Compositional Bias, Character-State Bias, and Character-State Reconstruction Using Parsimony

Compositional bias, the occurrence of the four bases in unequal proportions, is a common feature of nucleotide sequences. We analyzed patterns of character-state reconstruction using maximum parsimony in two empirical data sets exhibiting compositional bias. For each case in which the inferred reciprocal numbers of changes for a pair of bases differed substantially, the two bases also differed markedly in relative abundance and the asymmetry of reconstructed transformations favored changes from the common state to the rare state. In addition, the compositional biases of the inferred ancestral sequences were more extreme than that seen in the terminal taxa, having an excess of common states relative to the terminals. Both of these features suggested that patterns of character-state reconstruction might be systematically dis? torted when compositional bias results in unequal representation of character states within a character, a condition we term character-state bias. Character-state bias is essentially composi? tional bias within a character. Simulation studies showed that highly asymmetric patterns of character-state reconstruction can be produced in the face of an underlying symmetry of char? acter-state transformations in the presence of compositional bias. Rates of change are also im? portant. The asymmetry of transformations produced in the simulations matched the pattern found in empirical data sets, with transformations from the common state to the rare state being more abundant. Rare states tend to be autapomorphic, typically requiring a change to the rare state on a tree. Conversely, changes from rare states to common states are systematically lost. These results are significant for methods that rely on character-state reconstruction using max? imum parsimony, for example, to develop weighting schemes for phylogenetic analysis or to study patterns of correlated character evolution. Techniques that rely on character-state recon? struction may often be compromised by the distorting influence of character-state bias. (Com? positional bias; character-state bias; character-state reconstruction; parsimony.)