We conducted high-resolution HLA class I and II genotyping in a large, long-term prospective cohort of HIV-seropositive women, many of whom were enrolled in 1994, prior to the widespread use of HAART. This design allowed us to study HLA associations with HIV disease both before and after the introduction of HAART in a single population. Our results in women before they initiated HAART confirmed many of the previously reported associations between HLA and HIV disease in untreated patients, showing that our population was not dissimilar from those reported in previous studies of patients not using HAART. However, in treated patients, we observed few HLA associations. In fact, we detected only three significant HLA associations with virological response to HAART, each of them opposite to those that would be predicted based on prior results from untreated patients.
Specifically, B*57:01, B*58:01, and the Bw4-80I group were strongly associated with failure to control HIV replication following HAART initiation. Similar associations between these alleles and immunological response to HAART were also observed, although these relationships were nonsignificant, possibly because we had less extensive data to examine this endpoint. That is, the immunological response was only assessed among the subset of 176 women who were using effective HAART (i.e., those who had a virological response). The larger Bw4 allele group, which includes the Bw4-80I group (30
) and other HLA-B alleles with the Bw4 serological epitope (22
), also showed inverse associations with virological and immunological responses to HAART. In exploring these results further, however, we found that associations of B*57:01 and B*58:01 might have accounted for the associations of Bw4 and Bw4-80I with virological response. In contrast, though, we found no evidence that B*57:01 and B*58:01 accounted for the association of Bw4 with immunological response. We cannot therefore exclude the possibility that epistatic interactions between HLA and killer immunoglobulin-like receptor (KIR) genes may influence the response to HAART, as Bw4 alleles can act as a ligand for KIR.
We are aware of only three prior studies of HLA genotype and HIV disease in treated patients (2
). As in the current investigation, few associations between HLA and HIV disease were observed, and in two of these studies, B*57 was associated with poor CD4+
recovery following HAART initiation (2
). Bw4 homozygosity was also associated with poor CD4+
recovery in one of these studies (35
The current investigation was the first to examine the relation of HLA genotype with the long-term risk of AIDS events in treated patients and to examine HLA alleles other than those already reported to be significant in untreated patients. The cohort involved extensive person-years of observation and included a substantial number of AIDS events. Thus, if a relationship between HLA genotype and risk of AIDS-defining conditions in treated patients had been present, we would likely have detected it, but no such associations were found.
Overall, it increasingly appears that most of the HLA associations with HIV disease in untreated patients are not observed in treated individuals, and in fact, some of these relationships may be opposite one another. The reasons there may be different HLA genotype associations in treated versus untreated HIV disease progression are unknown. One possible explanation for the limited number of HLA associations in treated patients may reflect a reduced role for immune genes in the inhibition of HIV replication, because HAART has very strong antiviral effects regardless of a patient's HLA genotype. Second, HAART exerts strong selective pressure on the HIV quasispecies, and HIV variants associated with drug resistance, largely in the pol
), are common in patients exposed to antiretroviral therapies (17
). CTLs recognize epitopes in gag
, and pol
), so it is possible that therapy-associated selection pressures on pol
could change the distribution of HIV antigens presented to T cells, thereby altering established HLA associations. Even if this hypothesis is correct, though, it remains unclear why the few HLA associations in HAART users that we and others observed involved specific HLA genotypes that are highly protective in HAART-naïve patients but are high risk in treated patients.
Important limitations of this study must also be considered in the interpretation of the findings. First, we must consider the possibility that a bias related to study design explains the observed inverse associations. While our analyses adjusted for cofactors associated with both HLA genotype and HIV disease progression (i.e., conventional confounders), we cannot exclude the possibility that other factors unrelated to HLA could also have influenced the results through less conventional pathways (e.g., common effects) (see a recent review of this topic [20
]). It is also possible that women with HLA genotypes associated with slow untreated disease progression had delayed HAART initiation compared to other women; that is, women with protective HLA genotypes may have been HIV infected for longer periods of time prior to HAART initiation. While we controlled for the CD4+
count prior to HAART initiation, it is possible that other, independent factors associated with long-term HIV infection (e.g., chronic immune activation) may explain the observed inverse associations between protective HLA genotype and response to HAART. As in a prior study (35
), though, we could not directly examine this issue because the dates of HIV seroconversion for the vast majority of WIHS women are unknown. Cohorts in which duration of HIV infection is known or can be accurately estimated would be best suited to address this issue. One might additionally ask whether hypersensitivity to abacavir could explain why some patients with the B*57:01 genotype (and the Bw4-80I genotype, which includes B*57:01) do not respond to HAART (29
). In the current cohort, however, only three women with B*57:01 received abacavir during the time period studied, and exclusion of these women from the data set did not meaningfully change the results (data not shown).
Another important consideration in the interpretation of these results is the impact of survival bias. That is, women with rapid HIV disease progression may be underrepresented in the WIHS cohort because they did not live long enough to be enrolled and initiate HAART. The impact of this bias is that HLA genotypes that predispose to rapid disease progression with too little prevalence for analysis may have been underrepresented (we studied only HLA alleles with >3% prevalence). While this may have limited our ability to detect associations with certain alleles related to rapid progression, we note that all of the HLA alleles found to be strongly associated with HIV progression in prior studies were included in our analysis, having been found in >3% of women in our cohort. We also note that AIDS-defining conditions in WIHS are ascertained by participant self-report, and so it is possible that there was some misclassification in AIDS diagnoses, which may have attenuated the statistical power for our analyses of AIDS events. Even if this were the case, it is unlikely that misclassification completely explains the lack of HLA associations with AIDS in HAART users, since there were many AIDS events and a long duration of follow-up, as described above.
The current findings regarding HLA and HAART predominantly reflect data on HAART initiation from the late 1990s. HAART regimens have improved in recent years, and we cannot exclude the possibility that the results would be different in women using current regimens. Furthermore, most (77%) HAART initiators had used nucleoside reverse transcriptase inhibitor (NRTI) monotherapy or combination therapy prior to initiating HAART, and thus, the prevalence of NRTI resistance mutations may have been higher in the studied HAART initiators than would be expected in an antiretroviral naïve population. To address this issue, we controlled for prior antiretroviral use in additional sensitivity analyses of virological and immunological responses (as described above), but inclusion of this additional covariate did not meaningfully change the results (data not shown). Our study was also necessarily limited by the number of different analytic approaches that could be presented for outcomes for which there is no universally accepted definition (e.g., virological response to HAART). In the current study, for example, we a priori chose two definitions that we and others have used for virological response and found the results to be equivalent. Lastly, we could not conduct extensive subset analyses given the modest prevalences of many alleles.
In conclusion, the paucity of HLA associations with HIV disease progression in treated patients may reflect the effectiveness of HAART in suppressing viral replication irrespective of host genotype. However, virological and immunological nonresponses to HAART are not uncommon, and further study is warranted to determine the relation of this with HLA genotype and other host factors. Understanding why several alleles (notably alleles that are protective against HIV disease progression in untreated women) are associated with a greater risk of virological and immunological nonresponses to HAART could provide new insights into this important clinical issue.