The human phylome

Abstract

Background: Phylogenomics analyses serve to establish evolutionary relationships among organisms and their genes. A phylome, the complete collection of all gene phylogenies in a genome, constitutes a valuable source of information, but its use in large genomes still constitutes a technical challenge. The use of phylomes also requires the development of new methods that help us to interpret them.

Results: We reconstruct here the human phylome, which includes the evolutionary relationships of all human proteins and their homologs among 39 fully sequenced eukaryotes. Phylogenetic techniques used include alignment trimming, branch length optimization, evolutionary model testing and maximum likelihood and Bayesian methods. Although differences with alternative topologies are minor, most of the trees support the Coelomata and Unikont hypotheses as well as the grouping of primates with laurasatheria to the exclusion of rodents. We assess the extent of gene duplication events and their relationship with the functional roles of the protein families involved. We find support for at least one, and probably two, rounds of whole genome duplications before vertebrate radiation. Using a novel algorithm that is independent from a species phylogeny, we derive orthology and paralogy relationships of human proteins among eukaryotic genomes.

Conclusion: Topological variations among phylogenies for different genes are to be expected, highlighting the danger of gene-sampling effects in phylogenomic analyses. Several links can be established between the functions of gene families duplicated at certain phylogenetic splits and major evolutionary transitions in those lineages. The pipeline implemented here can be easily adapted for use in other organisms.

Figure 1

Schematic representation of the phylogenetic pipeline used to reconstruct the human phylome. Each protein sequence encoded in the human genome is compared against a database of proteins from 39 fully sequenced eukaryotic genomes (Table 1) to select putative homologous proteins. Groups of homologous sequences are aligned and subsequently trimmed to remove gap-rich regions. The refined alignment is used to build a NJ tree, which is then used as a seed tree to perform a ML likelihood analysis as implemented in PhyML, using four different evolutionary models (five in the case of mitochondrially encoded proteins). The ML tree with the maximum likelihood is further refined with a Bayesian analysis using MrBayes. Finally, different algorithms are used to search for specific topologies in the phylome or to define orthology and paralogy relationships.

Figure 2

The alternative phylogenetic relationships among the taxa involved in the three evolutionary hypotheses considered. (a) Placental mammals: primates, laurasatheria and rodents. (b) Ecdysozoa versus Coelomata hypothesis: relationships among arthropods, chordates and nematodes. And (c) the Unikont hypothesis: relationship among opisthokonts, amoebozoans and other eukaryotic groups. The numbers indicate the number of trees supporting each topology. For each alternative topology numbers on the top row refer to the total number of trees with a given topology, and what percentage of the total it represents; numbers in the middle row refer to those trees for which the posterior probabilities of the two partitions shown in the figure are 0.9 or higher. Numbers in the bottom row refer to the number and percentage of gene families supporting each topology.

Figure 3

Estimates for the number of duplication events occurred at each major transition in the evolution of the eukaryotes. Species abbreviations are the same as in table 1. Horizontal bars indicate the average number of duplications per gene. Boxes on the right list some of the GO terms of the biological process category that are significantly over-represented compared to the rest of the genome in the set of gene families duplicated at a certain stage. A full list of significantly over represented terms is given as a table in the supplementary material [22].

Figure 4

Benchmarking comparison of different orthology inference algorithms. The reference set used in the benchmark of Hulsen et al. [82] is taken as a gold standard to compute the number of true positives (TP), false positives (FP) and false negatives (FN) yielded by each method. For each method the sensitivity (S = TP/(TP+FN)) and the positive predictive value (P = TP/(TP + FP)) are computed. Methods described in [82] are indicated as BBH (Best reciprocal hits), MCL (OrthoMCL), ZIH (Z-score 1-hundred.), INP (Inparanoid), PGT (phylogeny-based algorithm used in [95]), KOG (Clusters of eukaryotic orthologous goups). 'Phylome' represents the results of our pipeline and algorithm, and Ensbl the orthology relationships predicted by Ensembl database.