In 292 initially human immunodeficiency virus (HIV)-1-serodiscordant and cohabiting Zambian couples, HLA-DRB1 and -DQB1 variants were associated with HIV-1 transmission events during a 7-year follow-up period. Initially seronegative partners with either DRB1*0301-DQB1*0201 (relative hazard [RH], 1.60; P = .009) or DRB1*1503-DQB1*0602 (RH, 1.67; P = .03) showed accelerated seroconversion. Carriage of DRB1*1301 in initially seropositive partners led to delayed transmission of HIV to their spouses (RH, 0.54; P = .05). The combined groups of seroprevalent and seroincident partners (n = 433) also differed from those who remained seronegative (n = 151), with regard to 2 common haplotypes, DRB1*1302-DQB1*0604 (relative odds [RO], 0.28; P = .003) and DRB1*1503-DQB1*0602 (RO, 1.81; P = .02). Statistical adjustments for other host factors (age, sex, genital ulcer, and index partner's virus load) known to influence transmission of HIV-1 seldom altered the genetic relationships. Overall, associations of HLA class II polymorphisms with both HIV transmission and acquisition are not as readily interpretable as are effects reported for other loci.
Sub-Saharan Africa has been experiencing the worst public health and socioeconomic problems associated with the HIV/AIDS pandemic [1, 2]. Studies of native Africans have uncovered important host and viral factors predictive of relative risk for heterosexual transmission of HIV-1 and subsequent pathogenesis [3–14]. Consensus findings from such research have suggested mechanisms that may eventually guide the design of effective biologic interventions [15–19].
Earlier analyses of initially HIV-1-serodiscordant and cohabiting Zambian couples [12, 14, 20] have identified multiple host factors with independent influences on HIV-1 infection and subsequent virus-host equilibration. A comprehensive analysis of both genetic and nongenetic host factors, in relation to heterosexual transmission of HIV-1, now suggests that HLA class II (DRB1 and DQB1) polymorphism may play a dual role in modulating both propagation of HIV by seropositive partners and acquisition of HIV by seronegative partners.
Subjects and Methods
Participants and HIV-1 identities. Initially HIV-1—serodiscordant couples (one partner seropositive and the other seronegative) were selected from the Zambia-UAB HIV Research Project (ZUHRP). Informed consent was obtained from all study participants, in accordance with the human-experimentation guidelines of the US Department of Health and Human Services. Detailed procedures for recruitment, counseling, quarterly follow-up visits, and laboratory testing have been described elsewhere [12, 21, 22]. Between February 1995 and December 2002, within-couple transmission of HIV was determined by phylogenetic analyses of subgenomic HIV-1 sequences corresponding to gag, env (gp120 and gp41), and the long terminal repeat regions [12, 20]. The vast majority (95%) of HIV-1 sequences derived from ZUHRP participants belonged to subtype C (HIV-1C), although other subtypes—such as A, D, G, and J—were also detected occasionally . Overall, virus isolates from each transmission pair (transmitter and seroconverter) were closely related, with a median of 1.5% and a maximum of 4.0% nucleotide substitutions . In contrast, the median nucleotide substitution rate was 8.8% for unlinked HIV-1C viruses from the same cohort or elsewhere.
Virological measurements. HIV-1 RNA levels in plasma samples from patients were measured universally by use of the Roche Amplicor 1.0 assay in a laboratory certified by the Virology Quality Assurance Program of the AIDS Clinical Trials Group. Additional comparison of 4 commercial viral assays (Chiron Quantiplex HIV-1 branched DNA, Amplicor version 1.0, Amplicor version 1.5, and Organon-Teknika nucleic acid sequence-based amplification HIV-1 RNA QT) suggested that each of them could successfully quantify cell-free HIV-1 RNA in Zambians; the modified (new-generation) Amplicor assay (version 1.5) with additional primer sets targeting non-B-subtype viruses showed slight (0.3 log10 copies/mL;P > .05) improvement, compared with the first-generation Amplicor assay (version 1.0) . The index partners with medium (104–105 copies/mL) and high (>105 copies/ mL) levels of HIV-1 RNA transmitted viruses more readily than did those with reduced viremia (<104 copies/mL) . These 3 categories of virus load were retained as a key predictive covariate in subsequent analyses.
DNA extraction and HLA typing. Genomic DNA was extracted from whole blood by use of the QIAamp Blood Kit and protocols recommended by the manufacturer (Qiagen). Allelic specificities of HLA class II genes DRB1 and DQB1 were resolved to their 4–5-digit molecular level by use of several techniques, including solid-phase DNA sequencing (Amersham Pharmacia Biotech) and polymerase chain reaction with sequence-specific primers, as described elsewhere for another cohort of native Africans (Rwandans) . Two-locus DRB1 and DQB1 haplotypes were assigned according to linkage disequilibria observed in populations of African ancestry [24, 25].
Statistical analyses. The overall genetic heterogeneity between groups of patients was measured by use of exact tests based on metropolis algorithms  and row-by-column contingency tables. Other analyses relied on several statistical routines in SAS (version 8.0; SAS Institute). More specifically, χ2 tests were applied, to compare effects of categorical variables—such as sex and the distribution of HLA alleles, and haplotypes—between groups of patients defined by the status of transmission of HIV-1. Student's t tests and F tests were used to compare continuous variables, including age and mean log10 HIV-1 load of patients. The univariate relative hazards (RHs) of HIV-1 infection for specific HLA variants, were estimated by use of Cox proportional hazards models, using subjects stratified by the HLA markers under investigation. Kaplan-Meier plots were used to show the time from enrollment to incident HIV-1 infection (transmission by seropositive index partners and acquisition by seronegative nonindex partners). Both Wilcoxon and log-rank tests of significance are shown, since the 2 tests differ in their emphasis on earlier and latter periods of follow-up. All HLA factors showing at least a marginal association (P < .10) with HIV-1 infection (in either cross-sectional or longitudinal analyses, described above) were tested in multivariable models. Further statistical adjustments were made for genetic markers identified at the HLA class I (M.T. Dorak, J.T., S.A., and R.A.K., unpublished data) and chemokine receptor (CCR2 and CCR5) loci (J.T., S.A., and R.A.K., unpublished data).
Characteristics of patients and key measures of outcome. Our analyses focused on a subset of 292 HIV-1-serodiscordant couples at relatively high risk for transmission of HIV, as reflected by unprotected sex (both self-reported and biologically proven), genital ulceration/inflammation, and a history of sexually transmitted infections . These couples were classified into 3 subgroups: 124 linked transmitting couples, 17 unlinked couples, and 151 persistently nontransmitting couples. Data censored in December 2002 were suitable for nested case-control analyses of viral and host factors related to transmission of HIV-1 and seropositivity (table 1) and for analysis of time to transmission. Overall, the sex ratio (F:M) was ∼0.90 when the 17 unlinked pairs were excluded, and the majority (∼87%) of fully analyzed subjects were ≤40 years old at enrollment. Virus load was measured once in >95% of all HIV-1-seropositive individuals, and high-resolution HLA class II typing was successful in all individuals.
Distribution of HLA-DRB1 and -DQB1 alleles and haplotypes in Zambians, in relation to HIV-1 seropositivity. At the 4-digit specificity level, 16 DRB1 and 10 DQB1 alleles were common in the Zambian cohort (table 2). These major alleles accounted for >90% of the total at each locus; they also produced 25 common 2-locus haplotypes (data available on request). The allelic and genotypic frequencies closely matched those expected for Hardy-Weinberg equilibrium (P > .20). Homozygosity frequencies for the major alleles were also within their expected ranges.
Stratification for HIV-1 infection status did not reveal any overall allelic or haplotypic heterogeneity at DRB1 and DQB1 loci (P = .113–.399, global exact tests; table 2). However, distribution of DRB1*0301, DQB1*0604, DRB1*1302-DQB1*0604, and DRB1*1503-DQB1*0602 differed betweenHIV-1-seropositiveand HIV-1-seronegative groups, in comparisons of individual alleles and haplotypes (P = .005–.042; table 2) and/or in univariate analyses of population (marker) frequencies (P =.004–.084; table 3). DRB1*0301 was found exclusively on the DRB1*0301-DQB1*0201 haplotype, but DQB1*0201 by itself showed similar frequencies in seropositive and seronegative groups (table 2). Differential distribution of these individual variants was not due to age group or sex (adjusted P =.003–.065; table 3).
Associations of DRB1 and DQB1 variants with transmission of HIV-1 from initially seropositive partners. In a comparison of transmitting couples (n = 124, excluding 17 unlinked) and nontransmitting couples (n = 151), initially seropositive index partners carrying the DRB1*1301 (n = 32) or the DRB1*1301-DQB1*0501 (n = 19) haplotype were less likely to transmit HIV-1 to their seronegative partners during the 7-year follow-up period (RH, 0.48; P = .034–.082; figure 1). The effects of DRB1*1301 remained stable (RH, 0.50; P = .044) when age, sex, and donor virus load (RH, 1.9; P < .0001) were treated as covariates. However, DRB1*1301 had no appreciable effect on log10 HIV-1 load in the seropositive partners, compared with other DRB1 alleles (P > .50, t test).
Associations of DRB1 andDQB1 variants with seroconversion, in initially seronegative partners. Among the initially seronegative partners, accelerated seroconversion was associated with 2 major haplotypes—that is, DRB1*1302-D QB1*0609 (n = 7; RHp3.03; P = .017; table 4) and DRB1*1503-DQB1*0602 (n = 60; RH, 1.68; P = .009; figure 2 and table 4). Adjustment for other host factors—including age, sex, and virus load—in seropositive index partners did not materially alter these relationships (RH, 4.09 and 1.44; P = .079 and P = .04, respectively), whereas the moremodest effect of DRB1*0301-DQB1*0201 on HIV-1 seroconversion was diminished after these adjustments (RH, 1.27–1.54; P = .046–.289).
DRB1 and DQB1 homozygosity and DRB1 lineages, in relation to transmission of HIV-1 and seroconversion. Homozygosity at the DRB1 and DQB1 loci was found in 11.3% and 19.9% of HIV-1 seronegative patients and in 9.7% and 16.6% of seropositive patients, respectively (P > .35). Homozygosity frequencies were similar between HIV-1-transmitting couples and HIV-1-nontransmitting couples (P > .50) and between HIV-1 seroconverters and persistently seronegative patients (P > .50). Analyses of DRB1 alleles alternatively classified into 5 known lineages (DR1, DR51, DR52, DR53, and DR8) revealed no trends toward association with transmission of HIV-1 or seroconversion. For example, DRB1*04, DRB1*07, and DRB1*09 alleles define the DR53 group, which was equally common in the HIV-1-seropositive group and the HIV-1-seronegative group (table 2). Likewise, the DR51 group, comprising DRB1*15 and DRB1*16 alleles, showed no separate association with transmission or seroconversion and did not better explain the effect of DRB1*1503.
DRB1 and DQB1 allele sharing, in relation to transmission of HIV-1 and seroconversion. In contrast to the clear effects of HLA class I sharing on transmission of HIV-1, in ZUHRP  and another African cohort , proportions (30%–42%) of couples who shared DRB1 or DQB1 alleles did not differ by HIV-1-transmission status (P > .50; data not shown). Likewise, sharing of DRB1 lineages was as common in transmitting couples as in nontransmitting couples (P > .50; data not shown). Ongoing analyses of extended HLA class I and class II haplotypes might yield more definitive findings.
Multivariable models for within-couple transmission of HIV-1. Several models were used to assess the simultaneous effects of multiple host and viral factors on within-couple HIV-1-transmission status (table 5). For seropositive partners, male sex (RH, 2.27–2.86; P < .0001) and elevated virus load (RH, 1.52–1.55; P < .01) were the significant cofactors for HIV-1 transmission. Genital ulcer/inflammation in both partners, whether initially seropositive (RH, 3.04–3.23; P < .0001) or seronegative (RH, 3.11–3.74; P < .0001), also served as a strong predisposing factor for HIV infection. With statistical adjustment for other genetic factors (including HLA class I and CCR2-CCR5 variants [M. T. Dorak, J. Tang, and R. A. Kaslow, unpublished data]), HLA class II variants (table 4) showing marginally significant (univariate P < .10) associations with transmission of HIV-1 (in seropositive partners) or seroconversion (in seronegative partners), for the most part, remained contributing factors in the respective models. In particular, DRB1*1301 in seropositive partners (RH, 0.16–0.18; P < .005) and DRB1*1302-DQB1*0609 in seronegative partners (adjusted RH, 2.69–3.85; P < .05) retained their relationships to virus transmission. The effect of DRB1*1503-DQB1*0602 was no longer statistically significant in the full model (model 1; RH, 1.41; P = .124), but the RH value remained similar to that observed in univariate analyses (table 4). Excluding DRB1*1503-DQB1*0602 from the reduced model (model 2) did not substantially alter the strength of other associations.
In further multivariable testing, increased duration of cohabitation reported at enrollment was accompanied by a reduction in subsequent transmission of HIV-1 in this cohort (RH, 0.95/year; adjusted P < .05). This effect was largely attributed to apparent confounding by other factors, especially the index partner's age, which correlated with both duration of cohabitation (Spearman r = 0.54; P < .0001) and age of the seronegative partner (Spearman r = 0.27; P < .0001). Alternatively, age could replace duration of cohabitation as another modest factor in the multivariable model (model 3; table 5)—that is, either younger age (RH, 1.44; adjusted P = .174) or shorter duration of cohabitation at enrollment (baseline) could be treated as a risk factor for within-couple transmission of HIV-1 in this cohort.
Effects of host genetic diversity on susceptibility or resistance to HIV-1 infection have been documented for several HLA loci [6, 28–31], but reported relationships have been less consistent for class II than for class I markers . The apparent protection against heterosexual HIV-1 infection by DRB1*1301 (or DRB1*1301-DQB1*0501) and DRB1*1302-DQB1*0604 here bears some similarity to the earlier observation of resistance to vertical transmission of HIV-1 in black and Hispanic infants with the DRB1*13 allele . The deleterious effect of DRB1*0301, found exclusively on the DRB1*0301-DQB1*0201 haplotype in Zambians, also resembled that seen in white individuals . On the other hand, the effects of DRB1*1503-DQB1*0602 contrasted with those of DRB1*1501, an allele commonly found on the DRB1*1501-DQB1*0602 haplotype in white individuals . For the rare haplotype DRB1*1302-DQB1*0609, its persistent and rather independent association with increased risk for HIV-1 seroconversion might indicate the influence of another variant in the neighboring loci, especially since the closely related haplotype DRB1*1302-DQB1*0604 had the opposite effect. Overall, these class II associations identified in Zambians could not be readily related to earlier, more-limited reports, in the context of heterosexual transmission of HIV-1.
Distinctive findings for HLA alleles and haplotypes more common in individuals of African ancestry, rather than those distributed more widely, should not be surprising for several reasons. First, HLA relationships may simply reflect the genetic background of the study population, the viruses circulating in it at the time of study, or other cohort characteristics. Indeed, several established genetic associations have been recognized more readily in selected, rather than all, ethnic groups studied, because of different marker frequencies or strength of relationships [33–35]. Nonetheless, multivariable analysis has reinforced our confidence that neither other known genetic factors (i.e., HLA class I or CCR2-CCR5 genotypes) nor nongenetic factors (i.e., sexual behavior, genital ulcer, and donor virus load) could entirely account for the association of HLA class II with transmission or acquisition of HIV-1. Second, application of Kaplan-Meier plots to the analyses of serodiscordant couples followed prospectively might be expected to reveal associations (table 4) not detectable in prior cross-sectional comparisons (table 3), especially since the mechanism by which HLA mediates infection may be better measured as a function of exposure time rather than as a cumulative phenomenon . Third, well-established and statistically significant associations of host genotypes with HIV-1 infection are often modest in magnitude, with estimates of RHs or relative odds only occasionally >2-fold [6, 30, 34, 36]. Such a pattern most likely implies either findings by chance or the involvement of multiple host factors acting in concert to mediate HIV-1 infection, as has already been shown for progression of HIV-1 disease [37–41].
HLA genes participating in adaptive immunity are the suspected determinants of numerous autoimmune and infectious diseases [42—44]. Conversely, HLA allelic and haplotypic diversity in human populations is apparently driven by a variety of old and new diseases [45–50]. For transmission of HIV-1, the mechanisms underlying HLA class II associations are not clear. Detection of HIV-1-specific antibodies and cytotoxic T lymphocyte responses, in exposed and uninfected individuals [10, 18], does suggest a possible role for HLA-mediated adaptive immunity in virus transmission, but there is scant evidence that HIV-1-specific immune responses mediated by certain HLA molecules would confer relatively higher or lower risk for infection [18, 51]. The unfavorable effect of the DRB1*1503-DQB1*0602 haplotype could, for example, reflect its tendency to promote another infection known to predispose to acquisition of HIV-1. The common allele DRB1*15 (or DR2 by serology) appears to be a risk factor for mycobacterial infections [52, 53]. Both Mycobacterium tuberculosis and M. avium have been shown to enhance HIV-1 replication and coreceptor expression [54, 55]. Therefore, DRB1*15-positive HIV-1 seroconverters among serodiscordant Zambian couples could be subclinically infected with mycobacteria capable of accelerating HIV-1 infection. Systematic analyses of extended HLA haplotypes in common disease models may help to define the extent to which both direct and indirect mechanisms operate in the process of transmission of HIV-1.
As in any immunogenetic studies, the search for host genetic determinants of HIV-1—related outcomes is often compromised by other issues, including type I error resulting from multiple statistical tests and confounding by closely related factors. The P values reported here were not “corrected” for multiple comparisons made or implied in our multifaceted analyses, and we view these statistical test results as a useful guide for future studies. Strong covariance among host, viral, and environmental factors was also of critical concern. For example, HIV-1 load as a key predictor of transmission of HIV-1 could vary by age, sex, and HLA class I genotypes [12, 14]. Thus, further efforts to replicate various findings beyond this single Zambian cohort may need to consider not only interpopulation differences in HLA allele and haplotype diversity but also behavioral risk, circulating HIV-1 subtypes, virus transmission mechanisms, and common coinfections.
Zambia-Uab HIV Research Project: Additional Investigators
Additional investigators in the Zambia-UAB HIV Research Project include the following: Elwyn Chomba and Alan Haworth, for epidemiology and clinical studies, and Francis Kasolo and Isaac Zulu, for virology (University of Zambia School of Medicine); Beatrice Hahn, Ulgen Fideli, and Eric Hunter, for virology, and Steffanie Sabbaj and Mark Mulligan, for immunology (University of Alabama at Birmingham); and Marylyn Addo, Marcus Altfeld, and Bruce Walker, for immunology (Harvard University).
We are indebted to C. A. Rivers and D. Munfus, for assistance with HLA genotyping; G. Cloud, R. Izurieta, and C. Flanigan, for data management; and S. Tang and A. Westfall, for help with SAS programming.