Principal component analysis of general patterns of HIV-1 replicative fitness in different drug environments

To detect general patterns and temporal trends of HIV-1 resistance, we apply principal component analysis (PCA) to in vitro fitness data. Twenty-eight thousand virus samples, obtained between 2002 and 2008, were assayed for fitness in 16 to 21 selective environments. Fitness measurements are based on replication capacity (RC), which quantifies in vitro viral replication in a single cycle of infection. RC is determined both in the absence of drugs and in the presence of 6–7 nucleoside analog reverse transcriptase inhibitors (NRTIs), 3–4 non-nucleoside reverse transcriptase inhibitors (NNRTIs), and 6–9 protease inhibitors (PIs). PCA shows remarkable structure in RC across the different environments, which reveals differences in the patterns of resistance and cross-resistance between drugs or between drug classes. To probe the causes of the observed patterns, we develop a model to generate simulated data and subject these simulated data to an equivalent analysis. By comparing the outcomes of PCA on the original and the simulated data, we quantify which part of the total variance of the original data is due to non-specific effects, class-specific effects, and drug-specific effects of resistance mutations. We find that relative fitness is mainly drug-independent and that drug-specific effects are substantially bigger than class-specific effects for NRTIs, but not for NNRTIs or PIs. The observed patterns remain remarkably stable over the period of observation. Comparison with known potent combination therapies suggests that PCA helps to identify combinations that act synergistically in preventing the emergence of resistance.