More diverse rhizobial communities can lead to higher symbiotic nitrogen fixation rates, even in nitrogen-rich soils

Abstract

Symbiotic nitrogen (N) fixation (SNF) by legumes and their rhizobial partners is one of the most important sources of bioavailable N to terrestrial ecosystems. While most work on the regulation of SNF has focussed on abiotic drivers such as light, water and soil nutrients, the diversity of rhizobia with which individual legume partners may play an important but under-recognized role in regulating N inputs from SNF. By experimentally manipulating the diversity of rhizobia available to legumes, we demonstrate that rhizobial diversity can increase average SNF rates by more than 90%, and that high rhizobial diversity can induce increased SNF even under conditions of high soil N fertilization. However, the effects of rhizobial diversity, the conditions under which diversity effects were the strongest, and the likely mechanisms driving these diversity effects differed between the two legume species we assessed. These results provide evidence that biodiversity–ecosystem function relationships can occur at the scales of an individual plant and that the effects of rhizobial diversity may be as important as long-established abiotic factors, such as N availability, in driving terrestrial N inputs via SNF.

1. Introduction

Symbiotic nitrogen (N) fixation (SNF)—the process through which certain plant species and endosymbiotic bacteria convert atmospheric N into bioavailable forms—constitutes one of the most important inputs of N into the biosphere [1], facilitating key biological processes such as primary productivity [2]. Current estimates suggest that SNF accounts for approximately one-third of all natural N inputs to the biosphere [3]. Legumes (plants in the family Fabaceae) and N-fixing rhizobial bacteria are, by far, the most common SNF mutualism, comprising an estimated 5000–8000 legume species worldwide [4] and a diverse suite of alpha- and beta-proteobacteria [5]. These mutualisms serve as particularly important N sources in grasslands, agricultural systems, and tropical forests, and are common in other terrestrial biomes as well. Given the importance of legume–rhizobia SNF to key ecosystem functions such as net primary productivity and carbon sequestration, there is intense interest in understanding the ecological factors that drive SNF rates in nature.

Historically, research focussing on the ecological controls of non-agricultural legume–rhizobia SNF has focussed on how abiotic environmental factors regulate SNF rates. Empirical and theoretical evidence have shown that soil N [6], phosphorus [7,8], light [9,10], temperature [11] and water [12] can all influence SNF rates, with soil N availability being among the most commonly cited regulators of SNF. Biotic controls on legume–rhizobia SNF have received far less attention but evidence suggests that the taxonomic identity and diversity of both legumes [13] and rhizobia [14] can have important impacts on SNF in natural systems. Indeed, substantial evidence in the agricultural literature shows large effects of different rhizobial taxa on SNF [15–17], and that these effects can differ between legume species [18]. These studies suggest that incorporating the diversity of the legume–rhizobia mutualism into our framework of SNF regulation may provide important context to established work on abiotic drivers of SNF [19,20].

The immense diversity of the legume–rhizobia mutualism and the critical ecosystem function that this mutualism provides make it an ideal study system for understanding the relationship between biodiversity and ecosystem function (BEF). Legumes have long played a central role in understanding BEF relationships [21,22], but their role has been largely confined to understanding the effects of adding a member of the ‘legume functional group’ to a plant species mixture. However, diversity among rhizobial taxa and differences in their ability to nodulate and fix N mean that BEF relationships may be occurring at a scale previously underappreciated—the diversity of rhizobial endosymbionts impacting the function (e.g. growth and SNF) of a single legume host [19,23].

Syntheses of the BEF literature show broadly positive relationships between biodiversity and ecosystem function, which can be driven by several potential mechanisms. Two of the most common BEF mechanisms include selection effects and complementarity effects [22]. If a particular rhizobia is highly efficient at SNF—as has been seen in several legume–rhizobia partnerships [15,24]—the selection effect posits that a more diverse mixture of rhizobia increases the probability of encountering that highly efficient taxa. The complementarity effect could occur if different rhizobia excel at different aspects of the legume–rhizobia mutualism (e.g. one rhizobia is a good nodulator but another is an efficient N fixer once in the nodule, as has been seen with Azospirillum–Rhizobium co-inoculation [25]), creating a scenario where diverse symbiont mixtures may increase SNF. Indeed, experimental evidence shows that more diverse rhizobial communities can increase legume growth (in some cases) [16,26,27], resistance to herbivory [28] and nodulation [16].

Alternatively, competition between rhizobial taxa could reduce symbiotic benefits to the legume host in more diverse rhizobial communities—particularly, if some rhizobial taxa are effective competitors but ineffective mutualistic partners [29,30]. Several studies show that co-inoculation with multiple rhizobial strains reduces growth and nodulation relative to single-strain inoculation [31–33] and that high-density, high-diversity rhizobial populations reduce symbiotic effectiveness for less selective legume hosts [20]. Given the observed positive [26,28], neutral [15] and negative [32,33] effects of diverse rhizobial communities on legume function, the balance between positive selection/complementarity effects and negative competitive effects remains poorly understood. Moreover, the vast majority of studies on rhizobial diversity focus on impacts on legume growth or nodulation rather than SNF itself. The impact of increasing rhizobial diversity on actual SNF rates, and thus N inputs, in non-agricultural ecosystems remains untested.

To better understand the impact of available rhizobial diversity on the growth and SNF of legume hosts, we conducted a large-scale greenhouse experiment where we assessed plant growth and SNF of legumes with access to different diversities of rhizobia and different amounts of soil N. Specifically, we asked: (i) how does the diversity of rhizobial bacteria available to a legume individual impact plant growth and SNF? Based on prevailing patterns in the biodiversity–ecosystem function literature from other ecosystems (and at larger scales), we expected positive relationships between the richness of rhizobial bacteria and legume growth and SNF rates; (ii) do rhizobial diversity effects differ under different N fertilization regimes? Given the well-established impact that soil N availability has on SNF, we predicted that rhizobial diversity effects would be the strongest under low N availability and weaken with increasing levels of N fertilization; and (iii) what is the primary mechanism of any observed rhizobial diversity effect? Based on previous evidence that some rhizobia can be highly effective N fixers, we expected rhizobial diversity effects to be driven primarily by the selection effect mechanism.

2. Material and methods

(a). Study species

Our study focussed on two species of leguminous shrubs, Spartium junceum (Spanish broom) and Ulex europaeus (gorse), which are both native to Europe and invasive across many other continents including North America and Australia. We focussed on these species owing to their cosmopolitan ranges, their potential economic impacts as invasive species, and the availability of isolated rhizobial partners for each. We also chose these species because previous work suggests that U. europaeus is a generalist with respect to the number and diversity of rhizobia with which it associates and that S. junceum is more specialized in the rhizobia with which it partners [34]. Given that we focussed on rhizobial taxa that are known to associate with each legume host, we expected the effects of rhizobial diversity to be less prominent in the specialist Spartium owing to its more tightly co-evolved relationship with each rhizobial partner leading to more efficient SNF with each rhizobial strain presented.

All plants were grown from surface-sterilized seeds (soaked in 3% sodium hydroxide for 30 s followed by 5× rinsing with sterile water) purchased from Sheffield’s Seed Co., NY, USA. For each legume species, we grew populations of five pure strains of archived rhizobial bacteria previously isolated from nodules collected in California [34]. Each rhizobia taxon was originally taken from surface-sterilized wild-collected nodules, cultured and twice restreaked to obtain single-cell initiated colonies, and archived in sterile glycerol at −80°C. When possible, we explicitly used rhizobia collected from different individuals and different locations during the original sample collection in an effort to maximize diversity within the rhizobia we used. To produce the rhizobia populations used for our study, we took approximately 2 μl of archived material for each rhizobia taxon, resuspended it in sterile yeast mannitol broth, and incubated it at 25°C and 100 rpm until it reached the log growth phase. Following incubation, each population was diluted to an optical density of 0.26 to ensure that each of our rhizobia inoculations contained approximately the same number of rhizobia cells. This optical density corresponds to approximately 1 × 10⁸ rhizobia cells ml⁻¹ (Joel Sachs, personal communication, 201). Sequencing data identifying each rhizobia taxon can be found in the electronic supplementary material, table S1 and a Neighbornet diagram of genetic divergence between rhizobia taxa can be found in the electronic supplementary material, figure S1. All of our rhizobial strains demonstrated substantial genetic differences in the nifD gene region except strains UE-C and UE-E (electronic supplementary material, figure S1), but genetic differences between these two strains in other gene regions are likely.

(b). Experimental design and growth conditions

Our greenhouse experiment consisted of a total of 1100 seedlings (550 of each legume species) grown in rinsed and autoclave-sterilized (121°C and 15 psi for 60 min) fine sand placed in bleach-sterilized (3% sodium hydroxide) 164 ml ‘Cone-tainers’. Plants were assigned to treatments of rhizobial diversity and soil N fertilization following a full-factorial experimental design. We created five N fertilization treatments by adding NH₄NO₃ at rates representing total N additions of 0.51, 2, 5, 10 and 20 g N m⁻² yr⁻¹. Fertilizer additions began at the time of planting and were applied on a biweekly schedule through plant harvesting. All plants received biweekly additions of sterilized N-free Jensen’s solution [35] to relieve limitation by phosphorus and micronutrients. To estimate the per cent of each plant’s N that it derived from SNF, we applied 98%^atm isotopically enriched ¹⁵NH₄¹⁵NO₃ at a rate of 0.51 gN m⁻² yr⁻¹ to each pot spread equally across our biweekly fertilizer treatments.

Within each N fertilization treatment, we established six rhizobial diversity treatments representing mixes of 0, 1, 2, 3, 4 and 5 rhizobial taxa (for 0 rhizobia treatments, sterile water was added to pots instead of rhizobia inoculum). For each rhizobia inoculation treatment, taxa were mixed in equal parts, maintaining the same total number of rhizobia cells (rhizobial density) across treatments but altering the taxa richness of rhizobia present in each inoculum. Uniform rhizobial density was accomplished by growing each rhizobial taxon separately to the same cell density (as described above) and then mixing equal parts of each taxon stock for different diversity treatments. To assess the effects of individual rhizobial taxa, we replicated individual taxa and taxa combinations (where logistically feasible) within each diversity treatment. For our 1-taxon diversity treatment (i.e. rhizobia monocultures), we replicated each rhizobial taxa in five pots (n = 25, monoculture pots for each legume species in each N treatment). For the two-strain diversity treatment, we replicated each of 10 possible pairwise combinations of rhizobial taxa four times (n = 40, 2-taxa diversity treatment pots for each legume species in each N treatment). Because the number of possible combinations of 3 and 4 rhizobial taxa exceeded the capacity of our experiment, we created 10 pots each for the 3-taxa and 4-taxa diversity treatments and randomly sub-sampled taxa mixes from all possible combinations (n = 10, 3-taxa and 4-taxa diversity treatment pots for each legume species in each N treatment). The 5-taxa diversity treatment was replicated 10 times for each legume species in each N treatment.

We established ‘control’ plants that were not inoculated with rhizobial bacteria (n = 20, for each legume species in each N treatment). A total of 73 control plants (36.5%) were contaminated with rhizobia over the course of the experiment and were removed from analyses. Contaminated control plants exhibited only minimal nodulation. Overall, a total of 550 individuals of each legume species were grown across all N and rhizobial diversity treatments.

All plants in the experiment (both legume species and all rhizobia and fertilization treatments) were arranged randomly within a single greenhouse using the sample function in R statistical software to randomly assign a position to each plant. Cone-tainer pots were set in racks with a 4 cm empty space between each pot to minimize rhizobial cross-contamination. Seeds were planted in January 2020 and N fertilizer treatments began immediately. In February 2020 immediately following seedling emergence, we replaced any plants that did not germinate and applied rhizobia inocula to all plants (except ‘control’ uninoculated plants). Plants received equal watering for 30 min four times daily via an automated mist watering system.

(c). Plant harvest and sample processing

Following a five-month growing period, plants were harvested over a one-week period in June 2020. Plants were carefully removed from the pot to avoid breaking roots and nodules, rinsed to remove residual sand from roots, and severed at the stem base to separate belowground and aboveground tissue. Nodules were hand counted for each plant and then aboveground and belowground tissues were separately dried for three days at 60°C and massed to determine biomass allocation between plant tissue pools.

We calculated SNF as the per cent of the plant’s N derived from fixation (%N_dfa) using ¹⁵N isotope dilution. Isotopic data for a subset of 496 plants were obtained using elemental analyser isotope ratio mass spectrometry at the Cornell Stable Isotope Laboratory. To calculate %N_dfa, we used a two-end-member mixing model as follows:

$${\displaystyle \mathrm{\%}{\mathrm{N}}_{\mathrm{d}\mathrm{f}\mathrm{a}}=\left[\frac{{\mathrm{\%}}^{15}{\mathrm{N}}_{\mathrm{s}\mathrm{a}\mathrm{m}\mathrm{p}}-{\mathrm{\%}}^{15}{\mathrm{N}}_{\mathrm{s}\mathrm{o}\mathrm{i}\mathrm{l}}}{{\mathrm{\%}}^{15}{\mathrm{N}}_{\mathrm{f}\mathrm{i}\mathrm{x}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}}-{\mathrm{\%}}^{15}{\mathrm{N}}_{\mathrm{s}\mathrm{o}\mathrm{i}\mathrm{l}}}\right]\times 100,}$$

where %¹⁵N_samp is the atm¹⁵% isotopic signature of the plant, %¹⁵N_fixation is the atm¹⁵% of N derived from fixation, and %¹⁵N_soil is the atm¹⁵% of soil-derived N [36]. To account for uncertainty in our soil and fixation end members, we calculated a distribution of %N_dfa for each plant using 1000 normally distributed values for both %¹⁵N_fixation and %¹⁵N_soil. Our %¹⁵N_fixation values were based on a mean of 0.3663 atm¹⁵% obtained from the literature and an s.d. of 0.05 (conservatively accounting for more than the actual variation %¹⁵N_fixation [9]). For %¹⁵N_soil, we used the mean and s.d. atm¹⁵% from uninoculated plants for each species and each N fertilization treatment, which did not have the ability to fix N and thus had an atm¹⁵% isotopic signature entirely reflecting soil-derived N.

(d). Statistical analyses

To assess the effects of available rhizobial diversity and N fertilization on plant biomass, %N_dfa, foliar %N and biomass allocation, we used a set of linear mixed-effects regression models using the above-listed variables as the dependent variable and rhizobial diversity and N fertilization treatment as numeric independent variables including a random effect for potting rack to account for variation in conditions within the greenhouse. We ran these models for each legume species independently, as it was not our intention to directly compare effects between legumes. We determined differences between treatment levels using Tukey HSD post-hoc testing incorporating a Bonferonni–Holm correction for multiple comparisons. All statistical analyses were conducted using the base, lme4 and multcomp packages in R statistical software [37–39].

3. Results

(a). Effects of rhizobial diversity and nitrogen fertilization

First, we observed that both Ulex and Spartium plants inoculated with rhizobia grew larger than control plants receiving no rhizobia symbionts (p < 0.001 for both Ulex and Spartium) and that this was true across fertilization treatments (inoculation x fertilization interaction, p > 0.1 for both species; electronic supplementary material, figure S2). For inoculated Ulex plants, increasing the diversity of available rhizobia (while keeping the amount of rhizobia constant) significantly increased biomass, %N_dfa, and the total amount of N fixed per plant (p < 0.001 for all; figure 1a,c,e). Increasing available rhizobial diversity also significantly increased the aboveground : belowground biomass ratio (AGB : BGB) for Ulex plants (p < 0.001; electronic supplementary material, figure S3a). Increasing N fertilization did not affect Ulex biomass, %N_dfa, or total N fixed per plant (p > 0.1 for all; electronic supplementary material, figure S4 a,c,e). N fertilization also had no significant effects on Ulex AGB : BGB (p = 0.066)

Figure 1.

The effects of increasing rhizobial diversity on plant biomass (a, b), the per cent of plant N derived from fixation (%N_dfa) (c, d), and total N fixed per plant (e, f) for U. europaeus (left column) and S. junceum (right column). Light grey boxes in (a) and (b) show the biomass of uninoculated control plants for reference. Per cent N_dfa and N fixed (c-f) were assumed to be 0 for non-nodulated control plants. Lines represent partial regressions based on linear mixed-effects models with shaded regions showing 95% confidence intervals. Panels without regression lines indicate no significant diversity effects on the response variable. Boxes represent the median, lower and upper quartiles of data for each rhizobial diversity treatment. Points are jittered horizontally for visualization. Sample sizes for each diversity treatment are shown in parentheses.

Inoculated Spartium plants showed different treatment responses than Ulex. Increasing diversity of available rhizobia did not significantly affect Spartium biomass, %N_dfa, or total N fixed per plant when considering all N fertilization treatments (p > 0.1 for all; figure 1b,d,f). Contrary to Ulex, Spartium AGB : BGB significantly decreased with increasing available rhizobial diversity (p = 0.021, electronic supplementary material, figure S3b). N fertilization significantly increased Spartium biomass (p = 0.021; electronic supplementary material, figure S4b) but did not affect AGB : BGB (p = 0.41).

(b). Comparing effects of rhizobial diversity and soil nitrogen

For Ulex, the effects of increasing available rhizobial diversity were stronger than the effects of our N fertilization treatments. Increasing available rhizobial diversity from 1 to 5 rhizobial taxa increased average Ulex biomass by 110% and average Ulex %N_dfa by 93%—effectively doubling both biomass and SNF. This resulted in a 250% increase in the total amount of N fixed per Ulex plant (figure 1e). Alternatively, increasing N fertilization from 0 to 20 g N m⁻² yr⁻¹ increased Ulex biomass by an average of only 43% and decreased %N_dfa by only 18.8%, neither of which were statistically significant (electronic supplementary material, figure S4a,c). For Spartium, available rhizobial diversity had no effect on biomass (figure 1b), but increasing N fertilization from 0 to 20 g N m⁻² yr⁻¹ increased average biomass by 35% (electronic supplementary material, figure S4b).

Contrary to our expectations, we found significant relationships between available rhizobial diversity and %N_dfa for both Ulex and Spartium plants in our high (20 g N m⁻² yr⁻¹) N fertilization treatment (figure 2). Available rhizobial diversity did not significantly affect %N_dfa for Spartium plants when all N fertilization treatments were combined (figure 1d), partly driven by the lack of diversity effects in the low N fertilization treatment (figure 2b). However, increasing available rhizobial diversity significantly increased %N_dfa in the highest fertilization treatment where plants were saturated with soil N (figure 2c,d).

Figure 2.

Rhizobial diversity induces luxury SNF. Increasing rhizobial diversity had variable effects on the per cent of N derived from fixation (%N_dfa) between (a) U. europaeus and (b) S. junceum in our severely N-limited treatment (0 g N m⁻² yr⁻¹). However, in our N-saturated fertilization treatment (20 g N m⁻² yr⁻¹) where plants have ample N available in the soil, increasing rhizobial diversity increased %N_dfa for both U. europaeus (c) and S. junceum (d). Colours, boxes and points are as in figure 1. Sample sizes for each diversity treatment are shown in parentheses.

(c). Biodiversity and ecosystem function mechanisms driving the rhizobial diversity effect

Overall, we found strong evidence for a positive selection effect driving the relationship between available rhizobial diversity and %N_dfa in Ulex. A single rhizobial taxa, ‘UE-A’, was associated with significantly higher plant biomass (p < 0.001; figure 3a) and %N_dfa (p < 0.001; figure 3c) than other rhizobia available to Ulex plants. Regardless of diversity, Ulex that were inoculated with UE-A rhizobia (either entirely or in mixture with other rhizobia), grew 26% larger (p < 0.001) and fixed 48% more N (p < 0.001) than Ulex that did not contain UE-A in their inoculum (figure 4a).

Figure 3.

Differences in plant biomass (a, b) and the per cent of plant N derived from fixation (%N_dfa) (c, d) for U. europaeus (left column) and S. junceum (right column) inoculated with different individual rhizobial taxa in monoculture. Letters above boxes indicate groupings from Tukey HSD post-hoc tests, where groups that do not share a letter are significantly different from each other. Boxes, points and sample sizes are as in figure 1.

Figure 4.

(a) The significant positive effect of including a particularly effective rhizobial partner, UE-A, on the per cent of N derived from fixation (%N_dfa) in U. europaeus plants. (b) The non-significant effect of including a particularly ineffective rhizobial partner, SJ-V, on %N_dfa in S. junceum plants. Data presented combine all rhizobial diversity and N fertilization treatments. Boxes, points and sample sizes are as in figure 1.

While Ulex demonstrated a particularly effective partnership with a single rhizobial taxon ‘UE-A’, Spartium demonstrated a particularly ineffective partnership with a single rhizobial taxon—‘SJ-V’, where %N_dfa was 30% lower for plants inoculated with SJ-V in monoculture than the average for Spartium plants given monocultures of the other rhizobia (figure 3d). However, we found no evidence for the negative selection effects of SJ-V in Spartium. For Spartium plants that included SJ-V in their inocula, they exhibited both lower biomass and %N_dfa if SJ-V was the only bacteria they were inoculated with (i.e. monocultures; figure 3b,d), but there was no significant effect of SJ-V on either biomass or %N_dfa for inocula mixtures of two or more rhizobia that contained SJ-V (figure 4b; electronic supplementary material, figure S6).

4. Discussion

The emergence of the study of BEF relationships has had an astounding impact on our understanding of ecosystem regulation over the past three decades. While positive diversity effects assessed at the plot scale are common in the BEF literature, our results provide an important empirical demonstration of the hypothesis that these effects can operate at the scale of a community of endosymbionts influencing the ‘ecosystem function’ of a single host [19,23]. We found clear evidence that available rhizobial diversity can have strong impacts on the growth of a host legume and provide direct evidence that available rhizobial diversity can increase N inputs via SNF.

When contextualizing the impacts of rhizobial diversity, it is important to consider the synergistic nature of BEF relationships on different aspects of host plant function. For example, the per cent of each Ulex plant’s N derived from fixation (%N_dfa) increased by 93% between 1- and 5-taxa rhizobial diversity treatments, but the total biomass (and thus the total amount of N in the plant) also increased by 110% across these diversity treatments. Combined, these diversity effects on plant biomass and %N_dfa result in a 250% increase in the total amount of N fixed per plant—a substantial increase in the ecosystem N input of an individual plant.

One of the most striking results of our study was how strong the effects of rhizobial diversity can be, even when compared with the well-established abiotic regulation of SNF by soil N fertilization. Most ecosystem theory on SNF assumes that SNF is downregulated when soil N is abundant because N uptake is energetically cheaper than SNF [6,40]. Recent work has shown that the density of rhizobia inoculation can have more important effects on legume growth than soil N availability [20]. Our data demonstrate that even when rhizobia density is kept constant, the difference between 1 and 5 rhizobial taxa provided to a plant can, in some cases, have much larger impacts on plant biomass and SNF than even agricultural levels of N fertilization (e.g. 10 g m⁻² yr⁻¹ or more; figure 2).

These data also show that under conditions of N abundance, where theory suggests plants should downregulate SNF (our high N fertilization treatments), plants exposed to more diverse communities of rhizobia exhibit higher SNF—a pattern seen in both of the legume species in our study (figure 2). This differs markedly from our expectations and contrasts with evidence from the agricultural literature that co-inoculation only impacts SNF in N-limited conditions [15]. If high rhizobial diversity induces ‘luxury fixation’ in legumes over large spatio-temporal scales in nature, it could provide a potential mechanism for ecosystem N surpluses such as those seen in many tropical forests [41] that harbour high rhizobial diversity [42]. Together, these data suggest that rhizobia-driven BEF effects on legume function are not merely scientific novelties, but could have important implications for plant growth and ecosystem N cycling.

The results showing available rhizobial diversity increasing ‘luxury SNF’ highlight an important aspect of our study: the direct measurement of N fixation activity in response to available rhizobial diversity. While much important work has looked at how different aspects of rhizobial diversity impact legume growth and nodulation [15,20,26,32,33], these metrics are often inadequate proxies of actual SNF. For example, our data showed extremely weak relationships between nodule number and SNF for the two legume species in our experiment (R² = 0.057; electronic supplementary material, figure S7), highlighting the need for future studies to incorporate direct measurements of SNF using approaches such as ¹⁵N tracing or acetylene reduction assays [36] rather than relying on nodule number to approximate this critical ecosystem function.

We note, however, that BEF effects were not ubiquitous even within the scope of our single greenhouse experiment. While Ulex demonstrated strong positive BEF effects, BEF effects in Spartium were context-dependent—occurring only in the highest N fertilization treatment (figure 2). BEF relationships often show substantial variability across ecosystems, and an analogous variability in these relationships may exist across different host–symbiont partnerships. Thus, it is possible that rhizobial diversity effects simply are not a strong driver of N-limited biomass or SNF in S. junceum. It is also possible that differences between Ulex and Spartium are either an artefact of the particular rhizobial taxa used in our experiment—which were a subset of the rhizobia that each plant actually partners within nature [34]—or differences in how selective the two legume species are for their rhizobial partners, which has been shown to influence effects of diverse rhizobial communities [20]. Thus, it is possible that BEF effects may be more important for Spartium growth and SNF in natural systems where they are exposed to a wider diversity of rhizobia.

Our data suggest that selection effects are the most likely mechanism for the positive BEF relationships we observed in Ulex plants. The positive influence of increasing available rhizobial diversity was primarily an effect of increasing the probability of including UE-A in more diverse mixtures of rhizobia (figure 4a), and this effect was also important in driving luxury SNF in Ulex plants (electronic supplementary material, figure S8). There was no significant effect of rhizobial diversity for all Ulex plants that included UE-A (electronic supplementary material, figures S5a and S8a), but plants that were not exposed to UE-A across all diversity treatment levels performed worse than those that were exposed to UE-A (figure 4). We found no evidence for complementarity effects (over-performance of particular rhizobia combinations) or particularly strong facilitative or antagonistic interactions between rhizobial strains for any pairs of rhizobia in either legume host species (electronic supplementary material, figure S9), which has been observed elsewhere [43]. However, it is important to note that we lack direct evidence for these mechanisms because we do not know exactly which rhizobia actually partnered with each plant in our experiment—only the taxa richness of the potential rhizobia pool the plant had access to. Our evidence for selection effects driving positive BEF patterns in a legume–rhizobia system aligns with theoretical expectations for mutualists that perform redundant functions for the host (such as rhizobia) being more likely to create positive BEF patterns via the probability of a particularly effective partner taxa [19,27].

Interestingly, we found that positive and negative selection effects [27] were not symmetric. The particularly effective rhizobia UE-A and ineffective rhizobia SJ-V had important positive and negative effects, respectively, on their host plants when these rhizobia were presented in monoculture. However, the particularly effective UE-A created important positive BEF effects in Ulex but the particularly ineffective rhizobia SJ-V did not lead to overall negative BEF effects in Spartium (figure 4b). That is, increasing rhizobial diversity in Spartium, which increased the probability of including the ineffective SJ-V, did not decrease plant function. Work in the clover–rhizobia system has found evidence for similar asymmetry in selection effects, where the influence of the most beneficial rhizobial strain predicts performance in diverse rhizobia mixtures, but the least beneficial rhizobial strain was a much poorer predictor of performance for diverse rhizobial mixes [44]. This suggests a functional rescue effect where more diverse rhizobial communities in the soil could mitigate the negative functional effects of a particularly ineffective partner strain. The most likely mechanisms for these patterns are host partner choice [45–47], host sanctioning [29,48,49] and reciprocal rewards [50]. If effective partnerships are promoted and ineffective partnerships are sanctioned within a plant, we would expect the inclusion of particularly effective rhizobial taxa (even in a diverse mix of taxa) to increase plant function while particularly ineffective rhizobial taxa should be inhibited from increasing its abundance and have minimal negative effects on plant function [27]. Thus, the diversity effects we see in both Ulex and Spartium species may provide indirect support for fundamental mechanisms that maintain mutualisms over evolutionary time [30,51,52].

It is important to acknowledge three potential caveats in our data. First, the positive effects of available rhizobial diversity were almost certainly owing to selection effects, which some do not consider to be effects of diversity per se, given that the effect is simply the occurrence probability of a single taxa. Because we were interested in understanding the effects of available rhizobial diversity in soils on the function and ecosystem impacts of legumes in natural systems, we consider the effects of a diverse group of available rhizobial partners to be equally important whether they are produced via selection effects, complementarity effects or another mechanism. Second, we did not sequence the rhizobial communities inside the nodules of our plants and thus only have direct knowledge of the diversity of rhizobia available to a legume, not the rhizobia that the legume actually partnered with. Again, our interests in this experiment were in understanding the effects of the potential pool of rhizobia in the soil that a legume could partner with, which our diversity inoculation treatments facilitated. Third, a portion (73) of our control plants formed a small number of nodules, indicating some cross-contamination of rhizobia between adjacent pots by the end of the experiment. While we have no ability to quantify the extent of this cross-contamination, any cross-contamination would only serve to homogenize our diversity treatments, potentially reducing any observed effects of available rhizobial diversity. This suggests that, if anything, the strong diversity effects we see in our data may be underestimates of the importance of available rhizobial diversity on legume growth and SNF.

Overall, our results provide important empirical evidence that BEF relationships can be important at the scale of individual mutualist hosts, promoting emerging lines of research in both the BEF and legume–rhizobia literature [19,27]. Important future avenues of research include understanding how these relationships manifest in nature, how common and important legume–rhizobia BEF relationships are across legume phylogeny and environmental conditions, how we can effectively and ethically use these relationships to improve land management, and how common these types of host–symbiont BEF relationships are across other symbiotic mutualisms. Thus, these results highlight the need to further explore the general role of partner diversity in mutualism function and specifically highlight the role of rhizobial diversity driving the impacts of N-fixing legumes on terrestrial ecosystem N inputs.

Ethics

This work did not require ethical approval from a human subject or animal welfare committee.

Data accessibility

All statistical analysis and figure code along with supporting data files are publicly available on Dryad [53].

Supplementary material is available online [54].

Declaration of AI use

We have not used AI-assisted technologies in creating this article.

Authors’ contributions

B.N.T.: conceptualization, data curation, formal analysis, funding acquisition, investigation, methodology, project administration, resources, visualization, writing—original draft, writing—review and editing; K.J.K.: conceptualization, formal analysis, funding acquisition, methodology, project administration, resources, supervision, validation, visualization, writing—review and editing.

Both authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Conflict of interest declaration

We declare we have no competing interests.

Funding

We thank the Maryland Native Plant Society for funding support. B.N.T. was supported by a Smithsonian Institute Postdoctoral Fellowship.