Epistatically interacting substitutions are enriched during adaptive protein evolution

Abstract

Most experimental studies of epistasis in evolution have focused on adaptive changes-but adaptation accounts for only a portion of total evolutionary change. Are the patterns of epistasis during adaptation representative of evolution more broadly? We address this question by examining a pair of protein homologs, of which only one is subject to a well-defined pressure for adaptive change. Specifically, we compare the nucleoproteins from human and swine influenza. Human influenza is under continual selection to evade recognition by acquired immune memory, while swine influenza experiences less such selection due to the fact that pigs are less likely to be infected with influenza repeatedly in a lifetime. Mutations in some types of immune epitopes are therefore much more strongly adaptive to human than swine influenza--here we focus on epitopes targeted by human cytotoxic T lymphocytes. The nucleoproteins of human and swine influenza possess nearly identical numbers of such epitopes. However, mutations in these epitopes are fixed significantly more frequently in human than in swine influenza, presumably because these epitope mutations are adaptive only to human influenza. Experimentally, we find that epistatically constrained mutations are fixed only in the adaptively evolving human influenza lineage, where they occur at sites that are enriched in epitopes. Overall, our results demonstrate that epistatically interacting substitutions are enriched during adaptation, suggesting that the prevalence of epistasis is dependent on the underlying evolutionary forces at play.

Figure 1. Phylogenetic tree of human and swine NP homologs.

The human and swine NP lineages in this tree are descended from a virus closely related to the 1918 virus. Swine viruses are highlighted in yellow; all other viruses are human. In red are the lines of descent to the human H3N2 strains Aichi/1968 and Texas/2012 from their most-recent common ancestor. In green are the lines of descent to the swine H1N1 strains swine/Wisconsin/1957 and swine/Indiana/2012 from their most-recent common ancestor. Overall, this tree shows NPs from the following lineages: human seasonal H1N1, human H2N2, human H3N2, and North American swine viruses. The tree is a maximum clade credibility summary of a posterior distribution sampled from date-stamped protein sequences using BEAST with a JTT substitution model. See http://jbloom.github.io/mutpath/example_influenza_NP_1918_Descended.html for code, input data, and detailed documentation.

Figure 2. Evolutionary trajectories of human and swine NP.

Mutational paths through protein sequence space along (A) the evolutionary trajectory from the human strain Aichi/1968 to Texas/2012 and (B) the evolutionary trajectory from swine/Wisconsin/1957 to swine/Indiana/2012. In the mutational paths, circles represent unique protein sequences, with areas and intensities proportional to the posterior probability that the sequence was part of the trajectory. Blue lines with black labels represent single mutations between sequences, with thicknesses and intensities proportional to the posterior probability that the mutational connection was part of the trajectory. When there is no single high-probability one-mutation connection between sequences, red lines and labels indicate that several mutations fixed in an unknown order. See http://jbloom.github.io/mutpath/example_influenza_NP_1918_Descended.html for code, input data, and detailed documentation. The trajectory in (A) is highly similar to that reported in , but is slightly longer and contains sequences from prior to 1968. The inclusion of these pre-1968 sequences is the reason why the first portion of the trajectory is slightly better resolved than that in .

Figure 3. Human and swine NP possess similar numbers of human CTL epitopes.

(A) The number of known human CTL epitopes for each residue for human and swine NP. (B) The distribution of number of epitopes per site. The curves in (B) are consistent with the null hypothesis that the human and swine per-site epitope counts are drawn from the same underlying distribution (Kolmogorov-Smirnov test, P = 1.00). The number of epitopes for each site was determined by downloading all human MHC class I epitopes with experimentally verified T-cell responses from the Immune Epitope Database , and identifying epitopes between 8 and 12 residues in length that aligned with Aichi/1968 or Texas/2012 (for human NP) or with swine/Wisconsin/1957 or swine/Indiana/2012 (for swine NP) with no more than one mismatch. Redundant epitopes for the same MHC allele were removed. The epitopes per site are listed in Table S1 and Table S2. See http://jbloom.github.io/epitopefinder/example_NP_CTL_epitopes_H3N2_and_swine.html for code, input data, and detailed documentation.

Figure 4. Human NP exhibits increased evolution in CTL epitopes relative to swine NP.

The number of CTL epitopes per site for all sites in NP versus those that substituted along the evolutionary trajectories for (A) human and (B) swine influenza. In human influenza, the substituted sites contain more epitopes than average sites – but in swine influenza, the substituted sites contribute to fewer epitopes than average sites. The P-values on the plots are the fraction of random subsets of all sites that contain as many (human NP) or as few (swine NP) total epitopes as the sites that actually substituted during the natural evolution of that homolog. The hypothesis of greatest interest is whether the substituted sites in the human NP contain more epitopes than do substituted sites in the swine NP. To test this hypothesis, we drew paired random subsets of sites from the human and swine NP homolog of the same size as the actual numbers of substituted sites for each homolog, and determined the fraction of these paired random subsets in which the number of epitopes for the human NP exceeded that for the swine NP by at least as much as for the actual data. This test gives a P-value of 0.008, supporting the hypothesis that human NP exhibits an increased rate of evolution in epitopes relative to swine NP. See http://jbloom.github.io/epitopefinder/example_NP_CTL_epitopes_H3N2_and_swine.html for code, input data, and detailed documentation.

Figure 5. Epistatically constrained mutations are fixed in human but not swine NP.

All single mutations that occurred along the evolutionary trajectories were introduced individually into the Aichi/1968 (human NP) or swine/Wisconsin/1957 (swine NP), and the impact of the mutation on the total transcriptional activity of the influenza polymerase was measured experimentally. (A) The effect of the mutations to human NP, as originally reported in . (B) The effect of the mutations to swine NP. Individual mutations that are strongly deleterious are classified as “epistatically constrained,” since their fixation during natural evolution required additional secondary mutations to counteract the deleterious effects. Three epistatically constrained mutations fixed along the human NP trajectory, but no epistatically constrained mutations fixed along the swine NP trajectory. The epistatically constrained mutations are colored red in the plot. The numerical data in Figure 5A are in ; the numerical data in Figure 5B are in Table S3.

Figure 6. Epistasis in human NP occurs at sites enriched in CTL epitopes.

The number of CTL epitopes per site for the sites of the epistatically constrained substitutions in the human influenza NP versus (A) all sites or (B) the full set of sites that substituted along the evolutionary trajectory. The P-values shown on the plots represent the fraction of random subsets that contain as many total epitopes as the actual sites of the epistatically constrained substitutions. See http://jbloom.github.io/epitopefinder/example_NP_CTL_epitopes_H3N2_and_swine.html for code, input data, and detailed documentation.