Oxaliplatin injection (50 mg/10 ml) was obtained from Sichuan Huiyu Pharmaceutical Co., Ltd. (Sichuan, China). A. muciniphila (ATCC^® BAA-835™; Product No. BNCC 323275) was supplied by BNCC^® (Henan Engineering Research Center of Industrial Microbiology, Henan, China). The bacterial suspension was stored at 4°C at a concentration of 2.0 × 10⁹ colony-forming units (CFU)/ml. NaB (CAS: 156-54-7; purity ≥ 98%) was sourced from Shanghai Macklin Biochemical Technology Co., Ltd. (Shanghai, China). Enzyme-linked immunosorbent assay (ELISA) kits targeting tumor necrosis factor-α (TNF-α; JX-1721A1), interleukin-1β (IL-1β; JX-1588A1), interleukin-6 (IL-6; JX-1731A1), interleukin-10 (IL-10; JX-1736A1), and neurofilament light chain (NfL; JX-11808A) were purchased from Yancheng Junxing Biotechnology Co., Ltd. (Yancheng, China).

This study was approved by the Laboratory Animal Welfare and Ethics Committee of the Zhejiang Academy of Traditional Chinese Medicine (Approval No. 2025002) and was conducted in strict compliance with the guidelines of the International Association for the Study of Pain (IASP).

Twenty-five male Sprague-Dawley rats (6–8 weeks old, weighing 250–300 g) were obtained from Hangzhou Medical College, Zhejiang, China (Certificate No. 20250421Aazz01000180033). The animals were housed under specific pathogen-free (SPF) conditions at a controlled temperature of 22 ± 2°C and a relative humidity of 50 ± 10%, under a standard 12-hour light/dark cycle with ad libitum access to food and water. Following a one-week acclimatization period, the rats were weighed and randomly allocated into five experimental groups (n = 5 per group): normal control (Group N), OIPN model (Group O), Akkermansia muciniphila intervention (Group A), sodium Butyrate intervention (Group B), and combined Akkermansia muciniphila and sodium Butyrate intervention (Group AB), as illustrated in Figure 2.

While intraperitoneal injection has been the predominant route for oxaliplatin administration in previous studies, intravenous delivery more accurately replicates the clinical administration pathway, despite greater technical demands. Accordingly, this study employed tail vein injection (30). Oxaliplatin was dissolved in 5% glucose to a concentration of 2 mg/mL and administered intravenously at 5 mg/kg (0.25 mL/100 g) once weekly for four consecutive weeks. Group N received an isovolumetric injection of 5% glucose solution via the same route.

As detailed in the Introduction, A. muciniphila and sodium butyrate were selected to target the gut-immune-nerve axis. By using these agents, we aimed to confirm whether targeting the gut-immune-nerve axis could effectively mitigate oxaliplatin-induced neurotoxicity. For drug administration, the A. muciniphila suspension was stored at 4°C, thoroughly mixed, and diluted with phosphate-buffered saline (PBS). Rats received a daily oral gavage of 1×10⁹ CFU suspended in 0.5 mL PBS. NaB was dissolved in PBS and administered daily via oral gavage at a dose of 100 mg/kg (in a volume of 0.5 mL/rat). Rats in the treatment groups (Groups A, B, and AB) received the corresponding interventions via oral gavage once daily for 34 consecutive days (from Day 1 to Day 34). On Day 35, 24 hours after the final administration, all rats were sacrificed for sample collection.

Body weight was monitored weekly from week 0 to week 5. Behavioral assessments for mechanical and cold allodynia were conducted on the first day of each week (weeks 1–5).

Rats were placed in individual transparent perspex boxes on an elevated wire mesh grid and allowed to acclimate for 30 minutes. A dynamic plantar aesthesiometer (Model KW-CT-1; Nanjing Calvin Biotechnology Co., Ltd., China) equipped with a 0.6 mm steel filament was used. The filament was applied vertically to the plantar surface of the left hind paw with increasing force. The force (in grams) at which the rat withdrew its paw was recorded as the Paw Withdrawal Threshold (PWT). Three measurements were taken for each rat with an inter-trial interval of 5 minutes, and the average was calculated.

Cold sensitivity was assessed using a cold plate apparatus (Model ZH-6C; Anhui Zhenghua Biological Instrument Co., Ltd., China) maintained at 4°C. Rats were placed individually on the cold surface, and the time elapsed until the first sign of nociceptive behavior (lifting, licking, or shaking of the hind paw) was recorded as the Paw Withdrawal Latency (PWL). To prevent tissue damage, a cut-off time of 30 seconds was imposed. If no response was observed within this limit, the latency was recorded as 30 seconds. A minimum interval of 10 minutes was allowed between tests to avoid sensitization.

On day 35 of the experiment, rats were administered an overdose of Zoletil^®50 via intraperitoneal injection to ensure deep anesthesia, followed by rapid cardiac blood collection after thoracotomy. Collected tissues included central plantar skin from the hind paws, DRG from the L4–L5 spinal segments. Blood samples were centrifuged to isolate serum, which was stored at –80°C until analysis.

Serum samples were collected at the end of the experiment (Day 35). The concentrations of TNF-α, IL-1β, IL-6, IL-10, and NfL were measured using rat-specific two-step sandwich ELISA kits according to the manufacturer’s instructions. All samples were analyzed in triplicate. Optical density (OD) was determined at 450 nm using a microplate reader (Infinite^® F50; Tecan Trading AG, Switzerland). Analyte concentrations were calculated based on their respective standard curves. The mean of the three technical replicates for each sample was used for statistical analysis.

L4–L5 DRG tissues were harvested, fixed in 4% paraformaldehyde, and subjected to standard processing, including dehydration, paraffin embedding, and sectioning. Sections were stained with hematoxylin and eosin (H&E) and digitized using a panoramic tissue scanner (KF-PRO-120; KFBIO, China). Pathological alterations in DRG neurons—including soma shrinkage, nuclear pyknosis, interstitial edema, and inflammatory cell infiltration—were evaluated under light microscopy. Using Visiopharm software (Visiopharm A/S, Denmark), the number of inflammatory cells in each rat DRG sample was analyzed across three different sections or three non-overlapping fields of view (400× magnification). The analysis specifically quantified lymphocytes, macrophages, and neutrophils within the loose connective tissue (interstitium) distributed between neurons. The analysis workflow followed a standardized protocol involving region definition, phenotyping, and algorithmic quantification. First, Regions of Interest (ROIs) were automatically delineated along the tissue boundaries, with manual adjustments applied where necessary to ensure anatomical precision. Second, for signal detection, the AI algorithm analyzed fluorescent expressions on cell nuclei to establish positivity thresholds; these settings were saved and consistently applied across all sections within the same batch to ensure reproducibility. Third, cellular identification was achieved by detecting DAPI-stained nuclei and algorithmically expanding the boundaries to define the cytoplasmic area. Finally, the software computed quantitative parameters—including positive cell counts—via high-resolution stepwise calculation within the defined ROIs. To ensure data accuracy, a senior pathologist, blinded to the experimental grouping, performed a systematic re-evaluation of the H&E-stained DRG sections to validate the automated counts. In instances where the automated quantification deviated from the pathologist’s assessment, the manual counts were recorded as the final data.

As the diagnostic gold standard for small fiber neuropathy recommended by the European Federation of Neurological Societies (31), IENFD was quantified. Plantar skin tissues were fixed, sectioned, and subjected to immunofluorescence staining using an antibody against the pan-neuronal marker PGP9.5 (Proteintech^®). Images were acquired using a Pannoramic SCAN scanner (3D Histech, Hungary). Using Visiopharm software (Visiopharm A/S, Denmark), 3 non-overlapping fields of view (400× magnification) were selected for each rat to count the nerve fibers crossing the basement membrane, and the mean value was calculated.

Genomic DNA was extracted from rat fecal samples using the Magnetic Soil and Stool DNA Kit (Tiangen Biotech) and quantified via Qubit. The V3–V4 hypervariable region of the bacterial 16S rRNA gene was amplified using primers 341F/805R, purified with AMPure XP beads, and paired-end sequenced on a DNBSEQ-G99 platform (LC-Bio Technology).

For the metabolomic analysis of short-chain fatty acids (SCFAs), samples were extracted with pre-cooled 80% methanol and derivatized using EDC and 3-NPH. The supernatant was transferred into injection vials for LC-MS/MS analysis. Target analytes were separated on an Agilent Poroshell 120 EC-C18 column using an AB Sciex Jasper UPLC system coupled to a 4500MD triple quadrupole mass spectrometer operating in negative electrospray ionization (ESI−) mode. Raw metabolomic data were normalized and log-transformed prior to Principal Component Analysis (PCA) and Pearson correlation analysis to identify functional metabolic associations.

Statistical analyses were performed using GraphPad Prism software (version 10.6.0). Continuous data are presented as mean ± standard deviation (SD). The normality of data distribution and homogeneity of variances were assessed using the Shapiro-Wilk test and Levene’s test, respectively. For comparisons between two groups, the unpaired Student’s t-test was used for normally distributed data with equal variances, whereas Welch’s t-test was applied when variances were unequal. Non-normally distributed data were analyzed using the Mann-Whitney U test. For multiple-group comparisons, one-way ANOVA followed by Tukey’s post-hoc test was performed when variances were homogeneous. In cases of heterogeneous variances, Welch’s ANOVA followed by the Games-Howell post-hoc test was employed. Longitudinal data were analyzed using repeated-measures ANOVA. Correlations were assessed using Pearson’s correlation coefficient for normally distributed data and Spearman’s rank correlation coefficient for non-normally distributed data. A two-sided P-value < 0.05 was considered statistically significant.

Body weight was comparable across all groups at baseline. Beginning in week 3, the OIPN model group (Group O) exhibited a progressive decline in body weight relative to the normal control group (Group N). This difference increased over time, reaching statistical significance at week 3 (P = 0.05) and week 4 (P = 0.04), with the most substantial reduction observed at week 5 (P = 0.0007). At the week 5 endpoint, body weights in all treatment groups (Groups A, B, and AB) remained lower than in Group N (536.6 ± 29.7 g). However, the combined intervention (Group AB) showed the least severe body weight reduction among the treatment groups. Specifically, Group AB recorded the highest mean body weight (461.4 ± 46.7 g) with a comparatively lower level of statistical significance versus Group N (P = 0.0397) than Group A (456.2 ± 62.7 g, P = 0.0265) and Group B (423.4 ± 25.5 g, P = 0.0018). These results indicate that while the OIPN model induced a progressive and significant decline in body weight that was not fully restored to baseline control levels by any intervention, the combined therapy was the most effective at mitigating this loss (Figure 3A).

Cold hypersensitivity was assessed by measuring the PWL on a 4°C cold plate. A two-way repeated-measures ANOVA revealed significant main effects of time (F (4, 100) = 31.82, P < 0.0001) and intervention (F(4, 100) = 27.46, P < 0.0001). The oxaliplatin model group (Group O) exhibited a marked and sustained decrease in PWL compared to Group N, confirming successful induction of cold allodynia (Figure 3B). At week 3, Group O demonstrated significant cold hypersensitivity compared to Group N (P < 0.0001; Figure 3D). Regarding therapeutic interventions, Group A (P = 0.0163) and Group AB (P = 0.0212) significantly increased PWL thresholds compared to Group O, whereas monotherapy with sodium butyrate did not yield a statistically significant improvement (P = 0.1188). At week 5, Group O demonstrated significant cold hypersensitivity compared to Group N (P < 0.0001; Figure 3E). Regarding therapeutic interventions, Group A (P = 0.0388) and Group AB (P < 0.0001) significantly increased PWL thresholds compared to Group O, whereas monotherapy with sodium butyrate did not yield a statistically significant improvement (P = 0.2048).

Behavioral assessment of PWT confirmed the successful induction of OIPN. Rats in the model group (Group O) developed progressive mechanical hypersensitivity, reflected in a sustained decline in PWT. A two-way repeated-measures ANOVA revealed a significant interaction between time and group (F(16, 100) = 6.842, P < 0.0001), indicating divergent temporal trajectories of nociceptive sensitivity among groups (Figure 3C). Simple effects analysis revealed no significant differences among groups during weeks 1–2. From week 3 onwards, Group O exhibited a continuous and significant decline in PWTs compared to Group N. Groups A, B, and AB showed transient reductions at week 3, followed by a rebound to baseline levels during weeks 4–5. Group AB exhibited a “V-shaped” recovery pattern: despite a marked reduction at week 3, thresholds showed partial recovery at week 4 and near-normal levels were restored by week 5.

Therapeutic interventions (Groups A, B, and AB) significantly attenuated the development of mechanical allodynia compared to Group O. At week 3, significant intergroup differences were observed (one-way ANOVA: F = 11.18, P = 0.0004); all treatment groups had higher PWTs than Group O, though values remained below those of the normal control group (Group N) (Figure 3F). By week 5, this pattern shifted: all treatments conferred robust protection, with PWTs not significantly different from Group N (all P > 0.05) but markedly higher than Group O (all P < 0.0001) (Figure 3G).

The combined intervention (Group AB) demonstrated the most pronounced effect at this endpoint. Compared to Group O, Group AB maintained a significantly higher mean PWT (55.5 ± 4.9 g vs. 21.9 ± 1.5g, Welch’s t-test: t = 14.72, P < 0.0001), with a mean difference of 33.6 g (95% CI: 27.6 to 39.5). The exceedingly large effect size (η² = 0.978) indicates that the mechanical hypersensitivity in Group AB was effectively reversed to a level comparable with normal control (Figure 3G).

Oxaliplatin administration triggered a severe systemic inflammatory imbalance. Compared to Group N, the model group (Group O) exhibited significant elevations in pro-inflammatory cytokines (TNF-α, IL-1β, IL-6) and a reduction in the anti-inflammatory cytokine IL-10 (all P < 0.0001), indicating a shift toward a pro-inflammatory state (Figures 4A–D).

Therapeutic interventions differentially modulated this response. Group B significantly reduced TNF-α and increased IL-10, but did not significantly affect IL-1β or IL-6. Group A significantly lowered IL-1β and IL-6 while elevating IL-10. The combined intervention (Group AB) demonstrated the broadest efficacy, significantly suppressing all three pro-inflammatory cytokines and restoring IL-10 to a level comparable with Group N (P > 0.05).

Concurrently, serum NfL, a biomarker of axonal injury, was markedly elevated in Group O compared to Group N (P < 0.0001; Figure 4E). All treatment groups showed lower NfL levels than Group O. Among them, Group AB achieved the greatest reduction, resulting in NfL levels significantly lower than those in Group B (P < 0.05) and comparable to those in Group A (P > 0.05). The Group AB reduced NfL concentrations by 82.58 pg/mL (95% CI [58.84, 106.1]; P < 0.0001) with a substantial effect size (η² = 0.921). Normalization analysis indicated that Group AB restored NfL levels to 77.63% of the physiological baseline, a recovery rate significantly superior to that of Group A (53.71%) and Group B (29.37%). These data demonstrate that the combined regimen of A. muciniphila and sodium butyrate exerts superior neuroprotective effects, minimizing axonal injury more effectively than monotherapies.

Correlation analyses further elucidated the biological significance of elevated serum NfL levels (Figure 5). Robust positive correlations were observed between NfL and pro-inflammatory cytokines, with the strongest association identified for IL-1β (r = 0.852, P < 0.0001), followed by IL-6 and TNF-α (r > 0.79 for both). Conversely, NfL levels displayed a strong inverse correlation with the anti-inflammatory cytokine IL-10 (r = −0.881, P < 0.0001), accounting for 77.6% of the variance.

Histological analysis of L4–L5 DRG and peripheral nerves confirmed profound oxaliplatin-induced structural damage. Compared to Group N, which displayed normal neuronal and axonal architecture with intact myelin sheaths, the model group (Group O) exhibited severe degenerative pathology (Figure 6). This included neuronal atrophy with nuclear pyknosis, disorganization of satellite glial cells, axonal swelling and fragmentation, extensive myelin sheath disruption (delamination and demyelination), and marked interstitial edema with inflammatory cell infiltration. To enhance clarity, we have provided high-resolution images featuring specific arrow annotations in Supplementary Figure 1. These enlarged views allow for a more detailed examination of tissue morphology and cellular density.

All therapeutic interventions (Groups A, B, and AB) attenuated these morphological alterations. Treatment groups showed preserved neuronal architecture with reduced nuclear pyknosis, more compact axonal fibers with diminished swelling, and largely restored myelin integrity. Interstitial edema and inflammatory infiltration were also markedly reduced compared to Group O. Oxaliplatin administration induced a marked elevation in inflammatory cell infiltration relative to the normal control (Group O: 8.1 ± 1.5 vs. N: 0.7 ± 0.5, P < 0.0001; Figure 6B). While monotherapy with sodium butyrate (Group B: 5.3 ± 1.4) significantly attenuated inflammatory cell counts compared to the Group O (P = 0.0114), intervention with A. muciniphila alone (Group A: 3.3 ± 1.5) exerted a more potent inhibitory effect (P < 0.0001). Notably, the combined intervention (Group AB: 1.7 ± 0.7) demonstrated the most robust efficacy (P < 0.0001 vs. Group O), successfully restoring the inflammatory profile to levels statistically distinguishable from the normal control group (Group AB vs. N, P = 0.6819).

To assess oxaliplatin-induced small-fiber neuropathy, we quantified IENFD in the hind paw plantar skin. Immunofluorescence imaging revealed a marked reduction of PGP9.5-positive nerve fibers in the model group (Group O) compared to normal control (Group N) (Figure 7).

Quantification confirmed a significant reduction in IENFD in Group O (555.2 ± 162.0 vs. 1432.0 ± 356.5 fibers/mm² in Group N). Therapeutic interventions significantly attenuated this denervation. All treatment groups exhibited higher IENFD compared to Group O. The combined intervention Group AB and Group A showed comparable and superior protective efficacy (1082.0 ± 182.2 and 917.2 ± 224.6 fibers/mm², respectively), both significantly exceeding the effect of Group B (762.8 ± 257.6 fibers/mm²). However, IENFD in all treatment groups remained lower than in Group N, indicating partial preservation.

To verify the efficacy of the gavage protocol, we performed 16S rRNA gene sequencing on fecal samples. As shown in Figure 8, the relative abundance of A. muciniphila was negligible (<1%) in both the Normal (N) and Model (O) groups. While the group receiving A. muciniphila alone (Group A) showed a modest increase in abundance (~3.5%) and the group treated with Sodium Butyrate alone (Group B) exhibited a slight elevation(~1.7%), the Combined Intervention group (Group AB) displayed a dramatic and statistically significant expansion of A. muciniphila, enabling it to become the dominant genus in the gut microbiome, reaching approximately 52% of total bacteria (P < 0.0001 compared to all other groups, Figure 8A). This confirms that the combined intervention successfully established high-abundance colonization.

Analysis of fecal SCFAs revealed that while the Model group showed no significant alteration in butyrate levels compared to the Normal group, the Combined intervention significantly elevated butyrate concentration (P < 0.005, Figure 8B), suggesting a functional boost of the neuroprotective microbial ecosystem. The metabolic profile of the Group AB—characterized by simultaneously elevated levels of both Acetate (a primary fermentation product of A. muciniphila) and Butyrate—supports the existence of a synergistic cross-feeding network. Quantification of fecal Acetate content (Figure 8C). The substantial production of acetate by the dominant A. muciniphila likely served as an abundant substrate for downstream butyrate-producing bacteria, thereby facilitating the overall restoration of SCFA levels. Furthermore, a correlation analysis indicated a positive trend between fecal acetate and butyrate levels (r = 0.3562), although this did not reach statistical significance (P = 0.0805, Figure 8D).