HIV cross-sectional incidence assays often measure characteristics of the antibody (Ab) response, such as titer, class, and avidity. These responses are affected by various factors including viral suppression. Since older guidelines recommended ART initiation later in infection, incidence algorithms often use viral suppression as a surrogate for non-recent infection. As new guidelines recommend ART initiation early in infection, viral suppression no longer indicates non-recent infection, and new approaches are needed to identify recent infections.
We analyzed Ab profiles for 258 samples from 57 individuals with HIV subtype C infection and known duration of infection (2 mo. – 8.7 yrs.) using phage immunoprecipitation sequencing (PhIP-Seq). PhIP-Seq quantifies Ab binding to 3384 peptides spanning the HIV genome. Our novel classifier for recent (2-6 mo.) and non-recent (18+ mo.) infection is based on the k-top scoring pairs (TSP) classifier. For each peptide pair, relative Ab abundances classify each sample as recent or non-recent. Overall sample classification is determined by a majority voting system. Optimal classification cutoffs and the number of voting pairs were identified using a training set of 176 samples from 38 individuals and subsequently tested on the remaining 82 samples from 19 individuals. We compared these results to results from a standard Limiting Antigen Avidity (LAg-Avidity) protocol.
In the final model with 4 voting pairs, 79% (71/90) recent samples and all 168 of non-recent samples were correctly classified. In contrast, the LAg-Avidity protocol classified 43% (35/81) recent samples and all 165 non-recent samples correctly. Comparison of TSP vs. LAg-Avidity showed that the TSP approach captured a greater proportion of recent infections with a mean window period of 217 days (95% CI: 183-257 days; see figure). In contrast, the mean window period of the LAg-Avidity protocol is 106 days (95% CI: 76-146 days).
We identified four 4 peptide pairs that outperformed the standard LAg-Avidity protocol for identifying samples from individuals with recent HIV infection. These peptides can be incorporated into a simple assay for field use. With a larger mean window period, the TSP classifier yields more precise incidence estimates without relying on viral load to identify non-recent samples. The TSP approach can also easily be applied to other populations and virus subtypes to identify novel peptide signatures for recent infection.