Comparative genomics reveals high rates of horizontal transfer and strong purifying selection on rhizobial symbiosis genes

Abstract

Horizontal transfer (HT) alters the repertoire of symbiosis genes in rhizobial genomes and may play an important role in the on-going evolution of the rhizobia–legume symbiosis. To gain insight into the extent of HT of symbiosis genes with different functional roles (nodulation, N-fixation, host benefit and rhizobial fitness), we conducted comparative genomic and selection analyses of the full-genome sequences from 27 rhizobial genomes. We find that symbiosis genes experience high rates of HT among rhizobial lineages but also bear signatures of purifying selection (low Ka : Ks). HT and purifying selection appear to be particularly strong in genes involved in initiating the symbiosis (e.g. nodulation) and in genome-wide association candidates for mediating benefits provided to the host. These patterns are consistent with rhizobia adapting to the host environment through the loss and gain of symbiosis genes, but not with host-imposed positive selection driving divergence of symbiosis genes through recurring bouts of positive selection.

1. Background

In bacterial populations, horizontal transfer (HT) is an important mechanism for introducing potentially adaptive genetic variation [1]. The potential advantage of horizontally transferred genes is that they can confer an adaptive phenotype more rapidly than might be possible through the accumulation of mutations within a single lineage. Such phenotypes may even be inaccessible through step-wise mutational processes, either because the population might go extinct in the time it would take for the new phenotype to evolve or because acquiring the phenotype would require a population to cross an adaptive valley [2].

There is strong evidence for the between-rhizobia transfer of at least some of the genes needed to establish a functional symbiosis between rhizobial bacteria and legume hosts [3]. Rhizobia are soil bacteria that form a facultative (i.e. can survive without the host) resource-based mutualism with legume host plants by infecting the roots of compatible legume species. Infected roots form structures called nodules, and inside of the nodules rhizobia convert atmospheric nitrogen into a plant-usable form. In exchange for this nitrogen, the bacteria obtain carbon resources from the host plant. However, not all rhizobia can form an effective symbiosis with all legumes. Rather, the ability of a rhizobium to form a symbiosis with a particular host is determined by a series of host- and rhizobia-specific nodulation molecules [4,5]. Functional analyses also have identified and characterized many of the genes that are needed for N-fixation and resource exchange after a nodule is formed [6].

Strict coevolution of rhizobia and legumes would result in parallel cladogenesis. However, host and bacterial phylogenies reveal evidence for extensive host switching. Rhizobia from some phylogenetically close lineages can form a symbiosis with distantly related plant hosts, and members of some distantly related bacterial lineages can form a symbiosis with the same host [7]. The non-parallel cladogenesis of legumes and rhizobia is suggestive of HT of symbiosis genes. This has been confirmed by multiple examples of individual symbiosis genes having relatedness that differs from that of the rhizobial species in which they are found [8–13]. Although there are many examples of the HT of symbiosis genes, we know relatively little about the evolutionary dynamics of HT among rhizobia, beyond that it occurs and can be important for shifting the host range.

Here, we use comparative genomic analyses to gain insight into the evolution of symbiosis genes and, by extension, the evolution of symbiosis. We had three main objectives. The first objective was to characterize the extent of HT of symbiosis genes. Certainly, rhizobia acquire some of their repertoire of symbiosis genes from other lineages [3]. Given that forming a symbiosis is expected to confer a fitness advantage, that HT potentially offers an opportunity for a rhizobium to expand its range of compatible hosts, and that many nodulation genes are found near insertion sequences or on plasmids [3,14–17], we expect that symbiosis genes are transferred at a higher rate than non-symbiosis genes. However, systematic genome-wide estimates of the HT rates of symbiosis relative to non-symbiosis genes across a wide range of rhizobial lineages are lacking.

Our second objective was to determine whether HT affects the evolution of some aspects of the symbiosis more than others. If HT primarily enables a symbiont to expand or alter its host range, then we would expect symbiosis genes acquired from different lineages would be over-represented by genes that are central to nodule formation. Alternatively, if post-establishment processes (e.g. fixation of nitrogen) are most affected by HT, then we would expect to see stronger evidence of HT among genes involved in N-fixation or in the exchange of benefits between host and bacteria.

Our third objective was to examine how selection has contributed to the divergence of symbiosis genes. Symbiosis genes might experience repeated bouts of positive selection in response to selection imposed by hosts to exclude less beneficial symbionts [18,19]. If this is the case, we would expect that symbiosis genes harbour signatures of positive selection having driven their divergence. Alternatively, symbiotic partners might be at an evolutionary stasis [18,20] because they are at or near a selective optimum. In this case, symbiosis genes would be expected to experience purifying selection, as has been found in the nod region in Ensifer medicae [9] and type III effector proteins in Bradyrhizobium japonicum and E. fredii [19].

To address these objectives, we analysed publicly available full-genome sequences from 27 species of alphaproteobacterial rhizobia. We examine four sets of symbiosis genes: those annotated as having a role in nodule formation, those annotated as involved in N-fixation, genes identified by genome-wide association (GWA) analyses as candidates underlying variation in the benefits plant hosts obtain from rhizobia, and GWA-identified candidates underlying variation in host-associated rhizobial fitness. For each group of genes, we estimate the phylogenetic signal of gene presence, the extent of HT among lineages, and past selection as measured by the rate of non-synonymous to synonymous nucleic acid changes (Ka : Ks).

2. Methods

We identified genomes for analysis by searching the NCBI Genome database for genomes from the named species or genera in the alpha- and betaproteobacterial rhizobial clades (species names from [21]). Because our objectives are to estimate the extent of HT among rhizobia, we include genomes from a widely related set of rhizobia species. We then used the NCBI ‘CladeIDs’ to identify a representative strain from every clade in these species or genera. In cases with different named species or biovars sharing the same CladeID (E. americanum CCGM7 and E. fredii NGR234; R. phaseoli N161, R. esparanzae N561, and biovars of R. etli; biovars of R. leguminosarum; and the three Mesorhizobium species), we included a representative of each of the named taxa (electronic supplementary material, table S1). We also included two genomes obtained from MaGe (mage.genoscope.cns.fr), Ensifer terangae (USDA4894) which was not in NCBI and E. meliloti USDA1106 because it was used in previous GWA analyses [22]. Our initial search identified 74 genomes. After running OrthoFinder through the ‘orthogroup' stage (see below), we excluded 43 genomes because manual inspection revealed that their assemblies were incomplete. Because we identified only three betaproteobacterial genomes, and beta and alpha lineages are highly diverged (core gene mean protein distances between Alphaproteobacteria strains = 0.36, and between Alpha- and Betaproteobacteria strains = 0.98), we limited our analyses to Alphaproteobacteria. After filtering, we were left with assemblies of 27 genomes (21 named species and 4 strains identified only to genus, electronic supplementary material, table S1), representing the major clades of alphaproteobacterial rhizobia (figure 1). These species include E. fredii NGR234, which can form nodules with members of 112 legume genera [23], and WSM2073, which resulted from the HT from a commercial inoculant to a native strain [24] and forms nodules with two species [25].

Figure 1.

Relationships among analysed strains and per genome number of symbiosis genes (one gene per column). The phylogenetic tree was constructed from concatenated single-copy core gene protein sequences using IQ-tree with the JTT + G4 model of evolution, scale bar is amino acid substitutions per site. Genus names abbreviated (in order from top to bottom: M. Methylobacterium, A. Azorhizobium, B. Bradyrhizobium, E. Ensifer, A. Agrobacterium, N. Neorhizobium, R. Rhizobium, M. Mesorhizobium). Genes in multiple categories are marked with a black oval. (Online version in colour.)

We identified homologous protein-coding genes by using OrthoFinder (v. 2.2.7, [26]) with the default settings except that the MCL (Markov Cluster algorithm [27]) inflation factor was set to 4. OrthoFinder groups genes by running all-versus-all BLAST (using diamond [28]), then using MCL [27] to form ‘orthogroups' on the basis of sequence similarity and then uses rooted, species-tree-reconciled gene trees (constructed using distance-based algorithms in FastME v. 2.1.5 [29] and STride [30] to group sequences. This analysis identified 31 853 phylogenetic clusters, hereafter referred to as genes. We aligned the amino acid sequences of each gene using Muscle v. 3.8.31 [31], then converted alignments to nucleotide sequences. The NCBI annotation indicated that 6878 sequences were pseudogenes and we removed these before subsequent analyses, leaving 29 671 genes and 176 566 sequences.

(a). Symbiosis genes

We identified four groups of symbiosis-related genes—two based on functional annotation and two based on genome-wide association (GWA) studies linking bacterial genotype to the benefit that either plants or rhizobia derived from symbiosis. We identified the annotated genes by searching the annotated gene names and product descriptions for: nod, noe, nol, nop, nfe, nodul, nif, fix, fixation and nitrogenase (see electronic supplementary material, table S2 for references and information on search terms). These searches identified 148 genes: 65 annotated nodulation genes, 55 annotated nitrogen fixation genes and 28 genes we removed because their relationship to symbiosis was unclear (details in electronic supplementary material, table S3).

The GWA candidates were from two studies with E. meliloti bacteria and M. truncatula hosts [22,32]. From these studies, we identified the 10 candidates showing the strongest statistical support for contributing to among-strain variation in the benefits plant hosts obtained in single-strain inoculations [32] (hereafter referred to as ‘host benefit’ genes) or contributing to variation in nodule-associated rhizobial fitness [22] (hereafter referred to as ‘rhizobia fitness’ genes). Due to linkage disequilibrium in the Ensifer panel used for the GWAS, the 10 rhizobia fitness candidates comprise 21 candidate genes (table 1; five rhizobia fitness candidates identified through GWA analyses were not included because they were in large LD groups, each of which harboured more than six genes). We note that the function of the genes identified by GWA may differ among lineages and may not be related to symbiosis outside of Ensifer. The rhizobia fitness genes and host benefit genes partially overlapped with the nodulation and fixation genes, but did not overlap with each other (table 1). Electronic supplementary material, table S3 contains the full list of the analysed symbiosis genes.

Table 1.

Number of genes in the 27 representative genomes and number of symbiosis genes.

	all genes	nodulation	N-fixation	host benefit	rhizobial fitness
number of genes	29 671	65	55	10	21
proportion genes that are single-copy	0.86	0.54	0.51	0.40	0.43
median family size	2	8	27	19.5	18
proportion of genes in all genomes	0.04	0.17	0.40	0.50	0.24
prop. of genes found in one genome	0.48	0.05	0.07	0.10	0.05
median number of genomes	2	6	22	19	17
mean copies/genome	1.07	1.22	1.26	1.37	1.45
number of genes also found in
nodulation		—	0	2	0
N-fixation			—	4	2
host benefit				—	0

(b). Phylogenetic distribution and horizontal gene transfer

We used three approaches to characterize the extent of HT. First, we examined the correlation between the amino acid divergence of symbiosis genes to the average inter-species phylogenetic distance, estimated as the median amino acid divergence (pairwise protein distances estimated using the JTT model in FastME v. 2.1.5 [29]) of 830 single-copy genes found in all 27 genomes. For these analyses, if a symbiosis gene was not found in a lineage, that lineage was excluded from the analysis. For lineages that contained multiple gene copies we included pairwise distances between all copies between species, but because we are interested in inter-species distance we did not calculate the pairwise distances between copies within a species. A high correlation between protein distance and species distance would be consistent with no HT or HT occurring only between closely related lineages. By contrast, a low correlation between gene and species distance would be consistent with extensive HT or highly variable evolutionary rates within lineages.

Second, we used the δ statistic of [33] to estimate the strength of the phylogenetic signal of gene presence. δ provides insight into whether the presence, or number of copies, of a gene is phylogenetically conserved but is not a specific test of HT; low δ could mean that genes have been lost from multiple strains, independent of those strains' relatedness whereas high δ could result from extensive loss and gain.

Finally, we used GeneRax [34] to specifically estimate the rate of HT of each gene. GeneRax is a species-tree aware method that estimates the rates of gene duplication, transfer and loss. For both δ and GeneRax analyses, the species tree was constructed using IQ-tree v. 1.6.10 [35] with the JTT + G4 model of protein evolution. IQ-tree was run on the concatenated alignments of the single-copy core genes that were trimmed with trimAl v. 1.4.1 [36]. The tree was manually rooted using FigTree v. 1.4.2 based on the positions of lineages in the OrthoFinder species tree that used Betaproteobacteria as an outgroup.

(c). Evolutionary rates

We estimated the rate of non-synonymous to synonymous site divergence (Ka : Ks) for each pair of sequences, including pairs found in the same genome using Comeron's method [37] implemented in the gestimator program in libsequence (v. 0.8.2, [38]). Sequences were excluded if Ks was too high for gestimator to estimate the number of substitutions (399 620 pairs in 7556 genes out of 2 076 943 pairs in 15 636 genes) or Ks = 0 (4977 pairs in 2225 genes); the sequence pairs with unestimated Ks values also had substantially higher Ka values (median = 0.71) than other sequence pairs (median = 0.35). In the text, we report Ka : Ks values only for genes with greater than or equal to 3 comparisons, although results were qualitatively similar when genes with only a single pair of sequences were included (electronic supplementary material, table S4). Summary statistics and annotations for every gene can be found in electronic supplementary material, table S5.

(d). Comparison data

We compared the phylogenetic distribution and evolutionary rates for each set of symbiosis genes to randomly selected sets of the same number of genes as well as to samples of non-symbiosis genes that were ‘matched' to be found in the same number of strains and have a similar number of copies per strain as the symbiosis genes. Using this matched set accounts for differences that might arise due to variation in gene representation. Comparison to randomly selected sets of genes, for which there is no control for gene occurrence, are shown in electronic supplementary material, table S4.

3. Results

The annotated protein-coding sequences in the 27 alphaproteobacterial genomes we analysed were grouped by OrthoFinder into 29 671 genes, 1216 of which were found in all 27 genomes, and 14 249 of which were found in only one genome (table 1). Relative to the entire genome, the symbiosis genes (electronic supplementary material, table S3) were more likely to be present in all 27 genomes (37 of 143 symbiosis genes, Fisher's exact test p < 0.001), less likely to be found in only a single genome (9 of 143, p < 0.001), and found in higher copy number in the genomes in which they were found (1.26 copies of symbiosis genes versus 1.1 copies of other genes per genome, p < 0.001; table 2). The higher mean copy number was driven by a few genes with very high copy number; both the symbiosis and non-symbiosis genes had a median of one copy per genome.

Table 2.

Symbiosis genes tend to be present in more genomes, have higher copy number, higher rates of horizontal transfer than other genes and less phylogenetic signal in gene presence/absence (lower δ) than non-symbiosis genes. Median values are shown. Duplication and transfer are estimates from GeneRax for genes with at least four genomes represented and at least six total copies. δ and Ka : Ks calculated only for genes with at least three sequences (numbers of gene in parentheses; results including all genes were similar—see electronic supplementary material, table S4). The Welch p-value is from a t-test comparing nodulation to fixation genes. The lower part of the table shows the probability that values are greater than a random sample of genes found in the same number of genomes and having approximately the same number of copies as the symbiosis genes.

	strains	copies/genome	duplication	transfer	median pairwise Ka : Ks	R2	δ
non-symbiosis	2	1	10−7	0.08	0.22 (n = 11 152)	0.76	3.3 (n = 9737)
nodulation	6	1	10−7	0.24	0.20 (n = 48)	0.23	1.6 (n = 35)
fixation	22	1	10−7	0.14	0.20 (n = 47)	0.54	2.2 (n = 23)
p-value Welch	<0.001	0.55	0.17	<0.001	0.9	0.007	0.56
benefit	19	1.08	0.002	0.20	0.15 (n = 7)	0.29	1.1 (n = 3)
fitness	17	1.13	10−7	0.18	0.23 (n = 19)	0.19	1.9 (n = 12)
probability of the estimated values being less than random samples of genes (%)
nodulation			0.2	0	99.7	99.7	99.9
fixation			0	0	99.2	88.6	99.2
benefit			6.7	0.6	99.6	96.5	95.9
fitness			0.8	4.1	73.4	85.5	94.5

Among the symbiosis genes, N-fixation genes were more likely to be found in all genomes than nodulation genes (tables 1 and 2; figure 1), about as likely to be found in only a single genome (4 fixation, 3 nodulation, p = 0.70), and harboured a similar number of copies in the genomes in which they were found (means of 1.26 fixation, 1.22 nodulation, p = 0.55; table 2). The number of nodulation gene copies in a genome was positively correlated with genome size (r = 0.56, p = 0.002) and the total number of annotated genes (r = 0.55, p = 0.003), but the number of fixation genes was not (r = 0.09, 0.17, respectively, both p > 0.3). The genome size and nodulation gene count correlation may reflect among-lineage variation in the efficiency of removing non-essential/non-advantageous genes. The host benefit and rhizobia fitness candidates were more likely to be found in all 27 strains than randomly selected genes (table 1; both p ≤ 0.001). They also tended to be found in greater copy number in the genomes in which they were present (tables 1 and 2; electronic supplementary material, table S4).

(a). Extent of HT

If genes are either not transferred between lineages or lost from multiple strains, we would expect gene presence to have a strong phylogenetic signal and divergence at the sequence level to be strongly correlated with genome-wide divergence. We used the δ statistic [33], to estimate the strength of the phylogenetic signal in gene presence. All four classes of symbiosis genes had lower median δ values than comparable sets of non-symbiosis genes (table 2; electronic supplementary material, table S4; figure 2). The amino acid divergence of symbiosis genes also tended to be less strongly correlated with genome-wide between species divergence than non-symbiosis genes (table 2).

Figure 2.

Distribution of median δ (a; only genes with greater than or equal to 3 sequences) and median log₂ Ka/Ks (b; only genes with greater than or equal to 3 comparisons) of 1000 randomly sampled sets of genes (bars) compared to the median values for symbiosis genes (dashed lines). Random samples are of genes found in the same number of strains and having similar copy number as symbiosis genes. δ is a measure of phylogenetic signal in the distribution among genomes with greater numbers indicating greater signal. (Online version in colour.)

The weaker phylogenetic signal of gene presence and the weaker correlations between gene and species divergences are both suggestive of symbiosis genes experiencing more HT than non-symbiosis genes. Estimates of the rate of HT, obtained from GeneRax, are consistent with this suggestion: median HT estimates of symbiosis genes were more than twice the median estimated rate of HT of non-symbiosis genes (table 2). The estimated HT rate of annotated nodulation genes was nearly twice as great as the rate obtained for annotated fixation genes (t-test, p < 0.001). Moreover, only four nodulation genes, of the 38 for which HT could be estimated, had rates of HT below the median of non-symbiosis genes and nearly 25% of the nodulation genes with estimated transfer rates were among the upper 10% of the non-symbiosis genes. The estimated duplication rates of symbiosis genes are also greater than that of non-symbiosis genes that are found in similar numbers of strains and copies as the symbiosis genes (table 2; electronic supplementary material, table S4).

(b). Evolutionary rates

The low phylogenetic signal, high estimates of HT, and high rates of duplication of annotated symbiosis genes (nodulation and N-fixation), suggest that these genes have highly dynamic evolutionary histories, at least from the perspective of their loss and gain from genomes. To determine whether these genes also experience atypical evolutionary rates at the sequence level, we examined the ratio of non-synonymous substitutions per non-synonymous site to synonymous substitutions per synonymous site, Ka : Ks. We found that symbiosis genes tend to harbour signatures of purifying selection that are stronger than the non-symbiosis genes (lower median Ka : Ks, p < 0.01 for all but the rhizobial fitness genes; table 2; electronic supplementary material, table S4). However, the mean Ka : Ks of symbiosis genes was slightly greater than the mean of the empirical null expectations (electronic supplementary material, table S4), although this difference was not unlikely by chance. The lower median, yet slightly higher mean, is consistent with most symbiosis genes having experienced stronger than average purifying selection but some genes having evolved under more relaxed or possibly even positive selection (electronic supplementary material, figure S2). In fact, four symbiosis genes have Ka : Ks values that are among the highest 0.2%: noeE (ortholog ID OG0006225), nolE (OG0014065), a nitrogenase (OG0014063) and fixK (OG0014068). The lower median Ka : Ks of the symbiosis genes is not because symbiosis genes are more likely to be transferred and transferred genes tend to be more highly constrained. Consistent with [39], genome-wide there is a positive correlation between the rate of transfer and Ka : Ks (r_{N = 7068} = 0.14) indicating that genes with higher transfer rates tend to have weaker signatures of purifying selection.

4. Discussion

HT is an important mechanism by which bacterial lineages gain adaptive genetic variants. In rhizobial bacteria, many symbiosis genes are on plasmids or flanked by insertion elements [3,14–17], suggesting that these genes experience high rates of HT. This expectation is supported by comparative phylogenetic analyses, sequence analyses of a handful of symbiosis genes, and genomic analyses of closely related lineages (e.g. [3,13]). Here, we expand upon those studies by estimating the extent of HT among a wide range of rhizobial lineages and determining whether different stages of the rhizobia–legume symbiosis are differentially affected by HT. Using publicly available, fully assembled and annotated genome sequences from 27 rhizobial strains, we found that symbiosis genes show a low correlation between the divergence of gene sequences and the genomes in which those genes are found, weak phylogenetic signal of occurrence, and high rates of HT among genomes. The genes annotated as involved in nodule formation are particularly dynamic, showing rates of transfer nearly three times that of non-symbiosis genes, more than 50% greater than that of N-fixation genes, and more than 20% greater than rhizobia genes that GWA analyses identified as affecting the benefits hosts derive from symbiosis and host-associated rhizobial fitness.

The high rates of HT of symbiosis genes suggests that symbiosis genes, particularly nodulation genes, experience environmentally dependent selection. Given that the function of at least some symbiosis genes is host-specific [40–43], environmentally dependent selection may not be unexpected. In some environments, presumably when specific hosts are present, these genes probably confer a selective advantage because they allow rhizobia to form a symbiosis (e.g. transfer of a symbiosis island from a commercial inoculant to M. australicum WSM2073 [24]). When compatible hosts are absent, however, the symbiosis genes may be lost (e.g. [44–46]), either through drift or because these genes are environmentally costly in the absence of a compatible host. The changes in host availability may change from site to site [20], from season to season in locations with mixed-species communities [47], and over longer time periods as plant ranges shift [48,49]. High rates of HT also may result from a combination of the selective advantage of losing symbiotic capability in certain abiotic environments or when hosts are not present (e.g. [44,45]) and the selective advantage of regaining symbiotic ability when host plants are present. Two population genetic studies provide support for environmentally dependent selection acting on symbiosis genes; a frequently introgressed region containing several symbiosis genes bears a signature of partial sweeps in clover-nodulating R. leguminosarum in Europe [13] and a genomic region in R. leguminosarum that bears a strong signature of adaptation in response to long-term N addition harbours several symbiosis genes as well as several presence–absence variants [45].

(a). Nodulation genes experience stronger purifying selection

Whereas symbiosis genes are clearly evolutionarily labile, in the sense that they are transferred among lineages at a high rate, most are conserved at the nucleotide level and bear a signature of purifying selection that is slightly stronger than the genome-wide average. Although identifying the sources of selection acting on rhizobial genes would require experimental work, it seems likely that host-imposed selection shapes the evolution of these genes. Purifying selection on nodulation genes likely reflects constraints related to the signalling molecules exchanged between hosts and rhizobia during nodule formation [4,5,43] (i.e. mutations that affect proteins involved in nodule formation may reduce the probability of forming a nodule). Similarly, purifying selection on the N-fixation and plant benefit genes suggests that mutations in genes that alter the exchange of benefits plants receive are selected against, which would be consistent with the action of sanctions [50] against rhizobia with reduced benefits.

There has been debate about whether symbiotic partners impose positive selection on one another, due to an arms-race scenario in which hosts impose selection for more beneficial symbionts or whether there is evolutionary stasis due to symbiotic partners being at local optima [18]. Our analyses do not allow us to make inferences about the fitness of symbionts or hosts, but they do show that the vast majority of genes we examined have not diverged in response to positive selection. Our data, therefore, seem to argue against arms-race coevolution being important in shaping the evolution of most genes involved in the legume–rhizobia symbiosis. The lack of positive selection on symbiosis genes is consistent with several studies that have compared signatures of selection in mutualistic and pathogenic lineages: an outer membrane protein that acts as an antigen in Rickettsiaceae showed evidence of positive selection in pathogenic but not in mutualistic lineages [51]; both the distribution among strains and the amino acid sequences of type III secretion system and effector proteins, which are involved in host interaction, are more conserved in Bradyrhizobium and Ensifer rhizobia than Pseudomonas plant pathogens [19]; and Rhizobium sp. sym plasmid genes appear to experience high rates of HT and purifying selection [12].

Although we do not find an overall signal of positive selection having driven the evolution of symbiosis genes, four symbiosis genes harbour Ka : Ks values that exceeded all but 0.2% of the non-symbiosis genes (electronic supplementary material, figure S2), suggesting that host-imposed positive selection may drive adaptation in these genes. Of course, we also cannot exclude the possibility that positive selection is important in driving the evolution of symbiosis genes not included in our analyses. Many of the genes we analysed were identified through forward genetic screens and likely play central roles in the establishment of a functional symbiosis. It is possible that other genes play important roles in shaping symbiont fitness and bear signatures of selection distinct from those characterized here. Finally, it is important to acknowledge that Ka : Ks will only identify genes that have experienced repeated bouts of adaptation [52]. The signal of purifying selection from Ka : Ks is, therefore, not incompatible with occasional bouts of positive selection within a population that drive a rapid increase in the frequency of alleles recently introgressed from other lineages (e.g. [13]). Due to their frequent transfer, the evolution of symbiosis genes may be de-coupled from the evolution of the populations of rhizobia in which they are found. Nevertheless, we found little evidence for antagonistic coevolution between legumes and rhizobia.

Although we have only looked at the symbiont here, a full understanding of host-symbiont coevolution would also consider the host. There is some evidence for selection on legume symbiosis genes from intraspecific studies in M. truncatula, but these signatures do not appear to be pervasive [53–55]. Moreover, the patterns we find may be dependent on the scope of sampling. We examined a wide range of rhizobial species with broad host ranges. A more narrow sampling of rhizobial diversity, or rhizobia that are all found on a single legume host might reveal different patterns of HT. For example, Tian et al. [56] found that symbiosis gene content overall is reflective of phylogeny, although even their restricted sample revealed evidence of HT of some symbiosis genes (e.g. nodC) and variation of symbiosis gene copies, with many symbiosis genes found in only a few genomes.

Data accessibility

Code and data are available in the electronic supplementary material and from the Dryad Digital Repository: https://doi.org/10.5061/dryad.2fqz612n2 [57].

Authors' contributions

Both B.E. and P.T. were actively engaged in all aspects of this work.

Competing interests

We declare that we have no competing interests.

Funding

This work was supported by the National Science Foundation under grant nos. IOS-1724993 and IOS-1856744.