LD patterns of SNPs with HLA alleles

We genotyped HIV⁺ and HIV⁻ subjects in the Wilford Hall Medical Center (WHMC) cohort for the SNP in HCP5 (rs2395029; designated as HCP5-T>G), the 5′-region of the HLA-C gene (rs9264942, designated as HLA-C5′-T>C), as well as for the seven SNPs in or near the ZNRD1 and RNF39 genes (Figure 1A). We had previously genotyped subjects from the WHMC HIV⁺ cohort for HLA class I alleles [38]. We evaluated the pair-wise patterns of LD between the SNPs, as well as between the SNPs and HLA class I alleles in HIV⁺ individuals. The seven SNPs near the ZNRD1 locus that were found to be associated with inter-subject differences in HIV disease course in the GWAS were all found to be in nearly 100% LD with each other (Figure 1B and Figure 1C, bottom). Predictably, consistent with these very high LD patterns, the genotype-phenotype associations reported below did not differ significantly when we analyzed each of the seven SNPs near ZNRD1 separately (data not shown). Hence, the results reported here are for a single SNP near ZNRD1 (rs9261174; third ZNRD1-RNF39 SNP shown in Figure 1B and Figure 1C, bottom), and this polymorphism is designated here as the ZNRD1-T>C SNP.

Download:

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

Figure 1. Polymorphisms analyzed in this study and their LD patterns in HIV⁺ subjects from the WHMC cohort.

(A) Genomic location of the polymorphisms studied on chromosome 6p21. Green ticks/boxes represent the HCP5, RNF39 and ZNRD1 genes and red ticks denote HLA loci. The relative location of the polymorphisms studied is shown by vertical arrows (along with their respective NCBI rs identifiers). The horizontal arrows under the genes point to the 5′→3′ orientation of the gene. The predicted size of the two extended haplotypes containing HLA-A10 (∼86kb) and B*57 (∼157kb) are also shown. (B) Linkage disequilibrium (LD) analysis of the studied alleles. D' value estimates are denoted by the intensity of the blue color shown on the scale on the top. (C) Estimated frequencies of the haplotypes based on possession of the HLA-B*57, HCP5 and HLA-C5′ polymorphisms (top), and separately for HLA-A10 and ZNRD1 (bottom) polymorphisms in HIV⁺ individuals from the WHMC cohort. Top panel shows the haplotype frequencies for HLA-B*57, HCP5 and HLA-C5′ alleles (red and purple colored bars are for EAs and AAs, respectively) while the lower panel is for haplotypes derived based on HLA-A10 and ZNRD1 polymorphisms (red and purple colored bars are for EAs and AAs, respectively). Bars represent the haplotype frequency while the numbers to the right of each bar represents the number of haplotypes found in the WHMC subjects. Numbers in parentheses, total number of haplotypes. As discussed in the text, because of the nearly 100% LD between ZNRD1 and RNF39 SNPs, rs9261129 was used to represent the ZNRD1 SNP in statistical analyses (SNP number 3 in panels B and panel C, bottom).

https://doi.org/10.1371/journal.pone.0003636.g001

The spatial proximity between the HLA-A locus and the ZNRD1 SNP (Figure 1A) prompted us to investigate whether the latter SNP was in LD with specific HLA-A alleles. We considered the results of de Bakker et al [32] who found a high LD between HLA-A*2601 and the SNP designated as rs2301751. The latter SNP is 244 bp from rs2301753, which is one of the seven ZNRD1-RFN39 SNPs that we genotyped. Furthermore, rs2301753 is reported to be in very high LD (D' = 1.0) with rs2301751 by the HapMap project in Utah residents of European descent. Accordingly, we found a very strong LD between the ZNRD1 SNP with not only HLA-A*26, but also with A*25, which along with HLA-A*34 and HLA-A*66 make up the serogroup HLA-A10 [40] (Figure 1B and Figure 1C, bottom). Among the WHMC HIV⁺ subjects, the HLA-A10 group was composed mainly of HLA-A*25 and HLA-A*26 alleles (89% of the alleles that categorize to the HLA-A10 serogroup) in European Americans (EAs), whereas HLA-A*34 and HLA-A*66 alleles (78% of the alleles that categorize to the HLA-A10 serogroup) were more common in African Americans (AA; Table 1 in Materials S1). This difference in the distribution of alleles that belong to the HLA-A10 serogroup across the two major ethnic groups in the WHMC HIV⁺ cohort was statistically significant (χ² = 72.7; P = 1.1×10⁻¹⁵).

Consistent with the results of the GWAS [27], a nearly 100% LD was noted between the HCP5 SNP and B*5701 (Figure 1B and Table 2 in Materials S1). Moreover, there was a high degree of LD between the HCP5-G, HLA-C5′-C and HLA-B*57 alleles, and the HLA-C5′-C/HLA-B*57/HCP5-G-containing haplotype was found almost exclusively in European Americans (Figure 1B and Figure 1C, top). No significant differences were found in the frequency of these SNPs among HIV-infected and -uninfected individuals (Table 3 in Materials S1).

Because of the observed LD patterns, in the analyses described below, we examined whether the influence on disease course of the SNPs in ZNRD1 or HCP5 and HLA-C were independent of HLA-A10 and HLA-B*57, respectively.

Effects of ZNRD1 SNP on HIV disease

In the entire cohort of WHMC HIV⁺ subjects, heterozygosity but not homozygosity for the ZNRD1-C allele was associated with disease-retardation (Figure 2A, left panel). We stratified the cohort based on whether subjects were EAs or African Americans (AAs), and the stratified analysis revealed that the disease-retarding effects associated with the ZNRD1-C allele was restricted mainly to EAs (Figure 2A, middle and right panel). These results are consistent with the observations by Fellay et al [27] who found that the ZNRD1-C allele is associated with a slower disease course in subjects of European descent. However, we also found that this allele does not influence HIV disease course in HIV-positive African Americans in the WHMC cohort.

Download:

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

Figure 2. Disease-influencing effects associated with HLA-A10/ZNRD1-containing genotypes.

(A) Plots are Kaplan-Meier curves for time to AIDS (1987 criteria). Disease-influencing effects of the ZNRD1 SNP in all subjects (left column), and in the European American (middle column) and African American (right column) components of the WHMC cohort. (B) Kaplan-Meier plots depicting the influence on rate of disease progression of alleles categorized into the HLA-A10 serogroup. The relative hazards (RH) and 95% confidence intervals (CI) were computed before (values above the complete horizontal line) and after (values below the complete horizontal line) adjustment for ethnic background, cohort membership across different therapy eras, CCL3L1-CCR5 genetic risk group status and whether date of seroconversion was known. (C) Six genotypic groups (enumerated, color-coded) were derived by accounting for HLA-A10 status and the ZNRD1 genotype. Number of subjects in each genotypic group are shown within the color-coded boxes. (D) Kaplan-Meier plot for the disease-influencing effects of the six genotypic groups (Gp) shown in panel C. RH, CI, and significance values are depicted to the right and were obtained by Cox proportional hazards modeling. n, number of subjects. Group 1 (Lacking HLA-A10 and ZNRD1-C) is used as a reference. (E) Association between HLA-A10/ZNRD1-containing genotypes with virologic-immunologic markers of HIV disease. Group 1 (lacking HLA-A10 and ZNRD1-C) is used as a reference and comparisons are made to HLA-A10/ZNRD1 genotypic groups 2 and 5. †M-W P refers to significance values obtained by Mann-Whitney test. bCD4, nCD4, and cCD4 represent baseline and nadir CD4⁺ T cell counts (cells/mm³, mean (±SE)), and cumulative CD4⁺ T cell count (x10⁵ cell-days/mm³, mean (±SE)), respectively; the latter parameter provides a measure of the change in CD4⁺ T cell count during disease course as described previously [39]. VL, steady-state viral load (log₁₀ HIV mRNA copies/ml, mean (±SE)). DTH, delayed type hypersensitivity skin test reactivity (an in vivo parameter of cell-mediated immunity) was examined as described previously [39] and data reflect the mean (±SE) number of positive skin tests out of the four applied.

https://doi.org/10.1371/journal.pone.0003636.g002

We next determined whether the disease retarding effects associated with the ZNRD1-C allele were due to those associated with alleles categorized to the HLA-A10 serogroup for three reasons. First, the ZNRD1-C allele was in high LD with HLA-A10 (Figure 1B and Figure 1C, bottom) and of the 60 HIV-positive EAs who possessed an allele corresponding to the serogroup A10, 52 subjects also possessed at least one ZNRD1-C allele. Second, the ZNRD1-C allele was associated with a reduced rate of disease progression to AIDS only in HIV⁺ EAs (Figure 2A). Similarly, only HIV⁺ EAs who possessed alleles corresponding to the HLA-A10 serogroup had a significantly slower rate of progression to AIDS compared to those who lacked alleles corresponding to the A10 serogroup (Figure 2B). An association between HLA-A10 and disease course was not detected in HIV⁺ AA (data not shown). Notably, the strength of the disease-retarding effects (relative hazards, RH) associated with HLA-A10 were similar before (RH = 0.62) and after (RH = 0.44) adjustment for explanatory covariates that in previous studies we had found influenced disease progression in this cohort ([39]; Figure 2B). The strength of the protective effects associated with HLA-A10 were also consistent when the analysis was restricted to subjects who had not received highly active antiretroviral therapy (HAART) (RH = 0.52; 95% confidence interval (CI) = 0.028–0.93; P = 0.028 for comparison of those possessing and lacking HLA-A10; Figure S1). Third, we surmised that if both HLA-A10 and ZNRD1-C each had independent effects on disease, then the protective effects of the ZNRD1-C allele that are not because of its LD with HLA-A10 should be evident in those 85 individuals who possessed at least one ZNRD1-C allele but lacked alleles corresponding to A10 serogroup. For these reasons, we stratified HIV⁺ EAs into six genotypic groups based on their ZNRD1-C allele and/or HLA-A10 status (Figure 2C). Those lacking both ZNRD1-C and HLA-A10 served as the reference category in the statistical analyses (Figure 2C, group 1).

Subjects possessing one ZNRD1-C allele but lacking HLA-A10 (Figure 2C, group 2) had a disease course that was indistinguishable from those who lacked both ZNRD1-C and HLA-A10 alleles (group 1). This suggested that in itself the ZNRD1-C allele was not associated with disease-retarding effects in the HIV⁺ EAs we examined (Figure 2D, compare survival curves for genotypic groups 1 and 2). By contrast, subjects who possessed both a ZNRD1-C and a HLA-10 allele (group 5) had a slower rate of disease progression (Figure 2D, compare survival curves for group 5 versus group 1 (P = 0.004) or group 2 (P = 0.057)). Additionally, although the number of subjects was small, those who were HLA-A10+ but lacked the ZNRD1-C allele (group 4) also had a slower disease course that was comparable to those in group 5 who were A10+ but possessed a ZNRD1-C allele (Figure 2D). These data indicated that the disease-retarding effect associated with the ZNRD1-C allele is evident only when it resides on a haplotype that also contains HLA-A10. The failure to detect an effect of the ZNRD1-C allele on disease course that is independent of HLA-A10 did not appear to be due to insufficient sample size because there were more HIV-positive EAs who possessed one ZNRD1-C allele but lacked HLA-A10 (n = 83) than those who had one allele each of ZNRD1-C and HLA-A10 (n = 50; Figure 2D). Furthermore, similar to what was observed for HLA-A10, the associations for the six genotypic groups defined by HLA-A10 and ZNRD1-C genotypes remained consistent when the analysis was restricted to subjects who had not received HAART (data not shown).

The survival curves associated with homozygosity for ZNRD1-C suggested that it might be associated with a rapid rate of disease progression (Figure 2A, middle; RH = 4.26; 95% CI = 1.36–13.3; P = 0.013). Although there were few subjects who were homozygous for the ZNRD1-C allele with (group 6) or without HLA-A10 (group 3), it was interesting to note that subjects assigned to either of these two groups had a progressive disease course (Figure 2D). The small sample sizes preclude firm inferences but the direction of the effects associated with ZNRD1-A10 genotypic groups 3and 6 suggest that homozygosity for the ZNRD1-C allele might associate with disease acceleration even in those instances when these genotypes contain HLA-A10 alleles.

The results shown in Figure 2D were derived from the genotype of HLA-A10 and ZNRD1-C alleles. To confirm that the aforementioned inferences derived from the associations of these compound HLA-A10/ZNRD1 genotypes are valid, we determined the disease-influencing effects of the individual A10-ZNRD1 haplotypes shown in Figure 1C in the EA component of the WHMC cohort. The results of these analyses conducted at the haplotype level (Figure S2) provide further confirmation that the effects associated with the ZNRD1-C allele occur only when this allele is present on the same haplotype on which alleles that categorize to the HLA-A10 serogroup exists. First, haplotype 2 bears the ZNRD1-C allele but lacks HLA-A10 (Figure 1C, bottom), and in the heterozygous state this haplotype is not associated with disease retardation and in the homozygous state it is associated with disease acceleration (Figure S2B; reference category are those lacking haplotype 2). Second, heterozygosity for the haplotype that bears both ZNRD1-C and HLA-A10 (haplotype 3; Figure 1C, bottom) is associated with a slower rate of disease progression (Figure S2C; reference category are those lacking haplotype 3). Third, those who are heterozygous for the haplotype that bears HLA-A10 but not ZNRD1-C (haplotype 5; Figure 1C, bottom) have a slower disease course than those lacking this haplotype (Figure S2D).

Taken together, the analyses conducted thus far at the compound genotype (Figure 2D) and haplotype/haplotype pair levels (Figure S2) together demonstrate that in the cohort we examined, the ZNRD1-C allele does not have independent disease protective effects, and instead, the associations observed for the ZNRD1-C allele occur because of its LD with HLA-A10, i.e, they are attributable to the disease-modifying effects of alleles corresponding to the HLA-A10 serogroup. Further underscoring the importance of HLA-A10 in AIDS pathogenesis in EAs is the observation that the HLA-A10-bearing haplotype that lacks ZNRD1-C, albeit low in frequency, is associated with disease retardation (Figure 2D, group 4, and Figure S2D). This suggests further that HLA-A10 influences AIDS pathogenesis independent of ZNRD1-C.

To further confirm that the ZNRD1-C allele does not influence disease independent of its association with HLA-A10, we conducted univariate and multivariate analyses. In univariate analyses, both heterozygosity for ZNRD1-C (RH = 0.69; P = 0.030) and HLA-A10 (RH = 0.46; P = 0.005) associated with a reduced rate of disease progression (Table 1 under the column designated as ‘Time to AIDS’). However, in multivariate models, HLA-A10 (RH = 0.51; P = 0.028) but not the ZNRD1-C (RH = 0.85; P = 0.409) allele was associated with a reduced rate of disease progression (Table 1).

Download:

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

Table 1. HIV disease-influencing effects associated with HLA-A10 status and ZNRD1 heterozygosity in European Americans.

https://doi.org/10.1371/journal.pone.0003636.t001

The aforementioned genotype-phenotype association analyses were conducted using AIDS (1987 criteria) as a phenotypic endpoint. Therefore, it was plausible that although the ZNRD1-C allele did not have a direct impact on disease progression rates to AIDS it influenced surrogate markers of disease (e.g., viral load, baseline CD4⁺ T cell count). However, we found that subjects who had one ZNRD1-C allele but lacked an HLA-A10 allele (group 2) were similar to those who lacked both ZNRD1-C and HLA-A10 alleles (group 1) with respect to their baseline, cumulative and nadir CD4⁺ T cell counts, as well as steady-state viral load and delayed type hypersensitivity skin test reactions (a measure of cell mediated immunity [39]) (Figure 2E). Subjects in group 5 (heterozygosity for the HLA-A10/ZNRD1-C-containing haplotype) were also similar to those in group 1 with respect to most of these laboratory and immunologic characteristics, except those in group 5 had lower cumulative CD4⁺ T cell counts, a parameter that reflects the amount of CD4⁺ T cell loss over disease course and is a parameter that is highly correlated with development of AIDS ([39]; Figure 2E). These findings conveyed two points: first, in the cohort examined, the ZNRD1-C allele was unlikely to have an impact on disease independent of its LD with HLA-A10; and second, the influence of the HLA-A10/ZNRD1-C-containing haplotype on disease course occurs independent of an impact on the viral load, and that the influence of this haplotype is restricted mainly to the rate and extent of CD4⁺ T cell loss (assessed by cumulative CD4⁺ T cell count; Figure 2E), and by extension, on AIDS progression rates (Figure 2D).

To clarify further the relationship between the ZNRD1-C allele and steady-state viral load, we determined the influence of ZNRD1-C and HLA-A10 on viral load using univariate and multivariate linear regression analyses. In these analyses, the alleles were considered as independent variables and the steady-state viral load was considered as the dependent variable. These analyses revealed that the impact of the ZNRD1 allele on viral load was minimal and HLA-A10 had no influence on viral load (as reflected by the low R² values and non-significant P values in Table 1 under column designated as ‘Steady-state viral load’).

Effects of HLA-C and HCP5 on disease progression

Because of (i) their proximity to HLA-B*57, (ii) the previously reported dominant protective role of B*57 alleles in virologic control and disease progression, and (iii) the high degree of LD between the HLA-C5′-C, HLA-B*57 and HCP5-G alleles (Figure 1B and Figure 1C, top), we determined whether the effects attributable to the SNPs in HLA-C and HCP5 were in part due to these LD patterns and their presence within HLA-B*57-containing genotypes. To this end, we used strategies that were complementary to those we used to clarify the relationship between ZNRD1 and HLA-A10 alleles. We first evaluated the associations and distributions separately for the HLA-B*57, HLA-C5′ and HCP5 alleles (Figure 3), and then because of their LD patterns, we determined the associations for B*57-containing genotypes that contained or lacked HLA-C5′ and HCP5 alleles (Figure 4). We took this analytical approach of first analyzing the genotype-disease-influencing phenotypes at the level of the haplotype and then haplotype pairs (genotypes), as ultimately the phenotypic effect is conveyed by the genotype.

Download:

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

Figure 3. Disease-influencing effects associated with the HLA-B*57, HLA-C5′ and HCP5 alleles in HIV-infected subjects from the WHMC cohort.

(A–C) Plots are Kaplan-Meier curves for time to AIDS (1987 criteria). Disease-influencing effects in all subjects (left column), and in the European American (middle column) and African American (right column) components of the WHMC cohort that are associated with (A) HLA-B*57 alleles, (B) the SNP in HLA-C5′ and (C) the SNP in HCP5. RH, relative hazards; CI, confidence interval; P, significance value obtained by Cox proportional hazards modeling; n, number of subjects. (D) Distribution of the indicated alleles in the WHMC HIV⁺ cohort stratified according to time to AIDS or duration of AIDS-free status. These are indicated as time to AIDS in <2, 2 to <5, and ≥5 years, and no development of AIDS during the first 10 years or after 10 or more years of follow-up. For example, within each panel the left-most bar corresponds to subjects who developed AIDS within 2 years while the right-most bar indicates subjects who did not develop AIDS after 10 or more years of follow-up. At the bottom of each panel is shown significance value (obtained using chi-square test for linear trend) for a linear trend in the proportion of subjects possessing the indicated allele.

https://doi.org/10.1371/journal.pone.0003636.g003

Download:

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

Figure 4. Disease-influencing effects associated with HLA-B*57/HCP5/HLA-C-containing genotypes.

(A) Six genotypic groups (enumerated, color-coded) were derived by accounting first for the presence or absence of HLA-B*57 alleles and the wild-type (T/T), heterozygous (T/C) or homozygous (C/C) state for the SNP in HLA-C. This group of subjects was stratified further based on presence or absence of the HCP5-G allele. Number of subjects in each genotypic group is shown within the color-coded boxes. “+” and “-” denotes “presence” and “absence”, respectively. (B) The disease-influencing effects associated with the six genotypic groups shown in panel A. (C) Kaplan-Meier plots depicting the influence of the HCP5 SNP on disease progression in subjects in genotypic group 2 stratified according to presence (group 2A) or absence (group 2B) of the HCP5-G allele. P values in panels B and C are by the logrank test. (D) HLA-B*57/HCP5/HLA-C genotypes that associate with nonprogressive disease (B*57-NPD, i.e., groups 1, 2B and 3 depicted in panels A–C) or progressive disease (B*57-PD; group 2A depicted in panel A and C) compared with genotypes that lack HLA-B*57. RH, relative hazards; CI, confidence interval; P, significance value obtained by Cox proportional hazards modeling; n, number of subjects. The RH was computed before (values above complete horizontal line) and after (below) adjustment for ethnic background, cohort membership across different therapy eras, CCL3L1-CCR5 genetic risk group status and whether date of seroconversion was known. All plots are Kaplan-Meier curves for time to AIDS development (1987 criteria). (E) Association of the genotypes shown in panel D with virologic and immunologic markers of HIV disease. †M-W P refers to significance values obtained by Mann-Whitney test for comparison between HLA-B*57-NPD genotype and the remaining cohort subjects. bCD4, nCD4, cCD4, VL and DTH are as described in Figure 2E. (F) The distribution of HLA-B*57-NPD or HLA-B*57-PD genotypes was examined according to time to AIDS or duration of AIDS-free status (left panel) as well as according to steady-state viral load (right panel). Data are from all subjects in the cohort. Time to AIDS or duration of AIDS-free status is shown on the bottom of the left panel and indicates time to AIDS in <2, 2 to <5, and ≥5 years, and no development of AIDS during the first 10 years or after 10 or more years of follow-up. Steady-state viral load shown on the right was categorized as <10,000, 10,000 to <20,000, 20,000 to <55,000 and ≥55,000 copies/ml. P value at the top of left panel is significance value for a linear trend obtained using chi-square test for linear trend.

https://doi.org/10.1371/journal.pone.0003636.g004

In univariate analyses, possession of a B*57 allele (Figure 3A, and Table 2, model 1), and heterozygosity and homozygosity for the HLA-C5′-C allele (Figure 3B and Table 2, model 2) were each associated with significant protective effects on the rate of disease progression to AIDS. The impact of the HLA-C5′-C allele on disease progression was greater in EAs than AAs (Figure 3B). In univariate analyses, the HCP5-G allele was associated with protective effects on disease progression, but this trend did not achieve statistical significance and was mostly restricted to EAs (Figure 3C and Table 2, model 3).

Download:

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

Table 2. Analysis of the disease-modulating effects of HLA-B*57, the HLA-C5′ and HCP5 alleles/genotypes in HIV⁺ subjects from the WHMC cohort.

https://doi.org/10.1371/journal.pone.0003636.t002

Consistent with the results of the univariate analyses, with increasing AIDS-free status, there was a statistically significant step-wise increase in the proportion of subjects possessing a HLA-B*57 allele, including the B*5701 allele (Figure 3D). A similar statistically significant trend for a shift in distribution with increasing AIDS-free status was also observed for the HLA-C5′-C allele (Figure 3D). Thus, compared to subjects who developed AIDS within 2 years, HIV⁺ subjects who were AIDS-free for more than 10 years were 12.7, 5.6 and 2.1 times more likely to possess a HLA-B*57, HLA-B*5701 and the HLA-C5′-C allele, respectively (Table 3). Increasing AIDS-free status was also associated with a trend for enrichment of the HCP5-G allele although this association was not significant (Figure 3D and Table 3). It was noteworthy that the degree of enrichment of subjects with the HCP5-G allele with increasing AIDS-free status was similar to that observed for the HLA-B*5701 subtype (Figure 3D). This observation is consistent with the LD between HLA-B*5701 subtype and the HCP5-G allele (D' = 0.91 for EAs and 0.73 for AAs, Figure 1B).

Download:

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

Table 3. Likelihood of possessing specific alleles in subjects with different rates of HIV disease progression.

https://doi.org/10.1371/journal.pone.0003636.t003

When placed in a multivariate model with HLA-B*57, heterozygosity and homozygosity for the HLA-C5′-C independently associated with disease-retardation (Table 2, compare models 2 and 4). However, the statistical significance of the protective effects associated with HLA-C5′-C in multivariate models were slightly lower than those observed in univariate models (Table 2, compare models 2 and 4). This was most prominent for homozygosity for HLA-C5′-C as the hazard ratios for the rate of progression to AIDS were 0.66 (P = 0.025) and 0.72 (P = 0.067) in univariate and multivariate analyses, respectively (compare models 2 and 4, Table 2). These results indicated that possibly because of its LD with HLA-B*57, the strength of the disease-retarding effects associated with HLA-C5′-C may in part be due to the effects conferred by the HLA-B*57 allele.

In univariate analysis, the hazard ratio for the rate of disease progression to AIDS associated with the HCP5-G allele was <1 (RH = 0.67, 95% CI = 0.39–1.14, P = 0.135), suggesting that this allele might convey weak disease-retarding effects (Table 2, model 3). However, when examined in the context of a multivariate model with B*57, the hazard ratio for the HCP5 allele was >1 (RH = 2.06; 95% CI = 0.98–4.33, P = 0.056; Table 2, model 5). Conversely, in the univariate analyses, the hazard ratio for rate of disease progression associated with the B*57 allele was 0.38 (P = 0.0002, Table 2, model 1), whereas when examined in the context of a multivariate model with HCP5-G, the hazard ratio was lower (RH = 0.25) and the strength of the association increased (P = 0.0001 Table 2, model 5). When HLA-C5′-C heterozygosity and homozygosity were also added to the multivariate model (Table 2, model 6), the strength of the observed detrimental and beneficial effects associated with the HCP5-G and B*57 alleles, respectively, on disease progression became even further accentuated (Table 2, compare models 5 and 6).

We inferred that the opposing direction of the disease-influencing effects associated with the HCP5-G allele in the univariate (RH<1, protective effects) and multivariate (RH >1, detrimental effects) analyses, and the accentuation of the protective effects of the B*57 in the multivariate models, convey the following: when examined at the haplotype level, those haplotypes which contain both B*57 and HCP5-G alleles are associated with a faster rate of disease progression compared to those B*57-containing haplotypes that bear the HCP5-T allele. Consequently, in multivariate models when the overall beneficial effects of B*57 are partitioned out or accounted for, the HCP5-G allele, which is in nearly 100% LD with a subset of HLA-B*57 alleles, associates with disease-acceleration. Conversely, in multivariate models when the effects of B*57/HCP5-G haplotypes are partitioned out, the overall beneficial effects associated with B*57 are accentuated. Concordant results (and inferences) were obtained when we studied the association of the HLA-C5′–HLA-B–HCP5 haplotypes shown in Figure 1C with rates of disease progression (Figure S3).

Influence of HLA-C and HCP5 on viral load

We next examined whether the direction and the relative strength of the effects observed for the HLA-B*57, HCP5 and HLA-C5′ alleles on rates of disease progression are similar to those observed when the steady-state viral load is used as a phenotypic endpoint. The hierarchal order of the extent to which the B*57, HLA-C and HCP5 alleles explained the variability in the steady-state viral load in the WHMC HIV⁺ cohort (R² estimates as percentages are in parenthesis; Table 2) was: B*57 (3.2%; model 1)>HLA-C (0.72%; model 2)>HCP5 (0.38%; model 3). These findings indicated that at least within the context of a natural history cohort, the SNPs in HLA-C and HCP5 explained only one-third to one-tenth of the variability in steady-state viral load than the B*57 allele, and several fold less than the reported estimates in the GWAS [27].

To further evaluate the influence of these alleles on viral load, univariate and multivariate linear regression analyses were conducted in which these alleles were considered as independent variables and the steady-state viral load was considered as the dependent variable. Univariate analysis revealed that (i) a HLA-B*57 allele, (ii) one (heterozygosity) or two (homozygosity) copies of the HLA-C5′-C allele, and (iii) a HCP5-G allele each associated with a lower steady-state viral load (as reflected by the negative values of the regression coefficients; Table 1, models 1 to 3). However, the viral load-mitigating effects observed for possession of the HCP5-G allele in the univariate analysis (Coeff = −0.25; P = 0.03, Table 1, model 3) was not evident when this allele was examined in a multivariate context with B*57 alone (Coeff = 0.46; P = 0.005, Table 1, model 5), or with the B*57 allele and HLA-C5′ genotypes together (Coeff = 0.56; P = 0.001, Table 1, model 6). Thus, the opposing direction of the viral load-influencing effects of the HCP5-G allele observed in the univariate (negative value for the regression coefficients in Table 1, model 3; Coeff = −0.25) and multivariate (positive value for the regression coefficients in Table 1, models 5 (Coeff = 0.46) and 6 (Coeff = 0.56)) analyses closely mirrored the opposing direction of the disease-influencing effects of the HCP5-G allele observed in univariate and multivariate analyses. Thus, once the protective effects associated with HLA-B*57 or HLA-C5′-C are partitioned out, the HCP5-G allele associated with not only a faster rate of disease progression (RH = 2.19; 95% CI = 1.07–4.46; P = 0.032), but also a higher viral load (Coeff = 0.56; 95% CI = 0.23–0.88; P = 0.001; Table 1, model 6).

Analyses of the final multivariate regression model also revealed that the HLA-B*57, HLA-C5′ and HCP5 alleles together explained 4.66% of the variability in the steady-state viral load in the WHMC HIV⁺ cohort (R² estimate for model 6 in Table 2). Additionally, the final model indicates that HLA-B*57 alleles might confer a greater protective effect on early viral replication than HLA-C5′ alleles (Coefficient values in model 6 in Table 2 for B*57 was −0.82, whereas it was HLA-C5′ −0.08 and −0.26 for heterozygosity and homozygosity for the HLA-C5′-C allele).

Effects of B*57-, HLA-C- and HCP5-containing genotypes

The aforementioned findings indicated that the ‘protective’ phenotype attributable to B*57 (e.g. Figure 2A) is a composite, i.e., an overall phenotype that is evident before accounting for the independent phenotypic effects of the SNPs in HLA-C and HCP5-G. Based on the LD patterns between the HLA-B*57, HCP5-G and HLA-C5′-C alleles (Figure 1B and Figure 1C, top) and associations described above for each of these alleles, we surmised that a more accurate assessment of the phenotypic effects attributable to B*57 alleles might be attained by evaluation of the genotype-phenotype relationships for B*57-containing genotypes that are defined based on whether or not they also contained the HCP5-G and/or HLA-C5′-C alleles. Conversely, such an approach we surmised would also provide insights into the extent to which the protective effects attributed to the HLA-C5′-C allele is due to its presence in B*57-containing genotypes. To investigate this, we categorized subjects into six genotypic groups which permitted an analysis of the phenotypic effects associated with HCP5 and HLA-C alleles in genotypes that contained or lacked a B*57 allele (Figure 4A).

These six genotypic groups together explained ∼3.1% of variability in steady-state viral load (as analyzed by R²) and conferred differential rates of disease progression (Figure 4B). Examination of the survival curves for these six genotypic groups revealed the following genotype-phenotype relationships.

Subjects lacking a B*57 allele were assigned to genotypic groups 4 to 6 (Figure 4A). Predictably, as group 4 lacked a B*57 allele as well as the protective HLA-C5′ allele, subjects within this group displayed the fastest rate of disease progression among the six genotypic groups (Figure 4B). It was noteworthy that in those subjects who lacked B*57, but were heterozygous (group 5) or homozygous (group 6) for the HLA-C5′-C allele had comparable rates of disease progression (Figure 4B). This contrasted with the observation that when LD patterns with the B*57 allele had not been accounted for, homozygosity for the HLA-C5′-C allele was associated with a greater protective effect against disease progression than heterozygosity for this allele (Figure 3B). The latter finding indicated that the maximal protective effects of HLA-C5′-C allele on HIV disease course might be evident when this allele is present in HLA-B*57-containing genotypes, and the data presented below affirm this possibility.

Subjects with a B*57 allele were assigned to genotypic groups 1 to 3, but despite containing a B*57 allele, these genotypic groups were associated with contrasting disease-influencing effects (Figure 4B). Compared to subjects assigned to genotypic groups 1 or 3, those in group 2 had a faster rate of disease progression (Figure 4B). The distribution of the HCP5-G allele in group 2 revealed that individuals assigned to this group could be stratified into two broad categories, with subgroups 2A and 2B possessing or lacking a HCP5-G allele, respectively (Figure 4A). Remarkably, none of the 10 subjects categorized to group 2B progressed to AIDS (Figure 4C), and they all had a genotype that contained a HLA-B*57 and a HLA-C5′-C allele but lacked a HCP5-G allele (Figure 4A). By contrast, the HCP5-G-carrying subjects categorized to group 2A had a progressive disease (Figure 4C). The differential disease outcomes associated with groups 2A and 2B indicated that genotypes which contained a HCP5-G allele associated with disease acceleration despite containing two separate ‘protective’ alleles namely a HLA-B*57 and HLA-C5′-C allele (group 2A). Thus, the genotype-phenotype relationship for group 2A (which contains both HLA-B*57 and HCP5G alleles) may explain why when examined in a multivariate model, possession of the HCP5-G allele was associated with both disease acceleration and higher viral loads (Table 2, model 6).

The phenotype of subjects in group 3 was also very instructive (Figure 4B). These subjects were homozygous for HLA-C5′-C and possessed B*57 and had among the slowest rates of disease progression (Figure 4B). This occurred despite the fact that most subjects in group 3 also possessed the HCP5-G allele. Thus, among those who possessed B*57, the disease-accelerating effects associated with the HCP5-G allele was evident in those with one (group 2A) but not two (group 3) HLA-C5′-C alleles (Figure 4B). These observations underscore our previous thesis [34], [41] that analyses conducted at the allele level obscure complex genotype-phenotype relationships that occur when the phenotypic effects of both haplotypes are accounted for.

These survival curves also underscore two other points: First, a HLA-B*57-containing genotype does not require the presence of a ‘protective’ HLA-C allele to convey a protective effect on disease progression. This is reflected by the observation that subjects in group 1 who have a B*57 allele and are homozygous for the HLA-C5′-T allele have a slow disease course. Second, the maximal ‘protective’ effects associated with possession of a HLA-C5′-C allele occurs only when it is present in a HLA-B*57-containing genotype. This is reflected by the observation that subjects assigned to B*57-containing genotypic groups 2B or 3 have a much slower disease course than those categorized to B*57-lacking genotypic groups 5 and 6 (Figure 4, B and C, compare survival curves for groups 2B and 3 vs those for groups 5 and 6).

B*57-containing genotypes and progressive disease

There is clinical, virologic and immunologic data indicating a conundrum: some subjects despite possessing a ‘protective’ B*57 have a progressive disease course [20], [22], [42]. We surmised that the genotype-phenotype relationships shown in Figure 4B and Figure 4C might address this paradox as they showed that subjects with HLA-B*57-containing genotypes classify into two categories.

The first group are those subjects with genetic features present in genotypic groups 1, 2B or 3 as these genotypic groups are associated with a non-progressive disease (NPD) course (Figure 4, B and C), and collectively these genotypes are designated here as B*57-NPD genotypes in Figure 4D. In the overall cohort, B*57-NPD genotypes conferred a nearly 83% slower rate of disease progression compared to genotypes that lack a B*57 allele (RH = 0.17, 95% CI = 0.07–0.41, P = 7.2×10⁻⁵; Figure 4D, left panel). These associations detected at the level of the entire cohort (Figure 4D, left panel) remained consistent after adjustment for covariates known to influence disease progression in this cohort, including ethnicity [39], [43].

The second group are those subjects categorized to group 2A and, compared to those with B*57-NPD genotypes, the clinical course of subjects in this group is characterized by a progressive disease (PD; designated as B*57-PD genotype in Figure 4D). Remarkably, the disease course of those with a B*57-PD genotype is similar to those who lack a B*57 allele (Figure 4D). Consistent genotype-phenotype relationships were also evident when similar analyses were conducted separately in HIV⁺ EAs and AAs before and after adjusting for covariates (Figure 4D; middle and right panels) or when the analyses were restricted to subjects who did not receive HAART (Figure S4). In these HAART-free subjects, those assigned to the B*57-NPD genotypic group had a nearly 80% lower risk of progressing to AIDS than those who lacked a B*57 allele (RH = 0.21; 95% CI = 0.08–0.57; P = 0.002).

Based on the aforementioned findings we surmised that the comparison of the immunologic (CD4⁺ T cell profiles and DTH skin test reactivity) and virologic (viral load) profiles would reveal a step-wise worsening in subjects categorized to B*57-NPD, B*57-PD and those lacking a B*57 allele. In general agreement with this, as a general rule, there was a step-wise decrease in the baseline, cumulative and nadir CD4⁺ T cell counts as well as DTH responses in subjects assigned to B*57-NPD, B*57-PD and those lacking B*57 (Figure 4E). Conversely, in subjects categorized to B*57-NPD, B*57-PD and those lacking B*57 there was a step-wise increase in the steady-state viral load (Figure 4E). The only exception to this general rule was that the baseline CD4⁺ T cell counts in subjects assigned to B*57-NPD and B*57-PD were similar, but they were higher than the cell counts found in those who lacked a B*57 allele (Figure 4E).

To validate that B*57-containing NPD and PD genotypes convey contrasting phenotypic effects during HIV disease course we used a complementary approach and determined the distributions of these genotypes as a function of increasing AIDS-free status and steady-state viral loads with the intent of testing two premises. First, although increasing AIDS-free status is associated with a stepwise enrichment of subjects with a B*57-containing genotype (Figure 3D), the greatest enrichment should be for those with B*57-NPD genotypes. Second, the B*57-NPD and not B*57-PD genotypes should be enriched for subjects with a low steady-state viral load. The data shown in Figure 4F (left panel) provides evidence in support for the first premise. Nearly 12 percent of those HIV-positive individuals who were AIDS-free after 10 years of follow-up possessed a B*57-containing genotype, and strikingly ∼90% of these subjects were those with a B*57-NPD genotype (Figure 4F, left panel). The data shown in Figure 4F (right panel) provides evidence in support of the second premise as among subjects with a steady-state viral load lower than 10,000 copies/ml, there was an overrepresentation of those with B*57-NPD genotypes. It was noteworthy that the proportion of subjects with a B*57-PD genotype was similar in those who had a steady-state viral load of <10,000 and ≥55,000 copies/ml (Figure 4F). Thus, these distribution patterns (Figure 4F) provide additional validity for the categorization of B*57-containing genotypes into B*57-NPD and B*57-PD genotypes as a means to improve the accuracy of identifying those B*57-carrying subjects who may have a slower disease course or control of viral replication during the early phases of infection.

Viral load-independent effects of B*57-NPD genotypes

Based on the well-established, strong relationship between steady-state viral load and AIDS development [28], [29], one interpretation of the aforementioned findings might be that the B*57-NPD genotypes convey disease-retardation by simply influencing the extent of early viral replication. However, in a recent study, we found that although the steady-state viral load is a strong predictor of AIDS risk, it accounted for only ∼12% of the variability in AIDS progression rates [39]. The findings by Rodriguez et al also suggest that the contribution of viral load to the explained variability in the rate of decline in CD4⁺ T cell counts might not be very large [44], [45]. These data indicated that parameters that are independent of viral load are also critical determinants of AIDS pathogenesis. We therefore used nested multivariate Cox proportional hazards models (Table 4) to assess whether HLA-B*57-NPD genotypes affected disease progression independent of the viral load.

Download:

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

Table 4. Viral load-independent disease-influencing effects of HLA-B*57-containing genotypes.

https://doi.org/10.1371/journal.pone.0003636.t004

Before accounting for any of the covariates, those HIV-positive persons categorized to B*57-NPD had an ∼83% lower risk of progressing to AIDS compared to individuals who lacked a HLA-B*57 allele (RH = 0.17; 95% CI = 0.07–0.41; Table 4, model 1). We next adjusted–individually and in unison–for the disease-influencing effects of several explanatory variables. These analyses revealed that even after adjustment for the disease-influencing effects of these covariates, the B*57-NPD genotypes remained strong independent predictors of a slow rate of disease progression (Table 4, models 2 to 10). By contrast, the rates of disease progression among those categorized to B*57-PD and those lacking a B*57 allele were not statistically different (Table 4, model 1), and this association remained after inclusion of the other covariates in the model (Table 4, models 2 to 10).

Study subjects

HIV-positive subjects were from the Department of Defense (DoD) HIV Natural History Study (NHS) cohort followed at Wilford Hall Medical Center (WHMC) and more recently at the Brooke Army Medical Center (BAMC), San Antonio, TX. The studied population is the local component of a prospective multisite observational cohort from the United States Military's Tri-Service AIDS Clinical Consortium (TACC) HIV Natural History Study. Unidentified cast-off blood from subjects participating in training at Lackland AFB, TX was used for the HIV-negative control population. The demographic and clinical characteristics of the HIV-positive WHMC cohort have been described extensively [34], [36], [37], [38], [39], [59], [60], [61]. The voluntary, fully written informed consent of the subjects studied in this research was obtained as required by Air Force Regulation 169-9 and additional approval from the Institutional Review Board (IRB) of the University of Texas Health Science Center, San Antonio, TX.

Genotyping

Genotyping for HLA-B*57 and HLA-B*5701 subtypes was undertaken in 1,143 HIV⁺ subjects. Polymorphisms in the HCP5 (rs2395029), HLA-C5′ (rs9264942) and seven SNPs in and around ZNRD1-RNF39 genes (Figure 1A) were genotyped in 1,230 HIV⁺ subjects and 1,129 HIV-negative subjects. Haplotypes based on HLA-A10 status and the seven ZNRD1 SNPs as well those based on HLA-B*57, HCP5 and HLA-C5′ alleles were generated using unphased data. For this purpose, we used the PHASE software. Detailed materials, methods and description of the study cohort are available as supporting online material (Materials S1).