Evidence for an ancient adaptive episode of convergent molecular evolution

Abstract

Documented cases of convergent molecular evolution due to selection are fairly unusual, and examples to date have involved only a few amino acid positions. However, because convergence mimics shared ancestry and is not accommodated by current phylogenetic methods, it can strongly mislead phylogenetic inference when it does occur. Here, we present a case of extensive convergent molecular evolution between snake and agamid lizard mitochondrial genomes that overcomes an otherwise strong phylogenetic signal. Evidence from morphology, nuclear genes, and most sites in the mitochondrial genome support one phylogenetic tree, but a subset of mostly amino acid-altering substitutions (primarily at the first and second codon positions) across multiple mitochondrial genes strongly supports a radically different phylogeny. The relevant sites generally evolved slowly but converged between ancient lineages of snakes and agamids. We estimate that approximately 44 of 113 predicted convergent changes distributed across all 13 mitochondrial protein-coding genes are expected to have arisen from nonneutral causes-a remarkably large number. Combined with strong previous evidence for adaptive evolution in snake mitochondrial proteins, it is likely that much of this convergent evolution was driven by adaptation. These results indicate that nonneutral convergent molecular evolution in mitochondria can occur at a scale and intensity far beyond what has been documented previously, and they highlight the vulnerability of standard phylogenetic methods to the presence of nonneutral convergent sequence evolution.

Fig. 1.

Squamate reptile phylogenetic tree. This Bayesian tree was estimated by using all 13 mitochondrial protein-coding genes and 2 nuclear genes. All nodes had 100% posterior probability support, except the 3 nodes indicated. In contrast to this topology, the agamid lizards are thought to form a group with the iguanid lizards (both in blue), as indicated by the red arrow. Trees based on mitochondrial genes tend to be similar to that shown (the MT topology). In contrast, trees based on nuclear genes place them with the Iguanidae (the NUC topology), in agreement with expectations from morphological studies.

Fig. 2.

Differences in site-specific likelihood support (ΔSSLS) for the MT and NUC topologies. Positive values of ΔSSLS indicate greater support for the NUC tree, and negative values indicate greater support for the MT tree. ΔSSLS across sites in all mitochondrial protein-coding genes are shown for (A) first codon positions, (B) second codon positions, (C) third codon positions, and (D) 4-fold degenerate sites. Values are shown in blue if the ΔSSLS magnitude is <0.5 and are shown in red if support levels are >0.5. This highlights strong support levels for one tree or the other. (E) ΔSSLS between the MT and NUC tree is broken down by relative rates of evolution for each of the 3 codon positions for all protein-coding mitochondrial genes.

Fig. 3.

Convergent evolution of protein sequences. The number of convergent and divergent substitutions in all pairs of branches along independent lines of descent was estimated (A) by using the ML marginal ancestral reconstructions, and (B) by using a Bayesian approach that calculated the posterior probability of all possible substitutions (see text). The numbers of convergent substitutions were related to the numbers of divergent substitutions by using orthogonal regressions (red line). The snake–agamid branch pair is well above the other branch pairs, regardless of the methodology used (red circles). The asymptotic calculation of the random expected fraction of neutral convergent substitutions, conditional on the ML parameter estimates from the observed data, is shown for reference (blue line in B; β̂ = 0.099). (C) Site-specific posterior probabilities of convergent substitutions between the snake–agamid branch pair for all codon positions using the Bayesian method. Sites with a high probability of having experienced convergent changes (red) are present in all protein-coding genes but are clustered particularly in COX1 and ND1. (D) Sliding window plots of the site-specific likelihood support in favor of the presumed false MT topology (blue) and the regional posterior probability of convergent substitutions (red). (E) Site-specific posterior number of substitutions versus the posterior probability of convergence per site; posterior substitutions were calculated to reduce the dependency of each site's rate estimate on the model of evolution. (F) Relationship between rates of evolution at a site, the probability of convergence, and the observed amino acid state space. Sites with posterior probabilities of convergence >80% are shown in red and >50% are shown in orange.