X Chromosomal Variation Is Associated with Slow Progression to AIDS in HIV-1-Infected Women

Monkey Samples

For the present study, retrospective data were compiled from different SIV infection studies on rhesus macaques of Indian descent, performed at the German Primate Center (“Deutsches Primatenzentrum,” DPZ, see Table 1 and Tables S1–S3 available online). The housing and treatment protocols of the DPZ follow the German Animal Protection Act, which in turn complies with European Union guidelines on the use of non-human primates for biomedical research and have been approved by the responsible public authority. All animals were housed, surveyed, and SIV-infected with standardized or highly similar protocols. The time point and route of SIV infection, the date of death, previous treatments, the dose of infecting virus, and the identity of the infecting virus strain itself were known for all monkeys. Necropsy results, hemograms and antibody titers were also generally available. Cell-associated viral-load data and viral RNA copy number were known for the majority of animals. Most monkeys had been infected with SIVmac251 or SIVmac239 (Tables S1–S3). Given that both strains are of similar pathogenicity,^12,13 confounding of our GWAS by strain identity appears to be unlikely. Furthermore, no significant association was observed between the infecting strain and the progression score that constituted the core phenotypic measure of our study (see below), neither in stage 1 (Fisher's exact p = 0.139) nor in stage 2 (p = 0.325) of the GWAS.

Definition of Progression Score

Some 46% of the monkeys in GWAS stage 1, and 32% of animals in stage 2, were euthanized before they had developed AIDS-related symptoms or were still alive by the end of the study. In addition, ∼40% of the monkeys were immunized with a prospective AIDS vaccine before infection. None of the vaccines tested reduced VL at set point substantially but may have prolonged survival. This implies that time of survival after SIV infection alone would have been an insufficient basis for the envisaged GWAS of disease progression. To overcome this caveat, we scored disease progression on a scale of 1 to 5 points, with adjustments made for immunization effects (Figure 1). Individual scores were based upon survival time, necropsy results, antibody titer, available viral-load data, CD4⁺ T cell counts, and neopterin values. This scoring system reflects our long-term experience in evaluating genetic, immunological, and virological data on the course of SIV infection in rhesus macaques. The scoring system not only divides the bulk of monkeys into fast (score 2), medium (score 3), and slow progressors (score 4) but also allows a more nuanced consideration of the extreme ends of the disease spectrum (ultrafast progression: score 1, long-term nonprogression: score 5).

The diagnosis of AIDS-related disease was based upon clinical, necropsy, and histopathology findings. For monkeys to be diagnosed with AIDS-related disease, they had to suffer from anorexia, incurable diarrhea, or neurological or lung dysfunction, evidenced by ataxia or respiratory sounds. Necropsy and histology findings had to reveal one or more of the following: severe hyperplasia of secondary lymphoid tissue, severe chronic active gastroenteritis, chronic interstitial pneumonia associated with Pneumocystis jirovecii infection, giant cell disease, or encephalopathy. Lymphomas, nonbacterial thrombotic endocarditis, or artheriopathy were regarded as additional AIDS-related complications and were used for confirmation of the primary diagnosis. Notably, CD4⁺ T cell count was not found to be an adequate diagnostic marker of AIDS because many rapid progressors still had physiological CD4⁺ T cell counts by the time of euthanasia.^14,15 For monkeys euthanized without a diagnosis of AIDS-related disease, necropsy and histopathology findings were used in determining whether they were at a pre-AIDS stage. The minimum prerequisite for this diagnosis was the presence of moderate to severe hyperplasia of lymphoid tissue in combination with moderate to severe chronic gastroenteritis. Animals presenting with either AIDS-related disease or at a pre-AIDS stage were scored according to their survival time after infection, taking into account whether or not they had been immunized prior to infection (Figure 1).

Monkeys that were still alive at the end of the study (all at > 100 weeks post infectio [wpi]), or that were euthanized without a diagnosis of AIDS-related disease or pre-AIDS, were scored either 4 or 5, depending upon their CD4⁺ T cell count and viral load (Figure 1). Score 5 (“long-term nonprogression”) was defined as the absence of any signs of immunodeficiency (i.e., a stable CD4⁺ T cell count), together with a cell-associated viral load < 3 infected cells per 10⁷ cells or a plasma viral load < 1000 RNA copies mL⁻¹ at set point. Long-term nonprogressors usually survive SIV infection substantially longer (>300 wpi) than slow progressors (score 4). Indeed, among the SIV-infected animals kept at the DPZ, this leap in survival time coincides perfectly with the differences in viral load and CD4⁺ T cell count defining the progression score, even at an early stage after infection.

Microsatellite Genotyping

In the screening stage (stage 1) of our microsatellite-based GWAS, 136 SIV-infected rhesus macaques were genotyped for a genome-wide set of 332 microsatellites. Human as well as Macaca mulatta-specific microsatellites that had been selected on the basis of published mapping information were used.¹⁶ All microsatellites, their chromosomal location, and the forward and reverse primers used for amplification are listed in Tables S4 and S5. For each of the 14 non-MHC markers selected for verification in 128 additional SIV-infected animals (stage 2 of our GWAS), we included two additional flanking markers located ∼250 kb up- and downstream of the original marker, respectively. These markers were identified through a systematic search for microsatellite sequences of the published Macacca mulatta genomic sequence (build 1.1). Assays were designed with the PRIMER 3 software (see Web Resources). Fragment sizes were measured by separation of PCR products on an ABI 3700 capillary sequencer (Applied Biosystems) and by subsequent data analysis with the Genemarker software (Softgenetics LLC).

Microsatellite-Based Association Analysis in Rhesus Macaques

Hemizygosity for X chromosomal microsatellites was verified in all male monkeys, and apparently heterozygous genotypes were excluded from subsequent analyses. The level of allelic association between individual markers and the progression score was assessed for statistical significance with Fisher's exact test. p values were calculated by Monte-Carlo Markov-Chain simulation. Bonferroni correction for multiple testing was carried out in stage 1 of the GWAS (Table S4 and Figure S6), although the actual selection of markers for verification in stage 2 (5% of the total markers) was based solely upon relative significance and resource availability rather than genome-wide statistical significance. In stage 2, each original marker and its two flanking neighbors were treated as one marker group (Table S5). p values were first corrected according to Holm¹⁷ within each marker group, and the most significant marker and its corrected p value were selected to represent the respective group. In a second step, the corrected p values of the group representative and the p values of three additional MHC markers were further corrected for multiple testing according to Holm.

Assessment of Population Stratification in the Monkey Verification Sample

To quantify the level of population stratification in the verification sample (stage 2), members of which originally came from four different centers, we calculated F_ST according to the definition of Weir and Cockerham¹⁸ by using an implementation in FSTAT (version 2.9.3; see Web Resources).

Data and Quality Control for the SNP-Based Association Analysis of the CHAVI Sample

We reanalyzed genotype data from a recent GWAS on host control determinants in HIV-1 infection, undertaken by the Centre for HIV/AIDS Vaccine Immunology (CHAVI).² VL at set-point data and a progression phenotype, defined by the CHAVI investigators as the time until either treatment initiation or a drop in CD4 cell count below 350 cells μL⁻¹, were available for 303 (93 females, 210 males) of the 486 original samples. For these individuals, we were provided the genotypes of all 13,821 X chromosomal SNPs generated in the CHAVI study.² Correct sex assignment was verified with the proportion of heterozygous genotypes at all markers (average percentage expected in males: <1%, in females: ∼25%). Average pair-wise identity by state (IBS) was used for the assessment of extreme genetic similarity and dissimilarity, pointing at cognate relatives or genetic outliers, respectively. For post hoc quality control after association analysis (see below), we required nominally significant SNPs to have a call-rate ≥ 95% and to be in Hardy-Weinberg equilibrium (p > 0.05, with an exact test).¹⁹ Quality control was carried out with the R software version 2.6.2.²⁰

Assessment of Population Stratification in the CHAVI Sample

To account for possible population stratification in the statistical analyses of the human genotypes, we used a principal component (PC)-based approach as implemented in EIGENSTRAT.²¹ Given that EIGENSTRAT can only handle autosomal genotypes, we assumed each male to be homozygous for each X chromosomal marker and females were doubled to ensure unchanged allele frequencies. A recent survey of genome-wide autosomal genetic variation in Europe revealed considerable variation only in the first five to six PCs.²² Therefore, we considered the first six PCs of the X chromosomal markers as sufficient to adjust our analyses of the CHAVI sample for potential population stratification as well.

Association Analysis in the CHAVI Sample

We normalized progression phenotype values by taking natural logarithms. Possible deviations from normality were tested for statistical significance with a Shapiro-Wilk test. To assess possible population stratification in the CHAVI sample, we subjected the first six PCs from the EIGENSTRAT analysis of the X chromosomal genotype data to a multiple linear regression analysis of the log-progression phenotype. AIC-based backward selection retained only the second and the sixth PC in the model. We then used ANOVA, stratified by sex and including the two significant PCs as covariates, to test for statistically significant associations between individual X chromosomal SNP genotypes and the log-progression phenotype and associations with VL at set point. All analyses were carried out with R version 2.6.2.²⁰ The association between SNP rs5968255 and VL at set point was tested for statistical significance with linear regression modeling,² including age at seroconversion and the two significant EIGENSTRAT PCs as covariates.

Linkage Disequilibrium Analysis

We used HaploView 4.1 separately for males and females to estimate linkage disequilibrium (LD) and to define tag SNPs.^23,24 We used the CEU (Utah residents with ancestry from northern and western Europe) genotype data of the HapMap project, release 23a (phase II, March 2008) to infer which SNPs were in LD with each other at a population level.²⁵

Kaplan-Meier Survival Analysis

In the monkey samples, we performed Kaplan-Meier survival analysis with log-rank statistics to assess the observed genotype-specific differences in survival after SIV infection for their statistical significance. In the CHAVI samples, Kaplan-Meier analyses of the progression phenotype were performed in addition to ANOVA to allow a more detailed illustration of the genotypic effects upon progression phenotype. Kaplan-Meier plots were created with R version 2.6.2 and with Sigmaplot 11 (Systat Software).

Synteny Analysis

Synteny between X chromosomal regions in humans and macaques was verified with the tuple plot program for pair-wise nucleotide sequence comparison with noise suppression.²⁶ The human Xq21.1 region, spanning coordinates 71,657,686 to 86,891,379 in NCBI Build 36.1, was compared to the rhesus X chromosomal region spanning coordinates 71,459,512 to 86,479,549 of Mmul_051212.

Genetic Variations in the MHC Class I Region and on Chromosome X Are Associated with Disease Progression in the Rhesus Macaque Model of AIDS

We performed a microsatellite-based GWAS of host factors that influence AIDS-free survival in SIV-infected rhesus macaques of Indian origin. The markers were derived mainly from the published linkage map of the rhesus macaque.¹⁶ An initial, pedigree-based linkage analysis of the available animals was dismissed after the number of genetically verifiable relationships turned out to be too small to achieve sufficient credibility (data not shown). We therefore reverted to a two-stage association analysis, by using 136 SIV-infected monkeys for screening in stage 1 (Table 1; Tables S1 and S2). Prominent association signals obtained in stage 1 were selected for verification in stage 2, in which an independent group of 128 unrelated SIV-infected animals was analyzed (Table 1; Table S3). In order to ensure consistent phenotyping, we used a scoring system for the progression rate toward AIDS-related disease (Material and Methods; Figure 1). The samples used in stages 1 and 2 were characterized by similar average progression scores and similar average AIDS-free survival times (Table 1).

A total of 332 microsatellite markers were genotyped in stage 1 (call-rate > 90%). The average intermarker genetic distance was ∼9 centiMorgans (cM). With a Monte-Carlo Markov-Chain implementation of Fisher's exact test, 83 markers were found to exhibit a nominally significant allelic association with the progression score (p ≤ 0.05). Four of these associations remained significant after Bonferroni correction (Table 2; Table S4), namely D6S2972 (p ≤ 10⁻⁶), D6S2704 (p = 1.2 × 10⁻⁴), D2S434 (p = 2.4 × 10⁻⁵), and MML9S30 (p ≤ 10⁻⁶). Markers D2S434 and MML9S30 are located on rhesus chromosomes 12 and 9, respectively; D6S2972 and D6S2704 are located in the rhesus MHC class I region on chromosome 4 (Table 2).

For verification of the results of stage 1, the top 5% of markers with the smallest p values (n = 17) were genotyped in a second sample of animals. This set of markers included 3 MHC markers, 13 markers from other autosomal loci, and 1 X chromosomal marker. For each of the non-MHC markers, two additional flanking polymorphisms ∼250 kb apart from the original marker were included in the genotyping. In total, 45 microsatellites were genotyped in stage 2 (Table S5). Of these, only X chromosomal microsatellite DXS986 (p = 3.2 × 10⁻³) and its two neighbors (DXS986+1, p ≤ 10⁻⁶; DXS986−2, p = 6.2 × 10⁻⁵) exhibited a statistically significant association with the progression score after multiple testing correction (Table 2). The two MHC markers D6S2704 and D6S2972 retained nominal significance, whereas the phenotype association of D2S434 and MML9S30 could not be confirmed in stage 2 (Table 2; Tables S4 and S5). The verification of the MHC association was not surprising because MHC class I genes are well-established determinants for disease progression in both rhesus macaques and humans.^1,13,27,28 Although some marginal population substructure became apparent in the verification sample (F_ST = 0.035), this was deemed too small to have inflated the significance of our results. Subgroup analyses of naive and immunized animals revealed that the association between DXS986 and the progression score achieved nominal significance in both subsets of stage 1 (p = 1.1 × 10⁻⁴ in naive animals, p = 0.027 in immunized animals). In stage 2, the two subgroup-specific p values were of borderline significance (0.05 < p < 0.10). These results argue against any notable confounding of the association between DXS986 and the progression score by the immunization state. A sex-specific subgroup analysis was not deemed meaningful because the number of females was too low in the two stages (35 and 18, respectively) to provide sufficient statistical power.

Analysis of X Chromosomal SNPs in HIV-1-Infected Patients

In order to investigate whether our results in macaques would also apply to HIV-infected humans, we obtained phenotypic and X chromosomal SNP genotype data from a recent human GWAS, carried out by the Centre for HIV/AIDS Vaccine Immunology (CHAVI).² In the CHAVI study, the progression phenotype was defined as either the time until treatment initiation or the time until the CD4 cell count was observed or extrapolated to have dropped below 350 cells μL⁻¹, a threshold recommended by many guidelines for the initiation of antiretroviral treatment. In the natural course of HIV-1 infection, this event often signifies the transition from a nonsymptomatic state to the phase of AIDS-defining illnesses and death, if not treated.²⁹ We studied 303 seroconverters (93 females, 210 males) for whom the progression phenotype and information on the VL at set point were available.

After synteny between the region surrounding DXS986 in macaques and Xq21.1 in humans had been confirmed (Figure S1), we performed an association analysis (Table 3; Table S6) by using 128 SNPs from a 15 Mb region centered at DXS986 (NCBI Build 36.1 coordinates 71,657,686 to 86,891,379). The region of interest was flanked at its proximal end by MMLXS5, the nearest nonsignificant marker in stage 1 of the rhesus GWAS (Table S4). After Bonferroni correction, sex-stratified ANOVA without adjustment for population structure revealed five SNPs from the DXS986 region to be significantly associated with the log-progression phenotype in females (data not shown). When the two significant PCs were included as covariates, only the association with rs5968255 remained significant. Notably, no significant association with the log-progression phenotype was observed for males. Multiple linear regression analysis of the log-progression phenotype and the five significant SNPs around DXS986, involving AIC-based backward selection with and without inclusion of the two significant PCs, confirmed rs5968255 as the only significantly associated marker in the DXS986 region after Bonferroni correction (corrected p = 1.3 × 10⁻²; nominal p = 8.8 × 10⁻⁵ in females, nominal p = 0.19 in males; Table S7).

Some of the SNPs analyzed in the CHAVI sample showed considerable allelic association with one another, thereby rendering Bonferroni correction for multiple testing overly conservative. We therefore repeated the initial association analysis of the CHAVI data with the assumption that the number of tag SNPs²⁴ in the region or interest represented a good approximation of the number of independent hypotheses tested. Pair-wise r² SNP tagging was carried out with HaploView 4.1.²³ Employing either 0.8 or 0.5 as tagging thresholds yielded 109 and 92 tag SNPs (out of 128 SNPs) in the DXS986 region, respectively. However, even when these figures were used for relaxed multiple-testing correction, the number of significant results remained virtually unchanged (data not shown).

SNP rs5968255 Is Associated with Slower Progression in Women Only

The allele and genotype frequencies of rs5968255 in the CHAVI sample, which consists of individuals of self-reported European ancestry,² were found to be similar to those in the European American HapMap sample (CEU). In both instances, T was the major allele (CEU frequency: 0.88) and C was the minor allele (CEU frequency: 0.12).²⁵ Data from chimpanzee suggest that T represents the ancestral state of this SNP. When the CHAVI samples were stratified by sex, 20 out of 210 males (9.5%) were found to be hemizygous for the C allele. Of the 93 females, 77 (82.8%) were homozygous TT and the remaining 16 women (17.2%) were heterozygous CT. No homozygous CC females were observed.

The progression phenotype was indicative of a significant delay of CD4 cell loss in CT heterozygous females compared to TT homozygotes (Kaplan-Meier analysis, log rank p = 0.001; Figure 2 and Figure S2). No such effect was seen in C hemizygous males. Furthermore, heterozygous females showed a higher median progression phenotype (8.0 years, IQR: 5.1–20.4 years) than homozygotes (2.2 years, IQR: 1.3–4.4 years). No such trend was apparent among males (hemizygous T median: 2.6 years, IQR: 1.4–3.9 years; C median: 1.8 years, IQR: 1.1–2.7 years). In a linear-regression analysis, SNP rs5968255 explained between 21% (coefficient of determination R²; no adjustment for population stratification) and 25% (inclusion of the two significant EIGENSTRAT PCs) of the variance in female log-progression phenotype, indicating a comparatively good model fit. The association analysis between rs5968255 and the VL at set point also yielded sex-specific results (see Table 4, Table S7, and Figure S3). Here, CT heterozygous females showed a lower average VL than TT homozygous females (p = 0.001), and no significant difference was seen between T and C hemizygous males (p = 0.2). For both findings, an artifact due to a hidden association between rs5968255 and the two previously established genetic markers of both VL and the progression phenotype in the CHAVI sample,² namely rs2395029 (a proxy for HCP5 [MIM 604676] and HLA-B^∗5701 [MIM 142830]) and rs9261174 (ZNRD1 [MIM 607525] and RNF39 [MIM 607524]), could be systematically excluded (Tables S8 and S9). Therefore, rs5968255 represents an X-linked genetic variant that independently impacts upon the progression phenotype exclusively in HIV-1-infected females.

SNP rs5968255 is located in a 500 bp conserved intergenic region (Figure 3), 3.9 Mb distal to DXS986. It resides 64 kb proximal to the ribosomal protein S6 kinase 6 gene (RPS6KA6 alias RSK4 [MIM 300303]) and 138 kb distally to the cylicin 1 gene (CYLC1 [MIM 300768]). The 3′ end of an alternatively spliced CYLC1 cDNA clone (accession number BC038787) is located only 233 bp proximally to rs5968255. Because rs5968255 itself may not be causal for the observed association, we identified those SNPs in the DX6986 region that showed at least moderate LD (r² ≥ 0.4) with rs5968255 in the HapMap CEU sample (Table S10).²⁵ This analysis did not yield any additional association signals that would have pointed toward possible candidates for causation. Using the Genevar database,³⁰ we also failed to obtain any evidence for an association between rs5968255 and the expression level of either CYLC1 or RPS6KA6 (data not shown).

HapMap Population Differences in the Allele Frequencies of SNP rs5968255

The HapMap project has been initiated to catalog common DNA sequence variation in different human population groups. The following populations have been included: YRI, Yoruba from Ibadan, Nigeria; JPT, Japanese from Tokyo; CHB, Han Chinese from Beijing, China; and CEU, CEPH (Utah residents with ancestry from northern and western Europe). Inspection of the respective genotype data revealed that C represents the minor rs5968255 allele among females in the CEU (frequency 0.15) and YRI (0.02) groups (Figure 4), but that it is a major allele in Asian females (CHB: 0.56; JPT: 0.61). Like in the CHAVI sample, no CC homozygotes were present among the CEU and YRI females. In contrast, homozygous and heterozygous female carriers of C were fairly frequent in CHB (CC: 0.35; CT: 0.43) and JPT (CC: 0.41; CT: 0.41). Similarly, C hemizygous males were a rare (0.1) group in CEU, and even rarer in YRI (0.03), but were fairly prevalent among males of Asian origin (CHB: 0.36; JPT: 0.54). Thus, on the basis of the hypothesis that the rs5968255 C allele exerts the same effect upon the progression phenotype in Asians as in Europeans (Figure 2), more HIV-1-infected females in China and Japan than in Europe would be predicted to show delayed CD4 cell decline.