Abstract
Background
Oxaliplatin-induced peripheral neuropathy (OIPN) is an important adverse reaction in patients with gastric cancer treated with oxaliplatin, but there is no objective biomarkers for changes in OIPN in patients after multiple rounds of chemotherapy. This research aimed to identify serum metabolic biomarkers using longitudinal untargeted metabolomics for early detection of OIPN progression in gastric cancer patients receiving repeated chemotherapy.
Methods
Eighty-four serum samples of the same gastric cancer patient (n=42) before and after receiving oxaliplatin chemotherapy twice were collected. The metabolic profiles of serum samples were acquired using an untargeted metabolomics approach based on ultra-high-performance liquid chromatography-Q-Exactive Orbitrap tandem mass spectrometry (UHPLC-Q-Exactive Orbitrap-MS/MS). Multivariate statistical analysis, receiver operating characteristic (ROC) curve analysis, SHapley Additive exPlanations (SHAP) analysis, and pathway enrichment analysis were used to identify potential biomarkers and metabolic pathways.
Results
A total of 16 differentially expressed metabolites (DEMs) were screened in discovery set, which belonged to amino acids and derivatives, lipids and derivatives, organic acids and derivatives, and others, mainly involved in amino acid metabolism, lipid metabolism, and nervous system metabolism. Four DEMs (including norepinephrine, 9,10-DHOME, 5-hydroxyindoleacetic acid, and procollagen 5-hydroxy-lysine) showed certain predictive ability for OIPN in the same gastric cancer patient before and after receiving oxaliplatin chemotherapy twice. Thirty-three DEMs were discovered in validation set, notably, norepinephrine emerged as a metabolite exhibiting consistent and notable statistical differences in both the discovery and validation sets.
Conclusions
These findings demonstrate the alterations of serum metabolic profiles in patients before and after receiving oxaliplatin chemotherapy, which may deliver valuable biomarkers for early identification and outcome prediction of OIPN progression.
Keywords: Oxaliplatin-induced peripheral neuropathy (OIPN), gastric cancer, biomarkers, untargeted metabolomics, SHapley Additive exPlanations analysis (SHAP analysis)
Introduction
Oxaliplatin is a platinum based chemotherapy drug, for gastric cancer (1), colorectal cancer (2), breast cancer (3), etc. Oxaliplatin has certain therapeutic effects, but its peripheral neurotoxicity has attracted widespread attention. According to reports, 80% of patients who receive oxaliplatin chemotherapy experience oxaliplatin-induced peripheral neuropathy (OIPN) related symptoms, such as sensitivity to cold objects and swallowing cold food, discomfort in the throat, muscle spasms, etc., which reduce people’s quality of life (4-6). In clinical practice, patients undergo different doses of oxaliplatin chemotherapy according to their condition. The OIPN symptoms produced after two rounds of chemotherapy are different, however, there is currently no research on the overall metabolic product differences of patients undergoing multiple rounds of chemotherapy. Therefore, it is urgent to establish an effective method to identify the differential biomarkers and metabolic pathways of OIPN in the same patient before and after multiple rounds of oxaliplatin chemotherapy, which could play a foundation for the study of the oxaliplatin toxicity mechanism.
Metabolomics, genomics, transcriptomics, and proteomics constitute modern systems biology. Genomics, transcriptomics, and proteomics tell us what may happen, while metabolomics tells you what actually happens. As a metabolomics study of downstream products, it can be used to primarily speculate on the pathological mechanisms of the body by analyzing the external effects of metabolic changes in affected cells, tissues, or organisms. Nowadays, metabolomics has been applied in multiple fields, including disease diagnosis, pharmaceutical research and development, nutritional food science, toxicology, microorganism, botany, and other fields closely related to human health care (7-10). In addition, through comprehensive analysis of multiple indicators, metabolomics can study the types, quantities, and relationships of endogenous small molecule metabolites. Metabolites, as products of metabolic pathways, play an important role in metabolism. Liquid chromatography-mass spectrometry (LC-MS) is the most widely used technique in metabolomics, with high sensitivity, wide range, and good resolution (11). Therefore, LC-MS metabolomics has great potential in identifying biomarkers for disease diagnosis, including such as Alzheimer’s disease (12), Parkinson’s disease (13), stroke (14), anxiety disorder (15), high-risk for psychosis (16) and drug induced toxicity assessment, including nephrotoxicity (17), hepatotoxicity (18), neurotoxicity (19). Therefore, metabolomics can predict early toxicity and help discover potential toxicity mechanisms. Xu et al. (17) used metabolomics and diversified data analysis to analyze the rat plasma metabolomic profile of oxaliplatin induced neurotoxicity, and screened for relevant biomarkers of neurotoxicity. However, that study only explored the occurrence of OIPN biomarkers in colorectal cancer rats after receiving oxaliplatin chemotherapy, and did not combine with clinical practice, considering the changes in OIPN in patients after multiple rounds of chemotherapy.
OIPN remains a dose-limiting complication of chemotherapy, with no validated biomarkers for early detection or outcome prediction. Current clinical assessments rely on subjective patient-reported scales or neurological examinations, which often fail to identify neurotoxicity until irreversible nerve damage occurs. While recent studies have explored genetic or proteomic biomarkers, the dynamic metabolic perturbations induced by oxaliplatin remain poorly characterized. Metabolomics offers a unique opportunity to uncover functional pathways linked to OIPN pathogenesis, yet no prior studies have systematically integrated untargeted metabolomics with machine learning to identify predictive serum biomarkers for OIPN in gastric cancer patients. This gap highlights the critical need for longitudinal metabolic profiling to address the clinical diagnostic dilemma of OIPN.
In this study, a metabolic profiling method based on ultra-high-performance liquid chromatography-Q-Exactive Orbitrap tandem mass spectrometry (UHPLC-Q-Exactive Orbitrap-MS/MS) coupled with multivariate statistical analysis and metabolic pathway analysis were employed to identify the potential biomarkers and major metabolic pathways to discriminate the serum samples of the same gastric cancer patients before and after receiving oxaliplatin chemotherapy twice. Additionally, SHapley Additive exPlanations (SHAP) values were employed to improve the model’s interpretability, highlighting the individual contributions of each differentially expressed metabolite (DEM). This study provides a deeper insight and comprehensive understanding for the underlying mechanism in the progression of OIPN. We present this article in accordance with the STARD reporting checklist (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-240/rc).
Methods
Study population
This research was part of a continuing prospective study carried out at the Affiliated Hospital of Jiangnan University. A total of 42 gastric cancer patients receiving oxaliplatin chemotherapy twice were enrolled from August 2022 and July 2023. The criteria for patient selection are as follows: histopathological confirmation of gastric cancer diagnosis; TNM staging ranging from I to IV; chemotherapy involving oxaliplatin-containing regimens; good general condition; Karnofsky Performance Status (KPS) score greater than 60; absence of other diseases causing peripheral neuropathy, such as diabetes; no current use of medications affecting peripheral nerves; age between 18 and 85 years, irrespective of gender; PS score ≤2 points; expected survival period of more than 3 months; normal liver, kidney, heart, bone marrow, and other functionalities; and patients with intact consciousness and the ability to clearly articulate their physical sensations. The study was conducted according to the Declaration of Helsinki and its subsequent amendments, and approved by the Ethics Committee of the Affiliated Hospital of Jiangnan University (No. LS2022080). The patients and participants provided their written informed consent to participate in this research.
Serum sample collection
In the period between August 2022 and July 2023, we collected serum samples of the same gastric cancer patient (n=42) before and after receiving oxaliplatin chemotherapy twice, including a total of 84 cases. Each patient’s blood sample was drawn on an empty stomach in the early morning and stored in EDTA tubes. The blood samples were immediately centrifugated at 1,500 rpm at 4 °C for 10 min, and the supernatants were then stored at −80 °C until required for analysis.
Sample preparation
The experimental samples were thawed at 4 °C. and the samples were vortexed for 1 min and evenly mixedy. 400 µL methanol was added to a 96-well plate and 100 µL sample was moved to 96-well plate and shaked for 5 min. After that, transferring the sample plate to the A200 positive pressure nitrogen blowing module, with the positive low mode pressure for 10 min, and the nitrogen blowing until the sample is completely dry. Accurately adding 150 µL of 2-chlorophenzylalanine solution (4 ppm) configured with 80% methanol solution, vortex oscillate for 5 min, transfer to the area to be tested for membrane sealing, for LC-MS detection.
UHPLC-Q-Exactive Orbitrap-MS/MS analysis
The LC analysis utilized a Vanquish UHPLC System from Thermo Fisher Scientific, USA. The chromatography was executed using an ACQUITY UPLC® HSS T3 column (2.1×100 mm, 1.8 µm) from Waters, Milford, MA, USA, with the column temperature set at 40 °C. The column was kept at 40 °C, with a flow rate of 0.3 mL/min and an injection volume of 2 µL. In LC-ESI(+)-MS analysis, the mobile phases were made up of (B2) acetonitrile with 0.1% formic acid (v/v) and (A2) water with 0.1% formic acid (v/v). The separation process followed this gradient: from 0 to 1 minute, 8% B2; from 1 to 8 minutes, 8% to 98% B2; from 8 to 10 minutes, 98% B2; from 10 to 10.1 minutes, 98% to 8% B2; and from 10.1 to 12 minutes, 8% B2. The analytes for LC-ESI(−)-MS analysis were conducted with (B3) acetonitrile and (A3) ammonium formate at a concentration of 5 mM. The separation process followed this gradient: from 0 to 1 minute, 8% B3; from 1 to 8 minutes, 8% to 98% B3; from 8 to 10 minutes, 98% B3; from 10 to 10.1 minutes, 98% to 8% B3; and from 10.1 to 12 minutes, 8% B3 (20).
MS data and statistical analysis
Mass spectrometric detection of metabolites was performed on Orbitrap Exploris 120 (Thermo Fisher Scientific, USA) with ESI ion source. Simultaneous MS1 and MS/MS (Full MS-ddMS2 mode, data-dependent MS/MS) acquisition was used. Parameters set were: 30 arb for sheath gas pressure, 10 arb for auxiliary gas flow, spray voltages of 3.50 kV for ESI(+) and −2.50 kV for ESI(−), a capillary temperature of 325 °C, an MS1 range of m/z 100 to 1,000, and an MS1 resolving power of 60,000 FWHM, each cycle consists of 4 data-dependent scans, MS/MS resolving power of 15,000 FWHM, 30% normalized collision energy, and dynamic exclusion time set to automatic (21).
Statistical analysis
According to Rasmussen et al. 2022 (22), the Proteowizard software package (v3.0.8789) was employed to transform the original mass spectrometry offline file into the mzXML format. Using the R XCMS (v3.12.0) (23) software package for peak detection, peak filtering, and peak alignment processing were processed to obtain the quantitative list of metabolites. The parameters were bw =2, ppm =15, peak width =c (5, 30), mzwid =0.015, mzdiff =0.01, method =centWave. The data matrix obtained mass to charge ratio, retention time (RT), and peak area. A total of 14,277 precursors were obtained in positive mode, and 13,438 precursors were obtained in the negative mode. The data of different orders of magnitude were compared. The based peak chromatogram (BPC) of gastric cancer patients who underwent oxaliplatin chemotherapy twice before and after in both positive and negative modes is shown in Figure S1.
The obtained table indicating the accurate mass of the compound was inserted into SIMCA-P 15.0 software (Umetrics, Umea, Sweden) and used for multivariate statistical
Analysis (24). Dimensionality reduction analysis on the sample data using principal component analysis (PCA) and orthogonal partial least squares discriminant analysis (OPLS-DA). Use the substitution test method to conduct overfitting tests on the model. R2X and R2Y indicate how well the model explains the X and Y matrices, respectively. The model’s predictive capability is shown by Q2, and a value near 1 suggests a superior fit. The more accurately the training set samples can be divided into their original attribution. Calculate the P value based on statistical testing, calculate the variable importance in projection (VIP) using the OPLS-DA dimensionality reduction method, and calculate the inter group difference multiple using fold change (FC) to evaluate the influence and explanatory power of each metabolite content on distinguishing sample classifications, and to aid in identifying marker metabolites. A statistically significant difference in metabolite molecules is considered when the P value is less than 0.05 and VIP is greater than 1 (25). Biochemical databases, Human Metabolome Database (HMDB), Metlin, massBank, LipidMaps, and mzCloud were used to identify the potential metabolites. Pathway analysis was carried out with MetaboAnalyst 4.0 (http://www.metaboanalyst.ca/), utilizing the Homo sapiens metabolic pathway database from Kyoto Encyclopedia of Genes and Genomes (KEGG). A hypergeometric test was employed for enrichment analysis, while relative centrality was used for topological analysis, with the findings displayed in a scatter plot (26). The area under the curve (AUC) was utilized to identify the best combination of significantly altered metabolites as potential diagnostic biomarkers for serum samples of the same gastric cancer patient before and after receiving oxaliplatin chemotherapy twice by GraphPad Prism 9.0.
SHAP analysis for biomarker selection
Conventional statistical methods overlook the interactive or modifying effects between various metabolites, potentially resulting in false positive outcomes (27). In our analysis, we leveraged SHAP values to determine the significance of features that demonstrated the highest predictive accuracy. SHAP, a cutting-edge technique for enhancing the interpretability of tree-based models, employs a game-theoretic approach to combine the local effects of each feature, thereby elucidating the model’s functioning across the entire dataset. This method is regarded as superior to other global approximation techniques. The SHAP algorithm not only quantifies the importance of features within the model but also delves into the specific influence of each feature on individual predictions (28). Therefore, we subsequently employed SHAP analysis workflow to identify biomarkers stably associated with OIPN. Employing the Random Forest model as a base, a SHAP analysis was conducted on the identified set of DEMs to compute the Shapley values associated with each. Data analysis was performed with the statistical software package R (v4.4.1).
Results
Patient characteristics
Patients were all recruited from the Affiliated Hospital of Jiangnan University. As shown in Table 1, in discovery set, we collected a total of 42 serum samples from 21 gastric cancer patients who received oxaliplatin chemotherapy twice before and after. Female patients accounted for 38.1%, while male patients accounted for 61.9%. 16 patients, accounting for 76.19%, developed OIPN after the first chemotherapy, and all patients developed OIPN after the second chemotherapy, which indicate that OIPN has a certain degree of accumulation. In the term of the symptoms of OIPN, patients showed more severe symptoms after the second chemotherapy. After the first chemotherapy, patients concentrated on numbness in their hands and feet. After the second chemotherapy, three patients showed numbness in their entire arms, hands, feet, mouth, and nose. In the term of symptom duration and symptom on set time, the OIPN of patients after the second chemotherapy is more severe than the first one, two patients had symptoms that lasted for more than 20 days. There were no significant differences in white blood cell (WBC) count, neutrophil (NEU), lymphocyte (LYM), monocyte (MON), eosinophil (EOS), red blood cell (RBC), hemoglobin (HGB), platelet (PLT), alpha-fetoprotein (AFP), carcinoembryonic antigen (CEA), fibrinogen (FIB), and D-Dimer between the first and second chemotherapy group. However, compared with the first chemotherapy group, the second chemotherapy group had a lower basophil (BAS) count (P<0.05) (Table 2). The detailed information was in available online: https://cdn.amegroups.cn/static/public/tcr-2025-240-1.xlsx.
Table 1. Demographic of the study population.
| Variable | Discovery set | Validation set | |||
|---|---|---|---|---|---|
| First | Second | First | Second | ||
| Cases (n) | 21 | 21 | 21 | 21 | |
| Female, n (%) | 8 (38.10) | 8 (38.10) | 6 (28.57) | 6 (28.57) | |
| Grade, n (%) | |||||
| 0 | 5 (23.81) | 0 (0.00) | 3 (14.29) | 0 (0.00) | |
| I | 3 (14.29) | 3 (14.29) | 15 (71.43) | 13 (61.9) | |
| II | 11 (52.38) | 15 (71.43) | 3 (14.29) | 5 (23.81) | |
| III | 2 (9.52) | 3 (14.29) | 0 (0.00) | 3 (14.29) | |
| IV | 0 (0.00) | 0 (0.00) | 0 (0.00) | 0 (0.00) | |
| Contact with cold objects, n (%) | 15 (71.43) | 19 (90.48) | 14 (66.67) | 16 (76.19) | |
| OIPN appeared, n (%) | 16 (76.19) | 21 (100) | 18 (85.71) | 21 (100) | |
| Symptoms, n (%) | |||||
| Numbness of fingers | 12 (57.14) | 15 (71.43) | 14 (66.67) | 16 (76.19) | |
| Numbness in both hands and feet | 4 (19.05) | 3 (14.29) | 3 (14.29) | 4 (19.05) | |
| Whole arm | – | 1 (4.76) | – | – | |
| Numbness of fingers and teeth | – | 1 (4.76) | 1 (4.76) | 1 (4.76) | |
| Numbness in hands, feet, nose, and mouth | – | 1 (4.76) | – | – | |
| Symptom duration, n (%) | |||||
| 0–10 min | 13 (61.90) | 16 (76.19) | 15 (71.43) | 13(61.9) | |
| 11–30 min | – | 2 (9.52) | 1 (4.76) | 5 (23.81) | |
| 31–60 min | 1 (4.76) | – | 2 (9.52) | 2 (9.52) | |
| >2 h | 1 (4.76) | 1 (4.76) | – | – | |
| Half a day | – | 1 (4.76) | – | 1 (4.76) | |
| Whole day | 1 (4.76) | 1 (4.76) | – | – | |
| Symptom onset time, n (%) | |||||
| One day | 6 (28.57) | 5 (23.81) | 10 (47.62) | 3 (14.29) | |
| Two days | 1 (4.76) | 1 (4.76) | – | – | |
| Three days | 4 (19.05) | 5 (23.81) | 3 (14.29) | 3 (14.29) | |
| Five days | 2 (9.52) | 2 (9.52) | – | 1 (4.76) | |
| Seven days | 1 (4.76) | 4 (19.05) | 2 (9.52) | 9 (42.86) | |
| Ten days | 1 (4.76) | 1 (4.76) | 1 (4.76) | 2 (9.52) | |
| Over 15 days | 1 (4.76) | 1 (4.76) | 2 (9.52) | – | |
| Over 20 days | – | 1 (4.76) | – | 2 (9.52) | |
| Two months | – | 1 (4.76) | – | 1 (4.76) | |
OIPN, oxaliplatin-induced peripheral neuropathy.
Table 2. Clinical characteristics of participants.
| Characteristics | First chemotherapy (n=21) | Second chemotherapy (n=21) | P value |
|---|---|---|---|
| WBC (109/L) | 5.748±4.301 | 5.743±4.307 | 0.82 |
| NEU (109/L) | 2.362±2.240 | 1.557±0.713 | 0.13 |
| LYM (109/L) | 0.719±0.490 | 0.462±0.282 | 0.08 |
| MON (109/L) | 2.471±3.580 | 3.567±4.233 | 0.33 |
| EOS (109/L) | 0.161±0.143 | 0.124±0.0995 | 0.44 |
| BAS (109/L) | 0.019±0.402 | 0.000±0.000 | 0.04 |
| RBC (1012/L) | 3.561±0.437 | 3.425±0.367 | 0.32 |
| HGB (g/L) | 108.429±13.920 | 106.191±13.563 | 0.64 |
| PLT (109/L) | 162.191±51.335 | 146.476±57.875 | 0.16 |
| AFP (ng/mL) | 5.887±9.887 | 4.395±3.754 | 0.90 |
| CEA (ng/mL) | 2.712±1.253 | 5.275±5.375 | 0.50 |
| FIB (g/L) | 2.547±0.594 | 2.417±0.319 | 0.30 |
| D-Dimer (mg/L) | 1.409±1.754 | 1.011±1.364 | 0.26 |
Values are given as mean ± SD. AFP, alpha-fetoprotein; BAS, basophil; CEA, carcinoembryonic antigen; EOS, eosinophil; FIB, fibrinogen; HGB, hemoglobin; LYM, lymphocyte; MON, monocyte; NEU, neutrophil; PLT, platelet; RBC, red blood cell; SD, standard deviation; WBC, white blood cell.
Multivariate statistical analysis
The data obtained were analyzed by PCA to produce an overview of the changes and determine the differences between the metabolic profiles of the gastric cancer patient before (first) and after (second) receiving oxaliplatin chemotherapy twice. After the Pareto scale of average centering, the data were displayed in the form of a fraction of the potential variable coordinate system, which is the result obtained from above samples. However, PCA is an unsupervised analysis form, which could not eliminate the intra-group error and random error irrelevant to the purpose of the study. This is not conducive to the discovery of inter-group differences and differential compounds. As shown in Figure 1A, PCA distribution between the two groups showed no significant trend of differences due to the limitations between first and second groups. To find the metabolites causing the differences between the first and second groups, OPLS-DA was employed. OPLS-DA is a guided model designed to minimize system noise and extract variable data (21,29). To test the robustness of our statistical model, we used the Y-matrix permutation method, generating 200 OPLS-DA models with a randomized Y-matrix. The values of R2 and Q2 were 0.95 and 0.02, respectively (Figure 1B), which indicated that the original model led to better sensitivity and predictivity that other random models, hence validating the observed result. As shown in Figure 1C, the score plots of OPLS-DA showed that first and second groups can be distinguished clearly with good model fitness and predictability. The model parameters were R2X=0.286, R2Y=0.957, Q2=0.627.
Figure 1.
The multivariate statistical analysis of first and second groups. The PCA score plots based (A). Overfitting analysis of the OPLS-DA model (B). OPLS-DA score plot (C). OPLS-DA, orthogonal partial least squares discriminant analysis; PCA, principal component analysis.
Analysis of differential expressed metabolites (DEMs)
Ion post-peaks with qualitative results considered inaccurate were systematically removed by applying the established reference threshold. The untargeted LC-MS analysis identified 340 metabolites, with 183 detected in positive ion mode and 157 in negative ion mode. The detailed information was in available online: https://cdn.amegroups.cn/static/public/tcr-2025-240-2.xlsx. Based on the obtained model, the metabolites that contribute the most to the differences between the two groups were selected for further analysis according to the VIP score. VIP score value >1 and P value <0.05 were selected as cut-off values. As shown in Table 3, 16 DEGs were identified between first and second groups in positive and negative modes. In these 16 DEMs, 10 DEMs (including catechol, norepinephrine, gabapentin, 9,10-DHOME, S-adenosylmethionine, 2,22-dideoxy-3-dehydroecdysone, shikimic acid, hippuric acid, 5-hydroxyindoleacetic acid, procollagen 5-hydroxy-L-lysine) were up-regulated in second group compared with first group, 6 DEMs (including diaminopimelic acid, farnesoic acid, desmosterol, 4-quinolinecarboxylic acid, N-acetyl-L-aspartic acid, 9,10-epoxyoctadecenoic acid) were down-regulated in the second group compared with first group. Significant differences are highlighted in the volcano plot shown in Figure 2A. Among the 16 DEMs, amino acids and derivatives were the most numerous (37.5%, 6), followed by lipids and derivatives (25%, 4) organic acids and derivatives (18.75%, 3), phenols (5%, 2), and others (6.5%, 1) (Table 3 and Figure 2B).
Table 3. The DEMs in the serum metabolic profiles of first and second groups in both positive and negative modes.
| Name | MZ | RT | FC | P | FDR | VIP | Trend | Classification |
|---|---|---|---|---|---|---|---|---|
| N-acetyl-L-aspartic acid | 174.04 | 45.00 | 0.41 | 0.048 | 0.99 | 1.97 | ↓ | Amino acids and derivatives |
| S-adenosylmethionine | 398.24 | 463.80 | 1.51 | 0.02 | 0.94 | 2.27 | ↑ | Amino acids and derivatives |
| Hippuric acid | 178.05 | 103.40 | 1.04 | 0.04 | 0.99 | 2.26 | ↑ | Amino acids and derivatives |
| Procollagen 5-hydroxy-L-lysine | 197.81 | 455.80 | 1.04 | 0.01 | 0.99 | 2.47 | ↑ | Amino acids and derivatives |
| Gabapentin | 172.13 | 574.00 | 1.14 | 0.03 | 0.94 | 2.26 | ↑ | Amino acids and derivatives |
| Diaminopimelic acid | 191.041 | 50.60 | 0.85 | 0.04 | 0.94 | 2.29 | ↓ | Amino acids and derivatives |
| Farnesoic acid | 236.16 | 388.80 | 0.55 | 0.049 | 0.94 | 2.30 | ↓ | Lipids |
| 9,10-DHOME | 315.25 | 626.00 | 1.30 | 0.02 | 0.94 | 2.48 | ↑ | Fatty acids and conjugates |
| Desmosterol | 384.35 | 439.10 | 0.76 | 0.04 | 0.94 | 2.09 | ↓ | Steroids and steroid derivatives |
| 9,10-epoxyoctadecenoic acid | 295.22 | 500.90 | 0.93 | 0.03 | 0.99 | 2.56 | ↓ | Fatty acids and conjugates |
| Shikimic acid | 172.99 | 237.00 | 1.35 | 0.02 | 0.99 | 2.66 | ↑ | Organic acids and derivatives |
| 4-quinolinecarboxylic acid | 173.12 | 300.10 | 0.51 | 0.03 | 0.99 | 2.26 | ↓ | Organic acids and derivatives |
| 5-hydroxyindoleacetic acid | 191.11 | 452.20 | 1.41 | 0.03 | 0.99 | 2.01 | ↑ | Organic acids and derivatives |
| Catechol | 110.02 | 369.80 | 1.16 | 0.043 | 0.94 | 2.03 | ↑ | Phenols |
| Norepinephrine | 169.98 | 73.60 | 1.34 | 0.02 | 0.94 | 2.02 | ↑ | Phenols |
| 2,22-dideoxy-3-dehydroecdysone | 430.29 | 386.50 | 1.18 | 0.03 | 0.94 | 2.49 | ↑ | Others |
↑, increase; ↓, decrease. DEMs, differentially expressed metabolites; FC, fold change; FDR, false discovery rate; MZ, mass-to-charge ratio; RT, retention time; VIP, variable importance in projection.
Figure 2.
Comparison of the abundance of 16 DEMs. Colored spots are metabolites that changed significantly. Blue: decreased levels; red: increased levels (A). The classification of 16 DEMs (B). DEMs, differentially expressed metabolites; FC, fold change; VIP, variable importance in projection.
As shown in Figure 3, the t-test generated estimation plots for 16 DEMs were analyzed using GraphPad Prism 9. The scatter on the left side of the plot displays the raw data of the first and second groups. The red square represents the second group, and the blue circle represents the first group, each circle or square represents each patient, and the left figure can visually show the difference in DEMs content of the same patient after two rounds of oxaliplatin chemotherapy, while the right side displays the magnitude of the effect (mean difference) and its 95% confidence interval.
Figure 3.
The t-test generated estimation plots for 16 DEMs (blue-first group, red-second group), green rod-shaped represents the magnitude of the effect (mean difference) and its 95% confidence interval. DEMs, differentially expressed metabolites.
Screening for biomarkers
SHAP analysis, grounded in the principles of game theory and local explanations, falls under the category of established post hoc interpretive methods. This approach enables the computation of Shapley values, which in turn are employed to quantify the individual contributions of each feature. Versatile and adaptable, SHAP is compatible with a wide range of machine learning algorithms, such as K-Nearest Neighbor, Random Forest, Support Vector Machine, Gaussian Naive Bayes, Logistic Regression, and Decision Tree. In our research, we applied SHAP analysis to a set of 16 DEMs, using a random forest model to ascertain their Shapley values, which serve as a measure of their significance. As depicted in Figure 4A, these Shapley values were presented for each sample. The importance of the metabolites is ranked based on the absolute average Shapley value, which was then used to normalize the quantitative data for each metabolite within the samples. This normalization helps illustrate how the importance of each feature data point influences the model’s outcomes. The vertical axis in the figure lists the 16 metabolites that exhibit differential effects, while the horizontal axis displays the Shapley scores predicted by the Random Forest model for the test set samples. These scores represent the contribution of each metabolite to the classification prediction of the sample. The magnitude of these scores indicates the relative contribution of the metabolite to the classification results, with higher absolute values suggesting greater importance. The color gradient in the visualization corresponds to the standardized (scaled) characteristic values (quantitative metabolite values) across different samples, with red denoting higher values and blue indicating lower values. As shown in Figure 4B, the most impactful DEMs that propel the model’s top five rankings included norepinephrine, 9,10-DHOME, 4-quinolinecarbolic acid, 5-hydroxyindoleacetic acid, and Procollagen 5-hydroxy-L-lysine.
Figure 4.
The analysis of biomarkers. The bee diagram of DEMs (A); the importance ranking of DEMs based on SHAP analysis (B). ROC curve analysis of 16 DEMs, the AUC of DEMs exceeded 0.7 were marked in red (C). AUC, area under the curve; DEMs, differentially expressed metabolites; ROC, receiver operating characteristic; SHAP, SHapley Additive exPlanations.
Receiver operating characteristic (ROC) curves were used to assess the discriminative power of the 16 metabolites that were differentially altered. To assess prediction accuracy, the area under the curve, known as the AUC, is utilized (30). Typically, AUC values ranging from 0.5 to 0.7 indicate low predictive accuracy, while those from 0.7 to 0.9 show moderate predictive accuracy, values reaching 0.9 and above reflect excellent predictive accuracy, but an AUC of 0.5 means the biomarker fails to predict the event, offering no predictive insight. In Figure 4C, The AUC values of 16 DEMs were all between 0.5 and 0.9, among which the AUC values of norepinephrine, 9,10-DHOME, 2,22-deoxy-3-hydroecdysone, 5-hydroxyindoleacetic acid, and procollagen 5-hydroxy-lysine are greater than 0.7, which indicated that these DEMs were likely to have some predictive accuracy. These metabolites may have potential applications in identifying OIPN in the same gastric cancer patient before and after receiving oxaliplatin chemotherapy twice. By integrating the findings from SHAP and ROC analysis, we have identified norepinephrine, 9,10-DHOME, 5-hydroxyindoleacetic acid, and procollagen 5-hydroxy-lysine as key biomarkers for OIPN in a single gastric cancer patient, both prior to and subsequent to undergoing two cycles of oxaliplatin chemotherapy.
Analysis of metabolic pathways
The DEMs were used to analyze the differential metabolic pathways. MetaboAnalyst 4.0 was used to analyze differential metabolic pathways to obtain biological information on metabolic pathway-related networks that changed in the same gas cancer patient before and after receiving oxaliplatin chemotherapy twice. KEGG pathway analysis for functional enrichment identified the top 20 metabolic pathways, as shown in Figure 5A,5B. Linoleic acid metabolism, adrenergic signaling in cardiomyocytes, sulfur relay system, gap junction, and synaptic vesicle cycle were the top 5 metabolic pathways in first and second groups. Out of the 20 leading metabolic pathways, six were linked to amino acid metabolism (including alanine, aspartate and glutamate metabolism, phenylalanine, tyrosine and tryptophan biosynthesis, lysine degradation, phenylalanine metabolism, and cysteine and methionine metabolism), three pathways were associated with lipids metabolism (linoleic acid metabolism, steroid biosynthesis, regulation of lipolysis in adipocytes), three pathways were associated with neurological metabolism (synaptic vesicle cycle, vascular smooth muscle contraction, neuroactive ligand-receptor interaction). The detailed information was in available online: https://cdn.amegroups.cn/static/public/tcr-2025-240-3.xlsx.
Figure 5.
The KEGG analysis of DEMs. Metabolic pathway influence factor bubble chart (A). Pathway analysis of the DEMs identified in first and second groups (B). DEMs, differentially expressed metabolites; KEGG, Kyoto Encyclopedia of Genes and Genomes.
Definition and verification of DEMs between two groups
During the validation phase, a thorough examination of 42 serum samples from another 21 gastric cancer patients (as outlined in available online: https://cdn.amegroups.cn/static/public/tcr-2025-240-1.xlsx) was performed to verify the consistency of metabolites showing differential changes. This process aimed to accurately identify key metabolites capable of effectively distinguishing the same gastric cancer patients before and after receiving two cycles of oxaliplatin chemotherapy, thereby establishing them as promising biomarkers.
As shown in Table 1, in validation set, we collected another 42 serum samples from 21 gastric cancer patients who received oxaliplatin chemotherapy twice before and after. Female patients accounted for 28.57%, while male patients accounted for 71.43%. Eighteen patients, accounting for 85.71%, developed OIPN after the first chemotherapy, and all patients developed OIPN after the second chemotherapy, which indicate that OIPN has a certain degree of accumulation. In the term of the symptoms of OIPN, patients showed more severe symptoms after the second chemotherapy. After the first chemotherapy, patients concentrated on numbness in their fingers. After the second chemotherapy, three patients showed numbness in their hands, feet, and mouth. In the term of symptom duration and symptom on set time, the OIPN of patients after the second chemotherapy is more severe than the first one, two patients had symptoms that lasted for more than 20 days. The results of the validation set were consistent with the discovery set. There were no significant differences in WBC, NEU, LYM, MON, EOS, RBC, HGB, PLT, AFP, CEA, FIB, and D-Dimer between the first and second chemotherapy group. However, compared with the first chemotherapy group, the second chemotherapy group had a lower BAS count (P<0.05) (Table S1). The detailed information was in available online: https://cdn.amegroups.cn/static/public/tcr-2025-240-1.xlsx.
As shown in Figure 6A, PCA score plot between the two groups showed no significant trend of differences due to the limitations between first and second groups. The permutation test results in Figure 6B further verified the model’s robustness and the OPLS-DA model in Figure 6C demonstrated significant separation between the first and second groups in the validation set. By applying the same analytical methodologies and statistical criteria as previously employed, 33 significantly DEMs were identified in the validation set (available online: available online: https://cdn.amegroups.cn/static/public/tcr-2025-240-4.xlsx). The volcano plots of 33 DEMs were in Figure 6D. Significantly, norepinephrine appeared as a metabolite with consistent and statistically significant variations across the discovery and validation datasets (Figure 6E). The AUC value of norepinephrine was 0.713, indicating that norepinephrine was likely to have some predictive accuracy (Figure 6F).
Figure 6.
The analysis of first and second groups in validation set. The PCA score plots based (A). Overfitting analysis of the OPLS-DA model (B). OPLS-DA score plot (C). The volcano plots of 33 DEMs in validation set (D). The t-test generated estimation plots for norepinephrine in validation set (E). The ROC curve analysis of norepinephrine (F). The top 20 metabolic pathways of DEMs (G). AUC, area under the curve; DEMs, differentially expressed metabolites; FC, fold change; OPLS-DA, orthogonal partial least squares discriminant analysis; PCA, principal component analysis; ROC, receiver operating characteristic.
Figure 6G displays the top 20 metabolic pathways identified through KEGG pathway analysis for functional enrichment. Similar to the discovery set, among these 20 metabolic pathways, four pathways were associated with amino acid metabolism (including arginine and proline metabolism, tyrosine metabolism, beta-alanine metabolism, and glutathione metabolism), two pathways were linked to lipid metabolism (comprising the regulation of lipolysis in adipocytes and the biosynthesis of unsaturated fatty acids), while vascular smooth muscle contraction was correlated with neurological metabolism.
Discussion
OIPN, an adverse effect of oxaliplatin, severely impacts the health and well-being of patients, making early identification of susceptible individuals essential in clinical practice. The objective of our research was to pinpoint potential biomarkers and key metabolic pathways to differentiate serum samples from the same gastric cancer patients before and after two rounds of oxaliplatin chemotherapy. From the clinical manifestations, the OIPN symptoms after the second chemotherapy are more severe than those after the first chemotherapy (including symptom duration, symptom onset time, and symptom manifestations). We used UHPLC-Q-Exactive Orbitrap-MS/MS to identify 340 metabolites, among which 16 DEMs were screened in discovery set. These DEMs belonged to amino acids and derivatives, lipids and derivatives, organic acids and derivatives, and others, mainly involved in amino acid metabolism, lipid metabolism, and nervous system metabolism, among them, four DEMs (including norepinephrine, 5-hydroxyindoleacetic acid, 9,10-DHOME, and procollagen 5-hydroxy-lysine) have certain predictive ability for OIPN in the same gastric cancer patient before and after receiving oxaliplatin chemotherapy twice. It is worth noting that the contents of these four DEMs were all higher in the second group than in the first group. Among the 33 DEMs identified in the validation set, norepinephrine emerged as a consensus biomarker shared between the discovery and validation sets, demonstrating robust diagnostic performance with AUC values exceeding 0.7. This finding confirms its significance as a potential biomarker for detecting OIPN in gastric cancer patients undergoing longitudinal monitoring before and after two cycles of oxaliplatin-based chemotherapy.
The identified metabolic pathways (including amino acid metabolism, lipid metabolism, and neurological metabolism) directly intersect with known OIPN mechanisms. The tryptophan metabolite kynurenine can induce central sensitization by activating N-methyl-D-aspartate receptors while promoting the generation of reactive oxygen species, thereby exacerbating neuroinflammation and allodynia. Oxaliplatin exacerbates oxidative stress by depleting glutathione, leading to dysfunction of neuronal mitochondria. Glutathione synthesis is dependent on metabolic pathways involving cysteine, glutamate, and glycine (31). Oxaliplatin induces lipid peroxidation products, such as 4-hydroxynonenal, which disrupt the phospholipid bilayer structure of neuronal cell membranes, leading to sodium/potassium ion channel dysfunction. This damage is linked to the suppression of the NRF2 signaling pathway, while NRF2 activation upregulates glutathione peroxidase to inhibit ferroptosis. Additionally, dorsal root ganglion neurons, which predominantly rely on fatty acid β-oxidation for energy production, exhibit compromised mitochondrial metabolism under oxaliplatin treatment. The drug inhibits peroxisome proliferator-activated receptor α, thereby reducing the expression of mitochondrial carnitine palmitoyltransferase 1, which prevents long-chain fatty acids from entering mitochondria for oxidation. This impairment exacerbates energy depletion and disrupts neuronal bioenergetics, further contributing to OIPN (31). In the OIPN model, impaired function of the KEAP1-NRF2 axis leads to downregulation of antioxidant enzymes such as GSTP1. formononetin activates NRF2 by binding to the His129/Lys131 sites of KEAP1, thereby restoring mitochondrial homeostasis and alleviating oxidative damage in vascular smooth muscle contraction. This finding suggests that antioxidant intervention may improve vasomotor function (32).
In the developing nervous system, neurotransmitters are crucial, and norepinephrine is especially thought to be a significant factor in brain development. During the early stages of development, norepinephrine is expressed and is responsible for regulating the development of noradrenergic neurons and their target areas (33). Norepinephrine induces vascular smooth muscle contraction by activating α1-adrenergic receptors (α1-AR), leading to peripheral microcirculatory disorders. In OIPN, oxaliplatin promotes the release of norepinephrine from sympathetic nerve endings, triggering persistent contraction of endoneurial blood vessels and exacerbating ischemic injury of dorsal root ganglia (DRG). Animal experiments show that α1-AR antagonists significantly improve oxaliplatin-induced decline in nerve conduction velocity. Excessive release of norepinephrine promotes lipolysis in adipose tissue by activating β3-adrenergic receptors, increasing free fatty acid (FFA) levels. Accumulation of FFAs in DRG neurons induces the opening of mitochondrial membrane permeability transition pores, leading to ATP synthesis dysfunction and increased reactive oxygen species production, thereby exacerbating axonal degeneration. Moreover, norepinephrine upregulates the PPARγ signaling pathway in adipose tissue macrophages, promoting the secretion of growth differentiation factor 15 (GDF15). GDF15 crosses the blood-brain barrier to act on the central amygdala and hypothalamic paraventricular nucleus, activating the hypothalamic-pituitary-adrenal axis and promoting the release of pro-inflammatory cytokines [such as interleukin 6 (IL-6) and tumor necrosis factor-alpha (TNF-alpha)], forming a vicious cycle of neuroinflammation (34).
Norepinephrine, as a biomarker for OIPN, has shown significant early predictive potential and dynamic monitoring value in research: its level significantly increases after the second oxaliplatin chemotherapy, and is positively correlated with the severity of OIPN. Its diagnostic performance across the discovery and validation sets is stable (AUC >0.7), which is superior to traditional biomarkers such as neurofibrillary light chain (NfL) (although NfL shows a dose-dependent increase after oxaliplatin exposure, individual differences are large and lag behind symptom onset) (35); in addition, SHAP analysis showed that norepinephrine had the highest contribution in differential metabolites, suggesting that it may directly participate in the “ischemia inflammation” vicious cycle of OIPN by regulating α 1-adrenergic receptor-mediated vasoconstriction and mitochondrial oxidative stress, with both mechanism explanation and potential intervention targets. However, the specificity of norepinephrine is limited by confounding factors such as sympathetic nervous system activity, cardiovascular disease, or drug interference, and its detection relies on high-sensitivity LC-MS/MS or electrochemical probe technology, resulting in high clinical promotion costs.
Conclusions
Employing an untargeted metabolomics strategy utilizing UHPLC-Q-Exactive Orbitrap-MS/MS, the current research showed a significant change in the serum metabolic profile of the same gastric cancer patients before and after receiving oxaliplatin chemotherapy twice. From the clinical presentation, the OIPN symptoms after the second chemotherapy were more severe than those after the first chemotherapy. Pathway analysis suggested that several pathways especially the amino acid metabolism, lipids metabolism, and nervous system metabolism differed significantly among the patients before and after receiving chemotherapy, which may be related to the pathological processing of worsening OIPN symptoms and functional deterioration. Additionally, norepinephrine with good diagnostic performance was identified in discovery and validation set. Overall, these findings could help clarify the molecular mechanisms involved in the progression of OIPN and enhance current methods for its early detection and outcome forecasting. Not with standing these findings, this study has several limitations that warrant rigorous investigation in future research. Firstly, validation of the identified DEMs requires multicenter, longitudinal cohort studies with expanded sample sizes, adjusting for potential confounders, to establish their clinical relevance and pathophysiological specificity for OIPN. Secondly, functional validation through preclinical models is imperative to causally link the dysregulated metabolic pathways to neuroaxonal injury mechanisms.
Supplementary
The article’s supplementary files as
Acknowledgments
None.
Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted according to the Declaration of Helsinki and its subsequent amendments, and approved by the Ethics Committee of the Affiliated Hospital of Jiangnan University (No. LS2022080). The patients and participants provided their written informed consent to participate in this research.
Footnotes
Reporting Checklist: The authors have completed the STARD reporting checklist. Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-240/rc
Funding: This work was supported by the Lean Medication and Stone Medicine Special Research Fund Supported Project (No. JY202235); Wuxi Municipal Health Commission Youth Project (No. Q202339); and the Medical Research Project of China Medical and Health Development Foundation (No. 2024-06).
Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-240/coif). The authors have no conflicts of interest to declare.
Data Sharing Statement
Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-240/dss
References
- 1.Wang FY, Huang XM, Cao YQ, et al. Comparison of PSOX (paclitaxel, oxaliplatin, S-1) and SOX (oxaliplatin, S-1) as postoperative adjuvant chemotherapy for stage II-III gastric cancer. World J Surg Oncol 2025;23:75 . 10.1186/s12957-025-03723-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Wang X, Xi M, Lu X, et al. The Role and Mechanism of Innate Immune Regulation in Overcoming Oxaliplatin Resistance and Enhancing Anti-Tumor Efficacy in Colorectal Cancer. Pharmaceuticals (Basel) 2025;18:317 . 10.3390/ph18030317 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Abhang K, Dighe S, Katari O, et al. Dual acting oxaliplatin (IV) prodrug loaded albumin nanoparticles for safer synergistic anticancer action against triple negative breast cancer. Drug Deliv Transl Res 2025. [Epub ahead of print]. doi: . 10.1007/s13346-025-01833-9 [DOI] [PubMed] [Google Scholar]
- 4.Jordan B, Jahn F, Jordan K. Peripheral neuropathy: from guidelines to clinical practise. Curr Opin Oncol 2025;37:168-74. 10.1097/CCO.0000000000001112 [DOI] [PubMed] [Google Scholar]
- 5.Yang Y, Zhao B, Gao X, et al. Targeting strategies for oxaliplatin-induced peripheral neuropathy: clinical syndrome, molecular basis, and drug development. J Exp Clin Cancer Res 2021;40:331 . 10.1186/s13046-021-02141-z [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Cheng F, Zhang R, Sun C, et al. Oxaliplatin-induced peripheral neurotoxicity in colorectal cancer patients: mechanisms, pharmacokinetics and strategies. Front Pharmacol 2023;14:1231401 . 10.3389/fphar.2023.1231401 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Zhou J, Hou D, Zou W, et al. Comparison of Widely Targeted Metabolomics and Untargeted Metabolomics of Wild Ophiocordyceps sinensis. Molecules 2022;27:3645 . 10.3390/molecules27113645 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Tian J, Zhu X, Wu H, et al. Serum metabolic profile and metabolome genome-wide association study in chicken. J Anim Sci Biotechnol 2023;14:69 . 10.1186/s40104-023-00868-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Gil-de-la-Fuente A, Mamani-Huanca M, Stroe MC, et al. Aspergillus Metabolome Database for Mass Spectrometry Metabolomics. J Fungi (Basel) 2021;7:387 . 10.3390/jof7050387 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Hua YJ, Xie F, Mao KJ, et al. Insights into the metabolite profiles of Rubus chingii Hu at different developmental stages of fruit. J Sep Sci 2023;46:e2300264 . 10.1002/jssc.202300264 [DOI] [PubMed] [Google Scholar]
- 11.Johnson CH, Ivanisevic J, Siuzdak G. Metabolomics: beyond biomarkers and towards mechanisms. Nat Rev Mol Cell Biol 2016;17:451-9. 10.1038/nrm.2016.25 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.MahmoudianDehkordi S , Arnold M, Nho K, et al. Altered bile acid profile associates with cognitive impairment in Alzheimer’s disease-An emerging role for gut microbiome. Alzheimers Dement 2019;15:76-92. 10.1016/j.jalz.2018.07.217 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Poewe W. Smelling Parkinson’s Disease: New Metabolomic Biomarker for PD. ACS Cent Sci 2019;5:575-6. 10.1021/acscentsci.9b00319 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Lee EJ, Kim DJ, Kang DW, et al. Targeted Metabolomic Biomarkers for Stroke Subtyping. Transl Stroke Res 2024;15:422-32. 10.1007/s12975-023-01137-5 [DOI] [PubMed] [Google Scholar]
- 15.Humer E, Pieh C, Probst T. Metabolomic Biomarkers in Anxiety Disorders. Int J Mol Sci 2020;21:4784 . 10.3390/ijms21134784 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Li Z, Zhang T, Xu L, et al. Plasma metabolic alterations and potential biomarkers in individuals at clinical high risk for psychosis. Schizophr Res 2022;239:19-28. 10.1016/j.schres.2021.11.011 [DOI] [PubMed] [Google Scholar]
- 17.Xu Y, Zhao Y, Guo X, et al. Plasma metabolic profiling analysis of neurotoxicity induced by oxaliplatin using metabonomics and multivariate data analysis. Toxicol Res (Camb) 2018;7:529-37. 10.1039/c7tx00345e [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Li Y, Wang L, Ju L, et al. A Systematic Strategy for Screening and Application of Specific Biomarkers in Hepatotoxicity Using Metabolomics Combined With ROC Curves and SVMs. Toxicol Sci 2016;150:390-9. 10.1093/toxsci/kfw001 [DOI] [PubMed] [Google Scholar]
- 19.Li Y, Deng H, Ju L, et al. Screening and validation for plasma biomarkers of nephrotoxicity based on metabolomics in male rats. Toxicol Res (Camb) 2016;5:259-67. 10.1039/c5tx00171d [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Wang C, Hu H, Liu J, et al. Study on the mechanism of brain injury caused by acute diquat poisoning based on metabolomics. Toxicol Appl Pharmacol 2025;494:117161 . 10.1016/j.taap.2024.117161 [DOI] [PubMed] [Google Scholar]
- 21.Hua Y, Lv J, Zhang Y, et al. LC-MS-based serum metabolomics analysis and potential biomarkers for oxaliplatin induced neurotoxicity in colorectal cancer. J Pharm Biomed Anal 2025;252:116492 . 10.1016/j.jpba.2024.116492 [DOI] [PubMed] [Google Scholar]
- 22.Rasmussen JA, Villumsen KR, Ernst M, et al. A multi-omics approach unravels metagenomic and metabolic alterations of a probiotic and synbiotic additive in rainbow trout (Oncorhynchus mykiss). Microbiome 2022;10:21 . 10.1186/s40168-021-01221-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Navarro-Reig M, Jaumot J, García-Reiriz A, et al. Evaluation of changes induced in rice metabolome by Cd and Cu exposure using LC-MS with XCMS and MCR-ALS data analysis strategies. Anal Bioanal Chem 2015;407:8835-47. 10.1007/s00216-015-9042-2 [DOI] [PubMed] [Google Scholar]
- 24.Lv J, Hua YJ, Ding YJ. Comprehensive Metabolomic Analysis of Lung Cancer Patients Treated with Fu Zheng Fang. Curr Pharm Anal 2022;18:881-91. [Google Scholar]
- 25.Hua Y, Dai B, Luo Y, et al. Integrated analysis of multiple metabolome and transcriptome revealed the accumulation of flavonoids and associated molecular regulation mechanisms in Rubus chingii Hu at different developmental stages. Plant Physiol Biochem 2023;204:108085 . 10.1016/j.plaphy.2023.108085 [DOI] [PubMed] [Google Scholar]
- 26.Hua YJ, Xie F, Liu XY, et al. Comprehensive metabolomics analysis of key taste components in different varieties of table grapes. J Sep Sci 2022;45:3700-13. 10.1002/jssc.202200137 [DOI] [PubMed] [Google Scholar]
- 27.Lu Q, Li Y, Ye D, et al. Longitudinal metabolomics integrated with machine learning identifies novel biomarkers of gestational diabetes mellitus. Free Radic Biol Med 2023;209:9-17. 10.1016/j.freeradbiomed.2023.10.014 [DOI] [PubMed] [Google Scholar]
- 28.Qi X, Wang S, Fang C, et al. Machine learning and SHAP value interpretation for predicting comorbidity of cardiovascular disease and cancer with dietary antioxidants. Redox Biol 2025;79:103470 . 10.1016/j.redox.2024.103470 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Xing L, Sun L, Liu S, et al. Comparative metabolomic analysis of the body wall from four varieties of the sea cucumber Apostichopus japonicus. Food Chem 2021;352:129339 . 10.1016/j.foodchem.2021.129339 [DOI] [PubMed] [Google Scholar]
- 30.Tu C, Gao Y, Song D, et al. Screening for Susceptibility-Related Biomarkers of Diclofenac-Induced Liver Injury in Rats Using Metabolomics. Front Pharmacol 2021;12:693928 . 10.3389/fphar.2021.693928 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Fang Y, Ye J, Zhao B, et al. Formononetin ameliorates oxaliplatin-induced peripheral neuropathy via the KEAP1-NRF2-GSTP1 axis. Redox Biol 2020;36:101677 . 10.1016/j.redox.2020.101677 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Zhang Z, Ye J, Liu X, et al. Huangqi Guizhi Wuwu decoction alleviates oxaliplatin-induced peripheral neuropathy via the gut-peripheral nerve axis. Chin Med 2023;18:114 . 10.1186/s13020-023-00826-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Saboory E, Ghasemi M, Mehranfard N. Norepinephrine, neurodevelopment and behavior. Neurochem Int 2020;135:104706 . 10.1016/j.neuint.2020.104706 [DOI] [PubMed] [Google Scholar]
- 34.Ye C, Zheng F, Xu T, et al. Norepinephrine acting on adventitial fibroblasts stimulates vascular smooth muscle cell proliferation via promoting small extracellular vesicle release. Theranostics 2022;12:4718-33. 10.7150/thno.70974 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Gehr NL, Mortensen C, Stage TB, et al. Neurofilament light chain as a marker for neuronal damage: integrating in vitro studies and clinical findings in patients with oxaliplatin-induced neuropathy. Cancer Chemother Pharmacol 2025;95:53 . 10.1007/s00280-025-04773-w [DOI] [PMC free article] [PubMed] [Google Scholar]