Exploring metabolic pathways in gastrointestinal cancer: new evidence of causality

Abstract

Background

In recent years, the development of metabolomics has provided new opportunities to explore the associations between hundreds of metabolites and cancer risk, and it is increasingly being utilized in cancer research. After the occurrence of cancer, there are often abnormal changes in the body’s metabolic processes, including alterations in glucose metabolism, lipid metabolism, amino acid metabolism, and nucleotide metabolism. These metabolomic changes provide new insights for the identification of effective tumor biomarkers and the development of targeted therapeutic approaches. However, whether blood metabolites have a causal impact on the development of gastrointestinal (GI) cancer remains to be elucidated. Therefore, the aim of this study is to investigate the potential causal effects of human blood metabolites on the risk of GI cancer through Mendelian randomization (MR) analysis.

Methods and results

We employed a two-sample Mendelian randomization approach to assess the unconfounded causal relationships between 275 blood metabolites and the occurrence of GI cancer. We conducted screening of exposure and outcome factors through GWAS databases and SNP selection, and employed various MR statistical methods, including inverse variance weighted (IVW) method, on the eligible instrumental variables. Additionally, a series of sensitivity analyses were performed to assess the heterogeneity and pleiotropy of the instrumental variables, ensuring the robustness of the results. Univariable MR analysis revealed that genetically predicted elevated levels of various metabolites, including Octanoylcarnitine, were causally associated with an increased risk of GI cancer. Conversely, elevated levels of several metabolites, including Laurate (12:0), were associated with a decreased risk of GI cancer. Moreover, certain metabolites exhibited potential causal relationships with multiple types of GI cancer simultaneously. Through MVMR analysis, we observed significant associations of 1-Stearoylglycerophosphocholine (p: 0.019), 1-eicosatrienoylglycerophosphocholine (p: 0.028), X-11,793–oxidized bilirubin (p: 0.013), and 1-arachidonoylglycerophosphocholine (p: 0.023) with GI cancer. Furthermore, we integrated nine metabolic pathways associated with GI cancer-related metabolites, suggesting that certain types of GI cancer may share common metabolic pathways.

Conclusions

We identified 36 human blood metabolites that are associated with GI cancer, thus confirming the significant role of blood metabolites in the pathogenesis of GI cancer. Furthermore, our findings provide novel potential biomarkers and targets for early screening and prevention of GI cancer.

Introduction

Gastrointestinal (GI) cancers, including esophageal, gastric, and colorectal malignancies, are among the most common cancers in humans. Although these cancers have different origins and clinical features, they share similarities [1]. According to data from the GLOBOCAN 2020 database, GI cancers accounted for 18.7% of newly diagnosed cancer cases and 22.6% of cancer-related deaths in 2020. The incidence and mortality rates of GI cancers are the highest among all cancer types, posing a significant global public health burden [2]. Among them, esophageal cancer ranks eighth globally in terms of incidence and sixth in terms of mortality [2]. Due to the insidious nature of early symptoms, most patients are already in advanced or late stages at the time of diagnosis, resulting in a generally poor prognosis with a 5-year survival rate of approximately 5%-34% [3]. Gastric cancer is the fourth most common cancer in humans and the fourth leading cause of death among all cancer types worldwide in 2020 [2, 4]. Despite breakthroughs in the etiology of gastric cancer (such as Helicobacter pylori) successfully reducing its incidence and mortality rates in the last century [2], the survival rate remains poor due to patients often seeking medical attention in the late stages of the disease. Colorectal cancer is the third most common cancer worldwide, accounting for approximately 10% of all cancer cases, and it is the second leading cause of cancer-related deaths globally. It is often diagnosed in advanced stages with limited treatment options. Due to the fact that the prognosis of these cancers primarily depends on early diagnosis and timely treatment, with early detection and diagnosis being pivotal for timely treatment, it is crucial to have a comprehensive understanding of the etiology of GI cancers for primary prevention of cancer occurrence.

In recent years, the development of metabolomics has provided new opportunities to validate the associations between hundreds of metabolites and cancer risk. Metabolomics, as an emerging discipline following genomics, transcriptomics, and proteomics, is a scientific field that investigates the types, quantities, metabolic pathways, influencing factors, and patterns of change of endogenous metabolites in organisms after perturbations or stimuli (such as genetic alterations or environmental changes). It reflects the pathological and physiological processes through the overall metabolic changes. Metabolomics has been widely used in various areas, including disease diagnosis and prevention, drug screening and development, and the study of disease and drug mechanisms. Generally, metabolomics primarily focuses on small molecular compounds (molecular weight < 1000 Da) that serve as substrates and products in various metabolic pathways. It aims to discover subtle changes occurring in organisms by examining the composition and abundance variations of these metabolites, thus identifying the relative relationship between metabolites and pathological and physiological changes. Metabolomics, positioned at the downstream end of the genomics-transcriptomics-proteomics-metabolomics systems biology framework, is closest to the biological phenotype, allowing for direct correlations with changes in the phenotype. Moreover, because minor variations in the upstream genomics and proteomics can be amplified at the metabolic level, the analysis of metabolites in biological fluids is more easily detectable and can more directly and accurately reflect the pathological and physiological status of organisms. As a result, metabolomics is increasingly being utilized in cancer research.

Following the occurrence of cancer, abnormal changes often accompany the metabolic processes in the organism, including alterations in glucose metabolism, lipid metabolism, amino acid metabolism, and nucleotide metabolism. Concurrently, small molecular substances within the body can undergo qualitative or quantitative changes. These metabolic variations in metabolomics can provide new insights for the identification of effective tumor diagnostic biomarkers and targeted treatment methods [5].

Currently, there is limited research on the metabolomic characteristics of gastrointestinal (GI) cancer patients, with the majority being cohort studies or observational studies. It is difficult to determine whether the selected metabolites have a potential causal relationship with the occurrence of GI cancer. Moreover, these studies are prone to selection bias and confounding factors, highlighting the need for a reliable method to investigate causal relationships.

Mendelian randomization (MR) analysis is currently the most commonly used method for causal inference in clinical research. It is based on summary statistics from large-scale genome-wide association studies (GWAS) and uses single nucleotide polymorphisms (SNPs) as instrumental variables (IVs) to simulate the randomization process of causal inference in randomized controlled trials (RCTs). This approach evaluates the causal impact of exposure factors on outcomes and provides a higher level of clinical evidence compared to RCTs [6]. According to the principles of Mendelian genetics and the random allocation of alleles, genotypes are randomly assigned from parents to offspring, independent of confounding factors. Therefore, MR analysis can reduce certain biases present in traditional research, overcome limitations of randomized controlled trials, and provide more convincing and reliable results. Currently, it has been widely adopted to explore potential causal relationships between environmental exposures and diseases. In this study, we conducted univariable and multivariable Mendelian randomization (MVMR) analyses to determine the causal relationships between various endogenous metabolites and GI cancer. Our aim was to uncover the potential causal impact of metabolomics on the risk of GI cancer, providing a theoretical basis for the early diagnosis and treatment of GI cancer.

Materials and methods

Selection of datasets

This study utilized a multi-batch two-sample Mendelian randomization (TSMR) approach to analyze the unconfounded causal relationship between human blood metabolites and the occurrence of GI cancer. The exposure factors were based on the summary results of a genome-wide association study (GWAS) of 275 blood metabolites from 7,822 adults in two European population studies conducted by Shin et al. in 2014. This study reported hundreds of associations and their metabolic backgrounds, providing a comprehensive molecular readout of human gene activity measured within the body [7]. Among them, these 275 metabolites can be categorized into nine metabolic groups: Carbohydrate, Energy, Nucleotide, Lipid, Fatty acid, Amino acid, Peptide, Cofactors and vitamins, and Xenobiotics [8].

The outcome data were obtained from FinnGen, a database that collects and analyzes genomic and health data from 500,000 participants in the Finnish Biobank, providing novel insights into medical and therapeutic aspects. We retrieved and obtained GWAS summary data for gastrointestinal (GI) cancers from the FinnGen database. The GWAS datasets with the IDs “finngen_R9_C3_OESOPHAGUS_EXALLC”, “finngen_R9_C3_STOMACH_EXALLC”, “finngen_R9_C3_COLON_EXALLC”, and “finngen_R9_C3_RECTUM_EXALLC” were used to investigate the genetic variants associated with esophageal, gastric, colon, and rectal cancers, respectively [9]. The four GI cancer datasets, including esophageal cancer, gastric cancer, colon cancer, and rectal cancer, consisted of 566, 1,307, 3,935, and 2,361 cancer cases, respectively. The control group comprised 287,137 individuals. A total of 20,167,370 single nucleotide polymorphisms (SNPs) were available for analysis. Table 1 provides the description of GWAS data related to GI cancers.

Table 1.

Description of Gastrointestinal (GI) cancers and metabolites related GWAS data

GI cancers
Traits	ID	ICD-10	Population	Year	Sample size	Number of cases	Number of controls	Number of SNPs
Malignant neoplasm of oesophagus (controls excluding all cancers)	finngen_R9_C3_OESOPHAGUS_EXALLC	C15.900	European	2023	287,703	566	287,137	20,167,370
Malignant neoplasm of stomach (controls excluding all cancers)	finngen_R9_C3_STOMACH_EXALLC	C16.900	European	2023	288,444	1,307	287,137	20,167,370
Malignant neoplasm of colon (controls excluding all cancers)	finngen_R9_C3_COLON_EXALLC	C18.900	European	2023	291,072	3,935	287,137	20,167,370
Malignant neoplasm of rectum (controls excluding all cancers)	finngen_R9_C3_RECTUM_EXALLC	C20.x00	European	2023	289,498	2,361	287,137	20,167,370

Metabolites
Category	Subcategory				ID	Author	Population	Year
Metabolites	Carbohydrate, Energy, Nucleotide, Lipid, Fatty acid, Amino acid, Peptide, Cofactors and vitamins, Xenobiotics				met-a-303 to 754	Shin	European	2014

Due to variations in race and population stratification, the allele frequencies of the same SNP may differ among populations with different ancestry. To minimize biases arising from population stratification and gender stratification, the exposed and outcome samples included in our study were derived from a European population, comprising both males and females [10]. Our analysis did not involve the collection of original data but rather utilized published studies or publicly available GWAS summary data. Therefore, ethical approval was not required.

Study design

We employed a two-sample Mendelian randomization (TSMR) design to investigate the causal effect of metabolites on the occurrence of GI cancers. We selected genetic variants (SNPs) associated with the exposure factor (human blood metabolites) as instrumental variables (IVs) to estimate the relationship between the exposure factor and the outcome factor (GI cancers). Subsequently, we conducted multivariable Mendelian randomization (MVMR) analysis to identify thg independent risk factors associated with GI cancers. The execution of MR analysis relies on three key assumptions: (1) the instrumental variables are significantly associated with the exposure factor (2), the instrumental variables are independent of any confounding factors that may be related to both the exposure and the outcome, and (3) the instrumental variables only affect the outcome indirectly through the exposure factor. We constructed a directed acyclic graph (DAG) incorporating the instrumental variables (SNPs), the exposure factor (human blood metabolites), and the outcome (GI cancers) to illustrate the aforementioned assumptions and summarize the study design (Fig. 1).

Fig. 1.

Study design. An overview of the study design

Selection of instrumental variable

Based on the aforementioned research strategy and the basic assumptions of MR, we performed a series of quality control steps on the obtained GWAS summary data to select eligible SNPs. Firstly, we extracted SNPs that were significantly associated with the exposure factor based on a significance level of p < 1 × 10 − 5. To ensure the independence of each SNP and reduce the risk of multicollinearity, we set a linkage disequilibrium (LD) factor (r2) < 0.001 and a clustering window width > 10,000 base pairs to exclude SNPs with LD. In cases where SNP data were missing in the outcome database, we selected proxy SNPs with high LD (r2 > 0.8) as substitutes. Weak instrumental variables with an F-statistic < 10 are insufficient to ensure the association between the SNP and the exposure and were excluded [11]. Ambiguous and palindromic SNPs during the two-sample matching process were excluded due to the inability to guarantee SNP specificity. Finally, we utilized the PhenoScanner database to examine the potential associations between each SNP and confounding factors. SNPs related to potential confounding factors such as alcohol intake [12, 13], smoking [14–16], obesity [12, 17], which may violate the independence assumption, were excluded [18]. Outliers were identified using the MR-PRESSO test for horizontal pleiotropy, and SNPs with incorrect causal estimate directions were removed using MR-Steiger analysis [19]. The SNPs that passed the rigorous screening process were considered as the final instrumental variables for MR analysis. The selection criteria for multivariable Mendelian randomization (MVMR) analysis were consistent with those for single-variable Mendelian randomization(SVMR) analysis.

Mendelian randomization analysis

We employed various MR statistical methods to assess the potential causal effect of exposure (human blood metabolites) on the susceptibility to the outcome (GI cancer), including inverse-variance weighted (IVW) method, MR-Egger regression, weighted median (WM) method, weighted mode method, and MR-RAPS method.

The IVW method is the most commonly used approach in MR analysis and the primary method employed in this study to provide the most precise estimates. It combines the ratio estimates calculated from each SNP through inverse-variance weighting and produces an estimate of the effect of the exposure factor on the outcome [20]. Following the guidelines for Mendelian randomization, in the absence of horizontal pleiotropy and heterogeneity, we selected the IVW random-effects model as the primary analytic approach [21].

MR-Egger regression is used to create a weighted linear regression between the outcome coefficients and exposure coefficients. Similar to the IVW method, it includes an intercept term in the regression model to assess horizontal pleiotropy. MR-Egger allows for one or more genetic variants to have pleiotropic effects. When there is evidence of horizontal pleiotropy among SNPs, we preferentially employ the MR-Egger method as the primary analysis. MR-Egger provides consistent causal effect estimates under the assumption of instrument strength independent of direct effect (InSIDE) and can address directional pleiotropy [22].

Weighted median (WM) method and weighted mode method are common consensus methods based on the majority valid assumption and the plurality valid assumption, respectively, for estimating causal effects [21, 23]. When there is heterogeneity among SNPs, the IVW method requires support from the median-based WM method to establish the significance of the conclusions. The WM method provides reliable effect estimates of the causal effects when at least 50% of the weights in the analysis come from valid instrumental variables. It improves precision, and even in the presence of some invalid instruments, the WM method can still correctly estimate causal relationships. When there is heterogeneity among SNPs, a comprehensive discussion will consider the results of both the IVW random-effects model and the WM model. MR-RAPS method directly simulates the pleiotropy of genetic variants using a random-effects distribution, providing stronger robustness against outliers [24].

The statistical power of this study was calculated using an online power calculator [25]. The final results were considered statistically significant at a significance level of p < 0.05 after correction for the false discovery rate (FDR).

Following univariable MR, we performed multivariable analysis on the significant metabolites using the IVW, MR-Egger, weighted median, and LASSO regression methods to identify independent and significant plasma biomarkers using the same set of parameters.

Sensitivity analysis

Firstly, we assessed the heterogeneity among SNPs using the IVW method and verified the heterogeneity using Cochran’s Q test, considering heterogeneity to be present when the p-value of the Q test was less than 0.05. Then, we employed the MR-Steiger model to verify the overall direction of the causal estimates and ensure the robustness of the results. Additionally, to determine whether the overall conclusions of the MR analysis were dependent on a specific SNP, we conducted leave-one-out sensitivity tests, where one SNP was removed from the MR analysis at a time, and the analysis was repeated.

Metabolic pathway analysis

The analysis of metabolic pathways based on the KEGG database was performed using MetaboAnalyst 5.0 (https://www.metaboanalyst.ca/) with the available set of metabolites. MetaboAnalyst 5.0 is a user-friendly online tool designed to simplify the analysis of metabolomics data.

Data visualization and statistical analysis software

For each MR analysis, scatter plots were generated to visualize the relationship between SNPs and the exposure factor as well as the outcome effect estimates. Regression curve plots were also created to illustrate the causal estimates. Forest plots were generated to display the effects of individual SNPs. Additionally, a circular heat map and forest plot were created to present the overall results. Some of the figures in this article were obtained from MedPeer (medpeer.cn).

All statistical analyses and data visualizations in this study were conducted using R software (version 4.1.2) with packages such as “TwoSampleMR,” “MR-PRESSO,” “mr.raps,” and “forestploter,” as well as several foundational R packages.

Results

After genome-wide significance level standard screening, we obtained a total of 6671 SNPs associated with human blood metabolites. No weak instrumental variables with F < 10 were found. This indicates a strong correlation between instrumental variables and the outcome. 120 SNPs were excluded from the outcome dataset due to missing data. 816 palindromic or ambiguous SNPs were excluded from the merged dataset. Subsequently, we applied the MR-Steiger test and did not identify any SNP with causal estimation direction errors. After Phenoscanner testing, we excluded 123 SNPs associated with confounding factors such as alcohol consumption, smoking, and obesity. MR-PRESSO testing revealed 9 SNPs with horizontal pleiotropy. After Bonferroni correction, 46 SNPs directly related to the outcome were removed. Finally, after rigorous screening, 5557 eligible SNPs were included as instrumental variables in the study.

Considering that certain SNPs may have other biological effects on the outcome. Therefore, after screening SNPs associated with the exposure factor, we calculated the intercept term in MR-Egger regression. No evidence of horizontal pleiotropy was found. Figure 2 presents the preliminary results of the circular heatmap, illustrating the relationship between genetic proxies of human blood metabolites and the risk of GI cancer. Additionally, we summarized all positive MR results in Tables 2, 3, 4 and 5. All estimates are presented as odds ratios (OR) per standard deviation (SD) increase in the corresponding exposure factor.

Fig. 2.

Preliminary MR estimates for the associations between metabolites and the risk of GI cancers. From the inner to outer circles, they represent the estimates of malignant neoplasm of stomach, malignant neoplasm of rectum, malignant neoplasm of oesophagus, and malignant neoplasm of colon, respectively. And the shades of color reflect the magnitude of the p-value

Table 2.

 MR results of causal effects between metabolites and risk of oesophagus cancer

Exposures	Method	OR(95% CI)	Adjusted P	Power%
Laurate (12:0)	IVW(MRE)	0.017 (0.001 ~ 0.186)	0.024	5.6
4-acetamidobutanoate	IVW(MRE)	0.025 (0.004 ~ 0.152)	0.002	7.4
1-stearoylglycerophosphocholine	IVW(MRE)	0.034 (0.004 ~ 0.293)	0.043	15.5
Octanoylcarnitine	IVW(MRE)	1.666 (1.250 ~ 2.219)	0.015	14.9
Cis-4-decenoyl carnitine	IVW(MRE)	2.273 (1.453 ~ 3.555)	0.011	25.5
Serine	IVW(MRE)	2.483 (2.138 ~ 2.885)	< 0.001	9.3
Myo-inositol	IVW(MRE)	5.062 (4.497 ~ 5.698)	< 0.001	26
1-palmitoylglycerophosphoethanolamine	IVW(MRE)	12.028 (6.873 ~ 21.049)	< 0.001	100
10-heptadecenoate (17:1n7)	IVW(MRE)	21.298 (3.327 ~ 136.328)	0.031	100
Stearate (18:0)	IVW(MRE)	30.677 (5.828 ~ 161.485)	0.002	100
Ornithine	IVW(MRE)	46.910 (4.291 ~ 512.774)	0.036	100
Palmitoyl sphingomyelin	IVW(MRE)	4439.714 (1788.385 ~ + 8)	< 0.001	100

Table 3.

MR results of causal effects between metabolites and risk of stomach cancer

Exposures	Method	OR(95% CI)		Adjusted P	Power%
Caprylate (8:0)	IVW(MRE)	0.006 (0.002 ~ 0.016)		< 0.001	6.4
Gamma-glutamyltyrosine	IVW(MRE)	0.006 (0.001 ~ 0.040)		< 0.001	7.1
Octadecanedioate	IVW(MRE)	0.078 (0.020 ~ 0.310)		0.007	15.7
ADSGEGDFXAEGGGVR	IVW(MRE)	0.410 (0.319 ~ 0.527)		< 0.001	31.7
1-arachidonoylglycerophosphocholine	IVW(MRE)	0.526 (0.418 ~ 0.663)		< 0.001	5.8
Taurochenodeoxycholate	IVW(MRE)	0.603 (0.435 ~ 0.835)		0.041	32.8
X-11,793–oxidized bilirubin	IVW(MRE)	0.609 (0.516 ~ 0.719)		< 0.001	15
X-13,431–nonanoylcarnitine	IVW(MRE)	1.726 (1.400 ~ 2.128)		< 0.001	53
Erythronate	IVW(MRE)	2.505 (1.460 ~ 4.299)		0.018	10.9
Ornithine	IVW(MRE)	2.851 (1.751 ~ 4.643)		0.001	26.7
1-eicosatrienoylglycerophosphocholine	IVW(MRE)	5.369 (1.917 ~ 15.036)		0.027	98.5
Phenylalanylphenylalanine	IVW(MRE)	6.358 (2.335 ~ 17.310)		0.007	100
Dihomo-linolenate (20:3n3 or n6)	IVW(MRE)	11.845 (10.266 ~ 13.666)		< 0.001	100
Linoleate (18:2n6)	IVW(MRE)	12.846 (5.697 ~ 28.966)		< 0.001	100

Table 4.

MR results of causal effects between metabolites and risk of colon cancer

Exposures	Method	OR(95% CI)	Adjusted P		Power%
Gamma-glutamyltyrosine	IVW(MRE)	0.283 (0.182 ~ 0.441)	< 0.001		8.3
4-androsten-3beta,17beta-diol disulfate 2	IVW(MRE)	0.583 (0.570 ~ 0.597)	< 0.001		13.8
Bradykinin, des-arg(9)	IVW(MRE)	0.728 (0.632 ~ 0.839)	< 0.001		73.7
X-14,189–leucylalanine	IVW(MRE)	0.763 (0.661 ~ 0.880)	0.006		11.2
X-13,431–nonanoylcarnitine	IVW(MRE)	0.785 (0.770 ~ 0.801)	< 0.001		18.1
X-11,793–oxidized bilirubin	IVW(MRE)	0.902 (0.869 ~ 0.937)	< 0.001		6.8
Mannitol	IVW(MRE)	1.316 (1.133 ~ 1.528)	0.009		72.1
1-stearoylglycerophosphocholine	IVW(MRE)	2.343 (1.388 ~ 3.956)	0.035		95.4
Xanthine	IVW(MRE)	6.285 (2.417 ~ 16.344)	0.006		100
Gamma-glutamylphenylalanine	IVW(MRE)	15.376 (4.975 ~ 47.526)	< 0.001		100

Table 5.

MR results of causal effects between metabolites and risk of rectum cancer

Exposures	Method	OR (95% CI)	Adjusted P	Power%
Citrulline	IVW(MRE)	0.005 (0.001 ~ 0.044)	<0.001	8.6
1-arachidonoylglycerophosphocholine	IVW(MRE)	0.063 (0.022 ~ 0.182)	<0.001	11
Asparagine	IVW(MRE)	0.178 (0.081 ~ 0.393)	0.001	12.6
1-eicosatrienoylglycerophosphocholine	IVW(MRE)	0.221 (0.091 ~ 0.536)	0.026	16.3
Fructose	IVW(MRE)	0.313 (0.178 ~ 0.552)	0.002	11.2
X-11793--oxidized bilirubin	IVW(MRE)	0.514 (0.343 ~ 0.771)	0.034	33.2
X-13431--nonanoylcarnitine	IVW(MRE)	0.620 (0.504 ~ 0.762)	<0.001	30.1
10-undecenoate (11:1n1)	IVW(MRE)	0.954 (0.929 ~ 0.980)	0.021	5.1
1,5-anhydroglucitol (1,5-AG)	IVW(MRE)	15.341 (5.127 ~ 45.900)	<0.001	100

In the absence of horizontal pleiotropy, we employed the IVW model as the primary analytical method. Other MR analysis methods were used as sensitivity analyses to assess their reliability. The IVW model revealed causal relationships between 36 metabolites and the occurrence of GI cancer.

• Oesophagus cancer

The IVW analysis revealed that elevated concentrations of Laurate (12:0) (odds ratio [95% confidence interval]: 0.017 [0.001–0.186], adjusted p-value: 0.024), 4-acetamidobutanoate (OR [95%CI]: 0.025 [0.004–0.152], adjusted p-value: 0.002), and 1-stearoylglycerophosphocholine (OR [95%CI]: 0.034 [0.004–0.293], adjusted p-value: 0.043) potentially reduce the risk of esophageal cancer. Conversely, elevated concentrations of Octanoylcarnitine (OR [95%CI]: 1.666 [1.250–2.219], adjusted p-value: 0.015), Cis-4-decenoyl carnitine (OR [95%CI]: 2.273 [1.453–3.555], adjusted p-value: 0.011), Serine (OR [95%CI]: 2.483 [2.138–2.885], adjusted p-value < 0.001), Myo-inositol (OR [95%CI]: 5.062 [4.497–5.698], adjusted p-value < 0.001), 1-palmitoylglycerophosphoethanolamine (OR [95%CI]: 12.028 [6.873–21.049], adjusted p-value < 0.001), 10-heptadecenoate (17:1n7) (OR [95%CI]: 21.298 [3.327-136.328], adjusted p-value: 0.031), Stearate (18:0) (OR [95%CI]: 30.677 [5.828-161.485], adjusted p-value: 0.002), Ornithine (OR [95%CI]: 46.910 [4.291-512.774], adjusted p-value: 0.036), and Palmitoyl sphingomyelin (OR [95%CI]: 4439.714 [1788.385-+∞], adjusted p-value < 0.001) were associated with a higherrisk of esophageal cancer.All positive results concerning oesophageal cancer are shown in Table 2.

• Stomach cancer

Elevated concentrations of Caprylate (8:0) (odds ratio [95% confidence interval]: 0.006 [0.002–0.016], adjusted p-value < 0.001), Gamma-glutamyltyrosine (OR [95%CI]: 0.006 [0.001–0.040], adjusted p-value < 0.001), Octadecanedioate (OR [95%CI]: 0.078 [0.020–0.310], adjusted p-value: 0.007), ADSGEGDFXAEGGGVR (OR [95%CI]: 0.410 [0.319–0.527], adjusted p-value < 0.001), 1-arachidonoylglycerophosphocholine (OR [95%CI]: 0.526 [0.418–0.663], adjusted p-value < 0.001), Taurochenodeoxycholate (OR [95%CI]: 0.603 [0.435–0.835], adjusted p-value: 0.041), and X-11,793–oxidized bilirubin (OR [95%CI]: 0.609 [0.516–0.719], adjusted p-value < 0.001) were associated with a decreased risk of gastric cancer. Conversely, elevated concentrations of X-13,431–nonanoylcarnitine (OR [95%CI]: 1.726 [1.400-2.128], adjusted p-value < 0.001), Erythronate (OR [95%CI]: 2.505 [1.460–4.299], adjusted p-value: 0.018), Ornithine (OR [95%CI]: 2.851 [1.751–4.643], adjusted p-value: 0.001), 1-eicosatrienoylglycerophosphocholine (OR [95%CI]: 5.369 [1.917–15.036], adjusted p-value: 0.027), Phenylalanylphenylalanine (OR [95%CI]: 6.358 [2.335–17.310], adjusted p-value: 0.007), Dihomo-linolenate (20:3n3 or n6) (OR [95%CI]: 11.845 [10.266–13.666], adjusted p-value < 0.001), and Linoleate (18:2n6) (OR [95%CI]: 12.846 [5.697–28.966], adjusted p-value < 0.001) were associated with a potential increased risk of gastric cancer.All positive results concerning stomach cancer are shown in Table 3.

• Colon cancer

Elevated concentrations of Gamma-glutamyltyrosine (OR [95%CI]: 0.283 [0.182–0.441], adjusted p-value < 0.001), 4-androsten-3beta,17beta-diol disulfate 2 (OR [95%CI]: 0.583 [0.570–0.597], adjusted p-value < 0.001), Bradykinin, des-arg(9) (OR [95%CI]: 0.728 [0.632–0.839], adjusted p-value < 0.001), X-14,189–leucylalanine (OR [95%CI]: 0.763 [0.661–0.880], adjusted p-value: 0.006), X-13,431–nonanoylcarnitine (OR [95%CI]: 0.785 [0.770–0.801], adjusted p-value < 0.001), and X-11,793–oxidized bilirubin (OR [95%CI]: 0.902 [0.869–0.937], adjusted p-value < 0.001) were associated with a decreased risk of colon cancer. Conversely, elevated concentrations of Mannitol (OR [95%CI]: 1.316 [1.133–1.528], adjusted p-value: 0.009), 1-stearoylglycerophosphocholine (OR [95%CI]: 2.343 [1.388–3.956], adjusted p-value: 0.035), Xanthine (OR [95%CI]: 6.285 [2.417–16.344], adjusted p-value: 0.006), and Gamma-glutamylphenylalanine (OR [95%CI]: 15.376 [4.975–47.526], adjusted p-value < 0.001) were associated with a higher risk of colon cancer.All positive results concerning colon cancer are shown in Table 4.

• Rectum cancer

Elevated concentrations of Citrulline (OR [95%CI]: 0.005 [0.001–0.044], adjusted p-value < 0.001), 1-arachidonoylglycerophosphocholine (OR [95%CI]: 0.063 [0.022–0.182], adjusted p-value < 0.001), Asparagine (OR [95%CI]: 0.178 [0.081–0.393], adjusted p-value: 0.001), 1-eicosatrienoylglycerophosphocholine (OR [95%CI]: 0.221 [0.091–0.536], adjusted p-value: 0.026), Fructose (OR [95%CI]: 0.313 [0.178–0.552], adjusted p-value: 0.002), X-11,793–oxidized bilirubin (OR [95%CI]: 0.514 [0.343–0.771], adjusted p-value: 0.034), X-13,431–nonanoylcarnitine (OR [95%CI]: 0.620 [0.504–0.762], adjusted p-value < 0.001), and 10-undecenoate (11:1n1) (OR [95%CI]: 0.954 [0.929–0.980], adjusted p-value: 0.021) were associated with a potential decrease in the risk of colorectal cancer, while elevated concentrations of 1,5-anhydroglucitol (1,5-AG) (OR [95%CI]: 15.341 [5.127–45.900], adjusted p-value < 0.001) were associated with a potential increase in the risk of colorectal cancer.All positive results concerning rectum cancer are shown in Table 5.

It can be observed that certain metabolites exhibit potential causal relationships with multiple GI cancer simultaneously, as shown in Table 6. Ornithine is associated with an increased risk of esophageal and gastric cancer. 1-Stearoylglycerophosphocholine is potentially associated with a decreased risk of esophageal cancer, but an increased risk of colorectal cancer. Gamma-glutamyltyrosine is associated with a reduced risk of both gastric and colorectal cancer. 1-arachidonoylglycerophosphocholine is associated with a decreased risk of gastric and colorectal cancer. 1-eicosatrienoylglycerophosphocholine is associated with an increased risk of gastric cancer but a potential decrease in the risk of colorectal cancer. X-11,793–oxidized bilirubin is associated with a reduced risk of gastric, colorectal, and rectal cancer simultaneously. X-13,431–nonanoylcarnitine is associated with an increased risk of gastric cancer but a potential decrease in the risk of colorectal and rectal cancer.

Table 7.

Metabolites with potential causal relationships to multiple GICs

Outcomes	ID of outcomes	Exposures	ID of exposures	nSNP	b	se	pval
Malignant neoplasm of oesophagus (controls excluding all cancers)	finngen_R9_C3_OESOPHAGUS_EXALLC	1-stearoylglycerophosphocholine	met-a-622	204	-2.3479	0.9993	0.0188
Malignant neoplasm of rectum (controls excluding all cancers)	finngen_R9_C3_RECTUM_EXALLC	1-eicosatrienoylglycerophosphocholine	met-a-602	153	-0.8901	0.4056	0.0282
Malignant neoplasm of rectum (controls excluding all cancers)	finngen_R9_C3_RECTUM_EXALLC	X-11793–oxidized bilirubin	met-a-543	153	-0.6166	0.2491	0.0133
Malignant neoplasm of rectum (controls excluding all cancers)	finngen_R9_C3_RECTUM_EXALLC	1-arachidonoylglycerophosphocholine	met-a-558	153	0.8410	0.3702	0.0231

Table 6.

Metabolites with potential causal relationships to multiple GICs

Exposures	Outcomes	Method	OR (95% CI)	Adjusted P
Ornithine	Malignant neoplasm of oesophagus	IVW	46.910 [4.291–512.774]	0.036
	Malignant neoplasm of stomach		2.851 [1.751–4.643]	0.001
1–1-Stearoylglycerophosphocholine	Malignant neoplasm of oesophagus	IVW	0.034 [0.004–0.293]	0.043
	Malignant neoplasm of colon		2.343 [1.388–3.956]	0.035
Gamma-glutamyltyrosine	Malignant neoplasm of stomach	IVW	0.006 [0.001–0.040]	< 0.001
	Malignant neoplasm of colon		0.283 [0.182–0.441]	< 0.001
1-arachidonoylglycerophosphocholine	Malignant neoplasm of stomach	IVW	0.526 [0.418–0.663]	< 0.001
	Malignant neoplasm of rectum		0.063 [0.022–0.182]	< 0.001
1-eicosatrienoylglycerophosphocholine	Malignant neoplasm of stomach	IVW	5.369 [1.917–15.036]	0.027
	Malignant neoplasm of rectum		0.221 [0.091–0.536]	0.026
X-11793–oxidized bilirubin	Malignant neoplasm of stomach	IVW	0.609 [0.516–0.719]	< 0.001
	Malignant neoplasm of colon		0.902 [0.869–0.937]	< 0.001
	Malignant neoplasm of rectum		0.514 [0.343–0.771]	0.034
X-13431–nonanoylcarnitine	Malignant neoplasm of stomach	IVW	1.726 [1.400–2.128]	< 0.001
	Malignant neoplasm of colon		0.785 [0.770–0.801]	< 0.001
	Malignant neoplasm of rectum		0.620 [0.504–0.762]	< 0.001

Firstly, there was no evidence of horizontal pleiotropy based on the intercept term of MR-Egger regression (all p-values > 0.05). Secondly, no significant heterogeneity was observed according to the Cochran’s Q test. Lastly, we applied the MR-Steiger model to confirm the correct direction of causality for each MR analysis. In conclusion, our MR analysis findings are overall robust and reliable.

In order to investigate the multivariable causal relationships between these 36 blood metabolites and GI cancers, we conducted MVMR analysis, and the results are presented in Table 7. In the MVMR analysis, 1-Stearoylglycerophosphocholine (p-value: 0.019) showed a significant association with esophageal cancer. 1-eicosatrienoylglycerophosphocholine (p-value: 0.028), X-11,793–oxidized bilirubin (p-value: 0.013), and 1-arachidonoylglycerophosphocholine (p-value: 0.023) exhibited significant associations with colorectal cancer.

• Metabolic pathway analysis

A total of 9 metabolic pathways were co-aggregated with GI cancers (Fig. 3), including 5 pathways associated with esophageal cancer. The most significant pathway was myo-Inositol in Galactose metabolism ( p < 0.001). Among the 4 metabolic pathways associated with gastric cancer, the most significant pathway was inoleate in Biosynthesis of unsaturated fatty acids ( p < 0.001). For colorectal cancer, “Citrulline in Arginine biosynthesis” ( p < 0.001) was the most significant pathway among the two pathways. Furthermore, we observed that certain types of GI cancers may share common metabolic pathways. For example, the pathways “Ornithine in Arginine and proline metabolism” and “Ornithine in Arginine biosynthesis” were shared by esophageal and gastric cancers.

Fig. 3.

9 important metabolic pathways associated with GIC

Discussion

Metabolic dysregulation is considered a common and crucial feature of tumors and cancer cells [26]. Studies have shown that tumors have the ability to induce metabolic alterations in the host organism, involving various intracellular processes such as energy, carbohydrate, lipid, and amino acid metabolism, which convert nutrients into cellular components [27]. Consequently, these altered metabolic pathways can influence various biological processes in which they are involved, such as cell proliferation and apoptosis.

In this study, we confirmed causal associations between 36 human blood metabolites and the risk of GI cancers. Using MVMR, we identified significant causal effects of 1-Stearoylglycerophosphocholine on esophageal cancer risk. For colorectal cancer, causal associations were observed with 1-eicosatrienoylglycerophosphocholine, X-11,793-oxidized bilirubin, and 1-arachidonoylglycerophosphocholine. Among them, 1-Stearoylglycerophosphocholine, 1-eicosatrienoylglycerophosphocholine, and 1-arachidonoylglycerophosphocholine belong to the lipid metabolism group, while X-11,793–oxidized bilirubin belongs to the coenzyme and vitamin metabolism group. These findings underscore the significant role of blood metabolites in the pathogenesis of GI cancers and provide valuable insights for further research on the early diagnosis and prevention of GI cancers.

• Lipid metabolomics and GI cancer risk

Lipid metabolites represent a class of molecules exhibiting significant structural and functional diversity, encompassing phospholipids (such as phosphatidylcholines, phosphatidylethanolamines, and phosphatidylinositols), triglycerides, cholesterol, and its esters. These metabolites serve not only as essential structural components of cellular membranes but also participate in critical biological processes, including energy metabolism, signal transduction, and the regulation of gene expression. In recent years, metabolomics studies have revealed that alterations in lipid metabolism constitute one of the most prominent metabolic changes in cancer, where heightened synthesis and uptake of lipids contribute significantly to rapid cancer cell proliferation, as well as tumor formation, colonization, and metastasis [28, 29]. Lipid metabolism is often enhanced at different stages of cancer development. These alterations not only provide energy for tumor cells but also trigger signaling and epigenetic events, as well as change the composition of membranes to facilitate metastasis [30].

Notably, we observed associations between specific lipid metabolites and GI cancer risk. For instance, 1-eicosatrienoylglycerophosphocholine and 1-Stearoylglycerophosphocholine may promote tumor progression by activating α-enolase (ENO1)—a glycolytic enzyme overexpressed in metastatic GI cancers. Analogous to the mechanism of oleic acid [31], these phospholipids enhance ATP generation in hypoxic tumor microenvironments and induce epithelial-mesenchymal transition (EMT). Experimental evidence supporting this mechanism indicates that ENO1 depletion significantly suppresses glycolysis in colorectal cancer cells and promotes ferroptosis via inactivation of the AKT/STAT3 signaling pathway [32].

1-arachidonoylglycerophosphocholine hydrolysis releases arachidonic acid (AA), which is subsequently metabolized by cyclooxygenase (COX) and cytochrome P450 (CYP) enzymes into pro-metastatic eicosanoids. In both plasma and colon tissues from AOM/DSS-induced colon cancer mice, CYP-derived metabolites such as EpOMEs and EETs were significantly elevated [33]. Furthermore, KEGG pathway analysis revealed that GI cancers share unsaturated fatty acid biosynthesis pathways (e.g., linoleic acid metabolism). This pathway intersects with the ENO1 activation mechanism: 1-eicosatrienoylglycerophosphocholine and 1-arachidonoylglycerophosphocholine may indirectly modulate ENO1 glycolytic activity by regulating linoleic acid derivatives (e.g., EpOMEs).

• Shared metabolites in colorectal cancer

Notably, despite differences in anatomical location and molecular characteristics between colon cancer and rectal cancer, our study identified shared metabolic biomarkers with cross-subtype significance. X-11,793-oxidized bilirubin exhibited significant protective effects in both cancer types. Multivariable MR analysis further confirmed its independent association with colorectal cancer. This consistency suggests its potential utility as a pan-subtype biomarker for colorectal cancer.

Bilirubin is the final breakdown product of heme, and some studies suggested that bilirubin can effectively inhibit tumor cell proliferation in vivo and exhibit cell-inhibitory and pro-apoptotic effects in vitro [34]. The anti-tumor mechanisms of bilirubin may explain its value across cancer subtypes. As a potent endogenous antioxidant, bilirubin suppresses DNA damage and the formation of an inflammatory microenvironment by scavenging reactive oxygen species (ROS) [35, 36]. In vitro studies have confirmed that unconjugated bilirubin (UCB) can penetrate the membranes of colorectal cancer cells, induce mitochondrial depolarization, mediate apoptosis via the ERK pathway, and inhibit tumor cell proliferation [37]. Our research extends these findings, suggesting that systemically elevated bilirubin levels may suppress tumorigenesis by modulating redox homeostasis and exerting direct pro-apoptotic effects. Notably, the combined pattern of X-11,793-oxidized bilirubin with other metabolites demonstrated a synergistic effect in MVMR analyses. This multi-metabolite panel holds promise as a biomarker for early colorectal cancer screening, potentially offering superior prognostic value compared to single biomarkers.

• Other metabolite analysis

Through univariate MR analysis, we also identified several blood metabolites from carbohydrate metabolism, nucleotide metabolism, fatty acid metabolism, amino acid metabolism, and peptide metabolism that have potential causal effects on the occurrence of GI cancer.

The intake of carbohydrates may influence the occurrence and progression of cancer [38]. Studies have shown that the utilization of the glycolytic pathway is a metabolic adaptation employed by cancer cells to enhance and sustain their survival and proliferative capacity, wherein tumor cells transition from aerobic cellular respiration to aerobic glycolysis as the primary mechanism for energy acquisition [39]. For instance, a study in 2020 indicated that fructose can be utilized in glycolysis to produce ATP, providing carbon for nucleotide and lipid synthesis, and it can also serve as a signaling cue to sustain the proliferation of cancer cells [40]. Our findings suggest that a high concentration of fructose potentially reduces the risk of colorectal cancer, indirectly confirming the high demand of cancer cells for fructose consumption to acquire substantial energy for their survival and proliferation.

Nucleotide metabolism is considered a crucial process in tumorigenesis and cancer cell replication [42]. The enrichment of nucleotide metabolism is utilized in various cancers to meet the uncontrolled and rapid self-replication demands of cancer cells [41]. Furthermore, this upregulation of nucleotide metabolism can lead to genomic instability, thereby exerting additional carcinogenic effects [42]. Xanthine oxidoreductase (XOR) is a crucial rate-limiting enzyme involved in the degradation of DNA and RNA [43]. Studies have indicated that the expression and activity of XOR are significantly lower in tumor tissues such as the gastrointestinal tract and colon compared to normal tumor tissues [44]. The low expression of XOR is associated with an increased risk of colorectal cancer (CRC), as well as histological differentiation, disease stage, and prognosis [45]. Our study revealed that a high concentration of xanthine is associated with an increased risk of colorectal cancer, and xanthine, a purine metabolite, can be converted to uric acid by XOR in purine metabolism, thus indirectly supporting the findings of the aforementioned researchers.

Fatty acids can provide fuel for the proliferation of cancer cells. It has been reported that cancer cells rapidly generate fatty acids to meet the urgent demands for membrane biosynthesis, cellular signal transduction, and energy consumption [46]. Studies have shown that in esophageal squamous cell carcinoma (ESCC), the ion intensity of fatty acids exhibits an increasing trend from muscle to epithelium and finally to cancerous tissue [47]. Currently, there is ample evidence indicating that dietary linoleic acid (LA) exacerbates the risk of colonic cancer in mice [48]. LA can be converted to epoxyeicosatrienoic acids (EpOMEs) by CYP monooxygenases, which mediate the pro-carcinogenic effects of LA [49]. Considering our findings that Dihomo-linolenate (20:3n3 or n6) and Linoleate (18:2n6) are associated with an increased risk of gastric cancer, it is reasonable to speculate that the mechanisms by which linoleic acid affects gastric cancer are similar to its effects on colonic cancer. However, the specific role of EpOMEs in increasing tumor burden remains unclear. They may promote tumors by exacerbating inflammation or by influencing cell proliferation or migration, thereby increasing tumor burden and/or metastasis [48], or they may contribute to tumorigenesis through microbiota-dependent mechanisms in the gut [49].

Amino acids are essential nutrients for the survival of all cell types [50]. Previous studies have indicated that an abundant supply of amino acids is crucial for sustaining the proliferative drive of cancer cells [51]. For example, a study in 2014 demonstrated that serine serves as a central hub in cancer metabolism and is an essential metabolite for cancer cells [52]. Furthermore, Sivashanmugam et al. have indicated that arginine plays a critical role in regulating cancer progression, and elevated levels of arginine are positively associated with various diseases [53]. Additionally, in a study conducted by Luo et al. in 2018, it was found that taurine promotes the migration, invasion, and intravasation of tumor cells at the primary tumor site. Moreover, there have been reports highlighting the intricate relationship between taurine and glutamine in promoting the survival, growth, and metastasis of tumor cells, underscoring the emerging role of metabolic adaptation in tumor development, metastasis, and progression [54]. All of the above findings are consistent with our viewpoint.

Research has indicated that non-coding RNAs (ncRNAs) encoding peptides or proteins can exert their anti-tumor functions by inhibiting cancer metabolic reprogramming, stabilization of oncogenic proteins, and epithelial-mesenchymal transition (EMT) process [55]. For instance, kinins are an octapeptide to decapeptide family structurally related to bradykinin (BK), which can exert various effects on the immune system, including macrophages, dendritic cells, T and B lymphocytes, and regulate the activation, proliferation, migration, and function of these cells [56]. Kinin B1 receptor, when activated by its agonist des-Arg9-bradykinin (DABK), can effectively stimulate immune responses in the host, acting on tumor cells to reduce the establishment of metastatic colonies [57]. This finding is consistent with our belief that Bradykinin, des-arg(9) can lower the risk of colon cancer. This is consistent with our finding that Bradykinin, des-arg(9) can reduce the risk of colorectal cancer. However, other studies have indicated that kinin receptors are highly expressed in other types of tumors, and their activation may contribute to the proliferation and migration of cancer cells [58]. Whether these effects may be attributed to cross-talk between B1 and B2 receptors remains unknown [59].

• Specific biomarkers in metabolic pathway integration analysis

Integrative metabolic pathway analysis revealed distinct, significantly enriched pathways for esophageal cancer, gastric cancer, and colorectal cancer: myo-Inositol in Galactose metabolism, Linoleate in Biosynthesis of unsaturated fatty acids, and Citrulline in Arginine biosynthesis, respectively. These pathway-specific biomarkers not only demonstrated significant associations with cancer risk but also possess mechanistically supported roles as evidenced by existing literature.

We observed that elevated myo-inositol levels were significantly associated with an increased risk of esophageal cancer, and its associated galactose metabolism pathway was the most significantly altered metabolic pathway in this malignancy. This finding aligns with the dual role of myo-inositol in cancer metabolism: while it can inhibit tumor growth via activation of the AMPK pathway [60], upregulated inositol synthase expression may promote cancer cell proliferation in esophageal squamous cell carcinoma [61]. This paradoxical effect suggests that dysregulation of myo-inositol metabolism may serve as a specific biomarker for esophageal cancer, with its dynamic changes potentially acting as an early risk indicator.

Increased levels of linoleate exhibited a strong positive association with gastric cancer risk, and the unsaturated fatty acid biosynthesis pathway in which it participates was the most significantly altered pathway in gastric cancer. Linoleate-derived epoxyoctadecenoic acids (EpOMEs), generated by CYP450 enzyme metabolism, have been demonstrated to promote tumor cell growth [62]. Notably, EpOME concentrations were significantly higher in gastric cancer tissues compared to adjacent normal tissues [63], indicating that linoleate metabolites could serve as specific circulating biomarkers for gastric cancer, with their pro-tumorigenic mechanisms showing an organ-specific propensity.

Elevated citrulline levels were significantly associated with a reduced risk of colorectal cancer, and its associated arginine biosynthesis pathway emerged as the core metabolic pathway in colorectal cancer. Arginine metabolism influences the tumor microenvironment by regulating nitric oxide (NO) synthesis and T-cell immune function [64, 65]. Furthermore, recent studies indicate that restricting arginine-rich foods or therapeutic arginine depletion using enzymes like arginine deiminase (ADI) or arginase can suppress colon cancer [66]. The protein arginine deiminase (PAD) enzyme family converts arginine to citrulline via citrullination; dysregulation of PADs, leading to abnormal citrullination, is implicated in various diseases, including CRC [67]. Additionally, colorectal cancer patients exhibit a significantly lower plasma citrulline-to-ornithine ratio compared to healthy individuals [53], suggesting its potential utility as a non-invasive biomarker for colorectal cancer screening.

Although there have been numerous studies on the potential mechanisms of blood metabolites influencing gastrointestinal carcinogenesis, there remains some controversy regarding the causal relationship between the two and their potential underlying mechanisms, partly due to limitations in research methods and the influence of various confounding factors. Our study provides a new and significant understanding in this regard.

In the absence of clinical evidence from randomized controlled trials, we have uncovered the potential causal impact of blood metabolites on the risk of gastrointestinal cancer, providing a theoretical basis for early diagnosis and treatment of GI cancer. These blood metabolites, including Octanoylcarnitine and others, which are considered to potentially increase the risk of GI cancer (see Results section for details), can serve as potential markers for assessing the risk of gastrointestinal carcinogenesis. They also signify the need for timely prevention and intervention against the occurrence of GI cancer, suggesting that theoretically it is feasible to screen early-stage patients by measuring the concentrations of the aforementioned metabolites in the blood. Furthermore, it is reasonable to infer that interventions such as appropriate dietary adjustments can be employed to regulate the concentrations of these metabolites, serving as a strategy for early prevention.

Interestingly, we also found that many metabolites, including Laurate (12:0) and others (see Results section for details), may reduce the risk of GI cancer. These findings provide valuable guidance for clinical translation and offer new avenues for the treatment of GI cancer. They can serve as biomarkers for GI cancer prevention or alternative therapeutic approaches to reduce the occurrence of GI cancer.

Our study has several strengths. Firstly, we employed Mendelian randomization (MR) analysis to investigate the causal relationship between different blood metabolite concentrations and the risk of GI cancer, thereby partially mitigating the influence of potential confounding factors. Secondly, the study included a large sample size available for analysis, effectively reducing sampling errors and providing sufficient power to estimate the causal effects of metabolites on GI cancer.

However, this study also has certain limitations. Firstly, although this study demonstrated potential causal effects of many metabolites on the occurrence of GI cancer, the underlying mechanisms remain unclear and further research is needed for clarification. Secondly, all the GWAS data included in this study are derived from European populations. While this reduces the impact of population stratification on the research, the generalizability of the findings to larger populations of different ethnicities still needs to be further validated, when MR studies primarily utilize data derived from European populations, the associations identified between lipid metabolism and GI cancer risk may exhibit significant population-specific differences, thereby limiting the generalizability of the findings. Firstly, substantial variations exist in genetic background across diverse ancestral populations, encompassing differences in allele frequencies, LDpatterns, and the magnitude of the effects of genetic variants on lipid metabolism. Secondly, dietary habits and environmental factors differ markedly across global regions, and these factors modulate the relationship between genes and phenotypes. Distinct dietary patterns—such as the Mediterranean diet, high-carbohydrate Asian diets, or Western diets rich in animal fats—influence lipid metabolic profiles and may consequently alter the magnitude or direction of the associations between lipid metabolites and gastrointestinal (GI) cancer risk. Additionally, since this study utilized summary data rather than individual-level data, we were unable to conduct stratified analyses based on variables such as gender. Finally, consistent with the vast majority of MR analyses based on publicly available GWAS summary data, this study assumed a linear relationship between metabolite concentrations and GIcancer risk. This assumption is primarily constrained by the nature of large GWAS databases, which typically provide only summary statistics for SNP-exposure associations and lack individual-level exposure-outcome distribution data, thereby precluding direct fitting of non-linear models. While the methods employed in this study—namely, IVWand MR-Egger—are predicated on linear regression frameworks, we fully acknowledge the potential for non-linear relationships. Consequently, future research should consider performing MR analyses stratified by metabolite concentration quantiles within cohorts providing individual-level data, and employing emerging non-linear MR methods, such as Quantile IV, to test for non-linear effects.Nevertheless, it is worth noting that despite these potential limitations, as long as the SNPs used in our study satisfy the three core assumptions of instrumental variables, our research findings remain valid. Furthermore, our conclusions have been supported by a series of rigorous quality control measures and sensitivity analyses, indicating that they are reliable and robust.

Conclusions

In summary, this study demonstrated a causal relationship between blood metabolite concentrations and the occurrence of GI cancer through two-sample Mendelian randomization analysis. Specifically, 36 blood metabolites, including Laurate (12:0), were found to have potential causal effects on the development of GI cancer.

Our study employed a perspective different from observational studies by using genetic variation as instrumental variables to simulate a randomized controlled trial (RCT) and identify this causal relationship. Our findings provide a theoretical basis for guiding clinical practice and further indicate directions for future research.

Abbreviations

1,5-AG

1,5-anhydroglucitol

BK

Bradykinin

CI

Confidence interval

CRC

Colorectal cancer

DABK

des-Arg9-bradykinin

EAC

Esophageal adenocarcinoma

EC

Esophageal carcinoma

EMT

Epithelial-mesenchymal transition

EpOMEs

Epoxyoctadecenoic acid

ESCC

Esophageal squamous cell carcinoma

FAP

Familial adenomatous polyposis

FDR

False discovery rate

FinnGen

Finnish biobank

GC

Gastric cancer

GI

Glycemic index

GIC

Gastrointestinal cancers

GL

Glycemic load

Gln

Glutamine

GWAS

Genome wide association study

IV

Instrumental variable

IVW

Inverse variance weighted

LA

Linoleic acid

LD

Linkage disequilibrium

MR

Mendelian Randomization

MVMR

Multivariate Mendelian Randomization

NCI

The National Cancer Institute’s

ncRNA

Non-coding RNA

NO

Nitric oxide

OR

Odds ratio

RCT

Randomized controlled trail

SD

Standard deviation

Se

Standard error

SNP

Single nucleotide polymorphism

TSMR

Two-sample Mendelian randomization

UCB

Unconjugated bilirubin

WM

Weighted median

XOR

Xanthine oxidoreductase

Author contributions

ZGW and XYG designed research and wrote the paper. ZGW, CW and XYG performed research and analyzed the data. ZGW and XYG reviewed and checked the manuscript.

Funding

No funding support for this work at this time.

Data availability

The original contributions presented in this study are included in the article/supplementary material. Further inquiries can be directed to the corresponding authors.

Declarations

Ethics approval and consent to participate

Since the study utilized the public GWAS database or summary-level data, no additional ethics approval was needed. All primary investigations included in this study received ethical approval from the relevant review boards, and all participants provided informed consent. Moreover, this study did not use any individual-level data.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.