Experimental evolution makes microbes more cooperative with their local host genotype

See allHide authors and affiliations

Science  23 Oct 2020:
Vol. 370, Issue 6515, pp. 476-478
DOI: 10.1126/science.abb7222

Microbial selection drives adaptation

Many legumes have a host-symbiote relationship with nitrogen-fixing bacteria, or rhizobia, that provides a benefit to both the plant and the microbe. Batstone et al. experimentally evolved the association between five legume accessions and different bacterial isolates. Rather than observe selection by the host for bacterial associations (host choice), mutations accumulated within a bacterial plasmid and increased the strength of the mutualism. Thus, local and recent associations between bacterial strains and plant genotypes are due to selection for bacterial adaptation.

Science, this issue p. 476


Advances in microbiome science require a better understanding of how beneficial microbes adapt to hosts. We tested whether hosts select for more-cooperative microbial strains with a year-long evolution experiment and a cross-inoculation experiment designed to explore how nitrogen-fixing bacteria (rhizobia) adapt to legumes. We paired the bacterium Ensifer meliloti with one of five Medicago truncatula genotypes that vary in how strongly they “choose” bacterial symbionts. Independent of host choice, E. meliloti rapidly adapted to its local host genotype, and derived microbes were more beneficial when they shared evolutionary history with their host. This local adaptation was mostly limited to the symbiosis plasmids, with mutations in putative signaling genes. Thus, cooperation depends on the match between partner genotypes and increases as bacteria adapt to their local host.

Host-associated microbiota are often beneficial, but we have a limited understanding of adaptation between partners in these mutualisms, especially at the genomic level. Mutualisms are sometimes viewed as reciprocal parasitism, potentially resulting in antagonistic coevolution that maintains genetic variation within populations. However, recent work has questioned the prevalence of fitness conflict in mutualisms (1), and concordant fitness interests between partners should lead to evolutionary stasis, reducing genetic variation in mutualistic traits like partner quality (2, 3).

Poor-quality microbes could be “cheaters” that increase their own fitness at their host’s expense (4). However, hosts often “choose” their microbiota (5) and preferentially associate with or reward more-cooperative microbes, selecting against would-be cheaters (6, 7). Such partner choice can be adaptive (8, 9) but is also paradoxical: If choosy hosts select high-quality symbionts, variation in symbiont quality should decrease, reducing the selective advantage of partner choice (10). Still, many hosts are choosy and many symbionts are not very beneficial (4, 11), making the persistence of ineffective microbes perplexing.

Ineffective microbes might simply be mismatched with their host. A high-quality microbe on one host genotype may be a low-quality microbe on another. Such genotype-by-genotype (GxG) interactions for symbiont quality are common, maintain variation in the benefits symbionts provide to hosts (3, 12), and are a prerequisite for coevolution (13, 14). Local adaptation, whereby partners from the same site outperform partners from different sites, can generate GxG interactions and occurs in mutualisms (12, 15), but it is not ubiquitous (16). Furthermore, the genetic mechanisms underlying local adaptation or GxG interactions remain largely unknown.

Legume-rhizobium interactions are economically and ecologically important and a model for studying mutualisms. Legumes trade carbon for nitrogen fixed by rhizobia, which they house in root nodules (17). Rhizobia have rapid generation times, can be cultured, and can acquire mutations through horizontal gene transfer (18). Rhizobia are also amenable to genome-wide association studies (GWASs) that can identify genomic variants associated with phenotypes (19). Adapting these approaches, Burghardt et al. (20) found stronger selection on rhizobia in hosts than on rhizobia free-living in soil.

Here, we resequenced and cross-inoculated rhizobia after they evolved on one of five host genotypes that vary in choosiness to test whether choosier hosts select for more cooperative symbionts or whether rhizobia adapt to their local host genotype. We used two rhizobia strains that differ in host benefits: ineffective nitrogen fixer Ensifer meliloti strain Sm1021 (referred to here as Em1021) and effective nitrogen fixer E. meliloti strain WSM1022 (Em1022) (9, 17). We paired both strains with one of five inbred lines of Medicago truncatula: Line 270 is indiscriminate, whereas the others all prefer Em1022 to Em1021, with line 267 almost exclusively partnering with Em1022 in past experiments (9, 17). After a year-long evolution experiment spanning five plant generations, we isolated “derived” rhizobia from nodules and compared them to the original “ancestral” strains.

Even though Em1021 started with a twofold advantage (17), it became nearly extinct on all host lines (Fig. 1). At the end of the experiment, Em1021 associated with only 15% of plants and occupied only 3% of 336 sampled nodules. In a linear model combining strain frequencies in soil and nodules (17), we found a marginally significant effect of generation (F3,96 = 2.40, P = 0.0730) but no effect of plant line (F4,96 = 0.552, P = 0.698). The effective symbiont, Em1022, outcompeted Em1021 on all hosts regardless of choosiness, indicating that partner choice was not an important selective force.

Fig. 1 High-quality microbial partner spread nearly to fixation in all treatment groups.

The percentage (mean ± 1 SE) of the effective N-fixing strain (Em1022) across plant generations in the evolution experiment was calculated from soil samples for generations 2 to 4, and nodules were dissected from plants in generation 5 (shaded in light orange). Colors indicate M. truncatula genotypes. All plants were initially inoculated with 33% Em1022.

We assayed Em1022 phenotypes by planting new seeds of the same five plant lines and singly inoculating each with either ancestral Em1022 or 1 of the 40 derived isolates, testing replicates of all possible combinations of host genotypes and isolates (17). We quantified the symbiotic quality of rhizobium isolates by measuring aboveground plant biomass and rhizobium fitness by measuring nodule number. We also sequenced whole genomes of the 40 derived isolates and nine replicates each of the ancestors and conducted GWASs for symbiont quality and fitness both on specific host genotypes and across all plant lines (17).

Variation in the benefits that experimentally evolved bacteria conferred to hosts was largely determined by the matches between rhizobium and host genotypes. Derived and ancestral Em1022 did not differ significantly in symbiont quality or fitness (tables S1A and S2A). However, derived bacteria provided greater benefits to the host genotype they were paired with during the evolution experiment (Fig. 2A), although a few isolates were high- or low-quality symbionts across plant lines (figs. S1 and S2). Derived rhizobia also achieved higher fitness when tested on their local host genotype, with the exception of isolates that evolved on the least choosy host genotype (line 270) (Fig. 2B). Linear mixed models found significantly positive effects of shared evolutionary history on both rhizobium quality and fitness (tables S1B and S2B).

Fig. 2 Bacteria adapt to their local host genotype.

(A and B) Mean ± 1 SE shoot biomass (A) and nodule number (B) for all combinations of 40 derived bacterial isolates from the evolution experiment and five plant genotypes (numbered across top). Bacteria share evolutionary history with the host genotype they evolved on during the evolution experiment.

Averaged across all hosts, there was a positive relationship between rhizobia quality and fitness (fig. S3), suggesting that none of the 40 derived strains evolved to cheat and that strong fitness feedback makes cooperation adaptive for rhizobia. Pairwise correlations between the same trait on different hosts were weaker and often not statistically significant (fig. S4), suggesting that microbial quality or fitness on one host poorly predicts quality or fitness on another.

Genomic sequencing of E. meliloti isolates uncovered 1330 genetic variants, including 547 de novo mutations (fig. S5A). Most de novo variants (305) were specific to rhizobia derived on the least choosy host line (line 270); fewer variants (16 to 59) were specific to rhizobia derived on other lines (fig. S5A). After filtering out most singletons and variants in high linkage disequilibrium, reducing our set to 363 variants (fig. S5B), GWASs identified 145 variants significantly associated with rhizobium quality or fitness (referred to as significant variants) in one or more host environments (table S3). These variants were mostly located on two symbiosis plasmids, pSymA and pSymB, which contain genes essential for the symbiotic and free-living phases of the rhizobium life cycle, respectively (21, 22); 62 variants were on pSymA, 68 on pSymB, and only 15 were on the bacterial chromosome.

Beta scores, which represent the strength of association between a genomic variant and a phenotype, were strongly positively correlated between symbiont quality and fitness (Fig. 3A), meaning that genomic variants that increased microbial fitness also increased microbial benefits to plants. Within each host environment, most of the significant variants evolved in isolates paired with that host genotype in the evolution experiment (i.e., local variants; red symbols in Fig. 3A and numbers before commas in Fig. 3B). We categorized variant effects as “cooperative” if beta scores for symbiont quality and fitness were both greater than zero, “defective” if beta scores were both less than zero, “altruistic” if beta scores were positive for symbiont quality but negative for symbiont fitness, or “cheater” if beta scores were positive for symbiont fitness but negative for symbiont quality. In all five host environments, most of the significant variant effects were cooperative and local (Fig. 3B and table S3).

Fig. 3 Most of the significant genomic variants were cooperative and local.

(A) GWAS beta scores for symbiont quality (shoot biomass) and fitness (nodule number) for the 363 variants (points) among the 40 derived isolates, averaged over all plant lines or for each host genotype separately. Larger symbols are variants with one or more statistically significant effects (squares indicate segregating in the ancestor; circles indicate de novo). For the panels showing each host genotype, red symbol color indicates that variants evolved in individual host environments and gray symbol color indicates that they did not. The region shaded in gray defines cheater variants. (B) Counts of significant variants in each host environment. (Category definitions are provided in the text.) Numbers before commas indicate local variants, and numbers after commas indicate variants that evolved in a different host environment. Plus (+) or minus (−) symbols indicate categories with more or fewer variants, respectively, than the null expectation. Orange shading indicates a high number of total variants relative to the other three quadrants; light red shading indicates a moderate number of variants relative to the other three quadrants.

We used permutations to assess whether this pattern could have occurred by chance by randomly assigning genomes to phenotypes 1000 times and recalculating beta scores (17). There were significantly more local-cooperative associations in all five host environments (Fig. 3B, plus symbols) and significantly fewer local-altruistic and local-cheater associations in some host environments (Fig. 3B, minus symbols) at a 5% false discovery rate. The numbers of associations in other categories (e.g., defective variants) did not differ from null expectations.

Local, cooperative variants were located in genes that putatively encode a calcium-binding protein, a flavin adenine dinucleotide (FAD)–binding oxidoreductase, and two Ti-type conjugative transfer relaxases, among others (table S3). These functions may be related to nodule initiation or rhizobium survival inside nodules (23). Variant effects were often host-specific and generally conditionally beneficial on their local host or deleterious on a nonlocal host (table S3), suggesting selection in local contexts and drift in nonlocal contexts. When variants exhibited significant effects on multiple hosts, they tended to be consistently beneficial or deleterious across hosts; only two variants had significant, opposite-sign effects across hosts, suggesting that trade-offs are rare (table S3). In summary, cooperation evolved more often, and cheating or altruism less often, than expected by chance as bacteria adapted to their local host genotype.

In our evolution experiment, we expected the choosiest host line to most strongly favor the better symbiont, Em1022. However, the ineffective strain, Em1021, went nearly extinct across all plant lines, even on indiscriminate hosts (Fig. 1), suggesting that this strain is a universally poor competitor. By contrast, Em1022 evolved previously uncharacterized variants that differed in both symbiont quality and fitness. Derived bacteria provided greater host benefits and generally achieved higher fitness on the host genotype with which they shared evolutionary history (Fig. 2), and we detected a significant excess of genomic variants with cooperative effects on local hosts (Fig. 3). These results suggest that local adaptation is a more important evolutionary force shaping microbial cooperation than is partner choice. Furthermore, when microbes are consistently paired with the same host and dispersal is limited, the resulting local adaptation likely leads to more, not less, cooperation. What emerges is a model of adaptation in host-microbe mutualisms: Microbes can rapidly adapt to a particular host genotype through standing or de novo variants that also benefit local hosts but have varying effects on nonlocal host genotypes, maintaining the variation in mutualist quality that we observe in host-associated microbiomes. Our results also imply that transplanting microbiota is more likely to be effective among closely related hosts or that one needs to give sufficient time for microbes to adapt to a new host environment.

Supplementary Materials

Materials and Methods

Figs. S1 to S6

Tables S1 to S4

References (2560)

MDAR Reproducibility Checklist

References and Notes

  1. Materials and methods are available as supplementary materials.
Acknowledgments: We thank volunteers J. Li, C. Chen, G. Forguson, F. Samad-zada, L. Wang, X. Zhang, D. Li, K. Ong, and J. Wong. The Frederickson Lab, especially E. Dutton and J. Laurich, as well as L. Burghardt, J. Lau, J. Stinchcombe, and C. Wood, provided feedback. I. Anreiter assisted with quantitative polymerase chain reactions, and A. Petrie, B. Cole, and C. Bonner managed the greenhouse. Funding: We acknowledge funding from an NSERC Discovery Grant and Accelerator Supplement (M.E.F.), Ontario Graduate (R.T.B.) and NSERC (T.L.H.) scholarships, and the University of Toronto. Author contributions: R.T.B. and M.E.F. conceived the study and designed experiments; R.T.B. conducted experiments; R.T.B. and M.E.F. analyzed phenotypic data; R.T.B., A.M.O., and T.L.H. conducted bioinformatics; and all authors wrote the manuscript. Competing interests: The authors declare no competing interests. Data and materials availability: Raw sequence reads have been deposited at NCBI BioProject ID PRJNA512862 ( All data and code for analyses are available on Zenodo (24).
View Abstract

Stay Connected to Science

Navigate This Article