Metapopulation dominance and genomic-island acquisition of Bradyrhizobium with superior catabolic capabilities

Abstract

Root nodule-forming rhizobia exhibit a bipartite lifestyle, replicating in soil and also within plant cells where they fix nitrogen for legume hosts. Host control models posit that legume hosts act as a predominant selective force on rhizobia, but few studies have examined rhizobial fitness in natural populations. Here, we genotyped and phenotyped Bradyrhizobium isolates across more than 800 km of the native Acmispon strigosus host range. We sequenced chromosomal genes expressed under free-living conditions and accessory symbiosis loci expressed in planta and encoded on an integrated ‘symbiosis island’ (SI). We uncovered a massive clonal expansion restricted to the Bradyrhizobium chromosome, with a single chromosomal haplotype dominating populations, ranging more than 700 km, and acquiring 42 divergent SI haplotypes, none of which were spatially widespread. For focal genotypes, we quantified utilization of 190 sole-carbon sources relevant to soil fitness. Chromosomal haplotypes that were both widespread and dominant exhibited superior growth on diverse carbon sources, whereas these patterns were not mirrored among SI haplotypes. Abundance, spatial range and catabolic superiority of chromosomal, but not symbiosis genotypes suggests that fitness in the soil environment, rather than symbiosis with hosts, might be the key driver of Bradyrhizobium dominance.

1. Introduction

Proteobacteria in the genus Bradyrhizobium are among the most cosmopolitan bacteria, thriving in multifarious free-living habitats and associating with diverse hosts [1–3]. Like other rhizobia, Bradyrhizobium spp. often exhibit a bipartite life cycle, alternating between free-living replication in the soil and symbiotic differentiation and N₂ fixation within the root nodules of legume hosts [4]. Bradyrhizobium and other rhizobia also exhibit bipartite genomes, with chromosomal loci largely expressed under aerobic free-living conditions, and symbiosis loci mainly expressed in planta [5,6]. Rhizobial symbiosis loci are grouped on megaplasmids or genomic islands that are transmitted horizontally between chromosomal backgrounds [7–10]. A key characteristic of both agricultural and natural rhizobial populations is that they are often overrepresented by one or few rhizobial genotypes [4,11,12], in some cases with evidence that a subset of genotypes have rapidly increased in frequency (i.e. selective sweeps [13]). Yet, relatively little is known about what factors might drive such variation in rhizobial genotypic frequencies.

The longstanding ‘host control’ paradigm of symbiosis predicts that hosts are a dominant selective force shaping populations of associated symbionts. Host control models of symbiosis posit that hosts must exhibit mechanisms to constrain exploitation in their associated symbionts, for instance via discrimination among symbionts during initial host colonization or via within host control over symbiont proliferation [14–17]. Consistent with these models, legume hosts have been shown to select certain rhizobial genotypes over others for nodulation [18,19] and favour beneficial rhizobia over less-effective strains in planta [20–23]. However, ex planta selection on rhizobia has remained poorly understood [4,24]. In particular, repeated attempts to leverage legume host control traits to use rhizobia as agricultural bioinoculants have revealed the so-called ‘rhizobial competition problem’, where highly beneficial introduced strains fail to compete against native strains and thus are unable to successfully invade populations [25–27]. The common failure of crop legumes to enrich soil populations with beneficial inoculated strains suggest that variation in rhizobial fitness in the soil might be able to overwhelm the hosts' ability to select beneficial strains [25,26,28].

Here, we investigated the population genetic structure of Bradyrhizobium isolated from a metapopulation of Acmispon strigosus (formerly Lotus strigosus), a common herb native to the southwestern USA. Bradyrhizobium thrives in soil and aquatic environments and colonizes wild legumes and major staple crops [1,2,4,19,29–32]. We genotyped Bradyrhizobium from more than 350 A. strigosus nodules from nine natural sites across California encompassing 72 plants collected over an 840 km range. Building on past studies focused on chromosomal loci [4,12], we sequenced eight loci (approx. 5.5 kb) distributed across the approximately 9 Mbp Bradyrhizobium genome including four loci on the chromosome (i.e. CHR) and four loci within the symbiosis island (i.e. SI), a large (approx. 680 kB) integrated genomic island that encodes nodulation and nitrogen fixation function [8,33]. To generate predictions about the relative importance of the soil versus the plant host in structuring rhizobial populations, we compared population genetic parameters of CHR and SI loci and examined rates of recombination across these two genomic regions. To analyse functional differences among Bradyrhizobium genotypes that could drive patterns of genotype frequency and spatial range, we chose 20 focal strains and used phenotypic microarrays to analyse utilization of 190 sole-carbon sources that are ecologically relevant to soil bacteria [34]. We investigated the fit of our dataset to four hypothetical scenarios of genome evolution, including: (i) selective sweeps restricted to the SI, predicted if host plants select for symbiosis loci that recombine among diverse CHR backgrounds [35]; (ii) selective sweeps restricted to the CHR, predicted if ex planta conditions select on CHR loci which acquire diverse SI genotypes; and (iii) whole-genome selective sweeps, predicted if selection affects both genome regions without recombination [36]. Finally, a scenario of stasis and whole-genome linkage is predicted in the absence of selective sweeps or recombination [37].

2. Material and methods

(a). Collection of nodule isolates

Root nodules were collected from A. strigosus host plants at nine field collection sites across California covering more than 800 km of the host's range, and including collections from previous studies [4,12]. Previous work showed that these sites vary in key soil parameters such as total soil nitrogen and mineral nitrogen content [21]. From each of the 72 plants collected, we sampled 1–26 nodules (mean approx. 5/plant). From each sampled nodule, we isolated a single clone of Bradyrhizobium following published protocols [4].

(b). DNA amplification and sequencing

Genomic DNA extracts were purified, PCR amplified and sequenced at four loci located on the Bradyrhizobium chromosome (CHR), including dnaK, glnII, ITS and recA, and four SI loci, including nifD, nodD-A, nodZ and nolL. PCR amplification followed previously published protocols [4,29,30,38–43].

(c). Phylogenetic analyses

Sequences for each gene were aligned and analysed separately or were concatenated per genomic region with Clustal Omega [44] including reference sequences from diverse Bradyrhizobium spp. and Mesorhizobium loti (which were used as outgroups; electronic supplementary material, S1). We used Akaike information criterion results from jModelTest 2 [45,46] to select the GTR model of nucleotide substitution all loci. Phylogenetic trees were reconstructed in PhyML v. 3.0 [47] using BioNJ as the starting tree with subtree pruning and regrafting. Branch support was estimated using the fast approximate likelihood ratio test (aLRT) with the Shimodaira–Hasegawa-like (SH-like) procedure [48]. Based on the CHR phylogeny, species-like clades of Bradyrhizobium were defined as highly supported, non-nested, monophyletic groups (SH support greater than or equal to 0.90) including no more than one reference species, attempting to follow past species demarcations that used some of the same loci [31,49]. A tanglegram associating phylogenetic trees of each genomic region was reconstructed in Treemap v. 3.0 [50]. Statistical significance of congruence between CHR and SI phylogenies was tested using AxParafit and AxPcoords [51] within CopyCat [52] using default parameters.

(d). Statistical analyses

Loci were analysed separately or grouped into genomic regions, recognizing the potential for horizontal transfer of the SI [35,39,53]. We estimated π (nucleotide diversity; [54]), H_d (haplotype diversity; [54]), k (average number of nucleotide differences per site; [55]), K_a/K_s (ratio of non-synonymous to synonymous substitutions; [56]), average absolute D’ (linkage disequilibrium) [19,57], R (recombination) [58], minimum number of recombination events [59] and Tajima's D [55]. Population differentiation was calculated using F_ST conducted on a base-pair basis using a Perl script and the Weir–Cockerham method [60]. Average F_STvalues were calculated between all pairs of collection sites for each genomic region. Isolation by distance was tested with a Mantel test [61] correlating F_ST and distance matrices between collection sites in PASSaGE [62]. We analysed collection sites and identified Bradyrhizobium clades separately when appropriate. We analysed inter-clade variation using the ratio of fixed to shared polymorphisms using DnaSP [63].

We identified isolates with identical haplotypes within one or both genomic regions using the ‘find redundant’ command in MacClade [64]. For each haplotype, we calculated raw abundance (number of times a haplotype was isolated), and an adjusted abundance (only counting identical haplotypes from unique GPS locations), which discounts repeated isolation of the same haplotype from closely neighbouring plants. Distances within collection sites were small (less than 300 m), thus we used the geographical midpoint at each collection site to calculate distance among collection sites. Measures of strain richness and dominance were estimated, which are akin to species richness and dominance [65,66]. Strain richness was calculated for each locus and genomic region by dividing the number of haplotypes by the number of isolates collected [11].

Within each genomic region, we analysed the number of times each haplotype was isolated (i.e. abundance; [11]) and the percentage of isolates each haplotype encompassed (i.e. strain dominance; [11]). Haplotypes were defined as ‘dominant’ at a field collection site if they constituted at least 10% of the isolates at that site (only including sites with greater than or equal to 50 isolates) and were defined as ‘epidemic’ if they were dominant in at least one site and were also found to have spread among multiple sites separated by at least 10 km. In population genetics, the term epidemic is used to describe microbial genotypes that reach extremely high frequency in populations composed mostly of rare genotypes [67,68].

(e). Carbon source utilization assays

We assayed growth upon 190 sole-carbon sources using phenotypic microarrays (PM1, PM2; Biolog Hayward, CA, USA). Carbon sources were categorized into amines and amides (n = 11), amino acids (n = 32), carbohydrates (n = 59), carboxylic acids (n = 55), polymers (n = 13) and miscellaneous (n = 20, [69]). The tested strains were incubated on modified arabinose gluconate (MAG) agar plates (29°C, approx. 96 h, approx. four plates per stain), plates were scraped and cultures individually re-suspended in liquid MAG [4], cell density was measured optically, cells were washed twice in sterile phosphate buffered saline buffer, then re-suspended in Biolog Buffer IF-O (with tetrazolium dye) and pipeted into phenotypic microarrays in duplicate (approx. 3.0 × 10⁷ cells well⁻¹; 100 µm). Microarrays were incubated (29°C, 120 h), absorbance was read at 570 nm using a Victor 2 plate reader, and readings were averaged among strain replicates. We employed a binary measure of carbon source utilization in which growth was considered positive with at least twofold absorbance relative to control wells (no carbon source). We first analysed these response variables using ANOVAs with strain and haplotype dominance as fixed effects, and treating strains as independent samples. We also analysed the response variables using haplotype (genotypic) means, which is more conservative and avoids phylogenetic non-independence among resampled haplotypes.

(f). Testing evolutionary-genomic scenarios

We tested for population genetic data consistent with four evolutionary-genomic scenarios including selective sweeps restricted to the SI (SI sweep; e.g. [35]), selective sweeps restricted to the CHR (CHR sweep), whole-genome selective sweeps (CHR-SI sweep; e.g. [36]), and stasis and whole-genome linkage (Stasis; e.g. [37]). Selective sweeps are predicted to result in reduced genetic diversity in the genome region where they occur [70,71]. To discriminate among these models, we compared GC%, π, haplotype number, H_d, strain richness, linkage and Tajima's D within and among genomic regions (see the electronic supplementary material, S2 for detailed predictions).

3. Results

(a). Genomic region evolution

We examined similar numbers of nucleotides and variable sites in the CHR and SI loci, but nonetheless these genome regions were characterized by dissimilar population genetic parameters. The CHR exhibited many fewer haplotypes (CHR, 138; SI, 225), lower strain richness (CHR, 0.39; SI, 0.63) and haplotype diversity than the SI (CHR, 0.947; SI, 0.993). The CHR also had slightly greater nucleotide diversity (CHR, 0.03; SI, 0.02) and differed by greater numbers of nucleotides per site (CHR, 0.030; SI, 0.019; electronic supplementary material, S3).

We found relatively high linkage among all loci (average |D’| > 0.9), with the SI exhibiting greater linkage on average than the CHR (average |D’| = 0.968 versus 0.925). High linkage values between the genomic regions (average |D’| = 0.937), and low estimates of recombination (R = 0.001 per gene; [58]) suggest that horizontal gene transfer (HGT) of the SI occurs infrequently (electronic supplementary material, S3). Nonetheless, the SI loci had reduced GC content compared with the CHR loci (except for ITS which encodes rDNA), consistent with sequenced Bradyrhizobium genomes USDA6 and USDA110 [8,33] and indicative of the SI's history of horizontal transfer (CHR, approx. 59% GC; SI, 55% GC; electronic supplementary material, S3). The ratio of non-synonymous to synonymous substitutions was low and varied little among loci (approx. 0.1–0.3), hence that most differentiation occurred via synonymous changes in the sequenced loci (electronic supplementary material, S3).

(b). Phylogenetic reconstruction

The CHR and SI trees exhibited contradistinctive topologies. Reconstruction of the CHR tree recovered six species-like clades (i.e. monophyletic; SH-like branch support more than 0.9; less than or equal to one reference species; higher ratio of fixed differences to shared polymorphisms among species [12]) including four previously defined species, Bradyrhizobium japonicum, Bradyrhizobium canariense, Bradyrhizobium retamae and Bradyrhizobium yuanmingense, and two unnamed clades, Bradyrhizobium sp. nov. I and Bradyrhizobium sp. nov. II (figure 1; electronic supplementary material, S4; [12]). Two isolates, 12LoS3_5 and 12LoS6_1 did not fit in any of the recovered clades. Collection sites varied substantially in relative frequencies of CHR clades (electronic supplementary material, S1). Most population genetic parameters were similar among the different CHR clades (electronic supplementary material, S5).

Figure 1.

Reconstructed phylogram of CHR haplotypes. Tree reconstructed in PhyML using concatenated dnak, glnII, ITS and recA loci. Zigzags indicate long branches shortened for visibility. Species-like clades are indicated in shaded portions. The size of the circles on the inner and our perimeter indicate the relative abundance and spatial range, respectively, of each CHR haplotype. (Online version in colour.)

Reconstruction of the SI tree recovered a single deep clade that encompassed all sequenced isolates as well as the reference strains Bradyrhizobium sp. WM9, B. canariense SEMIA928 and Bradyrhizobium cytisi LMG25866 (electronic supplementary material, S6). We compared topologies of the CHR and SI trees. A tanglegram analysis suggested a broad pattern of shared evolutionary history among the major CHR clades and four paraphyletic SI lineages (denoted SI lineages numbers 1–4; electronic supplementary material, S7). The CHR clades B. canariense, B. sp. nov. I and B. retamae were associated with SI lineages no. 4, no. 3, no. 1, respectively, and the paraphyletic CHR taxa B. japonicum and B. sp. nov. II was associated with SI lineage no. 2. At least seven independent HGT events are also evident among these defined lineages. We did not find significant support for congruence of the CHR and SI trees when we used the programs AxParafit and AxPcoords within CopyCat (ParaFitGlobal = 0.11097; p = 0.53), but this test is sensitive to poor phylogenetic resolution near branch tips [51,52].

(c). Spatial analysis of haplotypes

Most whole-genome haplotypes (i.e. all eight loci) were unique (86%), and only five were dominant at any single field site, comprising up to 14% of local isolates. No whole-genome haplotypes were found at multiple sites.

Ten of the 138 CHR haplotypes were categorized as dominant. Two CHR haplotypes were found at multiple sites greater than or equal to 10 km distant, and thus categorized as epidemic (figure 2; electronic supplementary material, S8). Among the 225 SI haplotypes, seven were categorized as dominant within a site, and none were found at multiple sites (electronic supplementary material, S8). The dominant SI haplotypes were always found to be paired with dominant CHR haplotypes.

Figure 2.

Map of dominant CHR haplotypes. Location of nine collection sites indicated by black dots. Proportional breakdown of dominant CHR haplotypes at each location indicated by piecharts. Potential distribution of epidemic haplotypes indicated by coloured dotted outlines. Beginning in the upper left corner and proceeding clockwise, the collection sites are Bodega Marine reserve, UC Riverside, Motte Rimrock Reserve, Burns Pinon Ridge Reserve, Anza Borrego Palm Canyon, Anza Borrego Roadside, Bernard Field Station, San Dimas Canyon and San Dimas Reservoir.

The CHR haplotype K01_G03_I01_R01 was found at all six collection sites that are not in desert habitat (a spatial range of 728.3 km) and was dominant at five of them. This haplotype encompassed greater than 17% of all isolates assayed, consistent with a massive clonal expansion of the CHR [11,40,68] (figure 1; electronic supplementary material, S2). A broad diversity of SI haplotypes were associated with K01_G03_I01_R01 (42 SI haplotypes, encompassing 61 nucleotide changes), revealing that the epidemic CHR haplotype acquired divergent SI haplotypes as it spread (figure 3).

Figure 3.

Tanglegram of CHR and SI phylogenies highlighting epidemic CHR haplotype. TreeMap cophylogeny tanglegram of CHR and SI haplotypes. The left cladogram is reconstructed with CHR loci (dnak, glnII, ITS and recA) and the right cladogram is reconstructed with SI loci (nifD, nodD-A, nodZ and nolL). Overall patterns in associations between CHR and SI haplotypes indicated by grey bars (Individual associations, see the electronic supplementary material, S1). Lineages of associated CHR and SI haplotypes bracketed and numbered. Lines connect the 42 SI haplotypes acquired by the epidemic CHR haplotype K01_G03_I01_R01. (Online version in colour.)

(d). Genetic differentiation among sites

The major CHR clades varied widely in geographical range. Bradyrhizobium canariense, which encompassed both epidemic haplotypes, had the largest range and was the only clade to be found in northern California (B. canariense range approx. 700 km; B. retamae, B. sp. nov. I approx. 150 km; B. sp. nov. II, B. japonicum less than 10 km; electronic supplementary material, S4). Clade diversity varied among sites and was greatest at San Dimas Canyon, which contained all sampled Bradyrhizobium clades and was the only site with B. yuanmingense. Three sites each only contained a single clade (Bodega Marine Reserve, Motte Rimrock Reserve, Anza Borrego Desert State Park—Roadside). Nucleotide and haplotype diversity roughly paralleled clade diversity among sampling sites in both the CHR and SI datasets (electronic supplementary material, S9). Differentiation among populations was lower for the SI than the CHR (mean F_ST = 0.08 and 0.20, respectively; electronic supplementary material, S9). Mean F_ST values for the SI varied little among populations (0.06–0.13). By contrast, two populations exhibited F_ST for the CHR loci that were well beyond this range (Anza Borrego Palm Canyon, mean F_ST = 0.25; Burns Piñon Ridge, F_ST = 0.83). We did not find support for isolation by distance within the CHR or SI datasets (CHR; Mantel test R = −0.00145; p = 0.99; SI, R = 0.09132; p= 0.72).

(e). Carbon source utilization

Twenty genetically diverse isolates were analysed for carbon utilization. Isolates were initially treated as independent data and binned based on haplotype abundance for each genome region. CHR haplotypes were classified as epidemic (n = 5) or not (n = 15) and SI haplotypes (none of which are epidemic) were classified as ‘frequent’ (sampled greater than or equal to 2x; n = 5) or unique (n = 15; electronic supplementary material, S10). Binary utilization scores were significantly higher for strains with epidemic CHR haplotypes (48.6 ± 8.8) than rare CHR haplotypes (26.1 ± 5.1; ANOVA; F_1,19 = 4.87, p = 0.04), and these higher utilization scores were most pronounced on carbohydrates (F_1,19 = 7.50, p = 0.013) and carboxylic acids (F_1,19 = 5.16, p = 0.036). Strains with frequent and rare SI haplotypes did not differ in binary utilization (frequent: 30.8 ± 9.3; rare: 32.1 ± 5.7; ANOVA; F_1,19 = 0.01, p = 0.91).

We analysed genotypic means for carbon utilization scores to take phylogenetic non-independence into account. These analyses generated similar results as in the categorical tests. The epidemic CHR haplotypes exhibited binary carbon utilization scores that were well above 99% confidence limits for the mean of the population distribution (α = 0.01; figure 4). The superiority of the epidemic CHR haplotype was most pronounced on carbohydrates and carboxylic acids (figure 4; electronic supplementary material, S10).

Figure 4.

Carbon utilization of epidemic CHR haplotype relative to population mean. Frequency distribution of binary carbon utilization scores for 20 strains tested. The population mean and 99% confidence intervals (CI) are indicated with solid and dashed lines, respectively. In each case, the genotypic mean of the five tested epidemic CHR haplotypes is above the upper 99% CI (1–0.99 = α). (Online version in colour.)

(f). Hypothesis testing of evolutionary-genomic scenarios

We examined the fit of our data to four potential evolutionary-genomic scenarios and tested models separately for each well-sampled Bradyrhizobium clade (i.e. minimum of 20 isolates; electronic supplementary material, S4). We had the best sampling for the B. canariense clade (n = 244), for which the data support the CHR sweep model (electronic supplementary material, S2). For the B. nov. I clade, n = 54, the next largest dataset, the data also support the CHR sweep model, except Tajima's D is not negative as would be expected following a selective sweep. The data for B. japonicum and B. yuanmingense do not strongly support any of the models, but both have relatively small sample sizes.

4. Discussion

Rhizobia are often studied in agricultural and pastoral settings where local populations are genetically diverse but often dominated by few chromosomally encoded genotypes [11]. A handful of studies in natural populations have mirrored these results, with chromosomal genotypes dominating local populations [4] or spreading among multiple locales [2,12,31,32]. However, there has been little understanding of what drives these patterns or whether they reflect whole-genome evolution. Studies of agricultural isolates can be confounded by tilling, flood irrigation, introduced or genetically altered plants, and biological soil amendments, any of which can transport rhizobia within and among sites. Moreover, rhizobial datasets often sample multiple legume host species and differences in host specialization can also confound patterns of rhizobial strain diversity and dominance [72]. We have focused on native host populations of a single legume species in natural soils to avoid these confounding effects [4,12].

The Bradyrhizobium populations we examined exhibited strikingly different population genetic parameters between the CHR and SI genomic regions. While we sampled similar numbers of nucleotide sites and variable sites in these two genome regions (electronic supplementary material, S3), we found fewer haplotypes, lower haplotype diversity and lower strain richness within the CHR loci. One hypothesis to explain these differences is additional pressure of natural selection on the CHR relative to the SI, which is also supported by the negative Tajima's D (found for the B. canariense clade; electronic supplementary material, S5). An alternative explanation would be that haplotype diversity and strain richness are elevated in the SI driven by local adaptation to host populations; for instance, if the host populations are genetic structured [19]. The genetics of A. strigosus are currently unknown, but given that we found similar levels of population subdivision across both CHR and SI loci, there is not much support for this latter hypothesis.

Focusing on the CHR dataset, we found that most field sites had one or a handful of dominant strains that were unique to that site. Among the two CHR haplotypes that were categorized as epidemic, one was dominant at the majority of sampled sites and ranged over a 700 km span to represent more than 17% of all isolates sampled (CHR haplotype, K01_G03_I01_R01; figure 2; electronic supplementary material, S4 and S8). This striking evidence of CHR clonal expansion was not mirrored in the SI dataset. Although we uncovered locally dominant SI haplotypes within 5 out of 9 collection sites, none had spread more than 10 km (not epidemic; electronic supplementary material, S8). The fact that the dominant SI haplotypes were always paired with dominant CHR haplotypes (but not vice versa) suggests that SI haplotypes only achieved local dominance via hitchhiking with the CHR.

We considered four potential models of Bradyrhizobium genome evolution and found the best support for a CHR sweep within the B. canariense lineage (other lineages were not as well sampled; electronic supplementary material, S2). Low nucleotide diversity in the chromosome, indicative of extensive hitchhiking following a selective sweep, was also found in Sinorhizobium melliloti [13]. In the case of our dataset, however, the presence of highly abundant and spatially widespread haplotypes was strictly limited to the CHR. Among the 42 SI haplotypes associated with the epidemic CHR haplotype in our dataset, only one is dominant (figure 3; electronic supplementary material, S8). Thus, the most likely explanation is that the epidemic CHR haplotype increased its spatial range and in the process acquired divergent SI haplotypes. Our data are inconsistent with the hypothesis of a genome-wide sweep followed by the SI accumulating variation because we found evidence of faster molecular evolution in the CHR versus the SI (0.149 versus 0.088 mutations per site, respectively, within the best sampled B. canariense/SI lineage 4; electronic supplementary material, S7).

The clonal expansion and geographical dissemination of a small subset of bacterial genotypes in a population suggests that these strains exhibit traits that engender superior fitness in their population [73–75]. Such clonal expansion events in bacterial pathogens are often associated with horizontal transfer of accessory DNA, for instance where acquisition of antibiotic resistance traits or vaccine-escape loci can result in epidemic spread of pathogenic strains [73,74]. Parallel processes are also possible in bacterial symbionts, with the host possibly promoting rather than countering bacterial spread. In rhizobial populations, host plants can favour beneficial over ineffective rhizobial genotypes [20,22,23] and thus select certain symbiosis alleles over others [18,19,32]. However, only under very specific conditions has there been evidence of plant selection promoting selective sweeps of symbiosis genotypes. One striking example comes from an agricultural site in which the legume hosts were planted, but were not being nodulated (e.g. no rhizobia were present with the capacity to form nodules). A single Mesorhizobium loti strain was inoculated and its SI genotype spread through a diverse population of non-symbiotic Mesorhizobium spp [53]. The recurrent acquisition of the SI in diverse non-nodulating Mesorhizobium strains is consistent with a fitness advantage of symbiotic versus a non-symbiotic lifestyle. By contrast, we found evidence of a CHR sweep. Our previous data showed that epidemic status of Bradyrhizobium genotypes occurs both in the presence and absence of the SI, suggesting that the fitness advantage encoded by these genotypes is unlinked to the symbiosis [12].

The rhizobial lineages both Bradyrhizobium and Mesorhizobium exhibit genomes with ‘expression islands’, wherein SI loci are primarily expressed in nodules and CHR loci are primarily expressed ex planta [5,6]. Our evidence of selective sweeps restricted to the CHR suggests that selection among CHR variants in the soil (i.e. outside the context of in planta symbiosis) might be the driving force structuring these populations. In further support of this hypothesis, we found that dominant CHR haplotypes had significantly enriched capacity to use sole-carbon sources, which can be important to bacterial fitness in the soil and soil–root interface [76]. However, more research is needed to uncover the portions of the rhizobial life cycle where selection of catabolism traits might be occurring. Superior capability to catabolize carbohydrates and carboxylic acids that we uncovered here could be favoured as a trait for persistence in the soil, competition on carbon-rich root surfaces, or even during growth within nodules that can be rich in these carbon compounds [24]. Moreover, the role of other hosts remains unknown in these populations, and the diversity of SI loci of the epidemic CHR haplotype suggests the possibility that other host legumes might be contributing to the fitness of these strains.

Harnessing natural rhizobial epidemics, such as we uncovered here, could represent a solution to the failed attempts to establish rhizobial inoculants as efficient symbionts of legume crops, known as the rhizobial competition problem [25,26,28]. The key challenge for inoculant strain establishment is that even when inoculum strains have superior nitrogen fixation traits, they often are inferior in competition with indigenous rhizobia for nodulation of the host root. The epidemic CHR genotypes could be particularly useful because they appear to express superior ex planta fitness in diverse soils over a large spatial range.

Data accessibility

All data used in this paper are publicly available in electronic supplementary material and in Dryad http://dx.doi.org/10.5061/dryad.nf981.

Authors' contributions

A.C.H. and J.L.S. designed the study. A.C.H., J.U.R., D.T., K.A.G.-C., R.B., A.B., D.M., J.P. and J.L.S. collected and analysed the data. A.C.H. and J.L.S. participated in drafting and critically revising the manuscript. All authors approved the final version of the manuscript.

Competing interests

Authors do not have competing interests.

Funding

The following grants supported this study: to A.C.H. a Herbert Kraft Scholarship and a UC Riverside Graduate Research Mentorship Fellowship, and to J.L.S. NSF DEB 0816663 and NSF DEB 1150278.