Unifying the molecular evolution and host population dynamics of rapidly evolving pathogens

Abstract

In light of evolution, we can make sense of the most incredible phenomena such as the mere existence of humankind. Among the smallest organisms, however, are the ones that evolve most rapidly, such that the rate at which they evolve can be directly measured from nucleotide sequences sampled sequentially over time. Their evolutionary dynamics shed light not only on their own phylogenetic history, but also on the evolution and population dynamics of their hosts. Many factors influence and interact with the processes that shape the viral genome. This thesis focuses on the development of phylodynamic methods that unify the molecular evolution and host population dynamics of rapidly evolving pathogens. In the Bayesian Markov chain Monte Carlo framework Beast2, three approaches are developed that infer evolutionary and epidemiological dynamics from sequence data. First, the birth– death skyline model provides insight into epidemiological dynamics without specification of an epidemiological model for the host population. Applied to an HIV epidemic sampled in the United Kingdom, the method indicates epidemic decline in the mid-1990s, which may be due to the introduction of highly active retroviral treatment. The analysis of the hepatitis C virus from Egypt supports the hypothesis that the Egyptian epidemic has been impelled, if not caused, by antischistosomal injection campaigns around 1920. Second, the birth–death SIR method employs the birth–death skyline likelihood to explicitly model the interaction between viral evolution and epidemiological host population dynamics. A compartmental SIR model of infection and recovery is incorporated in the phylogenetic inference and enables the reconstruction of incidence and prevalence. Phylodynamic analysis of five HIV clusters reflects the different stages at which each local epidemic was sampled. Third, the multi-type birth–death model incorporates the underlying population structure, such as geographic division, into phylogenetic analysis by allowing migration events between discrete locations. The analysis of a human influenza virus dataset from Australia and New Zealand demonstrates the applicability of this phylogeographic approach. Each of the approaches is based on stochastic birth–death–sampling processes. Through explicit modelling of the sampling process they provide a powerful basis for phylodynamic inference from contemporaneously and sequentially sampled sequence data.