Quantifying Cross-Species Similarity in Distributions of Amino Acid Compositional Divergence: Codon Partitioning and Pairwise Wasserstein Analysis of 85 Species

抄録

Our previous study showed that distributions of Jensen–Shannon distances between individual protein amino acid compositions and their species-specific mean compositions were unexpectedly similar across 85 diverse species, whereas analogous distributions based on CDS nucleotide composition were substantially more heterogeneous. Here, we examined how this contrast was expressed across nucleotide, codon, and amino acid representations derived from the same coding sequences, using a forward Kullback–Leibler (KL) framework that permits exact partitioning of total codon divergence into amino acid-composition and synonymous-codon-usage components.

We reanalyzed 2,095,689 retained CDS records from the same 85 species under an operational standard-code framework. Each CDS was represented by the frequencies of 61 standard sense codons, from which nucleotide and amino acid compositions were derived. Wasserstein-1 distances between species-specific empirical KL-divergence distributions were calculated for all 3,570 unique species pairs, first using the complete retained CDS set and then within three coding-length bins (100–199, 200–399, and 400–799 included codons). Pairwise distances were standardized by the median within-species interquartile range for the corresponding divergence type and analysis set.

In the whole-dataset analysis, the standardized mean pairwise distance was lowest for amino acid divergence (0.254), followed by total codon divergence (0.335), synonymous divergence (0.396), and nucleotide divergence (0.523). Amino acid divergence also showed the lowest standardized distance in every coding-length bin (0.258–0.311) and remained relatively stable across bins, whereas the standardized distances for the other divergence types increased with coding length.

Thus, the nucleotide–amino acid contrast previously observed using Jensen–Shannon distance was recapitulated under forward KL divergence and direct distribution-to-distribution comparison. Under the IQR-standardized comparison used here, cross-species similarity was greatest at the amino acid-composition level, although the relative contributions of codon-to-amino-acid dimensional reduction and biological constraints remain to be determined.