Background:  Although HIV-1 protease, reverse transcriptase (RT), and integrase are relatively conserved, all three genes show natural polymorphisms. We present an approach which allows us to detect selection at single amino acid sites, which we have applied to large sequence datasets to test whether selection is responsible for natural polymorphisms in the pol gene.

Methods:  Subtype B protease (n = 1094) and RT (n = 713) sequences isolated from antiretroviral-naïve patients were downloaded from the Stanford HIV Drug Resistance database. Subtype B integrase sequences (n = 261) were extracted from the Los Alamos HIV Sequence database. The number of nonsynonymous and synonymous substitutions that occurred at each site over the evolutionary history of the sample, dN and dS, were estimated by reconstructing ancestral sequences using maximum likelihood. Positive selection at a site was determined by testing whether dN>dS. We tested for different dN and dS at sites identified as secondary resistance mutation, at sites identified within cytotoxic T cell epitopes, and at sites in which amino acid mixtures were present.

Results:  Although all 3 genes were, on average, relatively conserved (dN/dS<1; protease = 0.2, RT = 0.1, integrase = 0.2), there was significant variation across sites both in dN (coefficient of variation, CV:  protease = 2.2, RT=2.6, integrase = 1.8), and dS (protease = 1.0, RT = 0.78, integrase = 1.2). Sites under positive selection (p <0.05) were detected in protease (12/99), RT (9/222), and integrase (5/288). There was no correlation between dN and identified epitopes. Secondary resistance mutations had higher dN (Wilcoxon test, p <0.001) and dS (p = 0.06) in protease but not RT. There was a positive correlation between dN (but not dS) and the number of amino acid mixtures at a site in both protease (Spearman's r = 0.73) and RT ( r = 0.65) (p <0.0001).

Conclusions:  Many natural polymorphisms in the pol gene at the population level are driven by positive selection, which may influence the outcome of protease-based therapy. The lack of overlap between selected sites and known CTL epitopes may reflect the use of peptide pools derived from heterologous rather than autologous virus. The correlation between selection pressure between hosts and amino acid mixtures, reflecting within-patient diversity, implies continuous viral evolution throughout infection rather than evolution solely during the transmission event. Incorporation of host HLA haplotype into these analyses may reveal more sites under selection.

Keywords: evolution; selection; resistance