Fine-mapping classical HLA variation associated with durable host control of HIV-1 infection in African Americans

Abstract

A small proportion of human immunodeficiency virus-1 (HIV-1) infected individuals, termed HIV-1 controllers, suppress viral replication to very low levels in the absence of therapy. Genetic investigations of this phenotype have strongly implicated variation in the class I major histocompatibility complex (MHC) region as key to HIV-1 control. We collected sequence-based classical class I HLA genotypes at 4-digit resolution in HIV-1-infected African American controllers and progressors (n = 1107), and tested them for association with host control using genome-wide single nucleotide polymorphism data to account for population structure. Several classical alleles at HLA-B were associated with host control, including B*57:03 [odds ratio (OR) = 5.1; P= 3.4 × 10(-18)] and B*81:01 (OR = 4.8; P= 1.3 × 10(-9)). Analysis of variable amino acid positions demonstrates that HLA-B position 97 is the most significant association with host control in African Americans (omnibus P = 1.2 × 10(-21)) and explains the signal of several HLA-B alleles, including B*57:03. Within HLA-B, we also identified independent effects at position 116 (omnibus P= 2.8 × 10(-15)) in the canonical F pocket, position 63 in the B pocket (P= 1.5 × 10(-3)) and the non-pocket position 245 (P= 8.8 × 10(-10)), which is thought to influence CD8-binding kinetics. Adjusting for these HLA-B effects, there is evidence for residual association in the MHC region. These results underscore the key role of HLA-B in affecting HIV-1 replication, likely through the molecular interaction between HLA-B and viral peptides presented by infected cells, and suggest that sites outside the peptide-binding pocket also influence HIV-1 control.

Figure 1.

Comparison of sequence-based and imputed classical HLA allele association results. For each HLA allele called by both sequencing and imputation, the point estimate of the beta (A–C) and Z-score (D–F) from logistic regression models using sequence types (x-axis) or imputed dosages (y-axis) is plotted. Pearson r² is given for each comparison and the red line indicates a perfect correlation.

Figure 2.

Association results (−log₁₀ P-value) for all variable amino acid positions in HLA-B. African American HIV-1 controllers were compared with progressors at each position using logistic regression including covariates to correct for populations structure. For amino acid positions accommodating more than two possible alleles (such as position 97), the multi-degree of freedom omnibus test P-value is shown. Positions are colored according to canonical pockets of the peptide-binding groove. Classical 4-digit HLA-B allele association results are shown for comparison on the left.

Figure 3.

Frequency and effect sizes for key HLA-B amino acids in HIV-1-infected African Americans. (A) Allele frequency differences between controllers (in orange) and progressors (in blue) for amino acids at positions 63, 97, 116 and 245 in HLA-B. Numbers above bars are ORs where OR >1 indicates a protective effect. (B) Beta estimates of alleles at positions 63, 97, 116 and 245 in the DoD HIV NHS sample. Asn97 was not observed in this sample. The beta estimates are from linear regression models including covariates and are given in log₁₀ units of virus load set point.

Figure 4.

Three-dimensional structure of HLA-B highlighting key amino acid positions. (A) The HLA-B protein based on Protein Data Bank entry 2BVP looking in to the peptide-binding groove (44). The independently associated pocket positions 63, 97 and 116 are highlighted. (B) Side view of the HLA-B molecule (blue) interacting with CD8 (green) based on Protein Data Bank entry 1AKJ with position 245 highlighted (45). Variation at position 245 has been shown to influence the binding kinetics between the CD8α subunit and the HLA molecule (32,33). The figure was prepared with UCSF Chimera (46).