Metabolomic Insights into the Predictive Landscape of Neoadjuvant Immunochemotherapy in Gastric Cancer: Towards Precision Medicine with Machine Learning

Metabolomic Insights into the Predictive Landscape of Neoadjuvant Immunochemotherapy in Gastric Cancer: Towards Precision Medicine with Machine Learning | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Metabolomic Insights into the Predictive Landscape of Neoadjuvant Immunochemotherapy in Gastric Cancer: Towards Precision Medicine with Machine Learning Wei Zhu, Xiaoyong Zhang, Xueli Hu, Kankai Zhu, Huidi Zhu, Shenghong Guan, and 10 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6365771/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Neoadjuvant Immunochemotherapy (NAIC) shows promising application prospects in the treatment of (Gastric Cancer, GC) patients. However, the differences between individual patients and the issue of treatment resistance significantly affect whether patients could truly benefit from this treatment. This study conducted metabolomic analysis of 369 plasma samples from 108 gastric cancer patients following NAIC treatment to characterize their metabolic profiles for more accurate therapeutic efficacy prediction. Machine learning was used to build a GC treatment response prediction model 21-PM based on the expression levels of baseline metabolites. Two efficacy monitoring models, 11-MMI and 13-MMP, were developed using R-specific metabolites based on imaging and histology outcomes, respectively. In conclusion, the outcomes of this study offered strong evidence for the advancement of precision medicine in GC by exposing the metabolic landscape of GC patients after NAIC treatment and efficiently creating models that could independently predict prediction and monitor treatment. Health sciences/Oncology/Cancer Health sciences/Biomarkers/Predictive markers Gastric cancer Neoadjuvant Immunochemotherapy Machine learning Metabolomics Prediction Model Monitoring Model Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Introduction Gastric cancer (GC) is the sixth most prevalent malignant tumor, accounting for over 33% of all cancer-related fatalities. Currently, 70–90% of GC patients are discovered at an advanced stage, significantly limiting their prognosis. Although surgery and chemotherapy have improved the survival rate of advanced GC patients, the overall survival (OS) rate for GC patients is still less than 40%, and more than half of GC patients have recurrence 1 , 2 . Following the advent of comprehensive treatment strategies such as perioperative chemotherapy, preoperative chemotherapy, and neoadjuvant chemotherapy, patients' survival and prognosis improved considerably 3 – 6 . However, overall, the prognosis for patients with locally advanced gastric cancer remains unsatisfactory. The advent of immune checkpoint inhibitors (ICIs) such as tremelimumab, toripalimab, sintilimab, and nivolumab had resulted in new advances in the treatment of metastatic cancer. Preoperative neoadjuvant immunochemotherapy (NAIC) has been shown in studies to considerably slow the growth of GC tumors. The CheckMate649 trial found that NAIC treatment (Nivolumab plus chemotherapy) had a positive therapeutic effect on patients, ushering in a new age of immunotherapy for GC 7 , 8 . Another prospective, single-arm, phase II clinical trial shown that NAIC achieved a 30% pathologic complete regression rate (pCR) and a 43% major pathologic regression rate (MPR), with good patient tolerance, demonstrating the clinical benefits of NAIC in gastric cancer treatment 9 . Furthermore, Verschoor et al. used neoadjuvant atezolizumab in conjunction with chemotherapy in gastric and gastroesophageal junction (G/GEJ) tumors, which demonstrated significant anticancer activity, with 70% (95% CI: 46%-88%) of patients reaching MPR and 45% (95% CI: 23%-68%) obtaining pCR 10 . Many patients with stomach cancer still do not benefit from NAIC, despite the fact that it has been demonstrated to successfully induce pathological regression in locally advanced gastric adenocarcinoma and can, to some extent, enhance patient survival 11 . Imaging, endoscopic, and histological exams were mostly used to assess the clinical prognosis and efficacy of GC 12 . The aforementioned techniques were used by clinical physicians to gather clinical features for efficacy evaluation, such as tumor location, TNM staging data, and Lauren categorization; however, the accuracy of these evaluation techniques was restricted 13 , 14 . Imaging examinations exposed patients to radiation, and while endoscopy was the gold standard for diagnosing GC, it was intrusive and costly. These restrictions hampered the quick monitoring of patient treatment efficacy 15 . This situation underscores the urgent need to identify reliable biomarkers for patient stratification during NAIC treatment. Tumor growth was frequently accompanied by the reprogramming of metabolic components in the body. This process was primarily triggered by the accumulation of metabolic alterations resulting from mutations in enzyme-coding genes during tumor cell proliferation. Epigenetic changes in proteins or inhibitors within mature tumor cells caused cancer while also having a direct impact on metabolism 16 . Furthermore, when cancers evolve with genetic factors in their microenvironment, tumor cells acquired metabolic alterations, which was one of the primary driving forces for intratumoral and intertumoral variability 16 . As a result, elucidating the metabolic reprogramming landscape during NAIC treatment of GC and identifying biomarkers that represent NAIC efficacy could improve NAIC's clinical efficacy, allowing for more timely and effective treatment strategies for GC patients. Metabolomics technology, as a systematic analysis, had the potential to better reflect the net outcomes of NAIC-tumor interactions. Previous research had mostly focused on molecular pathways to identify biomarkers linked with the occurrence, prognosis, and recurrence of GC. However, they lacked the characterization of the metabolic reprogramming profile of tumor patients undergoing NAIC, as well as the identification of metabolites associated with the efficacy of NAIC treatment for GC 17 . This significantly impedes real-time monitoring of the GC treatment process, making it harder for NAIC to achieve superior therapeutic outcomes. In this work, metabolomics was employed to assess the influence of NAIC reconfiguration on the metabolic profile of GC patients. To better identify biomarkers linked with NAIC therapy success, this study conducted a comprehensive examination across three distinct clinical evidence dimensions, as follows: First, an efficacy prediction model was created using baseline plasma samples in order to predict whether gastric cancer (GC) patients will benefit from NAIC therapy. Second, real-time clinical efficacy was dynamically assessed using imaging examinations to establish a foundation for forecasting future therapy response variations. Finally, more thorough histological examination data was used to evaluate patient clinical outcomes and find biomarkers associated with therapy monitoring. This study, which used metabolomics technology to describe the metabolic reprogramming profiles of patients with varying therapeutic outcomes, was based on the clinical empirical results of NAIC treatment for GC. It developed an efficacy monitoring model, identified efficacy-related characteristic biomarkers using machine learning, and assessed clinical efficacy based on various degrees of evidence. This served the goal of precision treatment for GC by offering a potent tool for forecasting and tracking the effectiveness of NAIC. Results 3.1 Study cohort, response information, adverse reactions, study design This study retrospectively included patients diagnosed with GC through pathology at the The First Affiliated Hospital, Zhejiang University School of Medicine. All patients received a NAIC regimen, following the standard dosing protocols described in previous studies. Initially, 154 patients were recruited, and after strict inclusion and exclusion criteria, a total of 108 patients were finally included in this study (Fig. 1 A). All patients involved in the study had baseline clinical information gathered, including as age, gender, tumor differentiation, Lauren classification, ECOG performance status score, and so on (Fig. 1 B). Among the 108 patients, the age distribution was 65.00 (58.00, 72.00) years, with 31 males (28.70%) and 77 females (71.30%). In terms of ECOG performance status scores, among patients scoring 0–3, 80.56% scored 0 and 17.59% scored 1. In the preclinical stage classification, Stage II patients accounted for 1.85%, Stage III patients accounted for 50.00%, and Stage IV patients accounted for 48.15%. In terms of treatment plans, 23.15% of patients received a chemotherapy regimen that included taxane drugs combined with PD1 inhibitors, while 76.85% of patients received a chemotherapy regimen that did not include taxane drugs combined with PD1 inhibitors. As of the analysis date of June 30, 2023, out of 108 patients who received NAIC, 65 patients showed a treatment response, with 64 achieving Partial Response (PR) and 1 achieving complete Response (CR). 43 patients did not show a treatment response, with 9 experiencing Progression Disease (PD) and 34 having Stable Disease (SD). A total of 42 patients underwent surgery after NAIC in the study, and post-operative pathological evaluation revealed pCR in 12 patients (28.57%). Becker1a/1b, Becker2, and Becker3 were observed in 21 cases (50.00%), 10 cases (23.81%), and 11 cases (26.19%) of patients, respectively. Detailed characteristics are summarized in Table 1 .The study also detailed the adverse reactions experienced by patients during treatment. In terms of the incidence of adverse reactions, the most common was nausea, with an incidence rate of 86.44%, followed by anemia at 80%, fatigue at 79.66%, anorexia at 74.58%, and leukopenia at 59.05%. It was worth noting that the frequency of grade 3 or higher severity among these adverse reactions was also relatively high. In addition, during the treatment of GC with NAIC, other adverse reactions were also observed, including diarrhea, vomiting, nausea, limb numbness, and other adverse reactions. For detailed information seeing in n Table 1 Baseline clinical features of the GC NAIC treatment population Variables Total (n = 108) NR (n = 43) R (n = 65) P value Age, n (%) 0.347 < 60 37 (34.26) 17 (39.53) 20 (30.77) ≥ 60 71 (65.74) 26 (60.47) 45 (69.23) Gender, n (%) 0.643 Female 30 (27.78) 13 (30.23) 17 (26.15) Male 78 (72.22) 30 (69.77) 48 (73.85) BMI, n (%) 0.238 18.5 ≤ BMI < 24 70 (65.42) 27 (64.29) 43 (66.15) 24 ≤ BMI < 28 19 (17.76) 5 (11.90) 14 (21.54) BMI < 18.5 15 (14.02) 9 (21.43) 6 (9.23) BMI ≥ 28 3 (2.80) 1 (2.38) 2 (3.08) ECOG, n (%) 0.214 0 87 (80.56) 32 (74.42) 55 (84.62) 1 19 (17.59) 9 (20.93) 10 (15.38) 2 1 (0.93) 1 (2.33) 0 (0.00) 3 1 (0.93) 1 (2.33) 0 (0.00) Differentiation status, n (%) 0.701 High 20 (18.52) 6 (13.95) 14 (21.54) Poor 85 (78.70) 36 (83.72) 49 (75.38) Unknown 3 (2.78) 1 (2.33) 2 (3.08) Lauren classification, n (%) 0.082 Diffuse/mixed 60 (55.56) 28 (65.12) 32 (49.23) Indeterminate 5 (4.63) 3 (6.98) 2 (3.08) Intestinal 43 (39.81) 12 (27.91) 31 (47.69) Tumor site, n (%) 0.063 GC 80 (74.07) 36 (83.72) 44 (67.69) GEJ 28 (25.93) 7 (16.28) 21 (32.31) HER2 status, n (%) 0.509 0 53 (52.48) 21 (52.50) 32 (52.46) 1+ 30 (29.70) 14 (35.00) 16 (26.23) 2+ 11 (10.89) 4 (10.00) 7 (11.48) 3+ 7 (6.93) 1 (2.50) 6 (9.84) MSI/MMR status, n (%) 1.000 MSI-H/dMMR 7 (7.07) 3 (7.69) 4 (6.67) MSS/pMMR 92 (92.93) 36 (92.31) 56 (93.33) CPS, n (%) 0.154 < 1 10 (24.39) 6 (37.50) 4 (16.00) 1–5 6 (14.63) 3 (18.75) 3 (12.00) 5–10 5 (12.20) 0 (0.00) 5 (20.00) ≥ 10 20 (48.78) 7 (43.75) 13 (52.00) Clinical stage, n (%) 0.200 Stage Ⅱ 2 (1.85) 2 (4.65) 0 (0.00) Stage Ⅲ 54 (50.00) 19 (44.19) 35 (53.85) Stage Ⅳ 52 (48.15) 22 (51.16) 30 (46.15) Clinical T category, n (%) 0.693 T2 3 (2.78) 2 (4.65) 1 (1.54) T3 37 (34.26) 14 (32.56) 23 (35.38) T4 68 (62.96) 27 (62.79) 41 (63.08) Clinical N category, n (%) 0.047 N+ 104 (96.30) 39 (90.70) 65 (100.00) N0 4 (3.70) 4 (9.30) 0 (0.00) Clinical M category, n (%) 0.409 M0 58 (53.70) 21 (48.84) 37 (56.92) M1 50 (46.30) 22 (51.16) 28 (43.08) Therapy, n (%) 0.059 Chemo(T) + PD1 25 (23.15) 14 (32.56) 11 (16.92) Chemo + PD1 83 (76.85) 29 (67.44) 54 (83.08) Surgery efficiency (Becker) < .001 Becker 1a 12 (28.57) 0 (0.00) 20 (40.00) Becker 1b 9 (21.43) 1 (8.33) 20 (26.67) Becker 2 10 (23.81) 3 (25.00) 7 (23.33) Becker 3 11 (26.19) 8 (66.67) 3 (10.00) Table S 1. Baseline plasma samples from 108 patients and 261 samples collected throughout the treatment process were gathered, with each post-treatment sample obtained at the end of each treatment cycle (up to a maximum of 6 months). The longitudinal cohort included in the study covered a total of 6 sampling time points during the treatment period. Next, a non-targeted metabolomics approach based on LC-MS/MS was used to obtain the metabolomic profile of the plasma samples. After metabolite data annotation, 1001 metabolites were obtained. Based on the baseline (T0) metabolomics of 108 patients, machine learning algorithms were applied to investigate the association between metabolic features and treatment response, resulting in the development of the 21-PM GC prediction model. This model was used to predict the response of GC patients to NAIC treatment, evaluated the impact of clinical characteristic factors, introduced risk stratification strategies, and assessed the model's ability to grade the level of risk. In order to deeply explore the changes in metabolites during the treatment process and identify potential monitoring biomarkers, the study focused on key efficacy nodes in the cohort. Machine learning was employed to construct a monitoring model (11-MMI) based on identified R-specific metabolites. Given the authority of pathological assessment in efficacy determination, the study included 42 patients whose treatment decisions were based on pathological findings. Potential monitoring markers for R group metabolites related to pathological efficacy specificity were further investigated, and a 13-MMP monitoring model was built using machine learning approaches. The development of these models aims to improve the accuracy of monitoring treatment responses in GC patients. 3.2 Dynamic Analysis of Metabolic Landscape To depict the dynamic changes in the plasma metabolic landscape of patients undergoing NAIC, plasma samples were collected from patients at baseline before surgery and across the treatment process. A total of 369 plasma samples were obtained across six defined treatment time points: pre-treatment (time point 1), first combined treatment (time point 2), second (time point 3), third (time point 4), fourth (time point 5), and fifth (time point 6) (Fig. 2 A). Using UHPLC-Q-Orbitrap-MS/MS technology, 369 plasma samples collected from 108 GC patients across treatment were analyzed. Following data batch normalization (Figure S 1) and metabolite annotation, a total of 1001 endogenous metabolites were identified. Through unsupervised principal component analysis (PCA), there was significant differences in metabolic profiles at the six time points, indicating that metabolites change as treatment progresses (Fig. 2 B). To track the dynamic changes in the metabolic landscape across the treatment process, the Mfuzz algorithm was applied to perform time series clustering analysis on metabolite data at multiple time points. The results showed that the trajectories of metabolite changes could be roughly divided into four trends. Among them, the metabolite abundance in C1 continuously decreased as the treatment progressed, while C3 exhibited a trend completely opposite to that of C1. The metabolite abundance in C2 decreased and then rebounded after the T4 time point, and C4 showed a plateau phase at the T1-T2 nodes (Fig. 2 C and Figure S 2A-B). Notably, after performing principal component analysis on each cluster separately, the differences in metabolic profiles at each time point among the different clusters became more apparent (Figure S 2C). The Kyoto Encyclopedia of Genes and Genomes (KEGG) ( https://www.metaboanalyst.ca/ ) analysis results showed that C1, C2, and C3 have high similarity in amino acid metabolism and energy metabolism, particularly in pathways such as Arginine biosynthesis and Valine, leucine, and isoleucine biosynthesis, where they were significantly enriched. In contrast, C4 had unique enrichment in pathways such as Biosynthesis of unsaturated fatty acids and Arachidonic acid metabolism (Fig. 2 B). Dynamic change line charts were created for the metabolites in the significantly disturbed metabolic pathways of each cluster over the course of treatment. These primarily included Leucine, L-Isoleucine, Threonine, Oxoglutaric acid, and Histamine in C1; Leucine, L-Isoleucine, L-Valine, Phenylethylamine, Methylimidazoleacetic acid, and Indoleacetaldehyde in C2; Pyridoxal, 4-Pyridoxic acid, DOPA, Gentisic acid, Glycocholic acid, and Glycochenodeoxycholic acid in C3; and 18-HETE, Arachidonic acid, 8,9-DiHETrE, 11,12-EET, Dihomo-gamma-linolenic acid, Eicosapentanoic acid, adrenaline, and Thyroxine in C4 (Fig. 2 D). 3.4 Prediction model for the therapy response To look for potential predictive biomarkers for treatment response, the baseline (T0) plasma metabolome was examined in 108 GC patients receiving NAIC therapy. In this study, patients with Complete Response (CR) and Partial Response (PR) are defined as having a treatment response (Response, R), while patients with Stable Disease (SD) and Progressive Disease (PD) are defined as having no treatment response (Non-Response, NR). PCA showed a significant difference between the R (n = 65) and the NR (n = 43) (Fig. 3 A). Preliminary differential analysis identified 61 differential metabolites ( P < 0.05, Figure S 3A). KEGG analysis showed that these metabolites were significantly enriched in the following signaling pathways: Primary bile acid biosynthesis, Valine, leucine and isoleucine biosynthesis, Vitamin B6 metabolism (Figure S 3B). Following sample balancing with the SMOTE algorithm (Figure S 4A), LASSO regression was used on the training cohort to build a plasma metabolome-derived predictive model that effectively distinguished treatment responders (R) from non-responders (NR), resulting in 21 predictive metabolites (21-PM) (Figure S 4B-C and Fig. 3 B). There were significant differences in 21 metabolites between the R and the NR, with 16 of them (Thymidine, Thialysine ketimine, S − Adenosylhomocysteine, N6,N6 − dimethyllysine, Hydroxytryptophol, Hydroxyprogesterone, Hexanoylcarnitine, Dihomo − gamma − linolenic acid, Asp − tyr, 15,16 − DiHODE, 13,14 − Dihydro PGF − 1α, 8,9 − DiHETrE, 4 − Hydroxy − 2−oxoglutaric acid, 3 − hydroxynonanoyl carnitine, 2 − Aminoadenosine, 3-Hydroxy-cis-5-tetradecenoylcarnitine) significantly upregulated in R, while 5 of them (Corticosterone, coenzyme Q2, 12S − HHT, 4 − Pyridoxic acid, and 3 − hydroxydecanoyl carnitine) significantly downregulated in R (figure violin). The main contributors include 13,14 − Dihydro PGF − 1a, 15,16 − DiHODE, 4 − Hydroxy − 2−oxoglutaric acid, 4 − Pyridoxic acid, 3 − hydroxydecanoyl carnitine, and 12S − HHT (Fig. 3 B). The model's AUC value was 0.9935 (95%CI: 0.9834-1, Accuracy: 97.67%, Sensitivity: 97.67%, Specificity: 97.67% (Figure S 4D-E). The prediction model was subsequently applied to the testing set, demonstrating robust performance with an AUC of 0.897 (95% CI: 0.7521-1), accuracy of 90.91%, sensitivity of 92.31%, and specificity of 88.89% (Fig. 3 C-D). A thorough evaluation of the predictive model's robustness was carried out by investigating the impact of major clinical parameters (gender, age, BMI, ECOG performance status, differentiation status, Lauren classification, and clinical stage) on prognostic markers. Chi-square tests were used to conduct differential analysis comparing responders (R) and non-responders (NR) in order to discover clinical factors related with treatment response. The results showed that there were no significant differences between R and NR in terms of gender, age, BMI, ECOG, differentiation status, and Lauren classification ( P > 0.05). Only the clinical N stage showed a significant difference between the groups (Table 1 and Figure S 5A). The predictive capability of clinical N staging for treatment response was investigated through analysis using AUC values as the evaluation metric. Clinical N staging alone demonstrated limited predictive value, with an AUC of 0.559 (95% CI: 0.5039–0.6138), which was far lower than the predictive ability of the 21-PM metabolite characteristic model (Figure S 5B). To further evaluate the influence of clinical N0 status on predictive performance, all four N0 patients (whom are in the NR) were excluded from analysis. This exclusion resulted in an AUC of 0.884 (95% CI: 0.815–0.953), representing a modest reduction compared to the 21-PM model's performance (AUC = 0.994) (Figure S 5C). The potential impact of sample size reduction was evaluated through a bootstrap analysis with 1,000 iterations, where four patients were randomly excluded in each iteration. The analysis yielded a median AUC of 0.891 (range: 0.879–0.918) (Figure S 5D). When the 1,000-iteration AUC values were compared with both the N0-patient-excluded group and the original complete dataset, the iterative AUC values were found to be significantly lower than those from the original complete dataset (p < 0.001), yet significantly higher than those from the N0-patient-excluded group alone (p < 0.001) (Figure S 5D and table S). These results indicated that while clinical N staging demonstrated some predictive capability for treatment response (AUC = 0.559), its impact remained within acceptable limits overall. Furthermore, the predictive efficacy of clinical N staging was shown to be markedly inferior compared to the 21-metabolite panel (21-PM) identified in the study. For enhanced clinical application of the 21-PM model in precision therapy, a risk stratification management strategy was implemented. The risk stratification management strategy was constructed through the following steps: First, risk scores were calculated for each patient based on the model's metabolite coefficients, serving as an independent predictive indicator. Second, patients were stratified according to the median score. Finally, distribution differences across response groups were determined for each stratum. The AUC value reached 0.891 (95% CI: 0.826–0.957) (Figure S 5E), and patients were divided into different risk groups (high-risk group and low-risk group) based on the median risk score. The predictive performance of the risk scoring model was evaluated for each patient in the test set. Results demonstrated that showed that most NR patients belonged to the high-risk group, while most R patients belonged to the low-risk group, which was related to the predictive ability of the risk scoring (Fig. 3 E). The Response Evaluation Criteria in Solid Tumours (RECIST, version 1.1) outcomes were further compared between the two risk groups, and the results showed a higher proportion of PD and SD in the high-risk group, while PR was more prominent in the low-risk group (Fig. 3 F). This indicated that the predictive model and the benefit-risk indicators formed by the model successfully identified patients who needed optimized treatment plans. Distribution differences of clinically relevant prognostic factors were analyzed between high- and low-risk groups. The intestinal type (Lauren classification), known for better prognosis, showed significantly higher prevalence in the low-risk group (Figure S 6F), and lower T stages were more prevalent in the low group (Figure S 6G). Furthermore, the results of the multivariate logistic regression indicated that the risk score was an independent predictor (Table S 2). At the same time, the 21-PM model could reflect the relationship between metabolite characteristics and treatment response, and this strategy could accurately reflect the relationship between risk scores and treatment response. In summary, the risk stratification management strategy could categorize patients by calculating risk scores, providing accurate clinical treatment strategies for patients, thereby better serving the precise treatment of GC. Given the neoadjuvant immunotherapy-chemotherapy combination regimen used in this study, additional analyses were conducted to evaluate immune response-related outcomes. First, regarding the expression of PD-L1, there was no significant difference in the CPS score of PD-L1 expression between R and NR, but there was a trend of higher CPS expression in the R group, which was also indicated by the model risk score results (Figure S 5H). Subsequent analysis focused on clinical immune cell profiles. A total of 30 patients' immune test results were collected, mainly including CD3, CD4, CD8, CD19, and the CD3/CD4 ratio. The results showed no significant differences in immune cell expression between the R group and the NR group at baseline (Figure S 6A). To examine potential relationships between immune markers and metabolic levels, correlations were analyzed between selected key metabolites and immune cell populations (Fig. 3 H). 13,14-Dihydro PGF-1α and hydroxytryptophol showed significant positive correlations with CD3 expression, while hydroxyprogesterone and hydroxytryptophol were positively correlated with CD8 levels (Figure S 6B). Negative correlations were identified between thialysine ketimine/hydroxyprogesterone and the CD3/CD4 ratio, as well as between 2-aminoadenosine, 3-hydroxydecanoyl carnitine, hydroxyprogesterone and CD19 expression (Figure S 6B). Correlation analyses were performed to evaluate potential associations between key metabolites and various clinical parameters. The results revealed that higher expression levels of 13,14-dihydro PGF-1α significantly correlated with more favorable clinical stages (Figure S 6C). Analysis of Lauren classification demonstrated elevated expression patterns of thymidine, S-adenosylhomocysteine, and 15,16-DiHODE in intestinal-type cases. Distinct metabolic profiles were observed across treatment groups, with 13,14-dihydro PGF-1α, dihomo-γ-linolenic acid, and 3-hydroxy-cis-5-tetradecenoylcarnitine showing increased expression in the Chemo + PD1 cohort ((Figure S 6C)). Furthermore, 15,16-DiHODE exhibited higher expression in gastroesophageal junction (GEJ) cases, while 8,9-DiHETrE demonstrated elevated levels in microsatellite stable/mismatch repair proficient (MSS/pMMR) patients (Figure S 6C). 3.5 A monitoring model based on Imaging diagnosis In order to explore the changes in the plasma metabolome during the treatment process, the treatment status was regularly evaluated based on imaging during the treatment of patients, and the first effective treatment node (PR, CR) and the final ineffective treatment node (SD, PD) were determined (Fig. 4 A-C). The plasma metabolic characteristics of 85 GC patients receiving NAIC treatment regimens at the baseline period (T0) and key treatment stages (Tx) were studied in detail. A comparative analysis was first performed to examine plasma metabolomic changes across different treatment response groups (R group and NR group). It was found that the R group experienced more significant plasma metabolomic changes than the NR group during the treatment phase, from baseline to a key decision point (Figure S 7A-B). Further comparison of the fold change (FC) value changes in metabolic levels before and after treatment between the two groups of patients revealed that the overall metabolic changes in the R group were more significant (Figure S 7C). For example, compared to the NR, the R group showed a significant increase in the plasma levels of 13,14 − Dihydro PGF − 1a after treatment, while the trend of change for 5 − Hydroxy − L−tryptophan was opposite between the two groups. The change in platelet-activating factor was more pronounced in the NR group. The above results indicated that the patient group with effective treatment exhibited more pronounced responses at the metabolomic level. The baseline-predicted 21-PM model demonstrated a certain predictive ability in the treatment response prediction of 85 patients, with an AUC value of 0.8440 (95% CI: 0.7607–0.9274), Accuracy: 0.7765, Sensitivity: 0.7037, Specificity: 0.9032 (Figure S 7D-E). However, the potential of these 21 metabolites in distinguishing between the R and the NR based on their changes from baseline to the critical treatment phase still needed further exploration. The FC values of the 21 metabolites were calculated between pre- and post-treatment timepoints, followed by PCA. The results showed that the FC values of these metabolites had a certain ability to distinguish between the R group and the NR group (Figure S 7F). Given the significant metabolic shifts identified in the R group, an in-depth investigation was performed to detect treatment-associated metabolite changes unique to responders. Metabolites that changed exclusively in the R group were combined with those that showed opposite trends in the R and NR groups, resulting in a panel of 156 R-specific metabolite (Figure S 7G). KEGG pathway analysis ( https://www.metaboanalyst.ca/ ) of the 156 characteristic metabolites revealed significant disturbances in multiple metabolic pathways such as Valine, leucine, and isoleucine biosynthesis, Taurine and hypotaurine metabolism, Cysteine and methionine metabolism, Arginine biosynthesis, Primary bile acid biosynthesis, and Citrate cycle (TCA cycle) (Figure S 7H). Analysis of FC values for these 156 metabolites revealed distinct dynamic patterns between groups, prompting further refinement through LASSO regression to identify metabolites with optimal discriminative power (Figure S 8A-B). This analysis identified 11 characteristic metabolites whose FC patterns effectively differentiated metabolic profiles between R and NR groups (Fig. 4 D). The model's AUC value was 0.9679 (95% CI: 0.9318-1, Accuracy: 91.18%, Sensitivity: 88.24%, Specificity: 94.12%) (Figure S 8C-D). In the test set, the AUC value was 0.8636 (95% CI: 0.6830-1, Accuracy: 82.35%, Sensitivity: 72.73%, Specificity: 100.00% (Fig. 4 E-F). The PCA results of the 11 metabolites showed that their ability to distinguish between the R group and the NR group was better compared to the 21 metabolites in the predictive model (Figure S 8E). Further analysis of the expression differences of these 11 metabolites between the R group and the NR group revealed that the changes in 13,14 − Dihydro PGF − 1a, Gentisic acid, and Tetrahydropersin were significantly higher in the R group compared to the NR group, while the changes in Glycerol 5 − hydroxydecanoate, Indolelactic acid, 3 − Methylxanthine, and His − Trp were more significant in the NR group (Fig. 4 G and Figure S 8F). Moreover, comparing the expression differences of the aforementioned 11 metabolites between the R group and NR group at baseline and during the critical treatment phase showed that Glycerol 5 − hydroxydecanoate, 3 − Methylxanthine, and His − Trp, which had significant differences at baseline, became indistinguishable during the critical treatment phase (Figure S 9A). In contrast, 13,14 − Dihydro PGF − 1a, Gentisic acid, and Indolelactic acid, which were indistinguishable at baseline, showed differences during the critical treatment phase (Figure S 9A). To characterize the temporal dynamics of the 11 signature metabolites during treatment, expression levels were quantified across T0-T5 timepoints in 85 patients. The results of the Kruskal-Wallis test showed that 13,14 − Dihydro PGF − 1α and Gentisic acid changed significantly at 6 time points (Figure S 9B). Finally, baseline plasma immune information from 21 patients (11 in the R group and 10 in the NR group) and post-treatment plasma immune information from 19 patients (12 in the R group and 7 in the NR group) were also collected. The Spearman correlation analysis showed that in the baseline data, in group R patients, the expression of His − Trp was significantly negatively correlated with CD3 and CD8, while significantly positively correlated with NK cells. Glycerol 5 − hydroxydecanoate was significantly negatively correlated with CD3, and Glutamine was significantly negatively correlated with the CD19 ratio (Figure S 10A). In the NR group of patients, Prostaglandin F2b (PGF2b) and Indolelactic acid were significantly positively correlated with the number of NK cells, while Glycerol 5 − hydroxydecanoate was negatively correlated with the number of NK cells (Figure S 10A). In the NR group of patients, His − Trp was negatively correlated with CD4 and CD4/CD8, and 13,14 − Dihydro PGF − 1a was negatively correlated with CD19 (Figure S 10A). The above results indicated the complex differential relationships between metabolite levels and immune cells in the R group and NR group. Next was 13,14 − Dihydro PGF − 1a. There was no correlation in the R and NR groups at the base and in the R group at the time point, but there was a significant positive correlation between the NR group at the time point and CD19. This indicated that the correlation between 13,14 − Dihydro PGF − 1a and immune cells changed before and after treatment. Therefore, the relationship between the FC of metabolite variations and the clinical immune expression of PD-L1 was explored. The results showed that the difference in the pre- and post-variation of 13,14 − Dihydro PGF − 1a in the R group was more significant in patients with high PD-L1 expression (CPS ≥ 10) (Figure S 10B). This suggested that 13,14 − Dihydro PGF − 1a might play a role in the immune microenvironment of GC, especially in patients with high PD-L1 expression. The expression of PGF2b in GC patients and its correlation with immune cells underwent significant changes during the treatment process. At baseline, PGF2b was significantly positively correlated with the number of NK cells in the NR, while after treatment, PGF2b was significantly correlated with the number of CD4 cells in the R. Additionally, the expression of PGF2b in the R group decreased from baseline to post-treatment, whereas it increased in the NR group (Figure S 10C). Further analysis of the relationship between PD-L1 expression and the FC of PGF2b revealed that in the R group, although the differences in PD-L1 expression levels were not significant (p > 0.05), higher PD-L1 expression was associated with lower PGF2b changes; conversely, in the NR group, higher PD-L1 expression was associated with higher PGF2b changes (Figure S 10B). These results indicated that changes in immune expression during treatment caused variations in PFG2b levels, with the decrease in PGF2b potentially being associated with a better treatment response. 3.6 A monitoring model based on Pathologic diagnosis In the process of diagnosing and evaluating the efficacy of GC, histopathological diagnosis usually requires obtaining tumor tissue samples through endoscopic biopsy or surgical resection. This process is invasive and may pose certain trauma and complication risks to the patient. However, despite these potential adverse factors, the Becker score based on histopathological information remains one of the most authoritative methods for evaluating the efficacy of GC 20 . Therefore, a subset of 42 surgical patients with available pathological outcomes was selected from the cohort, comprising 21 responders and 21 non-responders as classified by Becker criteria (Fig. 5 A-B and Table S 3). Compared to the imaging diagnosis, among these 42 patients, 28 were classified as response and 9 as non-response (Figure S 11A). This indicated that there was a certain bias in monitoring the efficacy response of patients through imaging diagnosis, while monitoring the efficacy of patients using histopathological information was more accurate. Considering the invasive procedures and potential trauma and complications associated with obtaining tumor tissue samples, the collection of plasma samples was particularly important due to its low invasiveness and ability to reflect the real-time impact of drugs on the body's metabolism. Therefore, supported by histopathological diagnostic evidence and plasma metabolomics data, the impact of NAIC on tumor metabolic reprogramming was systematically investigated, leading to the development of a GC therapeutic monitoring model. Compared to the efficacy monitoring model (11-MMI) based on imaging evidence, this model could reduce information bias caused by the method of efficacy evaluation to a certain extent, resulting in higher accuracy. During the patient recruitment process from October 2021 to June 2023, a total of 42 patients underwent surgical treatment after NAIC, preoperative tissue samples were collected, and the Becker score was used to assess the tumor progression status, serving as evidence to support clinical efficacy outcomes. Based on the clinical efficacy information supported by the aforementioned histopathological evidence, further studies were conducted on the baseline and preoperative plasma samples of these patients to explore the dynamic changes of characteristic metabolites in the treatment-effective group (R group) and the treatment-ineffective group (NR group) as determined by pathological results. PCA was initially performed on the 11-imaging efficacy-associated metabolites, demonstrating partial discriminative capacity between between the R group and the NR group to some extent (Figure S 11B). However, Evaluation of the 11-metabolite model's predictive performance in the histopathology-validated cohort yielded an AUC of 0.569 (95% CI: 0.390–0.748), indicating limited discriminative capacity (Figure S 11C), indicating its relatively low predictive ability. This indicated that biomarkers identified through imaging assessment had certain limitations in evaluating clinical efficacy. Therefore, it was very important to characterize the differences in plasma metabolite reprogramming between the R group and the NR group of patients based on histopathological evidence, mine the characteristic metabolites of the R group, and construct a clinical efficacy monitoring model. A comparative analysis of plasma metabolomic profiles was initially performed between different treatment responses (R and NR). The results showed that from baseline to preoperative, the metabolic profile of patients in the R group underwent more significant changes throughout the entire treatment process (Figure S 11D-E). The identical analytical methodology employed for the imaging cohort was applied to identify pathology-specific characteristic metabolites, resulting in the detection of 80 R-specific metabolites (Figure S 11G). The KEGG results for the 80 characteristic metabolites showed that metabolic pathways such as Tryptophan metabolism, Butanoate metabolism, and Alanine, aspartate, and glutamate metabolism are disturbed (Figure S 11H). For additional validation of the characteristic metabolites' predictive capability, patients were randomly allocated into training and validation sets at an 8:2 ratio. In the training set, LASSO regression was used to screen for characteristic metabolites. The results showed that 13 metabolites were selected (Fig. 5 D and Figure S 12A-B), with a model AUC value of 0.9653 (95% CI: 0.91585-1), Accuracy: 91.18%, Sensitivity: 88.24%, Specificity: 94.12% (Figure S 12C-D). The model performed welled on the test set, AUC: 0.9375 (95%CI: 0.76426-1), Accuracy: 87.50%, Sensitivity: 100.00%, Specificity: 75.00% (Fig. 5 E-F). These results indicated that the 13 metabolites selected through LASSO regression had high predictive accuracy in distinguishing between the R group and the NR group. The PCA results of the 13 metabolites also showed that their ability to distinguish between the R group and the NR group was superior to that of the 11 metabolites selected by the imaging group (Figure S 12E). To investigate immune-metabolite correlations during NAIC therapy, baseline plasma immune profiles were analyzed in 18 GC patients (9 R and 9 N]), with paired post-treatment data obtained from 14 patients (7 R and 7 NR). Notably, significant positive correlations were identified in treatment responders between key metabolites - eicosapentanoic acid (EPA), mhppa sulphate, and 13,14-dihydro-PGF-1α - and post-treatment immune cell populations (CD3+, CD4+, CD19+), suggesting their potential immunomodulatory role during NAIC therapy (Figure S 13A). The value of FC of EPA was substantially lower in the R group for patients with high PD-L1 expression (CPS ≥ 10) than for those with low PD-L1 expression (CPS < 10) (Figure S 13B). Notably, the R group's overall EPA levels grew significantly from the baseline to the preoperative period (Figure S 13B), indicating a possible connection between PD-L1 status and EPA dynamics. Figure S 13A showed that, in the NR group, at baseline, CD4 showed a significant negative correlation with Mhppa sulphate and CD3 showed a significant negative correlation with Tridecanoylglycine, while CD19 showed a significant positive correlation with 4 − Hydroxy − 2−oxoglutarate. After treatment, adrenaline was negatively correlated with CD3 and positively correlated with NK cells; 13,14 − Dihydro PGF − 1a was positively correlated with CD19; the correlation of 4 − Hydroxy − 2−oxoglutarate with immune cells changed significantly, being positively correlated with CD3 and CD8, while negatively correlated with the CD4/CD8 ratio and NK cell count; Isoferulate 3 − glucuronide was negatively correlated with NK cell count; Palmitoylcarnitine was negatively correlated with CD4. Discussion GC is a highly lethal malignant tumor globally, and the rise of immunotherapy had opened new pathways for its treatment. Among these, immune checkpoint inhibitors (ICIs) have become the first-line treatment for advanced gastric or esophageal adenocarcinoma 21 , 22 . Neoadjuvant therapy, or preoperative treatment, offers GC patients a new treatment option aimed at shrinking tumors, increasing resection rates, reducing the risk of intraoperative spread, and potentially improving patient survival and quality of life 21 , 23 . In neoadjuvant therapy, the combination strategy of immunotherapy and chemotherapy had garnered significant attention. This strategy leverages the dual effects of chemotherapy in releasing tumor antigens and immune checkpoint inhibitors in relieving immune suppression, thereby enhancing the immune system's ability to attack tumors 10 . Clinical trials have shown that this combination therapy can improve patients' pathological complete response rates and survival rates 8 , 24 – 26 . In this study, approximately 60.19% of patients responded (PR + CR) to NAIC, with a pCR rate of 28.57%. The toxic reactions were manageable, with most treatment-related adverse events (TRAE) being grade 1 or 2. The most common TRAEs were nausea, anemia, fatigue, anorexia, and decreased white blood cell count, which are consistent with previous research findings 8 , 24 – 29 . Furthermore, the toxicity associated with immunotherapy was often due to a decrease in the number of naive T cells and the overactivation of memory T cells. These activated memory T cells might have invaded peripheral organs, causing inflammatory damage, such as in the gastrointestinal tract 30 . According to the literature, in patients receiving PD-1/PD-L1 inhibitors combined with chemotherapy, the incidence of diarrhea of all grades ranged from 17–40%, while the incidence of grade 3–4 diarrhea was approximately 2% 31 . In our study, approximately 42.37% of patients experienced varying degrees of diarrhea after receiving NAIC treatment, with the proportion of grade 3 or higher diarrhea being 3.39%, a figure slightly higher than previously reported data. Most clinical trials used single drugs or single treatment regimens, whereas the patients in our cohort received a variety of chemotherapy regimens, including SOX, FLOT, XELX, and some patients also received taxane drugs. Reports indicated that the incidence of diarrhea caused by taxane drugs was higher than that caused by platinum-based drugs. At the same time, the types of PD-1/L1 inhibitors also varied, such as Trelagliptin, Tislelizumab, and Sintilimab. These factors might have been the reasons for the heterogeneity of adverse reactions to treatment. Moreover, by comparing the age and preclinical TNM staging of populations in various clinical trials, it was found that the population in this trial is relatively older and had higher TNM staging than those in other trials. This indicated that the physical condition of the subjects in this trial was slightly worse than that of the subjects in other trials, which might also have been one of the reasons for the slightly higher incidence of adverse reactions in this trial. The tumor microenvironment (TME) was considered a key determinant of tumor heterogeneity and plays an important role in regulating the tumor's response to various therapeutic interventions 32 – 34 . Metabolic reprogramming had become a hallmark of cancer, closely related to the tumor microenvironment. The dynamic crosstalk between immune cells, stromal cells, and tumor cells had a decisive impact on the growth and survival of tumor cells 35 , 36 . This study employed longitudinal cohort monitoring to systematically characterize NAIC-induced metabolic reprogramming dynamics throughout the treatment course, following initial efficacy assessment. Using the Mfuzz algorithm for clustering analysis of time series data, four distinct patterns of metabolite changes was successfully identified. Notably, significant perturbations in amino acid metabolism pathways were identified across clusters 1–3, with pronounced alterations observed in the pathway, such as leucine, valine, oxoglutarate, and 4-hydroxy-2-oxoglutarate, as well as biogenic amines related to inflammation and immune regulation, such as histamine, also showed significant changes in these clusters. In tumor cells, amino acids, as important proteins and signaling molecules, can influence tumor immunity by regulating the production of immune cells and immune factors 37 – 39 . Arginine, which is an important component of protein synthesis and also a precursor to polyamines, creatine, and nitric oxide, plays a crucial regulatory role in tumor angiogenesis and is essential for the growth and proliferation of tumor and immune cells 40 , 41 . Furthermore, arginine plays an important role in the TME, where there is competition for arginine among cancer cells, immunosuppressive cells, and anti-tumor cells, and when tumor cells consume a large amount of arginine in the tumor microenvironment, anti-tumor cells are inhibited, thereby promoting immune evasion 42 – 44 . Tryptophan (Trp) metabolism involves the kynurenine (Kyn) pathway, the serotonin (5-hydroxytryptamine, 5-HT) pathway, and the indole pathway, and its metabolic imbalance is associated with various cancers, primarily by promoting tumor growth and immune evasion through the generation of an TME 45 . Similar to arginine, the metabolism of tryptophan plays a crucial role in the immune regulation of tumor cells, and cancer cells and macrophages inhibit antigen-specific T cell responses by competitively consuming tryptophan in the TME 39 . Histidine catabolism also significantly affects the sensitivity of cancer cells to methotrexate by reducing the cellular pool of tetrahydrofolate 46 . The reprogramming of branched-chain amino acids (BCAAs) metabolism alters the levels of important metabolites including BCAAs, α-ketoglutarate (α-KG), glutamate, and reactive oxygen species (ROS), which play roles in protein synthesis and degradation, energy supply, signal transduction, and other processes, affecting the survival and growth of cancer cells 47 – 51 . Tyrosine metabolism is an important target in cancer treatment, particularly in enhancing the effects of chemotherapy and targeted therapy 52 , 53 . Relatively speaking, the metabolite changes in C4 were mainly concentrated in the biosynthesis of unsaturated fatty acids, tryptophan metabolism, and arachidonic acid metabolism pathways. Unlike other clusters, the abundance of these metabolites first increased during treatment, then entered a plateau phase during the T1-T3 stages, and finally decreased. This unique pattern of change might have been related to treatment resistance. Multiple studies have shown that the reprogramming of lipid metabolism reshapes the tumor immune environment, which is crucial for the survival, metastasis, and treatment resistance of cancer cells 54 – 56 . Lee and others found that the polyunsaturated fatty acid (PUFA) biosynthesis pathway plays a crucial role in ferroptosis 57 . The biosynthesis of unsaturated fatty acids mainly occurs through the action of various metabolic enzymes, such as desaturases and elongases, resulting in the production of important lipid molecules like oleic acid (OA), arachidonic acid (AA), docosahexaenoic acid (DHA), and EPA 58 , 59 . AA was an important polyunsaturated fatty acid and a key component of cell membrane phospholipids, and it was metabolized through the cyclooxygenase (COX) and lipoxygenase (LOX) pathways, significantly influencing tumor growth, angiogenesis, cell migration, and immune cell function 60 . Cui et al. enhanced the immune response against colorectal cancer by targeting the arachidonic acid metabolic pathway, activating CD8 + T cells, and inhibiting vasculogenic mimicry (VM) 61 . Furthermore, in non-small cell lung cancer, Chen et al. discovered that CYP4F2-mediated metabolism of arachidonic acid can promote stroma cell-mediated immune suppression in non-small cell lung cancer, and proposed a strategy to enhance the efficacy of immunotherapy by inhibiting CYP4F2 62 . Fan and others also found that inhibiting arachidonic acid metabolism can promote tumor growth and gemcitabine resistance in pancreatic cancer 63 . Despite the continuous advancements in the treatment of GC, the heterogeneity of tumor treatment remains a significant challenge. In this study, approximately 60% of patients exhibited positive treatment outcomes. Predicting potential treatment responses before starting therapy and subsequently customizing personalized treatment strategies was crucial for the prognosis of GC patients. Multiple studies have shown that early diagnosis, prognosis prediction, and treatment response monitoring of cancer patients based on various omics data have shown good results 17 , 19 , 64 – 68 . Therefore, this study developed a metabolite prediction model (21-PM) based on baseline plasma samples using machine learning. The results of the prediction model showed that the model's predictive performance was good and less influenced by clinical factors, highlighting the great potential of plasma metabolomics analysis in guiding patients to receive specific therapies (NAIC). The risk score was developed based on the 21-PM model to facilitate stratified management implementation. The risk score results showed that the higher the patient's score, the more likely they were to be sensitive to treatment. The introduction of the risk score not only enhanced our understanding of patients' treatment responses but also helps doctors assessed patients' prognosis before treatment and thereby formulate more precise treatment plans. Patients were stratified into high-risk and low-risk groups according to their risk scores, enabling comparative analysis of clinical characteristics and treatment outcomes across risk categories. Patients in the high-risk group might have required more aggressive treatment strategies, including more intense chemotherapy regimens, targeted therapy, or immunotherapy, while patients in the low-risk group might have benefited from more moderate treatments, reducing the risk of overtreatment. In summary, risk scoring, as a new prognostic tool, not only improves the accuracy of predicting the response of GC patients to NAIC treatment but also provides valuable information for clinical decision-making. Future research should further validate the universality and practicality of this model and explore its applicability in different patient populations. Among these 21 key metabolites, significant differential expression was observed in the responder (R) group, with 16 metabolites demonstrating marked upregulation and 5 showing pronounced downregulation. These changes in metabolites not only reflected alterations in tumor biological behavior but may also indicated the tumor's sensitivity to NAIC treatment. For example, among the upregulated metabolites, Thymidine, Thialysine ketimine, and S-Adenosylhomocysteine are involved in DNA synthesis and repair, amino acid metabolism, and methylation processes, which play crucial roles in tumor growth and response to chemotherapy drugs. Research on thymidine in the prognosis of GC mainly focuses on the expression levels of its metabolic enzymes. Multiple studies have demonstrated that the expression levels of thymidine phosphorylase and dihydropyrimidine dehydrogenase (DPD) are closely related to the prognosis of GC patients 69 . The upregulation of S-Adenosylhomocysteine may affect DNA methylation, thereby influencing gene expression and the tumor microenvironment. Studies have shown that in hepatocellular carcinoma (HCC), higher serum SAH levels are independently associated with poor prognosis in HCC patients 70 , 71 . The mechanism of action of the SOX, FLOT, and XELX treatment regimens used in this study primarily involves interfering with key processes such as DNA synthesis and cell division in tumor cells, which induces the accumulation of SAH 72 , 73 . Among the downregulated metabolites, corticosterone and coenzyme Q2 are related to stress response and energy metabolism, and their downregulation may reflect the metabolic inhibition of tumor cells under NAIC treatment. The downregulation of corticosterone may be related to the weakened stress response capability of tumor cells. Corticosterone had immunosuppressive effects and can inhibit the proliferation and function of various immune cells, such as T cells, B cells, and natural killer (NK) cells 74 . There were studies reporting that elevated levels of corticosterone might have been associated with poor prognosis in cancer patients receiving ICI treatment 75 . Hydroxytryptophol was a plasma metabolite of tryptophan, and studies have shown that it was expressed at higher levels in inflammatory diseases 76 . The study identified elevated hydroxytryptophol expression in R, potentially indicative of inflammatory activity within the tumor microenvironment 77 . The inflammatory response played an important role in the occurrence and development of tumors, and inflammatory cells and factors in the tumor microenvironment could promote the proliferation, invasion, and metastasis of tumor cells. Our further research found that there was a correlation between clinical T staging and the expression of Hydroxytryptophol, showing a trend where higher T staging corresponded to higher levels of Hydroxytryptophol expression. Clinical T staging, as an empirical indicator, had always been used by clinicians to assess the extent of tumor progression and predict patient prognosis. Therefore, by monitoring the levels of Hydroxytryptophol, it was expected to provide auxiliary evidence for assessing tumor invasiveness and determining the tumor's response to treatment, thereby aiding in the formulation of clinical treatment decisions. CD8 + T cells are the primary tumor-killing cells, and it has been specifically demonstrated that the density of pre-existing CD8 T cells at the edge of aggressive tumors (metastatic melanoma) was associated with the response to PD-1 inhibitor treatment 78 . The study identified a significant positive correlation between hydroxytryptophol levels and CD8 + T cell infiltration, suggesting its potential immunomodulatory role in antitumor responses. 13,14-Dihydro PGF-1α and 15,16-DiHODE emerged as high-contribution metabolites in the predictive model, with 13,14-dihydro-PGF-1α and 12S-HHT being characterized as cyclooxygenase-derived metabolites of arachidonic acid, implicating prostaglandin pathways in treatment responses. The upregulation of 13,14-Dihydro PGF-1α in the R group and the downregulation of 12S-HHT in the R group,15,16-DiHODE was a metabolite produced by the metabolism of arachidonic acid through the lipoxygenase pathway, which was upregulated in the R group, while 8,9-DiHETrE was a metabolite produced by the metabolism of AA through the cytochrome P450 pathway, which was also upregulated in the R group. Furthermore, the results of exploring the expression of immune cells at the early stage of treatment showed that 13,14 − Dihydro PGF − 1a was significantly positively correlated with CD3.Moreover, ECOG and clinical staging are often used in clinical practice for empirical judgment. Relatively speaking, patients with lower ECOG and clinical staging have better conditions and better treatment tolerance. Additionally, in our study, 13,14 − Dihydro PGF − 1a at baseline was negatively correlated with the potential prognostic characteristics of ECOG and clinical stage. In summary, the arachidonic acid metabolic pathway had significant biological implications and clinical application prospects in the baseline prediction of NAIC treatment. The prediction model (21-PM) established based on baseline data performed well in predicting treatment response. However, due to issues such as potential drug resistance during tumor treatment, relying solely on baseline predictions cannot monitor changes in real-time during the treatment process 79 . The study focused on characteristic R-spectific metabolites at key treatment nodes to characterize metabolic changes during therapeutic progression and investigate post-treatment biological alterations associated with treatment response. A more pronounced metabolic reprogramming was observed in R group compared to NR group when comparing pre- versus post-treatment profiles. 156 metabolites specifically altered in the R group highlighted significant disturbances in pathways such as the biosynthesis of valine, leucine, and isoleucine, the metabolism of taurine and hypotaurine, the metabolism of cysteine and methionine, the biosynthesis of arginine, the biosynthesis of primary bile acids, and the citric acid cycle (TCA cycle). Through LASSO regression analysis, the 11 key metabolites further screened based on 156 characteristic metabolites of the R group showed significant potential in distinguishing between the R group and the NR group. Among them, the metabolite 13,14-Dihydro PGF-1α showed predictive potential for treatment response at baseline and exhibited obvious changes at critical points during the treatment process. Metabolic expression data at multiple time points revealed that 13,14-Dihydro PGF-1α in the R group first increased and then decreased with the progression of treatment, and this trend of change suggests that it may be closely related to treatment sensitivity. In the R group, the initial increase of 13,14-Dihydro PGF-1α during treatment may be positively correlated with treatment efficacy, while the subsequent decrease in expression may imply the development of treatment resistance. A significant association was identified between 13,14-dihydro-PGF-1α dynamics and PD-L1 expression levels during treatment. PD-L1 was an important immune checkpoint molecule on the surface of tumor cells, and the level of its expression can affect the interaction between tumor cells and immune cells, thereby influencing tumor immune evasion and the efficacy of immunotherapy 80 . 3,14-Dihydro PGF-1α may be involved in shaping and regulating the tumor immune microenvironment by modulating the expression or function of PD-L1. When further analyzing the data of 13,14-Dihydro PGF-1α and immune cell expression, a significant correlation was found between this metabolite and CD19 in patients of the NR group after treatment. In addition, PGF2b was also a metabolite of particular interest in our study. PGF2b was produced through the COX pathway in arachidonic acid metabolism and participates in regulating the migration and activation status of immune cells, helping tumors to evade immune surveillance 81 . In our study, the decreased expression of PGF2b before and after treatment in the R group may be associated with increased tumor sensitivity to treatment, which indicates that the inhibition of PGF2b helps to enhance the treatment effect. In the tumor microenvironment, the high consumption of Glutamine by tumor cells leads to a decrease in the glutamine level in the microenvironment, thereby inhibiting the function of immune cells (such as T cells, natural killer cells, and macrophages) and promoting tumor immune evasion 82 . This was consistent with our results, where the expression of Glutamine increased before and after treatment, inhibiting tumor cell escape and making patients more likely to respond to treatment. The study further evaluated efficacy monitoring for patient cohorts with histopathological outcomes, delving into the accuracy and potential biases of different assessment methods (imaging and histopathology) to provide more precise treatment response predictions for clinical practice. Our analysis showed that compared to the imaging-based 11-MMI, the 13-MMP was more accurate in distinguishing between treatment responders and non-responders (with superior model performance). EPA 13,14-Dihydro PGF-1α, and other prostaglandin-like compounds still had significant predictive ability in analyses with tissue pathology as the outcome. And in the analysis with immune cells showed that in the R group of patients, EPA showed a significant positive correlation with the expression level of CD3 cells after treatment, and the FC value of EPA in patients with high PD-L1 expression (CPS ≥ 10) was significantly lower than that in patients with low PD-L1 expression (CPS < 10). This suggested that EPA might be negatively correlated with PD-L1 expression levels, implying that EPA might have played an inhibitory role in the TME with high PD-L1 expression, affecting the expression state of CD3 cells, and thereby potentially influencing the patient's treatment response. Furthermore, in the R group of patients, the expression level of EPA significantly increased before and after treatment. Due to the competition between EPA and AA for positions in the cell membrane, the presence of EPA reduced the amount of AA, inhibited the production of PGF2b, and weakened the tumor's ability to evade immune surveillance. This was also consistent with the results from our imaging group. Conclusion Overall, the study comprehensively revealed the reconstructive effect of NAIC on the metabolic profile of GC patients, clarifying four different dynamic change trajectories across treatment nodes, and provided valuable insights into the dynamic metabolic response of GC patients to NAIC. On this basis, a series of models with significant clinical value were constructed: The 21-PM model, which predicted treatment response based on baseline data, included 21 metabolites and their calculated risk scores, holding critical clinical significance for stratifying GC patients before treatment. The 11-MMI treatment response monitoring model, based on imaging, aided in real-time monitoring of efficacy and timely intervention during NAIC treatment. The 13-MMP treatment response monitoring model, based on tissue pathology, revealed biomarkers related to patient efficacy from pre-NAIC to pre-surgery from a more rigorous clinical evaluation perspective. The potential biomarkers for GC NAIC treatment discovered in this study could have been used for early prediction of treatment response and real-time monitoring of treatment efficacy, strongly promoting the development of personalized treatment and leading us towards an era of precision medicine driven by metabolomics. Methods Study population The study was approved by the Ethics Committee of The First Affiliated Hospital, Zhejiang University School of Medicine (IIT20230426B-R1), and all subjects provided written informed consent. This study recruited GC patients who received NAIC treatment at The First Affiliated Hospital, Zhejiang University School of Medicine from October 30, 2021, to June 30, 2023. All these patients had completed at least one cycle of NAIC before surgery. Inclusion criteria as following: histologically confirmed gastric adenocarcinoma by endoscopic biopsy before the treatment initiation; (2) locally advanced GC, regionally unresectable or clinical stage IV disease according to AJCC/UICC 8th staging system, which was mainly evaluated by computed tomography (CT); (3) patients receiving chemotherapy combined with immune checkpoint inhibitors with or without trastuzumab. The exclusion criteria were as follows: (1) history of another malignancy, except cured basal cell carcinoma of the skin or cured carcinoma in situ of the uterine cervix; (2) prior major stomach surgery; (3) previous systemic treatment or radiotherapy for GC. During the recruitment period of this study, a total of 447 plasma samples were collected from 154 patients diagnosed with GC. After applying the inclusion and exclusion criteria, 369 plasma samples from 108 patients were selected, including 108 preoperative baseline samples and 261 samples collected throughout the treatment process. In addition, clinical information including gender, age, BMI, Lauren classification, tumor location, differentiation status, MSI/MMR status, CPS score, clinical TNM stage, therapy, and treatment response was also collected. Plasma sample collection Patients had peripheral venous blood samples (baseline samples) collected before the start of treatment and after each treatment until the patients underwent surgery (including preoperative samples). All sample collection activities were completed by June 30, 2023. After fasting for 12 hours, blood was collected from the patients while they were in a fasting state. Plasma separation was performed according to the following procedure: The collected blood was centrifuged at 1200 rpm for 10 minutes at 4°C. Subsequently, the supernatant was collected and stored at -80°C until metabolite extraction was carried out. Metabolite extraction The procedure for metabolite extraction from plasma samples was as follows: First, plasma samples were thawed on ice. Then, 100 µL of plasma was mixed with 400 µL of pre-chilled methanol. The mixture was vortexed for 3 minutes and then centrifuged at 4°C and 13,000 rpm for 15 minutes. The supernatant was collected and evaporated using a centrifugal vacuum evaporator at 4°C. The dried metabolite samples were stored at -80°C until they were subjected to liquid chromatography-tandem mass spectrometry (LC-MS/MS) analysis. Quality Control (QC) samples, which were prepared by taking 6 µL from each plasma sample, were treated in the same manner as the study plasma samples. Plasma metabolites detection Using UHPLC-Q-Orbitrap-MS/MS technology, a comprehensive metabolomics analysis was carried out on 369 serum samples from 108 gastric cancer patients. First, 300 µL of the supernatant was taken, and 50 µL of the working solution containing internal standards (sulfamethoxypyridazine and ketoprofen) was added. Then, the mixture was vortexed for 30 seconds, followed by ultrasonic treatment on ice. After that, it was centrifuged at 4°C and 13,000 rpm for 15 minutes. Finally, 30 µL of the supernatant was taken for liquid chromatography-tandem mass spectrometry (LC-MS/MS) analysis. Mass spectrometry conditions: The liquid chromatography detection method mainly referred to the method developed in the literature. The chromatographic column used was ACQUITY UPLC HSS T3 (2.1 50 mm 1.7µm, Waters Corporation). The column temperature was maintained at 40°C. The mobile phase ratio consisted of A (0.1% formic acid in water) and B (acetonitrile). The flow rate was set at 0.35 mL/min, and the injection volume was 5 µL. The elution gradient is shown in Table 2.4 below: Table 2.4 Liquid Chromatography Gradient Elution Program Table gradient flow of LC solvent Time Solvent A(%) Solvent B(%) 0 95 5 3 80 20 5 60 40 9 40 60 16 35 65 18 20 80 21 5 95 23 95 5 25 95 5 Orbitrap Exploris 120 mass spectrometer (Thermo Fisher Scientific Inc.), with the following parameter settings: spray voltage, -2500V;Sheath gas(Arb), 50༛Aux gas༈Arb༉, 10༛Sweep gas༈Arb༉, 1༛Ion transfer tube temperature, 300℃༛Vaporizer temperature, 350℃༛orbitrap resolution, 60000, scan range, 200–1500༛Dynamic exclusion on༛ Metabolomics data processing To eliminate the noise and instrument fluctuation interference during the operation of the instrument, a QC sample was added during the plasma sample LC-MS/MS detection process, that is, a QC sample was inserted after an equal number of samples. Ultimately, the expression and coefficient of variation of metabolites in the QC samples were used as the basis to eliminate some of the interference caused by systematic noise. Afterwards, the Compound Discover version 3.2 (Thermo Fisher Scientific Inc.) software was used to analyze the data from the aforementioned plasma samples, and the metabolites were structurally identified using the mzCloud ( www.mzcloud.org/ ), mzVault ( https://mytracefinder.com/tag/mzvault/ ), KEGG ( www.kegg.jp/ ), HMDB ( https://hmdb.ca/ ), and ChemSpider ( www.chemspider.com ) databases. Using the MetNormalizer R package (version 1.3.02) with the support vector regression algorithm, potential loading/dispensing errors and batch effects were adjusted for 18 . After the features with RSD ≥ 30% in the QC samples had been excluded, 71% (5787/8167) of the positive ion mode features and 88% (3512/3980) of the negative ion mode features were retained. After annotation, 1001 metabolites remained. The raw data were log-transformed with base 2, and Z-scale normalization of metabolite features was performed before data analysis. Mfuzz clustering analysis Collected metabolic data from 369 plasma samples of 108 participants, standardized the average values of metabolites at six different time points as input data, and applied the Mfuzz fuzzy c-means algorithm (Mfuzz, R version 2.62.0) to cluster the metabolites and depict their dynamic changes. Pediction and monitoring model for the therapy response According to the outcome indicators, the cohort was randomly divided into a training set and a validation set in an 8:2 ratio, ensuring that the distribution of key clinical features in the training set and the validation set was similar. Before model construction, the SMOTE (DMwR, R, version 0.4.1) algorithm was used to address the data imbalance issue. Subsequently, a Least Absolute Shrinkage and Selection Operator (LASSO) regression analysis (glmnet, R, version 4.1-8) was used to construct the model. Based on 10-fold cross-validation to avoid overfitting, the parameters of the LASSO regression were finally determined, non-zero coefficient metabolites were selected, and the model was constructed. Subsequently, the metabolite feature information of plasma samples from the test set patients was applied to validate the established prediction model. The entire LASSO regression study was applied in three areas: 1. Efficacy outcome prediction model based on baseline samples; 2. Patient efficacy monitoring model based on imaging diagnosis during the course of treatment; 3. Patient efficacy outcome monitoring model based on pathological diagnosis information. Using ROC analysis (pROC package, R version 1.18.5), the predictive value was comprehensively assessed based on sensitivity, specificity, accuracy, and Area Under Curve (AUC). 1. Efficacy outcome prediction model based on baseline samples: First, based on the type of treatment response, 108 subjects were randomly divided into a training dataset (n = 86, 79.63%) and a test dataset (n = 22, 20.37%). The distribution of key clinical features in the training and validation sets was similar. After the SMOTE algorithm addressed the data imbalance issue, the differential metabolites (DMs) between R the NR were used as input data. The parameters for the LASSO regression were determined through 10-fold cross-validation, selecting the 1se model with a lambda value of 0.03574295 to construct the model. Subsequently, the metabolite characteristic information of plasma samples from patients in the test set was applied to validate the established prediction model. The clinical characteristics significantly related to patient treatment outcomes were explored through intergroup difference analysis, and the clinical N stage was identified as this differential feature. Following that, the predictive abilities of this factor and its impact on the model were thoroughly analyzed, with the AUC value serving as the primary metric for measuring predictive performance. The examination of the model's effect ability proceeded as follows: First, 4 N0 patients were excluded from the training set to evaluate the model's prediction accuracy. Following that, 1000 iterative experiments were carried out, each including the random removal of four patients and simultaneous performance assessment, to explore the influence of reduced sample size on predictive capability. In order to assess the benefit-risk ratio for patients after treatment, risk scores 19 . were calculated based on the metabolite coefficients in the model, and patients were classified into High-Risk and Low-Risk groups based on the median of the risk scores. To determine whether the Risk Score could independently affect the outcome, clinical factors with a baseline intergroup difference P -value < 0.1 were also included in the multivariate logistic regression model to comprehensively assess the relationship between various influencing factors and the outcome. 2. Patient efficacy monitoring model based on imaging diagnosis during the treatment process: First, paired difference analysis was conducted at baseline (T0) and key stages (Tx) for both the R and the NR, screening for metabolites that showed significant changes only in the R ( P < 0.05) and metabolites that exhibited opposite trends in the R compared to the NR. This allowed the construction of a characteristic metabolite set for the R. Subsequently, based on the type of treatment response, 85 subjects were randomly divided into a training dataset (n = 68, 80.00%) and a test dataset (n = 17, 20.00%). The distribution of key clinical features in the training and validation sets was similar. After the SMOTE algorithm addressed the data imbalance issue, the characteristic metabolites of R were used as input data. The parameters for LASSO regression were determined through 10-fold cross-validation, selecting the min model with a lambda value of 0.08584942 to construct the model. Then, the metabolite feature information from the plasma samples of the test set patients was used to validate the established monitoring model. 3. Patient efficacy outcome monitoring model based on pathological diagnosis information: Constructed a characteristic metabolite set for R with tissue pathology as the outcome using the same method. Subsequently, based on the type of treatment response, 42 subjects were randomly divided into a training dataset (n = 34, 80.95%) and a test dataset (n = 8, 19.05%), ensuring a similar distribution of clinical characteristics. After that, using the characteristic metabolites of R as input data, the parameters of the LASSO regression were determined through 10-fold cross-validation, selecting the min model with a lambda value of 0.07801261. Metabolites with non-zero coefficients were screened to establish the monitoring model. Then, the metabolite characteristic information from the plasma samples of the test set patients was applied to validate the established monitoring model. Statistical analyses Standard statistical methods were used to conduct statistical analysis on the clinical characteristic information and metabolite characteristics collected in the study. The normality of data distribution was assessed based on the Shapiro-Wilk test. Due to the non-normal distribution of most metabolite characteristics, statistical analysis of metabolite characteristics was conducted using non-parametric tests: the Wilcoxon rank-sum test was used for pairwise difference analysis, while the Kruskal-Wallis (K-W) test was used for multi-group difference analysis. For immune indicator information (CD3, CD4, CD8, etc.), parametric tests such as the T-test were used; if the data did not follow a normal distribution, non-parametric analysis such as the Wilcoxon rank-sum test was employed. For clinical categorical variable data (age, gender, ECOG performance status score, Lauren classification, etc.), the chi-square test or Fisher's exact test was used. The correlation coefficient was analyzed using the Spearman method. For all analyses, a two-sided P value of less than 0.05 was considered statistically significant. All data analyses were performed using R statistical software (version 4.3.3). Declarations Acknowledgments This work was supported by the Zhejiang Province Traditional Chinese Medicine Science and Technology Project (No.GZY-ZJ-KJ-23008), Key R&D Program of Zhejiang Province (No.2025C02171), Zhejiang Province Traditional Chinese Medicine Key Laboratory Project (No.GZY-ZJ-SY-2303), National Natural Science Foundation of China (No. 81972671) and Joint TCM Science &Technology Projects of National Demonstration Zones for Comprehensive TCM Reform (GZY-KJS-ZJ-2025-036). The authors acknowledge the support from the Shared Instrumentation Core Facility at the Hangzhou Institute of Medicine (HIM), Chinese Academy of Sciences. Author contributions Z.X.Y., H.X.L., Z.K.K., Z.W. and L.X.S. conceived the idea and designed the study; Z.X.Y., H.X.L and G.S.H. performed sample preparation and data acquisition; Z.X.Y., H.X.L. contributed to metabolite identification; Z.X.Y., H.X.L. performed data analysis with help of L.Y., W.X.D, D.Y, F.Y., G.T.Z., M.J.H. and L.C.Z.; Z.K.K, Z.H.D., J.H.L. and L.X.S, contributed to the clinical trial of this study, including patient recruitment, diagnosis, follow-up, and serum sample collection; Z.X.Y., H.X.L., and Z.K.K., wrote the paper; all authors read and approved the manuscript; L.X.S. led the clinical trial; L.X.S., T.J.K. and Z.W. supervised the project. References Liu, D. et al. The patterns and timing of recurrence after curative resection for gastric cancer in China. World J Surg Oncol 14 , 305 (2016). Jiang, Y. et al. Association of adjuvant chemotherapy with survival in patients with stage II or III gastric cancer. JAMA Surg 152 , e171087 (2017). Catenacci, D. V. et al. Toward a treatment sequencing strategy: A systematic review of treatment regimens in advanced gastric cancer/gastroesophageal junction adenocarcinoma. Oncologist 26 , e1704–e1729 (2021). Shapiro, J. et al. Neoadjuvant chemoradiotherapy plus surgery versus surgery alone for oesophageal or junctional cancer (CROSS): Long-term results of a randomised controlled trial. Lancet Oncol 16 , 1090–1098 (2015). Al-Batran, S.-E. et al. Perioperative chemotherapy with fluorouracil plus leucovorin, oxaliplatin, and docetaxel versus fluorouracil or capecitabine plus cisplatin and epirubicin for locally advanced, resectable gastric or gastro-oesophageal junction adenocarcinoma (FLOT4): A randomised, phase 2/3 trial. Lancet 393 , 1948–1957 (2019). Cunningham, D. et al. Perioperative chemotherapy versus surgery alone for resectable gastroesophageal cancer. N Engl J Med 355 , 11–20 (2006). Liu, T. et al. First-line nivolumab plus chemotherapy vs chemotherapy in patients with advanced gastric, gastroesophageal junction and esophageal adenocarcinoma: CheckMate 649 Chinese subgroup analysis. Int J Cancer 152 , 749–760 (2023). Janjigian, Y. Y. et al. First-line nivolumab plus chemotherapy versus chemotherapy alone for advanced gastric, gastro-oesophageal junction, and oesophageal adenocarcinoma (CheckMate 649): a randomised, open-label, phase 3 trial. Lancet (London, England) 398 , 27–40 (2021). Wang, L. et al. Intestinal subtype as a biomarker of response to neoadjuvant immunochemotherapy in locally advanced gastric adenocarcinoma: Insights from a prospective phase II trial. Clin Cancer Res 31 , 74–86 (2025). Yl, V. et al. Neoadjuvant atezolizumab plus chemotherapy in gastric and gastroesophageal junction adenocarcinoma: the phase 2 PANDA trial. Nature medicine 30 , (2024). Wang, L. et al. Intestinal Subtype as a Biomarker of Response to Neoadjuvant Immunochemotherapy in Locally Advanced Gastric Adenocarcinoma: Insights from a Prospective Phase II Trial. Clinical Cancer Research OF1–OF13 (2024) doi:10.1158/1078-0432.CCR-24-2436. Chen, D. et al. Prognostic and predictive value of a pathomics signature in gastric cancer. Nat Commun 13 , 6903 (2022). Ludwig, J. A. & Weinstein, J. N. Biomarkers in cancer staging, prognosis and treatment selection. Nat Rev Cancer 5 , 845–856 (2005). Smith, J. C. & Sheltzer, J. M. Genome-wide identification and analysis of prognostic features in human cancers. Cell Rep 38 , 110569 (2022). Suzuki, T., Kitagawa, Y., Nankinzan, R. & Yamaguchi, T. Early gastric cancer diagnostic ability of ultrathin endoscope loaded with laser light source. World J Gastroenterol 25 , 1378–1386 (2019). De Martino, M., Rathmell, J. C., Galluzzi, L. & Vanpouille-Box, C. Cancer cell metabolism and antitumour immunity. Nat Rev Immunol 24 , 654–669 (2024). Chen, Y. et al. Metabolomic machine learning predictor for diagnosis and prognosis of gastric cancer. Nat. Commun. 15 , 1657 (2024). Shen, X. et al. Normalization and integration of large-scale metabolomics data using support vector regression. Metabolomics 12 , 89 (2016). Tang Z. et al. Multiplex immune profiling reveals the role of serum immune proteomics in predicting response to preoperative chemotherapy of gastric cancer. Cell Reports. Medicine 4 , 100931 (2023). R, H. & M, C. Complications of diagnostic gastrointestinal endoscopy. Endoscopy 22 , (1990). Choi, S. J. et al. Gastric Cancer and Intestinal Metaplasia: Differential Metabolic Landscapes and New Pathways to Diagnosis. International Journal of Molecular Sciences 25 , 9509 (2024). Sharma, P. & Allison, J. P. The future of immune checkpoint therapy. Science 348 , 56–61 (2015). Liu, J. et al. Improved Efficacy of Neoadjuvant Compared to Adjuvant Immunotherapy to Eradicate Metastatic Disease. Cancer Discov 6 , 1382–1399 (2016). Jiang, H. et al. Efficacy and safety of neoadjuvant sintilimab, oxaliplatin and capecitabine in patients with locally advanced, resectable gastric or gastroesophageal junction adenocarcinoma: early results of a phase 2 study. J Immunother Cancer 10 , e003635 (2022). Kim, H.-D., Ryu, M.-H., Park, Y. S., Lee, J. S. & Kang, Y.-K. A phase 2 study of durvalumab plus DOS (docetaxel, oxaliplatin, S-1) neoadjuvant chemotherapy followed by surgery and adjuvant durvalumab plus S-1 chemotherapy in patients with resectable locally advanced gastric cancer: An interim efficacy report. JCO 41 , 408–408 (2023). Xu, J. et al. Sintilimab Plus Chemotherapy for Unresectable Gastric or Gastroesophageal Junction Cancer. JAMA 330 , 2064–2074 (2023). Li, N. et al. Efficacy and safety of neoadjuvant sintilimab in combination with FLOT chemotherapy in patients with HER2-negative locally advanced gastric or gastroesophageal junction adenocarcinoma: An investigator-initiated, single-arm, open-label, phase II study. Int J Surg 110 , 2071–2084 (2024). Liu, Y. et al. Camrelizumab combined with FLOFOX as neoadjuvant therapy for resectable locally advanced gastric and gastroesophageal junction adenocarcinoma: Updated results of efficacy and safety. JCO 39 , 4036–4036 (2021). Zhong, W. et al. Efficacy and safety of camrelizumab combined with oxaliplatin and S‐1 as neoadjuvant treatment in locally advanced gastric or gastroesophageal junction cancer: A phase II, single‐arm study. Cancer Med 13 , e7006 (2024). Waldman, A. D., Fritz, J. M. & Lenardo, M. J. A guide to cancer immunotherapy: from T cell basic science to clinical practice. Nat Rev Immunol 20 , 651–668 (2020). Nielsen, D. L., Juhl, C. B., Chen, I. M., Kellermann, L. & Nielsen, O. H. Immune checkpoint Inhibitor-Induced diarrhea and Colitis: Incidence and Management. A systematic review and Meta-analysis. Cancer Treat. Rev. 109 , 102440 (2022). Junttila, M. R. & de Sauvage, F. J. Influence of tumour micro-environment heterogeneity on therapeutic response. Nature 501 , 346–354 (2013). Bejarano, L., Jordāo, M. J. C. & Joyce, J. A. Therapeutic Targeting of the Tumor Microenvironment. Cancer Discov 11 , 933–959 (2021). Tsai, C.-H. et al. Immunoediting instructs tumor metabolic reprogramming to support immune evasion. Cell Metabolism 35 , 118-133.e7 (2023). Faubert, B., Solmonson, A. & DeBerardinis, R. J. Metabolic reprogramming and cancer progression. Science 368 , eaaw5473 (2020). Pavlova, N. N. & Thompson, C. B. The Emerging Hallmarks of Cancer Metabolism. Cell Metab 23 , 27–47 (2016). Liu, X. et al. The significant role of amino acid metabolic reprogramming in cancer. Cell Commun Signal 22 , 1–26 (2024). Wang, D. & Wan, X. Progress in research on the role of amino acid metabolic reprogramming in tumour therapy: A review. Biomedicine & Pharmacotherapy 156 , 113923 (2022). Siska, P. J. & Rathmell, J. C. T Cell Metabolic Fitness in Anti-Tumor Immunity. Trends Immunol 36 , 257–264 (2015). Chen, C.-L., Hsu, S.-C., Ann, D. K., Yen, Y. & Kung, H.-J. Arginine Signaling and Cancer Metabolism. Cancers (Basel) 13 , 3541 (2021). Holbert, C. E., Cullen, M. T., Casero, R. A. & Stewart, T. M. Polyamines in cancer: integrating organismal metabolism and antitumour immunity. Nat Rev Cancer 22 , 467–480 (2022). Peyraud, F. et al. Circulating L-arginine predicts the survival of cancer patients treated with immune checkpoint inhibitors. Ann Oncol 33 , 1041–1051 (2022). Mossmann, D. et al. Arginine reprograms metabolism in liver cancer via RBM39. Cell 186 , 5068-5083.e23 (2023). Canè, S., Geiger, R. & Bronte, V. The roles of arginases and arginine in immunity. Nat Rev Immunol 1–19 (2024) doi:10.1038/s41577-024-01098-2. Yan, J. et al. Molecular mechanisms and therapeutic significance of tryptophan metabolism and signaling in cancer. Molecular Cancer 23 , 241 (2024). Frezza, C. Histidine degradation boosts cancer therapy. Nature 559 , 484 (2018). J, W., W, W., F, Z. & Q, D. The role of branched chain amino acids metabolic disorders in tumorigenesis and progression. Biomedicine & pharmacotherapy = Biomedecine & pharmacotherapie 153 , (2022). Zhenyukh, O. et al. High concentration of branched-chain amino acids promotes oxidative stress, inflammation and migration of human peripheral blood mononuclear cells via mTORC1 activation. Free Radic Biol Med 104 , 165–177 (2017). Tönjes, M. et al. BCAT1 promotes cell proliferation through amino acid catabolism in gliomas carrying wild-type IDH1. Nat Med 19 , 901–908 (2013). Li, J.-T. et al. BCAT2-mediated BCAA catabolism is critical for development of pancreatic ductal adenocarcinoma. Nat Cell Biol 22 , 167–174 (2020). Mayers, J. R. et al. Tissue of origin dictates branched-chain amino acid metabolism in mutant Kras-driven cancers. Science 353 , 1161–1165 (2016). Nazemi, M. et al. The extracellular matrix supports breast cancer cell growth under amino acid starvation by promoting tyrosine catabolism. PLoS Biol 22 , e3002406 (2024). Li, J. et al. Tyrosine catabolism enhances genotoxic chemotherapy by suppressing translesion DNA synthesis in epithelial ovarian cancer. Cell Metabolism 35 , 2044-2059.e8 (2023). Zhang, L., Yao, Y. & Liu, S. Targeting fatty acid metabolism for cancer therapy. Fundamental Research (2024) doi:10.1016/j.fmre.2024.09.011. Tan, Z. et al. Targeting CPT1A-mediated fatty acid oxidation sensitizes nasopharyngeal carcinoma to radiation therapy. Theranostics 8 , 2329–2347 (2018). Lin, R. et al. Fatty Acid Oxidation Controls CD8+ Tissue-Resident Memory T-cell Survival in Gastric Adenocarcinoma. Cancer Immunol Res 8 , 479–492 (2020). Lee, J.-Y. et al. Polyunsaturated fatty acid biosynthesis pathway determines ferroptosis sensitivity in gastric cancer. Proceedings of the National Academy of Sciences 117 , 32433–32442 (2020). Hanna, V. S. & Hafez, E. A. A. Synopsis of arachidonic acid metabolism: A review. Journal of Advanced Research 11 , 23–32 (2018). He, B. et al. Identification of cytochrome b 5 CYTB-5.1 and CYTB-5.2 in C . elegans ; evidence for differential regulation of SCD. Biochimica et Biophysica Acta (BBA) - Molecular and Cell Biology of Lipids 1863 , 235–246 (2018). Wang, B. et al. Metabolism pathways of arachidonic acids: mechanisms and potential therapeutic targets. Signal Transduction and Targeted Therapy 6 , 1–30 (2021). Cui L. et al. Targeting arachidonic acid metabolism enhances immunotherapy efficacy in ARID1A-deficient colorectal cancer. Cancer Research (2024) doi:10.1158/0008-5472.CAN-24-1611. Chen, X. et al. CYP4F2-Catalyzed Metabolism of Arachidonic Acid Promotes Stromal Cell-Mediated Immunosuppression in Non–Small Cell Lung Cancer. Cancer Research 82 , 4016–4030 (2022). Fan, X. et al. TRIM21 promotes tumor growth and gemcitabine resistance in pancreatic cancer by inhibiting EPHX1-mediated arachidonic acid metabolism. Advanced Science n/a , 2413674. Li Y. et al. Longitudinal plasma proteome profiling reveals the diversity of biomarkers for diagnosis and cetuximab therapy response of colorectal cancer. Nat Commun 15 , 980 (2024). Harel, M. et al. Longitudinal plasma proteomic profiling of patients with non-small cell lung cancer undergoing immune checkpoint blockade. J. Immunother. Cancer 10 , e004582 (2022). Qian, L. et al. Proteomic landscape of epithelial ovarian cancer. Nat. Commun. 15 , 6462 (2024). Li, Y. et al. Longitudinal plasma proteome profiling reveals the diversity of biomarkers for diagnosis and cetuximab therapy response of colorectal cancer. Nat. Commun. 15 , 980 (2024). Wang, H. et al. Serum metabolic traits reveal therapeutic toxicities and responses of neoadjuvant chemoradiotherapy in patients with rectal cancer. Nat. Commun. 13 , 7802 (2022). Malier, M. et al. Tumor-associated macrophages confer resistance to chemotherapy (trifluridine/tipiracil) in digestive cancers by overexpressing thymidine phosphorylase. Cancer Letters 606 , 217307 (2024). Wusiman, M. et al. Serum S-adenosylhomocysteine, rather than homocysteine, is associated with hepatocellular carcinoma survival: A prospective cohort study. The American Journal of Clinical Nutrition 120 , 481–490 (2024). Guo, T. et al. LINC00662 promotes hepatocellular carcinoma progression via altering genomic methylation profiles. Cell Death Differ 27 , 2191–2205 (2020). Huang, W. et al. AHCYL1 senses SAH to inhibit autophagy through interaction with PIK3C3 in an MTORC1-independent manner. Autophagy 18 , 309–319. Uchiyama, N., Tanaka, Y. & Kawamoto, T. Aristeromycin and DZNeP cause growth inhibition of prostate cancer via induction of mir-26a. European Journal of Pharmacology 812 , 138–146 (2017). Goodman, R. S., Johnson, D. B. & Balko, J. M. Corticosteroids and cancer immunotherapy. Clin Cancer Res 29 , 2580–2587 (2023). Wang, Y. et al. Corticosteroid administration for cancer-related indications is an unfavorable prognostic factor in solid cancer patients receiving immune checkpoint inhibitor treatment. Int Immunopharmacol 99 , 108031 (2021). Wei, J. et al. Association between gut microbiome-related metabolites and symptomatic hand osteoarthritis in two independent cohorts. eBioMedicine 98 , 104892 (2023). Zhao, H. et al. Inflammation and tumor progression: Signaling pathways and targeted intervention. Sig Transduct Target Ther 6 , 1–46 (2021). Tumeh, P. C. et al. PD-1 blockade induces responses by inhibiting adaptive immune resistance. Nature 515 , 568–571 (2014). Yasuda, T. & Wang, Y. A. Gastric cancer immunosuppressive microenvironment heterogeneity: Implications for therapy development. Trends Cancer 10 , 627–642 (2024). Doroshow, D. B. et al. PD-L1 as a biomarker of response to immune-checkpoint inhibitors. Nat Rev Clin Oncol 18 , 345–362 (2021). Zelenay, S. et al. Cyclooxygenase-dependent tumor growth through evasion of immunity. Cell 162 , 1257–1270 (2015). Jin, J., Byun, J.-K., Choi, Y.-K. & Park, K.-G. Targeting glutamine metabolism as a therapeutic strategy for cancer. Exp Mol Med 55 , 706–715 (2023). Additional Declarations There is NO Competing Interest. Supplementary Files FigureandtableS.docx supplementary figure and table rs.pdf Reporting Summary Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6365771","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":456125466,"identity":"66a202a0-3010-4bea-a497-8e841b60d6ae","order_by":0,"name":"Wei Zhu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAuklEQVRIiWNgGAWjYDACCRBhYJPARqqWNJK1MBxOIF6H/OweM4kPBefz+MQOMH748YdB3pyQFsY5Z8wkZxjcLmaTTmCW7G1jMNzZQEALs0SO2W0eg9uJbdJA7/A2MCQYHCCghQ2k5Y/BObAWxj9/iNDCA9LCYHAArIWZh40ILRISaeU/ewySgVoSm6Vl2yQMNxDSIj8jebPBjz92ifNnJx/8+OaPjTxBW5AAYwMDLJpGwSgYBaNgFFAIAMduOTc4pyKQAAAAAElFTkSuQmCC","orcid":"","institution":"Hangzhou Institute of Medicine (HIM), Chinese Academy of Sciences","correspondingAuthor":true,"prefix":"","firstName":"Wei","middleName":"","lastName":"Zhu","suffix":""},{"id":456125467,"identity":"16a72a68-57d2-46c0-b0a2-097c2c2dd8b5","order_by":1,"name":"Xiaoyong Zhang","email":"","orcid":"","institution":"Hangzhou Institute of Medicine (HIM), Chinese Academy of Sciences","correspondingAuthor":false,"prefix":"","firstName":"Xiaoyong","middleName":"","lastName":"Zhang","suffix":""},{"id":456125468,"identity":"d84afd58-b242-4b3a-99e7-a5ac57d71eac","order_by":2,"name":"Xueli Hu","email":"","orcid":"","institution":"Hangzhou Institute of Medicine (HIM), Chinese Academy of Sciences","correspondingAuthor":false,"prefix":"","firstName":"Xueli","middleName":"","lastName":"Hu","suffix":""},{"id":456125469,"identity":"9f3e76bd-e100-4a1a-9f30-b87d5886e995","order_by":3,"name":"Kankai Zhu","email":"","orcid":"","institution":"the First Affiliated Hospital, Zhejiang University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Kankai","middleName":"","lastName":"Zhu","suffix":""},{"id":456125470,"identity":"0a6086bd-eb2d-4aa7-88be-36a2f610f4e2","order_by":4,"name":"Huidi Zhu","email":"","orcid":"","institution":"the First Affiliated Hospital, Zhejiang University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Huidi","middleName":"","lastName":"Zhu","suffix":""},{"id":456125471,"identity":"9d3c0d60-7321-4aa2-a5af-f935ee8371f3","order_by":5,"name":"Shenghong Guan","email":"","orcid":"","institution":"Hangzhou Institute of Medicine (HIM), Chinese Academy of Sciences","correspondingAuthor":false,"prefix":"","firstName":"Shenghong","middleName":"","lastName":"Guan","suffix":""},{"id":456125472,"identity":"e074822a-5b00-4c11-8567-788b84fc89bb","order_by":6,"name":"Hailong Jin","email":"","orcid":"","institution":"the First Affiliated Hospital, Zhejiang University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Hailong","middleName":"","lastName":"Jin","suffix":""},{"id":456125473,"identity":"5338c550-3876-4877-b570-99624e65c9ff","order_by":7,"name":"Yang Li","email":"","orcid":"","institution":"the First Affiliated Hospital, Zhejiang University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Yang","middleName":"","lastName":"Li","suffix":""},{"id":456125474,"identity":"4f33e767-c8e9-4048-9f7d-c7c0671189fb","order_by":8,"name":"Xiaodong Wang","email":"","orcid":"","institution":"the First Affiliated Hospital, Zhejiang University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Xiaodong","middleName":"","lastName":"Wang","suffix":""},{"id":456125475,"identity":"4904501f-c991-428d-a5d4-6ddcc5a2d4d6","order_by":9,"name":"Yang Dong","email":"","orcid":"","institution":"the First Affiliated Hospital, Zhejiang University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Yang","middleName":"","lastName":"Dong","suffix":""},{"id":456125476,"identity":"207e3938-1342-4d3c-aef8-05d72f50c387","order_by":10,"name":"Yue Feng","email":"","orcid":"","institution":"Hangzhou Institute of Medicine (HIM), Chinese Academy of Sciences","correspondingAuthor":false,"prefix":"","firstName":"Yue","middleName":"","lastName":"Feng","suffix":""},{"id":456125477,"identity":"07b929e6-5374-4851-81d5-693e6e0767bc","order_by":11,"name":"Tianzhe Gao","email":"","orcid":"","institution":"the First Affiliated Hospital, Zhejiang University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Tianzhe","middleName":"","lastName":"Gao","suffix":""},{"id":456125478,"identity":"44b0e387-e7de-44e4-bfa4-1a16558d46e4","order_by":12,"name":"Caizhao Lin","email":"","orcid":"","institution":"the First Affiliated Hospital, Zhejiang University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Caizhao","middleName":"","lastName":"Lin","suffix":""},{"id":456125479,"identity":"59d85cae-1ae9-4c9a-ba58-fd7b4b1caf76","order_by":13,"name":"Jing-kui Tian","email":"","orcid":"","institution":"Chinese academy of sciences","correspondingAuthor":false,"prefix":"","firstName":"Jing-kui","middleName":"","lastName":"Tian","suffix":""},{"id":456125480,"identity":"977d9e33-db2e-412a-b501-ee3f1705008f","order_by":14,"name":"Jiahui Ma","email":"","orcid":"","institution":"Chinese Academy of Sciences","correspondingAuthor":false,"prefix":"","firstName":"Jiahui","middleName":"","lastName":"Ma","suffix":""},{"id":456125481,"identity":"8d8ea8e8-90b1-49ed-aca5-9b8c368563a9","order_by":15,"name":"Xiaosun Liu","email":"","orcid":"","institution":"Zhejiang University","correspondingAuthor":false,"prefix":"","firstName":"Xiaosun","middleName":"","lastName":"Liu","suffix":""}],"badges":[],"createdAt":"2025-04-03 04:35:09","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6365771/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6365771/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":82709847,"identity":"8fb1ec81-b608-4615-914c-4eb3959e90e8","added_by":"auto","created_at":"2025-05-14 11:17:20","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":488730,"visible":true,"origin":"","legend":"\u003cp\u003eStudy cohort information.\u003c/p\u003e\n\u003cp\u003e(A) Flowchart of inclusion and exclusion criteria for the study cohort;\u003c/p\u003e\n\u003cp\u003e(B) Clinical characteristics of 108 GC patients.\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-6365771/v1/fa6e9fdcf9d22b62f2c8e684.png"},{"id":82709848,"identity":"ba0bbe08-d083-40f3-a165-a4091d4e9779","added_by":"auto","created_at":"2025-05-14 11:17:20","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":593529,"visible":true,"origin":"","legend":"\u003cp\u003eDynamic trajectory analysis of the metabolic landscape following NAIC therapy in gastric cancer.\u003c/p\u003e\n\u003cp\u003e(A) Flowchart of the collection and analysis process for 369 blood samples.\u003c/p\u003e\n\u003cp\u003e(B) PCA results, illustrating the dynamic changes in the metabolic landscape during NAIC therapy. Each data point represents the average metabolite levels across all patients.\u003c/p\u003e\n\u003cp\u003e(C) Overview of Mufzz time-series analysis, including clustering results, heatmaps of each cluster, and the corresponding top 5 significantly perturbed metabolic pathways.\u003c/p\u003e\n\u003cp\u003e(D) Dynamic trends of key metabolites within each cluster during the treatment process, with the y-axis representing the mean expression levels of metabolites normalized by Z-score.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-6365771/v1/7e9bb61fc6848ec71a1abaa6.png"},{"id":82709850,"identity":"8e624668-1c02-408a-9af1-770d9a70b69d","added_by":"auto","created_at":"2025-05-14 11:17:20","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":461017,"visible":true,"origin":"","legend":"\u003cp\u003eConstruction of a predictive model for the efficacy of NAIC therapy in GC based on baseline plasma metabolic profiles.\u003c/p\u003e\n\u003cp\u003e(A) PCA results between R and NR after NAIC therapy.\u003c/p\u003e\n\u003cp\u003e(B) Feature weights of metabolites in the 21-PM model.\u003c/p\u003e\n\u003cp\u003e(C) Performance evaluation of the 21-PM prediction model in the test set: receiver operating characteristic curve (ROC curve).\u003c/p\u003e\n\u003cp\u003e(D) Performance evaluation of the 21-PM prediction model in the Testing set: confusion matrix and evaluation metrics including accuracy, sensitivity, specificity, and recall rate.\u003c/p\u003e\n\u003cp\u003e(E) RECIST results in patients with high and low 21-PM risk scores.\u003c/p\u003e\n\u003cp\u003e(F) Comparison of treatment response rates between patients with high and low 21-PM risk scores. *indicates a \u003cem\u003eP\u003c/em\u003e-value \u0026lt; 0.05.\u003c/p\u003e\n\u003cp\u003e(G) Comparison of baseline plasma levels of 13,14-Dihydro PGF-1a, hydroxytryptophol, and hydroxyprogesterone between R and NR.\u003c/p\u003e\n\u003cp\u003e(H) Correlation network analysis between the 21 predictive biomarkers and immune cell phenotypes.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-6365771/v1/ce91d6ca2424d8dbafffa603.png"},{"id":82709849,"identity":"712cadb1-21db-46ae-a1b3-fcd560a68648","added_by":"auto","created_at":"2025-05-14 11:17:20","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":445681,"visible":true,"origin":"","legend":"\u003cp\u003eConstruction and validation of a radiomics-based treatment efficacy monitoring model.\u003c/p\u003e\n\u003cp\u003e(A) Workflow diagram for identification of R-specific metabolites.\u003c/p\u003e\n\u003cp\u003e(B) Representative pre- and post-treatment CT images from Responders and Non-responders.\u003c/p\u003e\n\u003cp\u003e(C) Longitudinal treatment response waterfall plot for the 108-patient cohort: Total bar length represents individual treatment duration; Bar color corresponds to RECIST v1.1 criteria classification; Key RECIST evaluation timepoints marked with distinct geometric symbols; Dashed lines indicate surgical interventions (length=timing; color=outcome).\u003c/p\u003e\n\u003cp\u003e(D) Weight coefficients of the 11-MMI model components.\u003c/p\u003e\n\u003cp\u003e(E) ROC curve analysis of the 11-MMI model in the test cohort.\u003c/p\u003e\n\u003cp\u003e(F) Confusion matrix evaluating the 11-MMI model's classification accuracy in the test dataset.\u003c/p\u003e\n\u003cp\u003e(G) Violin plots comparing serum concentrations of three signature metabolites between R and NR: Indolelactic acid、Glycerol 5-hydroxydecanoate、3-Methylxanthine.\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-6365771/v1/1ca6743e2a099f0c727c704e.png"},{"id":82709857,"identity":"e8d544b5-c9ac-4217-a8f0-d1c934000295","added_by":"auto","created_at":"2025-05-14 11:17:21","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":469074,"visible":true,"origin":"","legend":"\u003cp\u003eConstruction and validation of a pathology-based treatment efficacy monitoring model.\u003c/p\u003e\n\u003cp\u003e(A) Workflow diagram for identification of R-specific \u0026nbsp;\u0026nbsp;metabolites.\u003c/p\u003e\n\u003cp\u003e(B) Representative pre- and post-treatment pathological \u0026nbsp;\u0026nbsp;images comparing R an NR).\u003c/p\u003e\n\u003cp\u003e(C) PCA of metabolic profiles between R and NR groups \u0026nbsp;\u0026nbsp;based on pathological outcomes.\u003c/p\u003e\n\u003cp\u003e(D) Weight coefficients of metabolic biomarkers in the 13-MMP \u0026nbsp;\u0026nbsp;model.\u003c/p\u003e\n\u003cp\u003e(E) ROC curve analysis of the 13-MMP model in the \u0026nbsp;\u0026nbsp;validation cohort.\u003c/p\u003e\n\u003cp\u003e(F) Confusion matrix evaluating classification \u0026nbsp;\u0026nbsp;performance of the 13-MMP model in test dataset.\u003c/p\u003e\n\u003cp\u003e(G) Violin plots comparing serum \u0026nbsp;\u0026nbsp;concentrations of three signature metabolites between R and NR groups: \u0026nbsp;\u0026nbsp;eicosapentaenoic acid, epinephrine, and palmitoylcarnitine.\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-6365771/v1/82c9818abfa5e3e43dfe155d.png"},{"id":88882091,"identity":"c47e7ad6-15d7-4f98-b1cd-76a1e7eb6eda","added_by":"auto","created_at":"2025-08-12 11:19:19","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3819816,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6365771/v1/09311a20-ce91-4251-8161-cdf2e03039e7.pdf"},{"id":82710572,"identity":"4fad1ce9-96f9-4f51-9049-9adaf5317c55","added_by":"auto","created_at":"2025-05-14 11:25:21","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":5748564,"visible":true,"origin":"","legend":"supplementary figure and table","description":"","filename":"FigureandtableS.docx","url":"https://assets-eu.researchsquare.com/files/rs-6365771/v1/acc6785e13c1dadceaee5803.docx"},{"id":82709869,"identity":"0f4946b9-4444-45d0-8bf8-4c348a1c8cf9","added_by":"auto","created_at":"2025-05-14 11:17:21","extension":"pdf","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":3047924,"visible":true,"origin":"","legend":"Reporting Summary","description":"","filename":"rs.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6365771/v1/5fd9bdb12b8e17f83352d52d.pdf"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"Metabolomic Insights into the Predictive Landscape of Neoadjuvant Immunochemotherapy in Gastric Cancer: Towards Precision Medicine with Machine Learning","fulltext":[{"header":"Introduction","content":"\u003cp\u003eGastric cancer (GC) is the sixth most prevalent malignant tumor, accounting for over 33% of all cancer-related fatalities. Currently, 70\u0026ndash;90% of GC patients are discovered at an advanced stage, significantly limiting their prognosis. Although surgery and chemotherapy have improved the survival rate of advanced GC patients, the overall survival (OS) rate for GC patients is still less than 40%, and more than half of GC patients have recurrence\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e,\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e. Following the advent of comprehensive treatment strategies such as perioperative chemotherapy, preoperative chemotherapy, and neoadjuvant chemotherapy, patients' survival and prognosis improved considerably\u003csup\u003e\u003cspan additionalcitationids=\"CR4 CR5\" citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e. However, overall, the prognosis for patients with locally advanced gastric cancer remains unsatisfactory. The advent of immune checkpoint inhibitors (ICIs) such as tremelimumab, toripalimab, sintilimab, and nivolumab had resulted in new advances in the treatment of metastatic cancer. Preoperative neoadjuvant immunochemotherapy (NAIC) has been shown in studies to considerably slow the growth of GC tumors. The CheckMate649 trial found that NAIC treatment (Nivolumab plus chemotherapy) had a positive therapeutic effect on patients, ushering in a new age of immunotherapy for GC\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e,\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e. Another prospective, single-arm, phase II clinical trial shown that NAIC achieved a 30% pathologic complete regression rate (pCR) and a 43% major pathologic regression rate (MPR), with good patient tolerance, demonstrating the clinical benefits of NAIC in gastric cancer treatment\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e. Furthermore, Verschoor et al. used neoadjuvant atezolizumab in conjunction with chemotherapy in gastric and gastroesophageal junction (G/GEJ) tumors, which demonstrated significant anticancer activity, with 70% (95% CI: 46%-88%) of patients reaching MPR and 45% (95% CI: 23%-68%) obtaining pCR\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eMany patients with stomach cancer still do not benefit from NAIC, despite the fact that it has been demonstrated to successfully induce pathological regression in locally advanced gastric adenocarcinoma and can, to some extent, enhance patient survival\u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e. Imaging, endoscopic, and histological exams were mostly used to assess the clinical prognosis and efficacy of GC\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e. The aforementioned techniques were used by clinical physicians to gather clinical features for efficacy evaluation, such as tumor location, TNM staging data, and Lauren categorization; however, the accuracy of these evaluation techniques was restricted\u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e,\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e. Imaging examinations exposed patients to radiation, and while endoscopy was the gold standard for diagnosing GC, it was intrusive and costly. These restrictions hampered the quick monitoring of patient treatment efficacy\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e. This situation underscores the urgent need to identify reliable biomarkers for patient stratification during NAIC treatment. Tumor growth was frequently accompanied by the reprogramming of metabolic components in the body. This process was primarily triggered by the accumulation of metabolic alterations resulting from mutations in enzyme-coding genes during tumor cell proliferation. Epigenetic changes in proteins or inhibitors within mature tumor cells caused cancer while also having a direct impact on metabolism\u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e. Furthermore, when cancers evolve with genetic factors in their microenvironment, tumor cells acquired metabolic alterations, which was one of the primary driving forces for intratumoral and intertumoral variability\u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e. As a result, elucidating the metabolic reprogramming landscape during NAIC treatment of GC and identifying biomarkers that represent NAIC efficacy could improve NAIC's clinical efficacy, allowing for more timely and effective treatment strategies for GC patients.\u003c/p\u003e \u003cp\u003eMetabolomics technology, as a systematic analysis, had the potential to better reflect the net outcomes of NAIC-tumor interactions. Previous research had mostly focused on molecular pathways to identify biomarkers linked with the occurrence, prognosis, and recurrence of GC. However, they lacked the characterization of the metabolic reprogramming profile of tumor patients undergoing NAIC, as well as the identification of metabolites associated with the efficacy of NAIC treatment for GC\u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e. This significantly impedes real-time monitoring of the GC treatment process, making it harder for NAIC to achieve superior therapeutic outcomes. In this work, metabolomics was employed to assess the influence of NAIC reconfiguration on the metabolic profile of GC patients. To better identify biomarkers linked with NAIC therapy success, this study conducted a comprehensive examination across three distinct clinical evidence dimensions, as follows: First, an efficacy prediction model was created using baseline plasma samples in order to predict whether gastric cancer (GC) patients will benefit from NAIC therapy. Second, real-time clinical efficacy was dynamically assessed using imaging examinations to establish a foundation for forecasting future therapy response variations. Finally, more thorough histological examination data was used to evaluate patient clinical outcomes and find biomarkers associated with therapy monitoring.\u003c/p\u003e \u003cp\u003eThis study, which used metabolomics technology to describe the metabolic reprogramming profiles of patients with varying therapeutic outcomes, was based on the clinical empirical results of NAIC treatment for GC. It developed an efficacy monitoring model, identified efficacy-related characteristic biomarkers using machine learning, and assessed clinical efficacy based on various degrees of evidence. This served the goal of precision treatment for GC by offering a potent tool for forecasting and tracking the effectiveness of NAIC.\u003c/p\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Study cohort, response information, adverse reactions, study design\u003c/h2\u003e \u003cp\u003eThis study retrospectively included patients diagnosed with GC through pathology at the The First Affiliated Hospital, Zhejiang University School of Medicine. All patients received a NAIC regimen, following the standard dosing protocols described in previous studies. Initially, 154 patients were recruited, and after strict inclusion and exclusion criteria, a total of 108 patients were finally included in this study (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eA). All patients involved in the study had baseline clinical information gathered, including as age, gender, tumor differentiation, Lauren classification, ECOG performance status score, and so on (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eB). Among the 108 patients, the age distribution was 65.00 (58.00, 72.00) years, with 31 males (28.70%) and 77 females (71.30%). In terms of ECOG performance status scores, among patients scoring 0\u0026ndash;3, 80.56% scored 0 and 17.59% scored 1. In the preclinical stage classification, Stage II patients accounted for 1.85%, Stage III patients accounted for 50.00%, and Stage IV patients accounted for 48.15%. In terms of treatment plans, 23.15% of patients received a chemotherapy regimen that included taxane drugs combined with PD1 inhibitors, while 76.85% of patients received a chemotherapy regimen that did not include taxane drugs combined with PD1 inhibitors. As of the analysis date of June 30, 2023, out of 108 patients who received NAIC, 65 patients showed a treatment response, with 64 achieving Partial Response (PR) and 1 achieving complete Response (CR). 43 patients did not show a treatment response, with 9 experiencing Progression Disease (PD) and 34 having Stable Disease (SD). A total of 42 patients underwent surgery after NAIC in the study, and post-operative pathological evaluation revealed pCR in 12 patients (28.57%). Becker1a/1b, Becker2, and Becker3 were observed in 21 cases (50.00%), 10 cases (23.81%), and 11 cases (26.19%) of patients, respectively. Detailed characteristics are summarized in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.The study also detailed the adverse reactions experienced by patients during treatment. In terms of the incidence of adverse reactions, the most common was nausea, with an incidence rate of 86.44%, followed by anemia at 80%, fatigue at 79.66%, anorexia at 74.58%, and leukopenia at 59.05%. It was worth noting that the frequency of grade 3 or higher severity among these adverse reactions was also relatively high. In addition, during the treatment of GC with NAIC, other adverse reactions were also observed, including diarrhea, vomiting, nausea, limb numbness, and other adverse reactions. For detailed information seeing in n\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eBaseline clinical features of the GC NAIC treatment population\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVariables\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTotal (n\u0026thinsp;=\u0026thinsp;108)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNR (n\u0026thinsp;=\u0026thinsp;43)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eR (n\u0026thinsp;=\u0026thinsp;65)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cem\u003eP\u003c/em\u003e value\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.347\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;60\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e37 (34.26)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e17 (39.53)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e20 (30.77)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026ge;\u0026thinsp;60\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e71 (65.74)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e26 (60.47)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e45 (69.23)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGender, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.643\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFemale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e30 (27.78)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e13 (30.23)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e17 (26.15)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e78 (72.22)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e30 (69.77)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e48 (73.85)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBMI, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.238\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e18.5\u0026thinsp;\u0026le;\u0026thinsp;BMI\u0026thinsp;\u0026lt;\u0026thinsp;24\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e70 (65.42)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e27 (64.29)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e43 (66.15)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e24\u0026thinsp;\u0026le;\u0026thinsp;BMI\u0026thinsp;\u0026lt;\u0026thinsp;28\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e19 (17.76)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e5 (11.90)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e14 (21.54)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBMI\u0026thinsp;\u0026lt;\u0026thinsp;18.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e15 (14.02)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e9 (21.43)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e6 (9.23)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBMI\u0026thinsp;\u0026ge;\u0026thinsp;28\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3 (2.80)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1 (2.38)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2 (3.08)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eECOG, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.214\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e87 (80.56)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e32 (74.42)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e55 (84.62)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e19 (17.59)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e9 (20.93)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e10 (15.38)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1 (0.93)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1 (2.33)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0 (0.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1 (0.93)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1 (2.33)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0 (0.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDifferentiation status, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.701\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e20 (18.52)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e6 (13.95)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e14 (21.54)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePoor\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e85 (78.70)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e36 (83.72)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e49 (75.38)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUnknown\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3 (2.78)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1 (2.33)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2 (3.08)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLauren classification, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.082\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDiffuse/mixed\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e60 (55.56)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e28 (65.12)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e32 (49.23)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIndeterminate\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e5 (4.63)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3 (6.98)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2 (3.08)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIntestinal\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e43 (39.81)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e12 (27.91)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e31 (47.69)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTumor site, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.063\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e80 (74.07)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e36 (83.72)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e44 (67.69)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGEJ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e28 (25.93)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e7 (16.28)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e21 (32.31)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHER2 status, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.509\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e53 (52.48)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e21 (52.50)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e32 (52.46)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e1+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e30 (29.70)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e14 (35.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e16 (26.23)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e2+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e11 (10.89)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e4 (10.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e7 (11.48)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e3+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e7 (6.93)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1 (2.50)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e6 (9.84)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMSI/MMR status, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e1.000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMSI-H/dMMR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e7 (7.07)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3 (7.69)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e4 (6.67)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMSS/pMMR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e92 (92.93)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e36 (92.31)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e56 (93.33)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCPS, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.154\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e10 (24.39)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e6 (37.50)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e4 (16.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e1\u0026ndash;5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e6 (14.63)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3 (18.75)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3 (12.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e5\u0026ndash;10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e5 (12.20)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0 (0.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e5 (20.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026ge;\u0026thinsp;10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e20 (48.78)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e7 (43.75)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e13 (52.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eClinical stage, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.200\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStage Ⅱ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2 (1.85)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2 (4.65)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0 (0.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStage Ⅲ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e54 (50.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e19 (44.19)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e35 (53.85)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStage Ⅳ\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e52 (48.15)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e22 (51.16)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e30 (46.15)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eClinical T category, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.693\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eT2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3 (2.78)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2 (4.65)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1 (1.54)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eT3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e37 (34.26)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e14 (32.56)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e23 (35.38)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eT4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e68 (62.96)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e27 (62.79)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e41 (63.08)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eClinical N category, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.047\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eN+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e104 (96.30)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e39 (90.70)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e65 (100.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eN0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e4 (3.70)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e4 (9.30)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0 (0.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eClinical M category, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.409\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eM0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e58 (53.70)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e21 (48.84)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e37 (56.92)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eM1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e50 (46.30)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e22 (51.16)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e28 (43.08)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTherapy, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.059\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eChemo(T)\u0026thinsp;+\u0026thinsp;PD1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e25 (23.15)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e14 (32.56)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e11 (16.92)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eChemo\u0026thinsp;+\u0026thinsp;PD1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e83 (76.85)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e29 (67.44)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e54 (83.08)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSurgery efficiency (Becker)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;.001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBecker 1a\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e12 (28.57)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0 (0.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e20 (40.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBecker 1b\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e9 (21.43)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1 (8.33)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e20 (26.67)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBecker 2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e10 (23.81)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3 (25.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e7 (23.33)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBecker 3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e11 (26.19)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e8 (66.67)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3 (10.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eTable S 1. Baseline plasma samples from 108 patients and 261 samples collected throughout the treatment process were gathered, with each post-treatment sample obtained at the end of each treatment cycle (up to a maximum of 6 months). The longitudinal cohort included in the study covered a total of 6 sampling time points during the treatment period. Next, a non-targeted metabolomics approach based on LC-MS/MS was used to obtain the metabolomic profile of the plasma samples. After metabolite data annotation, 1001 metabolites were obtained.\u003c/p\u003e \u003cp\u003eBased on the baseline (T0) metabolomics of 108 patients, machine learning algorithms were applied to investigate the association between metabolic features and treatment response, resulting in the development of the 21-PM GC prediction model. This model was used to predict the response of GC patients to NAIC treatment, evaluated the impact of clinical characteristic factors, introduced risk stratification strategies, and assessed the model's ability to grade the level of risk. In order to deeply explore the changes in metabolites during the treatment process and identify potential monitoring biomarkers, the study focused on key efficacy nodes in the cohort. Machine learning was employed to construct a monitoring model (11-MMI) based on identified R-specific metabolites. Given the authority of pathological assessment in efficacy determination, the study included 42 patients whose treatment decisions were based on pathological findings. Potential monitoring markers for R group metabolites related to pathological efficacy specificity were further investigated, and a 13-MMP monitoring model was built using machine learning approaches. The development of these models aims to improve the accuracy of monitoring treatment responses in GC patients.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003e3.2 Dynamic Analysis of Metabolic Landscape\u003c/h3\u003e\n\u003cp\u003eTo depict the dynamic changes in the plasma metabolic landscape of patients undergoing NAIC, plasma samples were collected from patients at baseline before surgery and across the treatment process. A total of 369 plasma samples were obtained across six defined treatment time points: pre-treatment (time point 1), first combined treatment (time point 2), second (time point 3), third (time point 4), fourth (time point 5), and fifth (time point 6) (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e2\u003c/span\u003eA).\u003c/p\u003e \u003cp\u003eUsing UHPLC-Q-Orbitrap-MS/MS technology, 369 plasma samples collected from 108 GC patients across treatment were analyzed. Following data batch normalization (Figure S 1) and metabolite annotation, a total of 1001 endogenous metabolites were identified. Through unsupervised principal component analysis (PCA), there was significant differences in metabolic profiles at the six time points, indicating that metabolites change as treatment progresses (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e2\u003c/span\u003eB). To track the dynamic changes in the metabolic landscape across the treatment process, the Mfuzz algorithm was applied to perform time series clustering analysis on metabolite data at multiple time points. The results showed that the trajectories of metabolite changes could be roughly divided into four trends. Among them, the metabolite abundance in C1 continuously decreased as the treatment progressed, while C3 exhibited a trend completely opposite to that of C1. The metabolite abundance in C2 decreased and then rebounded after the T4 time point, and C4 showed a plateau phase at the T1-T2 nodes (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e2\u003c/span\u003eC and Figure S 2A-B). Notably, after performing principal component analysis on each cluster separately, the differences in metabolic profiles at each time point among the different clusters became more apparent (Figure S 2C). The Kyoto Encyclopedia of Genes and Genomes (KEGG) (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.metaboanalyst.ca/\u003c/span\u003e\u003cspan address=\"https://www.metaboanalyst.ca/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) analysis results showed that C1, C2, and C3 have high similarity in amino acid metabolism and energy metabolism, particularly in pathways such as Arginine biosynthesis and Valine, leucine, and isoleucine biosynthesis, where they were significantly enriched. In contrast, C4 had unique enrichment in pathways such as Biosynthesis of unsaturated fatty acids and Arachidonic acid metabolism (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e2\u003c/span\u003eB). Dynamic change line charts were created for the metabolites in the significantly disturbed metabolic pathways of each cluster over the course of treatment. These primarily included Leucine, L-Isoleucine, Threonine, Oxoglutaric acid, and Histamine in C1; Leucine, L-Isoleucine, L-Valine, Phenylethylamine, Methylimidazoleacetic acid, and Indoleacetaldehyde in C2; Pyridoxal, 4-Pyridoxic acid, DOPA, Gentisic acid, Glycocholic acid, and Glycochenodeoxycholic acid in C3; and 18-HETE, Arachidonic acid, 8,9-DiHETrE, 11,12-EET, Dihomo-gamma-linolenic acid, Eicosapentanoic acid, adrenaline, and Thyroxine in C4 (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e2\u003c/span\u003eD).\u003c/p\u003e\n\u003ch3\u003e3.4 Prediction model for the therapy response\u003c/h3\u003e\n\u003cp\u003eTo look for potential predictive biomarkers for treatment response, the baseline (T0) plasma metabolome was examined in 108 GC patients receiving NAIC therapy. In this study, patients with Complete Response (CR) and Partial Response (PR) are defined as having a treatment response (Response, R), while patients with Stable Disease (SD) and Progressive Disease (PD) are defined as having no treatment response (Non-Response, NR).\u003c/p\u003e \u003cp\u003ePCA showed a significant difference between the R (n\u0026thinsp;=\u0026thinsp;65) and the NR (n\u0026thinsp;=\u0026thinsp;43) (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e3\u003c/span\u003eA). Preliminary differential analysis identified 61 differential metabolites (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05, Figure S 3A). KEGG analysis showed that these metabolites were significantly enriched in the following signaling pathways: Primary bile acid biosynthesis, Valine, leucine and isoleucine biosynthesis, Vitamin B6 metabolism (Figure S 3B). Following sample balancing with the SMOTE algorithm (Figure S 4A), LASSO regression was used on the training cohort to build a plasma metabolome-derived predictive model that effectively distinguished treatment responders (R) from non-responders (NR), resulting in 21 predictive metabolites (21-PM) (Figure S 4B-C and Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e3\u003c/span\u003eB). There were significant differences in 21 metabolites between the R and the NR, with 16 of them (Thymidine, Thialysine ketimine, S\u0026thinsp;\u0026minus;\u0026thinsp;Adenosylhomocysteine, N6,N6\u0026thinsp;\u0026minus;\u0026thinsp;dimethyllysine, Hydroxytryptophol, Hydroxyprogesterone, Hexanoylcarnitine, Dihomo\u0026thinsp;\u0026minus;\u0026thinsp;gamma\u0026thinsp;\u0026minus;\u0026thinsp;linolenic acid, Asp\u0026thinsp;\u0026minus;\u0026thinsp;tyr, 15,16\u0026thinsp;\u0026minus;\u0026thinsp;DiHODE, 13,14\u0026thinsp;\u0026minus;\u0026thinsp;Dihydro PGF\u0026thinsp;\u0026minus;\u0026thinsp;1α, 8,9\u0026thinsp;\u0026minus;\u0026thinsp;DiHETrE, 4\u0026thinsp;\u0026minus;\u0026thinsp;Hydroxy\u0026thinsp;\u0026minus;\u0026thinsp;2\u0026minus;oxoglutaric acid, 3\u0026thinsp;\u0026minus;\u0026thinsp;hydroxynonanoyl carnitine, 2\u0026thinsp;\u0026minus;\u0026thinsp;Aminoadenosine, 3-Hydroxy-cis-5-tetradecenoylcarnitine) significantly upregulated in R, while 5 of them (Corticosterone, coenzyme Q2, 12S\u0026thinsp;\u0026minus;\u0026thinsp;HHT, 4\u0026thinsp;\u0026minus;\u0026thinsp;Pyridoxic acid, and 3\u0026thinsp;\u0026minus;\u0026thinsp;hydroxydecanoyl carnitine) significantly downregulated in R (figure violin). The main contributors include 13,14\u0026thinsp;\u0026minus;\u0026thinsp;Dihydro PGF\u0026thinsp;\u0026minus;\u0026thinsp;1a, 15,16\u0026thinsp;\u0026minus;\u0026thinsp;DiHODE, 4\u0026thinsp;\u0026minus;\u0026thinsp;Hydroxy\u0026thinsp;\u0026minus;\u0026thinsp;2\u0026minus;oxoglutaric acid, 4\u0026thinsp;\u0026minus;\u0026thinsp;Pyridoxic acid, 3\u0026thinsp;\u0026minus;\u0026thinsp;hydroxydecanoyl carnitine, and 12S\u0026thinsp;\u0026minus;\u0026thinsp;HHT (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e3\u003c/span\u003eB). The model's AUC value was 0.9935 (95%CI: 0.9834-1, Accuracy: 97.67%, Sensitivity: 97.67%, Specificity: 97.67% (Figure S 4D-E). The prediction model was subsequently applied to the testing set, demonstrating robust performance with an AUC of 0.897 (95% CI: 0.7521-1), accuracy of 90.91%, sensitivity of 92.31%, and specificity of 88.89% (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e3\u003c/span\u003eC-D).\u003c/p\u003e \u003cp\u003eA thorough evaluation of the predictive model's robustness was carried out by investigating the impact of major clinical parameters (gender, age, BMI, ECOG performance status, differentiation status, Lauren classification, and clinical stage) on prognostic markers. Chi-square tests were used to conduct differential analysis comparing responders (R) and non-responders (NR) in order to discover clinical factors related with treatment response. The results showed that there were no significant differences between R and NR in terms of gender, age, BMI, ECOG, differentiation status, and Lauren classification (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026gt;\u0026thinsp;0.05). Only the clinical N stage showed a significant difference between the groups (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e and Figure S 5A). The predictive capability of clinical N staging for treatment response was investigated through analysis using AUC values as the evaluation metric. Clinical N staging alone demonstrated limited predictive value, with an AUC of 0.559 (95% CI: 0.5039\u0026ndash;0.6138), which was far lower than the predictive ability of the 21-PM metabolite characteristic model (Figure S 5B). To further evaluate the influence of clinical N0 status on predictive performance, all four N0 patients (whom are in the NR) were excluded from analysis. This exclusion resulted in an AUC of 0.884 (95% CI: 0.815\u0026ndash;0.953), representing a modest reduction compared to the 21-PM model's performance (AUC\u0026thinsp;=\u0026thinsp;0.994) (Figure S 5C). The potential impact of sample size reduction was evaluated through a bootstrap analysis with 1,000 iterations, where four patients were randomly excluded in each iteration. The analysis yielded a median AUC of 0.891 (range: 0.879\u0026ndash;0.918) (Figure S 5D). When the 1,000-iteration AUC values were compared with both the N0-patient-excluded group and the original complete dataset, the iterative AUC values were found to be significantly lower than those from the original complete dataset (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), yet significantly higher than those from the N0-patient-excluded group alone (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) (Figure S 5D and table S). These results indicated that while clinical N staging demonstrated some predictive capability for treatment response (AUC\u0026thinsp;=\u0026thinsp;0.559), its impact remained within acceptable limits overall. Furthermore, the predictive efficacy of clinical N staging was shown to be markedly inferior compared to the 21-metabolite panel (21-PM) identified in the study.\u003c/p\u003e \u003cp\u003eFor enhanced clinical application of the 21-PM model in precision therapy, a risk stratification management strategy was implemented. The risk stratification management strategy was constructed through the following steps: First, risk scores were calculated for each patient based on the model's metabolite coefficients, serving as an independent predictive indicator. Second, patients were stratified according to the median score. Finally, distribution differences across response groups were determined for each stratum. The AUC value reached 0.891 (95% CI: 0.826\u0026ndash;0.957) (Figure S 5E), and patients were divided into different risk groups (high-risk group and low-risk group) based on the median risk score. The predictive performance of the risk scoring model was evaluated for each patient in the test set. Results demonstrated that showed that most NR patients belonged to the high-risk group, while most R patients belonged to the low-risk group, which was related to the predictive ability of the risk scoring (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e3\u003c/span\u003eE). The Response Evaluation Criteria in Solid Tumours (RECIST, version 1.1) outcomes were further compared between the two risk groups, and the results showed a higher proportion of PD and SD in the high-risk group, while PR was more prominent in the low-risk group (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e3\u003c/span\u003eF). This indicated that the predictive model and the benefit-risk indicators formed by the model successfully identified patients who needed optimized treatment plans. Distribution differences of clinically relevant prognostic factors were analyzed between high- and low-risk groups. The intestinal type (Lauren classification), known for better prognosis, showed significantly higher prevalence in the low-risk group (Figure S 6F), and lower T stages were more prevalent in the low group (Figure S 6G). Furthermore, the results of the multivariate logistic regression indicated that the risk score was an independent predictor (Table S 2). At the same time, the 21-PM model could reflect the relationship between metabolite characteristics and treatment response, and this strategy could accurately reflect the relationship between risk scores and treatment response. In summary, the risk stratification management strategy could categorize patients by calculating risk scores, providing accurate clinical treatment strategies for patients, thereby better serving the precise treatment of GC.\u003c/p\u003e \u003cp\u003eGiven the neoadjuvant immunotherapy-chemotherapy combination regimen used in this study, additional analyses were conducted to evaluate immune response-related outcomes. First, regarding the expression of PD-L1, there was no significant difference in the CPS score of PD-L1 expression between R and NR, but there was a trend of higher CPS expression in the R group, which was also indicated by the model risk score results (Figure S 5H). Subsequent analysis focused on clinical immune cell profiles. A total of 30 patients' immune test results were collected, mainly including CD3, CD4, CD8, CD19, and the CD3/CD4 ratio. The results showed no significant differences in immune cell expression between the R group and the NR group at baseline (Figure S 6A). To examine potential relationships between immune markers and metabolic levels, correlations were analyzed between selected key metabolites and immune cell populations (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e3\u003c/span\u003eH). 13,14-Dihydro PGF-1α and hydroxytryptophol showed significant positive correlations with CD3 expression, while hydroxyprogesterone and hydroxytryptophol were positively correlated with CD8 levels (Figure S 6B). Negative correlations were identified between thialysine ketimine/hydroxyprogesterone and the CD3/CD4 ratio, as well as between 2-aminoadenosine, 3-hydroxydecanoyl carnitine, hydroxyprogesterone and CD19 expression (Figure S 6B).\u003c/p\u003e \u003cp\u003eCorrelation analyses were performed to evaluate potential associations between key metabolites and various clinical parameters. The results revealed that higher expression levels of 13,14-dihydro PGF-1α significantly correlated with more favorable clinical stages (Figure S 6C). Analysis of Lauren classification demonstrated elevated expression patterns of thymidine, S-adenosylhomocysteine, and 15,16-DiHODE in intestinal-type cases. Distinct metabolic profiles were observed across treatment groups, with 13,14-dihydro PGF-1α, dihomo-γ-linolenic acid, and 3-hydroxy-cis-5-tetradecenoylcarnitine showing increased expression in the Chemo\u0026thinsp;+\u0026thinsp;PD1 cohort ((Figure S 6C)). Furthermore, 15,16-DiHODE exhibited higher expression in gastroesophageal junction (GEJ) cases, while 8,9-DiHETrE demonstrated elevated levels in microsatellite stable/mismatch repair proficient (MSS/pMMR) patients (Figure S 6C).\u003c/p\u003e\n\u003ch3\u003e3.5 A monitoring model based on Imaging diagnosis\u003c/h3\u003e\n\u003cp\u003eIn order to explore the changes in the plasma metabolome during the treatment process, the treatment status was regularly evaluated based on imaging during the treatment of patients, and the first effective treatment node (PR, CR) and the final ineffective treatment node (SD, PD) were determined (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e4\u003c/span\u003eA-C). The plasma metabolic characteristics of 85 GC patients receiving NAIC treatment regimens at the baseline period (T0) and key treatment stages (Tx) were studied in detail. A comparative analysis was first performed to examine plasma metabolomic changes across different treatment response groups (R group and NR group). It was found that the R group experienced more significant plasma metabolomic changes than the NR group during the treatment phase, from baseline to a key decision point (Figure S 7A-B). Further comparison of the fold change (FC) value changes in metabolic levels before and after treatment between the two groups of patients revealed that the overall metabolic changes in the R group were more significant (Figure S 7C). For example, compared to the NR, the R group showed a significant increase in the plasma levels of 13,14\u0026thinsp;\u0026minus;\u0026thinsp;Dihydro PGF\u0026thinsp;\u0026minus;\u0026thinsp;1a after treatment, while the trend of change for 5\u0026thinsp;\u0026minus;\u0026thinsp;Hydroxy\u0026thinsp;\u0026minus;\u0026thinsp;L\u0026minus;tryptophan was opposite between the two groups. The change in platelet-activating factor was more pronounced in the NR group. The above results indicated that the patient group with effective treatment exhibited more pronounced responses at the metabolomic level. The baseline-predicted 21-PM model demonstrated a certain predictive ability in the treatment response prediction of 85 patients, with an AUC value of 0.8440 (95% CI: 0.7607\u0026ndash;0.9274), Accuracy: 0.7765, Sensitivity: 0.7037, Specificity: 0.9032 (Figure S 7D-E). However, the potential of these 21 metabolites in distinguishing between the R and the NR based on their changes from baseline to the critical treatment phase still needed further exploration. The FC values of the 21 metabolites were calculated between pre- and post-treatment timepoints, followed by PCA. The results showed that the FC values of these metabolites had a certain ability to distinguish between the R group and the NR group (Figure S 7F).\u003c/p\u003e \u003cp\u003eGiven the significant metabolic shifts identified in the R group, an in-depth investigation was performed to detect treatment-associated metabolite changes unique to responders. Metabolites that changed exclusively in the R group were combined with those that showed opposite trends in the R and NR groups, resulting in a panel of 156 R-specific metabolite (Figure S 7G). KEGG pathway analysis (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.metaboanalyst.ca/\u003c/span\u003e\u003cspan address=\"https://www.metaboanalyst.ca/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) of the 156 characteristic metabolites revealed significant disturbances in multiple metabolic pathways such as Valine, leucine, and isoleucine biosynthesis, Taurine and hypotaurine metabolism, Cysteine and methionine metabolism, Arginine biosynthesis, Primary bile acid biosynthesis, and Citrate cycle (TCA cycle) (Figure S 7H). Analysis of FC values for these 156 metabolites revealed distinct dynamic patterns between groups, prompting further refinement through LASSO regression to identify metabolites with optimal discriminative power (Figure S 8A-B). This analysis identified 11 characteristic metabolites whose FC patterns effectively differentiated metabolic profiles between R and NR groups (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e4\u003c/span\u003eD). The model's AUC value was 0.9679 (95% CI: 0.9318-1, Accuracy: 91.18%, Sensitivity: 88.24%, Specificity: 94.12%) (Figure S 8C-D). In the test set, the AUC value was 0.8636 (95% CI: 0.6830-1, Accuracy: 82.35%, Sensitivity: 72.73%, Specificity: 100.00% (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e4\u003c/span\u003eE-F). The PCA results of the 11 metabolites showed that their ability to distinguish between the R group and the NR group was better compared to the 21 metabolites in the predictive model (Figure S 8E). Further analysis of the expression differences of these 11 metabolites between the R group and the NR group revealed that the changes in 13,14\u0026thinsp;\u0026minus;\u0026thinsp;Dihydro PGF\u0026thinsp;\u0026minus;\u0026thinsp;1a, Gentisic acid, and Tetrahydropersin were significantly higher in the R group compared to the NR group, while the changes in Glycerol 5\u0026thinsp;\u0026minus;\u0026thinsp;hydroxydecanoate, Indolelactic acid, 3\u0026thinsp;\u0026minus;\u0026thinsp;Methylxanthine, and His\u0026thinsp;\u0026minus;\u0026thinsp;Trp were more significant in the NR group (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e4\u003c/span\u003eG and Figure S 8F). Moreover, comparing the expression differences of the aforementioned 11 metabolites between the R group and NR group at baseline and during the critical treatment phase showed that Glycerol 5\u0026thinsp;\u0026minus;\u0026thinsp;hydroxydecanoate, 3\u0026thinsp;\u0026minus;\u0026thinsp;Methylxanthine, and His\u0026thinsp;\u0026minus;\u0026thinsp;Trp, which had significant differences at baseline, became indistinguishable during the critical treatment phase (Figure S 9A). In contrast, 13,14\u0026thinsp;\u0026minus;\u0026thinsp;Dihydro PGF\u0026thinsp;\u0026minus;\u0026thinsp;1a, Gentisic acid, and Indolelactic acid, which were indistinguishable at baseline, showed differences during the critical treatment phase (Figure S 9A). To characterize the temporal dynamics of the 11 signature metabolites during treatment, expression levels were quantified across T0-T5 timepoints in 85 patients. The results of the Kruskal-Wallis test showed that 13,14\u0026thinsp;\u0026minus;\u0026thinsp;Dihydro PGF\u0026thinsp;\u0026minus;\u0026thinsp;1α and Gentisic acid changed significantly at 6 time points (Figure S 9B). Finally, baseline plasma immune information from 21 patients (11 in the R group and 10 in the NR group) and post-treatment plasma immune information from 19 patients (12 in the R group and 7 in the NR group) were also collected. The Spearman correlation analysis showed that in the baseline data, in group R patients, the expression of His\u0026thinsp;\u0026minus;\u0026thinsp;Trp was significantly negatively correlated with CD3 and CD8, while significantly positively correlated with NK cells. Glycerol 5\u0026thinsp;\u0026minus;\u0026thinsp;hydroxydecanoate was significantly negatively correlated with CD3, and Glutamine was significantly negatively correlated with the CD19 ratio (Figure S 10A). In the NR group of patients, Prostaglandin F2b (PGF2b) and Indolelactic acid were significantly positively correlated with the number of NK cells, while Glycerol 5\u0026thinsp;\u0026minus;\u0026thinsp;hydroxydecanoate was negatively correlated with the number of NK cells (Figure S 10A). In the NR group of patients, His\u0026thinsp;\u0026minus;\u0026thinsp;Trp was negatively correlated with CD4 and CD4/CD8, and 13,14\u0026thinsp;\u0026minus;\u0026thinsp;Dihydro PGF\u0026thinsp;\u0026minus;\u0026thinsp;1a was negatively correlated with CD19 (Figure S 10A). The above results indicated the complex differential relationships between metabolite levels and immune cells in the R group and NR group.\u003c/p\u003e \u003cp\u003eNext was 13,14\u0026thinsp;\u0026minus;\u0026thinsp;Dihydro PGF\u0026thinsp;\u0026minus;\u0026thinsp;1a. There was no correlation in the R and NR groups at the base and in the R group at the time point, but there was a significant positive correlation between the NR group at the time point and CD19. This indicated that the correlation between 13,14\u0026thinsp;\u0026minus;\u0026thinsp;Dihydro PGF\u0026thinsp;\u0026minus;\u0026thinsp;1a and immune cells changed before and after treatment. Therefore, the relationship between the FC of metabolite variations and the clinical immune expression of PD-L1 was explored. The results showed that the difference in the pre- and post-variation of 13,14\u0026thinsp;\u0026minus;\u0026thinsp;Dihydro PGF\u0026thinsp;\u0026minus;\u0026thinsp;1a in the R group was more significant in patients with high PD-L1 expression (CPS\u0026thinsp;\u0026ge;\u0026thinsp;10) (Figure S 10B). This suggested that 13,14\u0026thinsp;\u0026minus;\u0026thinsp;Dihydro PGF\u0026thinsp;\u0026minus;\u0026thinsp;1a might play a role in the immune microenvironment of GC, especially in patients with high PD-L1 expression.\u003c/p\u003e \u003cp\u003eThe expression of PGF2b in GC patients and its correlation with immune cells underwent significant changes during the treatment process. At baseline, PGF2b was significantly positively correlated with the number of NK cells in the NR, while after treatment, PGF2b was significantly correlated with the number of CD4 cells in the R. Additionally, the expression of PGF2b in the R group decreased from baseline to post-treatment, whereas it increased in the NR group (Figure S 10C). Further analysis of the relationship between PD-L1 expression and the FC of PGF2b revealed that in the R group, although the differences in PD-L1 expression levels were not significant (p\u0026thinsp;\u0026gt;\u0026thinsp;0.05), higher PD-L1 expression was associated with lower PGF2b changes; conversely, in the NR group, higher PD-L1 expression was associated with higher PGF2b changes (Figure S 10B). These results indicated that changes in immune expression during treatment caused variations in PFG2b levels, with the decrease in PGF2b potentially being associated with a better treatment response.\u003c/p\u003e\n\u003ch3\u003e3.6 A monitoring model based on Pathologic diagnosis\u003c/h3\u003e\n\u003cp\u003eIn the process of diagnosing and evaluating the efficacy of GC, histopathological diagnosis usually requires obtaining tumor tissue samples through endoscopic biopsy or surgical resection. This process is invasive and may pose certain trauma and complication risks to the patient. However, despite these potential adverse factors, the Becker score based on histopathological information remains one of the most authoritative methods for evaluating the efficacy of GC\u003csup\u003e\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e. Therefore, a subset of 42 surgical patients with available pathological outcomes was selected from the cohort, comprising 21 responders and 21 non-responders as classified by Becker criteria (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e5\u003c/span\u003eA-B and Table S 3). Compared to the imaging diagnosis, among these 42 patients, 28 were classified as response and 9 as non-response (Figure S 11A). This indicated that there was a certain bias in monitoring the efficacy response of patients through imaging diagnosis, while monitoring the efficacy of patients using histopathological information was more accurate. Considering the invasive procedures and potential trauma and complications associated with obtaining tumor tissue samples, the collection of plasma samples was particularly important due to its low invasiveness and ability to reflect the real-time impact of drugs on the body's metabolism. Therefore, supported by histopathological diagnostic evidence and plasma metabolomics data, the impact of NAIC on tumor metabolic reprogramming was systematically investigated, leading to the development of a GC therapeutic monitoring model. Compared to the efficacy monitoring model (11-MMI) based on imaging evidence, this model could reduce information bias caused by the method of efficacy evaluation to a certain extent, resulting in higher accuracy. During the patient recruitment process from October 2021 to June 2023, a total of 42 patients underwent surgical treatment after NAIC, preoperative tissue samples were collected, and the Becker score was used to assess the tumor progression status, serving as evidence to support clinical efficacy outcomes. Based on the clinical efficacy information supported by the aforementioned histopathological evidence, further studies were conducted on the baseline and preoperative plasma samples of these patients to explore the dynamic changes of characteristic metabolites in the treatment-effective group (R group) and the treatment-ineffective group (NR group) as determined by pathological results. PCA was initially performed on the 11-imaging efficacy-associated metabolites, demonstrating partial discriminative capacity between between the R group and the NR group to some extent (Figure S 11B). However, Evaluation of the 11-metabolite model's predictive performance in the histopathology-validated cohort yielded an AUC of 0.569 (95% CI: 0.390\u0026ndash;0.748), indicating limited discriminative capacity (Figure S 11C), indicating its relatively low predictive ability. This indicated that biomarkers identified through imaging assessment had certain limitations in evaluating clinical efficacy. Therefore, it was very important to characterize the differences in plasma metabolite reprogramming between the R group and the NR group of patients based on histopathological evidence, mine the characteristic metabolites of the R group, and construct a clinical efficacy monitoring model.\u003c/p\u003e \u003cp\u003eA comparative analysis of plasma metabolomic profiles was initially performed between different treatment responses (R and NR). The results showed that from baseline to preoperative, the metabolic profile of patients in the R group underwent more significant changes throughout the entire treatment process (Figure S 11D-E). The identical analytical methodology employed for the imaging cohort was applied to identify pathology-specific characteristic metabolites, resulting in the detection of 80 R-specific metabolites (Figure S 11G). The KEGG results for the 80 characteristic metabolites showed that metabolic pathways such as Tryptophan metabolism, Butanoate metabolism, and Alanine, aspartate, and glutamate metabolism are disturbed (Figure S 11H). For additional validation of the characteristic metabolites' predictive capability, patients were randomly allocated into training and validation sets at an 8:2 ratio. In the training set, LASSO regression was used to screen for characteristic metabolites. The results showed that 13 metabolites were selected (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e5\u003c/span\u003eD and Figure S 12A-B), with a model AUC value of 0.9653 (95% CI: 0.91585-1), Accuracy: 91.18%, Sensitivity: 88.24%, Specificity: 94.12% (Figure S 12C-D). The model performed welled on the test set, AUC: 0.9375 (95%CI: 0.76426-1), Accuracy: 87.50%, Sensitivity: 100.00%, Specificity: 75.00% (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e5\u003c/span\u003eE-F). These results indicated that the 13 metabolites selected through LASSO regression had high predictive accuracy in distinguishing between the R group and the NR group. The PCA results of the 13 metabolites also showed that their ability to distinguish between the R group and the NR group was superior to that of the 11 metabolites selected by the imaging group (Figure S 12E).\u003c/p\u003e \u003cp\u003eTo investigate immune-metabolite correlations during NAIC therapy, baseline plasma immune profiles were analyzed in 18 GC patients (9 R and 9 N]), with paired post-treatment data obtained from 14 patients (7 R and 7 NR). Notably, significant positive correlations were identified in treatment responders between key metabolites - eicosapentanoic acid (EPA), mhppa sulphate, and 13,14-dihydro-PGF-1α - and post-treatment immune cell populations (CD3+, CD4+, CD19+), suggesting their potential immunomodulatory role during NAIC therapy (Figure S 13A). The value of FC of EPA was substantially lower in the R group for patients with high PD-L1 expression (CPS\u0026thinsp;\u0026ge;\u0026thinsp;10) than for those with low PD-L1 expression (CPS\u0026thinsp;\u0026lt;\u0026thinsp;10) (Figure S 13B). Notably, the R group's overall EPA levels grew significantly from the baseline to the preoperative period (Figure S 13B), indicating a possible connection between PD-L1 status and EPA dynamics.\u003c/p\u003e \u003cp\u003eFigure S 13A showed that, in the NR group, at baseline, CD4 showed a significant negative correlation with Mhppa sulphate and CD3 showed a significant negative correlation with Tridecanoylglycine, while CD19 showed a significant positive correlation with 4\u0026thinsp;\u0026minus;\u0026thinsp;Hydroxy\u0026thinsp;\u0026minus;\u0026thinsp;2\u0026minus;oxoglutarate. After treatment, adrenaline was negatively correlated with CD3 and positively correlated with NK cells; 13,14\u0026thinsp;\u0026minus;\u0026thinsp;Dihydro PGF\u0026thinsp;\u0026minus;\u0026thinsp;1a was positively correlated with CD19; the correlation of 4\u0026thinsp;\u0026minus;\u0026thinsp;Hydroxy\u0026thinsp;\u0026minus;\u0026thinsp;2\u0026minus;oxoglutarate with immune cells changed significantly, being positively correlated with CD3 and CD8, while negatively correlated with the CD4/CD8 ratio and NK cell count; Isoferulate 3\u0026thinsp;\u0026minus;\u0026thinsp;glucuronide was negatively correlated with NK cell count; Palmitoylcarnitine was negatively correlated with CD4.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eGC is a highly lethal malignant tumor globally, and the rise of immunotherapy had opened new pathways for its treatment. Among these, immune checkpoint inhibitors (ICIs) have become the first-line treatment for advanced gastric or esophageal adenocarcinoma\u003csup\u003e\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e,\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e. Neoadjuvant therapy, or preoperative treatment, offers GC patients a new treatment option aimed at shrinking tumors, increasing resection rates, reducing the risk of intraoperative spread, and potentially improving patient survival and quality of life\u003csup\u003e\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e,\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u003c/sup\u003e. In neoadjuvant therapy, the combination strategy of immunotherapy and chemotherapy had garnered significant attention. This strategy leverages the dual effects of chemotherapy in releasing tumor antigens and immune checkpoint inhibitors in relieving immune suppression, thereby enhancing the immune system's ability to attack tumors\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e. Clinical trials have shown that this combination therapy can improve patients' pathological complete response rates and survival rates\u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e,\u003cspan additionalcitationids=\"CR25\" citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eIn this study, approximately 60.19% of patients responded (PR\u0026thinsp;+\u0026thinsp;CR) to NAIC, with a pCR rate of 28.57%. The toxic reactions were manageable, with most treatment-related adverse events (TRAE) being grade 1 or 2. The most common TRAEs were nausea, anemia, fatigue, anorexia, and decreased white blood cell count, which are consistent with previous research findings\u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e,\u003cspan additionalcitationids=\"CR25 CR26 CR27 CR28\" citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e. Furthermore, the toxicity associated with immunotherapy was often due to a decrease in the number of naive T cells and the overactivation of memory T cells. These activated memory T cells might have invaded peripheral organs, causing inflammatory damage, such as in the gastrointestinal tract\u003csup\u003e\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e. According to the literature, in patients receiving PD-1/PD-L1 inhibitors combined with chemotherapy, the incidence of diarrhea of all grades ranged from 17\u0026ndash;40%, while the incidence of grade 3\u0026ndash;4 diarrhea was approximately 2%\u003csup\u003e31\u003c/sup\u003e. In our study, approximately 42.37% of patients experienced varying degrees of diarrhea after receiving NAIC treatment, with the proportion of grade 3 or higher diarrhea being 3.39%, a figure slightly higher than previously reported data. Most clinical trials used single drugs or single treatment regimens, whereas the patients in our cohort received a variety of chemotherapy regimens, including SOX, FLOT, XELX, and some patients also received taxane drugs. Reports indicated that the incidence of diarrhea caused by taxane drugs was higher than that caused by platinum-based drugs. At the same time, the types of PD-1/L1 inhibitors also varied, such as Trelagliptin, Tislelizumab, and Sintilimab. These factors might have been the reasons for the heterogeneity of adverse reactions to treatment. Moreover, by comparing the age and preclinical TNM staging of populations in various clinical trials, it was found that the population in this trial is relatively older and had higher TNM staging than those in other trials. This indicated that the physical condition of the subjects in this trial was slightly worse than that of the subjects in other trials, which might also have been one of the reasons for the slightly higher incidence of adverse reactions in this trial.\u003c/p\u003e \u003cp\u003eThe tumor microenvironment (TME) was considered a key determinant of tumor heterogeneity and plays an important role in regulating the tumor's response to various therapeutic interventions\u003csup\u003e\u003cspan additionalcitationids=\"CR33\" citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e\u003c/sup\u003e. Metabolic reprogramming had become a hallmark of cancer, closely related to the tumor microenvironment. The dynamic crosstalk between immune cells, stromal cells, and tumor cells had a decisive impact on the growth and survival of tumor cells\u003csup\u003e\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e,\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e\u003c/sup\u003e. This study employed longitudinal cohort monitoring to systematically characterize NAIC-induced metabolic reprogramming dynamics throughout the treatment course, following initial efficacy assessment. Using the Mfuzz algorithm for clustering analysis of time series data, four distinct patterns of metabolite changes was successfully identified. Notably, significant perturbations in amino acid metabolism pathways were identified across clusters 1\u0026ndash;3, with pronounced alterations observed in the pathway, such as leucine, valine, oxoglutarate, and 4-hydroxy-2-oxoglutarate, as well as biogenic amines related to inflammation and immune regulation, such as histamine, also showed significant changes in these clusters. In tumor cells, amino acids, as important proteins and signaling molecules, can influence tumor immunity by regulating the production of immune cells and immune factors\u003csup\u003e\u003cspan additionalcitationids=\"CR38\" citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e\u003c/sup\u003e. Arginine, which is an important component of protein synthesis and also a precursor to polyamines, creatine, and nitric oxide, plays a crucial regulatory role in tumor angiogenesis and is essential for the growth and proliferation of tumor and immune cells\u003csup\u003e\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e,\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e\u003c/sup\u003e. Furthermore, arginine plays an important role in the TME, where there is competition for arginine among cancer cells, immunosuppressive cells, and anti-tumor cells, and when tumor cells consume a large amount of arginine in the tumor microenvironment, anti-tumor cells are inhibited, thereby promoting immune evasion\u003csup\u003e\u003cspan additionalcitationids=\"CR43\" citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e\u003c/sup\u003e. Tryptophan (Trp) metabolism involves the kynurenine (Kyn) pathway, the serotonin (5-hydroxytryptamine, 5-HT) pathway, and the indole pathway, and its metabolic imbalance is associated with various cancers, primarily by promoting tumor growth and immune evasion through the generation of an TME\u003csup\u003e\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e\u003c/sup\u003e. Similar to arginine, the metabolism of tryptophan plays a crucial role in the immune regulation of tumor cells, and cancer cells and macrophages inhibit antigen-specific T cell responses by competitively consuming tryptophan in the TME\u003csup\u003e\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e\u003c/sup\u003e. Histidine catabolism also significantly affects the sensitivity of cancer cells to methotrexate by reducing the cellular pool of tetrahydrofolate\u003csup\u003e\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e\u003c/sup\u003e. The reprogramming of branched-chain amino acids (BCAAs) metabolism alters the levels of important metabolites including BCAAs, α-ketoglutarate (α-KG), glutamate, and reactive oxygen species (ROS), which play roles in protein synthesis and degradation, energy supply, signal transduction, and other processes, affecting the survival and growth of cancer cells\u003csup\u003e\u003cspan additionalcitationids=\"CR48 CR49 CR50\" citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e51\u003c/span\u003e\u003c/sup\u003e. Tyrosine metabolism is an important target in cancer treatment, particularly in enhancing the effects of chemotherapy and targeted therapy\u003csup\u003e\u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e52\u003c/span\u003e,\u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e53\u003c/span\u003e\u003c/sup\u003e. Relatively speaking, the metabolite changes in C4 were mainly concentrated in the biosynthesis of unsaturated fatty acids, tryptophan metabolism, and arachidonic acid metabolism pathways. Unlike other clusters, the abundance of these metabolites first increased during treatment, then entered a plateau phase during the T1-T3 stages, and finally decreased. This unique pattern of change might have been related to treatment resistance. Multiple studies have shown that the reprogramming of lipid metabolism reshapes the tumor immune environment, which is crucial for the survival, metastasis, and treatment resistance of cancer cells\u003csup\u003e\u003cspan additionalcitationids=\"CR55\" citationid=\"CR54\" class=\"CitationRef\"\u003e54\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e56\u003c/span\u003e\u003c/sup\u003e. Lee and others found that the polyunsaturated fatty acid (PUFA) biosynthesis pathway plays a crucial role in ferroptosis\u003csup\u003e\u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e57\u003c/span\u003e\u003c/sup\u003e. The biosynthesis of unsaturated fatty acids mainly occurs through the action of various metabolic enzymes, such as desaturases and elongases, resulting in the production of important lipid molecules like oleic acid (OA), arachidonic acid (AA), docosahexaenoic acid (DHA), and EPA\u003csup\u003e\u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e58\u003c/span\u003e,\u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e59\u003c/span\u003e\u003c/sup\u003e. AA was an important polyunsaturated fatty acid and a key component of cell membrane phospholipids, and it was metabolized through the cyclooxygenase (COX) and lipoxygenase (LOX) pathways, significantly influencing tumor growth, angiogenesis, cell migration, and immune cell function\u003csup\u003e\u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e60\u003c/span\u003e\u003c/sup\u003e. Cui et al. enhanced the immune response against colorectal cancer by targeting the arachidonic acid metabolic pathway, activating CD8\u0026thinsp;+\u0026thinsp;T cells, and inhibiting vasculogenic mimicry (VM)\u003csup\u003e\u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e61\u003c/span\u003e\u003c/sup\u003e. Furthermore, in non-small cell lung cancer, Chen et al. discovered that CYP4F2-mediated metabolism of arachidonic acid can promote stroma cell-mediated immune suppression in non-small cell lung cancer, and proposed a strategy to enhance the efficacy of immunotherapy by inhibiting CYP4F2\u003csup\u003e62\u003c/sup\u003e. Fan and others also found that inhibiting arachidonic acid metabolism can promote tumor growth and gemcitabine resistance in pancreatic cancer\u003csup\u003e\u003cspan citationid=\"CR63\" class=\"CitationRef\"\u003e63\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eDespite the continuous advancements in the treatment of GC, the heterogeneity of tumor treatment remains a significant challenge. In this study, approximately 60% of patients exhibited positive treatment outcomes. Predicting potential treatment responses before starting therapy and subsequently customizing personalized treatment strategies was crucial for the prognosis of GC patients. Multiple studies have shown that early diagnosis, prognosis prediction, and treatment response monitoring of cancer patients based on various omics data have shown good results\u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e,\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e,\u003cspan additionalcitationids=\"CR65 CR66 CR67\" citationid=\"CR64\" class=\"CitationRef\"\u003e64\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR68\" class=\"CitationRef\"\u003e68\u003c/span\u003e\u003c/sup\u003e. Therefore, this study developed a metabolite prediction model (21-PM) based on baseline plasma samples using machine learning. The results of the prediction model showed that the model's predictive performance was good and less influenced by clinical factors, highlighting the great potential of plasma metabolomics analysis in guiding patients to receive specific therapies (NAIC). The risk score was developed based on the 21-PM model to facilitate stratified management implementation. The risk score results showed that the higher the patient's score, the more likely they were to be sensitive to treatment. The introduction of the risk score not only enhanced our understanding of patients' treatment responses but also helps doctors assessed patients' prognosis before treatment and thereby formulate more precise treatment plans. Patients were stratified into high-risk and low-risk groups according to their risk scores, enabling comparative analysis of clinical characteristics and treatment outcomes across risk categories. Patients in the high-risk group might have required more aggressive treatment strategies, including more intense chemotherapy regimens, targeted therapy, or immunotherapy, while patients in the low-risk group might have benefited from more moderate treatments, reducing the risk of overtreatment. In summary, risk scoring, as a new prognostic tool, not only improves the accuracy of predicting the response of GC patients to NAIC treatment but also provides valuable information for clinical decision-making. Future research should further validate the universality and practicality of this model and explore its applicability in different patient populations.\u003c/p\u003e \u003cp\u003eAmong these 21 key metabolites, significant differential expression was observed in the responder (R) group, with 16 metabolites demonstrating marked upregulation and 5 showing pronounced downregulation. These changes in metabolites not only reflected alterations in tumor biological behavior but may also indicated the tumor's sensitivity to NAIC treatment. For example, among the upregulated metabolites, Thymidine, Thialysine ketimine, and S-Adenosylhomocysteine are involved in DNA synthesis and repair, amino acid metabolism, and methylation processes, which play crucial roles in tumor growth and response to chemotherapy drugs. Research on thymidine in the prognosis of GC mainly focuses on the expression levels of its metabolic enzymes. Multiple studies have demonstrated that the expression levels of thymidine phosphorylase and dihydropyrimidine dehydrogenase (DPD) are closely related to the prognosis of GC patients\u003csup\u003e\u003cspan citationid=\"CR69\" class=\"CitationRef\"\u003e69\u003c/span\u003e\u003c/sup\u003e. The upregulation of S-Adenosylhomocysteine may affect DNA methylation, thereby influencing gene expression and the tumor microenvironment. Studies have shown that in hepatocellular carcinoma (HCC), higher serum SAH levels are independently associated with poor prognosis in HCC patients\u003csup\u003e\u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e70\u003c/span\u003e,\u003cspan citationid=\"CR71\" class=\"CitationRef\"\u003e71\u003c/span\u003e\u003c/sup\u003e. The mechanism of action of the SOX, FLOT, and XELX treatment regimens used in this study primarily involves interfering with key processes such as DNA synthesis and cell division in tumor cells, which induces the accumulation of SAH\u003csup\u003e\u003cspan citationid=\"CR72\" class=\"CitationRef\"\u003e72\u003c/span\u003e,\u003cspan citationid=\"CR73\" class=\"CitationRef\"\u003e73\u003c/span\u003e\u003c/sup\u003e. Among the downregulated metabolites, corticosterone and coenzyme Q2 are related to stress response and energy metabolism, and their downregulation may reflect the metabolic inhibition of tumor cells under NAIC treatment. The downregulation of corticosterone may be related to the weakened stress response capability of tumor cells. Corticosterone had immunosuppressive effects and can inhibit the proliferation and function of various immune cells, such as T cells, B cells, and natural killer (NK) cells\u003csup\u003e\u003cspan citationid=\"CR74\" class=\"CitationRef\"\u003e74\u003c/span\u003e\u003c/sup\u003e. There were studies reporting that elevated levels of corticosterone might have been associated with poor prognosis in cancer patients receiving ICI treatment\u003csup\u003e\u003cspan citationid=\"CR75\" class=\"CitationRef\"\u003e75\u003c/span\u003e\u003c/sup\u003e. Hydroxytryptophol was a plasma metabolite of tryptophan, and studies have shown that it was expressed at higher levels in inflammatory diseases\u003csup\u003e\u003cspan citationid=\"CR76\" class=\"CitationRef\"\u003e76\u003c/span\u003e\u003c/sup\u003e. The study identified elevated hydroxytryptophol expression in R, potentially indicative of inflammatory activity within the tumor microenvironment\u003csup\u003e\u003cspan citationid=\"CR77\" class=\"CitationRef\"\u003e77\u003c/span\u003e\u003c/sup\u003e. The inflammatory response played an important role in the occurrence and development of tumors, and inflammatory cells and factors in the tumor microenvironment could promote the proliferation, invasion, and metastasis of tumor cells. Our further research found that there was a correlation between clinical T staging and the expression of Hydroxytryptophol, showing a trend where higher T staging corresponded to higher levels of Hydroxytryptophol expression. Clinical T staging, as an empirical indicator, had always been used by clinicians to assess the extent of tumor progression and predict patient prognosis. Therefore, by monitoring the levels of Hydroxytryptophol, it was expected to provide auxiliary evidence for assessing tumor invasiveness and determining the tumor's response to treatment, thereby aiding in the formulation of clinical treatment decisions. CD8\u0026thinsp;+\u0026thinsp;T cells are the primary tumor-killing cells, and it has been specifically demonstrated that the density of pre-existing CD8 T cells at the edge of aggressive tumors (metastatic melanoma) was associated with the response to PD-1 inhibitor treatment\u003csup\u003e\u003cspan citationid=\"CR78\" class=\"CitationRef\"\u003e78\u003c/span\u003e\u003c/sup\u003e. The study identified a significant positive correlation between hydroxytryptophol levels and CD8\u0026thinsp;+\u0026thinsp;T cell infiltration, suggesting its potential immunomodulatory role in antitumor responses. 13,14-Dihydro PGF-1α and 15,16-DiHODE emerged as high-contribution metabolites in the predictive model, with 13,14-dihydro-PGF-1α and 12S-HHT being characterized as cyclooxygenase-derived metabolites of arachidonic acid, implicating prostaglandin pathways in treatment responses. The upregulation of 13,14-Dihydro PGF-1α in the R group and the downregulation of 12S-HHT in the R group,15,16-DiHODE was a metabolite produced by the metabolism of arachidonic acid through the lipoxygenase pathway, which was upregulated in the R group, while 8,9-DiHETrE was a metabolite produced by the metabolism of AA through the cytochrome P450 pathway, which was also upregulated in the R group. Furthermore, the results of exploring the expression of immune cells at the early stage of treatment showed that 13,14\u0026thinsp;\u0026minus;\u0026thinsp;Dihydro PGF\u0026thinsp;\u0026minus;\u0026thinsp;1a was significantly positively correlated with CD3.Moreover, ECOG and clinical staging are often used in clinical practice for empirical judgment. Relatively speaking, patients with lower ECOG and clinical staging have better conditions and better treatment tolerance. Additionally, in our study, 13,14\u0026thinsp;\u0026minus;\u0026thinsp;Dihydro PGF\u0026thinsp;\u0026minus;\u0026thinsp;1a at baseline was negatively correlated with the potential prognostic characteristics of ECOG and clinical stage. In summary, the arachidonic acid metabolic pathway had significant biological implications and clinical application prospects in the baseline prediction of NAIC treatment.\u003c/p\u003e \u003cp\u003eThe prediction model (21-PM) established based on baseline data performed well in predicting treatment response. However, due to issues such as potential drug resistance during tumor treatment, relying solely on baseline predictions cannot monitor changes in real-time during the treatment process\u003csup\u003e\u003cspan citationid=\"CR79\" class=\"CitationRef\"\u003e79\u003c/span\u003e\u003c/sup\u003e. The study focused on characteristic R-spectific metabolites at key treatment nodes to characterize metabolic changes during therapeutic progression and investigate post-treatment biological alterations associated with treatment response. A more pronounced metabolic reprogramming was observed in R group compared to NR group when comparing pre- versus post-treatment profiles. 156 metabolites specifically altered in the R group highlighted significant disturbances in pathways such as the biosynthesis of valine, leucine, and isoleucine, the metabolism of taurine and hypotaurine, the metabolism of cysteine and methionine, the biosynthesis of arginine, the biosynthesis of primary bile acids, and the citric acid cycle (TCA cycle). Through LASSO regression analysis, the 11 key metabolites further screened based on 156 characteristic metabolites of the R group showed significant potential in distinguishing between the R group and the NR group. Among them, the metabolite 13,14-Dihydro PGF-1α showed predictive potential for treatment response at baseline and exhibited obvious changes at critical points during the treatment process. Metabolic expression data at multiple time points revealed that 13,14-Dihydro PGF-1α in the R group first increased and then decreased with the progression of treatment, and this trend of change suggests that it may be closely related to treatment sensitivity. In the R group, the initial increase of 13,14-Dihydro PGF-1α during treatment may be positively correlated with treatment efficacy, while the subsequent decrease in expression may imply the development of treatment resistance. A significant association was identified between 13,14-dihydro-PGF-1α dynamics and PD-L1 expression levels during treatment. PD-L1 was an important immune checkpoint molecule on the surface of tumor cells, and the level of its expression can affect the interaction between tumor cells and immune cells, thereby influencing tumor immune evasion and the efficacy of immunotherapy\u003csup\u003e\u003cspan citationid=\"CR80\" class=\"CitationRef\"\u003e80\u003c/span\u003e\u003c/sup\u003e. 3,14-Dihydro PGF-1α may be involved in shaping and regulating the tumor immune microenvironment by modulating the expression or function of PD-L1. When further analyzing the data of 13,14-Dihydro PGF-1α and immune cell expression, a significant correlation was found between this metabolite and CD19 in patients of the NR group after treatment. In addition, PGF2b was also a metabolite of particular interest in our study. PGF2b was produced through the COX pathway in arachidonic acid metabolism and participates in regulating the migration and activation status of immune cells, helping tumors to evade immune surveillance\u003csup\u003e\u003cspan citationid=\"CR81\" class=\"CitationRef\"\u003e81\u003c/span\u003e\u003c/sup\u003e. In our study, the decreased expression of PGF2b before and after treatment in the R group may be associated with increased tumor sensitivity to treatment, which indicates that the inhibition of PGF2b helps to enhance the treatment effect. In the tumor microenvironment, the high consumption of Glutamine by tumor cells leads to a decrease in the glutamine level in the microenvironment, thereby inhibiting the function of immune cells (such as T cells, natural killer cells, and macrophages) and promoting tumor immune evasion\u003csup\u003e\u003cspan citationid=\"CR82\" class=\"CitationRef\"\u003e82\u003c/span\u003e\u003c/sup\u003e. This was consistent with our results, where the expression of Glutamine increased before and after treatment, inhibiting tumor cell escape and making patients more likely to respond to treatment.\u003c/p\u003e \u003cp\u003eThe study further evaluated efficacy monitoring for patient cohorts with histopathological outcomes, delving into the accuracy and potential biases of different assessment methods (imaging and histopathology) to provide more precise treatment response predictions for clinical practice. Our analysis showed that compared to the imaging-based 11-MMI, the 13-MMP was more accurate in distinguishing between treatment responders and non-responders (with superior model performance). EPA 13,14-Dihydro PGF-1α, and other prostaglandin-like compounds still had significant predictive ability in analyses with tissue pathology as the outcome. And in the analysis with immune cells showed that in the R group of patients, EPA showed a significant positive correlation with the expression level of CD3 cells after treatment, and the FC value of EPA in patients with high PD-L1 expression (CPS\u0026thinsp;\u0026ge;\u0026thinsp;10) was significantly lower than that in patients with low PD-L1 expression (CPS\u0026thinsp;\u0026lt;\u0026thinsp;10). This suggested that EPA might be negatively correlated with PD-L1 expression levels, implying that EPA might have played an inhibitory role in the TME with high PD-L1 expression, affecting the expression state of CD3 cells, and thereby potentially influencing the patient's treatment response. Furthermore, in the R group of patients, the expression level of EPA significantly increased before and after treatment. Due to the competition between EPA and AA for positions in the cell membrane, the presence of EPA reduced the amount of AA, inhibited the production of PGF2b, and weakened the tumor's ability to evade immune surveillance. This was also consistent with the results from our imaging group.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eOverall, the study comprehensively revealed the reconstructive effect of NAIC on the metabolic profile of GC patients, clarifying four different dynamic change trajectories across treatment nodes, and provided valuable insights into the dynamic metabolic response of GC patients to NAIC. On this basis, a series of models with significant clinical value were constructed: The 21-PM model, which predicted treatment response based on baseline data, included 21 metabolites and their calculated risk scores, holding critical clinical significance for stratifying GC patients before treatment. The 11-MMI treatment response monitoring model, based on imaging, aided in real-time monitoring of efficacy and timely intervention during NAIC treatment. The 13-MMP treatment response monitoring model, based on tissue pathology, revealed biomarkers related to patient efficacy from pre-NAIC to pre-surgery from a more rigorous clinical evaluation perspective. The potential biomarkers for GC NAIC treatment discovered in this study could have been used for early prediction of treatment response and real-time monitoring of treatment efficacy, strongly promoting the development of personalized treatment and leading us towards an era of precision medicine driven by metabolomics.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eStudy population\u003c/h2\u003e \u003cp\u003e The study was approved by the Ethics Committee of The First Affiliated Hospital, Zhejiang University School of Medicine (IIT20230426B-R1), and all subjects provided written informed consent.\u003c/p\u003e \u003cp\u003eThis study recruited GC patients who received NAIC treatment at The First Affiliated Hospital, Zhejiang University School of Medicine from October 30, 2021, to June 30, 2023. All these patients had completed at least one cycle of NAIC before surgery. Inclusion criteria as following: histologically confirmed gastric adenocarcinoma by endoscopic biopsy before the treatment initiation; (2) locally advanced GC, regionally unresectable or clinical stage IV disease according to AJCC/UICC 8th staging system, which was mainly evaluated by computed tomography (CT); (3) patients receiving chemotherapy combined with immune checkpoint inhibitors with or without trastuzumab. The exclusion criteria were as follows: (1) history of another malignancy, except cured basal cell carcinoma of the skin or cured carcinoma in situ of the uterine cervix; (2) prior major stomach surgery; (3) previous systemic treatment or radiotherapy for GC. During the recruitment period of this study, a total of 447 plasma samples were collected from 154 patients diagnosed with GC. After applying the inclusion and exclusion criteria, 369 plasma samples from 108 patients were selected, including 108 preoperative baseline samples and 261 samples collected throughout the treatment process. In addition, clinical information including gender, age, BMI, Lauren classification, tumor location, differentiation status, MSI/MMR status, CPS score, clinical TNM stage, therapy, and treatment response was also collected.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003ePlasma sample collection\u003c/h2\u003e \u003cp\u003ePatients had peripheral venous blood samples (baseline samples) collected before the start of treatment and after each treatment until the patients underwent surgery (including preoperative samples). All sample collection activities were completed by June 30, 2023. After fasting for 12 hours, blood was collected from the patients while they were in a fasting state. Plasma separation was performed according to the following procedure: The collected blood was centrifuged at 1200 rpm for 10 minutes at 4\u0026deg;C. Subsequently, the supernatant was collected and stored at -80\u0026deg;C until metabolite extraction was carried out.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eMetabolite extraction\u003c/h2\u003e \u003cp\u003eThe procedure for metabolite extraction from plasma samples was as follows: First, plasma samples were thawed on ice. Then, 100 \u0026micro;L of plasma was mixed with 400 \u0026micro;L of pre-chilled methanol. The mixture was vortexed for 3 minutes and then centrifuged at 4\u0026deg;C and 13,000 rpm for 15 minutes. The supernatant was collected and evaporated using a centrifugal vacuum evaporator at 4\u0026deg;C. The dried metabolite samples were stored at -80\u0026deg;C until they were subjected to liquid chromatography-tandem mass spectrometry (LC-MS/MS) analysis. Quality Control (QC) samples, which were prepared by taking 6 \u0026micro;L from each plasma sample, were treated in the same manner as the study plasma samples.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003ePlasma metabolites detection\u003c/h2\u003e \u003cp\u003eUsing UHPLC-Q-Orbitrap-MS/MS technology, a comprehensive metabolomics analysis was carried out on 369 serum samples from 108 gastric cancer patients. First, 300 \u0026micro;L of the supernatant was taken, and 50 \u0026micro;L of the working solution containing internal standards (sulfamethoxypyridazine and ketoprofen) was added. Then, the mixture was vortexed for 30 seconds, followed by ultrasonic treatment on ice. After that, it was centrifuged at 4\u0026deg;C and 13,000 rpm for 15 minutes. Finally, 30 \u0026micro;L of the supernatant was taken for liquid chromatography-tandem mass spectrometry (LC-MS/MS) analysis.\u003c/p\u003e \u003cp\u003eMass spectrometry conditions: The liquid chromatography detection method mainly referred to the method developed in the literature. The chromatographic column used was ACQUITY UPLC HSS T3 (2.1\u003cem\u003e50 mm\u003c/em\u003e1.7\u0026micro;m, Waters Corporation). The column temperature was maintained at 40\u0026deg;C. The mobile phase ratio consisted of A (0.1% formic acid in water) and B (acetonitrile). The flow rate was set at 0.35 mL/min, and the injection volume was 5 \u0026micro;L. The elution gradient is shown in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2.4\u003c/span\u003e below:\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2.4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eLiquid Chromatography Gradient Elution Program Table gradient flow of LC solvent\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTime\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSolvent A(%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSolvent B(%)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e95\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e80\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e20\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e60\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e40\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e40\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e60\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e16\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e35\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e65\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e18\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e20\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e80\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e21\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e95\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e23\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e95\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e25\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e95\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e5\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eOrbitrap Exploris 120 mass spectrometer (Thermo Fisher Scientific Inc.), with the following parameter settings: spray voltage, -2500V;Sheath gas(Arb), 50༛Aux gas༈Arb༉, 10༛Sweep gas༈Arb༉, 1༛Ion transfer tube temperature, 300℃༛Vaporizer temperature, 350℃༛orbitrap resolution, 60000, scan range, 200\u0026ndash;1500༛Dynamic exclusion on༛\u003c/p\u003e \u003c/div\u003e\u003ch2\u003eMetabolomics data processing\u003c/h2\u003e\n\u003cp\u003eTo eliminate the noise and instrument fluctuation interference during the operation of the instrument, a QC sample was added during the plasma sample LC-MS/MS detection process, that is, a QC sample was inserted after an equal number of samples. Ultimately, the expression and coefficient of variation of metabolites in the QC samples were used as the basis to eliminate some of the interference caused by systematic noise. Afterwards, the Compound Discover version 3.2 (Thermo Fisher Scientific Inc.) software was used to analyze the data from the aforementioned plasma samples, and the metabolites were structurally identified using the mzCloud (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ewww.mzcloud.org/\u003c/span\u003e\u003c/span\u003e), mzVault (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://mytracefinder.com/tag/mzvault/\u003c/span\u003e\u003c/span\u003e), KEGG (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ewww.kegg.jp/\u003c/span\u003e\u003c/span\u003e), HMDB (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://hmdb.ca/\u003c/span\u003e\u003c/span\u003e), and ChemSpider (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ewww.chemspider.com\u003c/span\u003e\u003c/span\u003e) databases.\u003c/p\u003e\n\u003cp\u003eUsing the MetNormalizer R package (version 1.3.02) with the support vector regression algorithm, potential loading/dispensing errors and batch effects were adjusted for\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e. After the features with RSD\u0026thinsp;\u0026ge;\u0026thinsp;30% in the QC samples had been excluded, 71% (5787/8167) of the positive ion mode features and 88% (3512/3980) of the negative ion mode features were retained. After annotation, 1001 metabolites remained. The raw data were log-transformed with base 2, and Z-scale normalization of metabolite features was performed before data analysis.\u003c/p\u003e\n\u003ch2\u003eMfuzz clustering analysis\u003c/h2\u003e\n\u003cp\u003eCollected metabolic data from 369 plasma samples of 108 participants, standardized the average values of metabolites at six different time points as input data, and applied the Mfuzz fuzzy c-means algorithm (Mfuzz, R version 2.62.0) to cluster the metabolites and depict their dynamic changes.\u003c/p\u003e\n\u003ch2\u003ePediction and monitoring model for the therapy response\u003c/h2\u003e\n\u003cp\u003eAccording to the outcome indicators, the cohort was randomly divided into a training set and a validation set in an 8:2 ratio, ensuring that the distribution of key clinical features in the training set and the validation set was similar. Before model construction, the SMOTE (DMwR, R, version 0.4.1) algorithm was used to address the data imbalance issue. Subsequently, a Least Absolute Shrinkage and Selection Operator (LASSO) regression analysis (glmnet, R, version 4.1-8) was used to construct the model. Based on 10-fold cross-validation to avoid overfitting, the parameters of the LASSO regression were finally determined, non-zero coefficient metabolites were selected, and the model was constructed. Subsequently, the metabolite feature information of plasma samples from the test set patients was applied to validate the established prediction model. The entire LASSO regression study was applied in three areas: 1. Efficacy outcome prediction model based on baseline samples; 2. Patient efficacy monitoring model based on imaging diagnosis during the course of treatment; 3. Patient efficacy outcome monitoring model based on pathological diagnosis information. Using ROC analysis (pROC package, R version 1.18.5), the predictive value was comprehensively assessed based on sensitivity, specificity, accuracy, and Area Under Curve (AUC).\u003c/p\u003e\n\u003cp\u003e1. Efficacy outcome prediction model based on baseline samples: First, based on the type of treatment response, 108 subjects were randomly divided into a training dataset (n\u0026thinsp;=\u0026thinsp;86, 79.63%) and a test dataset (n\u0026thinsp;=\u0026thinsp;22, 20.37%). The distribution of key clinical features in the training and validation sets was similar.\u003c/p\u003e\n\u003cp\u003eAfter the SMOTE algorithm addressed the data imbalance issue, the differential metabolites (DMs) between R the NR were used as input data. The parameters for the LASSO regression were determined through 10-fold cross-validation, selecting the 1se model with a lambda value of 0.03574295 to construct the model. Subsequently, the metabolite characteristic information of plasma samples from patients in the test set was applied to validate the established prediction model. The clinical characteristics significantly related to patient treatment outcomes were explored through intergroup difference analysis, and the clinical N stage was identified as this differential feature. Following that, the predictive abilities of this factor and its impact on the model were thoroughly analyzed, with the AUC value serving as the primary metric for measuring predictive performance. The examination of the model\u0026apos;s effect ability proceeded as follows: First, 4 N0 patients were excluded from the training set to evaluate the model\u0026apos;s prediction accuracy. Following that, 1000 iterative experiments were carried out, each including the random removal of four patients and simultaneous performance assessment, to explore the influence of reduced sample size on predictive capability.\u003c/p\u003e\n\u003cp\u003eIn order to assess the benefit-risk ratio for patients after treatment, risk scores\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e. were calculated based on the metabolite coefficients in the model, and patients were classified into High-Risk and Low-Risk groups based on the median of the risk scores. To determine whether the Risk Score could independently affect the outcome, clinical factors with a baseline intergroup difference \u003cem\u003eP\u003c/em\u003e-value\u0026thinsp;\u0026lt;\u0026thinsp;0.1 were also included in the multivariate logistic regression model to comprehensively assess the relationship between various influencing factors and the outcome.\u003c/p\u003e\n\u003cp\u003e\u003cspan\u003e2. Patient efficacy monitoring model based on imaging diagnosis during the treatment process: First, paired difference analysis was conducted at baseline (T0) and key stages (Tx) for both the R and the NR, screening for metabolites that showed significant changes only in the R (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05) and metabolites that exhibited opposite trends in the R compared to the NR. This allowed the construction of a characteristic metabolite set for the R. Subsequently, based on the type of treatment response, 85 subjects were randomly divided into a training dataset (n\u0026thinsp;=\u0026thinsp;68, 80.00%) and a test dataset (n\u0026thinsp;=\u0026thinsp;17, 20.00%). The distribution of key clinical features in the training and validation sets was similar. After the SMOTE algorithm addressed the data imbalance issue, the characteristic metabolites of R were used as input data. The parameters for LASSO regression were determined through 10-fold cross-validation, selecting the min model with a lambda value of 0.08584942 to construct the model. Then, the metabolite feature information from the plasma samples of the test set patients was used to validate the established monitoring model.\u003cbr\u003e\u003c/span\u003e\u003cspan\u003e3. Patient efficacy outcome monitoring model based on pathological diagnosis information: Constructed a characteristic metabolite set for R with tissue pathology as the outcome using the same method. Subsequently, based on the type of treatment response, 42 subjects were randomly divided into a training dataset (n\u0026thinsp;=\u0026thinsp;34, 80.95%) and a test dataset (n\u0026thinsp;=\u0026thinsp;8, 19.05%), ensuring a similar distribution of clinical characteristics. After that, using the characteristic metabolites of R as input data, the parameters of the LASSO regression were determined through 10-fold cross-validation, selecting the min model with a lambda value of 0.07801261. Metabolites with non-zero coefficients were screened to establish the monitoring model. Then, the metabolite characteristic information from the plasma samples of the test set patients was applied to validate the established monitoring model.\u003cbr\u003e\u003c/span\u003e\u003c/p\u003e\n\u003ch2\u003eStatistical analyses\u003c/h2\u003e\n\u003cp\u003eStandard statistical methods were used to conduct statistical analysis on the clinical characteristic information and metabolite characteristics collected in the study. The normality of data distribution was assessed based on the Shapiro-Wilk test. Due to the non-normal distribution of most metabolite characteristics, statistical analysis of metabolite characteristics was conducted using non-parametric tests: the Wilcoxon rank-sum test was used for pairwise difference analysis, while the Kruskal-Wallis (K-W) test was used for multi-group difference analysis. For immune indicator information (CD3, CD4, CD8, etc.), parametric tests such as the T-test were used; if the data did not follow a normal distribution, non-parametric analysis such as the Wilcoxon rank-sum test was employed. For clinical categorical variable data (age, gender, ECOG performance status score, Lauren classification, etc.), the chi-square test or Fisher\u0026apos;s exact test was used. The correlation coefficient was analyzed using the Spearman method. For all analyses, a two-sided \u003cem\u003eP\u003c/em\u003e value of less than 0.05 was considered statistically significant. All data analyses were performed using R statistical software (version 4.3.3).\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was supported by the Zhejiang Province Traditional Chinese Medicine Science and Technology Project (No.GZY-ZJ-KJ-23008), Key R\u0026amp;D Program of Zhejiang Province (No.2025C02171), Zhejiang Province Traditional Chinese Medicine Key Laboratory Project (No.GZY-ZJ-SY-2303), National Natural Science Foundation of China (No. 81972671) and Joint TCM Science \u0026amp;Technology Projects of National Demonstration Zones for Comprehensive TCM Reform (GZY-KJS-ZJ-2025-036). The authors acknowledge the support from the Shared Instrumentation Core Facility at the Hangzhou Institute of Medicine (HIM), Chinese Academy of Sciences.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eZ.X.Y., H.X.L., Z.K.K., Z.W. and L.X.S. conceived the idea and designed the study; Z.X.Y., H.X.L and G.S.H. performed sample preparation and data acquisition; Z.X.Y., H.X.L. contributed to metabolite identification; Z.X.Y., H.X.L. performed data analysis with help of L.Y., W.X.D, D.Y, F.Y., G.T.Z., M.J.H. and L.C.Z.; Z.K.K, Z.H.D., J.H.L. and L.X.S, contributed to the clinical trial of this study, including patient recruitment, diagnosis, follow-up, and serum sample collection; Z.X.Y., H.X.L., and Z.K.K., wrote the paper; all authors read and approved the manuscript; L.X.S. led the clinical trial; L.X.S., T.J.K. and Z.W. supervised the project.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eLiu, D. \u003cem\u003eet al.\u003c/em\u003e The patterns and timing of recurrence after curative resection for gastric cancer in China. \u003cem\u003eWorld J Surg Oncol\u003c/em\u003e \u003cstrong\u003e14\u003c/strong\u003e, 305 (2016).\u003c/li\u003e\n\u003cli\u003eJiang, Y. \u003cem\u003eet al.\u003c/em\u003e Association of adjuvant chemotherapy with survival in patients with stage II or III gastric cancer. \u003cem\u003eJAMA Surg\u003c/em\u003e \u003cstrong\u003e152\u003c/strong\u003e, e171087 (2017).\u003c/li\u003e\n\u003cli\u003eCatenacci, D. V. \u003cem\u003eet al.\u003c/em\u003e Toward a treatment sequencing strategy: A systematic review of treatment regimens in advanced gastric cancer/gastroesophageal junction adenocarcinoma. \u003cem\u003eOncologist\u003c/em\u003e \u003cstrong\u003e26\u003c/strong\u003e, e1704\u0026ndash;e1729 (2021).\u003c/li\u003e\n\u003cli\u003eShapiro, J. \u003cem\u003eet al.\u003c/em\u003e Neoadjuvant chemoradiotherapy plus surgery versus surgery alone for oesophageal or junctional cancer (CROSS): Long-term results of a randomised controlled trial. \u003cem\u003eLancet Oncol\u003c/em\u003e \u003cstrong\u003e16\u003c/strong\u003e, 1090\u0026ndash;1098 (2015).\u003c/li\u003e\n\u003cli\u003eAl-Batran, S.-E. \u003cem\u003eet al.\u003c/em\u003e Perioperative chemotherapy with fluorouracil plus leucovorin, oxaliplatin, and docetaxel versus fluorouracil or capecitabine plus cisplatin and epirubicin for locally advanced, resectable gastric or gastro-oesophageal junction adenocarcinoma (FLOT4): A randomised, phase 2/3 trial. \u003cem\u003eLancet\u003c/em\u003e \u003cstrong\u003e393\u003c/strong\u003e, 1948\u0026ndash;1957 (2019).\u003c/li\u003e\n\u003cli\u003eCunningham, D. \u003cem\u003eet al.\u003c/em\u003e Perioperative chemotherapy versus surgery alone for resectable gastroesophageal cancer. \u003cem\u003eN Engl J Med\u003c/em\u003e \u003cstrong\u003e355\u003c/strong\u003e, 11\u0026ndash;20 (2006).\u003c/li\u003e\n\u003cli\u003eLiu, T. \u003cem\u003eet al.\u003c/em\u003e First-line nivolumab plus chemotherapy vs chemotherapy in patients with advanced gastric, gastroesophageal junction and esophageal adenocarcinoma: CheckMate 649 Chinese subgroup analysis. \u003cem\u003eInt J Cancer\u003c/em\u003e \u003cstrong\u003e152\u003c/strong\u003e, 749\u0026ndash;760 (2023).\u003c/li\u003e\n\u003cli\u003eJanjigian, Y. Y. \u003cem\u003eet al.\u003c/em\u003e First-line nivolumab plus chemotherapy versus chemotherapy alone for advanced gastric, gastro-oesophageal junction, and oesophageal adenocarcinoma (CheckMate 649): a randomised, open-label, phase 3 trial. \u003cem\u003eLancet (London, England)\u003c/em\u003e \u003cstrong\u003e398\u003c/strong\u003e, 27\u0026ndash;40 (2021).\u003c/li\u003e\n\u003cli\u003eWang, L. \u003cem\u003eet al.\u003c/em\u003e Intestinal subtype as a biomarker of response to neoadjuvant immunochemotherapy in locally advanced gastric adenocarcinoma: Insights from a prospective phase II trial. \u003cem\u003eClin Cancer Res\u003c/em\u003e \u003cstrong\u003e31\u003c/strong\u003e, 74\u0026ndash;86 (2025).\u003c/li\u003e\n\u003cli\u003eYl, V. \u003cem\u003eet al.\u003c/em\u003e Neoadjuvant atezolizumab plus chemotherapy in gastric and gastroesophageal junction adenocarcinoma: the phase 2 PANDA trial. \u003cem\u003eNature medicine\u003c/em\u003e \u003cstrong\u003e30\u003c/strong\u003e, (2024).\u003c/li\u003e\n\u003cli\u003eWang, L. \u003cem\u003eet al.\u003c/em\u003e Intestinal Subtype as a Biomarker of Response to Neoadjuvant Immunochemotherapy in Locally Advanced Gastric Adenocarcinoma: Insights from a Prospective Phase II Trial. \u003cem\u003eClinical Cancer Research\u003c/em\u003e OF1\u0026ndash;OF13 (2024) doi:10.1158/1078-0432.CCR-24-2436.\u003c/li\u003e\n\u003cli\u003eChen, D. \u003cem\u003eet al.\u003c/em\u003e Prognostic and predictive value of a pathomics signature in gastric cancer. \u003cem\u003eNat Commun\u003c/em\u003e \u003cstrong\u003e13\u003c/strong\u003e, 6903 (2022).\u003c/li\u003e\n\u003cli\u003eLudwig, J. A. \u0026amp; Weinstein, J. N. Biomarkers in cancer staging, prognosis and treatment selection. \u003cem\u003eNat Rev Cancer\u003c/em\u003e \u003cstrong\u003e5\u003c/strong\u003e, 845\u0026ndash;856 (2005).\u003c/li\u003e\n\u003cli\u003eSmith, J. C. \u0026amp; Sheltzer, J. M. Genome-wide identification and analysis of prognostic features in human cancers. \u003cem\u003eCell Rep\u003c/em\u003e \u003cstrong\u003e38\u003c/strong\u003e, 110569 (2022).\u003c/li\u003e\n\u003cli\u003eSuzuki, T., Kitagawa, Y., Nankinzan, R. \u0026amp; Yamaguchi, T. Early gastric cancer diagnostic ability of ultrathin endoscope loaded with laser light source. \u003cem\u003eWorld J Gastroenterol\u003c/em\u003e \u003cstrong\u003e25\u003c/strong\u003e, 1378\u0026ndash;1386 (2019).\u003c/li\u003e\n\u003cli\u003eDe Martino, M., Rathmell, J. C., Galluzzi, L. \u0026amp; Vanpouille-Box, C. Cancer cell metabolism and antitumour immunity. \u003cem\u003eNat Rev Immunol\u003c/em\u003e \u003cstrong\u003e24\u003c/strong\u003e, 654\u0026ndash;669 (2024).\u003c/li\u003e\n\u003cli\u003eChen, Y. \u003cem\u003eet al.\u003c/em\u003e Metabolomic machine learning predictor for diagnosis and prognosis of gastric cancer. \u003cem\u003eNat. Commun.\u003c/em\u003e \u003cstrong\u003e15\u003c/strong\u003e, 1657 (2024).\u003c/li\u003e\n\u003cli\u003eShen, X. \u003cem\u003eet al.\u003c/em\u003e Normalization and integration of large-scale metabolomics data using support vector regression. \u003cem\u003eMetabolomics\u003c/em\u003e \u003cstrong\u003e12\u003c/strong\u003e, 89 (2016).\u003c/li\u003e\n\u003cli\u003eTang Z. \u003cem\u003eet al.\u003c/em\u003e Multiplex immune profiling reveals the role of serum immune proteomics in predicting response to preoperative chemotherapy of gastric cancer. \u003cem\u003eCell Reports. Medicine\u003c/em\u003e \u003cstrong\u003e4\u003c/strong\u003e, 100931 (2023).\u003c/li\u003e\n\u003cli\u003eR, H. \u0026amp; M, C. Complications of diagnostic gastrointestinal endoscopy. \u003cem\u003eEndoscopy\u003c/em\u003e \u003cstrong\u003e22\u003c/strong\u003e, (1990).\u003c/li\u003e\n\u003cli\u003eChoi, S. J. \u003cem\u003eet al.\u003c/em\u003e Gastric Cancer and Intestinal Metaplasia: Differential Metabolic Landscapes and New Pathways to Diagnosis. \u003cem\u003eInternational Journal of Molecular Sciences\u003c/em\u003e \u003cstrong\u003e25\u003c/strong\u003e, 9509 (2024).\u003c/li\u003e\n\u003cli\u003eSharma, P. \u0026amp; Allison, J. P. The future of immune checkpoint therapy. \u003cem\u003eScience\u003c/em\u003e \u003cstrong\u003e348\u003c/strong\u003e, 56\u0026ndash;61 (2015).\u003c/li\u003e\n\u003cli\u003eLiu, J. \u003cem\u003eet al.\u003c/em\u003e Improved Efficacy of Neoadjuvant Compared to Adjuvant Immunotherapy to Eradicate Metastatic Disease. \u003cem\u003eCancer Discov\u003c/em\u003e \u003cstrong\u003e6\u003c/strong\u003e, 1382\u0026ndash;1399 (2016).\u003c/li\u003e\n\u003cli\u003eJiang, H. \u003cem\u003eet al.\u003c/em\u003e Efficacy and safety of neoadjuvant sintilimab, oxaliplatin and capecitabine in patients with locally advanced, resectable gastric or gastroesophageal junction adenocarcinoma: early results of a phase 2 study. \u003cem\u003eJ Immunother Cancer\u003c/em\u003e \u003cstrong\u003e10\u003c/strong\u003e, e003635 (2022).\u003c/li\u003e\n\u003cli\u003eKim, H.-D., Ryu, M.-H., Park, Y. S., Lee, J. S. \u0026amp; Kang, Y.-K. A phase 2 study of durvalumab plus DOS (docetaxel, oxaliplatin, S-1) neoadjuvant chemotherapy followed by surgery and adjuvant durvalumab plus S-1 chemotherapy in patients with resectable locally advanced gastric cancer: An interim efficacy report. \u003cem\u003eJCO\u003c/em\u003e \u003cstrong\u003e41\u003c/strong\u003e, 408\u0026ndash;408 (2023).\u003c/li\u003e\n\u003cli\u003eXu, J. \u003cem\u003eet al.\u003c/em\u003e Sintilimab Plus Chemotherapy for Unresectable Gastric or Gastroesophageal Junction Cancer. \u003cem\u003eJAMA\u003c/em\u003e \u003cstrong\u003e330\u003c/strong\u003e, 2064\u0026ndash;2074 (2023).\u003c/li\u003e\n\u003cli\u003eLi, N. \u003cem\u003eet al.\u003c/em\u003e Efficacy and safety of neoadjuvant sintilimab in combination with FLOT chemotherapy in patients with HER2-negative locally advanced gastric or gastroesophageal junction adenocarcinoma: An investigator-initiated, single-arm, open-label, phase II study. \u003cem\u003eInt J Surg\u003c/em\u003e \u003cstrong\u003e110\u003c/strong\u003e, 2071\u0026ndash;2084 (2024).\u003c/li\u003e\n\u003cli\u003eLiu, Y. \u003cem\u003eet al.\u003c/em\u003e Camrelizumab combined with FLOFOX as neoadjuvant therapy for resectable locally advanced gastric and gastroesophageal junction adenocarcinoma: Updated results of efficacy and safety. \u003cem\u003eJCO\u003c/em\u003e \u003cstrong\u003e39\u003c/strong\u003e, 4036\u0026ndash;4036 (2021).\u003c/li\u003e\n\u003cli\u003eZhong, W. \u003cem\u003eet al.\u003c/em\u003e Efficacy and safety of camrelizumab combined with oxaliplatin and S‐1 as neoadjuvant treatment in locally advanced gastric or gastroesophageal junction cancer: A phase II, single‐arm study. \u003cem\u003eCancer Med\u003c/em\u003e \u003cstrong\u003e13\u003c/strong\u003e, e7006 (2024).\u003c/li\u003e\n\u003cli\u003eWaldman, A. D., Fritz, J. M. \u0026amp; Lenardo, M. J. A guide to cancer immunotherapy: from T cell basic science to clinical practice. \u003cem\u003eNat Rev Immunol\u003c/em\u003e \u003cstrong\u003e20\u003c/strong\u003e, 651\u0026ndash;668 (2020).\u003c/li\u003e\n\u003cli\u003eNielsen, D. L., Juhl, C. B., Chen, I. M., Kellermann, L. \u0026amp; Nielsen, O. H. Immune checkpoint Inhibitor-Induced diarrhea and Colitis: Incidence and Management. A systematic review and Meta-analysis. \u003cem\u003eCancer Treat. Rev.\u003c/em\u003e \u003cstrong\u003e109\u003c/strong\u003e, 102440 (2022).\u003c/li\u003e\n\u003cli\u003eJunttila, M. R. \u0026amp; de Sauvage, F. J. Influence of tumour micro-environment heterogeneity on therapeutic response. \u003cem\u003eNature\u003c/em\u003e \u003cstrong\u003e501\u003c/strong\u003e, 346\u0026ndash;354 (2013).\u003c/li\u003e\n\u003cli\u003eBejarano, L., Jordāo, M. J. C. \u0026amp; Joyce, J. A. Therapeutic Targeting of the Tumor Microenvironment. \u003cem\u003eCancer Discov\u003c/em\u003e \u003cstrong\u003e11\u003c/strong\u003e, 933\u0026ndash;959 (2021).\u003c/li\u003e\n\u003cli\u003eTsai, C.-H. \u003cem\u003eet al.\u003c/em\u003e Immunoediting instructs tumor metabolic reprogramming to support immune evasion. \u003cem\u003eCell Metabolism\u003c/em\u003e \u003cstrong\u003e35\u003c/strong\u003e, 118-133.e7 (2023).\u003c/li\u003e\n\u003cli\u003eFaubert, B., Solmonson, A. \u0026amp; DeBerardinis, R. J. Metabolic reprogramming and cancer progression. \u003cem\u003eScience\u003c/em\u003e \u003cstrong\u003e368\u003c/strong\u003e, eaaw5473 (2020).\u003c/li\u003e\n\u003cli\u003ePavlova, N. N. \u0026amp; Thompson, C. B. The Emerging Hallmarks of Cancer Metabolism. \u003cem\u003eCell Metab\u003c/em\u003e \u003cstrong\u003e23\u003c/strong\u003e, 27\u0026ndash;47 (2016).\u003c/li\u003e\n\u003cli\u003eLiu, X. \u003cem\u003eet al.\u003c/em\u003e The significant role of amino acid metabolic reprogramming in cancer. \u003cem\u003eCell Commun Signal\u003c/em\u003e \u003cstrong\u003e22\u003c/strong\u003e, 1\u0026ndash;26 (2024).\u003c/li\u003e\n\u003cli\u003eWang, D. \u0026amp; Wan, X. Progress in research on the role of amino acid metabolic reprogramming in tumour therapy: A review. \u003cem\u003eBiomedicine \u0026amp; Pharmacotherapy\u003c/em\u003e \u003cstrong\u003e156\u003c/strong\u003e, 113923 (2022).\u003c/li\u003e\n\u003cli\u003eSiska, P. J. \u0026amp; Rathmell, J. C. T Cell Metabolic Fitness in Anti-Tumor Immunity. \u003cem\u003eTrends Immunol\u003c/em\u003e \u003cstrong\u003e36\u003c/strong\u003e, 257\u0026ndash;264 (2015).\u003c/li\u003e\n\u003cli\u003eChen, C.-L., Hsu, S.-C., Ann, D. K., Yen, Y. \u0026amp; Kung, H.-J. Arginine Signaling and Cancer Metabolism. \u003cem\u003eCancers (Basel)\u003c/em\u003e \u003cstrong\u003e13\u003c/strong\u003e, 3541 (2021).\u003c/li\u003e\n\u003cli\u003eHolbert, C. E., Cullen, M. T., Casero, R. A. \u0026amp; Stewart, T. M. Polyamines in cancer: integrating organismal metabolism and antitumour immunity. \u003cem\u003eNat Rev Cancer\u003c/em\u003e \u003cstrong\u003e22\u003c/strong\u003e, 467\u0026ndash;480 (2022).\u003c/li\u003e\n\u003cli\u003ePeyraud, F. \u003cem\u003eet al.\u003c/em\u003e Circulating L-arginine predicts the survival of cancer patients treated with immune checkpoint inhibitors. \u003cem\u003eAnn Oncol\u003c/em\u003e \u003cstrong\u003e33\u003c/strong\u003e, 1041\u0026ndash;1051 (2022).\u003c/li\u003e\n\u003cli\u003eMossmann, D. \u003cem\u003eet al.\u003c/em\u003e Arginine reprograms metabolism in liver cancer via RBM39. \u003cem\u003eCell\u003c/em\u003e \u003cstrong\u003e186\u003c/strong\u003e, 5068-5083.e23 (2023).\u003c/li\u003e\n\u003cli\u003eCan\u0026egrave;, S., Geiger, R. \u0026amp; Bronte, V. The roles of arginases and arginine in immunity. \u003cem\u003eNat Rev Immunol\u003c/em\u003e 1\u0026ndash;19 (2024) doi:10.1038/s41577-024-01098-2.\u003c/li\u003e\n\u003cli\u003eYan, J. \u003cem\u003eet al.\u003c/em\u003e Molecular mechanisms and therapeutic significance of tryptophan metabolism and signaling in cancer. \u003cem\u003eMolecular Cancer\u003c/em\u003e \u003cstrong\u003e23\u003c/strong\u003e, 241 (2024).\u003c/li\u003e\n\u003cli\u003eFrezza, C. Histidine degradation boosts cancer therapy. \u003cem\u003eNature\u003c/em\u003e \u003cstrong\u003e559\u003c/strong\u003e, 484 (2018).\u003c/li\u003e\n\u003cli\u003eJ, W., W, W., F, Z. \u0026amp; Q, D. The role of branched chain amino acids metabolic disorders in tumorigenesis and progression. \u003cem\u003eBiomedicine \u0026amp; pharmacotherapy = Biomedecine \u0026amp; pharmacotherapie\u003c/em\u003e \u003cstrong\u003e153\u003c/strong\u003e, (2022).\u003c/li\u003e\n\u003cli\u003eZhenyukh, O. \u003cem\u003eet al.\u003c/em\u003e High concentration of branched-chain amino acids promotes oxidative stress, inflammation and migration of human peripheral blood mononuclear cells via mTORC1 activation. \u003cem\u003eFree Radic Biol Med\u003c/em\u003e \u003cstrong\u003e104\u003c/strong\u003e, 165\u0026ndash;177 (2017).\u003c/li\u003e\n\u003cli\u003eT\u0026ouml;njes, M. \u003cem\u003eet al.\u003c/em\u003e BCAT1 promotes cell proliferation through amino acid catabolism in gliomas carrying wild-type IDH1. \u003cem\u003eNat Med\u003c/em\u003e \u003cstrong\u003e19\u003c/strong\u003e, 901\u0026ndash;908 (2013).\u003c/li\u003e\n\u003cli\u003eLi, J.-T. \u003cem\u003eet al.\u003c/em\u003e BCAT2-mediated BCAA catabolism is critical for development of pancreatic ductal adenocarcinoma. \u003cem\u003eNat Cell Biol\u003c/em\u003e \u003cstrong\u003e22\u003c/strong\u003e, 167\u0026ndash;174 (2020).\u003c/li\u003e\n\u003cli\u003eMayers, J. R. \u003cem\u003eet al.\u003c/em\u003e Tissue of origin dictates branched-chain amino acid metabolism in mutant Kras-driven cancers. \u003cem\u003eScience\u003c/em\u003e \u003cstrong\u003e353\u003c/strong\u003e, 1161\u0026ndash;1165 (2016).\u003c/li\u003e\n\u003cli\u003eNazemi, M. \u003cem\u003eet al.\u003c/em\u003e The extracellular matrix supports breast cancer cell growth under amino acid starvation by promoting tyrosine catabolism. \u003cem\u003ePLoS Biol\u003c/em\u003e \u003cstrong\u003e22\u003c/strong\u003e, e3002406 (2024).\u003c/li\u003e\n\u003cli\u003eLi, J. \u003cem\u003eet al.\u003c/em\u003e Tyrosine catabolism enhances genotoxic chemotherapy by suppressing translesion DNA synthesis in epithelial ovarian cancer. \u003cem\u003eCell Metabolism\u003c/em\u003e \u003cstrong\u003e35\u003c/strong\u003e, 2044-2059.e8 (2023).\u003c/li\u003e\n\u003cli\u003eZhang, L., Yao, Y. \u0026amp; Liu, S. Targeting fatty acid metabolism for cancer therapy. \u003cem\u003eFundamental Research\u003c/em\u003e (2024) doi:10.1016/j.fmre.2024.09.011.\u003c/li\u003e\n\u003cli\u003eTan, Z. \u003cem\u003eet al.\u003c/em\u003e Targeting CPT1A-mediated fatty acid oxidation sensitizes nasopharyngeal carcinoma to radiation therapy. \u003cem\u003eTheranostics\u003c/em\u003e \u003cstrong\u003e8\u003c/strong\u003e, 2329\u0026ndash;2347 (2018).\u003c/li\u003e\n\u003cli\u003eLin, R. \u003cem\u003eet al.\u003c/em\u003e Fatty Acid Oxidation Controls CD8+ Tissue-Resident Memory T-cell Survival in Gastric Adenocarcinoma. \u003cem\u003eCancer Immunol Res\u003c/em\u003e \u003cstrong\u003e8\u003c/strong\u003e, 479\u0026ndash;492 (2020).\u003c/li\u003e\n\u003cli\u003eLee, J.-Y. \u003cem\u003eet al.\u003c/em\u003e Polyunsaturated fatty acid biosynthesis pathway determines ferroptosis sensitivity in gastric cancer. \u003cem\u003eProceedings of the National Academy of Sciences\u003c/em\u003e \u003cstrong\u003e117\u003c/strong\u003e, 32433\u0026ndash;32442 (2020).\u003c/li\u003e\n\u003cli\u003eHanna, V. S. \u0026amp; Hafez, E. A. A. Synopsis of arachidonic acid metabolism: A review. \u003cem\u003eJournal of Advanced Research\u003c/em\u003e \u003cstrong\u003e11\u003c/strong\u003e, 23\u0026ndash;32 (2018).\u003c/li\u003e\n\u003cli\u003eHe, B. \u003cem\u003eet al.\u003c/em\u003e Identification of cytochrome \u003cem\u003eb\u003c/em\u003e5 CYTB-5.1 and CYTB-5.2 in \u003cem\u003eC\u003c/em\u003e. \u003cem\u003eelegans\u003c/em\u003e; evidence for differential regulation of SCD. \u003cem\u003eBiochimica et Biophysica Acta (BBA) - Molecular and Cell Biology of Lipids\u003c/em\u003e \u003cstrong\u003e1863\u003c/strong\u003e, 235\u0026ndash;246 (2018).\u003c/li\u003e\n\u003cli\u003eWang, B. \u003cem\u003eet al.\u003c/em\u003e Metabolism pathways of arachidonic acids: mechanisms and potential therapeutic targets. \u003cem\u003eSignal Transduction and Targeted Therapy\u003c/em\u003e \u003cstrong\u003e6\u003c/strong\u003e, 1\u0026ndash;30 (2021).\u003c/li\u003e\n\u003cli\u003eCui L. \u003cem\u003eet al.\u003c/em\u003e Targeting arachidonic acid metabolism enhances immunotherapy efficacy in ARID1A-deficient colorectal cancer. \u003cem\u003eCancer Research\u003c/em\u003e (2024) doi:10.1158/0008-5472.CAN-24-1611.\u003c/li\u003e\n\u003cli\u003eChen, X. \u003cem\u003eet al.\u003c/em\u003e CYP4F2-Catalyzed Metabolism of Arachidonic Acid Promotes Stromal Cell-Mediated Immunosuppression in Non\u0026ndash;Small Cell Lung Cancer. \u003cem\u003eCancer Research\u003c/em\u003e \u003cstrong\u003e82\u003c/strong\u003e, 4016\u0026ndash;4030 (2022).\u003c/li\u003e\n\u003cli\u003eFan, X. \u003cem\u003eet al.\u003c/em\u003e TRIM21 promotes tumor growth and gemcitabine resistance in pancreatic cancer by inhibiting EPHX1-mediated arachidonic acid metabolism. \u003cem\u003eAdvanced Science\u003c/em\u003e \u003cstrong\u003en/a\u003c/strong\u003e, 2413674.\u003c/li\u003e\n\u003cli\u003eLi Y. \u003cem\u003eet al.\u003c/em\u003e Longitudinal plasma proteome profiling reveals the diversity of biomarkers for diagnosis and cetuximab therapy response of colorectal cancer. \u003cem\u003eNat Commun\u003c/em\u003e \u003cstrong\u003e15\u003c/strong\u003e, 980 (2024).\u003c/li\u003e\n\u003cli\u003eHarel, M. \u003cem\u003eet al.\u003c/em\u003e Longitudinal plasma proteomic profiling of patients with non-small cell lung cancer undergoing immune checkpoint blockade. \u003cem\u003eJ. Immunother. Cancer\u003c/em\u003e \u003cstrong\u003e10\u003c/strong\u003e, e004582 (2022).\u003c/li\u003e\n\u003cli\u003eQian, L. \u003cem\u003eet al.\u003c/em\u003e Proteomic landscape of epithelial ovarian cancer. \u003cem\u003eNat. Commun.\u003c/em\u003e \u003cstrong\u003e15\u003c/strong\u003e, 6462 (2024).\u003c/li\u003e\n\u003cli\u003eLi, Y. \u003cem\u003eet al.\u003c/em\u003e Longitudinal plasma proteome profiling reveals the diversity of biomarkers for diagnosis and cetuximab therapy response of colorectal cancer. \u003cem\u003eNat. Commun.\u003c/em\u003e \u003cstrong\u003e15\u003c/strong\u003e, 980 (2024).\u003c/li\u003e\n\u003cli\u003eWang, H. \u003cem\u003eet al.\u003c/em\u003e Serum metabolic traits reveal therapeutic toxicities and responses of neoadjuvant chemoradiotherapy in patients with rectal cancer. \u003cem\u003eNat. Commun.\u003c/em\u003e \u003cstrong\u003e13\u003c/strong\u003e, 7802 (2022).\u003c/li\u003e\n\u003cli\u003eMalier, M. \u003cem\u003eet al.\u003c/em\u003e Tumor-associated macrophages confer resistance to chemotherapy (trifluridine/tipiracil) in digestive cancers by overexpressing thymidine phosphorylase. \u003cem\u003eCancer Letters\u003c/em\u003e \u003cstrong\u003e606\u003c/strong\u003e, 217307 (2024).\u003c/li\u003e\n\u003cli\u003eWusiman, M. \u003cem\u003eet al.\u003c/em\u003e Serum S-adenosylhomocysteine, rather than homocysteine, is associated with hepatocellular carcinoma survival: A prospective cohort study. \u003cem\u003eThe American Journal of Clinical Nutrition\u003c/em\u003e \u003cstrong\u003e120\u003c/strong\u003e, 481\u0026ndash;490 (2024).\u003c/li\u003e\n\u003cli\u003eGuo, T. \u003cem\u003eet al.\u003c/em\u003e LINC00662 promotes hepatocellular carcinoma progression via altering genomic methylation profiles. \u003cem\u003eCell Death Differ\u003c/em\u003e \u003cstrong\u003e27\u003c/strong\u003e, 2191\u0026ndash;2205 (2020).\u003c/li\u003e\n\u003cli\u003eHuang, W. \u003cem\u003eet al.\u003c/em\u003e AHCYL1 senses SAH to inhibit autophagy through interaction with PIK3C3 in an MTORC1-independent manner. \u003cem\u003eAutophagy\u003c/em\u003e \u003cstrong\u003e18\u003c/strong\u003e, 309\u0026ndash;319.\u003c/li\u003e\n\u003cli\u003eUchiyama, N., Tanaka, Y. \u0026amp; Kawamoto, T. Aristeromycin and DZNeP cause growth inhibition of prostate cancer via induction of mir-26a. \u003cem\u003eEuropean Journal of Pharmacology\u003c/em\u003e \u003cstrong\u003e812\u003c/strong\u003e, 138\u0026ndash;146 (2017).\u003c/li\u003e\n\u003cli\u003eGoodman, R. S., Johnson, D. B. \u0026amp; Balko, J. M. Corticosteroids and cancer immunotherapy. \u003cem\u003eClin Cancer Res\u003c/em\u003e \u003cstrong\u003e29\u003c/strong\u003e, 2580\u0026ndash;2587 (2023).\u003c/li\u003e\n\u003cli\u003eWang, Y. \u003cem\u003eet al.\u003c/em\u003e Corticosteroid administration for cancer-related indications is an unfavorable prognostic factor in solid cancer patients receiving immune checkpoint inhibitor treatment. \u003cem\u003eInt Immunopharmacol\u003c/em\u003e \u003cstrong\u003e99\u003c/strong\u003e, 108031 (2021).\u003c/li\u003e\n\u003cli\u003eWei, J. \u003cem\u003eet al.\u003c/em\u003e Association between gut microbiome-related metabolites and symptomatic hand osteoarthritis in two independent cohorts. \u003cem\u003eeBioMedicine\u003c/em\u003e \u003cstrong\u003e98\u003c/strong\u003e, 104892 (2023).\u003c/li\u003e\n\u003cli\u003eZhao, H. \u003cem\u003eet al.\u003c/em\u003e Inflammation and tumor progression: Signaling pathways and targeted intervention. \u003cem\u003eSig Transduct Target Ther\u003c/em\u003e \u003cstrong\u003e6\u003c/strong\u003e, 1\u0026ndash;46 (2021).\u003c/li\u003e\n\u003cli\u003eTumeh, P. C. \u003cem\u003eet al.\u003c/em\u003e PD-1 blockade induces responses by inhibiting adaptive immune resistance. \u003cem\u003eNature\u003c/em\u003e \u003cstrong\u003e515\u003c/strong\u003e, 568\u0026ndash;571 (2014).\u003c/li\u003e\n\u003cli\u003eYasuda, T. \u0026amp; Wang, Y. A. Gastric cancer immunosuppressive microenvironment heterogeneity: Implications for therapy development. \u003cem\u003eTrends Cancer\u003c/em\u003e \u003cstrong\u003e10\u003c/strong\u003e, 627\u0026ndash;642 (2024).\u003c/li\u003e\n\u003cli\u003eDoroshow, D. B. \u003cem\u003eet al.\u003c/em\u003e PD-L1 as a biomarker of response to immune-checkpoint inhibitors. \u003cem\u003eNat Rev Clin Oncol\u003c/em\u003e \u003cstrong\u003e18\u003c/strong\u003e, 345\u0026ndash;362 (2021).\u003c/li\u003e\n\u003cli\u003eZelenay, S. \u003cem\u003eet al.\u003c/em\u003e Cyclooxygenase-dependent tumor growth through evasion of immunity. \u003cem\u003eCell\u003c/em\u003e \u003cstrong\u003e162\u003c/strong\u003e, 1257\u0026ndash;1270 (2015).\u003c/li\u003e\n\u003cli\u003eJin, J., Byun, J.-K., Choi, Y.-K. \u0026amp; Park, K.-G. Targeting glutamine metabolism as a therapeutic strategy for cancer. \u003cem\u003eExp Mol Med\u003c/em\u003e \u003cstrong\u003e55\u003c/strong\u003e, 706\u0026ndash;715 (2023).\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Gastric cancer, Neoadjuvant Immunochemotherapy, Machine learning, Metabolomics, Prediction Model, Monitoring Model","lastPublishedDoi":"10.21203/rs.3.rs-6365771/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6365771/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eNeoadjuvant Immunochemotherapy (NAIC) shows promising application prospects in the treatment of (Gastric Cancer, GC) patients. However, the differences between individual patients and the issue of treatment resistance significantly affect whether patients could truly benefit from this treatment. This study conducted metabolomic analysis of 369 plasma samples from 108 gastric cancer patients following NAIC treatment to characterize their metabolic profiles for more accurate therapeutic efficacy prediction. Machine learning was used to build a GC treatment response prediction model 21-PM based on the expression levels of baseline metabolites. Two efficacy monitoring models, 11-MMI and 13-MMP, were developed using R-specific metabolites based on imaging and histology outcomes, respectively. In conclusion, the outcomes of this study offered strong evidence for the advancement of precision medicine in GC by exposing the metabolic landscape of GC patients after NAIC treatment and efficiently creating models that could independently predict prediction and monitor treatment.\u003c/p\u003e","manuscriptTitle":"Metabolomic Insights into the Predictive Landscape of Neoadjuvant Immunochemotherapy in Gastric Cancer: Towards Precision Medicine with Machine Learning","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-05-14 11:17:16","doi":"10.21203/rs.3.rs-6365771/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"07a5cd9d-2570-48e7-8be3-fd643234ccd9","owner":[],"postedDate":"May 14th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":48485329,"name":"Health sciences/Oncology/Cancer"},{"id":48485330,"name":"Health sciences/Biomarkers/Predictive markers"}],"tags":[],"updatedAt":"2025-08-12T11:11:11+00:00","versionOfRecord":[],"versionCreatedAt":"2025-05-14 11:17:16","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6365771","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6365771","identity":"rs-6365771","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}