The symbiotic capacity of rhizobium shapes root-associated microbiomes

Background: Root-microbiome interactions are of central importance for plant performance and yield. A distinctive feature of legumes in this context is that they engage in symbiosis with rhizobia, which are abundant in soils and include both symbiotic and non-symbiotic bacterial strains. If and how the capacity of rhizobia to form symbiosis modulates root-associated microbiomes are not well understood. Results: We address this question by inoculating soybean (Glycine max) plants with wild type (WT) or a noeI mutant of Bradyrhizobium diazoeciens. The noeI mutant produces a defective Nod factor and is thus compromised in its ability to establish functional symbiosis. Compared to soybean plants inoculated with WT rhizobia, plants inoculated with the noeI mutant showed a signicant decrease in nodulation and root-avonoid exudation, and exhibited strong changes in microbiome assembly in the rhizosphere and the rhizoplane. NoeI mutant-inoculated roots exhibited reduced diversity, co-occurrence interactions and a substantial depletion of benecial microbes on the roots. The effects of the noeI mutation were absent in soils without plants, demonstrating that they are plant dependent. Complementation experiments showed that avonoid supplementation is sucient to restore recruitment of benecial microbes. Conclusion: The results illustrate that the capacity of a rhizobium to form microbial symbiosis dramatically alters root-associated microbiomes, most likely by changing root exudation patterns. The results of this study have important implications for our understanding of the evolution of plant-microbiome interactions in the context of plant-bacterial symbioses.

The legume-rhizobial symbiosis begins with the secretion of avonoids by roots of the plants. Plant avonoid exudation is regulated by the presence of the bacteria, which speci cally recognize the compounds by their NodD receptors [35,36]. This in turn induces the bacteria to release Nod factors (NFs), which are signal molecules synthesized by their nodulation genes (nod, nol, and noe) [29,30]. Most rhizobial nodulation genes are located on transmissible genetic elements such as the symbiotic plasmids or islands and can be transferred horizontally at high frequency within the species [37]. On the plant side, lysin motif (LysM) receptor kinases recognize and bind the compatible NFs, and then initiate the accommodation of the rhizobia and the nodule-formation process [29,38]. Rhizobial nodulation genes and plant symbiotic signaling genes, including NF receptor genes and downstream common symbiotic signaling pathway (SYM) genes, which are shared with arbuscular mycorrhizal (AM) symbiosis, are necessary to establish the symbiotic relationship and nodule development [29,30,39]. Previous studies have shown that genetic variation in plant genes encoding the common SYM receptor of in Lotus japonicas [40,41], Glycine max [42,43], Medicago truncatula [44] and the non-leguminous Oryza sativa [45] drive the establishment of distinctive root-associated microbiomes. By contrast, the impact of microbial genetic variation such as in rhizobial nodulation genes on root-associated microbiomes remains unknown. This is particularly relevant in the light of recent ndings showing that rhizobia acquired the key symbiosis genes multiple times, and that the most recent common ancestor was able to colonize roots of many different plant species [46], begging the question if and how the evolution of symbiosis affects plant-microbiome interactions.
The common nod genes nodA, nodB, and nodC are responsible for synthesizing the core structure of the NFs and are necessary for most symbiosises, while other nodulation genes encode the speci c modi cations on the backbone of signaling compounds and have effects on host speci city [47]. The noeI gene is responsible for the methylation of the fucose moiety at the reducing end of the NFs [48]. Previous studies have found that noeI was not essential for Sinorhizobium fredii HH103 and Rhizobium sp. NGR234 nodulating several host plants [48,49]. However, a recent study conducted on B. diazoe ciens USDA 110 found that noeI has a vital role in maintaining nitrogen xation e ciency in soybean [50]. While nodulation phenotypes and host nitrogen status are known to have an impact on the structure of root-and shoot-associated microbiomes in soybean [42,51], the effect of genetic variation in nitrogen-xing symbionts is unknown.
In this study, we investigated the role of genetic variation in the noeI gene of B. diazoe ciens (strain USDA 110) in regulating the assembly of the soybean root-associated microbiota. We sampled ve compartments (rhizosphere, rhizoplane, endosphere, nodules and unplanted soil) to determine the direct and plant-mediated effects of the noeI gene mutation on the composition and diversity of root-associated bacterial communities. Further, we investigated the potential role of plant avonoids in triggering these effects. Our results reveal that noeI determines the composition of root-associated microbiota through plant-mediated effects such as increased avonoid exudation. These ndings shed light on the mechanisms underlying the relationship between root-microbe symbiosis and root-associated microbial communities.

Soil
Soil samples were collected from a perennially ooded paddy eld located in Leshan, Sichuan Province, China (29.2593 N, 103.9403 E). Surface soil was collected at a depth of 0 to 20 cm through a " ve points" sampling strategy in a 25 m × 25 m eld. All soil samples were transported immediately to the laboratory on ice and stored at 4 °C. Plant residues, roots, and stones were removed, and the soil was drained well enough to pass through a 2 mm sieve. These soils were used in greenhouse batch experiments; they were chosen as they contain no native compatible rhizobia that can nodulate with Glycine max variety C08. Greenhouse experiment and symbiotic phenotype testing The greenhouse experiment was of a complete factorial randomized block design (Fig. 1a) that consisted of two rhizobial genotype treatments and two planting patterns. The rhizobial genotype treatments included: 1) B. diazoe ciens USDA 110 wild type, isolated from soybean [52]; and 2) B. diazoe ciens USDA 110 noeI mutant, obtained from our previous study [50]. The two planting patterns were 1) planted with cultivated soybean (C08) and 2) intact soil without plants (unplanted). Planted and unplanted soils that were not inoculated with rhizobia were instead inoculated with sterile 0.8% NaCl (w/v) solution as negative control treatments. As such, the negative control of unplanted soil is also referred to as bulk soil.
Soybean seeds were selected for fullness and uniformity before being surface-sterilized in 95% ethanol for 30 seconds and then further sterilized with 2.5% (w/v) sodium hypochlorite (NaClO) solution for 3-5 minutes, after which they were rinsed seven times with sterilized deionized water. The sterilized seeds were germinated on 0.8% water-agar (w/v) plates in the dark at 28 °C for 36-48 h. Uniform germinated seedlings were selected and transferred into pots (10 by 12 cm height by diameter) containing 500 g of soil. Each treatment was inoculated with 1 mL of rhizobial culture (optical density at 600 nm [OD 600 ] concentration of 0.2, diluted with 0.8% NaCl solution), as described in our previous study [50]. Plants were grown in the greenhouse (day/night cycle 16/8 h, 25/16 °C and a relative humidity of 60%) and were harvested 45 days post-inoculation (dpi). Several symbiotic phenotypes were recorded for plants inoculated with the wild type and the mutant. Leaf chlorophyll concentrations were determined using a SPAD-502 meter (Konica Minolta, Osaka, Japan) [53]. Plant height, weight of fresh nodules and the number of nodules were measured after sampling and shoot and root weights were determined after being dried at 65 °C for 5 days. Nodule nitrogenase activity was measured using the acetylene reduction method as described in Buendiaclaveria et al. [54].
Sampling of unplanted soil, rhizosphere, rhizoplane, endosphere, and nodule The method for sampling unplanted soil, rhizosphere, rhizoplane, endosphere and nodules followed the protocol described Edwards et al. [55] with the following modi cations. Brie y, the plants were removed from each pot and the loosely attached soil on the roots was removed with gentle shaking, leaving the root-adhering soil layer (approximately 1 mm of soil). The soil collection steps were performed on ice. Firstly, the roots were placed in a sterile 50 mL falcon tube containing 30 mL of sterile pre-cooled PBS (phosphate-buffered saline) buffer (with pH 7.3-7.5) and vortexed for 15 s, and the turbid solution was ltered through a 100-μm aseptic nylon mesh strainer into a new 50-mL tube to remove root fragments and large sediments, followed by centrifuging for 5 min at 12,000 × g at 4 °C. The supernatant was discarded, and the soil washed from the roots was de ned as rhizosphere soil, which was then frozen with liquid nitrogen and stored at -80 °C. For rhizoplane samples, the washed roots were transferred to a falcon tube with 30 mL PBS and sonicated for 30 s at 50-60 Hz twice. The roots were then removed, and the rhizoplane samples was collected by centrifugation at 12,000×g for 5 min at 4 °C and stored at -80 °C until DNA extraction. The washed roots were cleaned and sonicated again as described before to ensure that all microbes were removed from the root surface. Two more sonication procedures using clean PBS solution were performed, and the sonicated roots were surface-sterilized in 70% (v/v) ethanol for 2 min and then in 2.5% (w/v) NaClO solution for 5 min, followed by washing with PBS solution for seven times. The root nodules were collected by separating them from roots using sterile blades. manufacturer's recommendations and index codes were added. The library quality was assessed on the Qubit@ 2.0 Fluorometer (Thermo Fisher Scienti c, MA, USA) and Agilent Bioanalyzer 2100 systems (Agilent Technologies, Waldbronn, Germany). Finally, the library was sequenced on an Illumina_Hiseq2500 platform and paired-end reads of length 250 bp were generated (Guangdong Magigene Biotechnology Co., Ltd. Guangzhou, China). The resulting paired sequence reads were then merged, trimmed, ltered, aligned, and clustered to de ne the operational taxonomic unit (OTU) using USEARCH v.11.06 [56]. Brie y, sequences with ≥ 97% similarity were assigned to the same OTU by the UPARSE-OTU algorithm in USEARCH; and chimera detection was performed with VSEARCH 2.11 [57]. Putative chimeric sequences and singletons were discarded.

Root exudate collection and UPLC-MS/MS analysis
Full and uniform soybean seeds were surface sterilized and germinated as described above. To enhance root growth, germinated seedlings were transferred to sterile pots containing sterile vermiculite and grown in the greenhouse for 7 days under the same conditions as described above. At harvest, the soybean plants were pulled from their pots and washed to remove the vermiculite, then four plants were transferred to a 9-well sponge lattice placed in a glass jar (12.6 cm in height and 8.5 cm in diameter) containing 100 mL 25% (v/v) of sterile nitrogen-free Rigaud-Puppo solution [58]. The plant roots grew through the holes of the lattice into the nutrient solution. These hydroponics systems were inoculated with 4 mL of USDA 110 WT and noeI mutant cultures as described above with 4 mL 0.8% NaCl added to the control samples. To provide an aerobic environment for rhizobia, oxygen was pumped into the nutrient solution; each treatment contained three replicate hydroponics systems. The systems were incubated for 7 days in a climate-controlled growth chamber (day/night cycle 14/10 h, 28/16 °C and relative humidity of 60%). To check the sterility of the hydroponics systems, aliquot of 500 µL from each system was spread and cultured on tryptone-yeast (TY) medium plates. Soybean root exudates were collected by centrifugation at 10,000 rpm for 20 min (5 °C), ltered using a 0.25-µm cellulose nitrate lter and then stored at -20 °C until further analysis.
Eleven standard avonoids (supplied by J&K or ANPEL) were determined during experiment: naringenin, hesperetin, genistein, daidzein, 7, 4′-dihydroxy avone, apigenin, chrysin, luteolin, isoliquiritigenin, morin, coumestrol; deuterated genistein was used as the internal standard. The calibration curve was prepared by the serial dilution of a mixture of eleven standards by methanol with concentrations as follows: 50, 25, 10, 5, 1, 0.5, 0.1 μg/L. The internal standard was also added to all samples to achieve a nal concentration of 10 μg/L. The calibration curve was obtained by plotting the peak area ratio (y) of the standard to the internal standard versus the ratio of their concentrations (x). The curve was tted to a linear function with a weight of 1/nx (R 2 > 0.99), with "n" being the calibration level. The concentrations of the compounds in the sample were determined by their peak area ratio with the internal standard and were determined using the calibration curve. All standards and samples were ltered through a PTFE syringe lter (0.22 μm) and stored at -80 °C until further analysis.
The internal standard was added to each hydroponics culture (100 mL) to give a concentration of 10 μg/L after which the solution was passed through a Resprep C18 solid-phase extraction cartridge [Sep-Pak Vac 6cc (500 mg), Waters, USA]. Flavonoids were eluted by 10 mL methanol and then freeze-dried with liquid nitrogen. For quanti cation, samples were resuspended in 1 mL of 50% (v/v) methanol solution and 10 µL aliquots were injected into a Waters ACQUITY I-class UPLC coupled with Xevo TQ-XS Triple Quadrupole Mass Spectrometer in the electrospray ionization negative mode (Waters, USA). Liquid chromatography was performed on a 100 mm × 2.1 mm BEH C 18 column with a particle size of 1.7 µm.
The mobile phase consisted of solvent A (water) and solvent B (100% acetonitrile) and the ow rate was 0.3 mL/min. The optimized linear gradient system was as follows: 0-1 min, 5% B; 1-10 min, 35% B; 10-12 min, 95% B; 12-15.5 min, re-equilibrium to 5% B. The parameters of the mass spectrometer were as follows: capillary voltage 2.5 kV, cone voltage 80 V, desolvation temperature 600 °C, desolvation gas ow 1100 L/h, cone gas ow 250 L/h, nebulizer gas ow 7 bar, and collision gas ow 0.15 mL/min of argon. A multiple reaction monitoring (MRM) mode was employed for quantitative analysis. Mass spectral parameters were optimized for each analyte and are shown in Supplementary Table S1.

Impacts of the mixture of avonoids on soil microbiome
To determine the effect of avonoids on the structure of the soil microbiota, watery solutions were prepared containing a mixture of the eleven avonoid standards according to the quantitative analysis of avonoids secreted by soybean. The nal concentration of daidzein was 1 µg/g, and the other ten avonoids were added following their ratios to daidzein. From the soil described above, 100 g were placed into pots and pre-incubated under the greenhouse conditions described above for one week to activate the soil microbiomes. 1 mL of the mixture solution was added into each pot twice a week for 4 weeks. The control treatment had the same volume of sterile water added; each treatment consisted of three replicates. All pots were watered twice a week during the incubation period. The soil samples were collected after incubation, with DNA extracted and the 16S rRNA gene sequenced and analyzed as described above.
Physicochemical characterization of soil The soil physicochemical characteristics of each treatment were measured following the methods described by Bao [59]. Soil pH was measured using a suspension of soil and deionized water at a ratio of 1:2.5 (w/v). Soil total C, N, H and S contents were determined separately using an elemental analyzer (Flash EA 1112, Thermo Finnigan). DOC and DON were measured using a TOC analyzer (Multi N/C 3100, Analytik Jena AG). Soil exchangeable Na, K, Ca and Mg were extracted with 1 M ammonium acetate and measured by atomic absorption spectrophotometry (NovAA300, Analytik Jena AG). CEC was measured in a continuous colorimetric ow system (Skalar SAN++ System, Netherlands).

Statistical analysis
The resulting OTU table was normalized by the negative binomial model using the package phyloseq [60] in R (version 3.6.0). Weighted UniFrac [61] distances were calculated from the normalized OTU tables using the R package vegan, Principal coordinate analyses (PCoA) utilizing the weighted UniFrac distances to assess the differences in microbial communities between treatments. To measure the βdiversity signi cance, permutational multivariate analyses of variance (PERMANOVA) was conducted using the function adonis in vegan [62]. Shannon, Chao 1 and Fisher indices and the number of observed species were calculated using the function diversity in R package vegan. Kruskal-Wallis tests followed by Dunn's multiple-comparison test were performed to assess differences between treatments. The statistical analysis of taxonomic and functional pro les (STAMP) was applied to identify different species associated with rhizobial treatments [63]. To explore the correlation between microbial communities and environmental properties, weighted UniFrac distance-based RDA (db-RDA) and Variation partitioning analysis (VPA) were performed using the function capscale and varpart in the package vegan, respectively. To determine OTU enrichment in each treatment, a generalized linear model (GLM) approach using edgeR [55] was conducted. Microbial co-occurrence networks were constructed based on Spearman correlations among 300 dominant OTUs. The nodes in this network represent OTUs and links indicate potential microbial interactions. We adjusted all P-values of the correlation matrix using the Benjamini and Hochberg FDR controlling procedure. The indirect correlation dependencies were distinguished using the network deconvolution method [64]. The subnetworks for various compartments were induced based on OTU-presenting in corresponding samples. The cutoff for correlation value was determined through random matrix theory (RMT)-based methods [65]. Network properties were calculated with the igraph [66] package in R and visualized in Gephi 0.8.2 [67]. Fisher's Least Signi cant Difference (LSD) test (p < 0.05) and Duncan multiple-comparison test (p < 0.05) using R package agricolae [68] were employed to analyze the difference of soybean symbiotic phenotypes and relative abundance of bacterial taxa, respectively. All gures in this study were generated using ggplot2 [69] in R and OriginPro 2017.

Results
A mutation in noeI of B. diazoe ciens suppresses soybean nodule formation Nodulation genes are essential for the establishment of symbiosis between legumes and rhizobia. To con rm the role of noeI in nodulation, we inoculated soybean roots with WT and noeI mutant of strain B. diazoe ciens USDA110 and then screened the roots for nodule formation (Fig. 1a). Inoculation with WT bacteria resulted in the formation of > 20 nodules a total weight of > 17 g, and a nitrogenase activity of > 45 nmol*h − 1 *mg − 1 per plant (Fig. 1b-e). The mutation of rhizobial noeI signi cantly impaired the nodulation e ciency of USDA 110 in soybean, with signi cantly lower nodule numbers, nodule weight and nitrogenase activity (Fig. 1b-e). The number of nodules was reduced to < 2 nodules per plant, and the nitrogenase activity dropped to < 0.5 nmol*h − 1 *mg − 1 , thus con rming that noeI is not strictly essential, but promotes nodule formation in soybean. No nodules formed in plants grown on soils treated with sterile control solution, and no nitrogenase activity was detected, showing that native nodule-forming rhizobia are absent in the experimental soil (Fig. 1b). As plants were well-fertilized, plant height, leaf chlorophyll content and shoot and root dry weights did not differ among treatments (Fig. 1f-i), thus allowing us to assess the impact of noeI-dependent symbiosis on microbial communities independently of plant performance.

Compartment-speci c modulation of microbial communities by noeI
To determine whether the noeI mutation altered the unplanted soil and/or soybean root-associated microbiomes, DNA was extracted from all compartments and bacterial community pro les were determined using amplicon sequencing of the 16S rRNA gene. After quality ltering and chimera removal, 6,302,405 sequences (mean, 68,504 per sample) were obtained from 92 samples and 5,667 microbial OTUs were identi ed at 97% sequence similarity. Alpha diversity was measured using Shannon, Chao1 and Fisher indices as well as with the number of observed OTUs (richness). Alpha diversity was highest in soil, rhizosphere and rhizoplane, intermediate in the root endosphere and lowest in root nodules (Fig. 2a, Supplementary Fig. S1a-c). In the rhizosphere and rhizoplane compartments, α-diversity was similar following WT-and control-inoculation, but signi cantly lower following inoculation with the noeI-mutant (p < 0.05) (Fig. 2a, Supplementary Fig. S1a-c). In the endosphere, α-diversity was higher in WT-and noeI mutant-inoculated samples than control samples (Fig. 2a, Supplementary Fig. S1a-c). No differences between treatments were found in unplanted soil and nodules (Fig. 2a, Supplementary Fig. S1a-c).
Taxonomy analysis revealed differences in the relative abundance of taxa at class level between WT and mutant treatments in the rhizosphere and rhizoplane; most bacterial classes were less abundant in the samples inoculated with the noeI mutant when compared to those inoculated with the WT strain ( Supplementary Fig. S2a, b). This relationship was not observed in unplanted soil ( Supplementary Fig.   S2a, b). Ktedonobacteria, Planctomycetia, Caldilineae, and Sphingobacteria classes differed signi cantly between WT and mutant treatments in the rhizosphere (p < 0.05) ( Supplementary Fig. S2a). The relative abundance of the predominant bacterial classes was signi cantly different between unplanted soil and endosphere compartments (p < 0.05), but the differences between WT and mutant treatments were not as distinct ( Supplementary Fig. S2c). As expected, a pattern of reduced microbial complexity and signi cantly different relative abundance was found in nodules compared to those of unplanted soil (p < 0.05) (Supplementary Fig. S2d). Taxonomic assignments at the family level using relative abundance revealed that the nodules in both treatments were dominated by bacteria belonging to the families Bradyrhizobiaceae and Nannocystaceae ( Supplementary Fig. S2e). Furthermore, the 16S rRNA sequences of B. diazoe ciens USDA 110 mapped to the most abundant OTU (OTU_77) and accounted for 67.85% and 69.70% of the nodule pro les inoculated with WT and the mutant strain, respectively ( Supplementary  Fig. S2f). These results show that the mutation in noeI has compartment-speci c effects on microbial communities.

NoeI affects niche differentiation in different rhizo-compartments
Enrichment analysis of OTUs using a generalized linear model con rmed differentiation of microbial communities across the rhizo-compartments. Compared to bulk soil, 49 bacterial OTUs mainly assigned to Proteobacteria (Alpha-, Delta-, Beta-, and Gamma-proteobacteria), and Firmicutes (Bacilli, Clostridia) were signi cantly enriched in the rhizosphere of soybean inoculated with WT strain (Fig. 3a). There was only one OTU (Bacilli) that was differentially enriched in the rhizoplane compared to the rhizosphere in the WT treatment. A total of 537 OTUs belonging to the phyla Proteobacteria, Bacteroidetes, Planctomycetes, Actinobacteria, Firmicutes and Chloro exi were also enriched in the endosphere compared to the rhizoplane. Overall, 171 OTUs mainly consisting of Proteobacteria, Firmicutes, Bacteroidetes, Actinobacteria and Chloro exi were enriched in nodules compared to the endosphere (Fig. 3a). The pattern of microbial community differentiation across the compartments in noeI mutantinoculated samples differed in the rhizosphere and rhizoplane (Fig. 3b). Speci cally, there were 148 OTUs enriched in the rhizosphere relative to bulk soil, which mainly belonged to Proteobacteria, Bacteroidetes and Actinobacteria. Compared to WT samples, the rhizoplane enriched a larger proportion of OTUs (27 vs 1) relative to the rhizosphere, which were mainly identi ed as members of Alphaproteobacteria, Betaproteobacteria and Clostridia. The microbial community differentiation between endosphere and nodule in mutant-inoculated samples was similar to WT-inoculated samples (Fig. 3b).
The STAMP method was performed to identify differences in taxonomic abundances between the WT and mutant treatments at the family level. Only Pseudomonadaceae were signi cantly enriched in the unplanted soil inoculated with the WT strain compared to that inoculated with the noeI mutant (Fig. 3c). A total of 11 families and 16 families were found to be signi cantly (p < 0.05) different between the inoculated soybean plants in the rhizosphere and the rhizoplane, respectively (Fig. 3c). Almost all of the differential families were enriched in samples inoculated with the WT strain, such as Micromonosporaceae, Streptomycetaceae, Clostridiaceae, Geobacteraceae, and Sphingomonadaceae. Strikingly, only one bacterial family in the endosphere samples showed a difference induced by the rhizobial treatments, with Bradyrhizobiaceae signi cantly enriched in the WT stain treatment (Fig. 3c). Finally, eight bacterial families were enriched in nodules inoculated with the WT strain; large differences were observed in Burkholderiaceae and Sphingobacteriaceae (Fig. 3c).

NoeI shapes microbial co-occurrence networks
To determine whether the noeI mutation affects co-abundance patterns between bacterial taxa across the different rhizo-compartments, we rst generated two full networks using bulk soil plus WT or bulk soil plus mutant samples using relative abundances of the 300 most abundant OTUs. We then constructed sub-networks for each rhizo-compartment from the corresponding full networks. In networks of WT-and noeI mutant-inoculated samples, the number of nodes and correlations in the sub-networks decreased from rhizoplane to nodule, with no differences between rhizosphere and rhizoplane (Fig. 4, Supplementary Table S3). Three dominant clusters were identi ed in all sub-networks. The rst cluster consisted of Bradyrhizobiaceae and Rhizobiaceae families; the second cluster contained taxa from Ktedonobacteraceae, and the third cluster contained families from Clostridiaceae_1. This third cluster exhibited a greater number of connections in the sub-network from samples inoculated with the WT than that in the noeI mutant treatment (Fig. 4). The topological features of sub-networks differed in both rhizocompartments and treatments (Supplementary Table S3). Speci cally, the average degree of subnetworks decreased from bulk soil to nodule; the sub-network in bulk soil had the lowest modularity, diameter and number of clusters. The average path length, betweenness centrality and modularity of subnetwork were highest in the nodule compared to the other rhizo-compartments (Supplementary Table S3).
In the rhizobial treatments, the average degree, connectivity and number of clusters of sub-networks were higher in samples inoculated with the WT rhizobium than those inoculated with the mutant, whereas the average path length and diameter were lower in the WT treatment (Supplementary Table S3). Thus, the noeI mutation alters network topologies features of microbial co-occurrence in different rhizocompartments.

Role of avonoid exudates in noeI-dependent effects
To determine whether NoeI may modulate microbial communities by changing avonoid exudation, we collected exudates from soybean plants following inoculation with WT and noeI-mutant of strain B. diazoe ciens USDA110 and analyzed them by UPLC-MS/MS. Eleven avonoids were identi ed and quanti ed in soybean root exudates (Fig. 5a). Compared to control roots, WT-inoculation increased the exudation of the six most abundant avonoids, including a 5-fold increase in daidzein (p < 0.05) (Fig. 5a). These increases were absent in exudates of plants inoculated with the noeI-mutant, whose avonoid exudation pro les were similar to control roots (Fig. 5a).
In a next step, we explored the relationships between avonoid exudation and rhizosphere bacterial community composition by weighted UniFrac distance-based redundancy analysis (db-RDA) (Fig. 5b). We also included several soil chemical factors into this analysis, including TC, TN, DOC and DON, pH, CEC.
Although the majority of these soil chemical factors were highly correlated with each other ( Supplementary Fig. S3a), daidzein was identi ed as the most signi cant factor that differentiated the rhizosphere microbial communities between the WT and mutant treatments. By contrast, soil exchangeable Mg 2+ explained differences in microbiomes between rhizospheres of control and inoculation treatments (p < 0.001) (Fig. 5b). To further assess the contribution of soil exchangeable Mg 2+ and daidzein to the diversity of microbial community in rhizospheres, variance partitioning analysis was applied; this metric indicated that soil exchangeable Mg 2+ and daidzein explained 28.6% and 4.80% of microbial community variation, respectively (Fig. 5c). These results indicate that avonoid exudation, and in particular daidzein, may be responsible for the differences in microbial community composition that are triggered by noeI.
To gain insight into the role of avonoids in determining microbiome structure, we performed an incubation experiment using the same soil and supplemented it with a mixture of avonoids, which contained 1 µg/g daidzein, and the other ten avonoids were added following their ratios to daidzein, twice a week. After four weeks of incubation, we determined changes in bacterial community structure using amplicon sequencing. Alpha diversity was lower in the soil treated with avonoids than that in the control treatment ( Supplementary Fig. S3b). PCoA using weighted Unifrac distances indicated a closer separation compared between soil treated with avonoids and rhizosphere inoculated with WT bacteria, with a peripheral distribution in the soil treated with water (control) and rhizosphere inoculated with mutant bacteria (Supplementary Fig. S3c). This was con rmed by PERMANOVA with 46.60% of variance (p < 0.001). STAMP analysis revealed that the families of Burkholderiaceae and Sphingobacteriaceae were signi cantly enriched in the soils treated with avonoids compared to the control (Fig. 5d).

Discussion
In this study, a genetic approach (gene mutation) is used to demonstrate that the capacity of B. diazoe ciens to form a symbiosis with soybean has major effects on root-associated microbiomes through plant-mediated effects. Here, we discuss these ndings in the context of root-microbiota interactions.
Our bacterial community sequencing approach con rmed a clear differentiation of bacterial community structure between unplanted soil, rhizosphere, rhizoplane, endosphere and nodule compartments, with a gradient of decreasing bacterial α-diversity from rhizosphere to endosphere and to nodules. This observation is consistent with previous studies in various plants of L. japonicas [40], soybean [70], pea [71], peanut [72] and rice [55]. Interestingly, disrupting symbiosis between B. diazoe ciens and soybean signi cantly reduced bacterial diversity in the rhizosphere and rhizoplane. This result is in line with a recent study on plant SYM mutants documenting a reduction of fungal diversity [41]. Together, these studies suggest that functional symbiosis favors a more diverse root microbiota. As it may change root exudate quality and thereby create opportunities for microbes to colonize the rhizosphere [51,73].
In contrast to the root-associated compartments, there was no effect of the noeI mutation on the diversity and composition of bacterial communities in the absence of soybean plants. This shows that the effect is plant-mediated. As a gene involved in nodulation, noeI is only expressed under induction of avonoids secreted from host plant [48]. Thus, the impact of the noeI mutation is indeed expected to be restricted to the interaction between the plant and B. diazoe ciens. Previous studies using SYM mutants in M. truncatula [44], L. japonicas [40] and soybean [42,43], found signi cant effects of these mutations on root-associated microbial communities' assemblages. Even in non-leguminous plants such as Oryza sativa, a mutated SYM pathway gene (CCaMK) has been found to structure distinctive root-associated microbiomes, as re ected by enrichment in Rhizobiales and Sphingomonadales [45], thus complicating the interpretation of these results in the context of legume symbiosis. Our work strengthens the notion that the successful establishment of legume symbiosis has substantial knock-on effects on native legume root-associated microbiota. It is likely that these changes will impact plant performance and soil legacy effects, thus in uencing plant productivity in nature and agriculture beyond the primary effect of the symbiosis. Understanding these consequences is an exciting prospect of this work.
Interestingly, the successful establishment of symbiosis resulted in an enrichment of OTUs that are associated with bene cial effects such as Micromonosporaceae and Streptomycetaceae. Previous studies revealed that Micromonosporaceae are widespread in nitrogen-xing nodules of different legume species and that these organisms enhance symbiosis e ciency when being co-inoculated with rhizobia [74][75][76][77][78][79]. The nifH-like gene sequences obtained from the nodule endophytic Micromonosporaceae strains were similar to nifH from Frankia, a nitrogen-xing actinobacterium that can develop symbiotic relationships with several woody dicotyledonous plants [75]. This similarity suggests that Micromonospora is capable of xing nitrogen [74,75]. Streptomycetaceae are reported to possess the ability to colonize the roots of Pisum sativum [80] and M. sativa [81] and they could also increase root nodulation e ciency and facilitated nutrient assimilation of their host plants. These ndings suggest that a functional symbiosis with effective nodulation and nitrogen xation in soybean may speci cally promotesthe enrichment of bene cial microbes.
In our study, two members of Clostridiales (Clostridiaceae_1, Clostridiales incertae sedis IV) were also enriched in the samples inoculated with the WT strain, which is consistent with previous experimentation where Clostridium was enhanced by rhizobial inoculation [82]. Bradyrhizobiaceae were depleted in roots of soybeans inoculated with the noeI mutant. This might be explained by a reduction in compatibility between the host plant and Bradyrhizobium caused by noeI mutation [50].
The bacterial families signi cantly enriched in nodules inoculated with WT strain and in soil supplemented with the avonoid mixture are presented in Fig. 3c and Fig. 5c. These families included Burkholderiaceae, which contained some species able to form symbiosis with a certain legumes from the Papilionoideae subfamily [83,84] and also some species dominate soybean nodule [85] or known as a plant growth-promoting strain in non-legume plants [86]. Our results are consistent with other studies that found also a depletion of Burkholderiales in the roots of Lotus symbiosis pathway gene mutants [40,41]. Therefore, we suggest that the effective symbiosis promotes the enrichment of bene cial microbes. In contrast, we found a signi cant depletion of Sphingobacteriaceae and Burkholderiaceae in the rhizosphere and rhizoplane of plants inoculated with the WT strain. This might be a consequence of potential niche replacement as a compensatory effect following the exclusion of Micromonosporaceae and Streptomycetaceae from these compartments.
Network analysis, an approach to visualize and examine microbial abundance patterns, con rmed in the different sub-networks the gradient of decreased diversity observed from soil to root and nodule compartments. This is also re ected in the topological features of the sub-networks. We noticed higher average degree, connectivity and number of clusters, and lower average path length and diameter for OTUs in the networks of the WT strain treatment compared to those for the noeI mutant treatment. This observation is possibly linked with the enhanced diversity seen in in root-associated compartments of the WT strain treatment. For instance, the higher average degree indicates that there are more potential bacterial connections in samples inoculated with the WT strain than those in the noeI mutant inoculated samples (Average degree measures the number of direct co-occurrence links for an each OTU in the network [87]). Our results are consistent with other work showing that rhizobia inoculation lead to an increase in soybean rhizobacterial network connections [88]. The increased modularity and number of clusters from bulk soil to nodule supports the conclusion that the nodule compartment is a highly selective niche [40]. Co-occurrence networks also identi ed several microbial clusters, which were composed of plant growth-promoting microbes such as Rhizobiaceae and Clostridiaceae_1 [89]. Speci cally, the formation of larger and stronger clusters by family of Clostridiaceae_1 in the subnetworks of samples inoculated with the WT strain than the mutant treatment might be the result of alpha diversity effects seen in the rhizosphere and rhizoplane compartments. Taken together, the network analysis suggested that a functional symbiosis enriches bene cial microbes and structures a more tighly connected bacterial network.
The avonoids daidzein, coumestrol and genistein have been found in exudates of most soybean cultivars [90,91]. Similarly, we found daidzein as the most abundant avonoid secreted by soybean variety C08, followed by coumestrol and genistein. This is consistent with a previous study on root exudates under similar conditions [92]. Previous studies reported that the amounts of secreted avonoids increased by inoculation with compatible symbionts or by treatment with Nod factors and were reduced by inoculation with nodC mutant rhizobium [35,36,93]. Accordingly, we found a signi cantly increased exudation of most avonoids when inoculating the WT strain. This has not been seen for the noeI mutant and is most likely due to the defective symbiosis.
Root exudates present a major organic carbon resource for soil microorganisms and drive the assembly of plant rhizosphere microbial communities. Speci c compounds in exudates are thought to promote or suppress speci c soil microbial members leading to the formation of distinctive root-associated microbiomes [92,94]. Soybean secretes the avonoids from the root surface to the surrounding rhizosphere. This is consistent with our results that the de cient symbiotic relationships, as mediated by the noeI mutation, affected the bacterial communities mainly in the rhizosphere and rhizoplane compartments. This is consistent with other studies that have revealed daidzein and genistein to shape soybean microbial communities [95,96]. We also found that the exogenous supplementation of avonoids affected soil microbiome diversity and signi cantly enriched bene cial microbes compared to the control-treated soil, which is consistent with a study of the relative abundance of B. diazoe ciens USDA110 increased in soybean treated with daidzein [92]. Redundancy analysis and variance partitioning analysis identi ed that soil exchangeable Magnesium (Mg 2+ ) and daidzein were signi cantly associated with a rhizosphere microbial shift. Mg 2+ plays an important role in the metabolism of rhizobia and the development of nodules, because nitrogen-xing requires ATP present as a Mg 2+ -complex [97]. Thus, wild-type rhizobial strains require more magnesium from the soil than symbiosis defective ones. Several studies have suggested that exchangeable magnesium has an impact on soil microbial communities [98,99].

Conclusions
In summary, our data point to the following model ( Fig. 6): noeI promotes functional symbiosis, which promotes the secretion of avonoids, which again shape the root-associated microbiome. We conclude that the symbiosis between legumes and rhizobia drives root microbiome assembly through plant-derived chemicals. The probable dual role of avonoids in the establishment of symbiosis and the structuring of microbial communities likely results in a direct link between legumes, rhizobia and root associated microbiomes. Understanding the consequences of this interplay for plant performance and the evolutionary dynamics of symbiosis are exciting prospects of this work.

Additional les
Additional le 1: Figure S1. α-diversity (Chao 1, observed OTUs and Fisher indices) and β-diversity among different rhizobial treatments in unplanted soil, rhizosphere, rhizoplane, endosphere and nodule. Figure  S2.  Table S1. Mass spectrometry parameters and ion patterns of tested compounds. Table S2. The effects of environmental variables on the microbial community assembly.