An increasing number of studies characterize plant microbiomes, moving us toward a more mechanistic and synergistic understanding of factors structuring these communities. Nevertheless, while many studies have looked at spatial variation or plant genetic variation, most studies do not simultaneously examine both soil origin and plant genotype in the same design, making direct comparisons among these effects difficult (but see [33, 62–64]). Our design, combined with sequencing microbial communities from both endosphere (root, nodule, leaf) and rhizosphere compartments, allows us to directly compare the effects of soil and genotype across these distinct “organs”. Here we show that both soil origin and plant genotype contribute to microbiome composition, but that the strength of these effects depend on the compartment - whether the microbes are inside (root endosphere) or outside (rhizosphere) plant tissues. Namely, plant genotype had much stronger effects on microbes within root tissues, while soil origin had stronger effects in the rhizosphere communities. Other interesting findings include: 1) that our results mirror those of recent studies in rice , poplar , and soybean  identifying a similar magnitude of effects and a larger role of soil origin than plant genotype in structuring microbiome variation, particularly in the rhizosphere, 2) nodule microbiomes, while containing more than the rhizobium that fixes nitrogen in Medicago nodules (Ensifer), were much less diverse than the rest of the root endosphere, and do not appear to harbor specialist microbial taxa, and 3) Ensifer was the dominant OTU, not only in the nodules, but throughout the entire plant. Medicago is a well-studied genetic model for plant-microbe symbiosis [60, 61, 65], and ours is the first NGS study of its native soil microbiome; therefore, we anticipate that our results will be of interest to many in the plant genetics community who are interested in building on our mechanistic understanding of 2-player plant-microbe interactions to better understand plant microbiomes. We discuss the major implications of our main results below.
Soil origin influences the microbiome: Soil microbial communities are remarkably diverse , and serve as a source reservoir for plant colonization. Throughout the range of a plant species, soil communities can vary considerably [67, 68], which could potentially confer variation in microbiomes. Here we find that soil origin had a larger effect than plant genotype overall and that rhizosphere communities, in particular, responded strongly to soil source. Finding significant soil origin variation suggests that we would discover additional rhizosphere taxa if we were to sample from more locations (though less so for internalized plant microbiomes, i.e., root, nodule, leaf endospheres). This is noteworthy, as rhizosphere communities are much more diverse than internalized plant communities, and rhizosphere dynamics can have profound impacts on plant fitness (reviewed by [1, 69]). Microbes are recruited to the rhizosphere by plant exudate production (reviewed by ). These exudates can provide a nutrient source and interact with edaphic conditions to generate a distinct environment from the surrounding soil that facilitates microbial growth, setting the stage for myriad microbe-microbe interactions that shape this dynamic community [1, 71]. These factors, which structure the rhizosphere, depend heavily on the abiotic and biotic facets of the soil; thus it is not surprising that soil variation impacts the rhizosphere microbiome. Despite this variation in rhizosphere communities, and the fact that the rhizosphere encapsulates the roots and is the source community for endosphere colonization, internalized plant microbiomes were consistent across soils, consistent with plant genetic, cellular, and/or biochemical mechanisms that restrict entry inside plant tissues (reviewed by [1, 16]).
Plant genotype structures the Medicago microbiome: Plant genetic variation played a role in structuring variation in the root endosphere community, but not in other compartments. Although we did not find that plant genotype structured rhizosphere communities, studies in other systems have identified such an effect [33, 63]. As mentioned above, root exudates mediate the rhizosphere community assembly, and these exudates are genetically determined . Examination of additional Medicago genotypes may very well reveal genetic variation for exudates which may confer rhizosphere variation, and/or nodule and leaf compartments. Importantly, growing plants in closed, bottom-watered boxes likely caused us to underestimate leaf endosphere diversity and miss key taxa; our leaf bacteria likely colonized through vertical migration via plant vasculature, whereas leaves in nature are often colonized from external sources [36, 37].
Plants in our experiment were grown in soils from the native range [58, 72]; therefore, we are likely capturing an ecologically-relevant and co-evolving set of microbes that colonize Medicago in nature. In particular, the root endosphere taxa that we sampled likely contain many of the “core” players within native Medicago plants, at least at the taxonomic scale sampled here (genus or above, see discussion below), because our results join many other studies showing that plants are robust filters of their environmental microbes [1, 73, 74], with soil origin having little effect on internal compartments. Indeed, in our study, microbiome diversity decreased moving from outside to inside the plant – from the rhizosphere to the root/leaf endosphere to the nodule endosphere. This filtering, operating at the boundary between the rhizosphere and the internal tissues, is likely the result of multiple selective processes , and our data join others suggesting that at least some of this filtering is plant genotype-dependent (see below).
Our data hint at the existence of quantitative genetic variation for niche breadth in plant microbiomes. We found that the similarity between the endosphere and rhizosphere varied among Medicago genotypes, suggesting that some genotypes might represent weaker filters than others. There is empirical evidence for variation along the specialist-generalist continuum within plant-microbe symbioses [75, 76], as well as among plant species, with potential applied implications for the spread of invasive legumes [77, 78]. Using five genotypes of the plant Boechera stricta, Wagner et al.  found significant genetic variation for metrics of microbiome diversity. Thus while considering such variation in the broader context of niche breadth theory can help us to make sense of plant-microbe symbiosis , our ability to interrogate the plant genes controlling microbiome diversity alongside those controlling microbiome composition grows as plant quantitative genetics and microbiome studies come together .
The variation in microbial community composition that can be partitioned among plant genotypes, and thus attributed to plant genetic variation, represents the natural variation upon which selection can act in nature and also the amount of standing genetic variation available to plant breeders interested in optimizing plant-microbe interactions. Although the main effect of plant genotype was small (~ 1.3% of total variation), the plant genotype x compartment interaction accounted for much more variation (~ 6.3%) because the effect of genotype was strong, but limited to the root endosphere compartment. These patterns generally mirror studies in other systems (e.g., [33, 62, 79, 80]). The mechanisms by which plant genotype influences the microbiome are still being elucidated, but genetic studies to date suggest that plant genes related to disease resistance, cell walls, and root hair structure may contribute [81–84].
The amount of variation explained by genotype in our experiment is likely an underestimate for multiple reasons. First, we only surveyed three plant genotypes; thus, we cannot account for any genetic variants not represented in these three genotypes. Second, our 16S survey represents species- or even genus-level variation; the long history of plant-microbe symbiosis literature has shown enormous within-species genotypic variation and genotype-by-genotype interactions for infection rates and abundance [85–89]; surely this variation is also present within at least some of the taxa in the less-studied members of the plant microbiome. Indeed a recent study showed that plant genotype-dependent shifts after multiple serial passages occurred at fine taxonomic scales among closely-related OTUs . We have little ability to incorporate these finer-scale genotypic effects using 16S surveys of community composition, though shotgun metagenomic methods for simultaneously addressing population genetics alongside community shifts are quickly evolving (e.g., [91, 92]).
The nodule microbiome: As expected, the nodule microbiome was dominated by Ensifer, and was also inhabited by a diverse community, albeit less diverse than the surrounding root endosphere and rhizosphere (Fig. 1). Despite evidence that nodule communities were deterministically sampled from the root endosphere, we did not find evidence that the nodule harbors unique microbial specialists, as no OTUs were found to be biomarkers for the nodule, besides Ensifer (Table S3). Nevertheless, we did identify multiple core nodule OTUs that were abundant throughout all nodules. The occurrence of Shewanella spp. within all nodules and at great abundance is intriguing, but further experimentation is needed to investigate the potential functional roles of this taxon.
Nodules seemingly represent a distinct environment from the adjacent root endosphere, yet many OTUs are found within nodules and across other compartments. Along these lines, there are numerous examples of non-classic rhizobial species (e.g., Pseudomonas spp., Agrobacterium spp., etc.) that possess nodulation genes, nitrogen-fixation genes, or both, being cultured from nodules, suggesting that these microbes may have a specialized role in the nodule . Nevertheless, culturing efforts have identified non-rhizobium species that act to increase nodulation . Additional unidentified, synergistic microbes of this type likely exist, but our findings suggest they will not be strictly restricted to nodules and could be cultured from root or rhizosphere communities.
Ensifer - a major actor in the Medicago microbiome: Beyond the nodule, our results suggest an extremely dominant role for Ensifer throughout the Medicago microbiome, both inside and outside the plant. Species in Rhizobiales have been found widely in plant microbiomes, including root and leaf tissues, and from a broad diversity of plants beyond legumes (e.g., [28, 79, 80, 93, 94]). Indeed we have identified Rhizobium and Bradyrhizobium in our study, and these taxa are major members of microbiome communities in multiple compartments (Table S2), yet they do not nodulate Medicago. Recent phylogenetic reconstructions  and mutant screens  suggest that such less-specific plant associations predate the origin of root nodule symbiosis in this group. Thus nodulating rhizobia may have evolved from commensal ancestors of plant microbiomes. In many rhizobia, including Ensifer, the majority of the genes governing nodulation are contained on symbiosis plasmids [97, 98], and symbiosis genes or entire plasmids can be lost as rhizobia evolve to a commensal lifestyle , though this might be unlikely, at least for pSymB, which is currently considered to be a chromid (rather than conjugative plasmid) and contains at least one essential gene .
These past observations raised the question of whether the Ensifer OTUs in the nodule were distinct from those in other plant compartments, in terms of sequence similarity as well as genome content. Further interrogation of a subset of communities using metagenomic shotgun sequencing suggested that leaf and root endosphere Ensifer likely retain their symbiosis plasmids and thus the ability to form nodules and fix nitrogen in the future. Beyond its presence outside the nodules, however, the dominance of Ensifer was striking – reaching more than 50% OTU relative abundance (Fig. 1b) even in the leaf tissue. While microbiome studies routinely identify various rhizobium taxa, this level of prevalence among all compartments is unique. One hypothesis is that Ensifer bacteria are particularly able to colonize and proliferate in host tissues due to its long-standing beneficial symbiosis with Medicago species , the result of a coevolutionary process within this group and throughout the legume phylogeny, and one that appears to have led to differentiation of signaling interactions among taxa while maximizing the signaling recognition within taxa . The ability to compete for and colonize root nodules in rhizobia is conferred by genes that act in a complex cascade of molecular “handshakes”, including nod factor, exopolysaccharides, effectors, etc. (e.g., ; reviewed by [46, 101]). It is possible that these molecules are used throughout the plant tissues to signal entry and allow Ensifer to proliferate in all compartments; this hypothesis could be tested by competing nod + and nod- strains of Ensifer (or other nodulation mutants) and testing their relative abundance across host compartments.
In our experiment, we did not find evidence that rhizobium genotype structured microbiome variation across endosphere compartments, or in the rhizosphere, although this question deserves further investigation. Abundant evidence exists demonstrating genetic variation for partner quality (i.e., the fitness benefits that the plant receives from interaction with a given rhizobium strain) among rhizobia genotypes [86, 87, 102, 103]. Given the abundance of rhizobia within plant microbiomes (beyond the nodule), it stands to reason that they could play a pivotal role in microbe-microbe interactions and thus influence plant fitness. Indeed, nodulation mutant plants have been demonstrated to elicit distinctive shifts in microbiome composition . Here we inoculated plants with individual strains of rhizobia, which is a common practice for agriculture and restoration . In future experiments, one could investigate the role of rhizobium genetic variation by inoculating plants with strains that are known to be of high vs. low partner quality, or manipulate strain identity for plants with highly specific strain preference (i.e., partner choice, which is known to vary in M. truncatula; .