Acetoin and 2,3-Butanediol Differentially Restructure Fungal and Bacterial Communities and Their Links to Host Transcription in the Rhizosphere of a Medicinal Plant

Simple Summary

This study explored how two common microbial volatile organic compounds—acetoin and 2,3-butanediol—affect the medicinal plant Pseudostellaria heterophylla and its rhizosphere microbial communities. Using a specialized pot system that allowed only volatile exchange, we found that both compounds reduced root growth and altered the composition of bacteria and fungi living around the roots. Fungal communities were particularly sensitive to 2,3-butanediol, which also triggered widespread changes in root gene expression, especially in pathways related to stress and secondary metabolism. By linking specific microbes to host gene activity, our findings reveal that volatile compounds can reshape both the root microbiome and plant molecular responses. These insights may inform sustainable strategies for improving medicinal plant cultivation.

Abstract

Microbial volatile organic compounds (VOCs) mediate rhizosphere plant-microbe interactions, yet their integrated effects on plant microbiome assembly and host transcriptional regulation remain unresolved. Here we address this gap by investigating how two common VOCs, acetoin (AC) and 2,3-butanediol (BD), influence growth, rhizosphere communities, and root gene expression in the medicinal plant Pseudostellaria heterophylla using a split-pot system. Bacterial and fungal communities were monitored across three developmental stages via amplicon sequencing, alongside root transcriptome profiling during tuber enlargement. Contrasting with widely reported growth-promoting effects of microbial VOCs, both compounds significantly reduced tuber number and biomass. Bacterial communities remained taxonomically stable, shaped primarily by species replacement, with modest VOC responses but clear shifts across developmental stages. Fungal communities exhibited marked compositional restructuring and greater treatment sensitivity, particularly under BD. Neutral community modeling indicated predominantly stochastic bacterial assembly, while fungal assembly—especially under BD—showed stronger influence of deterministic processes. BD associated with broader transcriptional reprogramming than AC, including downregulation of photosynthesis, specialized metabolism, and defense pathways. Cross-omics network analysis revealed discriminant genera (e.g., Granulicella, Harposporium) that correlated strongly with host genes involved in stress response, development, and epigenetic regulation, with fungal taxa showing tighter associations with host expression than bacteria. Together, these findings establish a mechanistic framework for how microbial VOCs shape rhizosphere communities and host responses, with implications for microbiome-based strategies in medicinal plant cultivation.

1. Introduction

Plant roots develop within a biologically active soil matrix where they interact with a dense assemblage of microorganisms collectively termed the rhizosphere microbiome [1,2,3]. Through prolonged co-evolution, roots and their associated microbes have formed integrated functional systems that influence plant growth, health, and stress tolerance via nutrient mobilization, phytohormone regulation, induction of systemic resistance, and pathogen suppression [4,5,6,7,8]. Elucidating the processes that govern rhizosphere microbiome assembly remains a central challenge in microbial ecology with implications for sustainable agriculture [9,10].

Among the rhizosphere microbial strategies that affect plant performance, volatile organic compounds (VOCs) have received increasing attention because they enable biological interactions without direct physical contact. These compounds, mainly low-molecular weight alcohols, ketones, and esters, can rapidly diffuse through soil pore spaces and act as long-distance chemical cues shaping plant physiology and development. For example, VOCs (dimethyl disulfide and dimethyl trisulfide) emitted by Rhodococcus ruber enhanced Arabidopsis thaliana growth and altered transcriptome profiles of growth-related genes [11]. Ye et al. [12] demonstrated that VOCs produced by Bacillus velezensis significantly promoted primary root elongation and lateral root development in A. thaliana by modulating auxin polar transport and auxin-responsive gene expression, highlighting a direct link between VOCs and phytohormone-regulated root architecture remodeling. Turksoy et al. [13] also revealed that the VOCs emitted by a synthetic rhizobacterial community composed of 16 strains can regulate plant growth and root architecture, and that this effect is not a simple summation of individual strain contributions but rather an emergent property exhibited by the microbial community as a whole. In parallel, microbial VOCs can modulate soil microbial communities through the selective inhibition or stimulation of neighboring taxa, thereby contributing to functional stability within the rhizosphere [14,15,16,17].

In particular, acetoin (AC) and 2,3-butanediol (BD)—two VOCs commonly produced by plant rhizobacteria—have been shown to regulate stomata by activating the salicylic acid and abscisic acid signaling pathways and inducing the accumulation of hydrogen peroxide and nitric oxide, revealing a new mechanism through which rhizobacteria enhance plant stomatal immunity to reinforce mutualistic symbiosis [18]. Notably, the same class of VOCs can exert contrasting effects in different plant species or under different environmental contexts, suggesting a context-dependent duality. Preliminary observations from our laboratory indicate that AC and BD may contribute to consecutive monoculture problems (also known as replant disease) in Pseudostellaria heterophylla, underscoring the need to elucidate their specific mechanisms for precision crop health management. Although the physiological roles of microbial VOCs have been extensively characterized in model plants and major crops, their ecological functions in medicinal plant systems remain poorly understood. Medicinal plants typically harbor complex and tightly regulated secondary metabolic networks, and increasing evidence suggests that their rhizosphere microbiomes influence not only biomass accumulation but also the biosynthesis and stability of pharmacologically active compounds. For example, Zhou et al. [19] identified key rhizosphere fungi (e.g., Serendipitaceae and Ceratobasidiaceae) and bacteria (e.g., Acetobacteraceae and Streptomycetaceae) in Dendrobium officinale that are associated with the accumulation of polysaccharides and flavonoids. Alterations in rhizosphere microbial composition were also found to significantly affect the accumulation of ginsenosides in Panax ginseng and phenolic acids in Salvia miltiorrhiza [20,21]. Despite these advances, no study has simultaneously examined how microbial VOCs reshape rhizosphere community assembly processes while also profiling host transcriptional responses across developmental stages in medicinal plants.

P. heterophylla is an important medicinal plant whose yield and therapeutic quality depend largely on the extent of storage root enlargement. The species is highly sensitive to continuous cropping and disturbances in the rhizosphere environment, making it a suitable system for examining plant–microbe interactions. To date, research on P. heterophylla has focused predominantly on its chemical composition and pharmacological properties, as well as on the effects of fertilization practices, pathogen pressure, and single-microbe inoculations on plant performance [22,23,24,25,26,27]. By contrast, the integrated roles of microbial VOCs in shaping rhizosphere community structure, governing community assembly processes, and regulating host transcriptional responses across developmental stages remain largely unexplored.

This study utilized a partitioned dual-chamber system to evaluate AC and BD effects on rhizosphere bacterial and fungal communities of P. heterophylla across developmental phases. We applied amplicon sequencing, β-diversity decomposition, community assembly modeling, and root transcriptomics to investigate two interrelated questions: (1) To what extent do AC and BD modify the structure and assembly processes of fungal versus bacterial communities? (2) How do compositional shifts in these communities correlate with host plant gene expression? Specifically, we hypothesized that: (1) fungal communities would exhibit greater sensitivity to VOC exposure than bacterial communities due to their stronger metabolic dependence on volatile signals and closer trophic interactions with plant roots; and (2) VOC-induced shifts in rhizosphere community composition would correlate with host transcriptional reprogramming in pathways governing secondary metabolism, stress responses, and storage organ development. By integrating ecological and molecular perspectives, this work expands knowledge of how VOCs mediate rhizosphere dynamics in medicinal plants and establishes a conceptual basis for microbiome-guided strategies in crop cultivation.

2. Materials and Methods

2.1. Rhizosphere Soil Sampling

Soil used for pot experiments was collected from a one-year-old P. heterophylla field in Chouling Village, Zherong County, Fujian Province, China (27°11′20″ N, 119°55′3″ E). After removing visible stones and debris, the soil was homogenized by sieving through a 10 mm mesh. To examine the effects of VOCs on plant growth, a vertically partitioned two-compartment system was established, separated by a sterile, permeable membrane that allowed VOC diffusion while preventing direct contact between the plants and the compounds. The lower compartment contained sterile Petri dishes loaded with 0.5 g of either acetoin (AC; ≥97% purity, Sigma-Aldrich, St. Louis, MO, USA) or 2,3-butanediol (BD; ≥98% purity, Sigma-Aldrich). This dose was selected based on preliminary experiments showing consistent biological effects without acute visible toxicity symptoms across a range of 0.1–1.0 g per dish, corresponding to approximately 5.7 mmol for AC (MW 88.1) and 5.6 mmol for BD (MW 90.1). While we acknowledge that the absolute VOC concentrations reaching the rhizosphere were not directly measured—a limitation that future dose-response studies using headspace sampling should address—the chosen dose range was based on preliminary experiments confirming consistent, non-phytotoxic biological effects. The interface between compartments was sealed with Parafilm^® to confine VOCs. A no-VOC treatment, consisting of a sealed split-pot with an empty lower compartment, served as the control (CK).

The experiment employed a completely randomized design with a total of 27 independent pots. This comprised 3 treatments (CK, AC, BD) × 3 destructive sampling time points (15, 45, and 90 days post-planting) × 3 biological replicates (pots) per treatment per time point. At each sampling point, three pots from each treatment were randomly selected and destructively harvested. Plants were maintained at 25 °C under a 14-h light (2000 lx)/10-h dark photoperiod and watered every three days, ensuring that no moisture entered the lower compartment. Plants were carefully uprooted, and bulk soil was removed by gentle shaking. The roots with tightly adhering soil were then placed into a sterile 50 mL centrifuge tube containing 20 mL of sterile phosphate-buffered saline (PBS). The tube was vortexed vigorously for 60 s to detach the rhizosphere soil from the root surface. Roots were then removed using sterile forceps. The resulting soil suspension was centrifuged at 3000× g for 5 min at 4 °C. The supernatant was discarded, and the soil pellet was retained as the operational definition of the rhizosphere sample for this study. This method, while not distinguishing between the rhizoplane and ectorhizosphere, provides a consistent and widely accepted approach for collecting the soil fraction most influenced by root activity

2.2. Soil DNA Extraction and Sequencing

Genomic DNA was extracted from 200 mg freeze-dried soil using the E.Z.N.A.^® Soil DNA Kit (Omega Bio-tek, Norcross, GA, USA). DNA quality and concentration were assessed via NanoDrop™ (Thermo Scientific, Waltham, MA, USA). Bacterial 16S rRNA (V3–V4) and fungal ITS1 regions were amplified with primers 341F/805R [28] and ITS1f/ITS2 [29], respectively. Each 25 μL PCR contained 12.5 μL Phusion^® Hot Start Flex 2× Master Mix (New England Biolabs, Ipswich, MA, USA), 2.5 μL each primer (20 μM), and 50 ng DNA. PCR conditions: 98 °C for 30 s; 35 cycles of 98 °C for 10 s, 54 °C for 30 s, 72 °C for 45 s; and 72 °C for 10 min. Products were purified, pooled, end-repaired, A-tailed, and ligated to Illumina adapters. Library quality was verified by Qubit, qPCR, and bioanalyzer (Agilent Technologies, Santa Clara, CA, USA). Paired-end sequencing was performed on an Illumina NovaSeq platform (Illumina, San Diego, CA, USA). Raw paired-end reads generated by Illumina sequencing were demultiplexed according to unique barcode sequences, and barcode and primer sequences were removed using a dedicated trimming tool. Paired-end reads were then quality filtered, trimmed, and denoised using the DADA2 pipeline [30], which includes error model learning, dereplication, paired-end read merging, and chimera removal, to generate amplicon sequence variants (ASVs). ASVs with fewer than five total reads across all samples were removed. Taxonomic assignment of bacterial ASVs was performed against the SILVA database (version 138) [31], while fungal ASVs were classified using the UNITE database [32]. Taxonomic classification was conducted using a naïve Bayesian classifier with a confidence threshold of 0.7. To ensure comparability across samples, all sequence reads were rarefied to an even depth (the minimum read count per sample) to generate a normalized ASV table and representative sequences, which were subsequently used for downstream analyses.

2.3. RNA Extraction and Transcriptome Sequencing

Roots of P. heterophylla were collected at the peak enlargement stage (90 days post-planting). Three biological replicates were prepared, each consisting of pooled tubers from a single pot. Samples were immediately snap-frozen in liquid nitrogen and stored at −80 °C. Total RNA was isolated using TRIzol reagent (Sigma-Aldrich, St. Louis, MO, USA) according to the manufacturer’s protocol. To remove contaminating genomic DNA, samples were treated with TURBO DNase (Ambion, Austin, TX, USA) and further purified using RNeasy MinElute spin columns (Qiagen, Hilden, Germany). RNA concentration and purity were determined using a NanoDrop 2000 spectrophotometer (Thermo Fisher Scientific, Waltham, MA, USA). RNA integrity was evaluated via 1% agarose gel electrophoresis and an Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA); only samples with an RNA Integrity Number (RIN) > 7.0 were used for library construction.

Strand-specific RNA-seq libraries were constructed using the dUTP-based method. Briefly, poly(A) mRNA was enriched from total RNA using oligo(dT) magnetic beads and fragmented into short inserts. First-strand cDNA was synthesized with random hexamer primers, followed by second-strand synthesis incorporating dUTP instead of dTTP to preserve strand specificity. Following end-repair, A-tailing, and adapter ligation, the dUTP-labeled second strands were selectively degraded using the USER enzyme (New England Biolabs, Ipswich, MA, USA). The resulting cDNA fragments were purified with AMPure XP beads (Beckman Coulter, Brea, CA, USA) and PCR-amplified to generate the final libraries. Library quality and insert size distribution were validated on an Agilent 2100 Bioanalyzer, and concentrations were quantified using a Qubit 3.0 Fluorometer (Life Technologies, Carlsbad, CA, USA). Sequencing was performed on the Illumina NovaSeq 6000 platform (Illumina, San Diego, CA, USA) in 150 bp paired-end (PE150) mode.

2.4. RNA-Seq Data Processing

Raw sequencing reads were quality-checked using FastQC (v0.11.9). Adapters, low-quality bases (Q < 20), and reads shorter than 50 bp were removed with Trimmomatic (v0.39). Cleaned reads were de novo assembled with Trinity (v2.13.2), yielding 196,599 transcripts. Transcripts shorter than 200 bp were filtered out using SeqKit (v0.16.1), and redundancy was reduced using CD-HIT-EST (v4.8.1) at 90% identity, resulting in 161,719 unigenes. Assembly completeness was assessed with BUSCO (v5.3.2; embryophyta_odb10), showing 94.6% and 94.2% completeness for the raw and filtered assemblies, respectively. Transcript abundances were estimated using Salmon (v1.9.0) with GC bias correction and mapping validation. Low-expression genes were filtered out, retaining 29,952 high-confidence transcripts. Open reading frames were predicted from these transcripts using TransDecoder (v5.5.0). Sample reproducibility was evaluated by Pearson correlation and principal component analysis.

For differential expression (DE) analysis, raw read counts per gene were obtained using Salmon and imported into R. DE analysis was performed using the edgeR 4.8.2 package, which is designed for count-based data following a negative binomial distribution. Genes with low counts (less than 10 reads in the minimum number of samples) were filtered out. Library sizes were normalized using the TMM (trimmed mean of M-values) method. The generalized linear model (GLM) likelihood ratio test was used to compare treatment groups (AC vs. CK, BD vs. CK). Genes with an absolute log₂ fold change > 1 and a false discovery rate (FDR) < 0.05 were considered significantly differentially expressed. For visualization of expression patterns and correlation analyses between replicates, and for use in the subsequent cross-omics network analysis, we used FPKM (Fragments Per Kilobase of transcript per Million mapped reads) values. FPKM values were calculated from the raw counts, providing a normalized measure of gene expression level that is comparable across samples. KEGG pathway enrichment analysis of differentially expressed genes was conducted using clusterProfiler (v4.6.0), with pathways considered significantly enriched at FDR < 0.05. Rich factor (number of DEGs annotated to a pathway divided by total genes in that pathway) was used to represent enrichment degree.

2.5. Statistical Analysis

Alpha diversity was assessed using the Shannon index for bacterial and fungal communities. Group comparisons were performed using Kruskal–Wallis tests followed by Dunn’s post hoc tests for multiple comparisons (p < 0.05). Significant differences among groups are indicated by distinct letters. For both bacterial and fungal communities, community composition was visualized using principal coordinate analysis (PCoA) ordination based on Bray–Curtis dissimilarity matrices, calculated from Hellinger-transformed ASV relative abundances. To test for differences in community structure among treatment groups (AC, BD, CK) and growth stages (FL, LS, RO), we performed permutational multivariate analysis of variance (PERMANOVA) using the adonis2 function in the R package Vegan v2.7-2. The model was specified as: Bray–Curtis dissimilarity ~ Treatment × GrowthStage, with permutations (n = 999) constrained by the GrowthStage factor to account for the temporal structure of the sampling design. Pairwise PERMANOVA comparisons were conducted post hoc where the overall interaction was significant. Differential abundance of bacterial and fungal ASVs was assessed using DESeq2. ASVs with |log2 fold change| > 1 and adjusted p < 0.05 (Benjamini–Hochberg) were considered significant. Volcano plots were generated using ggplot2, with red and blue points indicating ASVs enriched or depleted in the treatment groups, respectively.

To elucidate the mechanisms underlying microbial community assembly, total beta diversity (BDtotal) based on Jaccard dissimilarity was partitioned into two additive components—species replacement (Repl) and richness difference (RichDiff)—following the Podani framework [33]. This decomposition was performed on presence-absence ASV matrices using the adespatial v0.3-28 R package. The relative contributions of RichDiff, Repl, and community similarity were visualized using ternary diagrams via the ggtern v3.3.0 package. This approach allows for a simultaneous assessment of community turnover and nestedness-related patterns across different rhizosphere treatments.

To quantify the relative contribution of stochastic processes to rhizosphere microbial assembly, Sloan’s neutral community model (NCM) was applied [34]. The model predicts the relationship between ASV occurrence frequency and their mean relative abundance in the metacommunity, with the goodness-of-fit indicating the extent of stochastic influence. Dispersal potential was estimated by the parameter Nm (metacommunity size N times migration rate m). Calculations were implemented using the MicEco v0.9.19 R package.

To identify the root bacterial and fungal genera most characteristic of each treatment cohort, we performed sparse partial least squares-discriminant analysis (sPLS-DA) using the mixOmics v6.32.0 R package. This supervised dimensionality reduction was applied independently to bacterial and fungal community datasets. To ensure model robustness and optimize feature selection (the keepX parameter), we employed 50-fold repeated five-fold cross-validation, prioritizing the minimization of the balanced error rate. We defined the core discriminant microbial genera by selecting those with the highest VIP scores for the first two principal component analysis (PCA) components.

To elucidate the interplay between the rhizosphere microbiome and host molecular responses using the 9 samples from the RO stage, we integrated these microbial data with P. heterophylla root transcriptomes. RNA-seq expression levels, quantified as FPKM (Fragments Per Kilobase of transcript per Million mapped reads), were used for the integrative analysis. A subset of 2910 differentially expressed genes (DEGs) was prioritized for integrative analysis. To identify core discriminant microbial genera, we performed sPLS-DA and reported the final model’s balanced error rate (BER) along with its 95% confidence interval derived from 1000 bootstrap iterations. The stability of the selected features was assessed via a stability selection procedure involving 100 bootstrap runs. In each iteration, 80% of the samples were randomly resampled (stratified by treatment), and sPLS-DA was rerun.

To construct the cross-omics interaction network, we first applied centered log-ratio (CLR) transformation to both the relative abundances of core discriminant microbial genera (17 bacterial and 19 fungal genera identified via sPLS-DA) and the expression levels of the 2910 DEGs, thereby addressing the compositional nature of the microbiome data. Spearman’s rank correlation coefficients (ρ) were then calculated between the CLR-transformed datasets. To control for multiple testing, p-values were adjusted using the Benjamini–Hochberg false discovery rate (FDR), and associations with |ρ| > 0.6 and FDR < 0.01 were initially retained. To ensure robustness, we performed leave-one-out cross-validation (LOOCV) on all initially identified edges. For each edge, we sequentially removed one sample and recalculated the correlation using the remaining eight samples. Edges that maintained a consistent correlation direction (positive or negative) and retained significance (uncorrected p < 0.05) in at least seven (>75%) of the nine iterations were defined as robust associations. These robust edges were used for final network construction and visualization, which was conducted using the igraph v2.2.1 R package with the Fruchterman–Reingold layout.

3. Results

3.1. Plant Growth Responses

Exogenous application of acetoin (AC) or 2,3-butanediol (BD) did not significantly alter plant height or root length in P. heterophylla after 90 days of growth (p > 0.05; Table S1). In contrast, both treatments significantly reduced tuber number and biomass accumulation. The number of tuberous roots per plant decreased to 2.00 ± 0.58 (AC) and 3.00 ± 1.00 (BD), compared with 5.00 ± 1.00 in the control. Fresh and dry weights were correspondingly reduced under both treatments (p < 0.05). These results indicate that VOC exposure specifically affected belowground storage organ development rather than overall plant stature, consistent with the belowground focus of subsequent microbiome and transcriptome analyses.

3.2. Rhizosphere Microbial Diversity and Community Structure

The bacterial communities were dominated by a conserved assemblage of phyla, with Proteobacteria and Actinobacteria consistently comprising the most abundant groups, followed by Acidobacteria, Chloroflexi, Gemmatimonadetes, and Bacteroidetes (Figure S1A). Phylum-level community composition remained largely stable during the flowering (FL) and leaf and stem growth (LS) stages. Proteobacteria reached their highest relative abundance at the FL stage and exhibited moderate redistribution during the root tuber enlargement (RO) stage, with higher proportions observed under 2,3-butanediol (BD) treatment. At the class level, Alphaproteobacteria, Gammaproteobacteria, and Actinobacteria predominated across samples (Figure S1B). Bacteroidia was significantly enriched during the LS stage, whereas Actinobacteria showed a slight increase during the RO stage, particularly under the AC treatment.

Compared with bacterial communities, fungal communities showed a larger magnitude of compositional variation at high taxonomic levels (Figure S1C,D). At the phylum level, all samples were dominated by Ascomycota and Basidiomycota, although their relative abundances varied across the growth stages of P. heterophylla, without the emergence of alternative dominant phyla. In contrast, class-level composition varied markedly, particularly among Agaricomycetes, Sordariomycetes, Dothideomycetes, and Eurotiomycetes, with the greatest divergence observed during the RO stage.

External application of AC and BD influenced rhizosphere microbial α-diversity across the growth stages of P. heterophylla (Figure 1A,B). Analysis of the Shannon index showed no significant overall differences in α-diversity among treatments (AC, BD, and CK) for either bacterial or fungal communities. In contrast, stage-wise comparisons within each treatment revealed significant variations in α-diversity. Across all treatments, microbial α-diversity consistently increased with plant developmental progression. PCoA based on Bray–Curtis distances revealed clear separation of microbial community structures among treatments and growth stages (Figure 1C,D). In bacterial communities, PCoA1 and PCoA2 explained 22.99% and 8.96% of the variance, respectively; PERMANOVA showed no significant treatment effect (p = 0.272, R² = 0.084) but a significant growth stage × treatment interaction (p < 0.001, R² = 0.517). In fungal communities, PCoA1 and PCoA2 explained 30.03% and 13.60% of the variance, respectively; both treatment effect (p = 0.002, R² = 0.173) and interaction effect (p < 0.001, R² = 0.654) were significant.

Complementing the analysis of community membership, differential abundance analysis based on ASV counts revealed significant shifts in the relative abundances of specific taxa in response to the treatments (Figure 2). Relative to the control (CK), AC treatment significantly altered bacterial composition, with 19 ASVs enriched and 9 depleted (|log₂FC| > 1, p < 0.05; Figure 2A). A comparable but less extensive response was observed under BD treatment, which resulted in 12 enriched and 8 depleted bacterial ASVs (Figure 2B). Fungal communities exhibited pronounced shifts in response to both treatments (Figure 2C,D). Compared with CK, AC treatment led to the enrichment of 13 fungal ASVs and the depletion of 17 (Figure 2C). In contrast, BD treatment induced a stronger depletion effect, with 21 ASVs significantly reduced and only 8 enriched (Figure 2D). Taken together, both acetoin and 2,3-butanediol significantly reshaped rhizosphere bacterial and fungal communities, with BD exerting a more pronounced suppressive effect, particularly on fungal ASVs.

3.3. Partitioning of Beta Diversity

This incidence-based analysis reveals that regardless of treatment, the primary mechanism driving differences in community membership (which species are present) is the replacement of species (Repl), rather than simply a loss or gain of species (RichDiff). This indicates a dynamic turnover in community composition. Across all treatments (AC, BD, and CK), BDtotal was primarily governed by Repl, with RichDiff playing a subordinate role (Figure 3). Considerable compositional variation was observed between samples within each treatment group. In bacterial communities (Figure 3A–C), species turnover constituted the dominant mechanism of beta diversity. Under AC treatment, Repl contributed 85.5% of BDtotal versus 14.5% from RichDiff. In BD treatment, Repl accounted for 78.7% compared to 21.3% for RichDiff. While Repl remained predominant in the control (78.0%), RichDiff was somewhat more influential (22.0%) relative to both AC and BD conditions. For fungal communities (Figure 3D–F), RichDiff exerted a comparatively stronger effect, though Repl still prevailed overall. With AC exposure, Repl explained 88.7% of BDtotal, leaving 11.3% to RichDiff. Under BD, RichDiff rose to 16.2%, but Repl remained dominant (83.8%). The control group followed a similar pattern, with Repl contributing 87.3% and RichDiff 12.7%. These results suggest that while species replacement consistently drives community dissimilarity in both kingdoms, fungal assemblages are more sensitive to changes in species richness under BD treatment. In contrast, bacterial communities showed slightly reduced contributions from RichDiff under AC and BD treatments relative to CK, further underscoring the predominant role of species replacement in structuring rhizosphere microbiota under volatile compound exposure.

3.4. Community Assembly Processes

Fitting the neutral community model revealed that bacterial communities exhibited higher Nm and R² values than fungal communities across all treatments (Figure 4). Nm (metacommunity size × immigration rate) reflects dispersal connectivity, with higher values indicating greater potential for stochastic assembly. The R² value represents the proportion of community assembly explained by stochastic processes, with higher values indicating greater neutrality [34,35]. Among treatments, AC yielded the highest neutral fit for both groups (bacteria: Nm = 1254.11, R² = 0.396; fungi: Nm = 1047.07, R² = 0.371). Under BD treatment, bacterial neutrality remained high (Nm = 1684.19, R² = 0.390), whereas fungi showed the lowest fit among all treatment groups (Nm = 609.33, R² = 0.244). Across all treatments, the majority of ASVs (>84%) fell within the model’s 95% confidence interval, while 5–12% occurred more frequently than predicted. This proportion was highest under BD treatment (9.7% of fungal ASVs), indicating a greater departure from neutral expectations compared with AC and CK treatments.

3.5. Root Transcriptional Responses

Transcriptomic analysis confirmed high reproducibility across biological replicates in both treatment groups, reflected by strong pairwise correlations (AC vs. CK: r = 0.957, R² = 0.916; BD vs. CK: r = 0.929, R² = 0.864; Figure 5A,B). Differential expression analysis (|log₂FC| > 1, adjusted p < 0.05) identified 999 and 2493 significantly altered genes under AC and BD treatment, respectively, comprising 413 up- and 586 down-regulated genes under AC, and 1160 up- and 1333 down-regulated genes under BD. These results indicate a broader transcriptional reprogramming in roots exposed to 2,3-butanediol.

KEGG enrichment analysis of down-regulated differentially expressed genes revealed treatment-specific suppression of metabolic pathways. In both AC- and BD-treated roots, “Photosynthesis” and “Cutin, suberine and wax biosynthesis” represented the two most significantly enriched pathways in terms of gene number. Under AC treatment, additional pathways showing notable down-regulation included “Flavonoid biosynthesis”, “Tropane, piperidine and pyridine alkaloid biosynthesis”, “Isoquinoline alkaloid biosynthesis”, “Riboflavin metabolism” and “Galactose metabolism” (Figure 5C). By contrast, BD treatment led to pronounced suppression of pathways associated with “Linoleic acid metabolism”, “Glucosinolate biosynthesis”, “Carotenoid biosynthesis”, “Isoquinoline alkaloid biosynthesis”, “Brassinosteroid biosynthesis” and “Tyrosine metabolism” (Figure 5D).

Among up-regulated genes, AC treatment enriched pathways related to “Taurine and hypotaurine metabolism”, “Arachidonic acid metabolism”, “Flavonoid biosynthesis”, “Stilbenoid, diarylheptanoid and gingerol biosynthesis”, “Vitamin B6 metabolism”, “Neomycin, kanamycin and gentamicin biosynthesis” and “Cyanoamino acid metabolism” (Figure 5E). Conversely, BD-induced up-regulation was most strongly linked to “Polyketide sugar unit biosynthesis”, “Taurine and hypotaurine metabolism”, “DNA replication”, “Pentose and glucuronate interconversions”, “Stilbenoid, diarylheptanoid and gingerol biosynthesis” and “Biosynthesis of various plant secondary metabolites” (Figure 5F).

3.6. Rhizosphere Microbiota–Transcriptome Correlations

PCA showed clear separation among the three treatments, indicating that acetoin and 2,3-butanediol applications altered rhizosphere microbial community composition relative to the control during the tuber enlargement stage (Figure S2). Core discriminant genera were defined based on the highest VIP scores from the first two sPLS-DA components. This analysis identified 17 bacterial and 19 fungal genera contributing most strongly to treatment-level differentiation (Figure 6A,B). Bacterial genera with the highest discriminatory loadings included Janibacter, Acidothermus, Granulicella, Ramlibacter, Phycicoccus, Pseudonocardia, Conexibacter, Stenotrophomonas, and Saccharothrix, whereas Harposporium, Roussoella, Pochonia, Cyphellophora, and Ceratobasidium were the most discriminative fungal taxa.

These 36 discriminative microbial genera were integrated with 2910 root differentially expressed genes (DEGs) from P. heterophylla to construct a cross-omics interaction network. Edges were retained based on stringent correlation criteria (|Spearman’s ρ| > 0.6, FDR-adjusted p < 0.01). The resulting network comprised 70 nodes (11 fungal genera, 19 bacterial genera, and 40 DEGs) connected by 73 significant edges, including 22 bacterium–bacterium, 26 bacterium–transcriptome, 11 fungus–fungus, and 14 fungus–transcriptome associations (Figure 6C). The network exhibited sparse topology (density = 0.0302, average connectivity = 2.09), with fungus–transcriptome edges showing the highest mean absolute correlation coefficient, followed by bacterium–transcriptome (0.958), fungus–fungus, and bacterium–bacterium associations. Several DEGs functioned as highly connected hubs, indicating their central role in integrating microbial–host signals.

Hierarchical clustering of co-expression patterns further resolved modules linking specific microbial consortia with functionally coherent host genes (Figure 6D). Correlation analysis delineated targeted bacterial associations, such as Granulicella and Chthonomonas with NAC6, and Pirellula with both CYP707A2 and USP. Fungal genera displayed broader co-occurrence patterns, including Humicola with CYP707A2 and HSP90; Saccharothrix with LTI60 and WRKY40; Cyphellophora with CYP707A2; Harposporium with HSP90; Ceratobasidium with HSP70; and both Serendipita and Acremonium with NRS.

Developmental and metabolic regulators also showed significant microbial linkages: Granulicella and Chthonomonas correlated positively with NAC6; Chthonomonas, Chryseolinea, and Harposporium with DAR1; and Chryseolinea and Pirellula with CDKC-1. Negative correlations included Humicola with SAG13; Candidatus_Solibacter and Chryseolinea with PHN1; and Nakamurella with CDKC-1. Notably, Chthonomonas exhibited co-positive associations with both NAC6 and DAR1, whereas Chryseolinea displayed a divergent pattern, positively correlated with DAR1 and CDKC-1 but negatively with PHN1. Epigenetic regulators were likewise associated with specific taxa: Acidothermus and Janibacter correlated negatively with DDM1, Pseudonocardia negatively with AFC3, and the fungal genus Acremonium positively with AFC3.

4. Discussion

4.1. VOCs Inhibit P. heterophylla Root Growth

This study showed that AC and BD significantly reduced both the number and biomass of storage roots in P. heterophylla. At first sight, this appears to contrast with reports that bacterial VOCs stimulate shoot growth in Arabidopsis thaliana [35,36]. Rather than indicating a true inconsistency, these results underscore the context dependence of VOC-mediated effects. P. heterophylla is a perennial medicinal plant in which storage roots are the principal harvest organ, and its developmental priorities may respond VOC cues differently than annual model species.

KEGG enrichment indicated that both treatments significantly down-regulated pathways associated with cutin, suberin and wax biosynthesis and with photosynthesis. Cutin and suberin are integral to the endodermal Casparian strip and the exodermal suberin lamellae, which regulate water and solute movement and contribute to the defense against pathogen ingress [37,38]. Reduced biosynthesis of these polymers may weaken barrier properties and selective uptake, potentially contributing to the observed reduction in tuber biomass, though direct measurements of nutrient accumulation and barrier function are needed to confirm this mechanism. The repression of photosynthesis-related genes in roots—potentially reflecting altered assimilate translocation or changes in the expression of homologous genes—also may indicate perturbed carbon allocation between sources (leaves) and sinks (tuberous roots) [39,40]. Such shifts in partitioning represent a plausible mechanism that could contribute to the observed reduction in tuber biomass, a hypothesis that aligns with evidence from potato and warrants further investigation through direct measurements of carbon flux and sugar transport [41].

Treatment-specific metabolic reprogramming further suggests distinct modes of action. Under AC, secondary metabolic pathways including flavonoid and isoquinoline alkaloid biosynthesis were specifically down-regulated. Beyond their antioxidant roles, flavonoids modulate root development, mediate microbe–plant interactions and contribute to stress signaling [42]; their suppression could therefore alter rhizosphere chemistry and microbial recruitment. In contrast, BD induced broader metabolic inhibition, including pathways linked to brassinosteroid and carotenoid biosynthesis and to tyrosine metabolism. Brassinosteroids are key regulators of cell division and elongation [42], and enhanced brassinosteroid signaling can promote the growth of specialized organs [43]. Accordingly, reduced brassinosteroid biosynthesis may constrain cell expansion in storage roots, limiting tuber enlargement in P. heterophylla. Notably, both treatments up-regulated taurine and hypotaurine metabolism and stilbenoid, diarylheptanoid and gingerol biosynthesis. Taurine-related metabolism has been linked to osmotic adjustment and antioxidant responses [44,45], whereas stilbenoids and related compounds often confer antimicrobial activity and stress protection [46]. Thus, despite overall growth inhibition, P. heterophylla tuberous roots appear to activate targeted protective and acclimatory responses to VOC exposure. Thus, AC and BD reconfigured root metabolism in distinct but partially convergent ways, providing a mechanistic framework for the observed reduction in tuber number and biomass.

4.2. Rhizosphere Microbial Community Structure and Assembly Process

Alpha diversity increased throughout plant development, potentially driven by temporal shifts in root exudate composition and flux [47]. Neither VOC addition changed this trajectory, implying that AC and BD did not simply homogenize relative abundances; rather, they likely promoted compositional turnover. This interpretation is borne out by beta-diversity partitioning: for both bacterial and fungal communities, turnover (species replacement) accounted for most among- and within-treatment dissimilarity (contribution > 78%), whereas richness differences contributed comparatively little. Such dominance of turnover accords with evidence that root exudates shape rhizosphere assemblages by selectively enriching some taxa while excluding others [48,49]. For instance, exudate inputs from different cover crops generate distinct microbial metabolic profiles, including enrichment of bacterial nitrite oxidizers and activation of phytohormone-related functions at genus/family levels [50]. Likewise, stage-specific secretion of phenolic acids can recruit certain microbes while deterring others, yielding pronounced compositional change rather than simple species loss [51]. In this light, exogenous AC and BD—serving as additional carbon sources and/or signaling molecules—may have perturbed the native exudate milieu of P. heterophylla, imposing a comparable filter on community membership.

Fungal communities were particularly sensitive to BD, as indicated by the weakest fit to the neutral community model (NCM; R² = 0.244). This reduced fit suggests that, compared to other treatments, deterministic selection processes likely played an enhanced role in shaping the fungal community under BD exposure. This pattern is consistent with reports that some volatiles act as strong environmental filters [52]. VOC blends produced by Bacillus spp., for example, can markedly suppress fungal pathogens such as Sclerotinia sclerotiorum and Fusarium oxysporum [53,54,55]. BD has also been shown to reshape root-exudate profiles, with downstream effects on soil fungal communities [56,57]. Accordingly, we hypothesize that BD may have constrained BD-sensitive taxa in the P. heterophylla rhizosphere while favoring fungi able to tolerate and/or metabolize this compound (e.g., some Ascomycota), thereby potentially strengthening selection and increasing assembly predictability. However, this inference regarding the primacy of deterministic selection requires direct experimental validation, for example, by measuring the inhibitory or promoting effects of varying BD concentrations on isolated representative fungal strains. In contrast, AC-exposed communities—particularly bacterial assemblages—appeared more consistent with neutral dynamics, suggesting that AC may function as a relatively general substrate that relaxes niche structure and elevates stochasticity [58]. This aligns with studies showing that additions of low-molecular-weight organic compounds can increase the relative importance of stochastic processes in community assembly [59,60]. Together, these results indicate that distinct VOCs can differentially shift the stochastic–deterministic balance in the rhizosphere.

The lack of significant α-diversity differences alongside strong β-diversity shifts indicates that VOC exposure drove compositional turnover (species replacement) rather than net richness loss or gain. This pattern arises because species replacement—not richness differences—was the dominant mechanism of community change (accounting for 78–88% of total dissimilarity), allowing VOC treatments to alter which taxa were present while maintaining overall taxon counts. Functional redundancy within microbial communities likely enabled compensatory dynamics, where suppressed taxa were replaced by functionally similar VOC-tolerant species. Additionally, plant developmental stage exerted a strong influence on α-diversity across all treatments, potentially masking treatment effects on richness while β-diversity remained sensitive to compositional restructuring.

4.3. Associations Between Core Microbiota and Host Transcriptional Responses

The microbial–host gene association network developed here provides a quantitative framework that links community ecological variation with host molecular phenotypes, facilitating the detection of tightly co-varying response modules. Among the inferred connections, fungal–transcriptome edges showed the strongest mean correlations. This pattern is consistent with evidence that fungi—especially arbuscular mycorrhizal fungi and endophytes—engage in close physical and molecular interactions with plant hosts and can influence immunity and development via effector proteins or small-molecule signals [61]. For example, Glomus intraradices secretes the effector SP7, which interacts with the plant transcription factor ERF19 to attenuate immune responses and support biotrophic growth [62]. For example, Humicola abundance was positively correlated with the heat shock protein gene HSP90. Beyond its canonical role as a molecular chaperone, HSP90 acts as a central hub in immune signaling by interacting with R proteins and stabilizing their functional conformations [63]. Together, these observations raise the possibility that colonization by particular rhizosphere fungi influences the immune status of P. heterophylla through effects on HSP90-related homeostasis, potentially representing an adaptive adjustment to shifts in microbial community composition.

Bacterial genera including Granulicella correlated positively with the NAC transcription factor NAC6. Members of the NAC family regulate root development, lateral root initiation and diverse stress responses [64,65]. In rice, overexpression of OsNAC6 enhances drought and salinity tolerance but can impose growth penalties [66]. Accordingly, the positive correlation between Granulicella abundance and NAC6 expression suggests a possible link between this bacterial genus and host stress/development pathways, though functional validation is needed.

Notably, several microorganisms were correlated with epigenetic regulators such as DDM1. Epigenetic mechanisms allow plants to modulate gene expression in response to environmental signals, and accumulating evidence indicates that microbes can reshape plant DNA methylation patterns [67]. Rhizobacteria, for instance, can induce locus-specific methylation changes in Solanum nigrum and Phytolacca americana, with consequences for growth and stress resistance [68,69]. In this context, the negative correlations between Acidothermus and Janibacter with DDM1 are consistent with possible microbiome-associated epigenetic modulation, though this hypothesis requires direct testing.

Although correlation analyses based on abundance and expression levels cannot directly establish causality, they provide crucial hypothetical targets for subsequent functional validation. The network clearly reveals that the restructured rhizosphere microbiome following VOC treatment is not an independently changing assemblage but is intricately intertwined with the host’s internal complex molecular response network, forming a coordinated response system. VOCs may initially, either directly or indirectly (by altering root exudates), recruit specific microbial “responders.” These microorganisms, in turn, send feedback signals to the plant through their metabolic activities or physical presence, further modulating key physiological processes in the roots—such as stress response, hormone metabolism, and secondary metabolism—which collectively determine the plant’s growth phenotype.

4.4. Limitations and Future Perspectives

This study has several limitations. First, pot-based experiments cannot fully capture the ecological complexity of field systems; thus, microbiome responses and their yield implications require field validation. Second, while correlation networks reveal strong plant–microbe associations, the molecular mechanisms—particularly specific microbial metabolites involved—remain unresolved. Furthermore, the identification of stable discriminant microbial taxa via sPLS-DA, while enhanced by stability selection, was performed on a limited number of samples. Therefore, these taxa should be viewed as candidate biomarkers whose ecological significance and response to VOCs require validation in future studies with larger, independent cohorts and through targeted experimental manipulation.

Despite these limitations, our findings have applied implications for medicinal plant cultivation. The observed suppression of secondary metabolism pathways (flavonoid, alkaloid, carotenoid biosynthesis) under VOC exposure raises the possibility that soil VOC profiles could influence not only tuber yield but also accumulation of pharmacologically active compounds in P. heterophylla. Future studies should directly quantify secondary metabolite profiles in tubers exposed to VOC treatments. Additionally, the differential sensitivity of fungal communities to BD suggests that soil VOC monitoring could serve as a diagnostic tool for rhizosphere health in continuous cropping systems. If specific VOCs contribute to replant disease, targeted manipulation of soil VOC profiles—through organic amendments or antagonistic microorganisms—might alleviate cropping stress.

Several directions warrant further investigation. Mechanistically, comprehensive metabolomic profiling of root exudates under VOC exposure could identify chemical mediators driving tripartite interactions. Methodologically, combining spatial transcriptomics with microbial imaging could map root-associated microbial communities and their corresponding host transcriptional programs at high resolution. From an applied perspective, systematic monitoring of baseline soil VOCs in P. heterophylla fields subjected to continuous cropping—alongside microbial dynamics and disease incidence—would assess whether manipulating soil VOC profiles can mitigate replant disease.

5. Conclusions

This study demonstrates that two widespread microbial VOCs, acetoin and 2,3-butanediol, differentially influence rhizosphere microbiome assembly and root transcriptional programs in the medicinal plant Pseudostellaria heterophylla. Both compounds suppressed storage root formation, with BD exerting stronger effects on fungal community structure, deterministic assembly processes, and host gene expression—particularly in pathways related to secondary metabolism and stress responses. Correlation networks identified potential links between discriminatory microbial taxa and host functional genes, providing hypotheses for future mechanistic studies. These findings establish a conceptual framework for understanding VOC-mediated rhizosphere dynamics and suggest that soil volatile profiles may influence both the productivity and medicinal quality of cultivated crops.