Introduction

Cytotoxic T-Lymphocytes (CTL) eliminate virus-infected cells by recognizing virus-derived peptides (“epitopes”) presented by Human Leukocyte Antigen (HLA) class I molecules on the infected cell surface. The HLA-restricted CTL response is believed to play a major role in the immune control of HIV-1 infection [1], [2], [3], [4], [5], [6], and it is generally believed that an effective AIDS vaccine will have to elicit cellular as well as humoral (antibody) responses [7], [8], [9], [10], [11]. The genes encoding HLA class I are among the most polymorphic in the human genome [12]: each individual expresses up to six different class I alleles (two at each of the A, B and C loci) out of a pool of over two thousand allelic variants defined to date. Each unique HLA molecule is capable of presenting a broad but finite array of epitopes, defined by HLA allele-specific binding motifs. This extensive genetic diversity serves as a mechanism whereby the human immune system, on both the individual as well as on a population basis, is equipped to recognize a vast array of epitopes from a broad range of pathogens. In addition, this extensive diversity means that, at both the individual as well as the population level, the human immune response exerts a complex array of evolutionary selective pressures driving viral evolution [13], [14] in equally intricate, sometimes even conflicting [15], [16] ways.

One of the major mechanisms whereby HIV evades the cellular immune response is through the selection of HLA-restricted CTL escape mutations that allow the virus to evade immune recognition [17], [18], [19], [20]. Escape mutations may interfere with intracellular epitope processing [21], [22], disrupt peptide-HLA binding [23], [24], or disrupt recognition of the peptide/HLA complex by the T-cell receptor [25], [26].

Despite the extensive genetic diversity of both HIV-1 and HLA, recent studies indicate that viral evolution in response to immune selective pressures follows generally predictable patterns and kinetics [13], [14], [15], [27], [28], [29], [30]. For example, in B*57-expressing individuals, the B*57-associated T242N escape mutation in Gag is selected mere weeks after infection [31], [32], [33], whereas the B*27-associated R264K (Gag) may take years to develop despite strong continuous immune pressure in individuals expressing B*27 [24], [34]. Furthermore, both T242N and R264K are typically accompanied by a well-defined set of compensatory mutations [24], [35], [36]. The development of improved statistical methods [37] combined with the availability of large cohorts for which HIV sequences and HLA data are available has facilitated the systematic identification of HLA-associated CTL escape mutations, both within [15], [28], [29], [38] and across [27] HIV subtypes. Due to the extensive diversity of both HLA and HIV, the identification of mutational escape patterns requires large, well-powered datasets; thus, additional data are needed in order to refine existing escape maps. Indeed, just as the systematic identification of antiretroviral resistance mutations [39] has been of paramount importance to the design and monitoring of HIV therapies [40], the comprehensive elucidation of immune escape pathways will be of relevance to HIV vaccine research.

Here, we identify HLA-associated polymorphisms within the three HIV-1 proteins most commonly considered in cellular-based vaccine design strategies (Gag, Pol and Nef) in a combined analysis of three established cohorts totaling >1500 HIV-infected, antiretroviral-naïve individuals. We organize results into protein-wide escape maps illustrating the sites and pathways of immune-driven viral evolution, and hope that these maps will serve as useful reference material for researchers interested in CTL epitope discovery, host-pathogen co-evolution, and HIV vaccine design.

Materials and Methods

Results

HLA-associated polymorphisms were identified in HIV-1 Gag, Pol and Nef in a multicenter cohort of >1500 chronically-infected, treatment-naïve individuals using published methods featuring a correction for HIV phylogeny, HLA linkage disequilibrium and HIV codon covariation [27]. The false discovery rate [43] was used to account for multiple tests.

At the conservative threshold of q≤0.05, 282 HIV codons commonly mutating under HLA-associated immune pressure in Gag, Pol and Nef were identified. These polymorphisms were observed at 74 (of 206; 36%) Nef codons, 80 (of 500; 16%) Gag codons, and 128 (of 947; 14%) codons in Pol. At a more liberal threshold of q≤0.2, the total number of observed codons harboring HLA-associated polymorphisms increased to 442, which included 113 (55%), 130 (26%), 199 (21%) of codons in Nef, Gag and Pol, respectively. These data confirm the results of previous population-based studies reporting greater density of HLA-associated polymorphisms in Nef than in Gag or Pol [15], [29], [41], [48]. Moreover results underscore the observation that the effects of HLA-associated selection pressures on HIV-1 evolution are extensive and predictable.

All HLA-associated polymorphisms at q≤0.05 were organized into gene-wide “immune escape maps” (Figures 1, 2, 3, 4, 5) indicating their location, HLA restriction, specific amino acids, and their direction of association (“adapted” vs. “non-adapted”) with respect to the current HIV subtype B consensus sequence (http://www.hiv.lanl.gov). Published, optimally-described CTL epitopes [49] containing HLA-associated polymorphisms are also shown. In addition, the maps discriminate between HLA-associated polymorphisms directly attributable to selection pressure by the allele (meaning that they survive correction for HIV covariation), from those that may be better explained indirectly (meaning that their occurrence may be better explained by the presence of an HLA-associated covarying residue, rather than the allele itself). The full list of direct (covariation-corrected) plus indirect (covariation uncorrected) HLA-associated polymorphisms within each viral protein at q≤0.2 is provided in Table S1.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Gag Immune Escape Map.

Escape maps indicate the locations, specific residues and HLA restrictions of HLA-associated polymorphisms. The HIV-1 consensus B amino acid sequence is used as a reference. Alternating black and brown letters in the consensus amino acid sequence distinguish the different proteins in HIV-1 Gag (p17, p24, p2, p7, p1, p6). One hundred amino acids are displayed per line. Shaded vertical bars separate blocks of 10 amino acids. “Adapted” amino acids (those enriched in the presence of the HLA allele) are red. “Non-adapted” amino acids (those depleted in the presence of the HLA allele) are blue. UPPERCASE letters distinguish polymorphisms that survive correction for HIV codon covariation (“direct” associations), while lowercase letters distinguish polymorphisms that do not survive correction for codon covariation (“indirect” associations). Polymorphisms associated with the same HLA allele that occur in proximity to one another are grouped together in yellow boxes. Optimally-defined CTL epitopes containing HLA-associated polymorphisms are indicated above the consensus sequence. Note that the escape map does not list the locations of all published CTL epitopes. This is available at http://www.hiv.lanl.gov/content/immunology. The escape maps show all HLA-associated polymorphisms with q≤0.05. A complete listing of all HLA-associated polymorphisms with q≤0.2 is provided in Table S1.

https://doi.org/10.1371/journal.pone.0006687.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Protease Immune Escape Map.

https://doi.org/10.1371/journal.pone.0006687.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Reverse Transcriptase Immune Escape Map.

https://doi.org/10.1371/journal.pone.0006687.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 4. Integrase Immune Escape Map.

https://doi.org/10.1371/journal.pone.0006687.g004

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 5. Nef Immune Escape Map.

https://doi.org/10.1371/journal.pone.0006687.g005

As described in the methods, we also undertook an HIV codon-covariation analysis that, besides identifying direct HLA-associated polymorphisms, also identified all pairwise amino acid-amino acid (aa-aa) associations within a given HIV protein [27]. The HIV codon covariation analysis can be used to identify linked pathways of immune escape, as well as putative secondary and/or compensatory mutations associated with a primary escape site. The codon covariation analysis identified >7000 intra-protein aa-aa correlations occurring at >4500 codon pairs, illustrating the dense and complex networks of covarying amino acids in HIV (Table S2). Indeed, if one sums up the total number of codons harboring HLA-associated polymorphisms, plus the co-varying sites immediately associated with them, the total proportion of codons in Nef that are either directly or indirectly associated with HLA selection pressures reaches 77%. For Gag and Pol, the corresponding proportions are 55% and 44%, respectively.

The sheer density of the intraprotein codon covariation network renders the task of displaying these data rather challenging, but Carlson et al have developed an elegant tool for data visualization that is freely available [27]. Here, the amino acid sequence of a protein is displayed in a counterclockwise circle starting at the 3 o'clock position (Figures 6 and 7). Any HLA alleles associated with variation at those sites are indicated at the corresponding positions outside the circle, while covarying amino acids are joined together by arcs within the circle. The strength of the association (q-value) is indicated by the color of the arc.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 6. HLA-B*57-associated escape and covariation pathways in HIV-1 Gag.

The 500 amino acids of consensus B Gag are drawn counterclockwise, with the N-terminus of Gag at the 3 o'clock position. All direct (covariation-corrected) and indirect (covariation uncorrected) B*57-associated polymorphisms at q≤0.2 are identified at their respective positions along the circle's circumference, while covarying amino acids (also q≤0.2) are joined together by arcs within the circle. Note that this figure is limited to “one-hop” covarying amino acids only, meaning that only the codons directly associated with variation at a B*57-associated sites are shown. (Our analyses also identify, for example, codons associated with variation at the “one-hop” sites, and so on and so forth, but for simplicity we have limited the figure to the “one-hop” sites only. The strength of the association between two covarying codons (expressed in terms of q-value) is indicated by the color of the arc. The program used to construct these figures is available at http://research.microsoft.com/en-us/um/redmond/projects/MSCompBio/PhyloDViewer/ [27].

https://doi.org/10.1371/journal.pone.0006687.g006

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 7. HLA-B*27-associated escape and covariation pathways in HIV-1 Gag.

https://doi.org/10.1371/journal.pone.0006687.g007

HLA-associated intraprotein codon networks in Gag for HLA-B*57 and HLA-B*27 are shown in Figures 6 and 7, respectively. These two alleles were chosen as examples due to their association with slower HIV disease progression in numerous epidemiologic studies [5], [34], [50]. Similarly, Gag was chosen in light of accumulating evidence that Gag-specific CD8 T-cell responses may contribute substantially to HIV immune containment [51], [52], [53] as well as the observation that B*57 and B*27-associated escape mutations in Gag are associated with measurable costs to viral replication capacity [35], [36], [54], which may be partially rescued by compensatory mutations at secondary sites [35], [36]. All direct (covariation-corrected) and indirect (covariation uncorrected) B*57-associated Gag polymorphisms at q≤0.2 are identified at their respective positions along the circle's circumference: For B*57 (figure 6), this corresponds to codons 146, 147, 163, 173, 242, 248, 315, 340, 435 and 449. Within the circle, all q≤0.2 “one-hop” associations with these codons (meaning, Gag codons identified as covarying with them) are connected via arcs. For example, if the B*57-associated polymorphism at Gag position 242 is considered the “predictor variable” (see Table S2), then the residues positively associated with it (“adapted” associations) and/or negatively associated with it (“nonadapted” associations) are located at codons 146, 147, 215, 228, 230, 241, 243, 248, 256, 310, 340 and 373. If position 242 is considered the “target variable”, then the covarying residues positively and negatively associated with it are located at codons 109, 219, 292, 373, 469 and 473. It is important to note in the case of aa-aa associations, the use of “predictor” and “target” terminology should not be interpreted as suggesting a directional association between these polymorphisms or a specific temporal order of selection; rather, it is more appropriate to simply interpret these as codon-codon pairs. Therefore, if one is interested using Table S2 to look up all codons positively and/or negatively associated with Gag codon 242, one should investigate all “target” codons that appear when 242 is set as the “predictor” variable, and vice versa. The union of these two queries will provide a list of specific codons and residues that are positively and/or negatively associated with variation at codon 242.

Note that our analysis also identifies “two-hop” associations (meaning, codons that positively and/or negatively covary with the “one-hop” sites), however these are not shown on the figure due to the high density of the resulting networks. The full list of intraprotein covarying codons is provided in Table S2.

Discussion

HLA-associated polymorphisms were identified in HIV-1 Gag, Pol and Nef in a combined cohort of >1500 chronically-infected, treatment-naïve individuals from established cohorts in Canada, the USA and Western Australia. These cohorts have previously been independently investigated for HLA-associated polymorphisms; however by merging the data and re-analyzing as a whole, we achieved the highest-powered dataset to date to identify HLA associations in HIV subtype B. Indeed, where previous studies had employed a significance threshold of q≤0.2 when reporting associations, here we have lowered the threshold to q≤0.05, thus focusing on sites with the strongest statistical support for HLA-driven adaptation.

The current immune escape maps incorporate some improvements over previous iterations. Firstly, the maps cover all proteins in Pol (including RNAseH and Integrase), instead of just protease/RT as in previous studies [15]. Secondly, all associations, regardless of proximity to known epitopes, are displayed on a single map so that escape patterns in a protein can be visualized globally. Note that, in the case where an HLA-associated polymorphism does not fall within a known optimally-described epitope, we have not attempted to predict the likeliest epitope as has been done previously. This was done in order to avoid forcing an epitope prediction in the case where the HLA association may be attributable to another mechanism (for example a processing escape mutation occurring distant from a published epitope), and also to avoid favoring a particular epitope prediction algorithm among the many that are available (e.g.: MotifScan http://www.hiv.lanl.gov/content/immunology/motif_scan; Epipred http://atom.research.microsoft.com/bio/epipred.aspx [55]; SYFPEITHI http://www.syfpeithi.de/Scripts/MHCServer.dll/EpitopePrediction.htm, [56], NetCTL 1.2 http://www.cbs.dtu.dk/services/NetCTL/ [57], and various others [58], [59]). That being said, visual inspection of the maps reveals strong evidence for the existence of a number of novel CTL epitopes, particularly in Pol where epitope mapping initiatives may not have been as exhaustive compared to Gag and Nef. Thirdly, the incorporation of a multivariate correction for HLA linkage disequilibrium allows the identification of the HLA allele directly responsible for the association, rather than the manual assignment of the responsible allele using p-values post-hoc as employed in previous studies [15]. Finally, the incorporation of a multivariate correction for codon covariation represents an important step forward [27]. It allows us to discriminate HIV polymorphisms directly attributable to selection pressure by the HLA allele in question, from those who may be better explained indirectly (meaning that their occurrence may be better explained by the presence of an HLA-associated covarying residue, rather than possession of the allele itself). In addition, it allows us to comprehensively identify positively and negatively covarying amino acids across proteins (Table S2), thus providing candidate lists for secondary and/or compensatory mutations associated with known escape sites. Indeed, Carlson et al. [27] demonstrated that the codon covariation analysis accurately re-capitulates known pathways of B*57 and B*27-associated escape in Gag [35], [36], [60], supporting the use of this tool for the identification of secondary escape patterns for additional HLA-associated escape mutations.

Just as the standardized identification of drug resistance mutations [39] has been essential to both basic research as well as the clinical monitoring of HIV-infected individuals, we hope that the identification of immune escape pathways will be equally relevant to HIV immunology/virology research and AIDS vaccine design. Our results confirm the strong influence of immune escape on HIV diversity [13], [14], but more importantly underscore the reproducibility and predictability of immune escape in response to specific HLA pressures [15], [27], [28], [29], [38]. We hope these maps and tables will be useful to those interested in CTL epitope discovery, the effects of escape and compensatory mutations on viral replication and pathogenesis, the design of novel vaccines, as well as the broader question of host-pathogen co-evolution. Finally, we have assigned the name “International HIV Adaptation Collaborative” (IHAC) to describe the current multicenter cohort with the hope that this initiative may be expanded to include additional cohorts worldwide in the future. In particular, the merging of data and cohorts across different HIV-1 subtypes [27] will allow us to further explore similarities and differences in HLA-driven polymorphism patterns across subtypes.

Supporting Information

Acknowledgments

The authors wish to thank Bette Korber and the research team at the Los Alamos National Laboratory for inspiring the original design and layout of the immune escape maps. The authors wish to thank ACTG 5128 and 5142 team members and site staff, HLA and viral sequencing laboratory staff, and patient volunteers in all study sites who contributed to this study.