Two grams of soil samples (moist soil, 35°03′N, 112°61′E) collected from Jiyuan city, Henan Province, were added to a sterile 50 mL shake flask and shaken to prepare a soil suspension. Soil properties: 1.37 mg kg⁻¹ Cd, 97.6 mg kg⁻¹ Pb, pH 7.42, 23.56 g kg⁻¹ organic matter, 0.64 g kg⁻¹ available P, 1.45 g kg⁻¹ exchangeable Ca and 38.2 cmol(+) kg⁻¹ cation-exchange capacity. A 0.1 mL aliquot of each gradient dilutions was spread onto a solid nitrogen-containing medium plate using a coating rod (Zhang H. et al., 2024). The plates were incubated at 30°C for 5 days. Colonies with distinct morphological characteristics were isolated and purified to complete the preliminary screening of EPS-producing bacteria. The bacterial culture in the logarithmic growth phase was mixed with sterilized 80% glycerol (80 mL pure glycerol + 20 mL sterile water) at a volume ratio of 1:1 and stored in a − 80°C freezer for preservation.

Fifty milliliters of LB medium supplemented with 5 mg L⁻¹ Cd (Cd(NO₃)₂) and 10 mg L⁻¹ Pb (Pb(NO₃)₂) was prepared. The growth container was 250 mL conical flask (50 mL culture solution). A bacterial suspension (OD₆₀₀ = 1) was inoculated into the medium at a 2% ratio and incubated in a shaker at 30°C and 180 rpm for 48 h. The concentrations of Cd and Pb in the supernatant were measured via inductively coupled plasma atomic emission spectrometry (ICP–AES, ICPE-9820, Japan). The polysaccharide content in culture solution was determined using the sulfuric acid-anthrone colorimetric method (Wang et al., 2016). The strains were sequenced and identified via 16S rRNA analysis (Teng et al., 2019). Heavy metal resistance studies were conducted with Cd (50–500 mg L⁻¹, in 50 mg L⁻¹ increments) and Pb (1000–1800 mg L⁻¹, in 100 mg L⁻¹ increments) to determine the lethal concentration (LC₅₀, refers to the concentration of a chemical substance in the environment that causes the death or loss of metabolic activity in organisms). The effects of antibiotics on strain growth were also investigated (Subbaram et al., 2017). The determination of the ability of bacterial strains to secrete indole-3-acetic acid (IAA) was based on method of Jiang et al. (2008). The production of siderophores and 1-amino-1-cyclopropanecarboxylic acid (ACC) deaminase by the strains were determined according the approach proposed by Rajkumar et al. (2006) and Chretien et al. (2024).

Pseudomonas sp. H7 and Agrobacterium sp. Z22 were inoculated into LB liquid medium supplemented with 5 mg L⁻¹ Cd and 10 mg L⁻¹ Pb for a 9-day shake flask experiment. Foue concentrations of Cd²⁺ and Pb²⁺ (10, 20, 50, and 100 mg L⁻¹) were tested. Three treatment groups were established: a control group (CK), an experimental group inoculated with Pseudomonas sp. H7 (H7), and an experimental group inoculated with Agrobacterium sp. Z22 (Z22). Samples were collected on days 1, 3, 5, 7, and 9. The total heavy metal content in the culture medium (H₁) was calculated as the product of the total culture volume and the initial heavy metal concentration. For the heavy metal content in the supernatant (H₂), the culture was centrifuged, and the supernatant was collected for analysis using ICP-AES; H₂ was then determined by multiplying the supernatant’s heavy metal concentration by the total culture volume. To quantify the intracellular heavy metal content (H₃), the pelleted bacterial cells were washed three times with 5 mL of 10 mmol L⁻¹ EDTA-2Na solution, freeze-dried, and weighed. The dried cells were digested with 3 mL HNO₃ and 1 mL HCl, and the heavy metal concentration in the digestate was measured via ICP-AES. The extracellular heavy metal content (H₄) was derived by subtracting H₂ and H₃ from H₁ (H₄ = H₁ − H₂ − H₃) (Han et al., 2020).

Cell pellets were collected and fixed in 15 mL of 2.5% glutaraldehyde at 30°C for 3 h. The pellets were then dehydrated using a gradient of absolute ethanol. After drying and gold coating, the samples were analyzed using scanning electron microscopy coupled with energy-dispersive X-ray spectroscopy (SEM-EDS) (Zhu et al., 2016). Changes in surface functional groups were analyzed using Fourier-transform infrared spectroscopy (FTIR) (Chakravarty and Banerjee, 2012). X-ray diffraction (XRD) analysis was performed at a scanning speed of 2° min⁻¹ and a scanning angle of 5–100° (Chakravarty and Banerjee, 2012). Changes in the chemical forms of elements before and after heavy metal stress were characterized using X-ray photoelectron spectroscopy (XPS) (Zhang et al., 2022).

Soil samples (moist soil, 35°03′N, 112°61′E) were collected from farmland near a factory in Jiyuan City, Henan Province. Soil properties: 1.37 mg kg⁻¹ Cd, 97.6 mg kg⁻¹ Pb, pH 7.42, 23.56 g kg⁻¹ organic matter, 0.64 g kg⁻¹ available P, 1.45 g kg⁻¹ exchangeable Ca and 38.2 cmol(+) kg⁻¹ cation-exchange capacity. Each pot was filled with 4 kg of soil sieved through a 2 mm mesh. Four treatment groups were established: a control group (CK), groups inoculated with strain Pseudomonas sp. H7 (H7) or Agrobacterium sp. Z22 (Z22), and a group inoculated with both strains (H7 + Z22). Pakchoi seeds were sown, and after germination, seedlings were thinned to five per pot. 40 mL bacterial suspension (OD₆₀₀ = 1.0, 1×10⁸ CFU mL⁻¹) was added to the rhizosphere soil. The control group was added with the same volume of sterile deionized water. The experiment lasted 50 days. Soil samples were collected at depths of 5–15 cm, and the heavy metal content and pH of soil pore water and leachate were measured on days 0, 15, 30, and 50. Leachate was collected from the bottom of the pots, and pore water was extracted using a Rhizon MOM soil solution sampler (AgriEco Apptec (Shanghai) LLC, China). The pH of leachate and pore water was measured using a pH meter, and heavy metal concentrations were determined using ICP-AES. After washing the mature pakchoi, the edible parts were separated from the roots. The roots were soaked in 0.01 mmol L⁻¹ EDTA-2Na solution for 10 min to remove the adsorbed heavy metals on the surface. After being washed with deionized water, the whole plant was dried at 80°C to a constant weight, and the biomass of each part was measured. After crushing, 0.1 g of the sample was weighed into a polytetrafluoroethylene crucible. Mixed acid (HNO₃-HCl-HClO-HF) was added at a ratio of 4.5:1.5:2:2. The temperature was raised for digestion until nearly dry, and the volume was made up to 5 mL. The contents of Cd and Pb were determined by ICP-AES. The vitamin C content of pakchoi was determined by 2,4-dinitrobenzhydra (DNP) method (Koo et al., 2024). The soluble protein content was determined by coomassie brilliant blue method (Masih et al., 2002).

The water stability of soil aggregates was evaluated using a wet sieving method (Cheng et al., 2023). Microaggregates (<250 μm) and macroaggregates (>250 μm) were collected using a soil aggregate analyzer, and their dry weights were measured. EPS in soil aggregates were extracted using a cation exchange resin (CER) (Vardharajula and Shaik, 2014), and polysaccharide content was determined using the sulfate-anthrone method (Wang et al., 2016). Two grams of soil aggregates were mixed with 5 mL of deionized water, and the pH of the supernatant was measured using a pH meter. Organic matter content was determined using the potassium dichromate oxidation method (Wang et al., 2024a).

Five grams of soil aggregates were mixed with 25 mL of extraction solution (1.967 g diethylenetriamine pentaacetic acid, 13.3 mL triethanolamine (TEA), 1.11 g anhydrous calcium chloride, and 950 mL water, pH 7.3). The supernatant was digested with 3 mL nitric acid and 1 mL hydrochloric acid and determined Cd and Pb concentrations by ICP-AES. Tessier’s sequential extraction method was used to determine the contents of exchangeable Cd/Pb (EX-Cd/Pb), carbonate-bound Cd/Pb (CB-Cd/Pb), Fe-Mn oxide-bound Cd/Pb (Fe-Mn-Cd/Pb), organic matter-bound Cd/Pb (OMB-Cd/Pb), and residual Cd/Pb (RES-Cd/Pb) in rhizosphere soil (Tessier et al., 1979). The Cd and Pb contents in these extractions was also determined by ICP-AES.

The morphology of soil aggregates was analyzed using a JSM-7900F scanning electron microscope. The interaction between metal ions and EPS was investigated using an Aqualog fluorescence spectrophotometer to measure the 3D-EEM spectra of EPS (Peng et al., 2016).

Bacterial community analysis was performed on soil from fresh large aggregates (control group: CK-B; experimental groups: H7-B, Z22-B, H7 + Z22-B) and microaggregates (control group: CK-S; experimental groups: H7-S, Z22-S, H7 + Z22-S). Microbial DNA was extracted from soil aggregates using the E. Z. N. A.® soil DNA Kit (Omega Bio-tek, Norcross, GA, United States) according to manufacturer’s protocols. The V3-V4 region of the 16S rRNA gene was amplified using primers 338F (5′- ACTCCTACGGGAGGCAGCAG-3′) and 806R (5’-GGACTACHVGGGTWTCTAAT-3′). All samples were mixed with PCR products, and then subjected to electrophoresis on a 2% agarose gel. The gel was cut using AxyPrepDNA Gel Recovery Kit (AXYGEN Company) to recover the PCR products. These products were then quantified using the QuantiFluor™ -ST Blue Fluorescent Quantitative System (Promega Company). Subsequently, samples were mixed in proportion based on their sequencing requirements, followed by library construction, and finally sequenced at higher levels (Liang et al., 2022). Sequencing results were analyzed on the Meiji Biotech website.¹

Data were analyzed using Excel 2019 and SPSS 26.0. Mathematically processed results are presented in the form of M ± SE. Before performing Tukey’s multiple comparison test, Levene’s test was applied to assess the homogeneity of variances across treatment groups (significance level α = 0.05). Origin 2024 and Excel software were used for image processing. Advantage software was used for XPS data analysis and Matlab 2019a software was used for 3D fluorescence spectroscopy analysis. PCA analysis (Principal Component Analysis, R language 3.3.1) was used for the differences among samples of multiple sets of data. UPGMA (Unweighted Pairing-Group Method with Arithmetic Mean, Qiime 2020.2.0) is a clustering analysis method used to solve classification problems. LEfSe² is based on the taxonomy of the samples according to the different conditions of grouped linear discriminant analysis (LDA).

Eight bacterial strains were selected based on their ability to adsorb heavy metals and produce EPS. The Cd removal rates of these strains ranged from 64.87 to 86.29%, while the Pb removal rates ranged from 56.66 to 86.84% in solutions containing 5 mg L⁻¹ Cd and 10 mg L⁻¹ Pb (Supplementary Figure S1). The EPS production of these strains varied between 147.63 and 267.48 mg L⁻¹, with strains H7 and Z22 exhibiting EPS contents of 183.71 mg L⁻¹ and 267.48 mg L⁻¹, respectively (Figure 1). Consequently, strains H7 and Z22 were chosen as the target strains for further investigation. Based on phylogenetic analysis, strain H7 was identified as Pseudomonas sp. (PP784325), while strain Z22 was identified as Agrobacterium sp. (PP784326) (Supplementary Figure S2). The lethal concentrations (LC₅₀) of Cd and Pb for strain H7 were determined to be 400 mg L⁻¹ and 1700 mg L⁻¹, respectively, whereas the LC values for strain Z22 were 300 mg L⁻¹ and 1,600 mg L⁻¹, respectively (Supplementary Table S1).

By the seventh day of the experiment, strains H7 and Z22 significantly (p < 0.05) reduced the Cd concentrations in the solution by 68.2 and 53.6%, respectively, compared to the control (CK) group (Supplementary Figure S3a). Similarly, the Pb concentrations were significantly (p < 0.05) reduced by 74.6 and 49.8%, respectively (Supplementary Figure S3b). At low heavy metal concentrations (10 mg L⁻¹ and 20 mg L⁻¹), both strains primarily reduced heavy metals through intracellular enrichment (Supplementary Figures S3c,d). However, when exposed to higher concentrations of heavy metals (50 mg L⁻¹ and 100 mg L⁻¹), extracellular adsorption by strains H7 and Z22 became more prominent than intracellular enrichment (Supplementary Figures S3e,f).

Under conditions without heavy metal stress, the surfaces of strains H7 and Z22 appeared smooth (Figures 1a,b). However, in the presence of Cd and Pb, their surfaces became rough, with visible precipitates forming (Figures 1c,d). FTIR analysis revealed shifts in the peaks near 3,288 cm⁻¹ (O-H, N-H), 1,659 cm⁻¹ (C-H), and 1,067 cm⁻¹ (C=O) in the cell walls of strains H7 and Z22 after Cd and Pb adsorption, compared to cells without heavy metal exposure (Figure 1e). Specifically, the absorption peaks of C-H and C=O groups in strains H7 and Z22 shifted by 35 cm⁻¹ and 6 cm⁻¹, respectively, with strain H7 exhibiting more pronounced shifts than strain Z22 (Figure 1e). These results suggest that O-H, N-H, C-H, and C=O groups were involved in the immobilization of Cd and Pb. Furthermore, XRD analysis detected the presence of Fe₂Pb(PO₄)₂, CdCO₃, and Pb₂O₃ on the cell walls of strains H7 and Z22 under Cd and Pb stress (Figure 1f). Additionally, XPS analysis identified new peaks for Cd3d₃/₂, Cd3d₅/₂ (CdS), and metallic Cd in the Cd3d spectrum, as well as peaks for Pb4f₅/₂ and Pb4f₇/₂ (Pb₃O₄ and 2PbCO₃·Pb(OH)₂) in the Pb4f spectrum (Supplementary Figure S4). These findings indicate that strains H7 and Z22 facilitated the formation of precipitates such as CdS and Pb₃O₄.

Compared to the CK group, the inoculation with H7, Z22, and H7 + Z22 significantly (p < 0.05) increased the dry weight of the edible parts (56.1–81.8%) and roots (8.1–55.4%) of pakchoi (Figure 2a). Inoculation also led to a reduction in Cd (30.7–68%) and Pb (31.1–57%) content in the edible parts, as well as Cd (27.9–47.2%) and Pb (39.8–57.7%) content in the roots of pakchoi (Figure 2b). In the absence of inoculation, the soluble protein content in the edible parts of pakchoi was 7.63 mg g⁻¹. After inoculation with H7, Z22, and H7 + Z22, the soluble protein content significantly (p < 0.05) increased by 22.4, 12.6, and 32%, respectively (Figure 2c), while the vitamin C content increased by 29.6, 12.5, and 38.2%, respectively (Figure 2d). These results demonstrate that inoculation with polysaccharide-producing bacteria not only enhanced the growth of pakchoi but also improved its nutritional quality.

As the cultivation period progressed, the Cd and Pb content in the pore water of the H7, Z22, and H7 + Z22 treatment groups were significantly lower than those in the CK group (Figures 3a,b), indicating that strains H7 and Z22 effectively immobilized heavy metals and reduced their bioavailability. The pH of the pore water in the control group remained stable, whereas in the H7, Z22, and H7 + Z22 groups, it initially decreased from 7.68 to 7.02 and then increased to 8.12 (Figure 3c). Similarly, the Cd and Pb content in the soil leachate of the treatment groups were significantly lower than those in the CK group over time (Figures 3d,e). The pH of the soil leachate in the CK group showed no significant change, while in the treatment groups, it decreased from 7.54 to 6.76 and then increased to 7.79 (Figure 3f). These findings suggest that polysaccharide-producing bacteria influenced the pH dynamics of the soil leachate, thereby enhancing heavy metal immobilization in the soil.

Inoculation with strains H7 and Z22 significantly (p < 0.05) increased the proportion of macroaggregates while reducing the proportion of microaggregates compared to the control (Supplementary Figure S5a). In the H7 + Z22 group, the polysaccharide content in macroaggregates increased from 2.7 mg kg⁻¹ to 9.8 mg kg⁻¹, and in microaggregates, it increased from 2.8 mg kg⁻¹ to 12.3 mg kg⁻¹ (Supplementary Figure S5b). Additionally, inoculation with strains H7 and Z22 significantly increased the pH of soil aggregates across different particle sizes but had no significant effect on organic matter content (Supplementary Figures S5c,d). Overall, these results indicate that inoculation with strains H7 and Z22 enhanced the polysaccharide content and heavy metal retention capacity of soil aggregates.

In the CK group, the DTPA-extractable Cd content in macroaggregates was 0.047 mg kg⁻¹, while in microaggregates, it was 0.044 mg kg⁻¹. Inoculation with strains H7 and Z22 significantly reduced the DTPA-Cd content in both macro- and microaggregates (Supplementary Figure S6a). Similarly, the DTPA-Pb content in soil aggregates was also reduced following inoculation with strains H7 and Z22 (Supplementary Figure S6b), indicating that microaggregates exhibit a greater capacity for immobilizing Cd and Pb. Over time, the EX-Cd content decreased significantly in the H7- and Z22-inoculated groups, while the Fe-Mn-Cd and RS-Cd contents increased. In the H7 + Z22 treatment, the proportion of Fe-Mn-Cd increased from 18.3 to 29.4%, and the proportion of RS-Cd increased from 21.3 to 29.1% (Supplementary Figure S6c). These results suggest that strains H7 and Z22 facilitated the transformation of bioavailable Cd in macroaggregates into Fe-Mn oxide-bound and residual forms. Additionally, in microaggregates, the proportion of OM-Cd increased from 18.6 to 27.1%, and the proportion of RS-Cd increased from 23.3 to 31.3% in the H7 + Z22 treatment (Supplementary Figure S6c), indicating that strains H7 and Z22 promoted the conversion of bioavailable Cd into organic matter-bound and residual forms. Similar trends were observed for Pb, with strains H7 and Z22 inducing the transformation of bioavailable Pb in macroaggregates into Fe-Mn oxide-bound and residual forms, and in microaggregates into organic matter-bound and residual forms (Supplementary Figure S6d). In microaggregates, the percentage of C-C fitting peaks decreased, while the percentages of C-O-C, C=O, and HCO₃⁻ fitting peaks increased. Additionally, precipitates such as CdCO₃, PbCO₃, Cd₂(OH)₂CO₃, and 2PbCO₃•Pb(OH)₂ were detected, indicating that C-O-C, C=O, and HCO₃⁻ groups were involved in the immobilization of heavy metals in microaggregates (Figure 4; Supplementary Figure S7). Compared to the control, soil macroaggregates inoculated with H7 and Z22 exhibited denser particle aggregates with smoother surfaces, while microaggregates became looser with rougher surfaces and larger specific surface areas, providing more adsorption sites. This suggests that microaggregates have a stronger capacity for adsorbing Cd and Pb (Figure 5). Previous studies have demonstrated that three-dimensional fluorescence intensity is closely associated with EPS content (Ma et al., 2018; Rigby and Smith, 2020). The fluorescence intensity of microaggregates in the H7 and Z22 treatments was significantly higher than that of macroaggregates, indicating a greater EPS content in microaggregates (Figure 5).

The UPGMA algorithm revealed that in macroaggregates, the CK group and the H7 and Z22 inoculation groups clustered on the same branch, whereas in microaggregates, they formed distinct clusters (Figure 6a). Principal component analysis (PCA) further supported these findings, showing that the H7 and Z22 inoculation groups were closer to the CK group in macroaggregates but more distant in microaggregates (Figure 6b). Inoculation with strains H7 and Z22 increased the relative abundance of Proteobacteria, Acidobacteriota, and Actinobacterota in microaggregates while reducing the relative abundance of Chloroflexi and Myxococcota (Figure 6c). At the genus level, the dominant taxa included RB41, Bacillus, Sphingomonas, Gaiella, MND1, and Nocardioides. Following treatment with H7 and Z22, the relative abundance of Sphingomonas, Gaiella, and Nocardioides in microaggregates increased significantly (Figure 6d). In microaggregates, the key bacterial groups in the H7 + Z22 treatment included f_Planococcaceae,g_Arthrobacter, o_Rhodobacterales, f_Beijerinckiaceae, g_Microvirga, and g_Paracoccus (Figure 7). Compared to macroaggregates, microaggregates exhibited a greater number of significantly different bacterial populations after inoculation with H7 and Z22, indicating a more pronounced impact of these strains on bacterial community composition in microaggregates. Previous studies have reported that Sphingomonas possesses the ability to degrade various heavy metals and promote plant growth (Bin, 2011; Krishnan et al., 2016; Tangaromsuk et al., 2002). Additionally, Saccharimonadales abundance has been linked to polysaccharide content and exhibits synergistic effects with nitrogen cycling-related genes (Wang et al., 2022).