The Anna Karenina principle in the assembly of plant microbiome under pathogen stress

Abstract

The Anna Karenina Principle (AKP) posits that healthy microbiomes converge toward similar compositional states, whereas dysbiotic microbiomes diverge into distinct and system-specific configurations. Despite its broad recognition in microbiome research, systematic evidence remains scarce as to whether pathogen stress drives plant microbiome assembly in accordance with AKP. To address this knowledge gap, we examined 1,410 samples from multiple compartments (bulk soil, rhizosphere soil, roots, stems, and seeds) across a continental-scale, comparing healthy and Fusarium stalk rot-infected maize using 16S rRNA gene sequencing, complemented with metagenomic sequencing of 93 selected rhizosphere and stem samples. By integrating variations of bacterial community diversity, beta dispersion, average variation degree, and a modified stochasticity ratio, we demonstrated that pathogen-induced microbiome shifts conform to AKP predictions. Notably, AKP-conforming stochastic assembly enriched oligotrophic taxa, resulting in microbial communities with higher GC content, smaller average genome size, and reduced 16S rRNA operon copy numbers. Moreover, the selective enrichment of specific functional traits (including peptidoglycan biosynthesis and degradation, chromatin structure and dynamics, and lipid transport and metabolism) was closely associated with AKP. Our findings support AKP as a useful framework for understanding plant microbiome assembly under pathogen pressure and provide new insights into plant-microbiome-pathogen interactions.

Introduction

Over the course of the long-term co-evolution with plants, microorganisms have become indispensable partners in maintaining plant health and resilience against various stresses, including pathogen infections, pest infestations, drought, and salinity^1–3. Disruption of these microbial communities often leads to dysbiosis, with detrimental consequences for host fitness. Previous studies have shown that many stress-induced microbiome alterations follow stochastic patterns, resulting in destabilized community structures⁴. This phenomenon aligns with the Anna Karenina Principle (AKP), a biological concept inspired by Leo Tolstoy’s observation that “all happy families are alike; each unhappy family is unhappy in its own way”. In microbial ecology, AKP posits that dysbiotic communities exhibit greater compositional variation than healthy ones, reflecting the stochastic nature of stress-driven community assembly process⁵. Although the AKP is an important ecological process describing the stochastic assembly of microbial communities under stress, systematic quantification of this process remains insufficient.

Empirical evidence across human, mammalian, and aquatic invertebrate systems now converges on a unifying observation: host-associated microbiomes under pathogenic or environmental stress conform to the AKP. For instance, healthy cohorts display convergent community structure, whereas stressed individuals diverge idiosyncratically. Meta-analyses of 27 human diseases revealed that approximately 60% of dysbiotic conditions show elevated β-diversity, whereas healthy hosts cluster tightly in ordination space⁶. Comparable patterns occur in natural populations, such as Myodes glareolus exposed to heavy-metal contamination and the cnidarian Aurelia aurita under experimental perturbation, both exhibiting increased inter-individual microbiome variability and reduced temporal stability^7,8. In human mucosal niches, periodontitis and colorectal cancer demonstrate marked β-dispersion, with disease severity correlating positively with the magnitude of microbiome heterogeneity⁹. Collectively, these cross-kingdom observations demonstrate that pathogen-induced dysbiosis disrupts deterministic assembly, driving stochastic and unstable trajectories consistent with the AKP. While AKP-aligned stochastic assembly has been documented in animal and human microbiomes^6,7, its implications for plant microbiomes remain underexplored.

Alterations in the host microenvironment may be induced by pathogen infection. For instance, Verticillium dahliae facilitates its colonization by producing antimicrobial proteins that sculpt the host microenvironment¹⁰. Zymoseptoria tritici, upon infesting wheat, influences other bacterial communities through modulating the host metabolome and microbiome¹¹. Furthermore, pathogen infections trigger plant immune responses and cause substantial nutrient release, thereby dramatically altering the plant microenvironment. Such alterations may promote opportunistic colonization through shifts in root exudate profiles or pathogen-induced tissue necrosis^12,13. These environmental shifts prompt plant-associated microbiomes to undergo adaptive changes, requiring complex trade-offs between competing life-history strategies that involve coordinated adjustments in metabolic pathways, physiological functions, and genomic architecture¹⁴. While microbial life-history strategies in response to abiotic stresses have been well-documented, shifts in these strategies under pathogenic stress remain unclear.

To address this, we studied Fusarium stalk rot, a devastating maize disease causing 30–50% global yield losses while producing hazardous mycotoxins that compromise food and feed safety¹⁵. We analyzed healthy and Fusarium stalk rot-infected maize microbiomes via continental-scale amplicon and metagenome sequencing. We tested whether stalk rot induces AKP-aligned microbiome changes, examined associated microbial life-history strategies and genomic traits, and identified characteristic species and functional profiles. We hypothesized pathogen infection would drive AKP-conforming microbiome variation correlated with microbial life-history trade-offs.

Results

Pathogen-induced microbiome reassembly is compliant with AKP

In this study, we collected 1,410 continental-scale maize samples from different compartments including rhizosphere soil, roots, stems and seeds from both healthy and stalk rot-affected plants (Fig. S1, Supplementary Data 1). Pathogenicity assays confirmed the virulence of 25 Fusarium verticillioides isolates obtained from diseased maize stem (Supplementary Data 2). Microbial analysis revealed pathogen-induced, site-specific shifts in bacterial diversity, with pathogen infection significantly increasing diversity at multiple locations (Fig. 1A; rhizosphere: seven sites, root: six sites, stem: 27 sites, seed: five sites; P < 0.05). Permutational multivariate analysis of variance (PERMANOVA) revealed that bacterial community structure was collectively shaped by both the compartment niche and stalk rot disease at each sampling site, with the exception of site YL (Fig. S2, Supplementary Data 3). Random forest and correlation analyses demonstrated that environmental factors had a greater impact on bacterial community diversity under pathogen stress (Fig. 1B, Supplementary Data 4). Further linear mixed-effect models indicated that environmental factors (e.g., pH, AFe, AP, NH₄⁺-N, DOC) had a significant effect on diversity (Fig. S3, Supplementary Data 5). Soil pH increased bacterial diversity in the rhizosphere and roots of healthy maize but decreased it in the stems of diseased plants. AFe increased bacterial diversity in the rhizosphere, roots, and stems of healthy maize. DOC positively influenced bacterial diversity in both healthy and diseased maize seeds, whereas NH₄⁺-N consistently reduced diversity across the rhizosphere, roots, and stems, irrespective of plant health status. Mantel tests further confirmed enhanced environmental influence on bacterial community composition under pathogen stress (Fig. 1C; Supplementary Data 4).

Fig. 1. Effects of pathogen infection and environmental factors on the diversity and composition of bacterial communities.

A The effects of pathogen infection on the diversity of maize bacterial community varied in different compartments and sampling sites. Zhangye (ZY), Yongzhou (YZ), Yangquan (YQ), Yingkou (YK), Yanjiang (YJ), Xuzhou (XZ), Xiangyang (XY), Xinxiang (XX), Xuchang (XC), Shizuishan (SZS), Suiyang (SY), Shenqiu (SQ), Siping (SP), Sanming (SM), Suihua (SH), Shaoguan (SG), Qujing (QJ), Mianyang (MY), Liuzhou (LZ), Lu’an (LA), Jinan (JN), Jiujiang (JJ), Jinhua (JH), Harbin (HEB), Guizhou (GZ), Guyuan (GY), Fuyang (FY), Dali (DL), Changshu (CS), Changchun (CC), Baise (BS), Baoding (BD). B Contributions of environmental factors to the diversity of microbial communities and differences in pathogen infection and compartments based on correlation and random forest models. Circle size represents the variable importance (that is, proportion of explained variability calculated via random forest). Colors represent Spearman correlations. The abbreviations of soil properties accorded to the Method. C Contributions of environmental factors to the composition of microbial communities and differences in pathogen infection and compartments based on Mantel test. The thickness of the line represents the magnitude of the correlation, and the color of the line represents the significance. HRh healthy plants rhizosphere, DRh diseased plants rhizosphere, HRo healthy plants root, DRo diseased plants root, HSt healthy plants stem, DSt diseased plants stem, HSe healthy plants seed, DSe diseased plants seed. Asterisks denote significant differences (*P < 0.05; **P < 0.01; ***P < 0.001).

We quantified microbial community variation using four metrics: community richness standard deviation (SD) (richness SD), beta dispersion, average variation degree (AVD), and modified stochasticity ratio (MST). These metrics can respectively assess the degree of community variation from the perspectives of diversity, community structure, individual zero-radius operational taxonomic units (ZOTUs), and community assembly processes. Pathogen infection significantly increased bacterial community variability, with the most pronounced effects observed in stems (Fig. 2A, P < 0.001). Both root and stem compartments exhibited elevated beta dispersion under pathogen infection (Fig. 2B, P < 0.001). Furthermore, pathogen infection significantly enhanced AVD (Fig. 2C; root: P < 0.01, stem: P < 0.001, seeds: P < 0.05) and MST values (Fig. 2D; rhizosphere: P < 0.001). Beta Nearest Taxon Index (|βNTI | > 2, deterministic; |βNTI | < 2, stochastic) jointly with Raup-Crick based on Bray-Curtis (RC_bray) partitions stochasticity into four ecological processes: homogeneous selection (βNTI > 2), heterogeneous selection (βNTI < –2), homogenizing dispersal (|βNTI | < 2 and RC_bray < –0.95), dispersal limitation (|βNTI | < 2 and RC_bray > 0.95), and undominated processes (|βNTI | < 2 and |RC_bray | < 0.95). The results revealed that pathogen infection strengthened homogenizing dispersal processes in rhizosphere, root, and stem bacterial communities. Furthermore, pathogen infection enhanced the role of undominated processes (weak dispersal and ecological drift) in shaping seed bacterial communities (Fig. 2E). We developed a novel composite AKP score by integrating these four metrics to quantify community assembly conformity to AKP principles. These four metrics capture the core pattern of community stochastic from distinct yet complementary statistical perspectives (species richness, state heterogeneity, species-level variation, and the stochasticity of assembly processes). By standardizing and averaging them, we aimed to derive a robust index that comprehensively reflects the multidimensional features of the AKP effect, thereby avoiding reliance on any single metric. A higher AKP value indicates that the microbial community deviates further from a healthy steady state, exhibiting greater structural disorder and unpredictability. Furthermore, principal coordinate analysis (PCoA) and correlation analysis clarified the relationships among these different metrics. The results demonstrate that the AKP value effectively integrates these four metrics, showing strong and significant positive correlations with each of them (Fig. S4, Supplementary Data 6, r = 0.61–0.95, Adjusted P < 1×10^-27). Multiple regression and random forest analyses revealed that none of the assessed environmental factors showed a significant correlation with the AKP value (P > 0.05; Fig. S5, Supplementary Data 7), suggesting that the pathogen-induced microbiome restructuring follows the AKP and may be independent of abiotic conditions. Furthermore, analysis of two public datasets revealed that the bacterial community changes induced by Fusarium stalk rot in maize¹⁶, alongside the disease-associated shifts in the ginger bacterial community¹⁷, are largely consistent with the predictions of the AKP (Fig. S6–7).

Fig. 2. The AKP caused by pathogen infection.

A The effect of pathogen infection on diversity variation (SD) of bacterial communities. B The effect of pathogen infection on beta dispersion of bacterial communities. C The effect of pathogen infection on AVD of bacterial communities. D The effect of pathogen infection on MST of bacterial communities. E The relative influence of community assembly process among diseased and healthy plants in different compartments based on βNTI and RC_bray. Asterisks denote significant differences (*P < 0.05; **P < 0.01; ***P < 0.001) and ns denotes no statistical significance.

The relationship between microbial life history strategies and the AKP value

To investigate the life-history strategy shifts driven by microbial community assembly, we explored microbial life strategy based on r/K selection theory and 16S rRNA operon (rrn) copy number under pathogen stress. Pathogen infection differentially altered bacterial 16S rrn copy numbers across plant compartments, with significant reductions in roots (P < 0.01) and stems (P < 0.001), while increasing them in seed-associated communities (Fig. S8A). Taxa with high rrn copy numbers are typically r-strategists (copiotrophs), adapted for rapid growth under nutrient-rich conditions. Conversely, taxa with low rrn copy numbers are typically K-strategists (oligotrophs), adapted for slow, efficient growth under nutrient limitation. The specific thresholds (top 40%, ≥2.45 copies for copiotrophs; bottom 40%, ≤1.99 copies for oligotrophs, or middle) were determined based on the distribution within our dataset (Fig. 3A, Supplementary Data 8). The results showed that pathogen infection significantly decreased the relative abundance of copiotrophs across compartments (roots: P < 0.05; stems: P < 0.001; seeds: P < 0.01; Fig. 3B), while significantly increased the relative abundance of oligotrophs in stem (P < 0.001) and seeds (P < 0.05, Fig. 3C).

Fig. 3. The link between microbial life history strategies and Anna Karenina principles.

A Distribution of 16S rrn copy number of the maize associated microbiome. B Effect of pathogen infection on the relative abundance of copiotrophs. C Effect of pathogen infection on the relative abundance of oligotrophs. D Association between 16S rrn copy number and AKP value. E Association between relative abundance of copiotrophs and AKP value. F Association between relative abundance of oligotrophs and AKP value. RA relative abundance. Asterisks denote significant differences (*P < 0.05; **P < 0.01; ***P < 0.001) and ns denotes no statistical significance.

Functional traits serve as a good indicator of microbial adaptation to environmental stresses and resource fluctuations. Given the significant differences in microbial diversity and MST in the rhizosphere and stem between healthy and diseased maize, we performed metagenome sequencing on 93 representative samples from both healthy and diseased maize, yielding ~1.65 Tb of data. Rarefaction curves illustrate the impact of sequencing depth on both species and functional diversity (Fig. S9). Focusing specifically on the bacterial communities, the main interest of this study, we analyzed genomic traits following the removal of eukaryotic and viral sequences. Our results showed that pathogen infection significantly decreased the average genome size of the rhizosphere community (P < 0.05; Fig. S8B) while increasing GC content in stem-associated bacteria (P < 0.001, Fig. S8C). Notably, bacterial maximum growth rates remained unaffected (P > 0.05, Fig. S8D). Variance partitioning analysis (VPA) revealed that the variation in genome size was primarily shaped by community composition rather than environmental factors (Fig. S10, Supplementary Data 9). These observations suggest a significant association between pathogen infection and bacterial genome characteristics.

The relationships between microbial life-history strategies and AKP value were examined. The results revealed a significant negative correlation between mean 16S rrn copy numbers and AKP value (Fig. 3D, Fig. S11, Supplementary Data 10), particularly pronounced in roots (R = -0.54, P < 0.001) and stems (R = -0.31, P < 0.05). Additionally, AKP value exhibited a negative correlation with copiotroph abundance across compartments (root: R = -0.63, P < 0.001; stem: R = -0.47, P < 0.001; and seed: R = -0.84, P < 0.001; Fig. 3E). In contrast, AKP value showed a positive correlation with oligotroph abundance (root: R = 0.63, P < 0.001; stem: R = 0.48, P < 0.001; seed: R = 0.84, P < 0.001; Fig. 3F). Furthermore, the mean 16S rrn copy number as well as the relative abundance of copiotrophs and oligotrophs were more strongly correlated with AKP value in diseased plants compared to healthy plant microbiomes (Fig. S12). In addition, AKP value was significantly negatively correlated with bacterial average genome size and maximum growth rate (P < 0.001; Fig. S13, Supplementary Data 11), but positively correlated with GC content (P < 0.001; Fig. S13). These findings demonstrate that microbial community’s life-history strategy plays a significant role in the community assembly dynamics under pathogen-induced stress.

Classification and functional composition of microorganisms attributed to AKP

The multiple regression and correlation analysis revealed that AKP value showed significant positive correlations with several bacterial phyla, including Bacteroidota, Chloroflexi, Patescibacteria, Acidobacteriota, Gemmatimonadota, and Myxococcota (P < 0.001; Fig. 4A), while exhibiting a significant negative correlation with Proteobacteria (P < 0.001; Fig. 4A). This pattern was consistent across the different plant compartments (Fig. S14). These findings are consistent with the observed variations in community diversity metrics (richness SD, beta dispersion, AVD and MST), all of which showed significant phylum-level associations (P < 0.001; Fig. 4B). Notably, Firmicutes displayed a significant negative correlation with MST (P < 0.001; Fig. 4B), underscoring the complex interplay between AKP-aligned community assembly and specific bacterial lineages. Furthermore, at the genus level, a significant negative correlation was observed between the AKP value and the relative abundances of Enterobacter, Klebsiella, Pantoea, and Rouxiella within the roots. In the stem compartment, the abundances of Gluconobacter and Rahnella were also significantly negatively correlated with the AKP value (P < 0.05; Fig. S15).

Fig. 4. Taxa and functional compositions related to the AKP value.

A Contribution of relative abundance of microbial phyla to AKP value based on correlation and best multiple regression models. B Association of relative abundance of bacterial phyla with AKP-related life history strategies. C Module analysis of microbial functional network in maize rhizosphere. Nodes represent KOs, and edges represent significant correlations between them. The identified major modules are color-coded: Module #1 in orange, Module #2 in dark green, and Module #3 in light green. All other KOs that were not assigned to a dominant module are shown in gray. D Microbial functional modules associated with AKP. A linear mixed-effects model with two-sided tests was used for statistical analysis. The gray line is the regression line for each KO, and the colored line is the regression line for the mean of all KOs. E Differences in modules associated KO with AKP. F Module analysis of microbial functional network in maize stem. G Microbial functional modules associated with AKP. A linear mixed-effects model with two-sided tests was used for statistical analysis. The gray line is the regression line for each KO, and the colored line is the regression line for the mean of all KOs. H Differences in modules associated KO with AKP. Asterisks denote significant differences (*P < 0.05; **P < 0.01; ***P < 0.001).

To explore the link between bacterial-specific functional composition and AKP value, we constructed separate co-occurrence networks for the rhizosphere and stem microbial KEGG Orthologies (KOs). We identified 10,204 KOs through KEGG annotation and filtered them to retain only those present in ≥50% of samples, then performed pairwise Spearman correlation analysis (correlation coefficient |r | > 0.8 and a significance level of P < 0.001) to detect significant co-occurrences. The analysis revealed that the rhizosphere KOs consisted of 3,631 nodes which were organized into three main modules (Fig. 4C). Notably, module #1, with 1,733 nodes, and module #2, with 1,620 nodes, exhibited a significant negative correlation with AKP value (P < 0.0001; Fig. 4D). Further examination of the microbial functional network modules in the rhizosphere of healthy and diseased maize revealed that K11000 (callose synthase, glycosyltransferases) and K16048 (steroid degradation) were enriched in module #1 of the healthy maize rhizosphere (Fig. 4E). In module #2, K00104 (reactive oxygen species regulator) and K03200 (promotes bacterial adhesion to host cells) were enriched in the healthy maize rhizosphere (Fig. 4E).

Additionally, maize stem KOs had a total of 6019 nodes, forming three main modules (Fig. 4F). Among them, there was a significant positive correlation between module #1 (2480 nodes), module #2 (1822 nodes), and module #3 (1717 nodes) with AKP value (P < 0.0001; Fig. 4G). We selected the two modules with the highest conditional R² for the variance test and found that, in the healthy maize stem, module #1 exhibited upregulation of K06907 (bacteriophage tail sheath protein), K16709 (glycan metabolism), and K14188 (D-amino acid metabolism), while K08259 (lysostaphin; peptidoglycan biosynthesis and degradation proteins) and K01473 (amino acid metabolism) experienced downregulation (P < 0.05; Fig. 4H). In healthy maize stem microbial function module #3, K03574 (DNA repair and recombination proteins), K15725 (transporters), K05516 (Protein families: Genetic information processing) and K02190 (cbiK; Cofactor and vitamin metabolism) were upregulated, whereas K13571 (Protein families: Genetic information processing), K05364 (peptidoglycan biosynthesis) and K05565 (Protein families: Genetic information processing) were downregulated (P < 0.05; Fig. 4H). Consistent with the functional network, COG functional data indicated that microbiomes from healthy maize exerted more functions in cell cycle control, cell division, chromosome partitioning, amino acid transport and metabolism, nucleotide transport and metabolism, coenzyme transport and metabolism, carbohydrate transport and metabolism, replication recombination and repair, cell motility, intracellular trafficking, secretion and vesicular transport potential (Fig. S16). While diseased stems exhibited increased RNA processing and modification, chromatin structure and dynamics, lipid transport and metabolism, extracellular structures, and nuclear structure maintenances.

Overall, the community assembly process induced by pathogen infection follows the AKP. This assembly process selectively enriches oligotrophic taxa with smaller genomes and higher GC content, thereby driving shifts in the structural diversity and functional potential of bacterial communities.

Discussion

Elucidating the ecological principles governing microbiome assembly under pathogen stress is crucial for developing sustainable plant health strategies. While molecular mechanisms of individual microbial responses to pathogens are well-documented, the community-level ecological processes underlying microbiome reorganization during infection remain poorly understood. This study systematically examines how pathogen infection alters microbial community assembly, and provides empirical evidence supporting the AKP in plant microbiomes, demonstrating that stochastic processes dominate microbial assembly in diseased plants (Fig. 5). These findings contribute to our understanding of stress-driven microbiome dynamics and offer novel ecological perspectives on plant-microbe-pathogen interactions.

Fig. 5.

Schematic representation of the AKP caused by pathogens.

Microbial dysbiosis in plant holobionts serves as a key indicator of compromised health status, with diseased plants exhibiting greater environmental sensitivity and increased community variability compared to healthy counterparts. This phenomenon agrees with observations from animal and human microbiome studies, where pathogen stress similarly drives AKP-compliant community reassembly^4,9. In plant systems specifically, pathogen infection increases both alpha diversity (species diversity) and beta diversity (community dissimilarity)^18,19. Pathogen infection in plants disrupts the stability of the host–microbiome system. According to the second law of thermodynamics, the living organisms must expend increasing metabolic energy to counteract rising entropy and maintain functional order. When such compensatory efforts become overwhelmed under biotic stress, dysbiosis manifests as elevated compositional uncertainty. This entropic increase is conceptually analogous to the AKP pattern but should not be interpreted as a mechanistic cause²⁰. Our findings reveal varied compartment patterns of AKP compliance under the same pathogen stress, as evidenced in changes by altered diversity, elevated beta dispersion, increased AVD, and a shift toward stochastic community assembly. These findings demonstrate that pathogen stress drives plant microbial communities toward a stochastic and destabilized state, and showed that its predictions were generally validated in two independent studies. Furthermore, our study proposes a composite metric, the AKP score, which should be interpreted as an operational composite proxy for dysbiosis-associated variation in microbial communities, rather than as a direct measure of the underlying ecological processes. Our results leverage the AKP framework to interpret these shifts, offering mechanistic insights into microbiome assembly under biotic stress and supporting the predictable patterns of community assembly during pathogen infection. Nevertheless, future validation through more extensive sampling across broader geographical ranges, a wider variety of crop cultivars, and pathogens, coupled with a mechanistic dissection of specific ecological processes, remains necessary.

Our study also revealed that pathogen infection induces a systematic shift in microbial life-history strategies in plant-associated communities, indicated by reduced 16S rrn copy number and increased oligotroph prevalence. This shift may be consistent with a disruption of plant-mediated nutrient supply, potentially leading to increased microbial competition for resources²¹. The observed positive association between AKP value and oligotrophic life history strategies underscores how pathogen-induced potential nutrient limitation and metabolic disruption collectively shape stress-adapted community assembly²². By characterizing these community-level traits, our study contributes a theoretical framework for understanding how the microbiome responds to biotic stress through resource-conserving adaptations. Nevertheless, given the prevalence of microbial “dark matter” within the rhizosphere soil, the inference of bacterial growth rates from 16S rrn copy number and the subsequent classification into oligotroph and copiotroph lifestyles are proxy approaches and require future validation through direct metabolic measurements or integrated multi-omics approaches.

Bacterial genomic traits, including genome size and GC content, exhibit strong environmental sensitivity to abiotic factors, such as drought stress, nutrient availability, and pH^14,23. Our findings suggest pathogen infection mediates similar genomic adaptations in plant-associated microbiome, potentially via induced carbon limitation that selects for microbes with reduced genome sizes, elevated GC content, and other resource-conserving genomic strategies, characteristic of stress-adapted oligotrophs. Although bulk soil SOM and DOC did not show a consistent negative correlation with disease incidence across compartments, this does not preclude localized carbon limitation in the root or rhizosphere microenvironment. Pathogen-induced exudate changes or microbial metabolism may transiently alter carbon availability in ways not captured by bulk soil measurements²⁴. The reduction in genome size is closely associated with the decrease in genetic diversity and multifunctionality, and it is commonly observed that bacterial genomes tend to be smaller under nutrient-rich conditions compared with those in nutrient-poor conditions^14,25. In contrast, healthy plant microbiomes select for bacterial symbionts with larger genomes, which support a broad spectrum of metabolic capabilities, essential for sustaining mutualistic plant-microbe interactions^26,27. The GC content of bacterial genomes also varies with environmental conditions. For instance, microbial communities found in bare soils exhibit a higher GC content compared to those residing in vegetated soils²⁸, highlighting how habitat-specific selective pressures shape microbiome genome characters. Carbon limitation serves as a major selective pressure driving genomic adaptation in microbial communities, promoting characteristic adaptations, such as increased GC content and reduced genome size under nutrient stress²⁹. The stoichiometric advantage of GC base pairs (C:N ratio = 1.13) over AT pairs (C:N = 1.42) provides a plausible explanation for this phenomenon, as microbial communities may thus favor higher GC content and smaller genomes under carbon-limited conditions³⁰. While our study provides genomic and life-history evidence of potential nutrient limitation, future work incorporating metabolomic profiling would be valuable to directly quantify carbon availability in the plant-microbe interface. Furthermore, while these principles help explain observed patterns in pathogen-stressed plant microbiomes, significant knowledge gaps remain regarding: (1) the temporal progression of community assembly following pathogen invasion, and (2) the functional implications of these genomic adaptations for plant-microbe interactions under stress conditions.

The specific genera belonging to Proteobacteria observed in our study (e.g., Enterobacter and Klebsiella in roots; Gluconobacter and Rahnella in stems) exhibited a negative correlation with the AKP value. This phenomenon may be attributed to the capacity of healthy plants to actively shape their microbiome through immune responses and physiological status, such as antioxidant enzyme activity. By maintaining high physiological vitality, including robust osmotic regulation and antioxidant capacity, healthy plants likely create a supportive environment for the colonization of these beneficial Proteobacteria in the plants^31–33. This host phenotype-driven deterministic process may thus underlie the observed negative correlation with the AKP value. Functional traits related to AKP values highlight the fundamental importance of microbial life history strategies in community assembly under pathogen stress. Our analyses identified several key functional categories associated with AKP patterns, such as lipid transport and metabolism, nuclear structures, and peptidoglycan biosynthesis and degradation proteins are related to AKP value. These functions likely facilitate oligotroph adaptation to stressed plant environments by modulating membrane biophysical properties and preserving genomic integrity under stress conditions^28,34. In contrast, healthy maize microbiomes were enriched for functions supporting carbohydrate metabolism, amino acid metabolism, and cell motility, critical processes for maintaining symbiotic relationships and recruiting beneficial copiotrophs to plant surfaces^19,35. Notably, in healthy maize, there is an enriched function of hydrogen peroxide production, which regulates reactive oxygen species (ROS) generation for pathogen defense³⁶. Furthermore, the enrichment of vitamin B₁₂ synthesis genes (K02190) in healthy microbiomes underscores their metabolic versatility, as this cofactor is essential for numerous cellular processes including nucleotide biosynthesis, amino acid metabolism, and redox reactions^37,38.

Taken together, our findings demonstrate that pathogen infection drives stochastic assembly processes in plant-associated microbiome, resulting in increased variability in community diversity and structure, patterns that align with the AKP. These pathogen-induced stochastic assembly favors oligotrophic taxa, characterized by distinctive genomic signatures, including reduced genome sizes, elevated GC content, as well as diminished growth rates. Our study links the life history strategies of plant microbial communities to their assembly processes under pathogen stress. However, the functional significance of oligotroph enrichment for plant pathogen resistance remains an important open question. Additionally, as this study focused exclusively on bacterial communities, future work should examine whether fungal communities follow similar AKP-governed dynamics. By integrating microbial ecology with plant pathology, this work advances our understanding of tripartite pathogen-plant-microbiota interactions. Future studies should elucidate the molecular and ecological mechanisms underlying dysbiosis in plant-associated microbiota, thereby facilitating the development of innovative biocontrol strategies.

Methods

Sample collection and processing

In this study, 33 sites were sampled across a broad range of latitudes (23.48°N to 46.38°N) and longitudes (100.27°E to 126.57°E), with distances between sites ranging from 40 to 3317 km. The selected fields characterized by a history of long-term cultivation exhibited signs of maize stalk rot and the maize plants were in the milk stage during sampling. At each site, data collected included latitude, longitude, altitude and the incidence and disease index of maize. Each site was subdivided into five subplots measuring 10 × 10 m. From each subplot, three healthy and three diseased maize plants were sampled. Samples collected from each plant comprised bulk soil, rhizosphere soil, roots, stems, and seeds. Maize ears were harvested and stored in sterile bags, and stems were collected from nodes 2-4, totaling nine stems per sample. Rhizosphere soil and root samples were obtained by washing roots with adherent soil in phosphate-buffered saline (PBS) followed by centrifugation. Bulk soil samples were collected from areas between maize rows, intentionally avoiding regions directly impacted by the plants. Three portions of soil, at a depth of 5–15 cm, were collected and combined in sterile bags. Each subsample from all sites served as a replicate, yielding five replicates per site. Samples were maintained on ice and transported via a cold chain to the laboratory for analysis. Soil factors measured included soil organic matter (SOM), total nitrogen (TN), total phosphorus (TP), total potassium (TK), available potassium (AK), ammonium nitrogen (NH₄⁺-N), nitrate nitrogen (NO₃^--N), available phosphorus (AP), available iron (AFe), pH, soil water content (SWC), and dissolved organic carbon (DOC), all analyzed using standard soil testing methods. Climate data, such as mean annual temperature (MAT) and mean annual precipitation (MAP), were obtained from the WorldClim database (www.worldclim.org) corresponding to each sampling point’s coordinates. Additionally, soil types were identified using the China Soil Database (http://vdb3.soil.csdb.cn/), with black, brown, and red soils being the predominant types distributed longitudinally from north to south.

To prepare the root samples, they were initially placed in 50 mL centrifuge tubes containing 20 mL of sterile PBS and subjected to vortexing at 120 rpm for 20 min. Following this, the roots were removed using tweezers, and the remaining suspension was centrifuged at 8000 rpm for 5 min to isolate the rhizosphere soil. The roots were then disinfected using a 5% sodium hypochlorite solution and 75% ethanol, thoroughly washed with sterile water and ground into a fine powder with liquid nitrogen. For the preparation of stem and seed samples, these tissues were surface-disinfected with the same disinfectant solution, followed by a wash with sterile water. Subsequently, the maize tissues were blended with 200 mL of sterile water using a WARING lb20es blender. The resulting mixture was filtered through sterile gauze, and the filtrate was transferred to a 50 mL centrifuge tube. Finally, the stem and seed samples were collected by centrifugation at 8000 rpm for 5 min. Potential pathogenic fungi were isolated from the stems of diseased plants using a single-spore isolation method³⁹. Species identification utilized a phylogenetic analysis of the (translation elongation factor 1-alpha) tef1-α gene. PCR amplification with primers EF-1 (5’-ATGGGTAAGGARGACAAGAC-3’) and EF-2 (5’-GGARGTACCAGTSATCATG-3’) were performed. Sanger-sequenced products were compared against GenBank and validated via Maximum Likelihood trees. A subset of 25 representative F. verticillioides strains were subjected to pathogenicity tests. The pathogenicity of the Fusarium isolates was assessed following a previously described method⁴⁰.

DNA extraction, sequencing and data preprocessing

Microbial DNA extraction was conducted using the FastDNA® SPIN Kit for Soil (MP Biomedicals, Solon, USA), following the manufacturer’s protocols meticulously to ensure the integrity of the total DNA obtained. After extraction, the concentration and purity of the DNA were rigorously assessed to confirm its suitability for further analysis. This study characterized the bacterial microbiota from 1410 samples by amplifying the V5-V6 hypervariable region of the 16S rRNA gene with primers 799 F (5’-AACMGGATTAGATACCCKG-3’) and 1115 R (5’-AGGGTTGCGCTCGTTG-3’). Raw paired-end reads were processed in USEARCH v11⁴¹. Quality control involved trimming primers/adapters, filtering reads below Q30, and merging paired-end reads into consensus sequences. The UNOISE3 algorithm was applied for denoising to generate zero-radius Operational Taxonomic Units (zOTUs). An abundance filter removed zOTUs with a total read count below eight across all samples to minimize artifacts. Taxonomic assignment of representative zOTU sequences was performed against the SILVA v138⁴² database using a dual-algorithm approach: the ‘sintax’ algorithm for k-mer-based classification and the RDP Naive Bayesian Classifier, both with a confidence threshold of 0.8. From 169,849,682 reads, rarefaction analysis using the vegan package yielded a final dataset of 1,986,120 reads, representing 40,305 unique zOTUs for analysis.

For metagenomic sequencing, a total of 93 DNA samples from rhizosphere and stem tissues were sequenced using 150 bp paired-end reads on Illumina NovaSeq 6000 platforms. Each sample yielded approximately 15 GB of clean data. To eliminate host-derived sequences, we constructed a dedicated host genome database (Zm-B73-REFERENCE-NAM-5.0, NCBI reference sequence GCF_902167145.1) and employed Bowtie2 v2.4.1 for efficient alignment and removal⁴³. To specifically focus on bacteria, potential eukaryotic and viral contigs were detected and removed using EukRep v0.6.7 and BBMap v39.01. The average genome size of each sample was assessed using the MicrobeCensus pipeline⁴⁴, while GC% was calculated with Quast v5.2.0⁴⁵. Additionally, the minimal doubling time was estimated by analyzing codon usage bias with gRodon v2.3.0, which provided insights into bacterial maximum growth rates and responding minimal doubling times⁴⁶.

To analyze functional genes, the quality-controlled reads were subjected to de novo assembly using MEGAHIT (v1.2.9)⁴⁷. The assembly was performed with hardware acceleration (POPCNT and BMI2 support) to enhance computational efficiency. The assembly process employed a multi-k iterative strategy (k-mer list: 21, 29, 39, 59, 79, 99, 119, 141), allowing the algorithm to resolve regions with varying coverage and complexity by progressively increasing k-mer sizes. Finally, all assembled contigs from individual samples were pooled to create a comprehensive contig set for downstream gene prediction and functional annotation. Contig predictions were made using Prokka v1.14.5⁴⁸, and non-redundant gene catalogs were generated through CD-HIT v4.8.1 clustering, applying a similarity threshold of 0.95. For functional annotation, DIAMOND was used to align the gene catalog against the eggNOG databases (version 5.0), with Eggnog-Mapper v1.0.3 facilitating the mapping of annotations to various functional classification schemes⁴⁹. Specifically, annotation results were mapped to Kyoto Encyclopedia of Genes and Genomes (KEGG) Orthology (KO) profiles, providing insights into metabolic pathways and gene functions⁵⁰. Additionally, the data were categorized into Clusters of Orthologous Groups of proteins (COG) functional categories, offering a comprehensive view of protein families and their roles in cellular processes. Using the data derived from Salmon⁵¹, we associated the abundance of each gene with its corresponding KO/COG identifiers via a custom script. Unannotated genes were excluded from downstream analysis. The gene abundances were then aggregated by KO or COG identifier to generate a sample-specific count matrix, with rows representing functional categories (KO/COG), for differential comparison. To mitigate biases arising from variations in sequencing depth and technical artifacts, the raw counts were normalized to Reads Per Million (RPM).

Data statistics and analysis

All statistical analyses were performed using the R environment (https://www.r-project.org/). The richness and diversity of healthy and diseased maize bacterial communities were calculated using the “vegan” package⁵². Differences in the diversity of healthy and diseased maize bacterial communities across various sample sites were examined using the “wilcox.test” function from the “ggpubr” package. To explore the contributions of environmental factors to the diversity of microbial communities, as well as the differences in pathogen infection and compartments, the significance and correlation of these factors were calculated using the “randomForest” package. The influence of environmental factors on diversity was investigated using linear mixed-effects models (LMMs). To mitigate collinearity, environmental factors with a variance inflation factor (VIF) greater than 5 were excluded from the analysis. The effect of stalk rot disease on community structure within each sampling site was assessed by conducting a permutational multivariate analysis of variance. Mantel analysis in the “linkET” package was utilized to analyze the relationships between environmental factors and community compositions.

To measure the degree of community variability, the SD of bacterial community richness was calculated using the “dplyr”, the community beta dispersion using “vegan”, and calculated the AVD⁵³, and normalized stochastic ratio using “NST”⁵⁴. To minimize the influence of sampling sites, these metrics were calculated individually for each site. The microbial community assembly process in different subgroups was assessed using the “picante” null model package and beta nearest taxon index (βNTI and RC_bray)⁵⁵. To measure the AKP of the community, the four metrics were integrated by averaging the standardized values calculated using the formula STD = (X - Xmin) / (Xmax - Xmin), where STD represents the standardized variable and X, Xmin, and Xmax denote the target variable, sample minimum, and maximum, respectively. Given that the richness SD contained only a single value per group, the per site mean value of the other three metrics were calculated to ensure sample consistency and to account for potential site effects. To assess the relationships among all indicators, we performed PCoA and correlation analysis. The P-values from the correlation analysis were adjusted for multiple comparisons using the FDR method. The effects of environmental factors on AKP value were examined using multiple regression and random forest analyses. To mitigate collinearity, environmental factors with a VIF greater than 5 were excluded from the analysis. The random forest analysis was configured with 1500 trees (ntree = 1500) and 1000 replicates (nrep = 1000). The multiple regression analysis was performed using forward selection.

The 16S rrn copy number in each bacterial taxon was estimated using the rrnDB database^56,57. Based on 16S rrn copy number distributions, we classified taxa as copiotrophs (r-strategists; top 40%, ≥2.45 copies), oligotrophs (K-strategists; bottom 40%, ≤1.99 copies), or middle. Spearman’s correlation was calculated between the average 16S rrn copy number, the relative abundance of copiotrophs, the relative abundance of oligotrophs, and AKP value. Relationships between AKP value and relative abundance at the phylum and genus level were analyzed using Spearman’s correlation via the corr.test function in the “psych” package.

To investigate the functional differences in bacterial communities between healthy and diseased maize rhizosphere and stem compartments, we performed differential analysis of COG profiles. Statistical significance was assessed using the Wilcoxon rank-sum test, and the results were visualized via heatmaps. To investigate functional traits associated with the AKP, we constructed co-occurrence networks of KEGG Orthology terms from rhizosphere and stem compartments using the “psych” and “igraph” packages in R⁵⁸. Spearman rank correlation was applied, with a stringent threshold of |r | > 0.8 and FDR-corrected P < 0.001, to retain robust, biologically meaningful associations and minimize spurious connections. The resulting correlation matrices were imported into igraph to build undirected networks, where nodes represent KOs and edges indicate significant correlations. To identify cohesive modules, we applied a fast greedy modularity optimization algorithm “cluster_fast_greedy()”, which iteratively merges communities to maximize modularity. Node module membership was stored in the modularity attribute for downstream analysis. Regression analysis in the “nlme” package was employed to examine the relationship between AKP value and vertices in the largest modules. Differences in KO counts between healthy and diseased plants (FDR q < 0.05 and |log₂FC | ≥ 2) were explored using differential analysis implemented in the “edgeR” package⁵⁹. All the schematic figures have been created with https://biorender.com/.

Author contributions

D.L. conceived and designed the study under the supervision of L.C. and X.Z. D.L. collected the samples and performed the lab work experiment. D.L., Z.Q., C.W., and Z.P. contributed to the data analysis and visualization. D.L. wrote the manuscript; L.C. contributed substantially to revisions. All authors reviewed the manuscript.

Data availability

The raw sequence data reported in this paper have been deposited in the Genome Sequence Archive in National Genomics Data Center, China National Center for Bioinformation / Beijing Institute of Genomics, Chinese Academy of Sciences (GSA: CRA017629, CRA017732, CRA017841) that are publicly accessible at https://ngdc.cncb.ac.cn/gsa.

Code availability

The data visualization code can be found in https://github.com/lidas666/AKP.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41522-026-00964-2.