Coordination of microbe–host homeostasis by crosstalk with plant innate immunity

Plants grown in natural soil are colonized by phylogenetically structured communities of microbes known as the microbiota. Individual microbes can activate microbe-associated molecular pattern (MAMP)-triggered immunity (MTI), which limits pathogen proliferation but curtails plant growth, a phenomenon known as the growth–defence trade-off. Here, we report that, in monoassociations, 41% (62 out of 151) of taxonomically diverse root bacterial commensals suppress Arabidopsis thaliana root growth inhibition (RGI) triggered by immune-stimulating MAMPs or damage-associated molecular patterns. Amplicon sequencing of bacterial 16S rRNA genes reveals that immune activation alters the profile of synthetic communities (SynComs) comprising RGI-non-suppressive strains, whereas the presence of RGI-suppressive strains attenuates this effect. Root colonization by SynComs with different complexities and RGI-suppressive activities alters the expression of 174 core host genes, with functions related to root development and nutrient transport. Furthermore, RGI-suppressive SynComs specifically downregulate a subset of immune-related genes. Precolonization of plants with RGI-suppressive SynComs, or mutation of one commensal-downregulated transcription factor, MYB15, renders the plants more susceptible to opportunistic Pseudomonas pathogens. Our results suggest that RGI-non-suppressive and RGI-suppressive root commensals modulate host susceptibility to pathogens by either eliciting or dampening MTI responses, respectively. This interplay buffers the plant immune system against pathogen perturbation and defence-associated growth inhibition, ultimately leading to commensal–host homeostasis.

to interfere with RGI. RGI-suppressive activity was detected across all four phyla of the microbiota-Actinobacteria, Proteobacteria, Bacteroidetes and Firmicutes-but was overrepresented among Actinobacteria and Gammaproteobacteria commensals (Fig. 1a). Viable plate counting confirmed that the RGI-non-suppressive strains still colonize roots in mono-associations (Extended Data Fig. 1b). By contrast, only three strains, Streptomyces strains 107 and 187 and Pseudomonas 401, had detrimental impacts on Arabidopsis in monoassociations; Pseudomonas 401 most severely compromised plant growth (Extended Data Fig. 1c).
To examine whether root-derived bacteria were also able to suppress RGI elicited by an endogenous plant-derived danger-associated molecular pattern (DAMP), we treated plants with the DAMP Atpep1, which induces RGI and immune responses 27 . Using Atpep1-treated Col-0 wild type (WT) plants, we found that 12 out of 13 suppressive strains, representing members from diverse taxa, retained the ability to interfere with RGI, whereas none of the eight non-suppressive strains elicited this effect (Extended Data Fig. 2a). Thus, phylogenetically diverse root commensals can suppress both DAMP-and MAMP-induced RGI. One isolate, Caulobacter strain 342, suppressed flg22-but not Atpep1-mediated RGI (Extended Data Fig. 2b), suggesting the existence of at least two modes of RGI suppression: one interfering with both MAMP-and DAMP-induced RGI, and the other possibly specific to flg22 perception.
Although germ-free pWER::FLS2-GFP 25 plants respond to flg22 treatment with enhanced RGI compared with Col-0 on synthetic medium, no growth differences were noted between these two genotypes when grown on natural soil (Extended Data Fig. 2c). Given that root growth in natural soil probably proceeds in the face of chronic exposure to MAMPs and DAMPs, as well as colonization by both suppressive and non-suppressive commensals, we speculated that the aforementioned RGI suppression phenotype may act as a dominant community trait. To test this hypothesis, we composed four independent but taxonomically similar five-member SynComs with contrasting abilities for RGI suppression, that is, non-suppressive SynComs (SynCom NS1 and NS2) and suppressive SynComs (SynCom S1 and S2; Supplementary Table 1). We observed RGI-suppressive activity in plants inoculated with the suppressive SynComs, but not in plants inoculated with the non-suppressive SynComs. Furthermore, full RGI-suppressive activity was retained when these commensals were combined as ten-member SynComs (Fig. 1b). A recent study showed that auxin-mediated RGI could be rescued by Variovorax commensals 28 . However, our four tested SynComs did not induce RGI to a level comparable to flg22  Table 1), which differed in RGI-suppressive activity, on flg22-mediated RGI. Plants (aged 2 weeks) were germinated on plates supplemented with 1 μM flg22 and the indicated SynCom. Shapes represent three (SynCom NS1 + S1) and four (SynCom NS2 + S2) independent replicates in b. The n values indicate the total number of biological samples. Different letters indicate statistical significance, determined using two-sided Dunn's Kruskal-Wallis test (P < 0.05). The box plots centre on the median and extend to the 25th and 75th percentiles, and the whiskers extend to the furthest point within 1.5× the interquartile range. treatment, and the presence of Variovorax 434 in SynCom NS1 did not rescue the flg22-mediated RGI phenotype (Fig. 1b). Thus, we conclude that RGI is mainly caused by flg22 treatment, and is widely suppressed by At-RSPHERE members that function dominantly in our set-up.
We speculated that the co-occurrence of RGI-non-suppressive and suppressive strains might reflect a need for commensal microbes to dampen plant immunity to balance root growth and defence trade-offs. We therefore examined whether a single suppressive strain is sufficient to achieve full RGI suppression. We found that the addition of diverse individual suppressive strains to a five-member non-suppressive SynCom resulted in only partial RGI suppression (Extended Data Fig. 2d). This result suggests that the identity of suppressive commensals, as well as the input proportion of suppressive to non-suppressive strains, affect RGI-suppression capacity quantitatively.
The ability of specific strains to differentially suppress Atpep1and flg22-mediated RGI prompted us to investigate the mechanisms that underlie this biological process. Previously, commensal Pseudomonas spp. in monoassociations were shown to acidify the growth medium, rendering plants insensitive to flg22 (ref. 22 ). To determine whether acidification is responsible for RGI suppression by our SynCom, we measured the growth medium pH of plants that were co-inoculated with different SynComs. We observed average reductions in pH, ranging from pH 5.18 in mock treated plants to pH 4.62 and pH 3.97 in the presence of a SynCom S1 and NS1, respectively. This lack of correlation between RGI suppression and growth medium acidification suggests that this mechanism is unlikely to explain suppression in our community set-up. We next investigated whether type-III secretion, which is a well-characterized virulence mechanism among Gram-negative bacteria pathogens, is required for the suppressive activity of root commensals tested. Interestingly, hrcC-a gene that is essential for a functional type-III secretion system in pathogenic Pseudomonas-is dispensable for RGI suppression mediated by suppressive Pseudomonas strain 569 (Extended Data Fig. 2e,f). We next investigated whether root commensals can target the step upstream of flg22 perception. Only the culture filtrate of Janibacter 101, an Actinobacteria member, but not that of three other suppressive strains, derepressed both flg22 and Atpep1-induced RGI (Extended Data Fig. 3a-c). Heat treatment and filtration of the culture filtrate showed that the molecule(s) responsible for the suppressive activity retained in the supernatant of Janibacter 101 is heat-labile, and is larger than 3 kDa (Extended Data Fig. 3d). Mass spectrometry analysis further revealed that the filtrates of Janibacter 101 and the closely related suppressive Janibacter 563, but not three other tested suppressive commensals, elicited a significant reduction in intact flg22 peptide (Extended Data Fig. 3e). Thus, the ability of these two Janibacter strains to suppress MTI resembles the activity of pathogenic bacteria 29 , and is associated with an extracellular mechanism that can modify/ degrade flg22 peptide. Together, these data reveal that commensals use diverse mechanisms to suppress elicitor-mediated RGI.

Activation of immunity shapes root microbiota establishment.
To determine whether plant immunity affects microbiota establishment, we performed reconstitution experiments with gnotobiotic plants grown on an agar matrix. We designed three taxonomically similar SynComs with contrasting RGI suppression abilities for community profiling experiments at a strain-specific resolution (a total of six SynComs; the SynComs used in experiment 1 and 2 differ in two Gammaproteobacteria, and the SynComs used in experiment 3 are composed of entirely different strains; Supplementary Table 1). Principal coordinate analyses (PCoA) of Bray-Curtis dissimilarities revealed that root-associated bacterial communities were distinct from the corresponding unplanted or planted matrix samples (Extended Data Fig. 4), regardless of the SynCom composition and plant genotypes (Col-0 and pWER::FLS2-GFP). Constrained PCoA revealed that flg22 treatment elicited a consistent community shift in plants inoculated with non-suppressive SynComs, while samples from those inoculated with suppressive SynComs remained together. Consistent with a dominant effect of RGI suppression, roots inoculated with ten-member mixed communities (suppressive plus non-suppressive SynComs) were not affected by flg22 treatment (Fig. 2a-c and Extended Data Fig. 5a-c). This is consistent with another report showing that roots of fls2 mutant and Col-0 plants have similar community profiles, consisting of a mixed 32-member SynCom 30 .
To dissect the contribution of individual strains to the overall community shift, we quantified the relative abundance (RA) of individual strains. The detection of non-suppressive commensals as the most abundant strains in the mixed SynComs suggests that the ability to dominate in a community is not necessarily coupled to RGI suppression (Fig. 2d-f and Extended Data Fig. 5d-f). However, the RA of specific strains in a community was impacted by plant immunity. For example, flg22 treatment led to an altered RA of Pseudomonas 9/68 (increased) and Variovorax 434 (decreased), while Microbacteriaceae 60 was unaltered (experiments 1 and 2). Similarly, flg22 treatment altered the RA of Microbacteriaceae 1293 (increased) and Comamonadaceae 568 (decreased), while Pseudomonas 71 was unaffected (experiment 3; Fig. 2d-f). A similar trend was also detected in Col-0 (Extended Data Fig. 5d-f), although the effect was more pronounced in pWER::FLS2-GFP plants, possibly due to enhanced MTI and/or altered root architecture. Furthermore, we found that flg22 treatment reduced within-sample diversity of non-suppressive SynComs (experiments 1 and 2; Extended Data Fig. 5g), suggesting that immune activation can affect the distribution of specific strains in community contexts.
Root transcriptomic changes and dampening of immunity by suppressive SynComs. Although flg22-mediated RGI is closely associated with immune activation, its role as a bona fide immune output is unclear. Here, we sought to explore how inoculation with suppressive or non-suppressive SynComs affected the root transcriptome of plants treated with flg22 and grown on an agar matrix (Supplementary Tables 2-7). Principal component analyses (PCA) at the transcriptome level revealed distinct expression patterns between Col-0 plants inoculated with live bacteria, compared with germ-free plants (PC1, 20% variance; Fig. 3a). Interestingly, the transcriptional output of roots inoculated with these two taxonomically similar SynComs was clearly distinguishable after two weeks of co-cultivation, even in the absence of flg22 treatment (Fig. 3a, triangles). Furthermore, we observed a separation according to the immune status of the plants, triggered by flg22 exposure, in all of the samples treated with heat-killed bacteria as well as with the non-suppressive SynCom (PC2, 7% variance; Fig. 3a). By contrast, flg22 treatment of plants colonized by the suppressive SynCom did not elicit significant changes. Independent transcriptome experiments using pWER::FLS2-GFP plants confirmed these results (Extended Data Fig. 6 and Supplementary Tables 2-4).
Next, we performed k-means clustering of differentially expressed genes (DEGs) involved in the flg22 response, the SynCom response or both ( Fig. 3b and Supplementary Tables 5-7). We observed three large clusters (2,221 DEGs) that were induced (c4 and c5) or suppressed (c8) by live bacteria, independent of flg22 treatment (Fig. 3b,c). Gene Ontology (GO) enrichment analyses showed that the SynCom-responsive clusters were primarily enriched in functions related to detoxification, root development, nutrient transport and response to hypoxia (Extended Data Fig. 7). To determine whether similar GO terms could also be identified in experiments with more complex SynComs, we compared our data with two independent Arabidopsis root transcriptome studies that used SynComs consisting of both suppressive and non-suppressive commensals (35 members 31 and 115 members 13 ). Despite differences in technical set-ups and SynCom complexities, we identified 174 common SynCom-responsive DEGs in the absence of flg22 that were related to the same biological functions mentioned above (Extended Data Fig. 8 and Supplementary Tables 8-12).
Importantly, we found a flg22-inducible cluster (c3), which was significantly upregulated by the non-suppressive SynCom, but downregulated by the suppressive community (Fig. 3b,c), in a pattern matching the RGI phenotype of the plant (Fig. 1b) and the bacterial community shifts (Fig. 2). As expected, a portion of   defence-related genes were enriched in c3, such as PER5, FRK1 and RBOHD (70 genes; Fig. 3b). However, additional defence-related DEGs were found outside c3 and were upregulated by flg22 treatment, even in the presence of the suppressive SynCom (348 genes; Fig. 3b  systemic acquired resistance, for example, FMO1 (c5) and SARD1 (c1) 34,35 ; and endogenous peptides amplifying MTI, for example, PIP1 and PIP2 (ref. 36 ) (c5; Fig. 3b). Recent research showed that MAMP responsiveness in germ-free roots was gated by the expression of damage-induced PRRs 37 . However, the sustained expression of FLS2 (c1) in the presence of SynComs indicates that RGI suppression is not due to FLS2 downregulation (Fig. 3b). The ability of diverse root commensals to suppress Atpep1-mediated RGI (Extended Data Fig. 2a) also highlights the interference from FLS2-independent pathway(s). An independent study by Teixeira et al. also identified a cluster of DEGs that was highly induced in axenic Arabidopsis by treatment with flg22, but suppressed by the presence of a 35-member SynCom consisting of suppressive and non-suppressive root commensals 31 (Extended Data Fig. 9). Remarkably, this cluster showed the largest overlap with our cluster c3, with 58 common DEGs (at least 21 were defence-related) that were downregulated by both SynComs (Extended Data Fig. 9). Even though we have shown that two Janibacter strains can degrade/modify flg22 extracellularly, the downregulation of only a subset of flg22-mediated responses suggests that the direct removal of the flg22 peptide is insufficient to account for the differential suppressive activities observed.
We further validated our findings by examining the expression of two flg22-inducible defence marker genes 12,25,38 in roots of Arabidopsis using quantitative PCR (qPCR) in the presence of other suppressive SynComs. PER5 and FRK1 remained significantly elevated two weeks after co-inoculation with flg22 and a non-suppressive SynCom, but not with a suppressive SynCom (Extended Data Fig. 10a). A non-suppressive SynCom alone also significantly induced the expression of PER5 and FRK1, indicating that non-suppressive commensals stimulate specific root immune responses. As expected, a ten-member mixed SynCom, which was shown to suppress flg22-mediated RGI (Fig. 1b), did not significantly induce the expression of PER5 and FRK1 (Extended Data Fig.  10b). We next examined whether the suppressive SynCom exerted an effect on the root defence response at earlier time points after flg22 stimulation. Intriguingly, we observed a significant induction of PER5, FRK1 and MYB15 1 h after flg22 treatment. However, suppressive SynCom, in contrast to non-suppressive SynCom, specifically downregulated the expression of these three genes after 24 h (Extended Data Fig. 10c), suggesting that SynComs can modulate defence responses as early as 1 d after stimulation.
To determine whether MTI has a direct impact on commensal proliferation independent of any microbe-microbe interactions, we focused on transcription factors (TFs) and investigated the contributions of the top three candidates identified in our dataset-WRKY30, MYB15 and WRKY28 (cluster c3; Extended Data Fig. 10d,e). Null mutants of WRKY30 and WRKY28 have not been reported, and our attempts to knock out these TFs using CRISPR failed, suggesting that these genes are essential for plant viability 39,40 . We therefore focused on MYB15, a positive regulator of defence against the foliar pathogen PtoDC3000 (ref. 41 ). In myb15-1 plants, elimination of this single TF led to a significantly enhanced proliferation of the detrimental strain Pseudomonas 401, and the commensal Variovorax 434 (P < 0.05; Extended Data Fig. 10f,g), which also showed a reduced RA after flg22 treatment in community contexts (Fig. 2d,e). Together, amplicon sequencing and transcriptome data support the idea that colonization of specific root commensals is affected by host MTI, which can be attenuated by suppressive strains.
Next, we tested whether our suppressive SynCom can also suppress defence responses triggered by a non-proteinaceous elicitor, chitin, which is commonly found as a MAMP in the cell wall of eukaryotic fungi. We performed a time-resolved experiment to follow the expression of the defence marker genes FRK1, PER5 and chitinase (AT2G43620) 1 h, 6 h and 24 h after elicitor application. In contrast to flg22, chitin treatment only marginally induced FRK1 expression after 1 h, whereas PER5 and AT2G43620 were significantly induced up to 6 h (Extended Data Fig. 10h). This is consistent with a previous report showing that flg22 and chitin induce both overlapping and specific root responses 14 . No stimulation of marker gene expression was detected after 24 h chitin application. Interestingly, our SynComs exerted a cooperative effect on chitin-mediated signalling. For example, chitin-induced PER5 expression was further stimulated by a suppressive SynCom after 1 h but this stimulation was reversed such that the non-suppressive SynCom enhanced PER5 expression after 6 h. After 24 h, both suppressive and non-suppressive SynComs induced PER5 expression, while the chitin-triggered response was no longer detectable (Extended Data Fig. 10h), suggesting that our SynComs interact differently with flg22-and chitin-triggered responses.
Suppressive and non-suppressive commensals differentially impact plant susceptibility to opportunistic pathogens. As a subset of commensals dampens root immune responses, we hypothesized that colonization with a suppressive SynCom might render plants more susceptible to opportunistic pathogens. We identified three detrimental strains from At-RSPHERE. In particular, Arabidopsis plants inoculated with Pseudomonas 401 exhibited reduced growth and accumulated pigments in shoots reminiscent of stress-inducible anthocyanins (Extended Data Fig. 1c), which indicates its pathogenic potential in a laboratory environment. Consistent with the fact that Pseudomonas 401 was originally isolated from healthy and asymptomatic Arabidopsis roots colonized by a diverse microbial community, the detrimental effect was attenuated when plants were colonized by our SynComs. Interestingly, the attenuation was stronger when plants were co-colonized with the non-suppressive SynCom, compared with the suppressive SynCom (Fig. 4a).
Recent reports suggest that there is a positive correlation between disease progression in natural Arabidopsis populations and bacterial biomass 42,43 . To determine whether the virulence of Pseudomonas 401 is related to enhanced plant colonization, we quantified its absolute abundance on pWER::FLS2-GFP precolonized with suppressive or non-suppressive SynComs. Plants already colonized by suppressive SynComs harboured significantly higher Pseudomonas 401 titres compared with plants that were precolonized with non-suppressive SynComs (Fig. 4b,c). Interestingly, this SynCom-dependent difference seemed to be limited to roots, as Pseudomonas 401 growth in shoots was similarly restricted by co-colonization with either community (Fig. 4b,c). Even though we are not excluding an impact of microbe-microbe interactions through niche competition, none of the individual strains antagonized Pseudomonas 401 in vitro (Extended Data Fig. 10i). These data further suggest that the underlying growth differences are probably not the result of antibiosis.
To determine whether SynComs modulate plant susceptibility to a characterized opportunistic pathogen prevalent in natural A. thaliana populations, and exclude the possibility that differential impact of root commensals in roots and shoots is a result due to niche specialization of Pseudomonas 401 on roots only, we inoculated plants with the opportunistic Pseudomonas leaf pathogen OTU5 (isolate p5.e6) (Extended Data Fig. 1c). Plants colonized by suppressive SynComs supported higher growth of Pseudomonas OTU5 compared with plants colonized by non-suppressive SynComs, and this SynCom-specific effect was again observed only in roots and not in shoots (Fig. 4b,c). Together with the RNA-sequencing (RNA-seq) and targeted PCR data, these results suggest that precolonization with non-suppressive SynComs activated root immunity and this correlates with reduced growth of the tested opportunistic pathogens, whereas suppressive SynComs failed to provide pathogen protection.

Discussion
In nature, a subset of soil-dwelling bacteria colonizes roots seemingly without influencing host traits, and are therefore often considered to be commensals. Here, using a bottom-up approach, we show that phylogenetically diverse commensals, representing the core of the Arabidopsis root microbiota 20 , share the ability to suppress host defence responses, a microbial trait that is dominant in our community set-up and is therefore easily overlooked in nature. The functional redundancy of members of the root microbiota to interfere with the host's immune response is consistent with our observation that mixed communities consisting of non-suppressive and suppressive strains resist flg22-mediated community changes otherwise detected in non-suppressive SynCom-colonized plants (Fig. 2). This finding might explain why immune-related Arabidopsis mutants, tested in natural soil, show relatively mild changes in root microbiota composition 10 . Here we showed that the reduction in apoplastic pH and potential type-III secretion system-dependent mechanisms are insufficient to fully account for our MTI-suppression phenotype. Two closely related Janibacter strains, but not three other tested suppressive commensals, were shown to degrade/modify flg22. Even though we cannot rule out the possible involvement of specific plant metabolites produced after interactions with the suppressive bacteria, our findings suggest that root commensals can use multiple mechanisms to target host immune responses, rendering this community trait difficult to overcome by the plant host.
To date, information on Arabidopsis root transcriptomic changes evoked by commensals are limited to monoassociations 14,21 , leaving a gap in our understanding of how plant roots respond to commensal communities that can reach a steady-state as early as 13 d after inoculation 44 . We unexpectedly found that, after 2 weeks, root colonization by taxonomically similar commensal SynComs, differing in their ability to suppress RGI, elicited 2,221 DEGs (clusters c4, c5, c8) with remarkable overall similarity. These changes, which are associated with presumably steady-state SynComs, contrast with the subtle response to heat-killed SynComs or flg22 treatment alone, reflecting an impact of active commensal colonization on host transcriptional outputs beyond plant responses to chronic exposure to MAMPs. Furthermore, we observed robust enrichment of specific GO terms related to root development, nutrient transport, response to hypoxia and detoxification across experimental set-ups and SynCom complexities 13,31 ). Indeed, rhizobacteria alone are known to modulate root traits 21,45,46 . In return, root-secreted photoassimilates feed up to 20% of root-associated bacteria 47 . As photoassimilates have been shown to serve as sources of organic carbon that limit bacterial growth 48 , we speculate that the enrichment of these GOs is associated with altered nutrient flux, and reduced oxygen due to microbial respiration in roots. Although our SynComs are taxonomically diverse with predicted varied metabolic repertoires 20,49 , convergence to core transcriptomic outputs indicate integrated responses to a state of 'community commensalism' .
The zigzag model of the plant immune system proposes that effective resistance is the result of quantitative outputs above a certain threshold following MAMP perception 19 . Colonization by suppressive SynComs led to the downregulation of a subset of flg22-induced genes (Fig. 3, cluster c3), whereas colonization by non-suppressive SynComs alone stimulated these genes and further upregulated their expression together with flg22. Thus, the responsiveness of these defence-associated genes to SynCom colonization differs greatly with respect to the ability of the bacterial community to suppress RGI. However, roots in nature are co-colonized by both groups of commensals, and our experiments indicate a quantitative output that is dependent on their ratio. Intriguingly, recent studies reported that 42% (ref. 22 ) and 28% (ref. 31 ) of commensals from two other Arabidopsis root-derived culture collections quench early and late flg22-induced responses in monoassociations, respectively 22 . Together with our study, this confirms the potential of the root microbiota to modulate plant growth-defence traits.
We hypothesize that the imbalance between non-suppressive and suppressive commensals might reduce plant fitness under stress conditions. Indeed, plants that are precolonized by suppressive SynComs are as susceptible as germ-free plants to opportunistic Pseudomonas pathogens, whereas plants associated with non-suppressive SynComs are more resistant, but prone to MAMP-induced RGI. The observed defence-associated community shifts and potentially reduced alpha-diversity might hinder the provision of microbiota-derived beneficial services 50 , or exert a detrimental impact on the host under dysbiosis 51 . Thus, we propose a rheostat model (Fig. 4d) in which a balance between commensals with contrasting MTI-modulating activities constitutes an integral feature of the holobiont to buffer plant resistance to pathogen perturbation and defence-associated growth reduction.
It remains unclear which factors govern the state of equilibrium and the corresponding ratio between non-suppressive and suppressive strains. However, on the basis of the results obtained using three pairs of SynComs (Fig. 2), the initial input composition has a key role in defining the equilibrium state. The timing of colonization 44 and abiotic factors 52 probably also exert an influence. As a consequence, these complex interactions enable community coexistence, and ultimately establish microbe-host homeostasis. Accordingly, their ratio will impact the amplitude and/or might set the threshold for effective resistance in the zigzag model. Plants in nature are influenced by fluctuating stresses and are colonized by more diverse microbial communities that modulate plant physiology through multiple mechanisms, including the modulation of phytohormone signalling 28,53 . As our SynComs are constructed on the basis of their ability to suppress flg22-mediated responses, we found that they exert a synergistic effect on chitin-mediated responses. Selective modulation of chitin and flg22-mediated responses-for example, by the cytoplasmic receptor-like kinase PBL27-has been reported 54 . The characterization of the molecular mechanisms that underlie immunosuppression by root commensals may identify immunity components that are specific to one or integrate multiple upstream signalling pathways. Another future task will be to test whether the rheostat model also applies to communities with different traits to alleviate abiotic stresses.

Methods
Arabidopsis thaliana ecotype Columbia (Col-0, CS60000) and pepr1pepr2 were laboratory stocks. myb15 (ref. 41  Growth conditions for plants. Arabidopsis seeds were surface-sterilized in 70% ethanol twice for 5 min each followed by a brief wash with 100% ethanol. Seeds were then washed three times with sterile water. Cold-stratified seeds were sowed on agar plates (1%, Difco Agar Granulated, BD Biosciences, discontinued) or Bacto agar (BD Biosciences) prepared with half-strength Murashige and Skoog (MS) medium (Duchefa) and 0.1 g l −1 2-(N-morpholino)ethanesulfonic acid (pH 5.7). Sugars were not provided as an additional carbon source unless otherwise specified. Plants were grown under short-day conditions (10 h light-14 h dark) under a 21 °C-19 °C cycle, 65% relative humidity and a light intensity of 120 mE m −2 s −1 . For experiments involving myb15-1, surface-sterilized seeds were sowed on half-strength MS agar plates supplemented with 5 g l −1 sucrose.
Culture conditions for bacteria. Information on individual strains used can be found at At-RSPHERE (http://www.at-sphere.com/) 20 . OTU5 (p5.e6) 42 was provided by D. Weigel (Max Planck Institute for Developmental Biology, Tübingen, Germany). Bacterial strains were prepared by taking an aliquot from the glycerol stock, followed by incubation on 50% tryptic soy broth (TSB) agar plates (Sigma-Aldrich) at 25 °C for 1-4 d. Before the start of the experiments, strains were cultured in 50% TSB medium to saturation, and subcultured to log phase with fresh medium at a 1:5 ratio. Bacterial cultures were pelleted by centrifugation at 8,000g for 5 min, followed by two washes with 10 mM MgSO 4 .
Screening for RGI-suppressive strains in monoassociation. After washing, bacteria were diluted with 10 mM MgSO 4 to an optical density at 600 nm (OD 600 ) concentration of about 0.1. A total of 150 µl bacterial suspension was added to still warm 50 ml half-strength MS agar medium at a final bacterial concentration of OD 600 = 0.0005. A final concentration of 1 µM flg22 was added accordingly. Plates were dried for 2 h before approximately 15 surface-sterilized pWER::FLS2-GFP seeds were sowed on each plate. The expression of the flg22 receptor FLS2 in pWER::FLS2-GFP is limited to the root epidermis such that potential interorgan shoot-to-root signal after flg22 perception is minimized. The plates were sealed with 3M tape and transferred to the phytochamber for incubation. One week after germination, plants with delayed germination were removed and the plates were trimmed to about ten remaining plants. Pictures were taken 3 weeks after incubation and the primary root lengths were quantified using ImageJ. Shoots were separated from the roots and the fresh shoot weight of individual plants was taken. For experiments using 1 µM Atpep1, wild-type Col-0 plants were used instead.
A phylogenetic tree of selected strains from At-RSPHERE was performed previously 20 and visualized using iTOL 55 . Strains leading to a rescue of RGI with a root length longer than 2 cm (average root length of germ-free flg22-treated pWER::FLS2-GFP plants = 1.53 cm; n = 37) after coinoculation with 1 µM flg22, and exhibiting consistent suppressive activity across two biological replicates, were considered to be suppressive. Suppressive strains are indicated by a red triangle in Fig. 1a. For the inoculation of SynCom, each bacterium was inoculated to a final concentration of OD 600 = 0.0005, that is, for a five-member SynCom, the total bacteria added was OD 600 = 0.0025. The five-member SynCom is composed of Actinobacteria, Alpha-, Beta-and Gammaproteobacteria. Bacteroidetes and Firmicutes were not included in these SynComs as no strains with differential ability to suppress RGI were identified in these two phyla. The composition of SynComs used in this manuscript can be found in Supplementary Table 1. 16S amplicon sequencing and community profiling. For 16S community profiling, root samples were harvested and libraries were processed according to previously a published protocol 4 . In brief, plants were germinated with the indicated SynCom in the presence or absence of 1 µM flg22, and incubated for 14 d before harvesting. Plants were inoculated with SynCom NS1 and S1 for experiment 1 (Fig. 2a,d and Extended Data Fig. 5a,d); SynCom NS3 and S3 for experiment 2 (Fig. 2b,e and Extended Data Fig. 5b,e); and SynCom NS4 and S2 for experiment 3 (Fig. 2c,f and Extended Data Fig. 5c,f). Plant roots were separated from the shoots and pooled from three plates from each biological replicate. Roots were washed briefly with sterile water and blotted dry before being transferred into Lysing Matrix E tubes (MP Biomedicals) at −80 °C until processing. Samples were homogenized using a Precellys 24 homogenizer (6,200 r.p.m. twice for 30 s with 15 s pauses in between; Bertin Technologies). Total root and bacteria DNA was extracted using the FastDNA SPIN Kit for Soil (MP Biomedicals) according to the manufacturer's instructions, eluted in 80 μl elution buffer and quantified using the Quant-iT PicoGreen dsDNA Assay (Thermo Fisher Scientific). Samples were diluted to 3.5 ng μl −1 , and 3 µl samples were used in a three-step PCR amplification protocol as follows.
Step 1: the V5V7 region of the bacterial 16S rRNA gene was amplified in triplicate reactions using the primers 799F and 1192R in a 25 µl reaction volume containing 2 U DFS-Taq DNA polymerase (Bioron), 1× incomplete buffer, 2 mM MgCl 2 , 0.3% bovine serum albumin, 0.2 mM dNTPs (Life Technologies) and 0.3 μM forward and reverse primers. The same PCR parameters were used for each primer pair (initial denaturation at 94 °C for 2 min, denaturation at 94 °C for 30 s, annealing at 55 °C for 30 s, extension at 72 °C for 30 s, repeat steps 2-4 for 25 cycles, and final extension at 72 °C for 10 min). Primers and proteins were digested by adding 1 μl of Antarctic phosphatase, 1 μl exonuclease I and 2.44 μl Antarctic phosphatase buffer (New England Biolabs) to 20 μl of the pooled replicate reactions at 37 °C for 30 min, followed by enzyme deactivation at 85 °C for 15 min. Reactions were centrifuged for 15 min at 4,000 r.p.m. and 3 μl of supernatant was used for the second PCR step in triplicate reactions.
Step 2: PCR reactions were performed as stated above with the number of cycles reduced to ten using primer pairs 799F and individual reverse barcoded primers. PCR quality and quantity were estimated by loading 5 μl of each reaction on a 1.5% agarose gel. Approximately similar amounts of DNA sample from the same biological replicate were pooled, and the mixtures were loaded onto a 1.5% agarose gel. DNA bands with the correct size were cut out and purified using the QIAquick Gel Extraction Kit (QIAGEN).
Step 3: Gel-purified DNA was used as a template for the third PCR using forward barcoded primers and p7_pad_R with a total of ten cycles. PCR reactions were loaded onto a 1.5% agarose gel and DNA bands with the correct size were cut out and purified using the QIAquick Gel Extraction Kit (QIAGEN). Double-barcoded DNA was purified and concentrated using Agencourt AMPure XP beads. The concentration of the purified DNA was determined using Quant-iT PicoGreen dsDNA Assay (Thermo Fisher Scientific). Paired-end Illumina sequencing was performed using 20 ng µl −1 of the final library in-house, using the MiSeq sequencer and custom sequencing primers.
Amplicon data analysis. Forward and reverse sequencing reads were denoised and demultiplexed separately according to the barcode sequence using QIIME 56 using the following parameters: phred = 30; bc_err = 2. After quality-filtering and merging the paired-end reads, amplicon tags were next aligned to a reference set of sequences obtained from the whole-genome assemblies of every strain included in each experiment using USEARCH (uparse_ref command) 57 . A count feature table for each strain was generated using only perfect matches to the reference sequence from the genome collection. This count table was used for subsequent diversity and enrichment analyses. Alpha and beta diversities were calculated after normalizing count tables to the total number of reads per sample. The Simpson index was obtained using the diversity function in the vegan package. The Bray-Curtis dissimilarity index was calculated using the vegdist function in the vegan package 58 and used for unconstrained ordination by PCoA. All data were used except for biological replicate c of experiment 1 due to a potential contamination issue or PCR error. Constrained PCoA was performed using the vegan capscale function on the Bray-Curtis dissimilarity matrices, constraining by the interaction between flg22 treatment and SynCom variables and conditioning by technical parameters. Statistical significance of separation between community profiles according to flg22 treatment was determined using PERMANOVA with 999 permutations (anova.cca function in vegan). Finally, all amplicon data were visualized using the ggplot2 (ref. 59  Generation of the ΔhrcC mutant of Pseudomonas strain 569. The Pseudomonas 569 deletion mutant was generated using a homologous recombination protocol 60 . In brief, PCR fragments flanking the upstream and downstream region of the hrcC gene were PCR-amplified and cloned into pK18mobsacB. The FRT-flanked cassette from pCPP5209 was inserted between the two hrcC fragments. The resultant pK18mobsacB::ΔhrcC construct was electroporated into log-phase-grown Pseudomonas 569, which was prewashed and resuspended in 0.3 M sucrose, at 2.5 kV and 150 Ω using the Biorad electroporator. The transformant was selected on half of a TSB plate supplemented with 25 ng µl −1 gentamycin. The double-crossover deletion mutant was further confirmed by colony PCR and Sanger sequencing. Mass spectrometry. For in vitro detection of flg22, 1 µM flg22 was co-incubated with 1 ml supernatant for 1 h at room temperature. Half-strength MS medium without sugar was used as a control. Sample aliquots (100 µl) were mixed with 200 µl UA (8 M urea in 100 mM Tris-HCl pH 8.5) and adjusted to 10 mM dithiothreitol using 1 M stock. Samples were loaded onto 30 kDa spin filters (Vivacon 500, Sartorius) and centrifuged at 14,000g for 15 min. The filtrate was collected and loaded onto 2 kDa spin filters (Vivacon 500, Sartorius) and centrifuged at 14,000g for 30 min, after which 300 µl UA was added and the samples were centrifuged again (14,000g for 45 min, or until most liquid had passed through the filter). Next, 100 µl 55 mM chloroacetamide was added to the filter and samples were incubated for 30 min in the dark, after which they were centrifuged at 14,000g for 20 min. UA (300 µl) was added and the samples were centrifuged at 14,000g for 45 min. The samples were washed twice with 300 µl 100 mM Tris-HCl, pH 8.5, by centrifugation (14,000g for 45 min). For elution, 200 µl Tris-HCl was added, and the inverted spin filters were centrifuged at 2,000g for 2 min to collect eluate into a fresh tube. The eluates were desalted using StageTips with C18 Empore disk membranes (3M) 61 , and a final elution was performed using 40% acetonitrile and 0.1% trifluoroacetic acid. The samples were dried in a vacuum evaporator, and dissolved in 10 µl 2% acetonitrile (ACN) and 0.1% trifluoroacetic acid for analysis.
Liquid chromatography coupled with tandem mass spectrometry data acquisition and data analysis. Samples were analysed using the EASY-nLC 1000 system (Thermo Fisher Scientific) coupled to a QExactive mass spectrometer (Thermo Fisher Scientific). Peptides were separated on 16 cm frit-less silica emitters (New Objective, 0.75 µm inner diameter), packed in-house with reversed-phase ReproSil-Pur C18 AQ 3 µm resin (Dr. Maisch). Peptides were loaded into the column and eluted for 50 min using a segmented linear gradient of 5% to 95% solvent B (0 min, 5% B; 0-5 min, 5% B; 5-25 min, to 20% B; 25-35 min, to 35% B; 35-40 min, to 95% B; 40-50 min, 95% B) (solvent A, 0% ACN and 0.1% formic acid; solvent B, 80% ACN and 0.1% formic acid) at a flow rate of 300 nl min −1 . Mass spectra were acquired in data-dependent acquisition mode using a TOP10 method. Mass spectra were acquired in the Orbitrap analyzer with a mass range of 300-1,500 m/z at a resolution of 70,000 full width at half maximum, and a target value of 3 × 10 6 ions. Precursors were selected with an isolation window of 2.0 m/z. HCD fragmentation was performed at a normalized collision energy of 25. MS/MS spectra were acquired with a target value of 5 × 10 5 ions at a resolution of 17,500 full width at half maximum, a maximum injection time of 85 ms and a fixed first mass of 100 m/z. Peptides with a charge of 1, greater than 6 or with an unassigned charge state were excluded from fragmentation for MS 2 ; dynamic exclusion for 20 s prevented repeated selection of precursors.
Raw data were directly analysed at the MS1 level using Skyline (https://skyline. ms) 62 , against the sequence of the flg22 peptide. LysC specificity was required, and a maximum of two missed cleavages was allowed. The minimum peptide length was set to 7 amino acids and the maximum length was set to 25 amino acids. Carbamidomethylation of cysteine, oxidation of methionine and protein N-terminal acetylation were set as modifications. The results were filtered for precursor charges of 2 and 3. Peaks of the intact flg22 peptide precursor were integrated manually, and peak areas were exported for further processing.

Infection experiments.
To detect any Pseudomonas-401-dependent symptoms and the corresponding attenuation by SynComs, plants were germinated with Pseudomonas 401 alone (OD 600 = 0.0005; Extended Data Fig. 1c), or coinoculated with a SynCom for 2 weeks (Fig. 4a). For viable plate-counting experiments, plants were pregerminated on half-strength MS agar plates inoculated with the indicated strain or SynCom for 14 d (Fig. 4b,c). Plants were then flood inoculated with a bacterial suspension of Pseudomonas strain 401 or OTU5 (p5.e6) OD 600 = 0.0001) in 10 mM MgSO 4 supplemented with 0.005% silwet. Excessive liquid was removed 5 min after flood inoculation, and plants were transferred to a new half-strength MS agar for further incubation. After 1 d or 2 d, shoots and roots were separated, roots and shoots from five and three individual plants were pooled together after brief washing and blotted dry with sterile filter paper. The samples were homogenized with metal beads in 500 µl MgSO 4 using a Precellys 24 homogenizer (6,200 r.p.m. twice for 30 s with 15 s pauses in between; Bertin Technologies). Pseudomonas strain 401 and OTU5 (p5.e6) were transformed with pBBR1-MCS5 carrying a gentamycin-resistance cassette. Serial dilution was performed, and bacterial dilutions were spread on 50% TSB plates supplemented with 25 ng µl −1 gentamycin to select for the strain of interest until single colonies appeared.
In vitro halo-of-inhibition assay. Washed Pseudomonas strain 401 (100 µl) was inoculated into 50 ml warm 25% TSB medium with an initial OD 600 = 0.1. After solidification, 10 µl prewashed bacterial suspension prepared from 1 ml saturated overnight bacterial culture from individual strains was spotted onto the Pseudomonas-401-preinoculated plates. Any halo-of-inhibition was recorded up to 5 d after incubation at 25 °C.
Transcriptome experiments. Plants were germinated with the indicated SynCom in the presence or absence of 1 µM flg22 and incubated for 14 d before harvesting. For transcriptome experiments, Col-0 plants were inoculated with SynCom NS3 and SynCom S3 (Fig. 3); pWER::FLS2-GFP plants were inoculated with SynCom NS1 and SynCom S1 (Extended Data Fig. 6). For transcriptome experiments, plants were not transferred to minimize induced damage. Roots from three plates (minimum of 15 plants) were combined as one replicate and a total of three replicates were sampled for each condition. Roots were washed briefly with sterile water and blotted dry before being transferred into Lysing Matrix E tubes (MP Biomedicals) at −80 °C until processing. Roots were homogenized with Lysing Matrix E using a prechilled adapter and the TissueLyser II (QIAGEN, 20 pulses per s for 1 min). RNA was extracted using the Plant RNeasy Mini Kit (QIAGEN) according to the manufacturer's instructions. RNA quality was determined using a 2100 Bioanalyzer (Agilent Technologies). Preparation of Illumina sequencing libraries was conducted by the Max Planck Genome Center using an input of 1 μg total RNA. Sequences were generated using the Illumina HiSeq2500 platform. Approximately 6 million paired-end reads and 20 million single-end reads per sample with a length of 150 bp were generated for Col-0-and pWER::FLS2-GFP-based experiments.
RNA-seq data analysis. Raw Illumina RNA-seq reads were preprocessed using fastp (v0.19.10) 63 with the default settings for paired-end (Col-0 experiment) or single-end reads (pWER::FLS2-GFP experiment). For single-end reads, low-quality sequences from the head (8 bases) and tail (2 bases) were trimmed. High-quality reads were pseudoaligned to the TAIR 10 Arabidopsis thaliana transcriptome reference (Ensembl) 64 using kallisto (v.0.46.1) 65 . On average, 6.7 million paired-end and 18.1 million single-end reads per sample were mapped to the reference Arabidopsis transcriptome. After removing low-abundant transcripts that were absent in at least two replicates under each condition, count data were imported using the tximport 66 package.
Differential expression analyses were performed using the DESeq2 package 67 . First, raw counts were normalized with respect to the library size (rlog function) and log 2 -transformed. We tested for sample batch effects by surrogate variable (SV) analysis using the sva 68 package. Significant SVs were automatically detected and integrated into the model for differential analyses. PCA (prcomp function) on the basis of whole transcripts was performed and plotted to visualize the cluster and variance of biological replicates under each condition. The abundance of Arabidopsis latent virus-1 reads did not correlate with sample variances and therefore removed from downstream analyses. Pairwise comparisons were designed as: (1) flg22 treatment effect only; (2) non-suppressive and suppressive SynCom effect only; (3) flg22 treatment plus SynCom effects; and (4) living versus heat-killed bacteria. Transcripts with fold changes of >1.5 and adjusted P values corrected for multiple comparisons (Benjamini-Hochberg method) equal to or below 0.05 were considered to be significant.
The log 2 -scaled counts were normalized by the identified SVs using the limma 69 package (removeBatchEffect function), and transformed as median-centred z score by transcripts (scaled counts, scale function). Then, z scores were used to conduct k-means clustering for all transcripts. The cluster number (k = 10) was determined by the sum of squared error and Akaike information criterion. Next, confirmed transcripts with similar expression patterns were grouped in the same cluster. Differentially expressed transcripts (3,718 in pWER::FLS2-GFP and 4,450 in Col-0 experiments) and cluster results were visualized using heat maps generated using the ComplexHeatmap 70 package.
GO enrichment for each cluster using the whole Arabidopsis transcriptome as background were performed using the goseq 71 package with the consideration of transcript length. GO annotations were retrieved from the Gene Ontology Consortium 72,73 (September 2019). Significantly changed biological process GO terms (adjusted P < 0.05) were visualized in dot plots using the clusterProfiler 74 package. Defence-related genes were extracted on the basis of the GO-term annotation with manual curation. These genes were marked in Supplementary Table 7.
A GO gene network was built by connecting GO terms with shared differentially changed genes (Jaccard similarity > 0.2), such that GO terms holding close function annotations were gathered. Nodes in the network were coloured according to their representation in the k-means clustering analysis, while their size corresponded to the number of genes annotated in the corresponding GO term. GO gene networks were visualized in Cytoscape 75 with a modified configuration from metascape 76 .
DEGs from another two RNA-seq datasets, Teixeira et al. 31 and Harbort and Hashimoto et al. 13 , were used to confirm genes involved in SynCom response. The RNA-seq analyses pipeline was the same as described above. GO enrichment was conducted on the basis of the common significantly changed genes from this study and published datasets using clusterProfiler package. qPCR with reverse transcription. Plants were germinated on half-strength MS plate without sugar in the presence of the indicated SynCom. For chronic response, 1 μM flg22 was added to the agar plate at the beginning of the experiment. For acute response, plants (aged 2 weeks) were submerged with 1 μM flg22 solution in half MS medium for 1 h. Roots were then harvested after 1 h or 24 h. For chitin-related experiments, chitin was dissolved in sterile water at 10 mg ml −1 with gentle shaking together with metal beads for a few hours at 4 °C. Resuspended chitin solution was diluted to 1 mg ml −1 and autoclaved. The solution was centrifuged to obtain clear supernatant which was subsequently used to flood inoculate plants for 1 h, 6 h or 24 h. Roots from at least five plants (aged 2 weeks) were pooled and total RNA was extracted using the RNeasy Plant Mini Kit (QIAGEN) according to the manufacturer's instructions. Total RNA (200-500 ng) was treated with DNase, and then processed for first-strand cDNA synthesis using oligo dT primers and superscript II reverse transcriptase (Invitrogen). cDNA was diluted 10 times and 5 µl sample was used as a template for quantitative PCR analysis in a 20 µl reaction mixture supplemented with 1× iQ SYBR Green (Bio-Rad) and 0.2 µM primer each. UBQ5 (Extended Data Fig. 10a-c) or SAND (At2g28390) (Extended Data Fig. 10h) were used for internal normalization. The fold change values were determined using the 2 −ΔΔct method. Statistical analyses were performed on the log-transformed fold changes (Extended Data Fig. 10a-c) and fold changes (Extended Data Fig. 10h) as indicated. A list of the primers used in this study is provided in Supplementary Table 13.
Statistical analysis. Analyses were performed using the R environment. t-tests, Dunn's Kruskal-Wallis tests, Dunnett's tests and ANOVA were used to test for statistical significance. Unless otherwise indicated, P < 0.05 was considered to be significant.
Reporting Summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
Raw transcriptome and 16S rRNA amplicon sequencing data from this project were deposited at NCBI under the accession number GSE157128. Mass spectrometry data have been deposited to Panorama Public (https://panoramaweb.org/flg22_ RGI.url) and the ProteomeExchange (PXD020452). Source data are provided with this paper.

NATuRE PlANTS
Extended Data Fig. 1 | pWER::FLS2-GFP transgenic Arabidopsis plants are hypersensitive to flg22. a, Representative images of 3-week-old plants treated with 1 μM flg22 grown on agar plate. b, Bacteria load on 3-week-old WT roots inoculated with the indicated commensal. Boxes with filling background represent strains with RGI suppressive activity and colors correspond to phylogeny. n=total number of biological samples collected from one replicate. Asterisks indicate statistical significance (two-sided ANOVA, p < 0.05). The box plots centre on the median and extend to the 25th and 75th percentiles, and the whiskers extend to the furthest point within the range of 1.5× the interquartile range. c, Representative images of 2-week-old Col-0 germinated with the indicated strains with detrimental impact.