In our study of clade B HIV-1 infections among African Americans, HLA-B
but not HLA-A
supertypes were associated with disease control. Our analysis further demonstrated the importance of individual alleles at the B locus compared with those at the A locus. Associations of B7s and B44s were consistently disadvantageous for disease control overall and separately for VL and CD4 count. The latter are the most prevalent HLA-B
supertypes in both African Americans and Caucasians (34
). Our findings corroborated the few previous studies on the effects of HLA-B
supertypes broadly in Caucasians (34
) and of B44s specifically in native Africans (28
). The consistently favorable overall associations of B58s with disease control and with VL and the favorable association of B62s also agreed with published findings.
Associations between supertypes and HIV-1 outcomes have been demonstrated in previous studies of Caucasians with clade B infection (34
) and of native Africans with clade A or C infection (23
). Caucasians in one study showed an unfavorable effect of B7s and a favorable effect of B58s on HIV-1 control (44
); in another study, Caucasians with the relatively infrequent B58s, B27s, and B62s supertypes showed enhanced CTL responses, while the more frequent B7s and B44s supertypes were associated with diminished CTL responses (34
). Among native Africans, carriers of A2s, A24s, B27s, and B58s, but not B44s, had lower VLs and more vigorous CTL responses to total Gag, Gag p24, and Nef (28
Whether the effect of a particular supertype is driven by one, several, or all of the individual class I alleles currently assigned to that supertype has received little attention. In one population of European ancestry (44
), the advantage of B27s alleles persisted after excluding carriers with favorable B*27 alleles and the disadvantage of B7s alleles persisted after removing unfavorable B*54-56 (6
) and B*35 (12
) alleles. These findings suggested analogous outcomes associated with the seemingly cross-reacting alleles within each particular supertype. A large proportion of our African Americans with B7s remained “noncontrollers” of disease, even after exclusion of participants with B*35-Px and B*35-Py, both of which have appeared disadvantageous in at least one other HIV-1 clade B-infected North American population besides ours (3
The pattern of association with B7s described above has an exception: B*81 appeared to predispose individuals to more favorable HIV-1 outcomes here and elsewhere (8
As a largely African allele, B*81 was only recently assigned to B7s (39
), based on predictions from its three-dimensional protein structure of a preference for proline at position 2 of the binding epitope (7
). Important functional differences between various B7s alleles that present cross-reacting epitopes were previously detected among clade C-infected Africans (24
). Any of several mechanisms may explain this within-supertype allelic heterogeneity, including frequency, magnitude, immunodominance, or avidity of allele-specific CTL responses, as well as distinct T-cell receptor (TCR) usage by the individual alleles. In this study, B*81 accounted for only 6% of B7s alleles, and overall the remainder were disadvantageous.
As a whole, B44s (largely represented by B*44, B*45, and B*49) also appeared to be disadvantageous for disease control. All individual B44s alleles tended toward an association with poorer control (data not shown). These findings have extended the recognition of B44s as disadvantageous for multiple human ethnic groups and viral clades.
In contrast, the strongly protective B62s supertype was the least prevalent in both cohorts. The latter supertype includes B*13, an allele implicated repeatedly in protection against HIV-1 (4
). Interestingly, only 2 of 14 African Americans with B62s carried B*13 in our study; 9 carried B*52, an allele not previously associated with prognosis of HIV-1 infection. Since both B*52 and B*13 appear to share an affinity for glutamine at position 2 of binding epitopes, B*52 may successfully target an immunodominant CTL epitope(s) that cross-reacts with B*13 (e.g., epitope R19 in Nef) (18
). The low B62s population frequency, perhaps along with protective Bw4-80Ile specificity (10
) of its two common members, may have contributed to its favorable effect.
The observed relationships between supertype population frequencies and HIV-1 outcomes in the REACH and HERS cohorts separately and in aggregate are consistent with earlier findings (34
). On the other hand, infrequent individual HLA-B
alleles accounted in aggregate for only a modest portion of explainable VL variation across supertypes at the genotypic level; the magnitude of this effect was similar to that seen with more common individual HLA
alleles. Because any advantage conferred by infrequent individual HLA-B
alleles likely reflects past negative selective pressure by more than just a single pathogen, infrequent HLA-B
alleles might not be expected to exert great control of HIV-1 disease. Despite this lower magnitude of effect on HIV-1 outcomes, infrequent class I alleles remained consistent in their association with lower VLs (−0.26 Δlog10
< 0.005) in the multivariable analysis restricted to the carriers of the two most frequent HLA-B
supertypes (i.e., B7s and B44s).
The disproportionate contribution of a single B58s allele (B*57) to the VL heterogeneity among genotypic supertypes highlights the uniqueness of B*57 alleles in conferring highly effective natural immunity to HIV-1. In our study population, the number of carriers of the African B*5703 allele (n
= 28) outnumbered those with the Caucasian B*5701 allele (n
= 8), but the protective associations were quite similar. In additional exploratory analysis, the trend toward an association between B*1516 and -17 and lower VLs was in the direction predicted by other work (11
). Our earlier investigation of B58s alleles among clade A- and C-infected native Africans demonstrated contrasting effects of B*57 and B*58 alleles on HIV-1 control (23
). Here we found no appreciable effect of either B*5802, established as unfavorable, or B*5801, established as favorable.
Although B*27 alleles have repeatedly been associated with favorable outcomes (12
), B27s did not show a clear trend toward an association in either direction, for two reasons: B*27 alleles are relatively rare in individuals of African ancestry, and the effects of two other major B27s alleles (B*14 and B*1510) offset each other. Alleles of B*14, with or without its closely linked Cw*08 allele, have been associated with protection against HIV-1 (19
). In addition, B*14 appears to be more frequent among subjects with immunodominant CTL responses toward Gag p24 epitope SG20 (15
). B*1510 occurs more frequently among Africans than among Caucasians, but no prior epidemiologic association has been reported for this allele in the context of HIV-1. Our multilevel model (Table ) also reflected a mixed picture for B27s; genotypic supertypes containing it accounted for a modest 10% increase in viral load heterogeneity. Our results support the suggested reallocation of B*1510 from B27s into the novel B39 supertype (25
), which would leave the remaining B27s alleles showing uniformly favorable effects.
Our study had limitations. With 338 subjects, it was underpowered to study alleles at frequencies of <2%. Despite the sample size, it was the more frequent HLA-A supertypes that failed to show any evidence of association with disease control. Although the apparent heterogeneity of our two subcohorts (e.g., in median VL) might have precluded combining them, their close resemblance in the distribution of HLA markers (all with P values of >0.1) and in their median values for CD4 cell counts justified doing so. Moreover, the observed population differences in median VL were most likely due to the distinct methods for quantifying HIV-1 plasma RNA levels. Nevertheless, we further adjusted our analysis for individual cohort membership.
Our study was the largest to date to examine the relationship between HLA class I supertypes and HIV-1 outcomes in African Americans, the first to include a systematic assessment of the effects of individual class I alleles within each of the supertypes, and the first to analyze the contributions of individual class I alleles and their population frequencies to the differences in VLs among genotypic supertypes. Multilevel random-coefficient regression enabled us to address the hierarchical structure of the data in a flexible framework for analyzing the heterogeneity of VL, as predicted by both the HLA class I alleles (i.e., individual level) and their supertypes (i.e., group level).
The consistent demonstration of complementary influences of HLA-B supertypes and individual HLA-B alleles may focus attention on functional convergence in HLA polymorphism. Strong, consistent evidence for supertype effects could guide the development of multiepitope-based HIV-1 vaccines. Finally, the techniques introduced here to analyze the hierarchical relationships of HLA supertypes and their component alleles should be applicable to future studies of HLA in other infectious diseases.