Bipartite Network Analysis of Gene Sharings in the Microbial World

Abstract

Extensive microbial gene flows affect how we understand virology, microbiology, medical sciences, genetic modification, and evolutionary biology. Phylogenies only provide a narrow view of these gene flows: plasmids and viruses, lacking core genes, cannot be attached to cellular life on phylogenetic trees. Yet viruses and plasmids have a major impact on cellular evolution, affecting both the gene content and the dynamics of microbial communities. Using bipartite graphs that connect up to 149,000 clusters of homologous genes with 8,217 related and unrelated genomes, we can in particular show patterns of gene sharing that do not map neatly with the organismal phylogeny. Homologous genes are recycled by lateral gene transfer, and multiple copies of homologous genes are carried by otherwise completely unrelated (and possibly nested) genomes, that is, viruses, plasmids and prokaryotes. When a homologous gene is present on at least one plasmid or virus and at least one chromosome, a process of "gene externalization," affected by a postprocessed selected functional bias, takes place, especially in Bacteria. Bipartite graphs give us a view of vertical and horizontal gene flow beyond classic taxonomy on a single very large, analytically tractable, graph that goes beyond the cellular Web of Life.

Fig. 1.

Construction of the bipartite graphs and identification of the twins and articulation points. (A) Construction of the genome-CHG bipartite graphs. Top nodes represent genomes of cells and mobile genetic elements. Bottom nodes represent CHG: we display the corresponding connected component of the sequence similarity network (see Materials and Methods), edge color (from yellow to red) indicates increasing % ID. (B) Bottom twins and articulation points: bottom nodes forming a twin class and their incident edges are drawn in the same shade of gray. Nodes 7 and 8 have the same neighbors (nodes 1 and 2) thus form twin class 1. Twins 2 and 4 are trivial since they contain only one node. Node 9 is an articulation point since its removal disconnects the graph.

Fig. 2

Support composition of twins and CHGs. The taxonomy of the carriers of the genes in a CHG is called its support composition. For every similarity threshold (on the x-axis), we report how many CHGs (on the right) or twins of CHGs (on the left) are exclusively found in Archaea, in Bacteria, in both kinds of prokaryotes (“mixed prokaryotes”), in viruses only, in plasmids only, in both kinds of MGE (“mixed MGE”), or in both cellular and mobile elements (“externalized”).

Fig. 3

ESCHG and core. The “shared pangenome” of a lineage is composed of all CHGs that are shared by at least two genomes and contain at least one member of the lineage. (A) For each taxonomic group, we report the percentage of CHGs forming the shared pangenome that are core exclusive, exclusive but noncore and not exclusive. (B) For each taxonomic group, we report the percentage of the shared pangenome that is comprised of core exclusive, and core but not exclusive CHGs. The total height of the bar represents the core itself.

Fig. 4.

Percentage of gene externalization at ≥ 30% ID for the 382 prokaryotes in our data set. The proportion of externalized genes from Bacteria (green triangles) is significantly higher (Student’s t-test, p value < 10⁻¹⁶) than for Archaea (yellow dots), to the notable exception of Haloarchaea (red dots). It is moreover largely uncorrelated with genome size (regression lines with shaded confidence interval at 95%).

Fig. 5.

Distribution of functional categories among externalized genes. Color bars represent the percentage of a given COG category among externalized genes above a given identity threshold (according to the color code on the right side of the figure). On the upper bar, COG categories are grouped by large functional groups (including “Poorly characterized,” which includes COG categories R and S). The “No”/0 class (on the left) refers to the genes for which no COG category was attributed). Note the very conspicuous peak for the “L” category at thresholds ≥ 90% and 95% ID.