Core set approach to reduce uncertainty of gene trees

A genealogy based on gene sequences within a species plays an essential role in the estimation of the character, structure, and evolutionary history of that species. Because intraspecific sequences are more closely related than interspecific ones, detailed information on the evolutionary process may be available by determining all the node sequences of trees and provide insight into functional constraints and adaptations. However, strong evolutionary correlations on a few lineages make this determination difficult as a whole, and the maximum parsimony (MP) method frequently allows a number of topologies with a same total branching length.

Kitazoe et al. developed multidimensional vector-space representation of phylogeny. It converts additivity of evolutionary distances to orthogonality among the vectors expressing branches, and provides a unified index to measure deviations from the orthogoality. In this paper, this index is used to detect and exclude sequences with large deviations from orthogonality, and then selects a maximum subset ("core set") of sequences for which MP generates a single solution. Once the core set tree is formed whose all the node sequences are given, the excluded sequences are found to have basically two phylogenetic positions on this tree, respectively. Fortunately, since multiple substitutions are rare in intra-species sequences, the variance of nucleotide transitions is confined to a small range. By applying the core set approach to 38 partial env sequences of HIV-1 in a single patient and also 198 mitochondrial COI and COII DNA sequences of Anopheles dirus, we demonstrate how consistently this approach constructs the tree.

In the HIV dataset, we confirmed that the obtained core set tree is the unique maximum set for which MP proposes a single tree. In the mosquito data set, the fluctuation of nucleotide transitions caused by the sequences excluded from the core set was very small. We reproduced this core-set tree by simulation based on random process, and applied our approach to many sets of the obtained endpoint sequences. Consequently, the ninety percent of the endpoint sequences was identified as the core sets and the obtained node sequences were perfectly identical to the true ones.