Horizontal gene transfer drives the evolution of dependencies in bacteria

Abstract

Many naturally occurring bacteria lead a lifestyle of metabolic dependency for crucial resources. We do not understand what factors drive bacteria toward this lifestyle and how. Here, we systematically show the crucial role of horizontal gene transfer (HGT) in dependency evolution in bacteria. Across 835 bacterial species, we map gene gain-loss dynamics on a deep evolutionary tree and assess the impact of HGT and gene loss on metabolic networks. Our analyses suggest that HGT-enabled gene gains can affect which genes are later lost. HGT typically adds new catabolic routes to bacterial metabolic networks, leading to new metabolic interactions between bacteria. We also find that gaining new routes can promote the loss of ancestral routes ("coupled gains and losses", CGLs). Phylogenetic patterns indicate that both dependencies-mediated by CGLs and those purely by gene loss-are equally likely. Our results highlight HGT as an important driver of metabolic dependency evolution in bacteria.

Keywords: Computational molecular modeling; Microbiology; Molecular network.

Graphical abstract

Figure 1

Horizontal gene transfer adds new catabolic routes to bacteria (A) Schematic representation of our two-pronged approach: of combining phylogeny and bacterial metabolism. We used a well-known phylogenetic tree to infer the evolutionary relationships between the 835 bacterial species used in our analysis. For each extant species (shown on the tips of the tree), we used gene presence-absence data for 3,022 metabolic genes from the KEGG metabolic database. Filled circles indicate gene presence; empty indicate absence. (B) We inferred the gene presence-absence states of all internal nodes of the tree along each branch of the full tree (gray dashed circle in a); each branch connected an ancestor (anc) to a descendant (des). Along each branch, we inferred which genes were gained (green) and lost (red). (C) Bar plot showing the position of gained genes in bacterial metabolic networks. We split metabolic genes into catabolic (first or second reactions in a metabolic route) and anabolic (intermediate or biomass-synthesis reactions) based on the the chemical reactions they map to. Each green bar represents the average fraction of gained (horizontally transferred) genes at that position. Each black bar represents controls, i.e., the expected average fraction of gains at that position, given a random set of gene gains. Error bars show the SE, indicating the extent of variation across 1,669 phylogenetic branches.

Figure 2

HGT-enabled catabolic routes increase the likelihood of metabolic interactions (A) Distribution of the number of newly accessible catabolic routes (routes gained along each phylogenetic branch) across all 1,669 branches. The number of routes starting from nutrients are shown in blue, and those starting from byproducts are in red. New byproduct-driven routes would increase the chance of metabolic interactions with other bacteria (via their byproducts). We find that new catabolic routes are more likely to be byproduct-driven (median 56, versus 51 for nutrient-driven routes; $P<{10}^{-3}$; Kolmogorov-Smirnov test). (B and C) Pie charts comparing new routes with their corresponding ancestral routes on each of the 1,669 branches. We compare new and ancestral routes based on their (B) path length (i.e., is the shortest new path shorter, longer, or the same length as the shortest ancestral path) and (C) energy yield (i.e., does the most energy-yielding new path have a higher, lower, or equal yield than the best ancestral path).

Figure 3

HGT can affect dependency evolution via coupled gains and losses of genes (A) Schematic illustration of coupled gains and losses (CGLs), a new, alternate mechanism for dependency evolution driven by HGT. The gray box on the left indicates the environment, with a nutrient (nut, blue circle) and a byproduct (byp, purple triangle). Red arrows indicate the secretion and import of metabolites by bacteria. The three steps in the frame show how a bacterial species can evolve a dependency on another species that donates byp in the environment; by step 3, the acceptor species eventually depends on byp. Each step follows the modification of a part of the acceptor’s metabolic network. At step 1, the acceptor uses an ancestral route (black arrows) to convert nut in the environment to a key biomass component (bmc, yellow square). At step 2, it can alternatively use a newly gained route to convert byp to the same bmc. The ancestral and gained routes are coupled to each other, because they both produce bmc, which is crucial for survival and growth. At step 3, the acceptor loses the ancestral route (gray arrows) but can still produce bmc through the gained route. It thus becomes dependent on donors of byp for survival. (B) Schematic illustration of dependency evolution via pure gene loss, not driven by HGT. The environment now has bmc available as a byproduct, instead of byp (gray box on the left). In this mechanism, step 1 is the same as (A), but unlike (A), at step 2, the acceptor can lose the ancestral route straight away, without requiring an alternate coupled route. However, this requires a particular environment where bmc is available.

Figure 4

Metabolic dependencies are equally likely to emerge via CGLs and pure gene loss (A) Bar plot showing the fraction of 1,669 phylogenetic branches in which we observed gene gain-loss patterns consistent with coupled gains and losses (CGLs; green) and with pure gene loss (red). Each gray bar represents the corresponding controls, i.e., the expected fraction of branches with patterns consistent with CGLs and pure gene loss, given a random set of gene gains and losses. Error bars show the SE indicating the level of variation across all branches. (B) Line plot showing the likelihood of evolving dependency via CGLs (green) and pure gene loss (red) in simulated bacterial communities, as a function of the community diversity. The likelihood of dependency is the average fraction of communities in which the observed gains and losses along a branch led to a CGL-based on pure gene-loss-based dependency. Community diversity is the number of coexisting bacterial species in a simulated community. The gray region has $\ge 7$ species, where CGLs are more likely than pure gene loss.