Foliar metal micronutrients reshape rhizosphere soil multifunctionality by filtering microbial life-history strategies

Abstract

Foliar application of metal micronutrients is increasingly adopted in intensive cultivation systems, yet its potential ecological risks to rhizosphere functions remain poorly understood. Here, using the medicinal plant Panax notoginseng as a model, we conducted a gradient foliar amendment experiment with iron (Fe), zinc (Zn), and copper (Cu) to evaluate how aboveground metal inputs regulate rhizosphere soil multifunctionality (MF) through microbial life-history strategies. By integrating 16S rRNA amplicon sequencing, metagenomics, root transcriptomics, and a newly developed quantitative Yield-Acquisition-Stress tolerance (qYAS) framework, we disentangled the microbial mechanisms underlying divergent functional responses to metal amendments. Foliar Fe significantly enhanced multifunctionality, including nutrient provision and element cycling, while Cu and Zn reduced nutrient provision and element cycling, but enhanced plant pathogen abundances. These changes were closely associated with shifts in bacterial life-history strategies: Fe promoted Y-strategists characterized by efficient carbon use, streamlined genomes, and high network connectivity, whereas Cu and Zn enriched AS-strategists with larger genomes and negative associations with multifunctionality. Partial least squares path modeling (PLS-PM) further identified microbial strategies as key mediators linking foliar metal inputs, plant performance, soil properties, and multifunctionality. This study provides a trait-based microbial framework for evaluating foliar metal fertilization and guiding safer nutrient management.

Introduction

The rhizosphere, the narrow soil zone directly influenced by plant roots, plays a pivotal role in regulating plant growth, nutrient acquisition, and stress tolerance¹. Environmental inputs can profoundly reshape this microenvironment by altering its biogeochemical processes and functional capacities. For high-value crops, such as medicinal plants and vegetables, foliar micronutrient application has become an important agronomic strategy. While it is well-established that leaf-absorbed nutrients directly enhance photosynthetic efficiency, metabolic activity, and overall plant performance², their potential to drive shifts in rhizosphere health remains poorly understood. Given that foliar-applied metals can affect the rhizosphere either through deposition onto the soil or by regulating root transcriptional responses^3,4, it is essential to explore the response of rhizosphere biochemistry processes and functions to foliar fertilization.

Soil multifunctionality, defined as the concurrent performance of multiple biogeochemical and ecological processes, serves as a key indicator of rhizosphere health. Researchers have outlined that soil multifunctionality integrates element cycling functions (ECF), nutrient provision functions (NPF), plant pathogen suppression (PP), and the contributions of arbuscular mycorrhizal fungi (AMF), collectively reflecting the balance between nutrient cycling, stress regulation, and plant-microbe interactions⁵. Higher multifunctionality typically enhances nutrient turnover, increases resistance to environmental stress, and promotes faster plant growth^6,7. Because the availability of macro- and micronutrients is strongly mediated by microbial communities, shifts in nutrient inputs or stoichiometry can propagate through the soil-plant-microbe system to reshape broader ecological functions⁸. While soil-applied fertilizers are known to influence plant performance, the expanding use of foliar fertilization introduces an alternative pathway for micronutrients to reshape plant growth^9,10. Iron (Fe), zinc (Zn), and copper (Cu) are essential micronutrients that support both plant physiology and microbial metabolism. Because these elements participate in enzyme activation, electron transfer, and redox-related processes, their availability can directly affect plant growth and indirectly regulate rhizosphere multifunctionality through shifts in microbial activity and plant-microbe interactions. Recent studies have shown that foliar application of these metals can influence not only plant physiology directly, but also indirectly reshape rhizosphere physicochemical properties via canopy leaching and treatment-induced changes in plant performance^3,11. However, how foliar micronutrient inputs ultimately influence the rhizosphere microbiome and ecological multifunctionality remains largely unexplored.

Bacteria are major drivers of soil multifunctionality, regulating key biogeochemical processes involved in nutrient cycling, organic matter decomposition, and metal transformation. To better explain how bacterial communities respond to environmental change and how these responses scale up to soil functioning, trait-based ecological strategy theory has increasingly been incorporated into microbial ecology¹². Among these frameworks, the YAS (Yield - Acquisition - Stress tolerance) framework is particularly powerful. It categorizes microorganisms based on distinct resource-allocation patterns and genomic traits: Y-strategists prioritize rapid growth and high biomass yield (e.g., characterized by smaller genome size and preferences for simple substrates); A-strategists invest heavily in resource-foraging traits, such as the synthesis of extracellular enzymes to degrade complex polymers; and S-strategists allocate energy toward maintenance and defense mechanisms to survive under adverse conditions (e.g., low pH or high metal toxicity)^12–14. Researchers also recognized an intermediate AS strategy, representing taxa that simultaneously invest in nutrient acquisition and stress tolerance, enabling them to thrive under nutrient and metal stresses¹². This trait-based approach is crucial for soil ecology because it moves beyond mere taxonomic composition to directly link microbial physiological potentials to ecosystem processes. By identifying shifts in these strategies, it can be mechanistically predicted whether soil functions like nutrient mineralization will be accelerated or inhibited. Shifts in bacterial life-history strategies exert strong influences on soil multifunctionality by altering the rates and pathways of biogeochemical processes¹⁵. However, a robust approach for quantitatively classifying microbial life-history strategies is still lacking, which greatly limits our ability to explain soil functional responses to foliar fertilization from the perspective of microbial community structure and functional gene profiles.

Panax notoginseng (PN) is a highly valued medicinal herb predominantly cultivated in Wenshan Prefecture, Yunnan Province, in the Southwest of China. Foliar application of micronutrients is an important management strategy for PN. In the absence of exogenous foliar supplements, PN plants and their associated rhizosphere microbes must rely entirely on the inherent soil micronutrient pools (i.e., Cu, Fe, and Zn) to sustain their physiological functions and metabolic activities. While these native soil reserves provide a baseline supply, they are often insufficient to meet the high nutritional demands of PN during its growth or synthesis of key substances^16,17. Given the pivotal role of the rhizosphere in regulating plant growth, it is crucial to clarify how this method influences soil functions and bacterial communities^18–20. In this study, we established a gradient foliar fertilization experiment by applying different metals (Cu, Fe, and Zn) and concentrations of metals to PN leaves. By integrating 16S rRNA gene sequencing, metagenomics, root transcriptomics, and a newly developed quantitative life-history strategy framework (quantitative YAS, qYAS) based on functional gene profiles and ecological distributions, we investigated how metal inputs reshape rhizosphere soil multifunctionality and bacterial life-history strategies. Specifically, we aimed to: (i) determine the response of rhizosphere multifunctionality to foliar metal amendments; (ii) identify the dominant bacterial life-history strategies under different metal regimes; and (iii) link soil multifunctionality to microbial strategies and environmental variables. This integrative approach advances trait-based microbial ecology in the rhizosphere and provides a foundation for targeted soil and nutrient management in highly valued plant cultivation systems.

Results

Effects of foliar metal applications on rhizosphere soil multifunctionality

Potted P. notoginseng was exposed to a range of foliar-applied metal treatments (Cu, Fe, and Zn) in incubators for 160 days (Fig. 1a). We assessed ecosystem multifunctionality (EMF) across these treatments by integrating four key ecological functions: element cycling functions, nutrient provision functions, plant pathogen suppression, and arbuscular mycorrhizal fungi (Fig. 1b–f). Our results revealed that ecosystem multifunctionality, along with element cycling functions and nutrient provision functions, was significantly higher under CK and Fe treatments compared to Cu and Zn treatments. Specifically, ecosystem multifunctionality, nutrient provision functions, and element cycling functions increased with rising Fe concentration, whereas ecosystem multifunctionality, plant pathogen, and element cycling functions declined with increasing concentrations of Cu and Zn (P < 0.05). Interestingly, certain individual functional indicators, such as nutrient provision functions-related enzymes (e.g., available phosphorus, β-Glucosidase), and element cycling functions-associated available iron were more pronounced under Cu and Zn treatments (Fig. 1g). Collectively, these results indicate that Fe could partially enhance ecosystem multifunctionality, in stark contrast to the inhibitory effects of Cu and Zn.

Fig. 1. Effects of foliar metal applications on rhizosphere soil multifunctionality in Panax notoginseng.

a Experimental design. Potted Panax notoginseng was exposed to gradient metal amendments (Cu, Fe, and Zn) in foliar sprays for 180 days. Metal concentrations in treatments B, C, and D were two-, three-, and four-fold higher than in treatment A, respectively (detailed amendment amounts in Supplementary Table 1). Bar plots displaying the distribution of key rhizosphere multifunctionality metrics across treatments: b Ecological multifunctionality (EMF), c arbuscular mycorrhizal fungi (AMF) abundance, d Nutrient provision functions (NPF), e Plant pathogen suppression (PP), and f Element cycling functions (ECF). g Schematic summary diagram highlighting the differential distribution patterns of z-scored rhizosphere multifunctionality metrics under varying metal stress regimes.

The plant performance of Panax notoginseng, including plant height, total plant biomass, root biomass, and root/shoot ratio, were significantly influenced by the various element treatments (Fig. 1). Both total biomass and root biomass in the CK and Fe-amended treatments were significantly higher than those in the Cu and Zn treatments. Furthermore, the root/shoot ratio in the CK and Fe treatments (with the exception of the FeB treatment) was significantly higher than that in all Cu and Zn treatments (with the exception of the ZnA treatment) (P < 0.05) (Supplementary Fig 1).

The life-history strategy traits of soil bacterial communities

To investigate the microbiological mechanisms underlying the observed changes in ecosystem multifunctionality, we analyzed the distribution of life-history strategy genes in the rhizosphere microbial community (Fig. 2; Supplementary Table 1). Differential abundance analysis revealed that S-strategy genes were significantly enriched in Cu and Zn treatments compared to the CK and Fe treatments (P < 0.001). Conversely, the yielding strategy genes were enriched in the CK and Fe treatments (P < 0.05) (Supplementary Fig 2, 3). We further employed taxonomic and functional gene profiling of metagenomic contigs to infer microbial life-history strategies based on the relative abundance of different classes annotated to genes associated to life-history strategies (Supplementary Fig 4). At the phylum level, genes for Y and A strategies were predominantly associated with Chloroflexi and Acidobacteria, whereas S-strategy genes were linked to Gammaproteobacteria. Interestingly, Alphaproteobacteria were the major contributor to genes for both Y- and S-strategies. At the family level, the abundance of A- and S-strategy genes within Sphingomonadaceae and Caulobacteraceae (both Alphaproteobacteria) was 0.89 and 2.71-fold higher than that of Y-strategy genes. In contrast, families such as Rhizobiaceae, Bradyrhizobiaceae, and Xanthobacteraceae harbored a greater abundance of Y-strategy genes, which were 0.78 and 11.50 fold higher than A- and S-strategy genes (Supplementary Fig 5).

Fig. 2.

The relative abundance of life-history associated genes (Y: yield; A: acquisition; and S: stress-tolerance) across different foliar spray treatments and their Spearman’s correlations with multifunctionality (ECF Element cycling function, NPF Nutrient provision function, PP Plant pathogen, AMF Arbuscular mycorrhizal fungi, EMF Ecological multifunctionality). Significance levels: *P < 0.05, **P < 0.01, ***P < 0.001.

The distribution patterns of specific functional genes aligned with these trends. Genes for Y-strategy, including those involved in the pentose phosphate pathway (K00061 and K00030), tricarboxylic acid cycle (K00300), and ribosome function (K02026 and K02025), along with A-strategy genes for starch (e.g., K00844, K03434, and K01187) and cellulose degradation (K15855 and K01443), were enriched under CK and Fe treatments. These genes showed a significant positive correlation with multifunctionality (P < 0.05), consistent with the overall abundance of Y-strategy genes (Fig. 2). Conversely, S-strategy genes involved in core metabolism (e.g. K00902/K00901, K00253, and K00770) and stress/environmental adaptation (K01055 and K00901/K00902), as well as A-strategy genes related to the degradation of cellulose (K01233), hemicellulose (K22268), and lignin (K05972), were enriched in Cu and Zn treatments. These genes were significantly negatively correlated with multifunctionality (P < 0.05) (Fig. 2). Overall, these results highlighted the distinct responses of bacterial life-history strategies to Fe versus Cu/Zn amendments.

Responses of life-history strategy traits to metal amendments

Based on dimensionality reduction, affine transformation, and linear programming algorithms, we constructed a novel Yield-Acquisition-Stress tolerance (YAS)-based quantitative framework (qYAS) using the metagenomic and amplicon sequences from the rhizosphere. According to this framework, ASVs were classified into A (4516), AS (2203), S (150), and Y (429) strategies (Fig. 3a, b). Numbers and relative abundances of Y, A and AS strategists were higher than S strategists (Fig. 3c). Procrustes analysis was further conducted to explore the contribution of subcommunities to the variations of microbial community, showing that the m² of AS-strategy ASVs was the lowest, followed by Y and A-strategy (Fig. 3d). Community composition revealed that Alphaproteobacteria and Gammaproteobacteria were abundant in all the four subcommunities (Supplementary Fig 6). The Y-strategy subgroup demonstrated dominance of Bacteroidia (27.43%), Polyangia (7.85%), and Vicinamibacteria (6.59%), while the abundance of Alphaproteobacteria (34.27%), Gammaproteobacteria (27.73%), and Nitrospiria (1.67%) were larger in the AS-strategy subgroup (Supplementary Fig 6).

Fig. 3. Variation in different life-history strategists according to qYAS.

a Classification of life-history strategies using linear programming. a. Positioning of Y, A, and S strategy-related genes and ASVs within the RDA coordinate system derived from metal amendment data. b Classification outcomes of microbial life-history strategies. c Abundance and distribution of microbial taxa assigned to different life-history strategies across the datasets. d Procrustes permutation test (m² statistic, P < 0.001) demonstrating significant concordance between the total community and life-history subcommunities. e-g. Trends in strategy abundances (Y, A, AS, and S) under increasing Cu (e), Fe (f), and Zn (g) treatments. Duncan’s test results are shown with letters denoting significant differences. Quadratic and linear regression fitting comparisons for response variables under different metal treatments are presented in Table S3. h Relationships between the relative abundance of life-history strategies and multifunctionality (ECF Element cycling function, NPF Nutrient provision function, PP Plant pathogen, AMF Arbuscular mycorrhizal fungi, EMF Ecological multifunctionality) using Linear regression across treatment groups. Trend lines, R, and p values are provided for each pairwise comparison.

Metal amendment significantly altered both the composition and life-history strategies of bacterial communities. PCoA and PERMANOVA analyses revealed that communities in the Cu and Zn treatments clustered distinctly and differed significantly from those in the CK and Fe treatments (P < 0.001) (Supplementary Fig 7). In view of subcommunities with distinct life-history strategies, the Y-strategists exhibited higher Shannon diversity and ACE index in the CK and Fe treatments than those in the Cu and Zn treatments, while those of AS-strategy microbes were markedly lower (P < 0.05) (Supplementary Fig 8). Furthermore, PCoA ordination demonstrated significant separation between CK/Fe treatments and Cu/Zn treatments along the 1st axis for the A-, Y-, and AS-subcommunities (Supplementary Fig 9a, b). Community composition and indicator species-based analysis revealed that 92.35% of ASVs classified into the Y strategy were enriched in Fe treatments, while 82.52% and 72.22% of the AS-strategy ASVs were enriched in Cu and Zn treatments, separately (Supplementary Fig 10).

The dynamic changes of traits with metal amendment gradient exhibited large differences across Cu, Fe, and Zn treatments (Fig. 2e–g). For Cu and Zn, a strong positive correlation was observed between amendment amount and the abundance of AS-strategy sub-community, peaking at moderate concentrations before slightly declining at the highest dosage, on the contrast with the negative correlations of A and Y strategies (P < 0.001) (Figs. 2e, g; Supplementary Table 2). Fe amendment amount showed a robust positive correlation with the abundance of Y-strategy ASVs (R = 0.83, P < 0.001), but negative relationships with AS and A strategy (P < 0.05) (Fig. 2f; Supplementary Table 2). Linear regression analysis demonstrated distinct relationships between life-history strategies and ecosystem functions (Fig. 2h). Specifically, the A and Y groups exhibited significant positive correlations with overall EMF and specific individual functions (P < 0.001). Conversely, the AS and S groups were significantly negatively correlated with these metrics (P < 0.001). Furthermore, the AS and Y strategies showed markedly higher |R| than the A and S strategies, suggesting a more pronounced link to ecosystem multifunctionality.

Coexistence patterns and network stability of subcommunities under metal amendments

To uncover the potential coexistence patterns and ecological interactions among life-history strategy-defined subcommunities, we constructed a co-occurrence network across all samples, comprising 792 nodes and 4932 edges (Fig. 4a, b). Compared to AS and S strategy subcommunities, the A and Y strategy subcommunities exhibited significantly higher degree and eigenvector centrality, but lower betweenness centrality (Fig. 4c, d, and Supplementary Fig. 11). To investigate potentially coexistence patterns among strategies, the positive and negative associations of ASVs were analyzed. All intra-strategy associations were exclusively positive, and links between A and Y subcommunities, as well as between AS and S subcommunities, were consistently positive (Fig. 4e). In contrast, the AS strategy subcommunity exhibited predominantly negative correlations with both A and Y strategy subcommunities.

Fig. 4. Topological characteristics of the rhizosphere microbial co-occurrence network and core taxa identification.

a, b Microbial co-occurrence networks constructed from 16S amplicon datasets. Nodes are colored by microbial life-history strategy classification (a) and taxonomic class (b), respectively. Edges represent significant correlations (P < 0.05, r > |0.6|). Degree centrality (c) and eigenvector centrality (d) across life-history strategists. Significant differences (Duncan’s test; P < 0.05) are denoted by letters. e. Numbers of intra- and inter-community correlations for ASVs within each subcommunity. f Stability indices calculated from positive and negative cohesion across treatments. Bars with distinct letters indicate significant differences (Duncan’s test; P < 0.05). g Ratio of positive to negative connections for individual taxa. h Classification of ASVs based on within-module connectivity (Zi) and among-module connectivity (Pi). Categories: Module hubs (Zi > 2.5, Pi ≤ 0.62), Connectors (Pi > 0.62, Zi ≤ 2.5), Network hubs (Zi > 2.5, Pi > 0.62), Peripherals (Zi ≤ 2.5, Pi ≤ 0.62). i Proportions of module hubs and connectors (from h) classified by microbial classes. j Positive and negative cohesion values attributable to module hubs and connectors. Statistical significance was determined using Duncan’s test, with letters indicating significant differences.

Network stability varied under different metal treatments (Fig. 4e). Compared to CK and Fe treatments, the networks in Cu and Zn treatments showed a higher stability indice, while the indice declined with increasing Zn amendment in Zn treatments (Fig. 4f). Notably, Fe-treatment maintained network stability similar to CK (Fig. 4f). The divergence in connectivity ratios (negative/positive links) among strategy-defined subcommunities reflects fundamental trade-offs in resource allocation and stress tolerance. Our results revealed significant differences in microbial interaction patterns among strategy subcommunities, with AS and Y subcommunities showing higher ratios than A and S subcommunities (Fig. 4g). Zi-Pi analysis of topological roles identified that the Y subcommunity contained the highest proportion of module hubs (9 hubs, 56.25%), despite accounting for only 20.20% of total nodes (Fig. 4h, i). Only three connectors were observed, assigned to AS (2) and A (1) strategies. Distinct positive and negative cohesion patterns were also identified, showing that module hubs affiliated with the Y strategy exhibited higher positive cohesion compared to other hubs. Additionally, most module hubs were phylogenetically conserved, clustering mainly within Alphaproteobacteria (9), Polyangia (3), and Anaerolineae (2) (Fig. 4j).

Average genome size of microbial community

Genome size is a critical trait intrinsically linked to the metabolic strategies of microorganisms. In the present study, average genome size (AGS) varied significantly across different treatments (Fig. 5a). It was lower in the CK and Fe treatments compared to the Cu and Zn treatments, with the ZnC and ZnD treatments exhibiting significantly larger genomes than the CK and Fe treatments (P < 0.05). To elucidate the genetic underpinnings of this variation, we utilized a random forest model to identify key life-history strategy genes contributing to average genome size (Fig. 5b). The model identified 9 A-strategy, 5 S-strategy, and 3 Y-strategy genes as significant predictors (P < 0.05). Further correlation analysis showed that five A-strategy genes and all S-strategy genes were positively associated with average genome size (Fig. 5c). In contrast, the Y-strategy gene with the highest importance score was negatively associated with average genome size (P < 0.05). Subsequently, linear regression analyses revealed distinct relationships between average genome size and subcommunity abundances (Fig. 5d). The abundances of A- and Y-strategy sub-communities were negatively correlated with average genome size, whereas the AS-strategy sub-community abundance was positively correlated (P < 0.01). Conversely, multifunctionality indices, including element cycling functions, nutrient provision functions, plant pathogen, and overall ecosystem multifunctionality, were found to be significantly and positively correlated with average genome size (P < 0.05).

Fig. 5. Average genome size calculated using metagenomic data.

a The average genome size (Mb) for different experimental groups. Different lowercase letters (a–c) above bars indicate statistically significant differences between groups (p < 0.05, one-way ANOVA with Duncan test). b Random Forest regression quantifying the contribution of life-history strategy genes (Y, A, and S) to average genome size. The model’s goodness of fit is shown by R² = 0.591 (** indicates p < 0.01). Red and blue color represent positive and negative correlations with average genome size, respectively (P < 0.05). Relationships between average genome size and the relative abundance of life-history strategies (Y, A, AS, and S) (c) and multifunctionality (ECF Element cycling function, NPF Nutrient provision function, PP Plant pathogen, AMF Arbuscular mycorrhizal fungi, EMF Ecological multifunctionality) (d) using Linear regression across treatment groups. Trend lines, R, and p values are provided for each pairwise comparison.

Environmental and host factors driving the variations in subcommunities and multifunctionality

Soil physicochemical properties exhibited significant alterations in response to metal amendment levels (Supplementary Fig 12; Supplementary Table 3). Specifically, Fe amendment led to marked increases in SOC content and pH, whereas total Fe, Cu, and Zn concentrations decreased significantly (P < 0.01). These trends were diametrically opposed to those observed under Cu and Zn treatments, where metal accumulation correlated positively with dosage. Given the tight links between plant root activities and rhizosphere microbial communities, the gene expression profiles of P. notoginseng roots were quantified by RNA-Seq, which was used as host factors. PCoA analysis based on the root transcriptome of P. notoginseng confirmed significant inter-treatment differences in functional gene profiles (PERMANOVA, P < 0.001) (Supplementary Fig 13). WGCNA was then performed and 21 co-expression modules were identified, which provide basis for exploring linkages between host factors and rhizosphere microbial communities (Supplementary Figs. 14–16). Module distribution patterns and correlation analyses revealed that M3 abundance increased significantly with elevated Fe amendment levels (P < 0.001), whereas M2 and M12 expression exhibited dose-dependent upregulation under Cu and Zn treatments, respectively (P < 0.001) (Supplementary Figs. 17, 18).

Mantel tests revealed significant correlations between microbial community composition and various environmental factors across the four strategies (Fig. 6a). Among the physicochemical variables, soil pH, SOC, total Fe, and plant performance parameters consistently exhibited significant correlations with all four subcommunities (r > 0.2, P < 0.01), followed by EC, total Zn, and total Cu. Correlation analysis further showed that the ACE, Shannon diversity, and abundances of A- and Y-strategy taxa were significantly and positively correlated with pH, SOC, and plant performance parameters, but negatively correlated with total concentrations of three metals and EC, while these relationships for AS- and S-strategists were reversed (Supplementary Fig 19). Several WGCNA modules were also correlated with the variations in microbial subcommunities (P < 0.05) (Fig. 6a). Notably, M8 displayed significant correlations with the four subcommunities (P < 0.01), while other modules (M4, M5, M6, and M12) showed weaker but still significant correlations with subcommunities (p < 0.05), suggesting potential links between plant gene expression patterns and microbial community structure. Correlation analysis also identified transcriptional modules dominated by fatty acid synthesis pathways (e.g., M3, M4, and M8) positively associated with A/Y subcommunities (P < 0.05). However, M2, M12, and M16 were associated with S/AS subcommunities (P < 0.05), which were enriched in metal-chelating pathways such as phenylpropanoid biosynthesis, cysteine and methionine metabolism, and peroxisome activity (Fig. 6b, Supplementary Figs. 20,21). Modules positively correlated with A/Y strategies were enriched with numerous plant growth - related genes (LACS1, HXK2, and TUBB1) (P < 0.05), all of which exhibited higher abundances under the CK and Fe treatments (Supplementary Figs. 20,22). These results pointed out the key soil physiochemical and host factors dominating the variation of microbial life-history strategists among treatments.

Fig. 6. Effects of soil properties and root transcriptomic modules on variations in microbial subcommunities with distinct life-history strategies.

a Mantel test correlations revealing associations among factors and rhizosphere microbial subcommunities. Lines indicate Mantel associations of subcommunities with factors, while the heatmap represents relationships between factors. b Enrichment analysis of KEGG Level 3 pathways within root transcriptional modules associated with microbial subcommunities (P < 0.05). c Spearman’s correlations between environmental factors, host modules and multifunctionality (ECF Element cycling function, NPF Nutrient provision function, PP Plant pathogen, AMF Arbuscular mycorrhizal fungi, EMF Ecological multifunctionality). d Spearman’s correlations between phyla and multifunctionality Significance levels: *P < 0.05, **P < 0.01, ***P < 0.001.

To identify the key drivers of ecosystem multifunctionality, we performed a comprehensive Spearman’s correlation analysis (Fig. 6c). The results indicated that multifunctionality was significantly associated with a broad range of biotic and physicochemical parameters. However, its link to the root transcriptome was more selective, positively associated with M8, while positively associated with M12 and M16. We then delved into the taxonomic drivers at the class level, where Alphaproteobacteria and Gammaproteobacteria were significantly and negatively correlated with multifunctionality (P < 0.001) (Fig. 6d). Conversely, Bacteroidia and Polyangia were significant positive correlates (P < 0.001).

Regulatory routes of metal amendment on ecosystem multifunctionality mediated by microbial life-history strategists

To elucidate the microbial mechanisms by which metal amendments affect rhizosphere soil multifunctionality, we constructed PLS-PM incorporating microbial life-history strategists, soil properties, and host transcriptional profiles (Fig. 7a, b). The results revealed that metal amendments exerted significantly indirect effects on ecosystem multifunctionality. Specifically, Fe amendment positively drives ecosystem multifunctionality, as mediated by soil properties and the abundance of life-history strategists. Cu/Zn amendment negatively regulated ecosystem multifunctionality, which was mainly mediated by plant performance, followed by soil factors and the abundance of life-history strategists. A hierarchical analysis of influencing factors revealed that environmental parameters, including SOC and pH, exerted the strongest influence on ecosystem multifunctionality. Soil properties were the next most significant factor, followed by metal amendments and plant growth. Critically, the abundance of Y and AS strategy subcommunities were identified as a key mediator in these relationships (Supplementary Fig. 23). These findings underscore the strategy-specific microbial mediation effects on ecosystem multifunctionality and host health under metal stress and highlight the divergent regulatory routes of Fe versus Cu/Zn amendment.

Fig. 7. Partial Least Squares Path Modeling (PLS-PM) analysis.

a The validated PLS-PM illustrating the causal relationships among metal amendments, soil properties, host modules, the relative abundance of life-history strategies, and multifunctionality. Single-headed arrows indicate hypothesized directional effects, with standardized path coefficients displayed alongside (red: positive; blue: negative). The R² values indicate the proportion of variance explained for each latent variable. The overall model predictive power and fit are strong (Goodness-of-Fit, GOF = 0.835). b The direct, indirect, and total effects (standardized path coefficients) in (a).

Discussion

Metal addition caused variation of soil physicochemical conditions and significant alterations in plant performance and root transcriptional activity, reflecting the high environmental sensitivity of Panax notoginseng to foliar Fe, Cu, and Zn exposure. The contrasting effects of foliar metal application could be largely attributed to their fundamentally different physiological roles, redox activities, and absorption behaviors²¹. Fe acts as an essential cofactor in photosynthetic electron transport, chlorophyll biosynthesis, and the function of Fe-S proteins^16,22. Thus, supplying Fe directly to young expanding leaves rapidly alleviates latent Fe limitation and promotes carbohydrate transport to the roots². Consistently, our transcriptomic data revealed the enrichment of energy metabolism and root-development genes under Fe treatment, supporting that foliar Fe could improve plant vigor and root functionality. The strengthened root system might further facilitate the release of carbon, leading to increased SOC accumulation, as well as reduced mobility of metal in the rhizosphere through complexation and adsorption processes²³. In contrast, foliar Cu and Zn exhibit poor phloem mobility and therefore accumulate on the leaf surface, where elevated concentrations readily induce toxicity^24,25. The stress signals were reflected in the pronounced upregulation of stress-responsive and detoxification-related genes observed in the Cu and Zn treatments. The oxidative and photochemical damage collectively suppress photosynthesis and limit carbon allocation to the root system, thereby weakening the plant’s ability to shape rhizosphere environment, which is reflected by the decreased plant performance parameters, especially the root biomass and root/shoot ratio²⁶. Root-derived C is a major source of SOC, and direct experimental evidence indicates that C released during root growth could make a substantial contribution to SOC accumulation²⁷. Moreover, because they are inefficiently absorbed via foliar pathways, a considerable proportion returns to the soil surface with runoffs or natural falling, where they increase soluble metal pools. Soil pH is significantly positively correlated with plant performance and SOC, whereas it is significantly negatively correlated with metal content, which indicates that the formation of soil pH is a comprehensive result of complex environmental changes. For example, a decline in SOC may further reduce soil buffering capacity as an important contributor to cation exchange and acid buffering²⁸, while metal stress could lead to acidification by increasing the release of low-molecular-weight organic acids²⁹. The combination of plant physiological stress and rhizosphere contamination creates a positive feedback loop that further diminishes root health, as evidenced by the lower SOC and pH observed under Cu and Zn exposure. Overall, these findings demonstrate that Fe input initiates a beneficial plant-soil-microbe cascade, whereas Cu and Zn inputs trigger a toxicity-driven feedback loop that undermines both plant growth and rhizosphere ecological stability.

Although foliar application of metals could greatly alter soil properties, the impacts of these changes on rhizosphere microbiomes remain underexplored^30,31. Generally, soil abiotic factors are often more dominant deterministic filters in shaping rhizosphere microbial communities compared to plant signaling pathways^32,33. AS strategists combining nutrient acquisition with stress tolerance (e.g., antioxidant defense, metal sequestration) confer ecological advantages in metal-enriched soils due to their high functional gene diversity^34–36. In our results, the increase of AS strategist abundance in Cu and Zn amendments could be caused by the lack of SOM and the pressure of acidification and metal enrichment. In contrast, the substantial accumulation of SOM in the Fe treatment provided abundant nutrients for microbial growth, while the near-neutral pH and lower metal concentrations reduced environmental stress. These favorable conditions allowed microbes to allocate more resources toward growth-related functions, thereby driving the enrichment of Y-strategy taxa under Fe amendment. Furthermore, these life-history shifts were accompanied by marked differences in AGS among treatments. The larger AGS observed in the Cu and Zn treatments likely reflect an adaptive requirement for expanded genetic repertoires enabling stress mitigation, including genes involved in metal resistance, detoxification, membrane repair, and environmental sensing. Such expansion of functional capacity is widely recognized as a hallmark of stress-tolerant and facultatively adaptive microbes inhabiting chemically challenging environments^37,38. This mechanistic interpretation is consistent with our finding that average genome size was positively correlated with the abundance of AS strategists, which rely on broad metabolic and protective capabilities to cope with simultaneous nutrient limitation and metal stress. In contrast, average genome size was negatively associated with Y-strategy dominance, as Y-strategists typically possess streamlined genomes optimized for rapid growth and efficient resource utilization under favorable conditions^12,39. Therefore, the contrasting average genome size patterns across metal treatments further support a metal-induced shift from growth-oriented, genome-streamlined taxa under Fe amendment to stress-tolerant, genomically expanded taxa under Cu and Zn enrichment.

Our results further revealed clear lineage-specific shifts in microbial life-history strategies in response to metal amendments. Gammaproteobacteria were classified into AS-strategists, enriched under Cu and Zn treatments, predominantly as AS-strategy taxa, consistent with their ecological attributes as fast responders with high metabolic plasticity. Gammaproteobacteria are known to dominate in disturbed or chemically stressful environments because of their robust stress-defense systems and versatile carbon metabolism⁴⁰. In contrast, Y-strategies Verrucomicrobiae and Polyangia were enriched under Fe addition treatments. Verrucomicrobiae are widely recognized for their oligotrophic lifestyle, with specialized capacity to degrade high-affinity nutrients such as root-derived polysaccharides.^41,42. Meanwhile, Polyangia always shows cooperative metabolism and efficient conversion of organic substrates into microbial biomass⁴³. Interestingly, Alphaproteobacteria (most belonging to Rhizobiales) were distributed across both AS- and Y-strategy groups, implying that this lineage encompasses a wide functional spectrum capable of responding flexibly to contrasting environmental pressures. A plausible explanation for this pattern is that many Rhizobiales harbor lineage-specific transposable elements that facilitate rapid genomic innovation, which enabled them to simultaneously acquire nitrogen-fixation traits and stress-resistance functions, and allowed flexible shifts between different life-history strategies under fluctuating environments^44,45.

Microbial life-history strategies could play key roles in soil ecosystems, shaping biogeochemical processes across environmental gradients. The observed enhancement of nutrient provision functions under Fe amendment is likely attributable to the stimulation of Y-strategists, which could promote turnover of nutrients through rapid growth and metabolic versatility¹². In contrast, Cu and Zn amendments favored microbial traits aligned with resource acquisition and stress tolerance, thus leading to an increase in available metal contents and plant pathogen abundances. The positive relationships between AS strategies and available metal contents align with other studies, which were identified to contribute to biogeochemical cycling through specialized detoxification and redox processes^46,47. Soil acidification increases metal solubility in Cu and Zn treatments, particularly iron content⁴⁸. Moreover, the higher Fe bioavailability could result in the accumulation of plant pathogens by enhancing fungal iron acquisition through high-affinity siderophore systems, which has been identified to benefit pathogenicity, oxidative stress tolerance, and host colonization of pathogens^49,50. This supports the pivotal role of life history strategies in interpreting soil multifunctionality in the rhizosphere⁵¹.

By providing a quantitative framework (qYAS), This study demonstrates that foliar micronutrient fertilization could exert contrasting ecological effects on rhizosphere soil multifunctionality by reshaping microbial life-history strategies. Foliar Fe application partially promoted favorable soil conditions, enriched Y-strategy taxa with efficient metabolism and strong network connectivity, and enhanced multifunctionality, thereby contributing to rhizosphere functional resilience. In contrast, Cu and Zn inputs induced plant performance, soil acidification, and metal accumulation, favored stress-tolerant AS-strategists, and constrained ecosystem functions, indicating elevated ecological risks. By quantitatively linking metal exposure, microbial strategies, and ecosystem multifunctionality, we clarify the mechanistic basis for assessing the environmental consequences of foliar metal fertilization and support the development of safer and more sustainable nutrient management practices. However, several limitations of this study should be acknowledged. Firstly, this study was based on a 160-day controlled pot experiment with Panax notoginseng, and thus the observed responses may not fully represent long-term dynamics under heterogeneous field conditions. Moreover, ecosystem multifunctionality was assessed using four representative functional categories, which may not capture the full breadth of rhizosphere ecological functions. Although qYAS, co-occurrence network analysis, WGCNA, and PLS-PM consistently supported the proposed strategy-mediated mechanisms, these approaches remain largely inference-based and require further validation through direct functional experiments.

Methods

Experimental sites and collection of samples

The study was conducted in Wenshan Prefecture, Yunnan Province, southeast China, representing the most suitable growing area of Panax notoginseng (PQ) (the so-called Daodi area). The soil used in this experiment was collected from the topsoil layer (0-20 cm) of a P. notoginseng cultivation field in Panlong Town, Yanshan County, Wenshan Prefecture, before pot establishment. After collection, visible stones, plant residues, and other debris were removed manually, and the soil was thoroughly homogenized and sieved through a 2 mm sieve prior to use. The experimental soil is classified as Udic Ferrosol (FAO classification) and exhibited acidic properties with a pH of 5.59. Soil analysis revealed the following physicochemical characteristics: organic matter content of 2.66 g/kg, alkali-hydrolyzable nitrogen of 21.39 mg/kg, available phosphorus of 3 mg/kg, and available potassium of 75.02 mg/kg. A pot experiment with 13 treatments and 5 replicates per treatment was established using uniform-sized pots (60 cm × 60 cm, 0.36 m²) containing 12 kg of homogenized soil. Each pot was sown with 144 surface-sterilized PQ seeds arranged in a 12 × 12 grid. The control group (CK) received conventional fertilization: urea (N: 3.6 g/pot), calcium superphosphate (P₂O₅: 4.8 g/pot), and potassium sulfate (K₂O: 7.2 g/pot). Three micronutrient treatments were superimposed on the base fertilization: iron (FeSO₄·7H₂O), zinc (ZnSO₄), and copper (CuSO₄), each applied at four concentration gradients (Supplementary Table 1). Foliar spraying was conducted for four consecutive cycles (May 11, July 2, July 28, September 6) using 200 mL/pot per application, and we conducted it gently by using fine mist sprayers to ensure little direct runoff into the soil. We performed a small pot trial applying Cu, Fe, and Zn to PN leaves identically to the main experiment. Soil metal analysis 72 hours after application showed no significant difference from the CK group, which definitively excludes direct soil contamination from foliar runoff (Supplementary Fig 24). All agronomic practices were standardized across treatments to isolate micronutrient effects. Soil samples used for subsequent analyses were collected at the harvest stage. In this study, all soil analyses were conducted using rhizosphere soil rather than bulk soil. Briefly, whole plants were carefully removed from each pot, and loosely adhering soil was gently shaken off. The soil that remained tightly attached to the root surface was then brushed off and collected as rhizosphere soil. For each pot, rhizosphere soil from 10 plants was pooled to form one composite sample. After sampling, each soil sample was divided into two subsamples. One fresh subsample was immediately stored at −80°C for DNA extraction and microbial community analysis, whereas the other subsample was air-dried at room temperature, passed through a sieve, and used for soil physicochemical analyses. Standard test methods were adopted to measure soil pH, dissolved organic carbon (DOC), available nitrogen, available phosphorus (AP), available potassium (AK), available iron (AFe), available manganese (AMn), available copper (ACu), and available zinc (AZn), as previously described⁴. The activities of enzymes involved in carbon, nitrogen, and phosphorus cycling were determined using fluorometric microplate assays based on previously established protocols⁴. The target enzymes included β-glucosidase (BGL), cellobiohydrolase (CBH), β-xylosidase (BXL), and α-glucosidase (AGL) for carbon acquisition, N-acetyl-β-D-glucosaminidase (NAG) for nitrogen acquisition, and phosphatase (Pho) for phosphorus acquisition.

Soil DNA extractions, amplicon, and metagenomic sequencing

Genomic DNA was extracted from each soil microcosm using the E.Z.N.A.® soil DNA Kit (Omega Bio-tek, Norcross, GA, U.S.) according to the manufacturer’s instructions. DNA quality and purity were assessed using a NanoDrop 2000 UV-vis spectrophotometer (Thermo Scientific, Wilmington, USA). The primer pairs 338F/806R were used to amplify the bacterial 16S rRNA V3-V4 region⁵². Fungal-specific primers ITS1F-ITS2R were used for fungal DNA⁵³. The amplicons were subjected to sequencing on an Illumina MiSeq platform (Illumina, San Diego, USA) in paired-end mode (Majorbio Bio-Pharm Technology Co. Ltd., Shanghai, China). The same DNA was used for metagenomic library preparation as in the previous study. Briefly, a total of 65 DNA libraries were constructed using a NEXTFLEX Rapid DNA-Seq (Bio Scientific, Austin, TX, USA). The fragment lengths of DNA libraries ranged between 300-500 bp, which was determined with the Agilent 2100 High Sensitivity DNA Assay on an Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA). The size-selected libraries were sequenced on an Illumina HiSeq2000 platform in a paired-end mode (2 × 150 bp).

Amplicon and metagenomic data analyses

The raw reads were analyzed using the QIIME2 pipeline⁵⁴. Initially, the raw data were demultiplexed and quality-filtered by DADA2 with the following criteria: (i) the 250 bp reads were truncated at any site having an average quality score of < 20 over a 50 bp sliding window; in addition to the truncated reads shorter than 50 bp, reads containing ambiguous characters were discarded; (ii) overlapping sequences longer than 10 bp were assembled according to their overlap. Then, reads were clustered into amplicon sequence variants (ASVs), and chimeric sequences were identified and removed using DADA2, with the default quality score (truncQ = 2)⁵⁵. A representative sequence of each ASV was analyzed by RDP Classifier version 2.2 against the 16S rRNA database (Silva version 138), with a confidence threshold of 0.7⁵⁶. The relative abundances of plant pathogen suppression (PP) and arbuscular mycorrhizal fungi (AMF) of fungal ASVs were determined using the FUNGuild tool⁵⁷.

Trimmomatic was employed to remove adapters, ambiguous reads (“N” larger than 5%), and low-quality reads (quality score < 20)⁵⁸. Read lengths less than 100 bp were discarded. MEGAHIT was used to assemble filtered reads into contigs with k-mer parameters ranging from 27 ~ 141⁵⁹. QUAST was used to assess the quality of contigs and GC contents⁶⁰. The proteins of contigs were predicted using Prokka (version 1.14.6)⁶¹, then the proteins were searched against the NCBI-nr database with DIAMOND in blastp mode⁶², and annotated using the Kyoto Encyclopedia of Genes and Genomes (KEGG) database in the MEGAN6 Ultimate Edition^63,64. The average genome size (AGS) of each metagenome was estimated from metagenomic reads using a workflow described in a previous study⁶⁵.

RNA-seq and transcript assembly

Roots of PQ corresponding to rhizosphere soil samples were collected for RNA-seq analysis. The entire root was dug out, washed, and ground into powder using liquid nitrogen. Total plant RNA was extracted from bulk root using the Plant RNA Isolation Mini Kit (BioTeke, Beijing, China). The RNA-Seq library was prepared according to the Illumina instructions. Sequencing was performed on the Illumina HiSeq platform. After quality filtering, the FASTQ data were aligned to the reference genome of P. notoginseng using HiSAT2^66,67. The abundance of transcripts was counted using featureCounts⁶⁸. The transcript abundance was normalized as fragments per kilobase per transcript per million mapped reads (FPKM values).

Construction of the life-history trait model

To quantify microbial life-history strategies, we constructed an ecological classification model based on the YAS theoretical framework. Microbial taxa were categorized into Y-, A-, S-, and AS-strategy groups through redundancy analysis (RDA) integrating strategy-associated genes from metagenomes and microbial community profiles from 16S amplicon sequencing (Fig. 1a), which was conducted in the vegan package (v2.6-2) in R (v4.3.1)⁶⁹. This process involved sequential coordinate transformations and linear programming optimization, with detailed procedures as follows (Fig. 1b):

Quantification of life-history strategy indicators: Genes associated with specific microbial survival strategies (Y, A, and S strategies) annotated to the KEGG database were identified and aggregated in MEGAN6 (Supplementary Table 4). The relative abundance of genes corresponding to each strategy was summed to derive composite indicators, representing the functional potential of microbial communities for each life-history strategy.

Community ordination and geometric framework construction: Redundancy Analysis (RDA) was performed using ASV abundance data to establish a multivariate ordination space reflecting community structural variation. Life-history strategy indicators (Y, A, S) were incorporated as environmental variables and projected onto the RDA ordination. The terminal points of these indicator vectors in the RDA biplot were designated as the vertices of a triangular framework (YAS triangle), with the geometric proportions of its edges determined by their Euclidean relationships. Concurrently, the coordinates of all microbial species within the ordination space were extracted. This step established preliminary spatial relationships between microbial taxa and the life-history strategy framework.

Coordinate standardization via affine transformation: To address geometric irregularities in the original triangular framework, an affine transformation model was implemented to standardize coordinates. A least squares optimization approach was applied to solve transformation parameters (translation, rotation, scaling, and shearing) that minimized residuals between original coordinates (original coords) and target coordinates (std. coords). This transformation mapped all species coordinates and triangle vertices into a standardized equilateral triangle space, enabling quantitative strategy classification.

Geometric classification of survival strategies: Based on ecological classification principles adapted from methodological standards in a previous study¹², the standardized triangle was subdivided into four strategy zones. Two mutually perpendicular bisectors were constructed from the triangle centroid, maintaining equal Euclidean distances to the Y vertex along both the A-Y and S-Y edges. These bisectors partitioned microbial taxa into four functional groups: Y, A, S, and AS strategies. Spatial membership of microbial taxa within strategy zones was determined using the sf package in R (v4.3.1)⁷⁰.

Statistical analyses

Alpha diversity indices, including Shannon and ACE, were calculated using the vegan package in R. Principal Coordinate Analysis (PCoA) based on Bray-Curtis dissimilarity was performed using the ape package to visualize differences in microbial community structure among treatments⁷¹. Permutational Multivariate Analysis of Variance (PERMANOVA) was conducted using the vegan package to statistically test for significant overall differences in community composition between treatment groups. Significant differences in alpha diversity indices and specific microbial taxon abundances among treatments were identified using Duncan’s multiple range test (P < 0.05) implemented in the ‘agricolae’ package⁷². Bar plots were generated using the ‘ggplot2’ package⁷³. Indicator species analysis was performed using the ‘indicspecies’ package in R to identify microbial taxa significantly associated with specific environmental conditions or treatment groups⁷⁴. Statistical significance of indicator species was assessed through permutation tests with p-values adjusted for multiple testing using false discovery rate (FDR) correction. Indicator species were selected based on an indicator value (IndVal) > 0.7 and adjusted p-value < 0.05.

Weighted Gene Co-expression Network Analysis (WGCNA) was performed on root transcriptome data using the WGCNA package in R⁷⁵. Modules of co-expressed genes were identified using a dynamic tree-cutting algorithm. The soft threshold power for network construction was determined based on the scale-free topology criterion (R² = 0.8). Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses were conducted for genes within each significant module to identify overrepresented pathways.

Pearson correlation coefficients were calculated to assess relationships between microbial factors and surrounding factors (soil properties and host transcriptome modules). The Mantel test was performed using the vegan package to evaluate the correlation between the overall microbial community structure (Bray-Curtis dissimilarity) and the Euclidean distance matrix of soil environmental variables. Heatmaps depicting correlations between microbial taxa and environmental factors were generated using the pheatmap package⁷⁶. Relationships depicting the flow of ecological strategy groups, root transcriptome modules and KEGG pathways were visualized as Sankey diagrams using the ggalluvial and networkD3 packages⁷⁷.

Microbial co-occurrence networks were inferred based on Spearman correlation coefficients using the Hmisc and igraph packages in R⁷⁸. Microbial co-occurrence networks were visualized using Gephi software (v0.10.1). Topological properties of each node were calculated, including degree, eigenvector centrality, closeness centrality, clustering coefficient, and betweenness centrality⁷⁹. Keystone taxa were identified based on Zi-Pi analysis, which quantifies each node’s topological role in the microbial co-occurrence network^80,81. Here, two parameters were calculated: within-module connectivity (Zi) and among-module connectivity (Pi) using the igraph package in R. Nodes were classified as peripherals (Zi < 2.5, Pi < 0.62), connectors (Zi < 2.5, Pi ≥ 0.62), module hubs (Zi ≥ 2.5, Pi < 0.62), and network hubs (Zi ≥ 2.5, Pi ≥ 0.62). Keystone taxa were defined as connectors, module hubs, and network hubs.

Following the approach described by a previous study⁴, we categorized soil multifunctionality into 4 distinct components: element cycling functions (ECF), nutrient provision functions (NPF), plant pathogen suppression (PP), and arbuscular mycorrhizal fungi (AMF). The values of standardized enzyme activities, DOC, DON, and AP were used to represent NPF. Because soils in Wenshan are typically acidic, increases in bioavailable metals could further exacerbate soil acidification; therefore, in this study, the negative values of standardized bioavailable metal concentrations were used to represent the ECF.

A Partial Least Squares Path Modeling (PLS-PM) was constructed to explore direct and indirect pathways among soil properties, microbial strategies, and plant responses using the plspm package⁸².

Author contributions

Ye Liu: Conceptualization, Methodology, Investigation, Formal analysis, Writing original draft, Statistical methodology, qYAS framework construction, Visualization; Gao Xiong: Investigation, Data curation, Validation; Lifang Gao: Investigation, Chemical analysis; Ying Li: Methodology, Bioinformatics analysis; Xi Zhou: Data interpretation; Hui Yao: Funding acquisition, Methodology; Guangfei Wei: Investigation, Visualization; Miyi Yang: Sample preparation, Laboratory technical support; Yulong Yin: Metagenomic analysis; Jingjing Peng: Literature review, Writing review & editing; Linlin Dong: Supervision, Funding acquisition, Writing review & editing; Guozhuang Zhang: Conceptualization, Supervision, Funding acquisition, Writing review & editing.

Data availability

The raw shotgun metagenome sequencing data used in this study have been deposited in the NGSA BioProject database with the accession number PRJCA045557.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41522-026-01071-y.