CHARACTER-STATE WEIGHTING FOR CLADISTIC-ANALYSIS OF PROTEIN-CODING DNA-SEQUENCES

Nucleotide data are a restricted character system complex enough to confound phylogenetic analyses yet simple enough to permit establishment of probability models for sequence change and corresponding character-state weighting schemes. We have previously developed a general method for weighting DNA data that is here elaborated for protein-coding sequences. Included in the present model are corrections for (i) multiple substitution events, (ii) transition/transversion bias, and (iii) differential proportions of changes occurring at first, second, and third codon positions. This model is shown to be generally consistent for all phylogenetically useful data. Greater understanding of the properties of equal versus differential character-state weighting comes from consideration of numbers of terminal taxa and lengths of tree segments. With insufficient sampling of taxa, differential weighting attempts to correct for undetected multiple substitution events. Both equal and differential weighting should give the same result if sufficient numbers of terminal taxa permit the detection of historically misleading character-state changes. Nevertheless, spurious attraction of tree segments remains a systematic problem that is not easily resolved either by equal weighting or by our differential weighting model, which acts globally rather than adjusting for different probabilities of character-state change among tree segments. Artifactual segment attraction is best understood in terms of asymmetries in lambda (which represents state changes per character during a particular segment interval). We relate the consistency index to numbers of terminal taxa and lambda, illustrating its dependence upon numbers of potential tree segments. Prospects for phylogenetic reconstruction from protein-coding nucleotide data are discussed with reference to the robustness of equal weighting (given our own model) with adequate taxonomic sampling.