Bacillus subtilis biofilm matrix components target seed oil bodies to promote growth and anti-fungal resistance in melon

Beneficial microorganisms are used to stimulate the germination of seeds; however, their growth-promoting mechanisms remain largely unexplored. Bacillus subtilis is commonly found in association with different plant organs, providing protection against pathogens or stimulating plant growth. We report that application of B. subtilis to melon seeds results in genetic and physiological responses in seeds that alter the metabolic and developmental status in 5-d and 1-month-old plants upon germination. We analysed mutants in different components of the extracellular matrix of B. subtilis biofilms in interaction with seeds and found cooperation in bacterial colonization of seed storage tissues and growth promotion. Combining confocal microscopy with fluorogenic probes, we found that two specific components of the extracellular matrix, amyloid protein TasA and fengycin, differentially increased the concentrations of reactive oxygen species inside seeds. Further, using electron and fluorescence microscopy and metabolomics, we showed that both TasA and fengycin targeted the oil bodies in the seed endosperm, resulting in specific changes in lipid metabolism and accumulation of glutathione-related molecules. In turn, this results in two different plant growth developmental programmes: TasA and fengycin stimulate the development of radicles, and fengycin alone stimulate the growth of adult plants and resistance in the phylloplane to the fungus Botrytis cinerea. Understanding mechanisms of bacterial growth promotion will enable the design of bespoke growth promotion strains.

S eed germination is a complex biological process that is controlled by interconnected hormone-regulated pathways 1 . Seed inoculation with plant growth-promoting rhizobacteria is commonly used to enable bacterial colonization of plants and beneficial effects of plant growth-promoting rhizobacteria post germination 2 . A detailed understanding of the mechanisms that underpin growth-promoting host-microbe interactions is important to enable design of innovative biotechnological probiotics.
Bacillus subtilis and closely related species coexist with plants, providing multiple beneficial services 3 . Production of polyvalent secondary metabolites, sporulation or biofilm formation underpin bacterial fitness and bioactivity towards plants 4 . Biofilms are formed by bacterial cells that are embedded in a secreted extracellular matrix (ECM) 5 that is essential for efficient colonization of plant organs 6 . The ECM also comprises secondary metabolites that mediate bacterial communication with the plants by triggering physiological responses associated with defence or growth 7 .
Here we investigated the mechanism by which melon seeds respond to the stimulatory activity of B. subtilis and evaluated the functions of the ECM in this microbe-host interaction.

B. subtilis regulates metabolism and growth of colonized plants.
Beneficial bacteria can promote two different and genetically controlled stages of seed germination, namely germination itself and growth of the emergent radicle 8 . Melon seeds treated with B. subtilis NCIB 3610 produced larger radicles compared with those grown from untreated seeds (Fig. 1a). However, the germination rates (initial emergence of the radicle) of treated seeds did not change compared with untreated seeds. We analysed the expression levels of GA20ox1 and CYP07A1, which are two genes involved in the modulation of the determinant ratio of abscisic acid and gibberellins 9 , and found that they are statistically similar between treated and untreated seeds (Extended Data Fig. 1a,b). Analysis of transcriptomes of three pools of seeds 16 h after treatment with a suspension of B. subtilis did not identify any changes in the expression levels of the genes involved in the germination-related hormone signalling pathway. However, the upregulation of plant genes involved in carbon metabolism and photosynthesis, together with repression of plant heat shock proteins and structural proteins of lipid storage vesicles (oleosin and caleosin), was detected and enabled activation of seed metabolism ( Fig. 1b and Extended Data Fig. 1c,d) [10][11][12] .
Metabolomic profiles of treated and untreated seeds were compared to reveal increased abundance of energetic resources, such as glycerophospholipids and fatty acyls with analogy to triacylglycerides (TAGs) in the initial stages (0 h) after treatment with B. subtilis cells (Fig. 1c and Extended Data Fig. 2a). At 24 h after the treatment, metabolite levels were similar between both conditions, Bacillus subtilis biofilm matrix components target seed oil bodies to promote growth and anti-fungal resistance in melon growth-promoting activity might be triggered by bacterial cells that enter and colonize the seed storage tissues, which is a process that does not appear to rely specifically on any one type of bacterial motility.
Several studies have reported multifaceted contributions of bacterial biofilms to interactions with hosts 6,[19][20][21] . The cell densities of single structural mutants ∆tasA, ∆tapA, ∆eps and ∆bslA, which form altered biofilms, were substantially decreased in seed extracts compared with that of the wild-type bacteria (set at 100%) 5 d after seed treatment (Fig. 2d). However, this result was not associated with the attachment of bacterial cells to the seeds a few hours after treatments (Extended Data Fig. 3c).
A similar bacterial population pattern was monitored in the emergent radicles, which, as described above for WT, may reflect dragging along of bacterial cells (Extended Data Fig. 3d). The ∆srf, ∆pks or ∆pps strains, which do not produce the secondary metabolites surfactin, bacillaene or fengycin, respectively (also present in the ECM), had unchanged population dynamics compared with the WT strain ( Fig. 2d and Extended Data Fig. 3d). The decrease in the bacterial cell population size found in the structural mutants suggests that mutations led to a significant decrease in radicle growth promotion compared with treatment with the wild-type strain (Fig.  2e). In contrast to a correlation between bacterial cell density and plant growth promotion, the poorly persisting ∆tasA strain retained promoting activity, while the ∆pks and ∆pps strains failed to promote radicle growth despite considerable persistence on the seeds (Fig. 2d,e).
The absence of the amyloid protein TasA has recently been reported to provoke important cellular and physiological changes in B. subtilis 6 . To clarify whether the role of TasA in the structure of the ECM is related to TasA-dependent growth-promoting activity, we used B. subtilis strain JC81, which expresses a version of TasA that fails to restore biofilm formation but reverts the physiological status of ∆tasA cells to wild-type levels 6 . Treatment of seeds with JC81 failed to promote radicle growth, but the bacteria persisted on the seeds at a level comparable to that of the WT, ∆pks or ∆pps strains (Fig. 2d,e).
These results led us to propose three distinct and complementary contributions of ECM components to the promotion of seed radicle growth: first, EPS, BslA and TapA have a role in the persistence of B. subtilis cells in the seeds; second, fengycin (pps) and bacillaene (pks) have roles in the chemical dialogue of Bacillus with seeds; and third amyloid TasA is involved in both of these functions.
Comparison of the metabolic status of radicles 5 d after treatment of seeds with WT or various ECM mutant strains demonstrated a specific metabolic response of plants to chemicals present in the B. subtilis ECM (Extended Data Fig. 4a). The changes associated with bacterial treatment of seeds included a decrease in organooxygen radicle areas after seed treatments with B. subtilis. Statistical significance was assessed by a two-tailed t-test (n = 8; P = 0.0056). Right: representative radicles from untreated seed (left) and a B. subtilis-treated seed (right) 5 d after treatments. b, Volcano plot of DEGs identified by RNA-seq in bacterized seeds and untreated seeds 16 h after treatment. P values were calculated on the basis of the Fisher method using nominal P values provided by edgeR and DEseq2. Dashed lines represent the threshold defined for P (horizontal) and fold change (vertical) for a gene to be considered as DEG. Tags label the genes related to seed germination progress: OEE1, oxygen-evolving enhancer protein 1; OEE3-2, oxygen-evolving enhancer protein 3-2, chloroplastic; OEE2-1, oxygen-evolving enhancer 2-1, chloroplastic; Fd-like, erredoxin-like; PSAD2, photosystem I reaction centre subunit II, chloroplastic; Gdh, glutamate dehydrogenase; rbcL, ribulose bisphosphate carboxylase small chain; ppdK, pyruvate, phosphate dikinase; pckG, phosphoenolpyruvate carboxykinase; Clo, caleosin; Hsp70_2, heat shock 70 kDa protein_2; Hsp, class I heat shock protein; Hsp70, heat shock 70 kDa protein_1; and Hsp22, 22.0 kDa class IV heat shock protein. NS, not significant. c, Molecular families corresponding to fatty acyls analogous to triacylglycerides and glycerophospholipids differentially abundant in seeds 0 h after seed bacterization. Pie charts inside the nodes indicate the mean of the peak abundance of each metabolite in the corresponding condition. Node shape indicates the level of identification according to ref. 78 . The chemical structures of annotated features based on spectral matches to GNPS libraries are also presented for each molecular family. POPC, 1-palmitoyl-2-oleoyl-glycero-3-phosphocholine; PC, phosphocholine. d, Adult plants grown from seeds treated with B. subtilis (3610, right) or from untreated seeds (control, left). e, Circos plot showing the top 100 metabolites significantly more abundant in leaves of plants grown from bacterized seeds versus those in leaves of the control plants. Ribbon colours refer to the class of each metabolite, and thickness is proportional to the log 2 FC values. The curved colour bar shows the contribution of each metabolite (rectangles inside the bar coloured according to their chemical class) to the total abundance (100%), ordered from highest to lowest contribution. compound analogues of sphinganines and an increase in abundance of the prenol lipid and stilbene molecular families (Extended Data Fig. 4b). The presence of a functional ECM in the WT strain triggered the accumulation of fatty acyls belonging to two different clusters, and treatments with ECM mutants produced a clear decrease in organooxygen compounds and accumulation of specific prenol lipids (Extended Data Fig. 4b).
Lysophospholipids, glutathione and growth promotion. According to our data, TasA and fengycin are highly relevant molecules of the ECM with regards to growth promotion of B. subtilis in seeds. TasA, in common with other amyloids, is a polymorphic protein that can adopt a variety of structural conformations that have different biochemistries and functions 6,22,23 Fig. 2 | the extracellular matrix of B. subtilis is required for efficient seed colonization and radicle growth stimulation. a, B. subtilis dynamics (c.f.u. counts) in seed and radicle extracts during the first 5 d after seed treatment. Bars represent average values ± s.e.m. of total c.f.u. in seeds and radicles, and solid and dashed lines represent c.f.u. corresponding to the number of spores in seeds and radicles, respectively. Statistical significance was assessed by a one-tailed t-test between total c.f.u. at initial and final timepoints (n = 3, **P = 0.0062). b, C.f.u. counts of extracts from sectioned bacterized seeds separated at the micropylar (M) and chalazal (C) regions 4 h after treatment with the WT and motility-related mutants. Average values ± s.d. are shown (n = 3). c, C.f.u. counts of total extracts from bacterized seeds 4 h after treatment with the WT and motility-related mutants. Average values ± s.d. are shown (n = 3). d, Seed colonization of the ECM mutants relative to that of the WT bacteria (assumed to be 100%) measured in seed extracts 5 d after seed treatment. Average values ± s.e.m. are shown. Statistical significance was assessed by one-way ANOVA with Dunnett's multiple comparisons test (each treatment vs WT treatment) (n = 3, **P = 0.0027 in ∆eps, *P = 0.0135 in ∆bslA, **P = 0.0067 in ∆tasA, **P = 0.0013 in ∆tapA). e, Left: percentage of radicle growth increase in seeds treated with WT cells or ECM mutant cells (5 d after seed treatment) normalized to the average radicle area of untreated seeds (100%, dashed line). Mean ± s.d. Statistical significance was assessed by one-way ANOVA with post hoc Dunnett's multiple comparisons test (each treatment vs control treatment) (n = 17 in control and 3610, n = 18 in ∆eps, n = 12 in ∆bslA, n = 21 in ∆tasA, n = 3 in JC81, n = 22 in ∆tapA, n = 10 in ∆pks and ∆pps, n = 13 in ∆srf; ***P = 0.0002 in WT, **P = 0.0076 in ∆tasA, ***P < 0.0001 in ∆srf). Right: representative radicles from untreated seeds and seeds treated with WT or ECM mutant cells 5 d after treatment.   Right: representative radicles of untreated seeds or seeds treated with purified ECM 5 d after treatments. b, Molecular families corresponding to LysoPE and GSH. Pie charts represent the mean peak abundance of metabolites in each condition. Node shape indicates the level of identification according to ref. 78 . c, Top: representative radicles treated with water (C), LysoPE and GSH 5 d after treatment. Bottom: left Y axis, percentage of radicle growth promotion of seeds treated with water or purified LysoPE and GSH normalized to the control radicle area (100%, dotted line) 5 d after treatment. Statistical significance was assessed by one-way ANOVA with post hoc Dunnett's multiple comparisons test (each treatment vs control) (n = 24 in control, n = 8 in LysoPE and GSH; *P = 0.0451 in GSH, **P = 0.0021 in LysoPE); right Y axis, percentage of weight of adult plants grown from seeds treated with LysoPE and GSH, normalized to the weight of control plants (100%, dotted line) (n = 12 in control, n = 5 in LysoPE and n = 8 in GSH). Average values ± s.d. are shown. d, Representative CLSM images of the inner tissues of dihydrorhodamine-stained seeds immediately after seed treatments with water (control), fengycin or TasA. Graph represents the intensity of the fluorescence in each condition, measured as the mean grey value from n = 30 random sections from 3 different fields. Average values ± s.d. are shown. Statistical significance was assessed by one-way ANOVA with multiple comparisons test (P < 0.0001).
fied fengycin significantly increased the area of the radicles compared with untreated seeds (Fig. 3a). The lack of stimulatory activity of purified BslA or surfactin confirmed our previous findings using B. subtilis mutants (Figs. 2e and 3a).
The metabolome of individual radicles (n = 3) that emerged from seeds treated with fengycin or TasA was analysed to define their contributions to the metabolic signatures associated with the promotion of radicle growth. Results of principal component analysis and heatmap analyses indicated sample clustering in three groups (Extended Data Fig. 6a,b). Statistical analysis by partial least-squares discriminant analysis indicated that glycerophospholipids, fatty acyls, organooxygen compounds or carboxylic acids were discriminating metabolites in the case of TasA or fengycin treatments, and feature-based molecular networking of these features showed integration into molecular families composed of other metabolites that generally followed the same abundance patterns (Extended Data Fig. 6c-e). Further refinement of this analysis identified two molecular families of interest: lysophosphatidylethanolamine (LysoPE), mainly associated with TasA treatment, and an analogue of reduced glutathione (GSH) that accumulated in the radicles grown from seeds treated with fengycin (Fig. 3b). The treatment of seeds with commercially available LysoPE or GSH increased radicle growth as much as treatment with fengycin or TasA, but did not produce the long-term promoting effects observed in plants 1 month after treatment of seeds with B. subtilis (Fig. 3c).
GSH is an essential and polyvalent metabolite with special antioxidant functions, which maintains cellular redox homoeostasis and development, growth or response of plants to a variety of stimuli 24 . Further, an increase in the total pool of glutathione usually occurs 3 to 4 d after ROS exposure 24,25 , therefore we hypothesized that one possible outcome of fengycin treatment might be ROS production. The level of ROS in seeds after treatment with fengycin was statistically higher than that in untreated seeds. An intermediate increase in ROS, which was statistically less than that elicited by fengycin, was also observed in seeds treated with TasA (Fig. 3d).

TasA and fengycin differentially target seed oil bodies.
Accumulation of bacterial cells in the micropylar endosperm (Extended Data Fig. 3b), accumulation of TAGs (Fig. 1c) and changes in the lipid composition of the radicles and adult plants after treatment of seeds with bacteria or ECM components (Extended Data Figs. 2b and 6d,e) led us to hypothesize that seed oil bodies (OBs), which are nutrient reservoir organelles mainly composed of TAGs and other neutral lipids, are the target of fengycin and TasA. During germination, OBs are degraded by interaction with glyoxysomes that feed the plant embryo 26,27 . We applied CLSM to thin sections of seeds treated with a suspension of B. subtilis or untreated seeds using a neutral lipid stain 16 h after seed treatment and found an abundance of OBs in the endosperm of untreated seeds. In agreement with our hypothesis, certain regions of the endosperm accumulated high levels of disaggregated oil bodies that were surrounded by bacterial aggregates (Extended Data Fig. 7a). Treatment with purified TasA or fengycin promoted in vivo disaggregation of OBs, similar to the effect observed after treatment with Bacillus cells (Extended Data Fig. 8a). Transmission electron microscopy (TEM) analysis of thin sections of negatively stained seeds showed OB disaggregation and localization of OB-surrounding glyoxysomes in samples treated with TasA (Extended Data Fig.  7b,c). In fengycin-treated seeds, the most predominant changes were disaggregation of OBs and the presence of empty areas between the cell wall and the cytoplasm, suggestive of plasmolysis, a cellular response previously observed in the interaction of fengycin with membranes of fungal cells 28 (Extended Data Fig. 7c). The CLSM analysis of OBs purified from melon seeds treated with TasA or fengycin and double-stained with Fast Green FCF and Nile red indicated differences in the size and pattern of aggregation of OBs: fengycin induced disaggregation of OBs as observed in vivo, while TasA preferentially induced a higher level of aggregation of individual OBs than that observed in vivo (Fig. 4a). Higher intensity of the green fluorescence signal near OB aggregates in the samples treated with TasA suggested non-specific staining of the amyloid protein and its localization around the vesicles. TEM analysis of purified OBs and immunogold labelling with anti-TasA antibodies indicated the presence of gold particles decorating the perimeter of the OBs and interconnecting TasA fibres (Extended Data Fig. 7d). The results of pulldown assays of whole protein extracts of the seeds indicated co-elution of purified TasA with oleosins, which are structural proteins that modulate the stability and size of OBs, regulate lipid metabolism and have an important role in OB degradation 29,30 ( Fig. 4b and Supplementary Table 5). Interaction of the two proteins was confirmed by far-western blot analysis of purified TasA and purified oleosin 1 from Arabidopsis thaliana (Fig. 4c).
We propose that TasA-oleosin interaction is responsible for the aggregation of OBs, favouring the accumulation of lysophospholipids after the optimization of lipid catabolism in glyoxysomes. Fengycin has been described to efficiently interact with and disrupt artificial lipid monolayers or bilayer membranes in a concentration-dependent manner 31 . In addition to micellar concentrations of fengycin, specific disruption of the membranes relies on the lipid composition of the target 32 . Exact chemical composition of phospholipids of the membranes of OBs in the seeds is not known and varies between species 33 ; however, we propose that high affinity of fengycin for lipid membranes explains disaggregation and a reduction in the size of OBs.
Non-targeted metabolomic analyses provided additional information about the change in relative abundance of metabolites and potential (bio)chemical modifications of seed metabolites from 0 to 24 h after treatment with TasA or fengycin. The use of the chemical proportionality tool (ChemProp 34 ) showed putative changes in the dynamics of TAGs analogues after seed treatments (Fig. 4d). In OBs were stained with Nile red and membrane proteins were stained with Fast Green FCF. Colour bars below each image indicate the percentage of various statuses of OBs calculated by measuring each particle area (black, aggregation (area > 5 µm 2 ); red, individual OBs (0.5 µm 2 < area < 5 µm 2 ); grey, disaggregated OBs (area < 0.5 µm 2 )) in representative CLSM fields of view. b, Coomassie brilliant blue (CBB)-stained SDS-PAGE gel of elution fractions (E3 and E4) of pulldown assay of TasA with proteins in the extracts from melon seeds (left) and in absence of TasA as bait protein (right). The co-eluting band is marked with red arrows and identified as an oleosin by HPLC-ESI-MS/MS analysis. TasA bands are marked with red asterisks. c, Far-western blotting (F WB). Left: a CBB-stained SDS-PAGE gel of fractions of oleosin obtained during purification; BSA as a control, and fractions of BslA and TasA that were obtained during purification. Middle: immunoblot of purified proteins using an anti-TasA antibody (1:20,000). Right: far-western blot using an anti-TasA antibody (1:20,000) after renaturing of the proteins and incubation with TasA protein as a bait. d, Chemical proportionality analysis of the TAGs analogues molecular family in non-treated seeds (control) and seeds treated with fengycin or TasA from 0 h to 24 h after treatment. The size of the nodes is directly related to the abundance of the features at time 0 h. Arrows indicate chemical directionality of the modifications found and the colour scale of the arrow represents the ChemProp score. Floating bar plots represent the peak abundances of three selected TAGs analogues in non-treated (C) and treated seeds with fengycin (F) or TasA  control seeds, many modifications in this molecular family denoted an important catalytic activity occurring over key storage metabolites in this time period. Treatments with TasA and, more notably, with fengycin led to an increase in the initial abundance of many metabolites (Fig. 4d, floating bar plots) that would reflect the faster metabolic activity of these seeds. A decrease in the Chemprop score  TasA   Peak abundance   C0  C24  F0  F24  T0  T24  C0  C24  F0  F24  T0  T24  C0  C24  F0  F24  T0  T24   10   indicated that the significance of the changes between 0 h and 24 h would be directly related with the earlier initiation of the catalytic activity over these molecules. Similar changes to TAGs were also remarkable in the dynamics of glycerophospholipids (Extended Data Fig. 8b).
Fengycin promotes plant growth and resistance to Botrytis cinerea. Plants that emerged from seeds treated with fengycin were larger than plants grown from untreated or TasA-treated seeds (Fig.  5a). Adult plants grown from seeds treated with LysoPE or GSH, however, did not show a significant increase in growth over time compared with plants from untreated seeds (Fig. 3c). These findings indicate that the presence of different signalling events underpin short-term radicle growth mediated by TasA, fengycin or associated LysoPE and GSH molecules, as well as long-term growth of adult plants specifically associated with fengycin. GSH associated with treatments of the seed with fengycin did not produce a sustained long-term growth-promoting effect, suggesting that a constant endogenous trigger is required that can only be achieved by fengycin treatment. Both TasA and fengycin increased the level of   ROS inside seeds, but the highest response was elicited by fengycin (Fig. 3d). ROS production is a normal response after the first water uptake during seed imbibition and, in a certain range of concentrations, ROS has a well-recognized role in endosperm weakening, protection against pathogens, mobilization of seed reserves and interaction with plant hormones during the process of seed germination [35][36][37] . These results suggest that the increase in ROS levels after seed treatment with fengycin, but not TasA, may act as a beneficial stimulus for further plant development mediated by GSH accumulation.
Previous studies revealed that application of fengycin elicits plant defence responses in adult plants 7,38 . Therefore, we investigated whether treatment of seeds with fengycin immunized adult plants against aboveground pathogens by inoculating the third leaf of adult plants with a spore suspension of Botrytis cinerea. The size of necrotic lesions induced by the fungus in plants grown from fengycin-treated seeds was significantly smaller than that of control plants or plants grown from TasA-treated seeds (Fig. 5b). Transcriptomic analyses of infected leaves revealed a high basal level of transcripts, some of them related to plant defence mechanisms such as Allene oxide cyclase/synthase, directly implied in the biosynthesis of jasmonic acid 39 (Supplementary Table 5) in leaves of plants emerged from seeds treated with fengycin. Their expression was maintained or increased 48 h after inoculation with B. cinerea, while an increase was found in adult plants from non-treated seeds as a result of the response to the infection. Second, upregulation of the genes related to glutathione metabolism, specifically including glutathione S-transferases (GSTs), occurred 48 h after the challenge with B. cinerea ( Fig. 5c and Extended Data Fig. 9a). Participation of GSTs in antioxidant reactions, together with the important cel-lular antioxidant GSH, has been proven to mitigate oxidative stress caused by necrotrophic fungus B. cinerea in infected tissues 40 . Metabolomic data from the radicles of fengycin-treated seeds demonstrated an increase in the levels of certain molecules of the same cluster that included GSH (Fig. 3b), and adult plants grown from bacterized seeds accumulated certain molecules belonging to the GSH cluster, especially in the aerial region (Extended Data Fig. 9b). In addition, metabolomes of aerial parts of seedlings 5 d after treatment with fengycin also showed an increase in oxidized glutathione and glutathione-related metabolites in addition to defence-related compounds such as flavonoids (that is, mangiferin and 2-O-rhamnosilvitexin) (Fig. 5d and Extended Data Fig. 9c).

Discussion
In this study, we demonstrated that in addition to a structural role of bacterial ECM in biofilm formation, it is also essential for Bacillus colonization of melon seeds and growth promotion, including induction of anti-fungal resistance in the adult plant. Our findings indicate that the beneficial effects of B. subtilis treatment on melon seeds are associated with changes in the initial ROS increase during seed imbibition, and the contents and specific pools of metabolites released from storage tissues, which were mediated by at least two ECM components, fengycin and TasA.
We propose that specific interactions of TasA or fengycin with OBs defines two different physiological responses. First, a short-term effect on radicle growth, mediated by both components and related to changes in the dynamics of TAGs after differential targeting of oil bodies and lipid catabolism optimization. On one hand, the oleosin-TasA interaction determines the OB aggrupations around glyoxysomes and the subsequent accumulation of lysophospholipids acting as signal molecules for growth. On the other hand, the OB disaggregations caused by the action of fengycin over membranes determine a faster mobilization of their contents. Second, a long-term effect of growth promotion that persists in adult plants and is mediated by fengycin, through an initial increase in ROS that causes accumulation of GSH that is maintained over time and makes the plant more tolerant to biotic stresses (Fig. 6). Glycerophospholipids have both structural and signalling roles in plants, and this bifunctionality is in part due to continuous synthesis and turnover of endogenous pools 41 . Accumulating evidence suggests a role for lysophospholipid derivatives in signalling processes in plant cells. LysoPE treatment has been reported to delay fruit softening when used postharvest, mitigate defoliation effects of ethephon, and delay leaf and fruit senescence in tomato and potato 42,43 and is used commercially as a plant bioregulator to improve plant product quality 41,44 . On the other hand, glutathione is an essential metabolite with antioxidant properties, and is involved in cellular redox homoeostasis and many other physiological processes such as the development, growth and environmental response of plants 24 . The accumulation of glutathione-related metabolites in radicles and aerial parts of seedlings, and the higher expression of GSTs in leaves after seed treatment with fengycin, suggest that initial treatment causing an increase in ROS might act as a beneficial stimulus for plant development, and confer enhanced antioxidant capacity to mitigate an imbalance in the redox status in adult plants imposed by infection with B. cinerea.
We propose that only those seeds containing abundant OBs, and characterized by specific morphology in their primordial tissues, respond to the beneficial interaction with B. subtilis, and that this interaction is mediated at least in part by fengycin and TasA. According to this model, wheat or maize, which are monocotyledonous plants whose seeds are composed of starchy endosperm surrounded by a layer of living cells, did not react to the presence of fengycin, whereas cucumber or soybean seeds, which are anatomically different but do contain OBs, responded with a growth-promoting effect (Supplementary Table 6).
We note however that B. subtilis communities (cells and ECM) still retain stimulatory activity in these and other seeds plants, so it cannot be excluded that additional plant pathways or Bacillus-derived structural or chemical components, such as secondary metabolites or phytohormones, are specifically involved in maintaining these intricate interkingdom mutualistic interactions.

Quantitative analysis of bacterial growth and promotion of radicle growth.
Bacterial strains were grown in liquid LB at 30 °C overnight. The cells in the cultures were washed twice with sterile distilled water, and bacterial cell suspensions were adjusted to the same optical density (OD 600 ). When using a fluorescent inoculum, CellTracker CM-DiI dye (Invitrogen, C7000) was added to the bacterial suspension. Melon seeds (Rochet Panal -Fitó) were surface-sterilized with 0.1% sodium hypochlorite and bacterized with bacterial cell suspensions by bathing for 1 h at room temperature. The seeds were placed in Petri dishes with wet filter paper, covered with foil and maintained in growth chambers at 25 °C for 5 d. For quantitative analysis of the bacteria, the radicles were separated from the seeds. The radicles were placed in 1.5 ml microcentrifuge tubes with 1 ml PBS medium and 1 g of 3-mm-diameter glass beads, and attached bacteria were recovered by vortexing for 1 min. The seeds were ground using a mortar and a pestle in 1 ml PBS. Total c.f.u. were assayed by plating serial dilutions of resulting suspensions on solid LB medium. C.f.u. of the spores were assayed using the same suspension after thermal shock at 80 °C for 5 min. Seeds treated with water were used as controls.
Analysis of radicle growth-promoting effects was based on radicle areas measured 5 d after seed treatment using ImageJ software. The medium radicle area of the control treatment group was assumed as 100% growth and was used to normalize the percentage of growth-promoting effect corresponding to strain treatment groups. Treatments with compounds purified from the extracellular matrix were performed in a similar manner. BslA and TasA proteins were purified as described above while fengycin, surfactin and iturin were commercially acquired (Biosynth-Carbosynth AF35268, Sigma-Aldrich S3523 and I1774, respectively). Unless indicated, TasA was used at 3 µM and fengycin at 10 µM. Iturin was used for seed treatments at 260 µM, and lysophosphoethanolamine 16:0 (Avanti Polar Lipids, 856705) and reduced glutathione (Sigma-Aldrich G4251) were used at 0.2 mg ml −1 and 0.25 mM, respectively. For the evaluation of long-term growth promotion, seedlings from treated seeds were transferred to pots and maintained in a growth chamber at 25 °C.
Protein cloning and purification. TasA protein was purified as previously described 6 . BslA protein was purified from freshly transformed E. coli BL21-AI cells expressing the pDEST17 plasmid harbouring the bslA gene, which was initially subcloned into the donor vector PDONR207 using a Gateway recombination system. The primers designed to amplify the bslA gene were as follows: b sl AF w: 5 'G GG GA CA AG TT TG TA CA AA AA AG CA GG CT CA GC TG-GA ATCTACATCAACTAAAGC3'; bslARv: 5 'GGGGACCACTTTGTACAAGA AAGCTGGTTTAGTTGCAACCGCAAGGCT3' . The colonies of E. coli BL21-AI were picked and resuspended in 500 ml LB containing 100 µg ml −1 ampicillin and incubated at 37 °C in an orbital shaker to reach an OD 600 of 0.7. Then, the expression of recombinant protein was induced by 0.2% l-arabinose. After 4 h, the cells were collected and resuspended in buffer A (20 mM sodium phosphate, 500 mM NaCl and 20 mM imidazole) containing 400 µl 0.1 M phenylmethylsulfonyl fluoride, 40 µl lysozyme and 4 µl Cellytic 10x (Sigma-Aldrich). After a 45 min incubation with shaking at room temperature, the cells were disrupted by sonication (3 pulses of 1 min at an amplitude of 80%). The cell lysate was centrifuged at 12,000 g for 15 min at 20 °C, and the supernatant was filtered through a 0.45-µm-pore filter. The protein was purified using an AKTA Start fast protein liquid chromatography system (GE Healthcare). The soluble fraction was loaded into a 5 ml HisTrap HP column (GE Healthcare) previously equilibrated with buffer A. The protein was eluted from the column with elution buffer (20 mM sodium phosphate, 500 mM NaCl and 500 mM imidazole, pH 8). After affinity chromatography, the purified protein was loaded into a HiPrep 26/10 desalting column (GE Healthcare), and the buffer was exchanged for 20 mM Tris and 50 mM NaCl.
Oleosin protein was purified from inclusion bodies using freshly transformed E. coli Lemo21(DE3) cells expressing the pDEST17 plasmid harbouring the sequence of OLEOSIN1 from Arabidopsis thaliana, which was initially subcloned in the donor vector pENTR223 acquired from Arabidopsis Biological Resource Center using a Gateway recombination system. E. coli Lemo21(DE3) colonies were picked and resuspended in 500 ml LB containing 100 µg ml −1 ampicillin, 5 µg ml −1 chloramphenicol and 100 µM rhamnose, and incubated at 37 °C with orbital shaking to reach an OD 600 of 0.4. Then, the expression of the recombinant protein was induced by 400 mM isopropyl β-d-1-thiogalactopyranoside (IPTG). After 3 h, the cells were collected by centrifugation (5,000 g, 15 min, 4 °C), resuspended in buffer A (50 mM Tris and 150 mM NaCl, pH 8) and then centrifuged again. The pellets were stored at −80 °C until purification or processed after 15 min. After thawing, the cells were resuspended in buffer A, sonicated on ice (3 × 45 s, 60% amplitude) and centrifuged (15,000 g, 60 min, 4 °C). The supernatant was discarded because the protein was mainly expressed in inclusion bodies. The pellet was resuspended in buffer A supplemented with 2% Triton X-100, incubated at 37 °C with shaking for 20 min and centrifuged (15,000 g, 10 min, 4 °C). The pellet was extensively washed with buffer A, centrifuged (15,000 g, 10 min, 4 °C), resuspended in denaturing buffer (50 mM Tris, 500 mM NaCl and 6 M GuHCl) and incubated at 60 °C overnight for complete solubilization. The lysates were clarified by sonication on ice (3 × 45 s, 60% amplitude) and centrifugation (15,000 g, 1 h, 16 °C), and passed through a 0.45 µm filter before affinity chromatography. The protein was purified using an AKTA Start fast protein liquid chromatography system (GE Healthcare). Soluble inclusion bodies were loaded into a 5 ml HisTrap HP column (GE Healthcare) previously equilibrated with binding buffer (50 mM Tris, 0.5 M NaCl, 20 mM imidazole and 8 M urea, pH 8). The protein was eluted from the column with elution buffer (50 mM Tris, 0.5 M NaCl, 500 mM imidazole and 8 M urea, pH 8) and maintained under denaturing conditions. Apoplastic fluid extraction from the seeds. The apoplastic fluid (AF) of the seeds was collected by a standard technique allowing the recovery of the components present in the intercellular spaces based on vacuum infiltration and centrifugation 45 . Briefly, imbibed seeds were carefully decoated and immersed in infiltration buffer (50 mM Tris-HCl, pH 7.5, and 0.6% NaCl). Five vacuum pulses of 10 s (separated by 30 s intervals) were applied using a vacuum pump. Infiltrated seeds were recovered, dried on a filter paper, placed in plastic syringe barrels inside centrifuge tubes and centrifuged for 20 min at 400 g at 4 °C. The AF was recovered and stored at 4 °C until use at various concentration ratios (0, 1:2, 1:5, 1:10, 1:20 and 1:50) in TasA polymerization assays.
TasA polymerization assays. The kinetics of polymerization of TasA were monitored using thioflavin T (ThT) and detected by fluorescence 22 . The assay was carried out in a 96-well microtitre plate. Briefly, purified TasA protein was diluted in buffer (20 mM Tris and 50 mM NaCl, pH 7) or AF to final concentrations of 0.4 mg ml −1 or 0.2 mg ml −1 . ThT solution was added to the wells to a final ThT concentration of 20 μM in a final volume of 100 μl. The fluorescence signal was measured in a fluorescence microplate spectrophotometer reader (438 nm excitation and 495 nm emission, with a 475 nm cut-off) at 30 °C under shaking. Each protein concentration was assayed in triplicate.
Infection assays in the plants. Assays of B. cinerea infection were carried out in 5-6-week-old plants. Fungal conidia were collected from light-grown culture in sterile distilled water and filtered through a 40 µm cell strainer to remove remaining hyphae. For inoculation, the conidial suspension was adjusted to 10 5 conidia per ml in half-strength filtered (0.45 μm) grape juice (100% pure organic). Each leaf was inoculated with 5 μl droplets of conidial suspension. The pots were covered with a plastic dome and placed in a growth chamber. The leaves were imaged 72 h after the inoculation and the size of the lesions was determined using ImageJ software. The size of the lesions in experimental plants was normalized to a medium lesion size in the control plants, which was assumed as 100% disease percentage.
Purification of oil bodies from the seeds. Oil bodies were purified as previously described 46 , with some modifications according to ref. 47 . Briefly, the seeds were soaked in 0.1 M sodium bicarbonate at pH 9.5 (adjusted with 0.1 M NaOH) for 2 h and soaking medium was discarded. The soaked seeds were ground in the same medium with a mortar and a pestle. The slurry was transferred to 2 ml tubes and centrifuged at 10,000 g for 30 min at 4 °C. The upper layer was transferred to a clean tube, dispersed in washing solution (7 M urea, 1:4 w/w) and centrifuged (10,000 g, 30 min). To remove residue of washing solution, the fat pad was isolated and suspended in water (1:4 w/w) and centrifuged (10,000 g, 30 min). After centrifugation, the creamy layer was collected and stored at 4 °C. Analyses were completed within 24 h. Pulldown assay. To study possible TasA interactions with the proteins present in the seeds, we carried out a pulldown assay using a nickel-loaded affinity resin (Protino Ni-TED 2000, Macherey-Nagel); purified TasA was used as a bait protein for protein extract obtained from imbibed seeds. Imbibed and decoated seeds were ground to a fine powder using a mortar and a pestle in liquid nitrogen. The powder was transferred to a clean 15 ml tube and resuspended in extraction buffer (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 10 mM EDTA, 1% Triton X-100, 10% glycerol, 10 mM dithiothreitol (DTT), 1 mM phenylmethylsulfonyl fluoride, 1 mM NaF and 1% protease inhibitor cocktail). The samples were homogenized using a vortex, sonicated by 5 pulses of 20 s at 80% amplitude and centrifuged for 5 min at 9,000 g at 4 °C. The supernatant was transferred to a new tube and used immediately. Purified TasA was added to the seed protein extract to a final concentration of 50 µM and incubated at 4 °C overnight. Then, the mixture was incubated with equilibrated resin at 4 °C overnight. Control resin that did not contain bait protein was added to the seed protein extract. Flow-through, washing and elution fractions were collected and subjected to trichloroacetic acid protein precipitation. Precipitated proteins were resuspended in 1x Laemmli sample buffer (Bio-Rad) and heated at 100 °C for 5 min. The proteins were separated via SDS-PAGE through 12% polyacrylamide gels.
Mass spectrometry analysis of protein bands. The protein bands of interest were identified by high-performance liquid chromatography/electrospray ionization tandem mass spectrometry (HPLC-ESI-MS/MS). Briefly, the bands were cut out after electrophoresis, washed and destained. Subsequently, the disulfide bridges were reduced with DTT and cysteines were alkylated by iodoacetamide; in-gel trypsin digestion was performed to extract peptides from the protein samples. The entire process was carried out automatically using an automatic digester (DigestPro, Intavis Bioanalytical Instruments). The peptides were concentrated and desalted using a C18 ZORBAX 300SB-C18 capture column (Agilent Technologies; 5 × 0.3 mm, 5 µm particle diameter and 300 Å pore size) in a gradient of 98% H 2 O:2% acetonitrile (ACN)/0.1% formic acid (FA) at a flow rate of 20 μl min −1 for 6 min. The capture column was connected through a 6-port valve to a ZORBAX 300SB-C18 analytical column (Agilent Technologies; 150 × 0.075 mm, 3.5 µm particle diameter and 300 Å pore size). Elution of the samples from the capture column was performed by a gradient of 0.1% FA in water as a mobile phase A and 0.1% FA in 80% ACN/20% water as a mobile phase B. The LC system was coupled to a nanospray source (CaptiveSpray, Bruker Daltonics) and a three-dimensional (3D) ion trap mass spectrometer (amaZon speed ETD, Bruker Daltonics) operating in the positive mode at a capillary voltage of 1,500 V and a sweep range of m/z 300-1,500. Data-dependent acquisition was carried out in the automatic mode with sequential collection of the full scan MS spectra (m/z 300-1,400), followed by collection of MS spectra in tandem via collision-induced dissociation of the 8 most abundant ions. ProteinScape 3 software (Bruker Daltonics) coupled to Mascot 3.1 search engine (Matrix Science) was used for identification by matching the MS/MS data against the Cucurbit genomes databases 48 .
Far-western blotting. The interaction between TasA and oleosin was examined by far-western blotting 49 . Purified proteins were separated by SDS-PAGE through a 15% acrylamide gel and electrophoretically transferred onto PVDF membranes using a Trans-Blot Turbo transfer system (Bio-Rad) and PVDF transfer packs (Bio-Rad). The proteins were renatured by incubation of the membrane in denaturing and renaturing buffer (100 mM NaCl, 20 mM Tris, pH 7.6, 0.5 mM EDTA, 10% glycerol, 0.1% Tween 20, 2% skim milk powder and 1 mM DTT) containing decreasing concentrations of guanidine HCl (6, 3, 1, 0.1 and 0 M) and the membrane was blocked with blocking buffer (TBS containing 0.1% Tween 20 and 5% skim milk). Then, the membrane was incubated in protein-binding buffer (100 mM NaCl, 20 mM Tris, pH 7.6, 0.5 mM EDTA, 10% glycerol, 0.1% Tween 20, 3% skim milk powder and 1 mM DTT) containing 10 μg of purified TasA at 4 °C overnight. Excess protein was removed, and bound TasA was subjected to immunodetection. The membrane was probed with an anti-TasA antibody (rabbit) at a 1:20,000 dilution in blocking buffer. A secondary anti-rabbit IgG antibody conjugated to horseradish peroxidase (Bio-Rad) was used at a 1:3,000 dilution in the same buffer. The membranes were developed using Pierce ECL western blotting substrate (Thermo Fisher).
Gene expression analysis by RT-qPCR. Total RNA was extracted from the frozen tissues collected at various timepoints after the treatments by phenol/ chloroform extraction followed by precipitation with 1.5 M LiCl 50 and sodium acetate 51 . For quantitative PCR with reverse transcription (RT-qPCR) assays, the RNA concentration was adjusted to 100 ng µl −1 . Then, 1 µg of DNA-free total RNA was reverse transcribed into complementary DNA using SuperScript III reverse transcriptase (Invitrogen) and random primers in a final reaction volume of 20 µl according to the manufacturer's instructions. RT-qPCR was performed using an iCycler-iQ system and an iQ SYBR Green Supermix kit from Bio-Rad according to the manufacturer's instructions. The RT-qPCR cycle included the following steps: 95 °C for 3 min, followed by PCR amplification using a 40-cycle amplification programme (95 °C for 20 s, 56 °C for 30 s and 72 °C for 30 s), followed by a third step at 95 °C for 30 s. To evaluate the melting curve, 40 additional cycles were performed for 15 s starting at 75 °C, with stepwise temperature increases of 0.5 °C per cycle. The act7 gene was used as a reference for data normalization. The target genes were amplified using the primers listed in Supplementary Table 2. Primer efficiency tests and confirmation of the specificity of the amplification reactions were performed as previously described 52 . Relative transcript abundance was estimated using the ΔΔ cycle threshold (Ct) method 53 . Transcriptional data of the target genes were normalized to the act7 gene and are shown as the fold change in the expression levels of the target genes in each experimental treatment compared to those in the control treatment. The relative expression ratios were calculated as the differences between the qPCR threshold cycles (Ct) of the target gene and the Ct of the act7 gene (ΔCt = Ct gene of interest -Ct act7 ). The fold-change values were calculated as 2 −ΔΔCt , assuming that a single PCR cycle represents a two-fold difference in the template abundance 54,55 . RT-qPCR analyses were performed in technical triplicates using three independently isolated RNA samples (biological triplicates).

Total transcriptome analysis.
For RNA sequencing analysis, 100 bp single-end read libraries were prepared using a TruSeq stranded total RNA kit (Illumina). The libraries were sequenced using a NextSeq550 instrument (Illumina). Raw reads were preprocessed by NextSeq System Suite v.2.2.0. using specific NGS technology configuration parameters 56 . This preprocessing removes low-quality, ambiguous and low-complexity stretches, linkers, adapters, vector fragments and contaminated sequences, and preserves the longest informative parts of the reads. SeqTrimNext also discarded sequences shorter than 25 bp. Subsequently, clean reads of the BAM files were aligned and annotated by Bowtie2 57 using the Cucumis melo genome v.4.0 58 as the reference; these data were then sorted and indexed using SAMtools v.1.484110 59 . Uniquely localized reads were used to calculate the read number values for each gene by Sam2counts (https:// github.com/vsbuffalo/sam2counts). Differentially expressed genes (DEGs) in the treatment samples were analysed by DEgenes Hunter 60 , which provides a combined P value calculated on the basis of the Fisher method 61 using nominal P values provided by edgeR 62 and DEseq2 63 . This combined P value was adjusted by the Benjamini-Hochberg procedure 48 (false discovery rate approach) and used to rank all obtained DEGs. For each gene, a combined P value < 0.05 and log 2 fold change (FC) >1 or <−1 were considered significance thresholds. A heatmap and DEG clustering were generated using ComplexHeatmap in R Studio and Kobas 2.0 64 . Only profiles with a P value < 0.05 were considered in the present study. DEGs annotated using the Cucumis melo genome were processed to identify the Gene Ontology functional categories using sma3s 65 and TopGo software 66 . Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways and enrichment were estimated using Bioconductor packages (Bioconductor.org) GGplot2, ClusterProfiler, DOSE and EnrichPlot in R Studio. The data were deposited in the GEO database under the reference GSE175611.
CLSM. Bacteria, ROS and OBs inside the seeds were visualized by CLSM. Seeds were collected at the specified hours after treatment and were transversally cut with a scalpel. A drop of glycerol was applied to the sections, which were placed into 1.5-mm-thick cover glasses (22 × 22 mm). The bacteria were stained during inoculum preparation by CellTracker CM-DiI dye or in situ by Hoechst solution. Oil bodies and ROS were stained with Nile red (1:1,000 v/v) or dihydrorhodamine 123 (1:500 v/v), respectively. The images were acquired using a Leica SP5 confocal microscope with a ×HCX IRAPO L 25.0×0.95 WATER objective. To image the purified OB suspension, a drop of Nile red and Fast Green FCF-stained preparation was applied onto a patch of polymerized 1% agarose on a glass slide, which was covered with a coverslip. The images were acquired using a Leica SP5 confocal microscope with HCX PL APO lambda blue 63.0×1.40 OIL UV objective. Image processing was performed using Leica Application Suite Advance Fluorescence v.2.7.3.9723 (LCS Lite, Leica Microsystems) and FIJI/ImageJ software. Laser settings, scan speed, photomultiplier detector gain and pinhole aperture were constant for all acquired image stacks for each experiment.
TEM. For TEM analysis, 16 h after treatment, seeds were fixed directly using 2% paraformaldehyde, 2.5% glutaraldehyde and 0.2 M sucrose mixture in 0.1 M phosphate buffer (PB) overnight at 4 °C. After three washes in PB, the portions were excised from the micropylar endosperm region and postfixed in 1% osmium tetroxide solution in PB for 90 min at room temperature, followed by PB washes and 15 min of stepwise dehydration in an ethanol series (30%, 50%, 70%, 90% and 100% twice). Between the 50% and 70% steps, the samples were incubated in-bloc in 2% uranyl acetate solution in 50% ethanol at 4 °C overnight. After dehydration, the samples were gradually embedded in low-viscosity Spurr's resin (resin:ethanol, 1:1, 4 h; resin:ethanol, 3:1, 4 h; and pure resin overnight). The sample blocks were embedded in capsule moulds containing pure resin for 72 h at 70 °C. To image purified OBs in suspension or TasA protein under various polymerization conditions, carbon-coated copper grids were deposited over the sample drops and incubated at room temperature for 2 h. After incubation, the grids were washed in phosphate-buffered saline for 5 min, negatively stained with uranyl acetate (1%) for 20 s and washed once with water for 30 s.
For immunolabelling assays, carbon-coated copper grids were deposited over the samples of the corresponding OB treatments. After 2 h of incubation, the grids were washed in phosphate-buffered saline for 5 min and blocked with Pierce protein-free (TBS) blocking buffer (Thermo Fisher) for 30 min. An anti-TasA primary antibody was used at a 1:150 dilution in blocking buffer, and grids were deposited over the drops of antibody solution and incubated for 1 h at room temperature. The samples were washed three times with TBS-T (50 mM Tris-HCl, 150 mM NaCl, pH 7.5, and 0.1% Tween 20) for 5 min and then exposed to a 10-nm-diameter immunogold-conjugated secondary antibody (20 nm goat anti-rabbit conjugate, BBI solutions) for 1 h at a 1:50 dilution. The samples were then washed twice with TBS-T and once with water for 5 min each time. Finally, the grids were treated with 2% glutaraldehyde for 10 min, washed in water for 5 min, negatively stained with 1% uranyl acetate for 20 s and washed once with water for 30 s. The samples were left to dry and imaged under an FEI Tecnai G2 20 TWIN transmission electron microscope at an accelerating voltage of 80 kV. The images were acquired using TIA FEI Imaging Software v.4.14.
Metabolite extraction from plant tissues. Metabolites from the samples of adult melon plants or radicles (MSV000084674, MSV000084278 and MSV000086360) were extracted by adding 1 ml methanol and vigorous vortexing. At least 2 or 3 replicates were done for each experiment, except for adult melon plants where only 1 replicate was done due to the high number of sections analysed for each plant. Due to the large sample number, weighting and normalization by biomass was not feasible. However, the size of plant sections was kept constant. To enable relative quantitative comparison between samples, we normalized peak areas to total ion count, which is a robust and widely applied strategy in non-targeted metabolomics. In addition, sulfamethazine (1 µM) was added to all samples as an internal standard and was used for normalization. After incubation for 2 h, the extracts were centrifuged at 14,000 g for 30 min. The acquired supernatants were stored at −80 °C until use in liquid chromatography-tandem mass spectrometry (LC-MS/MS).
Metabolites from seeds and seedlings (MSV000088139) were extracted after two rounds of methanol incubation with the powder obtained after grinding the tissue in liquid nitrogen and freeze-drying the samples. Three replicates were done for each condition. The samples were transferred to weighted vials and the methanol was evaporated at room temperature. Extracts were then resuspended in 1 ml methanol and centrifuged at 14,000 g for 20 min at 4 °C. Vials were weighed after the extractions to adjust the final concentrations to 1 mg ml −1 , thus no internal standard was used.
In both cases, extraction controls were performed by adding 3 samples with pure methanol used for the extraction as a blank. Differences in abundance of metabolites were estimated on the basis of relative comparisons between equally treated samples due to the absence of authentic standards in the non-targeted metabolomics approach.
For the Q-Exactive instrument, 5 µl of the samples were injected for separation on a UHPLC with a C18 core-shell column (Kinetex, 50 × 2 mm, 1.8 μm particle size, 100 Å pore size, Phenomenex). For the mobile phase, we used a flow rate of 0.5 ml min −1 (solvent A: H 2 O + 0.1% FA; solvent B: acetonitrile + 0.1% FA). During the chromatographic separation, we applied a linear gradient from 0-0.5 min 5% B, 0.5-4 min 5-50% B, 4-5 min 50-99% B, followed by a 2 min washout phase at 99% B and a 2 min re-equilibration phase at 5% B. For positive mode MS/MS acquisition, the electrospray ionization was set to a 35 a.u. sheath gas flow, 10 a.u. auxiliary gas flow, 2 a.u. sweep gas flow and 400 °C auxiliary gas temperature. The spray voltage was set to 3.5 kV with an inlet capillary of 250 °C. The S-lens voltage was set to 50 V. MS/MS product ion spectra were acquired in data-dependent acquisition mode. MS1 survey scans (150-1,500 m/z) and up to 5 MS/MS scans per data-dependent acquisition duty cycle were measured with a resolution of 17,500. The C-trap fill time was set to a maximum of 100 ms or until the Automatic Gain Control target of 5 × 10 5 ions was reached. The quadrupole precursor selection width was set to 1 m/z. Normalized collision energy was applied stepwise at 20, 30 and 40%, with z = 1 as default charge state. MS/MS scans were triggered with apex mode within 2-15 s from their first occurrence in a survey scan. Dynamic precursor exclusion was set to 5 s. Precursor ions with unassigned charge states and isotope peaks were excluded from MS/MS acquisition.
For the QTOF instrument, 2 μl were injected into a Waters Acquity Peptide BEH C18 column (1.7 μm, 300 Å, 2.1 × 100 mm). The column was maintained at 40 °C and run at a flow rate of 0.5 ml min −1 , using 0.1% formic acid in H 2 O as solvent A and 0.1% formic acid in acetonitrile as solvent B. A gradient was employed for chromatographic separation starting at 5% B for 1 min, then 5% to 85% B for 9 min, 85% to 100% B for 1 min, and finally held at 100% B for 4 min. The column was re-equilibrated to 5% B for 3 min before the next run was started. The LC flow was switched to the waste for the first 0.5 min, then to the MS for 13.5 min, then back to the waste to the end of the run. The photodiode array detector acquisition was performed in the range 200-400 nm at 4.2 Hz, with 1.2 nm slit width. The flow cell was maintained at 40 °C. The MS system was tuned using standard NaI solution (Shimadzu). The same solution was used to calibrate the system before starting. System suitability was checked by including a standard sample made of 5 μg ml −1 thiostrepton, which was analysed regularly in between batches of samples.
All samples were analysed in positive polarity, using data-dependent acquisition mode. In this regard, full scan MS spectra (m/z 400-4,000, scan rate 20 Hz) were followed by 3 data-dependent MS/MS spectra (m/z 400-4,000, scan rate 20 Hz) for the 3 most intense ions per scan. The ions were selected when they reached an intensity threshold of 1,000, isolated at the tuning file Q1 resolution, fragmented using collision-induced dissociation with collision energy ramp (CE 10-40 eV) and excluded for 0.05 s (1 MS scan) before being reselected for fragmentation. The parameters used for the ESI source were: interface voltage 4 kV, interface temperature 300 °C, nebulizing gas flow 3 l min −1 and drying gas flow 10 l min −1 .
In both cases, a mixture of 10 mg ml −1 each of sulfamethazine, sulfamethizole, sulfachloropyridazine, sulfadimethoxine, amitriptyline and coumarin was run after every 96 injections for quality control.
Preprocessing for data analysis and MS/MS network analysis. Raw spectra were converted to .mzXML files using MSconvert (ProteoWizard). Feature-based molecular networking and spectral library search. Molecular networking and library search were performed using the GNPS environment with the settings listed in Supplementary Table 4 and visualized using Cytoscape software 72 . Putative annotation of detected features was obtained using automatic library search through the GNPS environment 73 , network annotation propagation 74 , chemical classification using ClassyFire 75 and MolNetEnahancer workflows 76 .
Mirror plots were done using GNPS and https://metabolomics-usi.ucsd.edu/ 77 , comparing mzspec of the selected features and the metabolites recorded in MS/MS databases (Extended Data Fig. 10). Annotations were done according to guidelines in ref. 78 (Supplementary Table 8 Proportionality score. The proportionality score was calculated between two directly connected nodes across the entire molecular network using the following equation: where Ns1 and Ms1 correspond to the peak area of the detected features N and M in sample S1, while Ns2 and Statistics and reproducibility. Statistical significance was assessed by appropriate tests (see figure legends). All analyses were performed using GraphPad Prism version 6. P values < 0.05 were considered significant. Asterisks indicate the level of statistical significance: *P < 0.05, **P < 0.01, ***P < 0.001 and ****P < 0.0001. Statistical analysis of metabolomic datasets was performed by Metaboanalyst v.5.0 (ref. 69 ), after data filtering by interquartile range (IQR). When needed, data were normalized to the feature corresponding to the internal standard of the samples. Experiments were repeated at least three times independently with similar results.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The RNA-seq data are deposited in the GEO database under the reference GSE175611. Metabolomics data are deposited at https://massive.ucsd.edu/ with the identifiers MSV000084674 (data from radicles from seeds bacterized with Bacillus subtilis subsp. subtilis NCBI 3610 and mutant strains), MSV000084278 (data from radicles from seeds treated with fengycin and TasA), MSV000086360 (data from adult plants grown from control seeds or seeds bacterized with Bacillus subtilis subsp. subtilis NCBI 3610) and MSV000088139 (data from seeds and seedlings after seed treatments with bacteria, fengycin or TasA). Source data are provided with this paper. Fig. 1 | Seed treatment with B. subtilis cells does not induce an increase in the germination rate but accelerates post germination progression. a Germination rates of untreated and bacterized seeds represented as the percentage of germinated seeds versus time. b Relative expression levels of the GA20ox1 and CyP707A1 genes in bacterized seeds 0 and 16 hours after the treatment compared with those in the untreated seeds. The average values of two biological replicates with three technical replicates are shown with error bars representing SD. c Heatmap of fold changes of selected DEGs identified by total transcriptomic analysis of bacterized and control seeds (untreated) 16 hours after the treatment. Colour scale indicates the Log2(FC) of each DEG. d KEGG pathway enrichment analysis of all DEGs identified by RNA-seq of bacterized and control seeds 16 hours after the treatment.

NATUrE MICroBIoloGy
Extended Data Fig. 2 | Carboxylic acids and lipids represent the major metabolic changes in bacterized seeds. a Top 20 features with high median weighted sum of absolute regression coefficient scores determined by PLS-DA and calculated using MetaboAnalyst, selected as features discriminating bacterized seeds from control seeds 0 hours after the treatment (left) or 24 hours after the treatment (right). Feature IDs are accompanied by chemical class according to Classyfire classification. Features in bold are selected as features discriminating any growth-promoting treatment from the control treatment. b Volcano plots showing the metabolites significatively increased or decreased in leaves (left) and stem (right) of plants emerged from bacterized seeds respecting control plants. Colours indicate the chemical class of each metabolite according to Classyfire. c Distribution of features identified by spectral match as L-tryptophan and cinnamic acid in 3D models representing the plants grown from the control or B. subtilis-treated seeds.

NATUrE MICroBIoloGy
Extended Data Fig. 6 | treatments of the seeds with tasA and fengycin influence the final metabolic patterns of emerged radicles. a PCA 3D score plot of the metabolome of the radicles showing clustering of the samples based on seed treatment. The percentage of variation explained by each principal component is indicated on the axes. b Heatmap of hierarchical clustering results of the top 50 features impacted in the radicles grown from TasA-and fengycin-treated seeds. The color code inside the heatmap depicts the relative fold change of each metabolite between groups. c Top 20 features with high median weighted sum of absolute regression coefficient scores determined by PLS-DA and calculated using MetaboAnalyst. Feature IDs are accompanied by chemical class according to Classyfire classification. Features in bold are selected as features discriminating any growth-promoting treatment from the control treatment. d, e Molecular family of features discriminating fengycin-treated seeds (D) or TasA-treated seeds (e) from the control seeds according to PLS-DA. The chemical structures of annotated features and their average mass based on spectral matches to GNPS libraries are also represented for the corresponding molecular families. Pie charts indicate the peak abundance of each metabolite in the corresponding condition. Node shape indicates the level of identification according to Sumner et al., 78 .
1 nature portfolio | reporting summary

March 2021
Corresponding author(s): Diego Romero Last updated by author(s): Apr 20, 2022 Reporting Summary Nature Portfolio wishes to improve the reproducibility of the work that we publish. This form provides structure for consistency and transparency in reporting. For further information on Nature Portfolio policies, see our Editorial Policies and the Editorial Policy Checklist.

Statistics
For all statistical analyses, confirm that the following items are present in the figure legend, table legend, main text, or Methods section.
n/a Confirmed The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement A statement on whether measurements were taken from distinct samples or whether the same sample was measured repeatedly The statistical test(s) used AND whether they are one-or two-sided Only common tests should be described solely by name; describe more complex techniques in the Methods section.
A description of all covariates tested A description of any assumptions or corrections, such as tests of normality and adjustment for multiple comparisons A full description of the statistical parameters including central tendency (e.g. means) or other basic estimates (e.g. regression coefficient) AND variation (e.g. standard deviation) or associated estimates of uncertainty (e.g. confidence intervals) For null hypothesis testing, the test statistic (e.g. F, t, r) with confidence intervals, effect sizes, degrees of freedom and P value noted

Software and code
Policy information about availability of computer code Data collection RNA-seq data were collected using NextSeq System Suite v2.2.0.
Metabolomics data were obtained with the commercial software of Thermo Scientific for UPLC systems. HPLC-ESI-MS/MS data corresponding to the identification of proteins in gel cut bands was acquired using ProteinScape 3 software from Bruker coupled to Mascot v3.1 (Matrix Science).

Data analysis
For statistics and representation of all experimental data, analysis were done using GraphPad Prism v.6.01.
Raw reads were preprocessed by NextSeq System Suite v2.2.0. using specific NGS technology configuration parameters. Clean reads of the BAM files were aligned and annotated using the Cucumis melo genome (v4.0) as the reference by Bowtie2; these data were then sorted and indexed using SAMtools v1.48411074. Uniquely localized reads were used to calculate the read number values for each gene by Sam2counts (https://github.com/vsbuffalo/sam2counts). Differentially expressed genes (DEGs) in the treatment samples were analyzed by DEgenes Hunter, which provides a combined p value calculated based on the Fisher method using nominal p values provided by edgeRand DEseq2. This combined p value was adjusted by the Benjamini-Hochberg (BH) procedure (false discovery rate approach) and used to rank all obtained differentially expressed genes. A heatmap and DEG clustering were generated using ComplexHeatmap in R Studio and Kobas 2.0. DEGs annotated using the Cucumis melo genome were processed to identify the Gene Ontology functional categories using sma3s and TopGo software. KEGG pathways and enrichment were estimated using Bioconductor packages (Bioconductor.org) GGplot2, ClusterProfiler, DOSE, and EnrichPlot in R Studio.
For LC-MS/MS data analysis, raw spectra were converted to .mzXML files using MSconvert (ProteoWizard). MS1 and MS/MS feature extraction was performed with Mzmine2.30. For molecular networking and spectrum library matching the .mgf file was uploaded to GNPS