Alleviation of water‐deficit inhibition of plant growth by rhizosphere microbiota conditioning

Summary

• Rhizosphere microbiota conditioning is a promising strategy to enhance plant growth. We conditioned the rhizosphere microbiota of Brassica juncea to water deficit to assess its impact on plant growth.

• In a glasshouse, plants were first grown under well‐watered conditions, then exposed to moderate (MD, pF = 2.3) or extreme (ED, pF = 3.5) water deficits. We extracted and inoculated the rhizosphere microbiota to new plants and repeated this process 10 times. Control plants were kept well‐watered. We monitored changes in plant phenotypes and in rhizosphere microbial communities (bacteria and eukaryotes).

• The initial water‐deficit growth inhibition of plants was successfully alleviated by 19.3% in MD and 29.4% in ED after conditioning (MD: from −35.6% to −16.3%; ED: from −56.8% to −27.4%). This beneficial effect on plants was not observed during the well‐watered phases, suggesting an active role of the microbiota when water became scarce. The increase in plant growth correlated with aggregated rhizosphere soil and significantly matched changes in the bacterial community, featuring reduced diversity and increased biofilm production capacity along the conditioning process.

• We showed that microbiota conditioning was a fast and efficient way to achieve better plant growth under adverse conditions, likely via the adaptation capabilities of the rhizosphere bacterial community.

Introduction

Climate change leads to a global temperature rise, possibly reaching +4.5°C by 2100 compared with the 1850–1900 period (Masson‐Delmotte et al., ²⁰²¹). This alters the distribution of rainfall (Pendergrass & Hartmann, 2014; Chen et al., ²⁰²¹), potentially increasing the frequency and intensity of droughts (Dai et al., ²⁰²⁰). Drought can significantly reduce agricultural yields (up to 90%; Giunta et al., ¹⁹⁹³). Plant breeding of drought‐resistant cultivars is one of the main strategies to address these challenges, but it faces development timelines that can be too long in the face of the rapid pace of global changes (Xiong et al., ²⁰²²). One alternative gaining increasing momentum for rapid plant adaptation is to rely on their beneficial association with microbial communities, and in particular the rhizosphere microbiota (Wallenstein, 2017; Jansson & Hofmockel, ²⁰²⁰). It is now recognized that the use of plant‐associated microbial communities has a significant potential for the short‐term adaptation of plants to environmental stressors (Trivedi et al., ²⁰²²). Unlike plants, microorganisms have important adaptation capacities driven by their short life cycles, their high genetic variability due to horizontal gene transfer, mutations, and rapid shifts in community structure (Lenski, 2017). Hence, they can effectively adapt and withstand environmental stresses, such as drought (Tan et al., ²⁰²²).

The rhizosphere microbiota comprises several microorganisms involved in beneficial effects on plant, including some that may enhance their tolerance to environmental stresses such as water deficit, via a range of well‐documented activities (de la Fuente Cantó et al., ²⁰²⁰). One of the microbial mechanisms improving plant tolerance to water deficit is biofilm production, a matrix made of extracellular polymeric substances mainly produced by bacteria, that improves their resistance to stress (Yang et al., ²⁰²¹). Biofilms improve soil aggregation and water/nutrient retention (Costa et al., ²⁰¹⁸), facilitating plant access to these resources (Ajijah et al., ²⁰²³). Interestingly, multispecies biofilm produce more matrix than monospecific ones (Madsen et al., ²⁰¹⁶; Liu et al., ²⁰¹⁹; Jacquiod et al., ²⁰²³). Thus, there is a real interest in studying the beneficial effects of multispecies biofilms on plants (Li et al., ²⁰²⁴). Still, most studies working on microbial‐mediated plant stress tolerance focus on the effect of single microbial species inoculation, despite the known limitations of such approaches (e.g. difficulties in colonizing and surviving; O'Callaghan et al., ²⁰²²). Hence, there is a real potential in harnessing microbial properties via the direct manipulation of entire rhizosphere microbiota for helping plants to face adverse conditions (Kong & Liu, 2022; Bell et al., ²⁰²⁵; Pradhan et al., ²⁰²⁵).

Direct manipulation of microbial communities can be achieved by experimental evolution via the iterative propagation of communities showing a property of interest (Blouin et al., ²⁰¹⁵; Raynaud et al., ²⁰¹⁹). This approach was applied to rhizosphere microbiota in order to improve plant phenotypes across multiple iterations. This can be performed using directed, human‐based selection of microbiota according to a specific plant phenotype target (e.g. flowering time: Panke‐Buisse et al., ²⁰¹⁵; leaf greenness: Jacquiod et al., ²⁰²²; salt tolerance: Mueller et al., ²⁰²¹; seedling drought tolerance: Jochum et al., ²⁰¹⁹). On the other hand, this can be performed without any human‐based selection, just by passing microbiota of plants to the following generations, either under adverse conditions (e.g. foliar plant pathogen resistance: Kalachova et al., ²⁰²²) or not (e.g. study of community assembly mechanisms: Morella et al., ²⁰²⁰). This approach is often referred to as ‘microbial conditioning’ (Monohon et al., ²⁰²¹). These experiments are often performed under controlled conditions in which the environment (glasshouse conditions with the same soil) and plant genotype (all seeds originate from the same initial production batch) are kept constant. Hence, any plant phenotype alteration may potentially come from changes in the microbial community structure and/or modifications in the way the rhizosphere microbiota interacts with its host plant. Nevertheless, a better understanding of the changes in plant–microbiota interactions and associated microbiota functions is needed to improve the reliability of the approach (Yu et al., ²⁰²³; Thomas et al., ²⁰²⁴).

In this study, we applied a water‐deficit conditioning of the rhizosphere microbiota, via repeated exposure of Brassica juncea to different intensities of deficit (moderate deficit or MD, pF = 2.3; extreme deficit or ED, pF = 3.5). We iteratively transferred the rhizosphere microbiota to a new set of plants (the so‐called generations) to evaluate the impact of this procedure on plant growth during the water‐deficit exposition. To ensure statistical robustness, we replicated three independent lineages for each water‐deficit modality (experimental design; Supporting Information Fig. S1). In a control, plants were kept well‐watered and did not received inoculations. We hypothesized that the rhizosphere microbial community would rapidly change due to the water‐deficit conditioning, with favorable consequences for plant growth, in particular the leaf surface area, compared with the well‐watered control. We expected that this effect could be mediated through belowground plant traits (e.g. root biomass), or traits resulting from the interactions between plant and rhizosphere microbiota (amount of aggregated rhizosphere soil around roots and biofilm production). Finally, we expected a more pronounced response of the bacterial community than other microbial groups of the rhizosphere microbiota (e.g. fungi), as they are reported to be sensitive to drought (Meisner et al., ^{2018, 2021}; Sarkar et al., ²⁰²²).

Materials and Methods

Soil

We used a sandy Cambisol with moor (CEREEP Ecotron research station, sampled December 2019, Saint‐Pierre‐Lès‐Nemours, France, 48°16′58.9″N, 2°40′19.0″E, composition: organic carbon = 14.7 g kg⁻¹, total nitrogen = 1.19 g kg⁻¹, sand = 74.1%, pH = 5.22, silt = 19%, clay = 6.9%), already used as a model soil by several laboratories (Milcu et al., ²⁰¹⁸). The sandy texture avoids cracks during drought and facilitates rhizosphere soil recovery with minimal root damaging. The soil was prepared following previous procedures (Wei et al., ²⁰²¹; Jacquiod et al., ²⁰²²). Briefly, soil was air‐dried at room temperature, sieved (4 mm), and autoclaved (115°C, 45 min, one drying cycle) before each plant generation. The purpose was not to sterilize the material but rather to limit batch effects between plant growth iterations and facilitate the establishment of inoculated communities. The soil was allowed to rest for 72 h to allow the removal of volatile compounds. The soil was used to fill up pots (350 g of dry soil per pot) and rewetted before seedling transplantation (60% soil gravimetric water content (GWC)).

Seedlings

Brassica juncea (L.) Czern., 1859, cv Etamine (brown mustard), was chosen for its sensitivity to water deficit (Saha et al., ²⁰¹⁶), especially in the selected soil (Thiour Mauprivez et al., ²⁰²³). All seeds used in the entire experiment came from a pure lineage of the same accession to avoid intraspecific variability and maximize the observation of microbiota‐driven effects on plant phenotype. They were germinated in a sealed translucid plastic box with wet blotting paper in the dark (18°C, 48 h, Fitoclima 600 PL/PLH; Aralab, Rio de Mouro, Portugal) before exposure to full light (21°C, 72 h). Homogeneous seedlings were transplanted individually into pots and placed into a glasshouse compartment. The starting point of the experiment was set by inoculating all seedlings of the first generation with the same initial microbial community, which was obtained from pooled rhizospheres of several B. juncea plants that were previously grown in the same soil (Fig. S1).

Growth conditions and water supply

Pots were placed in an automated glasshouse compartment (4PMI, Plant Phenotyping Platform for Plant and Micro‐organism Interactions, UMR Agroécologie Dijon, France), enabling automatic watering and randomization of plants four times per day. Plant growth was set in two distinct phases (Fig. S1): (1) a ‘normal watering phase’ for 2 wk (light intensity = 500 W m⁻²; day length = 16 h; day : night temperature = 22°C : 20°C; randomization and automatic watering at 60% GWC; no fertilization) followed by (2) a ‘water‐deficit phase’ for 2 wk in which water supply was suppressed until the soil GWC reached the setpoints at pF 2.3 and 3.5 (MD and ED, respectively), and thereafter maintained at these water levels. The setpoints were previously defined using water retention curves on the same soil (Thiour Mauprivez et al., ²⁰²³). The MD corresponded to a GWC of 12% (pF = 2.3; Fig. S2), and the ED to a GWC of 9% (pF = 3.5; close to the 4.2 wilting point; Fig. S2). In a well‐watered control group, 10 plants were maintained at 60% GWC during the whole growth period for each generation (hereafter ‘WW’), without microbiota inoculation (Fig. S1).

Conditioning of the rhizosphere microbiota

Three independent microbial lineages were initially set for each treatment (MD: MDL1, MDL2, and MDL3; ED: EDL1, EDL2, and EDL3) ensuring replication of our iterative passing (Fig. S1). Each lineage consists of a population of 20 plants, from which a random draw of three plants was carried out to prepare the inoculants for the next generation (sample function, R Core Team, ²⁰²³). The random selection procedure, adapted from the group selection literature, mimics random extinction events (Wade, 1976). The root systems were collected by carefully dismantling the pots and gently shaking off the nonadhering soil. We standardized the procedure by weighing the same amount of rhizosphere material obtained from all plants, according to the lighter root system (Jacquiod et al., ²⁰²²). Within each lineage, the three root systems were pooled and mixed with 200 ml of sterile water with a magnet bar stirring to generate an aqueous solution (30 min). The solution was used to inoculate 20 new seedlings of the next generation of the same lineage (5 ml per seedling). The remaining solution was centrifuged, the pellet was mixed with 3 ml sterile water, split in two, and stored for sequencing (−20°C) and long‐term storage (−80°C, 1 ml of 80% glycerol v/v). The rhizosphere pools of each generation are hereafter referred to as ‘inoculants’.

Sampling of individual rhizosphere microbiota

To match changes in individual plant phenotypes with their respective rhizosphere microbiota, six individual rhizospheres were sampled within each lineage of the MD and ED treatments at the first generation (G1) and from generation six onwards (G6–G10). This additional sampling at the end of the experiment was motivated by previous results with the same soil, showing a correlation between changes in the rhizosphere microbial community and plant phenotype after several generations (Jacquiod et al., ²⁰²²). Entire rhizospheres, sampled as described previously, were placed in 50‐ml Falcon tubes with 10 ml of sterile water, and vortexed (Vortex Genie 2, medium speed, 5 min). Two milliliters of homogeneous slurries were sampled and stored for sequencing (−20°C).

Plant traits acquisition

At the end of both the normal watering phases (d14) and the water‐deficit phases (d28), camera‐based phenotyping of all plant shoots was performed (Fig. S1) as described by Jacquiod et al. (2022). Several morphological traits were calculated based on plant images including maximum height, maximum width, convex hull area, convex hull perimeter, leaf density, projected leaf area, and projected leaf area perimeter. We assessed the impact of water deficit on traits by calculating the differences between values recorded at the end of the normal watering and water‐deficit phases (Δ = d28 − d14). We focused on the projected leaf surface area, whose reduction is known to be one of the first symptoms of water deficit (Farooq et al., ²⁰⁰⁹). Plant traits and their definitions are available as Dataset S1. At the last generation (G10), belowground traits were measured, including the amount of soil aggregated around the roots, and the fresh and dry root and shoot biomasses. The amount of aggregated rhizosphere soil was measured by weighing the entire system including both roots and adhering soil, and subtracting the fresh root weights obtained from cleaned root systems after rhizosphere soil sampling (see ‘Sampling of individual rhizosphere microbiota’ in the Materials and Methods section).

Biofilm production capacity

We selected microbial inoculant pools of one lineage in MD (MD‐L2) and one in ED (ED‐L2) for further in vitro estimation of biofilm production capacity. Due to technical limitations, only one lineage in each water‐deficit treatment could be used. We selected inoculants before (G1), in the middle (G4), and at the (G9) of the water‐deficit conditioning, giving a total of six pooled inoculants (G1‐MD‐L2, G4‐MD‐L2, G9‐MD‐L2, G1‐ED‐L2, G4‐ED‐L2, and G9‐ED‐L2). Briefly, 100 μl of the thawed inoculant glycerol stocks was mixed with 10 ml M408 medium, adapted for rhizobacteria growth (Bai et al., ²⁰¹⁵). For each inoculant, eight technical replicates of 100 μl were distributed in a 96 multiwell plate. One column was filled with M408 medium as a negative control. The plate was incubated (25°C, 96 h, no light; Stabilitherm Incubator; Thermo Fischer Scientific, Illkirch, France), emptied, and cleaned twice with physiological water (Well Wash Versa 96; Thermo Fischer Scientific). One hundred microliters of 0.05% crystal violet (CV) dye was added to stain surface biofilms in each well for 45 min. The excess CV dye was removed, and the plate was cleaned with physiological water four times. The CV dye that actually stained biofilms was solubilized (200 μl 96% ethanol, 4 h) and quantified by spectrometry (absorbance at 590 nm; Infinite M200 Pro; Tecan, Männedorf, Switzerland).

Analysis of soil microbial communities

Inoculants and rhizosphere samples were thawed on ice. Excess water was removed by centrifugation. Soil was weighed and resuspended in water to normalize soil slurry concentrations. Nucleic acids were extracted from 250 μl of normalized soil suspensions using the DNeasy PowerSoil HTP 96 kit (Qiagen). DNA concentrations were measured with the Quant‐iT™ PicoGreen™ kit (Thermo Fischer Scientific), with an average DNA yield of 3.49 ± 2.15 ng μl⁻¹. DNA extracts were stored at −20°C. Amplicons from the V3–V4 hypervariable regions of the 16S rRNA gene (primer pair: 341F (5′‐CCTACGGGRSGCAGCAG‐3′) and 805R (5′‐GACTACCAGGGTATCTAAT‐3′)) and the Internal Transcribed Spacer (ITS2) region (primer pair: ITS3F (5′‐GCATCGATGAAGAACGCAGC‐3′) and ITS4R (5′‐TCCTCSSCTTATTGATATGC‐3′)) were sequenced to analyze bacterial and eukaryote communities, respectively. A first PCR was performed (98°C for 3 min followed by 98°C for 30 s, 55°C for 30 s, and 72°C for 30 s; 25 cycles for 16S and 30 for ITS; final extension 10 min, 72°C) followed by a second PCR for sample indexing (98°C for 30 s, 55°C for 30 s, and 72°C for 30 s; 8 cycles for 16S and 10 for ITS; final extension 10 min, 72°C). Amplicon sizes were checked by electrophoresis, and their concentrations were normalized before pooling (SequalPrep™; Invitrogen). Sequencing was performed on a MiSeq platform (2 × 250 bp; Illumina, Paris, France), using the MiSeq v.2 reagent kit (500 cycles). The Casava software (Illumina) was used for demultiplexing and removing Illumina adapters and barcodes. Bioinformatics analysis was carried out using an in‐house Python pipeline (Spor et al., ²⁰²⁰). Since Amplicon Sequence Variant (ASV) assignment is not yet optimal with ITS (Estensmo et al., ²⁰²¹), we used an operational taxonomic unit (OTU)‐based pipeline to analyze our 16S rRNA and ITS2 sequences. Identity thresholds were based on a standard control community (OTU 16S rRNA = 97%; ITS2 = 97%). Taxonomic assignment for bacteria was performed using UCLUST (Edgar, 2010) and the SILVA database (v.138.1; Quast et al., ²⁰¹³), while for eukaryotes, taxonomic assignment was performed using Blast and the UNITE reference database (v.7‐08/2016; Abarenkov et al., ²⁰¹⁰). Potential contaminations from the host plant genome and mitochondrial DNA were removed by Blast homology (> 96% identity). To have a broader view of the soil eukaryote community, we kept all ITS OTUs in the profiles.

Statistical analysis

Statistics were performed using R 4.2.0 (R Core Team, ²⁰²³) in Rstudio, following the recommended best practices for microbial community analysis (Schöler et al., ²⁰¹⁷). As advised (Schloss, 2024), the sequencing depth was adjusted to the minimum level in each dataset through random resampling (16S/ITS inoculant datasets: n _min = 5800/7000; 16S/ITS individual rhizosphere datasets: n _min = 9300/5000; vegan, Oksanen et al., ²⁰²²). Alpha diversity analyses were conducted using the observed species richness (S) and the Shannon index (H). The normality and homoscedasticity of variables were assessed using Shapiro and Bartlett tests. Variance partitioning of normal data was performed with ANOVA followed by Tukey's honest significant difference post hoc test (agricolae; de Mendiburu, ²⁰²³), or the Scheirer–Ray–Hare test for non‐normal data (rcompanion; Mangiafico, ²⁰²³) followed by false discovery rate (FDR) P‐value adjustment. Temporal trends were analyzed using linear regressions. Microbiota structures were analyzed using distance‐based redundancy analysis (vegan, Bray–Curtis dissimilarity; Oksanen et al., ²⁰²²). Microbial taxa showing significant abundance changes during the water‐deficit conditioning were assessed by multiple correlation tests (Pearson's rho, FDR‐adjusted P < 0.05, |r| > 0.6; Dataset S2). Differential OTU abundance between treatments were identified using nonrarefied data through the quasi‐likelihood F‐test based on negative binomial distributions and generalized linear models (nb‐GLM, abundance fold‐change > 2, overall occurrence > 20 times, and FDR‐adjusted P‐value < 0.05, edgeR; Robinson et al., ²⁰¹⁰). The correspondence between the microbial community structure and plant traits was assessed using the sparse partial least squares discriminant analysis with blocks (block sPLS‐DA, mixomics; Rohart et al., ²⁰¹⁷). The method relies on the mutual discrimination of samples in both the plant trait dataset and the microbial community datasets using principal component analyses, followed by correlation tests between samples coordinates on the components to detect a significant association between datasets. Four separate block sPLS‐DA models were created: two for bacteria (MD and ED) and two for eukaryotes (MD and ED). Only OTUs whose abundance showed sufficient positive or negative correlations with the projected leaf area (|r| > 0.4) were considered.

Results

Impact of microbiota conditioning on the projected leaf surface area

We evaluated plant growth during the conditioning process. Leaf area being our proxy to assess the effects of water deficit changes on plants, only the projected leaf surface area is shown in the main text. In the following text, ‘plant growth’ will thus refer to the changes in projected leaf surface area. Other plant traits acquired through image analysis are shown in Fig. S3. To account for seasonal effects that occurred during the whole conditioning experiment (G1 started in June 2020 and G10 ended in February 2021, thus plants received more sunlight at the beginning than toward the end), the data were standardized using the z‐score method (centered and scaled using the mean and standard deviation of the well‐watered control for each generation; Fig. 1). The raw projected leaf surface area is presented in Fig. S4. Before microbiota conditioning, plants suffered a loss in projected leaf surface area growth of −35.6% in MD and −56.8% in ED, relative to the control (growth during the 2 wk of exposition to water deficit; Fig. 1a). After the microbiota conditioning, the negative impact of water deficit on plant growth was alleviated by −16.3% in MD and −27.4% in ED, relative to the control (Fig. 1a). Across generations, during the normal watering phases (the first 2 wk before deficit), rhizosphere microbiota manipulation had no effect on the projected leaf surface area, in neither MD nor ED treatments (P > 0.05, Fig. 1b). However, when looking at the evolution of plant growth after the 2 wk of water‐deficit exposition across generations, a significant positive effect of rhizosphere microbiota conditioning was observed for both MD (R ² = 0.37, P < 0.001) and ED (R ² = 0.46, P < 0.001; Fig. 1c). A model comparison analysis revealed that the projected leaf surface slope was significantly higher in ED than in MD (P < 0.05; Fig. 1c).

Fig. 1.

Growth of plant projected leaf surface area inoculated with rhizosphere microbiota exposed to repeated moderate (MD) and extreme (ED) water deficit. The data were standardized using z‐scores relative to the well‐watered plant controls of each generation (WW, dotted blue lines, raw data in Supporting Information Fig. S4). Panel (a) shows the amount of growth during the deficit phase (Δ) before (G1) and after (G10) of the conditioning process (G1: n _MD = 57; n _ED = 53; n _WW = 10; G10: n _MD = 60; n _ED = 56; n _WW = 10). Boxplot whiskers are representing the first and third quartiles, while dots are representing outliers based on the interquartile range. Panels (b, c) show the evolution of the projected leaf surface area growth for the first 2 wk before deficit exposition (b) and after the 2 wk of deficit exposition (c) across the 10 generations (n _MD = 553; n _ED = 516; n _WW = 93). The yellow and red lines represent linear models for MD and ED, respectively (MD model: y = 0.34x − 3.98; ED model: y = 0.48x − 6.01). The dashed blue lines represent the values of WW controls. Statistically significant differences are indicated by the different letters above the boxplots. Significance code: *, P < 0.05; **, P < 0.01; ***, P < 0.001. Plant species used: Brassica juncea (L.) Czern., 1859, cv Etamine.

Microbial conditioning effect on above‐ and belowground plant traits

We looked at the consequences of the rhizosphere microbiota conditioning on above‐ and belowground plant traits at the last generation G10 (Fig. 2). Plants of the ED treatment had lower dry shoot biomass than that of WW and MD (Fig. 2a). Rhizosphere microbiota conditioning had no effects on dry root biomass (Fig. 2b). The amount of aggregated rhizosphere soil was significantly increased for plants inoculated with MD microbiota compared with the others, while no differences were observed for plants inoculated with ED microbiota compared with WW (Fig. 2c). The aggregated rhizosphere soil per gram of dry root was higher in MD and ED treatments than in WW (Fig. 2d). A significant linear relationship between the aggregated rhizosphere soil and plant growth was observed during the water‐deficit phase (P < 0.001; Fig. 2e).

Fig. 2.

Above‐ and belowground plant traits measured at generation G10 after microbiota conditioning (a–e) and evolution of in vitro biofilm production capacity of inoculants during the conditioning process (f, g). Panels (a–d) show the dry shoot biomass (a), the dry root biomass (b), the mass of fresh soil aggregated to the roots (c), and the normalized mass of aggregated fresh soil per gram of dry root (d) for the moderate (MD) and extreme (ED) water‐deficit treatments and the well‐watered control (WW) at the last generation (G10). Statistical differences were inferred with ANOVA followed by a post hoc Tukey's honest significant difference test (P < 0.05) for normally distributed data (a, c), and multiple Kruskal–Wallis tests under false discovery rate P‐value adjustment (FDR‐adj P < 0.05) for non‐normal data (b, d). Boxplot whiskers are representing the first and third quartiles, while dots are representing outliers based on the interquartile range. Panel (e) shows the linear relationship between the amount of aggregated fresh soil on the roots and the growth of leaves during water deficit (WW = the amount of growth for the same time period), assessed with a linear mixed‐model effect (projected leaf surface area ~ aggregation + (deficit|lineage) = , n _MD = 30; n _ED = 30; n _WW = 9, y = 21.28x + 584.94). Panels (f, g) are showing the in vitro biofilm production capacity of the culturable fractions of the pooled inoculants of lineage MD‐L2 (f) and ED‐L2 (g), from generations G1, G4, and G9. The significance was verified using Kruskal–Wallis tests (FDR‐adj P < 0.05). Statistically significant differences are indicated by the different letters above boxplots. Plant species used: Brassica juncea (L.) Czern., 1859, cv Etamine.

Biofilm production capacity of the conditioned rhizosphere microbial community

To gain insight into potentially relevant microbial functions altered during the water‐deficit conditioning, we measured the biofilm production capacity of the inoculants in vitro. We selected one lineage per treatment (MD‐L2 and ED‐L2) to measure biofilm production in vitro on inoculants from the first generation (G1), a midterm generation (G4), and the second last generation (G9, since no inoculants were produced at G10). Biofilm production of inoculant pools was increased gradually during the conditioning process in both water‐deficit modalities (P < 0.001; Fig. 2f,g). For each tested generation, the water‐deficit intensity did not change the amount of biofilm production (P > 0.05).

Effect of water‐deficit intensity conditioning on the bacterial community

We looked at the effect of water‐deficit conditioning on the bacterial communities from the inoculants (Fig. 3). A significant decrease in observed richness and Shannon index along generations was found for both treatments (P < 0.001; Fig. 3a,b). These effects were robust when testing the three lineages separately (P < 0.001). Although the richness decreased at the same rate for both treatments (from 545 ± 7 to 454 ± 10 OTUs in MD and from 527 ± 28 to 428 ± 19 OTUs in ED), the overall averaged richness was higher in MD (511 ± 58) than in ED (478 ± 57, P < 0.05; Fig. S5a). Similarly, the Shannon index decreased at the same rate for both water‐deficit modalities (from 4.8 ± 0.1 to 4.4 ± 0.02 in MD and from 4.7 ± 0.14 to 4.3 ± 0.02 in ED), but the overall averaged Shannon index was higher in MD (4.7 ± 0.2) than in ED (4.6 ± 0.3, P < 0.05; Fig. S5b). The bacterial community structure was significantly altered (Adj. r ² = 0.39, P < 0.001; Fig. 3c,d), with an important conditioning effect between generations (45.1% of the variance explained, P < 0.001; Fig. 3d). Changes in bacterial community structure were more important at the beginning (G1–G5) than at the end (G6–G9) (Fig. 3d). This stabilization of the bacterial community structure was confirmed, as the beta dispersion of inoculant pools after G5 was significantly lower than before G5 (P < 0.01; Fig. S6a). A strong reproducibility between lineages was observed in terms of bacterial community structure for both MD and ED (P > 0.05; Fig. S7a,b). Proteobacteria were the most abundant taxa, followed by Bacteroidetes, Firmicutes, and Patescibateria (Fig. 3c). The abundance of taxa belonging to Proteobacteria and Bacteroidetes increased after G5, while some belonging to Firmicutes decreased (Spearman correlations, adj‐P < 0.05, |r| > 0.6; Dataset S2). The effect of water‐deficit intensity was significant, but limited (3.8%, P < 0.001; Fig. 3c,d). No interaction between generations and water‐deficit intensity was observed (P > 0.05).

Fig. 3.

Effect of water‐deficit conditioning on the rhizosphere bacterial community of the inoculants. Panels (a, b) show the changes in the observed richness (S) and Shannon index (H), respectively (significant linear models are represented by the colored lines). Panel (c) shows the changes in community composition at the phylum level (average between the three lineages). Panel (d) represents a distance‐based redundancy analysis of bacterial community structure (model: Bray–Curtis dissimilarity ~ conditioning time × deficit intensity, permutations = 10 000). In the panel (d), the colored dotted lines connect the different generations together in chronological order. The adjusted r ² values indicate the percentage of variance explained by the models. If significant, constrained coordinates are displayed (model P < 0.05; CAP: constrained analysis of principal coordinates; n = 54, with three lineages per generation per deficit modality). Significance code: *, P < 0.05; **, P < 0.01; ***, P < 0.001. MD, moderate water deficit; ED, extreme water deficit; G1–G9, conditioning generations. Plant species used: Brassica juncea (L.) Czern., 1859, cv Etamine.

Effect of water‐deficit intensity conditioning on the eukaryote community

For the eukaryotes, observed richness significantly increased during the conditioning exclusively in the ED condition (P < 0.01; Fig. 4a), going from 526 ± 57 observed OTUs at G1 to 643 ± 13 at G9. The Shannon index also significantly increased, still only in the ED condition (P < 0.001; Fig. 4b), going from 3.3 ± 0.3 at G1 to 4.1 ± 0.4 at G9. These effects were mainly driven by one lineage (EDL2, P < 0.01) while the others did not show any significant change during conditioning (P > 0.05). Hence, the averaged observed richness and Shannon index did not differ between the two water‐deficit treatments (Fig. S5c,d). The eukaryote community structure was also significantly altered during the experiment (Adj. r ² = 0.24, P < 0.001; Fig. 4c,d), with an important conditioning effect between generations (31.1%, P < 0.001; Fig. 4d). No stabilization of the community structure was observed during the experiment, as the beta dispersion was similar before and after G5 (P > 0.05; Fig. S6b). Concomitantly, a significant difference between the three independent lineages was observed (MD: 18.5%, P < 0.001; ED: 14.1%, P < 0.05; Fig. S7c,d), indicating that the eukaryote communities have diverged during the experiment (as seen by the high lineage dispersion; Fig. 4d). The community was dominated by Chytridiomycota, Ascomycota, and Chromista (Fig. 4c). Ciliophora also increased at the beginning, and then decreased after G5. The abundance of taxa belonging to Chytridiomycota decreased, while some from Ascomycota increased (Spearman's correlations, adj‐P < 0.05, |r| > 0.6; Dataset S2). The effect of water‐deficit intensity was significant, but limited (5.3%, P < 0.001; Fig. 4c,d). No interaction between generations and water‐deficit intensity was observed (P > 0.05).

Fig. 4.

Effect of water‐deficit conditioning on the rhizosphere eukaryote community of the inoculants. Panels (a, b) show the changes in the observed richness (S) and Shannon index (H), respectively (significant linear models are represented by the colored lines). Panel (c) shows the changes in community composition at the phylum level (average between the three lineages). Panel (d) represents a distance‐based redundancy analysis of eukaryote community structure (model: Bray–Curtis dissimilarity ~ conditioning time × deficit intensity, permutations = 10 000). In the panel (d), the colored dotted lines connect the different generations together in chronological order. The adjusted r ² values indicate the percentage of variance explained by the models. If significant, constrained coordinates are displayed (model P < 0.05; CAP: constrained analysis of principal coordinates; n = 54, with three lineages per generation per deficit modality). Significance code: *, P < 0.05; **, P < 0.01; ***, P < 0.001. MD, moderate water deficit; ED: extreme water deficit; G1–G9, conditioning generations. Plant species used: Brassica juncea (L.) Czern., 1859, cv Etamine.

Concomitant changes in plant traits and the microbial communities

We examined individual plant rhizosphere microbiota at the end of the experiment (G6 : G10) compared with the starting point (G1; Fig. S8). Significant effects were detected on the bacterial communities (Adj. r ² = 0.17, P < 0.001; Fig. S8a), mostly due to the conditioning (14.97%, P < 0.001), and marginally due to the water‐deficit intensity (2.48%, P < 0.001). Fifty‐two bacterial taxa displayed abundance change due to the intensity of the water deficit (Fig. S9). Significant effects were also detected in the eukaryote communities (Adj. r ² = 0.11, P < 0.001; Fig. S8b), still mostly due to the conditioning (8.94%, P < 0.001), and marginally due to the water‐deficit intensity (2.93%, P < 0.001). Twenty‐one eukaryote taxa (mostly fungi) displayed significant abundance change due to the intensity of the water deficit (Fig. S10).

Next, we identified microbial taxa whose abundance was associated with changes in projected leaf surface area using block sPLS‐DA. Using data from individual plants sampled from G6 to G10, we constructed several block sPLS‐DA models to explain covariations in the rhizosphere microbiota dataset and the plant trait dataset, for each water‐deficit intensity, respectively. Bacteria–plant MD and ED models exhibited strong correlations between the two component ones (MD R ² = 0.72, P < 0.001; ED R ² = 0.78, P < 0.001), both yielding a low classification error rate (< 10%; Fig. S11a,b) with accurate positioning of samples based on the generation membership using both datasets (Fig. S11c,d). Since the classification error rates of the eukaryote models were too high (> 50%), only bacterial models were used for further analysis. We extracted two bacterial subcommunities: one in MD featuring 49 OTUs and one in ED featuring 38 OTUs, whose abundances were correlated with higher projected leaf surface area (Fig. 5a,b). A stronger correlation was found in ED (R ² = 0.38, P < 0.001) compared with MD (R ² = 0.22, P < 0.001). The abundance of these positive bacterial subcommunities increased over generations, ranging from 5% to 10% at G1 up to 50% in MD and 40% in ED at G10 (Fig. 5a,b). Thirty‐one OTUs were specific to MD, 20 were specific to ED, and 18 were shared. For both deficit levels, this subcommunity was primarily composed of OTUs from the Massilia, Brevundimonas genus and the Burkholderiaceae family, accounting for the vast majority of sequences in both MD (94.08%; Fig. 5c) and ED (89.61%; Fig. 5d). It also included minor OTUs from Flavobacterium (Bacteroidetes) and Altererythrobacter. The most abundant OTU was from Massilia, whose abundance was < 10% at G1 and went up to 49% in MD and 32% in ED at G10.

Fig. 5.

Bacterial subcommunities associated with the increased projected leaf surface area in each water‐deficit intensity level. Panels (a, b) respectively show the linear relationship between the relative abundance of the bacterial subcommunities and the gains in projected leaf surface area of plants during the water‐deficit phase (n _MD = 105; n _ED = 101). The pie charts in panels (c, d) are showing the taxonomic compositions and relative abundance of the taxa in the bacterial subcommunities of each water‐deficit intensity level at G10, respectively (relative abundance in the sub‐community, not in the entire community). ED, extreme water deficit; G01–G10, conditioning generations; MD, moderate water deficit; WW, well‐watered control. Significance code: *, P < 0.05; **, P < 0.01; ***, P < 0.001. Plant species used: Brassica juncea (L.) Czern., 1859, cv Etamine.

Discussion

Conditioning of the rhizosphere microbiota promotes plant growth during water deficit

We hypothesized that rhizosphere microbiota conditioning via repeated exposition to water deficit could lead to a significant improvement of plant growth due to fast microbial adaptation. Our results could support this hypothesis, as we showed that the repeated rhizosphere microbiota exposition to water deficit led to the gradual increase in B. juncea growth during the deficit phases in MD and ED treatments. In only one year of conditioning, we obtained rhizosphere microbiota alleviating the effects of water deficit on plant growth in a consistent and reproducible manner in all lineages. The alleviation went from −35.6% to −16.3% in MD (+19.3%) and from −56.8% to −27.4% in ED (+29.4%). This illustrates how plant‐associated microbiota represent a real opportunity to help plant face environmental stresses in the short term (Trivedi et al., ²⁰²²). We demonstrated via amplicon sequencing that the rhizosphere microbial community structure has changed during the conditioning. This was particularly true for the bacterial community, whose alpha diversity was significantly reduced, while a specific subcommunity dominated by a Massilia OTU (Betaproteobacteria) gradually increased in abundance. This observation may be interpreted as a sign of selection due to the whole conditioning process, including the tangled effects of the water deficit and the repeated inoculations. This is in line with a previous artificial microbiota selection study on wheat seedlings, showing a significant delay in drought symptoms matching the selection of a bacterial community featuring a reduced diversity and an increased abundance of Betaproteobacteria (Jochum et al., ²⁰¹⁹). Likewise, another study showed the selection of beneficial bacterial species for plants growing in soils with a legacy of repeated drought (Gebauer et al., ²⁰²²). The effects we observed on plant growth became consistent after the fourth generation, concurring with a stabilization of the bacterial community structure starting after the fifth generation. This is consistent with previous work, showing that community stability is paramount for reproducibility of effects associated with selected rhizosphere microbiota (Jochum et al., ²⁰¹⁹; Jacquiod et al., ²⁰²²).

Conditioned microbiota increased aggregated rhizosphere soil and biofilm

We did not observe any change in projected leaf surface area of our plants inoculated with the conditioned microbiota during the 2 weeks before application of the water deficit. The microbial effect on plant growth was only detectable when the water became scarce, implying that it was actively triggered by the deficit. Hence, the conditioning process may have altered the microbial functional response to water deficit, potentially via changes in regulatory mechanisms, as often observed in experimental evolution on microbes (Long et al., ²⁰¹⁵). We expected that the conditioned communities will have an effect on belowground plant traits, potentially explaining the increase in aboveground traits. Based on the literature, microbial effects on root biomass and architecture via the production of hormones may be suspected (Eichmann et al., ²⁰²¹; Gholizadeh et al., ²⁰²⁴). Although we did not investigate root architecture, we did not observe any stimulation of root biomass by conditioned microbiota compared with the control. Interestingly, we found that the amount of fresh aggregated rhizosphere soil per gram of root was significantly higher in plants that received the conditioned microbiota than in the well‐watered control. This suggests that our conditioned microbiota may have a role on rhizosphere soil aggregation, or by modifying the way plant aggregates the soil in the root area. Hence, we suspected that biofilm formation might be involved, as it is known to affect soil aggregation (Costa et al., ²⁰¹⁸).

Microbial communities are producing biofilm to protect themselves against unfavorable conditions (Liu et al., ²⁰¹⁹), especially when water becomes scarce (Costa et al., ²⁰¹⁸). Biofilms, which are made of polymers with a high affinity for water (e.g. polysaccharides) are known to increase soil water retention (Li et al., ²⁰²⁴). A remarkable aspect of biofilm made by complex microbial communities is that they tend to produce more matrix and biomass through the establishment of complex interspecific interactions, as opposed to monospecific biofilms (Madsen et al., ²⁰¹⁶; Liu et al., ²⁰¹⁹; Jacquiod et al., ²⁰²³). This community property makes multispecies biofilm interesting for potential applications, such as the alleviation of plant drought stress (Yang et al., ²⁰²¹). However, investigating biofilm in situ is extremely challenging, especially in soil. Here, we have demonstrated via in vitro approaches that the amount of biofilm produced from the cultivable fraction of the rhizosphere community has significantly increased during the conditioning process. This shows that biofilm was likely one of the functional responses of the microbiota to the whole conditioning process, although we could not distinguish whether it was a consequence of the water deficit, an effect of the iterative inoculations, or both. Bacterial lifestyle decisions between motility and biofilm production are known to be regulated by secondary messengers (e.g. cyclic di‐GMP; Ueda & Wood, ²⁰⁰⁹), whose concentration in the cell is controlled by the sensing of environmental cues, such as stressing conditions (Römling et al., ²⁰¹³). Hypothetically, this could explain why no plant growth promotion was observed before the water deficits were applied, as bacteria might have started producing biofilm only after sensing that water started to become scarce during the deficit phases. A beneficial effect of biofilm production by a microbial community in the rhizosphere was already observed in Arabidopsis thaliana under drought conditions, leading to the alleviation of the stress (Yang et al., ²⁰²¹). The effect of biofilm on plant growth promotion is receiving increasing attention in the scientific community, as it has been often overlooked (Li et al., ²⁰²⁴). However, since no differences in biofilm production were observed between the two water‐deficit modalities, and since we did not measure biofilm production in the well‐watered controls, it was not possible to confirm that this increase was solely due to the water deficit, and not an effect of the conditioning procedure itself. Additional controls would be required to detangle the effect of water deficit from that of the repeated inoculation process. Still, the fact that the effects of the microbial conditioning became observable on plants only after the water deficit was applied (and not before) suggests that exposure to the water deficit is an important determinant.

Thanks to block sPLS‐DA, we were able to short‐list the bacterial subcommunity that was selected as the projected leaf surface area of plants increased during the conditioning process. The dominant OTU in this subcommunity belonged to the Massilia genus (Burkoldericaeae), which contains representatives with (1) phosphorous solubilization and plant growth promoting effect (Ofek et al., ²⁰¹²; Zheng et al., ²⁰¹⁷; Xiao et al., ²⁰²²), (2) auxin biosynthesis genes (Holochová et al., ²⁰²⁰), and (3) increased abundance in the rhizosphere of plants experiencing stresses, such as drought (Vescio et al., ²⁰²¹). Similarly, a species from the same family (Burkholderia sp. ADR17) was involved in water stress resistance in A. thaliana and Zea mays, notably via evapotranspiration reduction, antioxidant alterations, and influencing phytohormone levels (Huang et al., ²⁰¹⁷). In addition, inoculation of Burkholderia phytofirmans strain PsJN also increased growth and physiological status of maize under drought stress (Naveed et al., ²⁰¹⁴). The second most abundant OTU belongs to Brevundimonas, whose representatives were found in relation to plant drought tolerance (Tran et al., ²⁰²³). Other members included Altererythrobacter and Flavobacterium, both associated with enhancing drought resistance in plants (Gaete et al., ²⁰²¹; Kim et al., ²⁰²³). However, to the best of our current knowledge, no representatives of these taxa were previously identified in relation to biofilms in the rhizosphere. Additional work going beyond the scope of this study would be required to clearly identify the microbial taxa potentially involved in the in situ production of biofilm and other functions potentially involved in the beneficial effect observed on plant growth.

The water‐deficit intensity had limited effects on the microbiota, but not on the plant

During the deficit phase, we decided to maintain the GWC below strict levels, carefully chosen based on the water retention curves of this soil (Thiour Mauprivez et al., ²⁰²³), to generate pF values above 2.3 in MD and 3.5 in ED (the wilting point being at 4.2). Our distance based‐RDA results showed that water‐deficit intensity had significant, but very marginal effects on the microbial community. This was confirmed by the block sPLS‐DA, as both MD and ED subcommunities had very similar compositions. Finally, the deficit intensity had no consequences on the aggregated rhizosphere soil per gram of dry root, nor on the in vitro biofilm production capacity of the community. However, the relative alleviation of plant growth inhibition before and after conditioning was 10% higher in ED (+29.4%) than in MD (+19.3%), with a significantly more pronounced positive slope during the whole conditioning process in ED. Thus, while the microbial community has similarly adapted to both levels of water‐deficit intensities during the experiment, it seems that plants facing harsher conditions have benefited relatively more. To the best of our knowledge, the fact that the microbial alleviation of plant stress is modulated by the intensity of the stress itself was never evidenced and may deserve more attention. Indeed, it may have an importance for potential applications if inoculants efficiency depends on the stress intensity, and if these inoculants can be used in a range of different stress intensities.

Bacteria showed a clear and reproducible response to the conditioning

There is a large body of evidence in the literature suggesting that, overall, soil bacteria are very sensitive to drought (Meisner et al., ^{2018, 2021}; Sarkar et al., ²⁰²²). The main explanation is that bacteria, which are mostly aquatic organisms, need water for their nutrition, motility/dispersion, and communication through soluble molecules. This also explain why bacteria produce biofilms when water become scarce. We thus expected that the microbial conditioning would have more impact on the bacterial community than on the overall microbial eukaryote community. Although both bacterial and eukaryote communities showed a significant response to the water‐deficit conditioning, the sPLS‐DA revealed a better match between the variations of the plant traits and those in the bacterial community, hence confirming our initial assertion. Highly reproducible results were obtained among the three independent lineages in each water‐deficit modality for the bacterial community, featuring (1) a diversity reduction that was more pronounced in ED, (2) a stabilization of the community structure concurring with the observed beneficial effects on plant growth, and (3) a significant positive correlation between the abundance of a bacterial subcommunity and the observed beneficial effects on the plant during the conditioning. Regarding the eukaryote communities, we obtained different community structures among lineages (14–19% variance). We suspected that, unlike bacteria, eukaryote communities have been potentially more exposed to stochastic processes resulting in a drift effect among lineages. Similar variations among lineages for the eukaryote community were already observed in an artificial selection experiments on rhizosphere microbiota (e.g. 16% of the variance; Jacquiod et al., ²⁰²²), potentially due to the sampling and inoculation procedure.

To conclude, although we could not tell apart the effects of water deficit from those associated with the experimental procedure, we showed that iterative exposure of a rhizosphere microbiota to water deficit has rapidly conditioned the bacterial community, leading to the alleviation of plant growth inhibition. This work brings important elements toward the elaboration of new strategies based on the direct manipulation of rhizosphere microbiota for harnessing potentially beneficial plant–microbiota interactions that could increase plant tolerance to environmental stressors in the short term.

Competing interests

None declared.

Author contributions

SJ and MB conceptualized the study. SJ, RS and ML conducted the glasshouse experiments, plant picture acquisitions and analysis. RS and SJ collected the samples. CT‐M performed the molecular biology steps. OC and DG performed the biofilm analysis. VA analyzed the data, prepared the figures and wrote the manuscript. SJ and MB edited and reviewed the text. All authors agree with the content of the manuscript.

Disclaimer

The New Phytologist Foundation remains neutral with regard to jurisdictional claims in maps and in any institutional affiliations.

Supporting information

Dataset S1 Plant traits data obtained from image analysis during the conditioning process.

Dataset S2 Effect of the conditioning process on microbial operational taxonomic units (Pearson's correlation).

Fig. S1 Workflow of the rhizosphere microbiota conditioning to water deficit.

Fig. S2 Relationship between the gravimetric water content and the pF of the soil.

Fig. S3 Image analysis of morphological plant traits before and after water deficit exposition during the conditioning process.

Fig. S4 Effect of water‐deficit conditioning of the rhizosphere microbiota on leaf surface.

Fig. S5 Effect of the water‐deficit conditioning on the microbial alpha diversity of the pooled inoculants.

Fig. S6 Effect of the water‐deficit conditioning on microbial community structure stabilization.

Fig. S7 Conditioning reproducibility among microbial lineages.

Fig. S8 Effect of the water‐deficit conditioning on the community structure of individual plant rhizosphere microbiota.

Fig. S9 Discriminant bacterial operational taxonomic units between the two water‐deficit treatments.

Fig. S10 Discriminant eukaryote operational taxonomic units (OTUs) between the two water‐deficit treatments.

Fig. S11 Robustness of the bacterial block sparse least square discriminant analysis models.

Please note: Wiley is not responsible for the content or functionality of any Supporting Information supplied by the authors. Any queries (other than missing material) should be directed to the New Phytologist Central Office.

Data availability

The raw sequencing data were deposited in the Sequence Read Archive public repository (SRA, https://www.ncbi.nlm.nih.gov/sra) under the following accession nos.: 16S rRNA gene amplicon datasets (MD: PRJNA1121099; ED: PRJNA1121096) and ITS amplicon datasets (MD: PRJNA1121984; ED: PRJNA1121981). Plant trait data are provided in Dataset S1. The effect of conditioning on each individual OTU is provided in Dataset S2.