Estimating the Frequency of Horizontal Gene Transfer Using Phylogenetic Models of Gene Gain and Loss

Abstract

We analyze patterns of gene presence and absence in a maximum likelihood framework with rate parameters for gene gain and loss. Standard methods allow independent gains and losses in different parts of a tree. While losses of the same gene are likely to be frequent, multiple gains need to be considered carefully. A gene gain could occur by horizontal transfer or by origin of a gene within the lineage being studied. If a gene is gained more than once, then at least one of these gains must be a horizontal transfer. A key parameter is the ratio of gain to loss rates, a/v We consider the limiting case known as the infinitely many genes model, where a/v tends to zero and a gene cannot be gained more than once. The infinitely many genes model is used as a null model in comparison to models that allow multiple gains. Using genome data from cyanobacteria and archaea, it is found that the likelihood is significantly improved by allowing for multiple gains, but the average a/v is very small. The fraction of genes whose presence/absence pattern is best explained by multiple gains is only 15% in the cyanobacteria and 20% and 39% in two data sets of archaea. The distribution of rates of gene loss is very broad, which explains why many genes follow a treelike pattern of vertical inheritance, despite the presence of a significant minority of genes that undergo horizontal transfer.

Producing Polished Prokaryotic Pangenomes with the Panaroo Pipeline

“…These plots include a Multidimensional scaling (MDS) projection of pairwise Mash distances, interactive bar charts to investigate contamination, as well as gene and contig counts. In addition we have included post processing scripts for estimating gene gain and loss rates using both the infinitely many genes (IMG) model (37,48) and the finitely many genes (FMG) model of (8,48). These are preferable to the common practice of plotting accumulation curves to indicate pangenome size as they account for the diversity and timescale of the sampled isolates.…”

panX: pan-genome analysis and exploration

“…Individual gain and loss events are associated with branches based on this ancestral reconstruction. The gain and loss rates are optimized such that the likelihood for the observed presence/absence pattern of genes is maximized [32,33]. We found that optimal loss rates are always larger than the gain rates but their ratio is variable among species with a median ratio of 22 (inter-quartile range 9 to 35).…”

Peer Review #1 of "What is an archaeon and are the Archaea really unique? (v0.1)"

“…In general, systematic models of gene (or domain) gain-and-loss estimate significantly lower frequencies of HT(Zamani-Dahaj et al 2016), compared to HT estimates based on overall similarities(Nelson-Sathi et al 2015;Roettger et al 2009). HT estimates are consistently lower across the ToL: within Rickettsia, a genus(Murray et al 2016); within Cyanobacteria, a phylum and within Archaea(Zamani-Dahaj et al 2016). In addition, and importantly, these studies show that the tree-like pattern of inheritance of genomic loci is explained largely by the variation in rates of loss among lineages, and that the fraction of loci that are prone to HT is a minority…”

“…Most potential HT events inferred from anomalous placements of gene-OTUs are associated with systematic error, even among closelyManuscript to be reviewed of genomic loci by HT. Estimation of gains is better suited for methods that are designed to systematically model gains and losses in genomes(Klopfstein et al 2015;Zamani-Dahaj et al 2016). In general, systematic models of gene (or domain) gain-and-loss estimate significantly lower frequencies of HT(Zamani-Dahaj et al 2016), compared to HT estimates based on overall similarities(Nelson-Sathi et al 2015;Roettger et al 2009).…”

“…Estimation of gains is better suited for methods that are designed to systematically model gains and losses in genomes(Klopfstein et al 2015;Zamani-Dahaj et al 2016). In general, systematic models of gene (or domain) gain-and-loss estimate significantly lower frequencies of HT(Zamani-Dahaj et al 2016), compared to HT estimates based on overall similarities(Nelson-Sathi et al 2015;Roettger et al 2009). HT estimates are consistently lower across the ToL: within Rickettsia, a genus(Murray et al 2016); within Cyanobacteria, a phylum and within Archaea(Zamani-Dahaj et al 2016).…”