Microbiome manipulation by a soil-borne fungal plant pathogen using effector proteins

During colonization of their hosts, pathogens secrete effector proteins to promote disease development through various mechanisms. Increasing evidence shows that the host microbiome plays a crucial role in health, and that hosts actively shape their microbiomes to suppress disease. We proposed that pathogens evolved to manipulate host microbiomes to their advantage in turn. Here, we show that the previously identified virulence effector VdAve1, secreted by the fungal plant pathogen Verticillium dahliae, displays antimicrobial activity and facilitates colonization of tomato and cotton through the manipulation of their microbiomes by suppressing antagonistic bacteria. Moreover, we show that VdAve1, and also the newly identified antimicrobial effector VdAMP2, are exploited for microbiome manipulation in the soil environment, where the fungus resides in absence of a host. In conclusion, we demonstrate that a fungal plant pathogen uses effector proteins to modulate microbiome compositions inside and outside the host, and propose that pathogen effector catalogues represent an untapped resource for new antibiotics. A secreted protein effector from the fungal pathogen Verticillium dahliae has bactericidal properties. It allows the pathogen to modify the root microbiome in tomato and cotton, specifically eliminating plant-protective bacteria, to increase its own virulence.

T o establish disease, pathogenic microbes secrete a wide diversity of effector proteins that facilitate host colonization through a multitude of mechanisms 1 . Typically, pathogen effectors are defined as small cysteine-rich proteins that are secreted during colonization to manipulate host physiology or to deregulate host immune responses 2 . Consequently, effector proteins are predominantly studied in binary host-microbe interactions, while largely ignoring the biotic context in which these interactions take place. Higher organisms, including plants, associate with a plethora of microbes that collectively form their microbiome, which represents a key determinant for their health [3][4][5][6][7] . The most extensive microbial colonization of plants occurs at roots, where plants define rhizosphere microbiome compositions through secretion of exudates 8,9 and specifically attract beneficial microbes to suppress pathogen invasion [10][11][12] . Thus, we considered that plant pathogens evolved mechanisms to counteract this recruitment and modulate host microbiomes for successful infection, possibly through effector proteins 1,13 .
V. dahliae is a soil-borne fungus that causes vascular wilt disease on hundreds of plant species, including numerous crops 14,15 . V. dahliae survives in the soil through persistent resting structures called microsclerotia that germinate in response to nutrient-rich exudates released by nearby plant roots 16 . Subsequently, emerging hyphae grow through the soil and rhizosphere towards the roots where the fungus penetrates its hosts. Following root penetration, V. dahliae invades the xylem where it produces conidiospores that are spread throughout the vasculature by the sap stream. This systemic colonization causes chlorosis and necrosis of plant tissues, which is followed by plant senescence. V. dahliae then enters a saprophytic phase, emerges from the vasculature and colonizes the dead plant material where it produces new microsclerotia that are eventually released into the soil after tissue decomposition.
Using comparative population genomics, we previously identified the V. dahliae-secreted small cysteine-rich effector protein Ave1 that is recognized as an avirulence determinant by tomato plants that carry the corresponding Ve1 immune receptor 17 . However, on host plants lacking Ve1, VdAve1 acts as a virulence effector that promotes fungal colonization and disease development 17 . VdAve1 is homologous to plant natriuretic peptides (PNPs) that have been identified in numerous plant species, suggesting that VdAve1 was acquired from plants through horizontal gene transfer 17 . Whereas several of the plant PNPs were shown to act in plant homeostasis and (a)biotic stress responses 18,19 , the mode of action of VdAve1 to contribute to fungal virulence has remained unknown.

Results
Unlike most pathogen effector genes characterized so far, the V. dahliae effector gene Ave1 is not only highly expressed during host colonization 17,20 , but also during growth in vitro and under conditions mimicking soil colonization, suggesting a ubiquitous role throughout the fungal life cycle including life stages outside the host, and thus a role that does not primarily involve targeting host plant physiology (Extended Data Fig. 1). Our attempts to purify VdAve1 after heterologous expression in Escherichia coli, to facilitate functional characterization, repeatedly failed due to the formation of inclusion bodies (Extended Data Fig. 2a). The inability to obtain soluble protein using heterologous microbial expression systems can be attributed to a multitude of reasons, but is a well-known phenomenon when expressing AMPs 21 . Consequently, on the basis of ubiquitous expression of VdAve1 by V. dahliae, and our inability to purify soluble VdAve1 following expression in E. coli, we reasoned that VdAve1 may possess antimicrobial activity.
To obtain functional VdAve1, inclusion bodies were isolated from E. coli cells and denatured using guanidine hydrochloride. Next, VdAve1 was refolded by step-wise dialysis and functionality was confirmed through testing recognition by its immune receptor Ve1 (Extended Data Fig. 2b). To assess the potential antimicrobial activity of VdAve1, we developed an in vitro system in which we incubated a panel of plant-associated bacteria in tomato xylem fluid, to mimic a natural environment in which VdAve1 is secreted, namely tomato xylem vessels, and monitored their growth in presence and absence of the protein. VdAve1 selectively inhibited the growth of plant-associated bacteria (Fig. 1a). Whereas growth of all Gram-positive bacteria tested, namely Arthrobacter sp., Bacillus subtilis, Staphylococcus xylosus and Streptomyces sp., was strongly inhibited, Gram-negative bacteria displayed differential sensitivity to the protein. This differential sensitivity is not immediately explained by phylogenetic relationships of the tested isolates as even within bacterial orders/families differences are observed. For instance, whereas growth of the burkholderiales species Acidovorax is inhibited by VdAve1, growth of a Ralstonia isolate, which belongs the same order, is not. Similarly, treatment of two closely related rhizobiales, Rhizobium sp. and Agrobacterium tumefaciens, revealed differential sensitivity as VdAve1 affected growth of Rhizobium sp., but not of A. tumefaciens. Finally, growth of Pseudomonas corrugata and Serratia sp. was only slightly altered and unaffected, respectively, while growth of both Sphingobacterium sp. and Sphingomonas mali was affected on exposure to VdAve1. Treatment of a panel of fungal species with VdAve1 revealed no antifungal activity of the effector, whereas treatment of plant protoplasts could not reveal phytotoxic activity, suggesting that VdAve1 exclusively acts on bacteria (Extended Data Fig. 3a,b). Our initial observations with divergent, randomly chosen, plant-associated bacteria prompted us to further characterize the antimicrobial activity of VdAve1.
As a first step in the further characterization of the antimicrobial activity of VdAve1, we aimed to determine whether the effector protein is bacteriostatic or bactericidal by making use of electron microscopy to visualize the effect of protein treatment on bacteria. As a target species, the Gram-positive B. subtilis was chosen, considering its high sensitivity to VdAve1 treatment. By testing a concentration series of the VdAve1 effector protein, the minimum inhibitory concentration (MIC) was determined at 8 µM (Extended Data Fig. 3c). However, electron microscopy analysis revealed that sub-MIC concentrations of VdAve1 already induced blebbing and swelling of bacterial cells, followed by lysis and collapse, indicating that the antimicrobial activity of VdAve1 might be based on bactericidal activity (Fig. 1b).
To investigate whether the antimicrobial activity that is displayed by VdAve1 is more widely conserved among its homologues, we tested the only homologue that occurs in one of the sister species of the Verticillium genus, namely VnAve1, from the non-pathogenic species V. nubilum that displays 90% amino acid identity (Extended Data Fig. 3d). Also, this homologue displays antimicrobial activity, but it only inhibits a subset of the bacteria affected by VdAve1, and does not cause B. subtilis lysis (Fig. 1). Thus, the 13 amino acid polymorphisms between the two Ave1 homologues are responsible for differences in the activity spectrum. To investigate whether the antimicrobial activity also occurs among plant homologues, or is confined to microbial homologues and involves neofunctionalization after horizontal transfer, the more distant homologue AtPNP-A from Arabidopsis thaliana was tested as well. AtPNP-A completely arrests B. subtilis growth (Extended Data Fig. 3d,e). Collectively, these findings demonstrate that various Ave1 homologues possess antimicrobial activity, yet with divergent activity spectra, and indicate that the antimicrobial activity of VdAve1 did not result from neofunctionalization following horizontal gene transfer.
On the basis of the strong but selective bactericidal activity of VdAve1 in vitro, we proposed that V. dahliae exploits its effector protein to affect host microbiome compositions through the suppression of other microbes. Therefore, to determine the biological relevance of the observed bactericidal activity, we performed bacterial community analysis on the basis of the 16S ribosomal DNA profiling of tomato and cotton root microbiomes following infection with wild-type V. dahliae or a VdAve1 deletion mutant. Root microbiome compositions were determined during early V. dahliae infection stages, namely at 10 d post inoculation (dpi) when the fungus has just entered xylem vessels and initiated systemic spreading, to minimize indirect shifts in microbial compositions that result from severe disease symptomatology, rather than from direct shifts due to the presence of the effector protein. We did not observe major shifts in overall composition of bacterial phyla (Extended Data Fig. 4a) or total microbial diversity (α-diversity) (Extended Data Fig. 4b) on V. dahliae colonization of tomato and cotton. However, principal coordinate analysis based on Bray-Curtis dissimilarities (β-diversity) revealed a clear separation of root microbiomes ( Fig. 2a) (permutational multivariate analysis of variance (PERMANOVA), P < 0.01 for both tomato and cotton). The extent of V. dahliae colonization does not seem to determine the separation, as clustering of V. dahliae genotypes occurs in cotton although VdAve1 deletion hardly affects fungal virulence on this host plant (Fig. 2a). Thus, as anticipated on the basis of the potent, yet selective, antimicrobial activity, VdAve1 secretion by V. dahliae sophistically alters root microbiome compositions. However, despite the relatively small sample size of our 16S rDNA profiling, pairwise bacterial order comparisons on colonization by wild-type V. dahliae and the VdAve1 deletion mutant revealed differential abundances of Sphingomonadales, Bdellovibrionales and Ktedonobacterales for tomato ( Fig. 2b) (Supplementary Table 1). The finding that Sphingomonadales are repressed in the presence of VdAve1 indicates that this taxon is the most sensitive to VdAve1 activity. A similar comparison for cotton did not immediately reveal any differentially abundant orders, but agglomeration of amplicon sequence variants (ASVs) based on phylogenetic relatedness (patristic distance of <0.1) revealed eight differentially abundant taxa, including a taxon of the Sphingomonadaceae family ( Fig. 2b) (Supplementary  Table 2). Although this taxon only represents a small proportion of all Sphingomonadaceae in the cotton root microbiomes, it is exclusively and consistently found in the microbiomes of roots infected by the VdAve1 deletion mutant, and completely absent on infection with wild-type V. dahliae, again pointing towards the particular sensitivity of this taxon to VdAve1. Moreover, pairwise comparisons following the combination of tomato and cotton samples based on infection by the different V. dahliae genotypes, to identify differentially abundant bacterial orders that potentially remained unnoticed due to the limited sample size, again only revealed differential abundance of Sphingomonadales (P < 0.01, Extended Data Fig. 4c) (Supplementary Table 3). Given the fact that secretion of VdAve1 by V. dahliae during colonization of both tomato and cotton leads to a reduction of Sphingomonadales in the corresponding root microbiomes, we anticipated a broad efficacy of VdAve1 on bacteria within this order. Thus, to identify Sphingomonadales genera that are most sensitive to VdAve1, we identified ASVs with increased average relative abundance in the microbiomes with the VdAve1 deletion mutant when compared with wild-type V. dahliae, revealing Sphingomonas, Novosphingobium, Sphingopyxis and Sphingobium that are commonly referred to as Sphingomonads (Fig. 2c) 22,23 .
To test whether VdAve1 is indeed able to directly manipulate microbial communities, and to confirm that root microbiome compositional changes are not due to indirect effects through host manipulation, we treated a synthetic community comprising plant-associated bacteria with purified VdAve1 and determined changes in the microbial communities using 16S ribosomal DNA  profiling. As anticipated, VdAve1 clearly affected the composition of the microbial communities ( Fig. 2d and Extended Data Fig. 5). Pairwise differential abundance analyses of bacterial orders again revealed a significant repression of Sphingomonadales in the presence of VdAve1 (adjusted P < 0.05) ( Fig. 2e and Supplementary  Table 4). Collectively, these findings confirm the ability of VdAve1 to directly manipulate microbiome compositions, and further substantiate the previously observed impact on Sphingomonadales in planta as a direct consequence of VdAve1 activity. Transcriptional analysis revealed that VdAve1 is not only highly expressed in planta, but also during growth conditions mimicking soil colonization, suggesting that the antimicrobial activity of VdAve1 also facilitates V. dahliae niche colonization outside the host. Indeed, V. dahliae colonization assays performed in soil, or in Murashige and Skoog (MS) medium supplemented with soil, demonstrated that VdAve1 contributes to V. dahliae fitness (that is, biomass accumulation) in the presence of soil microbiota (Fig. 2f). No such contribution was detected in the absence of soil microbes, indicating that niche colonization promoted by VdAve1 is based on its antimicrobial activity. Accordingly, as inferred from principal coordinate analysis based on Bray-Curtis dissimilarities of the microbiomes in the MS medium, secretion of VdAve1 indeed affected community structures, an effect that could be detected less clearly in soil, arguably due to much stronger dilution effects ( To confirm the anticipated sensitivity, a panel of plant-associated Sphingomonads was incubated with VdAve1 in vitro 24,25 . In accordance with the previously observed effect on S. mali (Fig. 1a), treatment with VdAve1 was found to also inhibit growth of Sphingobium, Novosphingobium, Sphingopyxis and two other Sphingomonas species (Fig. 3a), indicating a broad sensitivity among the Sphingomonads. Given the selective efficacy of VdAve1 and the strong effect on Sphingomonads in the tomato and cotton microbiomes, we considered that these bacteria may act as antagonists and negatively affect V. dahliae growth in the absence of VdAve1. Indeed, cocultivation of V. dahliae with Novosphingobium sp. A, as well as with S. macrogoltabida, resulted in reduced fungal biomass of the VdAve1 deletion mutant when compared with the V. dahliae wild-type that secretes VdAve1 under these conditions, revealing that Sphingomonads comprise antagonists of V. dahliae, and explaining the importance of their inhibition by VdAve1 (Fig. 3b). Cocultivation with VdAve1-insensitive Agrobacterium, Pseudomonas and Ralstonia isolates did not affect biomass accumulation of the VdAve1 deletion mutant when compared with the wild-type, indicating that the detected differences in the presence of the Sphingomonads are indeed VdAve1-dependent (Extended Data Fig. 6a). Accordingly, pretreatment of surface-sterilized tomato seeds with Ralstonia sp. did not affect Verticillium wilt disease development. In contrast, and in line with previously described observations of plant-protective activities of Sphingomonad strains 24 , pretreatment of surface-sterilized tomato seeds with S. macrogoltabida negatively affected Verticillium wilt disease development as confirmed through biomass quantification of wild-type V. dahliae in the presence and the absence of the bacterium (Fig. 3c,d and Extended Data Fig. 7). Quantification of S. macrogoltabida in the presence of wild-type V. dahliae and the VdAve1 deletion mutant using 16S rDNA profiling and real-time PCR revealed that VdAve1 secretion significantly affects S. macrogoltabida proliferation to counter its protective effect ( Fig. 3e-g). Notably, this observation is not an indirect effect of differential host colonization by wild-type V. dahliae and the VdAve1 deletion mutant, as selection of tomato plants with equal levels of V. dahliae biomass (Fig. 3d, data points highlighted in red), reveals similarly impaired S. macrogoltabida proliferation in the presence of VdAve1 (Fig. 3g). Thus, these data underpin the hypothesis that V. dahliae secretes the VdAve1 effector to target antagonistic bacteria, including Sphingomonadales, during host colonization, although it needs to be acknowledged that ideally our hypothesis would be tested in experiments based on V. dahliae inoculation of germ-free tomato plants, a system we have not managed to establish thus far.
Our observation that V. dahliae secretes VdAve1 to suppress microbial competitors in the microbiomes of its hosts, prompted us to speculate about additional V. dahliae effector proteins involved in microbiome manipulation. Therefore, to query for the occurrence of additional effectors that aid in microbial competition, the predicted secretome of V. dahliae strain JR2 (ref. 26 ) was probed for structural homologues of known antimicrobial proteins (AMPs), revealing ten candidates (Supplementary Table 5). Most identified effectors share typical characteristics with canonical host-targeting effector proteins, such as being small and rich in cysteines. However, on the basis of previously performed RNA sequencing experiments, no expression of any of these candidates could be monitored during colonization of A. thaliana, Nicotiana benthamiana or cotton plants (Extended Data Fig. 8) 17,20,27,28 . Additionally, in vitro cultivation of V. dahliae in the presence of E. coli, B. subtilis or T. viride, or of peptidoglycan to mimic bacterial encounter, did not lead to induction of any of the effector candidate genes (Extended Data Fig. 8). Consequently, we proposed that these genes require other environmental triggers to be induced. Indeed, growth in soil extract consistently induced expression of candidate VdAMP2 (Fig. 4a) that shares structural similarity (confidence >90%) with amphipathic β-hairpins of aerolysin-type β-pore forming toxins (β-PFTs) (Extended Data Fig. 9a) 29 .
To test for potential antimicrobial activity of VdAMP2, we attempted heterologous production of the effector protein. . Roots with rhizosphere soil from three tomato or two cotton plants were pooled to form a single biological replicate. b, Differential abundance analysis of bacterial orders (tomato) and on agglomeration of ASVs (patristic distance <0.1) (cotton) through pairwise comparison between root microbiomes colonized by wild-type V. dahliae and a VdAve1 deletion mutant (Wald test, P < 0.01; Supplementary Tables 1 and 2). The average relative abundance (RA) of the differentially abundant taxa is indicated as a percentage of the total bacterial community in the corresponding root microbiome. c, Sphingomonads (Sphingomonas, Novosphingobium, Sphingopyxis and Sphingobium) are repressed by VdAve1. Dots represent single ASVs with increased abundance (average of three samples) in root microbiomes on colonization by the VdAve1 deletion mutant when compared with wild-type V. dahliae. d, Principal coordinate analysis based on Bray-Curtis dissimilarities reveals separation of synthetic community compositions based on treatment with 4 μM purified VdAve1 (n = 3). e, Differential abundance analysis of bacterial orders on agglomeration of ASVs through pairwise comparison of the synthetic communities treated with demineralized water (MQ) or 4 μM VdAve1 (Wald test, adjusted P < 0.05, Supplementary Table 4). f, VdAve1 contributes to soil colonization. V. dahliae biomass in soil samples (n = 15) or liquid MS medium supplemented with soil (n = 10) was determined by real-time PCR 7 or 3 d after inoculation with wild-type V. dahliae (WT) and the VdAve1 deletion mutant, respectively. P values indicate statistically significant differences according to unpaired two-sided Student's t-test. VdAve1 does not contribute to V. dahliae colonization in sterile soil (n = 10) and sterile MS medium with or without soil (n = 5). Whiskers of the boxplots display the upper and lower quartiles; the boxes display the interquartile range and the thick line displays the median. g, Principal coordinate analysis based on Bray-Curtis dissimilarities reveals clear separation of microbial community compositions in MS medium supplemented with soil (PERMANOVA, P < 0.01), but not in soil. n = 5 for all experimental conditions except for soil with V. dahliae WT for which n = 4.
However, since production in E. coli and Pichia pastoris repeatedly failed, production in V. dahliae under control of the VdAve1 promoter was pursued, resulting in high levels of VdAMP2 expression in vitro (Extended Data Fig. 9b-d). Proliferation of B. subtilis and of P. corrugata (Fig. 4b), but not of F. oxysporum and of T. viride (Extended Data Fig. 9e), was affected by filter-sterilized culture filtrate of the VdAMP2 expression transformant when compared with that of wild-type V. dahliae, suggesting that VdAMP2 exerts only antibacterial activity, similar to VdAve1, although with a different activity spectrum. Soil colonization assays using wild-type V. dahliae and a VdAMP2 deletion mutant (Extended Data Fig. 9f-h) demonstrated that VdAMP2 contributes to V. dahliae fitness in the soil as measured by biomass accumulation (Fig. 4c). Since this fitness contribution is not observed in sterilized soil, but is regained on supplementation with fresh potting soil, we conclude that VdAMP2 contributes to V. dahliae fitness through its efficacy in microbial competition ( Fig. 4d-f). As can be anticipated, the positive effect of VdAMP2 on biomass accumulation in the soil is reflected in disease development when plants are grown on this soil (Extended Data Fig. 10), demonstrating that VdAMP2 positively contributes to virulence of V. dahliae in an indirect manner.

Discussion
Microbial competition occurs in an extremely wide diversity of niches. It is nowadays generally appreciated that a host's microbiome plays a crucial role in its health and, consequently, that hosts actively shape their microbiomes to prevent or suppress disease development. It has also been well established that pathogens secrete effector molecules of various nature during attempted host ingress to promote disease development, many of which target essential components of the host immune system. In our study, we have demonstrated that the fungal broad host-range vascular wilt pathogen V. dahliae uses effector proteins that contribute to niche colonization through selective manipulation of local microbiomes, during host-associated as well as during soil-dwelling life stages. Thus, besides the known activities of plant pathogen effector proteins in targeting host physiology, including immune responses and self-defence against host-secreted defence molecules, we reveal a type of effector activity that involves microbiome manipulation. A wide array of microbially secreted molecules has been described to fulfil crucial functions in intermicrobial competition, including hydrolytic enzymes, secondary metabolites and AMPs. Some Gram-negative bacteria even use a specialized type VI secretion system (T6SS) to translocate AMPs into their microbial competitors 30 . In this manner, Vibrio cholerae, the causal agent of cholera, uses its T6SS to target members of the host commensal microbiota and hereby promotes colonization of the gut 31 . Similarly, the T6SS effector Hyde1 of the phytopathogen Acidovorax citrulli targets plant-associated bacteria in vitro and was speculated to play a role in microbial competition in planta 7 . This T6SS is analogous to the type III secretion system (T3SS) of Gram-negative bacteria that acts as a needle-like structure to directly inject effector proteins into host cells to promote disease 32 . Similar secretion machinery intended for host-microbe or microbe-microbe interactions has not been described for fungi and other filamentous microbes, which instead secrete their effector proteins by extracellular deposition. Consequently, effector molecules targeted towards host cells or towards microbial competitors cannot be discriminated on the basis of differential secretion motifs, such as those that determine type III versus type VI secretion in Gram negatives. Here, we have shown that the pool of effectors secreted by a fungal plant pathogen represents a diverse cocktail comprising proteins involved in the manipulation of the host as well as its microbiome. Consequently, the effectors reported here probably only represent a small proportion of a larger subset of the V. dahliae effector repertoire that is intended for microbiome manipulation. For instance, similar effectors might be crucial during advanced infection stages to prevent secondary infections by opportunistic microbes when host defences are impaired. Additionally, effector proteins can be anticipated to facilitate the survival of the V. dahliae resting structures that persist in the microbe-rich soil for years 33 . After all, possibly, fungal effectors with host microbiome-manipulating capacity initially evolved to limit bacterial growth in soil, as the advent of fungi on earth preceded land plant evolution and fungi initially probably coevolved with bacteria in soil to compete for organic carbon. The discovery of further molecules for microbiome manipulation secreted by V. dahliae and other microbes, and unravelling of underlying modes of action, may ultimately lead to the development of new antibiotics.

Methods
All experiments have been repeated at least three times.
Xylem fluid isolation. Tomato plants (Solanum lycopersicum cv. Moneymaker) were grown under controlled greenhouse conditions as described previously 34 . The stems of 6-week-old plants were cut to allow oozing of the xylem fluid, which was collected on ice with a vacuum pump. The collected xylem fluid was centrifuged for 10 min at 20,000g and filter sterilized using a 0.2-µm filter (Sarstedt). The sterilized xylem fluid was stored at −20 °C until use.

Soil extract preparation.
To prepare soil extract, 100 g of dry potting soil (substraat arabidopsis, Lentse Potgrond) was mixed with 500 ml of demineralized water and autoclaved for 15 min at 121 °C. Soil particles were pelleted through centrifugation and the supernatant was collected and stored at −20 °C until use.

Gene expression analysis.
Total RNA of V. dahliae strain JR2 was isolated from tomato roots seven days after root dip inoculation and following 5 d of in vitro growth in soil extract and potato dextrose broth (PDB) using the Maxwell76 LEV Plant RNA Kit (Promega). Real-time PCR was performed as described previously 17 to determine the expression of effector genes relative to VdGAPDH with primer pairs as shown in Supplementary Table 7.
Production and purification of recombinant effector proteins. The sequences encoding mature VdAve1 and VnAve1 were cloned into pET-15b with an N-terminal His 6 tag sequence (Novagen) (primer sequences, see Supplementary Table 7). The resulting expression vectors were confirmed by sequencing and used to transform E. coli strain BL21. For heterologous protein production, BL21 cells were grown in 1× YT liquid medium at 37 °C with constant shaking at 200 r.p.m. Protein production was induced with 1 mM isopropyl-β-d-thiogalactoside final concentration when cultures reached an optical density (OD 600 ) of 2 to ensure maximum yields. Following 2 h of protein production, the bacterial cells were pelleted and snap-frozen in liquid nitrogen and then washed with 100 mM NaCl, 1 mM EDTA and 10 mM Tris at pH 8.5. Cells were disrupted by stirring for 1 h in lysis buffer (100 mM Tris, 150 mM NaCl, 10% glycerol, 6 mg ml -1 lysozyme (Sigma), 2 mg ml -1 deoxycholic acid, 0.06 mg ml -1 DNaseI, protease inhibitor cocktail (Roche)) at 4 °C. Soluble and insoluble fractions were separated by centrifuging at 20,000g for 10 min. The insoluble protein pellets were washed with 10 ml of 1 M guanidine hydrochloride (GnHCl), 10 mM Tris at pH 8.0 and then denatured in 10 ml of 6 M GnHCl, 10 mM β-mercaptoethanol, 10 mM Tris at pH 8.0. Samples were incubated for 1 h at room temperature. Non-denatured debris was pelleted by centrifuging at 20,000g for 10 min and discarded. Denaturation was allowed to continue for additional 3-4 h. Proteins were purified under denaturing conditions by metal affinity chromatography using a column packed with 50% His60 Ni 2+ Superflow Resin (Clontech). The purified effector proteins were dialysed (Spectra/Por 3 Dialysis Membrane, molecular weight cut off of 3.5 kDa) step-wise against 20 volumes of 0.25 M ammonium sulfate, 0.1 M BisTris, 10 mM reduced glutathione, 2 mM oxidized glutathione, pH 5.5 with decreasing GnHCl concentrations for refolding. Each dialysis step was allowed to proceed for at least 24 h. Finally, proteins were dialysed against demineralized water. Final concentrations were determined using the BioRad Protein Assay (BioRad).
Functionality of refolded VdAve1 was confirmed through recognition by the corresponding tomato immune receptor Ve1. To this end, an overnight culture of A. tumefaciens strain GV3101 carrying the pSOL2092:Ve1 construct 35 was harvested by centrifugation and resuspended to OD 600 = 2 in MMA (2% sucrose, 0.5% MS salts (Duchefa Biochemie), 10 mM MES, 200 μM acetosyringone, pH 5.6) and infiltrated in the leaves of 5-week-old N. tabacum (cv. Petite Havana SR1) plants. After 24 h, 10 µM of purified and refolded 6xHis-VdAve1 was infiltrated in leaf areas expressing Ve1. Photos were taken 3 d after infiltration of the effector protein.

Generation of V. dahliae mutants.
To generate the VdAMP2 effector deletion construct, VdAMP2 flanking sequences were amplified using the primers listed in Supplementary Table 7 and cloned into pRF-HU2 (ref. 36 ). To allow expression of VdAMP2 under control of the VdAve1 promoter, the coding sequence of VdAMP2 was amplified and cloned into pFBT005. All constructs were transformed into A. tumefaciens strain AGL1 for V. dahliae transformation as described previously 37 .
V. dahliae culture filtrates. Conidiospores of V. dahliae strain JR2 and the VdAMP2 expression transformant were harvested from potato dextrose agar (PDA) and diluted to a final concentration of 10 4 conidiospores per ml in 20 ml of 0.2× PDB supplemented with 0.5× MS medium (Duchefa). Following 4 d of incubation at 22 °C and 120 r.p.m., the fungal biomass was pelleted and the remaining supernatants were filter sterilized and stored at −20 °C until use.
Bacterial isolates. Bacterial strains B. subtilis AC95, S. xylosus M3, P. corrugata C26, Streptomyces sp. NE-P-8 and Ralstonia sp. M21 were obtained from our in house endophyte culture collection. Strains used in this study were all isolated from the xylem vessels of tomato cultivars from commercial greenhouses, both from stem and leaf sections. All strains were identified on the basis of their 16S ribosomal RNA gene sequence using the primers 27F and 1492R (Supplementary Table 7). The 16S amplicons were sequenced by Sanger sequencing at Eurofins (Mix2Seq). The partial 16S rRNA gene sequences obtained were evaluated against the 16S ribosomal DNA sequence (Bacteria and Archaea) database from the National Centre for Biotechnology Information. Bacterial strains Acidovorax sp. In vitro microbial growth assays. Bacterial isolates were grown on lysogeny broth agar or tryptone soya agar at 28 °C. Single colonies were selected and grown overnight at 28 °C while shaking at 200 r.p.m. Overnight cultures were resuspended to OD 600 = 0.05 in xylem fluid supplemented with purified effector proteins or diluted using culture filtrates to OD 600 = 0.1. Additionally, fungal spores were harvested from a PDA plate and suspended in xylem fluid supplemented with purified effector proteins or the V. dahliae culture filtrates to a final concentration of 10 4 spores per millilitre. Then 200 µl of the microbial suspensions was aliquoted in clear 96-well flat-bottom polystyrene tissue culture plates. Plates were incubated in a CLARIOstar plate reader (BMG Labtech) at 22 °C with double orbital shaking every 15 min (10 s at 300 r.p.m.). The optical density was measured every 15 min at 600 nm.
Scanning electron microscopy. Samples for scanning electron microscopy were prepared as described previously with slight modifications 38 . In short, B. subtilis strain AC95 was grown overnight in lysogeny broth and resuspended in xylem fluid to an OD 600 = 0.05. Purified effector proteins were added to a final concentration of 6.5 μM (0.8× MIC, VdAve1) and bacterial suspensions were incubated for 0, 1, 3 and 7 h. Next, 20 μl of the bacterial suspensions was transferred to poly-l-lysine coated glass slides (Corning) and incubated for another hour to allow binding of the bacteria. Glass slides were washed using sterile MQ water and samples were fixed using 2.5% glutaraldehyde followed by postfixation in 1% osmium tetroxide. Samples were dehydrated using an ethanol dehydration series and subjected to critical point drying using a Leica CPD300 (Leica Mikrosysteme). Finally, the samples were mounted on stubs, coated with 12 nm of tungsten and visualized in a field emission scanning electron microscope (Magellan 400, FEI).
VdAve1 activity assay on cucumber protoplasts. Cucumber protoplasts were obtained by enzymatic digestion of 7-day-old cucumber cotyledons with cellulase and macerozyme, using mannitol as an osmostabilizer. The protoplasts were collected by 1 min of centrifugation at 100g and carefully resuspended in 1 M sorbitol, 10 mM MOPS pH 6.3, to a final concentration of 10 5 -10 6 protoplasts per ml. Following 30 min of incubation, intact protoplasts were quantified for the different treatments using a haemocytometer.

Root microbiome analysis.
Tomato and cotton inoculations were performed as described previously 34 . After 10 d, plants were carefully uprooted and gently shaken to remove loosely adhering soil from the roots. Next, roots with rhizosphere soil from three tomato or two cotton plants were pooled to form a single biological replicate. Samples were flash-frozen in liquid nitrogen and ground using mortar and pestle. Genomic DNA isolation was performed using the DNeasy PowerSoil Kit (Qiagen). Quality of the DNA samples was checked on a 1.0% agarose gel. Sequence libraries were prepared following amplification of the V4 region of the bacterial 16S rDNA (515F and 806R) and paired ends (250 bp) were sequenced using the HiSeq2500 sequencing platform (Illumina) at the Beijing Genome Institute (Hong Kong, China).
Sequencing data were processed using R v.3.3.2. as described previously 39 . Briefly, ASVs were inferred from quality filtered reads (Phred score >30) using the DADA2 method 40 . Taxonomy was assigned using the Ribosomal Database Project training set (RDP, v.6) and mitochondria-and chloroplast-assigned ASVs were removed. Next, ASV frequencies were transformed according to library size to determine relative abundances. The phyloseq package (v.1.22.3) was used to determine α-diversity (Shannon index) and β-diversity (Bray-Curtis dissimilarity) as described previously 39,41 . PERMANOVA was performed using the adonis function of vegan package v.2.5-6. Differential abundance analysis was performed using the DESeq2 extension within phyloseq 42 . To this end, a parametric model was applied to the data and a negative binomial Wald test was used to test for differential abundance of bacterial taxa with P < 0.01 as the significance threshold.
Soil colonization assays. Conidiospores of the V. dahliae strain JR2 and the mutants were harvested from PDA plate and a total of 10 6 or 10 7 conidiospores were added to 1 g of potting soil. Samples were incubated at room temperature in the dark for 1 week. Alternatively, conidiospores were diluted in sterilized (15 min at 121 °C) and untreated liquid 0.5 × MS medium supplemented with 3% sucrose and 2% potting soil to a final concentration of 10 6 conidiospores per millilitre. Samples were incubated at room temperature in the dark for 72 h.
Following incubation, DNA was extracted from the samples using the DNeasy PowerSoil Kit (Qiagen). Microbiome analysis on selected samples was performed as described above. V. dahliae biomass was quantified through real-time PCR using V. dahliae specific primers targeting the internal transcribed spacer (ITS) region of the ribosomal DNA (Supplementary Table 7). Primers targeting a conserved region of the bacterial 16S rRNA gene were used for sample equilibration.
To allow sample calibration when using sterilized potting soil or MS medium (15 min at 121 °C), the same amount of fresh potting soil was added to the samples before DNA extraction. Additionally, after 1 week of incubation of V. dahliae in the sterilized soil, serial dilutions were made and plated onto PDA to quantify colony forming units.
Treatment of synthetic community with purified VdAve1. A premixed synthetic community comprising a total of 137 bacterial strains from the At-SPHERE collection 25 was diluted in liquid R2A medium supplemented with demineralized water or 4 μM VdAve1 to a concentration of OD 600 = 0.002. The bacterial suspensions were grown overnight at 22 °C with constant shaking at 120 r.p.m. Next, bacterial cells were pelleted and DNA was extracted from the samples using the DNeasy PowerSoil Kit (Qiagen). Microbiome analysis was performed as described above using P adjusted at <0.05 as the significance threshold for the differential abundance analysis.
In vitro competition assay. Conidiospores of V. dahliae strain JR2 and the VdAve1 deletion and complementation mutants were harvested from a PDA plate using sterile water and diluted to a final concentration of 10 6 conidiospores per ml in liquid 0.5× MS medium (Duchefa). Next, overnight cultures of the bacterial isolates were added to the conidiospores to OD 600 = 0.05 and 500 µl of the microbial suspensions was aliquoted in clear 12-well flat-bottom polystyrene tissue culture plates. Following 48 h of incubation at room temperature, the microbial cultures were recovered and genomic DNA was isolated using the SmartExtract DNA Extraction Kit (Eurogentec). V. dahliae biomass was quantified through real-time PCR using V. dahliae specific primers targeting the ITS region of the ribosomal DNA (Supplementary Table 7). Statistical analyses (analysis of variance, ANOVA) were performed using SPSS 25.
In planta competition assay. To allow S. macrogoltabida and Ralstonia sp. colonization of in the absence of other microbes, tomato seeds were incubated for 5 min in 2% sodium hypochlorite to ensure surface sterilization. Next, surface-sterilized tomato seeds were washed three times using sterile water and transferred to a sterile Petri dish containing a filter paper premoistened with a S. macrogoltabida or Ralstonia sp. suspension in water (OD 600 = 0.05). The tomato seeds were allowed to germinate in vitro and eventually transferred to usual potting soil, 10-day-old seedlings were inoculated as described previously 34 . Tomato stems were collected at 14 dpi and lyophilized before genomic DNA isolation with a CTAB-based extraction buffer (100 mM Tris-HCl pH 8.0, 20 mM EDTA, 2M NaCl, 3% CTAB). V. dahliae biomass was quantified with real-time PCR on the genomic DNA by targeting the ITS region of the ribosomal DNA. The tomato rubisco gene was used for sample calibration. S. macrogoltabida biomass was quantified using Sphingopyxis specific primers (Supplementary Table 7) and normalized using the V. dahliae ITS. Statistical analyses (ANOVA) were performed using SPSS 25. Additionally, the relative abundance of the Sphingomonadales in three representative samples was determined by 16S ribosomal DNA profiling as described previously.
Disease assays using V. dahliae microsclerotia. V. dahliae microsclerotia were produced in a sterile moist medium of vermiculite and maize meal as described previously 43 . After 4 weeks of incubation, the vermiculite/microsclerotia mixture was dried at room temperature. Next, 150 ml of the dried mixture was mixed with 1 l of potting soil (substraat arabidopsis, Lentse Potgrond) and Arabidopsis seeds of the Col-0 ecotype were sown at equal distances on top of the mixture. The above-ground parts of the plants were collected at 27 dpi and V. dahliae biomass was quantified through real-time PCR using V. dahliae specific primers targeting the ITS region of the ribosomal DNA. The Arabidopsis rubisco gene was used for sample calibration (Supplementary Table 7).
Reporting Summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The metagenomics data have been deposited in the European Nucleotide Archive under accession number PRJEB34281. Source data are provided with this paper. Fig. 2 | Heterologously produced VdAve1 can be isolated from inclusion bodies. a, E. coli BL21 cells were grown in liquid YT medium and VdAve1 expression was induced using 1 mM IPTG. Following four hours of protein production at 28 o C, the presence of VdAve1 was confirmed by boiling the cells in 1% SDS, 2M urea, 1.25% β-mercaptoethanol, 2.5% glycerol, 15 mM Tris, pH 6.8. The band representing VdAve1 is indicated with an asterisk; limited solubility of the protein was detected upon sonication of the cells in the corresponding buffers (lanes 1-9). The presence of VdAve1 in the insoluble protein fractions was confirmed following denaturation of the insoluble proteins, indicating the formation of inclusion bodies. The experiment has been repeated three times with similar results, the gel image is from a single experiment. b, VdAve1 purified from the insoluble protein fraction under denaturing conditions was refolded by step-wise dialysis. Functionality of the protein was confirmed by infiltration into N. tabacum leaf sections overexpressing the corresponding tomato immune receptor Ve1, resulting in a hypersensitive response at three days post infiltration.

NAturE PlANtS
Extended Data Fig. 4 | Metagenomic characterization of tomato and cotton root microbiomes upon Verticillium dahliae infection. a, Relative abundance of bacterial phyla in the root microbiomes of tomato and cotton plants ten days after inoculation with wild-type V. dahliae (WT) and a VdAve1 deletion mutant as determined by 16S ribosomal DNA profiling. b, V. dahliae inoculation does not change α-diversity of host root microbiomes (one-way ANOVA and Tukey's post-hoc test; p<0.05; N=3). The plot displays the average Shannon index ± SD. c, Sphingomonadales are significantly enriched in the microbiomes of roots that are colonized by the VdAve1 deletion mutant. Differential abundance analysis of bacterial orders following combination of tomato and cotton samples based on infection by the different V. dahliae genotypes, only revealed the Sphingomonadales as differentially abundant (unpaired two-sided student's t-test, p<0.01; N=6). Relative abundances were normalized against the average relative abundance upon infection of the corresponding host by wild-type V. dahliae to correct for host-dependent differences. Whiskers of the boxplot display the upper and lower quartile; the boxes display the interquartile range and the thick line displays the median. Fig. 5 | Metagenomic characterization of soil microbiomes and a synthetic community upon Verticillium dahliae inoculation or VdAve1 treatment. a, Relative abundance of bacterial phyla or families in a synthetic community (Syncom), in MS medium supplemented with soil, and in soil after inoculation with wild-type V. dahliae (WT), a VdAve1 deletion strain, or treatment with purified VdAve1. b, Impact of V. dahliae inoculation or VdAve1 treatment on α-diversity in the microbiomes as shown in a (unpaired two-sided student's t-test; N=3) (one-way ANOVA and Tukey's post-hoc test; p<0.05; N=5 for all experimental conditions except for soil with V. dahliae WT for which N=4). The plot for the syncom displays the average Shannon index ± SD. Whiskers of the boxplots as shown for the microbial communities in MS + soil and soil display the upper and lower quartile; the boxes display the interquartile range and the thick line displays the median.

NAturE PlANtS
Extended Data Fig. 9 | VdAMP2 structure, antifungal activity assays and V. dahliae mutants. a, VdAMP2 shares structural homology with the amphipathic β-hairpins of aerolysin-type β-pore forming toxins. Predicted protein structure of part of VdAMP2 (aa 225-312) with the amphipathic β-hairpin highlighted in red. Protein structure was predicted using Phyre2 45 . b, In vitro expression of VdAMP2 in wild-type V. dahliae and the pVdAve1::VdAMP2 mutant after five days of cultivation in liquid 0.5x Murashige & Skoog (MS) medium. Plot displays the average expression of three biological replicates ± SD. c, d, In vitro expression of VdAMP2 does not affect V. dahliae growth. c, Growth of wild-type V. dahliae and the pVdAve1::VdAMP2 mutant in liquid 0.2x potato dextrose broth (PDB) + 0.5x MS. Graphs display the average OD 600 of three biological replicates ± SD. d, Morphology of wild-type V. dahliae and the pVdAve1::VdAMP2 mutant after five days of cultivation on potato dextrose agar (PDA). e, VdAMP2 does not inhibit fungal growth. Growth of F. oxysporum and T. viride in filter-sterilized culture filtrates from in vitro grown wild-type V. dahliae and the VdAMP2 expression transformant does not reveal antifungal activity of VdAMP2. Graphs display the average OD 600 of three biological replicates ± SD. f, Deletion of VdAMP2 was confirmed using PCR on genomic DNA of wild-type V. dahliae and the VdAMP2 deletion mutant (ΔVdAMP2), VdAve1 was used as genomic DNA control. The experiment has been repeated three times; the gel image is from a single experiment. g, Morphology of wild-type V. dahliae and the VdAMP2 deletion mutant at five, seven and ten days of cultivation on PDA. h, The VdAMP2 deletion mutant is not affected in microsclerotia formation (N=9). After ten days, colonies as shown in f were excised from plate and the tissue was ground to determine the number of microsclerotia per cm 2 using a haemocytometer. Whiskers of the boxplot display the upper and lower quartile; the boxes display the interquartile range and the thick line displays the median.

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 For manuscripts utilizing custom algorithms or software that are central to the research but not yet described in published literature, software must be made available to editors/reviewers. We strongly encourage code deposition in a community repository (e.g. GitHub). See the Nature Research guidelines for submitting code & software for further information.

Data
Policy information about availability of data All manuscripts must include a data availability statement. This statement should provide the following information, where applicable: -Accession codes, unique identifiers, or web links for publicly available datasets -A list of figures that have associated raw data -A description of any restrictions on data availability The metagenomics data have been deposited in the European Nucleotide Archive (ENA) under accession number PRJEB34281.

Field-specific reporting
Please select the one below that is the best fit for your research. If you are not sure, read the appropriate sections before making your selection.

Life sciences
Behavioural & social sciences Ecological, evolutionary & environmental sciences