Specific ecological niches and host resistance drive the assembly of associated bacterial communities
After inoculating F. graminearum into the root zones of 40 maize cultivars (to avoid injuring the roots) for two consecutive years, we identified pronounced and consistent differences in CSR disease indexes (Supplementary Table 1). Infected maize plants exhibit hollow stems and roots covered with red mycelium, which eventually cause the plant to break (Fig. 1a). To analyze the bacterial and fungal communities of maize plants that are resistant or susceptible to CSR, we collected samples from four resistant (disease incidence = 0%) and four susceptible (disease incidence > 30%) maize cultivars from five ecological niches: bulk soil (BS), rhizosphere soil (RS), root endosphere (RE), stem endosphere (SE) and grain endosphere (GE; Fig. 1b). We determined the disease incidence by recording the percentage of diseased seedlings relative to the total number of seedlings. We profiled the compositions of the bacterial and fungal communities based on sequencing of 16S ribosomal RNA (rRNA) gene fragments and internally transcribed spacer (ITS) sequences, followed by clustering into operational taxonomic units (OTUs; 97% identity).
We obtained 12,824,344 rRNA and 15,788,680 ITS high-quality reads from 198 samples, ranging from 30,162 to 94,013 reads per sample, with an average of 72,255 reads per sample, representing 4,624 bacterial and 1,471 fungal OTUs. We employed a linear mixed model (LMM) to assess which major factors shape the maize microbiota. To quantify species diversity within a microbial community, we also calculated the Shannon diversity index (SDI) for each community, with high values representing higher species diversity. We observed that ecological niches, together with resistant/susceptible genotypes, have a greater influence on bacterial SDI (P = 0.0087) than do differences in maize cultivars (P = 0.629; Supplementary Table 2).
We evaluated the assembly process of microbial communities, which is known to be strongly linked to the maintenance of plant health23, based on the β-nearest taxon index (βNTI), which measures the mean phylogenetic distance between taxa of a community. We detected marked differences in βNTI in bacterial communities between resistant and susceptible sample pairs among the BS, RS, RE and SE ecological niches. By contrast, we noticed no significant changes for fungal communities, suggesting that resistance has a stronger influence on the assembly of bacterial vs. fungal microbiota (Fig. 1c, Supplementary Table 3, Wilcoxon rank sum test, P < 0.05).
The bacterial communities in soil niches (BS and RS) exhibited a divergent assembly compared to the internal niches (RE, SE and GE), whereas we detected no significant differences in fungal community assembly among the five niches analyzed (Extended Data Fig. 1a). Furthermore, the proportion of deterministic (HS: homogeneous selection and VS: variable selection, |βNTI| > 2) and stochastic (DDH: dispersal, drift and homogenizing dispersal, |βNTI| < 2) processes was divergent across niches (Fig. 1d). The proportion of VS (βNTI > 2) of bacterial communities increased in the RE and SE niches of resistant cultivars relative to the other three niches, indicating that the phylogenetic turnover and interaction among microorganisms were higher than expected. Neutral community model analysis showed that the dispersal ability and habitat niche breadth of bacterial communities decrease gradually from soil to internal niches (Extended Data Fig. 1b, c). In the RS and RE niches, the dispersal limitation on bacterial communities was higher in resistant cultivars than in susceptible samples, as reflected by the lower Nm values (an estimate of dispersal between communities). Collectively, these findings indicate that the RE niche of resistant cultivars undergoes changes in its bacterial, rather than fungal, communities upon F. graminearum infection and that this niche plays a prominent role in host plant resistance.
Distinct patterns of microbial diversity characterize different maize niches
Assessments of Shannon diversity revealed significant differences between soil and plant niches in both bacterial and fungal communities (Tukey’s multiple comparisons test P < 0.05, Fig. 1e, Supplementary Table 4). As a whole, the richness of bacterial and fungal species was lower in the maize internal compartments than in the soil, with GE displaying the least diversity. The SDIs of bacterial communities showed no significant differences (Wilcoxon rank sum test, P > 0.05) between resistant and susceptible groups among the five niches. By contrast, the SDIs of the fungal community were lower in the SE than RE niches of susceptible groups, suggesting that the microenvironment is unstable in these two niches in susceptible cultivars (Extended Data Fig. 1d).
Principal coordinates analysis (PCoA) based on Bray–Curtis dissimilarity clearly separated bacterial communities in the soil from the plant internal samples along the first principal coordinate (Fig. 1f, Supplementary Table 5). In parallel, we observed significant differences in the composition of bacterial communities in the three internal niches along the second principal component. We noticed a similar positional variation pattern of fungal communities, as fungal communities in SE differed markedly from those in the other four niches. Moreover, we identified substantial differences in bacterial and fungal communities between resistant and susceptible samples in each niche, except for bacteria in the GE (Fig. 1g, Supplementary Table 6). Overall, these results indicate that maize niches and resistance characteristics have strong selective effects on the composition and assembly of microbiome communities.
Root-associated co-occurrence networks reveal stronger bacterial–fungal interactions in resistant maize cultivars
To investigate whether and how CSR resistance affects the complexity and stability of molecular ecological networks across the ecological niches, we constructed bacterial–fungal co-occurrence networks (Spearman’s correlation coefficient (ρ) was > 0.9 and P < 0.05). We detected a clear shift in the interkingdom network patterns across the five ecological niches. The number of nodes (N), including both bacterial and fungal taxa, was lower in internal samples (RE, SE and GE) than in soil samples (BS and RS; Fig. 2a, b). This finding was predictable, considering how soil tends to contain a greater diversity of microbes than the internal compartments of plants. Notably, the connectivity index (C) and average degree (D) in the RE, SE and GE niches did not decrease as might have been expected with the shift in nodes and edge, suggesting that the internal niches harbor stronger microbiological interactions. Compared to susceptible cultivars, roots (RE and RS) of resistant samples had more nodes and edges, with higher connectivity and average degree, representing stronger interactions. We observed an opposite pattern in maize stems (SE) compared with roots. Susceptible inbred lines had the most edges, highest connectivity and degree (Supplementary Table 7), and thus showed stronger bacterial–fungal interactions. In addition, the Bray–Curtis dissimilarity of root-associated networks (RS and RE) between resistant and susceptible groups was higher than that in the SE and GE niches, indicating that microbes in root niches were less stable than those in the stem and grain niches (Supplementary Table 8). These divergent network patterns between resistant and susceptible cultivars highlight the importance of root-associated bacteria in suppressing the growth of pathogens such as F. graminearum.
We calculated the robustness of each network by simulating species extinction. This analysis revealed marked divergence in robustness between resistant and susceptible groups by removal of either random taxa or targeted module hubs (Fig. 2c, d). Network vulnerability increased marginally, and the compositional stability and node persistence decreased along the bottom-up niches in both the resistant and susceptible groups (Fig. 2e-g). These results suggest that niches and resistance characteristics of the host plant greatly influence bacterial–fungal interactions, with the root niches of resistant cultivars conferring the greatest resistance to fungal pathogens.
Susceptible cultivars are exposed to rhizosphere microbiomes with functions related to cell wall degradation
Because the rhizosphere is a primary ecological niche for microbial, environmental and host interactions, we used metagenomic sequencing data from the rhizosphere microbiome to explore the functional variation of microbiomes associated specifically with resistant or susceptible maize cultivars. PCoA showed that CSR resistance has a pronounced effect on the functional composition (GO, CAZ, COG and KO) of rhizosphere microbes (Fig. 3a and Extended Data Fig. 2a, PERMANOVA P < 0.05). However, we did not observe persistent significant differences in functional diversity between data obtained in 2020 and 2021, as reflected by the SDI (Fig. 3b).
We then analyzed the metabolic pathways to explore the above functional alterations. We established that amino acid metabolism pathways (mainly including l-isoleucine, l-arginine and guanosine biosynthesis) are significantly enriched in the rhizosphere microbiome of resistant maize cultivars (Fig. 3c). Metabolic pathways associated with polysaccharides (such as d-glucosamine biosynthesis, d-glucarate degradation and sucrose degradation) and organic acids (such as fatty acid biosynthesis, unsaturated fatty acid biosynthesis and pantothenate and coenzyme A biosynthesis) were enriched in the susceptible maize cultivars. In addition, we identified pathways associated with the metabolism of amines (thiamine formation and thiamine salvage II) and aromatics (aromatic biogenic amine degradation) in the susceptible group.
Further analysis of the rhizosphere microbiome revealed that the abundance of 61 carbohydrate-related enzyme families (CAZy), including six carbohydrate-active enzyme classes, significantly differs between the resistant and susceptible groups (Wilcoxon, P < 0.05, Fig. 3d). Compared to the microbes associated with resistant samples (15 families), those associated with susceptible samples were enriched for CAZy families (20 families) associated with the degradation of major components of plant cell walls such as cellulose, hemicellulose or pectin. The presence of these enzymes would make it easier for pathogens to breach the cell wall barrier and invade the susceptible host plants. Three peptidoglycan-degrading (GT31, GT49 and GH103) and five chitin-degrading (GH18, GH19, GH23, CBM50 and CBM73) CAZy family members were uniquely enriched in the microbes associated with susceptible samples. The presence of these enzymes would facilitate the degradation of bacterial and fungal cell walls, respectively. Thus, the investigation of CAZy families revealed the more drastic fluctuations of carbohydrate-related family members and stronger cell wall degradation–related interactions among bacteria, fungi and maize roots in the rhizosphere microenvironments of susceptible plants.
Bacillus tends to be recruited across bottom-up niches
To define the core microbiota associated with CSR-resistant plants, we comprehensively analyzed the taxonomic compositions and relative abundances of their bacterial and fungal communities. Microbial community analysis at the phylum level showed niche specificity (Fig. 4a). The dominant bacterial phyla were Proteobacteria (63.23%), Acidobacteriota (13.10%) and Actinobacteriota (6.94%). Notably, Acidobacteriota mainly existed in the two soil niches, and the proportion of Actinobacteriota was higher in SE samples. A similar analysis of fungal communities showed that Ascomycota (65.29%) and Basidiomycota (24.93%) are the dominant phyla. Furthermore, measurement of the abundance of F. graminearum from ITS amplicon sequencing data showed that the proliferation of F. graminearum is significantly inhibited in the RS, RE and SE niches of resistant maize inbred lines (Fig. 4b).
We first identified the core taxa in resistant and susceptible groups through differential analysis, which demonstrated that four shared genera, i.e., Bacillus, Granulicella, Mucilaginibacter and Pantoea, showed significant differences among the RS, RE and SE niches (Wilcoxon rank sum test, P < 0.05, Fig. 4c). Relative abundance measurements revealed that only Bacillus tended to be recruited to the RS, RE and SE niches of CRS-resistant inbred lines (Fig. 4d). Linear discriminant analysis effect size (LEfSe) with a logarithmic LDA > 2.5 further indicated that Bacillus can serve as a useful biomarker for resistant maize cultivars (Fig. 4e and Extended Data Fig. 2b). To identify the core fungal taxa of CRS-resistant maize plants, we investigated fungal communities using the same method; however, the results were non-uniform (Extended Data Fig. 2c and d). Therefore, we selected Bacillus as a candidate core taxon for CSR-resistant cultivars and subjected it to a series of verification experiments.
The presence of Bacillus species enhances plant performance against F. graminearum
To explore the effects of the recruited bacteria on plant health, we purified 276 bacterial isolates, accounting for 43% of all the OTUs with more than five reads in RS samples, from the rhizosphere soil of resistant cultivars using gradient dilution and streak plate techniques (Supplementary Table 9). These 276 isolates, including 62 Bacillus isolates, were categorized into 17 families based on 97% identity of 16S rRNA sequencing results. In addition, we obtained 694 Bacillus isolates using the Bacillus-specific method, which clustered into 28 subgroups with 99% identity within a subgroup. We randomly selected 44 isolates from the 28 subgroups for de novo genome sequencing and protein prediction. Based on these results, we assigned accurate taxonomies by reconstructing a neighbor-joining (NJ) phylogenetic tree from 4,215 Bacillus genomes using the CVTree Standalone Version (Fig. 5a).
We identified three models representing the antagonistic phenotype of all the isolates against F. graminearum in dual culture assays, which we characterized as follows: inhibition by niche grabbing, inhibition by secreting antimicrobial compounds and no inhibition. Metagenomic mapping of 23 isolates showed an enrichment in resistant cultivars (marked by * in Fig. 5a, Wilcoxon rank sum test, P < 0.05) and represented two examples of niche grabbing, nine examples of secreting antimicrobial compounds and 12 isolates with no inhibition.
We generated three types of synthetic communities (SCs) by randomly selecting and mixing three isolates from each model in equal-volume suspensions (SC-I: inhibition by niche grabbing; SC-II: inhibition by secreting antimicrobial compounds; SC-III: no inhibition). We investigated the disease suppression activity of these SCs against F. graminearum in greenhouse experiments. Infected maize seedlings were characterized by hyphal diffusion at the base and the apparent yellowing of stems, resulting in wilting within 2 weeks (Fig. 5b, Extended Data Fig. 3a). Three individual pot experiments showed that SC-I, SC-II and SC-III significantly decreased CRS incidence in maize seedlings at 2 weeks after F. graminearum inoculation (Fig. 5c). In addition, F. graminearum treatments reduced the rate of seedling emergence and root growth, while additional inoculation with Bacillus at the same time, especially with SC-II, restored these seedling phenotypes (Extended Data Fig. 3b-d).
To gain insight into the disease resistance mechanisms of the host plant based on their responses to pure Bacillus treatment, we investigated the transcriptomic changes in maize roots treated with these three SCs. We obtained 534,236,556 high-quality reads, with an average of 44,519,713 reads per sample and a 94.6% mapping rate to the maize reference genome (B73-NAM-5.0). Compared to the control groups treated with sterile water, we identified 293, 387 and 171 differentially expressed genes (DEGs, |log2(fold-change) | > 1, Padj < 0.05) in maize roots treated with SC-I, SC-II and SC-III, respectively (Fig. 5d). PCoA based on an expression matrix of all genes showed that SC treatment did not greatly alter gene expression in maize roots (Fig. 5e), whereas PCoA based on DEGs (Fig. 5f) and functionally related genes (Fig. 5g) showed that SC-I and SC-II had similar effects on maize that were different from those observed for SC-III.
Gene set enrichment analysis in response to treatment with each SC revealed similar enrichment among DEGs for functional terms such as mitogen-activated protein kinase (MAPK) signaling and plant hormone signal transduction (Fig. 5h, Extended Data Fig. 4a, b, Supplementary Table 10). Notably, inoculation with any of the three SCs resulted in lower expression of WRKY33-homologous genes, which facilitate defense-related gene induction (Fig. 6a). In addition, SC-II and SC-III may enhance the biosynthesis of flavonoids, as evidenced by the induction of flavanone 3-hydroxylase1 (FHT1) expression.
We were most interested in the SC-III-specific functional pathway, since SC-III had no inhibitory effect on F. graminearum but significantly decreased the incidence of CSR. Furthermore, SC-III, rather than SC-I or SC-II, specifically facilitated the enrichment of maize genes related to sesquiterpenoid, isoquinoline alkaloid and betalain biosynthesis, as reflected by the normalized enrichment scores (NESs); these compounds are widely considered to be antibacterial metabolites (Fig. 5h). Both terpene synthase 6 (TPS6) and tyrosine decarboxylase 1 (TYDC1), which participate in the above pathways, were upregulated by treatment with SC-III. The functional terms phenylalanine and tyrosine metabolites, which are biosynthetic precursors of isoquinoline alkaloids, were also enriched by treatment with SC-III (Supplementary Table 10). We validated the expression patterns of the above-mentioned genes by RT-qPCR analysis (Fig. 6b and Extended Data Fig. 4c).
Isoquinoline alkaloids are enriched in resistant cultivars and suppress CSR
To investigate the biochemical composition of the root microenvironments, we determined the identities of metabolites in RS and RE samples collected from resistant and susceptible cultivars treated with F. graminearum in 2021 by liquid chromatography–tandem mass spectrometry (LC-MS/MS). We detected an average of 1,558 metabolites across all samples (Supplementary Tables 11–12). PCoA based on Bray–Curtis distance matrices revealed significant differences in both RS and RE metabolites between resistant and susceptible samples (P = 0.008, PERMANOVA by Adonis, Fig. 6c). We identified 947 significantly different metabolites in RE samples and 124 in RS samples (|log2FC| > 1, P < 0.05), demonstrating the presence of distinct chemical microenvironments in these two niches.
Consistent with the transcriptomic data (Fig. 5h), berberine (a natural isoquinoline alkaloid) and its isoquinoline precursor were enriched in the RE niche rather than the RS niche. The biosynthetic precursor l-phenylalanine appeared to be depleted from resistant samples (Fig. 6d, Extended Data Fig. 4d). Importantly, l-dopa and tyramine, members of the berberine biosynthesis pathway that are substrates used by TYDC1 to synthesize dopamine, were enriched in RE samples (Extended Data Fig. 4d). The antagonistic activities of berberine against F. graminearum were demonstrated in vitro. In potato dextrose agar (PDA) plate assays, F. graminearum growth was reduced as the concentration of berberine increased, with growth diameters decreasing by 37%, 52%, 54% and 58% in the presence of 5, 25, 50 and 100 µg/mL berberine, respectively (Fig. 6e and Extended Data Fig. 4e). Finally, pre-treatment of seed coats with berberine significantly reduced CSR disease severity (Fig. 6f). Together, these results indicate that SC-III treatment triggers disease-suppressive activity in the RE niche by inducing the accumulation of antifungal metabolites.