Probiotic Diversity Enhances Rhizosphere Microbiome Function and Plant Disease Suppression

Increasing Pseudomonas community genotypic richness correlated positively with community niche breadth (R² = 0.776, P < 0.0001, Fig. 1A), overlapping of the niche with the pathogen (R² = 0.709, P < 0.0001, Fig. 1B), and direct pathogen inhibition (R² = 0.389, P < 0.0001, Fig. 1C) in vitro.

Both disease incidence and pathogen density decreased significantly with increasing Pseudomonas community richness (Fig. 2A and B and Table 1). While all Pseudomonas monocultures reduced disease incidence to some extent, they offered only partial protection against bacterial wilt disease. In contrast, the 8-strain community provided almost complete protection against bacterial wilt, and 2- and 4-strain communities provided intermediate levels of protection (Fig. 2A and Table 1). The effect of Pseudomonas community richness on disease suppression increased with time (Fig. 2B, significant richness × time interaction, and Table 1): while community richness had no effect on disease suppression during the first 15 days after pathogen invasion, the 8-strain Pseudomonas community reduced pathogen density by 99% compared to the best performing monoculture on day 35 (Fig. 2B and Table 1).

At the initial stage, all Pseudomonas communities were able to colonize plant roots equally well regardless of the community diversity. However, only the 8-strain Pseudomonas communities were able to maintain high population densities in the rhizosphere throughout the whole experiment (Fig. 2C, significant richness × time interaction, and Table 1), reaching densities ca. 10 times higher that those seen with the most productive single-strain community at the end of the experiment (indicative of transgressive overyielding [23]) (see Table S1 in the supplemental material). Interestingly, none of the Pseudomonas strains showed a particularly strong identity effect on pathogen suppression (Table S1). This suggests that high Pseudomonas community richness increased its ability to colonize the rhizosphere microbiome due to synergistic effects between community members instead of inclusion of one particularly efficiently colonizing Pseudomonas strain.

We found that Pseudomonas community survival in the rhizosphere increased with the increasing niche breadth of the community, while pathogen density correlated negatively with the increasing inhibition activity of Pseudomonas communities measured in vitro (Table 2). Pathogen invasion success in the rhizosphere depended also on the density of the Pseudomonas community (Table 2). We used a structural equation modeling (SEM) approach to further study the relative levels of importance of different mechanisms linking Pseudomonas community composition to disease suppression. The final models fit the data well (both P > 0.05) and explained 72% of the variance in pathogen density and 37% of the variance in disease incidence at day 35 of the experiment (Fig. 3). Pathogen density decreased with in vitro antagonistic activity against the pathogen, higher strain richness, and wider niche breadth of the Pseudomonas communities. Accordingly, disease incidence decreased with increasing richness of the Pseudomonas communities.

We used eight fluorescent pseudomonad strains (CHA0, PF5, Q2-87, Q8R1-96, 1M1-96, MVP1-4, F113, and Phl1C2) as described previously (20) (for more information, see Table S2 in the supplemental material). All strains were stored at −80°C. Prior to the experiments, a single colony of each strain was selected randomly, grown overnight in lysogenic broth (LB), washed three times in 0.85% NaCl, and adjusted to an optical density at 600 nm (OD₆₀₀) of 0.5 using a spectrophotometer (Spectra Max M5; Molecular Devices, Sunnyvale, CA). We used Ralstonia solanacearum strain QL-Rs1115 (race 1 and biovar 3) as a pathogen. This strain was originally isolated from a tomato rhizosphere in Qilin (118°57′E, 32°03′N), Nanjing, China, is highly virulent, and is able to cause wilting of tomato, eggplant, pepper, and potato (13).

We created 48 communities by the use of eight different Pseudomonas strains, which we combined following a substitutive design as described previously (19) to obtain initial richness levels of 1, 2, 4, and 8 strains (Table S3). The diversity gradient was assembled so that each strain was drawn randomly, allowing disentangling the effects of strain identity and community diversity. We used a substitutive design so that the total biomass of every Pseudomonas community inoculant was kept the same in all treatments but the proportion of every single strain decreased with increasing community richness (100%, 50%, 25%, and 12.5% for 1-, 2-, 4-, and 8-strain communities, respectively).

In order to link biodiversity effects to bacterial resource competition, we assessed the resource use of the eight Pseudomonas species and R. solanacearum strains on 48 different single-carbon resources (Table S4) representative of tomato root exudates (5). Briefly, bacteria grown overnight in tryptic soy broth (TSB; tryptone at 15 g liter⁻¹, soy peptone at 5 g liter⁻¹, NaCl at 5 g liter⁻¹) were pelleted by centrifugation (4,000 × g, 3 min) and washed three times in 0.85% NaCl before their growth was measured on 96-well microtiter plates containing Os minimal medium (37) supplemented with a 10 mM concentration of a single resource representative of amino acids, organic acids, and sugars found in tomato root exudates (5). We used a total of 48 different single compounds as listed in Table S4. All microplate wells were inoculated with equal amounts of the specified bacterial mixtures (starting OD₆₀₀ = 0.05) and incubated for 48 h with agitation (170 rpm) at 30°C. Optical density (600 nm) was recorded at regular intervals with a spectrophotometer (Spectra Max M5; Molecular Devices, Sunnyvale, CA). Community-level resource use metrics were characterized using two indices, the niche breadth index and the niche overlap index, defined as the number of resources consumed by the Pseudomonas communities and the proportion of each resource used by R. solanacearum and the Pseudomonas community, respectively. Wells with an OD₆₀₀ greater than 0.05 were scored as representing positive growth on any given substrate.

In order to link biodiversity effects to direct inhibition of the pathogen, we quantified the pathogen growth in the presence of Pseudomonas supernatants. To avoid biases due to competition or facilitation between different Pseudomonas strains, we grew all eight Pseudomonas strains individually in nutrient broth for 30 h (30°C, 170 rpm), after which cells were pelleted by centrifugation (4,000 × g, 3 min). Cell-free supernatants were then mixed in proportions matching the diversity gradient of the communities (1-, 2-, 4-, and 8-strain richness levels; Table S3), and inhibition experiments were started immediately. Briefly, 20 μl of supernatant mix was added to a fresh culture (180 µl, OD₆₀₀ = 0.05) of the pathogen R. solanacearum in modified standard mineral salt agar (M-SMSA) media (38). Control treatments received 20 µl M-SMSA media. Bacteria were grown for 24 h (30°C, 170 rpm) before bacterial densities were measured as optical density at 600 nm using a spectrophotometer (Spectra Max M5 plate reader; Molecular Devices, Sunnyvale, CA). Pathogen inhibition was defined as the percentage of reduction in pathogen growth compared to pathogen growth in the control treatment.

The biocontrol efficiency of Pseudomonas bacterial communities was assessed in a 50-day-long greenhouse experiment (an overview of the protocol is presented in Fig. S1 in the supplemental material). The soil was collected from a tomato field in Qilin, a town of Nanjing, China (13), sieved at 5 mm, and homogenized. Please note that the homogenized soil contained the natural microbial community. We used the same 48 Pseudomonas community combinations as were used in the in vitro experiments (Table S3). Surface-sterilized tomato seeds (Lycopersicon esculentum, cultivar “Jiangshu”) were germinated on water-agar plates for 3 days before being sown into seedling plates containing cobalt-60-sterilized seedling substrate (Huainong, Huaian Soil and Fertilizer Institute, Huaian, China). Germinated tomato plants were transplanted to seedling trays containing natural, nonsterile soil at the three-leaf stage of growth (12 days after sowing). Twenty-four seedlings were transplanted into one seedling tray with 8 cells, each of which contained 500 g soil planted with three seedlings. Each tray was treated as one biological replicate. Two replicate seedling plates were used for all communities (and four replicate plates for a positive control). After 10 days of growth, plants were inoculated with Pseudomonas communities by the use of root drenching methods with a final concentration of 5.0 × 10⁷ CFU of bacteria g⁻¹ soil (39). After 5 days postinoculation of Pseudomonas communities, the pathogen R. solanacearum was inoculated at a final concentration of 10⁶ CFU of bacteria g⁻¹ soil. Tomato plants were then grown for 35 days in a greenhouse (with natural temperature variation ranging from 25°C to 35°C) and watered regularly with sterile water. Disease incidence per seedling plate was used as a disease index (13). Seedling plates were rearranged randomly every 2 days. Disease progression was monitored daily after the pathogen inoculation. The experiment was terminated 35 days after pathogen inoculation when all the plants given the positive-control treatment showed symptoms of wilting.

We performed destructive sampling to estimate pathogen and introduced Pseudomonas abundances 5, 15, 25, and 35 days after the pathogen inoculation. We removed two randomly chosen plants per community from one of the replicate seedling plates (total of 416 rhizosphere samples) at every time point. Rhizosphere soil was collected by first gently removing the plants from the pots before shaking off excess soil and collecting the soil attached to the roots. Samples were stored at −80°C for DNA extraction. Microbial DNA was extracted using a Power Soil DNA isolation kit (Mo Bio Laboratories, Inc., Carlsbad, CA) following the manufacturer’s protocol. DNA quality was checked by running samples by the use of 1% sodium boric acid agarose gel electrophoresis, and DNA concentrations were determined by using a NanoDrop 1000 spectrophotometer (Thermo Scientific, Waltham, MA). Extracted DNA was stored at −80°C for bacterial density analyses.

We used quantitative PCR (qPCR) to quantify the abundance of the introduced Pseudomonas bacteria and the pathogen in the rhizosphere soil. Pseudomonas bacterial density was estimated with primers B2BF (5′-ACC CAC CGC AGC ATC GTT TAT GAG C-3′) and B2BR3 (5′-AGC AGA GCG ACG AGA ACT CCA GGG A-3′) targeting the phlD gene (40), which is part of the phl operon responsible for the synthesis of the broad-spectrum antibiotic 2,4-diacetylphloroglucinol (DAPG). We use this gene as a reference because it is shared by all of the Pseudomonas strains that we used while being present at only a low background concentration in the reference soil (the background level is shown in all figures as a red dashed line). Pathogen density was quantified by using specific primers (forward, 5′-GAA CGC CAA CGG TGC GAA CT-3′; reverse, 5′-GGC GGC CTT CAG GGA GGT C-3′) targeting the fliC gene coding the flagellum subunit (41). The qPCR analyses were carried out with an Applied Biosystems 7500 real-time PCR system (Applied Biosystems, CA) using SYBR green I fluorescent dye detection in 20-μl volumes containing 10 μl of SYBR Premix Ex Taq (TaKaRa Bio Inc., Japan), 2 μl of template, and 0.4 μl of both forward and reverse primers (10 mM each). The PCR was performed by initially denaturizing at 95°C for 30 s, cycling 40 times with a 5-s denaturizing step at 95°C following a 34-s elongation/extension step at 60°C, and ending with melt curve analysis at 95°C for 15 s, at 60°C for 1 min, and at 95°C for 15 s. Each sample experiment was replicated three times.

For in vitro experiments, we used generalized linear models (GLM) to test whether Pseudomonas community richness affects niche breadth, niche overlap with the pathogen, and direct pathogen inhibition.

Data were analyzed in three ways. First, we used separate GLMs expressing disease incidence as well as pathogen and Pseudomonas community abundances as a function of the interactive effects of time and Pseudomonas community richness. Bacterial abundance data were log₁₀ transformed and disease incidence data square arcsine-transformed prior to analysis. Second, we attempted to link the dependent variables to changes in the characteristics of the Pseudomonas community, including resource competition metrics (niche breadth and niche overlap), direct pathogen inhibition (toxicity), and Pseudomonas community density in the rhizosphere. Due to potential correlations between different explanatory variables, a sequential analysis was used to uncover the most parsimonious GLMs. To this end, we used stepwise model selection based on Akaike information criteria (AIC) to choose the model with the best explanatory power [step () function in R]. We used both backward elimination starting with the full model and a forward-selection model (from simple to full model) to avoid selecting a local AIC minimum (42). Finally, we used structural equation modeling (SEM) to shed light on the mechanisms of disease incidence in tomato plants by accounting for multiple potentially correlated effect pathways. SEM analysis was chosen because it can disentangle the direct and indirect effects (43) of diversity and community characteristic parameters in vitro for determining the survival of Pseudomonas communities, the pathogen density in tomato rhizosphere, and the disease incidence in the greenhouse experiment. The initial model was based on previous knowledge (44) for assigning the exogenous variable “richness” and the endogenous variables “niche breadth,” “niche overlap,” “toxin production,” “Pseudomonas density,” “pathogen density,” and “disease incidence.” Due to the relatively low level of replication and the complex structural equation model, we ran separate models for “pathogen density” and “disease incidence.” The adequacy of the models was determined via chi-square tests, AIC, and root mean square error of approximation (RMSEA) (44). Model modification indices and stepwise removal of nonsignificant relationships were used to improve the models; however, only scientifically sound relationships were considered (43). Structural equation modeling was performed using Amos 5 (Amos Development Corporation, Crawfordville, FL).