Strength and tempo of selection revealed in viral gene genealogies

Abstract

Background: RNA viruses evolve extremely quickly, allowing them to rapidly adapt to new environmental conditions. Viral pathogens, such as influenza virus, exploit this capacity for evolutionary change to persist within the human population despite substantial immune pressure. Understanding the process of adaptation in these viral systems is essential to our efforts to combat infectious disease.

Results: Through analysis of simulated populations and sequence data from influenza A (H3N2) and measles virus, we show how phylogenetic and population genetic techniques can be used to assess the strength and temporal pattern of adaptive evolution. The action of natural selection affects the shape of the genealogical tree connecting members of an evolving population, causing deviations from the neutral expectation. The magnitude and distribution of these deviations lends insight into the historical pattern of evolution and adaptation in the viral population. We quantify the degree of ongoing adaptation in influenza and measles virus through comparison of census population size and effective population size inferred from genealogical patterns, finding a 60-fold greater deviation in influenza than in measles. We also examine the tempo of adaptation in influenza, finding evidence for both continuous and episodic change.

Conclusions: Our results have important consequences for understanding the epidemiological and evolutionary dynamics of the influenza virus. Additionally, these general techniques may prove useful to assess the strength and pattern of adaptive evolution in a variety of evolving systems. They are especially powerful when assessing selection in fast-evolving populations, where temporal patterns become highly visible.

Figure 1

Genealogical trees derived from hemagglutinin sequences of influenza A (H3N2) and measles virus. The influenza tree, sampled between 1968 and 2008, appears long and spindly with a distinct lack of deep branches. It only takes a few years for contemporaneous strains to find a common ancestor. The measles tree, sampled between 1979 and 2009, looks very different, harboring many deep branches. It takes approximately 50 years for contemporaneous measles strains to find a common ancestor.

Figure 2

Genealogical trees derived from simulated sequences at varying population sizes. In each case, sequence evolution was simulated using a neutral Wright-Fisher model, and approximately 200 samples were taken over the course of 10,000 generations (sampled at a Poisson rate of 0.02 per generation). In each frame, the x-axis is shown as intervals of 5000 generations, so that the temporal scale increases with N. As population size N increases, the time it takes for two lineages to coalesce increases proportionally, so that larger populations show a deeper most recent common ancestor. Population dynamics result in rapid coalescence when there are many contemporaneous lineages and slower coalescence when there are few, a detail especially apparent at larger population sizes.

Figure 3

Genealogical tree and skyline plots of coalescent statistics for sequences simulated under selective neutrality. There were 10,000 individuals in the simulation. The tree shown represents the maximum posterior tree sampled over the course of the MCMC analysis of the simulated data. For coalescent statistics, TMRCA, diversity and N_eτ, solid lines represent mean values and gray outlines represent 95% credible intervals across MCMC replicates. TMRCA, diversity and N_eτ are measured in units of generations.

Figure 4

Genealogical tree and skyline plots of coalescent statistics for sequences simulated under purifying selection. There were 10,000 individuals in the simulation, and a mutation rate μ of 10^-5 per site per individual per generation with each mutation having a 99% probability of being deleterious with a selective disadvantage of 0.01. The tree shown represents the maximum posterior tree sampled over the course of the MCMC analysis of the simulated data. Colors move from red to purple as fitness decreases. For coalescent statistics, TMRCA, diversity and N_eτ, solid lines represent mean values and gray outlines represent 95% credible intervals across MCMC replicates. TMRCA, diversity and N_eτ are measured in units of generations.

Figure 5

Genealogical tree and skyline plots of coalescent statistics for sequences simulated under constant positive selection. There were 10,000 individuals in the simulation, and an advantageous mutation rate μ_a of 0.002 per individual per generation with each mutation harboring a selective advantage s of 0.01. The tree shown represents the maximum posterior tree sampled over the course of the MCMC analysis of the simulated data. Colors move from purple to red as fitness increases. For coalescent statistics, TMRCA, diversity and N_eτ, solid lines represent mean values and gray outlines represent 95% credible intervals across MCMC replicates. TMRCA, diversity and N_eτ are measured in units of generations.

Figure 6

Genealogical tree and skyline plots of coalescent statistics for sequences simulated under episodic positive selection. There were 10,000 individuals in the simulation, and an advantageous mutation rate μ_a of 8 × 10^-7 per individual per generation with each mutation harboring a selective advantage s of 0.1. The tree shown represents the maximum posterior tree sampled over the course of the MCMC analysis of the simulated data. Colors move from purple to red as fitness increases. For coalescent statistics, TMRCA, diversity and N_eτ, solid lines represent mean values and gray outlines represent 95% credible intervals across MCMC replicates. TMRCA, diversity and N_eτ are measured in units of generations.

Figure 7

Relationship between population size N and genetic diversity at varying intensities of selection. Nucleotide diversity represents mean pairwise proportion of sequence differences between random individuals in the population. Diversity estimates are taken from 10⁵ generations across 3 replicate simulations. Neutral dynamics are shown as a dashed line. Selective dynamics are shown as solid lines. In each case, 1% of mutations are selectively advantageous, with a selective advantage s of 0.001, 0.01 or 0.1.

Figure 8

Genealogical tree and skyline plots of coalescent statistics for 1270 world-wide samples of HA sequences of influenza A (H3N2). The tree shown represents the maximum posterior tree sampled over the course of the MCMC analysis. Vaccine strains are shown on the tree as colored points. For coalescent statistics, TMRCA, diversity and N_eτ, solid lines represent mean values and gray outlines represent 95% credible intervals across MCMC replicates. TMRCA, diversity and N_eτ are measured in units of years.

Figure 9

Genealogical tree and skyline plots of coalescent statistics for 624 China and Southeast Asia samples of HA sequences of influenza A (H3N2). The tree shown represents the maximum posterior tree sampled over the course of the MCMC analysis. Vaccine strains are shown on the tree as colored points and included regardless of geographic origin. For coalescent statistics, TMRCA, diversity and N_eτ, solid lines represent mean values and gray outlines represent 95% credible intervals across MCMC replicates. TMRCA, diversity and N_eτ are measured in units of years.