Lateral Gene Transfer Shapes Diversity of Gardnerella spp

Abstract

Gardnerella spp. are pathognomonic for bacterial vaginosis, which increases the risk of preterm birth and the transmission of sexually transmitted infections. Gardnerella spp. are genetically diverse, comprising what have recently been defined as distinct species with differing functional capacities. Disease associations with Gardnerella spp. are not straightforward: patients with BV are usually infected with multiple species, and Gardnerella spp. are also found in the vaginal microbiome of healthy women. Genome comparisons of Gardnerella spp. show evidence of lateral gene transfer (LGT), but patterns of LGT have not been characterized in detail. Here we sought to define the role of LGT in shaping the genetic structure of Gardnerella spp. We analyzed whole genome sequencing data for 106 Gardnerella strains and used these data for pan genome analysis and to characterize LGT in the core and accessory genomes, over recent and remote timescales. In our diverse sample of Gardnerella strains, we found that both the core and accessory genomes are clearly differentiated in accordance with newly defined species designations. We identified putative competence and pilus assembly genes across most species; we also found them to be differentiated between species. Competence machinery has diverged in parallel with the core genome, with selection against deleterious mutations as a predominant influence on their evolution. By contrast, the virulence factor vaginolysin, which encodes a toxin, appears to be readily exchanged among species. We identified five distinct prophage clusters in Gardnerella genomes, two of which appear to be exchanged between Gardnerella species. Differences among species are apparent in their patterns of LGT, including their exchange with diverse gene pools. Despite frequent LGT and co-localization in the same niche, our results show that Gardnerella spp. are clearly genetically differentiated and yet capable of exchanging specific genetic material. This likely reflects complex interactions within bacterial communities associated with the vaginal microbiome. Our results provide insight into how such interactions evolve and are maintained, allowing these multi-species communities to colonize and invade human tissues and adapt to antibiotics and other stressors.

Keywords: Gardnerella spp.; bacterial vaginosis; evolution; lateral gene transfer; recombination.

Figure 1

Gardnerella core genome maximum likelihood phylogeny supports distinct species/clade structure. We inferred a maximum likelihood phylogeny from a core genome alignment of 106 Gardnerella isolates. Species/clade labels reflect classification schemes from Ahmed et al. (2012) and Hill et al. (2019). Newly named species indicated (Vaneechoutte et al., 2019). The phylogeny is midpoint rooted, and nodes with bootstrap values lower than 70 shown in red. Branch lengths are scaled by the number of substitutions per site.

Figure 2

Accessory genome of Gardnerella is structured by species/clade. Gene homologs identified using Roary with 75% amino acid threshold are plotted (excluding singleton genes). Order of species in the phylogenetic tree (left) corresponds to species order in Figure 1. Accessory content differs among Gardnerella spp. species, which is consistent with barriers to LGT between species.

Figure 3

Recombination occurs most often between closely related species/clades. FastGEAR inference of recombination from a core genome alignment. Briefly, FastGEAR uses a Hidden Markov Model approach to cluster isolates into lineages, detect ancestral and recent recombination, and measure the statistical strength of the recombination events. FastGEAR identified 8 clades in the dataset, consistent with Figure 1 and published delineations (Ahmed et al., ; Schellenberg et al., ; Vaneechoutte et al., 2019). Isolates are ordered according to core genome phylogeny and colored according to each of 8 clades identified by FastGEAR. Each horizontal line refers to an isolate's core genome with colors representing the inferred origin of that region. The clades/species colors are labeled in the legend. (A) Recent recombinant tracts identified with FastGEAR. Overall there are few recombination events between clades/species, which appear to be more common between G. vaginalis and G. piotii, than other combinations. (B) Ancestral recombination shows a similar pattern of species structured LGT. White fragments correspond to recent recombination events (shown in A) and are masked when inferring ancestral recombination. Recent recombination inferred with a Bayesian factor (BF) > 1 and ancestral recombination with BF > 10 shown.

Figure 4

Competence gene homologs are ubiquitous across Gardnerella. Presence absence matrix of gene homologs likely to be related to competence (right) and maximum likelihood phylogenetic tree (left). Most of the genes are found among all clades/species, indicating competence related machinery is conserved among Gardnerella spp. Gene homologs were identified using both Roary and PIRATE. Additionally, we looked in the annotations to identify homologs missed using Roary or PIRATE alone.

Figure 5

Purifying selection is the primary force shaping diversity in Gardnerella core, competence, and vaginolysin genes. (A) Diversity of core genomes, competence genes, and vaginolysin within G. vaginalis and G. piotii. Distributions shown are of dN/dS across core genomes, concatenations of competence genes, and vaginolyisn. The box spans the interquartile range, the median is represented by the middle line, and the whiskers extend to ±1.5 times the interquartile range. Data beyond the end of the whiskers are outlying points and plotted individually. All genes appear to be under purifying selection (dN/dS <1). (B) Proportion of sites under positive selection along branches of the comEA maximum likelihood phylogeny. Using the aBSREL test in HyPhy, we identified branches with significant evidence (p < 0.05) of selection. Branch specific omega values show little evidence of positive selection, suggesting purifying selection.

Figure 6

CRISPR/cas and prophage are not ubiquitous across Gardnerella clades/species. Presence absence matrix of CRISPR/cas genes and phage clusters. CRISPR/cas genes were identified using PIRATE output and Prokka genome annotations. We identified prophage regions using ProphET and calculated pairwise mash distances of the nucleotide sequences to define prophage clusters. Prophage clusters 1 and 2 are found across Gardnerella spp., while prophage clusters 3 and 4 are restricted to G. vaginalis and G. piotii, with the exception of one cluster 4 prophage found in G. swidsinskii. We identified prophage clusters in 70% of Gardnerella isolates.

Figure 7

Prophage clusters have distinct clade/species restrictions and patterns of transfer between clades/species. We used multi-dimensional scaling (MDS) of pairwise mash distances of prophage nucleotide sequences identified using ProphET to visualize prophage clustering across clades/species. Prophage clusters 2 and 3 cluster based on the core genome species designations (B,C), and prophage clusters 1 and 4 do not cluster according to core genome species designations (A,D), suggesting that the prophage have been transferred across clades. We did not include an MDS plot of cluster 5, as it is found in only 2 isolates.

Figure 8

Within species recombination is frequent in Gardnerella core genomes. Recombination in the core genomes of G. vaginalis (A) and G. piotii (B). Each row corresponds to the core genome of an isolate in the phylogenetic tree to the left. Blue segments represent laterally transferred fragments unique to an individual isolate. Red segments indicate laterally transferred fragments that are shared across multiple isolates. The proportion of the alignment affected by recombination (both red and blue fragments) is 0.94 for G. vaginalis and 0.92 for G. piotii.

Figure 9

G. vaginalis core genomes are more recombinant than G. piotii. Boxplots show proportion of each isolate's core genome affected by recombination, as estimated with Gubbins. The box spans the interquartile range, the median is represented by the middle line, and the whiskers extend to ±1.5 times the interquartile range. Data beyond the end of the whiskers are outlying points and plotted individually. To account for differences in sample size of G. vaginalis and G. piotii isolates, we subsampled G. vaginalis isolates to the number of G. piotii isolates and used Gubbins to identify recombination in the subsampled dataset. The mean of affected core genomes in the subsampled G. vaginalis his higher than that of G. piotii (Mann-Whitney-Wilcoxon test, W = 473, p = 0.0138).

Figure 10

Accessory genes are not maintained at similar frequencies in G. vaginalis and G. piotii. Heat map of pangenome gene frequencies in G. vaginalis and G. piotii. Accessory genes are not maintained at similar frequencies in the two species suggesting that selection pressures for the shared accessory genes are not the same across species. For example, some genes are maintained at high frequencies in one species, but low in the other. Number of genes are colored on a log scale.

Figure 11

Accessory gene flow occurs more often within Gardnerella species than between species. The distributions of nucleotide diversity (π) per shared accessory gene within and between G. vaginalis and G. piotii, were calculated and log transformed. The box spans the interquartile range, the median is represented by the middle line, and the whiskers extend to ±1.5 times the interquartile range. Data beyond the end of the whiskers are outlying points and plotted individually. Average gene π values differ significantly by group (Kruskal-Wallis test, H = 214.7, p < 2.2e-16). The distributions of average gene π values of G. vaginalis (Mann-Whitney-Wilcoxon, W = 373,928, p < 3.322e-13) and G. piotii (Mann-Whitney-Wilcoxon, W = 399,762, p < 3.37e-13) are lower than between species. In species that regularly exchange accessory gene content, we might expect to see similar levels of diversity in shared accessory gene content. However, we observed lower diversity within species, consistent accessory gene flow occurring more often within G. vaginalis and G. piotii than between them.

Figure 12

G. vaginalis has a larger accessory genome than G. piotii. Rarefaction curves of core and total gene content. G. vaginalis isolates were subsampled to the number of G. piotii isolates, and both species were iteratively sampled 100 times. The median value is shown. G. vaginalis has a larger accessory genome than G. piotii, consistent with acquisition of novel gene content from a diverse pool.