Of woods and webs: possible alternatives to the tree of life for studying genomic fluidity in E. coli

Abstract

Background: We introduce several forest-based and network-based methods for exploring microbial evolution, and apply them to the study of thousands of genes from 30 strains of E. coli. This case study illustrates how additional analyses could offer fast heuristic alternatives to standard tree of life (TOL) approaches.

Results: We use gene networks to identify genes with atypical modes of evolution, and genome networks to characterize the evolution of genetic partnerships between E. coli and mobile genetic elements. We develop a novel polychromatic quartet method to capture patterns of recombination within E. coli, to update the clanistic toolkit, and to search for the impact of lateral gene transfer and of pathogenicity on gene evolution in two large forests of trees bearing E. coli. We unravel high rates of lateral gene transfer involving E. coli (about 40% of the trees under study), and show that both core genes and shell genes of E. coli are affected by non-tree-like evolutionary processes. We show that pathogenic lifestyle impacted the structure of 30% of the gene trees, and that pathogenic strains are more likely to transfer genes with one another than with non-pathogenic strains. In addition, we propose five groups of genes as candidate mobile modules of pathogenicity. We also present strong evidence for recent lateral gene transfer between E. coli and mobile genetic elements.

Conclusions: Depending on which evolutionary questions biologists want to address (i.e. the identification of modules, genetic partnerships, recombination, lateral gene transfer, or genes with atypical evolutionary modes, etc.), forest-based and network-based methods are preferable to the reconstruction of a single tree, because they provide insights and produce hypotheses about the dynamics of genome evolution, rather than the relative branching order of species and lineages. Such a methodological pluralism - the use of woods and webs - is to be encouraged to analyse the evolutionary processes at play in microbial evolution.This manuscript was reviewed by: Ford Doolittle, Tal Pupko, Richard Burian, James McInerney, Didier Raoult, and Yan Boucher.

Figure 1

Genome network of E. coli at 100% identity. (A) Each node corresponds to a genome (blue for E. coli, purple for plasmid, orange for viruses, brown for E. histolytica, green for A. laidlawii and S. putrefaciens). Edges connect pairs of genomes sharing at least one gene with 100% identical sequence. The display is a spring-embedded layout. (B) Same dataset and same colour code for the nodes. The display was a group attributes layout, with three groups: viruses, plasmids and E. coli. Edges are coloured based on the dominant function of the shared genes: red for the replication and repair category, cyan for all the other COG categories and black for genes without known functions. Cytoscape [66] was used for both displays.

Figure 2

Selected connected components of the E. coli gene network. Nodes correspond to gene sequences (blue for E. coli, green for all other bacteria, orange for archaea, and pink for mobile genetic elements). Edges were drawn when sequences showed an homology with a BLAST score < 1e-20, > 30% identity, option false BBH. Cytoscape was used for the display. (A) Putative homoserine kinase type II. (B) Translation Initiation Factor I. (C) Predicted permeases. (D) Type V secretory pathway proteins. (E) restriction endonuclease S subunit.

Figure 3

Split decomposition graph of the E. coli strains. Visual representation of the conflict in the phylogenetic signals among 30 strains of E. coli, for (A) the core genes (n = 2317) and (B) the shell genes (n = 3511). The strains are tagged for pathogenicity with red nodes for PATH and blue nodes for NON-PATH E. coli. Splitstree4 http://www.splitstree.org/ was used for both display, with the Neighbor-Net algorithm [51].