Bacterial community richness shifts the balance between volatile organic compound-mediated microbe–pathogen and microbe–plant interactions

Abstract

Even though bacteria are important in determining plant growth and health via volatile organic compounds (VOCs), it is unclear how these beneficial effects emerge in multi-species microbiomes. Here we studied this using a model plant–bacteria system, where we manipulated bacterial community richness and composition and determined the subsequent effects on VOC production and VOC-mediated pathogen suppression and plant growth-promotion. We assembled VOC-producing bacterial communities in different richness levels ranging from one to 12 strains using three soil-dwelling bacterial genera (Bacillus, Paenibacillus and Pseudomonas) and investigated how the composition and richness of bacterial community affect the production and functioning of VOCs. We found that VOC production correlated positively with pathogen suppression and plant growth promotion and that all bacteria produced a diverse set of VOCs. However, while pathogen suppression was maximized at intermediate community richness levels when the relative amount and the number of VOCs were the highest, plant growth promotion was maximized at low richness levels and was only affected by the relative amount of plant growth-promoting VOCs. The contrasting effects of richness could be explained by differences in the amount and number of produced VOCs and by opposing effects of community productivity and evenness on pathogen suppression and plant-growth promotion along the richness gradient. Together, these results suggest that the number of interacting bacterial species and the structure of the rhizosphere microbiome drive the balance between VOC-mediated microbe–pathogen and microbe–plant interactions potentially affecting plant disease outcomes in natural and agricultural ecosystems.

1. Introduction

Soil microbiome research has focused mainly on the beneficial effects of root-associated microbes that reside in the near vicinity of the plants. However, microbes also interact with each other and plants over long distances by producing volatile organic compounds (VOCs) that are a broad group of lipophilic compounds with low molecular weight (100–500 Da), high vapour pressure and low boiling point [1]. These properties facilitate evaporation and diffusion of VOCs over long distances through the atmosphere or porous soils from the point of production [2]. The VOCs have been reported for distinct bioactive functions, which are as diverse as the chemical structures of VOCs shaping a wide range of bacteria–bacteria and bacteria–plant interactions, including cell-to-cell communication, plant growth, flowering and photosynthesis stimulation, inhibition of parasites and pathogens and activation of systematic plant resistance against biotic and abiotic stresses [3–6]. The composition of the emitted VOCs can also vary depending on the environmental conditions such as the substrate composition of the growth media [7]. While several VOCs have been shown to change pairwise interactions with plants and microorganisms [8,9], it is less clear how the presence of other microbes in multi-species communities affects the production and functioning of VOCs. Here we studied this directly by manipulating bacterial community richness and composition and determining subsequent effects on VOC production and VOC-mediated pathogen suppression and plant growth promotion.

Biodiversity is a key driver of several ecosystem functions [10] and the underlying bacterial interactions have been shown to affect the number, type and composition of produced antifungal VOCs [3,11]. Bacterial community diversity could affect VOC production in many ways. First, multi-species communities could produce higher amounts and a greater number of VOCs by reaching higher cell densities compared with species grown in isolation due to complementary [12] or facilitative [13] effects. Alternatively, it is possible that high bacterial community diversity could lead to increased antagonism within the bacterial community, which could then offset the VOC production by having a negative effect on the growth and overall metabolism of the community [14]. Increasing community diversity could thus either promote or constrain VOC production depending on the species interactions between the interacting community members that could be driven by competition for shared resources, cooperation, cheating or antibiosis [2,15]. Second, increasing the number of species in a community could increase the number of unique VOCs that are produced if each species produces a different subset of compounds [16]. High community diversity could thus increase the range of VOC-mediated functions. Third, intra- and interspecific bacterial interactions could lead to the expression of certain ‘emergent’ VOCs that are not produced in monocultures. One potential mechanism for this could be interference competition which is often stronger in diverse bacterial communities due to the production of a high variety of antimicrobial compounds [17]. While co-culturing two to five bacteria together has been shown to induce the production of novel antifungal VOCs [2,18], the effects of diversity on bacteria-specific VOCs have not yet been explored.

Theory and experiments suggest that increasing community diversity and richness could predictably affect the production of VOCs by bacterial communities. However, it is still largely unknown how these changes affect the type and strength of VOC-mediated functioning with bacterial pathogens and plants. To address this shortcoming, we used a model plant-bacteria system to causally test how the microbial community richness affects the VOC-mediated functioning in terms of Arabidopsis thaliana plant growth-promotion and the suppression of a widespread bacterial pathogen, Ralstonia solanacearum, capable of infecting many plant species [19]. To achieve this, we assembled VOC-producing model bacterial communities in different richness levels ranging from one to 12 strains using three ubiquitous, soil-dwelling bacterial genera: Bacillus, Paenibacillus and Pseudomonas. We then determined and classified the emitted VOCs by all bacterial communities and explored how this variation affected plant growth-promotion and pathogen suppression as a function of bacterial community richness.

2. Methods

(a). Bacterial strains

We used a total of 12 common soil bacterial strains belonging to Bacillus, Paenibacillus and Pseudomonas genera, which were isolated from the rhizosphere of different plant species (four strains from each genus; for more detail, see electronic supplementary material, table S1). The bacterial strains were selected based on the preliminary experiments, where we tested that pathogen suppression and plant growth-promotion were solely mediated by VOCs (electronic supplementary material, table S1). The bacterial strains were stored at −80°C in nutrient broth (BD Difco, Becton, Dickinson and Company, USA) containing 70% glycerol and routinely grown on nutrient agar medium (Bacto agar, cat. no. 214030, Becton, Dickinson and Company, USA). We used the Ralstonia solanacearum QL-Rs1115 strain isolated in China [20] as our target pathogen, which was stored at −80°C in casamino acid-peptone-glucose (CPG) medium (1 g casamino acid (BD Bacto, Becton, Dickinson and Company, USA), 10 g peptone (Sigma-Aldrich), 5 g glucose (Sigma-Aldrich) and pH 7.0) containing 70% glycerol [21].

During the experiments, R. solanacearum was grown on CPG agar medium.

(b). Assembly of model rhizosphere bacterial communities

Single colonies of 12 bacterial strains (electronic supplementary material, table S1) were grown separately in nutrient broth as monocultures for 24 h at 30°C before washing twice and adjusting to the final concentrations of 1 × 10⁷ colony forming units (CFU)/ml with 0.85% NaCl. The monoculture cell suspensions of bacterial strains were mixed in equal proportions (500 µl) to assemble 43 model communities with varying diversity (strain richness) levels and composition ranging from monocultures to 2, 3, 4, 6 and 12 species communities (electronic supplementary material, table S2) using broken stick design [22]. The final cell concentrations of monocultures and mixed co-culture communities were set to the same (1 × 10⁷ CFU ml⁻¹). Each bacterial strain was replicated two times at each richness level except for richness levels 1 and 12. The assays for each model community were conducted in triplicate.

In order to verify whether all three bacterial genera could coexist, we grew all the assembled bacterial communities in microtiter plates. Each well was filled with 195 µl of modified minimal salt medium amended with 1.5% sucrose, and 0.4% tryptone soy broth (w/v) and inoculated with 5 µl of bacterial communities, thereby mimicking the conditions used for VOC measurements later in the experiment. After 36 h at 30°C, total bacterial, Pseudomonas and Paenibacillus cell densities were determined by serial plating on nutrient agar medium, Pseudomonas selective agar (CFC) medium and Paenibacillus selective nutrient agar medium supplemented with 10 µg ml⁻¹ polymyxin B sulphate, respectively [23,24]. Bacillus densities were determined by subtracting the Pseudomonas and Paenibacillus densities from the total bacterial densities. Plating method was chosen over the qPCR method to include only living cells to our analysis. Potential negative effects of selective plates on target bacteria were also confirmed: Paenibacillus and Pseudomonas genera were not negatively affected by the selective media as similar colony numbers were observed when the same samples were grown on nutrient agar medium (electronic supplementary material, figure S1). The bacterial cell densities were represented as community productivity at different bacterial richness levels.

(c). Measuring volatile organic compound-mediated pathogen suppression and plant growth-promotion by monocultures and communities

We assessed the VOC-mediated inhibitory potential of each bacterial monoculture and constructed community on R. solanacearum pathogen using divided Petri dish and soil systems. Briefly, a single colony of R. solanacearum was grown in CPG medium for 24 h at 30°C before washing twice with 0.85% NaCl and adjusting to a final concentration of 1 × 10⁷ CFU ml⁻¹. Later, one half of the divided Petri dish (85 mm diameter) was filled with 15 ml of CPG agar medium and spot-inoculated with the cell suspension of R. solanacearum two locations at 5 cm apart (5 µl in each; electronic supplementary material, figure S2). The cell suspensions for 43 model communities (1 × 10⁷ CFU ml⁻¹) were prepared as described above and spot-inoculated at 5 cm apart two locations (5 µl in each) on the other side of the Petri dish containing minimal salt agar medium (same as above but with 15 g agar per litre; electronic supplementary material, figure S2). Petri dishes were incubated at 30°C for 12 h to initiate bacterial growth and then sealed with Parafilm and incubated for further 3 days at 30°C. Three replicates were set up for each community, including negative control treatment with R. solanacearum growing in the absence of VOC-producing communities. Later, R. solanacearum colonies were removed along with agar medium using a sterilized scalpel, suspended in 1 ml of sterilized water, diluted by 500 times and spread on CPG agar plates to count the CFU ml⁻¹ (cell densities) after incubation at 30°C for 2 days. The VOC effects were presented as the percentage increase or decrease in the pathogen suppression relative to the control treatment. Moreover, in a separate experiment, the effect of VOCs produced by R. solanacearum on the growth of monocultures of Bacillus, Paenibacillus and Pseudomonas bacterial strains was also evaluated in triplicate using the same method as described above including negative control treatments with bacterial monocultures growing separately in the absence of VOC-producing R. solanacearum. These results showed that the VOCs of R. solanacearum were not able to inhibit the cell densities of any of the bacterial strains from Bacillus, Paenibacillus and Pseudomonas genera (electronic supplementary material, figure S3).

The Petri dish assays were validated using a sterilized soil system as follows [25]. The soil (pH 6.5, organic matter 11.65 g kg⁻¹, and available N, P and K contents 41.3, 238.7, and 177.5 mg kg⁻¹, respectively), were collected from Yixing, China, and sterilized 121°C for 60 min. One millilitre of each bacterial monoculture and community (1 × 10⁷ cells ml⁻¹) was mixed with 7.5 g of soil (dry weight) and inoculated to one half of the divided Petri dish. The other half of the dish was filled with CPG agar and spot-inoculated with R. solanacearum as described above (electronic supplementary material, figure S2). Three replicates were set up for each treatment including negative control with R. solanacearum in the absence of VOC communities. Dishes were incubated at 30°C for 12 h to initiate bacterial growth and then sealed with Parafilm and incubated for 3 days at 30°C. The VOC effects on the cell densities of R. solanacearum was quantified similarly as described above.

We used the A. thaliana plant model system to assess whether changes in microbial community richness and composition affected plant growth via changes in VOC composition. The Petri dish system was used in a similar way as described above in triplicate, including a negative control treatment where A. thaliana grew in the absence of VOC-producing bacteria. The cell suspensions of 43 model communities (1 × 10⁷ CFU ml⁻¹) were spot-inoculated on one side of the Petri dish as described above and incubated at 30°C for 12 h to initiate bacterial growth (electronic supplementary material, figure S2). Later, three Arabidopsis Col-1 seedlings were placed onto the other half of the Petri dish containing half-strength Murashige and Skoog agar medium (0.8% agar and pH 5.7). Before that, Arabidopsis seeds were surface sterilized, vernalized for 2 days at 4°C in the dark on half-strength Murashige and Skoog agar medium with 1.5% sucrose and then placed in a growth chamber (22°C temperature, 12 h light, 12 h dark, 40 W fluorescent light) for 3 days. The Petri plates were sealed with parafilm and placed in a growth chamber. After two weeks, plants were gently removed from the medium, roots washed with sterilized water and the whole plant was blot dried and weighted to determine the plant fresh weight (mg plant⁻¹). To determine VOC-mediated plant growth-promotion in the soil, a similar system was used as when evaluating VOC-mediated pathogen suppression in the soil except that the pathogen was replaced with three Arabidopsis seedlings inoculated onto half-strength Murashige and Skoog agar medium. After two weeks, plant fresh weight (mg plant⁻¹) was determined as described above. The VOC effects were presented as the percentage increase or decrease in plant growth relative to control treatment.

(d). Analysis of volatile organic compound profiles produced by bacterial strains and assembled communities

To analyse the VOC profiles produced by all bacterial monocultures and communities, cell suspensions (1 × 10⁷ CFU ml⁻¹) were prepared as described above and two spots (5 µl each) inoculated on minimal salt agar medium (15 g agar l⁻¹) in a 100-ml vial and placed at 30°C. After 12 h of growth, vials were sealed and incubated for further 72 h at 30°C. Three replicates were set up for each treatment and vials without the inoculation of bacteria were used as controls. After incubation, 10 µl of (Z)-3-hexenyl acetate (5 mM) as an internal standard was added into the vial. Later, a solid-phase microextraction (SPME) fibre (Supelco (Bellefonte, PA) stable flex divinylbenzene/carboxen/polydimethylsiloxane (DCP, 50/30 µm)) was inserted into the vial and incubated further 30 min at 30°C and another 30 min at 50°C. The SPME fibre was then inserted into the injector of gas chromatography-mass spectrometry (GC-MS) (Finnigan Trace DSQ, Austin, TX, USA) and desorbed at 220°C (1 min) with an RTX-5MS column (30 m, 0.25 mm, inside diameter, 0.25 µm). The following oven temperature protocol was used: 33°C (3 min), 180°C (10°C min⁻¹), and 240°C (30°C min⁻¹) and finally for 5 min at 240°C. The mass spectrometer was operated at 70 eV and 220°C in the electron ionization mode with a scan from 50 to 500 m/z. Chromatographs were obtained and analysed by AMDIS 2.73 (National Institute of Standards and Technology, Gaithersburg, USA). The mass spectra of deconvoluted VOC peaks were compared with those in the NIST/EPA/NIH Mass Spectrometry Library with respect to the spectra in the Mainlib and/or Replib databases (Agilent Technologies, Santa Clara, CA, USA). Later, the Kovats retention indexes were calculated for each compound using an alkane calibration mix and compared with those found in NIST/EPA/NIH Mass Spectrometry Library. The compound was considered identified if its mass spectra matched well with a listed compound, had match factor greater than 800 and the difference between the retention index calculated for the detected compound and the listed compound (for a semi-standard non-polar column) was not larger than five. Except for 14 unidentified and four commercially unavailable VOCs (electronic supplementary material, dataset S1), the production of 67 identified VOCs was further confirmed by comparing with standard compounds (Sigma, Tokyo Chemical Industry Co., Ltd. (TCI, Tokyo, Japan) and Aladdin Reagent Database, Inc. (Shanghai, China)). The standards were mixed and measured using SPME fibres as described above. The peaks similar to the control treatment (without bacterial inoculation) were not considered for the identification of VOCs. The number of VOCs produced in each treatment were recorded and the chromatographic peak area was expressed as the relative peak area to (Z)-3-hexenyl acetate (internal standard) in arbitrary units (a.u.) as an indirect approach to estimate the relative amount (concentration) of each VOC.

(e). Classification of emitted compounds into pathogen-suppressing and plant growth-promoting volatile organic compounds

To evaluate the effect of different concentrations of identified VOCs (GC-MS analysis) on pathogen suppression, the Petri dish system was used in a similar way as described above. The cell suspension of R. solanacearum (1 × 10⁷ CFU ml⁻¹) was spot-inoculated at two locations (5 µl in each) on one side of the Petri dish and incubated at 30°C for 12 h to initiate bacterial growth. Later, stock solutions (20 µg ml⁻¹, 100 µg ml⁻¹, 500 µg ml⁻¹, 2 mg ml⁻¹ and 10 mg ml⁻¹) of 67 commercially available pure VOCs (electronic supplementary material, dataset S1) were prepared separately in methanol by serial dilutions and the other side of Petri dish was inoculated with 15 µl of stock solutions to give 0.3 µg, 1.5 µg, 7.5 µg, 30 µg and 150 µg final amount of each VOC on an approximately 10 mm diameter sterile filter paper disc (Whatman filter paper, 6 µm pore size), respectively. Petri dishes were sealed with Parafilm and incubated for 3 days at 30°C. The sterile filter paper discs inoculated with nothing or with methanol were used as control treatments (no difference found between these control treatments). The VOC effects on the cell densities R. solanacearum was quantified similarly as described above [6,24].

To evaluate the effect of VOCs on plant growth, the same methodology described above was used, with one exception: instead of the pathogen, three Arabidopsis Col-1 seedlings were placed onto the other half of the Petri dish containing half-strength Murashige and Skoog salt agar medium. After two weeks, plant fresh weight (mg plant⁻¹) was determined as described above. The VOC effects were presented as the percentage increase or decrease in plant growth relative to control treatment [6,24].

(f). Statistical analysis

The statistical differences between bacterial strains and genera were analysed using ANOVA and Tukey's tests. Linear regression analysis was used to analyse separately the VOC-mediated pathogen suppression and plant growth-promotion, relative amount of VOCs (sum of relative peak area to (Z)-3-hexenyl acetate of detected GC-MS peaks), number of VOCs (number of peaks) and VOC composition (first axis of the principal component analysis on non-transformed data), and total community abundance, genera abundances and community evenness (at genera level) as the function of bacterial community richness (factor with six levels); significance at p = 0.05. Similarly, to link VOCs production with VOC-mediated activity, we separately analysed the VOC-mediated pathogen suppression and plant growth-promotion as the function of the relative amount of produced VOCs, number of VOCs and VOC composition; significance at p = 0.05. To further link VOC profiles and community properties to functioning, we used separate models to explain plant growth promotion and pathogen suppression with bacterial genera, community abundances and community evenness, community richness and strain identity effects, and relative amount, number and composition of VOCs. To uncover the most parsimonious GLMs with the best explanatory power, and to avoid potential correlations between different explanatory variables, sequential analyses were performed using stepwise model selection based on the Akaike information criterion (AIC). Statistical analyses were conducted with SPSS version 19.0 statistical software (SPSS, Inc., Chicago, IL, USA).

3. Results

(a). Production, classification and activity of pathogen-suppressing and plant growth-promoting volatile organic compounds by bacterial species and genera

All 12 bacterial strains were effective at VOC-mediated pathogen suppression and plant growth promotion, though some bacterial strains were more effective than the others on agar medium and/or in soil (electronic supplementary material, figure S4a,b). Overall, these effects were similar regardless if they were measured on agar media or in the soil (F_1,70 = 0.02, p = 0.891 for pathogen suppression and F_1,70 = 2.20, p = 0.143 for plant growth promotion). As a result, VOC-mediated pathogen suppression and plant growth promotion observed on agar media and in the soil were highly positively correlated (R² = 0.20; p < 0.0001 and R² = 0.61; p < 0.0001, respectively; electronic supplementary material, figure S5), which suggests that VOCs activity on agar media provided a realistic estimate of VOC activity in the natural soil. At the genera level, Paenibacillus showed relatively lower pathogen suppression (F_2,33 = 14.73, p < 0.0001) and Bacillus genera relatively lower plant growth-promotion on agar medium (F_2,33 = 28.01, p = 0.001; electronic supplementary material, figure S6a,b), while no between-genera differences were observed in the soil (electronic supplementary material, figure S6a,b).

We next compared the relative amount and number of VOCs produced by different bacterial genera and strains. We found that Paenibacillus genera produced higher relative amount (F_2,33 = 263.3, p < 0.0001) and number (F_2,33 = 61.8, p < 0.0001) of total VOCs compared with Pseudomonas and Bacillus genera, which did not differ from each other (electronic supplementary material, figure S6c,d). However, bacterial strains showed significant variation in the relative amount (F_11,24 = 357.2, p < 0.0001) and number (F_11,24 = 54.6, p < 0.0001) of produced VOCs within each genus (electronic supplementary material, figure S7a–f).

When VOC effects were tested as pure compounds, most of the produced VOCs had pathogen-suppressing activity (52%; electronic supplementary material, figure S6a,b) and only 7% had plant growth-promoting activity (electronic supplementary material, figure S6c), while both pathogen-suppressing and plant growth-promoting activities were increased with the increase in the concentration of VOCs (electronic supplementary material, figure S8). At the genera level, we found that in total 49 VOCs produced by Paenibacillus genera showed pathogen suppression, while Pseudomonas and Bacillus genera produced 33 and 40 pathogen-suppressing VOCs, respectively (electronic supplementary material, figure S6c,d and dataset S1). As a result, the relative amount (F_2,33 = 46.9, p < 0.0001) and the number of pathogen-suppressing VOCs (F_2,33 = 34.6, p = 0.001) were the highest with Paenibacillus genera (electronic supplementary material, figure S6c,d). In contrast, only eight Paenibacillus, eight Pseudomonas and five Bacillus VOCs showed plant growth-promotion (electronic supplementary material, figure S6c,d and dataset S1). While the highest relative amount of plant growth-promoting VOCs was produced by Bacillus genera (F_2,33 = 42.6, p < 0.0001; electronic supplementary material, figure S6c), Paenibacillus and Pseudomonas genera both produced the most diverse selection of plant growth-promoting VOCs (F_2,33 = 10.5, p = 0.011; electronic supplementary material, figure S6d). These results suggest that while all bacteria from each genus produced both types of VOCs, most of the produced VOCs had pathogen-suppressing effect and that the Paenibacillus genera showed the highest relative VOC production in general.

(b). Effect of bacterial community richness on the volatile organic compound-mediated pathogen suppression and plant growth-promotion

We next explored how bacterial community richness affected the VOC-mediated pathogen suppression and plant growth-promotion using agar media assays (quantitatively similar results obtained in the soil; electronic supplementary material, figure S9a,b). We found that bacterial community richness and pathogen suppression showed a hump-shaped relationship (F_2,126 = 90.4, p < 0.0001) where pathogen suppression peaked at the intermediate community richness (four species) reaching 40% suppression efficiency and then decreasing to 8% efficiency at richness level 12 compared with non-VOC control (figure 1a). This pattern could be explained well with the relative amount (F_2,127 = 58.18, p < 0.0001; figure 1b), number (F_2,126 = 67.7, p < 0.0001; figure 1c) and composition (F_2,126 = 13.68, p < 0.0001; electronic supplementary material, figure S10a) of produced pathogen-suppressing VOCs, which all showed a similar hump-shaped relationship peaking at richness level 4 and then decreasing at richness levels 6 and 12. Together, pathogen suppression showed highly significant and positive relationships with the relative amount, number and composition of pathogen-suppressing VOCs (figure 2a,b; electronic supplementary material, figure S10b and table S3).

Figure 1.

Effect of bacterial community richness on volatile organic compound (VOC)-mediated pathogen suppression (PS) and plant growth promotion (PGP) and on the relative amount and number of produced pathogen-suppressing and plant growth-promoting VOCs. Top panels show the effect of bacterial community richness on VOC-mediated pathogen suppression (a) and on the relative amount (b) and number (c) of pathogen-suppressing VOCs. Bottom panels show the effect of bacterial community richness on VOC-mediated plant growth promotion (d) and on the relative amount (e) and number (f) of plant growth-promoting VOCs. The relative amount of VOCs shows the chromatographic peak area that was expressed relative to the peak area of (Z)-3-hexenyl acetate (internal standard) as an indirect approach to estimate the relative concentration of each VOC, while number of VOCs means the total number of VOCs produced at each community richness level. In all panels, each observation shows the effect of each replicate of each bacterial monoculture or community. The experiments were repeated twice in triplicate. (Online version in colour.)

Figure 2.

The relationship of volatile organic compound (VOC)-mediated pathogen suppression (PS) and plant growth promotion (PGP) with the relative amount and numbers of pathogen-suppressing and plant growth-promoting VOCs, respectively, produced by bacterial communities at different richness levels. Top panels show the relationship between VOC-mediated pathogen suppression and the relative amount (a) and number (b) of pathogen-suppressing VOCs. Bottom panels show the relationship between VOC-mediated plant growth promotion and the relative amount (c) and number (d) of plant growth-promoting VOCs. The relative amount of VOCs shows the chromatographic peak area that was expressed relative to the peak area of (Z)-3-hexenyl acetate (internal standard) as an indirect approach to estimate the relative concentration of each VOC, while number of VOCs means the total number of VOCs produced at each community richness level. In all panels, each observation shows the effect of each replicate in each bacterial monoculture or community. The experiments were repeated twice in triplicate. (Online version in colour.)

In contrast, the highest plant growth promotion was observed at low community richness levels (F_1,127 = 13.8, p < 0.0001). Specifically, a 67% increase in plant growth promotion observed at the richness level 1 decreased to 17% increase at richness level 4, and at richness level 12, an average of 33% decrease in plant growth promotion was observed compared with control treatment (figure 1d). Reduction in the plant growth promotion correlated clearly with a decrease in the relative amount of plant growth-promoting VOCs (F_1,127 = 39.9, p < 0.0001; figure 1e) resulting in 90% decrease between richness levels 1 and 12. However, similar to pathogen-inhibiting VOCs, the number of plant growth-promoting VOCs peaked at intermediate richness levels reaching up to 139% increase at the richness level 4 and then decreasing down to 19% increase at the richness level 12 compared with the richness level 1 (F_2,126 = 56.1, p < 0.0001; figure 1f). The composition of plant growth-promoting VOCs did not show any relationship with plant growth promotion (electronic supplementary material, figure S10c). As a result, plant growth showed a highly significant and positive relationship only with the relative amount of plant growth-promoting VOCs (figure 2c,d; electronic supplementary material, figure S10D and table S3).

(c). Linking pathogen suppression and plant growth-promotion with the production of volatile organic compounds

We next investigated if VOC-mediated functioning could be explained by the emission of certain VOCs. A total of 85 different VOCs were produced by all bacterial communities. Except for three VOCs (1, 2-ethanediol 1, 2-diphenyl; 9-decen-i-ol and 5-octadecene), the relative amount of VOCs varied significantly between communities with different richness levels (electronic supplementary material, dataset S1). Interestingly, 15 VOCs were produced only in communities. Similarly, 49 VOCs produced at richness levels 1–4 were absent from the VOC profiles of six and 12 species communities (electronic supplementary material, figure S11A and dataset S1). Out of 85 VOCs in total, 41 VOCs showed pathogen-suppressing activity. Of these, four pathogen-suppressing VOCs were not produced at the community richness level 1, and 26 pathogen-suppressing VOCs produced at richness levels 1–4 were absent from the VOC profiles of six and 12 species communities (electronic supplementary material, figure S11b and dataset S1). When chemical groups of VOCs were evaluated, 80% (61) of the identified VOCs produced by 12 bacterial strains belonged to alkane, alcohol, aldehyde, benzene, ketone and fatty acid groups. Almost all alcohol, aldehyde, benzene and ketone group VOCs showed pathogen-suppressing activity. Other VOC groups related to pathogen suppression included naphthalene, phenol, sulfur and nitrogen containing compounds (electronic supplementary material, figure S12).

Only six out of 85 VOCs were found to show plant growth-promoting activity (electronic supplementary material, figure S8C). Of these compounds, four VOCs were not produced at richness level 12, while tetradecane was only produced at richness levels 6 and 12 albeit in low relative amount (electronic supplementary material, figure S11c and dataset S1). Interestingly, two of the plant growth-promoting VOCs (indole, heptadecane) also showed antibacterial activity against R. solanacearum (electronic supplementary material, figure S8a,b). When chemical groups of VOCs were evaluated, plant growth-promoting VOCs mainly belonged to the alkane (4) group; while one VOC belonged to the diol and one to the nitrogen-containing compounds group (electronic supplementary material, figure S12). These results suggest that bacterial interactions within communities can trigger and abolish the production of certain pathogen-suppressing and plant growth-promoting VOCs.

(d). Linking bacterial community properties with pathogen suppression and plant growth-promotion

Lastly, we explored if richness-mediated VOC effects could be explained by certain underlying community properties such as community productivity, evenness, genera abundances or strain identity effects. While the community productivity increased with bacterial richness (F_1,127 = 36.8, p = 0.004; figure 3a), the relative abundance of all three genera showed a parabolic relationship with the richness reaching the lowest abundances at the intermediate richness levels and the highest abundances when grown in the low or high richness level communities (figure 3b). Moreover, while the community evenness of bacterial genera did not differ at the lower richness levels (in two to four species communities), it considerably decreased at the higher richness levels (F_4,88 = 41.00, p < 0.0001; figure 3c). As a result, bacterial community properties showed contrasting effects on VOCs functioning; while total community productivity was positively linked with pathogen suppression, it showed a negative effect on the plant growth promotion (electronic supplementary material, table S4). In contrast, while community evenness had no effect on the pathogen suppression, it was positively linked with the plant growth promotion (electronic supplementary material, table S4). Furthermore, while the densities of Pseudomonas and Paenibacillus genera showed a negative relationship with pathogen suppression, the densities of all three genera showed positive effects on the plant growth promotion (electronic supplementary material, table S4). Finally, some strains had strong and often opposing identity effects on both the pathogen suppression and plant growth promotion (electronic supplementary material, table S4). These results suggest that bacterial community properties had contrasting effects on VOC-mediated functioning, which probably constrained simultaneous expression of pathogen-suppressing and plant growth-promoting VOCs.

4. Discussion

While the role of individual VOCs on plant physiology and antimicrobial activity has been well described [2,3], their production and effects in complex microbial communities are poorly understood. Especially, VOC-mediated effects on bacterial pathogens and plants remain unclear. Here we investigated this by addressing how the composition and richness of bacterial communities affect the production of different VOCs and VOC-mediated functioning in terms of pathogen suppression and plant growth promotion. We found that the majority of produced VOCs were pathogen-suppressing and that bacterial strains from all genera produced both types of VOCs in monocultures. However, VOC production was dramatically changed when the strains were grown together in communities. Specifically, we found that pathogen suppression was maximized at intermediate community richness levels when the relative amount and number of produced pathogen-suppressing VOCs were the highest. In contrast, plant growth promotion was unaffected by the number of VOCs and maximized at low community richness levels when the relative amount of produced plant growth-promoting VOCs was the highest. Interestingly, community productivity and evenness had contrasting effects on the VOC functioning in this study: productivity promoted the pathogen suppression but constrained the plant growth promotion, while evenness promoted the plant growth promotion but constrained the pathogen suppression. Together these results suggest that species interactions within communities can change VOC-mediated functioning by affecting the amount and diversity of produced VOCs. VOC-mediated microbe–microbe and microbe–plant functions are thus likely to be optimized with contrasting community structures due to nonlinear and contrasting relationships with community diversity, productivity and evenness.

Of all the detected VOCs, 41 VOCs (52%) showed pathogen suppression and their relative amount and numbers peaked at the intermediate community richness levels, which was highly correlated with VOC-mediated pathogen suppression. Moreover, compared with monocultures, 14 unique VOCs, including four pathogen-suppressing VOCs, were produced in more diverse bacterial communities including two to four strains. These results suggest that the addition of new species probably increased the metabolic potential of the community by stimulating the production of antimicrobial compounds with greater chemical diversity and activity [14,26]. However, the relative amount and number of pathogen-suppressing VOCs decreased at higher richness levels and 26 VOCs including 10 pathogen-suppressing VOCs were not observed at 12 strain bacterial community. These results are in line with a previous study, which found a similar hump-shaped pattern between toxin production and bacterial community richness [27]. Bacteria often sense and respond to the presence of competitors by turning more antagonistic by upregulating secondary metabolism and by producing antimicrobial compounds like antibiotics [28,29]. The secondary metabolism is also the main driver of antimicrobial VOC production that has been shown to change in the presence of competitors [17,30]. It is thus possible that the presence of other bacterial strains promoted the production of pathogen-suppressing VOCs because they were also used in interference competition between VOC-producing species [17]. Some previous studies have also reported a relationship between increased VOC-mediated suppression of fungal pathogens and increasing microbial diversity [11,31]. However, in this study, increasing community diversity beyond four strains could have intensified interference competition to the extent that it led to a decrease in the production of pathogen-suppressing VOCs. In addition, quorum sensing, cross-talk between species, chemical cues from competitors (antibiotics), silence gene clustering or cross-feeding generating new metabolic pathways at community levels might also affect the production of VOCs [14,27,29,32]. While linking community effects on certain species is difficult, we found that community evenness decreased with richness and that Paenibacillus genera dominated at the 12-strain community (figure 3b,c). Interestingly, Paenibacillus polymyxa WR-2 strain had a strong negative effect on pathogen suppression in general, which suggests that it might have played an important role in reducing VOC-mediated pathogen suppression at high richness levels (electronic supplementary material, table S4). We also found that community productivity had a positive relationship with pathogen suppression, indicative of a positive link between bacterial metabolic activity and VOC-mediated pathogen suppression. However, most pathogen-suppressing VOCs were produced at intermediate richness levels when all genera were found to be at very similar abundances. As a result, intra- and inter-bacterial species interactions might be more important for the expression of pathogen-suppressing VOCs instead of bacterial growth and metabolic activity.

Figure 3.

Effect of bacterial community richness on (a) community productivity (total bacterial abundance), (b) genera abundances and (c) community evenness based on bacterial genera abundances. In panels a and b, CFU denotes for bacterial cell numbers per ml in terms of colony forming units. In panel b, black, dark grey and light grey data points represent Pseudomonas, Bacillus and Paenibacillus genera, respectively. In all panels, each observation shows the effect of each replicate in each bacterial monoculture or community. The experiments were repeated twice in triplicate.

Of all 85 detected VOCs, only six showed plant growth-promoting activity (7% of all VOCs). Moreover, and in contrast to pathogen-suppressing VOCs, plant growth promotion was the highest in bacterial monocultures and steadily decreased with increasing community richness, turning into plant growth inhibition in a 12-strain community. While a clear positive correlation was found with the relative amount of VOCs and plant growth promotion, the numbers or composition of plant growth-promoting VOCs had no effect. This is probably explained by the low number of plant growth-promoting VOCs produced in general and by the fact that all genera tended to emit them similarly. Moreover, some of the plant growth-promoting VOCs were not detected at higher richness levels, which could also partly explain the reduction in VOC-mediated plant growth promotion along the richness gradient. One potential explanation for this pattern is that the presence of other bacteria triggered a switch from the expression of plant growth-promoting to pathogen-suppressing VOCs due to bacterial competition, which has previously been shown to upregulate antibacterial activity including VOC production [18,29,30]. Moreover, we found that the community evenness and the abundance of all genera promoted, while community productivity constrained the VOC-mediated plant growth promotion.

These results clearly show that bacterial interactions within multi-species communities can affect the VOC production, which in turn can change VOC-mediated functioning in terms of pathogen suppression and plant growth promotion. Furthermore, VOC-mediated microbe–pathogen and microbe–plant interactions were optimized with different community structures due to nonlinear and contrasting relationships with community diversity, productivity and evenness. These results suggest that VOC-mediated interactions in communities cannot be predicted based on VOC expression patterns observed in bacterial monocultures [33]. Our results are in contrast with several previous studies. For example, Wagg et al. [10] and Hu et al. [34] have reported positive relationships between microbial diversity and plant performance in communities containing four and eight microbes, respectively. It is thus possible that diversity–functioning relationships between soil bacteria and plants are less predictable, especially when mediated through VOCs. Moreover, soil is a complex and heterogenous environment, and in reality, rhizosphere bacterial communities are composed of thousands of interacting bacterial strains. Because analysing this many interactions at the same time is practically impossible, we used small model communities consisting of 12 bacterial strains belonging to three genera. Even though our model system does not reflect the natural soil conditions, it can help to understand how interspecies bacterial interactions can change the production and activity of VOCs. In the future, it would be interesting to study the underlying ultimate mechanisms like quorum sensing, cross-talk, chemical cues (antibiotics), silence gene clustering or cross-feeding driving the VOC production within the communities. Moreover, it would be interesting to explore how the VOCs produced in the soil affect the microbiota residing in the aerial parts of the plant (for example in leaves and flowers) that could affect pollination [35]. Our results also show that bacterial communities can interact with plants and plant pathogens over long distances through VOCs, and, crucially, that bacterial interactions within communities change their effects on plants or pathogens in the absence of direct contact. Thus, it is important to move beyond plant rhizosphere microbiomes to explore microbe–microbe–plant interactions over larger spatial scales that also include VOC-mediated long-distance interactions in porous soils [36]. For example, plant root VOCs were reported to disperse over 12 cm distances mediating long-distance belowground interactions in the soil [1], indicative of interactions between microbial metapopulations. From the applied perspective, our study suggests that VOC-mediated functions could potentially be employed to manipulate rhizosphere microbiome composition to simultaneously improve multiple ecosystem functions including pathogen suppression and plant growth.

Data accessibility

All data generated or analysed during this study are included in this article and its supplementary information files. The supplementary data are available from the Dryad Digital Repository: https://doi.org/10.5061/dryad.dbrv15dxn [37].

Authors' contributions

W.R., J.W., A.J., V.-P.F., Z.W., X.M., S.W. and Q.S. developed the ideas; W.R., A.J., V.-P.F. and Z.W. designed the study; W.R., J.W., X.M., S.W. and Z.W. set up the experiment; Z.W., J.W. and W.R. collected data; W.R. and Z.W. analysed the data and wrote the manuscript; A.J., V.-P.F., Z.W. and Q.S. provided comments on the manuscript.

Competing interests

The authors declare no competing interest.

Funding

The work was supported by the National Natural Science Foundation of China (grant numbers 31601835, 41671248, 41671256); National Key Basic Research Program of China (grant number 2015CB150503), the Fundamental Research Funds for the Central Universities (grant numbers KYT201802, KJQN201745), 973 project (grant number 2015CB150500) and Jiangsu Science and Technology Department (grant numbers BK20171373, BK20170085). V.-P.F. is supported by the Wellcome Trust (reference no. 105624) through the Centre for Chronic Diseases and Disorders (C2D2) and Royal Society Research Grants (RSG\R1\180213 and CHL\R1\180031) at the University of York, UK. A.J. is supported by the Nederlandse Organisatie voor Wetenschappelijk Onderzoek (ALW.870.15.050) and the Koninklijke Nederlandse Akademie van Wetenschappen (530-5CDP18).