Identifying Orthologous Genes for Phylogenetic Analysis

We applied a stringent criterion for eliminating nonhomologous sequences and paralogous sequences, since both are likely to lead to false conclusions regarding the organismal phylogeny and frequency of LGT. In particular, the criterion of “best reciprocal hits” between sequences for a genome pair can lead to false conclusions of orthology because the resulting gene pairs are not always closest relatives phylogenetically (Koski and Golding 2001 (link)). Instead, we used a cutoff for the degree of similarity as reflected in the BLASTP bit scores (Altschul et al. 1997 (link)). The bit score is dependent upon the scoring system (substitution matrix and gap costs) employed and takes into account both the degree of similarity and the length of the alignment between the query and the match sequences. We used it to detect homologous genes, described as follows. A bank of all annotated protein sequences of all included species was created. A BLASTP (Altschul et al. 1997 (link)) search was performed for all the proteins in each genome against the protein bank. This implies that all proteins were searched against both their resident genome and those from the 12 other species. The match of a given protein against itself gives a maximal bit score. To determine a threshold to group genes into a family, we examined the distribution of the ratio of the bit score to the maximal (self) bit score based on the proteins of E. coli compared against proteins of the 12 genomes (Figure 6). In each case, the distribution showed a clear bimodal pattern with a first peak of low similarity values, which is constant among comparisons and therefore probably represents random matches, and a second peak of higher similarity values, representing true homologous genes. For comparisons of E. coli proteins with those of the most distant species in our set, such as Vibrio, Xanthomonas, Xylella, and Pseudomonas, the separation of the two portions of the distribution occurs at about 30% of the maximal bit score. Thus, in order to apply a stringent criterion for homology, we inferred as homologous genes those presenting a bit score value higher or equal to 30% of the maximal bit score. A protein was included in a family if this criterion was satisfied for at least one member. Our cutoff was chosen to minimize inclusion of nonhomologous sequences within a family; consequently, it may exclude some homologs, especially fast-evolving ones.
After establishing homolog families, we selected the set that contained a single sequence in each represented genome and regarded these as likely orthologs that could give information about the organismal phylogeny and the frequency of LGT affecting orthologs in this bacterial group.