Community Structure and Soil Environmental Drivers of Rhizosphere and Root Endophytic Microbiota of Polygonum divaricatum in a Temperate Grassland

Simple Summary

Grasslands are vital natural ecosystems that rely on close cooperation between plants and the microscopic life living in their roots and the surrounding soil. This study focused on a common grassland plant, Polygonum divaricatum, in the Hulunbuir Grassland to understand how different plant parts and soil conditions shape these hidden communities of bacteria and fungi. We found that plant roots act like a natural filter, allowing only certain microorganisms to live inside them, while the surrounding soil supports a different and more diverse group influenced by factors such as soil acidity and nutrient levels. Microorganisms in the soil mainly help break down organic matter, whereas those inside the roots are specialized in helping the plant absorb and transport nutrients. By revealing how these unseen partners respond to their environment, this study provides important knowledge for understanding how grasslands function. This information supports better land management and helps protect these ecosystems so they remain stable and resilient in the face of environmental change.

Abstract

Understanding the ecological drivers of plant-associated microbiota is essential for predicting grassland ecosystem resilience. This study aimed to characterize the community structure, functional potential, and soil environmental drivers of rhizosphere and root endophytic microbiota associated with Polygonum divaricatum across three Hulunbuir Grassland sites. A nested sampling design was applied with three replicated plots per site, from which paired rhizosphere soil and root samples were collected. Each sample represented a composite of 15 plants, yielding six samples per site (total n = 18) and allowing the separation of compartmental and environmental effects on community assembly. P. divaricatum plays a key role in nutrient cycling and soil stability; however, its rhizosphere and root microbiomes remain poorly characterized. Fungal diversity was consistently higher in the root endosphere, whereas bacterial diversity was greater in rhizosphere soils. Fungal assemblages were dominated by Ascomycota and Mortierellomycota, primarily represented by Mortierella and Trichoderma, while bacterial communities were dominated by Actinomycetota and Pseudomonadota, enriched in Bradyrhizobium and Pseudonocardia. Community differentiation reflected strong compartmental filtering and responses to soil pH, organic carbon, nitrogen, and enzyme activities. Functional prediction indicated clear compartmental partitioning: in the rhizosphere, bacterial communities were enriched in pathways related to carbon and nitrogen metabolism and secondary metabolite biosynthesis, whereas in the root endosphere, functional profiles were more associated with transport, uptake, and fermentation; fungal communities were dominated by saprotrophic and symbiotrophic guilds. These findings demonstrate that soil biochemical gradients and host-driven filtering jointly structure the P. divaricatum microbiome, providing ecological insights into plant–microbe–soil interactions and the maintenance of grassland ecosystem stability.

1. Introduction

Temperate grasslands are productive yet sensitive ecosystems in which plant–soil–microbe interactions underpin long-term stability [1]. Polygonum divaricatum L. (genus Polygonum, family Polygonaceae), a dominant perennial forb in the Hulunbuir Grassland, contributes substantially to soil stabilization and nutrient cycling [2,3]. Unlike common shallow-rooted grassland graminoids, P. divaricatum possesses a deep and highly branched taproot system that modifies belowground microhabitats. Furthermore, it synthesizes diverse secondary metabolites, including flavonoids and tannins, which may influence microbial assembly through allelopathic interactions and specialized carbon inputs [4,5].

Within this specialized belowground environment, the rhizosphere and root endosphere constitute two distinct but interconnected microbial compartments. The rhizosphere serves as a dynamic interface driven by root exudates [6], whereas the endosphere harbors internal, often asymptomatic, microbial residents. Together, these communities support key ecological functions, including nutrient solubilization, nitrogen fixation, and the induction of systemic resistance to environmental stresses [7,8]. In grassland ecosystems, such plant microbe feedbacks are critical for plant coexistence and ecosystem resilience [9,10]. However, while microbiome assembly patterns have been well documented for dominant graminoids, the mechanisms governing spatial compartmentalization in foundational non-graminoid forbs such as P. divaricatum remain largely unresolved [11].

Beyond host-specific traits, microbiome structure is strongly shaped by spatial environmental heterogeneity. Soil chemical properties (e.g., pH, organic carbon, and nutrient availability) consistently regulate microbial diversity and community composition [12,13,14], while soil enzyme activities (e.g., urease, acid phosphatase, and β-glucosidase) serve as sensitive indicators of nutrient transformation and microbial metabolic potential [15]. Yet, a critical knowledge gap exists regarding how the biological filters of P. divaricatum, including its extensive taproot architecture and diverse secondary metabolite profile, interact with varying soil biochemical gradients. It remains unclear whether this species exerts strong host-specific filtering on microbial communities or whether microbiome assembly is primarily structured by opportunistic responses to local soil fertility.

Addressing this question requires a holistic, multi-domain perspective. Most previous research on P. divaricatum has focused primarily on mycorrhizal fungi, which represent only a fraction of the total root-associated microbiota [16]. Existing approaches rarely integrate both rhizosphere and root endosphere compartments, nor do they simultaneously assess how bacterial and fungal communities respond to soil chemical properties and enzymatic activities [17,18]. To bridge these knowledge gaps, we conducted a comprehensive multi-site investigation across the Hulunbuir Grassland.

Therefore, this study aimed to (i) characterize the diversity, composition, and ASV-based predicted ecological functions of rhizosphere and root endophytic microbiota associated with P. divaricatum across three representative sites, and (ii) elucidate how soil chemical and enzymatic properties shape microbial community assembly and functional differentiation. Specifically, we hypothesized that: (i) microbial alpha diversity (Shannon and Simpson indices) and community composition (assessed by PCoA and PERMANOVA) would differ significantly between the rhizosphere and root endosphere, reflecting compartment-specific habitat filtering and host selection; (ii) both fungal and bacterial communities would exhibit site-specific variation driven by spatial heterogeneity in soil chemical and enzymatic properties, tested using constrained ordination analyses (RDA/db-RDA); and (iii) soil biochemical factors would differentially regulate inferred functional profiles between compartments, as predicted by FUNGuild and Tax4Fun2/PICRUSt2. By integrating high-throughput sequencing with soil biochemical analyses, this work provides new insights into the environmental regulation of plant-associated microbiota in temperate grassland ecosystems [19].

2. Materials and Methods

2.1. Study Area and Site Description

To examine how environmental heterogeneity shapes the P. divaricatum-associated microbiome, three representative sampling sites were selected in Chen Barhu Banner, Hulunbuir Grassland: Tenihe (S1), Xiertala (S2), and Hadatu Ranch (S3) (Figure 1). The region is characterized by a temperate continental climate, with a mean annual temperature ranging from −2.6 to −1.5 °C, annual precipitation of 200–400 mm, and 2800–2950 h of sunshine per year. Although the selected sites share a similar regional climate, they span a gradient of soil types from chernozem to sandy soils and differ in dominant plant communities, reflecting environmental variation across the distribution range of P. divaricatum in temperate grasslands. All sites were selected to share similar land-use history and management, including fencing and seasonal grazing over the past five years, to reduce anthropogenic variation. Detailed environmental characteristics of each site are provided in

Table 1. Environmental characteristics of the three sampling sites.

Site	Latitude and Longitude	Altitude (m)	Annual Rainfall (mm)	Annual Sunshine (h)	Dominant Species	Soil Type
S1	119°45′31″ E 49°14′02″ N	655.1	200–300	2800	Artemisia frigida, Carex duriuscula	chestnut and meadow soils
S2	120°17′24″ E 49°32′19″ N	714.0	300–400	2800	Agropyron cristatum, Filifolium sibiricum	chernozem and chestnut soils
S3	120°27′34″ E 50°14′31″ N	743.5	300–400	2950	Stipa krylovii, Leymus chinensis	chestnut and sandy soils

[20].

2.2. Sample Collection and Preparation

Field sampling was conducted in August 2023, corresponding to the peak growing season and flowering stage of P. divaricatum. A nested hierarchical sampling design was implemented at each site. Three spatially independent plots (50 × 50 m) were established as biological replicates and separated by at least 30 m to reduce spatial autocorrelation. Within each plot, three 10 × 10 m quadrats were randomly positioned. From each quadrat, five healthy P. divaricatum individuals at comparable developmental stages were excavated intact. Rhizosphere soil was defined as the 1–3 mm layer tightly adhering to root surfaces [21]. Loosely attached bulk soil was removed by gentle shaking, and the remaining rhizosphere soil was collected using sterile soft brushes into sterile polyethylene bags. To reduce within plot variability, rhizosphere soils from 15 individuals per plot were homogenized into a single composite sample, resulting in three biological replicates per site (RS1–RS3; n = 9 total). This compositing strategy was adopted to limit the influence of individual-plant variability and to strengthen site-level statistical comparisons. However, this approach does not allow direct assessment of plant-to-plant variability within plots. Each composite sample was divided into two subsamples: one was used for soil chemical and enzymatic analyses, and the other was flash-frozen and stored at −80 °C for DNA extraction and microbial community analysis. Root tissues were collected concurrently from the same plants by excising 3–4 healthy root segments (~6 cm each) per individual. Segments from 15 plants per plot were pooled to form one composite root sample (R1–R3; n = 9 total) [22]. All samples were transported to the laboratory on ice within 24 h. Root samples were processed under aseptic conditions using a surface sterilization protocol consisting of a 30 s rinse in 75% ethanol, followed by immersion in 2.0% sodium hypochlorite for 6–8 min, and five rinses with sterile distilled water. Sterilization effectiveness was confirmed by plating the final rinse onto nutrient agar and incubating at 28 °C for 72 h; no microbial growth was detected [23]. Surface-sterilized tissues were subsequently stored at −80 °C for downstream endophytic microbiome analyses.

2.3. Chemical Properties and Enzymatic Activities of Rhizosphere Soils

Rhizosphere soil chemical properties were determined according to standard analytical protocols [24]. Soil pH was measured in a 1:2.5 (soil:water) suspension using a glass electrode pH meter (FE28, METTLER-TOLEDO, Columbus, OH, USA). Organic carbon (OC) was quantified by the potassium dichromate–sulfuric acid oxidation method, and total nitrogen (TN) was determined by Kjeldahl digestion using a Kjeldahl nitrogen analyzer (KN-520, Alva, Shandong, China). Total phosphorus (TP) and total potassium (TK) were measured after NaOH fusion using the molybdenum antimony colorimetric and flame photometric methods, respectively. Available phosphorus (AP) was extracted with 0.5 M sodium bicarbonate (Olsen method) and determined colorimetrically after acid digestion using a digestion apparatus (SPH120, Alva, Shandong, China), while available potassium (AK) was measured by flame photometry after ammonium acetate extraction. Ammonium nitrogen (AN) and nitrate nitrogen (NN) were extracted with KCl and quantified spectrophotometrically. Soil enzyme activities were determined according to Guan (1986) [25]. Urease (URE) activity was measured by the phenol sodium hypochlorite colorimetric method and expressed as μg NH₄⁺-N g⁻¹ dry soil 24 h⁻¹. Catalase (CAT) activity was determined by the ammonium molybdate assay and expressed as μmol H₂O₂ decomposed g⁻¹ dry soil h⁻¹. Acid phosphatase (ACP) activity was assayed using the disodium phenyl phosphate method and expressed as mg phenol g⁻¹ dry soil 24 h⁻¹. Cellulase (CEL) activity was determined using the 3,5-dinitrosalicylic acid method and expressed as mg glucose equivalents g⁻¹ dry soil 24 h⁻¹. N-acetylglucosaminidase (NAG) activity was measured using p-nitrophenyl-N-acetyl-β-D-glucosaminide as substrate and expressed as μg p-nitrophenol g⁻¹ dry soil h⁻¹. Leucine aminopeptidase (LAP) activity was assayed using L-leucine-p-nitroanilide as the substrate, and the activity was expressed as nmol p-nitroaniline g⁻¹ dry soil h⁻¹, while β-glucosidase (β-GC) activity was determined using p-nitrophenyl-β-D-glucopyranoside and expressed as μg p-nitrophenol g⁻¹ dry soil h⁻¹. Enzyme activities were assayed using a microplate reader (SpectraMax 190, Molecular Devices, San Jose, CA, USA).

Analytical grade reagents were sourced from Nanjing Chemical Reagent Co., Ltd. (Nanjing, China), and all measurements were performed in technical triplicates to ensure analytical precision.

2.4. Sequence Processing and Taxonomic Assignment

Microbial genomic DNA was extracted from 0.5 g of rhizosphere soil and root tissues using the E.Z.N.A.^® Mag-Bind Soil DNA Kit (Omega Bio-Tek, Norcross, GA, USA) following the manufacturer’s instructions. DNA concentrations were determined with a Qubit 2.0 fluorometer (Thermo Fisher Scientific, Waltham, MA, USA). The fungal internal transcribed spacer (ITS) region was amplified using primers ITS1F (5′-CTTGGTCATTTAGAGGAAGTAA-3′) and ITS2R (5′-GCTGCGTTCTTCATCGATGC-3′) [26]. The bacterial 16S rRNA V3-V4 region was amplified with primers 341F (5′-CCTAYGGGRBGCASCAG-3′) and 806R (5′-GGACTACNNGGGTATCTAAT-3′) [27]. PCR reactions (30 μL) contained 15 μL of 2× Phusion Master Mix, 1 μL of each primer (1 μM), 10 μL of template DNA (1 ng/μL), and nuclease-free water. Amplification conditions were: 98 °C for 1 min; 30 cycles of 98 °C for 10 s, 50 °C for 30 s, 72 °C for 30 s; and a final extension at 72 °C for 5 min. PCR products were visualized on 2% agarose gels, purified with the Universal DNA Purification Kit (Tiangen, Beijing, China), and quantified using the Qubit fluorometer. Purified amplicons were pooled in equimolar ratios and used to construct sequencing libraries with the NEBNext^® Ultra™ DNA Library Prep Kit for Illumina (New England Biolabs, Ipswich, MA, USA). Library quality was verified using an Agilent 5400 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA), and paired-end sequencing (2 × 250 bp) was performed on an Illumina MiSeq platform (Illumina, San Diego, CA, USA). Raw reads were processed in QIIME2 (v2022.8) using DADA2 (v1.26) for quality control, denoising, and chimera removal to obtain high-quality amplicon sequence variants (ASVs) [28]. Representative ASVs were taxonomically classified in QIIME2 using the feature classifier plugin with a confidence threshold of 0.8. Bacterial and fungal taxa were annotated against the SILVA (v138.2) and UNITE (v10.0) databases, respectively [29]. All representative sequences were deposited in NCBI GenBank under BioProject accession numbers PRJNA1255875 (fungi) and PRJNA1255880 (bacteria).

2.5. Statistical Analyses and Data Visualization

For statistical analyses, site-level comparisons were based on three composite biological replicates per compartment (rhizosphere and root endosphere) at each site. Alpha diversity metrics, including the ACE richness index, Shannon diversity index, and Simpson index, were calculated in RStudio v2024.04.1 (Posit Software, PBC, Boston, MA, USA). Community composition was visualized using Venn diagrams and stacked bar plots to illustrate shared and unique taxa among sample groups. Beta diversity was assessed using Principal Coordinate Analysis (PCoA) based on Bray–Curtis dissimilarity matrices derived from genus-level relative abundance profiles. This analysis was implemented using the vegan package in R (https://cran.r-project.org/web/packages/vegan/index.html, accessed on 18 February 2026) to evaluate compositional variation across sampling sites and plant compartments. Differences in genus-level relative abundance were tested using non-parametric Kruskal–Wallis tests in SPSS v31.0 (IBM Corp., Armonk, NY, USA). Where applicable, pairwise post hoc comparisons were conducted, and p-values were adjusted using the Benjamini–Hochberg False Discovery Rate (FDR) correction to control for multiple testing [30]. To identify environmental factors associated with microbial community structure, Redundancy Analysis (RDA) was performed in CANOCO v5.0 (Microcomputer Power, Ithaca, NY, USA), incorporating soil physicochemical properties and enzyme activities as explanatory variables [23]. Functional guilds of fungal communities were predicted using FUNGuild, and bacterial functional profiles were inferred using PICRUSt2. Functional annotation results were visualized in OriginPro v9.8 (OriginLab Corp., Northampton, MA, USA). Spearman’s rank correlation analysis was used to evaluate associations between dominant microbial genera and environmental variables. Correlation patterns were visualized as hierarchical clustered heatmaps generated in SigmaPlot v16.0 (Systat Software, Inc., San Jose, CA, USA).

3. Results

3.1. Variations in Rhizosphere Soil Chemical and Enzymatic Profiles Across Sites

The chemical and enzymatic properties of P. divaricatum rhizosphere soils differed significantly among the three sampling sites (Figure 2). Soil pH ranged from slightly acidic to neutral at S1 and S3 (6.5–6.8), whereas S2 exhibited a significantly lower pH (p < 0.05). Organic carbon, total nitrogen, total phosphorus, available phosphorus, and available potassium concentrations were highest at S3 and lowest at S1 (p < 0.05). In contrast, mineral nitrogen levels were highest at S1, with significantly greater nitrate and ammonium concentrations compared to the other sites (p < 0.05). Total potassium concentrations were significantly higher at S1 and S2 than at S3. Soil enzyme activities also varied among sites. Activities of β-glucosidase, urease, leucine aminopeptidase, and catalase were significantly higher at S3 than at S1 and S2 (p < 0.05). In contrast, cellulase and acid phosphatase activities did not differ significantly among sites (p > 0.05).

3.2. Characterization of Microbial Diversity and ASV Distribution

High-throughput sequencing generated high-quality fungal and bacterial datasets from both the rhizosphere and root endosphere of P. divaricatum (Figure S1). A total of 548,634 ITS reads were obtained from rhizosphere soils, yielding 480,676 denoised, non-chimeric sequences. Using the DADA2 algorithm, 2407 amplicon sequence variants (ASVs) were identified, with 964, 693, and 750 ASVs detected at sites S1, S2, and S3, respectively. In the root endosphere, 553,362 reads yielded 483,318 denoised sequences and 3391 ASVs, including 1537, 1308, and 1322 ASVs at S1, S2, and S3. Venn diagrams revealed distinct ASV assemblages between compartments, with limited overlap between rhizosphere and root samples. A larger number of compartment and site-specific fungal ASVs were identified in the root endosphere compared with the rhizosphere (Figure S2a,b). For bacteria, 336,503 high-quality 16S rRNA reads were obtained from rhizosphere soils, producing 123,798 denoised, non-chimeric sequences and 15,688 ASVs. The ASV counts at S1, S2, and S3 were 5905, 5750, and 6304, respectively. In the root endosphere, 372,958 reads generated 253,749 high-quality sequences and 6270 ASVs, with 1991, 2001, and 3171 ASVs at S1, S2, and S3, respectively. Similar to fungi, bacterial communities shared only a small proportion of ASVs between the two compartments, while numerous site-specific taxa were detected in rhizosphere soils across all sites (Figure S2c,d).

Alpha diversity indices demonstrated clear differences between compartments and among sites (

Table 2. Alpha diversity metrics of fungal and bacterial communities associated with P. divaricatum across sampling sites and compartments.

Site	Rhizosphere			Root Endosphere
	ACE	Shannon Index	Simpson_1-D	ACE	Shannon Index	Simpson_1-D
Fungal communities
S1	500.33 ± 41.79 bA	3.09 ± 0.05 bB	0.88 ± 0.01 bB	818.67 ± 11.15 aA	3.94 ± 0.08 aB	0.92 ± 0.01 aB
S2	403.00 ± 28.14 bB	3.51 ± 0.11 bA	0.94 ± 0.01 aA	720.33 ± 32.35 aB	4.14 ± 0.08 aA	0.95 ± 0.01 aA
S3	419.00 ± 16.64 bB	2.73 ± 0.05 bC	0.75 ± 0.01 bC	706.67 ± 11.02 aB	3.90 ± 0.05 aB	0.95 ± 0.00 aA
Bacterial communities
S1	2775.21 ± 241.30 aA	7.14 ± 0.09 aA	1.00 ± 0.00 aA	1119.77 ± 6.75 bA	5.18 ± 0.04 bB	0.98 ± 0.00 bB
S2	2566.97 ± 670.35 aA	7.02 ± 0.25 aA	1.00 ± 0.00 aA	1065.50 ± 291.24 bA	4.65 ± 0.05 bC	0.96 ± 0.00 bC
S3	2797.47 ± 27.41 aA	7.13 ± 0.02 aA	1.00 ± 0.00 aA	1549.69 ± 206.71 bA	5.72 ± 0.06 bA	0.99 ± 0.00 bA

Data are presented as mean ± standard deviation (SD; n = 3). Different lowercase letters indicate significant differences between the rhizosphere and root endosphere within the same site (p < 0.05), whereas different uppercase letters denote significant differences among sites for the same parameter (p < 0.05).

). For fungi, richness (ACE) and diversity (Shannon, Simpson_1-D) were generally higher in the root endosphere than in rhizosphere soils, although the significance and magnitude of these differences varied among sites. S1 exhibited the greatest fungal richness in both compartments, whereas S2 displayed the highest diversity indices. In contrast, bacterial richness and diversity were generally higher in the rhizosphere than in the root endosphere, but the differences among sites were relatively small. Within the root endosphere, samples from site S3 exhibited the highest bacterial diversity, whereas S2 showed the lowest values. Based on these metrics, fungal diversity was greater in the root endosphere, while bacterial diversity was higher in rhizosphere soils.

3.3. Microbial Community Assembly and Taxonomic Composition

3.3.1. Fungal Community Structure and Biomarker Identification

Fungal community composition differed across sampling sites and between rhizosphere and root endosphere compartments (Figure 3a). At the phylum level, Ascomycota, Mortierellomycota, and Basidiomycota dominated across all samples, with notable variation in relative abundance among sites and compartments. In the rhizosphere, Ascomycota reached its highest relative abundance at S2 (81.17%), whereas Mortierellomycota (53.41%) and Basidiomycota (15.49%) were most abundant at S3. In the root endosphere, Ascomycota remained dominant at S2 (62.72%), while Mortierellomycota and Basidiomycota showed their highest relative abundances at S1 (49.83%) and S3 (12.74%), respectively. At the genus level, community composition exhibited site- and compartment-associated variation (Figure 3b). LEfSe analysis (LDA > 3.0) identified taxa with significant differential enrichment among site-compartment groups. In rhizosphere samples, Fusarium and Gibberella were significantly enriched at S1, Trichoderma and Trichocladium at S2, and Mortierella and Inocybe at S3. In root endosphere samples, Mortierella and Tomentella were significantly enriched at S1, Cephalosporium and Trichoderma at S2, and Gliophorus and Scutellinia at S3. Venn diagram analysis showed a higher number of compartment-specific fungal genera in root endosphere samples than in rhizosphere samples (Figure 4a–c). Trichocladium and Pseudogymnoascus were detected primarily in rhizosphere samples, whereas Monilia, Scutellinia, and Knufia were detected primarily in root endosphere samples.

3.3.2. Bacterial Community Structure and Biomarker Identification

Bacterial community composition varied across sites and between rhizosphere and root endosphere compartments (Figure 3c). At the phylum level, Actinomycetota (formerly Actinobacteriota) and Pseudomonadota (formerly Proteobacteria) dominated across all samples. In rhizosphere samples, Actinomycetota reached its highest relative abundance at S2 (61.82%), whereas Pseudomonadota peaked at S3 (27.27%). In root endosphere samples, Actinomycetota showed its highest relative abundance at S1 (49.95%), while Pseudomonadota reached its maximum at S3 (50.42%). LEfSe analysis identified bacterial taxa with significant differential enrichment among site compartment groups (Figure S3c,d). In rhizosphere samples, Bradyrhizobium and Inquilinus were significantly enriched at S1, Kribbella and Aeromicrobium at S2, and Phyllobacterium and Micromonospora at S3. In root endosphere samples, Pseudonocardia and Mesorhizobium were enriched at S1, Bacillus and Rubrivivax at S2, and Devosia and Steroidobacter at S3 (Figure 3d). Venn diagram analysis indicated limited taxonomic overlap among bacterial communities across sites and compartments (Figure 4d–f). Kaistobacter and Microlunatus were more frequently detected in rhizosphere samples, whereas Bosea, Cryptosporangium, and Steroidobacter were more frequently detected in root endosphere samples.

3.4. Structural Divergence and Responsive Taxa of Microbial Communities

Principal Coordinate Analysis (PCoA) based on Bray–Curtis distances revealed distinct compositional patterns in fungal and bacterial communities across sampling sites and plant compartments (Figure 5a). For fungal communities, the first two PCoA axes explained 81.42% of the total variance, with samples primarily separated by sampling site along Axis 1. PERMANOVA indicated that sampling site explained a larger proportion of variation in fungal community structure (R² = 0.61, F = 11.79, p < 0.01) than plant compartment (R² = 0.164, F = 3.147, p < 0.05). PERMDISP analysis showed significant differences in dispersion among sites (F = 40.246, p < 0.01), whereas dispersion between rhizosphere and root endosphere compartments was not significantly different (p > 0.05). For bacterial communities, the first two PCoA axes explained 70.50% of the total variance. In the ordination space, samples were primarily separated by plant compartment (Figure 5b). PERMANOVA results indicated that plant compartment explained a greater proportion of variance in bacterial community structure (R² = 0.45, F = 13.24, p < 0.01) than sampling site (R² = 0.28, F = 2.93, p < 0.05). PERMDISP analysis further indicated significant differences in bacterial dispersion across both plant compartments (F = 32.22, p < 0.01) and sampling sites (F = 4.269, p < 0.05).

To identify taxa associated with observed structural differences, genus-level Kruskal–Wallis tests with Benjamini–Hochberg False Discovery Rate (FDR) correction were applied (p < 0.05). Genera showing significant differences were visualized using a hierarchical clustering heatmap (Figure 6) with Ward’s linkage method. Relative abundance values were row-normalized using Z-score transformation to facilitate comparison across taxa. Among the top 30 fungal genera, 27 showed significant variation across groups, collectively accounting for 83.7–93.0% of total fungal sequence abundance. Genera including Trichoderma, Leptosphaeria, and Pseudogymnoascus exhibited among the lowest adjusted P-values. For bacterial communities, multiple genera also displayed significant abundance differences across compartments and sites. Kaistobacter and Microlunatus showed higher relative abundances in rhizosphere samples, whereas Devosia and Rhodoplanes showed higher relative abundances in root endosphere samples.

3.5. Soil Chemical and Enzymatic Drivers of Microbial Community Assembly

Spearman correlation analysis identified multiple significant associations between dominant microbial genera and soil physicochemical and enzymatic variables (Figure 7). For fungal communities, genus-level abundances showed significant correlations with soil pH, OC, TN, TP, TK, AP, NN, and enzyme activities including URE, CAT, LAP, and β-GC (p < 0.05). In the root endosphere, fungal genera exhibited significant correlations with AK, whereas no significant AK associations were detected in rhizosphere samples. For bacterial communities, significant correlations with soil pH and NN were detected in both rhizosphere and root endosphere samples (p < 0.05). In the root endosphere, bacterial genera showed significant associations with OC, TN, TP, TK, AP, AK, and multiple enzyme activities, whereas fewer significant correlations with these variables were observed in rhizosphere samples.

Redundancy analysis (RDA) revealed patterns of microbial community variation in relation to soil chemical and enzymatic parameters (Figure 8). Samples from S3 were distributed closer to multiple soil chemical and enzymatic vectors compared with S1 and S2. Permutation tests indicated that all environmental variables, except ACP, CEL, and NAG, were significantly associated with microbial community structure (p < 0.05). For fungal communities, the first two RDA axes explained 65.59% (RDA1) and 31.24% (RDA2) of the total variance, accounting for 96.83% cumulatively. In the ordination space, Mortierella and Archaeorhizomyces aligned with the positive directions of soil pH, NN, and LAP, and were oriented opposite to the TK vector. Tomentella, Gibberella, and Fusarium showed positive alignment with nitrogen-related variables, particularly AN and NN, and exhibited an overall orientation opposite to TK. In contrast, Trichoderma, Cladosporium, and Pseudogymnoascus were positioned toward the TK-associated direction and showed opposite orientations to pH and nitrogen-related vectors (e.g., pH, AN/NN, and LAP).

For bacteria, RDA1 and RDA2 together explained 73.24% of the total variance. Pseudomonas and Devosia were aligned strongly with the positive directions of AP, TN, and TP, and showed concordant orientation with enzyme activities URE and CAT. Kaistibacter and Nocardioides were closely aligned with AK and also oriented in the same direction as OC and β-GC, while being positioned opposite to inorganic nitrogen vectors, particularly AN (and for Nocardioides, also NN). In comparison, Rhodoplanes, Mesorhizobium, and Bradyrhizobium aligned strongly with NN and the soil pH gradient, with additional concordance to LAP. Bacillus and Streptomyces were positioned along the positive direction of TK and showed opposite orientations to the vectors for AP, TN, and TP, as well as enzyme activities URE and CAT, in the ordination space.

3.6. Potential Functional Profiles Inferred from Microbial Communities

Functional profiles were inferred based on taxonomic composition. Fungal trophic modes were annotated using the FUNGuild database, with more than 80% of fungal taxa assigned to functional guilds (Figure 9a). Across all samples, pathotrophs and saprotrophs represented the dominant trophic groups, whereas symbiotrophs accounted for a consistently low proportion (<1%). Significant site-dependent variation was detected in trophic mode composition. Saprotrophs exhibited significantly higher relative abundance at S3, while pathotrophs were significantly more abundant at S2. Trophic group distributions also differed between rhizosphere and root endosphere compartments. At S1 and S2, pathotrophs showed higher relative abundance in the rhizosphere, whereas saprotrophs were more abundant in the root endosphere. In contrast, at S3, pathotrophs were more abundant in root samples, while saprotrophs showed higher relative abundance in the rhizosphere.

Bacterial functional profiles were inferred using PICRUSt2 (Figure 9b). At the broad functional category level, major metabolic groups, including carbon metabolism, energy metabolism, and nitrogen metabolism, showed similar relative contributions across sampling sites. Differences in predicted functional composition were observed between rhizosphere and root endosphere compartments. The rhizosphere exhibited relatively higher representation of pathways associated with secondary or specialized metabolism and amino acid metabolism, whereas the root endosphere showed higher relative representation of pathways related to transport and uptake functions. At the MetaCyc pathway level, several pathways displayed differences in relative abundance between compartments. Pathways related to fermentation, denitrification, and secondary metabolite biosynthesis were more highly represented in root endosphere samples compared with rhizosphere samples.

4. Discussion

P. divaricatum, a perennial dominant species in the Hulunbuir grasslands, is considered capable of influencing plant community dynamics and belowground nutrient cycling through rhizosphere processes and associated plant–microbe interactions [31]. In this study, we observed pronounced compositional differentiation between rhizosphere and root endosphere microbiomes across sampling sites, supported by compartment-level turnover in ASV composition and clear separation patterns in ordination space. PERMANOVA further indicated contrasting contributions of site and compartment to bacterial versus fungal community variation: bacterial community structure was more strongly associated with compartment (R² = 0.45), whereas fungal community structure was more strongly associated with site (R² = 0.61). These patterns are consistent with stronger compartment-associated differentiation in bacteria, which is often interpreted as reflecting host-associated filtering in the endosphere and tighter coupling to host-derived substrates and signaling environments [5,32]. In comparison, fungal communities may be more closely associated with site-specific soil conditions, consistent with the close linkage between fungal hyphal growth, decomposition strategies, and soil chemical gradients such as pH and nutrient status [33,34]. Importantly, PERMDISP indicated significant heterogeneity of dispersion among sites for fungi, and for bacteria across both sites and compartments, suggesting that both centroid shifts and within group variability may contribute to the observed PERMANOVA patterns. This contrast is compatible with stronger compartment-associated selection in the root endosphere, whereas fungi may retain a clearer site-associated signal due to their close coupling to edaphic conditions and broader spatial integration in soils [35]. Nevertheless, given the dispersion heterogeneity detected by PERMDISP, these patterns should be interpreted as multivariate associations rather than definitive evidence of specific causal mechanisms.

Alpha diversity patterns and ASV overlap further supported strong compartmentalization of the microbiota. Fungal richness and diversity were generally higher in the root endosphere than in the rhizosphere, whereas bacterial communities showed the opposite trend, with higher richness and diversity in rhizosphere soils. In addition, Venn patterns indicated limited ASV sharing between rhizosphere and root endosphere samples for both kingdoms, suggesting substantial turnover between compartments. These contrasts are consistent with compartment-specific microhabitat differentiation and selective filtering [36]; however, the mechanisms underlying diversity differences cannot be resolved from amplicon data alone and may reflect a combination of colonization sources, habitat stability, and host-associated selection [37].

Microbial composition also varied among sites, consistent with significant associations with soil chemical properties and enzymatic activities [38]. In this study, RDA indicated that multiple edaphic variables were significantly associated with fungal community structure, and the first two axes captured clear separation among samples along key soil gradients. Notably, genera such as Mortierella and Archaeorhizomyces were positioned along the positive directions of soil pH, NN, and LAP, suggesting that fungal assemblages were associated with coordinated variation in acidity/alkalinity, inorganic nitrogen availability, and protein turnover potential. This pattern is consistent with the ecological attributes of these taxa, which are frequently reported as decomposition-associated lineages, and with evidence that pH and nitrogen availability are major correlates of fungal community turnover in grassland soils [39,40]. At the same time, because these inferences derive from ordination and correlation patterns, they should be interpreted as environmental associations rather than evidence of active host recruitment or direct functional activity.

The rhizosphere enrichment of Trichoderma was consistent with our previous culture-based observations [41], supporting concordance between sequencing-based patterns and culturable evidence. Genera such as Trichoderma and Fusarium include taxa frequently reported to participate in organic matter transformation and nutrient mobilization, whereas root endosphere-enriched taxa such as Tomentella belong to lineages often considered symbiotrophic, and may therefore be more closely associated with host filtering [42,43]. Nutrient-related gradients further structured these distributions. In the RDA biplot, Trichoderma, together with Cladosporium and Pseudogymnoascus, was oriented toward the TK-associated direction and opposite to the pH, NN, and LAP gradient, indicating that potassium-related variation co-occurred with shifts in dominant fungal taxa across samples [44]. This TK-associated axis may capture broader variation in soil nutrient balance and substrate-related conditions that can influence fungal ecological strategies, and may help contextualize the contrasting distribution patterns of dominant genera across sites and compartments [45]. Given that the Hulunbuir grassland is frequently water-limited, the alignment of Trichoderma with the TK gradient may be ecologically relevant. Because potassium supports drought-related plant physiology and Trichoderma spp. can enhance abiotic stress tolerance, the observed co-enrichment may indicate a potential nutrient microbe linkage, although this remains inferential [46,47].

Consistent with patterns commonly reported in grassland soils, bacterial communities were dominated by Actinomycetota and Pseudomonadota, a pattern widely reported in grassland ecosystems and often linked to organic matter transformation, nutrient cycling, and stress-tolerant ecological strategies [48,49]. Our RDA showed that bacterial community variation was significantly associated with edaphic gradients, with the constrained axes capturing major patterns related to soil chemical and enzymatic properties. These observations are consistent with established enzyme microbe feedback frameworks in grassland soils. Soil microorganisms are known to secrete extracellular enzymes that depolymerize recalcitrant organic matter, releasing bioavailable carbon and nutrients that can modify local microenvironmental conditions and potentially influence bacterial community assembly [50]. Such enzyme-mediated processes are considered fundamental to nutrient turnover and microbial community maintenance within the P. divaricatum root soil continuum [51].

In particular, bacterial assemblages appeared to align with multiple gradients rather than a single fertility indicator, including inorganic nitrogen availability and soil pH (e.g., NN and pH), as well as potassium-related variables (TK and AK). For example, Streptomyces was positioned toward the potassium-associated direction, whereas Rhodoplanes, Mesorhizobium, and Bradyrhizobium aligned with the NN and pH direction with additional concordance to LAP. Nocardioides showed contrasting associations with potassium- and carbon-related variables versus inorganic nitrogen, further supporting the role of multi- dimensional edaphic structure in shaping bacterial distributions across sites. In parallel, bacterial assemblages remained differentiated between the rhizosphere and root endosphere, indicating that compartment-associated filtering co-occurred with site-associated edaphic structure [52]. Rhizosphere-enriched genera such as Kaistobacter and Microlunatus have been reported to be associated with carbon turnover and rhizodeposition-related organic matter processing [53], whereas root endosphere-associated taxa, including Rhodoplanes and Bradyrhizobium, comprise lineages that have been implicated in nitrogen-related metabolic processes [54]. Their enrichment in the root endosphere is compatible with compartment microenvironments that may support nitrogen-related processes, although direct functional validation would be required to confirm specific mechanisms [55].

Fungal trophic modes varied among sites, indicating site-dependent differences in inferred functional composition. Saprotroph-assigned fungi were relatively more abundant at S3, which also exhibited higher organic carbon content and elevated activities of turnover-related enzymes. This pattern suggests a potential enzyme microbe feedback relationship, whereby greater carbon availability may favor saprotrophic proliferation, and the associated increase in extracellular enzyme production could further promote organic matter decomposition and nutrient turnover [56,57]. By comparison, the increased representation of pathotroph-assigned groups at S2 indicates a distinct community configuration that may relate to local edaphic conditions and host–microbe interactions, although causal drivers cannot be resolved from guild inference alone. Notably, a higher relative abundance of taxa annotated as pathotrophs does not necessarily indicate an active disease state, because many taxa taxonomically assigned to pathotrophic guilds can persist as symptomless endophytes in grassland plants [58]. The distribution of fungal guilds between the rhizosphere and root endosphere also varied by site. At S1 and S2, saprotroph-assigned taxa were relatively more represented in the root endosphere, whereas at S3, they were more enriched in the rhizosphere. This site-dependent compartment pattern is consistent with the possibility that differences in local soil biochemical conditions and root-associated microenvironments influence the compartmental distribution of saprotroph-assigned fungi across environments [59].

Bacterial functional profiles inferred by PICRUSt2 exhibited clear compartment-associated specialization. Although no significant differences were detected among sites at the level of major functional categories, rhizosphere communities consistently showed higher relative representation of pathways related to carbon and nitrogen metabolism as well as secondary/specialized metabolism, whereas root endosphere bacteria displayed relatively higher representation of functions associated with transport and uptake. Together, these patterns are consistent with compartment-level functional differentiation between the rhizosphere and root endosphere, potentially reflecting differences in substrate availability and selective pressures across root-associated microhabitats [60,61]. Notably, the predicted enrichment of fermentation- and denitrification-related pathways in the root endosphere is compatible with the possibility of more reduced or oxygen limited microconditions within root-associated microniches, and aligns with the compartment-associated taxonomic patterns observed in this study [62]. Such microconditions have been reported for plant internal habitats and may be facilitated by local oxygen consumption and diffusion constraints in root tissues [63]. However, because these results are derived from marker-gene-based functional prediction, they should be interpreted as indicative trends rather than direct measurements of in situ metabolic activity.

In summary, the rhizosphere and root endosphere microbiota of P. divaricatum exhibited distinct compartment and site-associated assembly patterns, consistent with the joint influence of edaphic filtering and host-associated selection [64]. In our dataset, fungal community patterns were more strongly associated with site-level edaphic variation, whereas bacterial assemblages showed a comparatively stronger compartment-associated signal. The root endosphere microbiome, which is commonly reported to be subject to stronger host filtering than the rhizosphere, may therefore more consistently reflect integrated plant-associated conditions; however, this interpretation remains inferential [65]. Furthermore, the inferred functional specialization, particularly the higher relative representation of nitrogen-transformation-related functions and anaerobic-metabolism-associated pathways in the root endosphere, suggests that this compartment may be relevant to plant-associated nutrient processing and nitrogen cycling, although direct validation (e.g., redox measurements or functional gene and transcript quantification) would be needed to confirm in situ activity [66]. From a management and restoration perspective, these findings suggest that assessments of grassland ecosystem functioning may benefit from jointly considering soil biochemical status and plant-associated microbiome patterns. More broadly, incorporating information on environmental gradients and compartment-associated assembly into a plant–soil–microbe framework may help prioritize restoration targets and monitoring indicators aimed at improving soil health and supporting plant performance in degraded grassland ecosystems, although the effectiveness of such approaches will require site-specific validation.

5. Conclusions

This study shows that the rhizosphere and root endosphere microbiomes of P. divaricatum exhibit distinct and asymmetric assembly patterns across the Hulunbuir grassland. Fungal community variation was more closely associated with site-level edaphic gradients, whereas bacterial communities displayed stronger compartment-related structuring, indicating differential ecological drivers between microbial domains. These patterns suggest functional complementarity across the soil root continuum, with nutrient cycling processes more prominent in the rhizosphere and transport-related functions enriched within root tissues. Future integration of shotgun metagenomics with controlled ecological experiments will be essential to validate these functional predictions and clarify causal plant–microbe–soil interactions. Such efforts will support the development of biologically informed strategies for the conservation and restoration of degraded temperate grassland ecosystems.