Unveiling gastric precancerous stages: metabolomic insights for early detection and intervention.

Background
Globally, gastric cancer (GC) is the fifth most common cancer and the third leading cause of cancer-related deaths [1]. Regrettably, the majority of patients receive a diagnosis at advanced or late local stages, leading to poor prognoses [2]. GC develops via a multi-step process, beginning with superficial gastritis and advancing through to chronic atrophic gastritis (AG), intestinal metaplasia (IM), low-grade intraepithelial neoplasia, and high-grade intraepithelial neoplasia, ultimately leading to invasive GC [3, 4]. According to epidemiological data, AG is a prevalent and persistent condition, often without symptoms, affecting over half of the world’s population at some stage [5]. IM significantly increases GC risks, doubling the likelihood of its occurrence [6]. Fortunately, at gastritis and IM stages, timely detection and intervention can effectively control or even reverse the disease. However, once the disease has progressed to different intraepithelial neoplasia levels, it is difficult to manage progression with non-surgical treatments [7]. Therefore, identifying high-risk populations at AG and IM stages is crucial for developing and implementing effective prevention and management strategies. While Helicobacter pylori (H. pylori) infection is strongly associated with gastritis and gastric disease advancement [8, 9], it does not directly cause GC [7]. Currently, limited biomarkers can accurately predict GPL progression. Therefore, to inform precise clinical intervention strategies, addressing GC prevention and control challenges require enhanced etiological research and the exploration of novel molecular features to delineate gastric lesion progression.
A fundamental feature of cancer cells is the reprogramming of energy metabolism, which promotes tumor cell proliferation [10–12]. Metabolic dysregulation significantly contributes to GC development, highlighting the need to characterize metabolic profiles associated with disease progression [13–16]. High-throughput metabolomics technology enables the identification and quantification of endogenous low-molecular-weight metabolites [17, 18], thereby highlighting complex interactions between the host, genes, and the environment [19]. Thus, metabolomics has become a promising tool for distinguishing GC molecular features and identifying biomarkers [13, 20].
Many studies have investigated GC metabolomic molecular features and core metabolic pathways. Kuligowski et al. used metabolomics to identify tryptophan, kynurenine, and phenylacetylglutamine as potential GC biomarkers, in their plasma metabolomic analysis of over 400 participants [21], Huang et al. showed that α-linolenic acid, linoleic acid, and palmitic acid were significantly associated with IM progression [22]. To our knowledge, most gastritis and IM studies have focused on blood, urine, and other biological samples [22–24], with no direct research on lesion sites. In this study, we conducted a comprehensive metabolomic analysis of 188 clinical tissue samples from 98 patients, focusing on AG and IM stages. Using pseudotargeted metabolomics technology, which provides higher sensitivity and broader dynamic monitoring ranges when compared to untargeted metabolomics [25], we identified differential metabolites between AG, IM, and relatively normal tissues, and explored the metabolic changes from normal to AG and progression to IM.

Materials and methods

Patients and tissue samples
Between April and June 2023, 26 patients from the Department of Gastroenterology at Jiangnan University Affiliated Hospital South Campus, who underwent gastroscopy or treatment, were enrolled for targeted metabolomics analysis. An additional 71 patients were recruited between July 2023 and April 2024 for potential metabolic biomarker verification. Patients were recruited with no history of gastric or other cancers and no other diagnosed specific diseases, including digestive system diseases, prior to sampling. Relatively normal tissue samples (controls) comprised of gastric mucosal tissues taken at 5 cm away from matched lesion sites. Samples were collected by two experienced gastroenterologists during endoscopic procedures, and histological identification was performed by pathologists. The Ethics Committee of Jiangnan University Affiliated Hospital approved the study and all patients provided written informed consent before enrollment. Ethics number for: LS2024300. All patients provided written informed consent before enrollment. The patients’ baseline clinical characteristics, including age, sex, and H. pylori infection status, are summarized in and Table 1, and the severity grading of H. pylori infection is provided in Supplementary Table 4 [26]. Data and software availability pseudotargeted metabolomics data on gastric have been deposited in the European Bioinformatics Institute under accession code MTBLS11835.

Sample preparation for pseudotargeted metabolomics analysis
Tissue samples were weighed, transferred to 2 mL tubes, and kept on ice. Next, 200 µL of cold methanol was added, and samples vortexed for 1 min. Two steel beads were added and samples ground at 50 Hz for 10 s; this procedure was performed 3–4 times under low temperature conditions. Samples then underwent 15 min of ultrasonic extraction in an ice bath, followed by centrifugation at 12,000 × g for 15 min at 4 °C. Supernatants (100 µL) were transferred to new tubes, freeze-dried, and stored at -20 °C. Metabolite extracts were redissolved in 100 µL of 20% methanol/water prior to Ultra-Performance Liquid Chromatography-Tandem Quadrupole Mass Spectrometry (UPLC-MS/MS).

UPLC-MS/MS conditions for pseudotargeted metabolomics
For metabolite analysis, we used the AB SCIEX Q-Trap 5500 triple quadrupole mass spectrometer (AB SCIEX, USA), using both positive and negative electrospray ionization modes with multiple reaction monitoring scanning. LC separation was performed using an ACQUITY UPLC HSS T3 column (2.1 × 100 mm, 1.8 μm) (Waters, USA) for positive ion mode and an ACQUITY UPLC BEH C18 column (2.1 × 100 mm, 1.7 μm) (Waters, USA) for negative ion mode. The Exion LC AD system (AB Sciex, USA) featured a binary gradient pump, using 0.1% formic acid in water (solvent A) and acetonitrile (solvent B) as mobile phases at a flow rate of 0.35 mL/min. The gradient started at 5% solvent B, linearly increasing to 99% in 11 min, holding for 2 min, then returning to 5% over the last 2 min. The column temperature was 50 °C and the injection volume was 5 µL.
MS parameters included an ion source temperature of 550 °C, a curtain gas pressure of 35 psi, a collision gas gas pressure of 10 psi, and ionspray voltages of 5,500 V in positive mode and − 4,500 V in negative mode. Ion gases 1 and 2 were both set to 60 psi. AB SCIEX Analyst 1.7.1 was used to acquire data.

Targeted metabolomics conditions
To create a standard curve, a mixed standard solution containing 1 ppm (1ppm = 1 mg/L) of glycerophosphocholine, sphingosine, malate, and γ-glutamylglutamic acid was prepared and diluted to 1, 5, 10, 50, 100, and 500 ppb (1ppb = 1/1000ppm). A separate tiglylcarnitine dilution series was prepared over the 0.05–10 ppb concentration range. Standard curves were established by plotting analyte-to-internal standard area ratios against concentration. The sample preparation as well as the metabolomics mass spectrometry and chromatographic conditions were consistent with the pseudotargeted metabolomics method, except that the column was ACQUITY UPLC HSS T3 (2.1 × 100 mm, 1.8 μm).

Reagents
Methanol (LC-MS grade), acetonitrile, and formic acid were obtained from ThermoFisher Scientific (USA). The 1-methyl-3-nitro-1-nitrosoguanidine (MNNG) (CAS: 70-25-7) and glycerophosphocholine (CAS: 28319-77-9) were sourced from Shanghai Yuanye Bio-Technology Co., Ltd (Shanghai, China). Sphingosine (CAS: 764-22-7), malate (CAS: 97-67-6), and γ-glutamylglutamic acid (CAS: 1116-22-9) were obtained from Shanghai Macklin Biochemical Co., Ltd (Shanghai, China). Tiglylcarnitine (CAS: 64681-36-3) was procured from Shanghai Zhenzhun Bio-Technology Co., Ltd (Shanghai, China).

Cell culture
GES-1 human gastric epithelial cells (iCell Bioscience Inc, Shanghai, China) were cultured to 60% confluence in six-well plates containing Roswell Park Memorial Institute 1640 medium plus 10% fetal bovine serum (FBS) and 1% Penicillin-Streptomycin solution. A humidified environment of 5% CO2 and 95% air was used to culture cells.

Cell viability assays
Cell viability was assessed using a Cell counting kit-8 (CCK8) kit (Nanjing BEB Laboratories Co., Ltd, NanJing, China). Cells (8 × 103) were seeded in 96-well plates, allowed to adhere, treated with 1% FBS for 24 h, and exposed to varying metabolite concentrations for another 24 h. Supernatants were then removed, CCK8 solution added, and cells incubated in the dark for 2–4 h. Next, to assess cell viability, the optical density at 450 nm was measured using a SpectraMax 190 light absorption enzyme labeler (Molecular Devices, USA).

Quantitative real-time polymerase chain reaction (qRT-PCR)
Total RNA was isolated from cells by using the TRIzol reagent. qRT-PCR was performed according to the previous protocol [27, 28]. Sequences of the specific primer sets are as follows: CDX2: Forward: 5’-TTCACTACAGTCGCTACATCACCA-3’, Reverse: 5’-CTGCGGTTCTGAAACCAGATT-3’, KLF4: Forward: 5’-GTGCCCCGAATAACAGCTCA-3’, Reverse: 5’- TTCTCACCTGTGTGGGTTCG- 3’, MUC2: Forward: 5’-GAGGGCAGAACCCGAAACC-3’, Reverse: 5’- GGCGAAGTTGTAGTCGCAGAG-3’. Glyceraldehyde-3-phosphate dehydrogenase (GAPDH) was used as a control (Supplement Table 3). The relative gene expressions were analyzed using 2−ΔΔCTmethod.

Statistical analysis
MetaboAnalyst 6.0 was used for unsupervised principal component analysis and hierarchical data clustering. Samples were normalized for their respective masses to visualize the metabolomic profiles of N, AG and IM participants. Multivariate analyses, such as partial least squares discriminant analysis (PLS-DA) and the orthogonal partial least squares discriminant analysis (OPLS-DA), were performed to identify key differentiating features between groups. Models were validated using 10-fold cross-validation, with results showing high R2 and Q2 values, indicating model reliability, and a permutation testing was performed 200 times to assess the risk of overfitting for the PLS-DA model. Metabolites with Variable Importance in the Projection (VIP) > 1, p < 0.05, and Fold Change (FC) > 1.5 or < 0.5 were selected as differential metabolites. GraphPad Prism 9.4.1 was used for univariate statistical analysis. One-way analysis of variance (ANOVA) with Tukey’s post hoc test was applied for multiple group comparisons. An FDR-adjusted p-value threshold of 0.05 was used to assess statistical significance. For two-group comparisons, Student’s t-test or the Mann-Whitney U test was applied as appropriate. The data was presented as mean ± standard error (SEM). p < 0.05 was considered statistically significant.

Results

Metabolic profiling of clinical gastric tissue samples
Given limited sample availability (approximately 5 mg), we performed pseudotargeted metabolomics. This approach provides greater sensitivity and a broader dynamic range when compared to non-targeted metabolomics, while eliminating complex feature detection or peak alignment [25].
First, we collected and prepared equal amounts of gastric mucosal tissue samples from clinically N, AG, and IM sites.AG Using UPLC-MS/MS pseudotargeted metabolomics, we identified 433 metabolites across both positive and negative ion modes. To assess instrument stability, we prepared tissue samples with varying concentration gradients and identified 152 and 147 metabolites in positive and negative ion modes, respectively, thereby establishing a database for targeted metabolomics analysis (Supplement Tables 1 and 2).

Metabolic differences between AG, IM, and normal tissues
Pseudotargeted metabolomics was performed on tissue samples (N = 20, AG = 20, and IM = 20) using UHPLC-MS/MS (Table 1, Cohort 1), with the relative quantification of 433 metabolites (Fig. 1A). A PLS-DA score plot showed that IM and Normal groups formed two distinct clusters, with the AG group overlapping to varying degrees with these groups. This observation is consistent with the Correa cascade, in which the disease progresses from N, AG to IM and eventually to gastric cancer (Fig. 2A). Thus, the PLS-DA model demonstrated good reliability, as evidenced by a 10-fold cross-validation yielding high R2 (0.83) and positive Q2 (0.57) values (Fig. 2B).

To elucidate metabolic alterations during disease progression, pairwise comparisons between Normal vs. AG, IM, and AG vs. IM were conducted. OPLS-DA supervised analysis identified distinct group separations, revealing distinct metabolic changes as the disease advanced from gastritis to IM (Fig. 2C–E). Using VIP > 1, P < 0.05, and FC > 1.5 or < 0.5 criteria, 18 metabolites were identified (Table 2), including organic acids, four amino acids and derivatives, three lipids, one glycerophospholipid, antibiotics, sphingolipids, and other compounds (Fig. 2F). Fatty acyl groups and glycerophospholipids were more abundant in normal gastric tissue when compared to GPL, whereas organic acids were more prevalent in GPL than in normal tissue. To determine major metabolic and signaling pathways associated with differential metabolites between N, AG, and IM tissues, we performed Kyoto Encyclopedia of Genes and Genomes enrichment analysis, which identified 18 metabolic pathways, encompassing citric acid cycle, pyruvate metabolism, Warburg effects, malate-aspartate shuttle, and pyruvaldehyde degradation pathways (Fig. 2G).

The identification of metabolic biomarkers to differentiate GPL from normal tissue
As differentiating AG from IM tissue was challenging, these groups were merged into a GPL group based on endoscopic observations. Then, a receiver operating characteristic (ROC) curve analysis was conducted on the 18 differential metabolites to evaluate their diagnostic accuracy in distinguishing normal from GPL groups. Of these metabolites, seven had an area under the curve (AUC) value > 0.7, including one with an AUC > 0.8, while the remainder ranged from 0.6 to 0.7 (Supplement Table 5). We cross-referenced this metabolite information and chemical abstracts service (CAS) numbers with the Human Metabolome Database and acquired standards for targeted validation. Ultimately, the following five standards successfully generated calibration curves: glycerophosphocholine, sphingosine, γ-glutamylglutamic acid, tiglylcarnitine, and malate (Fig. 3A-E), and their respective AUC values were 0.801 (95% confidence interval (CI): 0.682–0.920), 0.790 (95% CI: 0.669–0.911), 0.733 (95% CI: 0.592–0.873), 0.719 (95% CI: 0.589–0.849), and 0.665 (95% CI: 0.513–0.817) (Fig. S1A–E). The combined metabolite AUC value was 0.836 (95% CI: 0.714–0.959) (Fig. 3F). These results indicate that the combination of the five metabolites has the potential to serve as potential biomarkers for distinguishing between normal and GPL in gastric tissues.

Metabolic biomarker verification in GPL
To validate the effectiveness of these metabolites, including glycerophosphocholine, as biomarkers distinguishing normal and precancerous gastric tissues, we recruited an additional 71 patients who required endoscopic examinations (Table 1, Cohort 2) for targeted metabolomics (Fig. 1A). As shown, from absolute targeted metabolomics quantification results, inter-group metabolite variation trends, including glycerophosphocholine, generally matched the relative quantification trends in pseudotargeted metabolomics analysis, indicating that metabolite changes at different disease progression stages were consistent (Figs. 3A–E and 4A–E). Metabolite AUC values in normal and precancerous gastric tissues are shown: glycerophosphocholine (0.903, 95% CI: 0.834–0.972) (Fig. 4F), tiglylcarnitine (0.779, 95% CI: 0.688–0.870), malate (0.619, 95% CI: 0.512–0.727), sphingosine (0.602, 95% CI: 0.486–0.718), and γ-glutamylglutamic acid (0.643, 95% CI: 0.539–0.746) (Fig. S2A-D), (Supplement Table 6). The aggregated metabolite AUC value was 0.916 (95% CI: 0.851–0.981) (Fig. 4G). These findings demonstrate that glycerophosphocholine, tiglylcarnitine, malate, sphingosine, and γ-glutamylglutamic acid possess the potential to differentiate normal tissues from precancerous gastric lesions. The combination of these five metabolites effectively distinguishes normal and precancerous lesion tissues.

Clinically, accurately distinguishing between AG and IM stages has always been a challenge; therefore, we performed ROC curve analysis on AG and IM subgroups in PGLs. Only malate had an AUC value > 0.7, with an AUC of 0.772 (95% CI: 0.671–0.874) (Fig. 4H), while AUC values for glycerophosphocholine and tiglylcarnitine were 0.690 (95% CI: 0.577–0.803) and 0.697 (95% CI: 0.586–0.809), respectively (Fig. S2E-F), (Supplement Table 7). By combining absolute targeted quantification results, malate, glycerophosphocholine, and tiglylcarnitine showed significant differences not only between normal and IM groups, but also between AG and IM groups, suggesting that these metabolites could show potential as biomarkers distinguishing AG and IM tissues.

GPL cell model construction
To create a GPL-related GES-1 cell simulation model (Fig. 1A), GES-1 cells were exposed to 0, 10, 15, 20, 25, 30, 35, and 40 µM MNNG concentrations for 24 h to examine concentration and treatment duration effects. As shown (Fig. 5B), 20 µM MNNG significantly inhibited GES-1 cell proliferation, leading us to select the 15–25 µM concentration range, qPCR was then used to evaluate MNNG effects (0, 15, 20, 25 µM) on intestinalization progression in cells, and after 24 h of 20 µM treatment, GES-1 cells exhibited significant increases in CDX2, KLF4, and MUC2 expression (Fig. 5C–E). CDX2, KLF4, and MUC2 are universal indicators of expression when enterocytosis occurs [29]. From analyses, we determined that treating GES-1 cells with 20 µM MNNG for 24 h optimally induced intestinalization, thereby establishing a GPL cell model.

Key metabolite effects on MNNG-induced GPL
To investigate key metabolite roles and effects on GPL cell model viability following MNNG treatment, cells were starved for 24 h after adherence, pretreated with varying metabolite concentrations for 24 h, and then exposed to MNNG under optimal conditions. As shown (Fig. S3A-C), while metabolites did not significantly impact on cell proliferation, cells pretreated with glycerophosphocholine showed slightly decreased viability at 12.5 µM, as the concentration of glycerophosphocholine intervention increased, cell viability gradually returned to a level comparable to that of the MNNG modelling group alone. Cells pretreated with 62.5 µM malate showed some improvements in viability when compared to the model group, while cells pretreated with 3.125 µM tiglylcarnitine showed slightly increased viability when compared to the model group, despite CCK8 data showing no significant differences.
Although CCK8 data showing no significant differences, we selected altered concentration gradients for qPCR analyses to explore potential metabolite involvement in gene-level intestinalization processes. We observed that 50 µM glycerophosphocholine and 3.125 µM tiglylcarnitine significantly reduced CDX2 and KLF4 expression, while 125 µM malate significantly increased CDX2 expression (Fig. 6A–C). Although these metabolites did not induce differential MUC2 expression, as the concentration of malic acid increased, the expression of the MUC2 gene was observed to increase to a certain extent. Conversely, the intervention of tiglylcarnitine was found to reduce the rise in MUC2 gene levels to a certain extent (Fig. S4A-C). By combining metabolomics with in vitro data, our findings suggest that glycerophosphocholine and tiglylcarnitine may have the potential to delay intestinalization occurrence and progression in vitro, while malate may promote its development.

Discussion
Metabolomics, as a relatively mature detection technology, has shown significant developmental potential toward gastric diseases and GC at different stages [13, 22]. The metabolites discovered using this approach have shown broad research prospects and clinical applications in disease etiology, early diagnosis, and prognosis [30, 31]. Metabolomics-based analytical techniques have identified distinct and overlapping endogenous biomarkers across different gastric disease stages, encompassing glucose, amino acid, lipid, and nucleotide metabolism pathways. Metabolic alterations have crucial roles in disease onset and progression. However, in studies examining gastric diseases and GC at various stages, metabolic changes have been reported, but the results often vary [22, 23, 32, 33]. Most studies have used easily obtainable samples, such as patient blood, urine, etc., which reflect systemic metabolic changes. In contrast, in our study, we selected gastric tissue samples that more directly reflected metabolic changes at lesion sites. Using advanced metabolomics techniques, we analyzed the precious and limited gastric tissue samples obtained via endoscopy and have established our own gastric tissue metabolome database, comprising 433 metabolites collected in both positive and negative ion modes.
Using a pseudotargeted metabolomics analysis of the discovery cohort, 18 differential metabolites were identified in both normal and precancerous gastric tissues. ROC curve analysis was then used to identify potential differential metabolites as biomarkers, selecting those with AUC values > 0.7 for validation, and to distinguish between normal and precancerous gastric tissue. For this, we collected a larger validation cohort for absolute targeted quantification analysis. Ultimately, a combination of glycerophosphocholine, tiglylcarnitine, malate, sphingosine, and γ-glutamylglutamic acid potentially acted as powerful biomarkers differentiating normal from precancerous gastric tissue. Moreover, malate, glycerophosphocholine, and tiglylcarnitine effectively differentiated between AG and IM tissue subgroups, suggesting the timely identification and definition of different pathological stages in the gastric Correa cascade process.
Glycerophosphocholine can be effectively used to treat and prevent multiple biochemical diseases [34–38]. Other metabolomic studies have shown that glycerophosphocholine levels steadily decrease as GC progresses from gastritis to early GC and then onto advanced GC [24], consistent with our quantitative glycerophosphocholine results across different groups. During the progression from normal gastric tissue to gastric cancer, metabolic reprogramming occurs, including alterations in lipid metabolism. As a type of phospholipid, GPC levels may decrease due to dysregulation in lipid synthesis and degradation. Additionally, the expression levels of phospholipase A2 (PLA2) in gastric cancer tissues are significantly higher than in normal gastric mucosal tissues, and the elevated activity of PLA2 may lead to increased GPC degradation, resulting in reduced GPC levels. Malate is involved in multiple metabolic pathways and is a key intermediate in many biochemical reactions, with important roles in the body. Research has shown that downregulating malate dehydrogenase-3 may advance GC by influencing oxidative stress and stabilizing hypoxia-inducible factor-1α, with malate, a substrate of this enzyme, crucial in GC progression [39]. Sphingosine is a type of sphingolipid that helps mediate pro-inflammatory events in the gastrointestinal tract via sphingolipid signaling [40]. Previous studies have demonstrated that sphingosine plays a crucial role in apoptosis, inflammatory signaling pathways, and lipid metabolism. Notably, its metabolite sphingosine-1-phosphate (S1P) may have dual effects, either promoting cell survival or accelerating carcinogenesis. Furthermore, the S1P-S1P receptor axis has been reported to be a key player in gastric cancer progression, potentially activating PI3K/Akt, MAPK, and NF-κB signaling pathways to facilitate gastric epithelial cell proliferation, inhibit apoptosis, and enhance inflammation. Thus, dysregulated sphingosine metabolism may serve as a critical driver of gastric lesion development. Currently, there is limited research on associations between tiglylcarnitine and γ-glutamylglutamic acid and the stomach, thus further exploration and verification studies are required.
As we directly used clinical gastric tissue micro-samples, this approach better reflected metabolic changes at lesion sites when compared to blood, urine, and saliva samples used in previous studies [22–24, 41], maximizing the use of clinical samples and showing potential for translational medicine. We aim to identify potential beneficial metabolites that are abundant in normal tissues but significantly reduced in diseased tissues, with the goal of exploring the protective effects of supplementing these metabolites on the gastric mucosa. Through this approach, we established a gastric tissue metabolic database. By integrating metabolomics analysis of clinical endoscopic biopsy samples with measurements of glycerophosphocholine, tiglylcarnitine, malate, sphingosine, and γ-glutamylglutamic acid levels, we were able to predict the occurrence of gastric precancerous lesions to a significant extent. Functional validations of these biomarkers were also performed at the cellular level. We showed that glycerophosphocholine, malate, and tiglylcarnitine did not significantly impact GES-1 cell viability under MNNG treatments in our GPL model. However, glycerophosphocholine and tiglylcarnitine partially decreased gene expression related to intestinalization, while malate showed a tendency to enhance this expression.
Our study also has several limitations. First, although the study includes both discovery and validation cohorts, the overall sample size is relatively limited, and the study was conducted at a single center, which may impact the statistical power and generalizability of the findings. Second, the absence of a longitudinal cohort restricts our ability to capture the dynamic evolution of metabolic changes. Third, factors such as H. pylori infection, dietary habits, and medication use may influence the gastric tissue metabolome, but these variables were not strictly controlled in our study. To address the above limitations, we plan to expand the sample size in future studies, while strictly controlling for confounding factors such as H. pylori infection, diet, and medication use, and simultaneously conducting a longitudinal cohort study. In addition, we plan to adopt a multi-center design in future studies to further validate the robustness and clinical applicability of our results. We also prepare to collect paired tissue and peripheral biofluid samples (such as blood and urine), followed by metabolomics analysis of these samples, aiming to identify more stable and clinically relevant metabolic biomarkers, thereby enabling non-invasive screening and risk assessment. Furthermore, we plan to integrate machine learning and multi-omics approaches to enhance the predictive power of these biomarkers and explore their potential role in the early detection of gastric diseases and personalized medicine.

Conclusions
In summary, we have established the metabolic profiles of clinical precancerous gastric tissue samples for the first time, by focusing on the metabolic changes within the lesions themselves. We identified key metabolite molecules that can distinguish between normal, AG, and IM tissues, which provides new evidence and strategies for clearly defining the precancerous stages and offers a potential approach to distinguishing the pathological stages of AG and IM. However, further research is needed, including expanding the cohort, assessing the consistency of metabolic changes in different matrices, and conducting animal experiments to explore mechanisms and clinical applications.

Electronic supplementary material
Below is the link to the electronic supplementary material.