This study is divided into two parts. Initially, Mendelian randomization analysis was conducted to explore the causal relationship between gut microbiota and diabetic neuropathy. Subsequently, 16S rRNA gene sequencing was performed on fecal samples obtained from diabetic neuropathy patients and mice. The comparative analysis of microbial community structures discerned differential bacterial abundances, which served as external validation for the results of the Mendelian randomization analysis.

SPF-grade male C57BL/6J mice (6–8 weeks of age, body weight 21–27 g) were obtained from the Guangdong Medical Experimental Animal Center (License No. (2019) 05073). Mice were housed in a ventilated facility under controlled environmental conditions: temperature 21 ± 2 °C, humidity 60 ± 10%, a 12-hour light/dark cycle, with ad libitum access to food and water.

The study was conducted in accordance with the Declaration of Helsinki and was approved by the Medical Ethics Committee of the Third Affiliated Hospital of Sun Yat-sen University (No.2021-02-021-01, ChiCTR2100051493). Written informed consent was obtained from all participants prior to the survey. All animal experiments were approved by the Laboratory Animal Ethics Committee of South China Agricultural University (Approval No. 2021D081).

Following a 1-week adaptive acclimation period, mice were randomly allocated to three groups via a random number generator: control group (Con group, n=6), diabetes (DM group, n=10), and diabetic peripheral neuropathy (DPN group, n=10). Following 4 weeks of high-fat diet (HFD) feeding, DPN group received intraperitoneal injections of streptozotocin (STZ, 60 mg/kg) for 5 consecutive days, prepared in sodium citrate buffer (pH = 6.0) (23). DM group were subjected to 4-week HFD feeding followed by intraperitoneal STZ administration at 30 mg/kg for 5 days (24). Control group were fed a standard chow diet for 4 weeks and injected intraperitoneally with an equivalent volume of sodium citrate buffer for 5 consecutive days.

On days 7, 14, and 21 post-treatment, blood glucose levels were measured via tail vein blood sampling, and behavioral tests were conducted. The endpoint was set at 21 days after treatment, where mice in the DM group with random blood glucose levels ≥16.7mmol/L were selected. For the DPN group, in addition to blood glucose levels, a thermal withdrawal latency (TWL) of less than 20 seconds and log₁₀(50%MWT*10000) decrease ≥ 0.2 from baseline were considered indicative of successful model establishment (24–26). The detailed experimental design is elucidated in Figure 1 .

Behavioral assessments included TWL and mechanical withdrawal threshold (MWT) measurements, performed using a hot plate apparatus and Von Frey filaments, respectively.

Gut microbiota data were sourced from the NHGRI-EBI GWAS Catalog (https://www.ebi.ac.uk/gwas/), specifically selecting 473 microbial taxa ranging from GCST90032172 to GCST90032644 (27). This dataset was generated from a comprehensive genome-wide association study, assessing 7,979,834 human genetic variants in the FINRISK 2002 cohort, comprising 5,959 individuals. The participants in the dataset were aged 24 to 74, with an average age of 49.6, and were drawn from six regions across Finland, comprising 55.1% females and 44.9% males.

The genetic data of DPN patients were obtained from the FinnGen R11 database (https://www.finngen.fi/en). The GWAS dataset comprises 54,913 adult participants, including 3,503 cases and 51,410 controls from a prospective cohort study involving the European population Cases were defined by the presence of ICD-10 codes E10.4, E11.4, E12.4, E13.4, or E14.4. As the initial GWAS data had previously been granted approval by the relevant ethical and institutional review boards, ethical clearance for its use in this study was not required.

As the initial GWAS data had previously been granted approval by the relevant ethical and institutional review boards, ethical clearance for its use in this study was not required.

A total of 38 patients were enrolled between 2021 and 2023 from the Third Affiliated Hospital, Sun Yat-sen University, and allocated into three groups: 13 in the DM group, 15 in the DPN group, and 10 in the Con group. Detailed patient information was retrieved from the management information system and the medical records department of the Third Affiliated Hospital of Sun Yat-sen University. Inclusion criteria for diabetes mellitus were that the patients (i) met the diagnostic criteria for T2DM with random plasma glucose ≥11.1 mmol/L (200 mg/dL), (ii) Fasting plasma glucose (FPG) ≥7.0 mmol/L (126 mg/dL), (iii) 2-hour plasma glucose (2hPG) ≥11.1 mmol/L (200 mg/dL) during an oral glucose tolerance test (OGTT). Inclusion Criteria for Diabetic Neuropathy Patients were that the patients (i). Diagnosis of diabetes mellitus or prediabetes, (ii) Presence of peripheral neuropathy, confirmed by clinical examination and nerve conduction studies, (iii) Exclusion of other causes of neuropathy, including chemotherapy, infections, toxins, etc. Exclusion criteria were that the patients (i)Pregnant or lactating women (ii) Presence of acute diabetic complications (e.g., diabetic ketoacidosis, hyperosmolar hyperglycemic state, severe infection, or acute stress) (iii) Use of antibiotics, prokinetic agents, or prebiotics within 4 weeks prior to fecal sampling (iv) History of gastrointestinal diseases within 4 weeks prior to sampling (v) Gastrointestinal surgery within 6 months prior to enrollment (vi) Concurrent diagnosis of other organic diseases.

Human Sample Collection: Fecal collection tubes were used to collect approximately 10g of fresh fecal matter from the enrolled subjects, focusing on the inner layer of the midsection stool. During collection, urine and toilet bowl walls were avoided. Samples were stored at -80°C within 2 hours and labeled in advance with the patient’s name, gender, age, hospital admission number, and collection time.

Mice Sample Collection: Mice were placed in a clean transparent box lined with sterile filter paper to await defecation. Fecal samples were rapidly collected using sterile cryovials, with 3–5 pellets collected per mouse. In cases of insufficient fecal output or failure to defecate, mice were gently stimulated near the anus with a cotton swab to induce defecation.

Collected samples were immediately frozen in liquid nitrogen and later stored at -80°C.Fecal DNA was extracted using the MagPure Soil DNA LQ Kit (D6356-02, Magen). 100–200 mg of fecal samples were homogenized with 500 mg glass beads in 2-ml tubes containing 0.9 ml Buffer SOL and 90 μl Buffer SDS via bead beating, followed by bacterial lysis at 70 °C for 10 min. After centrifugation at 12,000×g for 1 min, 800 μl of the supernatant was transferred to 1.5-ml tubes, mixed with 150 μl Buffer PS, and centrifuged again at 12,000×g for 5 min. The resulting 450 μl supernatant was subjected to automated purification using the KingFisher Flex platform with the pre-programmed MagPure StoolDNA KF protocol. Extracted DNA was quantified by Qubit dsDNA BR Assay, validated via 1% agarose gel electrophoresis, sealed, and stored at -25 °C.

The V3–V4 hypervariable regions of 16S rDNA were amplified using primers 343F (5’-TACGGRAGGCAGCAG-3’) and 798R (5’-AGGGTATCTAATCCT-3’) through a two-round PCR approach. The first round included a 5-min denaturation at 94 °C, 26 cycles of 94 °C for 30 s, 56 °C for 30 s, 72 °C for 20 s, and a 5-min extension at 72 °C. After purification, the second round was performed with 7 cycles under identical conditions. Final PCR products were purified using Ampure XP beads, eluted, barcoded, and sequenced on an Illumina MiSeq platform with paired-end reads.

Raw paired-end sequencing reads were filtered to obtain clean data, which were then assembled using FLASH software. Subsequently, DADA2 in QIIME2 (2020.11) was employed for denoising, chimeric sequence removal, and amplicon sequence variant (ASV) generation.

For the analysis of gut microbial diversity, both alpha and beta diversity were assessed. Alpha diversity, which reflects species richness within a microbial community, was evaluated using the Chao1 index and the Shannon index. Beta diversity, which represents differences in microbial composition between groups, was analyzed via Principal Coordinates Analysis (PCoA).

Additionally, we analyzed the gut microbiota at the phylum level. The top ten most abundant phyla were selected, and statistical differences between groups were examined using the Kruskal–Wallis rank-sum test. Statistical significance was defined as P < 0.05.

Eligible genetic instrumental variables (IVs) were selected based on the following criteria: (1) a genome-wide significance threshold of P < 5 × 10^-5, (2) exclusion of SNPs in linkage disequilibrium (LD) with an r² threshold of P <0.001 and an LD window set to 10,000 kb; and (3) evaluation of the robustness of IVs through F-statistic analysis, where the F-statistic was calculated using the formula: F = F=beta²/se² , with a threshold of F > 10 (28, 29).

In this study, Mendelian randomization (MR) analyses were performed using multiple approaches to assess causal effects, including inverse-variance weighted (IVW), MR-Egger, weighted median estimator (WME), weighted mode (WM), and simple mode (SM) methods. Given its higher statistical power in simulation studies, the IVW method served as the primary analytical approach. Causal associations were considered statistically significant if they met the following criteria: (1) IVW analysis yielded P < 0.05, and (2) the effect direction was consistent across supplementary MR methods (30).

To evaluate the robustness of Mendelian randomization results, potential horizontal pleiotropy was assessed using MR-Egger regression and MR-PRESSO methods (31). Heterogeneity among instrumental variables was examined by applying Cochran’s Q test to the inverse-variance weighted estimates, with P < 0.05 suggesting significant heterogeneity (32).

All analyses were performed using R software (version 4.4.1). MR analysis was performed using the R-based package “TwoSampleMR” and the “MR-PRESSO” package was used to conduct multiplicity tests.

Statistical analyses were conducted using SPSS (Version 25). For normally distributed quantitative data, results are presented as mean ± standard deviation (SD), with independent sample t-tests used for between-group comparisons. Analysis of variance (ANOVA) was applied for multiple-group comparisons. Non-normally distributed data were described by median and interquartile range (IQR, P25–P75) and analyzed using non-parametric tests (Mann-Whitney U or Kruskal-Wallis test). Spearman,s rank correlation coefficient was used for correlation analysis. Statistical significance was defined as P < 0.05.

Following the IV selection criteria outlined in the Methods, 221 SNPs were selected as instrumental variables for gut microbiota. This study analyzed the relationship between 473 gut microbiota and diabetic neuropathy patients, Initially, circular heat maps were employed to visualize all data ( Figure 2 ). Subsequent screening identified 14 gut microbiota species exhibiting a causal relationship with diabetic neuropathy, as shown in Figure 3 . Among them, nine species demonstrated a protective effect against DPN development: Blautia sp000436935 (OR: 0.774, 95%CI: 0.672-0.891, P < 0.001), Borreliale (OR: 0.422, 95%CI: 0.222-0.801, P = 0.008), CAG-488 sp000434055 (OR: 0.680, 95%CI: 0.531-0.871, P = 0.002), Cetobacterium A (OR: 0.620, 95%CI: 0.464-0.830, P = 0.001), Clostridia (OR: 0.538, 95%CI: 0.294-0.984, P = 0.044), Escherichia flexneri (OR: 0.835, 95%CI: 0.718-0.972, P = 0.020), Ruminococcus C sp000437255 (OR: 0.843, 95%CI: 0.718-0.989, P = 0.037), UBA1417 sp003531055 (OR: 0.478, 95%CI: 0.272-0.838, P = 0.010), UBA9475 sp002161675 (OR: 0.633, 95%CI: 0.409-0.982, P = 0.041). Conversely, three species were associated with promoting DPN development: CAG-302 (OR: 1.144, 95%CI: 1.036-1.263, P = 0.008), QALR01 sp003150035 (OR: 1.820, 95%CI: 1.216-2.723, P = 0.004), Thermoprotei (OR: 4.308, 95%CI: 1.797-10.324, P = 0.001). However, following false discovery rate (FDR) correction, none of these associations retained statistical significance (P >0.05, Supplementary Table S5 ).

Leave-one-out sensitivity analysis confirmed the robustness of the initial causal associations, as the overall effect estimates remained consistent after sequentially excluding individual SNPs ( Supplementary Figure S1 ). No significant horizontal pleiotropy was detected by the MR-Egger regression intercept test (P > 0.05; Supplementary Table S2 ). While Cochran’s Q test suggested the presence of heterogeneity for the gut microbiota feature UBA1417 sp003531055 (P = 0.036; Supplementary Table S3 ), subsequent analysis using MR-PRESSO found no significant evidence of heterogeneity (P > 0.05; Supplementary Table S4 ).

During the experiment, all mice in the Con group maintained blood glucose levels ≤16.7 mmol/L. In the DM group, two mice did not meet the hyperglycemia criterion (blood glucose < 16.7 mmol/L), and one mouse showed a log₁₀-transformed 50% mechanical withdrawal threshold (MWT₁₀₀₀₀) decrease greater than 0.2, leading to their exclusion from the study. Similarly, one mouse in the DPN group was excluded due to a log₁₀-transformed 50% MWT₁₀₀₀₀ decrease less than 0.2. Consequently, the final sample sizes were 6 mice in the Con group, 7 in the DM group, and 9 in the DPN group.

Baseline comparisons of blood glucose levels and body weights among the three groups showed no significant differences (P > 0.05). At the experimental endpoint, as presented in Table 1 , body weights remained statistically unchanged across groups (p > 0.05). However, significant differences in blood glucose levels were observed (P < 0.05). Both the DM and DPN groups exhibited significantly higher blood glucose levels compared to the Con group (p < 0.05), while no significant difference was found between the DM and DPN groups (P > 0.05).

As shown in Table 1 , there were no significant differences in baseline in log10(50%MWT10000) and TWL values among the three groups of mice (P > 0.05). At the experimental endpoint, significant intergroup differences were observed in log₁₀(50%MWT*10000) and TWL (P < 0.05). The log₁₀(50%MWT*10000) of the DPN group was significantly lower than that of the Con and DM groups (P < 0.05), while no significant difference was found in log₁₀(50%MWT*10000) between the Con and DM groups (P > 0.05). Thermal nociception tests showed that compared with the Con group, the TWL of both the DM and DPN groups was significantly shortened (P < 0.05).

These results collectively demonstrate the successful establishment of both type 2 diabetes mellitus and DPN mice models.

To investigate alterations in gut microbiota composition associated with diabetes and diabetic neuropathy, fecal samples were analyzed via 16S rRNA gene sequencing. Statistical analysis of operational taxonomic unit (OTU) abundance per mouse revealed that compared with the Con group, the OTU abundance in both DM group and DPN group was significantly reduced (P < 0.01 Figure 4B ). Venn diagram showed that a total of 3,028 OTUs were shared among the three groups; among them, 1474, 275, and 373 unique OTUs were identified in the Con group, DM group, and DPN group, respectively ( Figure 4A ).

To evaluate differences in gut microbial community composition under diabetic and diabetic neuropathic conditions, alpha and beta diversity analyses were performed. In alpha diversity, compared with the Con group, both the DM group and DPN group exhibited a decreasing trend in Chao 1 index and Shannon index, while no statistically significant difference was observed between the DM group and DPN group ( Figures 4C, D ). Beta diversity analysis based on Bray-Curtis distance indicated that there were significant differences in gut microbiota structure among the three groups ( Figure 4E ).

Analysis of the microbial composition at the phylum level indicated that Firmicutes and Bacteroidetes were the dominant phyla in the gut microbiota across all three groups ( Figure 5A ). Relative to the Con group, both the DM and DPN groups displayed an increased relative abundance of Firmicutes and a decreased relative abundance of Bacteroidetes ( Figures 5C, D ). Consequently, the Firmicutes/Bacteroidetes (F/B) ratio was higher in both the DM and DPN groups compared to the Con group ( Figure 5B ).

The baseline data of the study participants are shown in Table 2. Subsequent analysis of gut microbiota composition in individuals with diabetes, diabetic neuropathy, and healthy controls revealed no statistically significant differences in OUT abundance among the three groups (P > 0.05, Supplementary Figures S2A, B ). In the α-diversity analysis, compared with the Con group, the Shannon index in both the DM group and DPN group was significantly reduced (P <0.05, Supplementary Figure S2C ), while there was no statistically significant difference in the Chao1 index among the groups(P > 0.05, Supplementary Figure S2D ). Principal Coordinates Analysis (PCoA) based on Bray–Curtis dissimilarity was used to visualize β-diversity, revealing significant differences in gut microbiota composition among the three groups (P <0.05, Supplementary Figure S2E ). Further analysis of the gut microbiota structure in the human cohort at the phylum level revealed that, consistent with the findings in mice. Firmicutes and Bacteroidetes were the dominant phyla across all three human groups ( Supplementary Figure S3A ). However, no statistically significant differences in their relative abundances were observed between groups ( Supplementary Figures S3B-D ).