This article is mentioned in:

Introduction

Upper gastrointestinal (UGI) tumors, characterized by high morbidity and mortality rates, present significant global health challenges, especially esophageal cancer (EC; 510,716 new cases in 2022) and gastric cancer (GC; ~1 million new cases in 2022) (1). The prognosis for patients with UGI cancer remains poor in multiple countries, primarily because of insufficient screening initiatives (2,3). Elucidating the mechanisms that initiate and advance UGI cancers is essential to develop successful prevention and therapeutic approaches. Growing evidence shows that metabolites, small molecules intermediately produced during multiple cellular metabolic reactions, are implicated in the pathways leading to UGI tumors. Progress in molecular biology and the introduction of diverse omics techniques have markedly advanced epidemiological studies at the molecular level in this field (4).

Metabolic imbalance is increasingly recognized as a pivotal factor in the development of UGI tumors (5). Beyond changes in glucose metabolism, exemplified by the well-documented Warburg effect, disrupted metabolism has been reported in nucleotides, lipids and amino acids in both laboratory and clinical studies (6–8). Metabolites are a group of end products arising from complex interactions between inherent metabolism, genetic predispositions and environmental influences. High-throughput metabolomics enables comprehensive identification and quantification of a vast array of low molecular weight metabolites (<1,000 Da) within a single sample. This method is instrumental in identifying novel biomarkers and providing insights into the mechanisms of cancer causation (9,10). In addition, high-throughput metabolomics facilitates the discovery of new preventive measures and therapeutic targets (11). Previous research has analyzed a wide range of metabolites in UGI tumors using human samples such as urine, plasma and tissue (12,13). Although substantial efforts have been made in the field of UGI cancer metabolomics (14,15), these studies are predominantly descriptive. With the increase in research over recent years, there is a critical need for a comprehensive analysis to enhance our understanding of metabolomic profiles in UGI cancer, aiming to identify specific metabolites and consistently involved pathways.

Numerous cross-sectional, cohort, or retrospective studies have examined the association between metabolites and UGI tumors. However, due to their observational nature, these studies were restricted to identifying the correlations rather than the causations (16,17). While randomized controlled trials (RCTs), could potentially establish causality, interventions designed to manipulate metabolites are generally neither feasible nor ethical, thus limiting their ability to determine causal relationships. Given the constraints of both observational and interventional studies, Mendelian randomization (MR) in human genetics offers a powerful tool for rigorously investigating potential causal associations between elevated metabolite levels and UGI tumors (16,17). Traditional observational epidemiological approaches are susceptible to biases, rendering the association results from these studies prone to confounding factors (such as sex and age) and reverse causality (such as lifestyle changes due to UGI cancer), which lead to unreliable causal inferences (18). MR has become a primary genetic epidemiological research method, using genetic variations such as single nucleotide polymorphisms (SNPs) as instrumental variables to examine exposure factors and conclude causal relationships between exposures and outcomes. Due to the principle of randomly distributing alleles to offspring, the causal association estimates obtained from MR studies are not influenced by confounding factors. Additionally, since genes are determined before birth and cannot be altered by diseases, MR research effectively controls for the effects of reverse causality (19).

In the present study, MR was employed to explore the roles of metabolites in both histophysiological and pathophysiological processes leading to UGI tumors, using insights from a recent statistical analysis based on metabolite-focused Genome-Wide Association Study (GWAS) data (20). The present study aimed to investigate the causal association between 1,400 metabolites and UGI cancer, particularly focusing on their roles in tumor initiation, progression and treatment resistance. Moreover, functional experiments were conducted to further validate the findings from the MR analysis. The present study sought to provide insights that could enhance future metabolomic methodologies and advance etiological research, thereby supporting the development of precise prevention and innovative therapeutic strategies. It is anticipated that the findings will contribute to the creation of personalized treatment plans that target the specific metabolic vulnerabilities of UGI tumors.

Materials and methods

Study design

The cause-and-effect association between 1,400 metabolites and UGI tumors was assessed using two-sample MR analyses. MR utilized genetic variations as proxies for metabolic risk factors. To ensure reliable causal inference, the instrumental variables (IVs) used in MR were required to satisfy three critical assumptions: i) Genetic variation must be directly associated with the exposure (metabolites); ii) genetic variant should not be linked to any confounders that might affect the exposure and the outcome; and iii) the genetic variation should affect the outcome solely through the exposure, without any alternative pathways involved. The present study excluded points with P<1×105 from the outcome when conducting MR analysis to satisfy the second assumption. In addition, it used methodologies such as MR Egger (21) and MR-pleiotropy residual sum and outlier (PRESSO) to test and found no pleiotropy in the results for the third assumption. Furthermore, before conducting MR analysis, the present study had already searched for the corresponding SNPs on the GWAS catalog and removed the sites with pleiotropy. Although a number of measures had been taken to avoid environmental and genetic factors, it was hypothesized that the external environment can still have an influence. Finally, an experiment was performed to validate the present results (Fig. 1).

Figure 1. Study design flowchart. The first assumption was that the IVs were strongly related to the exposure. The second assumption specified that the IVs are not associated with any confounders. The third assumption established that the IVs influence the outcome solely through exposure. IVs, instrumental variables; GWAS, Genome-Wide Association Study; SNPs, single nucleotide polymorphisms; MR, Mendelian randomization; PRESSO, pleiotropy residual sum and outlier; GI, gastrointestinal; EC, esophageal cancer; GC, gastric cancer.

Data sources for exposure and outcome

The statistical summary of GWAS data for each metabolite was taken from the European GWAS website (http://www.ebi.ac.uk/gwas/; accession no. GCST90199621-90201020) (20). Cancer-specific keywords were applied to search relevant data for each cancer type (https://gwas.mrcieu.ac.uk/). Specifically, ukb-saige-150 was selected for EC identification, while ukb-saige-151 was used for GC screening. The UK Biobank, used for this data retrieval, is a substantial biomedical database and research resource aimed at enabling the exploration of genetic, environmental and lifestyle factors influencing various diseases and health outcomes (22). This database includes health and genetic information from >500,000 participants in the UK, making it one of the most comprehensive biomedical resources of its type.

GWAS studies aim to identify common genetic variants linked to complex disorders, ultimately guiding the development of translational prevention and treatment strategies. Given their extensive coverage of common SNPs and relative cost-effectiveness, GWAS has emerged as a valuable tool for clinical and commercial genetic testing. Essentially, GWAS acts as a potent tool that has greatly enhanced our knowledge of the genetic foundations of complex traits and diseases, establishing a robust foundation for future research and medical innovations. GWAS studies typically involve collecting and analyzing DNA samples from individuals to identify genetic variations associated with particular traits or diseases. The biological material collected for GWAS usually comprises DNA extracted from blood, saliva, or other tissues from participants. Researchers extract DNA from these samples and employ genotyping techniques to detect genetic variations, such as SNPs that might be linked to the trait or disease of interest. In the current study, data were obtained from the UK Biobank (www.ukbiobank.ac.uk) using specific ICD-10 (International Classification of Diseases 10th revision; accession nos. ukb-saige-150), and (GC: ukb-saige-151) IDs for each type of cancer. This involved analyzing the association between 1,400 types of metabolites and UGI tumors. Specifically, data included 394,092 European individuals (720 case patients and 393,372 control participants) for EC and 393,926 European individuals (554 case patients and 475,308 control participants) for GC. The association between metabolites and each cancer type was subsequently analyzed according to these IDs.

Instrument selection

Considering the extensive number of SNPs demonstrating genome-wide significance (P<5×10−8) for metabolite traits, stricter correlation thresholds (P<5×10−9) were implemented for selecting genetic IVs. These IVs were categorized based on the reference panel of Linkage Disequilibrium from the 1,000 Genomes Project (23,24), with a threshold of R2<0.001 at a distance of 1,000 kb. Due to the relatively small size of the GWAS dataset for metabolites, a cutoff (P=5×10−8) and a less stringent clustering threshold (R2<0.001 at a distance of 1,000 kb) were employed (19). To ensure the reliability of the tools used, IVs with F>10 were selected and identified as strong elements for subsequent analyses. Next, these IVs were obtained from the summary statistics pertaining to UGI cancer outcomes, excluding any that displayed potential pleiotropic effects (P<10−5) on UGI cancer, in line with methodologies from previous research (25). To maintain consistency in this analysis, discrepancies in SNPs between the exposure and outcome datasets were synchronized to ensure uniform effect estimates for the same effect allele (26).

Cell lines

The esophageal cancer cell line (KYSE150) and gastric cancer cell line (HGC27) selected for the experiment were purchased from the cell bank of the Chinese Academy of Sciences.

Main reagents and instruments

Cell culture medium and reagents included: Serum-free DMEM (cat. no. PM150210; Procell Life Science & Technology Co., Ltd.), RPMI 1640 basic medium (cat. no. PM150110; Procell Life Science & Technology Co., Ltd.), fetal bovine serum (FBS; cat. no. 164210-50; Procell Life Science & Technology Co., Ltd.), pancreatic enzyme (cat. no. PB180226; Procell Life Science & Technology Co., Ltd.), fructosyllysine (cat. no. HY-129380; MedChemExpress), 2′-deoxyuridine (cat. no. HY-D0186; MedChemExpress), cytidine (cat. no. HY-D0158; MedChemExpress), carnitine (cat. no. HY-B0399A; MedChemExpress), Cell Counting Kit-8 (CCK-8; Shanghai Biyuntian Biotechnology Co., Ltd.), ELISA reader (BioTek; Agilent Technologies, Inc.) and an inverted microscope (Nikon Corporation).

Cell culture

KYSE150 cells and HGC27 cells were incubated in DMEM and RPMI 1640 medium, respectively. The media were supplemented with 10% FBS and 100 U/ml penicillin and streptomycin (Gibco; Thermo Fisher Scientific, Inc.). The cells were maintained in an incubator set at 37°C with a 5% CO2 atmosphere. Upon reaching 80–90% confluence, the cells were enzymatically dissociated for subculturing.

Detection of cell proliferation ability

Cells were digested with 0.25% trypsin and resuspended for counting when entering the logarithmic growth phase. The cells were then seeded in a 96-well plate at a density of 3,000 cells/well. After attachment and morphological expansion, four common metabolites identified from MR results, namely fructosyllysine, 2′-deoxyuridine, cytidine and carnitine, were added for treatment at 37°C for 1.5 h in darkness. The concentrations of the metabolites were 0, 50 and 100 µM, with triplicate wells for each concentration. Subsequently, each well received 10 µl of CCK-8 reagent and was incubated at 37°C for 1.5 h in darkness. The optical density was measured at 450 nm using an ELISA reader at time intervals of 0, 24, and 48 h following the introduction of the metabolites.

Cell scratch assay

Cells were digested with 0.25% trypsin and counted during the logarithmic growth phase. The cells were then plated in a 6-well plate and each well contained 1.0×105 cells. Once the cells adhered and expanded morphologically, reaching ~80% confluence, a scratch was made on the cell monolayer using a 10-µl pipette tip. After making the scratch, the wells were rinsed twice with PBS to remove any detached cells and then a serum-free medium and varying concentrations of metabolites was added for continued incubation. Images of the scratch were captured at 0, 24 and 48 h after adding the metabolites using a microscope (Nikon Ts2FL inverted microscope; Nikon Corporation). The area of the scratch was quantified using ImageJ software (version 2023; National Institutes of Health). To assess the rate of cell migration and healing, the formula used for calculating wound healing rate was [(Original scratch area-Final scratch area)/Original scratch area] ×100%.

Statistical analysis

After extracting the data concerning SNPs associated with metabolites, including details such as effect alleles and their corresponding β values. The formula established previously (27) was used to calculate the genetic variance for each metabolite. All data were presented and the number of replicates performed following the reported formula (20). In the present study, a range of genetic variants were employed as IVs, rather than relying solely on an allele score. This approach was selected to thoroughly examine key assumptions, uncover potential pleiotropy and enhance the sensitivity efficacy of the multivariable MR analyses (28). A total of four distinct MR methodologies, including the inverse variance weighted (IVW; random-effects model), weighted median, MR-Egger and MR-PRESSO, were used to evaluate the consistency of the current findings under different assumptions about heterogeneity and pleiotropy. The IVW method, employing a random-effects model, served as the primary analysis framework for all four sets of IVs. Heterogeneity was quantified using Cochran's Q statistic.

The present study also included analyses with more stringent conditions. Assuming that all genetic variants are valid, the IVW method might be subject to bias if a considerable number of SNPs are influenced by horizontal pleiotropy (29). By contrast, the weighted median approach, effective when <50% of variants exhibit horizontal pleiotropy, operated under the assumption that the majority of genetic variants were valid (30). For situations where >50% of variants were affected by horizontal pleiotropy, the strength of the present genetic tools was evaluated through F statistics, considering a mean F<10 to be indicative of weak IVs (31).

Furthermore, the MR-Egger method was applied to assess potential directional pleiotropy, where a significant intercept would indicate a violation of IV assumptions, suggesting directional pleiotropy (32). In addition, the MR-PRESSO method was implemented to minimize heterogeneity in causal effect estimates by excluding disproportionately influential SNPs (NbDistribution=1,500) (33). In addition, Steiger-filtering was used to identify and exclude genetic variants more strongly linked to the outcome than to the exposure, indicative of potential reverse causality (34).

All statistical analyses in the present study were performed using R (R Foundation version 4.3.1; 2023 version R-project.org/) and specific R packages (‘TwoSampleMR’ and ‘MR’) (35) designed for MR analysis (36). The TwoSampleMR package facilitated the provision of causal estimates across the four MR models: IVW, weighted median, MR-Egger and MR-PRESSO. Analysis of functional experiment data was conducted using the software ImageJ (version 2023; National Institutes of Health) and GraphPad Prism 8 (GraphPad; Dotmatics). Unpaired Student's t-tests were used to calculate differences between the two groups. For group comparisons, one-way ANOVA and Bonferroni post hoc analysis were used. Data are expressed as the mean ± SD. P<0.05 was considered to indicate a statistically significant difference.

Results

Causal estimation between metabolites and UGI cancer

The present study first assessed the causal impact of various metabolites on EC using the IVW method for a two-sample MR analysis. The assessment revealed that 57 metabolites were significantly associated with EC (Table I). Key findings highlighted notable associations with hub metabolites. Specifically, fructosyllysine levels showed a strong association with increased EC risk [odds ratio (OR)=1.450, 95% confidence interval (CI)=1.087–1.934, P=0.011]. Additionally, the ratio of 2′-deoxyuridine to cytidine (OR=1.464, 95% CI=1.111–1.929, P=0.007) and carnitine to protonylcarnitine (C3) (OR=0.655, 95% CI=0.499–0.861, P=0.002) were also significantly linked to EC risk (Fig. 2). Furthermore, a sensitivity analysis was conducted. Despite observing some heterogeneity and significant results from Cochran's Q test (P<0.05), the causal estimates remained robust under the random-effects IVW model. The P-values for the MR-Egger intercept were greater than 0.05, suggesting that no significant pleiotropy effects were found (Tables SI and SII). Additionally, the data were further assessed using scatter (Fig. 3A-C), funnel (Fig. 3D-F) and leave-one-out (Fig. 4A-C) plots and the potential influence of outliers and horizontal pleiotropy on the identified hub metabolites were excluded.

Figure 2. Causal estimation between metabolites and esophageal cancer. Inverse variance weighted was selected as a primary method. P<0.05 was considered to indicate a statistically significant difference. OR>1 indicated a risk factor, while OR<1 signified a protective factor. OR, odds ratio.

Figure 3. Scatter plot demonstrating the genetic associations of three metabolites with the risk of EC. (A) Fructosyllysine levels in EC. (B) 2′-Deoxyuridine to cytidine ratio in EC. (C) Carnitine to protonylcarnitine ratio in EC. The funnel plot represents instrumental variables for each significant causal relation between metabolites and EC. (D) Fructosyllysine levels, (E) 2′-deoxyuridine to cytidine ratio in EC and (F) carnitine to protonylcarnitine ratio in EC. EC, esophageal cancer; MR, Mendelian randomization; SNPs, single nucleotide polymorphisms.

Figure 4. Leave-one-out plot demonstrating the genetic associations of three metabolites with the risk of EC. (A) Fructosyllysine levels, (B) 2′-Deoxyuridine to cytidine ratio and (C) Carnitine to protonylcarnitine ratio in EC. EC, esophageal cancer; MR, Mendelian randomization.

Table I. Causal association between all metabolites and esophageal cancer. Inverse variance weighted was chosen as the primary method. P<0.05 was considered to indicate a statistically significant difference. OR >1 indicated a risk factor.

Table I.

Causal association between all metabolites and esophageal cancer. Inverse variance weighted was chosen as the primary method. P<0.05 was considered to indicate a statistically significant difference. OR >1 indicated a risk factor.

Metabolite	Method	Number of single nucleotide polymorphisms	P-value	OR	OR 95% lower CI	OR 95% upper CI
Maltotriose levels	IVW	24	0.027	1.28	1.027	1.603
Tartarate levels	IVW	22	0.029	0.73	0.557	0.969
1-stearoyl-2-arachidonoyl-GPI (18:0/20:4) levels	IVW	24	0.004	1.34	1.095	1.654
Stearidonate (18:4n3) levels	IVW	27	0.046	1.31	1.004	1.726
Hexanoylglycine levels	IVW	27	0.036	1.20	1.011	1.428
Beta-hydroxyisovaleroylcarnitine levels	IVW	36	0.027	0.76	0.605	0.97
1-ribosyl-imidazoleacetate levels	IVW	37	0.010	0.75	0.605	0.936
Thymol sulfate levels	IVW	20	0.048	1.34	1.002	1.808
2-hydroxyhippurate levels	IVW	20	0.028	0.70	0.520	0.964
2R,3R-dihydroxybutyrate levels	IVW	32	0.014	0.75	0.599	0.946
1-(1-enyl-palmitoyl)-GPC (p-16:0) levels	IVW	29	0.021	1.32	1.041	1.687
Docosadioate (C22-DC) levels	IVW	26	0.044	0.73	0.547	0.992
N-palmitoylglycine levels	IVW	22	0.021	1.28	1.036	1.582
Fructosyllysine levels	IVW	27	0.011	1.45	1.087	1.934
Sphingomyelin (d18:2/14:0, d18: 1/14:1) levels	IVW	24	0.019	1.49	1.067	2.098
Methyl-4-hydroxybenzoate sulfate levels	IVW	22	0.019	1.43	1.060	1.936
1,2,3-benzenetriol sulfate (2) levels	IVW	24	0.003	0.65	0.491	0.872
1,2-dilinoleoyl-GPC (18:2/18:2) levels	IVW	17	0.010	0.72	0.563	0.927
1-palmitoyl-2-stearoyl-gpc (16:0/18:0) levels	IVW	32	0.018	1.29	1.043	1.598
1-stearoyl-2-arachidonoyl-GPE (18:0/20:4) levels	IVW	32	0.006	1.24	1.062	1.450
1-(1-enyl-palmitoyl)-2-arachidonoyl-GPE (p-16:0/20:4) levels	IVW	22	0.017	1.40	1.060	1.850
1-palmitoyl-2-arachidonoyl-GPE (16:0/20:4) levels	IVW	19	0.044	1.18	1.003	1.398
Docosahexaenoylcholine levels	IVW	20	0.049	0.72	0.529	0.998
Sphingomyelin (d18:2/21:0, d16: 2/23:0) levels	IVW	35	0.022	1.34	1.043	1.744
Sphingomyelin (d17:1/14:0, d16: 1/15:0) levels	IVW	26	0.010	1.43	1.088	1.889
2,2′-Methylenebis(6-tert-butyl-p-cresol) levels	IVW	21	0.048	0.75	0.566	0.998
3-hydroxy-2-methylpyridine sulfate levels	IVW	21	0.030	0.72	0.548	0.969
4-acetylcatechol sulfate (1) levels	IVW	19	0.020	0.68	0.502	0.943
Hydroxy-N6,N6,N6-trimethyllysine levels	IVW	27	0.029	1.35	1.031	1.789
Alpha-tocopherol levels	IVW	28	0.016	1.40	1.063	1.862
Beta-hydroxyisovalerate levels	IVW	28	0.033	1.31	1.022	1.688
Eicosapentaenoate (EPA; 20:5n3) levels	IVW	27	0.0005	1.54	1.206	1.983
Cholesterol levels	IVW	21	0.018	1.44	1.064	1.964
Gluconate levels	IVW	24	0.007	0.71	0.553	0.911
Deoxycholate levels	IVW	27	0.032	1.32	1.023	1.717
Ornithine levels	IVW	28	0.013	0.72	0.563	0.936
Dimethylglycine levels	IVW	33	0.023	1.25	1.031	1.525
Tyrosine levels	IVW	29	0.014	0.71	0.539	0.934
N-stearoyl-sphinganine (d18:0/18:0) levels	IVW	16	0.003	0.64	0.478	0.867
X-12101 levels	IVW	25	0.007	0.74	0.604	0.926
X-22834 levels	IVW	22	0.041	0.73	0.549	0.988
X-23639 levels	IVW	23	0.043	1.44	1.011	2.057
X-23648 levels	IVW	19	0.024	1.41	1.045	1.927
X-23641 levels	IVW	30	0.001	0.71	0.580	0.872
X-23665 levels	IVW	20	0.048	0.70	0.495	0.997
X-23739 levels	IVW	28	0.011	0.75	0.604	0.938
X-25810 levels	IVW	36	0.029	0.77	0.619	0.974
3-methylcytidine levels	IVW	20	0.028	0.83	0.717	0.981
Adenosine 5′-diphosphate to Adenosine 5′-monophosphate ratio	IVW	26	0.018	1.27	1.042	1.557
Cortisone to cortisol ratio	IVW	20	0.008	0.64	0.46	0.892
Phosphate to N-palmitoyl-sphingosine (d18:1 to 16:0) ratio	IVW	24	0.049	0.75	0.571	1.000
Adenosine 5′-monophosphate to arginine ratio	IVW	17	0.032	0.65	0.44	0.964
2′-deoxyuridine to cytidine ratio	IVW	30	0.007	1.46	1.089	1.751
Acetylcarnitine to propionylcarnitine ratio	IVW	19	0.038	0.71	0.522	0.982
Carnitine to propionylcarnitine ratio	IVW	24	0.002	0.66	0.499	0.861
Alpha-ketoglutarate to trans-4-hydroxyproline ratio	IVW	20	0.020	1.42	1.055	1.936
Adenosine 5′-monophosphate to urate ratio	IVW	27	0.020	1.41	1.054	1.896

[i] OR, odds ratio; IVW, inverse variance weighted; CI, confidence interval.

Causal estimation between metabolites and GC

Using the IVW method as the primary analytic approach, a two-sample MR analysis was adopted to investigate the causal effect of GC on metabolites. The results identified associations with 58 metabolites (Table II). Notably, the GC risk was observed to be significantly associated with elevated levels of fructosyllysine (OR=1.70, 95% CI=1.240–2.346, P=0.001), the 2′-deoxyuridine to cytidine ratio (OR=1.464, 95% CI=1.094–1.957, P=0.010) and the C3 ratio (OR=0.664, 95% CI=0.486–0.906, P=0.010; Fig. 5). Although sensitivity analysis highlighted some heterogeneity, Cochran's Q yielded P<0.05 and the causality estimates were satisfactory when using a random-effects IVW approach. The P-values for the MR-Egger intercept were >0.05, indicating that no significant pleiotropy effects were found (Tables SIII and SIV). Additionally, scatter (Fig. 6A-C), funnel (Fig. 6D-F) and leave-one-out (Fig. 7A-C) plots excluded the likelihood inflicting by potential outliers and horizontal pleiotropy on the identified hub metabolites that were excluded.

Figure 5. Causal estimation between metabolites and gastric cancer. Inverse variance weighted was selected as a primary method. P<0.05 was considered to indicate a statistically significant difference. OR>1 indicated a risk factor, while OR<1 signified a protective factor. OR, odds ratio; MR, Mendelian randomization.

Figure 6. Scatter plot demonstrating the genetic associations of three metabolites with the risk of GC. (A) Fructosyllysine levels, (B) 2′-Deoxyuridine to cytidine ratio and (C) Carnitine to protonylcarnitine ratio in GC. Funnel plot representing instrumental variables for each significant causal association between metabolites and GC. (D) Fructosyllysine levels, (E) 2′-Deoxyuridine to cytidine ratio and (F) Carnitine to protonylcarnitine ratio in GC. GC, gastric cancer; EC, esophageal cancer.

Figure 7. Leave-one-out plot demonstrating the genetic associations of three metabolites with the risk of GC. (A) Fructosyllysine levels, (B) 2′-Deoxyuridine to cytidine ratio and (C) Carnitine to protonylcarnitine ratio in GC. GC, gastric cancer; MR, Mendelian randomization.

Table II. The causal association between all metabolites and gastric cancer. Inverse variance weighted was chosen as the primary method. P<0.05 was considered to indicate a statistically significant difference. OR >1 indicated a risk factor.

Table II.

The causal association between all metabolites and gastric cancer. Inverse variance weighted was chosen as the primary method. P<0.05 was considered to indicate a statistically significant difference. OR >1 indicated a risk factor.

Metabolite	Method	Number of single nucleotide polymorphisms	P-value	OR	OR 95% lower CI	OR 95% upper CI
Gentisate levels	IVW	27	0.036	1.44	1.024	2.028
1,5-anhydroglucitol (1,5-ag) levels	IVW	32	0.018	1.35	1.052	1.742
2-linoleoylglycerol (18:2) levels	IVW	23	0.016	1.43	1.069	1.921
3-carboxy-4-methyl-5-propyl-2-furanpropanoate (cmpf) levels	IVW	17	0.047	1.48	1.004	2.201
10-nonadecenoate (19:1n9) levels	IVW	13	0.018	0.57	0.363	0.912
Stachydrine levels	IVW	24	0.036	1.41	1.021	1.961
2-palmitoleoyl-GPC (16:1) levels	IVW	26	0.045	1.33	1.006	1.769
Isobutyrylglycine levels	IVW	30	0.027	1.3	1.029	1.666
Glycerophosphoethanolamine levels	IVW	25	0.048	1.34	1.002	1.807
Taurolithocholate 3-sulfate levels	IVW	23	0.018	0.69	0.510	0.939
Glycolithocholate sulfate levels	IVW	23	0.0009	0.59	0.435	0.808
Pregnenolone sulfate levels	IVW	36	0.023	0.72	0.543	0.955
S-methylmethionine levels	IVW	21	0.018	0.68	0.500	0.939
2,3-dihydroxyisovalerate levels	IVW	27	0.002	1.61	1.185	2.212
N-oleoyltaurine levels	IVW	21	0.008	1.46	1.103	1.956
2-aminooctanoate levels	IVW	29	0.037	1.23	1.011	1.495
3-acetylphenol sulfate levels	IVW	21	0.037	1.43	1.020	2.018
N-acetylalliin levels	IVW	30	0.001	0.69	0.557	0.872
Fructosyllysine levels	IVW	27	0.001	1.70	1.200	2.483
N-acetyltaurine levels	IVW	20	0.045	0.71	0.516	0.992
3-hydroxypyridine sulfate levels	IVW	24	0.014	1.50	1.084	2.081
Arabitol/xylitol levels	IVW	24	0.049	0.70	0.493	0.999
Tricosanoyl sphingomyelin (d18: 1/23:0) levels	IVW	29	0.006	0.64	0.469	0.882
1-stearoyl-2-arachidonoyl-GPE (18:0/20:4) levels	IVW	32	0.012	1.33	1.062	1.668
5-hydroxyindole sulfate levels	IVW	19	0.049	1.36	1.001	1.864
Nisinate (24:6n3) levels	IVW	18	0.003	1.51	1.145	2.002
Arachidonoylcarnitine (C20:4) levels	IVW	35	0.004	1.27	1.076	1.508
Heptenedioate (C7:1-DC) levels	IVW	16	0.016	1.69	1.103	2.605
Octadecenedioate (C18:1-DC) levels	IVW	23	0.013	0.75	0.606	0.943
Perfluorooctanoate (PFOA) levels	IVW	23	0.036	0.69	0.491	0.977
Glyco-beta-muricholate levels	IVW	25	0.017	0.76	0.617	0.954
8-methoxykynurenate levels	IVW	26	0.026	1.38	1.038	1.850
2-ketocaprylate levels	IVW	23	0.028	1.37	1.034	1.837
Palmitoyl-sphingosine-phosphoe-thanolamine (d18:1/16:0) levels	IVW	29	0.005	0.66	0.496	0.887
Picolinate levels	IVW	23	0.032	0.74	0.565	0.975
Erucate (22:1n9) levels	IVW	19	0.003	0.58	0.410	0.842
Cys-gly, oxidized levels	IVW	26	0.028	1.33	1.030	1.719
1-palmitoyl-2-linoleoyl-gpc (16:0/18:2) levels	IVW	35	0.040	0.78	0.622	0.989
Eicosapentaenoate (20:5n3) levels	IVW	27	0.040	1.34	1.012	1.791
Stearate (18:0) levels	IVW	21	0.033	0.67	0.469	0.97
Pseudouridine levels	IVW	23	0.004	0.59	0.418	0.85
X-11632 levels	IVW	26	0.014	0.68	0.508	0.928
X-12701 levels	IVW	15	0.006	0.6	0.426	0.871
X-18901 levels	IVW	30	0.006	1.5	1.119	2.018
X-21283 levels	IVW	17	0.007	0.72	0.58	0.917
X-22834 levels	IVW	22	0.026	0.71	0.525	0.961
X-23648 levels	IVW	19	0.028	1.48	1.043	2.113
X-23782 levels	IVW	20	0.027	1.57	1.051	2.365
X-24344 levels	IVW	17	0.006	1.54	1.129	2.116
X-24585 levels	IVW	19	0.013	1.58	1.099	2.29
Oleoyl-linoleoyl-glycerol (18:1 to 18:2) to linoleoyl-arachidonoyl-glycerol (18:2 to 20:4) ratio	IVW	25	0.021	0.80	0.661	0.968
Spermidine to adenosine 5′-diphosphate (ADP) ratio	IVW	25	0.039	0.78	0.624	0.987
Glycine to phosphate ratio	IVW	28	0.024	1.27	1.032	1.565
Adenosine 5′-diphosphate (ADP) to glutamate ratio	IVW	32	0.022	0.77	0.619	0.963
Cortisol to 4-cholesten-3-one ratio	IVW	21	0.020	1.53	1.068	2.192
2′-deoxyuridine to cytidine ratio	IVW	30	0.010	1.46	1.094	1.957
Carnitine to propionylcarnitine (C3) ratio	IVW	24	0.010	0.66	0.486	0.906
Arachidonate (20:4n6) to linoleate (18:2n6) ratio	IVW	26	0.017	1.40	1.062	1.865

[i] OR, odds ratio; IVW, inverse variance weighted.

Hub metabolites affect the proliferation and migration abilities of EC and GC cells

To verify the effects of four metabolites, fructosyllysine, 2′-deoxyuridine, cytidine and carnitine, on the proliferation and migration abilities of EC and GC cells, the present study established in vitro models of EC and GC using KYSE150 and HGC27 cells, respectively. CCK-8 assay showed that fructosyllysine and 2′-deoxyuridine promoted the proliferation of EC and GC cells, while cytidine and carnitine exhibited inhibitory effects on the proliferation of these two tumor cell lines. Thus, the 2′-deoxyuridine to cytidine ratio was associated with the proliferation promotion of EC and GC cells. Scratch assay demonstrated that EC and GC cells treated with fructosyllysine and 2′-deoxyuridine showed significantly faster scratch healing compared with the control group, indicating enhanced migratory capabilities. By contrast, cells treated with cytidine or carnitine displayed notably slower wound healing than those of the control group, suggesting a suppressive effect on cell migration. Thus, the scratch healing speed of EC and GC cells treated with a ratio of 2′-deoxyuridine to cytidine was significantly faster compared with the control group. In summary, the metabolites fructosyllysine and the 2′-deoxyuridine to cytidine ratio were found to enhance the proliferation and migration of EC and GC cells. Conversely, carnitine was observed to inhibit both the proliferation and migration of these tumor cell types (Fig. 8, Fig. 9, Fig. 10, Fig. 11).

Figure 8. Fructosyllysine promotes the proliferation and migration of both EC (KYSE150) and GC (HGC27) cell lines. (A) CCK-8 assay measured the effect of fructosyllysine on the proliferation of KYSE150 cells. (B) CCK-8 assay measured the effect of fructosyllysine on the proliferation of HGC27 cells. Cells were divided into three groups and treated with various concentrations of fructosyllysine (0, 50 and 100 µM). The absorbance values of cells at different time points (0, 24 and 48 h) were measured by CCK-8 assay and the proliferation of cells was evaluated by graphs. (C) Scratch assay determined the influence of fructosyllysine on the migration of KYSE150 cells. (D) Scratch assay determined the influence of fructosyllysine on the migration of HGC27 cells. scale bars, 200 µM. Magnification, ×40. KYSE150 and HGC27 cells were treated with different concentrations of fructosyllysine for 24 and 48 h and the scratch area at different time points was calculated to evaluate the migration ability of the cells. All experiments were repeated three times and the results were expressed as the mean ± SD. ns, no significant difference vs. control. *P<0.05, **P<0.01, ****P<0.001 vs. control. EC, esophageal cancer; GC, gastric cancer; CCK-8, Cell Counting Kit-8.

Figure 9. The 2′-deoxyuridine promotes the proliferation and migration of both EC (KYSE150) and GC (HGC27) cell lines. (A) CCK-8 assay measured the effect of 2′-deoxyuridine on the proliferation of KYSE150 cells. (B) CCK-8 assay measured the effect of 2′-deoxyuridine on the proliferation of HGC27 cells. Cells were divided into three groups and treated with various concentrations of 2′-deoxyuridine (0, 50 and 100 µM). The absorbance values of cells at different time points (0, 24 and 48 h) were measured by CCK-8 assay and the proliferation of cells was evaluated by graphing. (C) Scratch assay determined the influence of 2′-deoxyuridine on the migration of KYSE150 cells. (D) Scratch assay determined the influence of 2′-deoxyuridine on the migration of HGC27 cells. scale bars, 200 µM. Magnification, ×40. KYSE150 and HGC27 cells were treated with different concentrations of 2′-deoxyuridine for 24 and 48 h and the scratch area at different time points was calculated to evaluate the migration ability of the cells. All experiments were repeated three times and the results were expressed as the mean ± SD. ns, no significant difference vs. control. *P<0.05, ***P<0.005, ****P<0.001 vs. control. EC, esophageal cancer; GC, gastric cancer; CCK-8, Cell Counting Kit-8.

Figure 10. Cytidine inhibits the proliferation and migration of both EC (KYSE150 cell) and GC (HGC27 cell) cell lines. (A) CCK-8 assay measured the effect of cytidine on the proliferation of KYSE150 cells. (B) CCK-8 assay measured the effect of cytidine on the proliferation of HGC27 cells. Cells were divided into three groups and treated with various concentrations of cytidine (0, 50 and 100 µM). The absorbance values of cells at different time points (0, 24 and 48 h) were measured by CCK-8 assay and the proliferation of cells was evaluated by graphing. (C) Scratch assay determined the influence of cytidine on the migration of KYSE150 cells. (D) Scratch assay determined the influence of cytidine on the migration of HGC27 cells. scale bars, 200 µM. Magnification, ×40. KYSE150 and HGC27 cells were treated with different concentrations of cytidine for 24 and 48 h and the scratch area at different time points was calculated to evaluate the migration ability of the cells. All experiments were repeated 3 times and the results were expressed as the mean ± SD. ns, no significant difference vs. control. *P<0.05, **P<0.01, ***P<0.005 vs. control. EC, esophageal cancer; GC, gastric cancer; CCK-8, Cell Counting Kit-8.

Figure 11. Carnitine inhibits the proliferation and migration of both EC (KYSE150) and GC (HGC27) cell lines. (A) CCK-8 assay measured the effect of carnitine on the proliferation of KYSE150 cells. (B) CCK-8 assay measured the effect of carnitine on the proliferation of HGC27 cells. Cells were divided into three groups and treated with various concentrations of carnitine (0, 50 and 100 µM). The absorbance values of cells at different time points (0, 24 and 48 h) were measured by CCK-8 assay and the proliferation of cells was evaluated by graphing. (C) Scratch assay determined the influence of carnitine on the migration of KYSE150 cells. (D) Scratch assay determined the influence of carnitine on the migration of HGC27 cells. scale bars, 200 µM. Magnification, ×40. KYSE150 and HGC27 cells were treated with different concentrations of carnitine for 24 and 48 h and the scratch area at different time points was calculated to evaluate the migration ability of the cells. All experiments were repeated 3 times and the results were expressed as the mean ± SD. ns, no significant difference vs. control. *P<0.05, **P<0.01, ***P<0.005, ****P<0.001 vs. control. EC, esophageal cancer; GC, gastric cancer; CCK-8, Cell Counting Kit-8.

Discussion

MR analysis is increasingly recognized for its ability to delineate potential causal relationships between risk factors and diseases (19). The present study identified 57 metabolites associated with EC and 58 with GC. Notably, fructosyllysine levels, 2′-deoxyuridine to cytidine ratio and C3 ratio emerged as common metabolites influencing EC and GC. Among them, fructosyllysine levels and 2′-deoxyridine to cytidine ratio were identified as risk factors for both cancers, while the C3 ratio appeared as protective factors for GC and EC.

Fructosyllysine is a precursor of carboxymethyllysine, an advanced glycation end-product (AGE) presenting in heated foods, which is considered potentially harmful to human health (37). AGEs are produced through non-enzymatic glycation reactions between amino acids and reducing sugars, a process commonly known as the Maillard reaction (38). AGEs exist in both protein-bound and free forms in food, with the majority being absorbed through the gastrointestinal tract. According to previous research, AGEs contribute to the progression of various chronic diseases, including diabetes-related complications, cardiovascular, renal and neurodegenerative disorders, as well as aging (38–41). Carboxymethyllysine, the predominant component of AGEs, is formed through fructosyllysine oxidation (42), which is linked to inflammation and endothelial dysfunction. Notably, chronic inflammation in the gastrointestinal tract is closely associated with cancer (43). AGEs primarily mediate inflammation by binding to the receptor for AGE (RAGE). Expression of RAGE has been detected in human intestinal epithelial cells and colon tissues, where its engagement with ligands enhances the oxidase activity of nicotinamide adenine dinucleotide phosphate, leading to the transcription of NF-κB and the subsequent expression of pro-inflammatory genes, such as IL-6 and TNF-α (44). Thus, an elevated level of fructosyllysine may trigger the inflammatory response in the intestine, potentially inducing cancer over time. In the study by Raupbach et al (45), the expression of RAGE in colon cancer HCT116 cells was shown to be comparable to that in the intestinal epithelium and the absorption efficacy of fructosyllysine by HCT116 cells was largely related to their structural characteristics. However, reports linking fructosyllysine to cancer are limited. In the present study, increased fructosyllysine levels were identified as a risk factor for EC and GC, consistent with previous findings (44,45), although the mechanisms require further exploration, particularly concerning the role of inflammatory activation. Additionally, previous studies have shown that gut microbiota can degrade AGEs and their precursors under anaerobic conditions, reducing the exposure to dietary fructosyllysine or carboxymethyllysine in the gastrointestinal tract and thereby mitigating potential harm to humans (37). Investigating the role of gut microbiota presents a promising avenue for developing treatments for gastrointestinal tumors.

The present study indicated an increase in the 2′-deoxyuridine to cytidine ratio as a risk factor for GC and EC. Cytidine deaminase (CDA) is an enzyme involved in the pyrimidine recovery pathway, catalyzing the hydrolysis and deamination of cytidine and 2′-deoxycytidine to produce uridine and 2′-deoxyuridine (46). Research has demonstrated that CDA expression is often lost or downregulated in various types of cancer in breast, lung, ovarian, liver, colon and cervix, including EC (47). Low levels of CDA may contribute to cancer development (48) by suppressing the deoxyuridine synthesis. Deoxyuridine plays a role in mitigating reactive oxygen species (ROS)-induced endoplasmic reticulum stress, thereby supporting cancer cell survival; thus, its increased synthesis could potentially facilitate tumor development (49). In GC and EC, antitumor therapy commonly results in the accumulation of ROS, which promotes cancer cell apoptosis and inhibits tumor growth (50–53). The present results suggested that the 2′-deoxyuridine to cytidine ratio was a risk factor for EC and GC.

Carnitine serves as a potent antioxidant with multiple regulatory functions, such as boosting cellular respiration, promoting apoptosis and decreasing both tumor cell growth and inflammation. L-carnitine, the active form of carnitine, transports long-chain fatty acid acyl groups from the cytoplasm into the mitochondrial matrix. This transport facilitates the β-oxidation process, converting these acids into acetyl-coenzyme A, which then enters the citric acid cycle to produce energy. However, tumor cells proliferate rapidly and require notable energy and they rely on a distinct metabolic pathway known as the Warburg effect (54), but not the citric acid cycle, to proliferate. The accumulation of carnitine can promote the citric acid cycle, creating an environment that is harmful to the proliferation of cancer cells and even affects their stemness (55). Acylcarnitine is an ester produced by the combination of fatty acid (that is. acyl) and L-carnitine, which plays an important role in numerous cell metabolism pathways related to energy production. Acylcarnitine has been identified as an important indicator of multiple metabolism-related diseases, including cardiovascular diseases, diabetes, metabolic disorders, depression and certain tumors (56). Short-chain acylcarnitine is the most abundant acylcarnitine class in the human body and it has been reported to be closely associated with mental illness (57). Propionyl-L-carnitine is a type of carnitine donor that increases the transport of fatty acids across mitochondrial membranes and is used as a supplement for the treatment of depression (56). In the study by Zhang et al (58), a negative correlation between propionyl-L-carnitine and lung cancer was observed. In addition, L-carnitine has also been shown to improve cancer-related anorexia/cachexia by reducing inflammation (59). Agreeing with previous research, an increase in the ratio of C3 may promote the citric acid cycle, inhibit tumor cell proliferation and serve as a co-protective factor for EC and GC. The functional experiments in the present study also demonstrated that carnitine inhibited tumor cell proliferation.

The current two-sample MR analysis was designed to evaluate the causal relationship between metabolites and UGI tumors using large-sample GWAS and UK Biobank data. This approach overcame some limitations of traditional observational studies by reducing the influence of confounding factors and reverse causality. In addition, MR mitigated the representativeness and feasibility issues that are common in RCTs. However, the present study faced several challenges. First, non-fasting plasma samples were used for metabolomics profiling. Despite adjustments made for the time elapsed since the last meal or beverage, residual variability may still influence the results. Second, the study focused on gene-metabolite pairs identified from current expression data and biological understanding, specifically targeting effector genes. Nonetheless, the potential relevance of other highly heritable metabolites or ratios to the disease cannot be overlooked. Future studies should expand on this by including additional expression data and metabolic insights to identify effector genes for these additional metabolites and ratios. Third, the MR analyses used in the present study were constrained by the fact that most metabolites and metabolite ratios were associated with only a single IV. This constraint made it difficult to apply common MR sensitivity tests such as MR-Egger, which require multiple IVs. However, the selection of IVs that are closely linked to effector genes influencing metabolite levels in the present study mitigated the risk of horizontal pleiotropy. Additionally, assessments of metabolic pleiotropy were conducted to remove IVs linked to multiple metabolites not involved in the same metabolic pathways. Despite these measures, biases may not be fully eliminated due to limitations in metabolome profiling and uncertainties in metabolite-protein interaction databases. Finally, the present study primarily involved elderly individuals of European descent. Future research should explore the effects of identified genetic variations on metabolites and their ratios in a more diverse demographic spectrum. Moreover, the current study covered a broad range of metabolites; however, the metabolic pathway and mechanisms of certain metabolites in disease remain unclear, which limited the interpretability of MR findings.

In summary, this comprehensive MR analysis of metabolites has uncovered new gene-metabolite associations, offering an enhanced understanding of genetic regulation in human metabolism. These discoveries are expected to facilitate the identification of potential markers and targets that could aid in precision prevention and lead to behavioral modifications, innovative pharmaceutical interventions and personalized treatments. Such strategies would be specifically designed to address the unique metabolic vulnerabilities associated with UGI tumors.

Supplementary Material

Supporting Data

Acknowledgements

Not applicable.

Funding

The present study was supported by the Natural Science Foundation of Guangdong Province (grant no. 2023A1515012548) and the Science and Technology Program of Guangzhou (grant no. 2024A04J4802).

Availability of data and materials

The data generated in the present study may be found in the GWAS catalog and UKBiobank ICD PheWeb under accession number (GCST90199621-GCST90201020, inclusive; ukb-saige-150; ukb-saige-151) or at the following URL: (https://www.ebi.ac.uk/gwas/ and https://pheweb.org/UKB-SAIGE/, or https://pheweb.org/UKB-SAIGE/pheno/150 and http://pheweb.org/UKB-SAIGE/pheno/151).

Authors' contributions

PNo collected, analyzed and interpreted the data. contributed to conception, design and drafted the manuscript. CZ performed the cell scratch assay and the CCK-8 assay. SS, ST, SK, PNi, QK, YL, AP, WF, KD and JL designed, revised and supervised the study. PNo and KD confirm the authenticity of all the raw data. All authors read and approved the final manuscript.

Ethics approval and consent to participate

Not applicable.

Patient consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.

References