Tryptophan metabolism and bacterial commensals prevent fungal dysbiosis in Arabidopsis roots

We hypothesized that specific, host immune sectors might be necessary for beneficial, growth-promoting activities of microbial root commensals. In the gnotobiotic FlowPot system (39, 41), we recolonized germ-free A. thaliana Columbia-0 (Col-0, referred to as wild type [WT]) as well as a wide range of immunocompromised mutants with a synthetic multikingdom microbial community representative of naturally occurring root microbiomes (183 bacteria [B], 25 fungi [F], and 6 oomycetes [O]; BFO synthetic microbial community [SynCom], SI Appendix, Fig. S1 and Dataset S1) (3). The mutants include bak1/bkk1 (42), bak1/bkk1/cerk1 (31), efr/fls2/cerk1 (43), lyk5 (44), apex (45), and two other mutants (hub1 and hub2) lacking hub proteins in the Leucine-rich repeat receptor kinases network with putative immunomodulatory functions (45), wrky33/40 (46), dde2/ein2/pad4/sid2 - deps (47), pad4 (48), cyp79b2/b3 (49), 35S::BRI1 (50), bri1-301 (51), and rar1 (52) (Fig. 1A and Dataset S2). Consistent with previous work (39), the BFO SynCom significantly promoted shoot fresh weight (FW) compared to germ-free control plants 5 wk post-BFO inoculation in the FlowPot system (t test, P = 1.2e to 10; Fig. 1B). Notably, living microbes were needed for this plant growth–promoting activity, since heat-killed BFO SynCom members no longer promoted plant growth in this gnotobiotic system (SI Appendix, Fig. S2A). To analyze mutant-specific growth changes induced by the presence of the root microbiota, we first calculated the amplitude of the BFO effect between colonized and sterile plants for each genotype (mutants and WT; SI Appendix, Fig. S2B) and then normalized these differentials to that observed for WT (relative growth promotion, Kruskal–Wallis and Dunn control test with Bonferroni correction, Fig. 1C, see Materials and Methods). Remarkably, the vast majority of tested immunocompromised mutants (i.e., 64%) lost the ability to benefit from the multikingdom BFO SynCom (bak1/bkk1, bak1/bkk1/cerk1, lyk5, wrky33/40, wrky33, pad4, cyp79b2/b3, 35S::BRI1, rar1; Fig. 1C), indicating that several immune sectors were necessary for plant growth–promoting activities of microbial root commensals. It is noteworthy that several of these mutants also showed a growth defect relative to the WT control when grown in the Cologne agricultural soil (CAS) under greenhouse conditions, whereas this effect was not observed under sterile conditions (5 wk, bak1/bkk1, wrky33/40, cyp79b2/b3, 35S::BRI1, and rar1; SI Appendix, Figs. S2B and S3A). The results indicate a link between plant innate immune pathways and maintenance of beneficial plant–microbiota interactions.

We hypothesized that lack of BFO-mediated plant growth promotion in several of the immunocompromised mutants observed in the FlowPot system was associated with abnormal signatures in root microbiota composition. We harvested peat bulk soil, as well as roots of WT and mutant plants 5 wk post-BFO inoculation and monitored microbial community diversity and composition using amplicon sequencing of the bacterial 16S ribosomal RNA (rRNA) gene (B, 799F-1192R primers) and the fungal and oomycete internal-transcribed spacer (ITS, F: ITS1F-ITS2 primers, O: ITS1O-5.8sO primers). Alpha diversity analyses based on different metrics (Shannon index, Chao index, and observed 'operational taxonomic units' [OTUs]) did not reveal major effects of the different immune sectors on bacterial, fungal, and oomycete diversity in root samples (Fig. 2A and SI Appendix, Fig. S4). Constrained principal coordinate analysis (cPCoA) based on Bray–Curtis dissimilarities between samples revealed an overall significant yet subtle effect of the genotype on bacterial and fungal community composition in roots (B: 5.89%, P = 0.001; F: 7.62; P = 0.001, O: P = 0.145; Fig. 2 D and E), which was corroborated by permutational ANOVA (PERMANOVA, B: R² = 0.033; P = 0.013; F: R₂ = 0.078. P = 0.002; O: R² = 0.084, P = 0.03; Dataset S3). Further inspection of pairwise differences between WT and individual, immunocompromised mutants using cPCoA revealed significant differences in community composition for 6 out of 15 genotypes for bacteria (hub1, hub2, bri1-301, wrky33, bak1/bkk1/cerk1, and cyp79b2/b3) and only 1 out of 15 genotypes for fungi (bri1-301) and oomycetes (wrky33) (ANOVA, P < 0.05; Fig. 2 D–F and Dataset S4). Consistent with this, a genotype effect was only significant for bacterial community composition when the same mutants were grown in the CAS soil under greenhouse condition (B: 8.82%, P = 0.001; F: P = 0.525, O: P = 0.051), with bacterial assemblages in roots of the hub1, cyp79b2/b3, efr/fls2/cerk1, wrky33/40, deps, rar1, and pad4 mutants differing significantly compared to the WT control (ANOVA, P < 0.05; SI Appendix, Fig. S3 B–D). We next tested whether differentiation in root microbiota composition detected across the different genotypes in the FlowPot system could explain the observed variation in BFO-mediated plant growth promotion (SI Appendix, Fig. S5). Using linear regression models, we did not detect significant associations between relative growth promotion indices (mean relative FW of colonized versus sterile genotypes; see Fig. 1C) and community differentiation measured along the first and second axis of the principal coordinate analysis (n = 15, P > 0.05 for bacteria, fungi, and oomycetes; SI Appendix, Fig. S5). Consistently, genotype-dependent changes in microbial assembly with respect to WT were only subtle at strain-level resolution (edgeR, generalized linear model, P < 0.05; SI Appendix, Fig. S6). Our results suggest that inactivation of specific immune sectors mildly sculpt microbial assemblages in roots and that genotype-specific differences in the composition of the root microbiota are unlikely the primary cause driving variation in BFO-mediated plant growth promotion across mutants.

We hypothesized that total microbial abundance in roots, rather than shifts in microbial community composition, might explain variation in BFO-mediated plant growth promotion in the FlowPot system. Using the same root samples used for microbial community profiling, we quantified bacterial, fungal, and oomycete load relative to the plant DNA marker gene UBQ10 by qPCR (Fig. 3 A–C and Dataset S5). Specificity of all primer pairs was tested and cross-kingdom primer amplification was only observed between bacterial and plant DNA for the 799F-1192R primer pair. However, dilution series of pure bacterial DNA mixed with a fixed concentration of plant DNA indicated a linear amplification of the bacteria 16S rRNA gene, therefore suggesting a limited influence of plant DNA on bacterial quantification measurements (Materials and Methods and SI Appendix, Fig. S7). We detected significant, mutant-specific differences in bacterial and fungal but not oomycete load in plant roots with respect to WT (Kruskal–Wallis and Dunn control test with Bonferroni correction, P < 0.05; Fig. 3 A–C). Roots of the bak1/bkk1 mutant had a significantly higher bacterial load than WT control plants (Fig. 3A), whereas those of the efr/fls2/cerk1, wrky33, and cyp79b2/b3 mutants showed extensive fungal colonization (Fig. 3B). Inspection of fungal load in roots of the mutants grown in the CAS soil under greenhouse conditions revealed that the fungal load was the highest in the efr/fls2/cerk1, cyp79b2/b3, and lyk5 mutants, although the differences were not significant among genotypes (SI Appendix, Fig. S3 E and F). To determine whether total microbial load can more precisely explain variation in BFO-mediated plant growth promotion across mutants observed in the FlowPot system (see Fig. 1C), we employed a similar linear regression model as described above (Fig. 3 D–F). Remarkably, increase in fungal, but not bacterial or oomycete load in plant roots, was significantly correlated with lack of BFO-mediated plant growth promotion (n = 15, R² = 0.4196, P = 0.005374; Fig. 3E). Notably, these differences in fungal load measured across genotypes explained 42% of the between-genotype variation in BFO-mediated plant growth promotion (Fig. 3E). The results suggest that control of fungal load in plant roots by independent immune sectors is key for maintaining the beneficial activity of the multikingdom BFO SynCom.

Based on above-mentioned experiments, we observed that inactivation of two functionally redundant genes required to convert Trp into indole-3-acetaldoxime (IAOx, CYP79B2 and CYP79B3) (49) was sufficient to shift a beneficial plant–microbiota association from a homeostatic state into a dysbiotic state (Fig. 1C). IAOx is precursor of several types of known Trp derivatives, including indole glucosinolates (IGs), with their hydrolysis products, and camalexin and indole-3-carboxylic acids (ICAs), whose accumulation was strongly compromised in roots of cyp79b2/b3 plants grown in the FlowPot system, as compared with WT plants (SI Appendix, Fig. S8A). We also observed that levels of free indole-3-acetic acid (IAA, auxin) were significantly reduced with respect to WT, partly recapitulating reductions observed for CYP79B2/B3-dependent metabolites (SI Appendix, Fig. S8B). This is consistent with previous results observed for the same mutant but in leaves and at temperatures that were higher [26 °C (49)] than those measured inside our FlowPot system (i.e., oscillating between 18.9 °C, light phase, and 24.3 °C, dark phase; SI Appendix, Fig. S9). In contrast, levels of CYP79B2/B3-independent metabolites, such as aliphatic glucosinolates, flavonoids, coumarins, and phenolic acids did not vary significantly between cyp79b2/b3 and WT in root samples (SI Appendix, Fig. S8C), confirming the specific impact of cyp79b2/b3 mutations on Trp derivatives in roots of plants used in our experiments. We tested the extent to which the different branches of Trp metabolism could contribute to the maintenance of fungal homeostasis in roots and the BFO-mediated plant growth promotion using a set of mutants that, according to the literature, should be defective in the accumulation of camalexin [pad3 (53), cyp71a27 (25), and cyp71a12/a13 (54)], ICAs [cyp71a12/a13 (54)], IGs [myb34/51/122 (55)], and some of their hydrolysis products [pen2 (56) and pyk10/bglu21 (57)] (SI Appendix, Fig. S10A and Dataset S2). By repopulating these mutants and WT plants with the BFO SynCom in the gnotobiotic FlowPot system, we observed that none of the tested mutants phenocopied plant growth (SI Appendix, Fig. S10 B and C) and fungal load (SI Appendix, Fig. S10 D–F) phenotypes observed in the context of the cyp79b2/b3 mutant. To validate deficiency of tested lines in the accumulation of particular metabolites, we analyzed their accumulation in roots of these mutants inoculated with the fungal pathogen Plectosphaerella cucumerina, a species that is widespread in A. thaliana roots (3) and present in our fungal SynCom. This analysis proved lack of camalexin in roots of pad3 and cyp71a12/a13 lines as well as IG deficiency in myb34/51/122 mutant (SI Appendix, Fig. S11); however, it did not confirm partial ICA deficiency observed earlier in infected leaves of cyp71a12/a13 plants (58). Strikingly, we also found a cyp79b2/b3-like reduction in free IAA levels in roots of myb34/51/122 plants, which indicated that in A. thaliana roots significant amounts of this hormone can be derived from IAOx through IGs, as already postulated (59). Collectively, our metabolic analysis, combined with results on fungal load (SI Appendix, Fig. S10 D–F) and plant growth promotion (SI Appendix, Fig. S10 B and C), excluded individual contributions of IAA, IGs, and camalexin but not of ICAs to fungal overgrowth in cyp79b2/b3 plants.

To test the robustness of the dysbiotic phenotype (i.e., increased fungal load and altered plant growth) during the plant life cycle, we adapted our gnotobiotic FlowPot system to accommodate plant growth for up to 8 wk until reproductive stage and production of the first siliques (see Materials and Methods and SI Appendix, Fig. S12). By repopulating roots of WT and cyp79b2/b3 genotypes with the BFO SynCom, we observed that plant dry weight of the cyp79b2/b3 mutant was significantly reduced compared to sterile plants 8 wk post-BFO inoculation (Kruskal–Wallis and Dunn test with Bonferroni correction), which was not the case for WT plants (Fig. 4A; see BFO WT versus BFO cyp79b2/b3). Consistent with the aforementioned results obtained at the vegetative stage (Figs. 1C and 3B), variation in BFO-induced differential growth between WT and cyp79b2/b3 at the reproductive stage was associated with increased root fungal but not bacterial load (Fig. 4 D–F; see BFO WT versus BFO cyp79b2/b3) and changes in bacterial but not fungal community composition between the two genotypes (Fig. 4 G and H; see BFO WT versus BFO cyp79b2/b3). These results were validated by PERMANOVA (Dataset S6; see model BFO, genotype effect, B: R² = 0.1685, P = 0.001, F: R² = 0.068, P = 0.249). Our results indicate that the striking dysbiotic phenotype observed for the cyp79b2/b3 mutant at the vegetative stage was retained at the reproductive stage, irrespective of difference in residence time (vegetative stage: 5 wk and reproductive stage: 8 wk) and growth conditions (vegetative stage: light cabinet and reproductive stage: greenhouse). Notably, high fungal load and BFO-induced growth penalty observed in this mutant, compared to WT control plants, were not associated with significant differences in bolting time, number of first siliques produced per living plant (Fig. 4 B and C; see BFO WT versus BFO cyp79b2/b3), or other tested parameters (SI Appendix, Fig. S13), suggesting that the living plants harvested 8 wk post-BFO inoculation showed penalty on growth rather than on these initial indicators of reproductive fitness.

We previously showed that bacterial root commensals modulate fungal diversity and composition at the root interface, thereby promoting A. thaliana survival in the gnotobiotic FlowPot system (39). Therefore, we also tested the extent to which bacterial commensals and host Trp metabolism act in concert to modulate fungal growth in plant roots to promote A. thaliana health.

In the above-mentioned experiment conducted at the reproductive stage, we also recolonized roots of WT and cyp79b2/b3 genotypes with all possible combinations of single- and multikingdom microbial consortia (B, F, O, BO, BF, and FO). We observed that F, O, and FO communities negatively impacted the dry weight of WT plants (Fig. 4A) and triggered, in most cases, early bolting and silique production compared to the sterile control condition (Fig. 4 B and C). Dramatic consequences on growth, survival, and reproductive fitness were observed for the cyp79b2/b3 mutant since none of the plants survived in the absence of the bacterial community (see crosses in Fig. 4 A–C and SI Appendix, Fig. S13). In contrast, bacterial root commensals alone were not detrimental for the growth of both WT and cyp79b2/b3 genotypes (B WT and B cyp79b2/b3) and were able to fully rescue cyp79b2/b3 rosette dry weight to control level in the presence of oomycetes (cyp79b2/b3, O versus BO condition) but not when fungi were present in the SynComs (cyp79b2/b3, B versus BF and B versus BFO conditions, P < 0.05, ANOVA and Tukey’s post hoc test) (Fig. 4A). These results indicate that the presence of fungi rather than oomycetes or bacteria in the BFO SynCom was likely the cause of the dysbiotic phenotype observed for this mutant (Fig. 4A). Importantly, a similar experiment conducted at the vegetative stage with individual microbial groups or their combinations showed the same results, thereby strengthening this conclusion (SI Appendix, Fig. S14). Furthermore, we also assembled a second independent fungal SynCom using strains that were previously isolated from roots of A. thaliana harvested in a natural population in Saint-Dié, France (39), for repopulation experiments in the FlowPot system (F^sd: 23 members; Dataset S1). Although F^sd was less detrimental on plant growth than the original F SynCom, the fungal effect on WT and cyp79b2/b3 significantly differed in the absence of bacterial root commensals (P < 0.05, Kruskal–Wallis test and Dunn test; SI Appendix, Fig. S15), validating the hypersusceptibility of the cyp79b2/b3 mutants to at least two independent fungal communities isolated from geographically distant environments.

To determine how bacterial commensals supplement Trp metabolism to prevent fungal dysbiosis, we next inspected microbial load (qPCR; Fig. 4 D–F), as well as microbial diversity and microbial community composition in plant roots (amplicon sequencing; Fig. 4 G and H and SI Appendix, Fig. S16). Given that the cyp79b2/b3 mutant did not survive treatments with fungi only, we inspected these signatures in the context of WT plants. We observed that the presence of bacterial commensals led to a significant reduction in fungal load in plant roots (Kruskal–Wallis and Dunn test with Bonferroni correction, P < 0.05; Fig. 4E) but not diversity (Kruskal–Wallis and Dunn test, P > 0.05; SI Appendix, Fig. S16 A–C), as well as a dramatic shift in fungal community composition (Fig. 4H and SI Appendix, Fig. S16E) compared to conditions in which these fungi were inoculated in the absence of bacteria (see F versus BF conditions). These differences in fungal community composition in the presence of bacteria were corroborated by PERMANOVA (Dataset S6; global model, Treatment: F: R² = 0.287, P = 0.001). Taken together, the results indicate that modulation of fungal load by the host Trp metabolism and of fungal composition and fungal load by bacterial root commensals are both necessary and act additively to prevent fungal dysbiosis in A. thaliana roots.

In this study, a total of 23 A. thaliana lines were used. A. thaliana Col-0 was used as a WT, together with 22 mutants in the Col-0 background (Dataset S2).

Natural soil experiment was performed in the greenhouse in short day conditions (8-h light) in 9 × 9 cm square pots filled with natural CAS soil. Plants were grown for 5 wk, and the pots were randomized weekly within biological replicates. Around ∼2 wk after sowing, the extra seedlings were removed and only five plants were left per pot.

A total of 183 bacterial strains isolated from healthy A. thaliana roots (20, 74) were grown for 7 d in 600 μL 50% tryptic soy broth liquid media from a glycerol stock. A total of 100 μL of each strain was taken, combined together, centrifuged, and the pellet was resuspended in 10 mM MgCl₂. A total of 25 fungal and 6 oomycetes strains were grown individually on potato glucose agar media for 2 wk and harvested 1 d before the experiments (20). Harvested F and O mycelium (average of 50 mg/strain) was suspended in 1 mL 10 mM MgCl₂ inside a sterile 2-mL screwcap tube containing one stainless steel bead (3.2-mm diameter) and left at 4 °C overnight. On the day of the experiment, the mycelium was crushed for 10 min in a paint shaker (SK450, Fast & Fluid Management). A similar protocol was used to prepare the second fungal SynCom. Bacterial, fungal, and oomycete strains used in this study can be found in SI Appendix, Fig. S1 and Dataset S1. Note that over the course of experiments two oomycetes strains (namely 210 and 29) did not survive, and so, they were only used in the initial screen of innate immunity mutants.

To calculate the relative growth promotion index (Fig. 1C and SI Appendix, Fig. S10C), the first step was to compute the difference in shoot FW between BFO and sterile conditions per genotype (i.e., FW_BFO – mean FW_sterile). This allows to account for genotype-specific differences in vegetative growth. In this first step of the calculation, values will be above zero if there is a growth promotion and below if the growth is restricted compared to sterile conditions. Because the experiment was done in two batches, these differential growth values measured for WT and mutants were all normalized to that of WT by dividing them with the mean differential, observed in the WT context. For WT, this normalization resulted in a mean relative growth promotion value of 1. However, note that in the boxplots depicted in Fig. 1C and SI Appendix, Fig. S10C, the median is shown, not the mean.

Methodology of this experiment was the same as for vegetative stage experiment, with the use of only WT and an addition of “heat-kill” treatment. Heat-kill treatment was performed by taking the fully assembled BFO SynCom (prepared as described above) and subjecting it to two subsequent rounds of autoclaving (20 min at 121 °C for each round). Each FlowPot from heat-kill treatment was inoculated with 200 µL heat-killed bacterial, 200 µL heat-killed fungal, and 80 µL heat-killed oomycete communities.

DNA was isolated with FastDNA Spin Kit for Soil (MP Biomedicals) following the manufacturer’s instructions. Library for sequencing followed the protocol described in ref. 39. In short, after DNA isolation, DNA samples were diluted to 5 ng/µL based on PicoGreen measurements (Quant-iT PicoGreen dsDNA Assay-Kit, Invitrogen) and amplified in a two-step PCR with primers amplifying the bacterial 16S rRNA V5-V7 region (799F-1192R) and the fungal and oomycete ITS1 (fungi: ITS1F-ITS2 and oomycetes: ITS1O-5.8sO) (see ref. 39). Amplified bacterial products were purified on 1.5% agarose gel with QIAquick Gel Extraction Kit (Qiagen, category No. 28704) and fungal and oomycetes products with Agencourt AMPure XP beads (Beckman Coulter, category No. A63882). After purification, single bacterial, fungal, and oomycetes samples were pooled together within their respective microbial groups in equimolar concentrations, cleaned again with Agencourt AMPure XP beads, and finally pooled together into one final microbial library sample. Final pooling of bacterial, fungal, and oomycetes samples varied between 300 and 850 ng per microbial group, depending on the availability of the samples.

Prepared libraries were sequences on a MiSeq machine with pair-end Illumina sequencing (MiSeq reagent Kit v3, 600 cycle, category No. MS-102-3003). Primers used for sequencing are as described previously in ref. 39. Quality filtered and demultiplexed sequencing reads were mapped at 98% identity to the reference sequence database for bacteria, fungi, and oomycete using usearch (75). Unmapped reads were discarded. Count tables were derived from this mapping. Samples used for further analysis were filtered with the threshold of minimum 1,000 reads per sample for all microbiota-dependent analysis.

Primers tested for specificity are listed in Dataset S5. Tests for specificity were performed with the same PCR protocol used for amplicon library preparation [PCR I (39)]. Primers amplifying the A. thaliana UBQ10 showed the highest-primer efficiency and no signs of cross-amplification. Primers amplifying the bacterial 16S rRNA gene (V5-V7 region, 799F-1192R), and the fungal and oomycete ITS1 (fungi: ITS1F-ITS2, oomycetes: ITS1O-5.8sO) were chosen with the main advantage of being the same primer pairs used for microbial community profiling (Dataset S5). Subsequent PCR tests revealed that fungal and oomycetes primers are fully specific to their respective synthetic communities. Bacteria primers, although cross-amplifying the plant 16S rRNA gene, show a strong preference for bacterial DNA, since Cq readout was highly correlated with increase of bacterial load, regardless of the varying presence of plant DNA (SI Appendix, Fig. S7). Note that at least one oomycete strain used in our SynCom was colonized by a hyphae-associated bacterium, causing an unspecific cross-amplification with the primers used to amplify the bacterial 16S rRNA gene. The qPCR protocol as follows: 95 °C for 3 min, 40 cycles (95 °C for 15 s, 60 °C for 30 s, and 72 °C for 30 s), 95 °C for 10 s, and melting curve measurement from 55 °C to 95 °C with an increment of 0.5 °C. The total microbial load (relative to UBQ10) was calculated with the use of the following formula. Analysis includes one reference sample present on each plate in technical triplicates, serving as an interpolate normalization (named ref. in below formula).

MS medium with 1% Agar (BioShop) and 0.5 mM MES, pH 5.7, was poured into 120 × 120 mm square GREINER plates, and a 2-cm slide was cut out of the plates. Stratified sterile Arabidopsis seeds were imbibed in 10 mM MgCl₂ for 48 h in and transferred on the edge of the removed agar slide (10 seeds/plate). The plates were kept under short day conditions in PANASONIC MLR-352-PE phytochamber, as described earlier (73). Roots of 4-wk-old seedlings were spray inoculated with spore suspension (4 × 10⁶ spores/mL) of adapted Brigitte Mauch-Mani (BMM) isolate of P. cucumerina BMM. Around 48 h after inoculation, root samples (about 100 mg) were collected and frozen immediately in liquid nitrogen.

Frozen root samples were extracted with DMSO (Sigma-Aldrich) containing 0.5 mM camphorsulfonic acid and 0.5 mM lidocaine (Sigma-Aldrich), as described earlier (58). Samples were subjected to liquid chromatography–mass spectrometry (LC-MS) analyses performed using the UltiMate 3000 RS (Dionex, Thermo Fisher Scientific) attached to a TIMS-TOF mass spectrometer (Bruker Daltonics). Chromatographic separation was carried out on a BEH RP C18 column (2.1 × 150 mm, 1.7-µm particle size) at 30 °C with a mobile phase flow rate of 0.30 mL/min. The elution was conducted using water containing 0.1% formic acid (Sigma-Aldrich) (Solvent A) and acetonitrile (VWR Chemicals) containing 0.01% of formic acid (Solvent B) in the following gradient: 0 to 5 min from 10 to 30% B, 5 to 12 min to 100% B, 12 to 15 min maintained at 100% B, and up to 15.5 min the system was returned to starting conditions and reequilibrated for 5 min. The spectrometer was calibrated with sodium formate salt clusters prior to each analysis. MS was operated using the following settings: ion source voltage of −4.5 kV, nebulization of nitrogen at a pressure of 2.2 bar, and a gas flow rate of 10 L/min. Ion source temperature was 220 °C. The spectra were scanned in positive and negative ionization in fragmentation mode (data-dependent tandem mass spectrometry, ddMSMS) at a range of 95 to 1,000 m/z at a resolution >30,000 full width at half maximum (Dataset S7). Data acquisition was supervised by Compass HyStar version 5.1 software (Bruker Daltonics). Data were analyzed by Compass DataAnalysis version 5.3 (Bruker Daltonics). Data from both experiments, FlowPot and agar-based system, were processed separately by MS-Dial ver 4.24. Processing steps included conversion of raw LC-MS file to format suitable for MS-Dial software, transformation from profile to centroid data, peak detection, annotation to spectral MSMS public metabolomic library, adduct elimination, alignment, and gap filling by compulsion (Dataset S8). IAA, camalexin, and ICA peaks were identified based on LC-MS analysis of respective standard compounds. Other metabolites were putatively identified based on their mass-to-charge ratio (m/z value) and fragmentation spectra.

All statistical analyses were performed in R. Differences were considered as statistically significant when P < 0.05. For Kruskal–Wallis and Dunn test, FSA package was used, and for Dunn control test, PMCMR package was used. When the variables were normally distributed, the effect of the treatments and genotypes were assessed using linear models with ANOVA. Whenever necessary the response variable was root square or log transformed to ensure a normal distribution of the model’s residuals.

The models were constructed as follows: $\text{lm}\left(\text{log}\left(\text{bolt}\_\text{to}\_\text{flower}\_\text{days}\right)\sim \text{genotype}*\text{treatment}\right)$.

Generalized linear models (GLMs) were used when the transformation did not successfully normalize the variable. An example of GLM model with gamma distribution is presented below:

Differential abundance of strains between WT and Mutants was calculated by normalization of raw sequencing counts (TMM normalization, “calcNormFactors” from R package “EdgeR”) and fitting a generalized linear model (“glmFit”) including the replicate effects. Significantly different microbial taxa were then determined with a likelihood ratio test (“glmRT,” P < 0.05).

Variance partitioning between experiments, biological replicates, genotypes, and treatments were tested with a PERMANOVA approach with the Adonis function (R package vegan) in all experiments. For the root microbiota differentiation across mutants experiment, the effect of the genotype, the biological replicate, and the experimental replicate on microbial communities’ composition were tested in a global model, including the soil samples and a second model without soil samples. These models were constructed separately on each of the bacterial and fungal datasets using Bray–Curtis dissimilarity matrices between pairs of samples produced with the vegdist function (R package vegan). For the WT versus cyp79b2/b3 experiment, the effect of the genotype and treatment factors on microbial communities’ composition were tested in a global model. The effect of the genotype was then tested in separated models for each treatment separately and for bacteria and fungi separately. These models where constructed using Bray–Curtis dissimilarity matrices between pairs of samples produced with the vegdist function (R package vegan).