Machine-Learning Classification Suggests That Many Alphaproteobacterial Prophages May Instead Be Gene Transfer Agents

Abstract

Many of the sequenced bacterial and archaeal genomes encode regions of viral provenance. Yet, not all of these regions encode bona fide viruses. Gene transfer agents (GTAs) are thought to be former viruses that are now maintained in genomes of some bacteria and archaea and are hypothesized to enable exchange of DNA within bacterial populations. In Alphaproteobacteria, genes homologous to the “head–tail” gene cluster that encodes structural components of the Rhodobacter capsulatus GTA (RcGTA) are found in many taxa, even if they are only distantly related to Rhodobacter capsulatus. Yet, in most genomes available in GenBank RcGTA-like genes have annotations of typical viral proteins, and therefore are not easily distinguished from their viral homologs without additional analyses. Here, we report a “support vector machine” classifier that quickly and accurately distinguishes RcGTA-like genes from their viral homologs by capturing the differences in the amino acid composition of the encoded proteins. Our open-source classifier is implemented in Python and can be used to scan homologs of the RcGTA genes in newly sequenced genomes. The classifier can also be trained to identify other types of GTAs, or even to detect other elements of viral ancestry. Using the classifier trained on a manually curated set of homologous viruses and GTAs, we detected RcGTA-like “head–tail” gene clusters in 57.5% of the 1,423 examined alphaproteobacterial genomes. We also demonstrated that more than half of the in silico prophage predictions are instead likely to be GTAs, suggesting that in many alphaproteobacterial genomes the RcGTA-like elements remain unrecognized.

Selection for reducing energy cost of protein production drives the GC content and amino acid composition bias in gene transfer agents

“…Due to the relationship between codons and amino acids in the genetic code, the elevated GC1-and GC2-content of an open reading frame (ORF) translates into a biased amino acid composition of the encoded protein. Indeed, a significant amino acid composition bias in the GTA proteins has been demonstrated previously (11).…”

“…Consistent with this notion, energy savings for GTA proteins are as pronounced or even greater than those for highly expressed housekeeping genes that are known to utilize cheaper and smaller amino acids (20). Given that production of RcGTA-like particles in Alphaproteobacteria occurs in the stationary phase (2,30) and is associated with carbon depletion (5,11), the shift in GC-content of GTA genes and amino acid composition of their products likely reflects the adaptation for their efficient expression under such conditions. The change in the amino acid composition of GTA proteins was not uniform across the examined alphaproteobacterial lineages.…”

“…Moreover, the elevated GC3-content would not be limited to the genes in the RcGTA-like region but would be apparent in the adjacent genes as well. To test this hypothesis, we examined homologs of one RcGTA locus ('head-tail' cluster) in 212 alphaproteobacterial genomes (see Materials and Methods) (8,11). Although we analyzed homologs of only one locus from one GTA only, for brevity, we hereafter refer to these regions simply as "GTA regions", and to genes and encoded proteins in these regions as "GTA genes" and "GTA proteins".…”

“…The RcGTA production has been experimentally demonstrated to be stimulated by carbon depletion (5). Furthermore, knockout of the RcGTAlike genes in three alphaproteobacterial strains (24) resulted in a significant decrease in fitness of the mutants under growth conditions with alternative carbon sources that might not be utilized by these strains (11). If GTAs are indeed produced under conditions of limited carbon availability, the observed amino acid bias in the GTA genes might represent an adaptation in the GTAcontaining lineages to utilize energetically cheaper amino acids for GTA particle production.…”

“…Its production is directed by at least five loci that are scattered across the R. capsulatus genome, with 17 genes that encode most of the proteins necessary for the production of the RcGTA particles located in one locus (Table S1) (10). This locus, also known as the 'head-tail' cluster (2), is detectable in many alphaproteobacterial genomes (8,11). Across Alphaproteobacteria, the RcGTA-like 'head-tail' clusters appear to evolve relatively slowly (1), have an elevated GC-content relative to the host genome (8), and have skewed amino acid composition when compared to their viral homologs (11).…”