Host control by Acmispon strigosus constrains fitness gains of ineffective Bradyrhizobium symbionts in mixed infections

Plant hosts can gain significant growth benefits from symbiosis with microbes, but these benefits could be threatened by divergent fitness interests among partners. Here, we measured fitness outcomes in symbiosis, by varying the genotypes of both microbes and hosts, to examine scenarios that might favor uncooperative symbionts. We studied associations between Acmispon strigosus, an annual legume native to California, and its nitrogen fixing symbionts in the genus Bradyrhizobium. Bradyrhizobium symbionts form root nodules on compatible hosts, with strains varying from effective, fixing substantial nitrogen for the host, to ineffective strains that do not fix nitrogen and provide no benefit to host growth. We co-inoculated four A. strigosus plant lines with nine combinations of effective and ineffective Bradyrhizobium strains and measured the relative fitness of ineffective strains within individual nodules, as hosts must select against uncooperative symbionts to maintain benefits. In mixed infections, ineffective strains always had lower relative fitness in nodules compared to beneficial strains, consistent with efficient punishment of nonfixing rhizobia. However, ineffective strains exhibited genotypic variation in their fitness in nodules within individual nodules co-infected with a beneficial strain, suggesting a role for symbiont competitiveness in shaping this joint phenotype. Variation in symbiont fitness during co-inoculations did not measurably affect plant performance, suggesting that predicted conflict over the joint phenotype of rhizobia fitness has negligible effect on the host.

Methods
Data analysis – Generalized linear mixed models (GLMMs) were used to test hypotheses using JMP Pro 13.0.0 (SAS Institute Inc., Cary, NC, USA). Dependent variables were log10-transformed as needed to improve normality. Proportional data was logit-transformed after applying a linear transformation to account for zeros and ones in the dataset (i.e., 1% was added to all datapoints except ones, from which 1% was subtracted). All models included a random effect of harvest week; models using plant biomass data (and not nodule culturing data) also included a random effect of block nested within harvest week. For each GLMM, all possible interactions among main effects of interest were initially tested. Nonsignificant interactions were removed from the model if this reduced the corrected AIC (AICc) by at least 2 units, and results from trimmed models were reported. Significant differences among levels of main effects were assessed with pairwise t-tests (Tukey’s HSD) of least squares means. The Test Slices option was used to explore interaction effects when only specific contrasts were of interest. Mean values discussed below were backtransformed (if applicable) from raw means and presented alongside 95% confidence intervals.

Rhizobia strains were categorized as effective or ineffective depending on whether the total dry plant biomass of inoculated plants was significantly greater than that of the uninoculated control plants in the single inoculation experiment. Two different fitness proxies were estimated for rhizobia strains, including rhizobia population size per nodule (in the single inoculation experiment) and relative strain frequency within a nodule (i.e., proportion of CFU from a nodule of a particular strain in the co-inoculation experiment). Rhizobia population size (i.e., CFU) per nodule was averaged between replicate nodule cultures with the plant as the unit of replication (n = 4) and was tested for effects of strain genotype and host line. Nodule occupancy for a strain was quantified as its relative abundance (i.e., proportion of CFU/nodule) on each replicate plant (n = 4), for which the null expectation was 50% (i.e., relative abundance in the inoculum), and was tested for effects of the ineffective strain genotype, effective strain genotype, and host line.

Plant relative performance was examined in the co-inoculation experiment by dividing the total plant biomass of each co-inoculated plant by the biomass of plants singly inoculated with the effective strain. Relative performance less than one would indicate that plants performed worse during mixed inoculations than with the effective strain alone, suggesting a cost to encountering the ineffective strain. Relative performance of co-inoculated plants was tested for significant deviation from one based on whether the confidence interval overlapped with one. Plant performance was tested for effects of the effective strain genotype, ineffective strain genotype, and host line.