Host determinants of HIV-1 control in African Americans

Abstract

We performed a whole-genome association study on HIV-1 viral load setpoint in an African American cohort (n=515), and an intronic SNP in the HLA-B gene showed one of the strongest associations. Using a subset of patients, we show that this SNP reflects the effect of the HLA-B*5703 allele, which shows a genome-wide significant association with HIV-1 VL setpoint (p=5.6×10⁻¹⁰). These analyses therefore confirm a member of the HLA-B*57 group of alleles as the most important common variant influencing viral load variation in African Americans, consistent with what is observed in individuals of European ancestry in which the most important common variant is HLA-B*5701.

A recent-genome wide association studied carried out in individuals of European ancestry identified two polymorphisms associated with viral load at setpoint, and a third set of polymorphisms associated with a simple measure of disease progression [1]. One variant that was found to be associated with setpoint (rs2395029) encodes a nonsynonymous change in the HCP5 gene, and is also a tag for HLA-B*5701, which has been shown to associate with improved early outcomes after HIV-1 exposure [2,3]. The other variant that associated with setpoint (rs9264942) is 35kb upstream of the HLA-C locus and appears to be tagging a causative variant(s). A third variant (rs9261174) that associated with disease progression is located near the ZNRD-1 gene in the MHC region, although functional work on this gene has not yet identified a causal variant. Together, these variants are able to explain about 14% of the observed variation in outcome upon HIV-1 exposure.

A follow-up study investigated the impact of these same SNPs in an HIV-1 positive African-American cohort (n=121)[4]. Similar to the results in individuals of European ancestry, they found that the HLA-C associated variant (rs9264942) again associated with viral load, with the C “high expression” allele leading to lower viral load. They did not see an association with the G allele of rs2395029, however this allele is rare in African-Americans; the allele is in linkage disequilibrium (LD) with HLA-B*5701in people of European ancestry, which is also rare in people of African descent. However, an analysis of the HLA-B alleles present in the region showed an association between HLA-B*57 (comprised predominantly of HLA-B*5703) and favorable virologic outcome.

Although this and other studies [5,6] have assessed the impact of variants first identified in patients of European ancestry in African American individuals, there has not yet been any genome wide investigation of the most important common variants influencing viral load in patients of primarily African ancestry. Here we present the first genome-wide association study of determinants of HIV-1 control performed in a non-European population. Using a cohort of African American individuals (n=515), we sought to evaluate the associations previously reported and to discover novel or population-specific genetic variants that associate with HIV-1 control.

Methods

Samples

This study includesHIV-1 infected African American adult subjects enrolled in either the U.S. military (DoD) HIV Natural History Study (NHS)or in the Multicenter AIDS Cohort Study(MACS). This study was approved by local institutional review boards, and each subject provided written, informed consent.

The U.S. Military (DoD) HIV Natural History Study (NHS) (http://www.idcrp.org/hiv-natural-history-study.html) is an ongoing, prospective, continuous enrollment cohort study of consenting military personnel and beneficiaries with HIV infection and includes participants from the Army, Navy/Marines, Air Force, and their dependents. Since 1985, routine HIV testing (ELISA and confirmatory Western blot) has been used to exclude HIV-infected persons from enlisting for military service or from overseas deployment. Periodic testing among active duty members occurs every one to five years, resulting in a defined seroconversion window for incident HIV infection. Subjects with HIV infection are referred to military medical centers where they receive evaluation, and ongoing care, and are invited to enroll as participants in the DoD HIV NHS.

Those who consent to enroll in the DoD HIV NHS are seen every six months by an HIV specialist as part of the study, in addition to receiving routine clinical care. Data are collected on demographic characteristics, markers of HIV disease progression, medication use, and clinical events with medical record confirmation. Cells, plasma, and serum are collected at each visit and stored in a central repository.

Information was extracted from the database on HIV-infected African American individuals with 4 years between their last negative and first positive HIV tests, at least 5 million cells stored in the repository, and either one viral load result available between 3–12 months post seroconversion (n=140) or two viral loads within3 months -3 years post seroconversion (n=347). Ethnicity was self-identified. Seroconversion date was estimated as the midpoint between their last negative and first positive HIV tests.

The Multicenter AIDS Cohort Study (MACS) (http://www.statepi.jhsph.edu/macs/macs.html) is an ongoing prospective study of the natural and treated histories of HIV-1 infection in homosexual and bisexual men conducted by sites located in Baltimore, Chicago, Pittsburgh and Los Angeles. A total of 6,973 men have been enrolled; 3, 427 subjects were HIV-seronegative at study entry and were tested for seroconversion semiannually by ELISA, with confirmation of positive tests by Western blotting. Of the seroincident subjects, African Americans with DNA and viral load data available before treatment initiation were selected for inclusion in the current study.

Other cohorts referenced in this analysis include HIV-1 infected adult subjects of European ancestry collected by Euro-CHAVI (http://www.chuv.ch/imul/imu_home/imu_collaborations/eurochavi.htm) and MACS who were included in a previous whole genome association study (n=2362) (Fellay et al, submitted). The Euro-CHAVI cohort represents a consortium of 8 European cohorts and 1 Australian cohort that agreed to participate in the Host Genetic Core initiative of the Center for HIV/AIDS Vaccine Immunology (CHAVI).

Genotyping

All samples were genotyped using Illumina Human Hap 1M (n=368), Human Hap 1M-Duo (n=135) or Illumina Human Hap 550K (n=12) bead chips. All samples are brought into a single Bead Studio file using the standard Illumina cluster file. For quality control purposes, any sample that has very low intensity or a very low call rate using the Illumina cluster (<99%) is deleted. All SNPs that have a call frequency below 99% are put into a filter and reclustered (excluding those on the X-chromosome). The reclustering step creates SNP calling errors, but the following procedures are used to prevent the errant calls from being released in the final report:1) the SNPs with a cluster separation value below 0.3 are deleted and 2) any SNPs with a Het Excess value between −1.0 to −0.1 and 0.1 to 1.0 are deleted. The filter is then released and any SNP with a call frequency below 99% is deleted. These procedures resulted in a success rate of genotyping calls ranging from 99.20%–99.999%, and 1,212,217 SNPs were included in the analysis. Ten samples were excluded because of insufficient call rate.

Specification of gender check, cryptic relatedness check, low minor allele frequency (MAF) check, the Hardy-Weinberg equilibrium check, and a recheck of the genotyping quality were all performed as described in [1], and no samples were omitted at these steps. 129,723 SNPs were dropped due to a low MAF. We also required that all SNPs included in the study were successfully genotyped in at least 50% of the samples, hence202, 676 SNPs were dropped at this point, many of which were those not genotyped on all of the chips.

We have used the Illumina 1M and 1M-Duo bead chip data as input into the Penn CNV program [7], which allows us to look at deletions (0, 1 copy) as compared to wild type (2 copies) and duplications (3, 4 copies) as compared to wild type (2 copies). Due to the complications of hemizygosity in males and X-chromosome inactivation in females, this analysis was restricted to autosomes. Additionally, to ensure that we worked with high-confidence CNVs, we excluded any CNV for which the difference of the log likelihood of the most likely copy number state and the less likely copy number state was less than10 (generated using the -conf function in Penn CNV). We limited our analysis to CNVs that occurred in at least 3 people (MAF>0.003). Further quality control thresholds used in Penn CNV are detailed in Ge et al [8]. In this analysis, 497 subjects were included.

HLA-B allotypes were assigned by DNA sequencing, beginning with the amplification of genomic DNA using primers that flank exons 2 and 3. PCR products were cleaned using Ampure (Beckman Coulter). The cleaned products were cycle sequenced on the ABI9700. The cycle sequenced products were cleaned using Clean SEQ (Beckman Coulter) and then run on the ABI PRIZM 3730. Sequence analysis was carried out using Assign (Conexio Genomics).

The EIGENSTRAT method [9] was used to control for population stratification. Assessment of population structure in 616African Americans using EIGENSTRAT results in 73 significant axes of stratification after the removal of 35 population outliers. The first axis makes a larger contribution to the proportion of variation (0.6%) explained than the second axis (0.2%) and reflects the degree of African versus European ancestry in individuals. We therefore used only the first axis as a covariate in our association analyses to control for population stratification.

Phenotype

Setpoint viral load was defined as1) the mean viral load for samples with two or more viral load values that were collected at least 30 days apart, within 3 months –3 years post seroconversion, and were within a 1 log range, similar to what was done in the previous study of determinants of setpoint in subjects of European ancestry [1], or2) the first available viral load within 3 –12 months post seroconversion, as long as there was a corresponding CD4+ T cell count greater than 350/ul. All viral loads were prior to the initiation of antiretroviral therapy. Viral loads from the two definitions were highly correlated (r²=0.74) and the first definition was preferentially used when sufficient data was available.

Statistical Analysis

Viral load at setpoint was used as a quantitative trait in a linear regression using additive allelic effects. Because previous studies have found that gender and age may be associated with viral load, these factors were used as covariates the model [10,11]. The first Eigenstrat axis was used to control for population stratification and the cohort was also included in the model given baseline differences between the two groups. Individual regressions for each SNP were carried out using PLINK v1.06 [12,13]. Bonferroni correction was used to control for multiple comparisons; a p-value < 5×10⁻⁸ was considered genome wide significant.

All HLA-B allotypes were tested for association with viral load set point in a linear regression model, and were also evaluated to determine if they were responsible for associations observed for SNPs in the genome-wide SNP association analyses.

Results

From the DoD HIV NHS, 487subjects met inclusion criteria and 471 were successfully genotyped. From the MACS cohort, 158 subjects met the inclusion criteria and 145 were successfully genotyped. Thirty-five subjects were dropped due to the EIGENSTRAT correction for ancestry, and 66 subjects were not included because their viral load results did not meet the definition of setpoint described in the Methods section, leaving 515 subjects in the final dataset. Table 1 shows the baseline characteristics of subjects in the DoD and MACS cohorts as well as the bead chips which were used for genotyping. There were more females in the DoD cohort (p=0.006) and subjects in this cohort were on average younger at seroconversion (p<0.001); however, there was no difference between the cohorts in mean set point viral load (p=0.13).

Table 1.

Baseline characteristics of subjects in the MACS and DoD cohorts (n=515)

	MACS	DoD
Number of subjects	118	397
% Male	100%	94%
Mean Age at Seroconversion (range)	32.0 (20–55)	27.5(18–55)
Mean set point VL (log10, range)	4.00 (1.91–6.01)	4.12 (1.91–5.97)
Mean Year of Seroconversion (range)	1991 (1984–2005)a	1996 (1986–2003)
MAF for rs2523608	39.4%b	35.4%
HLA-B*5703 carriers	7.5%b	8.3%
Illumina 1M	53%	77%
Illumina 1M-Duo	37%	23%
Illumina 550K	10%	0%

^aSerocoversion date was only available for 52 of the 118 MACS subjects.

^bp>0.05 comparing MACS with DoD cohort

MAF = minor allele frequency

No single SNP had a genome-wide significant association (p<5×10⁻⁸) with viral load at setpoint. The tables list the top 20 genome-wide associations with setpoint (Table 2), the top 10 associations from the MHC region (Table 3A), and the top 10 functional SNPs that associate with set point (Table 3B). A functional SNP was defined as a SNP that would cause the gain or loss of a stop codon, a nonsynonymous coding change, or one that occurred in a splice site.

Table 2.

Top 20 setpoint associated outcomes in African Americans (n=515)

SNP	Rank	CHR	P-Value	Closest gene	Type	Frequency
rs454422	1	20	1.49E-06	MCM8	Intronic	0.258
rs2523608	2	6	2.29E-06	HLA-B	Intronic	0.366
rs6948404	3	7	3.41E-06	AOAH	Intronic	0.074
rs558718	4	19	3.71E-06	EVI5L	Intronic	0.110
rs1357339	5	11	4.58E-06	N/A	Intergenic	0.035
rs1413191	6	13	4.61E-06	GPC5	Intronic	0.181
rs236104	7	20	6.72E-06	MCM8	Intronic	0.252
rs7998089	8	13	7.41E-06	GPC5	Intronic	0.191
rs2593321	9	3	7.70E-06	AC023798.16	Intergenic	0.231
rs6492611	10	13	7.97E-06	GPC5	Intronic	0.208
rs4872511	11	8	9.25E-06	PPP3CC	Downstream	0.011
rs2280890	12	8	9.25E-06	SORBS3	Upstream	0.011
rs2789066	13	6	9.41E-06	RP11-100A16.1	Upstream	0.127
rs430374	14	18	1.01E-05	ST8SIA5	Intergenic	0.196
rs9910853	15	17	1.07E-05	ZNF652	Intronic	0.085
rs762372	16	21	1.10E-05	U6	Upstream	0.470
rs236106	17	20	1.23E-05	MCM8	Intronic	0.252
rs1348478	18	5	1.26E-05	PRR16	Intergenic	0.337
rs12103812	19	17	1.34E-05	ZNF652	3′ UTR	0.084
rs8014482	20	14	1.44E-05	AL355773.4-1	Intergenic	0.256

Age, gender, cohort, and one significant EIGENSTRAT axis were used as covariates.

Table 3.

Other setpoint associated outcomes in African Americans (n=515)

A. Top 10 setpoint associated outcomes in the MHC region in African Americans
SNP	Overall Rank	P-Value	Gene	Type	Frequency
rs2523608	2	2.29E-06	HLA-B	Intronic	0.366
rs34548063	23	1.57E-05	STK19	Stop Gained	0.028
rs2523933	29	2.03E-05	HLA-80	Intergenic	0.205
rs2844538	53	5.07E-05	ZDHHC20P2	Downstream	0.492
rs2596503	81	8.15E-05	HLA-B	Downstream	0.162
rs9266689	96	9.86E-05	ZDHHC20P2	In non-coding gene	0.485
rs4151650	99	0.0001	CFB, C2	Synonymous Coding	0.030
rs1736936	99	0.0001	HLA-G	Upstream	0.492
rs9378200	150	0.0002	UQCRHP	Intergenic	0.013
rs4713213	150	0.0002	OR5V1;OR12D3	Intronic	0.331
B. Top 10 functional SNPs that associate with setpoint
rs34548063	23	1.57E-05	STK19	Stop Gained	0.028
rs1034405	91	9.65E-05	C3orf18	Nonsynonymous Coding	0.332
rs4838865	99	0.0001	TUBGCP6	Nonsynonymous Coding	0.048
rs6542522	99	0.0001	C2orf76	Nonsynonymous Coding, Splice site	0.196
rs2280801	332	0.0004	BAT2	Nonsynonymous Coding	0.012
rs13146272	332	0.0004	CYP4V2	Nonsynonymous Coding	0.394
rs2273549	332	0.0004	TPC11L1	Nonsynonymous Coding	0.092
rs2278329	332	0.0004	OSMR	Nonsynonymous Coding	0.042
rs10423723	435	0.0005	AC020907.3	Nonsynonymous Coding	0.625
rs7258700	435	0.0005	AC020907.3	Nonsynonymous Coding	0.625

Age, gender, cohort and one significant EIGENSTRAT axis were used as covariates.

The most significant SNP in the MHC region for association with viral load setpoint in this African American cohort was rs2523608, located in the HLA-B gene (p=2.3×10⁻⁰⁶, Figure 1A, Table 3A). We found that the same SNP was also significantly associated with HIV-1 set point in a large sample of individuals of European ancestry (p = 1.1×10⁻⁰⁶, corrected for age and gender, Figure 1B). This association remains nominally significant (p=0.0083) after accounting for variants in the MHC region previously shown to associate with HIV-1 outcomes (rs2395029, rs9264942, rs9261174), and 12 significant EIGENSTRAT axes to control for population stratification in this cohort.

Figure 1. Distribution of mean viral load (VL) setpoint according to genotype.

The direction of the effect of the rs2523608 genotype is consistent in A) African Americans and B) Europeans. C) Samples with HLA-B*5703 have a lower HIV-1 viral load setpoint.

The rs2523608 variant is located in intron 5 (according to Ensembl transcript ENST00000376228), over 100bp from the nearest exon. Analysis of the HLA-B allotypes from 285 genotyped study subjects showed that this association was due to the association between rs2523608 and HLA-B*5703(D′=1, r²=0.075. The degree of linkage disequilibrium can be quantified using the D′ statistic [14]. This statistic compares the ancestral recombination patterns between two variants by standardizing allele frequencies. A D′=1 indicates that one variant always appears on the background of the other. On the other hand, the r² statistic is sensitive to allele frequency differences and assesses the degree to which the two variants appear together [15,16]). When considered alone, HLA-B*5703 was by far the strongest association with HIV-1 viral load setpoint for any HLA-B allotype, showing genomewide significance (p=5.6×10⁻¹⁰, with age, gender, cohort and one EIGENSTRAT axis as covariates)(Table 4, Figure 1C). Moreover, when HLA-B*5703 was included as a covariate, it was able to account for the effect of the rs2523608 genotype. This analysis shows that HLA-B*5703 is the most important common variant in influencing viral load in African Americans, explaining about 10% of the variation in viral load setpoint in this dataset, with an allele frequency of about 4.0%.

Table 4.

Top associations between HLA-B allotypes and setpoint(n=285)

Allotype	Rank	P-value	Frequency
HLA-B*5703	1	5.6E-10	0.040
HLA-B*3910	2	0.00032	0.006
HLA-B*1517	3	0.00040	0.006
HLA-B*4501	4	0.00084	0.062
HLA-B*1302	5	0.020	0.005
HLA-B*580101	6	0.021	0.016
HLA-B*5802	7	0.022	0.040
HLA-B*4201	8	0.022	0.040
HLA-B*140201	9	0.024	0.020
HLA-B*1801	10	0.025	0.016

Age, gender, cohort and one significant EIGENSTRAT axis were used as covariates.

CNV analysis

There were 8,724 SNPs that showed evidence of a duplication and 16,778 SNPs that showed evidence of a deletion. The CNV calls for each SNP were then run as genotypes in a regression using an additive genetic model, testing for association with HIV-1 setpoint. Gender, cohort and the first EIGENSTRAT axis were used as covariates. Using a Bonferroni correction (6×10⁻⁶ for duplications and 3×10⁻⁶ for deletions), no SNPs reached genome-wide significance for either deletions or duplications. Furthermore, when the association results from these two models were compared, no SNP associated with both deletions and duplications with a p-value of less than p=0.05.

Association of previously implicated variants

We also analyzed genetic variants that have previously been shown to have an effect on HIV-1setpoint. First, we tested the association of rs2395029, a nonsynonymous SNP in the HCP5 gene that is a tag for the functional allele HLA-B*5701 and found this SNP to show a weak association with viral load setpoint (p=0.030, Table 5). This SNP has a very low minor allele frequency in African Americans (MAF=0.008) due its virtual absence in West African populations. Power to detect an association in this cohort is, therefore, only 81% at p<0.05, assuming that the effect size in our cohort is comparable to that seen for individuals of European descent [1].

Table 5.

Association outcomes for genetic variants that have been previously reported to associate with setpoint

SNP	Gene	Location	Pa(AA)	MAF (AA)	Pb (European)	MAF (European)
rs2395029	HCP5	Nonsynonymous	0.030(n=513)	0.008	4.6×10−35 (n=2362)	0.048
rs9264942	HLA-C	Upstream	0.018(n=515)	0.286	6.0×10−32 (n=2362)	0.412
rs9261174	ZNRD1	Intergenic	0.352(n=511)	0.245	1.1×10−04 (n=2362)	0.141
rs333 (Δ32)	CCR5	Genic Deletion	0.484(n=502)	0.017	1.9×10−10 (n=2334)	0.099

^aP-values are corrected for age, gender, cohort, and one significant EIGENSTRAT axis.

^bP-values are corrected for age, gender and 12 significant EIGENSTRAT axes.

We then tested the association between rs9264942, a C → T polymorphism 35kb upstream of the HLA-C gene. This SNP itself is not causal, but is a tagging SNP for unknown causal variant(s). We observed a weak association between rs9264942 and setpoint in African Americans (p = 0.018, Table 5).

We also tested the associations between viral load at setpoint and rs9261174, a SNP located near the ZNRD1 gene in the MHC region, and CCR5-Δ32, a 32bp deletion in the CCR5 gene (rs333)that is rare in non-European populations. In our African American cohort, neither rs9261174 (p=0.352) nor CCR5-Δ32 (p=0.484) showed an association with viral load at setpoint (Table 5).

Discussion

This was the first genome-wide association study on HIV-1 outcomes to be carried out in an African-American cohort, the majority of who were infected with HIV-1 subtype B. We have shown that the intronic SNP rs2523608, the top associated SNP in the MHC region, is tagging HLA-B*5703. The D′ =1 between rs2523608 and HLA-B*5703, and a regression model shows that the HLA-B*5703 genotype is able to account for the effect of this intronic SNP. HLA-B*5703 strongly associates with viral load at setpoint, and reached whole genome-significance in the subset of samples for which HLA-B allotype data is available (p=5.6×10⁻¹⁰ in a model that also included age, gender, cohort, and the first EIGENSTRAT axis, n=285).

HLA-B*5701 is an important mechanism of HIV-1 viral control in the European population. It has an allele frequency of about 6.1% in a European population, but was not observed in a Yoruban population [17]. Its close relative HLA-B*5703 is absent in a European population, but has an allele frequency of about 5.8% in a Yoruban population [17]. Here, we have shown that African-American individuals who have HLA-B*5703 also show improved viral control, of a similar magnitude to that afforded by HLA-B*5701 in people of European descent. Thus, our results indicate that the general mechanism of genetic control of the HIV-1 virus is similar between African-Americans and Europeans: HLA-B*5701 accounts for about 6% of the observed variation in viral load setpoint in Europeans (Fellay et al, submitted), and HLA-B*5703 accounts for about 10% of the observed variation in viral load setpoint in African-Americans. There was also a small contribution to HIV-1 control by HLA-B*5701 (frequency=0.3%) in our African-American dataset, due to admixture.

In addition to HLA-B*5703, we also found that HLA-B*3910 and HLA-B*1517 may be playing a lesser role in viral control, although additional studies would be needed to confirm these observations. This pattern was similar to that observed in people of European descent, where HLA-B*5701 is the largest determinant of HIV-1 control but other HLA-B alleles (HLA-B*27, HLA-B*35 and others) also play a role (Fellay et al, submitted). The alleles that we have identified are further supported by an analysis of the MHC region conducted in southern African populations infected with HIV-1 subtype C, where HLA-B*5703 was found to be the HLA allele most strongly associated with a decreased viral load setpoint [3,18], with a weaker contribution by HLA-B*39[18].

We found a reduced or absent association with viral load setpoint when we explicitly checked variants that had been shown to associate in a cohort of European descent. Similar outcomes were seen in Shrestha et al. [4], who saw a reduced association between rs9264942 and setpoint, and no association between rs2395029 and setpoint. Our sample size was over four times larger than Shrestha et al, and revealed only weak associations between both variants and setpoint. So while both rs9264942 and rs2395029 definitively associate with viral load setpoint in European populations, neither study was able to replicate these associations in an African American cohort.

It is worth noting that it is only the HLA-B*5701 association in the previous study [1]where the causal site is thought to have been identified. In the cases of rs9264942 and rs9261174, it is not likely that the associated variants are themselves causal, but are rather markers of an as yet unidentified causal site or sites. It may therefore not be a coincidence that the HLA-B*57 association is the only one to show a similar effect in African Americans as in individuals of European ancestry. For the other two, the different and generally lower linkage disequilibrium in this region in African Americans (Figure 2) could mean that the causal sites are no longer being tagged by these variants.

Figure 2. Average linkage disequilibrium (LD) in the MHC region in people of European descent versus African Americans.

The average D′ and the average r² are both shorter in the African American genome as compared to the European genome.

A potential limitation of this study is that HIV-infected persons with rapid disease progression may have been excluded. Rapid progressors may not have had many research visits and therefore may not have had enough cells in the repository to be included in this study. Also, because of their rapid progression, they may not have had viral loads available that satisfied the definition of set point. Approximately ten percent of the subjects in this study progressed to a CD4 count < 200 cells/mm³ within 2years of seroconversion; therefore, there were at least some rapid progressors included.

Earlier studies have likewise suggested that HLA-B*5703 may also be involved in HIV-1 control in African-Americans, but our study shows that it is indeed the most significant common genetic factor affecting early viral control in this population. By using a genome-wide scan to implicate another allele of HLA-B*57 in HIV-1 control, in an entirely different ancestral background from where this association had previously been observed, we provide further support as to the important role of HLA-B*57 in HIV-1 control and the decreased fitness level of the viral mutants that are selected for by HLA-B*57. Given the increased burden of disease in African and African-American populations, and the paucity of common variants that clearly influence HIV-1 control, it is important to continue to investigate rare variants that may function specifically in these populations.