Potassium Phosphite Modulated the Soil Microbiome and Enriched the Antagonistic Bacteria Streptomyces Coelico avus and Paenibacillus Favisporus to Inhibit the Tomato Pathogen Ralstonia Solanacearum Synergistically


 Background: Application of certain agricultural chemicals could modulate the soil microbiome and induce potential antagonistic microbes. However, the specific selective effects of agricultural chemicals on soil bacterial functions and their co-occurrences are not well understood, and no studies have verified that the enriched potential antagonistic microbes could enhance the antagonistic functions of the soil microbiome.Results: Here, the effects of potassium phosphite (KP), an environment-friendly agricultural chemical, on the soil bacterial composition, co-occurrences and antagonistic functions were determined, and the potential antagonistic bacteria against the tomato bacterial wilt pathogen Ralstonia solanacearum were isolated to test their functions and associations among these strains. Our results showed that application of KP enriched Bacillus, Paenibacillus and Streptomyces. The positive links among the OTUs belonging to these genera were increased, and positive associations between these OTUs and predicted genes related to antagonistic substance production were revealed. Two strains, Streptomyces coelicoflavus F13 and Paenibacillus favisporus Y7, were isolated, and they inhibited the growth of R. solanacearum. Genomic sequencing showed that both strains harboured streptomycin synthetic genes, and P. favisporus Y7 also contained surfactin synthetic gene cluster. Synergistic inhibition of R. solanacearum growth by P. favisporus Y7 and S. coelicoflavus F13 was observed in soil. Genome-scale metabolic modelling showed that dextrin and lactic acid were potential cross-feeding metabolites. In addition, the KP-modulated soil microbiome could suppress R. solanacearum growth. Conclusions: Our results highlight that a KP-modulated soil microbiome has considerable potential for biocontrol and indicate a new mechanism for the inhibition of R. solanacearum by KP-enriched soil bacteria.


Background
Tomato bacterial wilt is a serious disease caused by Ralstonia solanacearum. This vascular pathogen not only causes large economic losses annually but also leads to imbalanced bacterial communities [1,2]. To manage tomato bacterial wilt disease, resistant cultivars, chemical bactericides and biocontrol agents have been used worldwide [3][4][5]. However, no effective methods have been found to control bacterial wilt, and prevention is still the major strategy. Moreover, after harvest, the root residuals in the soil provide the niche and nutrition for R. solanacearum to facilitate the next seasonal invasion [6]. Thus, decreasing the initial abundance of R. solanacearum may contribute to the prevention of bacterial wilt disease outbreak, for example, through the use of the fumigant chloropicrin [7]. However, this method is not environmentally friendly.
Agricultural chemicals are widely used in the control of soil-borne disease due to their direct inhibition of pathogens. The application of agricultural chemicals can signi cantly affect the soil microbial community [9]. Application of inorganic germicide copper depleted populations of bacteria, cellulolytic fungal species and Streptomyces in sandy soil [10]. The application of another inorganic germicide, sulphur, enriched acidophilic soil bacteria [11]. Moreover, application of certain agricultural chemicals may enrich potential bene cial microbes in soil [12][13][14]. The relative abundances of the bene cial bacteria Rhizobiales, Nitrosomonadaceae and Bryobacter and functional genes related to nitrogen metabolism, carbohydrate metabolism and cell processes and the numbers of network modules were enriched in soil amended with selenium [15]. The application of chloropicrin fumigation increased the relative abundance of the potential antagonistic microbe Actinobacteria, which may result in an enhanced soil antibacterial capacity against R. solanacearum [7]. However, the effects of these agricultural chemicals on the soil microbial ecology, speci cally for antagonistic functions and microbial associations are poorly understood. Furthermore, it is generally unknown whether the enriched bene cial bacteria could enhance the antagonistic ability of the soil microbiome.
Approximately 80% of antibiotics are known to be sourced from Actinobacteria [16]. Streptomyces has been widely used to suppress diverse plant pathogens in agricultural system [17,18]. Bacillus species are known for their tremendous resistance to adversity and antagonistic capacities [19,20]. Application of a Bacillus amyloliquefaciens strain that could produce iturin, fengycin and surfactin could effectively control tomato bacterial wilt, with a biocontrol e ciency of 97.6% [19]. Still, there is little understanding of whether the agricultural chemicals that were applied in soil can contribute to selective enrichment of Streptomyces and Bacillus strains that are bene cial for sustainable agricultural production.
Potassium phosphite (KP) has been used in agricultural systems and is receiving increasing attention because is environmentally friendly [21]. Several studies have shown that KP can control some soil-borne pathogens, such as bacterial wilt of geraniums caused by R. solanacearum [22] and Fusarium wilt in Monterey pines [23]. Moreover, KP can trigger the expression of host defence genes to resist plant pathogen invasion [24,25] and directly inhibit the growth of pathogens, such as Phytophthora [26,27]. The metabolic processes of Phytophthora related to phosphorus absorption are likely affected by KP. In addition, serval studies have shown that the application of phosphite can stimulate citrus yields [28] and increase the dry weights of the shoots and roots of cucumber plants [29]. Overall, KP may represent a promising chemical in agricultural systems.
Here, we determined the effects of KP on the composition, co-occurrence, and antagonistic functions of the soil bacterial community. Moreover, isolation and genomic analysis of potential antagonistic strains, determination of the antagonistic genes abundances as well as the associations of antagonistic strains, and a soil microbiome transfer experiment were performed, with the aims of identifying the agricultural chemicals that could contribute to selective enrichment of Streptomyces and Bacillus and of uncovering the mechanism of R. solanacearum inhibition by the enriched strains. We hypothesized that the KPmodulated soil microbiome may inhibit R. solanacearum indirectly.
The tomato bacterial wilt pathogen R. solanacearum ZJ3721 (biovar 3) used in the experiments was kindly provided by Professor Jianhua Guo; the KP solution was adjusted to a pH of 7.0 with potassium hydroxide; Luvisol soil (FAO) was collected from the upper soil layer (5-30 cm) of an open eld covered with grass at the Experimental Base of Nanjing Agricultural University (32.01´N, 118.85´E). The soil characteristics are listed in the Supplementary methods (Table S5). Two other soils Luvisol soil (FAO) from a grape plantation of Hebei province (36.96´N, 115.39´E) were collected from the upper soil layer (5-30 cm) of an open eld covered with grass (HN) and planted grape for two years (HG). All soils were airdried and sieved (20 mesh).

Effects of KP on the soil bacterial community
The experiment included one treatment (soil amended with KP) and a control (CK) (Fig. 7). Approximately 75 ml of sterilized water was added to 1.5 kg of dry soil from Nanjing to wet the soil. Next, 150 ml of an R. solanacearum cell suspension (cell density: 8×10 6 cfu ml -1 ) was added to the wet soil and mixed ( nal cell concentration: 8×10 5 cfu g -1 soil). R. solanacearum, as a soil-borne pathogen, can use root residuals to multiply; thus, tomato root tissues from healthy plants were added to the soils to simulate eld conditions. Approximately 15 g of dry root tissues cut into approximately 1-mm lengths was added to 1.5 kg of soil at a rate of 1.0% (w/w) [42]. Then, 750 g of soil containing root tissues was taken as the treatment and control. KP was applied at a concentration of 0.5% (w/w) [22] (KP treatment). The soil moisture of both the control and the KP treatments was adjusted to 45% of the soil capacity. Then, 750 g of soil was divided into 25 replicates consisting of 30 g of soil in a 50 ml centrifuge tube and then incubated at 30°C. The soils from the three replicates were randomly taken from all samples on days 7, 14 and 30 for soil DNA extraction.
The effects of KP on the numbers of culturable Actinobacteria were performed in two other soils (HN and HG). The same processes for the effects of KP on the microbial community of soil from Nanjing were performed. The three replicates from the two soils were randomly taken on days 7. One gram of the sampled soils was diluted and then spread on gauze No. 1 agar plates. The plates were incubated for 5 days at 28°C, and the numbers of culturable Actinobacteria were determined. The V4 hypervariable regions of the 16S rRNA gene were amplified using primers 515F (5-GTGCCAGCMGCCGCGGTAA-3) and 907R (5-CCGTCAATTCCTTTGAGTTT-3). Subsequently, 0.4 µl of the primers and approximately 10 ng of template DNA were analysed via PCR with the following thermal cycling conditions: an initial denaturation at 98°C for 1 min, followed by 30 cycles of denaturation at 98°C for 10 s, annealing at 50°C for 30 s, and extension at 72°C for 60 s, with a nal extension step at 72°C for 5 min after the cycling was complete. The PCR products were detected by electrophoresis in 1% (w/v) agarose gel and then puri ed with the GeneJET Gel Extraction Kit (Thermo Scientific). The puri ed PCR amplicons were sequenced using the Illumina HiSeq (250-bp paired-end reads) platform at the Novogene Bioinformatics Technology Co., Ltd. (Beijing, China).
The sequencing data were mainly processed on the USEARCH platform [43]. Brie y, sequences with a quality score lower than 0.5 or a length shorter than 200 nt and singletons were discarded. Noisy sequences were ltered, chimerism was inspected and an OTU cutoff was assigned at the 97% identity level. Representative sequences for each OTU were selected and classi ed according to the RDP database for bacteria (cutoff = 80%). To correct for differences in the sequencing depth, the bacterial read counts were rare ed to the lowest number of sequences present in each sample set. The raw sequences were submitted to the NCBI Sequence Read Archive (SRA) under BioProject accession PRJNA577427.
The functional genes of the bacterial community were predicted by PICRUSt [44]. The 16S sequences were used for closed-reference OTU selection with QIIME [45]. The resulting OTU table was used to predict the functional genes based on the metagenome inference work ow.

Bacterial isolation and Sanger sequencing of 16S rRNA genes
The sampled soil at 7 days from the KP treatment was serially diluted and then spread on nutrient agar (NA) and gauze No. 1 agar plates. The plates were incubated for 3-5 days at 28°C. Because our target microbes are dominant taxa in the results of high-throughput sequencing, 33 dominant colonies (numbers of similar morphology > 5 in each plate) were selected. A loop of bacterial cells was added to 500 µl of water and incubated for 15 min at 95°C. Next, the cells were cooled on ice for 1 min and centrifuged at 10,000 × g for 1 min to remove the cell debris. The supernatant of the cell lysate was used as a DNA template for the ampli cation of the 16S rRNA genes. Information on the detailed primers F27 and R1492 and the PCR steps are listed in Table S6 and Table S7, respectively. Sanger sequencing was performed by the Qin Ke Company (Nanjing, China). The sequences of these bacteria were classi ed against the 16S ribosomal RNA database using NCBI BLAST. The 16S rRNA sequences of the isolates were clustered to the OTUs at 100% sequence similarity in USEARCH.
Draft genomes of two strains of bacteria identi ed as Paenibacillus favisporus Y7 and Streptomyces coelico avus F13 The genomes of the two isolates Paenibacillus favisporus Y7 and Streptomyces coelico avus F13, which were highly abundant in the microbial community of the KP-treated soil, were sequenced. The sequences were analysed according to the method described in a previous study [46]. Brie y, the low-quality sequences were removed by adapter removal (version 2.1.7). After ltering, a total of 9,141,712 (98.65%) and 16,401,350 (97.68%) high-quality paired-end reads were obtained for strains Y7 and F13, respectively. All reads were quality corrected by SOAPec (version 2.0) based on the k-mer frequency, with the k-mer used for correction set to 17. The genome was assembled de novo using A5-miseq (version 20150522). The draft genome sequences of strains Y7 and F13 contained 17 and 59 contigs (a sequence length greater than 1 kb), respectively. The coding DNA sequences (CDs) in the draft genomes were predicted by GeneMarkS (version 4.32). The predicted CDs were searched against the NCBI NR protein database and the Kyoto Encyclopedia of Genes and Genomes (KEGG). The whole-genome shotgun project has been deposited in GenBank under the accession number WIBG00000000 (strain Y7) and WFLH00000000 Mutual stimulation between strains S. coelico avusF13 and P. favisporusY7 in vitro Strain P. favisporus Y7 and strain S. coelico avus F13 were cultured in 200 ml of one-tenth LB liquid at 30°C (strain Y7) for 2 days and 28°C (strain F13) for 5 days, respectively, at 170 rpm. Then, the culture suspensions were centrifuged at 10,000 × g for 5 min. The supernatants were sterilized with 0.22-mm sterile lter membranes. Next, 20 µl of cell suspensions of strains Y7 (approximately 1×10 8 cfu ml -1 ) was inoculated in 3 ml of one-tenth LB liquid containing the sterilized supernatants of strains F13 at concentrations of 10%, 30% and 50% (v/v). Approximately 20 µl of spore suspensions of F13 (approximately 1×10 8 cfu ml -1 ) was inoculated in 20 ml of one-tenth LB liquid containing the sterilized supernatants of strains Y7 at concentrations of 10%, 30% and 50% (v/v). The one-tenth LB liquid without strain supernatants was used as the control. Each treatment had four replicates. After 24 h of incubation at 30°C (strain Y7) and 28°C (strain F13) at 170 rpm, the cell density (OD 600 ) of strain Y7 and the dry biomass of strain F13 were measured.
Inhibition of R. solanacearum by combined P. favisporus Y7 and S. coelico avus F13 in sterile soil Approximately 50 ml of sterilized water was added to 1 kg of dry sterilized soil (2 × 99 min at 121 °C) to wet the soil. Approximately 50 ml of the R. solanacearum cell suspension (cell density at 2×10 6 cfu ml -1 ) and 10 g of sterilized root tissues (approximately 1 mm lengths) were added to the wet soils and mixed ( nal R. solanacearum concentration of 1×10 5 cfu g -1 dry soil). Then, all soils were divided evenly into 6 parts (per part: 166 g). Approximately 10 ml of the spore suspension of strain F13 at 1.66×10 6 cfu ml -1 (part 1), 10 ml of strain Y7 at 1.66×10 6 cfu ml -1 (part 2) and combinations of Y7 and F13 at different inoculation concentrations (Y7: F13 = 7:3 (part 3), 5:5 (part 4), 3:7 (part 5)) were added to the 166 g of soil to a nal density of 1×10 5 cfu g -1 dry soil. Approximately 10 ml of sterile water was added to 166 g of soil as a control (part 6). Before being added to the soil, all strains were washed 5 times. Approximately 30 g of the soil from each part was placed in a 50-ml sterilized centrifuge tube with 5 replicates and incubated at 30°C for 20 days; then, the soil samples were used for DNA extraction.
Inhibition of R. solanacearum by the KP-modulated soil microbiome A soil microbiome transfer experiment was performed to determine the function of the KP-modulated soil microbiome. The soil microbiome transfer experiment was performed based on a previous study [47] with slight modi cations. The same processes for determining the effects of KP on the soil microbial community were performed but without the addition of R. solanacearum in the soil from Nanjing. On day 7, 5 g of KP-treated soil and control soil were added separately to 45 ml of sterile water in a ask on a shaker at 200 rpm for 30 min followed by soni cation for 1 min at 47 kHz twice with shaking for another 30 min. Next, the soil suspension of each treatment was ltered with sterile lter paper (15 μm pore size) to remove soil particles. To remove water-soluble nutrients/chemicals, such as KP, the ltrate was centrifuged at 3,000 × g for 30 min, and the supernatant was discarded. The pelleted microorganisms were resuspended in 5 ml of sterile water.
Approximately 20 ml of sterilized water was added to 360 g of dry sterilized soil (2 × 99 min at 121 °C) to wet the soil; this soil was mixed with 9 ml of the R. solanacearum cell suspension (cell density at 8×10 7 cfu ml -1 ) and 3.6 g of dry sterilized tomato root tissues (approximately 1-mm lengths) and then divided evenly into 2 parts. Then, 3 ml of the cell suspension extracted from the KP treatment and control soils was applied separately to 180 g of sterilized soil containing root tissues. Each 180-g sample of the above soil was divided into 6 replicates of 30 g each in 50-ml sterilized centrifuge tubes, referred to as the MK microbial community from the KP-treated soil (MK) or the microbial community from the control soil (MC) treatments, and then incubated at 30°C. The soils from three replicates were randomly sampled on day 20 [48] for DNA extraction.
Quantitative PCR to measure the copies of functional genes The copy numbers of the iC (R. solanacearum), 16S rRNA (total bacteria), con31_49 (S. coelico avus F13), con2_67 (P. favisporus Y7), the main antagonistic genes for Bacillus (srf (surfactin), itu (iturin) and fen (fengycin)) were quanti ed by quantitative PCR. After sequencing the genomes of S. coelico avus and strain P. favisporus, the unknown functional genes from contig31_6449 (strain S. coelico avus) and contig2_3067 (strain P. favisporus) were found without homologous genes in the NCBI. Thus, the gene fragments were used to design primers (con31_49 for S. coelico avus and con2_67 for P. favisporus) in Premier 5. The designed primers were tested by primer-blast (www.ncbi.nlm.nih.gov/tools/primer-blast/), and no microbes were obtained.
Quantitative PCR (qPCR) assays were performed using the SYBR Premix Ex Taq TM (Perfect Real-Time) Kit (Takara Biotechnology Co., Dalian, China) with the ABI StepOne TM Real-Time PCR System (Applied Biosystems, USA). Each reaction was performed in a 20 µl volume. The detailed primer information and PCR steps are listed in the supporting information (Table S6 and Table S7). Standard curves were developed by serially diluting the plasmids with known positive inserts to nal concentrations of 10 2 to 10 7 gene copies µl −1 . The QPCR e ciencies ranged from 90% to 105%, and the R 2 values for all four assays were greater than 0.99.

Data analysis
The OTU tables were converted into a suitable input le for bacterial diversity analysis using Mothur [49]. Principal coordinate analysis (PCoA) of the bacterial community structure was performed by calculating the Bray-Curtis dissimilarity in R. Differences between groups were tested for signi cance by a permutation-based analysis of variance by using the adonis function of the vegan package in R. The signi cant genera (top50) between the control and the KP treatment and the MC and MK treatments were identi ed with DESeq in R, and the signi cant genera were shown in a circular treemap using the ggraph package in R.
Signi cant differences in the bacterial community diversity indices, the number of culturable Actinobacteria, the number of functional gene copies, colony diameters, inhibition zone diameters, the cell densities (OD 600 ) of R. solanacearum and strain P. favisporus Y7 and the biomass of strain S. coelico avus F13 from the inhibition level results were assessed using Student's t test; the data conformed to a normal distribution according to the Shapiro-Wilk test in R. Signi cant differences in the relative abundance levels of the PICRUSt-predicted genes were assessed using Welch's t-test.
The data on the cell density (OD 600 ) of strain P. favisporus Y7 and the biomass of strain S. coelico avus F13 from the mutual stimulation results were subjected to one-way ANOVA and then Tukey's test for multiple comparisons.
The OTU abundances (top 300) were used to build a microbial network by the function sparcc with 100 permutations in Mothur software. Edges whose p.adjust value was < 0.001 were retained. The link number of nodes (OTU degree) and visualized networks were assessed using Gephi software [50].
Phylogenetic trees of nodes were built by MEGA7. Moreover, the associations between the relative abundances of hubs (link numbers >50) from bacterial networks and the predicted genes related to production of antagonistic substances were determined by the cor.test function in R with the "pearson" method. The p-values were adjusted by the "FDR" method.
The gene annotation protein.faa les of strains Y7 and F13 were used to construct genome-scale metabolic models with "carve" functions, and the genome-scale metabolic models of strains Y7 and F13 were used to generate the microbial community models with the "merge_community" functions by CarveMe [51]. The metabolic substances and reactions of the microbial community models were obtained by the sybilSBML package in R, and the associations between metabolic substances and reactions were visualized using Cytoscape software [52].

Results
Effects of KP on R. solanacearum, total bacteria and the soil bacterial community The copy numbers of the iC (representing R. solanacearum) and 16S rRNA genes (representing total bacteria) were signi cantly reduced by approximately 1.91-fold to 2.44-fold and 1.61-fold to 4.81-fold, respectively, at all sampling times in the KP treatment ( Fig. 1a and Fig. 1b). The Shannon diversity of the soil bacterial community was slightly decreased in the soil amended with KP on days 7 and 14 (Fig. S1); however, the diversity was recovered at day 30. The microbial community structure of the control at different sampling times was signi cantly different from that of the KP treatment (P = 0.001, R = 0.45) (Fig. 1c). The difference in bacterial community structure between the control and KP treatment at day 7 was greater than that at days 14 and 30. Moreover, the bacterial community structure of the control at day 7 was different from that at days 14 and 30, suggesting that the time factor of tomato root residuals signi cantly affected the soil bacterial community.
Actinobacteria, Proteobacteria, Firmicutes and Bacteroidetes were the dominant phyla in the bacterial communities (Fig. S2). Actinobacteria (56.93%) was enriched, but Proteobacteria (16.21%) and Bacteroidetes (2.67%) were depleted in the KP treatment at days 7 and 14. However, the relative abundance of Proteobacteria (30.63%) in the KP treatment was higher than that in the control at day 30.
To increase the robustness of our results, we determined the number of culturable Actinobacteria in two other soils supplemented with KP. The numbers of culturable Actinobacteria in the HN and HG soils amended with 0.5% KP were 1.5-fold and 1.4-fold higher than those in the corresponding controls, respectively (Fig. S3).
At the genus level, the relative abundance of the dominant bacterium Streptomyces (25.29%) was signi cantly increased in the KP treatment for all sampling days (Fig. 1d, Fig. S4 and Fig. S5). Paenibacillus was signi cantly enriched in the KP treatment at days 7 and 14 ( Fig. 1d and Fig. S4). Bacillus was signi cantly enriched in the KP treatment only at day 7 (Fig. 1d). Speci cally, in addition to Micromonospora, Arthrobacter, Agromyces and Sporosarcina, the other signi cantly changed genera (top 50) belonging to Actinobacteria and Firmicutes, such as Actinoallomurus and Actinocatenispora, were signi cantly enriched in the KP treatment at day 7 (Fig. 1d). In addition to Sphingomonas, the other signi cantly changed genera (top 50) belonging to Bacteroidetes and Proteobacteria, such as Massilia and Flavobacterium, were signi cantly depleted in the KP treatment at day 7 (Fig. 1d). However, Bacillus and Flavobacterium were signi cant depleted and enriched in the KP treatment at day 14 and 30, respectively ( Fig. S4 and Fig. S5). Moreover, Sphingobium, Chitinophaga and Ensifer were signi cantly depleted in the soil of the KP treatment at all sampling times.
Effect of KP on the co-occurrence networks of the soil bacterial community Because of the difference in the composition of the bacterial community, we compared the bacterial networks between the KP treatment and control. The results showed that KP application slightly increased the link number of the bacteria community co-occurrence networks (Fig. S6).
The nodes with > 50 links represented hubs in the bacterial network. The top 10 hubs are shown in Table   S1. We found that the OTUs belonging to Streptomyces, Paenibacillus and Bacillus had the greatest link numbers. The link number of OTUs (Bacillus) was greatly higher in the KP treatment than in control, while the link number of OTUs (Paenibacillus and Streptomyces) was slightly lower in the KP treatment.
Due to the fact that Streptomyces, Paenibacillus and Bacillus were the dominant potential antagonistic bacteria and the top hubs, the associations between these potential antagonistic bacteria and other bacteria were determine. Unless otherwise noted, the target hub (link number > 50) refers hereafter to the nodes belonging to the potential antagonistic bacteria (Streptomyces, Paenibacillus and Bacillus). Our results showed that the positive links between the target hubs and nodes belonging to Actinobacteria and Firmicutes (not including Streptomyces, Paenibacillus and Bacillus) were increased 1.73-fold and 1.22fold, but the relative abundances of these nodes were decreased 1.36-fold and 2.28-fold in the bacterial network of KP treatment compared to those in the control network (Table S2). The low and high relative abundances of nodes that have positive associations with the target hubs were increased and decreased, respectively, in the bacterial network of KP treatment (Fig. S7). In addition, the opposite results were found in the negative links between the target hubs and these nodes (Table S3 and Fig. S4). These results suggested that the dominant target hubs may enhance the cooperation and competition with rare and dominant bacteria, respectively.
The associations between target hubs and the nodes belonging to Proteobacteria and Bacteroidetes were determined. We found that both the negative and positive link numbers between the nodes (Proteobacteria and Bacteroidetes) and the target hubs were decreased, suggesting that certain lowabundance bacteria belonging to Proteobacteria and Bacteroidetes may disappear with KP treatment, resulting in the decreased link numbers of these nodes (Table S2 and S3). In addition, the relative abundances of both rare and dominant nodes (Proteobacteria and Bacteroidetes) that had negative and positive links to the target hubs were also decreased in the KP-treated soil, suggesting that the target hubs corresponding to the potential antagonistic bacteria may inhibit the growth of the bacteria belonging to Proteobacteria and Bacteroidetes.
The associations among these target hubs showed that the numbers of the target hubs were increased 1.5-fold, and the positive link numbers among the associations of themselves were increased 3.19-fold in the KP-treated soil compared to those in the control soil (Fig. 2). Moreover, the relative abundances of the increased target hubs from the KP treatment were signi cantly lower by approximately 4.9-fold to 199.6fold than those of the dominant target hub (Streptomyces_OTU1). In summary, KP may enhance the cooperative associations among these target hubs.
The associations between the target hubs and predicted antagonistic genes from PICRUST were determined (Fig. S8). In addition to OTU38 (Streptomyces), the other OTUs (Streptomyces) and OTU39 (Paenibacillus), whose link numbers were more than 50, had signi cant and positive associations with the predicted antagonistic genes related to biosynthesis of 14-membered macrolides, type II polyketide products, vancomycin, butirosin, penicillin, and streptomycin, suggesting that these OTUs corresponded to bacteria that could produce these antagonistic substances. However, positive associations were not found between the OTUs (Bacillus) and these genes related to the biosynthesis of antagonistic substances described above.

Isolation and characterization of potential antagonistic microbes in soil
To test the results related to the positive associations among the target hubs and their antagonistic functions, we isolated 33 dominant strains from the KP-treatment soil and found OTU1 and OTU39, for which the sequences were clustered to the 16S rRNA sequences of the strains F13 (Streptomyces coelico avus) and Y7 (Paenibacillus favisporus) at 100% sequence similarity (Fig. 3). q-PCR results showed that the copy numbers of the speci c gene fragments con31_49 of S. coelico avus F13 and con2_67 of P. favisporus Y7 were signi cantly higher in the KP treatments than in the control (P = 0.01) at three sampling times (Fig. 3b), which con rmed the enrichment of the strains S. coelico avus F13 and P. favisporus Y7 in the soil amended with KP.
Based on the enrichment of the isolates P. favisporus Y7 and S. coelico avus F13 and the depletion of R. solanacearum in the KP treatment, we speculated that the inhibition levels of R. solanacearum, P. favisporus Y7 and S. coelico avus F13 by KP were different. Indeed, R. solanacearum barely grew in LB liquid medium amended with KP at a concentration of 0.5% (Fig. S9). KP also inhibited the growth of P. favisporus Y7 and S. coelico avus F13 in vitro, but the inhibition levels of P. favisporus Y7 and S. coelico avus F13 by KP were much lower than that of R. solanacearum.
We determined whether the enriched isolates P. favisporus Y7 and S. coelico avus F13 could inhibit R. solanacearum growth in vitro. Indeed, obvious inhibition zones of P. favisporus Y7 and S. coelico avus F13 to R. solanacearum were found on NA plates (Fig. S10).
Because a positive association between OTU1 (Streptomyces) and OTU39 (Paenibacillus) was found in the bacterial networks of KP treatment, we speculated that S. coelico avus F13 and P. favisporus Y7 may form a mutualistic association to inhibit the growth of R. solanacearum. The results showed that the fermentation broth of S. coelico avus F13 could signi cantly stimulate the growth of P. favisporus Y7 at all tested concentrations (P < 0.001), and the highest stimulation of P. favisporus Y7 was found at a concentration of 30% (v/v) fermentation broth of S. coelico avus F13 (Fig. S11). Similarly, the growth of S. coelico avus F13 was stimulated by the fermentation broth of P. favisporus Y7 at three tested concentrations. In addition, the cell suspension of S. coelico avus F13 was more homogeneous in LB liquid medium amended with the fermentation broth of P. favisporus Y7 than in the control.
Functional analysis based on the genomes of P. favisporus Y7 and S. coelico avus F13 We further determined the functions of the isolates from their genomes to support the above results. Due to their antagonistic ability against R. solanacearum, the KEGG metabolic pathways of the biosynthesis of other secondary metabolites (Fig. 4) were considered. The results showed that both strains contained the metabolic pathways related to the biosynthesis of prodigiosin, monobactam, streptomycin, phenylphopaniod and validamycin. Speci cally, the contig numbers of streptomycin biosynthesis were the highest in both strains. The operon (rfbA, rfbB, rfbC and rfbD) that encodes dTDP-L-rhamnose, which is an important branch related to the production of dTDP-L-dihydrostreptose in streptomycin biosynthesis, was found in both strains. Furthermore, distinct genes related to the production of surfactin for P. favisporus Y7 and germicidin and toxo avin for S. coelico avus F13 were found (Table S4).
Because the antagonistic gene stfAB was found in the genome of P. favisporus Y7, we speculated that these genes related to the production of surfactin were enriched in the soil amended with KP. Thus, the copy numbers of three genes related to the production of the main lipopeptides (srf, itu and fen) were determined in the soil (Fig. S12). We found that the copy numbers of srf, itu and fen were approximately 2.22-fold to 3.51-fold higher in the KP treatment than in the control at 7 days. Similar differences in the copy numbers of the antagonistic genes were found at 14 and 30 days between the KP treatment and control.
Because of the mutualistic associations between strain F13 and Y7, we constructed a genome-scale metabolic model of strains F13 and Y7 to identify cross-feeding metabolites. In general, cross-feeding metabolites are produced by one microbe and utilized by another microbe in extracellular environments. Thus, we focused on the metabolic reactions in extracellular environments. The results showed that 312 metabolic reactions were existed in both strain F13 and Y7, suggesting that certain nutrients could be utilized by both strains. Moreover, 134 and 55 distinct metabolic reactions were found in strains F13 and Y7, respectively, suggesting that certain distinct metabolites could be produced (Fig. S13). The detailed metabolites in extracellular environments were further determined, and the results showed that 65 main carbon and nitrogen sources, speci cally, cellobiose and xylobiose, could be utilized by both strains (Fig.   S14). Both strains harbour the operon (xylA, xylB and xylD) for D-xylose metabolism, suggesting that both strains can use certain nutrients from soil root residuals (Fig. S15 and Fig. S16). Moreover, dextrin could be produced via alpha amylase reaction in strain F13 and utilized by strain Y7 for dextrin transport via proton symport and maltodextrin glucosidase dextrin reactions (Fig. 5). L-lactate could be produced via lactaldehyde dehydrogenase in strain Y7 and utilized by strain F13 for lactate reversible transport via proton symport, L-lactate dehydrogenase and tagatose 1,6-diphosphate aldolase reactions.
Synergistic inhibition of R. solanacearum by Antagonistic strains Y7 and F13 in soil Since the cross-feeding existed between strain Y7 and strain F13, a soil culture experiment was performed to determine whether strains Y7 and F13 could inhibit R. solanacearum synergistically at different inoculation rates. According to the results of the soil culture experiment, P. favisporus Y7 and S. coelico avus F13 in combination or alone could signi cantly inhibit R. solanacearum growth in a sterile soil (Fig. S17). S. coelico avus F13 exhibited less inhibition than P. favisporus Y7. In addition, the lowest copy number of iC was found in the soil amended with P. favisporus Y7 and S. coelico avus F13 at an inoculation ratio of 7 to 3, suggesting that synergistic inhibition of R. solanacearum existed.

Inhibition of R. solanacearum by the KP-modulated soil microbiome
As the antagonistic bacteria Streptomyces and Paenibacillus were highly enriched in the soil amended with KP at 7 days, we speculated that the KP-modulated microbiome from 7 days may inhibit the growth of R. solanacearum. The results showed that the copy number of iC (R. solanacearum) was signi cantly lower by approximately 1.38-fold in the MK treatment (the soil amended with the microbiome from the KP treatment) than in the MC treatment (the soil amended with the microbiome from the control) (P = 0.004) (Fig. S18a). The qPCR results showed that the copy number of the speci c gene fragments con31_49 of S. coelico avus F13 and con2_67 of P. favisporus Y7 were signi cantly higher in the MK than MC treatments (P = 0.01) (Fig. S18b and c).

Discussion
KP affected the composition of the soil bacterial community KP, as an environment-friendly compound, is attracting increasing attention. Previous studies have focused on the function of KP in the induction of host resistance and the inhibition of pathogen growth [30]. In our study, we found that the relative abundance levels of Actinobacteria were increased, but Proteobacteria and Bacteroidetes were decreased in the soil amended with KP (Fig. S2). In agreement with our study, when the fungicide azoxystrobin was added to soil, the relative abundance of Actinobacteria was signi cantly increased [14]. The application of metam sodium fumigation could enrich the potential antagonistic microbes Actinomycetales and Bacilli [13]. However, Micromonospora and Sporosarcina belonging to Actinobacteria and Firmicutes, respectively, were signi cantly depleted in the KP-treated soil. It was possible that the enriched bacteria may indirectly inhibit the growth of Micromonospora and Sporosarcina. In addition, the differences in bacterial community structure were gradually decreased between the control and KP treatment from days 7 to 30, suggesting that the KPmodulated soil bacterial community could be gradually recovered (Fig. 1c). Likely, KP was gradually converted to phosphate in soil as a result of soil microbial activity [31].

KP affected the networks and antagonistic functions of the soil bacterial community
In our study, the numbers of positive links these target hubs were increased in the bacterial network of KPtreated soil, and the relative abundances of increased hubs in the KP treatment were much lower than that of the target hub OTU1 (Streptomyces) (Fig. 2). Similar results were also found in the positive associations between target hubs and the nodes belonging to Actinobacteria and Firmicutes (not including the target hubs) (Fig. S7), suggesting that a substantial content of metabolites from the dominant bacteria may have positive effects on the non-dominant bacteria. KeheJ et al. tested the associations among 60 strains and found that positive interactions commonly occurred between dominant and non-dominant bacteria [32]. Moreover, the relative abundances of the OTUs belonging to Proteobacteria and Bacteroides that had negative associations with the target hubs were signi cantly decreased in the KP treatment. In addition, the positive associations between certain target hubs, such as OTU1 (Streptomyces) and OTU39 (Paenibacillus), and the predicated antagonistic genes were found in the KP-treated soil (Fig. S8). This result suggested that the bacteria belonging to these target hubs may inhibit the growth of bacteria belonging to Proteobacteria and Bacteroides. In line with our study, the application of glucose and fructose in soil resulted in Streptomyces populations with greater niche widths and selectively enriched Streptomyces antagonistic phenotypes to inhibit the nutrient competitors [33]. In summary, application of KP may enrich the symbiotic antagonistic bacteria that play important roles in modulating the soil microbiome.
KP induced the antagonistic bacteria S. coelico avus F13 and P. favisporus Y7 to inhibit the growth of R. solanacearum synergistically To support the above results, we isolated 33 dominant strains in the soil amended with KP and found that S. coelico avus F13 and P. favisporus Y7 had stronger tolerance to KP than R. solanacearum (Fig. S9).
The previous study showed that because Microcystis aeruginosa (Cyanobacteria) had stronger tolerance to the fungicide azoxystrobin than Chlorella pyrenoidosa (Chlorella), azoxystrobin could favour M. aeruginosa growth through growth inhibition of C. pyrenoidosa in a freshwater ecosystem [34]. Most likely, the microbes that were susceptible to KP were depleted, and their niches were occupied by S. coelico avus F13 and P. favisporus Y7, which may be the main reason for the enrichment in the KPtreated soil. In addition, S. coelico avus F13 and P. favisporus Y7 (gram-positive bacteria) form a thick and dense peptidoglycan layer, and this layer may play an important role in resisting adversity. In agreement with our study, the application of metalaxyl had a transient stimulation effect on Actinomycetes [12]. On the other hand, both strains (F13 and Y7) contained genes related to the utilization of D-xylose, which may have helped them to occupy many niches in the soil amended with root residues (Fig. S15 and Fig. S16).
A previous study has shown that S. coelico avus can produce 1H-pyrrole-2-carboxylic acid to inhibit the quorum sensing of Pseudomonas aeruginosa [35]. P. favisporus isolated from the tomato phyllosphere can induce resistance in tomato plants to suppress root rot [36]. However, no previous studies have shown that these two bacteria can inhibit the growth of the tomato pathogen R. solanacearum. In our study, antagonistic genes related to the synthesis of streptomycin were found in P. favisporus Y7 and S. coelico avus F13 (Fig. 4), which was in line with the results from the predicated function of the soil microbiome. Streptomycin sulphate has been used to manage tomato bacterial wilt disease [37]. Antagonistic genes in P. favisporus Y7 related to the synthesis of surfactin, which can inhibit the growth of R. solanacearum [38], were also found. Indeed, the copy number of srf (surfactin) was signi cantly increased with KP treatment (Fig. S12). Moreover, certain antagonistic genes of Bacillus related to the production of other lipopeptides, such as iturin and fengycin, were also enriched in the soil of KP treatment, suggesting that other potential antagonistic Bacillus species may be enriched. Moreover, the genome-scale metabolic model showed that the distinct metabolites dextrin and lactic acid could be produced by strain Y7 and strain F13, respectively, and utilized by strain F13 and strain Y7, respectively ( Fig. 5) that may be the main reason for in the synergistic inhibition of R. solanacearum by strain F13 and strain Y7. Moreover, the KP-modulated soil microbiome cultivated in sterile soil could also inhibit the growth of R. solanacearum (Fig. S18), supporting the above results. Overall, a model was proposed that the symbiotic antagonistic bacteria S. coelico avus F13 and P. favisporus Y7 were enriched in soil amended with KP, resulting in the inhibition of R. solanacearum indirectly (Fig. 6).

Conclusion
Here, we showed that KP application enriched the potential antagonistic genera Streptomyces and Paenibacillus and enhanced the positive associations among these potential antagonistic genera as well as the antagonistic functions of the soil microbiome. Moreover, two antagonistic strains whose 16S RNA sequences were clustered to OTU1 (Streptomyces) and OTU39 (Paenibacillus) have stronger tolerance to KP than R. solanacearum. In addition, both strains have genes related to the production of streptomycin. The mutual stimulation between P. favisporus Y7 and S. coelico avus F13 in vitro may result in synergistic inhibition of R. solanacearum growth. The results of genome-scale metabolic modelling showed that dextrin and lactic acid were potential cross-feeding metabolites. Furthermore, the KPmodulated soil microbiome could suppress the growth of R. solanacearum. The results of this study suggest that the KP-modulated soil microbiome has considerable potential for biocontrol and indicate a new mechanism for inhibition of R. solanacearum by KP-enriched soil bacteria. Figure 1 Effects of KP on the soil bacterial communities Copy numbers of iC (a) and the 16S rRNA (b) between the KP treatment and the control. Statistical signi cance was determined based on Student's t test. * P < 0.05, ** P < 0.01, *** P < 0.001. Unless otherwise noted, "CK" and "KP" refer hereafter to control and KP treatment, respectively. (c) Principal coordinate analysis of the bacterial communities in the KP-treated soil and the control soil at different time points. (d) Signi cant differences in the relative abundance levels of bacterial genera between the KP treatment and the control at 7 days. The largest circles represent the phylum level. The inner circles represent the genus level. The colours of the circles indicate genera enriched in the KP treatment (green) or the control (red). The size of the circle represents the relative abundance of that genus.

Figure 2
The associations among the hubs belonging to the potential antagonistic bacteria (Streptomyces, Paenibacillus and Bacillus) in the of control (a) and KP treatment networks (b). The green and red lines represent negative and positive links among bacteria, respectively. The node size represents the relative abundances of the node. (a) The phylogenetic diversity of the isolates from the soil amended with KP at the species level. The isolates, for which the 16S rRNA sequences were clustered to sequences of target OTUs (Streptomyces and Paenibacillus) at 100% sequence similarity, are marked with a light red background. (b) Copy numbers of con31_49 of strain Streptomyces coelico avus F13 and (c) con2_67 of strain Paenibacillus favisporus Y7 in the soil amended with KP. Statistical signi cance was determined based on Student's t test. * P < 0.05, ** P < 0.01, *** P < 0.001. Statistical signi cance was determined by Student's t test. ** P < 0.01, *** P < 0.001.

Figure 4
Functional analyses of the genomes of the isolates S. coelico avus F13 and P. favisporus Y7. The biosynthesis contig numbers of the other secondary metabolites in S. coelico avus F13 (a) and P. favisporus Y7 (b). The KEGG metabolic pathway of streptomycin biosynthesis in S. coelico avus F13 and P. favisporus Y7 (c). Gene labels marked with black or red font indicate that these genes were found in both isolates or only the isolate S. coelico avus F13, respectively.

Figure 5
The potential cross-feeding metabolites dextrin (a) and L-lactate (b) in the genome-scale metabolic model between S. coelico avus F13 and P. favisporus Y7. Squares and circles represented reactions and metabolites, respectively. The green, red and pink circles represent the metabolites in the cytosol of strain Y7, cytosol of strain F13 and extracellular environments. The arrows and line segments represent the production and utilization of metabolites, respectively.  Flow diagram of the experimental design of this study