Can the right partner mitigate harm? Rhizobial strains vary in their mediation of herbicide stress in a plant-rhizobia mutualism

Abstract

Agriculture has intensified the presence of chemical stressors in the rhizosphere—the region surrounding roots where critical plant-microbe interactions occur, such as those between leguminous plants and nitrogen-fixing rhizobial bacteria. Particularly, rhizospheric pesticide exposure can disrupt the efficacy of the plant-rhizobia mutualism and reduce plant productivity. However, it is unknown whether genetic variation in plants (G_P), rhizobia (G_R), or interactions between them and the pesticide environment (E), i.e., G_{P or R}   E, or G_P   G_R   E, could mitigate these negative outcomes. We grew two genotypes of the leguminous plant Trifolium pratense in symbiosis with each of eight genetic strains of its rhizobial partner Rhizobium spp. symbiovar trifolii. We exposed symbionts to the contemporary synthetic auxin herbicide dicamba or a control in the rhizosphere, and evaluated the symbiotic interaction and plant growth. Our results provide new evidence that rhizobial genetic variation drives herbicide impacts on mutualism outcomes through G_R   E interactions. Rhizospheric herbicide delayed rhizobial colonization of plants via root nodule formation, but its effects on the number of nodules and fixed nitrogen produced varied depending on rhizobial strain. Similarly, while herbicide exposure reduced plant size on average, the degree of this effect was mediated by rhizobial partner, suggesting that rhizobia could potentially function as an “extended genotype” for defense against herbicide damage. As the use of herbicides, particularly synthetic auxins, continues to escalate, our findings have important implications for how certain rhizobia could be selected to improve plant fitness in the face of these anthropogenically-released chemicals.

Pesticide presence in the rhizosphere—the area surrounding plant roots—can disrupt beneficial plant-microbe symbioses, such as those between leguminous plants and nitrogen-fixing rhizobial bacteria. However, it is unknown whether genetic variation in plants, rhizobia, or their interaction could mitigate mutualism disruption. To address this gap, we grew two genotypes of the legume Trifolium pratense with eight genetic strains of Rhizobium spp. symbiovar trifolii, with and without rhizospheric exposure to a common herbicide, and evaluated the symbiosis. We found new evidence that rhizobial genetic variation is an important driver of herbicide impacts on plant-rhizobia interactions and plant growth. In particular, while herbicide exposure reduced plant size overall, this effect varied by rhizobial genotype, and was even mitigated by some rhizobial partners. Our findings therefore suggest that microbes could potentially function as an “extended genotype” of host plants, resulting in microbe-mediated plant tolerance of herbicide stress and mutualism evolution in agro-ecosystems.

Introduction

Symbiotic interactions between leguminous plants and rhizobial bacteria (hereafter “rhizobia”) constitute one of the most important mutualisms on Earth. Through biological nitrogen fixation (BNF), rhizobia housed within root nodules transform atmospheric nitrogen (N₂) into a plant usable form (NH₃) in exchange for carbon, thereby increasing nitrogen availability in terrestrial ecosystems (Gou et al., 2023). However, anthropogenically-released pesticides can contaminate the rhizosphere—the area surrounding plant roots where symbionts coexist—resulting in reductions in the efficacy of the plant-rhizobia mutualism in creating biologically-fixed N and promoting plant growth (Fox et al., 2007). In particular, rhizospheric contamination by synthetic auxin or “auxinic” herbicides (e.g., 2,4-D and dicamba), which stimulate abnormal and deleterious growth in target plants by mimicking the plant growth hormone auxin, is of concern. These pesticides have recently surged in use globally (Riter et al., 2021; USGS, 2024). Moreover, they frequently effectuate off-target movement, especially via “drift,” wherein a small fraction of the applied herbicide moves through the atmosphere.

The ramifications of off-target drift for wild plants and animals can be alarming (Baucom et al., 2025; Johnson et al., 2023). Yet, the less visible effects of auxinic herbicide exposures in the rhizosphere, and on the plant-rhizobia symbiosis, are just beginning to be understood (Iriart & Ashman, 2025). Some work has shown that low doses of auxinic herbicides can impede symbiosis establishment via nodule formation (i.e., nodulation; Fox et al., 2004) and/or BNF (Saraf et al., 1999; Zaidi et al., 2005). For instance, there is evidence that 2,4-D can impede symbiotic signaling between plants and rhizobia by blocking nodulation protein receptors and inhibiting gene expression related to nodulation in rhizobia, at least when one rhizobial strain was investigated (Fox et al., 2004). However, the biological forces that could mitigate these negative effects on symbiotic outcomes are largely unknown, yet they could inform on likely evolutionary trajectories of naturally-occurring plant-rhizobia mutualisms at the agro-ecological interface (Burdon & Thrall, 2008; Iriart et al., 2021). Here we explore three hypotheses to uncover these insights.

One hypothesis is that the effects of auxinic herbicides on symbiotic outcomes could be mitigated by genetic variation in leguminous host plants (G_P) through genotype-by-environment (G_P   E) interactions. For example, herbicide-tolerant plant genotypes might have more carbon reserves to invest in the mutualism than sensitive ones under herbicide exposure. In line with this possibility, Vaidya & Stinchcombe (2020) found that when legumes were deprived of light, shade-tolerant plant genotypes maintained or increased investment in rhizobial root nodules more often than less tolerant genotypes. While on the subject of herbicide stress, as opposed to other forms of abiotic stress, G_P   E has scarcely been characterized (but see France et al. (2022) and Iriart et al. (2024)), it may likewise be a potential mechanism determining mutualism trait expression when rhizospheres are exposed to these herbicides.

Alternatively, variation in rhizobia and the symbiotic traits they engender may be altered under auxinic herbicide exposure, resulting in influential G_R   E interactions. It has long been thought that rhizobial strains show context-dependency in the benefits they provide to leguminous hosts (e.g., fixed N) relative to their carbon costs, which can ultimately affect the net growth benefits plants gain from the symbiosis across environments (Burghardt & diCenzo, 2023; Heath & Tiffin, 2007). Moreover, microbial symbionts could act as “extended genotypes” of host plants, expanding their genetic repertoire and mediating plant trait expression under different environments (Henry et al., 2021). Notably, Iriart et al. (2024) previously found that the impacts of aboveground applications of drift-level dicamba on red clover growth depended on plant genotype to some extent, but more so on whether plants were inoculated with rhizobia, and the two rhizobial genotypes (strains) used led to different responses. However, studies such as this are scarce—and when rhizobial effects are considered, only a few strains are used. Therefore, the inferences concerning the importance of G_R   E in this context, i.e., whether rhizobia can function as an extended genotype for legumes under herbicidal conditions, are still quite limited.

Lastly, because the plant-rhizobia mutualism is a product of coevolution, it can vary greatly depending on the interaction between plant host and rhizobial genotypes. In addition, this can vary with the environment, resulting in G_P   G_R   E interactions (Heath & Tiffin, 2007). Particular combinations of legume and rhizobial genotypes could thus dictate quantity and/or quality of plant-rhizobia interactions in the presence of rhizospheric herbicide, similar to how they shaped nodule traits such as number and size in response to environmental N availability (Heath & Tiffin, 2007; Heath et al., 2010). Yet how this complex interaction is affected by an auxinic herbicide in the rhizosphere has never been addressed.

To test these hypotheses, we conducted a microcosm experiment where we used different genetic strains of Rhizobium spp. symbiovar trifolii (eight strains, representing G_R) to inoculate the ecologically and agriculturally-important red clover, Trifolium pratense (two widely used cultivars, representing G_P), in a rhizosphere with or without the auxinic herbicide dicamba (representing E). Trifolium pratense (Fabaceae) is an herbaceous legume that is cultivated or grows feral around the world, especially at agro-eco interfaces (Jones et al., 2020), and its growth is negatively impacted by direct foliar exposure to low-strength dicamba (Iriart et al., 2022, 2024). However, the impact on the plant when exposure occurs in the rhizosphere and the extent that rhizobial symbiont variation contributes to the plant response lacks exploration. Therefore, we sought to answer three primary research questions: (1) Does the presence of dicamba in the rhizosphere disrupt (a) key traits of the plant-rhizobia mutualism, particularly the timing of nodulation, the number of nodules formed, and BNF, and/or (b) the plant growth response across different pairings of rhizobial and plant genotypes? (2) Does G_R or G_P mediate the effects of dicamba on mutualism traits or plant growth via (a) G_R   E, (b) G_P   E, or (c) G_P  G_R   E interactions? And (3) does the presence of dicamba in the rhizosphere affect the overall strength of the plant-rhizobia mutualism from the plant perspective by making it more or less beneficial for plant growth compared to plants lacking rhizobial symbionts?

Methods

Selecting plant and rhizobial genotypes

We selected two cultivars (“genotypes”) of T. pratense (“Kenland” and “Mammoth”; Ernst Conservation Seeds; Meadville, PA, USA) for this study. As an outcrossing species, there is typically high genetic diversity across wild populations and within agricultural cultivars of T. pratense, like the ones selected (Nay et al., 2023). We paired T. pratense cultivars with eight strains of Rhizobium spp. symbiovar trifolii. Five Rhizobium leguminosarum strains (accessions 2063, 2087, 2141, 2214, and 2316) and one Rhizobium pisi (accession 2220) were acquired from the U.S. Department of Agriculture (USDA) Soybean Genomics and Improvement Lab (Beltsville, MD, USA). Two other R. leguminosarum strains were from the American Type Culture Collection (Manassas, VA, USA) (accession 14479) or the USDA Northern Regional Research Lab (Peoria, IL, USA) (accession 4386). We selected strains 14479 and 4386 because these previously had different symbiotic outcomes in T. pratense (Iriart et al., 2024). The others were selected because they could nodulate T. pratense or other Trifolium species, and thus belonged to the symbiovar trifolii group (Janczarek et al., 2024). To determine the genetic relatedness of strains, we constructed a phylogenetic tree (Supplementary Figure S1) from whole genome assemblies derived from Illumina NextSeq200 short reads by SeqCoast Genomics (Portsmouth, NH, USA) (for full methods, see Supplementary Methods S1). We also performed multiple sequence alignments of select genes of interest, including nifU and nodW, which are involved in regulating BNF and nodulation (Fu et al., 1994; Loh & Stacey, 2003), and acdS, which encodes the 1-aminocyclopropane-1-carboxylate (ACC) deaminase enzyme known for alleviating abiotic stress in plants (Glick, 2014). Interstrain genetic variation was apparent from these alignments and often mirrored patterns of relatedness suggested by the phylogenetic tree (Supplementary Figure S2).

Experimental set-up: herbicide treatments and rhizobial inoculation

A microcosm experiment was performed in two temporal blocks to optimize sample size: 1) June 13–July 25, 2023 (N = 288); 2) September 12–October 10, 2023 (N = 180). At the start of each block, we germinated surface-sterilized T. pratense seeds on 1.5% agar then randomly selected 18 seedlings per genotype for transfer to single-plant, agar-based “microcosms” made from 100mm petri dishes as in Jones et al. (2013). Half of the microcosms were pre-treated with dicamba herbicide to simulate off-target exposure in the rhizosphere. This was done by spreading 150L of dicamba (3,6-dichloro-o-anisic acid; Albaugh, LLC, Ankeny, IA, USA) at a concentration of 15 mg of active ingredient [a.i.]/L (i.e., ~0.5% of the field application rate [FAR] of 3 g a.i./L; Albaugh, 2018) over the surface of microcosms with sterile glass beads. This concentration is within the range of dicamba levels detected in plants and soil affected by off-target dicamba movement, e.g., via drift (Carlsen et al., 2006; Egan & Mortensen, 2012) or run-off (Hall & Mumma, 1994; Rice et al., 2010). The other half of microcosms were treated with DI water in the same way as a control. Both the dicamba and control treatments were sterilized with a 0.22m syringe filter prior to application. To confirm the efficacy of our dicamba stock, we transferred six additional seedlings (three per plant genotype, not inoculated with rhizobia) to microcosms pre-treated with 100% FAR dicamba and monitored survival.

We inoculated seedling roots of one plant genotype per microcosm type (i.e., with/without dicamba) with 100L of rhizobial cells (~1 10⁸ CFU/mL) of one of the eight Rhizobium spp. symbiovar trifolii strains. Cells were prepared by culturing strains in Modified Arabinose Gluconate (MAG) liquid media (Iriart, 2024). After ~2 days of bacterial growth, we centrifuged the cultures, resuspended the pellets in autoclaved DI water, measured the optical density at 600nm, and used the relationship between optical density and CFU for each strain (see Iriart, 2024) to dilute the CFU count, equalizing it among strains. To monitor contamination among inoculation treatments and obtain a baseline for the effect of rhizobia on plants, we did not inoculate one seedling per plant genotype per microcosm type (uninoculated treatment).

Finally, we sealed microcosms with parafilm and organized one experimental replicate (i.e., 2 plant genotypes 2 herbicide treatments [8 rhizobial inoculations + 1 uninoculated group] = 36 microcosms) into a tray (transparent, plastic, 41 58 15 cm). There were 13 trays across the two blocks, eight in the first and five in the second, resulting in a final sample size of 468 single-plant microcosms. Within a tray, microcosms were placed randomly and held up vertically by Styrofoam sheets as previously described (Iriart, 2024). Trays were stored in a growth chamber (22°C, 50% relative humidity, 16:8 hours light: dark, 100–160 M/m² light intensity). The seedlings exposed to 100% FAR dicamba (not inoculated with rhizobia) were stored similarly. Ultimately, we confirmed that both rhizobial inoculation treatments and herbicide treatments were effective: uninoculated plants did not form nodules, and plants within microcosms treated with 100% FAR dicamba died (N = 11/12) or did not grow (N = 1/12).

Measuring mutualism metrics and plant growth

We took weekly photographs of microcosms and used them to record the presence/absence of nodules each week post-inoculation up to Week 4 and count nodules at Week 4. In Week 6 of the first temporal block, we harvested shoots, dried them at 70°C for 48 hours, and weighed their biomass to the nearest pg (Cahn Model 31 Microbalance, Thermo Fisher Scientific Corp., Waltham, MA, USA) to measure final plant size. To assess BNF, we quantified foliar ¹⁵N (the isotopic ratio of ¹⁵N:¹⁴N in sample relative to atmospheric air) via an elemental analyzer at the Washington State University Stable Isotope Core Laboratory (Pullman, WA, USA). Foliar ¹⁵N estimates BNF as plants with lower foliar ¹⁵N typically have received high amounts of symbiotically-fixed N (enriched in ¹⁴N) compared to those receiving little or no fixed N (higher ¹⁵N content), due to the observation that the nitrogenase enzyme discriminatorily utilizes the lighter of the two N isotopes (Lindström & Mousavi, 2020). To obtain this data, we submitted Week 6 leaf samples from a random selection of five plants per genotype, herbicide treatment, and rhizobial inoculation (N = 160 total). We validated this method for estimating BNF by additionally analyzing foliar ¹⁵N of one uninoculated plant per genotype and herbicide treatment (N = 4 total). Confirming expectations, uninoculated plants had greater foliar ¹⁵N than plants inoculated with most (6/8) rhizobial strains (Supplementary Table S1). To account for the effects of nodule production on BNF, we also counted nodules on the Week 6 plants for use as a covariate during data analysis. Some plants (N = 75; 16% of the experiment) showed signs of a minor fungal infection at Week 4, and this was recorded. Six plants with severe fungal infection (1% of the experiment) were discarded.

Statistical analyses

We conducted all analyses in R version 4.2.2 (R Core Team, 2022) and created figures using the ggplot2 package (Wickham, 2016). To evaluate the timing of nodulation, we analyzed data of nodule presence/absence over time with mixed effects Cox proportional hazards regression models, suited for time-to-event analyses (Cox, 1972), using the coxme package (Therneau, 2024). We also graphed Kaplan-Meier curves (Rich et al., 2010), showing the proportion of plants lacking nodules at each week post-inoculation, to estimate mean time until nodulation by taking the area under the curves (Han & Jung, 2022) using the survival package (Therneau & Grambsch, 2000). When visualizing results (e.g., Figure 1), we plotted the proportion of nodulated plants (1 proportion of plants lacking nodules) for ease of interpretation.

Figure 1.

Rhizospheric herbicide exposure and rhizobial strain independently affected the timing of nodulation. Kaplan-Meier curves show the mean ± SE (bold line and shading) proportion of nodulated T. pratense plants following inoculation with rhizobia. A: proportions when plants were or were not treated with dicamba herbicide, averaged across rhizobial strain inoculations (No Dicamba: N = 206 plants, Dicamba: N = 202 plants). B: proportions when plants were inoculated with different rhizobial strains, averaged across herbicide treatments (strains 14479 and 2316: N = 51 plants, strain 2063: N = 50 plants, strains 2087, 2141, 2214, and 2220: N = 52 plants, strain 4386: N = 48 plants).

To assess all other response variables—nodule number, foliar ¹⁵N, and plant size, we built mixed-effects linear models (LMMs) using the lme4 package (Bates et al., 2015). Nodule numbers (based on Week 4 nodule counts) and plant sizes (based on Week 6 shoot biomass) were natural log- and square-root-transformed, respectively, to meet model assumptions of normality. We log-transformed nodule number data because LMMs with the log-transformation satisfied parametric test assumptions (i.e., passed diagnostics of residual plots), whereas Poisson and negative binomial generalized linear mixed models (GLMMs) did not (Warton et al., 2016).

We performed model selection using Akaike Information Criterion (AIC) (Akaike, 1973). Models included the explanatory variables (fixed effects): plant genotype (G_P), rhizobial strain (G_R), herbicide treatment (E), and all two-way and three-way interactions between them. Models also included the covariates: tray (random effect) and, if applicable, temporal block (fixed effect), and the extraneous variable: minor fungal infection (fixed effect), unless it worsened model fit (i.e., increased AIC; Zuur et al., 2009). Additionally, if a model that included the G_P   G_R   E or G_P   G_R interaction had a greater AIC than models that excluded them, we dropped them and only tested the G_P   E and G_R   E interactions. We also considered whether phylogenetic history improved model fit by building and selecting from phylogenetic models (package phyr; Li et al., 2020) accounting for rhizobial phylogeny (Supplementary Figure S1). We checked model assumptions and diagnostics using the DHARMa package (Hartig, 2022).

To determine whether G_P or G_R or their interactions mediate the effects of dicamba on mutualism traits or plant growth, we ran Type III sums of squares ANOVAs on best-fit models (Supplementary Table S2) using the car package (Fox & Weisberg, 2019). When we detected a significant G_P   E or G_R   E interaction for a given response variable, we calculated estimated marginal means (EMMs) using the emmeans package (Lenth et al., 2018) and conducted planned post hoc comparisons (Dunnett’s Test) between them to identify significant effects of rhizospheric dicamba at the plant genotype/rhizobial strain-level. Further, to discern whether a G_R   E effect resulted in a change in rhizobial genotypic rank order, we calculated Spearman rank correlations between rhizobial strain EMMs in the presence vs. absence of dicamba. A weak correlation (i.e., r   0.5) would suggest that inoculum rankings in each environment were unrelated, therefore a shift in rank order had occurred. If we found a significant G_P   G_R or G_P   G_R   E interaction, we compared strain EMMs across plant genotypes and (if applicable) environments of dicamba.

To determine whether dicamba modified the predicted benefits of the plant-rhizobia partnership from the plant perspective, we calculated the difference in size between inoculated plants and uninoculated plants for each plant genotype, rhizobial strain, herbicide treatment, and replicate (tray). If the mean rhizobial effect of strain A was > 0, then strain A’s effect was beneficial for plant growth. If < 0, it was costly. We assessed significance by performing a paired t-test on the mean rhizobial effect of strains in the presence vs. absence of herbicide.

Across all analyses, we found that models performed worse (AIC 2; Zuur et al., 2009) when the three-way G_P   G_R   E interaction was included, so this term was dropped during model selection. In addition, phylogeny-corrected models did not outperform standard models. Only the best-fit models are discussed below (see Supplementary Table S2 for model selection results).

Results

The presence of herbicide in the rhizosphere delayed nodulation

The timing of nodulation at four weeks post-inoculation was affected by herbicide treatment and rhizobial variation independently (Table 1A; Figure 1). Dicamba exposure increased the mean time until nodulation by 29%, equivalent to 0.74 weeks (Figure 1A; herbicide treatment effect: P < 0.001; Table 1A; Supplementary Table S3A). Additionally, the time until nodulation differed among rhizobial strain inocula by 0.4–50%, i.e., 0.01–1.29 weeks (Figure 1B; rhizobial strain effect: P < 0.0001; Table 1A; Supplementary Table S3B). However, neither the strain herbicide (G_R  E) nor the plant herbicide (G_P  E) interaction influenced nodulation timing. Instead, nodulation was similarly delayed by dicamba exposure across rhizobial strains and plant genotypes (Table 1A).

Table 1.

Type III sums of squares ANOVAs for traits related to the plant-rhizobia mutualism (A–C) and plant growth (D). ​​​​​​

		A. Time Until Nodulation		B. Nodule Number		C. BNF (foliar 15N)		D. Plant Size (mg)
Fixed Effect	df	2	P	2	P	2	P	2	P
Herbicide Treatment (E)	1	11.97	0.0005	0.41	0.52	2.12	0.15	6.92	0.0085
Rhizobial Strain (GR)	7	33.51	<0.0001	21.08	0.0037	12.80	0.077	8.68	0.28
Plant Genotype (GP)	1	2.39	0.12	0.14	0.71	0.33	0.56	0.07	0.80
Temporal Block	1	0.05	0.48	27.88	<0.0001	-	-	-	-
Minor Fungal Infection	1	-	-	8.43	0.0037	5.38	0.020	-	-
GR   E	7	8.65	0.28	18.29	0.011	15.03	0.036	18.47	0.010
GP   E	1	2.22	0.14	0.20	0.66	0.17	0.68	2.32	0.13
GP   GR	7	-	-	15.29	0.032	-	-	-	-

Rows correspond to fixed effect factors (“fixed effect”) and their degrees of freedom (“df”), , and P-values from the best-fit statistical model for each response variable (A–D; see Supplementary Table S2 for details regarding model selection). “Temporal Block” and “Minor Fungal Infection” account for variation attributed to whether plants were grown in the first or second randomized temporal block and whether plants were or were not affected by a fungal pathogen that contaminated a fraction of microcosms (see Methods). Dashes indicate that the fixed effect was not included in the analysis because it was either not applicable or it worsened model fit.

Rhizobial strains mediated the effect of herbicide exposure on nodule number and BNF

Mean nodule number varied among rhizobial strain inocula (G_R: P < 0.01; Table 1B), with as many as 32 nodules produced in four weeks (raw mean = 3.6 0.3). Unlike our results for nodulation timing, dicamba did not independently affect nodule number (E: P > 0.05); rather, the effect of dicamba on this trait was mediated by rhizobial strains (G_R  E: P < 0.05; Table 1B; Figure 2A). In the presence of dicamba in the rhizosphere, plants inoculated with half of the strains made significantly or marginally significantly fewer nodules (by 48–67%; P   0.062; Supplementary Table S4A). Yet, dicamba had negligible effects on nodule number for the other four strain inocula. Moreover, the rank correlation among strains for nodule number between herbicide environments was nonsignificant (Spearman’s r = 0.40, P = 0.33, N = 8), indicating that the strains which resulted in the most nodules without dicamba (e.g., 4386 and 2087) were often not the same as those that produced the most nodules with dicamba (e.g., 14479 and 2063). Additionally, the quantity of nodules formed was also affected by the rhizobial and plant genotype interaction (G_P   G_R; P < 0.05; Table 1B). Most strains resulted in similar nodule numbers on T. pratense genotypes, except that when paired with strain 2220, Kenland plants made significantly more (by 48%) nodules than Mammoth plants, whereas strain 2087 led Mammoth to make more nodules (by 61%) than Kenland (Supplementary Figure S3, Table S5). However, the result that the G_P   G_R   E interaction did not contribute substantially to model performance given the dataset and model selection criteria (see above), and that the G_P   E interaction was not significant (P > 0.5; Table 1B), indicates that these genotype-by-genotype patterns remained consistent across herbicide environments and that G_R more so than G_P drove the response of this mutualism trait to dicamba exposure.

Figure 2.

Herbicide exposure and rhizobial strain interacted to determine key traits of the plant-rhizobia mutualism and plant growth. Reaction norms show the change in EMM ( SE) averaged across T. pratense plant genotypes for nodule number (A), BNF (B; foliar ¹⁵N), and plant size (C; shoot biomass in mg) when plants were in symbiosis with different rhizobial strains and treated with rhizospheric dicamba herbicide. For ease of interpretation, EMMs of nodule number and plant size were back-transformed to the original scale of measurement, and EMMs of foliar ¹⁵N values were multiplied by −1, given the inverse relationship between foliar ¹⁵N and BNF.

We also found that rhizobial strains varied significantly in BNF as estimated by foliar ¹⁵N (G_R: P < 0.05; Table 1C). Similar to nodule number, dicamba in the rhizosphere did not have a main effect on BNF (E: P > 0.05), but it did significantly modify strain-specific BNF outputs (G_R   E: P < 0.05; Table 1C; Figure 2B). Interestingly, this significant G_R  E result was consistent even when nodule number was controlled for in a separate analysis (Supplementary Table S6; Figure S4), suggesting that BNF activity, despite being a product of nodulation, was not directionally related to the number of nodules produced out of symbiosis with rhizobial strains from this study. Overall, the greatest BNF activity was observed when plants were paired with strain 2220 in the absence of rhizospheric dicamba (¹⁵N = 0.320‰ 0.45, a 0.2-fold change in ¹⁵N compared to uninoculated plants), and the least occurred with strain 2316 in the presence of dicamba (¹⁵N = 2.066‰ 0.44, a 1.2-fold change in ¹⁵N compared to uninoculated plants). Nevertheless, both values fell within the range of BNF estimates previously recorded in T. pratense via foliar ¹⁵N in the field (Trněný et al., 2019). Post hoc tests did not detect statistically significant differences in BNF between dicamba-exposed vs. unexposed plants within any inoculum category (Supplementary Table S4B), suggesting that the G_R  E effect was more so driven by differences in the directional effects of dicamba on strain-specific BNF output rather than differences in the magnitude of these effects in any one direction. For example, half of the strain inocula showed slightly reduced BNF activity under dicamba (2316, 2220, 2214, and 2087: ¹⁵N increased by 0.404–1.221‰) and the remainder showed slightly increased BNF activity (4386, 14479, and 2141: ¹⁵N decreased by 0.469–0.977‰) or zero change (2063). Additionally, rank order of strain-specific foliar ¹⁵N in the absence of dicamba was not maintained in the presence of dicamba (Spearman’s r = 0.26, P = 0.54, N = 8), suggesting that herbicide exposure affected the identity of strains which resulted in the most BNF (without dicamba: 2220 vs. with dicamba: 14479) or the least BNF (without dicamba: 2214 vs. with dicamba: 2316).

Rhizobia mediated the effect of rhizospheric herbicide exposure on plant growth

The presence of rhizospheric dicamba reduced plant size at six weeks by 37% (P < 0.01; Table 1D). This effect, however, depended on rhizobial strain (G_R  E effect: P < 0.05; Table 1D). Significant size reductions from dicamba exposure were seen for all plants except those inoculated with strain 2214 or 2087 (Supplementary Table S4C).

The correlation between rhizobial benefit to plants in terms of size across strains, in the presence vs. absence of dicamba was low and nonsignificant (Spearman’s r = 0.29, P = 0.50, N = 8), suggesting that rhizospheric dicamba caused a rank shift in the inoculum which produced the largest plants. For example, without rhizospheric dicamba, the strains that were most beneficial for plant growth (i.e., highest ranking) were 2316 and 2220, but with rhizospheric dicamba, they became neutral (Figure 2C). Meanwhile, strain 2214, which was among the lowest ranking inoculum without dicamba, became the highest ranking with dicamba, suggesting that this strain tended to ameliorate the effects of dicamba on plant growth. In contrast, inoculation with strains 2063, 14479 and especially 4386 tended to worsen the effect of the herbicide—although these strains promoted average growth without dicamba, in the presence of dicamba these strains led to even smaller plants compared to their counterparts grown in the absence of dicamba (Supplementary Figure S5).

Finally, we found that the overall effect of rhizobial inoculation on plant growth was herbicide-dependent. Presence of dicamba in the rhizosphere weakened the mutualism by lessening the benefits of the partnership from the plant perspective (Figure 3). In the absence of dicamba, the average rhizobial effect on plant size was positive (mean rhizobial effect = +0.75 mg) as predicted by mutualism theory. Yet in the presence of dicamba, it turned somewhat costly (mean rhizobial effect = 0.26 mg; t = 3.82, df = 15, P = 0.0016).

Figure 3.

Herbicide exposure reduced the plant growth benefits of rhizobia. Reaction norms show the change in the mean rhizobial effect on plants, i.e., the difference in plant size with vs. without rhizobia, ( SE) according to herbicide treatment and rhizobial strain inoculum. Plant size was measured using shoot biomass at six weeks post-inoculation. The dashed line at 0 represents the baseline size for plants without rhizobia (uninoculated). Panels display results for different plant genotypes: Kenland (left) and Mammoth (right).

Discussion

Our results provide a new insight that genotype-by-environment (G E) interactions—driven primarily by rhizobial genetic variation—can mitigate the consequences of herbicide contamination in the keystone mutualism between leguminous plants and N-fixing rhizobia. Specifically, we showed that rhizospheric exposure to the widely-used auxinic herbicide dicamba, at a level relevant to contemporary off-target exposures, universally delayed rhizobial colonization of roots via nodulation, but its effects on the number of nodules produced and BNF depended on rhizobial partner. Concordantly, we provide a novel example for how microbial symbionts could serve as an extended genotype for their hosts and also a potential defense mechanism: while dicamba exposure was detrimental for plant size overall, the degree of this effect was mediated by rhizobial strains. Below, we discuss our findings, their potential causes, and implications for plant and mutualism evolution in the agro-ecological interface.

Firstly, our results demonstrated that a low-dose exposure to dicamba in the rhizosphere delayed the initiation of plant-rhizobia symbioses by almost a full week, a result that could possibly be explained by interference of symbiotic signal exchange by dicamba molecules. Although dicamba has yet to be specifically tested, other chemicals which mimic hormones, including the synthetic auxin 2,4-D, can interfere with rhizobial nod gene expression, which is essential for cross talk between plants and rhizobia (Fox et al., 2004). As nodules are the precursor to BNF, this effect on the timing of nodulation implies that auxinic herbicide-exposed ecosystems might experience a delay in receipt of fixed N, which could have variable effects on plant fitness, depending on species-specific nutritional requirements. Thus, these possible alterations in the initiation of plant-rhizobia interactions could have downstream effects on the assembly and evolution of agro-eco plant communities (Blackshaw et al., 2004; Wilson & Tilman, 1993), as well as the growth and yield of agricultural crops that utilize the plant-rhizobia symbiosis as a source of N (Fox et al., 2007). And if our microcosm experiments translate to natural settings, then our results suggest that these knock-on effects will be broadly seen, regardless of genetic variation among resident legumes or rhizobia.

Our study also provides rare yet convincing evidence that rhizobial genetic variation is a prominent genetic driver of other key mutualism attributes in response to off-target auxinic herbicide exposure in the rhizosphere. Nodule number, BNF activity, and plant growth were all highly dependent on the G_R   E interaction. Therefore, although the two T. pratense genotypes lacked variation in response to dicamba exposure (possibly because of high conservation in genes related to auxin metabolism; Busi et al., 2018), genetic variation among rhizobial partners gave rise to phenotypic variation upon which selection could potentially act to influence plant evolution. This finding was particularly transformative because although other studies have indicated that some rhizobia can mitigate plant stress caused by anthropogenic chemicals in the rhizosphere (Ahemad & Khan, 2010; Bianucci et al., 2017; Mårtensson, 1992), to our knowledge, ours is the first to provide evidence for this ability while also considering the possibility of additional biologically-relevant G_P and G_P   G_R   E effects at play. Moreover, of the limited studies that previously manipulated variation in both plants and rhizobia in other environments, most found that plant genetic variation (e.g., G_P   E_nitrogen in Porter & Simms, 2014; G_P   E_salt in Thrall et al., 2008) or its interaction with rhizobial variation (e.g., G_P   G_R  E_nitrogen; Heath & Tiffin, 2007; Heath et al., 2010) was driving mutualism responses. Here, our finding of rhizobia as a prominent genetic driver of the mutualism response to dicamba exposure is robust, even though our study was necessarily limited to two plant genotypes because: 1) we never found support for an influential G_P   G_R  E interaction based on our AIC-based model selection, and 2) in a previous study investigating the effects of dicamba drift aboveground, a strong G_R  E effect was also found across a greater number (N = 17) of plant genotypes (Iriart et al., 2024). More broadly, this result contributes to a rising body of research underscoring how microbial symbionts serve not only as an extended genotype for their hosts (Carthey et al., 2018), but also contribute to their hosts’ defense phenotype against various forms of stress (Bolin, 2025; Bolin & Lau, 2024; Yan et al., 2025).

By examining the particulars of the G_R  E interactions, we also discovered that our low dose of rhizospheric dicamba shifted the genotypic rank order of rhizobial strains that stimulated the most nodules, BNF output, or the largest plants (Figure 2). While nodule number determines the quantity of rhizobia released back into the environment, and thus reflects a component of rhizobial fitness (Burghardt & diCenzo, 2023), plant size reflects plant fitness (Younginger et al., 2017). Thus, if these fitness components fall out of positive alignment when auxinic herbicides contaminate natural rhizospheres, then evolution in response to long-term auxinic herbicide exposure could cause the mutualism to breakdown, as seen in response to other anthropogenic disturbances (Friesen, 2012; Weese et al., 2015). Additionally, for genes promoting BNF in rhizobia to be maintained, BNF must be positively correlated with rhizobial fitness. Here, we did not find a significant relationship between nodule number and BNF across both herbicide-treated and control plants (Supplementary Table S6). However, it is possible that BNF is correlated with other proxies of rhizobial fitness, such as nodule size or the number of reproductively viable cells released back into the soil. In Trifolium and other legumes, it is notable that there can be trade-offs between nodule number and nodule size and/or between nodule number and BNF activity (Gubry-Rangin et al., 2010; Iriart & Ashman, unpublished data). To better predict evolutionary trajectories, future work should focus on characterizing rhizobial fitness more precisely and elucidating genetic correlations between rhizobial fitness and BNF under varying types of herbicide environments. A quantitively rigorous approach would entail many rhizobial genotypes (i.e., ~50; Wood & Brodie, 2015).

We also found that, although there was a mitigating effect of dicamba exposure by some rhizobial strains, in general dicamba exposure negatively affected leguminous plant productivity (Figure 2C), and caused rhizobial interactions to become more costly (Figure 3). This finding expands upon recent results from Iriart et al. (2024) which showed that exposure to dicamba drift diminished the growth promoting benefits that two rhizobial strains provided to their legume partners. The biological market theory predicts that mutualisms can lead to this negative result if costs from one partner exceed the benefits returned (Schwartz & Hoeksema, 1998). Surprisingly, herbicide treatment did not have equivalently strong, negative effects on BNF across rhizobial inocula (Table 1C, Figure 2B); therefore, a reduction in N benefits to plants could not explain this particular outcome. However, considering that dicamba’s mode of action reduces photosynthesis (Gleason et al., 2011), the exposed plants in our microcosms may have lacked carbon reserves, which would make nodulation and plant facilitation of BNF more costly (Chaulagain & Frugoli, 2021; Minchin & Witty, 2005). In extrapolating these patterns to the plant population level, one would expect plant investment in rhizobial symbioses to decline if off-target auxinic herbicide exposures become common place. However, if at least a minority of resident rhizobial partners can mitigate auxinic herbicide stress, then selection could favor plant investment into these specific symbionts, resulting in possible evolutionary rescue of the mutualism (Bell, 2017). To test this hypothesis, it will be necessary for future experiments to compare the competitive fitness of strains in planta vs. in soil (free-living) under varying environmental levels of auxinic herbicide and determine what selective agents are driving rhizobial population dynamics and, by extension, rhizobial evolution.

Indeed, we identified one rhizobial partner (strain 2214) that was especially capable of mitigating the negative effect of dicamba on plants (Figure 2C; Figure 3). Interestingly, 2214 was neither a high-nodulating strain nor highly effective at BNF in either rhizospheric environment (Figure 2A-B); therefore, the ameliorating effects of this strain could be a result of reduced carbon costs, and potentially enhanced benefits unrelated to nodulation or BNF. For example, some rhizobia synthesize the enzyme ACC deaminase, which decreases levels of ethylene, a phytohormone that contributes to vegetative decay and is augmented by herbicides (Gleason et al., 2011). It has therefore been hypothesized that this enzyme plays a salient role in microbial mitigation of abiotic stress in plants, including potentially auxinic herbicide stress (Glick, 2014; Iriart & Ashman, 2025). While we did find interstrain variation in nucleotide identity for the gene that encodes ACC deaminase (acdS), strain 2214 did not possess this particular gene (Supplementary Figure S2). Nevertheless, moving forward it would be valuable to assess rhizobial gene expression using modern transcriptomic tools in order to identify which rhizobial genes are functional for alleviating auxinic herbicide stress in plants. A promising agricultural application of this work could be to engineer rhizobial inoculum that are enriched in these functional genes and, when applied, could help protect crops or other focal plants from instances of herbicide drift.

In sum, our results represent a critical first step in understanding the role of genetic variation in determining outcomes of the plant-rhizobia mutualism when both symbionts are simultaneously exposed to a contemporary herbicide in a way that simulates conditions at the agro-ecological interface. Our closed microcosm system allowed us to control the concentration of herbicide that organisms were exposed to and minimize extraneous variables. However, an important next step would be to replicate our methodologies in the field, in light of other relevant environmental factors (e.g., soil type, weather, resident microbial communities, etc.) that can influence herbicidal potency (Riter et al., 2021) and/or plant-microbe interactions (Granada Agudelo et al., 2023). Moreover, our work could be expanded by investigating a broader range of field-realistic herbicide concentrations as well as different symbiotic parings of plant and microbial species. These research pathways would give us greater power to define the ecological and evolutionary ramifications associated with continued herbicidal interference in the rhizosphere. Concurrently, they would give us greater knowledge to discern the extent to which microbial symbionts can biotically mediate the impacts of these common anthropogenic chemicals.

Data and code availability

Data and the accompanying code that underlie this work can be accessed on Mendeley Data (doi: 10.17632/whmzz32f5t.1). Sequence data of rhizobial strains have been deposited in the NCBI BioProject database (https://www.ncbi.nlm.nih.gov/bioproject/) under the following accession numbers: PRJNA1243723, PRJNA1243722, PRJNA1243721, PRJNA1243719, PRJNA1243716, PRJNA1243713, and PRJNA1241449.

Author contributions

V.I. and T-L.A. designed the study, which V.I. conducted under the guidance of T-L.A.. N. K. guided and assisted V.I. in performing genome assembly, genome annotation, and phylogenetic analysis of rhizobial strains. V.I. drafted the manuscript, which T-L.A. and N.K. edited.

Funding

Our research was made possible through generous funding provided by the Phipps Conservatory and Botanical Gardens Botany in Action Fellowship Program and the Sigma Xi Grants in Aid of Research Program (#G20230315-5016), and the DSA&S funding to T-L.A.. V.I. was supported by the National Science Foundation Graduate Research Fellowship (#1747452). N.K. was supported by the Ruth L. Kirschstein Predoctoral Individual National Research Service Award (#1F31AI179118-01). Additional funding sources include NIH U19AI158076 and the Pennsylvania Department of Health PA CURES Grant #4100085725. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health or the Pennsylvania Department of Health.

Conflict of interest

We declare no conflict of interest.