knockout in both strains (14 genes), 57% produced a phenotype (to varying levels and significance) in one of the two strains we tested. This indicates substantial epistatic interactions with the genetic background, such that the gene is important in one strain and but dispensable in another. Even more striking is the case of FLO1: knocking out the functional gene in YJM1444 produced a major benefit to that strain, whereas introducing the functional allele to YPS128 was very detrimental to SynH tolerance. Yet neither the allele nor the gene itself influenced SynH tolerance in the hybrid, because the hybrid is much less flocculant under these conditions (despite carrying functional YJM1444 FLO1 gene).
While it may not be surprising that gene knockouts result in quantitatively different phenotypes, we did not expect that most knockouts would have no detectible effect in specific backgrounds. It will be important to investigate the extent to which this effect is true in other organisms and for other phenotypes. However, evidence in the literature hints at the breadth of this result: several genes are required for viability in one yeast strain but not another [63, 78], while overexpression of other genes produces a phenotype in one background but not others [32]. Genetic background effects on gene contributions have been reported before, in yeast and other organisms [35, 79–84]; however, the extent to which different genes appear to be involved in toxin tolerance in the different strains studied here suggests an important consideration that is underappreciated in GWA analysis: that the network of genes contributing to the phenotype could be largely different depending on genomic context. Dissecting these epistatic interactions is likely to be daunting, since a major challenge in most GWA studies remains identifying the epistatic interactions due to the high statistical hurdle [34, 85, 86]. We propose that emerging network-based approaches to augment linear contributions will be an important area in identifying genetic contributions in the context of background-specific effects.
QTL mapping has allowed the characterization of the genetic architecture of industrially relevant stresses, including tolerance to ethanol [22, 87], acetic acid [23, 56], and plant hydrolysate [25] among many others [24, 88–90]. But while this method exploits the genetic diversity between two strains, with GWA we were able to study a much larger collection of genetic diversity, providing unique insights. SynH tolerance is clearly a complex trait, with many genes likely contributing. Previous studies have shown that part of the growth inhibition can be explained by a re-routing of resources to convert toxins into less inhibitory compounds [18, 19, 91–94] and to repair damage generated by reactive oxygen species in membranes and proteins [13, 14, 95]. One of the most significant effects was caused by deletion of LEU3, the transcription factor regulating genes involved in branched amino acid biosynthesis. Interestingly, weak acids have been shown to inhibit uptake of aromatic amino acids causing growth arrest [96], and it is possible that Leu3 is required to combat this effect. Chemical genomic experiments suggest an additional role for Leu3 in managing oxidative stress in the cell [97], which could relate to oxidative stressors in hydrolysate [13, 14, 32]. We also uncovered a gene, Mne1, that when deleted significantly increases ethanol production in SynH. Mne1 aids the splicing of COX1 mRNA [76] and has not been previously linked to stress tolerance. Interestingly, MNE1 mutants produced more ethanol per cell aerobically, but also grew substantially better in SynH anaerobically, raising the possibility that Mne1 plays an additional, unknown role in cellular physiology that can be utilized to increase fermentation yields. Finally, although flocculation has been previously shown to increase cell survival in hydrolysate [71], our study showed that flocculation reduced the rate of sugar consumption in the culture, likely because cells in the middle of the clump are nutrient restricted. Together, these results shed new light on SynH tolerance and mechanisms for future engineering.
Our results raise broader implications for strain engineering, based on the genetic architectures uncovered here. Given the implication of gene-by-background interactions, the best route for improving strain performance may be crossing strains for hybrid vigor [98–100]. Indeed, we unexpectedly generated a strain that outperformed the tolerant YPS128, by crossing two poor performers in SynH. This improved vigor could emerge if the hybrid complements recessive deleterious alleles in each strain, or if mating creates a new genetic background that changes the requirements (and fitness) of the strain. We believe that both models–weak but additive allelic contributions in the context of epistatic background effects–are at work in our study. For additive traits, GWA and genomic studies can have significant practical power, by predicting where individual strains fall on the genotype-phenotype spectrum and by suggesting which strains should be crossed for maximal phenotypic effect.
Strains used in the GWA are listed in S1 Table Gene knockouts were performed in strains derived from North American strain YPS128 and mosaic strain YJM1444. The homozygous diploid parental strains were first engineered into stable haploids by knocking out the homothallic switching endonuclease (HO) locus with the KAN-MX antibiotic marker [101], followed by sporulation in 1% potassium acetate plates and dissection of tetrads to attain heterothallic MATa and MATα derivatives. Gene knockouts were generated through homologous recombination with the HERP1.1 drug resistance cassette [102] and verified by 3 or 4 diagnostic PCRs (validating that the cassette was integrated into the correct locus and that no PCR product was generated from within the gene that was deleted). Most knockouts removed the gene from ATG to stop codon, but in some cases (e.g. kdx1) additional flanking sequence was removed, without removing neighboring genes. Genes from YPS128 or strains carrying the sensitive allele (S5 Table) were cloned by homologous recombination onto a CEN plasmid, taking approximately 1,000 bps upstream and 600 bps downstream from each genome, and verified by diagnostic PCR. Phenotyping of strains harboring alternate alleles on plasmids was performed in as previously described, except that the pre-culture was grown in YPD with 100 mg/L nourseothricin (Werner BioAgents, Germany) to maintain the plasmid expressing each allele. We note that plasmid-bourn expression of the gene complemented the gene-deletion phenotype, where applicable, in all cases tested (not shown). Allele specific effects were additionally tested by reciprocal hemizygosity analysis (RHA) [103]. The HO locus was replaced with the nourseothricin resistance cassette (NAT-MX) for each mating type of YPS128 and YJM1444. These were then crossed with the appropriate deletion strain of opposite mating type and harboring the KanMX cassette, selecting for mated cells resistant to both drugs, to generate heterozygous strains that were hemizigous for the gene in question (crosses shown in S6 Table).
Synthetic Hydrolysate (SynH) medium mimics the lignocellulosic hydrolysate generated from AFEX ammonium treated corn stover with 90 g glucan/L loading and was prepared as in Sardi et al. (2016). Two versions were prepared to represent the complete hydrolysate (SynH) and the hydrolysate without the hydrolysate toxin cocktail (HT) (SynH—HT), as previously published [32].
Phenotyping for GWA, gene deletion assessment, and RHA, was performed using high throughput growth assays in 24 well plates (TPP® tissue culture plates, Sigma-Aldrich, St. Louis, MO). To prepare the cultures, 10 μl of thawed frozen cell stock were pinned onto YPD agar plates (1% yeast extract, 2% peptone, 2% dextrose, 2% agar) and grown for 3 days at 30°C. Cells were then pre-cultured in 24 well plates containing 1.5 ml of YPD liquid, sealed with breathable tape (AeraSeal, Sigma-Aldrich, St. Louis, MO), covered with a lid and incubated at 30°C while shaking for 24 h. Next, 10 μl of saturated culture was transferred to a 24 well plate containing 1.5 ml of SynH or SynH-HT where indicated, and grown as the preculture for 24 h. Cell density was measured by optical density at 600 nm (OD600) as ‘final OD’. Culture medium collected after cells were removed by centrifugation was used to determine glucose concentrations by YSI 2700 Select high performance liquid chromatography (HPLC) and refractive index detection (RID) (YSI Incorporated, Yellow Springs, OH). Biological replicates were performed on different days.
For GWA, we used four different but related phenotype measures of cells growing in SynH or SynH–HTs: 1) final OD600 as a measure of growth, 2) percent of starting glucose consumed after 24 hours in SynH, 3) HT tolerance based on OD600 (calculated as the ratio of final OD600 in SynH versus final OD600 in SynH -HTs), and 4) HT tolerance based on glucose consumption (calculated as the ratio of glucose consumed in SynH versus in SynH -HTs). Strains and phenotype scores are listed in S1 Table. Initial phenotyping for GWA was performed in biological duplicates; knockout strains and hemizigous strains were phenotyped in five biological replicates to increase statistical power, whereas homozygous deletion strains were phenotyped in triplicate. Replicates for each batch of strains shown in each figure were performed on separate days, for paired statistical analysis.
Experiments done for allele replacements expressed on plasmids were performed in glass tubes using modified synthetic complete medium (SC) with high sugar concentrations and the toxin cocktail where indicated (Sardi et al 2016) to mimic SynH but with no ammonium to support nourseothricin selection [104] (1.7 g/L YNB w/o ammonia sulfate and amino acids, 1 g/L monosodium glutamic acid, 2 g/L amino acid drop-out lacking leucine, 48 μg/L leucine, 90 g/L dextrose, 45 g/L xylose). This was required since nourseothricin selection does not work in high-ammonium containing SynH. First, we precultured strains carrying plasmids in SC medium with nourseothricin (200 ug/ml) for 24 h. Next, we inoculated a fresh culture at a starting OD600 of 0.1 in 7 ml of the modified synthetic complete medium with nourseothricin (200 ug/ml) and HTs. Cultures were grown for 24 h and phenotyped as described above. Replicates were performed on different days, and thus samples were paired by replicate date for t-test analysis.
Anaerobic phenotyping was performed in the anaerobic chamber, where cells were grown in flasks containing 25 ml SynH or SynH-HT and maintained in suspension using a magnetic stir bar. Ethanol production was measured over time by HPLC RID analysis. Paired t-test analysis was performed to determine significance, pairing samples by replicate date.
We obtained publicly available whole genome sequencing reads from Saccharomyces cerevisiae sequencing projects [39, 42, 52, 60]. Sequencing reads were mapped to reference genome S288C (NC_001133, version 64 [105]) using bwa-mem [106] with default settings. Single nucleotide polymorphisms (SNPs) were identified using GATK [107] Unified Genotyper, analyzing all the strains together to increase detection power. GATK pipeline included base quality score calibration, indel realignment, duplicate removal, and depth coverage analysis. Default parameters were used except for -mbq 25 to reduce false positives. Variants were filtered using GATK suggested criteria: QD < 2, FS > 60, MQ < 40. A dataset with high quality SNPs was generated using VCFtools [108] by applying additional filters of a quality value above 2000 and excluding sites with more that 80% missing data. Genetic variant annotation was performed using SNPEff [109]. Principal component analysis and the neighbor-joining tree were performed with the R package Adegenet 1.3–1 [110] using the entire collection of high quality SNPs (486,302 SNPs). SNP data are available in the EBI under accession number PRJEB24747.
Correlations between genotype and phenotype were performed using a mixed linear model implemented in the software GAPIT [68]. Only SNPs with a minor allele frequency (MAF) of at least 2% were used for this analysis (282,150 SNPs). Multiple models, each incorporating a different number of principal components to capture population structure (from 0–3), were analyzed. The final model was manually chosen as the one with the greatest overall agreement between the distribution of expected and the observed p-values, i.e. based on QQ plots with the least skew across the majority of SNPs. We performed four analyses, one for each for the four related phenotypes measured. The model used to map SynH final OD600 and SynH percent glucose consumed used 0 principal components, with population structure corrected using only the kinship generated by GAPIT. The model used to map HT tolerance based on relative final OD600 used 2 principal components, and the model to map HT tolerance based on glucose consumed incorporated 1 principal component. The threshold for significance accounting for multiple-test correction was identified by dividing the critical p-value cutoff of 0.05 by the number of independent tests estimated by the SimpleM method [111], which decreased the number of tests from 282,150 to 137,398 to produce a p-value threshold of 3.6e-7 [112]. However, none of our tests passed this threshold, which is likely overly conservative. We therefore used a p-value threshold of 1e-04 to identify genes for detailed follow-up analysis. We realized that the extreme phenotypes of Asian/sake strains coupled with their strong population structure might be confounding the analysis [66]. Therefore, to further reduce the chance of false positives due to residual population influences, we reran the analyses without the 11 sake strains and removed from the original list of significant SNPs those with p>5e-3. For each locus carrying a significant SNP, we plotted phenotypic distributions for each possible genotype. We focused subsequent downstream analysis on individual SNPs whose effects were additive across strains that were heterozygous and homozygous at that site, assessed visually. Genes affected by each SNP were determined by the SNPEff annotation, which predicted the effect of variants on genes.
|
10.1371/journal.ppat.1000757 | IPS-1 Is Essential for the Control of West Nile Virus Infection and Immunity | The innate immune response is essential for controlling West Nile virus (WNV) infection but how this response is propagated and regulates adaptive immunity in vivo are not defined. Herein, we show that IPS-1, the central adaptor protein to RIG-I-like receptor (RLR) signaling, is essential for triggering of innate immunity and for effective development and regulation of adaptive immunity against pathogenic WNV. IPS-1−/− mice exhibited increased susceptibility to WNV infection marked by enhanced viral replication and dissemination with early viral entry into the CNS. Infection of cultured bone-marrow (BM) derived dendritic cells (DCs), macrophages (Macs), and primary cortical neurons showed that the IPS-1-dependent RLR signaling was essential for triggering IFN defenses and controlling virus replication in these key target cells of infection. Intriguingly, infected IPS-1−/− mice displayed uncontrolled inflammation that included elevated systemic type I IFN, proinflammatory cytokine and chemokine responses, increased numbers of inflammatory DCs, enhanced humoral responses marked by complete loss of virus neutralization activity, and increased numbers of virus-specific CD8+ T cells and non-specific immune cell proliferation in the periphery and in the CNS. This uncontrolled inflammatory response was associated with a lack of regulatory T cell expansion that normally occurs during acute WNV infection. Thus, the enhanced inflammatory response in the absence of IPS-1 was coupled with a failure to protect against WNV infection. Our data define an innate/adaptive immune interface mediated through IPS-1-dependent RLR signaling that regulates the quantity, quality, and balance of the immune response to WNV infection.
| West Nile virus (WNV) is a mosquito-transmitted RNA virus that has emerged in the Western hemisphere and is now the leading cause of arboviral encephalitis in the United States. However, the virus/host interface that controls WNV pathogenesis is not well understood. Previous studies have established that the innate immune response and interferon (IFN) defenses are essential for controlling virus replication and dissemination. In this study, we assessed the importance of the RIG-I like receptor (RLR) signaling pathway in WNV pathogenesis through analysis of mice lacking IPS-1, the central adaptor molecule of RLR signaling. Our studies revealed that IPS-1 is essential for protection against WNV infection and that it regulates processes that control virus replication and triggering of innate immune defenses. We found that IPS-1 plays an important role in establishing adaptive immunity through an innate/adaptive interface that elicits effective antibody responses and controls the expansion of regulatory T cells. Thus, RLRs are essential for pathogen recognition of WNV infection and their signaling programs help orchestrate immune response maturation, regulation of inflammation, and immune homeostasis that define the outcome of WNV infection.
| West Nile virus (WNV) is a neurotropic flavivirus and is an emerging public health threat. Infection with WNV now constitutes the leading cause of mosquito-borne and epidemic encephalitis in humans in the United States [1]. WNV is enveloped and contains a single strand positive sense RNA genome of approximately 11 kb in length that encodes three structural (C, prM/M, and E) and seven non-structural proteins (NS1, NS2A, NS2B, NS3, NS4A, NS4B, and NS5). It cycles enzootically between birds and Culex mosquitoes, with humans infected as dead-end hosts. WNV infection has been modeled in inbred mice wherein infection and pathogenesis recapitulate many of the features of human infection (reviewed in [2]). Following subcutaneous inoculation, WNV replicates in dendritic cells (DCs) at the portal of entry and in the draining lymph node. A primary viremia develops and virus spreads to visceral organs including the spleen, where further amplification occurs, leading to central nervous system (CNS) dissemination and encephalitis. In humans, WNV causes an acute febrile illness that can progress to severe and sometimes lethal neuroinvasive disease, especially in the elderly and immunocompromised [3]. However, healthy young adults are also afflicted with severe neurological disease [4],[5],[6], indicating that virulence can occur independently of immune deficiencies or aging.
Intracellular innate immune defenses and the actions of type I interferon (IFN) provide a first-line of defense against virus infection and are essential for the control of WNV replication, dissemination, and neurovirulence [7]. Innate antiviral immune defenses are triggered through the recognition of conserved pathogen associated molecular pattern (PAMP) motifs within viral products by intracellular pathogen recognition receptor (PRR) proteins in infected cells. PRR signaling directs downstream activation of latent transcription factors, including NF-κB, interferon regulatory factor (IRF)-3 and IRF-7, in a cell type-specific manner to induce antiviral response programs that include expression of proinflammatory cytokines, chemokines, type I IFN, and interferon stimulated genes (ISGs) [7],[8],[9],[10]. The ISG products induced through autocrine and paracrine actions of IFN confer antiviral activity by limiting virus replication and cell-to-cell virus spread. Modulation of IFN signaling has been identified as a virulence feature of pathogenic strains of WNV [11],[12].
The RLRs, retinoic acid inducible gene-I (RIG-I) and melanoma differentiation antigen 5 (MDA5) [13],[14],[15],[16], are PRRs that play critical roles in triggering immune defenses against RNA virus infection, including WNV. RIG-I and MDA5 are cytosolic RNA helicases that contain an amino terminal tandem caspase activation and recruitment domain (CARD). Upon engaging RNA substrates, the RLRs undergo a conformational change and bind to the mitochondrial associated protein, interferon promoter stimulator-1 (IPS-1) through a CARD-CARD interaction, leading to IPS-1-dependent signaling of IFN production and expression of immune response genes [17],[18]. RLR signaling and IPS-1 function have an essential role in triggering IFN defenses during WNV infection of mouse embryo fibroblasts (MEFs) and human cell lines in vitro. Cells lacking either RIG-I or MDA5 were attenuated in their ability to generate an effective innate immune response to infection, whereas cells lacking both RIG-I and MDA5 or those deficient in IPS-1 alone were unable to respond to infection with WNV and related flaviviruses [19],[20],[21],[22]. Recent studies examined the role of another class of pattern recognition receptors, Toll like receptor (TLR)3 and TLR7, and show that these receptors are also important PRRs of WNV infection, as they play a role in signaling IFN production and an inflammatory response upon viral ligand recognition [23],[24],[25]. TLR3 has been shown to contribute to both enhancement and protection of CNS inflammation and neurovirulence of WNV in vivo [23],[24], while TLR7-dependent signaling was shown to be essential for directing proper immune cell homing to sites of WNV infection during the adaptive immune response in vivo [25].
Type I IFN, a major product of PRR signaling, has been shown to link innate and adaptive immune responses. However, the specific PRR pathways that mediate this during acute WNV infection have not been delineated nor has the RLR pathway been evaluated in this context. The quantity and quality of the innate and adaptive immune responses after infection must be carefully regulated to avoid aberrant inflammation and immunopathogenesis. Regulatory T (Treg) cells and inflammatory dendritic cell (DC) subsets regulate inflammation during acute virus infection through T cell suppression and by modulating the trafficking and inflammatory cytokine production of immune cells into infected tissues [26],[27],[28]. Thus, the level of local and peripheral Treg cells, and the composition of local DC subsets that develop during WNV infection may determine immune control and WNV disease.
Here, we assessed the role of RLR signaling and IPS-1 in WNV infection and immunity. Our studies define IPS-1 as an essential modulator of immunity in vivo and demonstrate that IPS-1-dependent signaling orchestrates an innate/adaptive immune interface that regulates immune responses to effectively control WNV infection.
WNV infection of primary embryonic fibroblasts recovered from RIG-I−/− mice revealed that RIG-I was important in eliciting innate antiviral immune defenses early during infection, whereas MDA5 was important for enhancing and sustaining this response [21]. We further evaluated WNV infection of RIG-I−/− or MDA5−/− mice and confirmed that RIG-I serves a dominant role among the RLRs for the acute induction of innate immune defenses and protection against WNV infection in vivo (data not shown). Since the RLRs signal innate defenses through the IPS-1 adaptor protein [29], we also examined the role of IPS-1 in protection against WNV infection upon a sub-lethal virus challenge of wild type and IPS-1−/− mice. IPS-1−/− mice were highly susceptible to WNV infection and exhibited 100% mortality with an average survival time (AST) of 7.3 days as compared to wild type mice (38.5% mortality with an AST of 13.2 days; p<0.0001; Fig 1A). Thus, RIG-I and IPS-1-dependent signaling are essential for protection against WNV infection.
To define the role of IPS-1 in controlling WNV in vivo, wild type and IPS-1−/− mice were infected subcutaneously (s.c.) with 100 PFU of WN-TX and viral burden within peripheral tissues and the CNS was measured over time post-infection (pi). IPS-1−/− mice exhibited increased viremia compared to wild type mice (45.7 fold enhancement at day 1 pi, P<0.05) throughout the course of infection (Fig 1B). Similarly, viral loads in the spleen were elevated in the infected IPS-1−/− mice (Fig 1C). WNV infection of IPS-1−/− mice displayed an expanded tissue tropism as infectious virus was found in the kidneys, a tissue that is not normally permissive to infection in wild type mice (Fig 1D). WNV is typically detected in the CNS of wild type mice after s.c. challenge between 4 and 8 days pi [2]. Consistent with this time course, infected wild type mice exhibited detectable viral loads (average viral titer of 101.8 pfu/gram of tissue) in the brain by day 6 p.i., although virus was not detected in the spinal cord (Fig 1E and F). In contrast, WNV spread to the brain (Fig 1E) and spinal cord of IPS-1−/− mice (Fig 1F) by day 2 pi, with viral loads rising through day 6 pi. Together these results indicate that IPS-1, likely through RLR signaling of innate immune defenses, limits WNV replication, viremia, and peripheral spread, and is essential for the control of viral invasion of the CNS.
Myeloid cells, including tissue and lymphoid DC and macrophages (Mφ), are among the first cells to encounter WNV during infection and thus function to restrict the spread of virus to distant tissues and the CNS [2]. To define the role of IPS-1 in controlling virus replication and innate immunity in myeloid cells, we analyzed WNV infection and host responses in primary bone marrow-derived DC and Mφ recovered from wild type and IPS-1−/− mice. DC and Mφ were infected at an MOI of 1.0 (relative to viral plaque assay quantification of BHK-21 cells; see Methods) and evaluated for virus replication, IFN induction, and innate immune triggering of ISG expression (Fig 2). IPS-1−/− DCs sustained significantly higher WNV replication at 36 and 48 hours pi compared to wild type infected cells (Fig 2A). WNV infection of wild type DCs induced IFN-β secretion but this response was completely abolished in IPS-1−/− DCs (Fig 2B). The lack of IFN-β induction in IPS-1−/− DCs correlated with a lack of ISG expression including RIG-I, MDA5, and STAT-1 (Fig 2C). In addition, expression of ISG54 and ISG49, which are direct IRF-3 target genes [30],[31], were not induced during WNV infection of IPS-1−/− DCs (Fig 2C). Moreover, ISG56, another IRF-3 target gene [31], was induced late during infection and to lower levels as compared to ISG54 and ISG49 in wild type, infected DCs. WNV infection of IPS-1−/− Mφ resulted in significantly higher virus replication between 24 and 48 hours pi as compared to infected wild type cells (Fig 2D). Whereas wild type infected Mφ expressed IFN-β, this response was completely abolished in IPS-1−/− Mφ (Fig 2F). We also observed a differential expression of ISGs and IRF-3-target genes within WNV-infected Mφ. RIG-I, MDA5, and STAT-1 were not induced in IPS-1−/− Mφ, whereas, ISG56, ISG49, and PKR were expressed at reduced levels and with delayed kinetics. These data establish that IPS-1-dependent RLR signaling is the major innate immune signaling pathway that controls virus replication in conventional DCs and Mφ.
Neurons represent the target cell of WNV infection in the CNS and their death after infection is a key factor in pathogenesis and neurological sequelae [32],[33]. To define the role of RLR signaling in restricting virus replication in neurons, primary cortical neurons were generated from wild type and IPS-1−/− mice. Cells were infected at an MOI of 1.0 with WN-TX and virus yield, IFN-β induction, and ISG expression were evaluated. In the absence of IPS-1, WNV replicated faster and to higher levels resulting in a 2.2 and 4.2-fold (p<0.05) increase in viral production at 24 hrs and 48 pi, respectively as compared to infected wild type neuronal cells (Fig 3A). This relatively modest virologic effect in neurons compared to that observed in IPS-1−/− DC and Mφ was expected, as IFN-α or -β pre-treatment only inhibits WNV infection in cortical neurons to a maximum of 5 to 8-fold [12], suggesting that the IFN response is comparably less potent in neurons. IFN-β expression was induced to lower levels in IPS-1−/− neurons compared to wild type infected neurons at 24 (10-fold, p<0.05) and 36 hours pi (5-fold, p<0.05) despite the higher levels of virus replication (Fig 3A and 3B). Expression of ISGs, (including RIG-I and MDA5) and IRF-3 target genes (including ISG56 and ISG49) followed this pattern and were dependent on IPS-1 for rapid and high level expression (Fig 3C). The presence of IFN-β and ISG transcripts in IPS-1−/− cells at 48 hrs pi is consistent with the finding that TLR3 has an independent and subordinate role in triggering innate immune responses in cortical neurons at later time points after WNV infection [23]. These results demonstrate that the RLR signaling pathway controls virus replication and induces innate immune responses against WNV infection in cortical neurons.
To determine the role of the RLR pathway in protection of neurons against WNV pathogenesis in vivo, we conducted histological analysis of brain tissue from wild type and IPS-1−/− mice infected with WN-TX (Fig 4A). Analysis of brain sections from infected wild type mice revealed little or no inflammation or neuronal damage, with sparse and focal cell infiltrates restricted to the hippocampus and cerebral cortex on day 6 pi. By day 10 pi (a timepoint in wild type mice in which peak virus replication in the CNS occurs [34]) cellular infiltrates were present in the parenchyma and neuronal destruction was observed throughout the cortex and hippocampus. In contrast, brain sections from infected IPS-1−/− mice recovered on day 6 pi displayed extensive injury to neurons in the cortex and granular neurons of the hippocampus. Damaged neurons appeared pyknotic with vacuolation, degeneration and cell dropout. Somewhat surprisingly, we observed extensive inflammation in the brains from infected IPS-1−/− mice within the cortex, hippocampus, and cerebellum (data not shown) displaying prominent perivascular and parenchymal immune cell infiltrates.
To evaluate the composition and antigen-specificity of the inflammatory cells within the brains of WNV-infected mice, lymphocytes were isolated from infected brains on day 6 pi and were characterized from pools (n = 5) of wild type and IPS-1−/− infected mice. Brains from IPS-1−/− infected mice showed an 2.9-fold increase in the total number of immune cells as compared to wild type infected mice (Fig 4B), and this was associated with an increase in absolute numbers of infiltrating CD4+ and CD8+ T cells (Fig 4C). Among the brain CD8+ T cells isolated from IPS-1−/− mice, there was a remarkable 27-fold increase in the number of antigen-specific cells that expressed IFN-γ after treatment with an immundominant NS4B peptide (Fig 4D) [35],[36]. Analysis of microglia/Mφ cells, based on relative surface expression of CD45 and CD11b [37], revealed increased numbers of microglial cells (CD45+lo/CD11b+) and infiltrating macrophages (CD45+hi/CD11b+) within the brains of infected IPS-1−/− mice when compared to wild type mice (Fig 4E). Similar findings were observed in the spinal cords from infected IPS-1−/− mice (data not shown). Combined with the histological analysis, these results demonstrate that in the absence of IPS-1, WNV infection induces a strong inflammatory response in the CNS. While this response is likely associated with increased viral loads, the failure of this increased inflammatory response to elicit protection or control CNS pathology, in the absence of IPS-1, suggests a role for the RLR signaling pathway as a regulatory program that controls the quality of the inflammatory response to WNV infection.
To further characterize how IPS-1 modulates the inflammatory response to WNV infection, we measured levels of systemic type I IFN, proinflammatory cytokines, and chemokines present in the serum of WNV-infected mice at 1 and 4 days pi. Paradoxically, a trend towards more rapid induction and increased levels of type I IFN were observed in the serum of IPS-1−/− mice compared to wild type mice (Fig 5A). We note that in this case type I IFN was detected and quantified through a mouse-specific type I IFN bioassay, which does not differentiate between the IFN-α or -β species. This result is consistent with our recent studies showing that serum type I IFN levels accumulate during WNV infection in an IRF-7-dependent but IRF-3-independent manner [8],[9]. In this case IFN-α species are likely accumulating through a TLR7-dependent signaling pathway involving plasmacytoid DCs, which do not require the RLR pathway for IFN production [38]. More recently, Town et al. assessed the role of TLR7 and MyD88−/− during WNV infection and found that mice lacking MyD88 produced elevated levels of systemic IFN during WNV infection [25]. Thus, during WNV infection systemic IFN is regulated through RLR-dependent and independent processes. Correspondingly, when compared to wild type mice, IPS-1−/− infected animals, which show higher viremia (Fig 1B) produced increased serum levels of proinflammatory cytokines and chemokines in response to WNV infection. Elevated levels of systemic IL-6, TNF-α, CXCL10, and IFN-γ were observed at 1 and/or 4 days pi in IPS-1−/− mice (Fig 5B). Serum cytokine levels were also compared between uninfected wild type and IPS-1−/− mice and showed no differences in basal cytokine expression (data not shown). These results demonstrate that in the absence of IPS-1, greater proinflammatory cytokine and chemokine responses are induced during WNV infection.
WNV-specific antibody responses are essential for suppressing viremia and virus dissemination and limiting lethal WNV infection [39],[40]. To determine if a deficiency in IPS-1 modulated the quality and quantity of the humoral immune response, we characterized the antibody profile in sera during WNV infection. In wild type mice, neutralizing virus-specific IgM antibodies are typically detectable by day 4 pi with WNV and production of neutralizing virus-specific IgG antibodies follow between days 6 and 8 pi [40]. A time course analysis in wild type and IPS-1−/− infected mice showed that between 4 and 6 days pi, WNV-infected IPS-1−/− mice exhibited significantly higher levels of virus-specific IgM, IgG, and IgG subclasses as compared to infected wild type mice (Table 1). WNV-specific IgG1 antibodies were detected at low levels on day 6 pi in sera from wild type and IPS-1−/− mice. Additionally, we observed a ∼72.9-fold increase in WNV-specific IgG2a levels in infected IPS-1−/− as compared to wild type mice on day 6 pi and ∼2.2-fold increase on day 8 pi. Assessment of the virus-specific antibody responses through a PRNT assay revealed that neutralization titers in sera from wild type mice increased dramatically between 6 and 8 days pi. Sera from IPS-1−/− infected mice exhibited a modest increase in neutralization titer to 1∶1280, despite having much higher levels of virus-specific antibodies. This difference translated into a serum neutralization index that was ∼39-fold lower on day 6 pi in the infected IPS-1−/− mice compared to wild type mice. These results demonstrate that the humoral responses in WNV-infected IPS-1−/− mice are distinct from responses in wild type infected mice. Thus, RLR signaling and IPS-1 actions likely contribute to regulatory processes that govern the levels, IgG class switching, and neutralizing capacity of antibodies generated in response to WNV infection.
To characterize the immune parameters associating with the dysregulated inflammatory and humoral responses in WNV infected IPS-1−/− mice, we analyzed the immune cell composition in draining lymph node and spleen tissues. Wild type and IPS-1−/− mice were challenged with diluent alone or with WN-TX, and draining popliteal lymph node (DLN) and the spleen were harvested at 1 and 6 days pi, respectively. Analysis of the popliteal DLN provides insight into how IPS-1 modulates the inflammatory response immediately after infection whereas assessment of the spleen elucidates characteristics of the adaptive immune response prior to the infection endpoint. Immune cells were isolated from the popliteal DLN and were characterized by flow cytometry (Fig 6). Analysis of CD8α DC subsets, which are phenotypically the major antigen presenting cells within lymphoid tissues and are implicated in eliciting virus-specific CD8 T cell in response to acute WNV infection [41], showed that infected wild type and IPS-1−/− mice exhibited similar increases in the numbers of CD8α+ and CD8α− DCs compared to mock-infected mice (Fig 6A, B). However, a significant increase (∼3-fold; p<0.05) of a proinflammatory DC subset, characterized as CD11c+CD11bhiLy6C+, was observed within the popliteal DLNs of IPS-1−/− infected mice (Fig 6C). This DC subset is monocyte-derived and typically recruited to sites of acute inflammation where they propagate the inflammatory response [42]. We found that these DC subsets were not significantly expanded and showed no differences in their recruitment to the DLN in either wild type or IPS-1−/− infected mice at 12 hours pi (data now shown). Thus, as early as 24 hours pi, an elevated cellular inflammatory response is initiated in the IPS-1−/− mice. In contrast, similar increases in plasmacytoid DCs were observed within infected wild type and IPS-1−/− infected mice (Fig 6D), demonstrating that an absence of IPS-1 did not directly affect expansion and/or recruitment of this DC subset. Within the popliteal DLNs, mock-infected IPS-1−/− mice compared to wild type mice generally showed elevated numbers of B cells, CD4+ T cells (p<0.05), and CD8+ T cells (Fig 6E, F, and G). These results suggest that IPS-1 contributes to the homeostasis of lymphocyte populations within LNs. WNV infection of wild type mice increased the number of B cells (3.4 fold), CD4+ T cells (3.1 fold), and CD8+ T cells (2.3 fold; p<0.05) in the DLN within 24 hours pi. Similar increases in B cells were observed upon infection of IPS-1−/− mice. However, the number of CD4+ and CD8+ T cells was reduced in the DLN after WNV infection of IPS-1−/− mice. Thus, in the absence of IPS-1, WNV infection specifically increases the number of inflammatory Ly6c+ DCs but suppresses the overall expansion and/or recruitment of T cells in the DLN.
We further analyzed the lymphocyte composition of the spleen on day 6 after WNV infection (Fig 7). Gross pathologic analysis revealed that WNV infection of IPS-1−/− mice results in massive splenomegaly whereas infection of wild type mice induces only a slight increase in spleen size (Fig 7A). Cell analysis revealed increased numbers of total lymphocytes in the spleens of infected IPS-1−/− mice as compared to wild type mice (Fig 7B).
Regulatory T (Treg) cells have recently been shown to contribute to the dampening of inflammation and adaptive immune responses during acute virus infections [26],[43],[44]. Moreover, a reduction in the number of circulating Treg in mice leads to enhanced lethality after WNV infection [45]. A time course analysis of Tregs in wild type mice revealed that WNV infection results in a significant increase in the numbers of FoxP3+ T cells as compared to mock-infected mice beginning on day 4 and peaking by day 6 pi (Fig 7C), indicating the expansion of Tregs during acute WNV infection. Despite their marked increase in viral load, the infected IPS-1−/− mice did not display an increase in the numbers of FoxP3+ T cells at any timepoint analyzed. Thus, IPS-1 signaling directly or indirectly impacts Treg proliferation and does so independently of viral load.
We also observed that spleens from infected IPS-1−/− mice exhibited significantly increased numbers of CD8+ T cells in comparison to those from infected wild type mice, whereas the expansion of splenic CD4+ T cells in wild type and IPS-1−/− mice were not different (Fig 7D). The spleens from WNV-infected IPS-1−/− mice showed significantly higher numbers (3.9-fold; p<0.05) of WNV-specific CD8+ T cells producing IFNγ.
To further define the phenotype associated with WNV-induced splenomegaly in IPS-1−/− mice, we also assessed the numbers of NK cells and neutrophils. Spleens from infected IPS-1−/− mice contained greater numbers of NK cells (CD4−CD8−NK1.1+cells), NKT cells (CD4+/CD8+/NK1.1+ cells) and neutrophils (CD11b+Gr1+ cells) (Fig 7E). Although WNV-infected wild type mice infected displayed slight increases in the absolute numbers of these specific cell types, a deficiency of IPS-1 resulted in a more marked enhancement of these immune cell populations.
In this study, we establish a major role for RLR signaling in protection from WNV pathogenesis, and demonstrate that IPS-1 is critical for the control of WNV infection in vivo. Our studies indicate that IPS-1-dependent RLR signaling functions to establish balanced, effective, and protective innate and adaptive immune responses, and that IPS-1 links adaptive immune regulation with the innate immunity triggered by RLR signaling during WNV infection. In the absence of IPS-1 in vitro, innate immune defense programs of myeloid DCs and macrophages were ablated or severely attenuated. Moreover, in vivo analysis of infected IPS-1−/− mice showed altered IgG and IgM antibody responses with diminished virus neutralization activity. The inflammatory response to WNV infection is regulated by IPS-1-dependent processes such that a deficiency of IPS-1 resulted in elevated proinflammatory cytokine and chemokines and increased numbers of inflammatory DCs, antigen-specific T cells, natural killer cells, and neutrophils in lymphoid organs, and activated macrophage/microglial cells within the CNS. The dysregulated inflammatory response to WNV infection in IPS-1−/− mice was associated with a reduction in the numbers of Treg cells and their failure to expand during acute infection. These observations demonstrate the critical role of IPS-1 in mediating RLR signaling of innate antiviral immunity against WNV infection, and reveal novel features of IPS-1 function in regulating immune homeostasis, inflammation, and adaptive immunity to infection.
Although infection of primary DCs, macrophages, and neuronal cells failed to induce type I IFN upon WNV infection, WNV-infected IPS-1−/− mice showed enhanced systemic type I IFN responses. This finding agrees with previous studies that indicate both IPS-1-dependent and -independent pathways contribute to the systemic type I IFN production in vivo [8],[9],[23],[25]. Most importantly, the enhanced tissue tropism and rapid viral entry into the CNS observed in the IPS-1−/− mice is not affected by the elevated systemic IFN responses. This suggests that local tissue-specific and intracellular responses triggered by RLR-dependent signaling are more essential for reducing viral burden and dissemination. One possible explanation is that a distinct set of RLR-responsive genes function to control virus replication at the site of infection. This could explain, in part, the elevated levels of virus replication, enhanced tissue tropism and cell-to-cell spread in mice or cells deficient in IRF3 or IRF-7, each of which are downstream transcription factors of RLR signaling [8],[9],[10]. Additionally, it is likely that pDCs, which are specialized dendritic cells for producing systemic type I IFN during a viral infection [46], are likely contributing to the IFN responses observed during WNV infection. Studies by Silva et al. have shown that WNV triggers IFN induction in pDCs through a replication-independent manner [47]. Interestingly, within the DLN, we observed similar expansion of pDCs between wild type and IPS-1−/− infected mice, yet at the same timepoint (24 hours pi), elevated systemic type I IFN responses were observed in IPS-1−/− mice. This suggests two possibilities: 1) splenic pDCs or circulating pDCs are likely responding to the high levels virus in the serum from the IPS-1−/− infected mice to produce IFN at 24 hours pi and/or 2) various other cell types that express TLR3 and/or TLR7 are responding to WNV infection and contributing to systemic IFN responses. Taken together, these studies indicate that RLR signaling and the actions of IRF-3/7 are important in triggering IFN production and ISG expression to limit WNV replication and spread, and that TLR signaling may impart additional, RLR-independent defenses that regulate immunity against WNV infection.
The production of and response to type-I IFN is a major linkage point between innate and adaptive immunity, as IFN-α and IFN-β sustain B cell activation and differentiation [48],[49],[50], expand antigen-specific CD8+ T cells [51],[52], CD4+ T cells [53], and activation of NK cells [54]. One of the most intriguing aspects of this study was the global alteration of the immune response elicited in the IPS-1−/− mice, indicating that RLR signaling couples innate immunity with regulation of the adaptive immune response. Infection of IPS-1−/− mice exhibited increased IgM and IgG WNV-specific antibodies, enhanced WNV-specific CD8+T cell response, and increased expansion of neutrophils, NK cells and NK-T cells. One trivial explanation for these differences is that there is an increased antigen load in the absence of IPS-1 and, as a result, enhanced virus-specific (e.g. CD8+ T cells, IgG and IgM antibodies) and nonspecific (e.g. Neutrophils, NK cells) responses. However, there are several key findings from this study that argue against these responses simply being attributed to higher antigen load: (1) In the absence of IPS-1, the CD4 and CD8 T cells, which are protective against WNV infection [34],[35],[36],[55],[56],[57],[58], were significantly enhanced in the peripheral and CNS compartments but failed protect against infection. One explanation for this observation is that the expansion and migration of CD4 and CD8 T cells to different tissues was itself uncontrolled, resulting in T cell-mediated pathology rather than T cell-mediated protection. (2) While the quantity of virus-specific IgM and IgG antibody responses were greatly enhanced in the absence of IPS-1, there was a marked reduction in antibody quality in terms of neutralization capacity. In contrast deficiencies in TLR3 or MyD88 (data not shown) did not alter virus-specific antibody responses and neutralization capacities. Collectively, these findings suggest that RLR-dependent signaling coordinates effective antibody responses during WNV infection through as yet undefined pathway. (3) While systemic IFN responses provide a link between innate and adaptive immune responses, our studies suggest that the PRR signaling pathways (RLR-dependent vs –independent) and levels of IFN production in combination with production other proinflammatory cytokines or chemokines regulate the quantity and quality of the immune response during virus infection. Thus, in the absence of IPS-1 signaling, infected conventional DCs or Mφ, two integral cell types in establishing adaptive immunity, likely do not produce IFN or the normal array and level of proinflammatory cytokines/ chemokines. Instead, IFN and other mediators may be strictly produced by infected or bystander cells during WNV infection, occurring with altered kinetics and magnitude, through TLR-dependent pathways, such as TLR3 and/or TLR7 [23],[25]. (4) In the absence of IPS-1, the enhanced expansion of Ly6C+ “inflammatory” DCs failed to limit early WNV replication and dissemination. This inflammatory DC subset migrates to sites of infection, secretes pro-inflammatory cytokines, and promotes CD8+ T cell expansion during a secondary virus infection, suggesting it sustains the effector T cell response [59]. Our data indicate that Ly6C+ DC recruitment and/or expansion is governed by IPS-1-dependent events of RLR signaling. Thus, the aberrant recruitment/expansion of these inflammatory DCs may contribute to immunopathogenesis and limit development of an effective immune response to control WNV virus infection. (5) The lack of Treg expansion during WNV infection correlated with altered IFN levels, increased proinflammatory cytokines and chemokine levels, and an increased number and distribution of antigen-specific CD8+ T cells. These observations implicate an indirect or direct role for IPS-1 in regulating Treg levels during WNV infection, and provide evidence that links a lack of Treg expansion to immune dysregulation.
While their importance in autoimmunity is established [60], recent studies have implicated an integral role for Tregs in controlling inflammation and adaptive immune responses during acute virus infections [26],[43],[44]. During acute infection Tregs function to locally contain and control the immune response with the dual outcome of suppressing viral dissemination while decreasing the likelihood of immune-mediated pathology. In support of this model, infection studies with herpes simplex virus (HSV) and mouse hepatitis virus (MHV), two well established models of viral encephalitis, have demonstrated the importance of Tregs in limiting proinflammatory cytokine and chemokine responses to protect the CNS and enhance survival [26],[43]. Recent work also implicates Tregs in the control of WNV pathogenesis, wherein peripheral expansion of Tregs was associated with asymptomatic infection among WNV-infected blood donors but reduced Treg levels associated with WNV disease [45]. Furthermore, these studies revealed that the conditional depletion of Treg cells in mice results in enhancement of WNV virulence and expansion of antigen-specific CD8 T cells. Interestingly, from our studies, type I IFN does not appear to be the major contributor to Treg expansion during WNV infection, as Tregs failed to expand in the IPS-1−/− infected mice despite their elevated levels of systemic type IFN. These observations suggest that RLR signaling through IPS-1 provides essential signals that directly or indirectly impart the expansion of Tregs during WNV infection.
We propose that IPS-1 coordinates an innate/adaptive immune interface wherein IPS-1- signaling after RLR engagement regulates the quantity, quality, and balance of the subsequent immune response. The integrity of the innate/adaptive immune interface is central to the eliminating virus but also restricting immunopathogenesis and inflammation during infection. RLR signaling is essential for triggering the innate immune response to RNA viruses that cause human disease, including the influenza viruses, respiratory syncytial virus and other paramyxoviruses, picornaviruses, reoviruses, flaviviruses, and hepatitis C virus [14],[19],[22]. Thus, in addition to WNV, IPS-1-dependent RLR signaling will likely have a broad impact for the control of inflammation, immune response quality, and viral disease.
BHK21 and L929 cells were maintained in Dulbecco's modified Eagle medium (DMEM) supplemented with 10% fetal bovine serum (FBS), 2mM L-glutamine, 1 mM sodium pyruvate, antibiotic-antimycotic solution, and 1× nonessential amino acids (complete DMEM). WNV strain TX 2002-HC (WN-TX) was isolated by as previously described [11]. Working stocks of WN-TX were generated by a single round of amplification on Vero-E6 (ccl-81; ATCC) cells, and supernatants were collected, aliquoted, and stored at −80°C. Virus stocks were titered by a standard plaque assay on BHK21 cells as previously described [40].
IPS-1−/− (C57BL/6×129Sv/Ev) and their wild type littermate control mice have been published [38],[61] and were obtained as a generous gift from Dr. S. Akira (Osaka University, Osaka, Japan). Mice were genotyped and bred under pathogen-free conditions in the animal facility at the University of Washington. Experiments were performed with approval from the University of Washington Institutional Animal Care and Use Committee. The methods for mice use and care were performed in accordance with the University of Washington Institutional Animal Care and Use Committee guidelines. Age-matched six to twelve week old mice were inoculated subcutaneously (s.c.) in the left rear footpad with 100 PFU of WN-TX in a 10 µl inoculum diluted in Hanks balanced salt solution (HBSS) supplemented with 1% heat-inactivated FBS. Mice were monitored daily for morbidity and mortality.
For in vivo virus replication studies, infected mice were euthanized, bled, and perfused with 20 ml of phosphate-buffered saline (PBS). Whole brain, spinal cord, kidney, and spleen were removed, weighed, homogenized in 500ul of PBS, and titered by plaque assay.
Bone-marrow derived DC and Mφ were generated as described previously [9]. Briefly, bone marrow cells from wild type and congenic deficient mice were isolated and cultured for 7 days in either RPMI-1640 supplemented with granulocyte-macrophage-colony stimulating factor, and interleukin-4 (Peprotech) to generate myeloid DC or in DMEM supplemented with macrophage colony stimulating factor (Peprotech) to generate Mφ. On day 7, DC or Mφ were infected with WN-TX at an MOI of 1.0 and at 12, 24, 36, and 48 hours post-infection (hpi), supernatants were collected for titration of viral burden by plaque assay on BHK21 cells and levels of IFN-β (described below). Cells were collected in parallel for western blot analysis. Cortical neurons were isolated from 15-day-old embryonic mice and cultured as described previously [62]. On day 6 of culture, neurons were infected with WN-TX at an MOI of 1.0 and at 12, 24, 36, and 48 hpi, supernatants were collected for virus titration by plaque assay on BHK21 cells and cells were collected for RNA analysis by RT-qPCR (described below).
Cells were lysed in modified RIPA buffer (10mM Tris [pH 7.5], 150mM NaCl, 0.5% sodium deoxycholate, and 1% Triton X-100) supplemented with protease inhibitor cocktail (Sigma) and phosphatase inhibitor cocktail II (Calbiochem). Protein extracts (25 µg) were analyzed by immunoblotting as described previously [11]. The following primary antibodies were used to probe blots: mouse anti-WNV from the Center for Disease Control; rabbit anti-ISG56, rabbit anti-ISG54, rabbit anti-ISG49, kindly provided by Dr. G. Sen; mouse anti-PKR from Santa Cruz; rabbit anti-RIG-I and rabbit anti-MDA5 from IBL; mouse anti-tubulin from Sigma; and rabbit anti-STAT-1 from Cell signaling. Secondary antibodies included peroxidase-conjugated goat anti-rabbit, goat anti-mouse, donkey anti-rabbit, and donkey anti-mouse were from Jackson Immunoresearch.
For analysis of viremia, serum was separated (BD Microtainer tube SST) and RNA was extracted as previously described [8]. WNV RNA copy number was measured by RT-quantitative PCR (RT-qPCR) as previously described [63]. For cultured cells, total RNA was extracted using the RNeasy kit (Qiagen), DNase treated (Ambion) and evaluated for ISG49, ISG56, IFN-β, RIG-I, and MDA5 RNA expression by one-step SYBR Green RT-qPCR. Specific primer sets for ISG-49, ISG-56, RIG-I, and IFN-β have been described previously [30],[64]. Primer sets for MDA5 are: 5′-GTGGTCGAGCCAGAGCTGAT and 3′- TGTCTCATGTTCGATAACTCCTGAA.
IFN-α and -β were measured in sera using a biological assay as previously described [65]. Briefly, L929 cells were seeded at 3×104 cells/well in a 96 well plate one day prior to the addition of interferon standards or experimental samples. Mouse sera (diluted 1∶10 in L929 media) were treated with UV light for 20 minutes to eliminate residual virus. Duplicate sera samples were then added to the 96-well plates in two-fold dilutions along with a murine IFN-β standard. The following day, EMCV challenge virus was added to the cells in 50 µl/well at an MOI of 5.0. Twenty-four hours later, cytopathic effect was measured by a blinded scorer and IFN levels in the sera was calculated based on the IFN standard. IFN-β in cell culture supernatants was analyzed using mouse-specific ELISA kits from PBL Biomedical Laboratories according to the manufacturer's protocol.
WNV-specific IgM, total IgG, IgG1, and IgG2a levels were determined by an ELISA using purified recombinant E protein as previously described [55]. The neutralization titer of serum antibody was determined by using a previously described plaque reduction neutralization assay [40]. Briefly, sera samples from mock or WN-TX infected mice were diluted in DMEM followed by incubation at 56°C for 30 minutes to inactivate virus and complement factors. Sera were further diluted in two-fold increments and incubated with 100 PFU of WN-TX at 37°C for 1 hour. Standard plaque assays were performed on BHK21 cells and the dilution at which 50% of plaques were neutralized was determined by comparing the number of plaques formed from WNV-infected sera samples to mock infected sera samples.
WNV infected sera were analyzed for the presence and levels of TNF-α, IFN-γ, CXCL10 (IP-10), and IL-6 by a mouse-specific cytokine/chemokine Milliplex ELISA (Millipore).
Mock-infected or WNV-infected mice were exsanguinated and perfused with PBS, 4% paraformaldehyde, pH 7.3. Brains were embedded in paraffin and 10-µm sections were prepared and stained with hematoxylin and eosin (H&E) by the UW histology pathology laboratory. Sections were analyzed using a Nikon Eclipse E600 microscope (UW Keck microscope facility).
Draining lymph nodes from mice were isolated and digested with collagenase (Roche) and type I DNase in serum-free RPMI media at 37°C for 40 minutes with mechanical disruption. Cells were then incubated with RPMI media containing 10% FBS with EDTA and HEPES for 10 minutes at room temperature, pelleted, and resuspended in PBS containing 2% FBS and 0.1% sodium azide (FACS Staining buffer). Splenocytes were isolated, washed, and re-suspended in RPMI 1640 containing 10% FBS before in vitro stimulation. Cells were washed twice before FACS staining. For isolation of CNS immune cells, mice were euthanized and perfused extensively with PBS to remove residual intravascular leukocytes. Brains and spinal cords from 5 mice per experimental group were isolated and pooled. Tissues were minced in RPMI media, triturated, and digested with Liberase (Roche) and type I DNase in serum-free RPMI media at 37°C for 45 min. Immune cells were isolated after gradient centrifugation from a 37/70% Percoll interface and washed twice with FACS staining buffer. Immune cells were stained with antibodies specific to CD11c, CD11b, B220, CD3, CD25, CD4, CD8, NK1.1, Gr-1, siglec H, and CD45 (all reagents from eBiosciences). Intracellular FoxP3 staining was performed as previously described [26]. Intracellular IFN-γ staining was performed on splenocytes and CNS immune cells as previous described [35],[36]. Briefly, lymphocytes were stimulated with 1 µg/ml of the WNV NS4B peptide (SSVWNATTAI) for 4 h at 37°C. Cells were washed and stained for cell surface markers followed by permeabilization-fixation using the Cytofix-Cytoperm Kit (BD-PharMingen) and stained with a Pacific-Blue conjugated IFN-γ antibody (eBiosciences) at 4°C for 30 min, washed and analyzed by flow cytometry. Flow cytometry was performed on a BD LSRII machine using BD FACSDiva software. Cell analysis was performed on FlowJo (v.8.7.2) software.
For in vitro studies and immune cell analysis an unpaired student T-test was used to determine statistical differences. For in vivo viral burden analysis, Mann-Whitney analysis was used to determine statistical differences. Kaplan-Meier survival curves were analyzed by the log-rank test. A p-value≤0.05 was considered significant. All data were analyzed using Prism software (GraphPad Prism5).
|
10.1371/journal.ppat.1000628 | The Leishmania donovani Lipophosphoglycan Excludes the Vesicular Proton-ATPase from Phagosomes by Impairing the Recruitment of Synaptotagmin V | We recently showed that the exocytosis regulator Synaptotagmin (Syt) V is recruited to the nascent phagosome and remains associated throughout the maturation process. In this study, we investigated the possibility that Syt V plays a role in regulating interactions between the phagosome and the endocytic organelles. Silencing of Syt V by RNA interference revealed that Syt V contributes to phagolysosome biogenesis by regulating the acquisition of cathepsin D and the vesicular proton-ATPase. In contrast, recruitment of cathepsin B, the early endosomal marker EEA1 and the lysosomal marker LAMP1 to phagosomes was normal in the absence of Syt V. As Leishmania donovani promastigotes inhibit phagosome maturation, we investigated their potential impact on the phagosomal association of Syt V. This inhibition of phagolysosome biogenesis is mediated by the virulence glycolipid lipophosphoglycan, a polymer of the repeating Galβ1,4Manα1-PO4 units attached to the promastigote surface via an unusual glycosylphosphatidylinositol anchor. Our results showed that insertion of lipophosphoglycan into ganglioside GM1-containing microdomains excluded or caused dissociation of Syt V from phagosome membranes. As a consequence, L. donovani promatigotes established infection in a phagosome from which the vesicular proton-ATPase was excluded and which failed to acidify. Collectively, these results reveal a novel function for Syt V in phagolysosome biogenesis and provide novel insight into the mechanism of vesicular proton-ATPase recruitment to maturing phagosomes. We also provide novel findings into the mechanism of Leishmania pathogenesis, whereby targeting of Syt V is part of the strategy used by L. donovani promastigotes to prevent phagosome acidification.
| Upon their internalization by macrophages, Leishmania donovani promastigotes inhibit phagolysosome biogenesis. This inhibition is mediated by the virulence glycolipid lipophosphoglycan (LPG), attached to the promastigote surface. We recently showed that the exocytosis regulator Synaptotagmin (Syt) V controls early steps of phagocytosis, and remains associated to the phagosome during the maturation process. Here, we show that Syt V contributes to phagolysosome biogenesis by regulating the acquisition of the hydrolase cathepsin D and the vesicular proton-ATPase. Insertion of LPG into lipid microdomains of the phagosome membrane excluded Syt V from phagosomes, enabling L. donovani promatigotes to inhibit the recruitment of the vesicular proton-ATPase to phagosomes, preventing their acidification. Collectively, our results provide novel insight into the mechanism of vesicular proton-ATPase recruitment to maturing phagosomes and reveal how the virulence glycolipid LPG contributes to the mechanism of L. donovani pathogenesis by preventing phagosome acidification.
| Phagocytosis consists in the uptake and destruction of invading microorganisms, thereby playing an essential role in host defense against infection [1]. Following internalization, microbes end up in a vacuole, the phagosome, which engages in a maturation process involving highly regulated fusion and fission events with early and late endosomes, and with lysosomes [2],[3]. This leads to the acidification of the phagosome and the acquisition of an array of hydrolases, culminating in the generation of a highly microbicidal environment [4]. Soluble N-ethylmaleimide-sensitive factor protein attachment protein receptor (SNARE)-mediated membrane fusion events regulate phagosome maturation by facilitating interactions with the endocytic compartments [5]. Hence, VAMP3 and syntaxin 13 are present transiently on the young phagosome to regulate early maturation steps, whereas VAMP7 and syntaxin 7 remain associated with the phagosome to regulate interactions with late endosomes/lysosomes [6]–[8]. The lysosome-associated Synaptotagmin (Syt) VII, which controls membrane delivery to nascent phagosomes [9], is also involved in phagolysosome fusion [9],[10]. Other components and partners of these SNARE fusion machineries required during phagosome maturation remain to be identified.
Phagolysosome biogenesis is an important means of controling microbial growth. Yet, several pathogenic microorganisms have evolved mechanisms to subvert the phagosome maturation process, thus avoiding an encounter with the macrophage microbicidal machinery including exposition to reactive oxygen species and to acidification [4],[11],[12]. Protozoan parasites of the genus Leishmania cause a spectrum of diseases in humans, ranging from self-healing ulcers to potentially fatal visceral leishmaniasis, which affect millions of people worldwide. Leishmania is transmitted to mammals under its promastigote form during the bloodmeal of infected sand flies. Following phagocytosis by macrophages, promastigotes must avoid destruction to differentiate into amastigotes, the mammalian stage of the parasite that replicate inside acidic and hydrolase-rich parasitophorous vacuoles [13]–[15]. To avoid the microbicidal arsenal of macrophages, L. donovani and L. major promastigotes create an intracellular niche through the inhibition of phagolysosome biogenesis [16]–[19]. Genetic and biochemical approaches established that this inhibition is strictly dependent on the presence of lipophosphoglycan (LPG), an abundant surface glycolipid consisting of a polymer of Galβ1,4Manα1-PO4 units anchored into the promastigote membrane via an unusual glycosyl phosphatidylinositol [20],[21]. Hence, phagosomes harboring LPG-defective promastigotes quickly mature into functional phagolysosomes and coating of the Galβ1,4Manα1-PO4-defective mutant lpg2-KO with purified LPG conferred the capacity to inhibit phagosome-lysosome fusion [16],[17],[19],[22],[23]. LPG-mediated phagosome remodeling is characterized by a periphagosomal accumulation of F-actin [22],[23] and by the exclusion of cytosolic components of the NADPH oxidase from the phagosome membrane [24]. By creating an environment devoid of oxidants, L. donovani promastigotes evade a major microbicidal mechanism of macrophages and can initiate their differentiation into amastigotes. The ability of LPG to inhibit phagosome maturation is consistent with its role in the establishment of L. donovani and L. major promastigotes inside macrophages [24],[25].
A possible mechanism by which LPG exerts its action on phagosome maturation involves the transfer of LPG from the parasite surface to lipid microdomains present in the phagosome membrane, causing a disorganization of these structures and preventing their formation after phagocytosis [26]–[29]. Phagosomal lipid microdomains are essential for the recruitment/assembly of the NADPH oxidase and the vacuolar proton-ATPase and are involved in the regulation of phagosome-endosome fusions [27],[30],[31]. Disruption of lipid microdomains by microbial virulence factors is likely to facilitate the establishment of infection through an effect on phagolysosomal biogenesis, as described for the cyclic β-1,2-glucans of Brucella abortus and the lipoarabinomannan of Mycobacterium tuberculosis [32],[33]. How lipid microdomains regulate interactions between phagosomes and the endocytic system is unclear. The fact that proteins involved in membrane fusion such as SNAREs and Syts are located in lipid microdomains is consistent with these structures acting as fusion sites [34],[35].
Recently, we identified the exocytosis regulator Syt V [36]–[39] as a recycling endosome-associated protein that is recruited to the forming phagosome independently of the phagocytic receptor engaged [40]. Silencing of Syt V by RNAi revealed a role for this protein during phagocytosis, particularly under conditions of high membrane demand, possibly through the mobilization of recycling endosomes as a source of endomembrane. The association of Syt V with the phagosome throughout the maturation process raised the possibility that Syt V regulates interactions with the endocytic system [40]. Here, we provide evidence for a novel function of Syt V in phagolysosome biogenesis, where it controls the acquisition of cathepsin D and the vesicular proton-ATPase. We also provide novel insight into the mechanism of L. donovani pathogenesis with the demonstration that insertion of LPG into GM1-containing microdomains impairs the association of Syt V to phagosome membranes, enabling L. donovani promatigotes to inhibit the recruitment of the vesicular proton-ATPase to phagosomes, thereby preventing their acidification.
Syt V, a regulator of exocytosis, is recruited to the nascent phagosome and remains associated throughout the maturation process [40], suggesting that it may participate in the regulation of phagolysosome biogenesis. Maturing phagosomes sequentially interact with various endocytic organelles to acquire hydrolases such as cathepsins and the proton-vacuolar ATPase (V-ATPase), which is responsible for phagosome acidification [2],[41],[42]. To assess the potential role of Syt V in the acquisition of microbicidal features, we inhibited its expression by transfecting RAW 264.7 cells with a siRNA to Syt V [40] (Figure 1A) and we examined the localization of phagosomal markers following the internalization of Zymosan (Zym) or latex beads. Our results show that in the absence of Syt V, recruitment of both the early endosomal (EEA1) and the lysosomal (LAMP1) markers to Zym-containing phagosomes was normal (Figures 1B and S1A and B), whereas the acquisition of cathepsin D and the V-ATPase c subunit was inhibited (Figure 1B–E). Reduction in cathepsin D acquisition ranged from 25 to 35% for phagosomes containing beads and from 41 to 48% for phagosomes containing Zym, in five independent experiments. In the case of the V-ATPase c subunit, the reduction ranged from 30 to 50% for phagosomes containing beads and from 45 to 60% for phagosomes containing Zym in five independent experiments (Figure 1C–E). Interestingly, silencing of Syt V had no detectable effect on the acquisition of cathepsin B (Figure 1C). These results provide evidence that Syt V selectively regulates the phagosomal acquisition of cathepsin D and the V-ATPase c subunit.
Given their ability to inhibit phagosome maturation [16],[17],[24], we explored the impact of L. donovani promastigotes and their LPG on the phagosomal association of Syt V. Accordingly, we infected the mouse macrophage cell line RAW 264.7 stably expressing a Syt V-GFP fusion protein (Syt V-GFP RAW 264.7 cells) with either wild-type (WT) L. donovani promastigotes, the LPG-defective lpg1-KO mutant, the Galβ1,4Manα1-PO4-defective lpg2-KO mutant or the lpg2-KO add-back (lpg2-KO+LPG2). We used Zym as a positive control for the recruitment of Syt V to phagosomes [40]. Our results show that Syt V-GFP was present on over 80% of phagosomes containing either lpg1-KO promastigotes, lpg2-KO promastigotes, or Zym (Figure 2A and B). In contrast, we detected Syt V-GFP on 54 to 65% of phagosomes containing either WT or lpg2-KO+LPG2 promastigotes in three independent experiments. Quantification analyses revealed a three-fold reduction in the levels of Syt V-GFP present on those positive phagosomes with respect to phagosomes containing either lpg1-KO or lpg2-KO promastigotes (Figure 2C). These observations suggested that LPG impairs the phagosomal recruitment of Syt V.
To directly address the impact of LPG on the recruitment of Syt V to phagosomes, we fed bone marrow-derived macrophages (BMM) with either Zym or Zym coated with purified LPG (LPG-Zym) [22]. Consistent with previous observations [17],[22], we found a reduced acquisition of LAMP-1 on phagosomes containing LPG-Zym, whereas the recruitment of EEA1 to phagosomes containing Zym or LPG-Zym was similar (Figure 3A and B). In the case of Syt V, we detected its presence on 24 to 30% of phagosomes containing LPG-Zym compared to over 60% of phagosomes containing Zym at all time points tested in three independent experiments (Figure 3C and D). Quantification analyses showed that the levels of Syt V present on those positive phagosomes containing LPG-Zym was significantly lower than the Syt V levels on phagosomes containing Zym (Figure 3C and D). We obtained similar results with the Syt V-GFP RAW 264.7 cells (Figures 3E and S2). Furthermore, the signals for Syt V (green) and LPG (red) rarely superimposed on the phagosome membrane (Figure 4A), and fluorescence intensity line scans acquired along the periphery of phagosomes showed that the most intense LPG and Syt V signals never overlapped, at both 30 min and 120 min after the initiation of phagocytosis (Figure 4B). We made similar observations in Syt V-GFP RAW 264.7 cells (Figure 4C and D). Collectively, these results established that insertion of LPG into the phagosomal membrane caused the exclusion of Syt V in a very localized manner.
In rat brain synaptosomes, a fraction of Syt I and Syt II is present in lipid rafts [34]. To examine whether LPG-mediated exclusion of Syt V from phagosomes was related to the insertion of LPG into lipid microdomains [27],[29] (Figure 5D), we first determined whether phagosome-associated Syt V was present in these microdomains. Our results clearly show that a fraction of Syt V colocalizes with GM1-microdomains on Zym-containing phagosomes (Figure 5A, arrowheads). Consistently, cholesterol depletion by methyl-β-cyclodextrin inhibited the recruitment of Syt V (Figure 5B and C). Having established that phagosomal Syt V associates with GM1-containing microdomains, we examined the localization of LPG, Syt V, and GM1 on phagosomes containing either Zym or LPG-Zym. For phagosomes containing Zym, the signals for Syt V (blue) and GM1 (red) superposed to a large extent and fluorescence intensity line scans acquired along the periphery of a representative phagosome showed that most of the Syt V and GM1 signals overlapped (Figure 5E and F, top panel). In contrast, on phagosomes containing LPG-Zym, the signals for LPG and GM1 colocalized, whereas most of the remaining Syt V signal was not associated with GM1 (representative phagosome, Figure 5E and F, bottom panel). These results established that association of LPG with GM1-containing microdomains resulted in the exclusion or dissociation of Syt V from the phagosome membrane.
The demonstration that Syt V regulates acquisition of the V-ATPase led us to verify the hypothesis that exclusion or dissociation of Syt V from phagosomes containing L. donovani promastigotes may impair the recruitment of the V-ATPase to these phagosomes. At 2 h after the initiation of phagocytosis, our results from three independent experiments showed a reduction in the recruitment of the V-ATPase c subunit on phagosomes containing WT promastigotes, ranging from 54 to 62% with respect to phagosomes containing either lpg1-KO or lpg2-KO promastigotes (Figure 6A and B). Co-localization of the V-ATPase c subunit with LAMP-1 on phagosomes containing lpg1-KO promastigotes showed that the V-ATPase c subunit was present on the phagosome membrane (Figure S3). As expected, phagosomes containing lpg2-KO+LPG2 cells were similar to WT-phagosomes with respect to the presence of the V-ATPase. We next monitored the acidification of L. donovani promastigote-containing phagosomes using the lysosomotropic agent LysoTracker red as an indicator of phagosome pH. Our results showed a clear correlation between the presence of the V-ATPase c subunit and the association of LysoTracker red to phagosomes (Figure 6C). In Figure 1, we showed that silencing of Syt V inhibited recruitment of the V-ATPase c subunit to phagosomes containing Zym or latex beads. In Figure 6D, we show that silencing of Syt V abrogated recruitment of the V-ATPase c subunit to phagosomes containing lpg1-KO and lpg2-KO mutants. In the case of phagosomes containing either WT or lpg2-KO+LPG2 promastigotes, Syt V silencing had the same effect as the presence of LPG on the recruitment of the V-ATPase c subunit (Figure 6D). Collectively, these results show that LPG enables L. donovani promatigotes to inhibit phagosomal recruitment of the V-ATPase by a Syt V-dependent mechanism and to prevent acidification. Remarkably, at 24 h after the initiation of phagocytosis, we detected the V-ATPase c subunit on only 10 to 17% of phagosomes containing L. donovani promastigotes in three independent experiments, consistent with LPG still being present (Figure 7A and C). At this time point, we detected LysoTracker red on only 20% of phagosomes containing WT promastigotes (not shown), indicating that promastigotes remodel their intracellular niche to establish infection in a compartment that fails to acidify, at a time when differentiation into amastigotes takes place. In contrast, we detected the V-ATPase c subunit on 66 to 71% of phagosomes containing L. donovani amastigotes at both 2 h and 24 h after the initiation of phagocytosis (Figure 7B and C). This observation is consistent with the fact that amastigotes replicate in an acidic phagolysosomal compartment [14].
The exocytosis regulator Syt V is recruited to the nascent phagosome and remains associated throughout the maturation process [40], leading us to investigate its potential role in modulating interactions between the phagosome and endocytic organelles. Our results revealed that whereas silencing of Syt V had no effect on the recruitment of EEA1, LAMP-1, and cathepsin B, it inhibited the phagosomal acquisition of cathepsin D and of the V-ATPase c subunit. These findings indicated that Syt V plays a role in phagolysosome biogenesis, possibly by regulating the interaction between phagosomes and a subset of late endosomes or lysosomes enriched in cathepsin D and in the V-ATPase c subunit. Alternatively, Syt V may be needed to reach the level of phagosome maturation necessary to acquire the machinery that regulates the recruitment of cathepsin D and the V-ATPase c subunit. Our finding that acquisition of cathepsin B and cathepsin D is mediated by distinct mechanisms supports the demonstration that various hydrolases appear sequentially, at various time points during phagosome maturation [42]. This view is also consistent with evidence that various sub-populations of early endosomes, late endosomes, and lysosomes co-exist and that these compartments contain significant heterogeneity [43]. Together with previous findings [27], our results show that phagosomal acquisition of the V-ATPase and LAMP-1 are mediated through distinct mechanisms. Hence, the observations that LAMP-1 is recruited to phagosomes independently of Syt V and that L. donovani promastigotes (and LPG) impair the recruitment of LAMP-1 point to the existence of other inhibitory mechanisms and illustrate the complexity of phagolysosome biogenesis. The role of Syt V in regulating interactions between the phagosome and the endosomal compartments thus seems specific and further studies will be necessary to understand its precise role during phagosome maturation. Recent studies by Andrews and colleagues revealed that the lysosome-associated Syt VII, which controls membrane delivery to nascent phagosomes [9], is involved in phagolysosome fusion [9],[10]. It will be of interest to determine whether Syt V and Syt VII use similar mechanisms to regulate phagolysosome biogenesis.
To establish infection inside macrophages, L. donovani promastigotes, the form of the parasite transmitted to mammals by the sand fly vector, create an intracellular niche by inhibiting phagolysosome biogenesis [16]. Genetic and biochemical approaches revealed that this inhibition is mediated by the parasite surface glycolipid LPG [16],[17],[22]. Insight into the mechanism of this inhibition came from the observations that LPG transfers from the parasite surface to the nascent phagosome membrane [26], where it disrupts existing lipid microdomains and alters the formation of these structures after promastigote internalization [28],[29]. Whereas the precise mechanism remains to be elucidated, the current model is that LPG inserts into lipid microdomains via its GPI anchor, thereby allowing the negatively charged Galβ1,4Man-PO4 polymer of LPG to directly interfere with the clusterization of molecules into these microdomains. This model is consistent with the demonstration that alteration of membrane properties is dependent on the length of the Galβ1,4Man-PO4 polymer [16],[44]. Because of their role in clustering specific sets of proteins, membrane lipid microdomains are central to a wide variety of cellular processes, including regulated exocytosis [45],[46]. Our findings that Syt V was present in GM1-enriched phagosome microdomains and that LPG inserts into or associates with these structures to interfere with the phagosomal association of Syt V thus provides new insight into the mechanism of LPG-mediated inhibition of phagolysosome biogenesis.
Acquisition of an array of hydrolases and acidification of the phagosome enable the generation of a highly microbicidal environment [4] and the creation of a compartment competent for antigen processing and presentation [47]. To circumvent killing following uptake by macrophages, several intracellular microorganisms interfere with phagosome acidification and maturation [4],[12],[48]. The discovery that L. donovani promastigotes establish infection inside a compartment from which the V-ATPase is excluded may thus be favorable for parasite survival. Incidentally, a recent study showed that phagosome acidification is defective in Stat1−/− macrophages and this correlated with an increased survival of L. major promastigotes, suggesting a role for acidic pH in the control of intracellular Leishmania growth early during infection [49]. Furthermore, the finding that phagosomes containing L. donovani promastigotes fail to acquire the V-ATPase and acidify even at 24 hours post-infection provides new insight on our undestanding of Leishmania biology. Indeed, in the absence of data on the pH of promastigote-containing phagosomes, it has been assumed that promastigotes initiate infection in an acidic environment and that differentiation of promastigotes into amastigotes is mainly triggered by a rapid exposure to an acidic environment and elevated temperature [50]. Exclusion of the V-ATPase raises the possibility that L. donovani promastigotes initiate the differentiation process in a non-acidified environment. Further studies will be required to fully address this point. An issue that remains unsolved pertains to the acquisition of phagolysosomal features and acidification of parasite-containing vacuoles upon completion of the differentiation of promastigotes into amastigotes. Indeed, previous work by Antoine and colleagues [14] established that L. amazonensis amastigotes reside within an acidic vacuole (pH 4.7–5.2), in agreement with the notion that Leishmania amastigotes are internalized within a vacuole that rapidly acquires lysosomal features and in which amastigotes proliferate [13],[51]. Consistent with these previous reports, we showed the presence of LAMP-1 and the V-ATPase c subunit on phagosomes containing L. donovani amastigotes as early as 2 h after internalization. A possible explanation is that during the first few days post-infection, the presence of LPG in the phagosome membrane prevents acidification and maturation, allowing promastigote-to-amastigote differentiation to take place. The down-regulation of LPG biosynthesis below detectable levels in amastigotes [52] may enable phagosomes to gradually acquire lysosomal features and to acidify.
Little is known on the mechanisms that regulate recruitment of the V-ATPase to maturing phagosomes. The identification of Syt V as a regulator of this process and the fact that Syt V is present in microdomains of the phagosome membrane is consistent with the notion that these structures are important for the recruitment of the V-ATPase to the phagosome membrane [27]. Of interest, the V-ATPase c subunit has been previously identified in Triton X-100-resistant fractions from rat brain synaptic vesicles in association with synaptobrevin 2 and synaptophysin [53], leading the authors of that study to conclude that this interaction may play a role in recruiting the V-ATPase to synaptic vesicles. Whether Syt V is part of such a SNARE complex on phagosomes and the characterization of this complex are important issues that await further investigation.
In this study, we provided novel findings into the mechanism of Leishmania pathogenesis, whereby targeting of Syt V, which plays a role in the acquisition of phagosome microbicidal properties, is part of the strategy used by L. donovani promastigotes to create a niche propitious to the establishment of infection within mammalian hosts (see working model, Figure 8). Interestingly, phagocytosis of either zymosan or lpg2-KO promastigotes coated with the virulence glycolipid lipoarabinomannan from Mycobacterium tuberculosis, impaired the phagosomal association of Syt V (Figure S4). Whether other intracellular microorganisms use a similar mechanism to remodel their intracellular niche remains to be investigated.
All animals were handled in strict accordance with good animal practice as defined by the Canadian Council on Animal Care, and all animal work was approved by the Comité institutionel de protection des animaux of INRS- Institut Armand-Frappier (protocol 0811-08). BMM were obtained by growing bone marrow cells from female BALB/c mice at 37°C in 5% CO2 for 7 days in Dulbecco Modified Eagle Medium with L-glutamine (Life Technologies) supplemented with 10% heat-inactivated FBS (Hyclone, Logan, UT), 10 mM Hepes (pH 7.4) and antibiotics (complete medium) in the presence of 15% (v/v) L929 cell-conditioned medium as a source of colony-stimulating factor (CSF)-1 [54]. BMM were made quiescent by culturing them in the absence of CSF-1 for 18 h prior to being used. The murine macrophage cell line RAW 264.7 was grown in complete medium in a 37°C incubator with 5% CO2. Stably transfected RAW264.7 cells expressing Syt V-GFP (Syt V-GFP RAW 264.7 cells) were previously described [40]. Transfectants were cultured in complete medium containing 500 µg/ml G418 (Life Technologies).
Leishmania donovani promastigotes (Sudanese strain 1S) were grown at 26°C in RPMI 1640 medium supplemented with 20% heat-inactivated FBS, 100 µM adenine, 20 mM 2-[N-morpholino]ethanesulphonic acid (pH 5.5), 5 µM hemin, 3 µM biopterin, 1 µM biotin and antibiotics. The isogenic L. donovani LPG-defective mutants lpg1-KO and lpg2-KO were described previously [55]. The lpg1-KO mutant secretes repeating Galβ1,4Manα1-PO4-containing molecules, but lacks the ability to assemble a functional LPG glycan core [56], precluding synthesis of LPG. The lpg2-KO mutant expresses the truncated LPG Gal(α1,6)Gal>(α1,3)Galf(β1,3)[Glc(α1-P)]Man(α1,3)Man(α1,4)GN(α1,6)-PI, and does not synthesize repeating Galβ1,4Manα1-PO4 units [57]. The lpg2-KO+LPG2 add-back was grown in the presence of 50 µg/ml G418. For infections, promastigotes were used in late stationary phase of growth. L. donovani amastigotes (Strain LV9) were isolated from the spleen of infected female LVG Golden Syrian hamsters (Charles River, St-Constant, QC, Canada), as described [58].
The rabbit anti-Syt V spacer antiserum was raised against the cytoplasmic region between the transmembrane and the C2 domain (aa 71–216) [37] and was affinity-purified. The rat monoclonal antibody against LAMP-1 developed by J. T. August (1D4B) was obtained through the Developmental Studies Hybridoma Bank at the University of Iowa, and the National Institute of Child Health and Human Development. The rabbit antiserum against the 16 kDa proteolipid subunit (c subunit) of the V0 sector of the V-ATPase was kindly provided by Dr. Mhairi Skinner (University of Guelph, ON, Canada) [59]. The mouse monoclonal antibody against EEA1 was from BD Transduction Laboratories. The rabbit antiserum against cathepsin B was from Millipore and the rabbit antiserum against cathepsin D was from Upstate. The mouse monoclonal anti-LPG (CA7AE) was prepared from hybridoma supernatant [60]. Methyl-β-cyclodextrin (MβCD) was from Sigma (St-Louis, MO, USA). LPG was isolated from the log phase cultures of L. donovani promastigotes as previously described [61],[62]. Purified lipoarabinomannan (LAM) from H37Rv strain of Mycobacterium tuberculosis was from Colorado State University (Fort Collins, CO, USA).
Syt V silencing by RNAi was performed as previously described [40] using a small interfering RNA (siRNA) corresponding to nucleotides 94–112 of the Syt V cDNA [38], whereas a siRNA specific to GFP was used as a negative control [63]. Adherent RAW 264.7 cells were transfected with siRNA duplexes at a final concentration of 240 nM using OligoFectamine (Invitrogen) as described [63]. A BLAST search against the mouse genome sequence database was performed to ensure that the chosen siRNA sequences targeted only the mRNA of interest.
Cholesterol depletion was achieved by incubating macrophages with 10 mmol/L methyl-β-cyclodextrin (MβCD) (Sigma) in serum-free medium at 37°C for 1 h. Cells were washed with PBS before particle internalization.
Purified LPG and LAM were sonicated and added to the particles at a final concentration of 25 µM in PBS, pH 7.3, incubated at 37°C for 1 h. Particles were washed and resuspended in complete medium prior to phagocytosis experiments. The efficiency of LPG coating was assessed by immunofluorescence using the anti-repeating unit antibody CA7AE. Complement opsonization of L. donovani promastigotes was done as described [23] and complement opsonisation of beads and zymosan was carried out by incubating the particles in DMEM supplemented with 10% mouse serum for 30 min at 37°C prior to phagocytosis.
For synchronized phagocytosis assays, macrophages were incubated with particles at a particle-to-cell ratio of 15∶1 (unless otherwise specified) for 15 min at 4°C. Excess particles were removed by several thorough washes with DMEM and phagocytosis was triggered by transferring the cells to 37°C for the indicated time points before processing for microscopy.
Macrophages were fixed for 10 min in PBS containing 2% paraformaldehyde, permeabilized using 0.1% Triton X-100, and nonspecific binding to surface FcγR was blocked using 1% BSA, 2% goat serum, 6% milk, and 50% FBS. For immunostaining, cells were labeled with the appropriate combinations of primary antibodies or antisera (anti-Syt V, LAMP-1, EEA1, cathepsin D, cathepsin B, V-ATPase, LPG), and secondary antibodies (anti-rabbit, anti-mouse or anti-rat AlexaFluor 488, 568 or 647; Molecular Probes). DRAQ5 (Biostatus, Leicestershire, UK) was used to visualize macrophage and parasite nuclei and CTX-B-568 or 647 (Molecular Probes) was used to visualize GM1-enriched rafts. Syt V-GFP RAW 264.7 cells were fixed and directly incubated with DRAQ5 before being mounted or subjected to immunofluorescence. Of note, we used Syt V-GFP RAW 264.7 cells to localize Syt V following infection with L. donovani promastigotes because our antiserum against Syt V cross-reacts with Leishmania epitopes. All coverslips were mounted on glass slides with Fluoromount-G (Southern Biotechnology Associates). Detailed analysis of protein presence and localization on the phagosome was performed using an oil immersion Nikon Plan Apo 100 (N.A. 1.4) objective mounted on a Nikon Eclipse E800 microscope equipped with a Bio-Rad Radiance 2000 confocal imaging system (Bio-Rad, Zeiss). Images were obtained using appropriate filters, through the sequential scanning mode of the LaserSharp software (Bio-Rad Laboratories, Zeiss) with a Kalman filter of at least 6.
BMM were preloaded with the acidotropic dye LysoTracker Red (Molecular Probes, Eugene, OR) diluted in DMEM (1∶1000) for 2 h at 37°C. Cells were washed and infected with promastigotes for 2 h at 37°C as described in Phagocytosis assay. Cells were then rinsed, fixed with 2% paraformaldehyde for 10 min, washed and directly incubated 20 min with DRAQ5 before being mounted for confocal analysis.
To assess the recruitment of proteins of interest, we assessed the presence or absence of staining on the phagosome membrane for each protein, and at least 100 phagosomes were randomly scanned for each condition. To quantify the levels of Syt V and Syt V-GFP (Figures 2C, 3D and 3E), EEA1 (Figure 3A) or LAMP-1 (Figure 3B), we determined the relative staining intensity as follows. The 488 and 568 nm excitation channels (emission 515/30 and 600/40 respectively) were separated and the protein staining rim around each phagosome was manually traced with a one pixel width. The fluorescence intensity of individual pixels was determined using the software Image J and an average intensity was calculated for each fluorescence intensity profile. To normalize intensity values of all phagosomes, cytosol intensity was also evaluated in the proximity area of the phagosome under study but far enough from the phagosome membrane to avoid quantifying residual phagosome fluorescence. Final phagosome intensity was expressed as the ratio of phagosome intensity (P) on cytosol intensity (C), thus P/C. In all cases, we ensured that signal intensity was not at saturation and the 20 more intense staining for each condition were selected and the average compared for the intensity level of each protein.
Statistical analyses were performed using Student's two-tail two-sample unequal variance test.
|
10.1371/journal.pcbi.0030253 | Social Interactions in Myxobacterial Swarming | Swarming, a collective motion of many thousands of cells, produces colonies that rapidly spread over surfaces. In this paper, we introduce a cell-based model to study how interactions between neighboring cells facilitate swarming. We chose to study Myxococcus xanthus, a species of myxobacteria, because it swarms rapidly and has well-defined cell–cell interactions mediated by type IV pili and by slime trails. The aim of this paper is to test whether the cell contact interactions, which are inherent in pili-based S motility and slime-based A motility, are sufficient to explain the observed expansion of wild-type swarms. The simulations yield a constant rate of swarm expansion, which has been observed experimentally. Also, the model is able to quantify the contributions of S motility and A motility to swarming. Some pathogenic bacteria spread over infected tissue by swarming. The model described here may shed some light on their colonization process.
| Many bacteria are able to spread rapidly over the surface using a strategy called swarming. When the cells cover a surface at high density and compete with each other for nutrients, swarming permits them to maintain rapid growth at the swarm edge. Swarming with flagella has been investigated for many years, and much has been learned about its regulation. Nevertheless, its choreography, which is somewhat related to the counterflow of pedestrians on a city sidewalk, has remained elusive. It is the bacterial equivalent of dancing toward the exit in a crowd of moving bodies that usually are in close contact. Myxococcus xanthus expands its swarms at 1.6 μm/min, about a third the speed of individual cells gliding over the same surface. Each cell has pilus engines at its front end and slime secretion engines at its rear. Using the known mechanics of these engines and the ways they are coordinated, we have developed a cell-based model to study the role of social interactions in bacterial swarming. The model is able to quantify the contributions of individual motility engines to swarming. It also shows that microscopic social interactions help to form the ordered collective motion observed in swarms.
| Bacterial swarming, a coordinated motion of many bacterial cells, facilitates their spread on the surface of a solid medium, like agar [1]. Swarming may have evolved to permit the bacteria in a colony to expand their access to nutrients from the subsurface and to oxygen from above. When the surface is a tissue in a live host, pathogenic bacteria swarm to create a biofilm and to spread the infection. Swarming is observed in cells that are propelled by rotating flagella [2], by the secretion of slime [3], and by retracting type IV pili [4,5]. Bacterial swarming has been studied quantitatively in the modeling context of self-propelled particle systems [6–8]. Most models, such as those for Bacillus subtilis and Escherichia coli (see [8] for a review), are based on long-range cellular interactions facilitated by chemical gradient or nutrient level (chemotaxis). However, myxobacteria show no evidence of long-range communicating systems to guide their collective motion; they have only local contact signaling and use social interactions between neighboring cells for swarming [9]. How interactions between cells facilitate swarming is still an open question. Understanding this question might shed light on the self-organizing process in bacteria, such as the spreading of a biofilm in an infected tissue and the development of multicellular fruiting bodies [4,5]. In this paper, we describe a new cell-based model and study the effects of social interactions between cells, including the interaction mediated by slime trails and by type IV pili, on swarming. Type IV pili are found at one pole of a wide range of bacteria, including many pathogens that cause plant and animal disease.
We chose to examine Myxococcus xanthus because it swarms rapidly, has typical type IV pilus engines at the front end of cells, has slime secretion engines at the rear, and coordinates the two engines with each other. M. xanthus has been studied for more than a century; numerous swarming mutants have been identified and characterized. Myxobacteria are commonly found in cultivated soils, where they feed on other bacteria. On the surface of nutrient agar, they swarm away from a point inoculum, spreading outward at a constant rate for 2 wk. Although the bacteria are growing (and in fact they must grow to swarm), 90% of the swarm expansion rate is due to motility and to interactions associated with motility, as shown by the low spreading rate of nonmotile mutants [10,11].
Individual M. xanthus cells are rod shaped, roughly 5 μm in length and 0.5 μm in width. They have two types of molecular motors that provide the thrust necessary for their gliding movement over a surface [9]. At the leading end of the cell are retractile type IV pili, long and thin hairs responsible for S motility. When a cell is close to a group of other cells, the cell's type IV pili can attach to the fibrils, which cover the surface of the neighboring group of cells like a fisherman's net. After attachment, the pilus retracts, and the retraction force pulls the piliated cell forward, while the group hardly moves. This pilus-mediated interaction produces many asymmetric cell clusters that often have tips that are pointed at one end (arrowhead-shaped) and is characteristic of S motility. Arrowheads can be seen in the young swarms of A−S+ mutants [11]. S motility is found among pathogenic Neisseria and Pseudomonas, where it is called twitching motility [4,5].
At the trailing end of myxobacterial cells are several hundred pores, from which slime is secreted. There are roughly 150 pores scattered over the sides of the cell, also secreting slime that becomes a thin layer, protecting the cell from lysis by cell-wall digestive enzymes being secreted by all the cells [3]. Both the lateral and the polar slimes are thought to be the same polysaccharide that is part of A motility (hereafter simply referred to as slime). Importantly, slime is completely distinct from the fibril polysaccharide that serves S motility [11]. Slime secretion from the rear pushes the cell forward, leaving a trail of slime behind the cell [3,12] and generating movements called A motility. When a moving cell encounters a slime trail, it tends to turn through the acute angle to follow the slime trail. When an A motile cell collides with the side of another cell, the pushing of the slime engines at the rear causes the cell, which is flexible, to bend. The colliding cell thus reorients parallel to the other cell, producing a side-by-side cluster of cells. Such clusters are transient because the two cells do not adhere and often slide past one another.
The A and S motility engines, which are located at opposite poles of the rod-shaped cells, have engine-specific social interactions. During movement, a cell's polarity reverses regularly every 10 min or so [13,14], and reversal is required for swarming [15,16].
A wild-type cell (A+S+) expresses both A and S motilities. A+S− mutants express only A motility, while those with S motility but no A motility are called A−S+ mutants [9]. Because wild-type and A+S− mutants are self-propelled by A motility engines, a comparison can expose the social interactions specific to the type IV pili. In both cases, individual cells are observed to move, stop, and move again, sometimes slightly changing direction and regularly reversing [3]. To investigate the coordinated motion within M. xanthus swarms, culture droplets of each mutant were placed on agar plates, and the swarm expansion rates were measured [10]. Figure 1 shows the edge of a typical swarm of wild-type (A+S+) cells. It is observed that swarm expansion rates remain constant until the swarm covers the entire surface available [10]. The expansion rates for various initial cell densities in K-S units were measured and plotted against the cell densities. (K-S is Klett-Summerson unit; a measurement of cell density in suspensions [10]. A sample of cell suspension with 100 K-S units has approximately 4 ×108 cells/ml. Using the experimental data in [10], we find that 100 K-S units correspond to a close-packing arrangement of cells in a 2-D area.) The fitted functions of expansion rate data for the three cell types are shown as solid lines in Figure 2. To a first approximation, the velocity of individual cells, when they are moving, is the same for S− mutants (A+S−) and wild-type (A+S+) cells, about 4 μm/min, but their swarm expansion rates are different [10]. The A+S− and A−S+ mutants swarm with a maximum rate of 0.67 μm/min and 0.46 μm/min, respectively. Surprisingly, when S motility cooperates with A motility in wild-type M. xanthus (A+S+), the maximum swarming rate is 1.55 μm/min, about 2.3-fold larger than that of A+S− ([10], as shown in Figure 2).
Previously, we used a lattice-based model to study myxobacterial fruiting body development after starvation [17,18]. Swarming with sufficient nutrient supply has been studied using a continuous model in the form of partial differential equations (PDEs) [19]. The effects of engine mechanics and cell shape have yet to be taken into account. Recently, we introduced a simplified off-lattice stochastic description of swarming [20], and herein add our current understanding of engine mechanics to investigate swarming and the role of social interactions.
This paper is organized as follows. We start by describing the model of cell behavior and social interactions. Then, we present the simulation results and compare them with the experimental observations. We demonstrate a constant rate of swarm expansion and show that the model accounts for the significant difference in swarming rates between wild-type and A+S− myxobacteria arising from the loss of S motility. We also study in detail the order of collective motion in myxobacterial swarms. A detailed description of the computational model is given in the Methods section.
In this paper, we focus on the collective motion of a large number of cells in a swarm of high cell density, taking only the local, contact-mediated interactions between cells into account. We represent each cell as a string of N nodes in a 2-D space, following our earlier work [20] (Figure 3). The vector pointing in the direction from the tail node to the head node represents the orientation of a cell. We define an energy function (Hamiltonian) for the node configuration of a cell body and use it to constrain the cell length and the cell shape to a certain range. The active motion of an individual cell is then modeled as follows. After the head moves in a particular direction, a Monte Carlo approach [21] is used to reconfigure positions of other nodes in an attempt to minimize the Hamiltonian (see Methods). This allows the cell body to bend and to change its length by random fluctuations, which reflects the experimental observations [22].
As mentioned in the introduction, the measured velocities of individual cells vary over a wide range, but the average velocities of A+S− and A+S+ cell types are similar. To a first approximation, we take the cell velocity to be constant and the same for wild-type A+S+ and A+S− cells, with a magnitude of 4 μm/min [10]. The direction of cell movement is determined dynamically by the model, which takes the interactions between neighboring cells into account.
Frequently cells reverse their motion by 180°. Reversals are regulated by an internal biochemical clock that is not affected by collisions or other interactions between cells [15,16]. We model regular reversals of cell motility engines by switching roles of head and tail nodes in accordance with an internal clock (see Methods). The swarming efficiency (the ratio of the swarm expansion rate to the speed of individual cells) of myxobacteria primarily depends on social interactions between neighboring cells. The expansion rate of a swarm without social interactions would be zero, since the cells would move back and forth equally without any net displacement in the long run. Social interactions help a swarm of reversing cells to spread.
Ideally, interactions between all the more than 107 cells in a swarm would be considered, but that is not possible in practise. Instead, we try to identify for each cell a neighborhood within which a majority of its interactions are expected to be found. Social interactions arise in S motility when the type IV pili of one cell attach to the fibrils that surround other cells. Social interactions arise in A motility from the tendency of a cell to follow the trail of slime left by another cell, and from collisions between cells that cause a moving cell to stop and its engines to stall, or those cause a cell to change its direction. Using the experimental data of Figure 2, an area of interaction for each cell type, A+S−, A−S+, or A+S+, was defined as the statistically averaged area around a cell within which most of its social interactions occur. The interaction areas were taken to be proportional to the inverse of parameters in the exponential term of the formula obtained when an exponential curve was fitted to the experimental data in Figure 2. Fitting functions are specified in the legend to Figure 2. Each curve represents the observed swarm expansion rate as a function of the initial cell density of the culture. The interaction area for wild-type cells was found to be smaller than the sum of the interaction areas of the A+S− and A−S+ mutants. We suggest that this unanticipated finding results when both engines are working because the two engines on a wild-type cell are not statistically independent but are constrained by the structure of a cell to propel it in the same direction.
Pilus-mediated interactions depend on the dynamics of pilus retraction [23] and on the spatial distribution of the fibrils to which the pilus tips have attached [24,25]. Although these factors are mechanically complex and not yet understood in detail, the interaction has straightforward effects. Pilus retraction provides a driving force for cell movement that happens to be large, several times larger than the force developed by muscle acto-myosin. And, because the force is almost never directed along the cell's long axis, the force tends to reorient the direction of gliding. Because we are confined by the approximation that isolated cells move with constant speed, we need only consider the reorienting effect of pilus retraction. No effect on cell speed is considered, except that it drops to zero when one cell collides with another. Inasmuch as the fibrils tend to bundle groups of cells, as will be described below, the large size of the cell cluster prevents a significant reorientation of the bundle; only the cell whose pili have attached is reoriented. We model the reorientation effect of pilus-mediated interactions as driving the local alignment of cells (see area I of Figure 4 and Methods). Although we represent the interaction area by a rectangle, a circle or some irregular domain could have been used. The important quality of an interaction domain is its area. That area is proportional to the probability that a cell has an interaction. Swarms of wild-type cells cover a larger area than those of A+S− or A−S+ mutants [11]. Moreover, the peninsulas are denser with cells that are well-aligned side by side [10]. Both effects illustrate reorientation due to pilus retraction. Cell clusters tend to be narrow in the case of an A+S− mutant and wide in the case of wild-type bacteria.
A motility engines at the rear of the cell push it forward in the direction of their long axis. A motility also produces slime trails, and cells tend to follow them due to the adhesion of newly secreted slime to the older slime in the trail. The resulting alignment of the slime polysaccharide chains also reorients the direction of gliding. Slime trails are represented in the model by the paths that were taken by the last cells to have passed through area II (Figure 4). Further details are given in the slime orientation field described in Methods. Rod-shaped A motile cells, which are pushing at their tail ends, tend to form parallel arrays if they collide or come into close contact with each other. These effects are illustrated in area II of Figure 4 and are elaborated in Methods. Alignment results from inelastic collisions between cells that change their orientations. More generally, alignment in regions of high cell density arises spontaneously from the physical clustering of self-propelled rods [26].
For wild-type cells, we first model A and S motilities, individually. Then, we combine them under the approximation that isolated cells move at a constant rate, as described above. The persistent active motion is taken to be led by the head of the cell, no matter which engines are functioning. Finally, we model the reorientation due to pilus retraction and to the alignment of A motile cells with their neighbors, or with the slime field.
To test the consistency of the model, we simulated the motion of cells near the edge of the swarm, and studied the expansion of the swarm. Although M. xanthus swarms consist of many millions of cells, the radial symmetry of a swarm makes it possible to consider a small rectangular sector of the swarm (Figure 1). A rectangular area of 200 μm by 200 μm (Figure 5) was convenient.
To compare the simulation with experimental measurements (shown in Figure 2), we considered that growth in the center of the swarm was driving a net radial outflow of cells from their center [15], and that the swarm was expanding at a constant rate. A constant cell density near the swarm edge was observed experimentally as the edge moved out [10]. Skipping the early transient phases, we start the simulation after the steady state has been reached. Although cells in the initial area are oriented in all directions, the orientations are radially symmetric. Both conditions apply to the “Initial Area of Cells” in Figure 5.
Denoting cell density as p(r, θ) and the radial density as P(r), due to the symmetry and the steady state, we have:
This relation shows that the cell number flux across the lower boundary of the initial area (or the increase rate of total cell number in the whole simulation domain) is linearly correlated with the colony expansion rate in Figure 2. We calculate the cell number flux rather than expansion rate directly. Therefore, we do not have to increase the simulation domain or the total number of simulated cells. Further details of the simulation setup, implementation of the algorithm, and the choices of parameters are described in Methods and Table 1. Simulations show formation of long clusters (peninsulas) in both A+S+ and A+S− cases (see Figure 6A and 6B), which was observed experimentally [10].
Simulations were performed for cell densities ranging from 2 to 200 K-S units, and linear increase of cell number was observed in all cases (for example, see Figure 6C). This implies that the cell number flux is almost constant during the whole swarming process for a given initial cell density, in full agreement with experiments [10]. Figure 6 corresponds to an initial density of 50 K-S units, which is a near saturation density for the rate curves of Figure 2.
We have calculated linear fits for the cell number increase data at various cell densities, and taken the slopes to be the average cell number fluxes, as shown in Figure 6C. The results for both A+S+ and A+S− cells are plotted against cell density in Figure 7A. We found that the cell number flux of the wild-type cells (A+S+) is greater than that of the A+S− mutant at all cell densities. At densities higher than 50 K-S units, the cell number flux for A+S+ is 2-fold larger than the A+S−.
To see this effect more clearly, we fit the average cell number flux data into the first order exponential decay function, which is similar to the function used in Figure 2 from [10]. The fitting functions for wild-type A+S+ cells and A+S− mutants are found to be f(x) = 5.6 − 6.6 × exp(−x / 32.1) and g(x) = 2.8 − 3.1 × exp(−x / 29.4), respectively. The ratio of these two fitting functions is plotted against cell densities in Figure 7B. It is equivalent to the ratio of colony expansion rates since the cell number flux is linearly correlated with the expansion rate. The ratio first increases and then saturates around 2 for cell densities higher than 50 K-S units. Experimental data shows that the A+S+ rates are 2-fold to 2.5-fold larger at cell densities higher than 50 K-S units (Figure 2). Therefore, our result shows a significant difference in swarming rates between wild-type and A+S−, arising from the contribution of S motility that agrees with the experiment.
Collision occurs in the model whenever the head of one cell overlaps the area occupied by another cell. At this point, the moving cell stops; it is not permitted to glide on top of the other cell. As a consequence, at high cell densities the movement of individual cells is reduced. In reality, cells do glide over each other. Reduction becomes significant above 100 K-S units, because at 100 K-S units the average area occupied by an individual cell is close to the area of a cell body (i.e., the area is closely packed with cells). In practice, due to the tendency of cells to cluster, cell movement is reduced beginning at concentrations of 60 K-S units. This effect explains the decrease in cell flux observed at higher cell densities (>60 K-S units) for the wild-type cells in Figure 7A. The decrease results in a smaller value of maximum ratio (about 2-fold; see Figure 7B) than experimental data (about 2-fold to 2.5-fold).
Comparing the three curves of Figure 2 shows clearly that S motility contributes to the swarming of wild-type (A+S+) cells. Figure 2 also shows that A−S+ swarms expand without help from A motility, although the rate of expansion is less than one-third that of the wild-type cells at every cell density. With these data in mind, a puzzle takes shape: how are pili able to support expansion of an A−S+ swarm when there should be no fibrils to which the type IV pili might attach beyond the edge of the swarm? The surface ahead of the swarm edge never had cells upon it. Must the belief that pili attach to fibrils before they can retract be abandoned? This section describes an attempt to solve the puzzle by examining the evidence that pili bind fibrils specifically, by offering a mechanism whereby specific binding and retraction can bring about the expansion of an A−S+ swarm, and by testing the mechanism proposed.
Evidence for specific binding includes the observation that A−S+ cells move only when they are within a pilus length of another cell [10,27]. Fibrils are present in profusion, and they envelope clusters of adjacent cells (see Figure 2 from [28]). Although only half of the fibril mass is polysaccharide (the other half is protein [24]), several experiments have revealed that removing the protein has no effect on pilus binding [24,25,29–31]. Evidently, M. xanthus pili bind fibril polysaccharide.
Therefore, side-by-side clusters of M. xanthus cells, like the peninsulas in Figure 8, are viewed as a bundle that is enveloped by an elastic fisherman's net formed by association of polysaccharide fibrils that the cells have secreted. Bundling of cells by fibrils offers an explanation for the pointed shape, which A−S+ peninsulas tend to have. The points aim away from the swarm center (Figure 8 and [10,32]) and in the general direction of swarm expansion. The shape and orientation of the peninsula tips suggest that cells at the tip of the peninsula have been pushed into their position at the tip. Consider a cell within the body of the peninsula that happens to be moving toward the tip of the peninsula. This cell will have projected its pilus forward and attached it to the fibril network on cells ahead of it and closer to the tip. Retraction of that pilus could pull the cell forward and upward to add a new layer of cells to the peninsula. Indeed, most peninsulas have a second (or third) layer near their tips, which are evident in Figure 8. On other occasions, retraction would pull the piliated cell right up to the end of a cell in the bottom layer that lies just ahead of our piliated cell. Recalling the description of A motility in the Introduction, each cell is also covered by the slime polysaccharide, which protects them from autolysis. Since the network of fibrils that envelops cells of a peninsula bundles them, both the elasticity of the fibrils and the cohesion between the slime on adjacent cells would tend to prevent their separation, by wedging action of the rounded end of the pushing cell, from cells to their left and right in the tip of the peninsula. Consequently, complete retraction of the pilus would cause the moving cell to push the cell in the peninsula that is immediately ahead of it. The pushed cell might slide forward while adhering through its slime covering to the cells on either side. Localized sliding would be reflected in a sharpening of the tip contour to a point, as observed (Figure 8).
The hypothesis of pushing by S motile cells was tested by analysis of seven time-lapse movies of the advancing edges of A−S+ swarms, each movie of 1 h to 3 h in duration. Figure 8 is a single frame from one of the movies. In that frame, numerous single cells and many peninsulas of various sizes are evident. Several observations relevant to A−S+ swarming could be made from the movies (Key and Kaiser, unpublished data). First, almost all of the many thousands of cell movements were found within clusters of ten or more cells. No isolated cell moved significantly, unless the cell was within pilus-striking distance of another cell. This shows that the cells are moving with S motility alone. Second, although the peninsulas either elongated or moved forward, the translocation rate was much less than the rate of individual cell movement in the same field. A lower rate correlates the peninsula's advance to its being pushed from behind, because the hypothesis has the pushed cell sliding past its neighbors in the peninsula. The sliding friction would decrease the rate of advance. Finally, the movies show many examples of individual cells, which appear to be moving more or less randomly, behind an arrowhead or a peninsula. Individual cells advancing toward the rear edge of the peninsula could have pushed it. A quantitative analysis of cell movement in the movies will be published separately, but this qualitative analysis supports pushing.
In previous sections, we have shown that our model for social interactions is consistent with experimental results at the level of individual cells. In this section, we investigate how microscopic social interactions facilitate swarming at the population level. We demonstrate that social interactions lead to an increase in the order of collective motion, which is strongly correlated with swarming efficiency. We start by introducing an order parameter to characterize collective motion of bacteria in swarms with complex clustering patterns.
After analyzing experimental data and taking into account regular reversals of myxobacteria cells, we define the most ordered state as follows: all cells move side by side in close contact with each other in the same or opposite direction. The collective motion is considered purely nonordered when either one of the following criteria is satisfied: (1) the orientations of neighboring cells of any given cell are random (or uniformly distributed); and (2) any pair of cells is well-separated so that cells are not in direct contact.
Vicsek et al. [33] used the average velocity as a global order parameter for analyzing the motion of self-propelled particles. However, myxobacteria cells reverse regularly, and two opposite directions should be considered as being equivalent to each other. There are always cells moving in the direction opposite to the net motion of the whole cluster in most cell clusters in experimental movies. Also, as shown in the inset of Figure 1, the swarming pattern often exhibits localized clusters of aligned cells with different orientations of motion, and one would need to take local order into account when measuring global order of motion. Therefore, the average velocity is not the best way of measuring the nematic order in myxobacteria swarms.
We first define two local measuring components to describe the local orientational order and positional order of a given cell, denoted as Ψ and P, respectively. For a given cell k (k = 1,2…M, M is the total cell number), we choose the rectangular domain (of area s0) illustrated in Figure 9 as the local measuring domain (one cell length by two cell lengths), centered at the center of mass of a cell. We then measure the total area S occupied by neighboring cells within the local measuring domain and define the local positional order as the following:
We record the orientations θj, with j = 1,2,…n of the neighboring cells with either head node or tail node inside the local measuring domain, in a way used in Equation 11 in Methods. Then the angles between these orientations and the x-axis in Figure 9,
, with j = 1,2,…n, and
, are calculated. If cell k has no neighbors (j = 0), we define Ψ as 0. Otherwise, the local orientation order function is defined as follows:
with
and
Φk determines how ordered the distribution of
is. The most ordered state corresponds to the case when all
are equal and Φk has a maximum value of 1. Φk is rescaled to the range between 0 and 1. Therefore, in the case of uniform (random) distribution of
, the local orientational order function Ψk is equal to 0. In the most ordered state, all
are equal and Ψk is equal to 1.
Finally, we combine both the local orientational order and positional order components from Equations 3–5 to define the global order parameter for the collective motion of myxobacteria:
where M is the total number of cells.
The order parameter Ω has been specifically designed for myxobacterial swarming. Figure 10 shows values of Ω for the simulation of swarming near colony edge with initial cell density of 50 K-S units. We find that the order of collective motion in both A+S+ and A+S− swarms steadily increase, and that A+S+ cells achieve a much higher (about 2-fold) order than A+S− cells.
Further, we look at the order of cellular motion in the inner area of myxobacteria colony. In Figure 11A, cells are randomly distributed in a square area of size 167 μm × 167 μm with a density of 50 K-S units. All boundary conditions are periodic. This is different from the previous simulations for cells near the colony edge, because we do not assume a preorganized orientation distribution of cells.
Figure 11B and 11C are the simulation pictures after 3 h for A+S− mutant and wild-type cell (A+S+) swarms, respectively. We see that the pattern of A+S− mutant exhibits lower order, while wild-type (A+S+) cells form large clusters oriented in various directions. Plots of the parameter Ω are presented in Figure 11D. Again, we see that the order of motion in both A+S+ and A+S− cases increase with time, while A+S+ cells achieve a much higher order than A+S− cells.
Therefore, we demonstrate that social interactions lead to an increase in the order of collective motion. Type IV pilus-mediated interactions increase the order much greater than social interactions associated with A motility. This is consistent with the experimental findings by Pelling et al. [32], who observed higher-order patterns within wild-type (A+S+) swarms in comparison with motility mutant swarms.
Comparison of Figure 10B with the ratio of cell number fluxes (Figure 7B) indicates that the order of collective motion strongly correlates with the swarming efficiency. We suggest that higher order of motion results in greater swarming rates as observed in wild-type myxobacteria experiments. It explains the origin of the significant difference in swarming rates between wild-type and A+S− myxobacteria arising from the coupling of S and A motilities.
We have developed an off-lattice cell-based computational model to study the role of social interactions in bacterial swarming. The model is stochastic and is based on detailed description of the bacterial motility engines and their regulation. The model demonstrates how social interactions facilitate bacterial swarming, and provides an explanation to the significant difference in swarming rates between wild-type and A+S− mutants arising from the effects of S motility. Our simulations indicate that the order of collective motion strongly correlates with the swarming efficiency, which provides a connection between microscopic social interactions and population-level swarming behavior.
The model is two-dimensional and provides a very good approximation for the bacterial behavior near the edge of the swarming population. However, in experiments at higher densities, cells were observed to glide on top of each other, resulting in multiple cell layers just behind the edge of the swarm. As discussed in Results, the 2-D nature of our model causes a slight decrease in cell number flux at higher cell densities (>60 K-S units) for wild-type myxobacteria (Figure 7A), and results in a smaller value of maximum ratio (about 2-fold; see Figure 7B) than experimental data (about 2-fold to 2.5-fold). A 3-D extension of the model will avoid such affects, and allows us to study cell clustering inside of a swarm as well as during fruiting body development under starvation [17,18].
We did not quantitatively study the motion of mutants with impaired A motility (A−S+ mutants) in this paper. As discussed in Results, A−S+ cells only have persistent active motion when they are within a pilus length of other cells so that the type IV pili can attach to the fibril materials on the surfaces of other cells [34]. Wild-type and A+S− mutants both have A motility that can produce persistent active motion. The only difference between wild-type cells and the A+S− mutant is the effects of S motility, so it is more convenient to take wild-type cells and the A+S− mutant as the modeling systems. By comparing their movements, we could investigate the role of pilus–cell interactions during swarming, which was one of our aims. In Results, we have presented a qualitative analysis of A−S+ swarming, which demonstrated that the pushing of cells near the swarming edge can explain the expansion of A−S+ swarms. Preliminary simulations with the pushing mechanism show qualitative agreement with the experiment in terms of peninsula shape and cell ordering (unpublished data). Quantitative modeling of A−S+ swarming dynamics will require more knowledge of the distribution and mechanical properties of the fibril.
When studying the effects of social interactions, we have related the swarming efficiency with the order of collective motion. This order parameter may provide a novel perspective on quantifying the condition of bacterial swarming. Further experimental investigation of this concept will rely on advances in microscope and image processing of microphotographs. Such experiments require very-high-resolution imaging that can cover large areas of a live bacterial colony [35]. We defined an appropriate order parameter, which characterizes the combined local orientational and positional order. Not limited to the case of myxobacteria swarming, the order parameter provides a quantitative measurement of collective motion in nematic biological systems where local interactions play a dominant role.
We have shown that social interactions mediated by type IV pili, when coupled with active motion, have an alignment effect on neighboring cells and significantly facilitate swarming. Many pathogenic bacteria swarm within infected tissues and have type IV pili as virulence factors. It is likely that the ascent of Proteus mirabilis up the urinary tract is a result of growth and swarming with flagella [36]. Similarly, the spreading of Neisseria in infected tissue is related to swarming with type IV pili, since those pili are necessary for virulence [37]. The bacterial swarming model described in the paper may therefore shed light on the colonization and infection process of pathogens.
In the model, each cell is represented by a flexible string of N nodes (Figure 3) consisting of (N − 1) segments, each of length r. There are (N − 2) angles θi between neighboring segments. For each cell, we define the following energy function (Hamiltonian):
The first term in Equation 7 is the stretching energy determined by the cells' length. The second term is a bending energy. Kb and Kθ are stretching and bending dimensionless coefficients, analogous to the spring constants in Hooke's Law. They determine the extent to which the segment length and angles can change in the presence of fluctuations, respectively. They are the same for all segments and angles. r0 is the target length of a segment. In our simulations, we choose the number of nodes N = 3 (Figure 3), so that r0 is 2.5 μm (half-cell length). Kb and Kθ are set at 5 and 2, respectively, based on experimental observation that cells do not change their length a lot, but can bend rather easily.
Let's denote the dark cell in the center of Figure 4 as cell k. In the absence of cell–cell collisions, the velocity direction of cell k is determined by three contributions: a motility direction, orientation from slime trail, and orientation from type IV pili.
A motility direction. The cells secrete slime (polysaccaride) from their tail end, which expands as it leaves the cell body and pushes the cell directly forward [12]. We model this motility by trying to orient the cell along its long axis, which is the tail-to-head direction. The corresponding term in formulae below is denoted as
. Small deviations from the direction of long axis are observed [10]. This is modeled using a Monte Carlo reconfiguration algorithm.
Orientation from slime trail. When a moving cell encounters a slime trail, it tends to turn through an acute angle to follow the trail. We define a 2-D slime-orientation vector field
that records the slime trail orientation as a vector assigned to each position
. This vector coincides with orientation of a cell that passed through
most recently. We make a simplifying assumption of all orientation vectors having unit length. Once a slime trail is laid down at position
, it will be cleared after the slime aging time Ts.
Orientation from type IV pili. As discussed in an earlier section (Model of cell behavior and social interactions), type IV pilus-mediated interactions are assumed to align neighboring cells. For a particular cell k, we average the orientations of its neighboring cells within the pilus-cell interacting area (Figure 4, area I), and define this averaged direction as the contribution of pilus-mediated interactions to the head velocity direction of cell k. This term is denoted as
.
Cell velocity direction. When there are no collisions between cell k and its neighbors, the direction of its head velocity (denoted as
) is determined by the sum of A motility direction, orientation from slime trail, and orientation from type IV pili:
a motility direction:
orientation from type IV pili:
In Equation 8, C is a constant cell speed (4 μm/min); α, β, and γ are parameters representing the relative strength of each motility term. Denominators are used in all equations for normalization. The experiments suggest that the forces generated from A and S motilities are nearly the same, approximately 150 pN [9,12]. For an A+S+ cell, we choose α = γ = 1.0, and for an A+S− mutant, we choose α = 1.0 and γ = 0. The strength of slime orientation effect is set as β = 0.5. Note that the slime-orientation vector field
is recorded in a discrete 2-D lattice, with each lattice site having a slime-orientation vector. Slime trails interact with the newly secreted slime, not with the head of a cell. We analyze slime-orientation vectors at the lattice sites covered by the front half of a cell body, and take the direction, which occurs most frequently, as slime-orientation direction to be followed by the cell. It is denoted as slime
in Equation 8.
Equation 9 determines cell orientation, which is the direction from the tail node (
) to the head node (
) and which is considered the A motility direction. In Equation 10, n denotes the total number of neighboring cells of the cell k. We multiply the expression by the factor of n because we think that type IV pili have a stronger effect on the direction of motion of the head node (pili are located at the head of a cell), and that this effect depends on the number of neighboring cells. The terms cosθj and sinθj are the x and y components of the orientation vector
of the j-th neighboring cell. These vector components are then averaged (<cosθj> and <sinθj>) and are taken as the x and y components of the average direction.
and
denote unit vectors along the x and y axes. We model the alignment in such a way that cells orient with their neighbors to the acute angle. That is, if the dot product of the tail-to-head directions of cell k and its j-th neighbor cell is negative, we choose the opposite direction to
as its orientation. Therefore, we have:
This approach is different from that taken by Vicsek et al. [33]. The alignment is determined through acute angles because we use cell orientations instead of velocity directions.
Collision-resolving algorithm. When the head node of cell k collides with the body of cell j, this collision is resolved as follows:
Calculate distances between the head node of cell k and two end nodes of cell j;
If one of these distances is less then a cell width, choose at random a new direction such that the dot product of new direction of cell k and orientation of cell j is positive, and move;
Otherwise, take the average direction of both cells k and j as the new direction and stall until next time step. (The same method is used in the above alignment algorithm of type IV pilus-mediated interactions.)
Reversal of gliding direction. Each myxobacterial cell reverses its gliding direction every 10 min or so. (Reversal periods of myxobacteria follow a distribution with an average of about 10 min.) For simplicity, we choose the reversal periods in accordance with binomial distribution from 5 min to 15 min [38]. Each cell is assigned an inner reversal clock. The initial values of the clock are assigned at random. At each time step of a simulation, the clock value increases by a unit of time. Cell reverses when the clock reaches the value of the reversal period and the clock is reset to zero.
Simulation setup. The simulation domain is chosen in the form of a rectangle 200 μm by 200 μm (Figure 5). In simulations, a unit length is equal to 0.166 μm and one time step is equal to 0.2 min so that the initial cell length (5 μm) is equal to 30 units of length, and cell width (0.5 μm) is 3 units. As mentioned in the text, we approximate that myxobacteria move at the constant speed of 4 μm/min so that in the simulations, a cell moves a distance of 5 units in each time step.
Initially, cells are distributed within the “Initial Area of Cells” (see Figure 5). Cell centers are distributed at random, but cell orientations are distributed around the radial direction in accordance with the normalized distribution function f(x) with a peak at (π / 2):
From experimental observations, it follows that a steady rate of swarm expansion is reached only when most cells behind the swarm edge orient themselves outward along the radial direction. Ideally, one would need to choose the initial orientation distribution f(x) according to the experimental data measured at the beginning of the steady swarming. However, due to the lack of such data, we select the initial orientation distribution function f(x) in such a form that most cells initially point outward from the swarming edge. Cell growth and division are included in our model as maintaining the average density in the simulation domain near the edge.
Algorithm implementation. At each time step, we implement the following sequence of operations for each cell. First, check the inner reversal clock and decide whether to reverse polarity of the cell or not. Then, calculate the velocity direction of the head node according to the model for motility systems. If no collision occurs, move the head node at a distance of five units; otherwise, use the collision-resolving algorithm to resolve the collision. Then, apply Monte Carlo algorithm to reconfigure the positions of other nodes of the cell. Use the procedure suggested in [20]. After moving the head node to a new position, repeat the following operations for (integer part of 2.5N) number of steps (N is the number of nodes per cell): (i) choose node i at random and move it in the direction from node i to node (i − 1) at a distance of 5 unit lengths; (ii) calculate the energy change ΔE due to the relative position change of the nodes. Use the Metropolis algorithm [21] to determine the acceptance probability for the positional change of a node:
Then, record slime-orientation vectors in the end of individual cell movement at all positions passed through by the cell.
After all cells move, calculate the cell number flux through the boundary into the free space and add the same number of cells into the initial area to keep the cell number in the “Initial Area of Cells” constant. Table 1 provides values of modeling parameters.
Our model depends on two parameters characterizing properties of the slime trail: the slime aging time Ts and the relative strength of slime guidance. In this section, we describe simulation results for different ranges of these parameters to test the robustness of the model.
The slime aging time (Ts) is defined as the lifetime of a slime trail during which it has the ability to guide the motion of a bacteria. We used a value of 20 min in our simulations. In Figure 12, we simulate the swarming of wild-type cells and A+S− mutant at the density of 50 K-S units (the same simulation setup as in Figure 6), and varied Ts from 10 min to 200 min (the whole time span of the swarming simulations). We make linear fits for the data points and find that the value of Ts has little effect on simulation results (the cell number flux). This is because slime guidance is primarily a local effect, and slime trails will be washed out by other cells' slimes at short times when the cell density is high. Therefore, the parameter Ts is quite robust for the results in Figure 7B, which is the main validation of our model.
The relative strength of slime guidance is modeled by the parameter β in Equation 8. We used a value of 0.5 in the simulations (see Methods). Here, we varied β from 0 to 1.5, and calculate the cell number flux in swarms of wild-type cell and A+S− mutant type at the density of 50 K-S units (the same simulation setup as in Figure 6). The simulation data are plotted in Figure 13 along with the linear fits. We find that as the slime guidance effect gets stronger, the cell number flux increases. It increases slightly faster in the case of A+S− mutants than in the case of wild-type cells, with the slopes being 0.33 and 0.19 for A+S− mutant and wild-type cells, respectively. This result suggests that as the effect of slime guidance gets stronger, the local alignment of cells and the order of collective motion are both increased. However, this does not affect the results in Figure 7B much, since the increase of cell number flux in the case of A+S− mutants is only slightly faster than that in the case of wild-type cells. The ratio of two fitting functions remains greater than 2-fold until β = 18.5. This demonstrates robustness of our model with respect to the relative slime strength (see Figure 7B). |
10.1371/journal.pcbi.1005976 | Genetic programming based models in plant tissue culture: An addendum to traditional statistical approach | In this paper, we compared the efficacy of observation based modeling approach using a genetic algorithm with the regular statistical analysis as an alternative methodology in plant research. Preliminary experimental data on in vitro rooting was taken for this study with an aim to understand the effect of charcoal and naphthalene acetic acid (NAA) on successful rooting and also to optimize the two variables for maximum result. Observation-based modelling, as well as traditional approach, could identify NAA as a critical factor in rooting of the plantlets under the experimental conditions employed. Symbolic regression analysis using the software deployed here optimised the treatments studied and was successful in identifying the complex non-linear interaction among the variables, with minimalistic preliminary data. The presence of charcoal in the culture medium has a significant impact on root generation by reducing basal callus mass formation. Such an approach is advantageous for establishing in vitro culture protocols as these models will have significant potential for saving time and expenditure in plant tissue culture laboratories, and it further reduces the need for specialised background.
| Trials to find out the best combination of factors that contribute to the desired response takes up the chunk of time and resources in any plant tissue culture laboratory. The output of such experiments is analysed statistically to come to a conclusion. However, without prior statistical modifications, the results could be misleading. Recent reports from several labs point out the use of artificial neural networks to circumvent this. We have chosen to use a computational process that can predict the best combination of factors for the desired response after randomly testing the higher and lower limit of the factors with experiments. The magnitude of the desired response can be presumed at any concentration within this range using the models generated by symbolic regression. The procedure provides both optimum model function as well as the optimum variable values in the model. The variable sensitivity and percentage response add depth to the information thus obtained. The study indicated that these models would have significant potential for saving time and expenditure in plant tissue culture laboratories for the commercial establishment of in vitro protocols.
| Relatively more straightforward and efficient empirical modeling techniques based on input-output models are gaining popularity to conventional statistical methods across various disciplines [1]. This surge is due to its relative ease of use and understanding. Genetic programming (GP) is an approach which uses the concept of biological evolution to handle a problem with many fluctuating variables. Computational optimisation techniques have recently debuted in plant tissue culture research as studied in neural networks models [2]. Symbolic regression was one of the earliest applications of GP and continues to be widely considered [3]. A broad array of scientific fields like Biology, Chemistry, Environmental Science, Neurology and Psychology reports the use of symbolic regression [4–9]. However, plant tissue culture data has not yet been analysed using symbolic regression. The data generated from plant tissue culture experiments includes continuous, count, binomial or multinomial and predominantly the information is validated using analysis of variance method (ANOVA) [2,10]. ANOVA is adequate for normally distributed continuous data; but without prior manipulation, it is erroneous to analyze count, binomial or multinomial data [11]. Neuro-fuzzy logic is the standard practice by which computational modeling is achieved in plant tissue culture [2,12]. In this context, genetic algorithm based symbolic regression remains unevaluated. Unlike conventional regression analysis which optimises parameters for a pre-defined model, symbolic regression avoids imposing any apriori assumptions. In generalised linear model (GLM) regression, the dependent variable is represented as linear combination of the given set of basic functions and optimise the coefficients to fit the data. However, symbolic regression searches for both a set of basic functions and coefficients. The added value of symbolic regression, compared to GLM, lies in its ability to quickly and accurately find an optimal set of basic functions [13,14]. The algorithm infers the model from the data by combining variables and mathematical operators and generates an empirical formula which is a mathematical equation that predicts observed results derived from conducted experiments. GP combines previous equations and forms new ones. Thus it produces models with interpretable structure, relating to input and output variables from a data set without pre-processing and identifying critical parameters and hence shed insight into the underlying processes involved in a given system [15]. Symbolic regression can recognise and model complex non-linear relationships between the inputs and outputs of biological processes even in the presence of disturbances and potential for parallel processing. The preliminary data generated from experiments during rooting of in vitro regenerated plantlets in Wrightia tinctoria was employed to study the utility of symbolic regression to analyze plant tissue culture data. The effect of two variables - NAA and charcoal on root proliferation was considered. The datasets were subjected to usual statistical analysis as well as observation based modeling via symbolic regression. Moreover, we aimed to optimise the process by examining the influencing factors. We propound the use of symbolic regression-based model prediction as an addendum to data analysis method for plant tissue culture experiments.
The genetic variability was kept minimum by using a single field grown ortet, thus minimising statistical errors [16]. Nodal regions derived from the fresh flushes of growth from the ortet, two weeks after lopping one major branch served as the explants [17]. The nodal explants were conditioned over a period of 4 months (subculture/four weeks) on MS medium (1962) [18], pH 5.8 and 2 μM each of BAP and NAA for shoot multiplication. For rooting experiments individual shoots were transferred on MS medium containing 2 μM BAP with NAA (2, 4 and 6 μM, respectively) and charcoal (0.01, 0.03, 0.05, 0.07, 0.09 and 0.11%, respectively) in 250 ml culture flasks in 50 ml of sterilized medium (pH 5.8). The cultures were maintained at 25±2°C in a culture room with 40 μmolm−2 s−1 irradiances and a photoperiod of 8 hrs with 55±5% of relative humidity.
The plant tissue culture database, containing 21 conditions, followed a factorial design for two variables- concentration of NAA (2, 4 and 6 μM) and charcoal (0,0.01, 0.03,0.05,0.07,0.09 and 0.11%) in the medium. Each treatment consisted of 5-7 explants in a culture flask with three replicates. The subculture was done at the end of four weeks and five parameters were recorded to analyze the effects of the variables on rooting such as basal callus diameter (mm) (BC), the percentage of shoots rooted (R), the length of the longest root (cm) (RL), the number of roots (NR) and the number of lateral roots (NLR) (S1).
All experiments were conducted using Randomised Block Design (RBD). Continuous data were analysed using multiple linear regression in R and posthoc comparisons of pairs performed by Tukey's test (p>0.05). Count data were analysed using Poisson regression model. Pearson's Chi-squared test for count data was employed to access statistical significance of the variables.
Each of the observed parameters is modeled as a function of NAA and charcoal concentrations using symbolic regression and GLM for comparison. To obtain a global optimum, we have also modelled the combination (R+RL+NR+NRL-BC) by taking rooting factors together after normalisation by employing both GLM and symbolic regression. The optimum model for each case was generated by genetic programming based symbolic regression using the software package Eureqa (Version 0.98 beta) with 50% of the data randomly selected as training data, and 3-fold cross-validated with randomly selected 25% of the remaining data [19–21]. Corresponding to each symbolic regression model of the data partition, we have also obtained generalised linear model by including x, y, xy, 1/x, 1/y, sin(x), cos(x), sin(y), cos(y), xy sin(x), xy cos(x), xy sin(y), xy cos(y) into the set of basic functions and cross-validated similarly. The remaining 25% the data was used for testing and reporting error [19–21]. The Target expression used to generate the regression model was the minimal equation z = f(x, y) where ‘x' corresponds to NAA concentration and ‘y' corresponds to charcoal concentrations, and ‘z' represents each of the five observed parameters and their combination. The models were based on the primary and trigonometric building blocks, with the R2 goodness of fit as the error metric [22,23]. Root Mean Squared Error (RMSE) was calculated for the test data sets. Sensitivity represents the relative impact of the variable on the parameter studied within this model and was calculated by the local method using the partial derivatives [24]. Given a model equation of the form z = f(x, y), the influence metrics of x on z was;
Sensitivity=|∂z∂x|¯.σ(x)σ(z),evaluatedatallinputdatapoints;
The percentage positive was calculated as percentage of data points where σ(x)σ(z)>0 and percentage negative was calculated as percentage of data points where σ(x)σ(z)<0; where ∂z∂x was the partial derivative of z with respect to x, σ(x) was the standard deviation of x in the input data, σ(z) was the standard deviation of z, |x| denoted the absolute value of x, and x¯ denoted the mean of x [25]. The ‘fmin' function in MATLAB (R2012b) was used to obtain the maximum value of each of these functions.
The average values obtained for the five growth parameters observed during the study were given as the basal callus diameter (Table 1), the percentage of shoots rooted (Table 2), the length of longest roots (Table 3) and the number of roots and the number of lateral roots (Table 4). The miniscule alphabets within a column indicated the significant influence of charcoal and majuscule alphabets in the row represented the significant interaction of NAA. The shoots inoculated on MS medium with 0% charcoal (control) showed maximum basal callus formation (Fig 1). The shoots inoculated on MS medium supplemented with 4μM NAA and 0.07% charcoal showed the maximum percentage of rooting (Fig 2).
Multiple linear regression demonstrated a significant effect of NAA and its interaction with charcoal on basal callus (p>0.001), the percentage of shoots rooted (p>0.05) and root length (p>0.01) (Tables 1–3). The individualistic effect of NAA for the number of roots and lateral roots were found to be significant at p>0.05 and p>0.001 respectively (Table 4). The interaction of NAA and charcoal was not significant for the same parameters studied. Mathematical functions were successfully developed using symbolic regression to understand the correlation between the two variables for each of the parameters considered and is contrasted with those obtained by traditional regression models (Table 5). To analyze the effect of each of the variables on the parameter studied; variable sensitivity measures were calculated along with its percentage impact. Its sensitivity denoted the relative impact within this model that a variable has on the target variable. The individualistic effect of the two input variables on the output parameter was pointed out as percentage positive or negative of that input variable (Table 6). For the parameter basal calli diameter, the percentage positive value for variable ‘y' was zero. In other words, there was zero percent chance of basal calli mass increasing with increasing concentration of charcoal; or that basal calli mass decrease with increasing concentration of charcoal (Fig 3). The model predicted that increase in charcoal concentration had a consequent increase in root length and root number in 50% of all the trials while the same promoted rooting percentage and lateral root number in 75% of the trials. Root number and root length decreased with increasing concentration of NAA in 100% of the trials. Rooting percentage and lateral root numbers increased with increasing NAA concentration in 50% of all the trials. The function obtained and the 3D plots thus generated could be used to predict the combinations of input variables giving optimum results. The best response for rooting percentage was predicted at 3.7 μM NAA and 0.08% charcoal (Fig 4). The root length showed a non-linear pattern, and the highest value for its function was estimated with 2.8 μM NAA and 0.05% charcoal (Fig 5). The maximum root number was determined for 1.7 μM NAA and 0.06% charcoal (Fig 6). The maximum value of the function generated for lateral root number was with 6.3 μM NAA and 0.08% charcoal (Fig 7). The global optimum modelled upon the combination (R+RL+NR+NRL-BC) indicated the results as 2.44 μM NAA and 0.03% charcoal (Fig 8).
The conclusion obtained by traditional statistics suggested that charcoal had a positive and stimulatory effect in rooting of shoots by reducing basal callus (Table 1). Percentage of shoots rooted and root length showed a significant impact with the combination of NAA and charcoal (Tables 2 and 3). In the present study, NAA has a significant effect on rooting as shown by the number of roots and lateral roots (Table 4). Similar results were reported in Acacia leucophloea and Cinnamomum verum [26, 27]. With traditional statistics, we were not able to estimate the combination/s of both variables in producing the best results or able to identify the relative impact of a particular variable on the output parameter. Modeling of plant tissue culture data is practised using regression analysis where first an initial function is approximated and the data fitted to that function to obtain the optimum parameters [11,28,29]. In this procedure even when one gets the optimum parameter values, the model prediction was limited by the probable wrong selection of the model function. In contrast, symbolic regression procedures work simultaneously on model specification problem and the problem of fitting coefficients [30]. Thus it provides both optimum model function as well as the optimum variable values in the model. The simple relations derived from GP were more accessible to analyze the relationships between the input and output variables [31]. Observation-based predictive models using GP identified that the individualistic effect of charcoal was significant in all the output parameters. A previous investigation suggested basal callus mass formation as one of the primary constraints in the culture of this tree species [32]. In the present study, charcoal has a positive and stimulatory effect in rooting by reducing basal callus formation in shoots. For each of the functions, generated values can be obtained by increasing /decreasing the variables by a unit. After randomly testing the higher and lower limit of the additives with experiments, the magnitude of the observed parameters can be presumed at any concentration of the additives within this range using the models generated. It can be extended to analyze synergistic interactions between two parameters by testing whether increasing both variables by a unit, gives a higher or a lower value than the sum of the values obtained by increasing each individually by a unit. The basic requirement for any empirical model includes interpretability, robustness and reliability [33]. Symbolic regression gave comparably lesser RMSE values in comparison to multiple linear regression, thus adding validity to its use. In plant tissue culture obtaining an optimum model is crucial when one needs to find the optimum experimental parameters for large-scale production. The procedure adopted in the work can also be extended to similar experiments as it is general and computationally efficient. The analysis predicted the optimum concentration of medium for micropropagation of the selected tree species from the model plots derived from the preliminary experimental data. The study indicated that these models would have significant potential for saving time and expenditure in plant tissue culture laboratories for the commercial establishment of in vitro protocols in tree species.
|
10.1371/journal.pntd.0002252 | Geographic Distribution, Age Pattern and Sites of Lesions in a Cohort of Buruli Ulcer Patients from the Mapé Basin of Cameroon | Buruli ulcer (BU), a neglected tropical disease of the skin, caused by Mycobacterium ulcerans, occurs most frequently in children in West Africa. Risk factors for BU include proximity to slow flowing water, poor wound care and not wearing protective clothing. Man-made alterations of the environment have been suggested to lead to increased BU incidence. M. ulcerans DNA has been detected in the environment, water bugs and recently also in mosquitoes. Despite these findings, the mode of transmission of BU remains poorly understood and both transmission by insects or direct inoculation from contaminated environment have been suggested. Here, we investigated the BU epidemiology in the Mapé basin of Cameroon where the damming of the Mapé River since 1988 is believed to have increased the incidence of BU. Through a house-by-house survey in spring 2010, which also examined the local population for leprosy and yaws, and continued surveillance thereafter, we identified, till June 2012, altogether 88 RT-PCR positive cases of BU. We found that the age adjusted cumulative incidence of BU was highest in young teenagers and in individuals above the age of 50 and that very young children (<5) were underrepresented among cases. BU lesions clustered around the ankles and at the back of the elbows. This pattern neither matches any of the published mosquito biting site patterns, nor the published distribution of small skin injuries in children, where lesions on the knees are much more frequent. The option of multiple modes of transmission should thus be considered. Analyzing the geographic distribution of cases in the Mapé Dam area revealed a closer association with the Mbam River than with the artificial lake.
| Buruli ulcer (BU) is an infectious disease caused by Mycobacterium ulcerans that is affecting mostly children in endemic areas of West Africa. Proximity to slow flowing water is a risk factor, but the exact mode of transmission of BU remains unclear. Man-made environmental changes, such as sand mining, damming of rivers and irrigation have been implicated with increases in disease incidence. Here, we report findings from a survey for BU and continued case detection thereafter in the Bankim Health District of Cameroon. In this area, the local population believed that the damming of the Mapé River has led to the emergence of BU. In 28 months we identified 88 laboratory confirmed cases of BU. Studying these cases, we found that the age adjusted cumulative incidence of BU in the elderly is similar to that in children and that the distribution pattern of BU lesions neither matches mosquito biting patterns nor the distribution of small skin injuries. Multiple modes of transmission should therefore be considered. Our data further showed that the patients appear to have closer contact to the local Mbam River than to the artificial Mapé dam reservoir.
| Buruli ulcer (BU), a neglected tropical disease (NTD) of the skin, is caused by Mycobacterium ulcerans [1] and if untreated, can lead to disability. Worldwide, local BU incidence rates are highest in West Africa and Australia, where the classical lineage of M. ulcerans is found [2]–[4] and the disease occurs at different foci in the endemic countries. Both sexes can be affected by the disease and although individuals of all ages can get BU, most of the patients are less then 15 years old [5]. In Cameroon, BU was first described in 1969 in the Nyong river valley where during a cross-sectional survey in 2001, a total of 436 clinically diagnosed cases of active or inactive BU were found [6]. Since then, the Bankim Health District (HD) has been identified as an additional BU endemic area in Cameroon [7]. In this area, where our research has been carried out, the local population suspects that the creation of an artificial lake, by damming of the Mapé River in 1988, has led to an increase in BU incidence. Risk factors for BU include proximity to slow flowing water, poor wound care and not wearing protective clothing [8]. However, the exact mode of transmission has not yet been elucidated [9], [10]. Clinically, BU presents with symptoms ranging from nodules, plaques and oedemas to ulcers [11]. The cytotoxic and immunosuppressive toxin, mycolactone, uniquely produced by M. ulcerans, is believed to account for most of the pathology of BU [12]. The severity of cases is classified into three categories, with ‘1’ being patients with small (≤5 cm dimeter) lesions, ‘2’ patients with medium size lesions (5–15 cm) and ‘3’ being patients with large (>15 cm) lesions, multiple lesions or lesions at critical sites [13]. Many BU cases identified in rural areas are still diagnosed based on clinical symptoms only, although the use of laboratory diagnosis is highly recommended by the World Health Organization (WHO). In 2004, the WHO introduced the use of the combination of streptomycin and rifampicin given daily for 8 weeks as treatment [14]. However, surgery and wound management remain critical aspects of BU care [15], [16].
During our investigations of BU in the Bankim HD, we also examined the local population for two other NTDs of the skin, namely yaws and leprosy. Yaws is caused by Treponema pallidum (T. pallidum) subspecies pertunu, and is transmitted through skin and mucous membrane contact [17], [18]. After an initial single lesion, the disease progresses to secondary multiple lesions and in about 10% of cases it causes permanent disability [18]. Leprosy is caused by Mycobacterium leprae, which is believed to be transmitted by the respiratory route and can cause major disabilities through nerve damage. Diagnosis of yaws and leprosy relies mainly on physical examinations [17], [19] and treatment of both diseases is feasible with antibiotics [17], [20].
The objectives of the present study were i) to conduct an exhaustive survey for BU, yaws and leprosy in the Bankim HD; ii) to continuously monitor the occurrence of BU in the Mapé Dam area; and iii) to examine the age distribution, geographic origin and distribution of lesions of the real-time polymerase chain reaction (RT-PCR) confirmed cases of BU to underpin future environmental and social science studies.
Approval for the survey and the subsequent continuous enrolment of cases was obtained from the Cameroon National Ethics Committee (N°041/CNE/DNM/09 and N°172/CNE/SE/2011) and the Ethics Committee of Basel (EKBB, reference no. 53/11). Participation in all aspects of the study was voluntary and all patients, independent of their study participation, were treated according to national treatment standards. All clinically confirmed cases who participated in the study provided written informed consent.
The study was conducted in the Mapé Dam region of Cameroon (Figure 1) at two different geographical scales. The initial phase of the study was conducted in the Bankim HD which consists of seven Health Areas (HA): Atta; Songkolong, Somié, Nyamboya, Bandam, Bankim Urban and Bankim Rural. The health care infrastructure of the Bankim HD consists of one public district hospital, six primary and four private health centres (HC). All of these facilities employ two medical doctors and approximately 30 nurses. For the later part of the study, bordering regions in the 4 HD surrounding the Bankim HD (Nwa HD, Malantouen HD, Mayo Darle HD, Yoke HD) were also included in the study area. The main environmental features of the area are the Mapé Dam and the Mbam River.
In early 2010 (March 22 to April 19), we conducted an exhaustive cross-sectional house-by-house survey for BU, leprosy and yaws in the 88 villages of the Bankim HD (Figure 1). Eleven teams of three trained field workers, namely one local nurse and two local community relays, were employed to interview all inhabitants. Field workers were trained for two days on the use of the questionnaire and the clinical signs of the three diseases investigated. At the household level, demographic information of all inhabitants was collected and posters with photographs of the clinical presentations of the diseases were shown. Households with suspected cases were re-visited by staff with extensive experience in the diagnosis of BU, leprosy and yaws. From clinically confirmed BU cases, samples were collected for laboratory confirmation as follows. Two or three dry swabs were collected from ulcerative lesions or a fine needle aspirate (FNA) was drawn from non-ulcerative lesions [21]. To facilitate handling of FNA samples, they were transferred onto a swab.
Following the survey, we continued to monitor the occurrence of all new cases of BU in the Bankim HD by community and HC based case referral and regular supervision until the end of June 2012. For this, a health worker, trained and experienced in the diagnosis of BU, regularly visited all HC in the Bankim HD and areas of the adjacent Malantouen HD. During these visits, suspected cases who independently came forward or who were referred to the HC by community or family members, were evaluated and if clinically confirmed, asked to come for treatment. Before treatment, swabs or an FNA were collected for laboratory confirmation as described above. In addition to demographic and clinical information, the houses where the patients lived for at least a year before disease onset were mapped using a GPS device. From the GPS device, coordinates were only recorded once the GPS receiver showed an accuracy of below 10 m. Details of the location of the lesions on the patient's bodies were also collected and documented by photographs. Both clinically confirmed BU cases identified in the survey and during the continuous case detection were included in the cohort of patients investigated here.
Samples were locally stored at 4°C before transport to the laboratory where definite BU diagnosis was obtained by insertion sequence (IS) 2404 RT-PCR. Analysis was done according to the protocol developed by Fyfe et al. [22], [23]. In brief, swabs were transferred into glass bottles containing glass beads with 2–5 mL of PBS, and the bottle vortexed for 1.5 minutes. From 1 mL of the solution, DNA was extracted and RT-PCR performed. DNA was amplified in a StepOnePlus Real-Time PCR System (Applied Biosystems) and data analyzed using the Applied Biosystems StepOne Software (2.2.2).
Using published age specific relative body surface areas (RBSA) [24] and the number of patients in each of the age groups, the weighted average RBSAs of a model person (all ages), a model child (<15), and a model adult (≥15) were computed. If required to perform a Fisher's exact test, RBSA were converted to counts which add up to the observed number of lesions. The shape file used to analyze lesion localizations is found in Dataset S1.
Continuous variables were summarized as means and standard deviation or medians and interquartile ranges and categorical ones as counts and percentages. The Fisher's exact or Chi-squared tests were used to compare categorical characteristics between groups and Student t-tests or Mann-Whitney U-test in the case of continuous variables. Multiple comparisons were adjusted for using a Bonferroni correction. The software, SAS (SAS Institute, Cary, USA; release 9.3), RStudio (RStudio, Boston, USA, version 0.95.262) and R (The R Foundation for Statistical Computing; version 2.15.1) were used to perform the statistical analysis. Geographic data and the localisation of lesions were analyzed with ArcGIS ArcMap (Economic and Social Research Institute, Redlands, USA; version 10.0).
In the course of the survey, a total of 48 962 individuals in 9 344 households (Figure 1) were interviewed. The population of one village (approximately 550 people) refused to participate in the study. Assessing demographics and living conditions in the Bankim HD (Table 1), we found that the local population is very young with an average age of 19.3±17.0 (median = 14.0, interquartile range = 6.0 to 28.0), that 51.4% of the population are women and that overall, 61.2% of the population have attended school at some point in their lives. We further observed that Christianity is the most common (64.9%) religion and that, apart from the young members of society which are either students (32.2%) or children (23.5%), the most common professions in the district are farming (16.9%) and household work (17.4%). In terms of living conditions we found that there are on average 5.2 individuals living in each household and 26.8% of the households have a mosquito net. Further, our data showed that only 38.3% of the population have access to clean drinking water that comes at least from a fortified well and that the roofs and floors of the local houses are often very poorly constructed. Table 1 also shows that, the main local differences in the level of development in the HD exist between the six rural HA and the Bankim Urban HA (BA HA), which includes the town of Bankim (77.4% of the BA HA population) and nine small settlements around it. The higher level of development in the BA HA is reflected by the significantly higher percentage of people having gone to school at some point in their lives (p-value<0.0001) or by the significantly better access to clean drinking water (p-value<0.0001). Furthermore, in the BA HA the proportion of houses with better flooring (p-value<0.0001) and walls (p-value = 0.0037) is also significantly higher compared to the other HA.
In the survey, we identified 32 cases of leprosy, 29 cases of yaws and 25 cases of BU based on clinical symptoms. With 32 cases of leprosy, the population-adjusted prevalence was at 6.5 cases per 10'000. The majority (70%) of the identified leprosy cases suffered from the multibacillary form of the disease and 22% of them were previously known but had abandoned their treatment and needed treatment re-initiation. Of the 29 yaws cases identified, 28% presented with the advanced symptom of hyperkeratosis and of the BU cases, 23% (6 cases) could be re-confirmed by RT-PCR.
In the five months after the survey (April 2010 to August 2010), only two new RT-PCR reconfirmed BU cases were identified (Figure 2). Following this lag, between September 2010 and June 2012 (22 months) there was a steady flow of about 2.5 new RT-PCR confirmed BU cases per month from the Bankim HD. During this period, RT-PCR confirmed BU patients from the surrounding HDs (about 1.2 per month) also reported to BU treatment facilities in the Bankim HD. Overall, our study identified 157 clinically confirmed cases of BU of which 88 (56%) could be confirmed by RT-PCR. Of the non-confirmed patients, 48 (31%) tested negative in RT-PCR and of 21 patients (13%) no samples were collected. Gender ratio, age distribution, average disease duration prior to consultation, and proportion of category 3 cases were comparable between the RT-PCR positive and negative patients. Only age differed significantly (p-value = 0.034) between the RT-PCR confirmed and the non-confirmed cases with the average age of the confirmed cases being 21.2 and that of the non-confirmed ones being 29.3. To ensure the reliability of our conclusions we focused the remaining analysis only on the 88 RT-PCR confirmed BU cases. Age distribution (p-value = 0.4754) and professions (p-value = 0.5161) did not differ significantly between the population of the Bankim HD and the confirmed BU patients. The gender distribution among patients was moderately different (p-value = 0.061) from that of the overall population with a larger proportion of males among the confirmed BU cases.
To better describe BU epidemiology in the Mapé basin we set out to identify the exact geographic origin of all 88 laboratory confirmed cases in our cohort. Based on information from the patients or their close relatives we were able to determine the HD of origin for 86 (98%) of the cases (Figure S1). For the remaining 2 cases we could only determine that they did not live in the Bankim HD for the year before the onset of symptoms, but we could not conclusively determine which HD they were from. Studying the distribution of cases by HD, we found that the proportion of category 1 cases among the patients originating from inside the Bankim HD (24/62) was significantly higher (p-value = 0.039) compared to the cases from the surrounding HDs (5/26; Figure S1). For the 62 cases that originated from within the Bankim HD we were also able to determine their HA of origin. Using the population data as collected by the survey, we were then able to calculate the cumulative incidence rate of BU per HA in the Bankim HD during our study. As shown in Figure 3A, the cumulative incidence rate of BU in the Bankim HD is highest in the BR HA (5.08/1'000). The cumulative incidence rate in this HA is significantly higher compared to all other HA in the HD (p-value<0.001). Interestingly the cumulative incidence rate is also significantly higher in the southern HA (BR, BA, BD, NY) compared to the northern HA (AT, SO, SG) of the Bankim HD (p-value<0.001) (Figure 3A). Finally, for more detailed spatial analysis, the exact domiciles of 79 (89.8%) of the confirmed BU cases were mapped (Figure 3B). For 7 of the remaining cases (Bankim HD: 3 from the Bandam HA, 1 from the Somié HA; surrounding HD: 1 from the Malantouen HD, and 2 of unknown origin) we could not conclusively identify the exact house where they lived before the onset of BU. An additional two cases (1 each from the Nwa HD and Mayo Darle HD) are not considered in the analysis because they originated from outside of the Mapé basin. Based on the known exact origin of the cases that came from within the Bankim HD (n = 58) and who were therefore identified by the same case finding strategy, a Kernel function was used to compute the density of BU in the Bankim HD (Figure 3B). This BU density map shows that most of the cases occur in the southern part of the Bankim HD, particularly along the Mbam River and in the area between the Mapé Dam reservoir and that river. The exact origins of 80.8% (21/26) of the BU cases from outside of the Bankim HD, indicate that the local BU focus expands outside of the Bankim HD, in particular westwards into the Malantouen HD (Figure 3B).
The median age of the 88 RT-PCR confirmed cases was 12.5 (interquartile range = 8.0 to 30.0). The age of patients ranged from 0.5 to 73, 52 out of 88 (59.1%) were children (age <15) and 11 (12.5%) were older than 50. The gender ratio of all cases was 1.44 male/female. In children this ratio was 1.89, in the 15 to 50 year olds it was 0.79 and in the above 50 year olds, it was 1.75. The age dependent variation in the gender ratio was not statistically significant (p-value = 0.20).
With the ages of the 62 (70.5%) cases of BU which originated from within the Bankim HD and the population age distribution as collected in the survey (Figure 4A and Table 1), we computed the age adjusted cumulative incidence rate of BU in the Bankim HD for the period of the study. As shown in Figure 4B, we observed a low age adjusted cumulative incidence rate of BU in individuals aged below four years. The rate then peaked in children aged between four and <14 years of age, with the 12 to <14 year olds particularly affected (34.4 cases per 10'000 inhabitants). Interestingly, the age adjusted cumulative incidence rate peaks again in the over 50 year olds (27.0 cases per 10'000 inhabitants; Figure 4B).
In the laboratory confirmed BU patients studied here, 49/88 (55.7%) lesions occurred on the lower limbs, 27/88 (30.7%) on the upper limbs, 2/88 (2.3%) on the head and neck and 10/88 (11.4%) on the trunk. One of the trunk lesions occurred on the genitals. Two patients had multiple lesions and only the initial lesion was considered for analysis. The distribution of lesions differed significantly (p-value<0.001) from the relative body surface area (RBSA; Table S1). Interestingly, most of the lesions (52.3%) occurred in close proximity to joints with clusters around the ankles (19.2%) and elbows (15.9%; Figure 5A, 5B and Table S2). When analyzing the occurrence of lesions on the different body parts, we did not observe any statistically significant difference between lesions occurring on the right or left or front or back of the patients. However, when analyzing the occurrence of lesions on the joints, we did observe a statistically significant difference between the lesions on the front or back of the joints (p-value = 0.012), in particular there was a significant difference between the occurrence of lesion on the front or back of the elbow (p-value = 0.005). No such difference was observed when analyzing the joint lesions on the right of left of the patients' bodies. Analyzing the distribution of the lesions by body part, we found a moderately significant difference between males and females (p-value = 0.076) with the percentage of lesions on the trunk being significantly higher (p-value = 0.033) in males (Table S1).
The distribution of lesions by body parts in children (Figure 5C and Table S1) was significantly different (p-value = 0.009) from the RBSA of children. Interestingly, only children (n = 2) had lesions on the head and neck. Overall, the lesions appear more dispersed in children (Figure 5C and 5D). While, the difference in the general lesion distribution by body parts between adults and children (Table S1) was not statistically significant (p-value = 0.154), there was a significant difference (p-value = 0.011) between the distribution of lesions at joints in children and adults. In particular, most joint lesions in adults occurred at the ankle (36.1%), whereas most joint lesions in children occurred at the elbow (19.2%). Finally in adults, lesions occurred mainly on the front and back of the feet and the distribution also differed significantly (p-value = 0.004) from what is expected based on the RBSA of adults (Figure 5C).
The 2011 Cameroon Demographic Health Survey (DHS) examined approximately 22'000 adults (>14 years old) and found that the population is very young with roughly 24% being 15–19 years old [25]. In the Bankim HD, we also observed a population that is strongly skewed towards young individuals and we found that living conditions and access to clean drinking water are very poor. Given the basic health infrastructure, these factors pose big challenges when addressing any health care related issues [26], [27].
Although Cameroon has achieved nationwide leprosy elimination as defined by the WHO (<1 case per 10'000 inhabitants) [28], our data showed that leprosy remains endemic in the Bankim HD. The substantial proportion of leprosy patients that had previously abandoned treatment further demonstrated that the oral treatment regimen requires better patient monitoring to achieve good compliance. In 2010 Cameroon reported 800 cases of yaws [29] and our survey confirmed that the Bankim HD is a yaws focus. Studies on the use of oral antibiotics have again raised hope for the eradication of yaws [30]–[32]. However, until eradication is possible, the focus of leprosy and yaws care should be early detection, complete cure and prevention of disabilities. To achieve this, front line medical staff needs to be trained on clinical diagnosis and efficient case management.
Characteristics of BU and the remote areas where it occurs have been suggested to necessitate active case searches for early case detection [33], [34]. Indeed, house-by-house surveys have helped to elucidate BU epidemiology in Ghana and Ivory Coast [35], [36]. In Cameroon a study around the Nyong River, identified 135 PCR confirmed cases of BU [6]. In the survey described here, the number of RT-PCR confirmed BU cases identified was smaller than expected. However, the lag of new cases during the first months after the survey indicated that the survey identified the cases present at that time. It cannot be excluded however, that a proportion of BU patients seeks to avoid contact with the formal health system. By continuous case detection, also accounting for the trust needed for cases to come forward [37], we identified 157 clinically diagnosed cases of BU (from April 2010 to June 2012). To increase validity of the findings [38], [39], our analysis focused on the 88 (56.1%) RT-PCR confirmed cases among them. False negative RT-PCR results are unlikely since we nalysed multiple samples from each patient (data not shown). Although accurate BU clinical diagnosis is possible [40], misdiagnosis rates of up to 40% have been reported emphasizing the pressing need for a point-of-care rapid diagnostic test [5], [38], [39], [41]–[43].
Based on the number of BU cases in each of the HA in the Bankim HD and their respective populations, the BR HA was determined to have the highest cumulative incidence rate of BU in the Bankim HD. Furthermore, by detailed mapping of cases and through the use of a geographic information system (GIS), we identified hot-spots of BU transmission along the Mbam River. With only few cases living in the immediate proximity of only the Mapé Dam reservoir, our data does not support the suspected direct importance of this man-made lake. This does not exclude that environmental changes associated with the damming of the Mapé River may have had a more indirect effect on the spread of BU in the wider area. Whether the relatively large proportion of patients living in the town of Bankim (11/79 GPS mapped cases), contracted BU there, remains to be investigated. By also mapping cases from outside of the Bankim HD, we found that the local BU endemic area is larger than previously described [7]. Indeed it is possible that, because of the differences in case finding strategy inside and outside of the Bankim HD, our findings from outside the HD under represent the true degree of BU endemicity in the areas surrounding the Bankim HD. Further studies are therefore needed to investigate BU endemicity in the entire Mapé basin in more detail. Ongoing environmental and social science research at the identified hot-spots of disease is aiming to further elucidate the mode of transmission of BU.
BU affects individuals of all ages [15], [34] but in the African endemic regions most patients are children [9], [44]. However, when adjusting for the population age distribution, studies in Benin [45] and in Australia [46] showed that 75 to 79 year olds or the ≥74 olds, respectively, have the highest risk of contracting BU. Our data similarly showed that the age adjusted risk of BU is as high in the >50 year olds as in children, a trend possibly associated with immunosenescence, the gradual deterioration of the immune system associated with natural age advancement [46], [47]. It is interesting to note that cases among very young (<5) children, which make up an even larger part of society than the 5–10 year olds, are relatively rare. This may indicate that compared to older children the very young children are less exposed to risk factors due to a smaller movement radius away from the house [10]. In the exposed individuals, host factors are likely to contribute to the degree of susceptibility [48]; seroepidemiological studies indicate that only a small proportion of exposed individuals develop clinical disease [49], [50].
Detection of M. ulcerans DNA-positive mosquitoes in an Australian BU focus [46] as well as identification of the failure to wear protective clothing as a risk factor and of the use of mosquito repellent as a protective factor for BU [8], support the hypothesis that insects are involved in M. ulcerans transmission [10]. Most biting arthropods selectively feed at specifics sites based on visual, physical or chemical cues such as distance of the ground, breath and skin temperature of the bait [51]–[55]. The resulting feeding patterns are often focused either on the feet and ankles or the head of the human subject [52]. Interestingly for vector transmitted parasitic diseases with local manifestations such as cutaneous leishmaniasis and filariasis, it has been found that the lesion distribution correlates with the biting sites of the responsible vectors [56], [57]. BU lesions occur mostly on the lower limbs [15], [45], [58]–[60] and in adults, a focus on joints, specifically the elbows and ankles, has been reported [15], [58]. Studies on the distribution of lesions also show that they are usually equally distributed between the left and right side of the body and compared to adults, children tend to have more lesions on the trunk [45], [60]. Using GIS methodology we observed in this study that lesions cluster at specific locations on the limbs. We found that, particularly in adults, lesions occur mostly at locations where the skin is not commonly protected with clothing. As previously described, in females, which are more likely to cover their upper body with clothing, we found that there are less lesions on the trunk. In rural African villages children may often have their upper body exposed explaining the more dispersed distribution of their lesions.
Detection of M. ulcerans DNA in the environment [10] and identification of poor wound care and failure to wear protective clothing as risk factors for BU [8] have led investigators to speculate that transmission may alternatively occur by skin trauma and direct contact with M. ulcerans contaminated environment [10]. A study in Canadian children found that children 9 months and older have on average >3.5 recent skin injuries [61]. In 5 to 17 year olds injuries most often occur where the bones are close to the skin, i.e. at shins, knees, elbows and forearms. Injuries on the head were most common in children less than 5 years of age and lesions on genitals were rare in all ages [61]. While this study may have, due to differences in dress code and activities, limited relevance for Cameroonian children, it is remarkable that both BU lesions on the head in our cohort occurred in patients under the age of 5 [61]. While BU lesion distribution does not seem to correlate closely with the published distribution of insect bites, inoculation of skin injuries by a contaminated environmental source should lead, for example, to more lesions on the knees. Based on these data, the option of multiple modes of transmission should be considered.
|
10.1371/journal.pgen.1000584 | EPHA2 Is Associated with Age-Related Cortical Cataract in Mice and Humans | Age-related cataract is a major cause of blindness worldwide, and cortical cataract is the second most prevalent type of age-related cataract. Although a significant fraction of age-related cataract is heritable, the genetic basis remains to be elucidated. We report that homozygous deletion of Epha2 in two independent strains of mice developed progressive cortical cataract. Retroillumination revealed development of cortical vacuoles at one month of age; visible cataract appeared around three months, which progressed to mature cataract by six months. EPHA2 protein expression in the lens is spatially and temporally regulated. It is low in anterior epithelial cells, upregulated as the cells enter differentiation at the equator, strongly expressed in the cortical fiber cells, but absent in the nuclei. Deletion of Epha2 caused a significant increase in the expression of HSP25 (murine homologue of human HSP27) before the onset of cataract. The overexpressed HSP25 was in an underphosphorylated form, indicating excessive cellular stress and protein misfolding. The orthologous human EPHA2 gene on chromosome 1p36 was tested in three independent worldwide Caucasian populations for allelic association with cortical cataract. Common variants in EPHA2 were found that showed significant association with cortical cataract, and rs6678616 was the most significant in meta-analyses. In addition, we sequenced exons of EPHA2 in linked families and identified a new missense mutation, Arg721Gln, in the protein kinase domain that significantly alters EPHA2 functions in cellular and biochemical assays. Thus, converging evidence from humans and mice suggests that EPHA2 is important in maintaining lens clarity with age.
| Cataract is the leading cause of blindness. Cataract may form at any age, but the peak incidence is bimodal—in the perinatal period or later than 50 years of age. The early onset forms follow Mendelian inheritance patterns and are rare. Age-related cataract accounts for 18 million cases of blindness and 59 million cases of reduced vision worldwide. Among three types of age-related cataract, cortical cataract is known to be highly heritable, although few genes have been linked to its etiology. We report here that EPHA2 is associated with cortical cataract. EPHA2 is expressed in mouse and human cortical lens fiber cells, and homozygous deletion of Epha2 in two independent strains of mice led to development of cataract that progressed with age. Common and rare variants including a missense mutation in the EPHA2 gene were associated for cortical cataract in three different Caucasian populations. Our study identified EPHA2 as a gene for human age-related cataract and established Epha2 knockout mice as a model for progressive cortical cataract.
| Cataract is the leading cause of visual impairment worldwide, with approximately 37 million people affected, accounting for 48% of global blindness [1]. Cataract is defined as any opacity of the crystalline lens resulting from either alteration in lens cell structure or changes in lens proteins or both. Morphologic classification of age-related cataract can be divided into four categories: nuclear, cortical, posterior subcapsular cataract and mixed type. Among Caucasian Americans aged 65 years or older, the reported prevalence of cortical cataract was 24% [2]. Incidence of cataract surgery in the United States is highest amongst those over 70 years, with an annual increase of about 14% [3].
Age-related cataract has consistently been attributed to risk factors such as old age, female gender, diabetes, hypertension, smoking, UV light exposure, and heavy alcohol intake [4]. Recent studies showed excess clustering of disease in families [4]. Heritability for cortical cataract in female twins was estimated as high as 58%, and at least 11% of total variance was accounted for by age [5]. In segregation analysis in the Beaver Dam Eye Study (BDES), a single major gene accounted for 58% of the variability of age- and gender-adjusted measures of cortical cataract [6]. Genes involved in the oxidative stress pathway [7], hypergalactosemia [8], hyperferritinemia [9], diabetic complications [10], and neurodegenerative disease [11],[12], mediate development of juvenile or age-related cataract.
A genome-wide scan for cortical cataract after adjusting for covariates in a subset of the BDES showed evidence of linkage in 1p36 [13]. This region is one of the most replicated loci for inherited cataract including the Volkman (pulverulent) cataract from a large Danish pedigree [14], a type of posterior polar cataract [15], and a total congenital cataract from a six-generation Australian family [16]. The Eph receptor A2 gene (EPHA2) that resides under the linkage peak on 1p36 was selected as a candidate gene because knockout mice for Epha2 developed age-related cortical cataract. In this study, we extensively examined characteristics of the EPHA2 protein using knockout mice. We investigated allelic association by re-sequencing and genotyping of single nucleotide polymorphisms (SNPs) in coding and non-coding conserved regions in large population-based human studies.
During the course of characterizing mice with deletion of Epha2 gene, we observed development of progressive cortical cataract in a significant fraction of homozygous knockout mice. This line of knockout mice (Epha2−/−) with deletion of exons 6–17 was generated using a secretory gene trapping strategy [17],[18]. Initially noticed on FVB/NJ genetic background, visible bilateral lens opacity in Epha2−/− mice appeared by five months, which was confirmed as cataract by slit lamp examination (Figure 1A). The incidence and severity increased with age, affecting over 80% of mice by 12 months (Table 1). Inspection of dissected lens revealed significant lens opacity between three to four months, and mature cataracts with lens rupture occurred between six to eight months (Figure 1B and Figure S1A). Dense opacity initiated around the equator and progressed to involve the entire lens (Figure 1B). As early as one month after birth, prior to the onset of lens opacity, retroillumination examination revealed clusters of subcapsular vacuoles in the anterior cortex of the Epha2 knockout but not in heterozygous or wild-type lenses (Figure 1C), which are indicative of structural and osmotic alterations in the lens [19].
To further confirm a link between loss of Epha2 gene and cataractogenesis, we examined another Epha2 mutant mouse line on C57Bl/6 genetic background, where a retrovirus is inserted in the first intron, inactivating expression of Epha2 [20]. The homozygous mutant mice developed cataracts with similar morphological changes as the secretory gene trapping mice on FVB/NJ background described above. Thus, neither the strain genetic backgrounds nor the secretory trapping of partial EPHA2 ectodomain fused to β-gal is likely to be a major contributing factor in the lens phenotypes of the Epha2 mutant mice. Interestingly, in cohorts of mice subject to two-stage chemical skin carcinogenesis studies [17], the penetrance of cataract increased to 100% by six months, suggesting that similar to humans, environmental factors could significantly influence cataractogenesis in Epha2-null mice.
We found that EPHA2 was expressed in lens homogenates of wild type mice, and the level of expression progressively decreased with age (Figure 1D). The relative level of expression is compatible with other tissues known to have high levels of EPHA2 such as the skin [17]. Immunofluorescence analysis revealed that EPHA2 was most abundantly expressed in cortical lens fiber cells but not nuclear fiber cells (Figure 1E). In lens epithelial cells, EPHA2 expression was low in the anterior region and became upregulated when epithelial cells underwent differentiation at the lens equator (Figure 1E–1K). Much weaker staining was seen in the nuclei of the lens. To verify these findings, we separated lens cortex and nuclei from two week and five month old mice and subjected the homogenates to immunoblot. EPHA2 was highly expressed in cortical fiber cells at two weeks, which fell significantly by five months (Figure S1F). Histochemical staining for LacZ reporter gene cassette confirmed the elevated EPHA2 expression near the equator (Figure S1XPATH ERROR: unknown variable "checknextn".). Consistent with our observations in the mouse, EPHA2 was readily detectable in human lens fiber cells by immunoblot and immunohistochemistry (Figure S2).
Eph kinases bind to membrane-anchored ligands called ephrins and mediate cell contact-dependent bidirectional signaling [21],[22]. We found that ephrin-A1, a ligand for EPHA2, displayed a similar expression pattern with strong expression in lens fiber cells as well as lens epithelial cells near the equator (Figure 2A and 2B). The expression of EPHA2 and ephrin-A1 in the overlapping regions suggests EPHA2-ephrin-A1 interaction at these sites, which could have signaling and/or structural roles in the lens. Homozygous deletion of Epha2 gene led to significant alteration in ephrin-A1 localization (Figure 2C and 2D) concomitant with changes in lens structures (Figure 2D). Staining for expression of N-cadherin (Figure 1L and 1M), a known component of cell adhesion junctions in lens fiber cells, further confirmed structural defects in the mutant lens (Figure 2E and 2F).
Disruption of lens structure could be caused by cellular stresses. We found that HSP25, a small heat shock protein and the murine homologue of human HSP27, was significantly overexpressed in the Epha2 knockout lens from day two after birth, long before the onset of cataract (Figure 2G–2I). Expression of HSP90, on the other hand, was not affected. Prior to the onset of lens degeneration at old age (6 months and beyond), expression of major crystallins did not show significant change either (Figure S3A). HSP25 is closely related to and directly interacts with αB-crystallin, being expressed during oxidative stress in human cataractous lenses [21]. The overexpressed HSP25 in mutant lenses is in an underphosphorylated form that is known to form large aggregates with misfolded proteins (Figure 2J) [23],[24]. By trapping misfolded proteins in large aggregates, HSP25 could contribute to cataractogenesis. While the exact mechanisms of cataractogenesis remain to be fully elucidated, these results together suggest that Epha2 deletion probably causes elevated stress responses in the lens, which may lead to increased protein misfolding, cellular structural damages and eventual lens opacity.
The human EPHA2 gene is located on 1p36 where we and others had previously reported linkage with cortical and progressive juvenile onset cataract, respectively [13]–[16]. This, together with the Epha2 knockout mouse data described above, motivated us to investigate association between cortical cataract and EPHA2 in humans. For discovery, we used a population-based family sample from the US Beaver Dam Eye Study (BDES) cohort (Table 2), which was initially used for linkage analysis of cortical cataract [13]. To select individuals for re-sequencing, we identified families linked to markers on 1p36 from our previous set. We re-sequenced all 17 exons and intron-exon boundaries of EPHA2 in 34 individuals from the linked families (Table S1). We identified four new coding variants (Ser74Ser, Arg721Gln, Lys728Lys, and Ile779Ile) and one new non-coding variant (A/G in intron 12-13) (Table 3). These variants were rare (minor allele frequency MAF<1%) and have never been reported before. We also confirmed one non-coding insertion-deletion polymorphism (rs35519704: -/G) and seven other coding variants, including rs6678618 (Ala190Ala), rs6678616 (Leu191Leu), rs35484156 (Ser277Leu), rs2230597 (Pro329Pro), rs55655135 (Leu632Leu), rs10907223 (Leu661Leu), and rs3754334 (Ile958Ile). These polymorphisms were previously reported in public databases.
Among variants identified via re-sequencing, two non-synonymous variants (Arg721Gln and rs35484156) and two synonymous variants (Ile779Ile and rs6678616) were pursued by genotyping all available individuals (Table 4) in the discovery BDES set (494 families, N = 1401). The Arg721Gln variant demonstrated the strongest association (P values: cortical = 2×10−8, severe cortical = 8×10−5) in the BDES. We performed extensive quality control metrics for the final model containing Arg721Gln in the discovery dataset to confirm both the consistency and the validity of the association result (Figure S4). We did not discover any bias in the SNP association caused by other factors.
The risk allele for Arg721Gln was the rare ‘A’ allele (Table 5). We estimated that after adjusting for covariates the risk allele for Arg721Gln can cause a net increase in the cortical score of 9.2% under the dominant model (i.e. if an individual carries one ‘A’ allele for the Arg721Gln variant, the cortical cataract score increases by 9.2% compared to carrying ‘GG’ alleles). Since we did not have Arg721Gln ‘AA’ homozygotes in our data, we did not assess the effect size for a recessive model. Using the severe cortical cataract trait the association signal at Arg721Gln was less significant (Table 5), showing that smaller sample sizes led to reduced power.
The sequencing data for Arg721Gln confirmed that the risk allele ‘A’ segregates with disease in a BDES family (Figure 3). We were unable to investigate segregation for heterozygous individuals in the other families, because missing data caused these individuals to be singletons or unconnected. We also genotyped the Arg721Gln variant in two replication datasets (Table 2), the United Kingdom Twin Eye Study (UKTS; 185 individuals from 172 families) and the Australian Blue Mountains Eye Study comprising unrelated persons (BMES; N = 1470). The three datasets were comparable with respect to measurement of the trait and the way the data were defined for analysis (Figure S5). The frequency of the rare ‘A’ allele is 0.6%, 0% and 0.2% in the BDES, UKTS, and BMES, respectively. These frequencies are not significantly different in the BMES and UKTS (Table S2); increased frequency of the ‘A’ allele in the BDES was probably due to enrichment of the same risk allele in these families (i.e. a founder effect), and we believe that 0.1%–0.2% represents a true population estimate for this rare variant. The risk allele has 78% penetrance in heterozygous individuals whose age is 70 years or more. We were unable to test association for the Arg721Gln in the replication sets because either the rare allele was not present (UKTS) or its frequency was too small (BMES) to be used in the statistical models.
Functionally, the Arg721Gln mutation changes arginine, a positively charged amino acid, to glutamine. Examination of the EPHA2 kinase domain crystal structure [25] revealed that Arg721 in the αE helix forms a salt bridge with Asp872 in the αI helix (Figure 4A). Interestingly, both Arg721 and Asp872 are concordantly conserved among different members of human Eph kinases and among EPHA2 proteins across different species (Figure 4B and 4C). The Arg721Gln mutation may disrupt the salt bridge and affect its conformation and function. To test this possibility, wild type (WT) and Arg721Gln mutant EPHA2 were expressed in HEK 293 cells that have low endogenous EPHA2 [26]. Although the mutant receptor still remained responsive to ligand stimulation in both kinetic and dose-response studies (Figure 5A and 5B), it displayed significantly higher basal activation than WT-EPHA2 in the absence of ligand stimulation. Consistent with our previous report that activated EPHA2 inhibited the Ras/ERK1/2 signaling cascade [26], the higher basal activation of EPHA2 was correlated with dramatically reduce basal ERK1/2 activities compared to WT-EPHA2 or vector control, suggesting altered signaling by the mutant EPHA2. In a clonal growth assay, HEK 293 cells expressing Arg721Gln-EPHA2 were significantly growth-inhibited cells by ephrin-A1, whereas WT-EPHA2 expressing cells were refractory (Figure 5C). In addition, we observed stochastic intracellular retention of the Arg721Gln mutant EPHA2, but not WT-EPHA2 when expressed in mouse embryonic fibroblast (MEF) cells derived from Epha2−/− embryos, affecting about 40% of total cell populations (Figure 5D). The effects were cell type-specific, as they were not observed in HEK 293 cells. While the molecular mechanism for the cytosolic retention is unclear at present, the data in aggregate demonstrate that the Arg721Gln mutation significantly alters EPHA2 signaling and cellular regulation in vitro. Because EPHA2 is essential in maintaining lens clarity in the mouse knockout model (Figure 1), it is possible that such perturbation of EPHA2 functions by the Arg721Gln mutation can predispose the carrier to cataractogenesis. Further studies, including knock-in of the mutant allele in mice, will be necessary to determine if the mutation is sufficient to predispose to lens opacity in vivo.
In order to find common variants associated with disease, we selected 15 SNPs based on tagging parameters and phylogenetic conservation (Figure 6A–6C) in addition to 4 coding variants identified via re-sequencing, resulting in 19 additional SNPs that covered at least 90% of the gene (Table 4). We genotyped these SNPs in three worldwide Caucasian populations (Table S2). Genotyping in all three datasets demonstrated that the EPHA2 gene contained at least two blocks of linkage disequilibrium (LD), one at the 5′ and another at the 3′ end that were relatively independent; the LD in the BDES, BMES, and UKTS was very similar to that in the CEU HapMap sample (Figure S6). The markers within each block are highly correlated, while those between the 5′ and 3′ blocks show low correlation (Figure S6). The highest pairwise correlation (r2) between SNPs in 5′ and 3′ LD blocks did not exceed 0.5 in all three datasets (Figure S6).
We performed association analyses using percent of lens affected with cortical cataract as a dependent variable. We also tested association using affection status only, classifying individuals as clearly affected (percent of cortical score ≥25) or unaffected (percent of cortical score <1), and eliminating all those with intermediate scores. If the whole range of scores was considered, the average cortical score in the BDES, UKTS, and BMES was 7.6%, 16.3%, and 3.2%, respectively, while the proportions with severe cortical cataract in the BDES, UKTS, and BMES were 10%, 29%, and 3%, respectively. Prior to association analysis, the quantitative trait of percent of lens affected and the binary trait, affected with cataract (yes/no), were adjusted for significant covariates using regression. The values obtained after regression (residuals) were inverse rank transformed to normalize their distribution. Association testing was performed on values of the trait before and after inverse rank transformation. Multiple genetic models (additive, dominant, and recessive) were tested for each SNP (Table S3 and Table S4), but for brevity only the dominant models are presented in the main text (Figure 6 and Table 5); the remainder of the models are presented in the supplementary materials. In the discovery family data, six SNPs, rs924201, rs7548209, rs11260721, rs3768293, rs6603867 and rs6678616 showed significant association (nominal P<4.4×10−4, calculated as 0.05 divided by 3 models×2 traits×19 genotyped SNPs), either with the full range of cortical cataract scores or with severe cortical cataract. Except for rs6678616, P values at the other five SNPs (rs924201, rs7548209, rs11260721, rs3768293 and rs6603867) were smaller when using the categorical yes/no definition of cortical cataract, despite the decrease in sample size (Table 5). Differences in the ability to detect association as a result of changed in phenotypic definition are not unexpected, as association is dependent on allele frequency, as well as on sample size.
Because the average cortical cataract score of the cases in the UKTS was 10% greater than the other two studies, the proportion of severe cases in the UKTS is 3 times and 10 times higher than that in the BDES and BMES, respectively (Table 2). Thus, in the replication dataset from the UKTS, P values for the severe cortical cataract trait were in general smaller than the full range of cortical cataract scores (Figure 6). Four of the same common SNPs observed in the BDES, rs7548209, rs3754334, rs3768293, and rs6678616, showed replication (smallest P value with rs3754334 = 1×10−4). In the UKTS we also identified significant association for severe cortical cataract with another rare variant Ile779Ile (P = 3×10−5). Rigorous diagnostic testing of the final models in the two family datasets (BDES and UKTS) confirmed that the results were valid (Figure S4).
We conducted single SNP and haplotype association analyses in the BMES (Table S5). As the BMES cohort is slightly younger, if we selected an age-matched case-control set, restricting controls to 70 years or greater, and defining cases as those with a cortical score≥25% and controls with cortical score<1% (Table 5) we observed replication using the trend test (best P value at rs7548209 = 0.003). Without these restrictions, the association was less significant because the chance of misclassifying disease is higher at younger ages (Table 5). Using the five most significant SNPs, rs7548209, rs3754334, rs3768293, rs6603867 and rs6678616, we performed a haplotype association test for cortical cataract using a moving window approach with different window sizes from 2 to 5. Haplotype analyses showed stronger evidence than single marker tests in the BMES (Table 5). Three haplotypes showed significant association at a nominal P<0.05 with window size 2 (Table 5). The most significant haplotype encompassed rs6603867-rs6678616 (r2 = 0.7) with C-A as the risk haplotype (β = 6.6, P = 1×10−6) and explained 6% of the total variance in cortical cataract scores. This result was consistent with the single SNP association tests in the two family studies (Figure 6E). The next best association is with C-G as the risk haplotype (β = 2.4, P = 0.0024) and G-G as the protective haplotype (β = −0.7, P = 0.020) at markers rs7548209-rs3754334 (r2 = 0.8), suggesting that the ‘C’ allele at rs7548209 causes increased risk; this haplotype explained 2% of the variance in the quantitative trait for cortical cataract. In total, these haplotypes accounted for 8% of the total variation. The risk alleles in the haplotype association test above are similar to those in the single SNP association in the two family datasets (Table 5).
Joint (meta-)analysis was conducted to combine the results of the two family studies and then to combine all three studies (Table S6). The most significant association for the full range of cortical cataract trait values using the two family sets and all three datasets was found with rs6603867 (P = 4×10−5) and rs6678616 (P = 1×10−4), respectively (Table 5). There was no significant heterogeneity for cortical cataract scores under the dominant model (P>0.05) with these two significant SNPs (Table S6A). For severe cortical cataract, rs7548209 (P = 8×10−9) and rs3768293 (P = 0.0037) were the most significant markers when combining the two family datasets and when combining all three datasets, respectively (Table 5). We noted that the SNPs for the severe cortical cataract trait showed heterogeneous effect sizes across studies (Table 5). Nevertheless, the direction of the association for the full spectrum of cortical cataract was the same among all three studies (Tables S3, Table S4, and Table S5). Diversity in effect sizes among the three datasets for severe cortical cataract is not surprising, because the proportion of severe cases in each study is different (Table 2). As a result, meta-analysis of all three datasets did not improve the overall significance of the results (Table 5). However, single SNP association, haplotype association, and meta-analysis consistently identified that two independent association signals are present in the 3′ (rs7548209 and rs3754334) and the 5′ (rs3768293, rs6603867, and rs6678616) regions of the gene. The maximum pairwise correlations between SNPs in the 3′ and the 5′ regions is less than 0.4 (Table 5), arguing for more than one susceptibility variant being present in these samples.
We report here that EPHA2 is essential in maintaining the clarity of crystalline lens in aging mice and is associated with age-related cortical cataract in humans. EPHA2 protein is expressed in the cortical lens fiber cells, and homozygous deletion of Epha2 in mice caused progressive cortical cataract. At the molecular level, cataractogenesis was preceded by accumulation of underphosphorylated HSP25, indicating elevated stresses and protein misfolding in Epha2−/− lens. In human studies involving three independent Caucasian populations, we discovered common polymorphisms, as well as a rare variant that significantly altered the function of EPHA2 kinase activities and cellular functions. Therefore, converging evidence from genetically engineered mice and human population studies, coupled with in vitro cell-based assays, strongly suggest that EPHA2 is the first major genes associated with the common form of cataract. The involvement of a receptor tyrosine kinase in cortical cataract etiology points to future directions in designing new preventive and therapeutic approaches for cortical cataract by targeting EPHA2.
Prior to this study, the only noticeable phenotype reported in Epha2 knockout mice is the kinked tails, which develops during early embryogenesis [20]. For yet unknown reasons, the penetrance of this phenotype has been quite low in the large cohorts of the same strain of knockout mice maintained in our facility on either FVB/NJ or C57Bl/6 genetic background. In another stain of Epha2 knockout mice generated by the secretory trapping strategy [17],[18], the incidence of kinked tails was also very low. This is in contrast with cataract phenotype reported here that is observed in significant fraction of both strains of homozygous knockout mice.
There are several unique approaches in our study compared with other studies to identify genes involved in complex disease. First, we utilized two independent lines of Epha2 knockout mice that were extensively investigated for potential molecular implications in cataractogenesis. This effort provided confidence in subsequent studies pursuing causal variants in human populations. Second, we investigated association with polymorphisms of EPHA2 in worldwide Caucasian populations from the United States, United Kingdom, and Australia. Phenotypic measurements were comparable among the three datasets, but there was variability in the extent of severity of cataracts in these populations. The most severe forms of cataract were associated with rare variants that were unique to each population and there was little replication between populations. Common variants could be replicated across populations with a greater degree of certainty. To ensure validity of the results, we extensively examined diagnostics of the final models in the family data. A candidate gene study using family data, unlike a genome-wide association study using unrelated data, requires special techniques to ensure the validity, i.e. normality of the model residuals expected asymptotically and required for the P values that presume normality to be valid. We applied the George-Elston transformation to both sides of the regression equation in the association tests [27]. We confirmed the normality of the residuals in the final models for all SNPs (Figure S4).
Our studies led to the successful identification of a non-synonymous variant, Arg721Gln, residing in the protein kinase domain of EPHA2. Moreover, the variant showed increased spontaneous activation in the absence of ligand stimulation, suggesting that the coding variant alters the spontaneous and ligand-triggered cellular signaling. The altered signaling is associated with enhanced growth inhibition by ligand, and stochastic intracellular retention in primary fibroblasts. The reason for the cytoplasmic retention in only some, but not all cells expressing the mutant is not clear at present. Regardless of the mechanisms, cytoplasmic trapping is likely to interrupt both signaling and structural functions of EPHA2/ephrinA system in the lens fiber cells. Because EphA2 is essential for maintaining lens clarity as evidenced by cataractogenesis in homozygous knockout mice, functional alterations in EphA2 is likely to impair its physiological roles in the lens, predisposing the carriers to increased risk of cataract development.
In addition to this rare variant, common polymorphisms were also discovered and replicated in worldwide Caucasian populations, suggesting that EPHA2 could be involved in predisposition to cataractogenesis in the general population. The most significant common variant for the quantitative cortical cataract trait was rs6678616 in the region coding for the ligand binding domain of the EPHA2 gene. Bioinformatic investigation revealed that rs6678616 is located in an exonic splicing enhancer element, and nucleotide changes of the SNP may potentially affect the affinity of splicing enhancer factors up to 14-fold (data not shown). The most significant SNP, rs7548209, for severe cortical cataract is present in a downstream conserved non-coding region of the EPHA2 gene. Characteristics of this variant are not known. It is possible that such a variant may affect EPHA2 expression during aging.
While this manuscript was under preparation, it was reported that mice with homozygous deletion of Efna5 encoding a ligand (ephrinA5) for EPHA2 also developed cataract [28]. These data nicely complement the data described in this paper, and demonstrate that both the receptor (EPHA2) and the ligand (ephrinA5) are required for maintaining lens clarity. In addition, recent genetic studies involving a relatively small number of affected individuals have found a link between congenital cataract and variants of EPHA2, although none of rare variants were functionally characterized [29]. These recent results on rare congenital cataract support our conclusion on the link between EPHA2 and the much more common age-related cataract, based on large human population studies and genetically engineered mice.
Since aging was the most important risk factor for age-related cataract, age-dependent reduction of EPHA2 protein expression in normal lens and age-dependent deterioration of lens clarity in Epha2 knockout mice suggests an important role of EPHA2 kinase in cataractogenesis during aging. Previous studies have shown that EPHA2 regulates epithelial cell morphogenesis and homeostasis in part by interacting with components of adherens and tight junctions, such as claudins and cadherins [30]. Notably, the lens is primarily composed of lens fiber cells where intricate cell-cell interactions are essential for structural and functional integrity. Changes of EPHA2 expression or function due to aging or genetic predisposition can impair junctional structure, leading to excessive cellular stress and eventual opacity.
Recent studies show that there are common genes and pathways among late onset neurodegenerative diseases, including Alzheimer's disease, multiple sclerosis, diabetes, and age-related eye disease. Genes involved in neurodegenerative diseases, kinesin light chain 1 [11] and apolipoprotein E [12], showed association with age-related cataract. Also, there is evidence of a link between longevity and age-related cataract [31]. Successful aging, defined as preserved cognition, is associated with reduced age-related cataract among the elderly [31]. Microarray expression analysis proved that lens expresses brain specific molecules, including β-amyloid secretases and degrading enzyme [32], synapsin and synaptic vesicle protein [33], and brain-specific miRNAs [34]. This suggests that common genes and pathways may be involved in both age-related cataract and neurodegenerative diseases affecting the elderly population. Since the EPH/ephrin system is known to be critical for neural development [35] and vision processes in the midbrain [36], the EPH and EFN genes may be potentially involved in a broad spectrum of age-related eye and neurodegenerative diseases. Further functional investigation of these genes will elucidate pathogenic mechanisms leading to age-related cataract and other late onset diseases.
In summary, converging evidence from mouse and humans demonstrates that EPHA2 has an important role in maintaining lens transparency. While genetic predisposition is known to be a major contributing factor to age-related cortical cataract, our study has identified and characterized genetic variants of the EPHA2 gene in human, and both common and rare variants confer risks for cortical cataract.
The KST085 line of Epha2 knockout mice was generated through secretory gene trapping as described previously [18]. The secretory trapping vector was inserted at the boundary between exon 5 and intron 6, leading to the truncation of exons 6 to 17 encoding the second fibronectin type III repeat in the extracellular domain all the way to the carboxyl terminal end. The remaining ectodomain encoded by exons 1 to 5 is fused to neomycin resistance and β-galactosidase reporter cassette (β-geo) and is trapped inside the cells in secretory vesicles, presumably in inactive form. These mice, which were generated on a C57Bl/6/129 genetic background, were then backcrossed with FVB/N mice. Mice from N4 or higher backcrosses were bred with each other to generate Epha2+/+, Epha2+/−, and Epha2−/− mice that were used in subsequent studies. Another line of mice on C57Bl/6 background was provided by Drs. M. Asano and Y. Iwakura [20]. In these mice, a retrovirus was inserted in the first codon of Epha2, leading to abolishment of EPHA2 expression.
All procedures involving mice were performed in accordance with guidelines set forth by the American Association for Accreditation of Laboratory Animal Care and the USPHS “Policy on Humane Care and Use of Laboratory Animals.” Studies were approved and supervised by The Case Western Reserve University Institutional Animal Care and Use Committee.
For histological analyses, mouse eyes were enucleated and fixed in neutral buffered 10% formalin solution for 24 hours. Eyes were dehydrated through a graded alcohol series and embedded in paraffin. Sections (5 µm) were cut through papillary-optic nerve axis and stained with hematoxylin and eosin (H&E).
For immunofluorescence analyses, postnatal day 14 eyes were dissected, fixed with 4% paraformaldehyde for 25 min. Eyes were washed with ice cold Phosphate Buffered Saline (PBS) and kept in 15% sucrose (prepared in PBS) overnight. Tissue freezing medium (Triangle Biomedical Science, Durham, NC) embedded eyes were cut on a cryostat (Leica, Germany), and air dried onto Superfrost Plus slides (Fisher Scientific). After washing with PBS, the sections were blocked with 50 mmol/L NH4Cl and permeabilized with 0.3% NP40 for 10 minutes. The sections were then incubated with primary antibodies at room temperature for 1 hour followed by detection with donkey secondary antibodies conjugated with FITC or Texas-red (Jackson ImmunoResearch, West Grove, PA) at room temperature for 30 minutes.
EPHA2 and Ephrin-A1 proteins were detected in human and mouse lens paraffin-embedded sections. Briefly, sections were deparaffinized, hydrated, and immersed in citrate buffer (10 mM sodium citrate, 0.05% Tween 20, pH 6.0) for 10 minutes at 95°C. Primary antibodies were incubated with samples and then detected by Biotinylated secondary antibody and avidin–biotin–peroxidase system (Vector). Color immunostaining was revealed using diaminobenzidine (Vector). Images were taken using a fluorescent Leica DM-IRE2 microscope equipped with a SPOT RT-SE digital camera (Diagnostic Instrument) and analyzed with MetaMorph software 6.1r4 (Universal Imaging).
Antibodies used include: goat anti-mouse EPHA2 ectodomain, goat anti-human EPHA2 (R&D Systems, Minneapolis, MN), rabbit anti-EPHA2 and anti-ephrin-A1, goat anti-HSP25 and mouse anti-phospho-ERK, rabbit anti-ERK (Santa Cruz Biotechnology, Santa Cruz, CA), rabbit anti-phospho-HSP25, anti-phospho-AKT, anti-Akt, anti-GAPDH (Cell Signaling), mouse monoclonal anti-N-cadherin (BD Biosciences). Rabbit-anti-α,β,γ-crystallin were kindly provided by Dr. Zigler (National Eye Institute).
Lenses were dissected from eyes and homogenized in ice-cold lysis buffer containing 20 mmol/L Tris (pH 7.4), 125 mmol/L NaCl, 10% glycerol, 1% Triton X-100,0.5% DCA, 0.1% SDS, 20 mmol/L NaF, 1 mmol/L phenylmethylsulfonyl fluoride, µg/mL aprotinin, 4 µg/mL leupeptin, and 1 mmol/L Na3VO4. Then samples were centrifuged at 13,000 g for 10 minutes at 4°C. Protein concentrations in supernatant were assessed using the bicinchoninic acid protein assay kit (Bio-Rad, Hercules, CA). Equal amounts of protein extracts were resolved by 4–20% SDS-PAGE and electrotransferred onto polyvinylidene difluoride membranes (Millipore, Bedford, MA), which were then blotted with the indicated antibodies.
Mice were anesthetized by an intramuscular injection of Tribromoethanol at 200 mg/kg body weight followed by induction of mydriasis with tropicamide (1%) and phenylephrine (10%). The lenses were examined with slit lamp and the observations were recorded by digital photography.
We selected 34 individuals for re-sequencing, 19 of whom had severe cortical cataract (worst cortical score ≥25% of lens involved), 15 were unaffected family members with no evidence of cataract formation (worst cortical score <1%) at ≥70 years of age, representing 18 families. Our rationale for choosing this set of individuals for re-sequencing was that they were previously shown to be linked to markers at 1p36 [13] and we hoped to find coding variants associated with disease, in addition to finding neutral variants.
To identify linked families, sib pairs from the SIBPAL program (version 5.3.1) in the candidate region (1p36) were ranked for linkage informativity [45]. For each sibling pair, a score based on the squared sib-pair difference and the estimated sib-pair marker allele sharing was computed:where and are the trait values for sibs 1 and 2, is the average of over the whole sample, is the estimated mean allele sharing for the two sibs. This score is large (positive) either when the squared sib pair difference of the trait value is small and is large or when the squared sib-pair difference is large and is small, which means the sibs are similarly alike, in terms of deviation from the mean, for both their traits and allele-sharing; otherwise the score will tend to be small (negative). Therefore, all the sibpairs with positive values were considered to be contributing to linkage.
All 17 EPHA2 exons along with 50–100 base pairs of the surrounding intronic junctions were sequenced with the polymerase chain reaction (PCR) with primers designed using the PrimerSelect program (version 4.05) in the DNASTAR software (Table S1). For each amplicon, the PCR conditions for amplification are described in Table S1. After PCR, the product was purified using ExoSAP-IT enzyme (1×; 2 µl of the ExoSAP-IT enzyme with 10 µl PCR product for a final concentration of 5 ng/µl) and the product sizes were confirmed on a 1.5% agarose gel. The amplified PCR products along with primers were sent to McLab (San Francisco, CA) for sequencing. Electropherograms were examined using the ChromasPro program (version 1.4) and new variants not previously reported in the literature were confirmed by sequencing in the reverse direction.
We selected SNPs for full-scale genotyping via three strategies: re-sequencing, tag-based selection using the linkage disequilibrium (r2), and phylogenetic mining. Based on the results of re-sequencing, three rare coding variants and a common SNP were selected for follow up in the remaining individuals of the discovery data. For the tag-based approach, linkage disequilibrium (LD) in EPHA2 was investigated by examining SNPs from 10 Kb upstream (+10 Kb) to 10 Kb downstream (−10 Kb) of the gene (16,323,419–16,355,151 bp) in HapMap data (CEU; Centre d'Etudes du Polymorphisme Humain; Utah residents with ancestry from northern and western Europe, HapMap Data, Data Rel 21 a/phaseII Jan07, on NCBI B36.3 assembly, dbSNP b130) using the HAPLOVIEW program (version 3.1). We selected at least one SNP in each block based on HapMap data to get four tagging SNPs.
Phylogenetic footprinting [46] and phylogenetic shadowing [47] were used for additional SNP selection. Under the assumption that non-coding regulatory regions are conserved through evolution, direct genotyping of non-coding conserved SNPs may increase power in chromosomal regions with low LD. We examined mouse-human sequences to maximize the number of inter-species conserved regions. The EPHA2 sequences from human (Homo sapiens) and mouse (Mus musculus) were aligned using zPicture. We used window size width of ≥100 bp with ≥70% identity between human and mouse [48]. In the phylogenetic shadowing approach, using multiple sequence comparisons among primates, sequence elements (blocks) that did not change among primates are identified as regulatory elements [47]. We identified the non-coding conserved regions in EPHA2 by comparing the human EPHA2 gene as a reference with the orthologous sequences in two primates, the chimpanzee (Pan troglodytes) and the rhesus monkey (Macaca mulatta) using eShadow [47]. Within each conserved block one SNP with the highest minor allele frequency (MAF) was selected for genotyping.
Based on the hybrid selection strategy for SNPs, we first selected four non-redundant tag based SNPs that had MAF>5%. These SNPs were supplemented with 11 conserved non-coding SNPs identified through phylogenetic approaches. Finally, a set of intriguing coding and regulatory SNPs (N = 4) identified through re-sequencing were also genotyped. In total, 19 SNPs were genotyped in all three datasets, the BDES, UKTS, and BMES.
We used two different techniques, TaqMan and SNPlex assays, to interrogate SNPs. The genotyping procedure was performed according to the manufacturer's protocol. Clustering algorithms for genotype assignment used three quality control steps: (1) removing wells (individuals) with a significant number of outlier SNPs, (2) setting a SNP pass value by assigning a pass or fail status for each SNP, and (3) for each passing SNP, a confidence value was calculated (call rate). A confidence value of at least 95% was required to retain a genotype call. We further cleaned the data by removing individuals with low quality DNA across the SNPs (at least 90% of SNPs met the threshold quality value). We calculated the error rate and no-call rate using 5% replicate plating of DNA prior to these filtering steps. The average error rate and no-call rate for SNPlex genotyping was 0.7% and 3.9%, respectively. The average error rate and no-call rate for TaqMan genotyping was 0.3% and 3.7%, respectively. Cleaned data were used in association analyses.
All genotyped SNPs met Hardy-Weinberg equilibrium with a nominal P>0.05. We estimated allele frequencies using the FREQ program in S.A.G.E. (version 5.4.2), and 95% confidence intervals were reported (Table S2). The program estimates allele frequencies using a maximum likelihood formulation that assumes a random mating population. Standard errors are computed by numerical double differentiation of the log likelihood (Table S2). We also examined linkage disequilibrium (LD) blocks in each dataset using the HAPLOVIEW program (version 3.32).
|
10.1371/journal.pgen.1007079 | Common, low-frequency, and rare genetic variants associated with lipoprotein subclasses and triglyceride measures in Finnish men from the METSIM study | Lipid and lipoprotein subclasses are associated with metabolic and cardiovascular diseases, yet the genetic contributions to variability in subclass traits are not fully understood. We conducted single-variant and gene-based association tests between 15.1M variants from genome-wide and exome array and imputed genotypes and 72 lipid and lipoprotein traits in 8,372 Finns. After accounting for 885 variants at 157 previously identified lipid loci, we identified five novel signals near established loci at HIF3A, ADAMTS3, PLTP, LCAT, and LIPG. Four of the signals were identified with a low-frequency (0.005<minor allele frequency [MAF]<0.05) or rare (MAF<0.005) variant, including Arg123His in LCAT. Gene-based associations (P<10−10) support a role for coding variants in LIPC and LIPG with lipoprotein subclass traits. 30 established lipid-associated loci had a stronger association for a subclass trait than any conventional trait. These novel association signals provide further insight into the molecular basis of dyslipidemia and the etiology of metabolic disorders.
| Lipid and lipoproteins are heritable traits that differ in content and size and are correlated with coronary heart disease and mortality. To identify genetic variants associated with different subclasses of lipoproteins, we conducted a genome-wide association study of 8,372 Finnish men. We curated a dataset of all genetic variants known to be associated with lipid or lipoprotein subclasses and used these data to conduct rigorous analyses to identify new associations in the same gene region or new ones. We identified five new signals at established lipid-associated loci revealing possible complex regulatory mechanisms underlying the signals. Using the contribution of rare coding variants predicted to be protein truncating or missense, we uncovered novel associations for a set of variants at LIPC and LIPG with HDL subclasses. Investigating the genetic association of lipoprotein subclass traits may help lead to a better understanding of the etiology of cardio-metabolic diseases, and provide novel therapeutic targets.
| Genome-wide association studies (GWAS) have identified hundreds of common (MAF>0.05) variants associated with conventional lipid and lipoprotein traits: high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), total cholesterol (TC), and triglycerides (TG)[1–4]. While some low-frequency (0.005<MAF≤0.05) and rare variants (MAF≤0.005) have been associated with lipid and lipoprotein traits, additional loci remain to be identified[2,5,6]. High-throughput proton nuclear magnetic resonance (NMR)-based measurements of lipid and lipoprotein subclasses provide a more comprehensive view of particle size and composition than conventional blood lipid profile measurements[7], and these expanded sets of traits have been associated with metabolic and cardiovascular diseases[8–10]. For example, HDL subclasses are differentially associated with incidence of coronary heart disease, and VLDL particle size is negatively associated with mortality[11,12].
Previous association studies for lipid traits have identified several genomic regions of <1 Mb that contain more than one association signal for which the lead variants are not in strong linkage disequilibrium (LD) (r2<0.8)[2,3,13]. Fine-mapping with higher density variants and conditional analyses can determine which signals are distinct (remain significant after conditional analysis) and which are independent, which we define here as r2<0.01. For example, Teslovich et al. used conditional analysis at 95 lipid loci to identify 26 loci that harbor at least two distinct association signals[3]. Association signals at the same locus can be population-specific or shared across populations, with potentially different effect sizes and/or lead variants[13]. Multiple association signals at a locus may indicate allelic heterogeneity in gene function or regulation or that more than one gene at the locus affects the trait[14]. Furthermore, identifying and accounting for additional independent association signals increases the variance in traits that can be explained by genetic loci[15,16].
In this study, we performed genome-wide single-variant and gene-based association analyses of 68 NMR lipid and lipoprotein subclass traits and four conventional traits (TC, TG, HDL-C, and LDL-C) in 8,372 non-diabetic Finnish men from the METabolic Syndrome In Men (METSIM) study[17]. To identify novel associations, we performed analyses with and without conditioning on lipid-associated variants at loci previously described in array- and sequence-based GWAs. We identified the most strongly associated lipid and lipoprotein subclass traits at established loci for conventional lipid and lipoprotein traits. Since several subclasses are associated with cardiovascular and metabolic diseases, identifying the variants that influence these traits is the first step to develop novel clinical treatments. These expanded association results have the potential to lead to advances in determining the etiological role of the variants and genes in cardiovascular and metabolic disease.
To identify genetic variants associated with the 72 lipid and lipoprotein traits, we analyzed 15.1M genotyped and imputed variants in 8,372 non-diabetic Finnish men (S1 Table, S1 Fig). Each trait was adjusted for age, age2, lipid-lowering medication use, and smoking status. Inverse normalized trait residuals were tested for association with each variant assuming additive allelic effects using a linear mixed model to account for relatedness among study participants[18]. Many of the traits are highly correlated with each other, with 104 trait-pair comparisons having a pairwise Pearson correlation greater than 0.98 (S2 Fig). We used a genome-wide significance threshold of P≤5×10−8, consistent with previous association studies of this scale and high trait correlation[19]. We note where associations meet a conservative experiment-wide Bonferonni-corrected P-value (P≤4.6×10−11). We identified 32,524 variant-trait associations (Pdiscovery<5×10−8) for the 72 lipid and lipoprotein traits (S3 Fig). 30,348 (93%) of the 32,524 associations were for one of the 68 subclass traits and 2,176 (7%) for one of the four conventional lipid traits (TC, TG, HDL-C, LDL-C). More than half the associations were with the VLDL- (38%) or HDL-subclass (29%) traits. 3,784 unique variants comprise the total 32,524 trait-variant associations (S2 Table). 73% (2,780) of the 3,784 variants had a greater association with one of the 68 subclass traits, and 27% (1,004) were more highly associated with at least one of the four conventional traits (S2 Table). These variants cluster into 42 loci that were associated with at least one of the 72 traits (S1 Table). For example, at the well-characterized APOA5 locus on chromosome 11, rs964184 was significantly associated (Pdiscovery<5×10−8) with 43 of the 72 lipid and lipoprotein traits. At CETP, rs12446515 was significantly associated with 34 of the 72 traits. For such loci, the high correlation between the traits obscures identification of a causal trait underlying the signal.
To identify novel associations not reported previously for any conventional or subclass lipid or lipoprotein trait, we identified and curated a list of previously known associated variants to use in genome-wide conditional analyses (Methods). We identified 1,714 variants (S3 Table) that we clustered based on stringent LD (r2>0.95) into 885 representative variants (S4 Table). After genome-wide conditional analysis using these 885 variants, we defined novel association signals using a significance threshold of Pconditional<5×10−8 (S5 Table). Consistent with highly correlated traits, we observed that most of the associated variants were associated with multiple correlated traits. We considered variants located within 1 Mb of an established lipid or lipoprotein signal to be an additional signal in the region, and we define a locus as the region 1 Mb up- and downstream of a signal. We considered additional signals independent if the signal was not in LD (r2<0.01) with known lipid/lipoprotein signals, and remained significant (Psingle<5×10−8) after single-variant conditional analyses. Associated variants with MAF<0.01 were validated by direct genotyping or sequencing (see Methods). Using this genome-wide conditional approach, we identified five novel signals near established lipid and lipoprotein loci (Table 1).
Common variant rs73059724 (MAF = 0.09), associated with decreased (β = –0.14) concentrations of phospholipids in small VLDL, is located 3.5 kb upstream of HIF3A (hypoxia inducible factor 3, alpha subunit) and 1.4 Mb from APOE (S4A and S5A Figs). Additionally, this signal is associated with decreased VLDL subclass traits (S6 Fig). This signal achieved significance after conditioning on known lipid GWAS variants (Pdiscovery = 3.8×10−7, Pconditional = 1.4×10−8) (Table 1, S6 Table). When adjusted for total triglycerides, the strength of the association of rs73059724 with phospholipids in small VLDL was reduced (P = 3.6×10−2, S7 Table). This signal is located in a gene-dense region on chromosome 19 that includes 10 previously reported lipoprotein-associated variants within 1 Mb of the index variant (S6 Table)[20]; none of these variants exhibited LD (r2>0.02) with rs73059724. Further analysis of the APOE locus with additional samples may be necessary to elucidate the haplotype relationships between these signals. Twenty-nine proxy variants in LD (r2>0.7) with rs73059724 span a 25-kb region including the promoter and intron 1 of HIF3A, and five of these variants overlap ≥5 liver and adipose regulatory element (histone marks of transcriptional regulation and open chromatin) datasets (S8 Table). Hyper-methylation at HIF3A is associated with increased adiposity and BMI in Asian infants and children[21,22]. HIF3A is a known negative regulator of HIF1A (hypoxia inducible factor 1, alpha subunit)[23], which has been shown to regulate the cellular uptake of cholesterol esters and VLDL by creating hypoxic conditions[24]. One or more of the associated variants may affect HIF3A transcription or other genes in the region, leading to fewer phospholipids in small VLDL particles.
We identified two new signals with low-frequency variants located near ALB and SYS1 (Table 1, S4B and S5B Figs). At the ALB locus, the low-frequency allele of rs187918276 (MAF = 0.017) located in intron 1 of ANKRD17 was associated with increased (β = 0.60) concentration of small LDL particles (Pdiscovery = 6.3×10−22, Pconditional = 3.2×10−11) and 26 additional traits, including increased TC, LDL-C, esterified cholesterol, free cholesterol, and IDL/LDL/VLDL subclasses (S6 Fig). When adjusted for total cholesterol, the strength of the association of rs187918276 with small LDL particles was reduced (P = 9.1×10−7, S7 Table). Variants in LD (r2>0.7, METSIM) with this variant span >1.2 Mb (S5B Fig, S8 Table), consistent with long haplotypes previously described in Finns[25]. The 885 variants used for the conditional analysis included established TC-associated signals at rs60873279 and rs182616603, located 337 kb and 1 Mb away; these variants exhibited low (r2<0.01) and moderate (r2 = 0.39) pairwise LD with rs187918276 (S6 Table). When conditioned on rs182616603, the association with rs187918276 was reduced but still highly significant (Psingle = 5×10−15), suggesting the signals are distinct. An additional variant at this locus, rs115136538, was reported previously to be associated with albumin levels[5]. rs115136538 is located 710 kb away from and is not in LD with rs187918276 (r2<0.01 in METSIM), and the association of rs187918276 with small LDL particles was essentially unchanged when conditioned on rs115136538 (S6 Table). Taken together, the ALB region contains three distinct signals for lipid traits (rs60873279, rs182616603, and now rs187918276).
ALB encodes albumin, which is responsible for shuttling cholesterol in the blood to the lipoprotein particle acceptors; deletion of Alb in mice led to a hyperlipidemic condition[26,27]. One of the 12 variants in LD (r2>0.7) with rs187918276, chr4:74265673, is located 4.3 kb upstream of the ALB transcription start site (TSS), and is the only variant that overlapped any epigenomic marks of transcriptional regulation from the adipose, blood, and liver datasets (S8 Table). This variant may mediate a regulatory effect on ALB to increase the plasma concentration of small LDL particles, or another of the candidate variants spanning 1.2 Mb may act on this or another nearby gene.
In an intergenic region downstream of PIGT, we identified the low-frequency allele of lead variant rs184392658 (MAF = 0.008) associated with the increased (β = 0.45) concentration of large HDL particles (Pdiscovery = 2.3×10−7, Pconditional = 2.5×10−9, Table 1, S4C Fig and S5C Fig). When adjusted for HDL-C, the association of rs184392658 with large HDL particles was reduced (P = 4.1×10−5, S7 Table). Two previously established lipid-associated variants are located within 1 Mb of rs184392658: rs1800961 near HNF4A and rs6065904 near PLTP. rs184392658 was not in LD (r2<0.015) with either of these established variants, and conditioning on the individual known variants did not substantially change the association signal (all Psingle<3.7×10−6, S6 Table). Thus, rs184392658 represents a new distinct signal in this region. Of six variants in high LD (r2>0.7) with lead variant rs184392658, only rs149985455 overlaps multiple epigenomic marks of transcription regulation from liver, blood, and adipose tissue datasets (S8 Table). This variant is located 2.2 kb upstream from SYS1 (Sys1 Golgi trafficking protein), which may have a role in lipid metabolism through an interaction with GTPases[28]. SYS1 targets ARFRP1 (ADP-ribosylation factor-related protein 1) and forms a complex in the Golgi membrane[29]; deletion of Arfrp1 in mouse adipocytes led to lipodystrophy caused by failure in lipid droplet formation[30]. rs149985455 may mediate a regulatory effect on SYS1 to increase the plasma concentration of large HDL particles, or another of the candidate variants spanning >500 kb may act on this or another nearby gene.
We identified additional novel independent signals with rare variants near LCAT and LIPG (Table 1). The rare allele (MAF = 0.005) of the missense variant rs199717050 (Arg123His) in exon 3 of LCAT (lecithin-cholesterol acyltransferase) was associated with decreased (β = –0.72) HDL-C levels (Pdiscovery = 5.9×10−10, Pconditional = 2.5×10−12, Table 1, S4D Fig and S5D Fig). This signal was not significantly associated with any of the HDL subclass traits or other traits from this study (S6 Fig). The association of rs199717050 with HDL-C was nominally reduced (P = 2.9×10−8) when adjusted for total cholesterol (S7 Table). Six variants at this locus, within 1 Mb of rs199717050, were reported previously to be associated with HDL-C[2,4] (S6 Table). However, these six variants all show low pairwise LD with rs199717050 (r2<0.01), and single-variant conditional analyses using any one of the six variants did not substantially change the association of rs199717050 with HDL-C (Psingle ≤1.9×10−9, S6 Table). rs199717050 may be nearly specific to Finns; the Exome Aggregation Consortium (ExAC) database shows a total allele count of 16: fifteen in Finns and one in a non-European population. LCAT is responsible for cholesterol esterification for eventual transfer into the lipoprotein core, and facilitates the transport of cholesterol into the liver[31]. rs199717050 is predicted to be deleterious (SIFT, 0.02) or possibly damaging (PolyPhen, 0.55)[32], consistent with a plausible functional effect on LCAT to decrease levels of HDL-C.
Another novel signal was located at the well-established HDL-C-associated LIPG locus (Fig 1)[33]. The rare allele (MAF = 0.004) of lead variant rs538509310 is located 3.6 kb upstream from ACAA2, and was most strongly associated with increased (β = 0.72) levels of phospholipids in medium-size HDL (Pdiscovery = 1.7×10−9, Pconditional = 3.2×10−10, Table 1, Fig 1A and 1B). This signal was also significantly associated with increased levels of four other HDL subclass traits and apolipoprotein A-I (S6 Fig). When adjusted for HDL-C, the association of rs538509310 with phospholipids in medium-size HDL was reduced (P = 4.5×10−5) (S7 Table). rs538509310 is in near complete LD (r2 = 0.98) with rs201922257, which encodes a missense substitution (Ala172Val) in exon 4 of LIPG. At least four previously described HDL-C variant association signals are located within 1 Mb of this variant, including rs74558535 (P = 2×10−10), rs10438978 (P = 7.7×10−36), rs77960347 (P = 3.6×10−11), and rs2156552 (P = 2×10−12). The new signal is not in LD (r2<0.043) with the previously described variants and remained significant after single-variant conditional analyses (S6 Table, Fig 1C). LIPG encodes endothelial lipase (EL), which catalyzes HDL phospholipids and aids in the sequestration of HDL from circulation, and is expressed in several tissues and organs including the liver[34–36]. The association with phospholipids in medium-size HDL is consistent with the known phospholipase of EL[37]. Several variants in LIPG have been shown to decrease endothelial lipase levels and increase HDL-C[38]. Based on the direction of effect in these previous studies, missense variant (A172V) may decrease function of LIPG, leading to increased phospholipids in medium-size HDL and other HDL subclasses.
To test the association between lipid and lipoprotein subclasses and sets of coding variants within a gene, we performed gene-based tests of association using SKAT-O with four variant masks (Methods) based on the predicted function of the coding variants. Sets of variants in LIPC (Pgene = 7.1×10−11) and LIPG (Pgene = 3.8×10−17) were associated with lipid and lipoprotein subclasses using the gene-based method; these results remained significant after adjusting for nearby noncoding signals (LIPC P<1.3×10−10 and LIPG P<1.2×10−17) (Fig 2, S9 Table).
At LIPC, the set of five rare missense variants, R138C, A145T, R208H, R281Q, and R329H, showed the strongest association using the protein truncating variant (PTV)+missense mask with triglycerides in very large HDL (Fig 2A, Pgene = 7.1×10−11). Of the five single-variant tests of association with triglycerides in very large HDL, A145T was individually the most significant (Pdiscovery = 5.3×10−8). Four of the variants (R138C, A145T, R208H, and R281Q) showed higher trait levels (β = 0.72 to 1.8) and were predicted to be deleterious by Variant Effect Predictor (VEP), while R329H, observed in one individual, showed a modestly lower trait level (β = –0.24) and was predicted to be benign[32]. While rare, A145T had 1.7-fold higher allele frequency in Finns (0.003%) than other populations[39]. Three of the variants, A145T, R138C, and R208H, were associated with increased HDL-C in a previous gene-based association study, consistent with our results[40]. Among the other variants, the relatively high trait values for R281Q suggest that it may also increase HDL-C. Based on previous data that decreased LIPC expression can result in increased large HDL levels[41], the rare alleles may lead to reduced LIPC function. Consistent with the gene-based test, deficiency in hepatic lipase activity resulted in increased concentration of triglycerides in plasma HDL[42].
At LIPG, the PTV+missense mask showed five variants with the strongest association with phospholipids in medium-size HDL (Fig 2B, Pgene = 3.8×10−17). Of the five single-variant tests, a rare missense (A172V) variant rs201922257 was the only one significantly associated (Pdiscovery = 8.6×10−9) with the subclass trait, and in three of four transcripts the amino acid substitution is predicted by VEP to be ‘deleterious’ and ‘probably damaging’ in most of the transcripts (Fig 2B). This variant is in LD (r2 = 0.98) with the non-coding index variant rs538509310 for phospholipids in medium HDL (Table 1, Fig 1A). The other associated variants may also affect LIPG function despite less-significant P-values. A splice variant rs200435657 (MAF = 0.0035, Pdiscovery = 4.0×10−6) is located at the 3’ end of intron 1; this variant has only been observed once (1/121,029; 0.0008%) in non-Finnish ExAC samples. Based on position, this splice variant is predicted to cause skipping of exon 2, which would lead to four aberrantly coded amino acids and a stop codon in exon 3. The remaining missense variants are predicted by VEP to be deleterious except for E391K. N396S and E391K have been reported previously to be associated with increased HDL-C levels[43–45]. However, our data suggest that all five variants increase phospholipids in medium HDL (β = 0.01 to 0.75) (Fig 2B). Together, the gene-based tests suggest that additional rare variants may influence LIPG function and HDL-C subclass levels.
We next asked whether any of 157 previously known loci associated with one or more of the four conventional lipid and lipoprotein traits exhibited stronger evidence of association with one of the lipid or lipoprotein subclass traits. Among the 157 loci associated (P<5×10−8) here with at least one subclass trait, 30 showed stronger association with a subclass trait than any conventional trait (Table 2, S7 Fig). For example, at PLTP (phospholipid transfer protein), rs4812975 was much more strongly associated with HDL diameter (Psubclass = 1.4×10−15) than with HDL-C (Pconventional = 2.6×10−3), consistent with PLTP mediating the net transfer of phospholipids between lipoproteins and uptake of phospholipids into the HDL-C core[46]. In addition, at ANGPTL3 (angiopoietin-like 3), ANGPTL4 (angiopoietin-like 4), and LPL (lipoprotein lipase), the variants were all more strongly associated with VLDL subclass traits than with the conventional traits (Table 2), consistent with studies showing that mouse Angptl3 knockout and Angptl4 overexpression may act via Lpl to decrease or increase VLDL, respectively[47,48].
At less well-characterized and gene-dense loci, lipid and lipoprotein subclass associations may help suggest target genes or biological roles. At the gene-dense MTCH2-NUP160 locus, rs4752801 was >3 log units more strongly associated with decreased free cholesterol in large HDL levels (Psubclass = 1.4×10−9) than any conventional trait (HDL-C, Pconventional = 8.2×10−6, Table 2). The pattern of association of rs4752801 with all 72 subclass traits (S7 Fig) is similar to the pattern of association and direction of effect for at least two other signals, rs737337 at ANGPTL8 and rs1129555 at GPAM. ANGPTL8 and GPAM are both regulated directly or indirectly by LXR, encoded by NR1H3,[49,50] which is a positional candidate gene at this locus[49,50]. Thus, the global pattern of association supports a contribution of NR1H3 at the MTCH2-NUP160 locus and suggests that the lipid and lipoprotein subclass traits can be a useful tool to help determine which genes underlie association signals.
We performed a similar analysis of lipoprotein associations at coronary artery disease (CAD) loci (S10 Table). Variants at the APOA5/APOA1 locus were 5.2 log units more strongly associated with triglycerides in small VLDL than total triglycerides, and APOE/APOC1 was 2.3 log units more strongly associated with ratio of apoA-I/apoB than any conventional trait. APOA5 has been shown to affect VLDL concentrations and TG-rich particle metabolism, and the stronger association with the subclass trait is consistent with the known functions of these genes[51].
In this study we conducted GWAS for 72 lipid and lipoprotein subclass traits in 8,372 Finnish men participating the METSIM study, and focused on identifying association signals that had not been identified previously with any lipid or lipoprotein trait. From the literature of existing lipid and lipoprotein association studies, we identified 1,714 cholesterol, TG, lipid, and lipoprotein-associated variants. We trimmed this list based on LD (r2>0.95) to 885 variants to account for multiple known signals in a genome-wide conditional analysis. With this approach, we identified five novel signals at established lipid loci. We confirmed that signals were independent by reciprocal conditional analyses.
This analysis focused on NMR measurements of 72 lipid and lipoprotein subclasses, including four conventionally measured lipid traits: TC, TG, HDL-C, and LDL-C. 892 of the association signals were located at or near loci previously associated with one or more of the four conventional traits. Lipid and lipoprotein subclass traits have been linked to metabolic and cardiovascular diseases, which underline their clinical importance[8–10]. We identified variants at 30 loci that showed a more significant association with a subclass trait than one of the conventional lipid traits, consistent with previous observations[52].
The identification of multiple independent association signals at established GWAS loci can provide supporting evidence to identify target genes, as with monogenic disorders. Loci that harbor more than one association signal that affect transcriptional regulation of the same gene, or more than one coding variant that affect the same gene’s function, provide stronger evidence for a gene’s role in determining trait variability. Multiple signals can be critical to understanding the relationship between genetic variants and gene function, quantitative traits, and disease[53]. Multiple association signals at established loci can also be used to detect molecular interactions between coding and regulatory variants on protein levels[54]. In addition, multiple signals at the same locus may suggest that more than one nearby gene affects trait variation, and the association signals may represent different routes of transcriptional regulation. Further study of the multiple association signals at a locus may more precisely define the functional genetic mechanisms.
The gene-based tests of association at LIPC and LIPG identified new rare coding variants that may alter the function of these genes, and of lipid and lipoprotein subclass levels. While the missense variants identified here all have mean normalized lipoprotein trait values above the population mean, this type of analysis can help distinguish variants that lead to loss or decrease vs gain or increase of gene function[53]. As well, the rise of whole exome sequencing will likely uncover many more rare coding variants, including variants with unknown significance on gene function. While it is still unclear which variants included in the gene-based tests for LIPC and LIPG truly affect gene function, the comparison of trait values between carriers of different variants may be used to help interpret the potential role of these variants in individual carriers.
In conclusion, this GWAS of 72 lipid and lipoprotein subclass traits in 8,372 Finnish participants in the METSIM study identified associations with 42 loci previously identified only with the conventional lipid and lipoprotein traits[2], five novel signals associated with lipoprotein subclasses, and eight rare, potentially functional, coding variants at LIPC and LIPG. Our use of a dense reference panel of >15M variants combined with the high-throughput NMR-measured traits allowed us to conduct higher-resolution genetic analyses than reported previously. Functional analysis of the variants identified in this study is the next step to determine which variants and genes are affected, and replication of these lipid and lipoprotein subclass associations in women and in other ancestry groups will be useful to better understand the genetic architecture of lipid and lipoprotein metabolism.
The METSIM study was performed in accordance with the Helsinki Declaration and was approved by the Research Ethics Committee, Hospital District of Northern Savo (number 171/2004). All study participants gave their written informed consent.
Among the 10,197 participants in the METSIM study, we analyzed 8,372 non-diabetic individuals (mean age 57±7 SD years and BMI 26.8±3.8 kg/m2)[55].
We measured the 72 lipid and lipoprotein traits from blood serum samples by proton NMR, as previously described[56]. Briefly, lipid samples are extracted and measured by proton NMR, and the NMR-spectra and automated phasing are compared to plate, background, and serum controls. Regression modeling is used to quantify the spectral areas to produce the quantified molecular data. The samples included 60 lipoprotein subclasses, 6 cholesterol and triglyceride measures, 3 cholesterol diameter measures, and 3 apolipoprotein measurements (S1 Table). Definition of the subclass traits has been previously described[56,57]. We visualized the Pearson correlation matrix between lipoprotein traits using a corrgram with the ellipse (https://cran.r-project.org/web/packages/ellipse/) and lattice packages (http://lattice.r-forge.r-project.org/) within R (S2 Fig)[58].
We genotyped the study samples using the HumanOmniExpress-12v1_C BeadChip and Infinium HumanExome-12 v1.0 BeadChip, resulting in 631,879 and 236,849 variants, respectively. Imputation was performed using the GoT2D reference panel of >19M variants (SNPs, in-dels, and large deletions) based on whole-genome sequence of 2,657 Europeans consisting of German, Swedish, Finnish, and British participants; with the majority of the cohort comprised of Finns[59]. The resulting 15,144,991 variants were subjected to quality controls including sample- and variant-level controls for detecting sample contamination, sex and relatedness confirmation, and detection of sample outliers using principal-component analysis. To exclude samples with evidence of DNA contamination, we used BAFRegress v0.9 (http://genome.sph.umich.edu/wiki/BAFRegress). Based on principal component analysis, eighteen exome array sample duplicates, one individual each from seven monozygotic twin pairs, and twelve population outliers were removed from analysis. Due to sex chromosome inconsistencies, fourteen OmniExpress samples were removed. Samples with low genotype call rate (<95%) for either array were removed. Variants with low-mapping quality to build hg19, low genotype completeness (<95% for OmniExpress and <98% for exome array), or multi-allelic variants were removed. The remaining high quality variants were phased using Shape-It v2[60].
We tested for association using imputed dosages for all variants with summed minor allele count dosage >1 with each of the 72 lipid and lipoprotein traits assuming an additive model and accounting for cryptic relatedness using the EMMAX linear mixed model approach as implemented in EPACTS (http://genome.sph.umich.edu/wiki/epacts). Traits were adjusted for age, age2, smoking status, and lipid lowering medication. Residuals were inverse normalized. To assess the level of genomic inflation, we calculated the genomic control statistic (λGC) for all of the trait-variant associations using R[58] (S1 Fig). Reported effect size regression coefficients (betas) sizes are given in standard deviation units. The rare lead associated variants that were imputed and had MAF<0.01 were tested for genotype accuracy by using TaqMan assays (Thermo Fisher Scientific) or Sanger sequencing in 499 METSIM participants who carried one or more rare alleles at these variants. Variants that had >10% discordance between the imputed genotype and the sequenced genotype in the examined individuals were removed from the analysis. Variants with MAF<0.001 were excluded from analysis (S5 Table).
To identify variant association signals distinct or independent from those reported previously, we identified variants previously reported to be associated with any cholesterol, lipid, lipoprotein, or triglyceride trait. We performed a literature review of GWAS and sequencing studies using PubMed (https://www.ncbi.nlm.nih.gov/pubmed/) and Google Scholar (https://scholar.google.com/), screened a GWAS Catalog (http://www.ebi.ac.uk/gwas/), and used SNIPPER (https://csg.sph.umich.edu/boehnke/snipper/) to query publicly accessible databases. The resulting curated list contained 1,714 variants, at >150 loci from 33 studies (S3 Table). The resulting curated list contained 1,714 variants, at >150 loci from 33 studies (S3 Table). We used this list to represent the known genome-wide lipid and lipoprotein-associated variants.
We LD-pruned (r2>0.95) the compiled list of 1,714 variants (S3 Table) to 885 variants (S4 Table) and we used this list (n = 885) in genome-wide conditional tests of association; this stringent LD threshold facilitates conditioning on multiple known signals at a locus. Signals that remained significant (Pconditional<5×10−8) after genome-wide conditional analysis were considered novel and further tested by single-variant conditional analyses to determine independence. At each of the five loci (Table 1), variants within 1 Mb up- and downstream of the lead variant and on the compiled list of 1,714 variants were included in single-marker conditional analyses (S6 Table). Signals that remained significant (Psingle<5×10−8) after single-variant conditional analysis were considered independent. Signals that only achieved a significance threshold of Psingle<5×10−6 after single-variant conditional analysis were considered distinct. The single-variant conditional analyses considered variants within 1 Mb of the signal, which accounts for <1% of the genome. Therefore, the significance thresholds for the distinct and independent additional signals are conservative. At each locus, we validated that signals were distinct/independent by reciprocal conditional analysis with the putative novel lead associated variant for the trait. Additionally, the association data for the novel signals was adjusted for each of the four conventional traits (HDL-C, LDL-C, TC, and TG), and the effect of the association is reported in S7 Table.
For each of the 885 lipid/lipoprotein-associated variants (S4 Table), we determined whether the variant showed stronger association with one of the 68 subclass traits compared to the four conventional lipid traits (TC, TG, LDL-C, HDL-C). Variants were included for comparison if the variant association with a subclass trait satisfied P<5×10−5 and if–log10pvalue for subclass trait association was greater than any of the four conventional lipid traits.
To determine the contribution of rare coding variants, we used the Optimal Sequence Kernel Association Test (SKAT-O) with EMMAX, as implemented in EPACTS, to test for gene-based associations with the 72 lipoprotein traits[61]. Only coding variants directly genotyped on the OmniExpress or Exome array were included, resulting in 709,600 variants. Since SKAT-O requires no missing data, we imputed missing genotype data with the variant mean genotype. We annotated coding variants using VEP. These annotations were the basis for four masks that we implemented in the gene-based tests, as previously described[53]. Briefly, the four masks were: Protein-Truncating Variants (PTV): no MAF limit; variants are nonsense, frameshift, or essential splice variants. PTV+missense: MAF<1%; all PTVs and missense variants. PTV+Nonsynonymous strict (NSstrict): no MAF limit; all PTVs and missense variants predicted as deleterious by five variant annotation algorithms: LRT, Mutation Taster, PolyPhen2-HumDiv, PolyPhen2-HumVar, and SIFT. PTV+NSstrict+NSbroad: MAF<1%; all variants in PTV+NSstrict and variants predicted to be deleterious by any of the five algorithms above. Only genes containing two or more variants in a given mask were tested. We conducted gene-based conditional analyses to determine whether a single variant or a net-effect of multiple variants could explain the observed association signal.
To better characterize the novel signals from this study, we determined whether the associated lead variants and LD proxies (r2>0.7 in METSIM) at each signal were within ChIP-seq peaks of epigenomic transcriptional regulatory elements (S8 Table). We built lists of such elements using data from the ENCODE Consortium[62] and Roadmap Epigenomics Project[63]. We used datasets from three lipid and cholesterol relevant tissues (adipose, blood, and liver datasets) that were comprised of experimentally defined regions of transcription factor binding sites (ChIP-seq), open chromatin (DNase- and FAIRE-seq), and histone modification marks (H3K4me1, H3K4me2, and H3K4me3, H3K27ac, and H3K9ac).
|
10.1371/journal.pgen.0030126 | Substitution as a Mechanism for Genetic Robustness: The Duplicated Deacetylases Hst1p and Sir2p in Saccharomyces cerevisiae | How duplicate genes provide genetic robustness remains an unresolved question. We have examined the duplicated histone deacetylases Sir2p and Hst1p in Saccharomyces cerevisiae and find that these paralogs with non-overlapping functions can provide genetic robustness against null mutations through a substitution mechanism. Hst1p is an NAD+-dependent histone deacetylase that acts with Sum1p to repress a subset of midsporulation genes. However, hst1Δ mutants show much weaker derepression of target loci than sum1Δ mutants. We show that this modest derepression of target loci in hst1Δ strains occurs in part because Sir2p substitutes for Hst1p. Sir2p contributes to repression of the midsporulation genes only in the absence of Hst1p and is recruited to target promoters by a physical interaction with the Sum1 complex. Furthermore, when Sir2p associates with the Sum1 complex, the complex continues to repress in a promoter-specific manner and does not spread. Our results imply that after the duplication, SIR2 and HST1 subfunctionalized. The single SIR2/HST1 gene from Kluyveromyces lactis, a closely related species that diverged prior to the duplication, can suppress an hst1Δ mutation in S. cerevisiae as well as interact with Sir4p in S. cerevisiae. In addition, the existence of two distinct protein interaction domains for the Sir and Sum1 complexes was revealed through the analysis of a chimeric Sir2–Hst1 molecule. Therefore, the ability of Sir2p to substitute for Hst1p probably results from a retained but reduced affinity for the Sum1 complex that is a consequence of subfunctionalization via the duplication, degeneration, and complementation mechanism. These results suggest that the evolutionary path of duplicate gene preservation may be an important indicator for the ability of duplicated genes to contribute to genetic robustness.
| Gene duplication is an important force in evolution, as it provides a source of new genetic material. However, the mechanisms by which duplicated genes are retained and diverge are understudied at the experimental level. We have examined a pair of duplicated histone deacetylases Hst1p and Sir2p from baker's yeast, which are important for distinct types of gene repression. In this study, we show that before the duplication the ancestral histone deacetylase had both Hst1p- and Sir2p-like functions, and after the duplication Sir2p and Hst1p subfunctionalized, giving rise to two distinct proteins with normally nonoverlapping functions. Despite having partitioned the ancestral functions after the duplication, Sir2p can substitute for Hst1p in its absence by interacting with the normal partner of Hst1p. This study suggests that the evolutionary path of duplicate gene preservation may be an important indicator for the ability of duplicated genes to substitute for one another and hence protect the organism against deleterious mutations.
| The evolutionary role of gene duplication presents a paradox. Gene duplication provides a source of new genetic material that is free of selective constraint and can evolve novel functions [1,2], but at the same time, gene duplication provides genetic robustness against deleterious mutations through redundant function [3–5]. How duplicated genes protect against null mutations while continuing to evolve different functions is at the core of the paradox. Deletion of duplicated genes results in less severe fitness phenotypes than deletion of singleton genes [6]. It has been hypothesized that duplicate gene pairs with high sequence similarity are more likely to be functionally redundant and contribute to genetic robustness against deleterious mutations, whereas duplicate gene pairs with low sequence similarity have diverged to such an extent to no longer be able to functionally complement each other. However, there is no correlation between sequence similarity between duplicates and their contribution to genetic robustness [7]. Indeed, regardless of sequence divergence, gene duplicates arising from a whole genome duplication in S. cerevisiae are less likely than singleton genes to be essential [8]. However, it remains unclear how duplicated genes that have diverged from each other in sequence and function can provide genetic robustness against deleterious mutations. Previous genome-wide studies have been limited in their ability to deduce a molecular mechanism for gene duplication in genetic robustness because phenotypes were assessed without regard to gene function. In this study, we have investigated in detail how the nonredundant duplicated gene pair HST1 and SIR2 in Saccharomyces cerevisiae functions to provide genetic robustness against null mutation.
In S. cerevisiae, Hst1p is an NAD+-dependent histone deacetylase that acts with the protein Sum1p to repress a subset of midsporulation genes [9–12]. Hst1p deacetylates histones H3 and H4 [9], and this deacetylation is thought to be important for its repressive function. Sum1p is a DNA binding protein that associates with the middle sporulation element (MSE), a conserved sequence found primarily in midsporulation gene promoters [12–14]. The third member of the Sum1 complex, Rfm1p, is a small protein thought to serve an architectural role by associating with both Sum1p and Hst1p [10]. There are two noteworthy differences among phenotypes of sum1Δ, rfm1Δ, and hst1Δ null mutations. First, the subset of midsporulation genes derepressed in these backgrounds differs. One group of genes requires Rfm1p and Hst1p in addition to Sum1p; the other group requires only Sum1p for repression [10]. Second, for genes that are repressed by both Sum1p and Hst1p, there is a difference in the level of derepression of target loci between sum1Δ and hst1Δ strain backgrounds [10]. Deletion of Sum1p results in a strong derepression of target midsporulation genes, whereas deletion of Hst1p results in a modest derepression of the same target midsporulation genes. It has been unclear what contributes to this difference in phenotypes.
Hst1p is a member of the Sir2 family of NAD+-dependent deacetylases. These enzymes have a highly conserved catalytic core domain and variable terminal extensions. Deacetylases of the Sir2 family are present ubiquitously across life, with family members in bacterial, archael, plant, fungal, and animal species [15]. Biological functions of these family members are diverse, with roles in transcriptional silencing, chromosome stability, cell cycle progression, and aging [16]. The distinct and variable functions of Sir2 family members are a result of multiple duplication events with subsequent diversification of substrates and functions. In S. cerevisiae there are five NAD+-dependent deacetylases: Sir2p, the founding member of the entire family; Hst1p; Hst2p; Hst3p; and Hst4p [17–19]. Hst2p is a predominantly cytoplasmic protein [20], but may have a cell cycle-specific nuclear localization [21,22]. Hst3p and Hst4p deacetylate lysine 56 on histone H3 and are involved in cell cycle and DNA damage checkpoints that modulate chromatin, enabling replication and condensation to occur properly [17,23,24].
Of these five NAD+-dependent deacetylases in S. cerevisiae, HST1 is the most closely related to SIR2. HST1 and SIR2 arose in a whole genome duplication in the ancestry of Saccharomyces species, which occurred approximately 100 million years ago [25–27]. Overall sequence conservation between SIR2 and HST1 is 63% (76% similar) [17] with three conserved regions: the well-conserved catalytic core domain with 82% sequence identity (92% similarity) and lesser-conserved regions in the N terminus and the extreme C-terminal tail [28]. Despite their sequence similarity, HST1 and SIR2 have non-overlapping functions [10,17,18]. We have used SIR2 and HST1 as a case study to understand diversification of the Sir2 family through duplication.
In contrast to the promoter-specific mechanism of transcriptional repression in which Hst1p participates, Sir2p is involved in long-range silencing. Sir2p acts with Sir3p and Sir4p to generate a special chromatin structure that silences the mating-type loci and telomeres [29]. Cis-acting silencer elements recruit the four Sir proteins. Then, Sir2p, Sir3p, and Sir4p spread along the chromosome [30–32]. The histone deacetylase activity of Sir2p is required for the spreading of all three Sir proteins [30,32]. Sir3p and Sir4p bind preferentially to deacetylated tails of histones H3 and H4 [33]. Sir2p deacetylates nearby nucleosomes, creating new high affinity binding sites for Sir3p and Sir4p, which in turn recruit additional Sir2p to the newly deacetylated nucleosomes. As the Sir proteins spread, they generate a specialized chromatin structure that is restrictive to transcription and independent of DNA sequence. Sir2p is also part of the RENT complex (Regulator of nucleolar silencing and telophase exit), which modulates chromatin structure in the rDNA repeats [34]. The RENT complex does not contain the other Sir proteins, and its mechanism of action is less well understood.
We examined whether the difference in derepression of midsporulation genes between sum1Δ and hst1Δ strains is a consequence of Sir2p, the closest paralog to Hst1p substituting for Hst1p in the absence of Hst1p. We have shown through genetic and biochemical means that Sir2p can substitute for Hst1p, and this phenomenon is a product of the path of evolutionary divergence after the duplication of these two genes.
Gene expression data indicate that deletion of HST1 derepresses target genes modestly, compared to the level of derepression observed in a sum1Δ background (unpublished data) [10]. These results suggest that either deacetylation is not critical for gene repression or another deacetylase acts at these promoters in the absence of Hst1p. To identify other deacetylases that may function in the absence of Hst1p, the four other known NAD+-dependent deacetylases, SIR2, HST2, HST3, and HST4 were deleted in combination with HST1. To assay levels of expression in these double deletion backgrounds, a pGAS2-HIS3 reporter was used. The GAS2 promoter is not strongly induced in the absence of Hst1p but is greatly induced in the absence of Sum1p [10]. In addition, the promoter contains a MSE and is reported to bind Sum1p [35]. If another deacetylase contributes to repression at this promoter in the absence of Hst1p, then deletion of both deacetylases should derepress the pGAS2-HIS3 reporter to a greater extent than deletion of Hst1p alone. Increased expression was observed in the hst1Δ sir2Δ double deletion strain compared to the hst1Δ strain (Figure 1A). The other double deletions, hst1Δ hst2Δ, hst1Δ hst3Δ, and hst1Δ hst4Δ, did not display any difference in derepression compared to the single hst1Δ background.
To extend this observation and examine more quantitatively the difference between hst1Δ and hst1Δ sir2Δ derepression phenotypes, gene expression levels of DTR1 and SPS1, two midsporulation genes repressed by Sum1p and Hst1p [10], were measured by quantitative reverse transcriptase (RT)-PCR in wild-type, hst1Δ, and hst1Δ sir2Δ strains. DTR1 and SPS1 were modestly induced in an hst1Δ background (Figure 1B) in accordance with previous observations [10]. Consistent with the results of the pGAS2-HIS3 reporter (Figure 1A), the induction of DTR1 and SPS1 in an hst1Δ sir2Δ strain was dramatically greater than in an hst1Δ strain (Figure 1B). It should be noted that although derepression of midsporulation genes in an hst1Δ sir2Δ background was greater than was observed in an hst1Δ background, this derepression was not to the level observed in a sum1Δ strain (unpublished data). These results indicated that Sir2p contributed to the repression of midsporulation genes in the absence of Hst1p.
To determine whether the increased expression of Hst1p-repressed loci in an hst1Δ sir2Δ background resulted specifically from the loss of Sir2p or was an indirect effect due to the disruption of Sir-mediated silencing, the induction of DTR1 and SPS1 was examined in an hst1Δ sir3Δ background. If the observed increased expression resulted from the loss of Sir-mediated silencing, then the hst1Δ sir3Δ strain should have the same level of DTR1 and SPS1 induction as the hst1Δ sir2Δ strain. On the other hand, if the increased gene expression observed in the hst1Δ sir2Δ strain resulted specifically from the loss of Sir2p, then retaining Sir2p while disrupting Sir-mediated silencing should resemble the hst1Δ phenotype rather than the hst1Δ sir2Δ phenotype. The level of DTR1 and SPS1 induction in the hst1Δ sir3Δ strain was comparable to the hst1Δ strain and dramatically less than the hst1Δ sir2Δ strain (Figure 1B). We conclude that it was the absence of the Sir2p deacetylase and not disruption of Sir-mediated silencing that contributed to the elevated level of DTR1 and SPS1 gene expression in the hst1Δ sir2Δ background.
It is possible that Sir2p always contributes to the repression of the midsporulation genes. Alternatively, the absence of Hst1p could provide an opportunity for Sir2p to associate with the Sum1 complex, such that Sir2p only contributes to this repression in the absence of Hst1p. To test the latter hypothesis, we characterized DTR1 and SPS1 expression in a strain in which Hst1p was enzymatically inactive, such that the mutant Hst1p could not contribute to deacetylation yet was present and could physically block Sir2p from acting in its place. To inactivate Hst1p, a single amino acid substitution, N291A (described in [36]), was used, analogous to a characterized substitution in Sir2p (N345A), which has been shown to be enzymatically inactive [37] but structurally intact [38]. This point mutation in Hst1p reduced deacetylation in vivo (as discussed in a subsequent subsection of the results). The hst1-N291A strain displayed significantly greater induction of DTR1 and SPS1 compared to the hst1Δ strain (Figure 1B). This observation suggests that Sir2p may be acting in the absence of Hst1p, but not when the mutated Hst1-N291Ap is present. Furthermore, the increased induction in the hst1-N291A strain compared to the hst1Δ strain indicates that the main function of Hst1p in repression is deacetylation.
To examine whether Sir2p normally contributes to repression of DTR1 and SPS1, the expression profiles of DTR1 and SPS1 in a sir2Δ background were analyzed. If Sir2p has no role in Hst1p-mediated repression when Hst1p is present, then deleting SIR2 alone should have no discernable phenotype compared to wild-type yeast, and repression of DTR1 and SPS1 should be maintained. Repression of DTR1 and SPS1 was maintained in a sir2Δ background (Figure 1B), and the pGAS2-HIS3 reporter also remained repressed in a sir2Δ strain (unpublished data). These results suggest that Sir2p does not normally play a role in Sum1p-mediated repression when Hst1p is present.
To further test the hypothesis that Sir2p substitutes for Hst1p but does not normally act with Sum1p, the association of Sir2p with repressed promoters in the presence and absence of Hst1p was examined. If the substitution model is correct, Sir2p should not be enriched at repressed promoters when Hst1p is present (in wild-type or hst1-N291A strains) but should be recruited to these promoters in an hst1Δ background. Chromatin immunoprecipitation (ChIP) was used to detect HA-Sir2p or Hst1p-HA at the DTR1 promoter in wild-type and hst1Δ strains. In a wild-type background, there was a high level of Hst1p-HA enrichment but no detectable enrichment of HA-Sir2p at the promoter of DTR1 (Figure 2A). There was also no enrichment of HA-Sir2p observed in the hst1-N291A background (Figure 2A). These results are consistent with the model that Sir2p is absent from these promoters when Hst1p is present. However, when Hst1p was absent, there was a modest enrichment of HA-Sir2p at the DTR1 promoter (Figure 2A). The enrichment of Sir2p was not as robust as wild-type Hst1p at these loci, suggestive of a weaker interaction between Sir2p and the Sum1 complex.
To examine directly whether the recruitment of Sir2p to repressed promoters is due to an interaction with the Sum1 complex, co-immunoprecipitation experiments between Sum1p and HA-Sir2p in an hst1Δ background were performed. Hst1p and Sum1p are part of a stable complex that coprecipitates (Figure 2B) [10,11,39]. If Sir2p substitutes for Hst1p via a similar interaction with the Sum1 complex, then a physical association between these two proteins should be detectable. Sum1p was immunoprecipitated, and the immunoprecipitation samples were probed for HA-Sir2p by immunoblotting. Consistent with the substitution model, Sir2p associated with Sum1p in the hst1Δ background (Figure 2B). This coprecipitation of Sir2p with Sum1p was weaker than the precipitation observed for Hst1p from an equivalent amount of cell extract. This qualitative comparison is consistent with the Hst1p–Sum1p interaction being more robust than the Sir2p–Sum1p interaction and in accordance with the reduced enrichment of Sir2p compared to Hst1p observed at the promoter of DTR1 (Figure 2A).
To test the hypothesis that the presence of Hst1p physically blocks the association of Sir2p with the Sum1 complex, the Sir2p–Sum1p interactions in wild-type and hst1-N291A backgrounds were examined. In the presence of Hst1p, Sir2p would not be expected to interact with the Sum1 complex, and indeed Sir2p was not observed to coprecipitate with Sum1p in wild-type yeast. There was a faint band in the hst1-N291A background that could be indicative of Sir2p interaction with the Sum1 complex, however this band was considerably less robust than that observed in the hst1Δ strain (Figure 2B). Therefore, we conclude that Sir2p is recruited to Hst1p-repressed loci through an interaction with the Sum1 complex and this recruitment only occurs in the absence of Hst1p. Presumably, Hst1p outcompetes Sir2p for association with the Sum1 complex because Hst1p has a higher affinity for the Sum1 complex.
The results in the previous section suggested that Sir2p can substitute for Hst1p but does not normally act with Sum1p. Gene expression data (Figure 1), in addition to the physical interactions described above (Figure 2), do not support the hypothesis that Sir2p plays a role in Sum1p-mediated repression when Hst1p is present. Instead, these results support the hypothesis that Sir2p and Hst1p have nonoverlapping functions in wild-type backgrounds [10,17,18].
To investigate whether Sir2p acts as a deacetylase at Sum1p-repressed promoters, ChIP experiments were performed with two different histone H4 antibodies, one specific for acetylated lysine 8 (K8) and the other specific for acetylated lysine 16 (K16). The changes in acetylation of K8 or K16 at the DTR1 promoter in hst1Δ, hst1-N291A, hst1Δ sir2Δ, and sir2Δ strains relative to a wild-type strain were analyzed. Loss of deacetylation by Hst1p and Sir2p at DTR1, such as in hst1-N291A and hst1Δ sir2Δ backgrounds, should result in increased acetylation of K8 and K16. Indeed, increased levels of acetylation of both K8 and K16 were observed in both hst1-N291A and hst1Δ sir2Δ backgrounds (Figure 3). These results parallel the patterns observed in our gene expression profiles. The single sir2Δ deletion did not display elevated levels of acetylated K8 or acetylated K16 at DTR1, providing further support for the model that Sir2p does not normally act at Hst1-repressed loci when Hst1p is present.
Interestingly, when the hst1Δ and wild-type strains were compared, changes in acetylation were different for K8 and K16 (Figure 3). A modest increase of acetylation at K8 was observed in the hst1Δ background, whereas no detectable change (compared to a wild-type strain) in acetylation of K16 was noted. These results suggest that K16 is more efficiently deacetylated by Sir2p than K8 because changes in K16 acetylation were only revealed when both Hst1p and Sir2p were absent (hst1Δ sir2Δ). These data are consistent with published reports that Sir2p preferentially deacetylates H4 K16 in vitro [37,40]. Nevertheless, K16 must also be a target for deacetylation by Hst1p because K16 acetylation increased when Hst1p was nonfunctional (hst1-N291A) (Figure 3). In conclusion, these results indicate that Sir2p acts as a deacetylase at Sum1p-repressed promoters in the absence of Hst1p.
To determine whether there is still sufficient Sir2p available to silence the mating-type loci (a primary function of Sir2p) when Sir2p is substituting for Hst1p, the ability of wild-type, sir2Δ, and hst1Δ strains to mate was assessed. If Sir2p recruitment to Sum1p-repressed loci in an hst1Δ background reduces the pool of available Sir2p, then silencing at the mating-type loci might be reduced, leading to diminished mating. Alternatively, if the preferred function of Sir2p is to silence the mating-type loci, then there should be no defect in mating ability in an hst1Δ background, even though Sir2p is substituting for Hst1p. There was no observable defect in mating ability in an hst1Δ background compared to a wild-type strain (Figure 4A) [17,18]. Therefore, Sir2p is more likely to silence the mating-type loci than to substitute for Hst1p. Furthermore, these results suggest that Sir2p has a higher affinity for the Sir complex than the Sum1 complex, because the ability to mate is not perturbed in the absence of Hst1p, whereas repression of midsporulation genes is not complete when Sir2p is substituting for Hst1p.
If the majority of Sir2p is involved in silencing the mating-type loci (and telomeres), and only a few molecules of Sir2p are available for recruitment to Hst1p-repressed loci in the absence of Hst1p, then additional copies of Sir2p may enhance repression of Hst1p-repressed loci. Overexpression of Sir2p has been reported to reduce β-Galactosidase activity from an MSE-containing promoter driving lacZ expression in an hst1Δ background [12]. To further characterize this observation, the amount of Sir2p in the cell was varied to determine whether overexpression of Sir2p enhanced its ability to substitute for Hst1p. To assay repression, a reporter construct consisting of the Sum1-repressed PES4 promoter fused to the open reading frame of HIS3 was utilized. In the absence of Hst1p, the PES4 promoter is derepressed to a greater extent than the GAS2 promoter described previously (Figure 1A), enabling an enhancement of repression to be detected. hst1Δ cells were transformed with low copy plasmids expressing HST1-HA or HA-SIR2 and a high-copy plasmid expressing SIR2. The relative levels of Sir2p are shown in Figure 4C. Expression of the pPES4-HIS3 reporter was monitored on medium lacking histidine and uracil (to ensure plasmid retention). A wild-type strain displayed no growth on selective medium, indicating that the PES4 promoter was repressed as expected (Figure 4B). In an hst1Δ background, cells were able to grow on selective medium as a result of derepression of the pPES4-HIS3 reporter (Figure 4B), demonstrating that the reporter assay is functional. Note that endogenous levels of Sir2p are present in all strains (Figure 4C). The addition of Sir2p on a low copy plasmid resulted in an enhancement of repression of pPES4-HIS3, and overexpression of Sir2p from a high copy plasmid enhanced repression of pPES4-HIS3 to an even greater extent. Despite the enhancement in repression observed upon overexpression of Sir2p, repression of pPES4-HIS3 was not complete in the absence of Hst1p. This incomplete suppression probably results from the relatively weaker interaction of the Sum1 complex with Sir2p compared to Hst1p.
It is thought that Sir2p associates directly with Sir4p but not Sir3p [41]. Therefore, additional Sir2p might become available by deleting Sir4p, which would result in a stronger repression phenotype than observed in an hst1Δ background. However, gene expression analysis of DTR1 in an hst1Δ sir4Δ strain showed roughly equivalent levels of DTR1 induction to an hst1Δ strain (unpublished data).
A key distinguishing feature between Hst1p and Sir2p is that Sir2p is normally part of the Sir-silencing complex that spreads along the chromosome [29], whereas the Hst1p–Sum1p complex does not spread [36]. We were interested to determine whether the Sir2p–Sum1p complex was able to spread, although the Hst1p–Sum1p complex does not spread, indicating some intrinsic property in Sir2p to promote spreading. To assess the ability of Sum1p to spread, the distribution of myc-Sum1p across the DTR1 locus was analyzed by ChIP when the Sum1 complex was interacting with Hst1p (wild-type cells) or Sir2p (hst1Δ cells). There is a probable MSE sequence in the promoter of DTR1 to which Sum1p is thought to bind (Figure 5A) [35]. When the Sum1p–Hst1p complex is present, myc-Sum1p should associate most strongly with the MSE DNA sequence and should have reduced association with the surrounding sequences (approximately 200 bp upstream and downstream of the MSE). Due to the technical limitations of shearing DNA by sonication, sequences near the binding site are also enriched in immunoprecipitated material, even if the protein does not spread. If Sir2p causes Sum1p to spread when it substitutes for Hst1p, myc-Sum1p should be more broadly distributed across the DTR1 promoter and into the open reading frame. However, if Sir2p does not confer the ability to spread, then the distribution of myc-Sum1p across DTR1 should not be appreciably different in HST1 and hst1Δ strains. The distribution of myc-Sum1p across the DTR1 locus remained the same regardless of which deacetylase was interacting with Sum1p (Figure 5B). Therefore, Sir2p did not cause noticeable spreading of Sum1p when substituting for Hst1p, and Sum1p continued to act as a promoter-specific repressor.
We extended this analysis to examine whether Sir2p itself can spread across the DTR1 locus, even though Sum1p does not spread. The distribution of Hst1p-HA (in a wild-type background) and HA-Sir2p (in an hst1Δ background) across the DTR1 locus was assessed by ChIP. As expected, Hst1p-HA had a distribution centering around the MSE and did not extend into the open reading frame (Figure 5C), indicating that Hst1p is not spreading at repressed midsporulation genes. The localization of HA-Sir2p had a similar distribution that was centered at the MSE and did not extend into the open reading frame (Figure 5D). These results demonstrate that Sir2p can act in a promoter-specific manner to repress gene expression when associated with the Sum1 complex.
How do Hst1p and Sir2p maintain nonoverlapping functions when both deacetylases are present, despite considerable sequence identity and the ability of Sir2p to substitute for Hst1p? One possibility is that Hst1p and Sir2p have unique determinants that confer specificity for the Sum1 complex and the Sir complex, respectively. Because the N terminus is less conserved than the catalytic core (Figure 6A), this region may have evolved distinct specificities for either the Sir or Sum1 complex. To determine whether such determinants exist, a chimeric Sir2–Hst1p molecule was constructed in which the N terminus of Sir2p was fused to the catalytic core of Hst1p. The junction of the Sir2–Hst1 chimera was at the start of the catalytic core domain, such that amino acids 1–255 of Sir2p were fused to amino acids 201–503 from Hst1p (Figure 6A), generating HA-Sir21−255-Hst1201−503p. This chimeric gene was expressed from the SIR2 promoter. We also constructed the reverse chimera, HA-Hst11−200-Sir2256−562p, but were not able to detect protein expression by immunoblotting (unpublished data) and continued analysis only with HA-Sir21−255-Hst1201−503p.
This chimeric protein was tested for its ability to function like Sir2p and Hst1p. If specificity for the Sir complex (in Sir2p) and specificity for the Sum1 complex (in Hst1p) is established by the N terminus, then HA-Sir21−255-Hst1201−503p should only function like Sir2p. Alternatively, if specificity for the Sir or Sum1 complex is determined by the C terminus, then HA-Sir21−255-Hst1201−503p should associate with the Sum1 complex and function like Hst1p. Finally, it may be that specificity for the Sir and Sum1 complex may be determined in entirely different regions of Sir2p and Hst1p. If this were true, then it may be possible that HA-Sir21−255-Hst1201−503p can associate with both the Sir and Sum1 complexes and function in an Hst1p- and Sir2p-like manner or interact with neither complex, and HA-Sir21−255-Hst1201−503p would be nonfunctional in both Sir- and Sum1-mediated repression.
To determine whether HA-Sir21−255-Hst1201−503p has Hst1p-like function, the ability of this chimera to repress pPES4-HIS3 was examined. HA-Sir21−255-Hst1201−503p completely suppressed an hst1Δ mutation (Figure 6B). In fact, the chimeric protein was more effective than Sir2p at repressing pPES4-HIS3 (Figure 6B compared to Figure 4B). However, Sir2p is still present in these cells and could affect the results. To directly compare the abilities of HA-Sir21−255-Hst1201−503p, Hst1p-HA, and HA-Sir2p to function in an Hst1p-like manner, each of these proteins was expressed in an hst1Δ sir2Δ strain, and DTR1 expression was assayed by quantitative RT-PCR. The chimera was better at repressing DTR1 expression than Sir2p and was equally able to repress DTR1 expression as wild-type Hst1p (Figure 6C).
To determine whether this repression mediated by HA-Sir21−255-Hst1201−503p resulted from a strong interaction with the Sum1 complex, co-immunoprecipitation studies were performed. When Sum1p was immunoprecipitated, HA-Sir21−255-Hst1201−503p coprecipitated to an extent comparable to, if not greater than, wild-type Hst1p and was much greater than Sir2p (Figure 6D). From these results we propose that unique features in the C terminus of Hst1p specify an interaction with Sum1p.
To determine whether HA-Sir21−255-Hst1201−503p has Sir2p-like function, the ability to silence the mating-type loci was examined by mating assays. HA-Sir21−255-Hst1201−503p enabled the cells to mate in the absence of Sir2p to a level comparable to that seen with wild-type Sir2p (Figure 7A). The extent of mating was greater with HA-Sir21−255-Hst1201−503p than in cells expressing only wild-type Hst1p (Figure 7A). To test if the mating ability of HA-Sir21−255-Hst1201−503p resulted from an association with the Sir complex, we co-immunoprecipitated Sir4p with HA-Sir2p, Hst1p-HA, and HA-Sir21−255-Hst1201−503p. Results from these experiments showed an interaction between HA-Sir21−255-Hst1201−503p and Sir4p comparable to that of wild-type Sir2p (Figure 7B). We conclude that there is a critical component in the N terminus of Sir2p that specifies Sir2p to interact with the Sir complex. The ability of the chimeric Sir2–Hst1 protein to suppress both hst1Δ and sir2Δ mutations suggests that different regions of the protein are involved in conferring specificity for the Sir and Sum1 complexes.
A recent study [42] also analyzed Sir2p and Hst1p interaction domains by using chimeric molecules and obtained similar results. This study determined that amino acids 12–209 in the N terminus of Sir2p were important for recruiting the protein to the Sir complex, consistent with our chimera analysis. Additionally, it was shown that two nonconserved amino acids in the catalytic core of Hst1p, Q324, and I325, were critical for recruitment to the Sum1 complex. Together, these results strongly indicate the presence of two different domains in Sir2p and Hst1p that confer substrate specificity for the Sir or Sum1 complex.
As outlined in the introduction, SIR2 and HST1 arose by gene duplication, and it is possible that the ancestral deacetylase interacted with both the Sum1 and Sir complexes. To test this model, we examined the function of the single SIR2/HST1 gene from K. lactis, a species known to have diverged from S. cerevisiae before the whole genome duplication [25,26]. If the ancestral SIR2/HST1 gene possessed only the function of ScSIR2 or ScHST1 and the other function evolved after the duplication, the gene having the new function would be expected to have experienced accelerated evolution compared with the gene retaining the original function. However, there appears to have been no accelerated evolution of either ScSir2p or ScHst1 compared to KlSir2p (Figure 8A) [26], an observation more consistent with a partitioning of functions after the duplication.
The initial identification of KlSIR2/HST1 (referred to hereafter as KlSIR2) reported that overexpression of KlSIR2 in S. cerevisiae was able to partially suppress a sir2Δ mating defect [43]. We did not observe suppression of a sir2Δ mating defect by KlSir2p (unpublished data); however, this could be due to differences in expression between our work and previously reported findings. Nevertheless, we could detect a weak interaction between KlSir2p and ScSir4p in co-immunoprecipitation experiments (Figure 8D). Subsequent studies of KlSIR2 revealed a role in silencing the mating-type loci in K. lactis [44]. Therefore, it has clearly been demonstrated that KlSIR2 has SIR2-like function both in K. lactis as well as in S. cerevisiae.
To test whether KlSIR2 is able to function in Hst1p-mediated repression in S. cerevisiae, KlSIR2 was cloned into a low-copy plasmid such that KlSIR2 has an N-terminal HA epitope tag and is expressed from the ScSIR2 promoter. This plasmid was used to transform the hst1Δ pPES4-HIS3 strain of S. cerevisiae. Immunoblot analysis showed that KlSIR2 was stably expressed in S. cerevisiae (Figure 8D). There was complete repression of the pPES4-HIS3 reporter by HA-KlSIR2, with no observed difference from the wild-type level of repression (Figure 8B). We also examined the ability of KlSir2p to repress midsporulation genes in an hst1Δ sir2Δ background by analyzing DTR1 expression levels and found that KlSir2p repressed DTR1 to a level comparable to ScHst1p and better than ScSir2p (Figure 8C).
To test if this repression was a result of an interaction with the S. cerevisiae Sum1 complex, we immunoprecipitated ScSum1p and found that HA-KlSir2p coprecipitated (Figure 8D). From this data, as well as studies described previously [43,44], we conclude that the preduplicated KlSIR2 carries out both SIR2- and HST1-like functions. These data provide further evidence that the ancestral SIR2/HST1 had dual functions that diverged after duplication.
In this study we provide evidence that in S. cerevisiae, the deacetylase Sir2p substitutes for Hst1p in its absence. Deletion of both HST1 and SIR2 results in dramatically greater derepression of Sum1p-repressed genes than observed in an hst1Δ background. This additional derepression is not observed when other related NAD+-dependent deacetylases or other silencing factors are deleted (Figure 1). Furthermore, Sir2p localizes to the promoters of Hst1p-repressed loci through an interaction with the Sum1 complex (Figure 2) and acts as a histone deacetylase at these loci in the absence of Hst1p (Figure 3). Based on these results, we propose a substitution model rather than genetic redundancy to explain the overlapping roles of this duplicate gene pair. We consider this phenomenon as substitution and not redundancy because Sir2p-mediated repression at Hst1-regulated genes is not as proficient as wild-type Hst1p-mediated repression and only occurs when Hst1p is absent. We propose that this substitution by Sir2p in an hst1Δ background accounts for some of the difference in derepression observed between hst1Δ and sum1Δ strains, although Sir2p substitution did not account for the entire difference in derepression phenotypes between hst1Δ and sum1Δ strains. It is possible that another deacetylase may also have limited ability to substitute, or it could be that Sum1p has repressive properties that are independent of a deacetylase.
It has also been observed by others that Hst1p substitutes for Sir2p in a sir2Δ background. Overexpression of Hst1p from a high copy plasmid partially suppresses sir2Δ mating defects in MATα cells [17]. However, this suppression does not completely restore function, as genetic redundancy would predict, because mating efficiency is still about 30-fold lower than in a wild-type strain. Given that overexpression of Hst1p is required to observe this effect, Hst1p may be less capable of substituting for Sir2p than Sir2p is capable of substituting for Hst1p. The requirement for multiple Sir2p–Sir4p complexes to silence a single HMR locus may also reduce the ability of Hst1p to substitute for Sir2p. This is in contrast to what is considered to be a single Hst1p–Sum1p complex required for repression, which would make it easier for Sir2p to substitute for Hst1p, even if the affinity of Hst1p for the Sir complex were comparable to the affinity of Sir2p for the Sum1 complex. Regardless, this previously published result is consistent with our model of the duplicated SIR2–HST1 gene pair acting as substitutes for each other. This type of biological phenomenon has been proposed previously as the imposter model, with some controversy, for the MAP kinases Fus3p and Kss1p in S. cerevisiae [45,46]. However, our study has further developed this model to consider the evolutionary relationships between substituting proteins.
This substitution is likely a consequence of SIR2 and HST1 originating by duplication. Duplication has been proposed to be a strong evolutionary force because it generates a source of new genetic material that is free of selective constraint [1]. Duplicated genes have two ultimate fates: degeneration or preservation in the genome. Clearly SIR2 and HST1 have been retained. Two models have been proposed outlining the steps towards preservation. The classical model proposed that the only way to preserve duplicated genes is through neofunctionalization, in which one of the duplicate genes evolves a new function by acquiring beneficial mutations, while the other gene retains the original function. In such a case, it is predicted that the gene with the new function will experience a more rapid change in sequence, i.e., “accelerated evolution,” compared with the duplicate retaining the original function. A more recent paradigm for the preservation of duplicated genes has been proposed [47,48] to account for the much larger retention of duplicate genes than the classical model would predict. This new model of duplication, degeneration, and complementation (DDC) states that if the ancestral gene had multiple functions, duplicate genes can each lose one of the original functions by degenerative mutations, while still retaining a different ancestral function. The DDC mechanism was originally proposed in the context of cis-regulatory elements of duplicated gene pairs. However, our work suggests that the DDC mechanism can also act on protein coding sequences.
This study provides evidence to suggest that the ancestral SIR2–HST1 gene provided both SIR2- and HST1-like functions. After the duplication, SIR2 and HST1 subfunctionalized to evolve into distinct SIR2 and HST1 genes with non-overlapping functions. By using K. lactis as a representative nonduplicated species, we found that the single HST1/SIR2 gene completely suppressed an hst1Δ mutation in S. cerevisiae (Figure 8). Previous studies have reported that KlSIR2 contributes to silencing the HM loci in K. lactis [44] and can partially suppresses a sir2Δ mating defect in S. cerevisiae [43]. We have extended this analysis to show that KlSir2p can interact with Sir4p in S. cerevisiae (Figure 8). Together, these results indicate that the preduplicated HST1/SIR2 is likely to have had both functions. It is probable that the Sir2 family has diversified by this type of mechanism.
Results from our functional characterization of the chimeric Sir21−255-Hst1201−562 as well as those reported elsewhere [42] provide preliminary evidence that the evolution of SIR2 and HST1 may have followed a DDC mechanism. Two different specificity determinants in Sir2p and Hst1p have been found; a domain specific for determining an interaction with the Sum1 complex residing in the C terminus of Hst1p (Figure 6), specifically Q324 and I325 [42], and a domain specific for conferring an interaction with the Sir complex in amino acids 12–209 of the N terminus of Sir2p (Figure 7) [42]. These interaction domains have likely been conserved subsequent to the duplication. We propose a scenario in which, after the duplication, Hst1p acquired degenerative mutations in the N-terminal domain that interacts with the Sir complex, leading to the loss of affinity for Sir4p, yet maintaining its ability to interact with the Sum1 complex. Sir2p, on the other hand, acquired degenerative mutations in the C-terminal domain required for interaction with the Sum1 complex, leading to reduced affinity for the Sum1 complex, while maintaining a strong interaction with the Sir complex. Nevertheless, Sir2p has retained an interaction domain for the Sum1 complex, although it has a weaker affinity for this complex than Hst1p (Figure 2). Interestingly, of the two amino acids Q324 and I325, important for Hst1p specificity for the Sum1 complex in S. cerevisiae, only the isoleucine is conserved in K. lactis (K434 and I435). However, KlSir2p can fully suppress an hst1Δ mutation in S. cerevisiae (Figure 8). There may be additional residues in Hst1p that confer an interaction with the Sum1 complex but are conserved between ScHst1p, ScSir2p, and KlSir2p.
In this study we suggest that the particular evolutionary path taken as duplicated genes diverge from one another may be an important indicator of their potential contribution to genetic robustness. Duplicates that have subfunctionalized through a DDC mechanism may be more likely to substitute for each other than duplicates that display accelerated evolution or neofunctionalizion. SIR3 and ORC1 represent a pair of duplicated genes arising from the whole genome duplication that, in contrast to SIR2 and HST1, experienced accelerated evolution [26]. Orc1p is an essential component of the origin recognition complex. Deletion of ORC1 results in lethality, and Sir3p cannot complement an orc1 mutation. Likewise, Orc1p cannot suppress a sir3Δ mating defect [49]. ORC1 and SIR3 are clearly an example of a duplicated gene pair that does not provide genetic robustness.
These results illustrate how gene duplication can provide genetic robustness against null mutations. It has been shown in S. cerevisiae that genes with duplicates are significantly more likely to have a weaker fitness defect phenotype compared to nonduplicated genes [6,50]. Here, we present data revealing that duplication provides genetic robustness through substitution not redundancy. This is an important distinction because about 550 duplicated gene pairs in S. cerevisiae were retained after the genome duplication [25,26], and many of these duplicates have diverged from each other [26]. It is quite likely that there are other duplicate genes, in addition to SIR2 and HST1, which in wild-type backgrounds have nonoverlapping functions, yet, are able to substitute for one another in the event of a deletion. The biological significance of this phenomenon will be reflected in a null phenotype that underestimates or masks the real function of the deleted gene. Thus, one should apply caution in interpreting deletion phenotypes, particularly if it is known that the gene of interest has a retained duplicate. Our study also demonstrates that, in the case of an enzyme, the use of an inactivating mutation that abolishes enzymatic activity may be more useful in characterizing protein function than a complete deletion because such inactivating mutations retain the protein in the cell and thereby prevent an alternative protein from taking its place.
Finally, we can draw some conclusions about the relationship between different transcriptional repression mechanisms. It is clear from this study that deacetylation is an important component of Sum1p-mediated repression, as it is in Sir-mediated silencing. However, there is no intrinsic property of the deacetylase that determines whether it will act in a promoter-specific or regional manner (Figure 6). The results described here are consistent with our previous results indicating that a mutant form of Sum1p does spread and that this spreading requires the deacetylase activity of Hst1p rather than Sir2p [36]. Therefore, the tendency for a repressor complex to spread or not to spread is probably a function of the DNA or histone binding proteins with which the histone deacetylase associates. Sir2p is able to spread because its partners, Sir3p and Sir4p are able to spread. In fact, Sir3p and Sir4p can spread in the absence of Sir2p deacetylase activity when the histone tails mimic a deacetylated state [51], supporting the model that the role of Sir2p is to provide a substrate for its partners to bind. In contrast to Sir2p, Hst1p does not spread because its partner, Sum1p, normally does not spread [36]. This model is consistent with the hypothesis that the single ancestral histone deacetylase associated with both spreading (Sir) and nonspreading (Sum1) complexes.
Strains used in this study were all derived from W303-1a (Table 1). The hst1Δ::KanMX, HST1-HA, myc-SUM1 [39], and hst1-N291A [36] alleles were described previously. The sir3Δ::LEU2 and sir2Δ::URA3 alleles were obtained from J. Rine (unpublished data). The sir2Δ::TRP1, hst2Δ::TRP1, hst3Δ::TRP1 and hst4Δ::TRP1 alleles were complete deletions of the open reading frames generated by one-step gene replacement. To generate the pGAS2-HIS3 and pPES4-HIS3 reporter alleles, the open reading frames of GAS2 and PES4 were replaced precisely with the HIS3 open reading frame. The correct integration was confirmed by PCR using primers flanking the sites of recombination. These alleles were moved into various genetic backgrounds (as described in Table 1) through standard genetic crosses.
Plasmids used in this study are described in Table 2. The plasmid containing HST1-HA (pLR30) has been previously described [39]. To generate plasmid pLR488 expressing the chimeric SIR2–HST1 protein, the N terminus of SIR2 (amino acids 1–255) was amplified from genomic DNA, with the 5′ primer containing the recognition site for EcoRI and the 3′ primer containing 20 base pairs of overlapping homology to the start of the catalytic core of HST1. The C terminus of HST1 (amino acids 201–503) was amplified from genomic DNA, with the 5′ primer containing 20 base pairs of overlapping homology to the SIR2 sequence just upstream of the catalytic core sequence and the 3′ primer containing the recognition sequence for AgeI restriction endonuclease. A second PCR reaction was performed in which equimolar amounts of the SIR2 N terminus amplicon and the HST1 C terminus amplicon were pooled in a 25-μl PCR reaction and allowed to run in the thermocycler for five cycles, after which an additional 25 μl of reaction mix containing the 5′ oligonucleotide used previously for the SIR2 amplification and the 3′ oligonucleotide used previously for the HST1 amplification were added to the initial PCR reaction and allowed to cycle for 25 more rounds. The PCR product was cloned into the EcoRI and AgeI sites of pRO298, thereby replacing the SIR2 open reading frame with the chimeric SIR2–HST1 gene while retaining the N-terminal HA tag. The correct plasmid was verified by restriction enzyme analysis and sequencing. Expression of the HA-Sir21−255-Hst1201−503 chimeric molecule was confirmed by immunoblotting. To generate plasmid pLR490 containing KlSIR2, KlSIR2 was amplified from genomic DNA from a wild-type K. lactis strain (SAY45, from S. Astrom). The 5′ primer contained an MfeI site, and the 3′ primer contained an AgeI site. The resulting PCR product was cloned into the EcoRI and AgeI sites of pRO298, thereby replacing the ScSIR2 sequence with KlSIR2. The correct plasmid was verified by restriction enzyme analysis and sequencing. Expression of HA-KlSir2p in S. cerevisiae was confirmed by immunoblotting.
RNA was isolated from two independent logarithmically growing cultures of each strain as previously described [52]. To remove DNA, 3 μg of RNA was treated with 3 units DNAseI (Promega, http://www.promega.com) and 1× DNase Buffer (Promega) in a final volume of 30 μl and incubated for 30 min at 37 °C. The DNase was inactivated by the addition of 3 μl of STOP solution (Promega) and incubation at 65 °C for 15 min. To verify that there was no contaminating DNA, 1 μl of DNAse-treated material was used in a PCR reaction containing primers to amplify the ACT1 transcript. A lack of product indicated successful removal of DNA. We used 1 μg of DNA-free RNA for cDNA synthesis by addition of 1 μl 10 mM dNTPs and 1 μl oligo dT16 (500 ng/μl) and incubation at 65 °C for 5 min, followed by a quick chill on ice. A master mix of 4 μl 5× first strand buffer (Invitrogen, http://www.invitrogen.com), 2 μl 0.1 M DTT, and 1 μl RNAseOUT (Invitrogen) was added to the samples. The resulting reaction was incubated at 37 °C for 2 min at which point 1 μl of M-MLV-RT (Invitrogen) is added to the reaction and incubated for 50 min at 37 °C followed by a 15-min incubation at 70 °C to inactivate the enzyme. We subsequently analyzed one-fortieth of the resulting cDNA by real time-PCR in the presence of SYBR Green on a Bio-Rad iCycler (http://www.bio-rad.com) to quantify the relative amounts of mRNA transcripts. Duplicate qPCR reactions were performed to ensure consistency. The standard curve was generated with genomic DNA isolated from the wild-type strain (W303-1a). Oligonucleotide sequences are provided in Table 3. Data were analyzed with iCycler iQ Optical System Software (Bio-Rad). DTR1 and SPS1 transcript levels were normalized to ACT1 transcript levels. To determine fold-induction, DTR1 and SPS1 transcript levels were normalized to the wild-type strain. Results reflect the average fold induction (relative to a wild-type strain) of two independent cultures for each strain background, each analyzed in duplicate qPCR reactions. The standard deviation was calculated from the difference in fold induction of the two independent cultures from the mean.
ChIPs were performed as previously described [39] using ten optical density equivalents of cells and 2–4 μl anti-HA tag antibody (Upstate Biotechnology 05–902, http://www.upstate.com), 2 μl of antibodies against acetyl-lysine 8 or acetyl-lysine 16 of histone H4 (Upstate Biotechnology 07–328 and 06–762), or 3 μl anti-myc tag antibody (Upstate Biotechnology 06–549). For immunoprecipitation of HA-Sir2p and Hst1p-HA, a second crosslinking agent was used [53]. A total of 50 optical density units of cells were harvested by centrifuging at 2,700 rpm for 5 min and resuspended in 1× ice cold DMA (10 mM dimethyl adipimidate, 0.1% DMSO, and 1× PBS) and crosslinked for 45–60 min at room temperature. After crosslinking with DMA, cells were washed twice with cold 1× PBS, resuspended in 50 ml 1× PBS, and treated with 1% formaldehyde for 45–60 min at room temperature. The DNA was sheared by sonication to an average size between 500 to 1,000 bp in all experiments. ChIP samples were analyzed by qPCR using a standard curve prepared from input DNA. The amounts of the immunoprecipitated DNA at the experimental promoter (DTR1) and a control promoter (ATS1) were determined relative to the input DNA, and then the enrichment of the DTR1 promoter was determined relative to the control locus ATS1. Enrichment is considered significant if the ratio of experimental to control region equals two or higher. Oligonucleotide sequences are provided in Table 3. Results reflect the relative immunoprecipitation of two independent cultures for each strain background, and the standard deviation was calculated from the difference in fold induction of the two independent cultures from the mean. To determine the relative acetylation level of Lys8 and Lys16 of H4 (Figure 3) in various strain backgrounds, normalized DTR1 IP levels were quantified relative to the wild-type strain. To determine differences in nucleosome occupancy in the various strain backgrounds, an independent ChIP using an antibody against the H4 core domain (Upstate Biotechnology 07–108) was performed. Results in Figure 3 depict relative acetylation levels of Lys8 and Lys 16 of H4 for each strain that accounts for strain differences in nucleosome occupancy.
Co-immunoprecipitations were performed as previously described [39] using 30 optical density equivalents of cells grown in media lacking uracil to ensure plasmid retention. The whole-cell lysates were incubated for 4 h at 4 °C with 5 μl of antibody. For Sum1p immunoprecipitations, serum from a guinea pig inoculated with a C-terminal fragment of Sum1p was used (Pocono Rabbit Farm, http://www.prfal.com). For Sir4p immunoprecipitations, serum from a rabbit inoculated with Sir4p was used [54]. After incubation with the antibody, 60 μl of protein A agarose beads (50% slurry from Upstate) were added and samples were rotated at 4 °C overnight. Samples and whole-cell extracts were electrophoretically fractionated on 7.5% polyacrylamide–SDS gels, transferred to nitro-cellulose membranes (Amersham, http://www.amersham.com), and probed with mouse polyclonal α-HA antibody (Upstate Biotech 05–904).
Whole-cell protein samples were prepared from three optical density equivalents of cells grown in medium lacking uracil to ensure plasmid retention. Trichloroacetic acid (TCA) was added to the culture medium to a final concentration of 10%, and the cells were incubated on ice for 20 min. Cells were pelleted, washed with Tris pH 8.0, and resuspended in 75 μl 3× protein sample buffer. Cells were lysed by vortexing in the presence of glass beads and subsequently incubated at 95 °C. Whole-cell protein extracts were electrophoretically fractionated on 7.5% polyacrylamide-SDS gels, transferred to nitro-cellulose membranes (Amersham), and probed with rabbit α-Sir2p (from J. Rine), rabbit α-HA antibody (Upstate Biotech 05–902), or 3-phosphoglycerate kinase (Molecular Probes/Invitrogen A-6457).
One optical density equivalent of cells was collected from logarithmically growing cultures by centrifugation and resuspended in 100 μl YM (yeast minimal) medium. For each strain, ten-fold serial dilutions were prepared, and 3 μl of each sample in the dilution series was spotted onto a YPD plate to monitor growth. To assay mating, an equal volume of the tester strain LRY1022 (MATα his4) at 10 OD equivalents /ml in YPD was mixed with each sample in the dilution series, and 3 μl of this mixture was spotted onto YM plates to select for the growth of prototrophic diploids. Yeast were grown at 30 °C for 3–4 d and subsequently photographed.
One optical density equivalent of logarithmically growing cells was collected by centrifugation in a microcentrifuge and subsequently resuspended in 100 μl YM medium. For each strain, ten-fold or five-fold serial dilutions were prepared, and 3 μl of each sample in the dilution series was spotted onto complete medium to monitor overall growth and medium lacking histidine or lacking histidine and uracil to monitor Hst1p-mediated repression. Uracil was omitted to maintain the plasmids in the pPES4-HIS3 reporter assays. Yeast were grown at 30 °C for 3–4 d and subsequently photographed. |
10.1371/journal.pcbi.1000138 | Identification and Rational Redesign of Peptide Ligands to CRIP1, A Novel Biomarker for Cancers | Cysteine-rich intestinal protein 1 (CRIP1) has been identified as a novel marker for early detection of cancers. Here we report on the use of phage display in combination with molecular modeling to identify a high-affinity ligand for CRIP1. Panning experiments using a circularized C7C phage library yielded several consensus sequences with modest binding affinities to purified CRIP1. Two sequence motifs, A1 and B5, having the highest affinities for CRIP1, were chosen for further study. With peptide structure information and the NMR structure of CRIP1, the higher-affinity A1 peptide was computationally redesigned, yielding a novel peptide, A1M, whose affinity was predicted to be much improved. Synthesis of the peptide and saturation and competitive binding studies demonstrated approximately a 10–28-fold improvement in the affinity of A1M compared to that of either A1 or B5 peptide. These techniques have broad application to the design of novel ligand peptides.
| Breast cancer is one of the most frequently diagnosed malignancies in American females and is the second leading cause of cancer deaths in women. Several improvements in diagnostic protocols have enhanced our ability for earlier detection of breast cancer, resulting in improvement of therapeutic outcome and an increased survival rate for breast cancer patients. However, current early screening techniques are neither comprehensive nor infallible. Imaging techniques that improve breast cancer detection, localization, and evaluation of therapy are essential in combating the disease. Cysteine-rich intestinal protein 1 (CRIP1) has been identified as a novel marker for early detection of breast cancers. Here, we report the use of phage display and computational molecular modeling to identify a high-affinity ligand for CRIP1. Phage display panning experiments initially identified consensus peptide sequences with modest binding affinity to purified CRIP1. Using ab initio modeling of binding peptide structures, computational docking, and recently developed free energy estimation protocols, we redesigned the peptides to increase their affinity for CRIP1. Synthesis of the redesigned peptide and binding studies demonstrated approximately a 10–28-fold improvement in the binding affinity. The combination of computational and experimental techniques in this study demonstrates a potentially powerful tool in modulating protein–protein interactions.
| Cysteine-rich intestinal protein 1 (CRIP1) belongs to the LIM/double zinc finger protein family, which includes cysteine- and glycine-rich protein-1, rhombotin-1, rhombotin-2, and rhombotin-3. Human CRIP1, primarily a cytosolic protein, was cloned in 1997 [1] using RT-PCR of human small intestine RNA and oligonucleotides whose sequence was derived from the human heart homolog of this protein, CRHP [2]. Recently CRIP1 has been identified as a very exciting biomarker for human breast cancers [3],4, cervical cancers [5],[6], pancreatic cancers [7],[8] and potentially other cancers [4],[9]. In experiments comparing CRIP1 expression in human breast cancer to matched normal breast tissue the mRNA for this target was overexpressed 8–10-fold in approximately 90% of both invasive and ductal carcinoma in situ [3]. Furthermore, in situ hybridization studies demonstrated close association of the expression with the ductal carcinoma cells [3]. CRIP1 overexpression has also been demonstrated to be the most highly differentially expressed gene in invasive cervical carcinomas; 100-fold up-regulation relative to normal cervical keratinocytes measured in 34 cervical tissues from different clinically defined stages [5],[6]. CRIP1 was also found to have high levels of expression in pancreatic adenocarcinoma, lung cancers and colorectal cancers [7]–[9]. These data strongly support the development of imaging probes targeting CRIP1 to improve cancer detection.
Phage display technology is a robust methodology for identifying peptides that bind relatively tightly to target proteins. This is especially true if the targeted protein's function is to bind peptides in vivo. In these applications, the first generation peptides have a generally lower Kd (10–100 µM) for their target and typically need to be structurally altered to improve binding before the peptides exhibit robust binding suitable to image the target protein. If structural data for the targeted protein exists, it should be feasible to utilize the data to help redesign in silico the Phage display-identified peptides thereby increasing their binding affinity. This approach is much more cost efficient than exhaustive screening of structured phage libraries or expansion of screening assays to include other types of phage display libraries.
Despite the potential utility of CRIP1 [10] as an imaging target, significant efforts to develop CRIP1-specific ligands have not been attempted. Here we utilized phage display techniques [11]–[20] to identify peptide ligands with micromolar binding affinity for purified human CRIP1 and exploited rational protein redesign [21]–[27] to increase the peptide's binding affinity. This approach has yielded a peptide that has approximately 10–28-fold improved binding affinity as measured by in vitro saturation and competitive binding assays. This study is a significant advance to the ultimate goal of synthesizing imaging probes that report CRIP1 expression levels in vivo.
We used phage display technology to identify peptides that bind relatively tightly to CRIP1. Then, we utilized NMR structural data of CRIP1 and computational methods to increase the peptide binding affinity to CRIP1.
CRIP1 was initially cloned into a mammalian expression vector and subsequently into pHAT10 for expression in bacteria. The pHAT10/CRIP1 vector encodes a naturally occurring polyhistidine epitope tag with the sequence of nonadjacent histidines that enable purification of expressed proteins under native conditions at neutral pH 7.0 (details of construct can be found in Figure S1 and Figure S2). Bacterial expression was chosen since it is a robust expression system and presumably CRIP1 does not require post-translational modifications for function. Cultures derived from these bacteria were induced to express CRIP1 using IPTG. We then isolated purified CRIP1 (see Methods). SDS-PAGE analysis of the cell lysate and fractions containing eluted CRIP1 show a single band for chimeric CRIP1 running approximately at the calculated molecular weight for the chimeric protein, 12.8 KDa (Figure S2 and Table S1). The yield of CRIP1 protein was approximately 10 mg of recombinant protein per liter of culture.
In order to generate CRIP1 protein that was as similar as possible to endogenous CRIP1, enterokinase cleavage was performed on purified CRIP1. Uncleaved contaminating HIS-tagged CRIP1 as well as HIS-tagged peptides were removed by re-running the digest over the CellThru resin and retaining the flow thru. This manipulation of the chimeric protein resulted in a polypeptide almost completely devoid of other “non-CRIP1” amino acids and was used as the bait for phage display studies. After four rounds of positive selection against enterokinase-truncated CRIP1, 29 phage DNA inserts were sequenced using a 96 gIII primer (5′-HOCCC TCA TAG TTA GCG TAA CG-3′). Sequencing verified that 18 of the 29 phagotopes were from the cysteine-constrained phage library, Table 1. Many of the peptide sequences contained similar motifs and six sequences occurred in more than one phagotope. The peptides A1 and C5 were identified four times, C1 three times, and A9, B1, and B5 twice. However, even accounting for conserved amino acid substitutions, no clear motif could be identified.
To select clones for further analysis, we used ELISA to determine the relative binding affinities of selected individual phage clones to purified CRIP1 (see Methods and Figure S3). For these studies the purified chimeric CRIP1 was not reacted with enterokinase. Clone A1 and Clone B5 were measured to possess higher relative affinity. Although it occurred with the same frequency as clone A1, clone C5 exhibited lower binding affinity. With these results, inserts from the two clones with the highest affinities (A1 and B5) were further investigated as potential ligands to CRIP1.
The computational optimization of the binding affinity of the peptide initially identified from phage display involved three stages. We first constructed a structural model of the cyclic peptide A1, and then we identified putative binding sites on CRIP1 by docking. Lastly, we searched for new peptide sequences that optimize the stability of the peptide-CRIP1 complex.
Shown in Figure 1C is the molecular model of the cyclic peptide A1 (see Methods). To remove the bias on the docking that may be introduced by using only one backbone peptide conformation, we first generated several peptide backbone conformations from snapshots of equilibrium molecular dynamics simulations. Each peptide was docked to the 48 conformations of CRIP1 derived from NMR [10]. By clustering the location of the peptides on the CRIP1 surface, we were able to identify and rank the putative binding sites (Figure 2). Interestingly, the peptides preferably bind to one face of CRIP1 (Figure 2). This side of CRIP1 contains two grooves, one formed by helix H3 and S6–S7 loop and another by S2–S3 loop and the N-terminal loop. The binding site of the successfully redesigned A1M is formed by Glu46, His45, Phe60, Tyr56, and Lys48 (Figure 3).
In the second stage of the redesign, we searched for peptide sequences that optimized the binding free energies of the peptide-CRIP1 complexes using heuristic algorithms and a physical force-field (see Methods). The methodology employs rapid side-chain packing and backbone relaxation to calculate the free energy change due to a mutation. For a given CRIP1-peptide complex, we determined a set of mutations in the bound peptide that resulted in the lowest free energy change, and thus, the highest predicted increase in binding affinity. All CRIP1-peptide complexes were subjected to redesign. All redesigned peptides were then grouped according to their starting peptide backbone conformation (1-ns, 9-ns, or 10-ns), and according to their putative binding site. The redesigned sequence CLDGGGKGC, which we denote here as A1M (“modified A1”), corresponds to a peptide with the lowest binding free energy ΔΔG among the redesigned sequences in the highest-ranked binding mode. In Table 2 we list representative peptide sequences with high binding affinity but located in other putative binding sites and featuring backbone conformations other than the 1-ns.
To identify the dominant motifs in the redesigned peptides, we show in Figure S5 the dominant sequence motifs in the top three candidates binding sites for each peptide model. There is a prevalence of Gly, presumably due to the strongly curved backbone that prefers more flexible Gly over any other residue when the peptide is in the context of the protein but not when the peptide is isolated (Figure S6). The redesigned sequences also exhibit a preference for charged residues (mostly Asp, Glu, and Lys) in at least two positions (Figure S5). These charged residues, we believe, are what attributes the redesigned peptides their specificity to CRIP1. In particular, the designed sequence A1M (Figure S5), which is a member of the largest cluster in the CRIP1 and 1-ns peptide complexes, exhibits a preference for either Lys or Glu in the 2nd position, Asp in the 5th, Lys in the 7th, and Gly in the rest.
A closer inspection of the specific energy contributions to the ΔΔG of A1M (Table S2), we found that the largest contributions to ΔΔG arises from more favorable van der Waals interaction between the peptide and CRIP1, which we believe is reflected in the preference for Gly in some sites of the binding peptide. The CRIP1-peptide complex also exhibits more favorable solvation energy after the redesign. This observation is also reflected structurally in Figure 3. In particular, the peptide side chains in A1 (such as 4D, 5N, 6H, and 8S) that point toward the CRIP1 surface are replaced by Gly, while those pointing to solution (3K and 7R) retain their polar nature.
We computationally redesigned A1 resulting in a peptide with a new sequence (denoted as A1M) predicted to bind to CRIP1 with higher affinity. To test this prediction, A1, B5 and A1M peptides were all synthesized and labeled with FITC for binding studies. Since the peptides encoded by the C7C phage library are at the N-terminus of the minor phage coat protein pIII followed by a short phage encoded spacer Gly-Gly-Gly-Ser, we included this 4-mer in the synthesized peptide. An additional C-terminal Lys was also included in order to enable fluorescent labeling of the peptide. Thus, the different selected mimotopes were produced as synthetic peptides with Gly-Gly-Gly-Ser-Lys and then labeled by adding a fluorescent molecule to the C-terminal lysine. We synthesized the cyclic form of the peptides, A1, B5 and A1M and determined their ability to bind CRIP1 using saturation binding experiments. The value for the apparent equilibrium dissociation constant (Kd apparent) of the FITC-A1M peptide determined by saturation binding was 2.6 µM, Figure 4A. This was substantially lower than that obtained for either the parent A1 peptide (Kd apparent = 34.4 µM) or the estimate for the Kd apparent of the B5 peptide (Kd apparent = 62.5 µM) derived using similar assays, data not shown. To directly compare the affinity of the A1 and A1M peptides for CRIP1 protein, we performed a competitive binding assay and determined the IC50 for each of the peptides using FITC-A1M as the ligand. These studies demonstrated that the binding affinity of the A1M peptide to CRIP1 was approximately 27.5 times better than that of the original A1 peptide, Figure 4B (A1 peptide IC50 = 8.8 µM, A1M peptide IC50 = 0.32 µM). Since each peptide was effective at displacing FITC-A1M and reached the same minimal binding these data also suggest that ligand binding to CRIP1 occurs at a single site. Further analysis of the binding data with multiple binding site models clearly showed that the best fit of the data was obtained with a one binding site model. Both experimental results are further supported by the predicted binding sites for each peptide to CRIP1 as depicted in Figure 3. Interestingly, when the Ki for the A1M peptide is calculated (Ki = 0.067 µM), it is not the same as the apparent Kd for FITC-A1M (Kd = 2.6 µM) determined by saturation binding experiments. Based on this observation, the FITC label likely reduces the affinity of the peptide for CRIP1, which is not uncommon with labeled peptides. However, this observation does not alter the interpretations of the data comparing the affinity of the unlabeled peptides A1 and A1M.
From the apparent equilibrium dissociation constants, we calculate the experimental free energy change to be ΔΔG = RT ln Kd,A1M − RT ln Kd,A1 = −1.6 kcal mol−1, which is smaller than the estimated computational free energy change ΔΔG = −83 kcal mol−1 (Table 2). This difference between the experimental and computational free energy changes is primarily contributed by the van der Waals repulsion term (Table S2), suggesting initial clashes in the docking of the A1 peptide to the CRIP1 structure. However, since the docking protocol (ZDOCK) is consistently implemented, we still expect strong correlation between the computational and experimental free energy changes, that is, those redesigned peptides with lower computational ΔΔG is also expected to have low experimental ΔΔG, although the absolute values may not be directly comparable. In a separate study benchmarking the Medusa force field [28],[29], experimental and computational ΔΔG values exhibited a correlation of 0.75 (P = 10−108).
CRIP1 is an extremely compelling marker to exploit for enhanced detection of breast and other cancers. However, its cytosolic expression makes it hard to measure by conventional means, e.g., antibodies. The cell membrane increases the pharmacological barriers that must be overcome to bind and consequently image the expression of this protein in cancer cells. Thus, we developed methodologies to generate high affinity peptides to purified cytosolic proteins with the ultimate aim of designing these peptides to cross membranes and serve as imaging ligands. To rapidly identify peptides that will bind to CRIP1, we utilized phage display technology and purified CRIP1 protein. This technology identifies relatively low affinity (10–100 µM) ligands to target proteins. To increase the affinity of the peptides identified using phage display, we developed a protocol for rational peptide redesign that utilizes computational techniques. This protocol successfully increased peptide affinity by approximately 10–28-fold.
Computational design methods have been employed to modulate protein-protein interactions. Major challenges in protein design include (1) identification of ligand-peptide binding site and (2) optimization of affinity of the peptides that bind to that particular protein [30]. In practice, sequence and conformational space need to be adequately sampled [30]. There is also the need for accurate energy functions that identifies protein sequences corresponding to the global free energy minimum of a given protein conformation [30]. Several studies have been reported to identify protein interaction specificity [31]–[35]. For example, Shifman and Mayo computationally redesigned the promiscuous binding site of calmodulin to increase its specificity to one of its ligand peptides [36]. The authors performed iterative optimization of the rotamers. In another study, Reina et al. computationally engineered a small protein-protein interaction motif of the PDZ domain to bind novel target sequences [37]. The study demonstrated that by combining different backbone templates with computer-aided protein design, PDZ domains could be engineered to specifically recognize a large number of proteins [37]. Another example of successful redesign was the engineering of coiled-coil interfaces that direct the formation of either homodimers or heterodimers [38]. The design protocol involved both positive design, stabilization of desired interaction, and negative design, the destabilization of undesired interactions [38].
The problem of redesigning ligand peptides initially identified from phage display is challenging because the structure of the peptides are not known and the peptides do not have a known binding site in CRIP1. While there have been successes in the redesign of protein-protein interfaces and location of binding site through computational docking, there is yet no study where the system being designed face these two major challenges simultaneously. We computationally modeled the cyclic peptide and performed molecular dynamics to find the equilibrium conformation of the peptide. To diversify the backbone conformation of the peptide included in the redesign, we selected 3 peptides from the equilibrium molecular dynamics and docked them to 48 CRIP1 conformations from NMR. Interestingly, this procedure of diversifying protein and ligand peptide conformation is sufficient to identify putative binding sites on the protein. We believe that the cyclic structure of the peptide was an important factor to the success of the procedure, because the error from enthalpy-entropy compensation is reduced when docking a cyclic peptide compared to docking a linear peptide.
Another important factor that contributed to the success of the peptide design is the conformational sampling introduced in the design steps to maximize the coverage of sequence-structure space available to the CRIP1-peptide complex. First, we performed multiple docking simulations that allowed us to identify various poses for binding. Second, we allowed backbone of the peptide to be flexible during sequence design procedure, thereby significantly diversifying the designed sequences [28],[29]. Hence, the combination of the restricted conformational space available to a peptide due to circularization and our flexible-backbone sampling technique [28],[29] allowed us to sufficiently sample the conformational space of the peptide during design, thereby contributing to a successful peptide binder to CRIP1. Our approach can be further extended to other systems of interest.
In this study, we combine empirical and computational approaches to develop a novel paradigm to improve ligand affinity when limited structural information is available. CRIP1, a potentially powerful biomarker for several cancers, was purified and used in an empirical phage display assay to identify short amino acid peptides with modest affinity for the protein. The resulting peptides were then structurally modeled, based on the structures of other known but unrelated peptides of similar size. Using the limited NMR structure available for CRIP1 the modeled peptides were then computationally docked to CRIP1 resulting in identification of several potential structural motifs responsible for the binding interaction. The modeled interactions were then optimized and peptides were redesigned based on these data.
Interestingly, even after 4 rounds of phage display isolation, no consensus sequence for CRIP1 binding peptides emerged. These data might possibly suggest that a strong binding “natural” peptide did not exist on the CRIP1 protein. Remarkably, however, computational manipulation of the amino acids contained within the peptide, based on energy minimization, significantly increased the affinity of the peptide. This suggests that: (1) conditions for phage-CRIP1 binding were not optimal for peptide identification; (2) the phage library did not contain all possible combinations of amino acids; and/or (3) the library was not exhaustively screened. In any of these cases, however, the use of computational redesign combined with empirically derived initial binding data significantly improved the quality of final peptide ligands. As our database of redesigned peptides and resulting Kd's accumulates, the approaches described here potentially can be generalized and could be implemented for peptide ligand generation routinely. The resulting peptide from these studies, A1M, will be further developed as an imaging probe.
The coding region of the CRIP1 cDNA was removed from CRIP1 in pcDNA3.1+ using BamH1 and Xba1 restriction enzymes and subsequently subcloned into the BamHI and EcoRI sites of the vector pHAT10 (BD Clontech) which contains an N-terminal histidine affinity tag. The construct was confirmed by sequencing.
Bacterial cells expressing the pHAT10-CRIP1 were cultured in LB media containing 50 µg/ml ampicillin until reaching OD of 0.6 at which time they were induced to express the protein by adding IPTG to a final concentration of 0.5 mM IPTG. The bacteria were then harvested and resuspended in Equilibration/Wash Buffer (50 mM sodium phosphate pH 7.0, 300 mM NaCl) containing 0.75 mg/ml lysozyme and 0.0174 mg/ml PMSF and sonicated with three 10 s pulses (medium power, Sonic Dismembrator Model 100, Fisher Scientific), with a pause for 30 s on ice between sonication cycles. Following sonication, the lysates were cleared by centrifugation, and incubated with TALON CellThru Resin (BD Biosciences, Palo Alto, CA) in Extraction/Wash Buffer. The tagged protein was eluted from the washed column with 0.15 M imidazole in Extraction/Wash Buffer. The purity of CRIP1 in fractions was confirmed by SDS-PAGE [39]. The concentration of CRIP1 in fractions was determined by Bradford Assay using IgG as a standard [40].
CRIP1 was digested with enterokinase (Roche Diagnostics, Inc.) to remove the His tag and then was used as bait for 4 rounds of panning with the Ph.D.-C7C Phage Display Peptide Library (New England Biolabs). The nucleotide sequence of the gene III insert was determined by sequencing the phage, and the amino acid sequence of the insert was deduced from the nucleotide sequence, shown in Table 1.
Computational optimization of the peptide binding affinities consists of three major steps: (1) structural modeling of cyclic peptides initially identified from phage display experiments, (2) finding putative binding sites of the peptides on CRIP1, and (3) searching for sequences that optimize the stability of the peptide-CRIP1 complex.
The binding affinity of the peptides for CRIP1 protein was determined by saturation binding experiments [54],[55]. Ninety-six well plates were coated with 150 µl of PBS buffer containing 100 µg/ml of CRIP1 and incubated overnight at 4°C. The wells were then washed three times with 50 mM Tris, 150 mM NaCl, pH 7.5 (TBS) containing 0.1% Tween-20 (TBST), and then each well filled completely with blocking buffer (TBS containing 0.5% BSA), incubated at least 1 hour at 4°C, and then rapidly washed 3 times with TBST. Following washing 100 µl of binding buffer containing different concentrations of FITC-labeled peptides (ranging from 50 nM to 100 µM) were added to the CRIP1 containing wells and incubated for 1 hour at 37°C with rocking. After incubation, the plates were washed three times with binding buffer. The fluorescence intensity in each well was determined on Infinite M200 Tecan Instrument (Tecan, NC) (Excitation wavelength: 494 nm, Emission wavelength: 530 nm). The apparent equilibrium dissociation constant, Kd,apparent, was calculated by non-linear regression using GraphPad Prism (GraphPad Prism 4.0 Software, San Diego, CA). Each data point is the average of three determinations. Nonspecific binding was defined in the presence of 1 mM unlabeled peptide. All binding experiments (saturation binding and competitive binding experiments) were conducted under equilibrium binding conditions and under conditions where total ligand added was essentially equivalent to the amount of free ligand after the binding reaction occurred.
The binding affinity of the A1 and A1M peptides for CRIP1 protein was directly compared by a competitive binding experiment [55],[56]. Labeled A1M peptide (FITC-A1M) was competed with increasing concentrations of either unlabeled A1M peptides or A1 peptide and the IC50 for each peptide calculated.
Ninety-six well plates were coated with 150 µl of PBS buffer containing 100 µg/ml of CRIP1 and incubated overnight at 4°C. The wells were then washed 3 times with 50 mM Tris, 150 mM NaCl, pH 7.5 (TBS) containing 0.1% Tween-20 (TBST), and then each well filled completely with blocking buffer (TBS containing 0.5% BSA), incubated at least 1 hour at 4°C, and then rapidly washed 3 times with TBST. Following washing 150 µl of binding buffer containing FITC-A1M peptides of 10 µM and appropriate dilutions of unlabeled A1 and A1M peptides (ranging from 0 to 300 µM) were added to the CRIP1 containing wells and incubated for 1 hour at 37°C with rocking. After incubation, the plates were washed three times with binding buffer. The fluorescence intensity in each well was determined on Infinite M200 Tecan Instrument (Tecan, NC) (Excitation wavelength: 494 nm, Emission wavelength: 530 nm). Ki was calculated by non-linear regression with one binding site using GraphPad Prism (GraphPad Prism 4.0 Software, San Diego, CA). Each data point is the average of three determinations, shown in Figure 4B. Data was analyzed using several different binding models and was found to only fit a one binding site model. When no competitor was added the data point was graphed as 0.1 nM to satisfy software requirements. This has no effect on calculations of the IC50's.
|
10.1371/journal.pcbi.1007168 | Molecular noise of innate immunity shapes bacteria-phage ecologies | Mathematical models have been used successfully at diverse scales of biological organization, ranging from ecology and population dynamics to stochastic reaction events occurring between individual molecules in single cells. Generally, many biological processes unfold across multiple scales, with mutations being the best studied example of how stochasticity at the molecular scale can influence outcomes at the population scale. In many other contexts, however, an analogous link between micro- and macro-scale remains elusive, primarily due to the challenges involved in setting up and analyzing multi-scale models. Here, we employ such a model to investigate how stochasticity propagates from individual biochemical reaction events in the bacterial innate immune system to the ecology of bacteria and bacterial viruses. We show analytically how the dynamics of bacterial populations are shaped by the activities of immunity-conferring enzymes in single cells and how the ecological consequences imply optimal bacterial defense strategies against viruses. Our results suggest that bacterial populations in the presence of viruses can either optimize their initial growth rate or their population size, with the first strategy favoring simple immunity featuring a single restriction modification system and the second strategy favoring complex bacterial innate immunity featuring several simultaneously active restriction modification systems.
| Mathematical understanding of how randomness at the molecular scale, also known as molecular noise, ultimately affects the fate of organisms and whole populations is widely recognized as a challenging problem in multi-scale modeling. Here, we develop an analytical framework for analyzing how the randomness of individual reaction events in single cells propagates to higher levels of biological organization and affects population and ecology scale dynamics. We deploy our mathematical results to study an example from the ecology of bacteria and bacteriophage viruses. Bacteria defend themselves against viruses by a simple innate immune system composed of a pair of enzymes. Due to molecular noise, however, viruses sometimes escape this immunity, causing bacterial populations to plummet. Noise can also cause the immune system to turn against its own host. By analyzing how such costs and benefits of bacterial immunity balance at the ecological level, we predict the optimal parameters for bacterial innate immune systems. While the focus of this work is on bacteria-phage ecologies, we expect that our results will generally help to better understand population and ecology scale consequences of stochastic cell fate decisions in diverse biological domains ranging from cell differentiation in developmental biology to studies of the microbiome and consequences of stochasticity in apoptotic responses for anti-cancer therapies.
| One of the major challenges in biology is to understand how interactions between individual molecules shape living organisms and ultimately give rise to emergent behaviors at the level of populations or even ecosystems. At the very bottom of this hierarchy, inside single cells, interacting biomolecules such as DNA or proteins are often present in small numbers, giving rise to intrinsic stochasticity of individual reaction events [1, 2]. As a result, genetically identical organisms occupying identical environments can express different phenotypes [3, 4] and make different decisions when presented with identical environmental cues [5, 6]. This molecular noise is known to be the cause of biologically and medically important traits of bacteria such as persistence in response to antibiotics [7, 8] and competence during acquisition of heterologous DNA [9]. However, while its causes and consequences are relatively well-studied at the organismal level [10–13], how molecular noise propagates to higher scales of biological organization to affect the ecology and evolution of organisms remains mostly unknown [4]. Recently it has been shown that ecosystems can follow surprisingly deterministic trajectories despite the prevalence of stochastic events [14, 15], yet these trajectories could themselves be strongly influenced by molecular noise. Thus, the extent to which ecological interactions are affected by molecular noise, and the extent to which these ecological consequences feed back to reshape individual traits, remain to be explored.
Perhaps the most prevalent biological systems in which molecular noise is thought to play an important role are restriction-modification (RM) systems [16]. Present in nearly all prokaryotic genomes [17], RM systems are a highly diverse class of genetic elements. They have been shown to play multiple roles in bacteria as well as archaea, including regulation of genetic flux [18] and stabilization of mobile genetic elements [19], but have most frequently been described as primitive innate immune systems due to their ability to protect their hosts from bacterial viruses [20]. When a virus (bacteriophage or phage) infects a bacterium carrying a RM system, the DNA of the phage gets cleaved with a very high probability, thus aborting the infection. With a very small probability, however, the phage can escape and become immune to restriction by that specific RM system through epigenetic modification, leading to its spread and potentially death of the whole bacterial population in absence of alternative mechanisms of phage resistance [21, 22]. Thus, in the context of RM systems, molecular noise occurring at the level of individual bacteria can have profound ecological and evolutionary consequences. Because RM systems are ultimately based on only two very well characterized enzymatic activities (restriction and modification) [23], they represent a simple and tractable biological system in which we can investigate propagation of effects of molecular noise across different scales of biological organization.
Here, we mathematically model the action of RM systems from individual molecular events occurring inside a single cell, through individual bacteria competing in a population, to interactions between populations of bacteria and phages in a simple ecological setting, as shown in Fig 1. We demonstrate that, by imposing a tradeoff between the efficiency and cost of immunity, molecular noise in RM systems occurring at the level of individual bacteria has consequences that propagate all the way up to the ecological scale, and that the ecological consequences in turn imply the existence of optimal bacterial defense strategies against phages.
RM systems consist of two enzymes, a restriction-endonuclease R, that recognizes and cuts specific DNA sequences (restriction sites), and a methyl-transferase M, that recognizes the same DNA sequences and ensures that only invading phage DNA can be cut by the endonuclease while the bacterial DNA remains methylated and protected. However, since chemical reactions occur stochastically, RM systems can produce errors and fully methylate invading phage DNA before it is cut and degraded (phage escape, typically occurring with a probability ranging between 10−2 and 10−8) [24, 25]. Similarly, it is possible that newly replicated restriction sites on the bacterial DNA, which are originally unmethylated, are accidentally cleaved instead of methylated (self-restriction) [26].
Inside a single cell, the probability of such self-restriction events depends on the total activity, r, of all restriction endonuclease molecules R, the total activity, m, of all methylase molecules M, as well as the bacterial replication rate λ, since λ determines the rate at which new unmethylated restriction sites are generated. To investigate how self-restriction depends on these parameters, we model the corresponding biochemical reactions at each individual restriction site on the bacterial DNA with the stochastic reaction network displayed in Fig 2a (see S1 Appendix Section S.1). The time τS until the first self-restriction event in a given cell—i.e., until that cell’s death or substantial reduction in growth rate—can be obtained as the time when the first restriction site is cut, that is as τS = mini∈{1,…,NS} τi, for bacterial DNA with NS restriction sites, where τi, i = 1,…,NS are the waiting times for cutting events at individual sites. It can be shown that all τi follow a phase-type distribution (see [27] and Fig 2b and 2c):
f ( τ i ) = p Q exp ( B τ i ) c 1 , where B = [ - ( r + m ) m 0 λ / 2 - ( m + λ / 2 ) m 0 λ - λ ] and c 1 = [ r 0 0 ] , (1)
with pQ = [p0 p1 p2] being the initial methylation configuration, i.e., the proportion of restriction sites that are unmethylated (p0), hemi-methylated (p1) and doubly-methylated (p2); see S1 Appendix Section S.1.
Eq [1] allows us to derive the expected time until self-restriction of a single site as
E [ τ i ] = - p Q B - 1 1 , where 1 = [ 1 1 1 ] ⊤ ; (2)
more generally, Fig 2b shows how the distribution of waiting times depends on the restriction rate r (increasing the probability of the site getting cut when it is unmethylated) and the magnitude of m relative to λ (which decreases the probability that the site is unmethylated in the first place).
Fig 2c shows that time to self-restriction at a single site depends essentially on an unknown quantity, the methylation configuration pQ. We will now proceed to show that when we consider an exponentially growing population of bacterial cells, the configuration pQ can no longer be freely chosen, and has to be determined self-consistently instead. Intuitively, this is because when the bacterial population is in steady-state growth, new unmethylated sites are constantly replenished by replication, while cells with more unmethylated sites are simultaneously and preferentially being removed, as illustrated in Fig 3a and required by Eq [2]. These two forces, generation of new unmethylated sites and their preferential removal, will push any initial pQ towards a unique steady state equilibrium.
Mathematically, assuming that the methylation dynamics in all cells are equilibrated and that cells cannot be distinguished, the internal methylation configuration of any randomly chosen cell at any time during growth of the population can be derived from the quasi-stationary distribution pQSD(r, m) of the individual-site methylation process in Fig 2a (see S1 Appendix Section S.1). pQSD(r, m) is the equilibrium distribution of the stochastic process conditional on it not having reached the absorbing state where the DNA is cut and the cell has died (Fig 3a); in short, methylation and growth equilibrate “in all directions except the one leading towards self-restriction”. Then, setting pQ = pQSD(r, m) in Eq [1] reduces the phase-type distribution f(τi) for the time τi until self-restriction at an individual restriction site to a single exponential, implying further that the waiting time τS = mini∈{1,…,NS} τi until self-restriction of any site in the cell is also exponentially distributed. Consequently, we are led to the main result of this section: growth with self-restriction can be rigorously modeled at the population level with a Markov birth-death process for which the expected population size n(t) follows a simple ordinary differential equation
d d t n ( t ) = ( λ - μ ( r , m , λ ) ) n ( t ) = λ e n ( t ) , (3)
where λe(r, m, λ) = λ − μ(r, m, λ) is the effective growth rate and μ(r, m, λ) is the rate of self-restriction, defined as the inverse of the per-cell expected waiting time until self-restriction
μ ( r , m , λ ) = E [ τ S ] - 1 = N S - p QSD B - 1 1 = - γ 1 N S , (4)
with γ1 being the largest eigenvalue of B (an explicit stochastic simulation validating this analytical result is provided in the S1 Appendix Section S.2).
Eq [4] allows us to straightforwardly evaluate the reduction in the population growth rate due to random self-restriction events in single cells for any given pair of enzyme activities, r and m. To study possible qualitative effects of self-restriction, we explore in Fig 3b a wide range of enzyme activities for a system with NS = 5 restriction sites (chosen, for illustration purposes, significantly smaller than the typical number of sites recognized by real RM systems). We find that the main determinant of self-restriction is the activity m of the methyl-transferase and that the effects of molecular noise can be suppressed by sufficiently increasing m. Furthermore, so long as m is large enough such that unmethylated restriction sites are only rarely available, μ(r, m, λ) lies on a large plateau of low self-restriction and changes only little with r and m, suggesting that stochastic fluctuations in enzyme activities would only have minor consequences for the population, especially when they are positively correlated, as would be the case if R and M enzymes were expressed from the same operon (S1 Appendix Section S.3).
The (r, m) plane in Fig 3b contains a transition region that separates the large plateau with low self-restriction from the plateau where self-restriction is severe enough to stop the population growth altogether. We have chosen our reference (red) parameter values (rref,mref) to lie in this transition region, and explored the regime with an e-fold higher rates (“large r & m”, indicated by green), and with 2e-fold lower rates (“small r & m”, indicated by blue) in Fig 3b and 3c. The comparison of these three regimes in Fig 3c is most clear when the effective growth rate is shown as a function of λ, the rate at which the cells, and thus the restriction sites, are replicated. In the “small r & m” regime, self-restriction is so infrequent that it can easily be outgrown by replication (except at very low λ). In the “large r & m” regime, m is sufficiently high to keep the restriction sites protected and thus self-restriction is rare, except at extremely large λ, where the green curve falls below the blue curve. In the reference regime, r is too large and m not high enough to protect, so self-restriction can not be “outgrown”; effective growth thus falls significantly below λ. Our numerical analyses further show that the self-restriction rate μ(r, m, λ) grows faster-than-linearly with λ (S1 Appendix Section S.1), causing the effective population growth to slow down and ultimately drop to zero at high enough λ.
We end this section by highlighting a non-trivial interaction between the single-cell and population-scale processes. While increasing the activity r of the endonuclease always decreases the effective growth rate of the population due to self-restriction, the effect can be smaller than expected from the single-cell analysis (dashed lines in Fig 3c). This is because high values of r feed back through the population scale to bias the steady-state distribution of methylation configurations away from cells with lots of unmethylated sites, as shown in Fig 3a, making self-restriction less likely. Implicit feedback effects of this type frequently give rise to complex dynamics in multi-scale models.
RM systems lower the growth rate of the population due to self restriction, especially when the activity m of the methyl-transferase is small. Upon infection by a phage, however, small values of m are advantageous, making it less likely that the unmethylated phage DNA will get methylated and escape the immune system before it can be cut by the restriction endonuclease.
Assuming that all restriction sites are identical and independent, the probability of phage escape can be calculated [28] as
p V ( r , m ) = ( m r + m ) N V , (5)
where NV is the number of restriction sites on the phage DNA. From Eq [5] it is straightforward to see that pV(r, m) is monotonically increasing in m and decreasing in r. One might therefore expect that the balance between avoiding self-restriction that favors high m, Eq [4], and minimizing phage escape that favors low m, Eq [5], would impose a tradeoff and thus lead to an optimal value of m. However, this is not the case, because the phage escape probability pV(r, m) and the population self-restriction rate μ(r, m, λ) can both approach zero so long as r and m both increase to infinity but r does so faster. While mathematically possible, this limit is, however, not biologically relevant: large enzyme expression levels should incur a cost (metabolic or due to toxicity presumably caused by non-specific protein-DNA interactions in the case of RM systems) for the cells [29, 30], which we sought to incorporate into our model by including a growth rate penalty proportional to the activity of restriction and methylation enzymes, i.e., λe(r, m, λ) = λ − μ(r, m, λ) − crr − cmm. Interestingly, it can be verified that our reasoning is valid only because two subsequent demethylation events need to occur to create a restriction-susceptible site on the bacterial DNA (S1 Appendix Section S.1). If hemi-methylated sites could be recognized by the restriction endonuclease, or if both methyl groups could be lost in a single event, our initial expectation about the existence of the tradeoff would be correct, and a particular choice of r and m values would simultaneously minimize the phage escape and self-restriction, even in the absence of the expression cost for R and M.
Our model can be generalized to multiple coexisting RM systems that recognize different restriction sites and operate in parallel, as is often observed for bacteria in the wild [17]. This provides increased protection from phages since the phage has to escape all RM systems to infect successfully. However, multiple RM systems also imply that the bacteria either have to pay higher expression and self-restriction costs or that they have to re-balance the expression levels of the enzymes such that lower self-restriction rates per RM system are obtained with the same overall enzyme activity. Allowing bacteria to have multiple RM systems, but assuming for the sake of simplicity that these systems are all equivalent in terms of enzyme activities and number of recognition sites, we obtain the phage escape probability for k RM systems as p V ( r , m , k ) = ( m r + m ) k · N V, with the corresponding growth rate being
λ e ( r , m , k , λ ) = λ - k · ( μ ( r , m , λ ) + c r r + c m m ) . (6)
What is the combined effect of phage escape and self-restriction in simple bacteria-phage ecologies? To investigate this question, we first extended an established deterministic model of bacteria-phage ecology [31] to track the population dynamics of bacteria with and without RM systems and both susceptible and methylated phages (see S1 Appendix Section S.4.2). By numerically integrating this population model for more than a million parameter combinations for the activity of restriction (r) and methylation (m) enzymes, we find that whether or not phages will ultimately take over the population depends on the ecological parameters (e.g. phage adsorption rate, rate of spontaneous phage inactivation, etc.) but is completely independent of RM system efficiency. This result might seem surprising at a first glance, but closer analysis reveals that for efficient RM systems the phage population reaches levels that are so small that they should be considered as extinction from a biological perspective. Nevertheless, even in these cases methylated phage eventually takes over the population. This is because phages cannot go extinct in the mathematical sense and the phage population always remains at levels that are strictly larger than zero if ecology models based on ordinary differential equations are used for the analysis. While this clearly limits the practical relevance of such models, the finding that RM systems apparently cannot provide long term protection if phage escape probability and phage population size remain strictly larger than zero is still interesting since it suggests that the task of RM systems cannot be to prevent phage escape but only to delay it as much as possible to give bacteria enough time to develop alternative mechanisms of phage resistance through genetic mutations [21, 22].
In line with the above reasoning, we decided to represent phage escape events stochastically and to focus in more detail on how RM systems impact exponentially growing bacterial populations until the first phage escape event. To explore this question, we formulated several efficiency measures that quantify how RM systems can help bacterial populations before the first phage escape event:
Here we will show that questions (i)-(iii) can be answered rigorously if we assume that the size of the phage population remains approximately constant until the first phage escape event. An example of an ecological scenario where this assumption is realistic is that of bacteria colonizing a phage-dominated environment in which the number of phages is much larger than the number of bacteria such that the reduction in the phage population size due to unsuccessful infections is negligible. More generally, any ecological scenario in which the phage population size is for some reason in equilibrium at least until the first phage escapes on a bacterium carrying a RM system, will fulfill this assumption.
Mathematically, we consider a bacterial population of initial size n0 trying to colonize an environment containing a phage population of size v. As we have shown before, the bacterial population will initially grow exponentially at rate λe until the time τp at which the first phage escape event occurs. Interpreting these events as random, the crucial unknown is therefore τp, the random time to first phage escape, characterized by its probability distribution, f(τp), which we find to be given by (see S1 Appendix Section S.4.3):
f ( τ p ) = ρ v n 0 p V exp ( ρ v n 0 ( 1 - e λ e τ p ) p V λ e + λ e τ p ) . (7)
Specific examples of this probability distribution are visualized in Fig I in S1 Appendix. In general, larger values of any of the parameters ρ, v, n0, pV or λe will imply that phage escape is likely to occur at earlier times. Importantly, the waiting time distribution until first phage escape, f(τp), allows us to analytically answer questions (i)-(iii), as summarized in Table 1 (see S1 Appendix Section S.4.3). We note that despite the somewhat intricate form of f(τp) the “bacterial performance” metrics derived for all three efficiency criteria turn out to be remarkably simple, depending only on some of the parameters that define f(τp). More concretely, by examining these metrics, we can make two important observations:
First, assuming a fixed mutation rate cmut, expressions for bacterial performance in Table 1 are functions of λe and pV, which depend solely on the restriction rate r, the methylation rate m, and the number of concurrently active RM systems, k. This means that optimal bacterial strategies at the ecological level can be found mathematically —and possibly tuned evolutionarily— by adjusting the three parameters, r, m, and k, defined at the single-cell level.
Second, despite the dependence of the time to phage escape on the initial population size n0, the performance of the bacterial population is independent of n0 according to all criteria. This has the important consequence that optimal defense strategies against phages do not depend on the size of the bacterial population and that there exists a single unique best defense strategy that is constant in time: if phage escape has not happened until a certain time during which the bacterial population has grown to a new size, the same defense strategy continues to be optimal with the initial size taken to be the new size, with no need to re-balance the activity levels r and m of the RM enzymes, or the number of RM systems, k. For cases (ii) and (iii), we further observe that the results are independent of the effective growth rate λe. Faster growth leads to quicker increases in the probability that immunity conferring mutations happen but this is exactly compensated by the increase in probability of a phage escape event.
An in-depth study of the consequences and implications of case (i) is presented in the following section while questions (ii) and (iii) are treated in the S1 Appendix (Section S.4.4). Taken together, these results show how the mathematical framework developed in this paper can be readily adjusted to analyze trade-offs between the efficiency and cost of immunity in different ecological contexts. For the concrete scenarios that we considered here, we find that (i) and (iii) imply overall similar results in which bacteria can relatively directly trade cost for efficiency and vice versa. The results for (ii), however, are qualitatively different since (ii) implies that increasing the cost beyond a certain point provides only diminishing returns in terms of efficiency (Fig J in S1 Appendix). We conclude that analyzing such trade-offs in practice will require careful consideration of the efficiency criteria according to which bacteria might have been shaped in a particular ecological context. Reversely, different trade-offs and optimal strategies for different efficiency criteria imply that the criterion on which evolution might have been operating in a given ecological context can, in principle, be reverse engineered from observations of phage defense strategies that are employed by the bacteria.
Can bacteria tune the single-cell parameters over evolutionary timescales in order to minimize the cost of RM systems, that is maximize the growth rate λe(r, m, k), while also maximizing their efficiency, quantified here as the increase in population size before the first phage escape, n s ( r , m , k ) : = E [ n ( τ p ) ] - n 0 in (i), that is determined by λe(r, m, k)/pV (r, m, k)? Eq [6] and Table 1 assert that cost and efficiency are necessarily in a tradeoff and cannot be optimized simultaneously. This tradeoff is the first key result of the section. With no single optimum possible, we look instead for Pareto-optimal parameter combinations, (r, m, k), i.e., solutions for which λe cannot be further increased without reducing ns and vice versa [32, 33]. Different Pareto-optimal solutions trace out a “front” in the plot of λe vs ns in Fig 4a that jointly maximizes growth rates and population sizes to the extent possible. Points in the interior of the front are sub-optimal and could be improved by adjusting parameter values, while points beyond the front are inaccessible to any bacterial population. Which Pareto-optimal solution ultimately emerges as an evolutionary stable strategy depends on the actual bacterial and phage species considered as well as their biological context. Rather than focusing on specific examples, we next establish several general results of our analysis, contrasting in particular “fast growth” bacterial strategies that maximize λe with “large size” strategies that maximize ns.
We start by examining in Fig 4b the optimal enzyme activities, mopt and ropt, along the Pareto fronts. For the “large size” regime at low λe, the bacterial population primarily needs to defend against phage escape, favoring low m and high r, even at the cost of self-restriction. As we move towards the “fast growth” regime, r can drop to decrease the cost, but m must increase to protect against self-restriction, until maximal mopt is reached. For even higher λe, it is optimal to “shut down” the RM systems altogether to save on the cost, by tuning r and m simultaneously to zero. Numerical analysis (S1 Appendix S.4.5) reveals that along the Pareto front of Fig 4a, the total cost of running the RM systems varies in precise inverse linear relationship with λe. Pareto-optimal solutions are further characterized by the fact that the reduction in growth rate, λ − λe, is split equally between the cost of running RM systems, c(r + m), and self-restriction. If this were not the case and the cost were larger (or smaller) than self-restriction cost to growth, cells could always down- (or up-)regulate the RM system activity to trade cost for self-restriction and obtain an overall smaller total growth reduction. This universal equality of cost of running RM systems and self-restriction at optimality is the second key result of the section.
A detailed examination of the Pareto front in Fig 4a reveals a striking shift in the structure of optimal solutions as we move from “fast growth” to “large size” regime. In situations where fast growth is favored, we observe that a single RM system (k = 1) is optimal. In contrast, large bacterial population sizes favor kopt > 1 RM systems, with the optimal number, kopt, set by the costs, cm and cr, of operating the RM systems. These results are quantitatively robust to changes in replication rate, λ, as shown in Fig 4c, where Pareto fronts for different λ are nearly rescaled versions of each other. These results are also qualitatively robust to changes in the cost c = cr = cm so long as the cost is nonzero, as shown in Fig 4d.
Establishing that “fast growth” regime favors simple innate immunity with a single RM system while “large size” regime favors complex innate immunity with multiple RM systems is the third key result of this section. This result can be understood intuitively by considering under what conditions, if any, multiple RM systems could be optimal at “fast growth”. If costs for R and M enzymes are vanishingly small, a single RM system can provide arbitrarily good protection, as we showed previously. If the costs are not vanishingly small, multiple RM systems must be more costly than a single system at comparable phage escape and self-restriction rates: to keep self-restriction constant with k RM systems, not only does the cell require k times more M molecules than at k = 1, but their individual activities need to be higher as well, leading to a higher cost for M and thus a lower effective growth rate; thus, k > 1 cannot be optimal for “fast growth” and can only be tolerated in the “large size” regime where protection from phages is more important than fast growth.
Lastly, we sought to put our results into perspective by relating them to a typical E. coli strain. Recent measurements [26] quantified the self-restriction rate in a bacterial population with the EcoRI system replicating at λ = 0.017 min−1 to be around μ ≈ 10−3 min−1. The cost of RM systems was not detectable in WT strain but could be detected in strains overexpressing M enzymes. Treating the cost c as unknown and assuming that E. coli is Pareto-optimal in light of criterion (i) in Table 1 (black dots in Fig 4c and 4d), would lead us to predict the following parameter values for the RM systems: cost c ≈ 3.7 ⋅ 10−7, enzyme activities r ≈ 1.2 ⋅ 103 min−1, m ≈ 1.5 ⋅ 102 min−1, with the optimal number of RM systems being at the boundary between k = 1 and k = 2. Clearly, this prediction depends on the chosen measure of the efficiency of RM systems, which is determined by the considered ecological scenario and the particular objective that bacteria have in this scenario. Consequently, the concrete numbers presented here should not be understood as general results, but rather as a demonstration of how our framework can be used to calculate optimal bacterial strategies given different modeling assumptions about the phage-bacteria ecology.
Despite the ubiquity of RM systems in prokaryotic genomes [17], basic ecological and evolutionary aspects of these otherwise simple genetic elements are poorly understood [20]. Although RM systems have been discovered more than six decades ago due to their ability to protect bacteria from phage [34] and this is often assumed to be their main function [35], only a few experimental studies focused on the ecological and evolutionary dynamics of interactions between RM systems and phage [36, 37]. Similarly, effects of RM systems on their host bacteria, such as their cost in individual bacteria due to self-restriction, began to be addressed quantitatively only recently [26, 38]. In this work, we bridged these two scales using mathematical modeling. Our model captures the stochastic nature of RM systems originating at the level of interacting molecules in individual bacteria and extends it all the way to the dynamics of interactions between bacterial and phage populations.
Using this approach, we analytically described tradeoffs between the cost and the efficiency in different ecological contexts of immunity conferred by RM systems. The existence of such tradeoffs was previously indicated by quantitative single-cell experiments with two RM systems isolated from E. coli [26] and has since then been reported in the context of other RM systems [39]. We used our mathematical framework to quantify these tradeoffs and to study their ecological consequences, as well as the implications that these consequences have for optimally tuning the R and M enzymatic activities at the level of single cells. Our results for different ecological scenarios suggest that we should expect observed expression levels and enzymatic activities of naturally occurring RM systems to represent adaptations to specific environmental pressures. Such “tuning” of expression levels towards optimality has previously been directly experimentally shown in different molecular systems [29]. The expression levels of both R and M should be readily tunable by mutations in the often complex gene-regulatory regions [40].
With optimal bacterial defense strategies depending on the ecological scenario and the particular objective of the bacteria (see S1 Appendix Section S.4 and Table 1), making general predictions on R and M expression levels or numbers of concurrently active RM systems that we should expect to find in bacteria in the wild is difficult. However, we want to highlight that, in a given context, assuming optimality of the bacterial defense strategy allows one to make clear and quantitative predictions about enzymatic activities and the number of RM systems, and improving these predictions to take into account more relevant biological detail (if needed and known) remains only a technical, rather than conceptual challenge. Second, for the ecological scenario that we investigated in detail in this paper, parameter values measured for an E. coli RM system put optimal solutions into a regime that permits a large variation in the optimal number of RM systems, between one to six, with relatively small changes in the effective growth rate. This observation allows us to advance the following hypothesis: the number of RM systems in different bacterial strains and species is not a historical contingency, but an evolutionary adaptation to different ecological niches. In other words, the tradeoff between the cost and the efficiency of immunity can be partially alleviated in bacteria employing multiple RM systems. It is therefore interesting to note that many bacterial species carry multiple RM systems and the number of RM systems varies significantly among bacteria with different genome sizes and lifestyles [16, 17]. Our results indicate that different numbers of RM systems would be optimal in populations under different selection pressures (phage predation/resource limitation).
The analytical model presented here makes several simplifying assumptions. First, we consider only interactions between a single species of bacteria and a single species of phage. In natural environments, many bacterial and phage species interact and this diversity will certainly impact the resulting ecological end evolutionary dynamics [36, 41–43]. Second, we assumed the key parameters such as the numbers of restriction sites in bacterial and phage genomes to be constant in time and thus disregarded the long-term evolutionary dynamics. Bioinformatic studies have shown that many bacteria and phage avoid using restriction sites in their genomes [44–46]. Restriction site avoidance can represent an adaptive mechanism for increasing the probability of escape in phages [45, 47] and decreasing the probability of self-restriction in bacteria [26, 48]. The stochastic nature of RM systems observed at the level of individual cells is thus likely to critically shape the ecological and evolutionary dynamics of interactions between bacteria, RM systems and phage.
|
10.1371/journal.pntd.0007016 | General contextual effects on neglected tropical disease risk in rural Kenya | The neglected tropical diseases (NTDs) are characterized by their tendency to cluster within groups of people, typically the poorest and most marginalized. Despite this, measures of clustering, such as within-group correlation or between-group heterogeneity, are rarely reported from community-based studies of NTD risk. We describe a general contextual analysis that uses multi-level models to partition and quantify variation in individual NTD risk at multiple grouping levels in rural Kenya. The importance of general contextual effects (GCE) in structuring variation in individual infection with Schistosoma mansoni, the soil-transmitted helminths, Taenia species, and Entamoeba histolytica/dispar was examined at the household-, sublocation- and constituency-levels using variance partition/intra-class correlation co-efficients and median odds ratios. These were compared with GCE for HIV, Plasmodium falciparum and Mycobacterium tuberculosis. The role of place of residence in shaping infection risk was further assessed using the spatial scan statistic. Individuals from the same household showed correlation in infection for all pathogens, and this was consistently highest for the gastrointestinal helminths. The lowest levels of household clustering were observed for E. histolytica/dispar, P. falciparum and M. tuberculosis. Substantial heterogeneity in individual infection risk was observed between sublocations for S. mansoni and Taenia solium cysticercosis and between constituencies for infection with S. mansoni, Trichuris trichiura and Ascaris lumbricoides. Large overlapping spatial clusters were detected for S. mansoni, T. trichiura, A. lumbricoides, and Taenia spp., which overlapped a large cluster of elevated HIV risk. Important place-based heterogeneities in infection risk exist in this community, and these GCEs are greater for the NTDs and HIV than for TB and malaria. Our findings suggest that broad-scale contextual drivers shape infectious disease risk in this population, but these effects operate at different grouping-levels for different pathogens. A general contextual analysis can provide a foundation for understanding the complex ecology of NTDs and contribute to the targeting of interventions.
| Variation in infectious disease risk between groups of individuals represent health inequalities: reducing these inequalities, alongside reductions in infection prevalence, is a major focus for public health interventions. Despite this, it is rare that general contextual effects, or measures of within-group correlation or between-group heterogeneity, are reported as substantive outcomes from community-based studies of infectious disease risk, including for the NTDs. This reflects wider issues around a lack of social epidemiological perspectives, or consideration of the effects of contextual drivers, in communicable disease research, particularly in low-income settings. The aim of this study was to measure general contextual effects on human infection risk for a number of endemic helminth, protozoal, bacterial and viral pathogens in a rural farming community in western Kenya. Using this approach, we reveal clustering at a range of administrative and geographic levels and are able to show that the magnitude of clustering, and the hierarchical grouping level at which it occurs (from the household to administrative constituency), varies substantially between pathogens. Greater within-group correlation and between-group heterogeneity in infection risk was observed for the helminth NTDs and HIV than for Entamoeba histolytica/dispar, Plasmodium falciparum or Mycobacterium tuberculosis. Quantification of general contextual effects can inform the design of interventions that aim to reduce health inequalities within a population and can provide actionable targets for assessing the short- and long-term impact of interventions.
| People living in rural areas in sub-Saharan Africa are often at high risk of infection with a range of pathogens [1–3]. The burden of preventable infectious disease in many of these communities can perpetuate poverty [4], reduce well-being [5,6], and contribute to high rates of mortality [7]. An individual’s risk of infection with any pathogen depends on a complex interplay of factors that relate to their exposure and susceptibility [8]. The individual-level characteristics that determine the likelihood of encountering a particular pathogen, and of infection following exposure, are often greatly influenced by the social, cultural, political, economic and/or environmental contextual conditions in which a person lives [9–11]. Since individuals living in the same geographic, administrative or institutional setting are generally exposed to the same contextual conditions (although not necessarily in the same way), adverse health outcomes commonly cluster within particular grouping levels. Hence, all else being equal, two people living in the same group will tend to be more similar in their health status than two people living in different groups [12]. Such clustering effects are often large for infectious diseases, and particularly so at the household-level for pathogens that are spread through poor sanitation, contaminated water, endophagic vectors, and unhygienic practices [13–18].
Clustering of infection within groups, and the contextual effects that drive it, such as marginalization, poverty and access to health services, is integral to the conceptualization of an infectious disease as ‘neglected’ [19]. However, it has been suggested that effects acting at the group-level are often forgotten in the epidemiological study of NTD infection risk [20,21], or indeed of infectious disease risk more broadly [22,23]. This apparent deficit in “contextual thinking” has occurred despite the widespread use of multi-level models, also called random effect or hierarchical models, in community-based studies of infectious disease risk in low income settings. One possible explanation for the absence of explicit contextual thinking is that the condition that necessitates the use of random effects in these multi-level models, namely the presence of within-group correlation in the outcome of interest, is almost never reported as an outcome of substantive interest in studies of NTD risk. Intra-group correlation (and the need for group-level random effects) is evidence that the grouping level chosen, be it household or geographic or administrative area, has a role in shaping variation in risk, and therefore points to the importance of group-level effects on individual infection.
A number of authors have described the value (and, it could be argued, the need [24]) of considering and reporting measures of general contextual effect, such as within-group correlation or between-group heterogeneity, from multi-level studies of disease risk [12,24–29]. Such effects are described as “general” as they refer only to influence of the cluster boundaries, rather than the specific contextual characteristics of the cluster [28]. Quantification of the extent and level at which infection risk varies between these clusters of individuals can contribute to the development of research questions that are explicitly contextual, and which therefore seek to better understand how the conditions in which people live impact upon their health [27,28]. Moreover, if health inequalities can be defined as differences in health status between groups of individuals [30], estimating general contextual effects (GCE), such as the median odds ratio or the intra-cluster correlation coefficient, can also provide a simple and standardized means with which to quantify and compare health inequalities within and between populations, and for different health outcomes [31]. Estimation of these group-level effects is straightforward to integrate into the multi-level analysis of community-based disease risk [32–35], and can provide fundamental information on the levels of variation that exist within a population.
Here, we describe a general contextual analysis that seeks to quantify the role of group-level effects in shaping variation in endemic NTD risk at a range of levels of aggregation in a rural farming community in Kenya. Since the NTDs commonly co-occur with HIV/AIDS, tuberculosis (TB) and malaria [36], we compare the GCE observed for NTDs with infection with pathogens causing these three diseases. In addition to describing the levels of variation in helminth, bacterial, protozoal and viral infection risk that exists within a single population, our aim is to use this analysis to demonstrate the value that can be added to the multi-level analysis of NTD risk through the quantification of GCE.
Data were collected as part of the ‘People, Animals and their Zoonoses’ (PAZ) study [37]. This was a large cross-sectional survey of all eligible and consenting members of 416 randomly selected households in a single, mixed farming community in western Kenya. In total, 2113 people of all ages meeting the inclusion criteria (≥ 5 years and without conditions that may have made blood sampling harmful) were included and sampled between September 2010 and July 2012. Samples from participants were tested for current infection with a range of pathogens. A questionnaire was conducted with all recruited participants in their preferred language (Kiswahili, Dholuo, Kiluhya or English).
Sampled individuals were nested within randomly selected households. These represented family groups living within a single compound, sharing meals and a common water source. The average reported household size was 7.6 people (range 1 to 30), from which our average sample size was 5.1 (range 1 to 21). Households were selected from within sublocations, the smallest administrative unit in Kenya. We sampled between 1 and 8 households in all 141 sublocations in the study area. The PAZ study focused on zoonotic disease risk, and the number of households selected per sublocation was proportional to the cattle population (see [37] for further details). Sublocations were nested within constituencies, the level at which government funding for development, and particularly for poverty alleviation, is allocated in Kenya. There were a total of 13 constituencies in the study area. On the basis of the 2009 census (OpenData, http://www.opendata.go.ke), sublocations in the study area had a median total population of 4,809 (range 1,187–33,352) and a median area of 10.8 km2 (range 0.96–64.6). Constituencies were made up of between 7 and 22 sublocations. The geographic distribution of sampled households, sublocations and constituencies is shown in Fig 1.
Several infectious agents, including a number that cause NTDs, were highly prevalent in the population under study. Here we focus on those with an individual-level prevalence after adjustment for the complex study design (see [37] for details) greater than 5%. These were Ancylostoma duodenale and/or Necator americanus (hereafter, hookworm) (36.3% (95% CI 32.8–39.9)); Entamoeba histolytica/dispar (30.1% (95% CI 27.5–32.8)); Plasmodium falciparum (29.4% (95% CI 26.8–32.0)); Taenia spp. (causing taeniasis) (19.7% (95% CI 16.7–22.7)); Taenia solium (causing cysticercosis) (5.8% (95% CI 4.4–7.2)); Ascaris lumbricoides (10.4% (95% CI 8.1–12.7)); Trichuris trichiura (10.0% (95% CI 8.2–11.7)); Mycobacterium tuberculosis (8.2% (95% CI 6.8–9.6)); Schistosoma mansoni (5.9% (95% CI 3.7–8.1)); and HIV (5.3% (95% CI 4.2–6.3)).
Individuals were classified as infected with P. falciparum, the only agent of malaria identified in the study area, if parasites were observed by light microscopy on thick or thin blood smears stained with Giemsa. Infection with the soil-transmitted helminths (hookworm, A. lumbricoides, T. trichiura) and S. mansoni was defined as the presence of at least one egg in a single faecal sample examined following preparation using the Kato-Katz (KK) [38] and formal ether concentration (FEC) techniques [39]. Infection with E. histolytica/dispar was defined as the presence of at least one cyst in a single faecal sample prepared using the FEC technique. M. tuberculosis infection was determined using a gamma-interferon assay (QuantiFERON-TB test, Cellestis) and HIV infection diagnosed using a rapid strip test (SD Bioline HIV 1/2 3.0, Standard Diagnostics). Infection with Taenia species (causing taeniasis, or the presence of an adult tapeworm in the gastrointestinal tract) was defined on the basis of a non-species specific copro-antigen ELISA [40], whilst cysticercosis due to T. solium (the presence of encysted larvae) was determined using a HP10-Ag ELISA on serum [41].
Ethical approval for this study was granted by the Kenya Medical Research Institute (KEMRI) Ethical Review Board (SCC1701). All participants or their guardians provided written informed consent. Individuals found to be infected with helminths or protozoa (including P. falciparum) were offered treatment free of charge by study clinical officers. Referral to local health facilities was provided where necessary.
The entire sample of 2113 people was used for the general contextual analysis. Missing-ness was present in all outcome measures and ranged from 0.05% (for P. falciparum) to 11.1% (for M. tuberculosis). Missing-ness was related to an absence of a particular sample type (blood or faeces), typically due to inadequate volumes collected or because of participant unwillingness to provide it.
Four-level logistic regression models were specified with infection as a binary outcome (infected/not infected) for each pathogen. Probability of infection was related to a set of predictors at the individual-level and random effects at the household-, sublocation- and constituency-levels. These models estimated the log odds of individual infection together with the variance at the intercept for the household (σ2H), sublocation (σ2SL) and constituency (σ2C) levels for an individual i living in household j in sublocation k in constituency l. The regression equation can be summarised as logit(πijkl) = β0 + βX + H0jkl + SL0kl + C0l. Our primary motivation for this analysis was to quantify general (rather than specific) contextual effects operating at each of the three grouping levels. However, age, sex, education status and ethnicity were included as fixed effects, X, at the individual level in order to assess the impact of within-household composition on between-group variation. Models with and without fixed effects were estimated for each pathogen. A quadratic term was included for the continuous predictor age (recorded as 5 year intervals) based on the expectation of non-linear relationships with infection risk for several pathogens [37]. The continuous age variable was scaled to have a mean of zero and standard deviation of one.
Models were estimated for each pathogen in WinBUGS 1.4.3 (http://www.mrc-bsu.cam.ac.uk/software/bugs/) using weakly informative normal priors for all fixed and random effects. The standard deviation for each of the group-level random effects was defined using a wide uniform hyper-prior (i.e. Uniform(1,100)). Model convergence was confirmed by visual assessment of MCMC chains. Inference was based on 3 chains that were allowed to run for at least 70,000 iterations after a burn-in of at least 30,000 with a thinning interval of at least 10. We derived the median and 2.5th and 97.5th percentiles from posterior distributions of each parameter for point estimates and 95% credibility intervals, respectively. All data manipulation was performed in R statistical environment (R version 3.1.1, http://cran.r-project.org/) with logistic regression models estimated via the R2WinBUGs package [42]. Estimation was performed within a Bayesian framework based on MCMC to reduce bias in the estimates for random effect parameters [43], and for ease of estimation of the associated uncertainty for GCE.
Geographic effects not captured in the non-spatial multi-level logistic regression models were identified by testing the standardised sublocation level residual log odds for evidence of spatial clustering in high or low values using the spatial scan statistic [46]. The default maximum cluster size of 50% of the sample was chosen using a circular spatial window. The sublocation was used as the highest contextual level for the exploration of spatial clustering due to the small number of groups at the constituency level (n = 13). We used a normal model in SatScan version 9.4.4 (www.satscan.org). To account for differences in sample sizes, the number of individuals sampled in each sublocation were included as model weights [47]. Sublocation residuals for spatial analysis were drawn from a three-level logistic regression model (with random effects for household and sublocation only) with and without adjustment for within-household compositional effects.
The variation in prevalence of each infectious agent across the range of variables included as fixed effects is shown in Table 1. Variation in prevalence of infection between self-reported members the different ethnic groups was particularly apparent, and most notably so for A. lumbricoides, T. trichiura, Taenia spp. (causing taeniasis) and HIV. Heterogeneity in the prevalence of infection with each of these pathogens, and with S. mansoni and T. solium (causing cysticercosis), was also evident between constituencies.
Co-efficients from the adjusted models (M2) for each pathogen are shown in Table 2 (STH and S. mansoni), Table 3 (E. histolytica/dispar, Taenia spp. and T. solium) and Table 4 (HIV, P. falciparum, M. tuberculosis). Male gender was associated with increased odds of hookworm and S. mansoni infection, with weaker evidence for taeniasis. Females had greater odds of T. trichiura, E. histolytica/dispar, and HIV infection and T. solium cysticercosis. There was no evidence of a relationship between sex and A. lumbricoides (Table 2), M. tuberculosis, or P. falciparum (Table 4) infection. Hookworm (Table 2), M. tuberculosis and HIV (Table 4) infection increased with age, with evidence in each case of a negative quadratic effect. Infection declined with age for T. trichiura, A. lumbricoides (Table 2) and P. falciparum (Table 4). Having an education beyond primary school tended to reduce odds of infection for the majority of pathogens under study, although this was only significant in the case of hookworm (Table 2). There were strong relationships between ethnicity and infection for several pathogens, including substantially reduced odds among people of Samia and Teso ethnicity for A. lumbricoides compared to the Luhya baseline. Odds of T. trichiura infection were reduced among people of Teso ethnicity and elevated among people of Luo ethnicity when compared to the Luhya baseline. The odds of HIV infection was also higher among individuals of Luo ethnicity than the Luhya baseline.
The posterior distribution of household-, sublocation- and constituency-level variance, VPCs, MORs and PTVs for the gastrointestinal nematodes and S. mansoni are shown in Table 2, in Table 3 for E. histolytica/dispar and Taenia species, and in Table 4 for HIV, P. falciparum and M. tuberculosis. Some degree of clustering at the household-level was apparent for all pathogens. This was consistently highest for the helminth parasites (Fig 2), for which there was substantial heterogeneity in risk of infection between individuals in different households, as evidenced by MORs which exceeded 3.5 for each helminth infection in both the null and adjusted models (Table 2 and Table 3). To put these effects into context, we would expect that were an individual to permanently move from one household to another with higher risk anywhere in the study area, their odds of infection with the helminth parasites under study would change by at least 3.5 times. This household clustering effect was particularly large for S. mansoni (Table 2) and T. solium cysticercosis (Table 3). The partitioning of group-level variation was generally largest at the household-level, although the greatest proportion of individual variation was partitioned at the constituency level (VPCc) in null models for T. trichiura (Table 2) and HIV (Table 4), and the sublocation-level for T. solium cysticercosis (PTVSL) (Table 3). Using MORs, these higher-level contextual effects could be interpreted as an almost five- and three-fold change in the odds of infection for an individual that permanently moves to a higher risk constituency for T. trichiura and HIV, respectively. Similarly, the median odds of an individual permanently moving to a higher risk sublocation could be expected to increase by around eight times for T. solium cysticercosis. Control for individual-level fixed effects resulted in declines in within-constituency correlation (VPCC) and between-constituency heterogeneity (MORC) for infection with several of the pathogens under study, most notably for A. lumbricoides and T. trichiura (Table 2) and HIV (Table 4).
The spatial distribution of sublocations with evidence for clustering in high or low values of residual log odds of infection is shown in Fig 3. Large spatial clusters of both high and low values were observed from null models for T. trichiura, S. mansoni, A. lumbricoides, and Taenia spp.. There was substantial overlap in clusters for all of these pathogens and a large cluster of sublocations with elevated risk of individual HIV infection. We found no evidence of spatial structuring in the sublocation-level residual log odds of infection with M. tuberculosis or T. solium and relatively small clusters for P. falciparum, hookworm and E. histolytica/dispar (Fig 3). The spatial extent of the clusters of both high and low sublocation residual log odds was reduced when controlling for individual-level fixed effects in the case of HIV. Adjustment for these fixed effects resulted in a loss of significance in spatial clusters of both high and low values from the model for A. lumbricoides, and of high values for T. trichiura. Only the spatial cluster of positive sublocation residual log odds remained significant in the case of S. mansoni (Fig 3).
In this general contextual analysis, we demonstrate the value of summarizing variation in individual infectious disease risk at one or more biologically relevant grouping levels using the outputs from multi-level regression. Deriving statistics such as the MOR and VPC (or ICC) as part of an exploratory analysis of infectious disease risk is straightforward, and can contribute important information about the heterogeneity that underlies population-level averages, such as prevalence [26–28,33]. Using this approach, we show that variation in individual infection risk is partitioned at the household, sublocation and constituency-levels for a range of NTDs in a rural population in Kenya. These findings point to the importance of social and/or environmental contextual conditions in shaping infection at each of these levels, and which may provide actionable targets for public health interventions seeking to reduce both the prevalence of infection and the health inequalities observed.
An important limitation that should be recognised when interpreting these findings, and particularly when making comparisons between pathogens, is the lack of precision in many of our estimates of GCE, particularly at higher contextual levels. Hence, whilst estimates of VPC and MOR at the constituency-level were substantially different between, for example, hookworm and S. mansoni infection, the 95% credibility intervals overlap. This is a limitation of the sample available, both in terms of number of individuals and number of individual groups at the higher contextual levels. The magnitude of the MOR or VPC provides useful information on the importance of a particular level in structuring risk [28], and for the example of hookworm and S. mansoni, strongly suggests contextual drivers operating at the constituency level are more important for the latter than the former. However, when interpreting differences between pathogens at these higher contextual levels, or between different contextual levels for the same pathogen, it should be noted that the statistical support for many of the differences we observed was often limited.
A general contextual analysis can provide a tool for exploring the levels at which pathogen transmission occurs within a population [16]. For example, we show that the majority of variation in individual hookworm infection was partitioned at the household level, with comparatively smaller amounts at sublocation and constituency levels. This suggests clustering at higher contextual levels is less important for this parasite in this population than for the other STHs. Individual infection with A. lumbricoides, for example, was partitioned at both the household- and constituency-levels, and therefore household clusters of infection can also be considered to cluster by constituency. Household clustering was less important for T. trichiura, but there was substantial variation in infection between constituencies, and to a lesser extent between sublocations within constituencies. Understanding these patterns of partitioning in infection risk may assist in the design of interventions that seek to reduce both the prevalence and health inequalities observed. For pathogens with limited evidence for higher level GCE, such as hookworm or E. histolytica/dispar, it is likely that households in all parts of the study area would need to be targeted. Interventions in high risk constituencies are likely to be more cost effective for T. trichiura, A. lumbricoides and S. mansoni, potentially including a focus in high risk sublocations for the latter two pathogens. The general contextual analysis approach described here could be particularly valuable in monitoring the effectiveness of an intervention, such as mass drug administration. For example, a decline in population-level prevalence but persistence of, or increase in, general contextual effects at particular grouping-levels would point to ongoing or new health inequalities. Moreover, such a finding would suggest the presence of hotspots of transmission that may impact elimination [48]. Wider usage of general contextual analysis in the study of NTD risk could therefore contribute to the post-2020 NTD roadmap that sees a transition from monitoring programme coverage to measuring impact [49].
Clustering in T. solium cysticercosis and Taenia spp. taeniasis was observed at both the household and sublocation levels. This was particularly large at the sublocation level for T. solium cysticercosis, but not between constituencies. Hence, while spatially heterogeneous factors appear to influence cysticercosis risk, these effects are likely to operate at small spatial scales (i.e. at the sublocation-level). Cases of human cysticercosis commonly cluster around human tapeworm carriers [50], and Okello et al [51] reported hyper-endemic hotspots for T. solium infection in Lao PDR. The importance of non-spatially-structured sublocation effects in our own study area could therefore be hypothesised to reflect small-scale differences in pork consumption practices, or the existence of slaughterhouses in particular sublocations with inadequate meat inspection practices. Sublocation-level residuals for taeniasis showed substantial spatial structuring on the basis of the spatial scan statistic, and the lack of a similar finding for cysticercosis may point to a preponderance of the beef tapeworm, T. saginata (which does not cause human cysticercosis) over T. solium in the study area.
The nesting of variation in individual HIV infection at the constituency level supports the growing recognition that HIV epidemiology can be characterized as a number of diverse epidemics, often with substantial variation in prevalence even at small spatial scales [52,53]. In this part of western Kenya, individual risk of HIV infection was most concentrated in constituencies in the south-western part of the study area. Further work is needed to explore the important clustering observed, including the compositional effect of ethnicity; the Luo community who, as a group, have been previously been described to be heavily burdened by HIV [54], reside primarily in the southern part of the study area [37]. Schistosoma haematobium, which we did not test for but which is known to be an important co-factor for HIV infection in sub Saharan Africa [55], is also likely to be common in the swampy area around Lake Victoria [56], and may also contribute to the clustering observed. There were substantial overlaps in the spatial distribution of HIV infection risk and that for several NTDs, most notably S. mansoni, A. lumbricoides and T. trichiura. This supports earlier analysis of the same data that showed overlapping spatial clustering in household-level infection with these pathogens [37]. The observed co-distribution of these pathogens may point to the existence of shared environmental, cultural, behavioural or social conditions leading to poly-parasitism [19]. Alternatively, it may suggest immunological interactions between HIV and these helminth parasites that influence transmission dynamics, a hypothesis supported by a growing number of field and laboratory based studies [57].
Interestingly, between-group levels of variation were considerably lower for P. falciparum and M. tuberculosis than for any of the NTDs, with the exception of infection with E. histolytica/dispar. Previous studies on M. tuberculosis have suggested that the majority (>80%) of transmission events for the pathogen occurs in the public (or community) rather than domestic domain [58–60]. The comparatively small levels of individual variation partitioned at the household-level (particularly compared to the helminth pathogens under study) provides further support for these findings. Moreover, in the absence of higher level GCEs, we show there is little variation in community-level transmission between different parts of the study area for M. tuberculosis. Although we found evidence for a small cluster of sublocations with reduced risk of P. falciparum infection, the absence of higher-level contextual effects (at the sublocation- and constituency-level) for this pathogen suggests geographic or administrative place of residence does not have a major influence on infection risk. This is supported by a recent study from neighbouring Eastern Uganda which, using highly sensitive molecular-based diagnostic tests, demonstrated that the vast majority of community residents, regardless of age, demography and geographic location, were infected with malaria parasites [61].
We have explored only a limited set of fixed effects at the individual level in this analysis, and no specific contextual effects (i.e. predictors operating at group-level). Having demonstrated the importance of these grouping-levels in structuring infectious disease risk, the next analytical step would be to integrate specific contextual effects, including household, sublocation and constituency-level indicators of social or environmental conditions that may explain the variation observed. The inclusion of individual-level predictors resulted in substantial decreases in the variation at higher contextual levels for pathogens such as A. lumbricoides, T. trichiura and HIV. There were large, overlapping spatial clusters for each of these pathogens, the size of which was reduced or made to be non-significant following the inclusion of individual level predictors. All of these pathogens had strong relationships with ethnicity, which is known to be highly spatially structured in the study area [37]. Disentangling the importance of individual level cultural and behavioural practices and local social and environmental conditions would therefore help to better understand the general contextual effects observed.
Quantification of general contextual effects provides a means to evaluate the importance of social and environmental conditions in structuring infectious disease risk within a population. Such an approach encourages the explicit consideration of group-level, contextual effects on individual health and can form the basis for subsequent analyses that seek to explain the variation observed. Using a general contextual analysis, we have demonstrated the existence of important place-based contextual effects for a range of pathogens in a rural farming community in Kenya and show that these are particularly large for the NTDs and HIV. This study provides evidence for important variation in infectious disease risk in this underprivileged population that point to the existence of health inequalities at a range of grouping-levels.
|
10.1371/journal.ppat.1005389 | Bacillus thuringiensis Crystal Protein Cry6Aa Triggers Caenorhabditis elegans Necrosis Pathway Mediated by Aspartic Protease (ASP-1) | Cell death plays an important role in host-pathogen interactions. Crystal proteins (toxins) are essential components of Bacillus thuringiensis (Bt) biological pesticides because of their specific toxicity against insects and nematodes. However, the mode of action by which crystal toxins to induce cell death is not completely understood. Here we show that crystal toxin triggers cell death by necrosis signaling pathway using crystal toxin Cry6Aa-Caenorhabditis elegans toxin-host interaction system, which involves an increase in concentrations of cytoplasmic calcium, lysosomal lyses, uptake of propidium iodide, and burst of death fluorescence. We find that a deficiency in the necrosis pathway confers tolerance to Cry6Aa toxin. Intriguingly, the necrosis pathway is specifically triggered by Cry6Aa, not by Cry5Ba, whose amino acid sequence is different from that of Cry6Aa. Furthermore, Cry6Aa-induced necrosis pathway requires aspartic protease (ASP-1). In addition, ASP-1 protects Cry6Aa from over-degradation in C. elegans. This is the first demonstration that deficiency in necrosis pathway confers tolerance to Bt crystal protein, and that Cry6A triggers necrosis represents a newly added necrosis paradigm in the C. elegans. Understanding this model could lead to new strategies for nematode control.
| Necrosis contributes to many devastating pathological conditions, such as neurodegenerative diseases and microbial pathogenesis. Bacillus thuringiensis crystal proteins are effective biopesticides. Our study reveals that B. thuringiensis Cry6Aa protein triggers the necrosis pathway using Caenorhabditis elegans as a model. We show that aspartic protease ASP-1 is required for Cry6Aa protein-induced necrosis, whereas intrinsic insults induce necrosis mediated by ASP-3 and ASP-4. Our findings contribute to the understanding of the mechanism of Bt crystal protein action and host-pathogen interactions. Because necrosis mechanisms are conserved from nematodes to humans, the fact that necrosis can be induced by Cry6Aa provides a model system for studying necrosis mechanisms in human diseases.
| Cell death plays critical roles in development and in pathological conditions. Apoptosis and necrosis are the two major modes of cell death [1]. Apoptosis, the most well- known mode of the cell death, plays a significant role in development, tissue homeostasis, and host defense [2,3]. Unlike apoptosis, necrosis is characterized by loss of plasma membrane integrity [4,5]. Necrotic cell death can contribute to many pathological conditions, such as inflammation [6], human neurodegenerative and aging-associated diseases [5,7]. Moreover, Necrosis plays an important role in microbial pathogenesis. In some cases, necrosis plays a significant role in antiviral/antibacterial host defense [2,8]; in others, necrosis is utilized as pathogen survival strategy to aid its spread [2].
Bacillus thuringiensis (Bt) is one member of the Bacillus cereus group of bacteria [9]. An important characteristic of Bt strains is that they produce insecticidal crystal proteins (Cry) during the sporulation phase. These proteins are highly specific to their target insects and nematodes and are harmless to non-target animals and humans, thus, they represent a viable alternative for the control of pests in agriculture and of important disease vectors in human public health [10].
The classical pore-forming model remains the widely accepted mode of action of the three-domain crystal protein (3d-Cry). When susceptible larvae ingest the 3d-Cry protoxin, it is activated by gut proteases. Upon activated toxin binding to cadherin receptor, toxins forms oligomers. Subsequently, these oligomers bind to a second group of receptor proteins. Finally, they generate toxin pores in the cell membrane that leading to midgut cells lysis [10–13]. There is little information on the cellular mechanisms of the classical pore-forming model. An alternative signaling pathway model of the 3d-Cry action has also been reported. This model disregards pore formation and proposed that crystal toxins activate a Mg2+-dependent adenylyl cyclase (AC)/protein kinase A (PKA) signaling pathway by interacting with receptor [14,15]. Nevertheless, this signaling pathway has only been identified in one insect cell line. There is no published data showing a signaling pathway involved in whole larvae death and the mode of action by which crystal toxins to induce cell death is not completely understood.
Crystal proteins have also been shown to intoxicate nematode parasites of animals [16] and plants [17–19]. Nematicidal activity has been found in several families of Bt crystal proteins, including Cry5, Cry6, Cry12, Cry13, Cry14, Cry21, and Cry55 [17]. Cry5Ba and Cry6Aa represent two distinct families. The structure of Cry5Ba is similar to that of insecticidal 3d-Cry, and shows the conserved three-domain (3-d) architecture responsible for pore formation in insecticidal crystal proteins [20]. However, Cry6Aa does not show the typical 3-d architecture. In contrast to the insect model, the mode of action of nematicidal crystal toxins has been investigated only in C. elegans using the Cry5Ba [21]. As previously shown, the receptors for crystal protein Cry5Ba in C. elegans are invertebrate-specific glycolipids [21]. The Cry5Ba-resistant glycolipid mutants are sensitive to Cry6Aa [22]. The above information implies that the Cry6Aa may utilize a different toxicity pathway from Cry5Ba.
In C. elegans, necrotic cell death can be induced by extreme environmental stimuli or intrinsic insults, including hypoxia [23], ionic imbalance, heat stroke, bacterial infection, and hypoosmotic shock [24]. In this study, we found that necrotic cell death can also be induced by B. thuringiensis crystal protein Cry6Aa. This pathway induced by Cry6Aa is mediated by aspartic protease (ASP-1). Additionally, ASP-1 protects the crystal protein Cry6Aa from over-degradation by C. elegans.
Proteomic approaches have previously been used to identify B. thuringiensis toxin binding proteins [25] and toxin receptors [26] and to understand insect resistance to B. thuringiensis [27]. Cry6Aa binding proteins in C. elegans were identified using 2-DE proteomics and ligand blotting. Separation of the C. elegans proteins by isoelectric focusing with a pH 4–7 IPG strip (18 cm) revealed that most proteins detected in 2-DE separations were smaller than 175 kDa and fell in the pH range of 4–6 (S1A Fig). Two-dimensional gel blots were probed with biotin-labeled Cry6Aa to identify C. elegans proteins that bound to Cry6Aa (S1B Fig). Fourteen silver-stained protein spots were detected (S1A Fig) and selected for identification. Four of these proteins were identified by mass spectroscopy and peptide mass fingerprinting (PMF) (S1 Table), whereas the other spots were not identified. Spot 1 matched aspartic protease (ASP-1) from C. elegans, and Spots 2, 3 and 4 matched an ATP synthase subunit family member (ATP-2), actin and F58E2.4, respectively.
ASP-1 is mainly distributed in the intestinal cells of C. elegans [28]. The toxicity of crystal protein Cry6Aa toward C. elegans is believed to involve intestinal damage [29]. Therefore, ASP-1 is likely to be involved in Cry6Aa toxicity against C. elegans. To further characterize the interaction between ASP-1 and Cry6Aa, a ASP1-GST fusion protein was over-expressed in Escherichia coli (Fig 1A, lane 3) and was purified by affinity chromatography with glutathione Sepharose 4B (Fig 1A, lane 4). Ligand blot experiments showed that crystal protein Cry6Aa can bind to purified ASP1-GST fusion protein (Fig 1B, lane 6). In the control experiment, Cry6Aa did not bind to purified GST (Fig 1B, lane 7).
To facilitate the measurement of Cry6Aa toxin binding under non-denaturing conditions, purified ASP1-GST was dotted in increasing amounts on a membrane filter, and the filter was probed with biotin-labeled Cry6Aa. GST was a negative control, dot blot showed 3.2 μg GST did not bind to Cry6Aa (Fig 1C). As the amount of dotted ASP1-GST protein increased, more biotin-labeled Cry6Aa was bound, and excess unlabeled Cry6Aa (1000-fold) competed for biotin-labeled Cry6Aa binding, which indicated that the binding between Cry6Aa and ASP1-GST was specific (Fig 1C). To measure the binding affinity of Cry6Aa to ASP1-GST, a competitive ELISA was performed. The dissociation constant (Kd) for ASP1-GST binding to Cry6Aa was 145.7 nM (126.8–164.7 nM) (Fig 1D). To assess whether ASP-1 and Cry6Aa interact directly, we determined the binding affinity between ASP-1 and Cry6Aa by isothermal titration calorimetry (Fig 1E). The estimated Kd was 126.4 nM (99.0–153.8 nM) (Fig 1E).
To test the role of Asp-1 in vivo, we compared the difference in larval growth inhibitory activities (GI) and mortality after treatment with Cry6Aa between wild type C. elegans N2 and mutant asp-1 (tm666). For the quantitative growth test, the GI50 values for N2 and asp-1 (tm666) were 4.9 ± 0.6 μg/mL and 45.1 ± 3.7 μg/mL, respectively (S2 Table and Fig 2A). Their LC50 values were 63.7 ± 6.8 μg/mL and 715.6 ± 69.3 μg/mL, respectively (S3 Table and Fig 2B). Lifespan measurements were performed in N2 and asp-1(tm666) upon exposure to Cry6Aa. As expected, the fraction of asp-1(tm666) alive was significantly higher than that of N2 (Fig 2C). The effects of different concentrations of Cry6Aa on the growth of L1 larvae of C. elegans N2 (top panels) and mutant asp-1(tm666) (middle panels) are summarized in Fig 2D. L1 larvae of N2 (Fig 2D, top panels) were not able to progress to adulthood after exposure to 30 μg/mL Cry6Aa. However, the L1 larvae of mutant asp-1(tm666) (Fig 2D, middle panels) were able to progress to adulthood after the same exposure. These experiments indicated that the mutation of asp-1 reduces the nematode’s susceptibility to Cry6Aa.
When we transformed asp-1 (tm666) with the whole asp-1 gene, the rescued L1 larvae of asp-1 (tm666) were unable to progress to adulthood after exposure to 30 μg/mL Cry6Aa (Fig 2D, bottom panels, and Fig 2E), which indicates that the whole asp-1 gene was sufficient to restore susceptibility to Cry6Aa in the asp-1 (tm666) mutant background. Taken together, these results indicate that ASP-1 is required for Cry6Aa-induced C. elegans death.
Necrosis in C. elegans requires an increase in intracellular Ca2+ levels mediated by the inositol triphosphate receptor ion channel (ITR-1), the calcium-dependent cysteine protease mediated by calpain TRA-3, lysosomal lysis for cytosolic acidification mediated by the vacuolar proton translocating ATPase (VHA-12), and the destructive release of lysosomal killer cathepsin proteases mediated by ASP-3 [44] or ASP-4 [43,44].
To examine whether necrosis is implicated in host defenses or for disease susceptibility during Cry6Aa targeting nematode, we compared the difference in mortality after treatment with Cry6Aa between wild type C. elegans N2 and necrosis mutants. Relative to the wild type, mutations in itr-1(sa73), tra-3(e1107), and vha-12(ok821) reduced the sensitivity of nematodes to Cry6Aa since their LC50 values were 614.9 ± 70.6 μg/mL, 539.8 ± 81.4 and 594.2 ± 68.5 μg/mL, respectively (S3 Table and Fig 7A). Lifespan measurements were performed in N2 and these mutations upon exposure to Cry6Aa. As expected, the fraction of itr-1(sa73), tra-3(e1107), and vha-12(ok821) alive was significantly higher than that of N2 (Fig 7B). These experiments indicated that these mutations are tolerant to the crystal protein Cry6Aa. However, neither the necrosis mutant asp-3 (tm4559) nor asp-4 (ok2693) significantly suppressed the induction of nematode death by Cry6Aa (S3 Table and Fig 7C). Heat stroke was a positive control, as expected, relative to the wild type, mutations in itr-1(sa73), tra-3(e1107), vha-12(ok821), asp-3 (tm4559) and asp-4 (ok2693) reduced the sensitivity of nematodes to heat stroke (Fig 7D).
Taken together, these results indicate that a deficiency in the necrosis pathway confers tolerance to Cry6Aa.
The classical, pore-forming model is the widely accepted model for describing the mode of action of 3d-Cry. This model elucidates crystal protein action at the biochemical level, which includes crystal protein activation, receptor binding, pore formation, and cell lysis [10–13]. The identified receptors in the pore-forming model, such as cadherin, aminopeptidase (APN), and alkaline phosphatase (ALP), mediate insect resistance to 3d-Cry [45]. However, these receptors were not identified in C. elegans by the Cry6Aa-ligand blotting assay. These results indicate that the mode of action of non-3d crystal protein Cry6Aa against C. elegans may be different from the classical pore-forming model. The signal transduction model proposed that crystal proteins activate a Mg2+-dependent adenylyl cyclase (AC)/protein kinase A (PKA) signaling pathway [14,15]. This work show that crystal protein Cry6Aa triggers the Ca2+-dependent calpain–cathepsin necrosis pathway in C. elegans. Thus, in contrast to the classical pore-forming model, which elucidated at the biochemical level, the present findings reveal that crystal toxin triggers cell death by necrosis signaling pathway. This is the first demonstration that deficiency in necrosis pathway confers tolerance to Bt crystal protein.
Apoptosis occurs in embryonic and larval development. The CED-3 caspase, which is regulated by pro-apoptotic (CED-4, EGL-1) and anti-apoptotic (CED-9) factors, is involved in apoptosis [36,46]. However, we found that neither ced-3 (n717), ced-4 (n1162), nor ced-9 (1950) mutation changed Cry6Aa-induced death (S9 Fig). Therefore, cell death induced by Cry6Aa does not depend on the apoptotic machinery.
Autophagy is required for necrosis induced by prolonged hypoxia in C. elegans [47]. Conversely, autophagy protects C. elegans against necrosis during Pseudomonas aeruginosa infection [48]. Lgg-1 encodes the C. elegans ortholog of Atg8/LC3 that facilitates autophagic vesicle growth, and lgg-1 is involved for autophagy [38,49]. Unc-51 encodes the serine threonine kinase ortholog of yeast autophagy protein Atg1 [50]. However, we found that neither unc-51(e1189) nor lgg-1(bp500) mutations suppressed or promoted Cry6Aa-induced death (S10 Fig). The above information suggest that autophagy plays different role respond to different pathological conditions.
An increasing number of bacterial pathogens have been shown to induce necrosis in host cells [2]. For example, Erwinia carotovora, Photorhabdus luminescens, and Enterococcus faecalis are reported to induce necrosis in C. elegans [51]. β-Toxin from Clostridium perfringens, alpha-toxin from Clostridium septicum, and Helicobacter pylori VacA [2] are reported to cause necrosis. Our study reveals that B. thuringiensis toxin triggers necrosis pathway in C. elegans. That Cry6A triggers necrosis represents a newly added necrosis paradigm in the C. elegans. Our findings contribute to the understanding of the mechanisms of host-pathogen interactions in higher species.
Necrosis plays an important role in host-pathogen interactions which can be used by both sides [2]. In some cases, necrosis plays a significant role in antiviral/antibacterial host defense [2,8]. For example, herpes simplex virus 1 (HSV-1) protein is reported to trigger an effective host-defense mechanism by activating RIP3/MLKL-dependent necrosis [52]. In others, necrosis does not as a host defense, but as pathogen survival strategy to aid its spread [2]. For example, cIAP2-dependent antagonism of RIPK3-mediated programmed necrosis critically protects the host from influenza infection [53]. Besides, it is reported that E. carotovora and P. luminescens used the necrosis to develop their effective virulence [51]. Our study reveals that deficiency in necrosis pathway confers tolerance to Bt crystal protein. It is possible that necrosis plays different role respond to different pathogens or toxins.
In the present work, evidence is presented that Bt crystal protein Cry6Aa triggers the necrosis pathway in C. elegans mediated by ASP-1, not by ASP-4 or ASP-3. To test whether ASP-1 is specifically required for Cry6A-induced necrosis, the asp-1(tm666) nematodes are exposed to two other death-inducing stimuli, heat stroke and hypoxia. We found that both mutant asp-1(tm666) and the wild-type N2 were sensitive to either heat stroke or hypoxia at the same level (S11 Fig). Thus, there is some difference in the necrosis pathway induced by Cry6Aa compared to other stressors, and ASP-1 is specifically required for Cry6A-induced necrosis. Previous studies of the cellular necrosis pathway have largely focused on neurodegeneration [43], where intrinsic insults induce necrosis mediated by ASP-3 and ASP-4 [44]. Recently, this pathway in the nematode intestine was reported, where lethal stress induced necrosis in C. elegans mediated by ASP-4 [43]. We recently found a two-domain protein named Nel, which is composed of a necrosis-inducing phytophthora protein 1-like domain found in phytopathogens and a ricin B-like lectin domain, induced necrosis mediated by ASP-4 [54]. Therefore, we speculate that different aspartyl proteases in the necrosis pathway respond to different death-inducing stimuli.
Our study show that Cry6A binds to ASP-1 and that ASP-1 mediates the protection and stabilization of Cry6Aa. This stabilization effect could be similar to that cadherin-mediated protection of Cry1Fa toxin from protease degradation in the insect gut [55]. ASP-1 is mainly distributed in the intestinal cells of C. elegans [28]. To explain how an intracellular ASP-1 protein binds Cry6Aa, we fed rhodamine labeled Cry6A to C. elegans, and monitored the signal of rhodamine-labeled crystal proteins with confocal microscope. As expected, the labeled Cry6A toxin was internalized into intestinal cells (S12 Fig).
A phylogenetic tree of the aspartyl proteases were constructed. ASP-1 from C. elegans belongs to the nematode-specific aspartyl protease class; however, ASP-3 and ASP-4 from C. elegans belong to a distinct branch that also includes mammalian proteases [44]. These results provide a plausible explanation for the lack of toxicity of Cry6Aa toward mammals. We hypothesize that the Bt non-three-domain Cry6Aa evolved to recognize the nematode specific ASP-1 and thus to target nematodes.
Griffitts et al. reported that C. elegans glycolipid mutants with resistance to Cry5Ba were not resistant to Cry6Aa [22]. In the present study, the C. elegans necrosis mutants with tolerant to Cry6Aa, were not tolerant to Cry5Ba. The above observations indicate that pathway of Cry6Aa against nematodes is different from that of Cry5Ba. A possible explanation for this difference is that the sequence and structure of the Cry6Aa toxin are different from those of Cry5Ba. The structure of Cry5Ba is similar to that of insecticidal 3d-Cry, and shows the typical conserved three-domain (3-d) architecture responsible for pore formation in insecticidal crystal proteins [20]. However, the amino acid sequence of Cry6Aa is completely different from that of 3d-Cry, and Cry6Aa does not show the typical 3-d architecture. Accordingly, further research is necessary to determine the structure of Cry6Aa.
Both B. thuringiensis and nematodes coexist in the soil ecosystem. We recently reported that nematodes are an alternative dominant host that contributes to the persistence, growth, and transmission of B. thuringiensis [56]. Some strains of B. thuringiensis have evolved to become nematode pathogens and use the host to reproduce, other non-nematicidal strains can reproduce on nematodes killed by other means or can still use nematodes as a means to reach their true host [56]. Bt has evolved crystal proteins to help it reproduction inside the host nematode [29]. On the other hand, C. elegans has evolved some conserved pathways to protect against B. thuringiensis crystal proteins [57,58]. Our study reveals that B. thuringiensis toxin triggers cell death by the necrosis pathway. Therefore, our findings have potential implications for understanding the co-evolution between nematodes and B. thuringiensis.
In conclusion, the present study reveals that crystal toxin triggers the necrosis pathway using Cry6Aa-C. elegans toxin-host interaction system, which involves an increase in concentrations of calcium, lysosomal lyses, uptake of propidium iodide, and burst of death fluorescence (Fig 11). We show that aspartic protease ASP-1 is required for Cry6Aa induced necrosis, whereas intrinsic insults induce necrosis mediated by ASP-3 and ASP-4. Understanding this model could lead to new strategies for nematode control. Further research is necessary to establish the existence of additional steps in Cry6Aa action: for example, what is the functional receptor in the intestinal surface and the role of receptor in crystal toxin action? Identification of the receptor would provide new avenues for studying the nematicidal mechanisms of Cry6Aa.
Maintenance of C. elegans strains was performed as previously described [59]. The mutated and transgenic strains used in this study include the following (S4 Table): ced-3(n717), ced-4(n1162), ced-9(n1950), itr-1(sa73), tra-3 (e1107), vha-12(ok821), asp-4(ok2693), unc-51(e1189), lgg-1(bp500), pwIs50[lmp-1::GFP + Cbr-unc-119(+)], and rnyEx109 [nhx-2p::D3cpv + pha-1(+)] were provided by the Caenorhabditis Genetics Center (http://www.cbs.umn.edu/CGC/). The asp-3(tm4559) and asp-1 (tm666) (mutagen: TMP/UV) was obtained from Tokyo Women’s Medical College (http://www.wormbase.org/db/gene/variation?name=tm666;class=Variation). C. elegans was cultured using standard techniques, and the Bristol N2 strain was the wild type [59]. RnyEx109 is an integrated transgene that directs expression of calcium indicator d3cpv under the control of intestine-limited promoter Pnhx-2 [42,43]. PwIs50 is an integrated transgene with the intestinal lysosomal marker LMP-1:: GFP [39,40].
Cry6Aa and Cry5Ba proteins were purified according to our previously described method [60]. All purified protein samples were then solubilized in 20 mM HEPES (Calbiochem BB0364) (pH 8.0), quantified [61], and stored at -80°C. The purified Cry6Aa protein was labeled with N-hydroxysuccinimide-rhodamine (Pierce) or N-hydroxysulfosuccinimide ester-PC-biotin (Pierce), respectively, according to the manufacturer’s instructions.
The growth assay was performed according to the protocol described previously reported [62]. The LC50 assay was performed according to the protocol reported by Marroquin et al [22]. All lifespan measurements were performed on NGM agar plates of E. coli express Cry6Aa at 20°C [48]. The lifespan was monitored at 24 h, 48 h, 72 h, 96 h, 120 h, 144 h. Nematodes that did not move when they were gently prodded and displayed no pharyngeal pumping were marked as dead [48].
Preparation of the total protein of C. elegans was conducted according to the protocol described by Schrimpf et al [63]. 2-DE was performed according to the protocol described by Krishnamoorthy et al [25]. Strips of 18 cm (pH 4–7) in length were used. After 2-DE, Gels were either silver stained and scanned using a GS-800 calibrated densitometer (Bio-Rad), or transferred to polyvinylidene difluoride Q (PVDF) membrane filters (Immobilon P, Millipore) for ligand blotting.
Ligand blotting was performed according to the protocol reported by Fernandez-Luna et al [26]. Total C. elegans proteins were separated by 2-DE, and gels were transferred to a PVDF membrane in transfer buffer (20% methanol, 25 mM Tris-base, 192 mM glycine) for 60 min. Filters were blocked overnight in PBST (0.1% Tween 20 in phosphate buffered saline, pH 7.4) containing 3% BSA. Blocked filters were incubated with biotinylated Cry6Aa (10 nM) for 2 h at room temperature, and then washed three times using PBST. The bound protein was detected by 1 μg/ml streptavidin- horseradish peroxidase (HRP) conjugate (Sigma, S5512). The membrane was visualized using enhanced chemiluminescence substrate (SuperSignal West Pico, Pierce) following the manufacturer’s instructions.
C. elegans treated with Cry6Aa were harvested and subsequently washed five times in M9 medium. Treated C. elegans samples were grinded with liquid nitrogen and were then subjected to sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) and transferred onto a PVDF membrane. Primary antibodies were anti-ASP-1 antibody (1:5000 dilution) for detection of ASP-1 expression. The secondary antibody was HRP-coupled anti-rabbit antibody (1:5000 dilution). The membrane was visualized as described above for the ligand blotting assay.
To monitor the size of Cry6Aa that is found inside the nematode, N2 and asp-1(tm666) were fed purified Cry6Aa proteins. Total proteins were then extracted from crystal protein treated nematodes, separated by SDS-PAGE, transferred onto a PVDF membrane. Primary antibodies were anti-Cry6Aa antibody (1:5000 dilution) for detection the size of Cry6Aa. The secondary antibody was HRP-coupled anti-rabbit antibody (1:5000 dilution). The membrane was visualized as described above for the ligand blotting assay.
Different quantities of ASP-1 were dotted onto a nitrocellulose (NC) membrane (Millipore, Bedford, MA, USA). After blocking with 3% bovine serum albumin (BSA) in PBST, NC membrane was bathed in biotinylated-Cry6Aa (10 nM) for 2 h at room temperature, washed three times using PBST. Unlabeled Cry6Aa (1000-fold excess) was used in the competition assays. The bound protein was detected by 1 μg/ml streptavidin-horseradish peroxidase (HRP) conjugate (Sigma, S5512). Finally, the signal was visualized using 3, 3'-diaminobenzidine tetrahydrochloride (DAB) substrate (Pierce) following the manufacturer’s instructions.
Selected spots were excised from the stained 2-DE gel and digested with trypsin, and the resulting peptide fragments were examined by mass spectroscopy according to the method described by Krishnamoorthy et al [25]. Peptides were examined using matrix-assisted laser desorption ionization-time of flight mass spectrometry (MALDI-TOF-MS) system (4700 Proteomics Analyzer, Applied Biosystems). A database search of spectral data was conducted using a mascot search engine (Matrix Science) to search the C. elegans databases.
RNA was extracted from C. elegans using a total RNA isolation system (Promega, Madison, WI, USA). First strand cDNA was synthesized from total RNA according to the manual (Takara, Tokyo, Japan). The ASP-1 full-length coding gene was amplified from synthesized cDNA by polymerase chain reaction (PCR) using a pair of primers based on the sequences reported for the aspartic protease gene (GenBank NM_171587): P1, 5′- CGCGAATTCATGCAGACCTTCGTTTT-3′, EcoRI restriction site is underlined; P2, 5′-CTGCTCGAGCTTACAATCCCTTGTGG-3′, XhoI restriction site is underlined. The amplified fragments were purified, digested with EcoRI and XhoI, and cloned into vector pGEX-6P-1 to create a recombinant vector pASP1-GST. Then, plasmid pASP1-GST was transformed into E. coli strain BL21 (DE3) (Amersham Biosciences, Uppsala, Sweden), and positive transformants were selected on an LB plate containing 100 μg/ml ampicillin. The ASP-1 was overexpressed in E. coli as a GST-tagged fusion protein ASP1-GST. ASP1-GST was purified according to the GST bind kit protocol.
pPD49.26 was used as the vector backbone for test constructs. The rescuing fragment comprising the entire 2715 bp asp-1 gene (GenBank AF210248) contains 1421 bp 5′-flanking DNA of asp-1, 1191 bp asp-1 cDNA, and 103 bp 3′-UTR of asp-1. The 2.7 kp rescuing fragment was made by PCR amplification with Pfu polymerase using a pair of primers designed based on the sequences reported for the aspartic protease gene: P3, 5′- CGCGAATTC CCAAAATGGGTCTTACC-3′, EcoRI restriction site is underlined; P4, 5′-CTGCTCGAG ATCAGAAATTAAAGATT-3′, XhoI restriction site is underlined. The amplified fragment was purified, digested with EcoRI and XhoI, and cloned into vector pPD49.26 to create a recombinant vector pASP1RES. This construct was co-injected at 10 ng/ml with the dominant rol-6 marker (pRF4) at 50 ng/ml into the gonads of nematode asp-1 (tm666) by standard techniques [64]. The eggs of roller hermaphrodites were allowed to hatch to obtain L1 larvae, and then the L1 larvae were treated with Cry6Aa. The nematode resistance to Cry6Aa was evaluated based on the ability of L1 larvae to progress to adulthood during the course of the experiment [65].
The purified Cry6Aa was biotinylated using N-hydroxysulfosuccinimide ester-PC-biotin (Pierce, Rockford, IL) according to the manufacturer’s instructions. ELISA plates (high-binding, 96-well, Immulon 2HB; Thermo Fisher Scientific Inc., Waltham, MA) were incubated at 4°C for 12 h with 0.5 μg of ASP-1/well in 20 mM HEPES buffer (pH 8.0). The plates were then blocked at room temperature for 2 h in 100 μL PBST containing 3% BSA. For the binding assays, ELISA plates coated with ASP-1 were incubated with increasing concentrations of biotinylated Cry6Aa. For the competition assays, a 1000-fold molar excess of nonlabeled Cry6Aa was added to a solution that contained biotinylated Cry6Aa. Specific binding was determined by subtracting nonspecific binding (with 1000-fold molar excess of unlabeled Cry6Aa) from total binding (without excess unlabeled Cry6Aa). The other reaction conditions and the data analysis were conducted following the method described by Zhang et al [66]. Data were analyzed using SigmaPlot 12.0 software.
Isothermal titration calorimetry experiments were performed at 25°C using the TAM Thermal Activity Monitor system (Thermometric AB, Sweden). Recombinant proteins ASP-1 were buffer-exchanged into 20 mM HEPES (pH 8.0) by G-25 spin-column chromatography. A solution containing 21.16 μM Cry6Aa in the same buffer was used as titrant, and solutions containing 2.20 μM ASP-1 were used in the calorimetry cell. The heat of reaction per injection was determined by integration of the peak areas using the Origin 8.6 software. ITC titration experiments were carried out with 20 injections, 10 μL per injection, and 75 s between each injection. Cry6Aa was titrated into 20 mM HEPES buffer to account for heat released due to dilution. Data were analyzed in Origin 8.6 software after subtracting the heat released from titrating Cry6Aa alone into buffer. The dissociation constants (Kd) were calculated from the plots of the total heat versus the molar ratio of Cry6Aa to ASP-1 [67].
The experiments were based on the methodology of Fragoso et al. [68] with some modifications. Approximately 500 mg of N2 or asp-1(tm666) were triturated in 500 μL of acidic buffer (0.1 M sodium acetate and 0.5% v/v Triton-100; pH 4.8) at 4°C, centrifuged at 12,000g for 20 min, and the supernatant was used as crude protein extract. For proteolytic assays, 50 μL Cry6Aa was incubated with 50 μL crude protein extracts from wild type N2 or asp-1(tm666) for 2 h at 37°C. For protease protection assays, Cry6Aa was preincubated with ASP-1 or GST (control), and then treated with crude protein extracts from asp-1(tm666). For the toxicity of the digested Cry6Aa, the sample buffer was replaced by HEPES buffer (pH 8.0), and the nematodes N2 were exposed to digested Cry6Aa for lifespan analysis.
Nematodes were imaged on a 2% agarose pad prepared on glass slides, and images were acquired using an Olympus BX63 microscope. Death fluorescence was observed through a DAPI filter (λex/λem 345 nm/455 nm). The fluorescence densities of nematodes were quantified using computerized image analysis with Olympus cellSens imaging software.
In vitro changes in the concentrations of cytoplasmic calcium ([Ca2+]i) can be assessed by measuring the cytoplasmic fluorescence with the calcium indicator Fluo-4 AM (Molecular Probes) [36]. Nematodes were placed on a 2% agarose pad prepared on a glass slide, and then imaged under at 400 magnification on an Olympus BX63 microscope. The excitation and emission wavelengths were 489 nm and 508 nm, respectively. The relative fluorescence (F1/F0) was calculated using computerized image analysis with Olympus cellSens imaging software, where F1 was the fluorescence of C. elegans exposed to 63 μg/mL Cry6Aa protein for 6 days or heat stroke, and F0 was the fluorescence of C. elegans without Cry6Aa protein or heat stroke [36].
In vivo calcium levels were visualized using the calcium indicator d3cpv expressed from the intestine-limited promoter Pnhx-2 in transgenic (rnyEx109) nematodes [42,43]. Nematodes were imaged under Olympus FV1000 inverted confocal IX81 microscope. CFP (405 excitation, 480 emission), and FRET (405 excitation, 535 emission) filters and the FV10-ASW software were used to collect the FRET data. The FRET ratio was calculated by (FRETint−FRETbkgnd)/(CFPint−CFPbkgnd), where FRETint and CFPint represent the fluorescent intensities of the FRET and CFP channels of nematode gut, and FRETbkgnd and CFPbkgnd are the fluorescent intensities of FRET and CFP in the background region [41,42].
In vitro lysosomal rupture can be assessed by labeling with lysotracker [43] (Life Technologies, USA). The excitation and emission wavelengths were 555 nm and 580 nm, respectively. In vivo, a C. elegans transgenic (pwIs50) strain expressing the intestinal lysosomal marker LMP-1::GFP [39,40] was used to examine lysosome integrity. Nematodes were placed on a 2% agarose pad prepared on a glass slide, and then imaged under 400 magnification on an Olympus BX63 microscope. The excitation and emission wavelengths were 489 nm and 508 nm, respectively.
Nematode heat stroke, hypoxia and propidium iodide staining assays were performed according to the protocol reported by Nikos Kourtis et al [35]. Pepstatin A (an aspartyl proteases inhibitor) and Z-Val-Phe-CHO (a calpain inhibitor) were used as necrosis inhibitors [44] in this study. After propidium iodide (Sigma) staining, worms were visualized using an Olympus BX63 microscope. The excitation and emission wavelengths were 555 nm and 580 nm, respectively.
The results presented in each figure are the average of three independent experiments. LC50 values were calculated using PROBIT analysis [69]. The three independent LC50 were averaged and showed as mean ± SD. The significance of the differences between two datasets was assessed by Student’s t test.
|
10.1371/journal.pgen.0030218 | A RNA Interference Screen Identifies the Protein Phosphatase 2A Subunit PR55γ as a Stress-Sensitive Inhibitor of c-SRC | Protein Phosphatase type 2A (PP2A) represents a family of holoenzyme complexes with diverse biological activities. Specific holoenzyme complexes are thought to be deregulated during oncogenic transformation and oncogene-induced signaling. Since most studies on the role of this phosphatase family have relied on the use of generic PP2A inhibitors, the contribution of individual PP2A holoenzyme complexes in PP2A-controlled signaling pathways is largely unclear. To gain insight into this, we have constructed a set of shRNA vectors targeting the individual PP2A regulatory subunits for suppression by RNA interference. Here, we identify PR55γ and PR55δ as inhibitors of c-Jun NH2-terminal kinase (JNK) activation by UV irradiation. We show that PR55γ binds c-SRC and modulates the phosphorylation of serine 12 of c-SRC, a residue we demonstrate to be required for JNK activation by c-SRC. We also find that the physical interaction between PR55γ and c-SRC is sensitive to UV irradiation. Our data reveal a novel mechanism of c-SRC regulation whereby in response to stress c-SRC activity is regulated, at least in part, through loss of the interaction with its inhibitor, PR55γ.
| Protein Phosphatase type 2A (PP2A) represent a family of holoenzyme complexes involved in wide range of activities such as growth, differentiation, and cell death. The PP2A holoenzyme complex is made up of a catalytic, a structural, and one of various “B” subunits. These “B” subunits are thought to provide the substrate specificity required for PP2A activity. Previous work on PP2A has mostly been derived by inhibiting the catalytic subunit through chemical inhibition, as such inhibiting all of the pathways associated with PP2A. To identify individual “B” subunits involved in specific cellular processes we have generated a “B” subunit gene knockdown library, which allows us to inhibit each of the known “B” subunits individually. One of the many pathways regulated by PP2A is the c-Jun NH2-terminal kinase (JNK) kinase pathway, which, depending on stimulus, can affect either cell survival or cell proliferation. Here we report that the “B” subunit PR55γ acts as a negative regulator of JNK activity and cell death. We show that PR55γ influences JNK activity by inhibiting one of its upstream regulators, the proto-oncogene c-SRC, through dephosphorylation at one of the key residues on c-SRC, a site we show to be critical for c-SRC activation following cell stress. Overall our work describes the novel function of a specific PP2A subunit involved in cell survival and identifies a novel mechanism of c-SRC regulation.
| The Src family of nonreceptor tyrosine kinases are integral players in the mediation of various physiological processes such as cell motility, adhesion, proliferation, and survival [1]. Members of the Src family share a conserved structure consisting of four Src homology (SH) domains, a unique region, and a short negative regulatory tail. The amino terminal SH4 domain is myristoylated and targets the protein to the membrane, while the carboxy-terminal SH1 domain functions as a tyrosine kinase domain [2]. c-SRC activation is negatively regulated by Carboxy Src Kinase (CSK) or its homologue CHK through Tyrosine 527 (Tyr527) phosphorylation [2]. This inhibitory phosphorylation promotes the assembly of the SH2, SH3, and kinase domains into a closed conformation [2]. Following stimulation by various stresses and growth factors c-SRC activation is initiated by dephosphorylation of the Tyr527 residue by the protein-tyrosine phosphatase PTPα [3] and PTP1B [4]. Alternatively, c-SRC is activated by the binding of tyrosine-phosphorylated proteins to the SH2 domain, resulting in destabilization of the intermolecular interaction between Tyr527 and the SH2 domain [2]. Subsequently, c-SRC is autophosphorylated at Tyrosine 416 (Tyr416), a site within a segment of the kinase domain termed the activation loop, promoting a conformational change that allows the kinase to adopt an open active confirmation [2].
c-SRC is overexpressed or activated in a wide variety of tumors [5,6]. However, overexpression of c-SRC by itself has only minor oncogenic potential [7] and mutations in c-SRC in cancer have only been found sporadically [8]. This led to the hypothesis that c-SRC has a supportive function in tumorigenesis rather than a role in the actual transformation process [9]. Overexpression of v-Src, a constitutively active form of c-SRC lacking the c-terminal part containing the inhibitory Tyr527, is a potent activator of c-Jun NH2-terminal kinase (JNK), a growth-regulatory enzyme that can control cell proliferation and cell survival both positively and negatively, depending on the stimulus and the cellular context [10,11]. Furthermore, SRC activity is essential for JNK activation following a number of different stress stimuli, including UV irradiation [12–14].
Protein Phosphatase 2A (PP2A) is a serine/threonine phosphatase that can influence the phosphorylation state of many signaling enzymes [15,16], and inhibition of this phosphatase can affect cellular responses such as growth, differentiation, and apoptosis [15,17]. The holoenzyme generally exists as a core dimer, consisting of a 36-kDa catalytic subunit (PP2Ac) and a 65-kDa scaffold subunit (PR65) that associates with a variety of regulatory subunits. These regulatory B subunits can modulate the activity of the PP2Ac/PR65 core unit, thus allowing specific temporal targeting of a wide range of PP2A substrates. To date 15 genes coding for more than 26 (B) regulatory subunits have been identified that are subdivided into five different subfamilies [17].
The variable PP2A B subunits are targeted by a number of viral oncogenes, which thereby compete for interaction with the PR65/PP2Ac core dimer. This suggests that specific PP2A holoenzymes play a role in viral propagation and oncogenic transformation [18,19], which is further supported by the finding that general inhibitors of PP2A can cause tumor growth on the skin and liver of rodents [20–23]. Understanding the precise manner in which PP2A is involved in the regulation of these different signaling cascades and its role during oncogenic transformation requires the identification of the specific holoenzymes involved in these processes. Interpretation of a large amount of data using general PP2A inhibitors has been limited by the pleiotropic inhibition of all PP2A holoenzyme complexes by the inhibitors used. Furthermore, ectopic expression of the various B subunits can lead to competition with other subunits for binding to the holoenzyme, making it difficult to draw firm conclusions from the data [24,25]. Using a gene family knockdown library targeting all deubiquitinating (DUB) enzymes, we previously identified the familial tumor suppressor gene CYLD as a novel regulator of the NF-κB signaling pathway [26] and USP1 as the deubiquitinating enzyme of the FANCD2 DNA repair protein [27]. To study the role of the various PP2A complexes in specific pathways we have constructed a library of 61 independent vectors expressing short hairpin RNAs (shRNA) targeting the PP2A regulatory B subunits for suppression. Using this knockdown library in a screen for enhancers of JNK activation following cellular stress, we identified a number of PP2A B subunits as novel regulators of JNK activation, most notably PR55γ and PR55δ. Furthermore we demonstrate that the PP2A B subunit PR55γ negative regulates the JNK effector pathway by acting as a stress sensitive inhibitor of c-SRC activity.
To identify the specific PP2A holoenzyme complexes involved in pathways known to be modulated by PP2A, we constructed a gene family knockdown library targeting all putative human PP2A regulatory B subunits for suppression. We retrieved the cDNA sequences for each of the PP2A subunit family members from the ENSEMBL database and designed two to four unique 19-mer sequences for each transcript for cloning into pSuper and pRetro-Super [26,28]. In total 61 knockdown vectors were generated, which were then subsequently pooled into 16 sets of two to four vectors per transcript with each set targeting one of the regulatory B subunits or a specific transcript variant (Figure 1A; Table S1). To validate the pooled knockdown vectors, we tested six randomly chosen pools of vectors for their ability to effectively knockdown the target proteins. All pools tested show a notable reduction in target protein expression levels (Figure 1B).
Studies using viral proteins that target the regulatory B subunits of the PP2A holoenzyme complex indicate that JNK and the proto-oncogene c-Jun can be regulated by PP2A [29]. This suggests that specific PP2A regulatory B subunits are involved in PP2A-mediated regulation of the JNK pathway. To directly assess the putative role of PP2A in JNK regulation we asked if suppression of one or more PP2A regulatory subunits by RNA interference could affect JNK activity following UV irradiation. U2-OS cells were transfected with the different library pools and then assayed by western blotting for the efficiency of UV induced JNK activation as judged by threonine-183/tyrosine-185 phosphorylation. Unsurprisingly, we found that suppression of a number of the B subunits appeared to enhance the levels of phosphorylated JNK following UV. Of these, PR55γ consistently yielded the strongest effect and was chosen for further validation (Figure 1C and unpublished data). To evaluate which of the four individual knockdown vectors in this pool were active against PR55γ, we transfected cells with HA-tagged PR55γ and determined the protein levels of HA-PR55γ in lysates of transfected cells in the presence or absence of the individual PR55γ knockdown vectors. As depicted in Figure 2A, all four shRNA vectors (A–D) in this pool were able to suppress HA-PR55γ expression levels, whereas no effect was detected on a cotransfected green fluorescent protein (GFP) (Figure 2A). A shRNA targeting the mouse-specific B subunit PR59 was used as a negative control in all experiments. Vectors A and C were more efficient in suppressing HA-PR55γ protein levels than vectors B and D (Figure 2A). A fifth knockdown vector (E) was designed, which like vector C, induced strong suppression of ectopic PR55γ expression (Figure 2A). shRNAs C and E will be referred from here on as shRNA#1 and shRNA#2, respectively.
To test whether these shRNAs #1 and #2 could inhibit endogenous PR55γ levels we performed quantitative real-time PCR (QRT-PCR). We found that both shRNAs efficiently suppressed endogenous PR55γ mRNA levels (Figure 2B). Furthermore, inhibition of PR55γ with both validated knockdown vectors could efficiently enhance the activation of JNK by UV irradiation (Figure 2C), arguing against an off-target effect of the shRNAs. This result underscores the validity of the screen and suggests that endogenous PR55γ is a repressor of stress-induced JNK activation.
To determine whether the activation of JNK after transfection of PR55γ knockdown vectors was a consequence of the loss of PR55γ expression, we performed an add-back experiment. To do this we restored PR55γ levels to the control situation using a PR55γ construct (ΔPR55γ) containing two noncoding mutations within the region targeted by knockdown vector #2, rendering it refractory to shRNA-mediated suppression (Figure 2D). We found that expression of ΔPR55γ completely abolished the enhanced activation of UV-induced JNK observed with shRNA vector #2, but not with shRNA vector #1, which targets a region that was not mutated in ΔPR55γ (Figure 2E). These results argue that the effects of the knockdown vectors targeting PR55γ for shRNA-mediated suppression on JNK activation are the result of loss of PR55γ.
To investigate whether the enhanced JNK activation upon PR55γ knockdown is specific for UV irradiation, we asked whether other stimuli that lead to the activation of the JNK pathway might also be enhanced by loss of PR55γ. We found that TNFα, insulin, and osmotic stress-mediated JNK activation could all be enhanced by suppression of PR55γ but not EGF-mediated JNK activation (Figure 2F). These results suggest that PR55γ is a regulator of the JNK signaling pathway when activated by diverse stimuli.
It has previously been established that activation of JNK by UV irradiation can enhance apoptosis in cell culture [30]. Since knockdown of PR55γ leads to enhanced JNK activation, we asked whether knockdown of PR55γ could enhance apoptosis following UV irradiation. UV-induced apoptosis was indeed significantly enhanced in PR55γ-depleted cells (Figure 3A) as determined by measuring the mitochondrial membrane potential with a fluorescent dye (3,3′-dihexyloxa-carbocyanine iodide, [DiOC6 (3)]). Figure 3B represents three independent DiOC6 experiments demonstrating the percentage of apoptosis with or without UV in presence of knockdown vectors targeting PR55γ or a control vector. We also observed an increase in caspase 3 cleavage, a primary executioner of apoptosis, in lysates of cells exposed to UV irradiation, when PR55γ was suppressed (Figure 3C). Similar results were obtained with a second shRNA targeting PR55γ (unpublished data).
To investigate whether PR55γ regulates the JNK pathway upstream of JNK we asked if loss of PR55γ affected MKK4, the kinase acting directly upstream JNK [31]. We indeed found also that MKK4 activity was significantly enhanced in cells with depleted PR55γ (Figure 4A). These data suggest that PR55γ does not directly affect JNK phosphorylation levels. We therefore asked whether the suppression of PR55γ had an effect on the other MAPK pathways, p38 and ERK. Indeed, western blot analyses indicated that knockdown of PR55γ resulted in increased phosphorylation of both JNK and p38, but not of ERK following UV irradiation (Figure 4B). Thus indicating that PR55γ acts on a key regulatory protein required for activation of both JNK and p38. Of note, no direct interaction was found between PR55γ and components of the MAPK and JNK kinase pathways including the previously described PP2A interacting proteins JNK, MKK4, p38, or c-RAF as determined by coimmunoprecipitation assays (unpublished data) [32–35].
One of the major contributors to the activation of the JNK pathway is the nonreceptor tyrosine kinase c-SRC [12–14]. It was previously shown that the polyoma middle t (MT) antigen, which binds to c-SRC and has been suggested to compete with the PP2A regulatory B subunit for binding to the holoenzyme complex [36–38], is also a potent activator of the JNK signaling cascade [38]. It has recently been described that of the ubiquitously expressed SRC family members only c-SRC [39] and LYN [40] play decisive roles in UV-induced JNK activation. Consistent with this, only c-SRC and Lyn have putative PKC sites in the N-terminal region. However, LYN appears to exclusively regulate JNK kinase but not p38 or ERK. Since knockdown of PR55γ in our system regulates not only JNK but also the MAPK p38 (Figure 4B), it would suggest that c-SRC may be the critical target of PR55γ in negatively regulating the JNK pathway in U2-OS cells following stress.
To test whether the enhanced activation of JNK after suppression of PR55γ is dependent on c-SRC, we cotransfected a dominant negative version of c-SRC, which has a lysine 295 to methionine mutation, resulting in a kinase deficient c-SRC (Src295M) [41]. UV irradiation-induced JNK phosphorylation was attenuated in the presence of Src295M, in agreement with the earlier finding showing that JNK is activated by both c-SRC independent and c-SRC dependent pathways [39]. However, the enhancing effect of PR55γ knockdown was completely abolished upon coexpression of Src295M (Figure 4C). Likewise inhibition of c-SRC by the generic Src family inhibitor PP2 also inhibited the enhanced JNK activity caused by suppression of PR55γ (Figure 4D). Similarly, cotransfection of a hairpin targeting PR55γ with an shRNA targeting c-SRC completely abolished the enhancing effects of PR55γ on JNK activity (Figure 4E).
To further investigate whether PR55γ can influence the levels of phosphorylated JNK by a non c-SRC family kinase–associated stimulus, we cotransfected a constitutively active form of the GTPaseCdc42 (Cdc42V12), which functions upstream of MKK4 in the JNK pathway, in the presence or absence of PR55γ. As expected, transfection of Cdc42V12 resulted in activation of JNK [42]. However, the knockdown of PR55γ had no significant affect on JNK activation, whereas it did enhance phosphorylation of JNK following exposure to UV, which served as a control (Figure 4F). Since suppression of a number of the B subunits appeared to enhance the levels of phosphorylated JNK following UV (Figure 1C), we wanted to determine whether the increased levels of phosphorylated JNK observed with knockdown of the other B subunits are dependent on c-SRC. As expected knockdown of PR55δ enhanced the activity of JNK following UV irradiation. Furthermore, like PR55γ, the increased JNK activity was completely attenuated upon cotransfection with kinase dead Src295M (Figure 4G). Together these results suggest that PR55γ and PR55δ negatively regulate JNK signaling in a c-SRC-dependent manner. Since PR55γ is primarily expressed in neuronal tissues and PR55δ is more ubiquitously expressed, it may be that PR55γ and PR55δ mediate the same biochemical responses to stress in different tissues.
Several studies have suggested a role for PP2A in the regulation of c-SRC [43–45]. For instance, both polyoma MT antigen and adenovirus E4orf4 were previously shown to interact with both c-SRC and PR55α independently, but the relevance of these interactions remained elusive [46–48]. To further address the functional relationship between PR55 and c-SRC, we asked whether PR55γ could physically interact with c-SRC. To investigate this we performed coimmunoprecipitation experiments. We found that immunoprecipitation of c-SRC from lysates of cotransfected cells resulted in coprecipitation of PR55γ (Figure 5A). We also detected this interaction reciprocally by immunoprecipitating GFP-tagged PR55γ with a GFP antibody and then probing the blotted precipitate with a c-SRC antibody (Figure 5B). Importantly, endogenous c-SRC also coimmunoprecipitated with GFP-PR55γ (Figure 5B). Together, these data suggest that PR55γ and c-SRC can form a complex in vivo.
Since PR55γ binds to c-SRC we asked if PR55γ could form physical complexes with other SRC family members. We cotransfected FLAG-PR55γ with either the c-SRC family kinases LYN or FYN. We found that LYN and FYN do not share the ability of c-SRC to interact with PR55γ in coimmunoprecipitation assays (Figure 5C and 5D)
To test if the PR55γ/c-SRC interaction was specific for the B subunit PR55γ we cotransfected c-SRC with PR55γ or the PP2A B'' subunit PR72. As shown in Figure 5E, c-SRC physically associated with PR55γ but failed to coimmunoprecipitate with PR72.
Moreover, the specific binding of c-SRC to the B subunit PR55γ suggests that PR55γ is able to recruit the holoenzyme complex to c-SRC. We therefore asked if PR55γ could mediate binding of the PR65/PP2Ac core dimer to c-SRC. We transfected HEK293 cells with constructs expressing FLAG-SRC, HA-PR65, and HA-PP2Ac in the presence or absence of GFP- PR55γ and performed coimmunoprecipitation assays for FLAG-SRC. We found that c-SRC formed a complex with the PP2A holoenzyme exclusively in the presence of PR55γ, indicating that PR55γ is required as bridging factor between c-SRC and the PR65/PP2Ac core dimer (Figure 5F). These observations demonstrate that PR55γ specifically interacts with c-SRC and mediates the recruitment of the PR65/PP2Ac core dimer to c-SRC.
The physical interaction between PR55γ and c-SRC suggests a role as a modulator of c-SRC activity. Since c-SRC activity is increased following UV irradiation, we asked whether UV irradiation could affect the interaction between PR55γ and c-SRC. We followed the interaction between PR55γ and c-SRC after UV irradiation by performing immunoprecipitation experiments. We found that the interaction between PR55γ and c-SRC was gradually lost over time (Figure 5G) demonstrating that the interaction between c-SRC and PR55γ is sensitive to UV irradiation.
Since PR55γ appears to regulate JNK activation at the level of c-SRC, we examined the role of PR55γ on c-SRC- activated transcription of a JNK responsive luciferase reporter. We found that suppression of PR55γ enhanced the ability of c-SRC to activate this reporter (Figure 6A). Conversely, overexpression of PR55γ represses the ability of c-SRC to activate this reporter (Figure 6B). Consistent with these results, western blot analyses demonstrate that overexpression of c-SRC causes an increase in JNK phosphorylation after UV (Figure 6C). Moreover, when we cotransfected short hairpins targeting PR55γ in the presence of c-SRC, we observed that suppression PR55γ enhanced the levels of phosphorylated JNK compared to c-SRC alone (Figure 6D). Consistent with this, ectopic expression of PR55γ inhibited the synergistic activation of JNK mediated by c-SRC and UV (Figure 6E). These results demonstrate that PR55γ is able to influence c-SRC-mediated signaling to the JNK pathway.
To assess whether PR55γ directly modulates c-SRC kinase activity, we evaluated c-SRC Tyr416 phosphorylation, a hallmark of its activity [2], by western blot analyses. We found that knockdown of PR55γ could further enhance c-SRC Tyr416 phosphorylation following stimulation by UV irradiation (Figure 6F). In contrast overexpression of PR55γ could reduce c-SRC Tyr416 phosphorylation following stimulation by UV irradiation (Figure 6G). Together, these results suggest that PR55γ is an inhibitor of c-SRC activity.
Several studies have indicated that pretreatment with the PP2A inhibitor okadaic acid (O.A.) induces the phosphorylation of the PKC phosphorylation site, Ser12 on c-SRC, while simultaneously stimulating c-SRC kinase activity [45,49]. To further investigate if PR55γ alters the phosphorylation status of SRCSer12, we stimulated U2-OS cells with UV irradiation in the presence of [32P] orthophosphate and performed a 2D tryptic phospho-peptide analysis of phosphorylated c-SRC. Indeed, comparisons with tryptic phosphopeptide maps indicate that overexpression of PR55γ decreased the levels of the phosphorylated Ser12 peptide while okadaic acid slightly increased the phosphorylation levels of the Ser12 peptide, when compared to control samples (Figure 7A–7E). Of note peptide maps showed similar patterns to those performed by Moyers et al. [47]. Similar results were observed when the cells were treated with phorbal 12-myristate 13-acetate (PMA) and Forskolin, potent activators of PKC and PKA, respectively (unpublished data). From these data we conclude that treatment with either UV or PMA induces the phosphorylation of the PKC site Ser12 on c-SRC and that this specific phosphorylation event is significantly diminished in cells overexpressing PR55γ. Of note, no direct interaction was observed between PR55γ and PKCδ, the kinase that directly phosphorylates Ser12 of c-SRC, as determined by coimmunoprecipitation assays (unpublished data).
Since phosphorylation of a specific target sequence has been suggested to be one of the requirements for the targeting of the B regulatory subunit and the subsequent recruitment of the PP2A holoenzyme to the substrate, we wanted to address whether recruitment of PR55γ to c-SRC is dependent upon the phosphorylation status of residue Ser12. To answer this question we generated a gain-of-function (SRCS12D) and a loss-of-function (SRCS12A) mutant of this phospho-site. To test if phosphorylation of Ser12 was required for the binding of PR55γ we cotransfected HA-PR55γ with either: FLAG-SRC, FLAG- SRCS12D, or FLAG- SRCS12A in the presence or absence of UV irradiation and performed coimmunoprecipitation assays. As shown in Figure 7F, mutation of serine 12 to an alanine decreased the association with PR55γ compared to wild-type c-SRC, while the association of PR55γ with SRCS12D significantly increased. However, the association of PR55γ with either wild-type c-SRC or the mutant forms of SRCSer12 were completely abolished following UV irradiation. Together these results suggest that phosphorylation of Ser12 is one of the factors determining PR55γ affinity towards c-SRC and demonstrates that the interaction between PR55γ and c-SRC is sensitive to UV irradiation regardless of the presence of a phospho-moiety on Ser12.
To determine whether the effects of PR55γ on SRCSer12 phosphorylation were the cause of PR55γ-mediated c-SRC regulation, we examined the role of SRCSer12 on a JNK-responsive luciferase reporter. We found that overexpression of the SRCS12A mutant significantly inhibited c-SRC's ability to activate a JNK-responsive luciferase promoter (Figure 7G). Conversely, overexpression of SRCS12D enhanced c-SRC's ability to activate the JNK-responsive luciferase promoter (Figure 7G).
Since SRCS12D and SRCS12A significantly enhanced and diminished c-SRC's ability to activate the JNK-responsive luciferase promoter respectively, we wanted to determine whether phosphorylation of this site affects c-SRC kinase activity. Kinase activity was assayed by monitoring the levels of phosphorylated enolase as an exogenous substrate. As shown above (Figure 6F and 6G), exposure to UV irradiation increases the activation of c-SRC compared to nonstimulated cells (Figure 7H). However, the c-SRC kinase activity was severely crippled in cells expressing SRCS12A. Furthermore, c-Src kinase activity was significantly enhanced in SRCS12D cells compared to controls in unstimulated cells (Figure 7H). Similar results were detected when measuring the autophosphorylation of c-SRC at Tyr416 (Figure 7I). Taken together these results demonstrate that phosphorylation of Ser12 on c-Src is one of the requirements for full activation of the protein following stress.
To determine if the observed effect on JNK activity by PR55γ following UV stimulation is dependent on Ser12, we cotransfected hairpins targeting PR55γ with wild-type c-SRC or the SRCS12A mutant and measured the levels of phosphorylated JNK by western blotting following UV irradiation. As expected knockdown of PR55γ intensified the effect on JNK phosphorylation compared to c-SRC alone. However, the enhancing effect of PR55γ knockdown on JNK phosphorylation was completely attenuated upon coexpression on the SRCS12A mutant (Figure 8A). In agreement with this result, cotransfection of SRCS12D interfered with PR55γ's ability to inhibit JNK phosphorylation following exposure to UV (Figure 8B). Our data collectively demonstrate that modulation of SRCS12 phosphorylation by PR55γ is critical for PR55γ's effects on JNK activation.
We next asked whether SRCS12A could abrogate the enhanced apoptosis observed with knockdown of PR55γ. We cotransfected hairpins targeting PR55γ with either SRCS12A or wild-type c-SRC and quantified apoptosis using DiOC6 staining following exposure to UV irradiation. In line with results above (Figure 3A), knockdown of PR55γ increased UV-induced apoptosis in the presence of wild-type c-SRC. However, coexpression of SRCS12A completely curtailed the enhancing effect of PR55γ suppression (Figure 8C). In contrast, coexpression of PR55γ and c-SRC repressed UV-induced apoptosis compared to c-SRC alone, while these effects were completely abolished when PR55γ was coexpressed with SRCS12D (Figure 8D). Taken together, these results demonstrate that the regulation of c-SRC by PR55γ and its subsequent effects on cell survival are mediated through regulation of Ser12 phosphorylation on c-SRC.
It has previously been proposed that the stress response to environmental stimuli mediated by the JNK pathway is regulated by PP2A protein phosphatase activity [33,50,51], although the way in which this pathway is regulated and the specific PP2A holoenzyme responsible for this regulation have not been identified. Via an RNAi-mediated gene family–knockdown screen of regulatory PP2A B subunits, we now identify PR55γ as a specific negative regulator of stress-induced JNK activation. We find that PR55γ regulates the JNK pathway through negative regulation of c-SRC kinase activity. Importantly, we identify here c-SRC serine 12 as a critical residue for the regulation of the c-SRC kinase activity during stress signaling. We show that PR55γ physically interacts with c-SRC and has a higher affinity for c-SRC when it is phosphorylated on serine 12. Since the interaction of the PP2A holoenzyme complex with c-SRC is dependent on PR55γ, this would suggest a transient interaction between PR55γ-containing PP2A holoenzyme and c-SRC, which is reduced as soon as serine 12 dephosphorylation has occurred.
Previous work has indicated that PP2A might play a role in the regulation of c-SRC activity, since treatment of cells with okadaic acid, a chemical inhibitor of PP2A [22], resulted in enhanced c-SRC activity [45], and PP2A can inactivate c-SRC in vitro [43]. Interestingly, polyoma MT is able to compete with the PP2A B regulatory subunit for interaction with the PR65/PP2Ac core dimer [38], and overexpression of polyoma MT is able to activate c-Jun kinase by virtue of its interaction with PP2A [29]. Furthermore, polyoma MT was also found to interact with c-SRC [36] leading to its activation [37]. Moreover, it was reported that adenovirus E4orf4 can interact with both c-SRC and PR55α independently and that the interaction with c-SRC is required for E4orf4 to induce apoptosis [46,52]. Overexpression of E4orf4 phenocopies loss of PR55 in yeast [53], allowing the possibility that inhibition of PR55α is a prerequisite for E4orf4-induced apoptosis in mammalian cells. Our data identifying PR55γ as a negative regulator of c-SRC are in agreement with these studies and could suggest that these viral proteins may function to displace PR55γ from c-SRC.
It has previously been reported that JNK is activated by both c-SRC independent and c-SRC dependent pathways [39]. This present study confirms and extends these results by demonstrating that the inhibition of SRC by PR55γ does not completely inhibit JNK activation but rather results in an overall decrease, similar to the effects observed with a kinase dead mutant of c-SRC. In contrast, knockdown of PR55γ increases SRC kinase activity following UV resulting in enhanced levels of phosphorylated JNK. These results suggest that modulation of one of the upstream activator pathways may result in a prolonged and amplified JNK effect.
It has previously been shown that the majority of c-SRC is present in the perinuclear region where it was found to be inactive as judged by Tyr 416 phosphorylation [54]. c-SRC was also found in the cytoplasm at lower levels correlating with increasing activity and moved to the membrane in response to various stimuli where it was fully active [54]. Similarly we found PR55γ to be primarily expressed in the perinuclear region indicating that PR55γ may colocalise with c-SRC (unpublished data). These data suggest that PR55γ may interact with c-SRC within the perinuclear region inhibiting the induction of c-SRC by PKC by limiting the phosphorylation status of Ser12. Since PR55γ did not decrease the overall levels of other phosphorylation sites within the unique region of c-SRC primarily SER17, phosphorylation of which has been shown to be involved in SRC dependent ERK signaling [55], it suggests that the selective response of c-SRC following PKC phosphorylation at Ser12 may reflect the restricted activation of the JNK downstream effector pathway through either, phosphorylation dependent changes in subcellular localization, as suggested by Liebenhoff et al. who demonstrated that cytoskeletal association of pp60c-src is dependent on phosphorylation of pp60c-src at Ser12 by PKC [56] or by regulation of the binding of proteins that may function to regulate the activity of c-SRC towards JNK.
One of the intriguing findings of this study is that upon treatment of cells with UV irradiation the interaction between c-SRC and PR55γ is lost. We propose a model in which we suggest that in response to stress c-SRC is activated in part by losing the interaction with its inhibitor allowing c-SRC to be localized to the plasma membrane and subsequent activation of the downstream JNK effector pathways (Figure 9). Similarly, it was described for PR55α that its interaction with the PP2Ac/PR65 dimer is sensitive to gamma irradiation [57]. Further work will be required to reveal the mechanism of UV-induced dissociation of the c-SRC/PR55γ in response to stress.
pcDNA3- FLAG-SRC, pEGFP-SRC, pMT-SRC, pMT-SRC(527), pMT-SRC(295), pcDNA-LYN, and pcDNA-FYN (Table S2) were kindly provided by P. Stork, G. Superti-Furga, W. Molenaar, and J. Borst. All other Flag-, GFP-, and HA-coding constructs were generated using pcDNA (Invitrogen). Detailed cloning information will be provided upon request. PP2A knockdown library vectors were generated by annealing the individual oligonucleotide primer pairs and cloning them into pSuper as described in [58].The bacterial colonies of each B subunit were then pooled and used for plasmid preparation. The extra shRNA (E) that gave the most efficient knockdown against PR55γ as described in Figure 2A was generated by ligating synthetic oligos (Sigma) against the target sequence 5′-CATGGAGGCAAGACCCATAG-3′ into pSuper. The c-SRC knockdown sequence was obtained from Gonzalez et al. and cloned into pSuper [59]. Antibodies anti-p-JNK, anti-p-MKK-4, anti-p-Src(416), and cleaved caspase-3 were from Cell Signaling; anti-SRC, anti-JNK (C-17), anti-MKK4, HA (Y11), anti-GFP, and anti-FYN were purchased from Santa Cruz Biotechnology Inc. The anti-LYN antibody was a kind gift from J. Borst.
All cells were cultured in Dulbecco's modified Eagle medium (DMEM) supplemented with 10% fetal calf serum, D-Glutamate, and Penicillin/Streptomycin. U2-OS cells were divided in 10-cm dishes 1 d prior to transfection. Subconfluent cells were transfected using the calcium phosphate transfection method [60]. Cells were incubated overnight, washed in PBS, and puromycin selected (1.5 μg/ml) for 48 h. When required cells were serum starved for 48 h prior to stimulation. The cells were not allowed to reach confluency. For the screen and subsequent knockdown experiments, U2-OS cells were cotransfected with 20 μg of pooled PP2A shRNAs and 1 μg of pBabe-puro. After 72 h, selected cells were trypsinized and 5 × 105 cells were plated out in a 10-cm dish. After incubating overnight cells were exposed to UV irradiation (100 J/m2) and incubated for a further 60 min in the same medium. The following agents were used to stimulate cells: 50 ng/ml EGF (Upstate), 10 ng/ml TNF (Sigma), 10 ng/ml Insulin (Sigma), 500 mM NACL, or UVC (254 nm, 100 J/m2). Luciferase assays were performed using the Dual luciferase system (Promega). AP1 luciferase vector (300 ng) was transfected in the presence of CMV-c-SRC (0.5 μg) or a control vector and CMV-Renilla (0.25 μg). For loss of function, 2.5 μg of pSuper vector [58] was cotransfected, and luciferase counts were measured 72 h post-transfection using a TD-20/20 Luminocounter (Promega). For gain-of-function assays, 0.5 μg of CMV construct or control vector (empty CMV) was cotransfected, and luciferase counts were measured 48 h post-transfection.
For detection of apoptotic cells, selected cells were incubated for 72 h, trypsinized, and incubated for another 10–12 h in new media. The cells were washed twice in PBS and incubated for 18–24 h following UV treatment (50 J/m2), trypsinized, washed once with PBS, and resuspended for 10–15 minutes in 250 μl PBS containing 40 nM DiOC6 (3). After incubation the cells were analyzed by FACS analysis.
Cells were lysed in solubilizing buffer (50 mM Tris [pH 8.0], 150 mM NaCl, 1 % NP-40, 0.5% deoxycholic acid, 0.1% SDS, 1 mM Sodium Vanadate, 1 mM pyrophosphate, 50 mM sodium fluoride, 100 mM β-glycerol phosphate), supplemented with protease inhibitors (Complete; Roche). Whole cell extracts were then separated on 7%–12% SDS-Page gels and transferred to polyvinylidene difluoride membranes (Millipore). Membranes were blocked with bovine serum albumin and probed with specific antibodies. Blots were then incubated with an HRP-linked second antibody and resolved with chemiluminescence (Pierce). For coimmunoprecipitations, cells were lysed in ELB (0.25 M NaCl, 0.1%NP-40, 50 mM HEPES [pH 7.3]) supplemented with protease inhibitors. Lysates were then incubated for 2 h with 2 μg of the indicated antibodies conjugated to protein A or protein G sepharose beads, washed three times in ELB buffer, and separated on SDS-PAGE gels. When appropriate cell lystates were immunoprecipitated with ANTI-FLAG M2 Affinity Gel (Sigma).
For tryptic phosphopeptide analysis U2-OS cells were cotransfected with 4 μg Flag-Src or 4μg or Flag-Src (12A) and 20 μg PR55γ or control vector. Cells were phospho-starved for 45 min and 2 mCi of [32 P] orthophosphate was then added to the cells and incubated an additional 3 h. PMA at a final concentration of 200 nM was added for 30 min at 37 °C or the cells were treated with UV irradiation (100 J/m2) and incubated for a further 30 min at 37 °C. c-SRC was immunoprecipitated with Flag antibody (Sigma) as described above. The entire sample was loaded onto an SDS-PAGE gel, run, and then dried. The film was then exposed for 3 h at room temperature. The radioactive bands were then isolated, proteins eluted, digested with trypsin, and phosphopeptide mapping was performed as described previously [61,62]. For phospholabeling analysis HEK 293 cells were cotransfected with 4 μg Flag-PR55γ, 10 μg HA-PR65, and 10 μg HA-PP2Ac. Cells were phospho-starved for 45 min, UV stimulated as above, and then 2 mCi of [32 P] orthophosphate was added to the cells and incubated for a further 2 h in the same medium. c-SRC was immunoprecipitated with Flag antibody (Sigma) and an HA antibody (Y11, Santa Cruz) as described above. The entire sample was loaded onto an SDS-PAGE gel, run, and then dried. |
10.1371/journal.pcbi.1002958 | A Discriminative Approach for Unsupervised Clustering of DNA Sequence Motifs | Algorithmic comparison of DNA sequence motifs is a problem in bioinformatics that has received increased attention during the last years. Its main applications concern characterization of potentially novel motifs and clustering of a motif collection in order to remove redundancy. Despite growing interest in motif clustering, the question which motif clusters to aim at has so far not been systematically addressed. Here we analyzed motif similarities in a comprehensive set of vertebrate transcription factor classes. For this we developed enhanced similarity scores by inclusion of the information coverage (IC) criterion, which evaluates the fraction of information an alignment covers in aligned motifs. A network-based method enabled us to identify motif clusters with high correspondence to DNA-binding domain phylogenies and prior experimental findings. Based on this analysis we derived a set of motif families representing distinct binding specificities. These motif families were used to train a classifier which was further integrated into a novel algorithm for unsupervised motif clustering. Application of the new algorithm demonstrated its superiority to previously published methods and its ability to reproduce entrained motif families. As a result, our work proposes a probabilistic approach to decide whether two motifs represent common or distinct binding specificities.
| Transcription factors play a central role in the regulation of gene expression. Their interaction with specific elements in the DNA mediates dynamic changes in transcriptional activity. Databases store a growing number of known DNA sequence patterns, also denoted as DNA sequence motifs that are recognized by transcription factors. Such databases can be searched to find a match for a newly discovered pattern and that way identify the potential binding factor. It is also of interest to cluster motifs in order to examine which transcription factors have similar binding properties and, thus, may promiscuously bind to each other's sites, or how many distinct specificities have been described. To gain deeper insight into the similarities between DNA sequence motifs, we analyzed a comprehensive set of known motifs. For this purpose we devised a network-based approach that enabled us to identify clusters of related motifs that largely coincided with grouping of related TFs on the basis of protein similarity. On the basis of these results, we were able to predict whether two motifs belong to the same subgroup and constructed a novel, fully-automated method for motif clustering, which enables users to assess the similarity of a newly found motif with all known motifs in the collection.
| An important goal of biological research is to understand the mechanisms that control gene expression. Of key interest are transcription factors (TFs) that bind to specific functional elements in the DNA and from there regulate expression of target genes. Binding site sequences recognized by individual TFs often exhibit distinct patterns of more or less stringent nucleotide preferences at different positions, also denoted as DNA sequence motifs. There are commercial and public databases like Transfac® (public or commercial) [1] and Jaspar (public) [2] that maintain libraries of DNA sequence motifs in the form of Position-specific Frequency Matrices (PFMs). The PFM is a 4×L matrix whose columns describe nucleotide preferences at corresponding binding site positions by their absolute or relative frequencies.
In recent years there has been increased interest in methods to quantitatively compare DNA sequence motifs. There are two eminent applications for such methods in the current literature. One is to search a library of known motifs with a newly discovered pattern to check its novelty or to derive hypotheses about TF families that could be assigned to the search pattern. This database search application is of increasing importance for the widely adopted ChIP-seq and ChIP-chip assays that enable computational extraction of DNA sequence motifs from large sets of genomic regions bound by a transcription factor of interest [3], [4]. In the second application, quantitative comparison forms the basis to define groups or families of motifs. The growing body of known binding motifs for different transcription factors has stimulated interest to assign patterns to groups representing distinct specificities. While DNA sequence motifs in databases are typically defined for a narrow selection of proteins such as a group of isoforms, a subfamily or a complex, motif families may widen the scope to represent the DNA-binding properties, e.g., of a whole class of transcription factors.
A number of methods have been developed for motif comparison. Kielbasa et al. [5] proposed a combination of Chi2 distance and correlation coefficients of Position-specific Weight Matrix (PWM) scores to group highly similar binding specificities. Mahony et al. [6] compared global and local alignment algorithms as well as column-wise similarity metrics with respect to their ability to recognize motifs belonging to the same transcription factor class and developed methods to cluster PFMs into representative Familial Binding Profiles (FBPs) [7]. By now, many tools are available for motif comparison and clustering such as MatCompare [8], STAMP [6], [9], T-Reg Comparator [10], MATLIGN [11], Tomtom [12], Mosta [13], or KFV [14].
A large group of methods compares motifs on the basis of column-wise scores that scale the similarity or dissimilarity of aligned motif positions. Column-wise scores that have been described for DNA sequence motif analysis include Chi2 statistics [5], Kullback-Leibler divergence [10], Pearson correlation [6], Fisher-Irwin test P-values [8], absolute, squared or Euclidean distances [15], [7], [12], generalized log-odds scores [16], [17], Bayesian methods [18], or fuzzy integral techniques [19]. One advantage of column-wise scoring is its straightforward application within standard local or global alignment algorithms, e.g. [6]. Other methods assess motif similarity on the basis of how binding sites are predicted by corresponding PWMs. Similar to the score correlation approach described in [5], the Mosta algorithm analyzes the tendency of binding sites to overlap when they are predicted with two PWMs at a certain score threshold and for a certain background distribution of nucleotides [13]. Finally, the alignment-free KFV method evaluates the similarity of fixed-length k-mer vectors to which motifs are converted [14]. In this work we present the information coverage (IC) criterion as a further enhancement of column-wise scoring. The IC evaluates the fraction of information of compared motifs that is covered by an alignment. Alignments between related and unrelated motifs exhibit different IC distributions. Combination of the IC with existing motif alignment scores improved their motif classification performance.
Despite the great interest in classification and clustering of DNA sequence motifs, little progress has been made to define families of motifs that methods aim to identify. Validation of motif clustering results mainly addressed their homogeneity with respect to structural classes of TFs, such as ETS, homeobox or nuclear receptor proteins. On the other hand, inference of clusters relied on ad-hoc cut-offs to prevent potential false merges of PFMs into common groups or cut hierarchical clustering trees at an optimal height that balanced inter- and intracluster variability, see e.g. [6], [13], [14]. Neither of these strategies used information about known motif families to define such thresholds.
In this work we therefore undertook a first step to compile a comprehensive collection of motif families that can be used as a goal set for motif clustering methods. We denote as motif family a (sub)set of motifs from the same TF class with a common, distinct binding specificity. Methods developed in this work aimed at identifying clusters of motifs that correspond to such motif families and to propose a representative FBP. Our analyses used a set of 1001 Transfac matrices that were assigned to 35 motif classes mainly corresponding to distinct classes of DNA-binding protein domains [20]–[22]. To subgroup them into motif families, we next devised a network analysis-approach. This procedure constructed networks of Transfac matrices that revealed families of similar motifs as modules of highly connected nodes. Computational graph-cluster analysis confirmed our manual observations based on network visualizations. Furthermore, we examined the concordance between extracted motif clusters and phylogenies of corresponding DNA-binding domains as well as experimental knowledge regarding specificities of certain types of transcription factors. According to this assessment, the motif clusters matched protein domain families as well as prior expectations about DNA-binding properties of some well-described transcription factors. A set of motif families assembled on the basis of network analysis results was then applied to train a probabilistic classifier. The classifier was designed to assign a probability to the hypothesis that two PFMs belong to the same motif family given their similarity score and offers a natural decision threshold. We integrated the new classification function into a novel algorithm for unsupervised motif clustering and demonstrate its ability to extract meaningful motif clusters that are represented by Familial Binding Profiles.
Our workflow for the general goal of clustering DNA sequence motifs depicted in this article can be summarized as follows. We first describe novel information coverage-scores and their validation. We then illustrate the use of the best score for further analysis of motif networks and extraction of motif families. Finally, we report on the development and validation of a new probabilistic classifier that enabled us to conduct motif clustering in an unsupervised fashion and accurately reproduced the entrained motif families.
Our motif alignment program m2match [17] was designed to search for pairwise ungapped local alignments between PFMs. The algorithm selects an optimal alignment according to the score which is the sum of individual column-column scores (column-wise scoring). For this study we developed new composite scores that integrate an alignment feature denoted as information coverage. Information coverage refers to the fraction of information of the motifs that is covered by their alignment. The information of a DNA sequence motif is determined by probability distributions over nucleotides in each of its positions. Figure 1A shows alignments with different information coverage. The alignment of basic helix-loop-helix (BHLH) matrices for transcription factors E47 and MyoD (Fig. 1A top) reaches out over most of the informative positions, whereas the (local) alignment of the E47 motif with a PFM for the MADS transcription factor RSRF (Fig. 1A bottom) omitted several informative positions (gray logo positions). In our study set from the Transfac database alignments between matrices of the same class (intra-class alignments) exhibited a pronounced peak at high IC values which is absent in the IC distribution obtained from inter-class alignments (Fig. 1B).
We subsequently derived new scores that take into account the information coverage of alignments. The new scores extend previously described Euclidean distance (ED) [12] and sum-of-squared distance (SSD) [7] metrics by information coverage terms and are straightforward to compute. Specific variants implemented in m2match are denoted as ED.ave, ED.sqr, SSD.ave, and SSD.sqr (Material and Methods). We carried out a comparison of existing and new methods with respect to two different performance statistics as well as two different libraries of PFMs, Transfac and Jaspar [1], [2].
Figure 2 shows best hit and class-depth statistics achieved by different methods for the 12 largest Transfac classes with at least 20 PFMs. Overall, integration of IC indeed improved ED as well as SSD scores, with ave and sqr variants showing similar performance. Differences were rather small according to the best hit assessment. The ability to recognize other class members increased most strongly with regard to the class-depth statistic where differences up to 5% were recorded for the median values (see also Table 1 below). In few cases, e.g. in the homeobox (HOX) or MADS classes, the ED score was slightly better than ED.ave and ED.sqr scores according to class-depth. However, the improvements visibly outweigh minor performance decreases. Best hit statistics for SSD.ave and SSD.sqr scores were similar or slightly worse than for the SSD score, whereas consideration of IC again improved class-depth statistics in most classes. Some score methods excelled on some classes, but at the same time exhibited difficulties with other classes. For instance, Mosta did not perform as well as other methods on the STAT class according to best hits, and on the HOX class according to class-depth, but the method was ahead on the FORKHEAD class according to class-depth. In contrast, we observe that results of the IC-extended ED and SSD scores were consistently at a high level without bearing remarkable weaknesses for particular TF classes.
Table 1 summarizes our results on the Transfac data set for different sets of motif classes. The values show that inclusion of the information coverage led to an overall improvement of ED scores, especially according to class-depth. Based on the summary values, results were similar for SSD scores, but inclusion of IC did not accomplish as strong improvements as for ED scores. Average values over the six leftmost columns confirm that ED.ave and ED.sqr scores achieved the best overall performance among all of the compared methods.
The strongest methods of the previous comparison were selected to further compete on the Jaspar CORE database. Here we calculated best hit and class-depth statistics for the five largest Jaspar families as well as the Jaspar families with at least 10 motifs, including the zinc finger family (see Material and Methods). Results are summarized in Table 2. As for the Transfac data set, integration of information coverage improved motif classification by ED and SSD scores and the extended scores were competitive to the other state-of-the-art methods. Notably, the advantage of SSD.ave and SSD.sqr scores over the SSD score is more pronounced on the Jaspar data set than on the Transfac collection. On the set of families with at least 10 motifs, the ED.sqr achieved a 6% better performance than the ED score with respect to class-depth. Again ED.sqr and ED.ave scores attained highest average values over best hit and class-depth criteria (Table 2), which is in concordance with the Transfac results. We therefore carried out further analysis of motif relationships using m2match with the ED.sqr score.
Network analysis was applied to further split motif classes into clusters of closely related binding specificities. We compiled networks connecting each motif with other class members that achieved a higher score than non-class members. Finally, we applied the Markov Clustering Algorithm (MCL) [23] to each motif network containing at least 5 motifs. This network-based approach was motivated by our class-depth analysis. The class-depth statistic assumed distinct, motif class-specific levels across methods that participated in the comparison (Fig. 2). For instance, class-depth values were below 20% in the two largest classes, HOX and C2H2 zinc fingers (ZFC2H2), whereas most methods achieved a class-depth over 50% for the classes ETS, FORKHEAD, and E2F. However, the four smallest classes STAT, MADS, REL, and HMG were associated with lower values (Fig. 2B), which rules out that class-depth levels depended on motif class sizes. We conjectured that these class-specific levels originated from the existence of motif families that formed subgroups of highly similar matrices within classes.
Network analysis predicted in total 125 and 135 clusters (including disconnected singletons) when using ED.sqr or ED scores, respectively (Table S1). No connections between matrices were obtained in the ATHOOK class (Table S2). In ten TF classes comprising 6 to 60 matrices all PFMs were drawn together into a single cluster, not taking into account disconnected motifs (Table S2). These classes encompass well-characterized TF classes such as basic helix-span-helix (BHSH), ETS, FORKHEAD, or GATA zinc fingers (ZFGATA). Networks of another ten classes were each split into two clusters by MCL (Table S3). Finally, between 2 and 12 clusters were identified in the classes REL, basic leucine-zipper (BZIP), BHLH, nuclear receptor zinc fingers (ZFC4-NR), ZFC2H2, and HOX. Thus, C2H2 zinc finger and homeobox classes exhibited an outstanding number of different binding specificities, whereas other TF classes comprised much fewer different motif types (1–3 without singletons, 1–6 with singletons).
We compared motif network clusters to phylogenies of DNA-binding domains for the classes BHLH, BZIP, HMG, MADS, REL, SMAD, STAT and ZFC4-NR. A detailed discussion of several of these classes is provided in the supplement (Text S1). Overall, the extracted motif clusters were closely correlated with subtypes of DNA-binding domains. Strongest departures between motif clusters and protein domain phylogenies were observed in BZIP and STAT classes and, according to our assessment, induced by different types of spacers or different numbers of half-sites covered by PFMs (Text S1).
Motif clusters often correlated with broader protein families or subfamilies such as BHLH-Zip, CREB/ATF, SMAD factors in BHLH, BZIP and SMAD classes, respectively. SREBP matrices in the BHLH class and 3-Ketosteroid receptors of the nuclear receptor class presented exceptions to this trend. In compliance with the dual binding specificities of SREBP [24], network analysis assigned its motifs to two clusters, with one reserved exclusively for the SREBP-specific pattern. In the nuclear receptor class, motif clusters accurately distinguished the half-site specificity of 3-Ketosteroid (NR3C) receptors from other nuclear receptors, whereas the protein phylogeny reflects the standard grouping of Estrogen and Estrogen-related receptors (NR3A and NR3B) with those of the NR3C type [25] (Fig. 3). However, half-sites recognized by NR3A and NR3B proteins resemble the pattern bound by non-NR3 receptors and therefore Estrogen receptor matrices were allocated in one cluster with PFMs of non-NR3 receptors (Fig. 3). The molecular causes of different DNA-binding preferences within the nuclear receptor class have been described in detail by Zilliacus et al. [26].
In summary, the network-based analysis delivered meaningful results for a wide range of transcription factor classes. Also in the large and diverse HOX and ZFC2H2 classes the method proposed groups of motifs dominated by closely related transcription factors. In addition, some cases could be highlighted where computational predictions accurately fit prior experimental knowledge such as for SREBP factors or nuclear receptors.
In the following we used clusters of Transfac matrices derived through motif network analysis to train a classifier for motif families. It was the ultimate goal of our study to predict common motif family membership purely by computational means. The conceived classifier accomplished this on the basis of the motif similarity score without requiring information about TF classes. For this we compiled a list of 47 Transfac matrix sets for 26 motif classes (Table S4). These were used as representatives of motif families for the classifier training. Some minor modifications were made to the raw MCL clusters in order to omit some potential false positive or uncertain cluster members which are described in the supplement. For instance, we discarded the V$NMYC_02 matrix that was falsely assigned to the BHLH-only cluster.
To make alignment scores for PFM pairs of different lengths comparable we estimated the dependence of mean and variance of inter-class scores on the space of possible alignments (Fig. 4A). Raw ED.sqr scores were subsequently adjusted according to the following formula:(1)In (1), x is the raw score, La and Lb are the lengths of matrices a and b, μ() is the conditional mean and σ() is the conditional standard error estimated by non-parametric regression. Next we compared distributions of intra-class scores, intra-family and inter-class scores (Fig. 4B+C). While distributions of intra- and inter-class scores strongly overlapped (Fig. 4B), the intra-family distribution exhibited a smaller overlap with and a different mode than the inter-class distribution (Fig. 4C). To utilize this information in a classification framework, we trained a logistic regression model with positive examples comprising intra-family alignment scores and with inter-class scores as negative examples. We considered inter-class instead of inter-family scores as a careful choice for the negative set. This follows from our results of motif network analysis where we also used inter-class scores to place or omit edges between PFMs. Hence, for inter-class alignment scores we were certain that they belonged to pairs of unrelated motifs. The resulting classifier estimates the probability that two matrices belong to the same motif family (F = 1) given the adjusted score of their alignment:(2)The logistic regression (LR) classifier is both discriminative and probabilistic. The approach moreover provides for a natural threshold of P(F = 1|Sadj)>50% to decide in favor of the hypothesis that compared motifs belong to the same family. Parameter estimates reported by R's glm function [27] for adjusted ED.sqr scores were β1 = 5.294,β0 = −3.3296.
We incorporated the classification function into a novel motif clustering algorithm as part of m2match. The goal of motif clustering is to identify a non-redundant set of Familial Binding Profiles by clustering a given collection of motifs [6], [7]. In the context of our work we can now formulate the more precise objective that the motif clusters inferred by the algorithm shall match defined motif families. Hence, the motif clusters and respective FBPs are predictions about motif family assignments. Our algorithm accomplished this as follows. For a set of TF matrices m2match first calculated a distance matrix for subsequent agglomerative clustering. Off-diagonal entries of the distance matrix were set to complementary motif family probabilities (1-P(F = 1|Sadj)). The distance matrix was applied in hierarchical average-linkage clustering. During the clustering process each cluster was represented by a Familial Binding Profile, where the input set of TF matrices was regarded as initial set of FBPs. At each clustering step, the program examined whether the alignment score of the FBPs representing two clusters satisfied the motif family threshold. Furthermore it was tested that a newly formed FBP detected all original TF matrices above the same cut-off. The motif family threshold was set to the natural classification threshold P(F = 1|Sadj)>50%. A merging was considered valid if it satisfied the described criteria and the new FBP subsequently replaced the two FBPs from which it had been derived. If either of the two criteria was not satisfied a cluster and its representative FBP were marked as invalid and could not further contribute to forming valid clusters. A new FBP was derived from the alignment of two predecessor FBPs based on a weighted average of aligned matrix positions. Empty (unaligned), flanking positions of the alignment were filled with uniform nucleotide distributions and were assigned a weight of 1. The weight of real matrix positions was the square root of the number of underlying binding site sequences. We imposed a maximum of 200 for the number of binding site sequences to be taken into account in order to accommodate ChIP-assay derived matrices, whose underlying binding site alignment may sometimes cover several hundred or over a thousand genomic sites. This maximum was therefore adopted to prevent matrices derived from a very large number of binding sites from overriding the contribution of other matrices to the FBP.
Figure 5 shows the clustering result for the set of 71 non-zinc finger Jaspar motifs that was also used in previous studies, see e.g. [14]. The set was split into 33 FBPs by m2match using the ED.sqr score. This is higher than reported in other studies, where the number of 16 FBPs was obtained in [14]. A striking difference between our method and those of previous studies is that all motif families returned by m2match were homogeneous with respect to the TF class. One reason why other methods achieved fewer clusters is therefore that merges happened between motifs from different TF classes. To our knowledge no other method has before produced a clustering of this particular data set with perfect class homogeneity. Moreover, the FBPs formed from multiple Jaspar PFMs and several of the matrices that m2match left as singletons correspond to separate motif families as identified in the course of our motif network analysis. For instance, MCL extracted separate clusters comprising PBX-, NKX-, and PAX-type motifs, which we observe as singletons in the Jaspar clustering result. In both the Transfac and Jaspar PFM set, MEF2 and SRF motifs formed separate motif families (Table S3). Interestingly, plant MADS matrices of the Jaspar set were assigned to one further FBP (Fig. 5). The Transfac PFMs collection used in this study consisted exclusively of matrices for vertebrate transcription factors and did not contain any plant motifs. For comparison, Mosta merged all Jaspar MADS matrices into a single cluster [13]. The FBP of SOX/SRY-type motifs (Fig. 5) matches the results of our analysis of Transfac HMG motifs. We think that it is plausible to assign the Jaspar HMG-1 matrix to a separate motif family as suggested by m2match, because, unlike HMG box factors of the SOX/SRY-type, the sequence specificity of the HMG-1 factor was shown to be limited to oligomers enriched in certain dinucleotides [28]. The androgen receptor matrix Ar was not clustered with other nuclear receptor motifs, since it was the only member of the NR3C family in this data set. Finally, m2match faithfully grouped all the ETS and REL motifs of the Jaspar set into one FBP for each class, which again correlates perfectly with prior results on the Transfac data set and was not achieved by some other previously published methods (clustering of REL dl_1 reported in [14]). Note that Jaspar REL motifs encompassed only the Rel/NF-kappaB subset, so that allocation of their PFMs into one single FBP agrees with our earlier results.
We also applied our method to the motif classes of the Transfac study set. A summary of the results is provided in Table S1. For comparison, we also included results obtained with ED scores. Rand indexes for the clusterings by network analysis and by m2match show that m2match was able to closely reproduce the network cluster results, achieving a Rand index of about 91% on average (Table S5). Among the classes with lowest agreement between the methods, m2match joined IRF or RUNT matrices into one single FBP for each class, whereas the TBP class was split into two clusters with the ATATA_B matrix as singleton (see also Table S3). Conversely, m2match partitioned the group of non-NR3C motifs within the nuclear receptor class into several smaller clusters, but otherwise assigned all NR3C motifs to a common group with the exception of V$AR_Q6 (Figure S4). Generally, we observe that the hierarchical clustering approach had a tendency to produce more motif clusters than MCL applied to motif networks, especially in the large HOX and ZFC2H2 classes. In the HOX class, m2match perfectly recovered the IRX motif family. Other training motif families were partially restored (Table S6). Furthermore, the method detected FBPs predominated by matrices for certain protein subfamilies which selected for HNF1-, PBX-, or SIX-type motifs, respectively. In the ZFC2H2 class, m2match re-identified all eight motif families. The program assigned one more Helios A matrix (V$HELIOSA_02) to family #40 (Table S4) and predicted new clusters with high protein subfamily-homogeneity that comprised EGR motifs or ZIC and GLI matrices (Table S7), which were part of one large cluster in the MCL/motif network result (Text S3). The ETS and FORKHEAD classes show that the algorithm is able to detect that a motif set consists largely of one single FBP, albeit it did not to join the FOXO1 matrix with the large FORKHEAD cluster (Figure S4). FBPs inferred for BHLH and BZIP classes also closely resembled motif clusters identified during network-based analysis. Matrices for AHR factors were not allocated with other BHLH-Zip motifs but formed a separate FBP, separating the CACGCG-consensus of AHR motifs from the CACGTG-consensus of other E-boxes in the BHLH-Zip group. In addition, the selectivity for a particular factor subfamily suggests that this finding is biologically meaningful. In the BZIP class our method produced new clusters of Maf-type matrices and of VBP/HLF/E4BP4 matrices. Several matrices previously assigned to larger groups were isolated. These comprise unclear or false assignments in clusters derived from motif networks, e.g. V$CEBP_01, V$DBP_Q6, and V$TAXCREB_02, so that we regard their separation from other motifs as an improvement of the previous solution.
This study developed novel solutions for some important problems in motif classification and clustering. First, we presented novel motif similarity scores that make use of the information coverage criterion and showed improved performance in retrieving related motifs of the same class. Then, two new methods for clustering of DNA-sequence motifs were developed, one network-based approach and one based on hierarchical clustering. Both motif clustering methods demonstrated their ability to propose motif clusters that were biologically meaningful as validated with respect to protein domain phylogenies and prior knowledge about distinct binding specificities.
An important aspect of the IC extension is its evaluation of a local alignment as a whole. In the presented formulations it is not restricted to distance metrics used in this work, but can be combined with other alignment scores as well. This development therefore motivates exploration of further possibilities to improve motif alignment scoring apart from improving column-wise scoring metrics.
It was previously noted that some scoring methods can report high scores for aligned PFM positions regardless of their information content [18], [19]. This induces a potential source of false positives, because it is disregarded whether aligned positions confer specificity. Column-wise scores based on Bayesian and fuzzy integral approaches have been developed that did not suffer from that flaw [18], [19]. Also the LSO score has the property of assigning less extreme scores to less informative positions [16]. On the contrary, ED and SSD metrics do not differentiate between PFM columns with respect to their information content. Although the IC criterion was conceived from the perspective of distinguishing between intra- and inter-class alignments, it also addresses the handling of informative and non-informative columns. In contrast to other solutions our treatment of information coverage did not directly reduce the contribution of less or non-informative motif positions to an alignment score, but was designed to favor alignments extending over as much information of compared motifs as possible. It is therefore in our interest to further explore IC as an alternative or additional strategy to attribute more importance to informative motif positions.
Motif network analysis enabled us to compile a set of motif families, which were required as input for subsequent classifier training. This part of our study highlighted the diversity among C2H2 zinc finger and homeobox motifs. We think that further study of the causes of the exceptional positioning of these classes as well as the relative homogeneity with regard to the number of different binding specifities in other classes can elucidate new aspects of the evolution of cellular regulatory systems. Furthermore, inspection of motif clusters and corresponding protein phylogenies showed that distinct binding patterns can appear at different levels of primary sequence divergence. It is of great interest to identify the changes necessary to generate a new binding specificity within a transcription factor class and the results of our study can be explored in that direction. As a computational tool, the network-based analysis of motif clusters was not purely unsupervised, because it used information about class membership. In practice, this is not a significant burden as the classes of PFMs collected in large databases are usually known. As a particular advantage, the devised method did not require any further choice of parameters (MCL was invoked with default parameters).
The motif families derived by network analysis enabled us to develop another novel approach for motif clustering on the basis of the logistic regression model. Important novelties of this method are its discriminative training with positive and negative examples of motif alignment scores and its integration of a probabilistic decision threshold. Specifically, at the natural 50% threshold the devised algorithm was able to produce meaningful motif clusters. Therefore, unlike other methods it can rely on an entrained decision function, e.g. it can be applied to a set consisting of only two PFMs, where estimation of clustering indexes or empirical thresholds may be difficult or error prone. The proposed classifier offers an intuitive, probabilistic quantity to assess the similarity of two motifs and to decide whether they present common or distinct binding specificities. Hence, the obtained results motivate exploration of other machine learning methods to the problem of motif classification and clustering. As another practical advantage, all the motif clustering methods developed in this work automatically determined the number of clusters. Nevertheless, we see room for improvement, particularly with regard to the treatment of spacers (gaps) or different numbers of half-sites. These issues may be addressed by corresponding alignment algorithms as well as alternative IC formulations, possibly in combination with a hierarchical classification of motifs.
Motif clustering predicted between 125 and 197 motif families for vertebrate transcription factors from 35 motif classes. The smaller numbers, 125 or 135, were obtained with motif network analysis. In comparison to each other, motif network analysis revealed PFM clusters on a broader scope, whereas m2match sometimes split these further into narrower subsets. For a manually revised motif classification our results suggest an arrangement on three levels. Below the class level, motif families as defined here represent distinct specificities, e.g. different half-sites. The third level (motif subfamilies) can group more specific arrangements. The treatment of heterogeneous complexes remains to be determined for now. In addition, one could allocate classes into superclasses following the classification for TF proteins [20]. Our study has provided a good foundation of data sets and tools to work towards a honed motif classification. A possible application is the study of similar specificities across transcription factor classes, which can lead to further insights regarding interactions or interference of signaling pathways or other regulatory systems e.g. in host-pathogen interactions. Combined with a classification of transcription factor proteins, a motif classification can also support prioritization of poorly characterized TF subfamilies for experimental investigation of their binding properties. A next goal is to make the computational methods available as freely accessible web tools for applications outlined in the beginning.
This study used a set of 1001 PFMs from the Transfac database [2] version 2011.3, which we classified into 35 classes on the basis of DNA-binding domain annotations and manual revision. The classes, their sizes, as well as the assigned matrices are listed in Table S8. All of the motif classes correspond to distinct types of protein DNA-binding domains with the exception of the GENINI class, which contains initiator motifs. For ETS, IRF, and MYB as well as for FORKHEAD and RFX families (Table S8) we digressed from the protein classification [20] by focusing on the narrower family level instead of the transcription factor class level.
We manually assigned matrices of transcription factors having multiple domains with a DNA-binding property to a single binding class. All matrices of TFs with both HOX and POU domains were added to the HOX class. The HOX/ZFC2H2 motifs V$AREB6_01, V$AREB6_02, V$AREB6_03, V$AREB6_04, V$DELTAEF1_01 were classified as ZFC2H2 motifs, because Ikeda and Kawakami have shown in the respective study that DNA-binding specificities of AREB6 (ZEB1) are mainly determined by the zinc finger domains [29]. Further, we added all PFMs corresponding to TFs with a PAX domain or to TFs with both PAX and HOX domains to the HOX class in order to investigate the similarity between these motifs and other HOX PFMs (see Results). Finally, several Transfac motifs were associated with factors containing both a BZIP and a ZFC2H2 domain, e.g. V$CREBP1CJUN_01, V$CREBP1_Q2, V$CREBP1_01, V$CREBATF_Q6. Since to our knowledge the zinc finger domain in these proteins does not contribute directly to DNA-binding, these matrices were treated together with other BZIP matrices.
A second set of PFMs was obtained from the Jaspar CORE database version 2009. Since the redundant and non-redundant matrix libraries differed in size by only 17 entries, we used the redundant set of 476 motifs for the assessment of motif comparison methods. For motif clustering we compiled the set of 71 non-zinc finger matrices following previous studies, where the matrix Athb-1 was replaced by the one named ATHB-5 (MA0110.1). The data set is listed in Dataset S1. Computational experiments carried out with Jaspar used the Jaspar families and matrix assignments.
Our program m2match implements a local ungapped alignment algorithm for DNA sequence motifs. Here we compare motifs described by the standard Position-specific Frequency Matrix model, a 4×L matrix whose elements are the frequencies or probabilities of individual nucleotides in each of position (column). The algorithm searches for the best (highest scoring) alignment consisting of at least min(5,Lx,Ly) consecutive columns in each of two motifs x and y with lengths Lx and Ly, respectively. The score of the alignment of two PFMs is determined by the sum of aligned column scores. For this study we implemented the column-wise scoring methods listed in Table 3.
In Table 3, px(b) and py(b) are the nucleotide probability distributions in aligned columns of matrices x and y. Probabilities were estimated on the basis of raw PFM entries. We denote as raw PFM the model which is stored in a database and typically requires further transformation in order to be used for binding site prediction or motif comparison. A pseudo-count of 1 was added to counts of nucleotide occurrences. If a raw PFM did not contain nucleotide counts we set the normalized frequencies so that a minimal value >0 and not less than 10−3 was guaranteed. Like Mahony et al. [6] we took care to remove uninformative flanking positions. Here we trimmed positions for which the highest difference between any two nucleotide frequencies was less than .25. The selection of α parameters used by ED and SSD scores is described below.
The Euclidean distance score was previously applied by Gupta et al. in the Tomtom tool [12]. Further, the SSD and Pearson correlation coefficient (PCC) metrics were analyzed in detail by Mahony et al. [6], whereas the SSD metric for motif comparison was introduced in [7]. The Log-sum-of-odds (LSO) score has been successfully applied to comparison of HMMs modeling protein alignments [16]. Its extension by the Kullback-Leibler (KL) divergence has been used to characterize conserved non-coding sequence motifs [17].
We extended ED and SSD alignment scores (sums of individual ED or SSD column scores) with the information coverage. The IC quantifies the proportion of the information contained in both aligned PFMs which is covered by the alignment. The information I(px) of a PFM column was defined in terms of the entropy as:We calculated the information coverage as follows:where sx is the start position of the alignment in matrix x and w is the width of the ungapped alignment. Here, we added indices i and k for corresponding columns within the entire motif. To combine the information coverage with ED and SSD scores we defined the following extended scores:where and denote the sum of ED and SSD scores for the alignment with start points sx and sy in matrix x and y as well as width w, respectively. Both sqr and ave extensions multiply the total alignment by a value in the interval [0,1]. Hence, the IC moves the raw score towards 0 the less information of the motifs is covered by the alignment.
Motif comparison methods were evaluated on both the set of classified Transfac PFMs as well as the Jaspar CORE data set (see above). Following previous studies the best hit statistic provides for a way to assess the ability of a method to identify members of the same TF class for an uncharacterized input motif [6]. A leave-one-out test is performed where each motif is removed and compared to all remaining motifs in the database. One then records which proportion of held-out motifs matched a pattern from the same transcription factor class as best hit. Since our Transfac data set was considerably larger than the one used in [6] and contained many similar motifs, we in addition calculated the class-depth statistic. We have developed this statistic in order to record for each held-out motif which proportion of PFMs from the same class can be detected before the first false positive. Since this approach yields several proportion values for each class, we calculated robust statistics consisting of upper and lower quartiles as well as the median.
Aside from a list of column-wise score methods the comparison included Mosta [13] and KFV [14] as third-party tools. Mosta was invoked with two GC contents of 40% (the program default, here denoted as Mosta.GC.4) and of 50% (denoted as Mosta.GC.5). Column-wise scores were implemented in m2match and encompassed LSO, LSO.KL, ED, SSD, and PCC scores. Both SSD and PCC scores are also available in the STAMP tool [9].
Note that matrices from the family named Other in the Jaspar CORE data set, which gathers potentially unrelated motifs, were considered in determining false positive matches, but PFMs from that family were not used as hold-out set.
We determined α parameters for ED, ED.sqr, ED.ave, as well as SSD, SSD.sqr, and SSD.ave scores that were optimal with respect to best hit and class-depth statistics obtained on the Transfac PFM set. The results are illustrated in Figure 6. The graph for each method shows best hit (red lines) and class-depth (blue lines) statistics over a range of α values. We also considered different subsets of TF classes, which were the 5 largest classes only (dashed lines), classes with at least 20 motifs (solid lines) and classes with at least 10 motifs (dotted lines). According to our assessment, optimal alpha values were 0.5 for ED.ave and ED.sqr scores, 0.55 for the ED score, 0.25 for SSD.ave and SSD.sqr scores, as well as 0.3 for the SSD score (gray dotted lines, Fig. 6). These values were kept for all subsequent analyses.
Motif networks were constructed for each class of the Transfac data set. Further analysis focused on motif classes with at least 5 PFMs. In the networks each motif was connected to all other motifs which were detected with a higher score than the first non-class member. The Markov Cluster algorithm (MCL) [23] was then applied to extract clusters from motif networks. The program was used with default values. All edge weights were equal, so that the algorithm clustered motifs on the basis of the graph topological properties of the motif network. Network visualizations were created with the help of yED [30]. Alignments of transcription factor DNA-binding domains represented by at least one classified motif were compiled in [22]. Phylogenetic trees were calculated using Tree-Puzzle [31].
Network and m2match clusters were compared on the basis of the Rand index [32]. For two clusterings U and V over a set of N items the Rand index RI is defined aswhere #C is the number of item pairs in a common cluster and #S is the number of item pairs in different clusters in both clusterings. RI is a quantity in the [0,1]-interval and equals 1 for perfect agreement between two clusterings of the same set of items.
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.