Research Interests

I work on the evolutionary epidemiology of HIV and particularly the evolution of virulence. We showed that virulence, as measured by the set-point viral load (the stable value of viraemia in asymptomatic infection) is significantly determined by viral genetic factors. We also showed that the average set-point viral load in a well-studied cohort in Uganda declined over the last 20 years, a finding that can be explained by viral adaptation to a low fitness optimum. We now study global patterns of virulence evolution across many cohorts, and develop methods to find the genomic changes underlying the trends in virulence.

I am also interested in the evolution of antibiotic resistance in bacterial species such as Streptococcus pneumoniae and Escherichia coli. Surveillance data shows that average levels of resistance are often stable over decades. Resistances that have emerged more recently, such as third-generation cephalosporin resistance in E. coli, also tend to stabilise (around 5 to 15% in European countries). The reasons why resistance genes or alleles stabilise at an intermediate level, instead of going to fixation, are unclear: this has been called the "coexistence problem". Using very finely resolved temporal data on levels of antibiotic consumption and resistance in Israel, we show that rapid seasonal changes in resistance, at timescales of a few months to years, can be predicted, and that indeed a strong force stabilises resistance to an intermediate level. We are now developing new mathematical models to better describe resistance evolution in a structured host population.

The dynamics of adaptation crucially depends on the relationship between genotypes and fitness: the fitness landscape. The fitness landscape determines the genetic variation that is available for selection. We studied theoretically Fisher's Geometric model, a fitness landscape model. We developed theoretical predictions for the distribution of fitness effects, the epistasis coefficients, and higher order interactions between mutations, under Fisher’s Geometric Model. We found Fisher’s model is sufficiently flexible to generate a great diversity of genotypic landscapes, from smooth to rugged, and in particular reproduces patterns of epistasis and higher order interactions observed in several experimental landscapes. By analysing a large number of published experimental fitness landscapes, we found Fisher’s Geometric model was able to explain several statistical properties of the landscapes, but it was often unable to explain the full structure of fitness landscapes and the fitness of specific combinations of mutations.