Distinct rhizosphere microbiomes and metabolomes mediate Fusarium crown rot resistance across wheat cultivars Authors: Runlai Xu, Chenghang Du, Yutian Gao, Kunhu Chen, Jieru Guo, Jun Ma, Yinghua Zhang, Zhimin Wang, Zhencai Sun Published: 2026 ABSTRACT Soil-borne Fusarium crown rot (FCR) poses a major challenge to wheat production. Plants can recruit beneficial microbes through a “cry-for-help” strategy in response to pathogen attack, yet how different wheat genotypes modulate this response remains unclear. In FCR-affected wheat fields, we systematically investigated the interactions among wheat, pathogens, rhizosphere microbes, and metabolites using a combination of phenotypic assessment, molecular detection, and microbiome and metabolome profiling. Our results showed that wheat genotype and plant health jointly shaped rhizosphere microbial communities and metabolite profiles. Beneficial Bacillus species were key taxa in healthy rhizosphere communities, whereas the susceptible cultivar additionally relied on Pseudomonas to maintain plant health. In vitro assays and a pot experiment confirmed that these microbes antagonize the pathogen and enhance wheat performance, thereby alleviating FCR. Furthermore, the tolerant cultivar exhibited more complex rhizosphere microbial co-occurrence networks and harbored conserved taxa, such as Devosia. Microbe–metabolite interaction networks revealed strong associations between beneficial rhizosphere microbes (e.g., Bacillus and Pseudomonas) and specific metabolites, such as tritrans-heptacis-undecaprenyl phosphate, PE(14:1(9Z)/20:5(5Z,8Z,11Z,14Z,17Z)), and 1,2-Dioleoyl-sn-glycerol-3-phosphate. Collectively, our results provide novel evidence that rhizosphere metabolite-mediated plant–microbe interactions contribute to the suppression of soil-borne diseases. Introduction Fusarium crown rot (FCR) in wheat (Triticum aestivum L.) is a devastating soil-borne fungal disease that poses a serious threat to cereal production worldwide1,2. It is particularly prevalent in arid and semi-arid regions, such as the North China Plain (NCP), where yield losses can exceed 50%3. In the NCP, FCR is primarily caused by F. pseudograminearum (Fp)4, which infects roots and stem bases, causing browning and eventually resulting in shriveled or empty grains5. The disease has become an increasing concern in the winter wheat–summer maize rotation system, partly because maize residues returning to the field carry the mycelia or spores of Fusarium species, thereby contributing to the buildup of pathogen inoculum in soils under wheat cultivation6. To date, no complete resistance to FCR has been reported in wheat7,8. Although several functional genes involved in FCR resistance have been identified5,9,10, the resistance remains partial and depends on complex regulatory interactions influenced by genetic background and environmental conditions. In natural environments, plants live and evolve in close association with microorganisms, forming integrated holobionts11. Throughout their life cycle, plants interact with microbes across distinct compartments, among which the rhizosphere serves as the first line of defense against soil-borne diseases and a hotspot for plant–microbe communication12,13. Under pathogen stress, plants can actively recruit beneficial rhizosphere microorganisms, a process commonly termed the “cry for help” response14,15, leading to the formation of disease-suppressive soils that can persist as a microbial legacy and benefit subsequent plant generations16. Previous studies in the wheat–Fp system have documented the recruitment of beneficial rhizosphere microbes following FCR infection and highlighted their roles in improving host growth and disease defense17–19. Microbiome recruitment and assembly are influenced by both host type and plant compartment20,21, thereby linking plant phenotypes with microbial community dynamics. Plant genotype can shape the composition and function of its associated microbiome22–25. Resistant varieties, for example, often show more complex community structures and a capacity to recruit beneficial microbes from surrounding soils to combat soil-borne pathogens26,27. Integrating host genetic variation with associated microbiome characteristics provides deeper insights into the mechanisms underlying disease resistance than considering either factor alone. Nevertheless, the variation and functional roles of wheat-associated microbiomes across cultivars following Fp infection remain largely unknown. Root exudates play a crucial role in the “cry for help” strategy by mediating host-driven manipulation of the rhizosphere microbiome28,29. Upon pathogen attack, exudates change in composition and abundance, serving as both signaling molecules and nutrient sources to selectively shape and assembly rhizosphere microbial communities by recruiting beneficial microbes30–33. This process enhances our understanding of how plants exploit microbial resources under pathogen stress. In addition, root exudates and rhizosphere metabolites link plant genotypic variation with differences in the composition and structure of rhizosphere microbial communities34. This suggests that genotype-specific patterns of rhizosphere secretion can alter microbial community assembly, which may contribute to differences in disease resistance. Future research should focus on the interaction mechanisms among host genotype, root exudates or metabolites, and the associated microbiome. In the FCR-prevalent wheat cultivation system, we aimed to investigate how wheat genotypes and plant health shape rhizosphere microbiomes and metabolomes, and how these profiles and their potential interactions may contribute to disease resistance. We hypothesized that the structure and composition of the rhizosphere microbial community are shaped by genotype and health status; that key microbial taxa are recruited as beneficial members to maintain plant health, with the enrichment patterns differing among wheat genotypes; and that wheat orchestrates a crop–microbiome–metabolite interaction network in response to infection. To test these hypotheses, two wheat cultivars with contrasting FCR susceptibility were selected, and severely infected plants and apparently healthy plants were collected. Rhizosphere and root endosphere microbial communities were profiled using 16S amplicon sequencing. Potential beneficial bacteria were isolated, and their effects on plant growth and disease resistance were validated through in vitro and pot experiments. In addition, rhizosphere metabolite profiles were characterized to reveal microbe–metabolite interactions. Integrating these analyses provides insights into metabolite-mediated mechanisms by which wheat reshapes the rhizosphere microbiome to combat FCR. Results Field occurrence and impact of FCR on different wheat cultivars In the field experiment, disease scoring, yield assessment, and quantification of Fp content in stem bases were conducted to evaluate the impact of FCR on different winter wheat cultivars (Fig. 1A–C). Consistent patterns were observed across these measurements. Cultivar JM22 exhibited the highest disease index and Fp biomass in stem bases, and subsequently showed the lowest grain yield. Compared with the least diseased cultivar XM26, the yield of JM22 was reduced by 16.4% (Fig. 1B). These susceptible (JM22) and tolerant (XM26) cultivars, which exhibited the greatest contrast in disease severity, were therefore selected for further analyses. Diseased plants (scores 3–4) and healthy plants (score 0) were collected separately, resulting in four groups: JMD (diseased samples of JM22), JMH (healthy samples of JM22), XMD (diseased samples of XM26), and XMH (healthy samples of XM26) (Fig. 1D). Fp abundance in the rhizosphere soils was quantified to assess pathogen load in healthy and diseased samples (Supplementary Fig. 2). The results showed that Fp content in diseased samples was ~2.13-fold and 2.46-fold higher than that in healthy samples of JM22 and XM26, respectively. When genotype effects were considered alone, Fp biomass in the rhizosphere of JM22 was 51.1% higher than that of XM26. Plant defense-related genes encoding allene oxide synthase (AOS) and phenylalanine ammonia-lyase (PAL) were significantly expressed in stem bases of diseased plants (Fig. 1E). AOS and PAL are involved in the jasmonic acid (JA) and salicylic acid (SA) signaling pathways. PAL expression in JMD and XMD was 12.74-fold and 18.55-fold higher than in the respective healthy samples. AOS expression in JMD and XMD increased by 13.13-fold and 27.91-fold compared with the corresponding healthy samples. Among diseased samples, PAL expression was higher in JM22 than in XM26, whereas AOS expression showed no significant difference between the two cultivars.Fig. 1Disease investigation, yield assessment, and molecular detection in different winter wheat cultivars under field conditions.A Disease index of wheat Fusarium crown rot (FCR) at jointing and grain filling stages. B Grain yield at maturity. The diamond in the violin plots indicates the mean value. C Relative abundance of Fp (F. pseudograminearum) in the stem bases of four winter wheat cultivars. 04z36: 04 zhong 36, GY2018: Gaoyou 2018, JM22: Jimai 22, XM26: Xinmai 26. D The susceptible (JM22) and tolerant (XM26) wheat cultivars were divided into healthy (level 0 of disease score) and diseased (level 3–4 of disease score) samples, respectively. JMD: diseased plants of JM22, JMH: healthy plants of JM22, XMD: diseased plants of XM26, XMH: healthy plants of XM26. E The relative expression of genes encoding phenylalanine ammonia-lyase (PAL) and allene oxide synthase (AOS) in the healthy and diseased stem bases of JM22 and XM26. The data in the bars are presented as the mean ± standard error. Different lowercase letters indicate significant differences (P < 0.05) based on Duncan’s test. Disease investigation, yield assessment, and molecular detection in different winter wheat cultivars under field conditions. A Disease index of wheat Fusarium crown rot (FCR) at jointing and grain filling stages. B Grain yield at maturity. The diamond in the violin plots indicates the mean value. C Relative abundance of Fp (F. pseudograminearum) in the stem bases of four winter wheat cultivars. 04z36: 04 zhong 36, GY2018: Gaoyou 2018, JM22: Jimai 22, XM26: Xinmai 26. D The susceptible (JM22) and tolerant (XM26) wheat cultivars were divided into healthy (level 0 of disease score) and diseased (level 3–4 of disease score) samples, respectively. JMD: diseased plants of JM22, JMH: healthy plants of JM22, XMD: diseased plants of XM26, XMH: healthy plants of XM26. E The relative expression of genes encoding phenylalanine ammonia-lyase (PAL) and allene oxide synthase (AOS) in the healthy and diseased stem bases of JM22 and XM26. The data in the bars are presented as the mean ± standard error. Different lowercase letters indicate significant differences (P < 0.05) based on Duncan’s test. Wheat-associated bacterial diversity in the rhizosphere and root endosphere The bacterial microbiomes in the rhizosphere and root endosphere were distinctively assembled by wheat cultivars (JM22 and XM26) and plant health statuses (diseased and healthy samples) (Fig. 2A and Supplementary Figs. 3–5). Venn diagrams illustrated the shared and unique bacterial ASVs across wheat cultivars and health statuses (Supplementary Fig. 3). The rhizosphere microbiomes contained a higher proportion of core ASVs shared across the four groups (97.49% of total ASVs). In contrast, the root endosphere harbored a lower proportion of shared ASVs (76.03% of total ASVs), with more cultivar-specific microbes, including 25 ASVs unique to JM22 and 78 ASVs unique to XM26. The alpha diversity indices of bacterial communities in the rhizosphere differed significantly between the two cultivars, whereas no differences were observed in the root endosphere communities (Supplementary Fig. 4). The richness and Shannon diversity of the rhizosphere microbiomes were significantly higher in XM26 than in JM22. The bacterial beta diversity was assessed using Bray–Curtis distance-based principal co-ordinate analysis (PCoA) (Fig. 2A, B and Supplementary Table 2). In the rhizosphere, bacterial communities were strongly influenced by wheat cultivar (R2 = 0.28, P < 0.001), plant health status (R2 = 0.11, P < 0.015), and their interaction (R2 = 0.1, P < 0.013), whereas only the cultivar effect was significant in the root endosphere microbiomes (R2 = 0.09, P < 0.02). We compared the Bray–Curtis dissimilarity of microbial communities across cultivars and health statuses (Supplementary Fig. 5). In the rhizosphere, JMD exhibited higher community dissimilarity than JMH, whereas no such pattern was observed for XM26. Across cultivars, diseased JM22 plants showed higher rhizosphere dissimilarity than diseased XM26 plants. No differences were detected in the root endosphere microbiomes. Lower Bray–Curtis dissimilarity reflects a more stable community structure.Fig. 2Beta diversity and recruitment of bacterial communities in the rhizosphere and root endosphere.A Principal coordinates analysis (PCoA) of the rhizosphere communities. B PCoA of the root endosphere communities. PCoA was based on Bray–Curtis distance. C Relative abundance of different bacterial phyla in the rhizosphere. D Relative abundance of different bacterial phyla in the root endosphere. E Linear discriminant analysis (LDA) effect size (LEfSe) analysis showing bacterial biomarkers from phylum to genus levels among different cultivars and health statuses in the rhizosphere communities. F Bacterial biomarkers identified in the root endosphere communities. Bars represent the log-transformed LDA score >3 (P < 0.05). Beta diversity and recruitment of bacterial communities in the rhizosphere and root endosphere. A Principal coordinates analysis (PCoA) of the rhizosphere communities. B PCoA of the root endosphere communities. PCoA was based on Bray–Curtis distance. C Relative abundance of different bacterial phyla in the rhizosphere. D Relative abundance of different bacterial phyla in the root endosphere. E Linear discriminant analysis (LDA) effect size (LEfSe) analysis showing bacterial biomarkers from phylum to genus levels among different cultivars and health statuses in the rhizosphere communities. F Bacterial biomarkers identified in the root endosphere communities. Bars represent the log-transformed LDA score >3 (P < 0.05). Community composition and bacterial biomarkers across wheat cultivars and plant health statuses Analyses of community composition revealed that several bacterial phyla showed significantly different relative abundances across wheat cultivars and plant health statuses in the rhizosphere and root endosphere (Fig. 2C, D, and Supplementary Fig. 6). In the rhizosphere (Supplementary Fig. 6A), Acidobacteriota, Chloroflexota, Gemmatimonadota, and Planctomycetota showed a higher relative abundance in XMD than in JM22, whereas Pseudomonadota exhibited the opposite trend. Candidatus Saccharimonadota and Pseudomonadota were more abundant in JMH than in XMH. The relative abundance of Bacillota differed significantly across cultivars and plant health statuses. It was higher in healthy rhizosphere microbiomes than in diseased ones for both cultivars, and regardless of health status, its relative abundance in XM26 was higher than in JM22. The composition of root endosphere microbiomes was largely similar at phylum level, with the exception of Fibrobacterota, which showed higher relative abundance in XMH than in XMD (Supplementary Fig. 6B). A linear discriminant analysis (LDA) effect size (LEfSe) analysis focusing on lower taxonomic levels revealed bacterial biomarkers from phylum to genus (LDA score >3; Fig. 2E, F). The results showed that 13 and 36 taxonomic clades were detected in the rhizosphere bacterial communities of healthy and diseased plants, respectively. In JMH, the genera Pedobacter, Pseudomonas (and its subordinate order Pseudomonadales and family Pseudomonadaceae), and Ensifer were identified as biomarkers. For XMH, biomarkers included Bacillus (and its subordinate phylum Bacillota, class Bacilli, order Bacillales, and family Bacillaceae), an unassigned genus of Cytophagaceae, and Erythrobacteraceae along with its corresponding genus. Meanwhile, 22 and 13 biomarkers were detected in the root endosphere microbiomes of healthy and diseased samples, respectively. Bacteroidia, Bacteroidales, Prevotellaceae, Cellvibrionales, Cellvibrionaceae, Kitasatospora, Cellvibrio, Janthinobacterium, Saccharothrix, Fibrobacterota, Fibrobacteria, Fibrobacterales, Fibrobacteraceae, and possible_genus_04 exhibited greater relative abundances in XMH. Additionally, Gammaproteobacteria, Cytophagia, Cytophagales, Cytophagaceae, Streptosporangiaceae, Streptosporangiales, and unassigned genera of Cytophagaceae and Chitinophagaceae were enriched in JMH. Differential enrichment patterns and potential functions across wheat cultivars and plant health statuses Differential enrichment analyses identified bacterial ASVs that were enriched or depleted between different wheat cultivars, as well as between healthy and diseased status (Fig. 3A, Supplementary Fig. 7, and Supplementary Table 3). In the rhizosphere microbiome, 74 and 48 ASVs were enriched in the healthy samples of JM22 (JMH group) and XM26 (XMH group), respectively, whereas 70 and 116 ASVs were depleted. Among the enriched ASVs, eight were shared by both cultivars: ASV 113, ASV 83, ASV 811, ASV 64, ASV 597, ASV 2094, ASV 1410, and ASV 1335. With the exception of ASV 113, all of these belong to the genus Bacillus (Fig. 3B and Supplementary Table 3). Regarding specific cultivars, ASV 153 (Pseudomonas) and ASV 147 (Bacillus) were the most highly enriched in JMH, whereas ASV 64 (Bacillus) exhibited the highest enrichment in XMH (Fig. 3A). The relative abundances of these ASVs were significantly higher in healthy samples than in diseased samples (P < 0.001; Supplementary Fig. 8). Additionally, ASV 50 (Algoriphagus) and ASV 63 (Sphingobacterium) showed the strongest depletion in the healthy rhizosphere of XM26 (Supplementary Table 3). We further analyzed cultivar-specific ASVs that were enriched in both healthy and diseased samples of the same genotype, which indicates a conserved enrichment pattern. In the rhizosphere, JM22 harbored 72 cultivar-specific ASVs, whereas XM26 harbored 63 ASVs (Fig. 3B). Among these, ASV 11 (Devosia) and ASV 12 (Parapedobacter) showed high enrichment in XM26 (Fig. 3A). Furthermore, ASV 52 and ASV 122, which belong to the genus Devosia, were identified as XM26-specific ASVs enriched in the root endosphere microbiome (Supplementary Fig. 7). The genus-level classification of ASVs enriched in the healthy rhizosphere was further summarized (Fig. 3C). In the rhizosphere of JMH, Bacillus, unassigned taxa, and Pseudomonas accounted for 57.12%, 21.84%, and 6.78% of the relative abundance of enriched ASVs, respectively. In contrast, no enriched ASVs affiliated with Bacillus or Pseudomonas were detected in the rhizosphere of diseased JM22 plants (JMD), and more than 50% of the enriched abundance was classified as “Others”. For XM26, Bacillus represented the largest proportion of enriched ASV abundance in the rhizosphere of XMH (33.15%). In the rhizosphere of diseased XM26 plants (XMD), however, the contribution of Bacillus decreased to only 0.15%, while the majority of the enriched abundance (75.03%) was attributed to “Others”. Notably, when examining the distribution of enrichment across the healthy rhizosphere of both cultivars, JMH and XMH contributed 72.56% and 27.02% of the total abundance of enriched Bacillus ASVs, respectively, whereas the enrichment of Pseudomonas was almost exclusively associated with JMH (92.93%).Fig. 3Differential enrichment analysis of bacterial ASVs and validation assays for beneficial Bacillus and Pseudomonas.A Volcano plots displaying the differential enrichment of bacterial ASVs in the rhizosphere across cultivars and health statuses. Colored points represent ASVs with |log2(fold change)| > 1 and an FDR-adjusted P < 0.05. B Upset plots showing the differentially enriched ASVs compared between wheat cultivars (top) and between health statuses (bottom). Single blue dot indicates the unique enrichments. The line between dots indicates the shared enrichments. Enriched size: total number of enriched ASVs. Intersection size: the number of shared or unique enrichments. C Sankey plot showing genus-level classification (top) of enriched ASVs in JMD vs JMH and XMD vs XMH comparisons (bottom). The flow indicates the relative abundance of the differentially enriched ASVs. “Others” is defined as genera represented by fewer than 10 enriched ASVs. D Dual-culture assays of four bacterial isolates against pathogen Fp on potato dextrose agar. The red line represents 1 cm. E Pot experiment: FCR disease index, shoot height, shoot fresh weight, and root fresh weight of wheat seedlings. CK: no inoculation with isolates; RL83, RN92, RN186 and RN91: inoculation with the isolates; NFP: no inoculation with Fp; FP: inoculated with Fp. Different lowercase letters indicate significant differences (P < 0.05) based on Duncan’s test. Differential enrichment analysis of bacterial ASVs and validation assays for beneficial Bacillus and Pseudomonas. A Volcano plots displaying the differential enrichment of bacterial ASVs in the rhizosphere across cultivars and health statuses. Colored points represent ASVs with |log2(fold change)| > 1 and an FDR-adjusted P < 0.05. B Upset plots showing the differentially enriched ASVs compared between wheat cultivars (top) and between health statuses (bottom). Single blue dot indicates the unique enrichments. The line between dots indicates the shared enrichments. Enriched size: total number of enriched ASVs. Intersection size: the number of shared or unique enrichments. C Sankey plot showing genus-level classification (top) of enriched ASVs in JMD vs JMH and XMD vs XMH comparisons (bottom). The flow indicates the relative abundance of the differentially enriched ASVs. “Others” is defined as genera represented by fewer than 10 enriched ASVs. D Dual-culture assays of four bacterial isolates against pathogen Fp on potato dextrose agar. The red line represents 1 cm. E Pot experiment: FCR disease index, shoot height, shoot fresh weight, and root fresh weight of wheat seedlings. CK: no inoculation with isolates; RL83, RN92, RN186 and RN91: inoculation with the isolates; NFP: no inoculation with Fp; FP: inoculated with Fp. Different lowercase letters indicate significant differences (P < 0.05) based on Duncan’s test. Additionally, we performed functional prediction of the rhizosphere microbiomes, and the top 20 functional groups were included in the analysis. (Fig. 4D). Differences in microbial functions were observed across cultivars and health statuses. In the rhizosphere communities of XM26, functions related to global and overview maps, signal transduction, and energy metabolism were upregulated, whereas functions associated with amino acid metabolism and metabolism of cofactors and vitamins showed higher relative abundance in JM22. Regarding health status, lipid metabolism and metabolism of terpenoids and polyketides was more abundant in the rhizosphere of healthy plants than in diseased samples for both cultivars, while carbohydrate metabolism and xenobiotics biodegradation and metabolism were specifically upregulated in the healthy rhizosphere of XM26. In the rhizosphere of diseased plants, biosynthesis of other secondary metabolites was upregulated in both cultivars. Additionally, the prokaryotes cellular community was specifically enriched in JM22, whereas the antimicrobial function were upregulated in XM26.Fig. 4Co-occurrence network analysis of two wheat cultivars’ microbiomes.A Bacterial co-occurrence network patterns in the rhizosphere. B Bacterial co-occurrence network patterns in the root endosphere. Nodes are colored according to bacterial phyla, and node size represents degree. Red and green edges indicate positive and negative correlations, respectively, with edge thickness reflecting the strength of the correlation. C Zi (within-module connectivity)–Pi (among-module connectivity) plots visualizing keystone hubs in the rhizosphere networks. D keystone hubs identified in the root endosphere networks. Co-occurrence network analysis of two wheat cultivars’ microbiomes. A Bacterial co-occurrence network patterns in the rhizosphere. B Bacterial co-occurrence network patterns in the root endosphere. Nodes are colored according to bacterial phyla, and node size represents degree. Red and green edges indicate positive and negative correlations, respectively, with edge thickness reflecting the strength of the correlation. C Zi (within-module connectivity)–Pi (among-module connectivity) plots visualizing keystone hubs in the rhizosphere networks. D keystone hubs identified in the root endosphere networks. Beneficial microbes alleviate FCR stress and improve plant performance We used various media to isolate potentially beneficial microbes, such as Bacillus and Pseudomonas, from the rhizosphere soil of healthy wheat plants. A total of 367 bacterial strains were obtained: 237 from JM22 and 130 from XM26. Among these, five strains were identified as Bacillus sp. and one as Pseudomonas sp. via 16S rRNA gene sequencing (Supplementary Table 4). After removing duplicates with 100% sequence identity, four distinct strains were retained: RL83 (Bacillus sp.), RN92 (Bacillus sp.), RN186 (Bacillus sp.), and RN91 (Pseudomonas sp.). Strain RN186 displayed 100% nucleotide similarity to ASV 64 (Bacillus), whereas RL83 and RN92 showed 96.29% and 96.98% sequence identity to ASV 811 (Bacillus; shared enrichment) and ASV 2872 (Bacillus; enriched in JMH), respectively. In addition, RN91 exhibited 98% sequence similarity to ASV153 (Pseudomonas) (Supplementary Fig. 10). To validate the in vitro antagonistic abilities of the isolated strains against the pathogen Fp, dual-culture assay was performed on Potato Dextrose Agar (PDA). Strains RL83, RN92, and RN186 significantly inhibited Fp mycelial growth (Fig. 3D), exhibiting inhibition rates of 72.59% (P = 0.00236), 73.70% (P < 0.001), and 75.56% (P < 0.001), respectively, compared to the negative control (Supplementary Fig. 11A, B). All isolated strains grew rapidly in the spread-plate assay and inhibited the extensive spread of Fp mycelia (Supplementary Fig. 11C). Meanwhile, the cell-free supernatant of the 1-day bacterial fermentation broth failed to inhibit fungal growth (Supplementary Fig. 11D). We conducted a pot experiment with wheat seedlings to evaluate the ability of beneficial bacteria to alleviate disease stress and improve plant performance (Fig. 3E). Inoculation with isolates RN186, RL83, RN92, and RN91 reduced the disease index by 52.55%, 56.20%, 40.88%, and 51.82%, respectively (P < 0.05). All beneficial bacteria alleviated the inhibitory effects of Fp on shoot height (P < 0.05) and shoot fresh weight (P < 0.05), while only RN91 mitigated the reduction in root fresh weight, providing an alleviation of 80.74% (P < 0.05). In the absence of Fp, RL83 increased aboveground fresh biomass by 21.84% (P < 0.05), whereas RN91 increased root fresh weight by 37.14% (P < 0.05). Furthermore, the root system architecture (RSA) of wheat inoculated with or without the beneficial bacteria and/or Fp was analyzed (Supplementary Fig. 12). Total root length (TL), total root surface area (SA), and total root volume (RV) were negatively affected by Fp inoculation, showing reductions of 53.24%, 53.09%, and 56.46%, respectively (P < 0.05), whereas root average diameter (AD) was not significantly affected. Under pathogen-free conditions, inoculation with the beneficial bacteria enhanced RSA traits to varying degrees. RN91 exhibited the greatest effect, increasing TL, SA, AD, and RV by 41.65%, 45.75%, 8.40%, and 51.98%, respectively (P < 0.05). Additionally, bacterial inoculation alleviated Fp-induced stress on RSA traits; the Pseudomonas strain RN91 showed particularly strong effects, producing more than a 1.5-fold increase in TL, SA, and RV relative to Fp inoculation alone (P < 0.05). The three Bacillus strains also enhanced such traits, with average increases of 72%, 61%, and 90%, respectively (P < 0.05). Bacterial co-occurrence networks of two wheat cultivars The microbial co-occurrence networks were constructed for two wheat cultivars in the rhizosphere (Fig. 4A, C and Supplementary Fig. 13A) and root (Fig. 4B, D and Supplementary Fig. 13B). The networks of different wheat cultivars and compartments exhibited distinct topological characteristics (Supplementary Table 5). Regarding the number of nodes, the network of XM26 contained more nodes than that of JM22 in both the rhizosphere (+13.47%) and root endosphere (+15.08%). Larger cultivar-dependent differences in edge number and average degree were observed in rhizosphere networks, with XM26 exceeding JM22 by 51.41% and 33.43%, respectively, while differences in root endosphere networks were minor. Comparisons between compartments revealed substantially more edges (+351.08%), higher average degree (+200.87%), and lower modularity (−53.01%) in the rhizosphere networks compared with the root endosphere networks. Furthermore, the network nodes were classified by bacterial phylum (Fig. 4A, B) and module (Supplementary Fig. 13) to analyze their distribution patterns. The dominant phyla (with higher node degree) in the wheat rhizosphere networks were Pseudomonadota, Bacteroidota, Actinomycetota, and Bacillota, which occupied 88.40% of all nodes in the average of two cultivars. While in the root endosphere network, Pseudomonadota occupied nearly 50% of nodes. As for modularity (Supplementary Fig. 13), JM22 and XM26 rhizosphere networks were mainly clustered by modules 2, 3, and 4, whereas the root endosphere networks showed a more dispersed modular structure with a greater number of modules, corresponding to higher modularity values (Supplementary Table 5). Node topologies were identified in the networks based on within-module connectivity (Zi) and among-module connectivity (Pi) values (Fig. 4C, D). Peripherals dominated both the rhizosphere and root endosphere networks, accounting for an average of 93.80% of all nodes. The two cultivars showed similar numbers of connectors in the rhizosphere networks, whereas a higher number of connectors was observed in the root endosphere networks. Module hubs were identified exclusively in the bacterial networks of XM26. Shifts in rhizosphere metabolic profiles among wheat cultivars and plant health statuses We further analyzed the rhizosphere metabolite profiles across cultivars and plant health statuses (Fig. 5). Rhizosphere metabolite composition was distinct between genotypes (R2 = 0.39, P < 0.001) and between health status (R2 = 0.12, P < 0.01), and their interaction also showed an effect (R2 = 0.19, P < 0.001) (Fig. 5A). A total of 1996 compounds were annotated in the rhizosphere metabolome, among which organic acids, lipids, and alcohols and amines were the most abundant classes, together accounting for 37.87% of the total relative abundance (Fig. 5B). Unique and shared differentially abundant metabolites (DAMs) were identified across cultivars and plant health statuses (Fig. 5C–F and Supplementary Table 5). Compared with diseased samples, 40 and 16 DAMs were uniquely detected in JMH and XMH, respectively, among which only tritrans, heptacis-undecaprenyl phosphate (fatty acyls) was upregulated in both cultivars. In XMH, amino acids and glycerophospholipids (GP) were the dominant DAM classes, collectively accounting for 61.53% of the total abundance (Supplementary Fig. 14A). PE(14:1(9Z)/20:5(5Z,8Z,11Z,14Z,17Z)) was the most abundant DAMs in XMH (+23.26 fold change; Fig. 5E). In JMH, alcohol and amines constituted the major DAM class, accounting for 22.76% of the total abundance (Supplementary Fig. 14A). Glyceryl tripalmitoleate (+183 fold change), 5-methoxytryptophan (+22.07 fold change), and pantethine (+19.76 fold change) were the most abundant healthy-associated DAMs in JM22 (Fig. 5E). Regarding cultivar-specific metabolites, 11 and 61 DAMs were found in XM26 and JM22, respectively (Fig. 5D, F). Phenolic acids, and benzene and substituted derivatives were the dominant DAM classes in the rhizosphere of JM22, being consistently upregulated in both healthy and diseased samples compared with XM26, and together accounting for 67.93% of the total abundance (Supplementary Fig. 14B). In contrast, alkaloids and nucleotides and derivatives were the predominant conserved DAM classes in XM26, jointly representing 89.69% of the total abundance (Supplementary Fig. 14B). Among XM26-specific DAMs (Fig. 5F), MW0135243 (organic acids), 10-hydroxycamptothecin (alkaloids), and glutathione reducedform (amino acids and derivatives) were the top three upregulated metabolites. In contrast, MW0129760 (benzene and substituted derivatives), MW0130688 (benzene and substituted derivatives), along with Ononin (flavonoids), were the top three cultivar-specific DAMs in JM22 (Fig. 5F).Fig. 5Rhizosphere metabolome profiling and differentially abundant metabolites (DAMs).A Principal component analysis (PCA) of rhizosphere metabolites. PERMANOVA results show the effects wheat cultivar (C), plant health status (S), and their interaction (C × S). Asterisks indicate significant differences: *P < 0.05, **P < 0.01, and ***P < 0.001. B Treemap showing the classification of rhizosphere metabolites. The size of tiles indicates the average relative abundance of metabolites. FA fatty acyls, GL glycerolipids, GP glycerophospholipids, SL saccharolipids. C Venn diagram showing shared and unique DAMs between healthy and diseased samples. D Shared and unique DAMs between the two cultivars. E Heatmap showing the regulatory patterns of DAMs between healthy and diseased samples. F Regulatory patterns of DAMs between the two cultivars. Asterisks indicate significantly differential compounds with VIP > 1, |log2 fold change| > 1, and P < 0.05. Detailed information on DAMs is listed in Supplementary Table 6. Rhizosphere metabolome profiling and differentially abundant metabolites (DAMs). A Principal component analysis (PCA) of rhizosphere metabolites. PERMANOVA results show the effects wheat cultivar (C), plant health status (S), and their interaction (C × S). Asterisks indicate significant differences: *P < 0.05, **P < 0.01, and ***P < 0.001. B Treemap showing the classification of rhizosphere metabolites. The size of tiles indicates the average relative abundance of metabolites. FA fatty acyls, GL glycerolipids, GP glycerophospholipids, SL saccharolipids. C Venn diagram showing shared and unique DAMs between healthy and diseased samples. D Shared and unique DAMs between the two cultivars. E Heatmap showing the regulatory patterns of DAMs between healthy and diseased samples. F Regulatory patterns of DAMs between the two cultivars. Asterisks indicate significantly differential compounds with VIP > 1, |log2 fold change| > 1, and P < 0.05. Detailed information on DAMs is listed in Supplementary Table 6. Microbe-metabolite interaction networks in the rhizosphere The associations between beneficial microbes and metabolites in the rhizosphere were explored using Spearman correlation–based interaction networks (Fig. 6 and Supplementary Fig. 15). Bacterial ASVs and DAMs enriched in JMH and XMH (Fig. 6A and Supplementary Fig. 15A–C), as well as taxa and compounds specifically enriched in the tolerant cultivar XM26 (Fig. 6B and Supplementary Fig. 15D), were selected for network construction. The results showed that tritrans, heptacis-undecaprenyl phosphate, the only DAMs shared in the healthy rhizospheres of both cultivars, was positively associated with all Bacillus ASVs (average Spearman r = 0.72). In XMH, PE(14:1(9Z)/20:5(5Z,8Z,11Z,14Z,17Z)) also showed positive correlations with Bacillus ASVs (average Spearman r = 0.66). In JMH, glyceryl tripalmitoleate was associated with ASV 17 (Bacillus), whereas 5-methoxytryptophan was linked to multiple Bacillus ASVs (ASV 1410, ASV 811, ASV 1335, and ASV 147). In addition, pantethine correlated with ASV 1410 (Bacillus) and ASV 147 (Bacillus), while ASV 153 showed a positive correlation with 1,2-Dioleoyl-sn-glycerol-3-phosphate (Spearman r = 0.57). For tolerant cultivar-specific interactions, reduced form glutathione exhibited the strongest correlations with ASV11 (Devosia) and ASV12 (Parapedobacter) (Spearman r = 0.79 and 0.82, respectively), whereas MW0135243 and 10-hydroxycamptothecin showed relatively weaker associations with these ASVs (average Spearman r = 0.66).Fig. 6Interaction networks of beneficial microbes and metabolites in the rhizosphere.A Microbe–metabolite interaction network in the rhizosphere of healthy plants of two wheat cultivars. B Microbe–metabolite interaction network in the rhizosphere of the tolerant cultivar. Interaction networks show correlations between differentially enriched ASVs and DAMs in the rhizosphere. Edges indicate the correlations (Spearman r). Full name of MW0135243 is listed in Supplementary Table 6. Interaction networks of beneficial microbes and metabolites in the rhizosphere. A Microbe–metabolite interaction network in the rhizosphere of healthy plants of two wheat cultivars. B Microbe–metabolite interaction network in the rhizosphere of the tolerant cultivar. Interaction networks show correlations between differentially enriched ASVs and DAMs in the rhizosphere. Edges indicate the correlations (Spearman r). Full name of MW0135243 is listed in Supplementary Table 6. Discussion Our study conducted disease surveys, yield assessments, and quantification of pathogen load and disease-resistance-related genes across different wheat cultivars grown in the field, revealing variation in defense responses and resource allocation under FCR stress. However, interactions between crops and their associated microbiomes are often overlooked. Upon pathogen invasion, plants activate salicylic acid (SA) and jasmonic acid (JA) signaling pathways. These pathways have also been reported to be triggered by beneficial microbes recruited by the plant17 or artificially inoculated35, enhancing host disease resistance. Nevertheless, in Fp-infected plants, disease symptoms appear even when genes involved in SA and JA pathways are upregulated. In healthy individuals, Fp content in the rhizosphere soils is relatively low, suggesting a potential role of the associated rhizosphere microbiome. Soil pathogen abundance is considered closely linked to disease severity and yield loss36. The variation in disease susceptibility among cultivars is likely attributable to differences in microbial community assembly and the recruitment of beneficial microbes24. Therefore, characterizing wheat-associated rhizosphere microbiomes is crucial for a true understanding of plant responses to pathogen invasion under complex environmental conditions. Fp-tolerant cultivar exhibited higher alpha diversity in the rhizosphere microbiomes. Microbial community diversity is crucial for soil and plant resilience37,38, and higher diversity has been associated with reduced disease severity39, possibly due to impeded pathogen invasion40. In this study, the structure of the rhizosphere microbiome was influenced by both cultivar and plant health status, whereas the root endosphere microbial community was shaped solely by cultivar. Genotype-driven microbiome specificity is well established34,41. The effect of plant health on the microbiome appears to be compartment-specific, likely reflecting the fact that Fp primarily infects the wheat coleoptile and leaf sheath42, which may influence rhizosphere microbiome assembly while exerting limited impact on root endosphere communities. Certain microbial taxa showed higher relative abundance in the rhizosphere and may play roles in disease suppression and plant growth promotion, such as the phylum Bacillota43. This phylum includes well-known beneficial microbes, such as Bacillus, which can suppress pathogens either through direct antagonism or by activating host systemic resistance27. In our study, Bacillus was widely recruited in the rhizosphere of healthy plants, and the isolated Bacillus strains demonstrated antifungal effects and alleviate FCR stress, as confirmed by both in vitro assays and the pot experiment. We also identified Pseudomonas as another potentially beneficial microbe, which has been widely applied in biocontrol44 and was found to mitigated FCR while promoting wheat aboveground and root growth. A previous study further proposed that non-pathogenic derivatives of pathogenic Pseudomonas could trigger a “cry for help” response to recruit other beneficial microbes without causing disease45. Earlier reports suggested higher Pseudomonas abundance in resistant cultivars46, which contrasts with our findings. This discrepancy may be attributable to our simultaneous consideration of both host genotype and health status, as well as the influence of complex environmental factors47,48. By analyzing the differential enrichment of microbiomes across different cultivars and plant health statuses, we suggested that the tolerant cultivar relies on Bacillus to maintain plant health, whereas the susceptible cultivar depends on both Bacillus and Pseudomonas. Studies with synthetic communities (SynComs) have demonstrated that the combined presence of Bacillus and Pseudomonas enhances disease suppression49. Therefore, cultivars with weaker resistance may rely more on the effects of SynComs, which should be addressed in future research. However, the antagonistic or synergistic interactions among microbial members, as well as their actual performance within the soil-crop system, still need to be considered. In addition, we identified cultivar-specific microbial taxa of wheat. Complete disease resistance is considered to result from the combined effects of host genotype and associated microbiomes50. However, the relationship between FCR resistance in wheat cultivars and their associated microbiomes remains unclear. Devosia, identified as a tolerant cultivar-specific bacterial taxon in our study, has been reported to degrade the mycotoxin deoxynivalenol (DON) and promote plant growth51,52. Moreover, we observed that Devosia exhibited compartmental adaptability, as it was also detected in the root endosphere microbial community. Given that no Devosia strains were recovered in this study, selective culture media could be used in future research to isolate these bacteria. Co-occurrence networks reflect the interactions among microbial members53. The microbial network differed across cultivars and compartments. Plant genotype is considered to play an important role in shaping microbial co-occurrence networks54, and complex networks can reinforce the host’s resistance to diseases46,55. Our findings supported this notion, as the tolerant cultivar harbored more connected and robust microbial networks. Network complexity and stability decreased from the rhizosphere to the root endosphere, reflecting host-mediated filtering56, indicating that the rhizosphere permits more extensive interactions influenced by host genotype. Key network nodes play an important role in maintaining network stability57. Our study identified connector nodes affiliated with Bacillus in the rhizosphere networks of the susceptible cultivar. Connectors link multiple modules within a network to sustain connectivity and robustness, and the loss or reduction of such nodes may disrupt network integrity, potentially leading to a shift from a healthy to a diseased state. Module hubs were detected only in the bacterial networks of the tolerant cultivar’s microbiome. These key nodes are highly connected to other nodes within their respective modules58, implying greater internal module stability in the tolerant cultivar’s network, which may in turn support better plant performance. The rhizosphere is a hotspot for the exchange of substances and energy between plants and microbes. Plants regulate microbial community diversity and function through rhizosphere metabolites or root exudates, while microbes in turn enhance host resource utilization59,60. Root exudates are produced as part of a “cry for help” strategy triggered when plants are invaded by pathogens 28,33, and are influenced by host genotype34,61. Our analysis of the rhizosphere metabolome indicated that the composition of metabolites was jointly shaped by cultivar and plant health status, consistent with the patterns observed in the rhizosphere microbiome. This suggests that different wheat cultivars exhibit distinct rhizosphere metabolic strategies and microbiome assembly in response to Fp stress. The upregulation patterns of specific rhizosphere metabolites may facilitate the recruitment of beneficial microbes, thereby supporting plant growth and disease suppression. For example, root exudates such as fatty acids and amino acids secreted under disease stress can recruit beneficial Pseudomonas, protecting both the host and future generations30. Specific host regulators, such as RIN, can reprogram root exudate composition to selectively recruit beneficial rhizosphere microbes for disease resistance32. However, how FCR and cultivar affect rhizosphere metabolite composition, and the interactions between metabolites and microbiomes, remain unclear. Tritrans heptacis-undecaprenyl phosphate was commonly enriched in the rhizosphere of healthy plants from both cultivars. Its core structure, undecaprenyl phosphate, is implicated in bacterial cell wall and peptidoglycan biosynthesis62, potentially linking it to the co-enriched Bacillus in healthy rhizosphere. We also revealed cultivar-specific metabolites regulatory patterns. In the healthy rhizosphere of tolerant cultivar, PE(14:1(9Z)/20:5(5Z,8Z,11Z,14Z,17Z)) showed specifically high abundance, and its class phosphatidylethanolamine often serves as an essential component of plant and microbial membranes, participating in plant growth, membrane homeostasis, and signaling62–64. Moreover, we found that glyceryl tripalmitoleate, 5-methoxytryptophan, and pantethine were most significantly enriched in the healthy rhizosphere of the susceptible cultivar, with their precursors or derivatives involved in plant and microbial growth and stress tolerance65–67. Notably, melatonin synthesized from 5-methoxytryptophan is reported to reprogram soil microbial communities, creating a healthy environment for the host68. 1,2-Dioleoyl-sn-glycerol-3-phosphate has not been specifically studied, but its class, phosphatidic acid, may be essential for hormone signaling-mediated plant–microbe interactions69, potentially linking it to cultivar-specific enrichment of Pseudomonas. Conserved rhizosphere metabolites in wheat cultivars are also of interest. The reduced form of glutathione maintains cellular oxidative homeostasis and modulates root exudate composition and stress-related signaling pathways, indirectly promoting the colonization or activity of disease-suppressive microbes70. Future research is essential to elucidate the mechanisms underlying crop–metabolite–microbe interactions. Taken together, our findings emphasize the importance of integrating host–microbe–metabolite interactions to enhance wheat resilience, providing new perspectives for controlling soil-borne diseases and promoting sustainable agriculture. Methods Site description and design of field experiment The field experiment was conducted at the Wuqiao Experimental Station of China Agricultural University (37°41′N, 116°37′E) during the 2021–2022 wheat-growing season. Total precipitation and mean temperature during this period were 126 mm and 9.5 °C, respectively (Supplementary Fig. 1). The main physicochemical properties of the 0–20 cm soil layer were as follows: light loam texture, organic matter content of 19.14 g kg⁻¹, total nitrogen of 1.01 g kg⁻¹, available phosphorus of 39.92 mg kg⁻¹, available potassium of 113.81 mg kg⁻¹, and a pH of 8.27. The site has a long-term winter wheat–summer maize rotation, with a prevalence of severe Fusarium crown rot (FCR) caused by F. pseudograminearum (Fp). The field was artificially inoculated with Fp in 2020 using colonized grains as inoculum71, and tilled before sowing in 2021 to ensure uniform pathogen distribution. The field trial included four winter wheat cultivars: Jimai 22 (JM22), Xinmai 26 (XM26), Gaoyou 2018 (GY2018), and 04 Zhong 36 (04Z36). Each cultivar was planted in four replicate plots, each measuring 18 m² (1.5 × 12 m) and separated by a 1 m buffer zone. In total, 16 plots (4 cultivars × 4 replicates) were arranged in a randomized complete block design and sown on 24 October. The sowing density was 625 plants m⁻², adjusted among cultivars according to thousand-kernel weight. Irrigation was applied only prior to sowing to maintain soil moisture at ~85% of field capacity. All fertilizers were applied as basal inputs, including 180 kg ha⁻¹ N, 138 kg ha⁻¹ P₂O₅, and 113 kg ha⁻¹ K₂O. All other agronomic practices followed local conventional management. Disease investigation and yield measurement Disease severity was investigated at the jointing (Feekes 6) and grain filling stage (Feekes 11.1)72. A total of 150 plants were surveyed in each replicate plot, obtained through zigzag sampling. Scoring of FCR were performed by assessing the degree of browning at the stem base (from the crown to the first node), using a scale ranging from 0 (no visible symptoms) to 5 (area of browning ≥75%)73. The disease index (DI) was calculated as follows: DI = (∑nX/5 N) × 100 (X: the disease scale of the plant; n: the number of plants of each scale; N: the total number of plants). Wheat grains were harvested at the maturity stage (Feekes 11.4), with each plot comprising three 1.2 m² replicates. Grain yield was converted based on a grain moisture content of 13%. Plant and soil sample collection Soil and plant samples were collected at the grain filling stage following a previous study74. Briefly, the whole wheat plant was dug up from the field by the roots using a fire-heated and cooled shovel. Following a rapid assessment of disease severity, the 150 plants in each replicate plot were divided into two sections. Plants in the first section were used to determine the relative Fp content in stem bases across the four cultivars. Plants in the second section were further classified as healthy (level 0 on the FCR scale, with no visible symptoms and normal growth) or diseased (levels 3–4 on the FCR scale, exhibiting severe infection but no plant death). Stem bases were cut using sterile scissors and immediately stored in ultra-low temperature freezers at −80 °C. Roots were vigorously shaken to remove bulk soil, and soil tightly adhering to the root surface was washed off in a 50 mL centrifuge tube containing 25 mL 1× phosphate-buffered saline (PBS; Solarbio, Beijing, China) at 200 rpm for 10 min. Rhizosphere soil was collected by centrifugation at 4000 × g for 15 min to obtain a pellet. Roots were then transferred to a new tube, surface-sterilized with 75% ethanol, and rinsed three times with sterile ddH2O. To verify the effectiveness of surface disinfection, clean roots were placed on potato dextrose agar (PDA) medium and incubated at 37 °C for 2d; the absence of colony formation confirmed successful sterilization. Both rhizosphere soil and root samples were divided into two subsamples: one was stored at −80 °C for DNA extraction, and the other was temporarily stored at 4 °C for the isolation of beneficial bacteria. Based on field disease scoring, the two cultivars exhibiting the most severe (JM22) and least severe (XM26) infection were selected for further analyses (Fig. 1D). Accordingly, four groups were defined: (1) JMD: diseased samples of JM22, (2) JMH: healthy samples of JM22, (3) XMD: diseased samples of XM26, and (4) XMH: healthy samples of XM26. DNA extraction from rhizosphere, root, and stem base samples Pre-frozen samples of wheat roots and stem bases were ground into a powder in liquid nitrogen. Total DNA was extracted from 0.5 g of rhizosphere soil and 1 g of plant (root and stem base) samples using a PowerSoil™ DNA Isolation Kit (Mo Bio Laboratories, Carlsbad, CA, USA). Then the DNA extract was checked by 1% (w/v) agarose gel electrophoresis (AGE), and quantified concentration and purity using a NanoDrop 2000 spectrophotometer (Thermo Fisher Scientific, Wilmington, DE, USA). Template DNA was temporarily stored at −20 °C for further analysis. Quantification of Fp in stem bases and rhizosphere soils The relative content of Fp in stem bases was calculated as the proportion of Fp DNA (Tri5 gene) to wheat DNA (wheat actin-binding protein (WABP) gene) using real-time quantitative PCR (RT-qPCR) (Supplementary Table 1). In brief, quantification of fungal and plant genes was performed on a CFX96 Real-Time PCR Detection System (Bio-Rad) using Hieff qPCR SYBR Green Master Mix (YEASEN, Shanghai, China). Each biological replicate includes three technical replicates. Raw data were analyzed using the Bio-Rad CFX Maestro software (v4.0.2325.0418). Fp biomass was calculated as follows: Relative content = Effungal-Ct/Efwheat-Ct (Ef: PCR amplification efficiency; Ct: Cycle threshold). Fp abundance in the rhizosphere soil was estimated as copies calculated from the Ct values using a standard curve constructed from serially diluted Tri5 amplicons. Detection of defense-related genes Total RNA from stem bases was extracted using a RNApure Total RNA Kit (Aidlab, Beijing, China) following the manufacturer’s instructions. The concentration and quality of total RNA were checked with 1% (w/v) AGE and a NanoDrop 2000 spectrophotometer (Thermo Fisher Scientific, Wilmington, DE, USA). Afterwards, cDNA was synthesized using TRUEscript RT Kit (Aidlab, Beijing, China) according to the manufacturer’s protocols. The expressions of the defense-related genes allene oxide synthase (AOS) and phenylalanine ammonia-lyase (PAL) were quantified using a CFX96 Real-Time PCR Detection System (Bio-Rad) with Hieff qPCR SYBR Green Master Mix (YEASEN, Shanghai, China). AOS and PAL are key genes involved in the jasmonic acid (JA)75 and salicylic acid (SA)76 signaling pathways, respectively. The WABP gene was used as the reference gene (Supplementary Table 1). RT-qPCR was performed in a 20 μL reaction mixture containing 10 μL of SYBR Mix, 0.4 μL of each of the forward and reverse primers, 2 μL of cDNA (eluted to 50 ng/μL), and sterile deionized water was used to bring the total volume to 20 μL. The thermocycler settings of RT-qPCR were as follows: 95 °C for 5 min, followed by 40 cycles of 10 s at 95 °C, and 30 s at 60 °C. Finally, the melt curve was generated according to the default setting. Each amplification was performed in triplicate. The relative expression of these defense-related genes was calculated using the 2–ΔΔCT method. 16S rRNA amplicon sequencing and bioinformatic analysis A total of 32 samples [groups: 2 cultivars (JM22 and XM26) × 2 health statuses (healthy and diseased) × 2 compartments (rhizosphere and root endosphere) × 4 replicates] were processed for amplicon sequencing. The V3–V4 region of the 16S rRNA gene in soil samples was amplified using primer pair 338F/806R, while the V5–V7 region was targeted in wheat root samples using primer pairs 799F/1193R to exclude the effect of host sequences (Supplementary Table 1). PCR reactions were performed in a 50 μL reaction mixture containing 25 μL of 2× Premix Taq, 1 μL of forward/ reverse primers, 50 ng of template DNA, and sterile ddH2O. Amplification reactions were carried out in triplicate as following thermal cycling conditions: 95 °C for 5 min, followed by 30 cycles of 30 s at 94 °C, 30 s at 52 °C, 30 s at 72 °C, and 10 min at 72 °C for final extension. PCR products were detected by 1.5% AGE and mixed according to concentration. Products were extracted and purified using E.Z.N.A.® Gel Extraction Kit (Omega, USA). Amplicon libraries were constructed using NEBNext® Ultra™ II DNA Library Prep Kit for Illumina® (New England Biolabs, USA) following the manufacturer’s protocols. Then Purified amplicons were performed on the Illumina Nova 6000 platform (Guangdong Magigene Biotechnology Co., Ltd., Guangzhou, China). The 16S rRNA gene sequencing data was deposited in the National Genomics Data Center (NGDC)77,78 under accession number CRA022986. Raw paired-end sequences were split and quality filtered using an open-source tool Quantitative Insights Into Microbial Ecology 2 (QIIME2) (v2020.11)79. High-quality sequences were denoised to generate amplicon sequence variants (ASVs) using DADA2 (v1.16.0, https://benjjneb.github.io/dada2/)80. The “truncLen” value was set as 220 and 210, while other parameters were set default. Also, the chimeras were filtered. Representative sequences of ASVs were taxonomically identified by using the SILVA database (v132)81. ASVs classified as mitochondria or chloroplast were removed. In total, 1,151,818 and 1,059,283 sequences were assigned to 2705 and 1919 ASVs in the rhizosphere and root, respectively. The ASV table generated from amplicon sequencing was rarefied to the minimum sequencing depth using R (v4.3.3, https://www.r-proje ct.org/) package “Vegan” (v2.6.4)82, and then used to calculate alpha diversity indices (e.g., richness and Shannon) and to perform other downstream analyses. Principal Coordinates Analysis (PCoA) based on the Bray–Curtis distance was analyzed by vegdist function in “Vegan” package to assess beta diversity of bacterial communities. The statistical difference among the two cultivar and two health status was calculated based on PERMANOVA (permutation = 999) using adonis2 function in the “Vegan” package. Community structure dissimilarity based on the Bray–Curtis distance was calculated to evaluated differences within groups. Linear discriminant analysis (LDA) effect size (LEfSe) analysis was conducted using the “microeco” package (v0.20.0)83 to identify bacterial biomarkers from the phylum to genus levels, with significance defined as Wilcoxon P value < 0.05 and logarithmic LDA score > 3. Differential enrichment of bacterial ASVs between different cultivars and health statuses were analyzed using “DESeq2” package (v1.40.2)84. The enriched and depleted ASVs were identified by the threshold of |Log2(Fold change)| > 1 and FDR-adjusted P < 0.05. The “Tax4Fun2” package (v1.1.5) was used to predict the functional profile of microbial communities85. Co-occurrence networks of microbial communities were constructed for two cultivars based on the Spearman’s correlation. The Spearman’s r was calculated using “WGCNA” package (v1.72.1)86, and P value was adjusted by FDR method. ASVs with an average relative abundance of less than 0.05% and those absent from any sample within each cultivar were removed. Only robust (Spearman’s r > 0.65 or r < −0.65) and significant (P < 0.05) correlations were conserved for further network construction. Co-occurrence networks were built using the “igraph” package (version 1.6.0)87, and topological parameters were calculated. The network connectivity characteristics, including within-module connectivity (Zi) and among-module connectivity (Pi) were generated from “igraph” workflow and used to categorize nodes to peripherals (Zi < 2.5 and Pi < 0.62), connectors (Zi < 2.5 and Pi > 0.62), module hubs (Zi > 2.5 and Pi < 0.62), and network hubs (Zi > 2.5 and Pi > 0.62)58. The network was visualized using Gephi (v0.9.7) software (https://gephi.org/). Rhizosphere metabolite sample collection and analysis A protocol adapted from previous studies was used for rhizosphere metabolomic sample collection27,88. Briefly, rhizosphere soil samples collected from the field at the grain filling stage were subjected to untargeted metabolomic profiling using liquid chromatography–tandem mass spectrometry (LC–MS/MS). For metabolite extraction, 250 mg of soil was transferred into a centrifuge tube, and 500 μL of a 70% methanol–water internal standard extraction solution (pre-cooled to −20 °C) was added. Samples were vortexed for 3 min, ultrasounded in an ice water bath for 10 min, vortexed again for 1 min, and then incubated at −20 °C for 30 min. The extracts were centrifuged at 13,500 × g and 4 °C for 10 min, and the supernatants were filtered through a 0.22 μm PTFE membrane for further analysis. The LC and MS conditions and the MS data preprocessing workflow were provided by Metware Biotechnology Co., Ltd. (Wuhan, China). Aliquots were analyzed under positive and negative ion conditions. Specifically, the LC analysis was performed on a Shimadzu LC-30A system (Japan) equipped with an ACQUITY Premier HSS T3 column (1.8 μm, 2.1 × 100 mm, Waters) maintained at 40 °C. The injected sample (4 μL) was eluted using 0.1% (v/v) formic acid in water (solvent A) and 0.1% (v/v) formic acid in acetonitrile (solvent B) at a flow rate of 0.4 mL/min in the following gradient: 0–2 min, 5–20% B; 2–5 min, 20–60% B; 5–6 min, 60–99% B; 6–7.5 min, 99% B; 7.5–7.6 min, 99–5% B; 7.6–10 min, 5% B. Data of MS was acquired on a SCIEX TripleTOF 6600 system (Foster City, CA, USA) using Analyst TF 1.7.1 Software (Sciex, Concord, ON, Canada) in the information-dependent acquisition (IDA) mode. The source parameters were set as follows: ion source gas 1 (GAS1), 50 psi; ion source gas 2 (GAS2), 50 psi; curtain gas (CUR), 25 psi; temperature (TEM), 550°C; declustering potential (DP), 60 V (positive mode), or −60 V (negative mode). And spray voltages were set for 5000 V and −4000 V in positive and negative ion, respectively. The product ion scan parameters were set as follows: collision energy, 30 (positive mode) or −30 V (negative mode); collision energy spread, 15. The raw data were converted to mzXML format using ProteoWizard software (version 3)89. Peak detection, alignment, and retention-time correction were performed using the R package “XCMS” (version 3)90. Features with missing values > 50% within any treatment group were removed, and blank values were replaced by K-nearest neighbor (KNN) method using “impute” package (v1.56.0)91. Then features were finally corrected by Support Vector Regression (SVR) method. Metabolite identification was conducted by matching the processed features against an in-house database (metware database), integrated public databases, predicted spectral libraries, and the metDNA workflow. Metabolites with a comprehensive identification score >0.5 and coefficient of variation (CV) < 0.5 in quality control (QC) samples were retained. Modes of positive and negative ion were subsequently merged. Identified metabolites were annotated using the Kyoto Encyclopedia of Genes and Genomes (KEGG) Compound database (Release 103.1, September 1, 2022) and the Human Metabolome Database (HMDB) (v5.0). R package “MetaboAnalystR” (v3.0) was used to perform data normalization and generate Orthogonal Partial Least Squares-Discriminant Analysis (OPLS-DA)92. Principal Component Analysis (PCA) were used to assess the separation of metabolomes across different groups. The adonis2 function in the R package “Vegan” was used to statistically evaluate the effects of treatments based on PERMANOVA (permutation = 999). Differential metabolites were initially screened by Variable Importance in Projection (VIP) > 1 and P value < 0.05 (ANOVA), while VIP values were extracted from OPLS-DA. Subsequently, differentially abundant metabolites (DAMs) were selected by |Log2Fold change| > 1 and P < 0.05. Bacterial isolation and identification Potentially beneficial bacteria were isolated from the rhizosphere soil of healthy plants (cultivars JM22 and XM26). Briefly, a sterile centrifuge tube containing 1 g of rhizosphere soil, 9 mL of 1× PBS, and 3–4 sterile glass beads was shaken at 200 rpm for 30 min to obtain a soil suspension. Serial dilutions (10−4 to 10−7) were prepared and spread onto 90-mm Petri dishes containing different media: 1/10 tryptophan soy agar (TSA), nutrient agar (NA), or Luria–Bertani (LB) agar (Aoboxing Biology Technology Co., Ltd., Beijing, China). Each combination of medium and dilution was prepared three times. All plates were incubated at 30 °C for 2–5 d. Individual bacterial colonies exhibiting distinct morphologies and colors were selected and purified by streaking onto new LB solid medium. The purified bacteria were stored in 25% (v/v) glycerol at −80 °C for subsequent studies. To obtain fermentation broth, the isolated strains were cultured in LB liquid medium for 24 h at 200 rpm and 30 °C. Fresh liquid cultures were then sent for identification via 16S rRNA gene sequencing at Beijing Tsingke Biotech Co., Ltd (Beijing, China). Briefly, bacterial genomic DNA was extracted using the PowerSoil™ DNA Isolation Kit (Mo Bio Laboratories, Carlsbad, CA, USA) following the manufacturer’s protocol. The 16S rRNA gene was subsequently amplified by PCR using the primer pair 27F/1492R (Supplementary Table 1). The obtained sequences were aligned with the sequences of enriched ASVs using the BLAST tool provided by the National Center for Biotechnology Information (NCBI, https://www.ncbi.nlm.nih.gov/). Antifungal assays of isolated strains against Fp The in vitro antagonistic activity of the isolated strains against the pathogen Fp was tested using a dual-culture assay. Mycelial plugs were prepared by culturing Fp on potato dextrose agar (PDA) for 7d and then excised using a sterile cork-borer. A 6-mm diameter agar plug containing fungal mycelia was placed in the center of each PDA plate, and 1 μL of bacterial fermentation broth was spotted on two opposite sides, each 2 cm away from the Fp. Plates treated with sterile ddH2O served as a negative control. Each treatment was performed in triplicate, and all plates were incubated at 28 °C for 7d. The radial growth of Fp was measured with a ruler. The inhibition rate (IR) was calculated as: IR = (D − d)/D × 100, where D and d are the colony diameters of the pathogen in the negative control and the bacterial treatments, respectively. Spread-plate and cell-free supernatant assays were further performed following a previous procedure93. The sterile supernatant was prepared by centrifuging the isolated strains at 7000 × g for 20 min and then filtering them through a 0.22 μm membrane filter. Two hundred microliters of cultured fermentation broth or cell-free supernatant was spread onto a PDA plate and allowed to dry, followed by inoculation with an Fp plug and incubation as above described. Fp inoculum preparation Fungal inoculum was prepared according to Gao et al.71. First, 200 g of wheat grains were soaked in a sterilized plastic bag with 200 mL of water and heated at 105 °C for 3 min. The grains were then autoclaved twice at 121 °C and 0.1 MPa for 30 min each and cooled to room temperature. Five agar plugs containing Fp mycelia (cultured as described above) were added to the plastic bag and incubated at 25 °C for 10 d under a 12 h light/12 h dark photoperiod. The bags were shaken daily to ensure thorough fungal colonization. The colonized grains were air-dried prior to inoculation. Pot experiment A pot experiment was conducted to validate the capacity of the isolated strains to inhibit FCR and/or improve wheat performance. The treatments included four types: (1) NFP + CK: no Fp and isolates were inoculated, (2) FP + CK: inoculated with Fp but not with isolates, (3) NFP+isolates: inoculated with isolates but not with Fp, and (4) FP+isolates: inoculated with both Fp and isolates. Each treatment was performed in triplicate. The pot experiment was conducted following a protocol modified from previous studies17,94. Briefly, experiments were performed in plastic containers (10 × 7.8 × 11 cm) containing 300 g soil in a greenhouse (20 ± 5°C; 70% air relative humidity; 12 h/12 h, light/dark; about 144 μmol m−2 s−1 light intensity). Soils collected from the field of Wuqiao Experimental Station were autoclaved for 1 h at 121 °C and 0.1 MPa, and mixed with sterile vermiculite in a 3:1 (v:v) ratio to prevent compaction. Wheat seeds were surface sterilized with 75% ethanol, rinsed in sterile ddH2O, and then germinated on the filter paper at 25 °C for 55 h. Before sowing, the soil was thoroughly irrigated and drained to field capacity. Then, 12 germinated wheat seeds were planted evenly in each pot at a depth of 2 cm. At the two-leaf stage, plants were thinned to ~8 uniform seedlings per pot. One Fp-colonized wheat grain of uniform size was placed beside the crown of each seedling as inoculum. The surface was then covered with a 1 cm layer of sterilized soil. For the bacterial inoculation treatment, the isolated bacterial strain was cultured in LB liquid medium for 1 d (30 °C, 200 rpm). The fermentation broth was then centrifuged at 7000 ×g, and then the resulting pellet was resuspended in sterile 1× PBS to adjust the concentration to an OD600 of 1.0. Before placing the colonized grains, 45 mL of the bacterial suspension was added directly to the soil, resulting in a final concentration of ~7 × 10⁶ CFU g-1 of soil for Bacillus sp. and 7 × 10⁷ CFU g−1 of soil for Pseudomonas sp. For control plants not inoculated with bacterial isolates, an equal volume of sterile PBS was applied instead. Plants were irrigated daily for the first 7d, and subsequently watered whenever the soil surface dried. Irrigation was conducted from the bottom using trays with sterile ddH2O. Disease assessment and phenotype survey were performed at 35d after Fp inoculation. Disease scoring of FCR followed a 0–5 scale method95, with 0 indicating no visible symptoms and 5 indicating severe to complete necrosis. The height and fresh weight of the shoot of each wheat seedling were measured, and the underground section was additionally cut for fresh weight measurement and scanning. The roots were washed to remove visible debris, and the roots system was scanned with an EPSON (V750) scanner. The trait of root system architecture (RSA), including root average diameter (AD), total root length (TL), total root volume (RV), and total root surface area (SA), were extracted by WinRHIZO (2013A PRO) software. Statistical analysis Differences between two groups were assessed using Student’s t tests or the nonparametric Wilcoxon test. Differences among multiple groups were analyzed by one-way analysis of variance (ANOVA) followed by Duncan’s multiple range test using the “agricolae” package (v1.3.6)96. Statistical significance was set at P < 0.05. Unless otherwise specified, the statistical analyses in this study were conducted using R (v4.3.3). Plots were visualized with the “ggplot2” (v3.5.1)97, “UpSetR” (v1.4.0)98, “treemap” (v2.4.1)99, and “pheatmap” (v1.0.12)100 packages. The interaction network between microorganisms and metabolites was constructed based on Spearman’s correlation using the “igraph” package (version 1.6.0). For network building, weak (Spearman’s r < 0.5 or r > −0.5) and statistically non-significant (P > 0.01) correlations were discarded. The interaction network was visualized using “ggraph” (v2.1.0)101 and “tidygraph” (v1.2.3)102 packages. Schematic illustrations were created with BioRender (https://www.biorender.com). Supplementary information Supplementary Information Supplementary Table 3 Supplementary Table 6 Supplementary Information Supplementary Table 3 Supplementary Table 6