Engineering multifunctional rhizosphere probiotics using consortia of Bacillus amyloliquefaciens transposon insertion mutants

While bacterial diversity is beneficial for the functioning of rhizosphere microbiomes, multi-species bioinoculants often fail to promote plant growth. One potential reason for this is that competition between different species of inoculated consortia members creates conflicts for their survival and functioning. To circumvent this, we used transposon insertion mutagenesis to increase the functional diversity within Bacillus amyloliquefaciens bacterial species and tested if we could improve plant growth promotion by assembling consortia of highly clonal but phenotypically dissimilar mutants. While most insertion mutations were harmful, some significantly improved B. amyloliquefaciens plant growth promotion traits relative to the wild-type strain. Eight phenotypically distinct mutants were selected to test if their functioning could be improved by applying them as multifunctional consortia. We found that B. amyloliquefaciens consortium richness correlated positively with plant root colonization and protection from Ralstonia solanacearum phytopathogenic bacterium. Crucially, 8-mutant consortium consisting of phenotypically dissimilar mutants performed better than randomly assembled 8-mutant consortia, suggesting that improvements were likely driven by consortia multifunctionality instead of consortia richness. Together, our results suggest that increasing intra-species phenotypic diversity could be an effective way to improve probiotic consortium functioning and plant growth promotion in agricultural systems.


Introduction
Bacterial rhizosphere microbiome diversity has been strongly associated with beneficial effects on plant growth in both natural and agricultural environments (1,2).Such beneficial diversity effects are often associated with increased functional diversity and community multifunctionality (3)(4)(5)(6)(7), where different taxa play complementary roles in plant growth-promotion, specializing in nutrient solubilization, plant immune priming or for example by interacting with beneficial or pathogenic rhizosphere microbes (8,9).Moreover, high microbial diversity can have other benefits such as providing functional redundancy in the case of species extinctions (10), community stability (11) or provide emergent consortia-level properties that cannot be predicted based on the sum of individual species (12)(13)(14)(15).While several studies have tried to harness positive effects of bacterial diversity for crop production (16)(17)(18), diverse inoculants often fail to produce the desired assembling consortia from strains that use different niches or specialize in different functions within the same niche (42,43).This might allow more efficient expression or production of different compounds at the consortia level and help to overcome any antagonistic pleiotropic effects experienced at the individual strain level (i.e., trade-offs between plant growth-promotion traits).Moreover, interactions between near clonal specialist genotypes could allow division of labor, where shared workload by two specialists leads to higher productivity relative to one generalist as has been demonstrated with Bacillus subtilis bacterium during biofilm matrix production (42).Such overperformance (or overyielding) could also be driven by other mechanisms such as resource partitioning, abiotic facilitation, or biotic feedbacks (44), leading to higher consortia functioning than predicted based on the sum of its individual members.
Here, we used a combination of genetics, molecular biology and biodiversity-ecosystem functioning theory (3,11) to test if increasing phenotypic diversity of a single plant growth-promoting Bacillus amyloliquefaciens T-5 bacterium offers a viable strategy to improve bioinoculant consortia multifunctionality.
We chose B. amyloliquefaciens T-5 strain as our model species because it originates from the tomato rhizosphere (45) and has previously been shown to protect plants from various diseases, including bacterial wilt caused by phytopathogenic Ralstonia solanacearum bacterium (46).To increase functional diversity within a single species, we first created a B. amyloliquefaciens mutant library using TnYLB-1 transposon mutagenesis (47), resulting in 1999 unique insertion mutants.A representative subset of 479 mutants were chosen for high throughput phenotyping in vitro in the lab regarding four important plant growth-promotion traits: biomass production, biofilm formation, swarming motility and pathogen suppression (9,48,49).After phenotyping, a subset of 47 mutants that performed better or equally well as the wild-type strain were taken forward to plant experiments to determine if in vitro phenotyping could predict mutant success in terms of tomato root colonization and plant protection from R. solanacearum in vivo.Finally, we tested if plant growth-promotion could be improved by combining phenotypically distinct mutants into multi-strain consortia.We predicted that increasing mutant consortia richness would lead to improved root colonization and plant protection if mutants are phenotypically dissimilar, which could result in positive diversity-ecosystem relationships or trait multifunctionality.These two hypotheses were tested by comparing the performance between phenotypically dissimilar and randomly assembled mutant consortia.Our results demonstrate that transposon insertion mutagenesis is an effective way to improve and identify genes underlying plant growth-promotion traits in B.
amyloliquefaciens.Importantly, we show that in vitro phenotyping can be used to optimize inoculant consortia functioning in vivo and that diverse mutant consortia are better at colonizing and protecting tomato plants when they are assembled based on phenotypic dissimilarity.

Effects of transposon insertions on B. amyloliquefaciens T-5 mutant traits measured in vitro and in vivo
We first quantified the phenotypic effects of transposon insertions on B. amyloliquefaciens T-5 mutant traits across the whole mutant library (1999 mutants in total, Figure 1 -figure supplement 1, Supplementary file 1a).
To make the number of mutants more manageable for in vivo experiment, we randomly selected a subset of 479 mutants for further analyses (Supplementary file 1b).While we likely lost certain unique mutants in the process, the sampled subset was phenotypically representative of the original collection based on four measured traits (Mantel test; r = 0.7591, p = 0.04167).Within this subset, most insertions had negative effects on the four measured phenotypic traits, with more than half of the mutants showing reduced swarming (58.7%), biomass production (67.2%) and biofilm formation (60.8%) compared to the wild-type strain (Fig. 1A).
In contrast, the median effect of insertions affecting the pathogen suppression was neutral, and 51.1% of the mutants showed only a moderate increase in their suppressiveness (Fig. 1A, Supplementary file 1b).In line with this finding, the distribution of effects of insertions on each trait was skewed, where beneficial mutations resulted mainly in a moderate improvement, while harmful mutations often led to severe reductions in measured traits (Fig. 1A).Moreover, several insertions caused trade-offs, where improvement in one trait led to a reduction in another trait (Fig. 1B).For example, swarming motility correlated negatively with biomass production, while biomass production led to a trade-off with both biofilm production and pathogen suppression (Fig. 1B).These results thus suggest that transposon insertions constrained the simultaneous improvement of multiple traits, leading to specialized B. amyloliquefaciens T-5 mutants, which could be clustered in three phenotypic groups based on K-means clustering (Adonis test: R 2 = 0.5283, p < 0.001, Fig. 1C, Figure 1 -figure supplement 2).Compared to the other two clusters, mutants belonging to the cluster 1 showed significant increases in biofilm formation and pathogen suppression but reduced biomass production (Fig. 1D, Supplementary file 2a).Mutants in the cluster 2 showed improved swarming motility and reduced pathogen suppression, while mutants in the cluster 3 had poor performance overall, showing highly reduced swarming motility and pathogen suppression (Fig. 1D, Supplementary file 2a).
To test if the mutants clustered in different phenotypic groups also differed in their tomato root colonization and ability to protect plants from R. solanacearum infections, 47 mutants representing a subset of three clusters were randomly selected for a greenhouse experiment (the specific effects of insertions on biological processes, cellular components and molecular function for all mutants are shown in Figure 1 -figure supplement 3B and Supplementary file 1c).Compared to the wild-type, 57% of these mutants (27/47) reached lower population densities in the rhizosphere (30 days post-pathogen inoculation (dpi)), and this was especially clear for mutants belonging to clusters 2 and 3.In contrast, mutants belonging to the cluster 1 retained more efficient root colonization compared to mutants from the other clusters, and more than half (16 of 27) of the cluster 1 mutants showed improved root colonization relative to the wild-type (30 dpi, Fig. 1E, Supplementary file 2b).Around 93% of all mutants (44/47) exhibited reduced plant protection relative to the wild-type strain.However, mutants from the cluster 1 showed higher plant protection compared to the other two clusters, and specifically, two of the cluster 1 mutants showed improved plant protection relative to the wild-type strain (30 dpi, Fig. 1F, Supplementary file 2b).Together, these results show that while most transposon mutants had reduced performance relative to the wild-type strain, some of the mutants showed improvement in at least in one plant growth-promotion trait, which often resulted in trade-offs with some other traits.

Phenotypic trait variation explains mutant success in rhizosphere colonization and plant protection in vivo with tomato
To test if phenotypic trait variation measured in vitro correlates with beneficial effects on plants in vivo, the rhizosphere colonization and plant protection of 47 phenotypically distinct mutants was quantified after 5-, 15and 30-days post pathogen inoculation (dpi) in a greenhouse experiment.Results showed that B. amyloliquefaciens inoculations led to approximately 20.6% mean reduction in bacterial wilt disease incidence at the final time point of the experiment (30 dpi, Fig. 2A-B).To establish a link between phenotypic variation measured in vitro and in vivo, we correlated mutant trait variation with root colonization and plant protection during different phases of the experiment.Trait correlations with the root colonization and plant protection became more significant in time and the most significant correlations were observed at the final time point (30 dpi), followed by middle (15 dpi) and early (5 dpi) sampling time points (Table 1).Specifically, swarming motility predicted the root colonization during the seedling stage (5 dpi, Table 1; Fig. 2C, F 1; Fig. 2D, 2I).As a result, the mean performance of mutants ('Monoculture average performance' index based on mean of all traits; see methods) was significantly positively correlated with both root colonization (Fig. 2G, F 1,46 = 7.77, R 2 = 0.1259, p = 0.0077) and plant protection (Fig. 2L, F 1,46 = 28.26,R 2 = 0.3671, p < 0.001) at the flowering stage (30 dpi).Together, these results suggest that while mutants with high trait values in biofilm formation, swarming motility and pathogen suppression had positive effects on root colonization and plant protection, their relative importance varied depending on the growth stage of the plant.

Designing and testing the performance of mutant consortia in vitro and in vivo
As transposon insertions mainly improved the performance of mutants regarding only one phenotypic trait, we tested if B. amyloliquefaciens T-5 performance could be improved by using mutants as phenotypically diverse consortia.To this end, we chose eight mutants that showed improved performance relative to the wild-type strain regarding one of the plant-beneficial traits measured in vitro (Figure 3 3 -figure supplement 3 for detailed composition of consortia).We hypothesized that consortia could show improved performance due to phenotypic complementarity or multifunctionality, where different mutants would 'specialize' respective to one of the four phenotypic traits, overcoming trade-offs and potential antagonistic pleiotropy experienced at the individual strain level (Fig. 1B).
We first tested the consortia performance regarding the four traits measured in vitro.We found that relative to wild-type strain, only a few consortia showed improved performance regarding swarming motility (8 of 37), biomass production (2 of 37), biofilm formation (4 of 37) or pathogen suppression (15 of 37) (Figure 3 -figure supplement 3A-D).
Moreover, consortia performance did not show clear relationship with increasing richness regarding to any of the measured traits (Figure 3 -figure supplement 4).We further tested if the consortia performance could be predicted based on the trait averages of individually grown mutants, assuming that mutant performance is not affected by interactions between the consortia members.Only one significant positive relationship was found between the predicted pathogen suppressiveness, and the size of the inhibition halo observed in vitro lab measurements (Fig. 3A-D).This suggest that individually measured mutant traits poorly predicted observed consortia performance in vitro except for the pathogen suppression.
We next tested the consortia performance regarding root colonization and plant protection in vivo.While only 7 of 37 of consortia showed improved rhizosphere colonization, around half of them (18 of 37) exhibited a clear increase in plant protection compared to the wild-type strain (at 30 dpi, Figure 3 -figure supplement 3E-F).While root colonization or plant protection could not be predicted based on the consortia performance measured in vitro (Fig. 3E-F), increasing consortia richness improved both root colonization (Fig. 3G, F 1,35 = 6.52,R 2 = 0.1330, p = 0.0152) and plant protection (Fig. 3H, F 1,35 = 18.64,R 2 = 0.3289, p < 0.001), which were also positively correlated with the consortia average performance measured in vitro (root colonization: F 1,35 = 6.47,R 2 = 0.1319, p = 0.0156; plant protection: F 1,35 = 8.82, R 2 = 0.1786, p = 0.0053; Fig. 3I-J).We also analyzed the significance of mutant identity effect on the consortia performance in vivo by excluding each strain from the dataset and comparing model fit and significance of explanatory variables.The presence of M54 mutant (efficient in biofilm formation) significantly increased consortia root colonization, while the presence of mutants M59 (efficient in biomass production) and M143 (efficient in biofilm formation) significantly improved plant protection (Figure 3 -figure supplement 5, Supplementary file 2e).Crucially, the effect of consortia richness remained significant after sequential removal of each mutant and refitting of the model, which demonstrates that the diversity effect was robust and relatively more important in explaining root colonization and plant protection compared to mutant identity effects (Supplementary file 2f).Together, these data suggest that mutant consortia diversity was positively linked with consortia performance in vivo, indicative of positive diversity-ecosystem functioning relationship.
To test if the positive diversity effect was potentially driven by consortium multifunctionality, we compared the performance of the 'optimized' 8-mutant consortium (no.37; Figure 3 -figure supplement 3) used in the diversity-ecosystem functioning experiment with eight randomly assembled 8-mutant consortia, which could be considered as phenotypically 'unoptimized'.We found that the 'optimized' consortium was more effective at both plant root colonization and plant protection compared to 'unoptimized' 8-mutant consortia (Fig. 4A, B).We also correlated the relative performance of all 8-mutant consortia with both in vivo traits and found that both root colonization (Fig. 4C, F 1,7 = 26.56,R 2 = 0.7616, p = 0.0013) and plant protection (Fig. 4D, F 1,7 = 6.39,R 2 = 0.4026, p = 0.0394) improved along with the increase in the relative performance of consortia.Together, these results suggest that in vitro and in vivo phenotyping could reliably predict the root colonization and plant protection of 8-mutant B. amyloliquefaciens consortia.

Discussion
In this work, we tested if increasing intra-species diversity of B. amyloliquefaciens T-5 bacterium via mutagenesis could offer a viable strategy for improving mutant consortia multifunctionality and plant health (Fig. 3G-H).Our results show that mutations that improved bacterial performance regarding one trait often led to specialism and reduced performance regarding other traits.Such trait trade-offs experienced at the individual genotype level could be overcome by assembling consortia of phenotypically distinct mutants, that showed increase in average trait performance.Crucially, the consortia richness and average trait performance correlated positively with increased root colonization and plant protection, indicative of increased consortia multifunctionality and improved plant health.Especially, diverse 8-mutant consortium that consisted of phenotypically distinct mutants performed better compared to randomly assembled 8-mutant consortia consisting of phenotypically similar mutants.Together, these findings suggest that increasing intra-species functional diversity could offer an easy solution for improving the performance of bacterial consortia.
We specifically focused on four B. amyloliquefaciens traits that have previously been linked to plant growth-promotion: swarming motility, biomass production, biofilm formation and direct pathogen suppression via antibiosis (50)(51)(52)(53).While most transposon insertions reduced the strains' performance, some of them improved at least one measured trait.However, all trait improvements were costly and reduced mutants' performance regarding other traits, indicative of trade-offs and antagonistic pleiotropy (54).Such costs of adaptation are common with microbes and have previously been linked to a wide range of functions, including metabolism, antibiotic production, motility and stress resistance (55,56).Overall, transposon insertions were identified in several genes associated with broad range of functions (Figure 1 -figure supplement 3, Supplementary file 1c).With eight phenotypically distinct mutants that were used for consortia assembly experiment, increased swarming motility was associated with insertions in parE (DNA topoisomerase IV subunit B) and DeoR (DNA-binding transcriptional repressor) genes, which has previously been linked to antibiotic resistance and bacterial deoxyribonucleoside and deoxyribose utilization (57), respectively.While it remains unclear how these genes were linked with swarming motility, their disruption affected also other traits as evidenced by reduced biofilm formation.Increased biomass production was linked to disruption of comQ (competence protein) and hutI (imidazolonepropionase) genes and trade-offs with the other three measured traits.ComQ gene controls the production of ComX pheromone (58) and has recently been linked to antimicrobial activity (59), which could explain reduction in the pathogen suppression by this mutant.Insertion in histidine utilization (hut) system gene, hutI, could have potentially impaired catabolite and amino acid repression, resulting in improved biomass production with one of the mutants (60,61).Interestingly, increased biofilm formation was also linked to insertion in hut system (hutU) with one mutant, while the other mutant had insertion in YsnB gene, which encodes for a putative metallophosphoesterase.While both insertions were linked to trade-offs with swarming motility and biomass production, they are not commonly associated with biofilm formation in Bacillus (42,62).Finally, moderate improvement in pathogen suppression was observed with two mutants that had insertions in nhaC (sodium-proton antiporter) and dfnG (difficidin polyketide synthesis) genes.The nhaC gene is known to act as repressor for Pho regulon in Bacillus (62), while mutations in Pho regulon have been linked to increased antibiotics production with several Streptomyces species (63).
Moreover, the dfnG gene controls the production of difficidin antibiotic, which has previously been linked to biocontrol activity against fire blight and Xanthomonas oryzae rice pathogen (64) and could have also suppressed the growth of R. solanacearum.Similar to the other phenotypically distinct mutants, insertions associated with increased pathogen suppression led to trade-offs with other measured traits.While more work is required to unequivocally link these mutations with associated traits at the molecular level, our findings show that all above trait improvements achieved via mutagenesis resulted in trade-offs with other traits and that this technique could be used to identify genes underlying plant growth-promotion and pathogen suppression.
To overcome trait trade-offs experienced at the individual mutant level, we tested if we could improve the plant growth-promotion by combining phenotypically distinct mutants into multifunctional consortia.We found that mutant performance measured in monocultures was a poor predictor of B. amyloliquefaciens performance in consortia in vitro, except for the pathogen suppression.This suggest that while the selected mutants did not show direct antagonism towards each other, they likely interacted in other ways leading to unpredictable trait expression when growing together (for example via certain emergent effects ( 65)).Despite this, we found clear diversity effects, where consortia richness and average performance were positively associated with both plant root colonization and plant protection from R. solanacearum pathogen infection.
While important mutant identity effects were also observed, omission of these strains did not change the significance of underlying diversity effects, highlighting the importance of interactions between the consortia members in determining the positive effects on the plant health.Together, these results suggest that consortia that performed well regarding all measured traits on average had improved ability to colonize rhizosphere and suppress the pathogen.To test if these effects were driven by diversity per se or underlying trait variation between consortia members, we compared the optimized 8-mutant consortium with eight randomly assembled 8-mutant consortia that were phenotypically more similar.We found that randomly assembled consortia performed less well on average, while consortia functioning improved along with consortia relative performance, suggesting that diverse consortia performed better only when they had been assembled from phenotypically dissimilar mutants.While similar positive diversity-ecosystem functioning relationships have previously been found in more complex Pseudomonas (5,11,17), leaf bacterial (66) and grass-land soil microbiomes (67), we here show that this pattern also holds along with intra-species diversity gradient.While the measured phenotypic traits are considered to be robust to inoculum densities, it will be important to evaluate in the future if the absolute abundances of each mutant play a role in the consortia functioning.
Positive diversity effects have previously been explained by facilitation, ecological complementarity, and division of labor, which can reduce competition between the consortia members within or between niches (42,43,68).Moreover, high diversity could provide stability for consortia functioning via insurance effects by increasing the likelihood of certain consortia members surviving in the soil after the inoculation (69,70).While our experiments were not designed to disentangle the relative importance of these potential mechanisms, we analyzed which mutant traits could significantly explain the dynamics of root colonization and plant protection at seedling, vegetative and flowering stages of the tomato growth by focusing on 47 mutants.While the effect of Bacillus biomass production was consistently non-significant, both motility and biofilm formation were positively associated with the root colonization.However, motility was significant only at the seedling stage, while biofilm became significant during vegetative and flowering stages.In line with ecological succession often taking place in the rhizosphere (50), high motility might have allowed faster colonization of relatively 'sterile' young roots by Bacillus, while biofilm formation could have promoted stress tolerance and resource competition in more diverse and mature microbial communities during the later stages of tomato growth (71,72).Interestingly, increase in pathogen suppression, biofilm formation and motility were positively associated with improved plant protection at the flowering stage, which suggests that all these traits were positively associated with the ecosystem functioning in terms of plant health.It is thus possible that consortia were together able to overcome the trait trade-offs experienced at the individual mutant level, leading to improved and more stable ecosystem functioning for the whole duration of tomato growth cycle.In addition, it is possible that some of the measured plant growth-promotion traits might act as public goods (73, 74), which could have been shared between different mutants, leading to overall improvement in the mutant population fitness.Such micro-scale mechanisms could be potentially validated in the future using transcriptomics and barcoded Tn-seq mutants, which would allow estimating activity and changes in mutant frequencies during bioinoculation.
In conclusion, we here demonstrate that the beneficial effects provided by a single B. amyloliquefaciens bacterium can be improved by increasing consortia functional diversity using transposon mutagenesis.Our approach highlights the importance of intra-species genetic diversity for the ecosystem functioning and provides a trait-based approach for designing microbial communities for biotechnological applications.Our approach does not require a priori knowledge on specific genes or molecular mechanisms, but instead relies on generation of trait variation which is screened and selected by the experimenter.Our method can also help to identify novel functional roles of previously characterized and uncharacterized genes.While the benefit of this method was here demonstrated in the context of agriculture, it could be applied in other biotechnological contexts, such as biofermentation, waste degradation and food manufacturing.Future work focusing on the population dynamics, metabolism and gene expression of different mutants will help to understand the relative importance of ecological complementarity, division of labor, insurance effects, population asynchrony (93) and facilitation for the consortia ecosystem functioning.

Generation of Bacillus amyloliquefaciens T-5 transposon mutant library
To increase the intra-species diversity of B. amyloliquefaciens, we generated a random transposon insertion mutant library by using a TnYLB-1 transposon derivative, carried in the thermosensitive shuttle plasmid pMarA (Supplementary file 2g), which was electro-transformed to bacteria as previously described by Zakataeva et al. (78,79).The cells with intact pMarA plasmid contained resistance cassettes to both erythromycin and kanamycin, while the cells with integrated transposons were resistant only to kanamycin.Transposon mutant library was created as follows.An overnight B. amyloliquefaciens T-5 cell culture grown in neutral complex medium (NCM, 17.4 g L -1 K 2 HPO 4 , 11.6 g L -1 NaCl, 5 g L -1 glucose, 5 g L -1 tryptone, 1 g L -1 yeast extract, 0.3 g L -1 trisodium citrate, 0.05 g L -1 MgSO 4 .7H 2 O, and 91.1 g L -1 sorbitol, pH = 7.2) was diluted 25-fold with fresh NCM medium supplemented with 5 mg mL -1 of glycine and grown at 30 °C for 3 h on a rotary shaker (170 rpm).After 1 h incubation (at an optical density OD 600 ∼ 0.8), cells were cooled on ice, harvested by centrifugation (8,000 × g for 6 min at 4 °C) and washed four times with ice-cold electrotransformation buffer (ETM, 0.5 M sorbitol, 0.5 M mannitol, and 10% glycerol).Resulting pellets were resuspended in ETM buffer supplemented with 10% PEG 6000 and 1mM MgCl 2 , yielding approximately 10 10 cells mL -1 .Cells were then mixed with 500 ng of plasmid DNA in an ice-cold electrotransformation cuvette (2-mm electrode gap), and after 1-3 min incubation at room temperature, exposed to a single electrical pulse using a MicroPulser Electroporator (Bio-Rad Laboratories) at field strength of 7.5 kV cm -1 for 4.5-6 ms.Immediately after the electrical discharge, cells were transferred into 1 mL of LB, incubated with gentle shaking at 30 °C for 3-8 h, and plated on LB agar containing 10 μg mL −1 erythromycin.Transformants were selected after 36-48 h incubation at 30 °C.To generate final transposon library, erythromycin-resistant colonies with plasmids were individually transferred to fresh LB and incubated overnight at 30 °C, after cultures were diluted, spread on LB plates supplemented with 10 μg mL −1 kanamycin, and incubated for 24 h at 46 °C.As the plasmid cannot replicate at 46°C, only cells with an integrated transposons grew and could be separated.A total of 1999 transformed colonies were isolated and individually cryopreserved in 30% glycerol at -80 °C.

Phenotypic characterization of B. amyloliquefaciens T-5 mutant library in vitro
The wild-type strain and 1999 mutants were phenotyped for following plant-growth promoting traits: swarming motility, biomass production, biofilm formation and pathogen suppression via production of antimicrobials (see below).These traits were selected due to their known importance for B. amyloliquefaciens competitiveness in the rhizosphere and their involvement in pathogen suppression (50)(51)(52)(53).To prepare bacterial inoculants, frozen colonies were picked and pre-grown overnight in LB at 37 °C, washed three times in 0.85% NaCl and adjusted to initial OD 600 of 0.5 (∼ 10 7 cells mL -1 , based on OD vs CFU calibration curve, Figure 1 -figure supplement 4).In addition to each individual trait, we also calculated the average of all measured traits and used the resulting 'monoculture average performance' (80) index to compare mutants' overall performance.Of the 1999 phenotyped mutants, a subset of 479 mutants were randomly selected for more detailed analysis and probiotic bioinoculant design.While we likely missed certain mutants with this method, the 479 mutants represented a similar phenotypic diversity as the 1999 mutant collection (Mantel test; r = 0.7591, p = 0.04167), indicating that our sampling captured a phenotypically representative subsample of mutants (Supplementary file 1a, 1b).
Swarming motility was measured using a previous method described by Kearns (81).Briefly, 2 μL of each B.
amyloliquefaciens mutant was inoculated into the center of 0.7% agar LB plates supplemented with 10 μg mL -1 of kanamycin.After 24 h incubation at 30 °C, swarming motility was evaluated as the radius of the colony.
Three replicates were used for each mutant.
Biofilm formation was assessed as described previously (83)   amyloliquefaciens mutants by drenching, resulting in a final concentration of 10 7 colony forming units (CFU) g -1 soil ( 89).The R. solanacearum strain was inoculated using the same method one week later at a final concentration of 10 6 CFU g -1 soil.Positive control plants were treated only with R. solanacearum, while negative control plants received no bacterial inoculants.Three replicated trays were set up for each treatment, with 20 seedlings (in individual cells) per tray.Each tray was considered as one biological replicate.Tomato plants were grown for 30 days after pathogen inoculation with natural temperature (ranging from 25°C to 35°C) and lighting variation (around 16 h of light and 8 h of dark).Seedling trays were rearranged randomly every second day and regularly watered with sterile water.

Quantifying B. amyloliquefaciens mutants' root colonization and plant protection in the rhizosphere
The root colonization and plant protection of 47 B. amyloliquefaciens T-5 mutants was quantified individually as a change in their population densities in the tomato rhizosphere after 5, 15 and 30 days of R. solanacearum pathogen inoculation (days post pathogen inoculation, i.e., 'dpi').At each sampling time point, three independent plants per inoculated mutant were randomly selected and sampled destructively by carefully uprooting the plant and gently removing the soil from the root system by shaking.After determining plant fresh weight, the root system of each plant was thoroughly ground in 5 mL of 10 mM MgSO 4 •7H 2 O using a mortar, and serial dilutions of root macerates were plated on a semi-selective Bacillus medium consisting of 326 ml L -1 vegetable juice (V8, Campbell Soup Co., USA), 33 g L -1 NaCl, 0.8 g L -1 dextrose, 16 g L -1 agar (pH 5.

Assembly of phenotypically dissimilar B. amyloliquefaciens mutant consortia
To test if the performance of B. amyloliquefaciens T-5 mutants could be improved by using consortia of phenotypically dissimilar mutants, a subset of eight best-performing mutants excelling at different phenotypic traits were selected (Figure 3 -figure supplement 1, Supplementary file 2c).Specifically, these included two mutants that showed high swarming motility (M108: pare and M124: DeoR), high biomass production (M59: comQ and M109: hutI), high biofilm formation (M54: hutU and M143: YsnB), and slightly improved pathogen suppression (M38: nhaC and M78: dfnG) relative to the wild-type strain (Figure 3 -figure supplement 1, Supplementary file 2c).To test the effect of transposon insertions on potential antagonism between the mutants, we conducted two types of assays: direct growth inhibition by 1) spotting each strain on top of the others using agar overlays, and by 2) growing each strain in the supernatant of the other strains.With agar overlay assays, 2 μL of each mutant with density OD 600 of 0.5 (∼ 10 7 cells mL -1 ) was spotted on the soft agar overlay of the other mutants and direct antagonistic effect was measured as the size of the inhibition halo observed on the soft agar plates (91).For the supernatant assay, we first cultured each mutant in liquid LB for 2 days and collected supernatants by using 0.22 μm filters.In the growth assays, 2 μL of each strain with initial concentration of 10 7 cells mL -1 was mixed with 20 μL of each supernatant and 178 μL of 50% LB.The growth of each strain was measured after 24h as optical density (OD 600 ), and inhibition calculated as the relative growth of each strain in its own or other strains' supernatant compared to strains' growth in the fresh 50% LB (diluted with sterile water).Here, the reduced growth in other strains' supernatant relative to the growth in the fresh medium was deemed as inhibition between mutants.A following formula was used where OD 600 sup and OD 600 LB denote for mutants' growth in other mutants' supernatant or in 50% fresh LB after 24h: These eight mutants were then used to assemble a total of 29 consortia with 2, 4 or 8 mutants, following a substitutive design where each mutant was equally often present at each community richness level (see left panel key of Figure 3 -figure supplement 3 for detailed consortia assembly).Mutants were mixed in equal proportions in each consortium with final total bacterial density OD 600 of 0.5 (e.g., 50:50% or 25:25:25:25% in two and four mutant consortia, respectively; ∼ 10 7 cells mL -1 ).This design has previously been used to investigate biodiversity-ecosystem functioning relationships in plant-associated bacterial communities (11,92), allowing disentangling the effects due to consortia richness, composition and mutant strain identity.In addition, to compare the performance of optimized 8-member consortium (assembled based on phenotypic dissimilarity; see above) with non-optimized 8-mutant consortia, we assembled eight additional 8-mutant consortia randomly from the 479 mutant collection, which were used in in vitro lab and in vivo greenhouse experiments.

Phenotypic characterization of B. amyloliquefaciens consortia performance in vitro and consortia root colonization and plant protection in the tomato rhizosphere
The performance of each mutant and assembled consortium was assessed in vitro in the lab by measuring traits as mono-and co-cultures following the same methods as described previously (swarming motility, biomass production, biofilm formation and pathogen suppression).Mutant strains were prepared individually from frozen stocks by growing overnight in liquid LB, pelleted by centrifugation (4,000 × g, 3 min), washed three times with 0.85% NaCl and adjusted to OD 600 of 0.5 (10 7 cells mL -1 ).Consortia were then assembled following the substitutive design describe earlier (Figure 3 -figure supplement 3) by mixing mutants in equal proportions for each consortium with total bacterial density OD 600 of 0.5 (10 7 cells mL -1 ; e.g., 50:50% or 25:25:25:25% in two and four mutant communities, respectively).Consortia traits were characterized as described previously and compared with the ancestral B. amyloliquefaciens wild-type strain.The root colonization and plant protection of B. amyloliquefaciens T-5 consortia were quantified in greenhouse experiments following previously described methods.Predicted performances were calculated following the additive model, equaling the sum of different trait values of each member divided by the richness value of the given consortium.To link the performance of single mutant with functioning of consortia, we used the relative performance measure, which included the magnitude and direction of difference relative to the wild-type strain.The difference and direction in magnitude to the wild-type strain were calculated based on the euclidean distance and average performance using following formula: D i , euclidean distance between each consortium member and wild-type based on four traits; AP i , average performance of each community member; AP wt , average performance of wild-type; n, community richness.

Statistical analyses
Data were analyzed with a combination of analysis of variance (ANOVA), principal component analysis (PCA), linear regression models, unpaired two-sample Wilcoxon tests and student's t-test.Individually measured mutant traits data was normalized between 0 and 1 across the all collection using min-max normalization (93).
In addition, the different phenotypic traits were combined into a 'Monoculture average performance' index, which was calculated as the mean of the four standardized traits for each mutant.Monoculture average performance and consortia traits values were also min-max normalized between 0 and 1 for subsequent analyses.To classify mutants into different functional groups, K-means clustering algorithm ('fviz_nbclust' in 'factoextra' package and 'kmeans' function) was used and clusters were visualized using PCA ('princomp' in 'vegan' package) based on multivariate trait data.The phenotypic dissimilarity between the same mutants and the wild-type strain were calculated using 'vegdist' based on 'Euclidean' algorithm.The B. amyloliquefaciens T-5 abundance data measured in root colonization assays were log 10 transformed and disease incidence data were arcsine square root-transformed prior the analyses.Linear regression models were used to explain root colonization and plant protection with mutant traits, average performance, consortia richness and consortia relative performance.Treatment mean differences were analyzed using two-sample Wilcoxon test ('wilcox.test'function) or student's t-test ('t.test' function) depending on the unequal or equal sample sizes, respectively.
The temporal effects of four traits on root colonization and plant protection were assessed separately for different time points using ANOVA ('aov' function).All statistical analyses were performed using R 3.5.2(R core Development Team, Vienna, Austria).All code used in this study is available on request from corresponding authors.amyloliquefaciens consortia average performance measured in vitro, respectively.In all panels, the black dashed lines show the performance of the wild-type strain, while red dashed lines in panels F, H and J show the disease incidence of pathogen-only control treatment.In all panels, shaded areas show the confidence interval around the mean.
) and Bacillus amyloliquefaciens T-5 biocontrol(45) strains as our model bacterial species.The B. amyloliquefaciens T-5 can suppress the growth of R. solanacearum QL-Rs1115 by competing for space and nutrients in the rhizosphere(77) and by producing various antibacterial secondary metabolites(55).Both bacterial stocks were cryopreserved at -80 °C in 30% glycerol stocks.Prior starting the experiments, active cultures were prepared as follows: B. amyloliquefaciens T-5 was grown at 37 °C in Lysogeny Broth (LB-Lennox, 10.0 g L -1 Tryptone, 5.0 g L -1 yeast extract, 5.0 g L -1 NaCl, pH = 7.0) and R. solanacearum QL-Rs1115 was grown at 30 °C in Nutrient Broth (NB, 10.0 g L -1 glucose, 5.0 g L -1 BLASTX and BLASTN available at the NCBI, and against the complete ancestral B. amyloliquefaciens T-5 genome sequence (Accession: CP061168, Figure1-figure supplement 3A, Supplementary file 1c).The functional classification of disrupted genes for all 47 transposon mutants is summarized in Figure1-figure supplement 3B.Assessing the performance of individual B. amyloliquefaciens T-5 mutants in a greenhouse experimentAll selected 47 mutants and the wild-type strain were individually screened for their ability to colonize tomato rhizosphere and protect plants against infection by R. solanacearum QL-Rs1115 pathogen strain in a 50-day long greenhouse experiment.Surface-sterilized tomato seeds (Lycopersicum esculentum, cultivar "Jiangshu") were germinated on water agar plates in the dark at 28 °C for 2 days, before sowing to sterile pots containing wet vermiculite (Huainong, Huaian soil and fertilizer Institute, Huaian, China).Ten-days old tomato seedlings (at three-leaves stage) were then transplanted to seedling trays containing natural, non-sterile soil collected from a tomato field in Qilin Town, Nanjing, China(50).Plants were inoculated with individual B.

2 ,
adjusted with NaOH) supplemented with 45 mg L -1 cycloheximide and 22.5 mg L -1 polymyxin B (90).This media was used to count the densities of B. amyloliquefaciens T-5 wild-type, and the same media supplemented with h and bacterial densities expressed as CFU per gram of root biomass.The effect of B. amyloliquefaciens wild-type and mutants on plant protection was measured as the reduction of bacterial wilt disease symptoms during the experiment (based on the proportion of plants showing wilting symptoms).The first wilting symptoms appeared 7 dpi and the proportion of diseased plants quantified at 5, 15 and 30 dpi were used in analyses.Plant protection was expressed as the relative reduction in the number of wilted plants compared to the positive control (only R. solanacearum inoculated in the absence of B. amyloliquefaciens T-5 mutants or wild-type).

Figure 1 .
Figure 1.The effects of transposon insertions on the traits of 479 B. amyloliquefaciens T-5 mutants measured in vitro and in vivo.Panel A shows distribution of fold changes regarding mutants' swarming motility, biomass production, biofilm formation and pathogen suppression relative to the wild-type strain (black dashed line equaling 1).Panel B displays pairwise covariance matrix between individual traits, where red cells indicate negative trait correlations (trade-offs) and blue cells positive traits correlations; white cells indicate no correlation between the given traits.Panel C shows principal coordinates analysis showing the clustering of all mutants and the wild-type (black point).Mutants were assigned to different clusters based on K-means algorithm of four measured traits.Panel D shows mean trait differences between clusters based on unpaired two-samples Wilcoxon test.The wild-type was assigned in the cluster 1 based on K-means clustering and its trait values are shown as dashed black lines.Panel E displays root colonization of representative 47 B. amyloliquefaciens mutants from clusters 1-3 relative to the wild-type strain (black dashed line) based on cell densities in the root system 30 days post pathogen inoculation (dpi).Panel F, shows plant protection of representative 47 B. amyloliquefaciens mutants from clusters 1-3 relative to the wild-type strain (black dashed line), and negative 'pathogen-only' control (red dashed line), quantified as bacterial wilt disease incidence 30 dpi.Shaded areas in panel E and F represent the mean ± SEM.Since panel D displays the normalized trait

Figure 2 .
Figure 2. Regression analysis explaining root colonization and plant protection of representative 47 B. amyloliquefaciens mutants based on their trait values measured in vitro at different sampling time points.Panels A and B show the dynamics of root colonization and plant protection, respectively.The red and black lines in panel B shows the disease incidence of pathogen-only control and B. amyloliquefaciens mutant treatments, respectively.Panels C-G and H-L show root colonization and plant protection, respectively, correlated with different traits at 5 dpi (grey), 15 dpi (light blue) and 30 dpi (dark blue) time points.Significant relationships and R-squared values are shown in panels with colors corresponding to the sampling time points ('ns' denotes for non-significant relationship).

Figure 3 .
Figure 3.The relationships between predicted and observed consortia performance measured in vitro and correlations between plant performance, consortia richness and consortia average performance measured in vivo.Panels A-F show correlations between predicted and observed consortia performance regarding swarming motility, biomass production, biofilm formation, pathogen suppression, root colonization and plant protection, respectively (blue dashed lines show 1:1 theoretical fit and solid black lines show the fitted

Figure 4 .
Figure 4. Optimized and randomly assembled 8-mutant consortia show contrasting effects on root colonization and plant protection.Panels A-B compare differences between optimized (white bar) and randomly assembled (grey bars) B. amyloliquefaciens 8-mutant consortia on root colonization and plant performance based on students' t-test: *** denotes for statistical significance at p < 0.001; ** denotes for statistical significance at p < 0.01; * denotes for statistical significance at p < 0.05.In both A and B panels, Y-axes show the consortia performance as a fold change relative to wild-type strain and shaded areas represent the mean ± SEM.Panels C-D show positive and negative correlations between consortia relative performance (calculated based on average performance and trait deviance, see methods) with root colonization (C) and disease incidence (D); optimized and randomly assembled 8-mutant consortia are shown as white and black circles, respectively (shaded area shows the confidence interval around the mean).

Figure 1 -figure supplement 2 .
Figure 1 -figure supplement 2. Optimal number of K-means clusters suggested by gap statistic method.

Figure 1 -figure supplement 3 .
Figure 1 -figure supplement 3. Panel A shows graphical circular map of Bacillus amyloliquefaciens T-5 genome.Moving from the outside to the center, different circles denote for: genes of the forward strand (colored by COG categories), genes of the reverse strand (colored by COG categories), RNA genes (tRNAs green, rRNAs red, other RNAs black), GC content and GC skew.Panel B shows the distribution of disrupted gene functions of a subset of 47 mutants based on GO categories and K-mean cluster identity (detailed gene sequences of the disrupted genes is provided in Supplementary file 1c).

Figure 1 -figure supplement 4 .
Figure 1 -figure supplement 4. Positive and linear relationships between optical density (OD 600 nm ) and cell counts (CFUs) of wild-type (black), M78 (representative mutant with decreased biomass production, red) and M109 (representative mutant with increased biomass production, blue) strains.

Figure 3 -figure supplement 1 .
Figure 3 -figure supplement 1. Phenotypic characteristics of the eight B. amyloliquefaciens mutants used for assembly of consortia richness gradient.Blue, red and grey cells denote for normalized trait values of mutants that showed increase, decrease or no change (ns) relative to the wild-type strain (students' t-test at p < 0.05; see details in Supplementary file 2c).The rightmost column (*) shows the functional categories (based on clusters of orthologous genes; COG) of different genes in parentheses, where capital letters denote for: C: Energy production and conversion; E: Amino acid transport and metabolism; R: General function prediction only; Q: Secondary metabolites biosynthesis, transport and catabolism; L: Replication, recombination and repair; K: Transcription.

Figure 3 -
Figure 3 -figure supplement 2. Pairwise interactions between eight mutants used in diversity-ecosystem functioning experiment based on supernatant culture assays (A) and agar overlay spot assays (B).In panel A, Y-axis denotes the growth of different mutants on their own (diagonal) and other mutants' supernatant(mutants on X-axis) relative to growth in the fresh 50% LB; the magnitude is shown as color gradient from red (negative effects) to blue (positive effects).In panel B, no inhibition halos were found when mutants were spotted on top of each soft agar overlays.

Figure 3 -
Figure 3 -figure supplement 3. The performance of single mutants and mutant consortia relative to wild-type strain regarding four traits measured in vitro and root colonization and plant protection measured in vivo.Panels show: (A) Swarming motility, (B) biomass production, (C) biofilm formation, (D) pathogen suppression, (E) root colonization and (F) plant protection.In panels A-F, the black vertical dashed lines represent the performance of the wild-type strain, while the vertical red line in panel F represents pathogen-only control treatment; shaded areas represent the mean ± SEM; consortia richness gradient isshown on a grayscale as symbol colors from light to dark (low to high richness).Mean differences between consortia and the wild-type strain were analyzed using student's t-test: *** denotes for statistical significance at p < 0.001; ** denotes for statistical significance at p < 0.01; * denotes for statistical significance at p < 0.05.

Figure 3 -figure supplement 4 .
Figure 3 -figure supplement 4. The relationship between B. amyloliquefaciens mutant consortia richness and measured consortia trait performance in vitro.The panels denote for swarming motility (A), biomass production (B), biofilm formation (C), and pathogen suppression (D) and consortia average performance (E;mean of all traits).In all panels, Y-axis show the normalized trait values, the black dashed line represents the performance of the wild-type strain and solid black line shows the fitted regression; ns denotes for non-significant relationship.

Figure 3 -figure supplement 5 .
Figure 3 -figure supplement 5. Analysis of mutant identity effects on consortia performance in vivo.The mutant identity effects were analyzed comparing consortia root colonization (A) and plant protection (B) in the absence and presence of each mutant.In both panels, the black dashed lines represent the performance of the wild-type strain, while red dashed line in panel (B) represents disease incidence in pathogen-only control treatment.Shaded areas represent the mean ± SEM.The mutants' 'trait specialism' is shown in parentheses on X-axis.Differences were analyzed using unpaired two-sample Wilcoxon test, where * denotes for statistical significance at p < 0.05; ns denotes for no significant (See details in Supplementary file 2e).

Table 1 . ANOVA table summarizing the effects of mutant traits measured in vitro on the root colonization and plant protection.
Separate models were run for each dependent variable at different time points(5, 15,and 30 dpi) and all response variables were treated as continuous variables (bacterial abundances were log-transformed before the analysis).Table data represent only the most parsimonious models based on the