Serum Untargeted Metabolomics Profiling of Esophageal Squamous Cell Carcinoma in High-Incidence Areas of China

Introduction

Esophageal cancer (EC) is recognized as a critical global health issue. Based on the GLOBOCAN 2022 estimates, there were 510,716 new EC cases and 445,129 related deaths worldwide.¹ ESCC is the dominant histological subtype, accounting for more than 85% of all EC cases, with a disproportionately high prevalence in low‑ and middle‑income countries.² ESCC incidence shows substantial geographic heterogeneity, with Hebei Province in northern China recognized as a typical high‑incidence region.^3–5 Although multidisciplinary therapeutic strategies have improved in recent years, the clinical outcomes of ESCC are still dismal, as the 5‑year overall survival ranges from 25% to 49%.⁶ The persistently poor survival, coupled with distinct geographic disparities, underscores the urgent need for etiological investigations. Conducting case-control or cohort studies in high-incidence areas is crucial for identifying key risk factors, informing targeted prevention strategies, and ultimately reducing the disease burden of ESCC.

Metabolic reprogramming stands as a core hallmark of malignant tumors.⁷ As a key systems biology platform, metabolomics enables precise detection and quantification of small molecules (<2000 Da) via liquid chromatography-mass spectrometry (LC-MS)-based untargeted profiling.⁸ Of note, it has prominent advantages in deciphering molecular mechanisms of disease, particularly in biomarker discovery, early diagnosis, and prognosis monitoring, all of which hold substantial clinical translational value.^9–11

In ESCC research, several metabolomics studies have yielded key clinical translational findings: Wang et al¹² identified 15 differential metabolites (eg, all-trans-13,14-dihydroretinol) forming a robust early prediction model, and found that high monoacylglycerol (MG) (20:4) isomers combined with low 9,12-octadecadienoic acid and L-isoleucine correlate with poor prognosis; another study constructed a diagnostic model using 5 metabolites (eg, tryptophan) to distinguish early ESCC from HCs;¹³ most notably, Liu et al¹⁴ developed a model based on 22 serum metabolites (eg, n-methylproline) with excellent early warning capabilities up to five years pre-ESCC diagnosis. Collectively, these findings highlight the potential of serum metabolites for early ESCC screening. Nevertheless, published metabolomics data show obvious platform-dependent variations, necessitating high-sensitivity instrumentation for comprehensive metabolic profiling.¹⁵ Furthermore, metabolic characteristics are susceptible to geographic environment, lifestyle factors, and host physiological status, particularly in high‑incidence regions.¹⁶ Therefore, further metabolomics investigations focusing on ESCC are warranted to improve the reproducibility and generalizability of related findings.^15,17

To address the need for further research, particularly in populations from high-incidence regions, we performed the present serum untargeted metabolomics study. While sharing the same patient cohort as our previously published cross-sectional gut microbiota study in Chinese ESCC high-incidence areas,¹⁸ this work represents the first characterization of serum metabolic profiles for this cohort. Using distinct technical platforms, we focused on host systemic metabolic alterations to identify complementary molecular mechanisms underlying ESCC pathogenesis. Our findings are independent of those prior results and offer a multi-omics perspective for elucidating ESCC development and progression.

In the present study, 70 ESCC patients and 72 HCs were enrolled from a high-incidence region of ESCC in China. Serum untargeted metabolomics profiling was performed using UHPLC-Orbitrap-MS/MS, followed by bioinformatics analysis. We hypothesized that serum metabolic profiles differ significantly between ESCC patients and HCs, with the aim of identifying differential metabolites and key metabolic pathways closely associated with ESCC initiation.

Materials and Methods

Research Design

The patient cohort of this serum untargeted metabolomics study overlaps with that in our previously published gut microbiota study of ESCC patients from the same high-incidence region.¹⁸ Although this study shares the same study population, it focuses exclusively on serum metabolomic profiling instead of gut microbial community analysis. We aim to explore the etiologies of ESCC from distinct and independent molecular levels, thereby providing essential exploratory work and data support for subsequent integrated research.

Initially, 117 histopathologically confirmed ESCC patients with no prior anti-tumor treatment were recruited from the Fourth Hospital of Hebei Medical University from September 2023 to September 2024. ESCC was histopathologically confirmed according to the 2019 WHO Classification of Tumors: Digestive System Tumors.¹⁹ All histological diagnoses were independently reviewed and verified by at least two senior gastrointestinal pathologists. Any discrepancies were resolved by consensus discussion to ensure diagnostic accuracy. Clinical stages were categorized according to the 8th edition of the American Joint Committee on Cancer (AJCC) staging system:²⁰ stage I (n = 16), stage II (n = 19), stage III (n = 58), and stage IV (n = 24). Concurrently, 78 healthy volunteers with normal laboratory test results were recruited from the Health Examination Center as controls.

Given that advanced ESCC patients with distant metastasis are frequently accompanied by malnutrition, cachexia, and severe systemic complications, these factors may significantly interfere with the host metabolic status, so only patients with clinical stage I–III ESCC were ultimately included in the subsequent analysis. According to the principle of approximate 1:1 matching, 70 ESCC cases and 72 healthy individuals were selected as the final study participants. Baseline data were collected, including 1) sociodemographic and lifestyle characteristics: gender, age, body mass index (BMI), smoking, and drinking habits; and 2) the clinical staging of ESCC²⁰ and tumor lesion location.

Criteria for Participant Enrollment

The eligibility criteria were adapted from previous studies with minor modifications.²¹ The following were regarded as inclusion criteria: 1) participants aged 18–85 years; 2) histopathologically confirmed primary ESCC (untreated); 3) an Eastern Cooperative Oncology Group (ECOG) performance status score not exceeding 2; and 4) patients with clinical stage I–III ESCC.

Exclusion criteria were as follows: 1) prior tumor treatment (chemoradiotherapy or surgery); 2) secondary malignancies; 3) history of severe cardiovascular events (myocardial infarction or stroke); 4) exposure to antibiotics, probiotics, or proton pump inhibitors (PPIs) within the past two months; 5) history of major gastrointestinal surgery; 6) chronic intestinal inflammatory diseases (inflammatory bowel disease (IBD) or irritable bowel syndrome (IBS)); 7) metabolic or mental health disorders (eg, diabetes or depression); and 8) patients complicated with malnutrition, cachexia, or severe systemic complications. All participants were long-term residents of Hebei Province (≥10 years).

Subject Sample Collection and Processing

For serum collection, professional nurses performed standardized venipuncture to obtain fasting venous blood in the morning; for ESCC patients, this was completed within 48 hours before treatment. Subsequently, blood specimens were placed into anticoagulant-free vacuum tubes and stored at 4°C for no longer than 2 hours. Finally, serum was harvested by centrifugation at 4°C and 4000 rpm for 15 minutes, then aliquoted and preserved at −80°C until assay.²² All samples were managed under a cold chain to preserve integrity.

Serum Sample Preparation

Serum sample preparation followed established protocols. Serum samples were thawed at room temperature from −80°C storage. For each sample, 100 μL serum was combined with 400 μL of pre‑cooled extractant (acetonitrile:methanol = 1:1, containing isotopically-labeled internal standards). The mixture was subjected to three cycles of vortexing‑ultrasonication (30 s/10 minutes, ice bath), incubated at −40°C for 1 h, then centrifuged at 4°C and 12,000 rpm (RCF = 13,800×g) for 15 minutes. The resulting supernatant was subjected to LC‑MS analysis.

Quality control (QC) samples were prepared by pooling equal volumes of supernatants from all samples. During analysis, QC samples were interspersed at regular intervals to monitor instrument stability and signal drift. Analytical stability was assessed by evaluating Total Ion Current (TIC) overlap (Supplementary Figure 1), relative standard deviation (RSD) of internal standards (median RSD < 10%) (Supplementary Table 1), clustering in principal component analysis (PCA) (Supplementary Figure 2), and correlation among QC samples (Supplementary Figure 3). All samples were analyzed in randomized order to reduce systematic bias. Batch effects were controlled by internal standard normalization, real-time QC monitoring, and data preprocessing, including RSD filtering (removing features with RSD > 30% in QC samples) and missing value imputation (using K-nearest neighbors).

UHPLC-Orbitrap-MS/MS Data Analysis

UHPLC-Orbitrap-MS/MS data acquisition was conducted based on previously described methods with minor modifications.²³ Analysis was performed on a UHPLC system (Vanquish, Thermo Fisher Scientific) coupled with an Orbitrap Exploris 120 mass spectrometer (Thermo), using a Waters ACQUITY UPLC BEH Amide column (2.1 mm × 50 mm, 1.7 μm). Mobile phase A was 25 mmol/L ammonium acetate containing 25 mmol/L ammonia water (pH 9.75), and mobile phase B was acetonitrile. The column was maintained at 25°C, while the auto-sampler was set to 4°C, with an injection volume of 2 μL. Mass spectrometry detection was conducted on the Orbitrap Exploris 120 system, controlled by Xcalibur software (Thermo), in information-dependent acquisition (IDA) mode, where full-scan MS spectra were continuously evaluated by the acquisition software. Both positive and negative ionization modes were applied simultaneously in this study. The ESI source parameters were set as follows: sheath gas flow 50 Arb, auxiliary gas flow 15 Arb, capillary temperature 320°C, full MS resolution 60,000, MS/MS resolution 15,000, normalized collision energy (NCE) 20/30/40, and spray voltage ±3.8/3.4 kV.

Metabolomics Data Preprocessing and Annotation

Prior to multivariate statistical analysis, internal standard normalization was performed to correct systematic variations. Isotopically labeled internal standards (deuterated compounds) were spiked into each sample at a fixed concentration during metabolite extraction. Metabolite peak areas were normalized to the corresponding internal standard, or to the median response of all internal standards when no matched internal standard was available. The stability of internal standards was validated with RSD < 10% (median RSD < 5%) across QC samples. Raw data were converted to mzXML format using ProteoWizard (https://proteowizard.sourceforge.io/) and processed with an in-house R script based on XCMS (v3.7.1) for peak detection, alignment, and integration. Metabolite identification was performed using BiotreeDB (V3.0) with a confidence threshold of 0.3.^23,24

Metabolomics Data Analysis

The raw dataset containing 89,383 metabolic features was preprocessed to ensure data quality. Metabolic features were initially filtered out if their RSD in quality control samples exceeded 30%, resulting in 52,103 features retained for further analysis. Half-minimum value imputation was applied to address missing values. Prior to performing multivariate statistical analysis, log-transformation was applied to the data to achieve an approximate normal distribution. For PCA, data were Pareto‑scaled (divided by the square root of the standard deviation) to reduce the dominance of high‑abundance metabolites. For supervised orthogonal partial least squares discriminant analysis (OPLS-DA), data were auto‑scaled (unit variance scaling) to minimize the influence of high‑variance features. Multivariate analysis was conducted using SIMCA software (v18.0.1, Sartorius Stedim Data Analytics AB, Umea, Sweden). Initially, an unsupervised PCA model was established to visualize intrinsic clustering and detect outliers (Supplementary Figure 4). Subsequently, supervised OPLS-DA was employed to maximize group separation. Validation of the OPLS‑DA model was achieved through 7-fold cross‑validation (evaluation of R² and Q² values) and a 200‑permutation test, thereby guarding against potential overfitting. Differential metabolites were screened according to the variable importance in projection (VIP) score from the OPLS-DA model greater than 1.0 and a p-value from a Student’s t-test less than 0.05.²⁵ Finally, KEGG pathway enrichment analysis (http://www.genome.jp/kegg/) was carried out to elucidate the significantly perturbed metabolic pathways. The identities of the seven key fatty acids were verified by LC-MS/MS mirror spectra matching between experimental samples and authentic standards (Supplementary Figure 5).

Statistical Analysis

Different statistical methods were employed for various data analyses as follows: Mean ± standard deviation was used to express continuous data, with independent-samples t-tests applied for statistical analysis, whereas categorical data were given as proportions and analyzed via chi-square tests (SPSS 26.0, IBM, USA). Significant differential metabolites were identified using multivariate analysis (performed in R 4.2.0), defined as those simultaneously meeting the dual thresholds of VIP > 1 and P < 0.05 in t-tests. For comparisons of differential metabolites among the three groups, the Kruskal–Wallis H-test with Bonferroni-corrected post-hoc pairwise comparisons was performed. For two-tailed tests, the significance level was set to P < 0.05. Diagnostic models for distinguishing ESCC patients from healthy controls were established using logistic regression analysis. Ten-fold cross-validation was performed to evaluate the stability and diagnostic efficacy of the models, with AUC, sensitivity, specificity, and accuracy as evaluation indicators.

Results

Clinical Information

All subjects enrolled in this study were Han Chinese residents from Hebei Province, China. No significant discrepancies were observed in baseline characteristics between the two participant groups (all P > 0.05), thus ensuring baseline comparability. Detailed information is presented in Table 1.

Table 1 Baseline Demographic and Clinical Profiles of the Study Cohorts

Raw Data Analysis of Metabolomics

Untargeted metabolomics identified 3587 annotated metabolites from serum samples. The predominant class was lipids and lipid-like molecules (19.15%), followed by organic heterocyclic compounds (12.85%), benzenoids (10.48%), and organic acid derivatives (10.29%) (Figure 1A).

Figure 1 Composition of identified metabolites and analysis of significantly differential metabolites. Distribution of identified metabolites classified by the HMDB super class (A). Score plot of the OPLS‑DA model (B); model validation with 200 permutation tests (C); and validation plot for assessing model overfitting (D). Volcano plot of differential metabolites screened by combining VIP values and Student’s t‑test (E). Dot plot of differential metabolites after Log2 transformation (F).

Screening of Differential Metabolites Based on OPLS-DA

A pronounced separation between the metabolic profiles of HCs and ESCC patients was observed by OPLS-DA, with most data points located within the 95% CI (Figure 1B), indicating distinct metabolic phenotypes. The model was validated by 200 permutation tests, which yielded R²Y = 0.933 and Q² = 0.715 (Figure 1C). Moreover, the intercepts of the regression lines were 0.68 for R²Y and −1.15 for Q², indicating no overfitting and reliable model performance (Figure 1D). Applying a dual-threshold criterion (VIP > 1 and P < 0.05), we identified 713 significantly altered metabolites, with 418 (58.6%) upregulated and 295 (41.4%) downregulated (Figure 1E and Supplementary Table 2). After log₂-transformation, quantitative analysis highlighted key regulatory changes: the lipid (1β,2α,3α)-1,2,3,24-Tetrahydroxy-12-oleanen-28-oic acid was the most significantly upregulated, whereas the organic heterocyclic compound Sapidolide_A was the most significantly downregulated (Figure 1F).

Hierarchical Cluster Analysis of Differential Metabolites

Hierarchical cluster analysis was performed on the four most abundant classes of differential metabolites, as classified by the HMDB “Super Class” system. The analyzed classes, in order of abundance, were: lipids and lipid-like molecules (115 species, 16.1%), benzenoids (70 species, 9.8%), organic heterocyclic compounds (61 species, 8.5%), and organic acids and derivatives (49 species, 6.8%). The resulting heatmaps delineated distinct expression patterns for these metabolite classes between the HC and ESCC groups (Figure 2A–D). Long-named differential metabolites were labeled A–S in the heatmaps, with detailed information listed in Supplementary Table 3.

Figure 2 Hierarchical cluster heatmap analysis reveals the distribution patterns of differential metabolites. Based on the “super class” classification of the HMDB database, the distribution patterns of differential metabolites in HC and ESCC groups are shown, including lipids and lipid-like molecules (A), benzenoids (B), organoheterocyclic compounds (C), and organic acids and derivatives (D). The shade intensity in each region corresponds to the significance of metabolite changes (red indicates upregulation, while blue indicates downregulation).

Differential Metabolic Pathways and Metabolites

To clarify the functional relevance of the differential metabolites, we annotated and classified them against the KEGG database. The analysis revealed distinct distribution patterns across metabolic pathways. Most notably, the “Biosynthesis of unsaturated fatty acids” subclass within lipid metabolism exhibited the greatest proportion of differential metabolites (18.42%) (Figure 3A). Pathway enrichment analysis further identified eight statistically significant pathways (Supplementary Table 4). Among these, the “Biosynthesis of unsaturated fatty acids” was the most significantly enriched (P < 0.001) (Figure 3B) and contained seven differential metabolites that were all significantly upregulated (Table 2). Their identities were further validated by LC-MS/MS mirror plots comparing experimental spectra with authentic standards (Supplementary Figure 5).

Table 2 Differential Metabolites Participating in the Biosynthesis of Unsaturated Fatty Acids

Figure 3 Investigation of differential metabolic pathways. Distribution of differential features in metabolic pathways (A) and KEGG pathway enrichment analysis of the dysregulated pathways (B).

Association of Seven Key Metabolites in Biosynthesis of UFAs with ESCC Clinical Stages and HCs

We further analyzed the association of seven pivotal metabolites involved in UFA biosynthesis with ESCC clinical stages (stage I–II and stage III) and HCs. The Kruskal–Wallis H-test with Bonferroni-corrected post-hoc pairwise comparisons was adopted for intergroup analysis across the three cohorts. Relative to HCs, the levels of cis-11-eicosenoic acid, cis-11,14-eicosadienoic acid, docosapentaenoic acid, erucic acid, linoleic acid, oleic acid, and palmitic acid were all significantly elevated in both early-stage (stage I–II) and locally advanced (stage III) groups (all P < 0.05). No significant differences in these seven metabolite levels were observed between early-stage and locally advanced ESCC patients (all P > 0.05) (Figure 4A–G and Table 3). These findings indicated that dysregulated UFA biosynthesis occurs at an early stage of ESCC tumorigenesis and persists during disease progression, suggesting its potential as an early diagnostic biomarker.

Table 3 Comparison of Differential Metabolites Involved in Unsaturated Fatty Acid Biosynthesis Among Three Groups

Figure 4 Comparison of the seven differential metabolites across clinical stages. Box plots with overlaid jitter points show the levels of cis-11-eicosenoic acid (A), cis-11,14-eicosadienoic acid (B), docosapentaenoic acid (C), erucic acid (D), linoleic acid (E), oleic acid (F), and palmitic acid (G) in the HC group, clinical stage I–II ESCC, and clinical stage III ESCC. T1–2, patients with clinical stage I–II ESCC (n=26); T3, patients with clinical stage III ESCC (n=44); *P<0.05, **P<0.01, ***P<0.001.

Construction of a Diagnostic Model Distinguishing ESCC Patients from HCs

The diagnostic potential of the seven key differential metabolites in ESCC diagnosis was assessed through receiver operating characteristic (ROC) curve analysis. The AUC values for individual metabolites were as follows: cis-11-eicosenoic acid (AUC = 0.76, 95% CI 0.681–0.842), cis-11,14-eicosadienoic acid (AUC = 0.77, 95% CI 0.692–0.854), docosapentaenoic acid (AUC = 0.71, 95% CI 0.620–0.797), erucic acid (AUC = 0.74, 95% CI 0.660–0.824), linoleic acid (AUC = 0.76, 95% CI 0.675–0.841), oleic acid (AUC = 0.78, 95% CI 0.696–0.856), and palmitic acid (AUC = 0.72, 95% CI 0.633–0.804) (Figure 5A–G). To improve diagnostic performance, a combined model incorporating all seven biomarkers was established. The AUC for this model was 0.804 (95% CI 0.729–0.880), demonstrating superior discriminatory efficiency (Figure 5H). The seven-metabolite panel achieved favorable diagnostic efficacy for ESCC, indicating its promise as a novel noninvasive serum biomarker signature.

Figure 5 AUC values and box plots of the seven key differential metabolites. AUC values and box plots for cis-11-eicosenoic acid (A), cis-11,14-eicosadienoic acid (B), docosapentaenoic acid (C), erucic acid (D), linoleic acid (E), oleic acid (F), and palmitic acid (G), respectively. AUC of the combined diagnostic model incorporating the seven metabolites (H).

Discussion

China records the highest global incidence of ESCC, which exhibits striking regional clustering, particularly in central and northern provinces such as Hebei.^26,27 Traditional risk factors (smoking, alcohol intake, dietary habits) are well-established.²⁸ Advances in omics technologies have confirmed aberrant epigenetic regulation as a key molecular mechanism underlying ESCC onset and progression.²⁹ Such progress improves the mechanistic understanding of ESCC etiology and enables systematic dissection of its molecular regulatory network through systems biology and omics strategies. As a core downstream omics approach of systems biology that closely mirrors disease phenotypes, metabolomics captures metabolic changes driven by host genetics and environmental exposures, and acts as a key platform for deciphering ESCC molecular features and discovering specific metabolic biomarkers.³⁰ Therefore, we conducted this untargeted serum metabolomics study focusing on high-incidence regions of ESCC.

Our results showed that serum metabolomic profiles were markedly perturbed in ESCC patients relative to HCs. Among all differentially regulated metabolite classes, lipids and lipid-like molecules accounted for the largest proportion (16.1%). Enrichment analysis revealed that biosynthesis of UFAs was the most significantly altered pathway (P < 0.001) and harbored seven key differential metabolites. Notably, these seven fatty acids were rigorously validated via LC–MS/MS mirror spectral matching between serum samples and authentic chemical standards (Supplementary Figure 5), confirming the accuracy and reliability of metabolite identification in this study. All seven UFAs were significantly upregulated in both early-stage (Stage I–II) and locally advanced (Stage III) ESCC patients compared with HCs, with no obvious differences between the two ESCC subgroups (all P > 0.05). Each metabolite alone showed moderate diagnostic potential (AUC > 0.7). The combined diagnostic model based on the seven metabolites achieved an AUC of 0.804 (95% CI: 0.729–0.880), indicating promising value for early ESCC screening. These findings highlight a critical role of dysregulated lipid metabolism in ESCC pathogenesis,³¹ consistent with the well-recognized functions of lipids in signal transduction, membrane homeostasis, and energy metabolism.³² In the tumor microenvironment, cancer cells frequently reprogram lipid metabolism to sustain malignant phenotypes such as excessive proliferation and invasion, especially under nutrient deprivation.²³ Accordingly, perturbed lipid metabolism is closely implicated in the initiation and progression of ESCC,^12,33 as well as other malignancies.^10,34

Consistent with the aforementioned enrichment analysis, UFA biosynthesis was identified as the most significantly altered pathway, in which seven metabolites were upregulated. Of note, all differential fatty acids (FAs) identified in this study were UFAs with the only exception of palmitic acid, a saturated FA. This metabolic pattern is highly consistent with the characteristic metabolic reprogramming of cancer.³⁵ UFAs are essential for maintaining membrane fluidity, sustaining rapid cell proliferation, and regulating redox balance in cancer cells.³⁶ In contrast, palmitic acid serves as a central intermediate in de novo FA synthesis and provides precursors for elongation and desaturation into unsaturated FAs.^36,37 The dominance of unsaturated FAs thus reflects the adaptive metabolic state in ESCC, which facilitates membrane biosynthesis, signal transduction, and tumor progression.^35,36 Among these biologically and statistically significant differential metabolites, several have been previously reported in serum or plasma metabolomic studies of ESCC. For instance, elevated linoleic acid is a consistent finding across multiple cohorts.^12,33,38 Notably, alterations in linoleic acid metabolism have also been implicated in various inflammation-mediated cancers, particularly those involving immune damage and cell proliferation.^39,40 Mechanistically, studies indicate that cytochrome P450–mediated lipid oxidation of linoleic acid may contribute to the generation of bioactive lipid mediators and oxidative stress, thereby further amplifying tumor-promoting inflammatory signaling. These bioactive oxidized lipids exhibit pro-inflammatory and pro-angiogenic properties, which have been shown to accelerate malignant tumor proliferation and enhance metastatic potential.⁴⁰ In line with these findings, other metabolites, including oleic acid and palmitic acid, have also been documented to be markedly increased in the serum or plasma of ESCC patients.^22,41,42

The robustness of our finding regarding dysregulated UFA biosynthesis is supported by consistent results across previous studies using diverse sample types, including feces, tissues, and cells.^43,44 Intriguingly, prior serum metabolomics studies have further associated ESCC progression with enriched glycerophospholipid metabolism,²³ wherein UFAs act as major structural components.⁴⁵ As key elements of phospholipids, triglycerides, and sphingolipids, UFAs regulate core cellular functions such as membrane homeostasis and signal transduction, providing a functional basis for their role in cellular pathological states.⁴⁵ From the perspective of the overall FA metabolic network, cellular FA uptake, de novo synthesis, and oxidation collectively regulate key oncogenic events, including angiogenesis, metastasis, and apoptosis evasion.⁴⁶ Central to this network is FA synthesis, which not only converts precursors into energy storage, membrane components, and signaling molecules to support cancer cell survival,⁴⁷ but also exemplifies the broader metabolic reprogramming increasingly recognized in tumorigenesis. For instance, FAs have been identified as key mediators in cancer progression and metastasis by remodeling the tumor microenvironment.⁴⁸ Dysregulated FA metabolism sustains abnormal cancer cell proliferation,⁴⁹ and metabolic studies in early-stage EC patients have revealed characteristic patterns of FA metabolic reprogramming.⁵⁰

Notably, the predominant enrichment of differential metabolites and the top altered pathway converge on UFA biosynthesis rather than other canonical metabolic pathways, which can be explained by distinct biological and population-specific mechanisms. First, ESCC is a typical inflammation-driven malignancy, and long-term chronic mucosal irritation in high-incidence areas continuously activates lipid metabolic reprogramming, which preferentially upregulates FA desaturation to support rapid membrane synthesis, inflammatory microenvironment remodeling, and sustained tumor cell proliferation.¹² Second, among all lipid subtypes, UFAs are the core structural components of cell membrane phospholipids and the key precursors of lipid signaling molecules; thus, cancer cells selectively enhance the biosynthesis of UFAs to maintain membrane fluidity, redox balance, and oncogenic signal transduction, leading to the consistent upregulation of oleic acid, linoleic acid, palmitic acid, and other key metabolites observed in this study.⁵¹ Third, compared with other metabolic pathways, UFA metabolism is highly sensitive to regional living patterns, long-term dietary habits, and chronic inflammatory stimulation in high-incidence regions of northern China, which explains why the majority of significant metabolic alterations in this ESCC cohort specifically converged on UFA biosynthesis rather than amino acid, glucose, or other energy metabolism pathways.¹² Collectively, this inherent tumor metabolic dependency and regional population metabolic characteristics jointly lead to the dominant dysregulation of UFA metabolism in ESCC, further supporting that this pathway is not only a statistical enrichment result but also a core pathogenic metabolic signature of ESCC in high-risk populations.^12,51

Despite the promising diagnostic performance of the 7-metabolite panel (AUC = 0.804), several limitations should be acknowledged. First, this was a single-center study with a relatively small sample size, which limits the generalizability of our findings. Second, no multiple testing correction was applied during differential metabolite selection, which may increase false positive rates. Third, the OPLS-DA model carries a potential risk of overfitting and requires further external validation. In addition, serum metabolomics reflects systemic rather than tissue-specific metabolic status. Pathway enrichment results should be interpreted with caution because serum metabolites (as well as plasma metabolites) may be affected by dietary intake, lipid mobilization, or degradation rather than de novo biosynthesis in tumor cells.⁵² Finally, this cross-sectional study only demonstrates correlations, not causal relationships between UFA metabolism and ESCC pathogenesis.

Future studies should first validate this 7-UFA metabolite panel in large, multicenter cohorts to improve generalizability and explore its prognostic value in ESCC. More stringent statistical thresholds or multiple testing correction should be applied to reduce false positives, and independent external validation will help mitigate OPLS-DA model overfitting. Further mechanistic investigations using tissue metabolomics and isotope tracing will establish causal relationships, while multi-omics and functional experiments will elucidate the upstream regulatory networks.

Conclusion

In this study conducted in Hebei, a high‑incidence region of ESCC in northern China, serum untargeted metabolomics using UHPLC‑Orbitrap‑MS/MS revealed significant metabolic alterations in ESCC patients compared with healthy controls. Lipids and lipid‑like molecules were the most dysregulated class. The biosynthesis of UFAs was the top enriched pathway (P < 0.001), involving seven key metabolites that were all significantly upregulated. Compared with HCs, these seven metabolites were elevated in both early‑stage (stage I–II) and locally advanced (stage III) ESCC patients, with no significant difference between the two patient groups. A diagnostic model combining these seven metabolites achieved an AUC of 0.804 (95% CI: 0.729–0.880), indicating favorable discriminatory performance. These findings provide novel insights into the metabolic characteristics of ESCC in high-risk populations and offer a promising non-invasive diagnostic biomarker panel for early ESCC screening.