This experiment was conducted at the P. notoginseng Science and Technology Park in Miao Township, Yanshan County, Yunnan Province, China (104.39° E, 24.71° N). The region features a subtropical plateau monsoon climate with an average annual temperature of 16.1°C and annual precipitation of 1008 mm.

Uniformly cultivated 7-month-old P. notoginseng seedlings were selected and transplanted into pots containing 10 kg of cultivation soil, with a total of 45 seedlings per pot. Eight Cd treatments 0 mg kg^–1(CK), 10 mg kg^–1 (N1P), 20 mg kg^–1 (N2P), 30 mg kg^–1 (N3P), 40 mg kg^–1 (N4P), 60 mg kg^–1 (N5P), 80 mg kg^–1 (N6P), 100 mg kg^–1 (N7P) were applied, each with three replicates. Cd was applied as CdCl₂⋅2.5H₂O solutions uniformly mixed into the soil. During the experimental period, plants were maintained under natural field conditions with supplemental shade netting (70% light transmission) and were watered every 2–3 days to maintain soil moisture at 60–70% of field capacity, with no additional fertilizers applied. The Cd concentration gradient was designed to cover environmentally relevant contamination levels as well as low- and high-dose ranges, enabling assessment of both hormetic and toxic responses.

After 75 days of treatment (from August 15, 2022, to October 31, 2022), P. notoginseng plants were harvested as whole plants, including both aerial parts and roots. The whole plants were thoroughly washed with deionized water. For each treatment, 15 plants were randomly selected for enzyme activity assays, while the remaining plants were used for agronomic trait measurements. After agronomic trait measurements, the stems, leaves, and roots were separated and initially heat-treated at 105°C for 30 min, followed by drying at 65°C for 6 h until reaching a constant weight. The dried samples were then ground into fine powder for Cd content determination.

Fresh soil samples were collected and immediately placed in sterile centrifuge tubes, transported to the laboratory in an icebox, and stored at –80°C for 16S rRNA sequencing. The remaining soil samples were air-dried indoors and sieved through 40-mesh and 100-mesh screens for enzyme activity and cadmium content analysis, respectively.

Agronomic traits were measured using a ruler with a precision of 0.1 cm. Plant height was determined as the length from the cutting site to the petiole base. Leaf length (central leaflet length), leaf width (central leaflet width), and leaf area (calculated as leaf length × leaf width × 0.6134) were recorded. Root fresh weight was measured using an analytical balance with a precision of 0.01 g.

Enzyme activity assays were conducted using commercial assay kits (Beijing Solarbio Science & Technology Co., Ltd., Beijing, China) following the manufacturer’s protocols. Antioxidant enzyme activities in plant tissues, including superoxide dismutase (SOD), peroxidase (POD), and catalase (CAT), were determined using the nitroblue tetrazolium (NBT) photoreduction method, the guaiacol method, and the ammonium molybdate method, respectively. Soil enzyme activities, including acid phosphatase (S-ACP), urease (S-UE), and catalase (S-CAT), were measured using the disodium phenyl phosphate method, the indophenol blue colorimetric method, and the hydrogen peroxide decomposition method, respectively. Enzyme activities were expressed on a fresh weight basis for plant tissues and a dry weight basis for soil samples, according to the instructions provided with each assay kit.

The Cd concentrations in soil and P. notoginseng samples were analyzed using graphite furnace atomic absorption spectrometry (GFAAS) (Beijing Purkinje General Instrument Co., Ltd., Beijing, China). Survival rate was defined as the percentage of plants remaining alive at the end of the 75-day Cd treatment period relative to the initial number of plants per treatment. The bioaccumulation coefficient (BCF) was calculated as the ratio of Cd concentration in individual plant tissues to the corresponding Cd concentration in soil. The translocation coefficient (TC) was calculated as the ratio of Cd concentration in aboveground tissues to that in belowground tissues.

To isolate rhizosphere soil, whole plants were gently uprooted, and loosely adhered soil was removed by carefully shaking the roots. Rhizosphere soil tightly attached to the roots (defined as soil within 1–2 mm of the root surface) was then collected by brushing the root surface with a sterile brush and shaking vigorously for 2 min in 10 mL of phosphate-buffered saline (PBS). The resulting soil suspension was centrifuged at 5,000 rpm for 10 min, and the pellet was used for downstream DNA extraction.

Total genomic DNA was extracted from 0.25 g of rhizosphere soil using the TIANGEN^® Soil DNA Kit (TIANGEN, Beijing, China), following the manufacturer’s protocol based on CTAB/SDS lysis chemistry. This kit has been widely used for microbial DNA extraction from soil samples with high efficiency and purity (Zhou et al., 1996). DNA concentration was measured using a Qubit^® 2.0 Fluorometer (Thermo Fisher Scientific, United States), and quality was verified by 1% agarose gel electrophoresis. DNA was diluted to 1 ng μL^–1 with sterile water for downstream amplification.

The V3–V4 hypervariable region of the bacterial 16S rRNA gene was amplified using the primer pair 341F (5’-CCTAYGGGRBGCASCAG-3’) and 806R (5’-GGACTACNNGGGTATCTAAT-3’), which target a broad range of bacterial taxa. PCR reactions were conducted in triplicate using Phusion^® High-Fidelity PCR Master Mix with GC Buffer (New England Biolabs, United States) (Avershina et al., 2013). PCR products were purified using AMPure XP beads (Beckman Coulter, United States), quantified with a Qubit dsDNA HS Assay Kit (Thermo Fisher Scientific, United States), and pooled in equimolar concentrations. Sequencing libraries were prepared using the TruSeq^® DNA PCR-Free Sample Preparation Kit (Illumina, United States) and sequenced on the Illumina NovaSeq 6000 platform using a paired-end 2 × 250 bp configuration.

Raw reads were quality-filtered, demultiplexed, and processed using QIIME2 (v2023.2). Operational taxonomic units (OTUs) were clustered at 97% sequence similarity using the Uparse algorithm (USEARCH v7), and taxonomy was assigned using the Mothur method against the SILVA 138.1 reference database with a confidence threshold of 0.8. Rarefaction curves were generated to assess sequencing depth. Alpha diversity indices (Chao1, Shannon) and beta diversity metrics (Bray-Curtis, weighted UniFrac) were calculated. Principal coordinate analysis (PCoA) and hierarchical clustering (UPGMA) were used to visualize differences in community composition among treatments. Differential microbial biomarkers were identified and microbial co-occurrence networks were constructed following established analytical frameworks described by Banerjee et al. (2018). Differences in beta diversity among treatments were assessed using PERMANOVA (adonis) based on UniFrac distance matrices (p < 0.05).

Differential microbial taxa among treatment groups were identified using LEfSe (Linear Discriminant Analysis Effect Size), implemented on the Galaxy web platform.¹ The analysis followed the standard pipeline: first, the Kruskal–Wallis rank-sum test was used to detect features with significant differential abundance (α = 0.05); this was followed by unpaired Wilcoxon rank-sum tests for pairwise comparisons. Finally, Linear Discriminant Analysis (LDA) was applied to estimate the effect size of each differentially abundant feature, using an LDA score threshold of 4.0. LEfSe was applied to genus-level relative abundance data, unless otherwise specified. Results were visualized using bar plots and cladograms to highlight the most discriminative microbial biomarkers across groups.

Co-occurrence network analysis was performed using SparCC to assess microbial interactions under cadmium stress. Networks were constructed separately for control (N0P: 0 mg kg^–1 Cd) and high Cd stress (N7P: 100 mg kg^–1 Cd) conditions to evaluate Cd-induced changes in bacterial community interactions. The N6P treatment was selected as it showed the strongest Cd-induced community shifts. Bacterial genus-level OTU data were filtered to include genera with > 0.1% relative abundance in ≥ 50% of samples per treatment. SparCC correlations were computed with 100 bootstrap replicates, and only statistically significant correlations (| ρ| > 0.8, p < 0.05) were retained. Network topology parameters (node degree, betweenness centrality, modularity) were calculated using igraph in R. Networks were visualized in Gephi v0.9.2 with Force Atlas 2 layout. Nodes represent genera (colored by phylum, sized by degree centrality), while edges indicate positive (red) or negative (blue) correlations with thickness reflecting correlation strength. Network topology differences between treatments were quantitatively compared to assess heavy metal impacts on microbial community structure.

Data were analyzed using SPSS 22.0, with one-way ANOVA applied to assess significant differences (p < 0.05). Results were represented as means with corresponding groupings indicated by lowercase letters.

The growth performance of P. notoginseng showed a clear hormetic response to Cd exposure. Plant height, leaf length, leaf width, leaf area, and root fresh weight increased under moderate Cd levels (N2P–N3P; ∼30 mg kg^–1 Cd), but declined markedly under high Cd (N7P) (Table 1 and Supplementary Figure 1). Survival rate followed a similar non-linear trend: moderate Cd (N3P) improved survival to 95.34 ± 0.65%, whereas excessive Cd (N7P) reduced it to 65.07 ± 5.91% (Figure 1). Survival rate, defined as the percentage of plants remaining alive after 75 days of Cd treatment, varied among Cd treatments, with the highest survival observed at the moderate Cd level (N3P) and a marked decline under excessive Cd stress (N7P) (Figure 1).

Cd accumulated primarily in roots, indicating a sequestration strategy to protect aerial tissues (Figure 2a). The BCF decreased with increasing Cd, while the TC peaked at N4P (1.04) before stabilizing (Figure 2b). Antioxidant enzyme activities (CAT, POD, and SOD) showed non-linear and tissue-specific responses to Cd stress, with marked induction at low Cd levels (N1P–N3P) and altered regulation under higher Cd exposure (N5P–N7P) (Figures 2c–e). Collectively, the results indicate that moderate Cd exposure elicits activation of defense-related physiological responses, while excessive Cd stress is associated with growth inhibition and pronounced changes in antioxidant enzyme activity patterns.

Soil enzymatic responses to Cd stress were non-linear and threshold-dependent (Figure 3). S-CAT activity declined steadily across treatments, whereas S-URE and S-ACP increased under moderate Cd (N3P–N5P) before decreasing under higher Cd (N7P). Such biphasic patterns suggest that low Cd may transiently enhance nutrient cycling through stimulated microbial activity, while high Cd suppresses key biochemical processes necessary for soil fertility.

In parallel, microbial alpha diversity indices (Shannon, Chao1, ACE) peaked under moderate Cd and declined sharply at N7P (Figures 3d–g), whereas the Simpson index remained relatively stable. Together, these data suggest that moderate Cd (30–60 mg kg^–1) is associated with enhanced microbial and enzymatic activity but excessive exposure compromises rhizosphere stability and nutrient availability.

Beta-diversity analyses (weighted and unweighted UniFrac) demonstrated distinct community clustering across Cd gradients (Figure 4). Control (CK) samples were clearly separated from Cd-treated groups, and PERMANOVA (adonis) confirmed significant compositional differences among treatments. These findings indicate that These results suggest that Cd-induced community differentiation is largely associated with shifts in relative abundance rather than extensive turnover of major taxa.

Beta-diversity analyses (weighted and unweighted UniFrac) showed clear ordination shifts across the Cd gradient (Figure 4). PERMANOVA (adonis) further indicated moderate-to-large effect sizes of Cd treatment on community dissimilarity across pairwise contrasts (R² = 0.215–0.510 for weighted UniFrac and 0.490–0.785 for unweighted UniFrac; Supplementary Table 1), suggesting that Cd exposure is associated with substantial community differentiation. Overall, the observed separation is consistent with Cd-related changes in community composition, largely reflected by shifts in the relative abundance and/or presence–absence patterns of dominant lineages rather than wholesale replacement of major phyla.

Hierarchical and phylogenetic clustering (Supplementary Figure 2 and Figure 5) further supported this pattern by showing clear differentiation of microbial communities among Cd treatments, with high-Cd samples (N7P) generally tending to separate from the lower-Cd groups. Notably, some variability among biological replicates was observed within certain Cd treatment groups, suggesting heterogeneous community responses under Cd stress. Overall, these results indicate that Cd stress is associated with pronounced restructuring of community composition, which is largely reflected by shifts in the relative abundance of dominant lineages across treatments.

Cd stress altered the rhizosphere bacterial co-occurrence networks of P. notoginseng (Figure 6 and Supplementary Figure 6). Compared with the control, the high-Cd (N7P) network exhibited a higher proportion of negative correlations, with the proportion of negative correlations increasing from 36.7% in CK to 47.2% in N7P, as supported by quantitative network edge statistics (Supplementary Table 2). These quantitative differences indicate altered interaction patterns under Cd stress, consistent with a reorganization of the rhizosphere microbial network.

Along the Cd gradient, positive correlations declined while negative correlations increased, accompanied by higher modularity, indicating weakened connectivity and reduced buffering capacity under high Cd conditions. A global comparison (Supplementary Figure 7) further revealed that several persistent core taxa, particularly Granulicella and Gaiella, maintained high network connectivity across treatments. These taxa have been widely recognized as central members of soil microbial networks (e.g., Banerjee et al., 2018; Herren and McMahon, 2018), although their associations tended to become more antagonistic under Cd stress. Overall, these results suggest that Cd stress may transform rhizosphere microbial networks from more cooperative and redundant structures to increasingly specialized and competitive ones, with potential implications for rhizosphere ecological functioning under heavy metal contamination.

Functional predictions based on PICRUSt2 and FAPROTAX suggested a non-linear response of microbial functional profiles to Cd stress (Figure 7). At low Cd levels (N1P), the relative abundances of major functional categories remained comparable to the control, indicating a potential functional threshold. With increasing Cd exposure (N3P–N5P), subtle variations were observed in the relative abundances of functions associated with carbohydrate metabolism and xenobiotic-related processes, although these changes were not statistically significant, reflecting potential adaptive adjustments in microbial metabolism. Under high Cd stress (N7P), functions associated with environmental sensing and signal transduction tended to decline, whereas core metabolic functions, including energy, amino acid, protein, and xenobiotic metabolism, remained relatively stable across treatments.

Consistent with these trends, PICRUSt2-based statistical analysis showed that carbohydrate and energy metabolism did not differ significantly among treatments (one-way ANOVA, p > 0.05; Figure 7 and Supplementary Table 3), whereas signal transduction exhibited a significant decline with increasing Cd stress (p < 0.05), indicating that microbial environmental sensing and regulatory capacity is more sensitive to Cd exposure than core metabolic potential.

FAPROTAX ecological annotation further supported functional resilience, showing that chemoheterotrophy and aerobic chemoheterotrophy remained stable across the Cd gradient, ensuring continuous carbon turnover. Nitrogen cycling functions displayed adaptive plasticity: nitrogen fixation increased under high Cd stress, while nitrate and nitrogen respiration tended to be higher at low Cd levels, whereas fermentation and parasitic processes were reduced.

Together, these results suggest that functional redundancy and the maintenance of core metabolic functions may contribute to partial functional resilience in the P. notoginseng rhizosphere.

As the first level of Cd perception, plant physiological responses provide the primary driver shaping belowground processes. The present study showed that P. notoginseng exhibited a canonical hormetic response when exposed to Cd: moderate Cd (∼30 mg kg^–1) promoted growth, improved leaf morphological traits, and stimulated antioxidant enzyme activities, whereas high Cd concentrations led to oxidative damage, biomass decline, and physiological inhibition. This pattern aligns with the broader consensus that hormesis is a fundamental response class in plants under environmental contaminants (Calabrese and Baldwin, 2003; Li R. et al., 2025; Saitanis and Agathokleous, 2019). Moderate Cd exposure may activate redox-linked signaling cascades (e.g., MAPK/ROS signaling loops) that prime antioxidant defense and transiently enhance photosynthetic efficiency and resource allocation (Carvalho et al., 2020). However, once Cd input exceeds detoxification capacity, excessive ROS accumulation overwhelms antioxidant buffering, leading to membrane peroxidation, impaired metabolic processes, and growth inhibition, as previously observed in other medicinal crops (Wang et al., 2023; Chen et al., 2024).

The strong root-biased Cd accumulation observed here suggests that P. notoginseng employs a retention strategy in which the roots function as the primary sequestration compartment, thereby restricting the translocation of Cd into aerial tissues, a trait of considerable economic importance given the medicinal use of its underground organs. Similar spatial partitioning of heavy metals between roots and shoots has been reported in Arabidopsis and other Cd-tolerant species (Lux et al., 2011; Ghori et al., 2019; Jia et al., 2020). At moderate Cd (∼30 mg kg^–1), SOD, POD and CAT activities were markedly elevated, demonstrating that an effective stress-induced activation of antioxidant defense enzymes operates in P. notoginseng. Under high Cd however, these enzymes declined, consistent with enzymatic inactivation, protein carbonylation, and redox-homeostasis collapse (Lu et al., 2023; Mahmood et al., 2025). This exhaustion phase appears to coincide with a physiological transition where Cd enters beyond-threshold bioavailability, disrupting metabolic efficiency even in the presence of a rhizosphere microbial buffering effect.

Root-based Cd sequestration and antioxidant regulation not only determine plant tolerance thresholds but also modulate the chemical and biological conditions of the rhizosphere. Although hormesis is classically defined at the level of organismal physiology, the present results suggest that plant hormetic responses to moderate Cd stress may be accompanied by coordinated shifts in rhizosphere enzyme activities and microbial community structure. Importantly, these microbial and enzymatic responses are interpreted as coupled feedbacks to plant stress and altered root exudation, rather than as direct evidence of microbial hormesis per se.

The nonlinear responses of rhizosphere enzymes and microbial alpha diversity indicate that Cd modulates belowground processes via threshold-dependent biochemical and ecological feedbacks. Moderate Cd (∼30 mg kg^–1) enhanced urease and acid phosphatase activities and increased Shannon and Chao1 diversity, implying a transient acceleration of nutrient cycling, likely mediated by shifts in root exudation and stress-induced recruitment of beneficial microorganisms (Li N. et al., 2025; He et al., 2024). Rhizosphere network studies have shown that the architecture of trophic interactions largely determines the capacity of bacterial communities to buffer perturbation and resist pathogen invasion (Wei et al., 2020). However, high Cd sharply reduced enzyme activity and microbial richness, indicating that once toxicity exceeds functional buffering capacity, facilitative interactions weaken and rhizosphere metabolic turnover becomes impaired. The observation that beta diversity separation occurred mainly through abundance redistribution rather than phylum-turnover further supports the view that rhizosphere microbiomes are compositionally resilient yet functionally plastic only within definable tolerance boundaries (Griffiths and Philippot, 2013).

Co-occurrence network analyses further revealed that Cd stress shifted rhizosphere microbial networks from cooperative-redundant architecture to more modular and competitive configurations, accompanied by reduced ecological connectivity. The densified positive interactions observed in control soils, which are characterized by cooperative taxa such as Granulicella and Gaiella, suggest the presence of a stable and metabolically buffered core network. These taxa are widely recognized as dominant and functionally important members of soil microbial communities, with Acidobacteria (including Granulicella) actively participating in carbon turnover and energy metabolism and contributing to soil organic matter processing (Kalam et al., 2020; Giguere et al., 2021; Delgado-Baquerizo et al., 2018). In addition, Actinobacteria, including Streptomyces, are well known for their roles in secondary metabolite production and nutrient cycling in soils, thereby supporting ecosystem functioning and resilience under non-stressed conditions (van der Meij et al., 2017).

Under high Cd stress, negative correlations increased and stress-tolerant hubs became more prominent, while PGPR genera such as Pseudomonas and Bacillus were selectively enriched, highlighting their well-characterized roles in Cd tolerance and metal immobilization (Rajkumar et al., 2009; Sun et al., 2022; Kang et al., 2024). Despite this apparent structural simplification, functional prediction indicated that key processes such as carbon turnover and nitrogen fixation remained largely maintained, demonstrating strong functional redundancy embedded within the rhizosphere system even when taxonomic and network structures were perturbed by Cd stress.

Collectively, these findings indicate that P. notoginseng operates under a coupled plant–microbe stress buffering mechanism: moderate Cd triggers beneficial feedback loops that enhance physiological performance and microbial facilitation, whereas excessive Cd breaks network connectivity, reduces cooperative interactions, and leads to functional decline. The preferential enrichment of Pseudomonas and Bacillus suggests that these taxa may play contributory roles in plant stress mitigation under Cd exposure and highlights their relevance as candidate PGPR for future investigation and experimental validation in Cd-affected soils (Kumar et al., 2023; Ding et al., 2025).