Exploring Metabolomic Drivers of Colorectal and Gastric Cancer: A Mendelian Randomisation Study

By Yiming Wang¹, Mi Jian², Jinchen Hu²

Affiliations

1. Department of Clinical Medicine, Shandong Second Medical University, Weifang, China
2. Department of Gastrointestinal Surgery, The Affiliated Yantai Yuhuangding Hospital of Qingdao University, Yantai, China

doi: 10.29271/jcpsp.2025.02.203

ABSTRACT
Objective: To evaluate the causal relationship between 1,400 metabolites and colorectal and gastric cancer.
Study Design: Mendelian randomisation study.
Place and Duration of the Study: The Affiliated Yantai Yuhuangding Hospital of Qingdao University, Yantai, China, from July to August 2024.
Methodology: Metabolite genome-wide association study (GWAS) data and genetic data from the Canadian Longitudinal Study on Aging (CLSA) as well as the expensive FinnGen project, respectively, were sourced. Suitable instrumental variables were chosen based on their association with metabolites at a genome-wide significance level, thus ensuring a high degree of reliability in the causal inferences drawn. Inverse variance weighting (IVW) was used for initial analysis. Sensitivity analyses were conducted using MR Egger regression and weighted median methods to validate findings and assess potential pleiotropy or bias.
Results: Metabolites were included in the study of 8,299 individuals. Gastric cancer included 1,307 cases and 287,137 controls; while colorectal cancer included 6,509 cases and 287,137 controls. The research identified sixty-nine metabolites associated with varying degrees of risk enhancement or mitigation. For gastric cancer, a more focused discovery highlighted two metabolites with significant causal links—associated with increased risk as well as a protective effect. Sensitivity analyses confirmed the validity of these findings.
Conclusion: By elucidating specific metabolites that exert direct causal effects on colorectal and gastric cancer risk, the study marked a significant advancement in the understanding of the metabolic pathways involved in cancer development.

Key Words: Mendelian randomisation, Colorectal cancer, Gastric cancer, Metabolites, Genetic variants, Genome-wide association studies, Causal inference.

INTRODUCTION

Colorectal cancer ranks as the third most common cancer globally. In 2020, there were more than 1.9 million new cases and approximately 935,000 deaths.¹ Gastric cancer was fifth in incidence and fourth in mortality, causing over 1 million cases and 769,000 deaths annually.² The high incidence and mortality rates associated with gastrointestinal cancers underscore the urgent need to further investigate their pathogenesis, which is crucial for early screening, accurate diagnosis, effective treatment strategies, and prognostic assessment.

A recent study indicates that metabolic dysregulation is one of  the key drivers in the development of colorectal and gastric tumours.³

Aberrant metabolism is not merely a downstream consequence of malignant transformation but appears to play an active, causal role in tumour development and progression.⁴ Therefore, identifying the involved metabolic pathways can assist in the discovery of new biomarkers, chemo-preventive targets, and therapeutic strategies. Recent advancements in metabolomics have provided deeper insights into cancer biology, particularly in colorectal and gastric cancers. By capturing the complete metabolic profile of an organism, metabolomics has highlighted key pathways that may influence tumourigenesis. For example, aberrations in energy metabolism, including the Warburg effect, underscore the importance of glycolysis and mitochondrial function in cancer. Disruptions in lipid metabolism, such as increased fatty acid synthesis, have been linked to tumour growth and metastasis. Such findings emphasise the potential of metabolic reprogramming as a hallmark of cancer.

Large-scale studies combining genetic and metabolic data have enabled the identification of novel biomarkers. Recent Mendelian randomisation studies have associated specific lipid species and amino acid levels with gastrointestinal cancer risk, providing a compelling evidence for causal relationships. Nevertheless, a gap remains in systematically evaluating the entire metabolome to uncover new biomarkers and elucidate the mechanisms underlying these associations.

Mendelian randomisation (MR) analysis explores causal relationships between exposures and outcomes by utilising genetic variations as instrumental variables.

MR simulates a natural randomised controlled trial by utilising the random allocation of alleles during gamete formation. Single nucleotide polymorphisms (SNPs) closely associated with specific metabolites are not influenced by confounding factors or disease processes, making them valuable alternative indicators for lifelong metabolite exposure. Associations between these genetic instruments and cancer outcomes are thus likely to reflect the causal influence of lifelong metabolite levels rather than reverse causation. While previous MR studies have examined metabolite-cancer associations for specific metabolites,⁵ a comprehensive analysis across the metabolome could paint a complete picture of the metabolic drivers of colorectal and gastric oncogenesis.

This study aimed to investigate the potential causal relationships between the risk of colorectal and gastric cancer and 1,400 blood metabolites through a hypothesis-free large-scale Mendelian randomisation analysis. The primary objective of this study was to identify metabolites that may have causal relationships with both colorectal and gastric cancer, providing new targets for the treatment of these cancers.

METHODOLOGY

The study evaluated the causal relationships between 1,400 metabolites and colorectal cancer as well as gastric cancer using two-sample Mendelian randomisation (MR) analyses, focusing on their intersections. MR utilises genetic variations as proxies for the risk factors, and effective instrumental variables (IVs) must satisfy three critical assumptions: First genetic variation is directly associated with the exposure; second genetic variation is independent of any confounding factors that may influence both the exposure and the outcome; and the third genetic variation does not affect the outcome through pathways other than the exposure.^6,7 Data on colorectal cancer (6,509 cases, and 287,137 controls) and gastric cancer (1,307 cases, and 287,137 controls) were obtained from the FinnGen project.

Summary genome-wide association study (GWAS) statistics for each metabolite were sourced from the GWAS catalogue (registration numbers GCST9019621 to GCST90201020). This broad-scale GWAS study encompasses data from 8,299 individuals in the Canadian Longitudinal Study on Aging (CLSA), covering 1,091 metabolites and 309 metabolite ratios.

Instrumental variables (IVs) were selected based on a significance threshold of 1 × 10⁻⁵ for metabolite associations. The study applied a linkage disequilibrium (LD) threshold of R² <0.001 and an aggregation distance of 10,000 kb to identify independent sites for IVs using the Two-SampleMR package. For colorectal cancer and gastric cancer, the significance thresholds for IVs were adjusted to 5 × 10⁻⁸ and 5 × 10⁻⁶, respectively, with the same LD thresholds. Inclusion criteria for SNPs included genome-wide significance (p <5 × 10⁻⁸) and LD thresholds (R² <0.001). SNPs with ambiguous strand orientation or poor imputation quality were excluded. To ensure robustness in Two-SampleMR analysis, metabolite data were required to come from non-overlapping cohorts.

Statistical analyses were performed using R 4.2.1. Causal relationships between metabolites and cancer outcomes were primarily assessed using inverse variance weighting (IVW) and weighted median estimation. The IVW method combines Wald estimates from multiple genetic variants, weighted by the inverse variance of each SNP outcome association, to produce robust causal estimates. Weighted-median and mode-based methods were applied as complementary approaches. Cochran's Q test was used as a sensitivity analysis to evaluate heterogeneity among IVs. The TwoSampleMR R package facilitated the estimation, testing, and sensitivity analyses of causal effects in MR studies.

Egger regression was employed to assess and adjust for pleiotropy by examining the relationship between estimated causal effects and their standard errors. A non-zero intercept in Egger regression indicates unaccounted pleiotropy, suggesting potential bias in the causal estimates, while a non-significant intercept indicates minimal pleiotropy. Additionally, the PRESSO (Pleiotropy RESidual Sum and Outlier) model was used to identify and adjust for outlier SNPs that might distort causal estimates. This model operates in two steps by detecting SNPs with significant deviations from expected causal effects and then correcting the causal estimate by excluding these outlier SNPs to minimise pleiotropic effects. By using both Egger regression and the PRESSO model, the study ensured the validity of this analysis of causal estimates (Figure 1).

Figure 1: Flow diagram for quality control of the instrumental variables (IVs) and the entire Mendelian randomisation (MR) analysis process.
SNPs, Single-nucleotide polymorphisms; IVW, Inverse variance weighting; MR, Mendelian Randomisation; MR Presso, Mendelian randomisation pleiotropy RESidual sum and outlier.
 

Table I: Use metabolites that are positive for colorectal cancer to screen for positive metabolites that have a causal relationship with gastric cancer.

Traits	Methods	Pval	Or	Or_lci95	Or_uci95
Gulonate levels	MR Egger	0.015585	0.442591	0.243346	0.804972
	Weighted median	0.116418	0.762262	0.543152	1.069762
	IVW	0.015523	0.729135	0.564535	0.941726
	Simple mode	0.433233	0.762492	0.392597	1.480896
	Weighted mode	0.389017	0.773311	0.436652	1.369533
Dihomo-linoleate (20: 2n6) levels	MR Egger	0.554095	1.183007	0.687059	2.036952
	Weighted median	0.232248	1.257708	0.863368	1.832163
	IVW	0.036052	1.332168	1.018829	1.741875
	Simple mode	0.255066	1.470852	0.776309	2.786785
	Weighted mode	0.616609	1.164563	0.649409	2.088369

Table II: Metabolites with reverse causal relationship with colorectal cancer.

Traits	Methods	Pval	Or	Or_lci95	Or_uci95
Andro steroid monosulfate C19H28O6S (1) levels	MR Egger	0.799532	1.015832	0.900783	1.145576
	Weighted median	0.080138	1.060526	0.992969	1.132679
	IVW	0.007439	1.072755	1.018979	1.129369
	Simple mode	0.635307	1.033312	0.903605	1.181639
	Weighted mode	0.439169	1.045686	0.935102	1.169347
Sphingomyelin (d17:2/16:0, d18:2/15:0) levels	MR Egger	0.017414	0.887585	0.808731	0.974127
	Weighted median	0.362441	0.972433	0.915656	1.032731
	IVW	0.0278	0.95642	0.919199	0.995149
	Simple mode	0.90845	1.007384	0.889537	1.140844
	Weighted mode	0.905766	1.007384	0.892709	1.13679

RESULTS

The causal effect of metabolite on colorectal cancer risk: Sixty-nine metabolites were found to have a causal relationship with the development of colorectal cancer with a p-value of less than 0.05. The PRESSO model identified 12 outlier SNPs, which were excluded to ensure unbiased causal estimates. Cochran’s Q-test indicated minimal heterogeneity among instrumental variables, validating the robustness of this analysis findings. MR Egger regression showed no significant evidence of directional pleiotropy, reinforcing the reliability of causal estimates.

To investigate metabolites that simultaneously influence gastric cancer and colorectal cancer, the study used 69 metabolites with established causal relationships to colorectal cancer to conduct a two-sample MR. The results revealed that Gulonate levels were negatively causally associated with gastric cancer (p = 0.015, OR = 0.729, 95% CI = 0.564~0.941). On the other hand, Dihomo-linoleate (20: 2n6) levels were positively causally associated with gastric cancer (p = 0.036, OR = 1.332, 95% CI = 1.018~1.741, Table I). The PRESSO model revealed that certain metabolites, such as Gulonate and Dihomo-linoleate, had minimal pleiotropic influences after adjusting for outlier SNPs. The exclusion of these outliers resulted in a stronger causal signal for Gulonate's protective role in gastric cancer (OR = 0.729, p = 0.015).

The study further conducted a two-sample MR analysis to assess potential reverse causality, with colorectal cancer as the exposure and 69 metabolites as the outcomes. The primary approach employed was the IVW method, supported by additional methodologies. The results indicated that Andro-steroid monosulfate C19H28O6S (1) levels were positively associated with gastric cancer (p = 0.007, OR = 1.072, 95% CI = 1.018~1.129). Sphingomyelin (d17:2/16:0, d18:2/15:0) levels were negatively associated with gastric cancer (p = 0.027, OR = 0.956, 95% CI = 0.919~0.995) as shown in Table II. There is no reverse causal relationship between other metabolites and colorectal cancer. There is no reverse causal relationship between the two metabolites that are causally related to colorectal cancer and gastric cancer at the same time.

The study used double sample MR analysis, with gastric cancer as the research object and two metabolites as the research results, primarily using the IVW method supplemented by additional methods. The results indicate that there is no interference from reverse causal relationships.

DISCUSSION

The use of genetic instruments minimises the influence of confounding and reverse causation, supporting potentially aetiological links between particular metabolic perturbations and gastrointestinal malignancies. These findings carry important clinical implications, offering insights into cancer prevention, early detection, prognosis, and novel therapeutic opportunities.

The discovery of 69 metabolites causally related to colorectal cancer risk significantly advances the understanding of meta-bolic pathways that may drive or reflect colorectal carcino-genesis. While prior research has connected alterations in amino acid, lipid, carbohydrate, and xenobiotic metabolism to colorectal cancer,⁸ the current analysis provides directional, causal evidence on specific-implicated compounds. Key metabolites such as Gulonate and Dihomo-linoleate showed strong causal associations with cancer risk. Gulonate was protective against both colorectal and gastric cancers, likely due to its antioxidative properties and role in ascorbate metabolism. In contrast, Dihomo-linoleate, a pro-inflammatory lipid, was associated with increased gastric cancer risk, potentially via its role in promoting NF-κB activation and oxidative stress pathways.

Several amino acids emerged with significant causal asso-ciations, including increased risk with higher levels of serine and 3-methoxytyrosine along with decreased risk with higher kynurenine and homoarginine. Serine supports proli-feration in colorectal cancer cells, while 3-methoxytyrosine indicates increased catecholamine turnover.⁹ Elevated homoarginine, synthesised from lysine, has been linked to reduced colorectal cancer risk, potentially via inhibitory effects on NF-κB signalling.¹⁰ Lower kynurenine hints at reduced tryptophan catabolism along the kynurenine pathway, which may limit immuno-suppressive metabolite production.¹¹

Multiple causal lipid metabolites also appear salient to colorectal oncogenesis. Increased colorectal cancer risk was associated with higher levels of arachidonate, 1-arachidonoylglycerol, 1-arachidonoyl-GPC, 1-arachidonoyl-GPE, and various glycerophospholipids containing arachidonate. This implicates pro-inflammatory eicosanoid signalling pathways as potential metabolic culprits.¹² On the other hand, metabolites containing linoleate showed inverse associations, along with ratios such as arachidonate: Linoleate. This aligns with the known chemopreventive role of n-3 polyunsaturated fatty acids such as linoleate in suppressing arachidonate-derived eicosanoids.¹³

Several sphingolipids were also causally related to increased colorectal cancer risk, congruent with their emerging role as oncogenic second messengers influencing key pathways such as mTOR, Wnt/β-catenin, and NF-κB.¹⁴ Meanwhile, the decreased risk was linked to higher levels of plasmalogens such as 1-stearoyl-2-linoleoyl-GPC, which have anticancer properties such as antioxidant effects.¹⁵ The causal links between colorectal cancer and specific lipids, thus converge upon the disruption of bioactive lipid mediators as a potential metabolic driver.

Beyond amino acids and lipids, higher levels of gulonate (ascorbate metabolite) and mannonate (mannose metabolite) were causally associated with lower colorectal cancer risk. This aligns with prior research showing the chemopreventive effects of vitamin C and mannose supple-mentation on colorectal carcinogenesis.¹⁶ Higher bilirubin derivatives also showed inverse causal links, fitting with bilirubin's known anticancer effects.¹⁷

Several novel xenobiotic metabolites possessed previously unidentified causal connections to colorectal cancer, includ-ing andro-steroid monosulfate, 16a-hydroxy DHEA sulfate, and dihydrocaffeate sulfate. Elucidating their underpinning biological mechanisms could reveal new facets of colorectal cancer metabolism. Interestingly, multiple causal associations emerged among metabolite ratios, suggesting that altered metabolic flux could contribute to carcino-genesis. Overall, these causal findings provide critical clues into metabolic pathways that may influence colorectal tumour development and progression. From amino acid turnover to bioactive lipids to carbohydrate metabolites, a complex metabolic signature appears capable of modulating colorectal cancer risk.

Within the broad panel of colorectal cancer-related metabolites, gulonate and dihomo-linoleate (20: 2n6) also showed significant causal associations with gastric cancer risk. Higher gulonate levels were linked with reduced risk, similar to the colorectal cancer results. This highlights gulonate's potential as a pan-gastrointestinal cancer chemopreventive agent warranting further exploration.

Meanwhile, higher levels of dihomo-linoleate (20: 2n6) were causally associated with increased gastric cancer risk. Dihomo-linoleate can be converted to arachidonate and possesses pro-inflammatory properties.¹⁸ This metabolite could contribute to gastric oncogenesis via mechanisms such as increasing oxidative stress and NF-κB activation.¹⁹ Causal relationship between dihomo-linoleate and gastric cancer risk represents a novel finding distinguishing the gastric cancer metabolome from colorectal cancer.

The more limited metabolic associations observed for gastric cancer could reflect its higher anatomical heterogeneity compared to colorectal cancer. However, lipids appear to be a unifying metabolic axis, with dihomo-linoleate implicating inflammatory eicosanoid signalling analogous to the colo-rectal cancer results. Expanded GWAS of gastric cancer sub-types could reveal further causal metabolite associations specific to tumours arising from the cardia versus non-cardia stomach.

The metabolites causally implicated in colorectal and gastric cancer risk by this MR study offer tangible clinical trans-lation opportunities. Firstly, causal metabolites represent prime biomarker candidates for cancer screening, diagnosis, and prognostic risk stratification. Compounds such as 1-arachidonoylglycerol and dihomo-linoleate could be readily measured in blood using mass spectrometry.²⁰ Their circulating levels may correlate with colorectal and gastric cancer susceptibility or prognosis, providing clinically valuable metabolic biomarkers. This study represents the first comprehensive Mendelian randomisation analysis of over 1,400 metabolites, providing novel causal insights into their role in colorectal and gastric cancer risk. By utilising hypothesis-free methods and combining genetic and metabolic data from large cohorts, this research advances the understanding of metabolic drivers in gastrointestinal oncogenesis.

If validated, these biomarkers could improve risk prediction models to enable personalised screening or therapeutic intervention. Metabolite biomarkers could also help diagnose early-stage tumours based on subtle systemic metabolic perturbations that antedate clinical disease. Identifying high-risk subjects via screening could facilitate life-saving interventions such as colonoscopies for colorectal cancer.

Beyond biomarkers, causally implicated metabolites provide natural chemopreventive and therapeutic targets. Supplements or medicines that favourably modulate levels of protective versus risky metabolites could short-circuit cancer development. For instance, vitamin C supplementation to increase gulonate could curb colorectal and gastric cancer risk.²¹ Lipid mediators such as arachidonate offer particularly promising chemopreventive targets. Agents blocking arachidonate synthesis (such as acetylsalicylic acid) or signalling (such as celecoxib) already show efficacy for colorectal cancer prevention.²² Causal signals from this MR analysis could guide more potent and selective interventions.

Agents mimicking or inhibiting causally implicated metabolites also offer provocative treatment possibilities for gastrointestinal cancers. For instance, administering bilirubin mimetics such as bilirubin ditaurate could provide anti-tumour effects.²³ Conversely, inhibitors of oncogenic lipids such as sphingomyelin warrant investigation as novel therapeutics. The causal links unveiled by this study provide a promising starting point for developing metabolite-centered prevention and treatment strategies.

Future research should explore these clinical translation opportunities. Causally-implicated metabolites require validation as blood biomarkers capable of risk stratification via nested case-control studies. Their utility for screening or early diagnosis can be evaluated through prospective clinical trials. Efforts to develop safe and effective modulators of risky versus protective metabolites for cancer prevention represent an intriguing direction. Ultimately, the metabolite-causal associations discovered by this MR study can inspire the next generation of personalised strategies to reduce the burden of lethal gastrointestinal cancers.

This study also has certain limitations. For example, the conclusions of this analysis have not yet been validated through experimental methods. In the future, the study will strive to address this limitation.

CONCLUSION

This study identified several metabolites with significant causal links to gastrointestinal cancers, offering novel bio-markers for risk assessment and potential therapeutic targets. The findings underscore the importance of metabolic reprogramming in cancer pathogenesis and highlight Gulonate and Dihomo-linoleate as potential candidates for clinical translation.

ETHICAL APPROVAL:
The GWAS database is a database of publicly available datasets, where each study has been approved by local institutional review boards and ethics committees.

PATIENTS’ CONSENT:
Not applicable.

COMPETING INTEREST:
The authors declared no conflict of interest.

AUTHORS’ CONTRIBUTION:
YW: Drafted the manuscript.
MJ: Collected and analysed the data.
JH: Provided financial support.
All authors approved the final version of the manuscript to be published.

REFERENCES

1. Sung H, Ferlay J, Siegel RL, Laversanne M, Soerjomataram I, Jemal A, et al. Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 Countries. CA Cancer J Clin 2021; 71(3): 209-49. doi: 10.3322/caac.21660.
2. Bray F, Ferlay J, Soerjomataram I, Siegel RL, Torre LA, Jemal A. Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin 2018; 68(6):394-424. doi: 10.3322/caac.21492.
3. Pavlova NN, Thompson CB. The emerging hallmarks of cancer metabolism. Cell Metab 2016; 23(1):27-47. doi: 10.1016/j.cmet.2015.12.006.
4. Hanahan D. Hallmarks of cancer: New dimensions. Cancer Discov 2022; 12(1):31-46. doi: 10.1158/2159-8290.Cd-21- 1059.
5. Holmes MV, Korpela MA, Smith GD. Mendelian randomisation in cardiometabolic disease: Challenges in evaluating causality. Nat Rev Cardiol 2017; 14(10):577-90. doi: 10.1038/nrcardio.2017.78.
6. Gupta V, Walia GK, Sachdeva MP. Mendelian randomisation: An approach for exploring causal relations in epidemiology. Public Health 2017; 145:113-9. doi: 10.1016/j.puhe.2016. 12.033.
7. Yarmolinsky J, Wade KH, Richmond RC, Langdon RJ, Bull CJ, Tilling KM, et al. Causal inference in cancer epidemiology: What is the role of mendelian randomisation? Cancer Epidemiol Biomarkers Prev 2018; 27(9):995-1010. doi: 10.1158/1055-9965.Epi-17-1177.
8. Inamura K. Colorectal cancers: An update on their molecular pathology. Cancers (Basel) 2018; 10(1):26. doi: 10.3390/ cancers10010026.
9. Li AM, Ye J. Reprogramming of serine, glycine and one-carbon metabolism in cancer. Biochim Biophys Acta Mol Basis Dis 2020; 1866(10):165841. doi: 10.1016/j.bbadis. 2020.165841.
10. Villar VH, Nguyen TL, Delcroix V, Teres S, Bouchecareilh M, Salin B, et al. mTORC1 inhibition in cancer cells protects from glutaminolysis-mediated apoptosis during nutrient limitation. Nat Commun 2017; 8:14124. doi: 10.1038/ ncomms14124.
11. Peyraud F, Guegan JP, Bodet D, Cousin S, Bessede A, Italiano A. Targeting tryptophan catabolism in cancer immunotherapy era: Challenges and perspectives. Front Immunol 2022; 13:807271. doi: 10.3389/fimmu.2022. 807271.
12. Nedeva C. Inflammation and cell death of the innate and adaptive immune system during sepsis. Biomolecules 2021; 11(7):1011. doi: 10.3390/biom11071011.
13. Calder PC. Omega-3 fatty acids and inflammatory processes: From molecules to man. Biochem Soc Trans 2017; 45(5):1105-15. doi: 10.1042/bst20160474.
14. Larrauri AG, Presa N, Herrera AD, Ouro A, Trueba M, Munoz AG. Role of bioactive sphingolipids in physiology and pathology. Essays Biochem 2020; 64(3):579-89. doi: 10.1042/ebc20190091.
15. Paul S, Lancaster GI, Meikle PJ. Plasmalogens: A potential therapeutic target for neurodegenerative and cardiometa-bolic disease. Prog Lipid Res 2019; 74:186-95. doi: 10. 1016/j.plipres.2019.04.003.
16. Lee WJ. Vitamin c in human health and disease: Effects, mechanisms of action, and new guidance on intake. London: Springer; 2019. Available from: httpss://api. semantic-scholar.org/CorpusID:199491079.
17. Ziberna L, Praznikar ZJ, Petelin A. Serum bilirubin levels in overweight and obese individuals: The importance of anti-inflammatory and antioxidant responses. Antioxidants (Basel) 2021; 10(9):1352. doi: 10.3390/antiox10091352.
18. Dicks LMT. How important are fatty acids in human health and can they be used in treating diseases? Gut Microbes 2024; 16(1):2420765. doi: 10.1080/19490976.2024. 2420765.
19. Sugimoto MA, Sousa LP, Pinho V, Perretti M, Teixeira MM. Resolution of inflammation: What controls its onset? Front Immunol 2016; 7:160. doi: 10.3389/fimmu.2016.00160.
20. Wishart DS. Emerging applications of metabolomics in drug discovery and precision medicine. Nat Rev Drug Discov 2016; 15(7):473-84. doi: 10.1038/nrd.2016.32.
21. Li Y, Lin Q, Lu X, Li W. Post-diagnosis use of antioxidant vitamin supplements and breast cancer prognosis: A systematic review and meta-analysis. Clin Breast Cancer 2021; 21(6):477-85. doi: 10.1016/j.clbc.2021.09.001.
22. Skriver C, Maltesen T, Dehlendorff C, Skovlund CW, Schmidt M, Sorensen HT, et al. Long-term aspirin use and cancer risk: A 20-year cohort study. J Natl Cancer Inst 2024; 116(4):530-8. doi: 10.1093/jnci/djad231.
23. Llido JP, Jayanti S, Tiribelli C, Gazzin S. Bilirubin and redox stress in age-related brain diseases. Antioxidants (Basel) 2023; 12(8):1525. doi: 10.3390/antiox12081525.