All chemicals used for metabolite extraction and sample preparation for DIMS analysis were of the highest commercially available purity. Methanol, ethyl acetate, formic acid, ammonium acetate (Optima® grade), and water (HPLC grade) were purchased from Fisher Scientific Company (Ottawa, ON, Canada).

Starter cultures of the mycoparasite Stachybotrys elegans (Pidoplichko) W. Gams (ATCC 18825) and the pathogen Rhizoctonia solani AG-3 (ATCC 10183) were revived from pre-colonized oat kernels on 1% potato dextrose agar (PDA; Difco Laboratories, Michigan, USA) and incubated at 24°C for 7 and 5 days, respectively. Induction and collection of S. elegans conidia were performed as previously described (Chamoun and Jabaji, 2011).

Dual-culturing of S. elegans and R. solani was conducted in 9 cm Petri plates containing 20 mL of minimal synthetic medium (MSMA) composed (g L⁻¹) of: MgSO₄.7H₂O, 0.2; K₂HPO₄, 0.9; KCl, 0.2; FeSO_4.7H₂O, 0.002; MnSO₄, 0.002; ZnSO₄, 0.002; NaNO₃, 1.0; biotin, 10 mg; gellan gum, 1% (composed of glucose, glucuronic acid and rhamnose in the molar ratio of 2:1:1) (Phytagel, Sigma, St. Louis, USA).

Agar plugs (8 mm) of a 5-day old R. solani culture were grown on MSMA for 48 h and then sprayed with 100 μL of a suspension of S. elegans conidia (10⁶ mL⁻¹ water) using a Badger 350 air brush and MC-80 mini air compressor calibrated at 1 kg cm⁻². The control treatments consisted of spraying 100 μL of S. elegans conidia on non-inoculated MSMA plates and R. solani-inoculated MSMA plates sprayed with sterile distilled water. Additionally, a negative control representing the MSMA medium was used to determine compounds of non-biological origin. All culture plates were incubated at 24°C for 4 or 5 days following dual and pure strain cultivation. These time points were chosen based on a priori knowledge to capture the infection and colonization of R. solani hyphal cells by S. elegans (Chamoun and Jabaji, 2011). Five replications were performed per treatment.

To associate the metabolic changes with the progress of the mycoparasitic process, agar pieces (5 × 5 mm) from interaction zones of dual-culture plates and from pure cultures of both fungal partners were collected in a time course. Sections from interacting zones were stained with lactophenol blue or water and viewed under an optical microscope. Presence of hyphal coils, penetration pegs and intracellular colonization of the pathogen was digitally documented with the Moticam 2300 digital camera (GENEQ Inc. QC, Canada).

Four plugs (8 mm in diameter × 7 mm in height) were collected from the interaction zones of dual-cultures, pure cultures of each fungal partner after 4 or 5 days of cultivation and from the negative control (MSMA) plates. Plugs were placed in glass autosampler screw thread vials (2 mL, Fisher Scientific, ON, Canada). Quenching was instantly performed by adding twice 2 mL of liquid N₂, and samples were stored at -80°C until further processing. Extraction was performed as previously described (Aliferis et al., 2014). Briefly, 1 mL of a mixture of methanol-ethyl acetate (50:50, v/v) was added to the vials, followed by sonication for 25 min. Samples were further extracted for 2 h under continuous agitation (250 rpm) at 25°C and filtered through 0.2-μm filters (Millex-FG; Millipore, MA, USA). The volume of samples was adjusted to 1 mL and subsequently divided into two equal portions (0.5 mL) for DIMS analyses in positive (ESI⁺) and negative (ESI⁻) electrospray modes. Finally, extracts were dried using a Labconco CentriVap refrigerated vacuum concentrator equipped with a cold trap (Labconco, MO, USA).

For DIMS and DIMS/MS analyses, an LTQ Orbitrap MS Classic (Thermo Scientific, CA, USA) was used acquiring in the ESI⁺ or ESI⁻ modes (Aliferis et al., 2014). All experimental events were controlled by the software Xcalibur v.2.2 (Thermo Scientific). The analyzer was equipped with a heated electrospray ionization probe (HESI-II, Thermo Scientific), a quadrupole linear ion trap, and an Accela pump (Thermo Scientific). For analysis in ESI⁺ and ESI⁻, 100 μL of a mixture of methanol/formic acid (0.2% v/v) (50–50, v/v) or methanol/ammonium acetate (4 mM) was added to the dried samples, respectively. Extracts were then transferred to glass microinserters (150 μL), which were consecutively placed into 2 mL glass autosampler vials. Samples (10 μL) were injected at a flow rate of 10 μL min⁻¹ using a 100 μL syringe (Hamilton, NV, USA). Full scan mass spectra were acquired in the range between 50 and 1200 Da at a rate of 0.6 scans/s and a mass resolution of 60,000 at 400 m/z. The source and capillary voltages were set to 3.2 kV and 5.0 V for ESI⁺ and 4.0 kV and −35 V for ESI⁻, respectively. The capillary temperature for both modes was set to 275°C. Sheath gas flow was set to 10 (ESI⁺), and 20 (ESI⁻) whereas no auxiliary and sweep gases were used. For selected samples, MS/MS spectra were recorded using previously described settings (Aliferis et al., 2014).

Mass spectra were processed using the freely available software MZmine 2 (Pluskal et al., 2010) following the procedures recommended by the developers after optimization of the obtained data. Cumulative spectra were collected between 0.8–1.3 min for ESI⁺ and 0.6–1.1 min for ESI⁻. Metabolic features were detected using the centroid algorithm and the noise level was optimized for each sample. The Fourier transform mass spectrometer (FTMS) shoulder filter was then applied at a mass resolution of 8000 using the Lorentzian extended model function. Chromatogram built, alignment and gap-filling were performed using an m/z tolerance (Δppm) <3. Alignment was performed using the “Join aligner” option, whereas gap filling was performed in two steps; first using the “Peak finder option” and then the “Same RT and mz range gap filter” (Pluskal et al., 2010). This procedure accounted for the presence of missing peaks in the matrix as a result of the performance of the peak detection algorithm or possible mistakes in the alignment. Subsequently, the matrices were subjected to filtering by removing rows with more than 50% missing values among the biological replications of the same treatment. Following alignment, metabolic features of non-biological origin corresponding to the negative control samples (MSMA) and also detected in the biological samples were removed and were excluded from further analysis.

The obtained aligned matrix was then exported to Microsoft Excel for further processing. Finally, the matrix composed of identified secondary metabolites detected in ESI⁺ and ESI⁻ was exported to the SIMCA-P+ v.12.0.1 software (Umetrics, MKS Instruments Inc., MA, USA) for multivariate analysis (Aliferis and Jabaji, 2012). The discovery of biomarker-ions was based on partial least squares-discriminant analysis (PLS-DA) regression coefficients (P < 0.05). Based on the variability in the model parameters encountered in the different cross-validation cycles, standard errors were calculated with 95% confidence interval using Jack-knifing (Efron and Gong, 1983).

The identification of metabolites was performed following a biologically-driven approach performing searches against the targeted in-house species-specific metabolic databases for Rhizoctonia and Stachybotrys. The libraries were constructed acquiring information from the literature and publicly available databases such as, KNApSAcK (http://kanaya.naist.jp/KNApSAcK/) and PubChem (http://pubchem.ncbi.nlm.nih.gov/). Identification of metabolites was based on mass accuracy (<2 ppm) and where available, on isotope and/or MS/MS fragmentation patterns (Supplementary Data Sets 1–4) using data from the databases of METLIN (http://metlin.scripps.edu/index.php) and mzCloud (https://www.mzcloud.org/) and the literature. In addition, the heuristic rules of Kind and Fiehn (2007), which are implemented in the MZmine 2 (Pluskal et al., 2010), were applied. These rules provide a valuable tool for reducing the number of candidate molecular formulae for a given ion. Detection of mass errors was confirmed by Xcalibur v.2.2 (Thermo Scientific).

Additionally, since the majority of the secondary metabolites have unique structures, the assignment of metabolites to the corresponding producing fungus during mycoparasitism was a feasible task at the applied mass resolution.

In dual-cultures on Petri plates, S. elegans conidia germinated within 24 h, made contact with hyphal cells of R. solani and overgrew over the pathogen colonies after 4 days of dual-cultivation (Figure 1). Therefore, 4 and 5 days were selected as the time points to study the induced production, involvement and changement of secondary metabolites during mycoparasitism. Conspicuous accumulation of S. elegans aerial hyphae over R. solani colonies was observed and accompanied by heavy coiling and formation of infection pegs and intracellular colonization of R. solani cells (Figure 1). In the presence of the mycoparasite, the cytoplasm of R. solani infected cells appeared disorganized and devoid of granules. R. solani pure cultures appeared less pigmented compared to the parasitized cultures (Figure 1), which manifested a change in the color from white to dark brown of underneath medium, corresponding to the biosynthesis and diffusion of fungal metabolites into the growth medium as a result of the interaction (data not shown).

The lack of chromatographic separation in DIMS analysis makes the identification of metabolites challenging, even with high mass accuracy (e.g., <2 ppm). In addition, the possible presence of metabolites with identical molecular formulae or isomers makes their absolute identification even more complex. In this context, the identification of metabolites during the mycoparasitic interaction being studied, was based on searches against the two species-specific metabolite libraries for Stachybotrys and Rhizoctonia (Supplementary Tables 1, 2) for commonly occurred adducts (Supplementary Table 3). Identities were assigned to 36 metabolic features of the obtained metabolite matrix combining results of ESI⁺ and ESI⁻ analyses, 30 of which were unique (single metabolite) (Supplementary Data Set 5). Such approach not only facilitates the robust identification of fungal secondary metabolites, which represents a bottleneck for high-throughput fungal metabolomics, but additionally, it enables the assignment of the origin of the recorded metabolic features in their dual-cultures. The latter is facilitated largely by the unique structures that the identified metabolites of both fungal species have (e.g., none of the metabolites of the two target libraries share the same molecular formula).

For the detection of trends within the obtained matrix and corresponding biomarkers of mycoparasitism, the metabolic profiles of R. solani-S. elegans dual-cultures were compared to those of pure cultures for both time points (Figure 2 and Supplementary Figures 1–4) applying multivariate analysis (MVA). Initially, application of the unsupervised principal component analysis (PCA), revealed a tight clustering between the biological replications of the same treatment in the corresponding PC1/PC2 score plots and the absence of outliers (P < 0.05) (Supplementary Figure 5). This is indicative of the robustness of the applied bio-analytical protocol, data processing, instrument's performance and of the substantial differences between the metabolic profiles of Rhizoctonia-Stachybotrys dual-cultures and their corresponding pure cultures at both time points. Additionally, the number of identified Rhizoctonia-derived metabolites was substantially higher than that of Stachybotrys-derived ones for both time points (Figure 3). Plausibly this is due to the fact that Rhizoctonia was established in the media prior to treatments, which gave it more time to synthesize and release metabolites.

In a second step of MVA, PLS-DA and hierarchical clustering were applied for the discovery of trends within treatments (Figure 4). Similarly to PCA, both analyses showed a very strong discrimination between the recorded metabolic profiles of pure and dual-cultures and tight clustering between biological replications. As an indication of the dynamics of its biosynthetic activity, the metabolite profiles of Rhizoctonia grown in pure and in dual-cultures were substantially different at both time points.

It is noteworthy, that for metabolomics, the absence of a metabolite from a treatment in the matrix means either no detection (e.g., below the limits of detection) or elimination following filtering (missing values >50%). The latter could be attributed either to the variation in the biological samples or signals with intensity near the limits of detection of the instrument.

Results revealed the substantial impact of Stachybotrys mycoparasitic activity on Rhizoctonia's metabolism (Figures 4, 5). The biosynthesis of the vast majority of the identified Rhizoctonia-derived metabolites were significantly down-regulated, whereas only a handful was up-regulated or remained unaffected in response to mycoparasitism. This is indicative of the general disturbance of the pathogen's metabolism in response to the invasion of the mycoparasite, which plausibly represents the evidence for the outcome of such interaction.

The diketopiperazines (DKPs) cyclo(S-Pro-S-Leu)/cyclo(S-Pro-S-Ile), ethyl 2-phenylacetate, and 3-nitro-4-hydroxybenzoic acid (Supplementary Figure 6) were induced in Rhizoctonia 4 days following dual-cultivation with Stachybotrys, whereas only the latter was found to be up-regulated 1 day later.

DKPs exhibit antifungal and antibacterial properties, and inhibit the biosynthesis of mycotoxins (Martins and Carvalho, 2007; Huang et al., 2010; Borthwick, 2012). They have been isolated from fungal species such as, Aspergillus spp. (Li et al., 2004; Wang et al., 2008), Alternaria spp. (Musetti et al., 2007), Fusarium oxysporum (Trigos et al., 1995), and R. solani (Pedras et al., 2005). In the latter study, the DKPs cyclo(S-Pro-S-Leu)/cyclo(S-Pro-S-Ile) were isolated from R. solani cultures, which is in accordance with our data. Cyclo(S-Pro-S-Leu)/cyclo(S-Pro-S-Ile) do not exhibit phytotoxicity (Pedras et al., 2005), however, based on the resemblance of their structure with other DKPs with established antimicrobial and mycotoxin inhibitory action, it is plausible to suggest a role of cyclo(S-Pro-S-Leu)/cyclo(S-Pro-S-Ile) in the defense mechanism of Rhizoctonia against the stress imposed by the invasive mycelia of Stachybotrys, a known producer of mycotoxins (Deng et al., 2003). In contrast, the biosynthesis of cyclo(S-Pro-S-Val) (Figure 5) was suppressed in the presence of the mycoparasite at both time points.

Phenylacetic acid (PAA) and its derivatives are the first studied bioactive metabolites of Rhizoctonia (Aoki et al., 1963). These metabolites share a functional phenyl group and a carboxylic acid (Supplementary Figure 6), and are known not only for their phytotoxicity but also for their antimicrobial activities (Hwang et al., 2001; Mao et al., 2006; Ding et al., 2008; de Lima Mendonça et al., 2009). Interestingly, here, the biosynthesis of PAA, phenylethyl-2-phenylacetate, and mandelic acid was suppressed in the presence of the mycoparasite; whereas ethyl 2-phenylacetate and 1-menthyl phenylacetate were the most induced metabolites in response to Stachybotrys attack (Supplementary Figure 6). A major response of Rhizoctonia to mycoparasitism was the increased biosynthesis of 3-nitro-4-hydroxybenzoic acid at both time points. Information in the literature supporting the antimicrobial activity of benzoic acid and its derivatives against economically important plant pathogens is established (Sopheareth et al., 2013). Its mode of action is attributed to lowering the intracellular pH of fungal cells leading to inhibition of glycolysis (Krebs et al., 1983). Whether 3-nitro-4-hydroxybenzoic acid has the same role during the mycoparasitic process of R. solani by S. elegans requires further study.

The success of Stachybotrys to overcome the defense mechanisms of Rhizoctonia and parasitize it is indirectly linked to the suppression of the biosynthesis of the majority of its metabolites with well-established bioactivity. Among these, melatonin is an antioxidant and free radical scavenger in many organisms, including fungi (Hardeland et al., 2006; Tamura et al., 2012). Interestingly, N6-acetyl-L-lysine and (S)-2,3,4,5-tetrahydropiperidine-2-carboxylate that belong to the lysine degradation pathway were detected in lower amounts during mycoparasitism compared to pure cultures. Both metabolites control the synthesis of glutamate which contributes to the synthesis of the stress-related molecules γ-aminobutyric acid, polyamines, and nitric oxide (Galili, 2002).

Fungal species are also sources of pigments (Gessler et al., 2013). Among them are octaketide pigments with structure based on the anthraquinone skeleton (Velíšek and Cejpek, 2011). Here, a decrease in anthraquinone was observed in the dual-cultures compared to the pure cultures 4 days following treatment, whereas the metabolite was not detected in the dual-cultures 1 day later. Bioassay studies performed on several anthraquinones derivatives that were isolated from various fungi have highlighted their antibacterial, antiparasitic, antiviral and fungicidal activities (Kanokmedhakul et al., 2002; Srinivas et al., 2007; Zhou et al., 2014). The metabolite slaframine, known to be produced by R. leguminicola, the causal pathogen of the black patch disease of red clover (Li et al., 2012), has been detected also in decreased amount in pure cultures compared to dual-cultures at both time points. Slaframine is an indolizidine alkaloid responsible for locoism and leads to economical losses in animals (Croom et al., 1995).

The benzophenone derivative rhizoctonic acid was present only in pure cultures and suppressed during mycoparasitism. In addition to its isolation from R. solani (Ma et al., 2004), it has been isolated from endophytic fungi such as, Guignardia and Penicillium sp. and has been reported to exhibit antimicrobial activity against human pathogens (Ma et al., 2004; Wang et al., 2010).

Stachybotrys genus includes diverse species having the ability to produce a wide range of bioactive secondary metabolites (Deng et al., 2003). Several mycotoxins of Stachybotrys were detected during Rhizoctonia's mycoparasitism 4 and 5 days following dual-culturing (Figure 6 and Supplementary Data Set 5), which probably suggesting their involvement in the mycoparasitism process. The production of mycotoxins by the genus Stachybotrys as well as other unrelated fungi such as, Fusarium, Trichoderma, Trichothecium, Verticimonosporium, and Cephalosporium is well documented (Bräse et al., 2009; McCormick et al., 2011; Kramer and Abraham, 2012).

The majority of the identified metabolites belongs to trichothecenes, a well-studied class of sesquiterpenes (Rocha et al., 2005; McCormick et al., 2011). They are able to passively move across cell membranes (McCormick et al., 2011) and bind to ribosomes and trigger mitogen-activated protein kinases (MAPKs) (Pestka et al., 2004). Their bioactivity is mainly attributed to the epoxide that they contain (C12-C13) (Sudakin, 2003) (Supplementary Figure 6).

The identified trichothecenes are simple (e.g., trichothecin andtrichodermol), with the exception of the macrocyclicroridin A-trichoverrin A/B (Supplementary Data Set 5). Studies performed on human and plant cells have revealed that trichothecenes act by inhibiting the synthesis of nucleic acids and protein synthesis (Rocha et al., 2005; McCormick et al., 2011). Additionally, trichothecenes have been reported to generate hydrogen peroxide, alter cell division and membrane function (Shifrin and Anderson, 1999; Nishiuchi et al., 2006; Yazar and Omurtag, 2008). Reports on trichothecenes' activity against plant pathogens are less common (Ayer and Miao, 1993). Trichodermol produced by Stachybotrys cylindrospoa exhibited strong activity against the blue stain fungus Ophiostoma crassivaginatum in confrontation assays (Hiratsuka et al., 1994). In a similar analogy to their documented bioactivity on human and plant cells, we hypothesize that the presence of trichothecenes in dual-cultures is triggered by the pathogen and results in the alteration of its metabolism and ultimately its growth and development.

In addition to trichothecenes, the Stachybortys-produced atranones D/E, F and H, were identified during mycoparasitism, with the latter being present only during mycoparasitism and not in pure cultures (Supplementary Data Set 5). This indicates its de novo or substantially increased biosynthesis during mycoparasitism. This toxin is an analog of atranones A, B, and I (Hinkley et al., 2003) and a precursor of atranone J (Jarvis, 2003). Atranones are diterpenoids with unique structures (Supplementary Figure 6), produced by species such as, Stachybotrys spp. and Myrothecium verrucaria (Bräse et al., 2009). However, in contrast to trichothecenes, atranones do not exhibit significant bioactivity (Jarvis, 2003).

Finally, a small number of identified Stachybotrys metabolites were detected only in pure cultures (Supplementary Figure 7). Among these, spirodihydrobenzofuranlactam 4 (Deng et al., 2003) act as protein synthesis inhibitors and protein antagonists (Roggo et al., 1996); the spirocyclic drimane stachybotrylactone exhibit antiplasmodial activity (Wang et al., 2014); and the sesquiterpenoid trichothecolone exhibits cytotoxic activity (Wang et al., 2015). This finding plausibly indicates their decreased biosynthesis and thus, minor importance for mycoparasitism. This could be attributed either to a “preference” for the biosynthesis of other bioactive metabolites or inhibition of their biosynthesis as a result of the action of Rhizoctonia-derived metabolites, and needs further investigation.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.