Identifying predictors of time-inhomogeneous viral evolutionary processes

Various factors determine the rate at whichmutations are generated and fixed in viral genomes. Viral evolutionary ratesmay vary over the course of a single persistent infection and can reflect changes in replication rates and selective dynamics. Dedicated statistical inference approaches are required to understand how the complex interplay of these processes shapes the genetic diversity and divergence in viral populations. Although evolutionarymodels accommodating a high degree of complexity can now be formalized, adequately informing thesemodels by potentially sparse data, and assessing the association of the resulting estimates with external predictors, remains amajor challenge. In this article, we present a novel Bayesian evolutionary inferencemethod, which integratesmultiple potential predictors and tests their association with variation in the absolute rates of synonymous and non-synonymous substitutions along the evolutionary history.We consider clinical and virologicalmeasures as predictors, but also changes in population size trajectories that are simultaneously inferred using coalescentmodelling. We demonstrate the potential of ourmethod in an application to within-host HIV-1 sequence data sampled throughout the infection ofmultiple patients.While analyses of individual patient populations lack statistical power, we detect significant evidence for an abrupt drop in non-synonymous rates in late stage infection and amore gradual increase in synonymous rates over the course of infection in a joint analysis across all patients. The former is predicted by the immune relaxation hypothesis while the lattermay be in line with increasing replicative fitness during the asymptomatic stage.