Exploring metabolomics for colorectal cancer risk prediction: evidence from the UK Biobank and ESTHER cohorts

Abstract

Background: While metabolic pathway alterations are linked to colorectal cancer (CRC), the predictive value of pre-diagnostic metabolomic profiling in CRC risk assessment remains to be clarified. This study evaluated the predictive performance of a metabolomics risk panel (MRP) both independently and in combination with established risk factors.

Methods: We derived, internally validated (IV), and externally validated (EV) a metabolomics risk panel (MRP) for CRC from data of the UK Biobank (UKB) and the German ESTHER cohort. Baseline blood samples were assessed for 249 metabolites using nuclear magnetic resonance spectroscopy analysis. We applied LASSO Cox proportional hazards regression to identify metabolites for inclusion in the MRP and evaluated the model performance using the concordance index (C-index). We compared the performance of the MRP to an environmental risk panel (ERP; sex, age, body mass index, smoking status, and alcohol consumption) and a genetic risk panel (GRP; polygenic risk score).

Results: The study included 154,892 participants of the UKB cohort (mean age at baseline 54.5 years; 55.5% female) with 1879 incident CRC and 3242 participants of the ESTHER cohort (mean age 61.5 years; 52.2% female) with 103 CRC cases. Twenty-three metabolites, primarily amino acid and lipid-related metabolites, were selected for the MRP, showing moderate predictive performance (C-index 0.60 [IV] and 0.54 [EV]). The ERP and GRP showed superior performance, with C-index values of 0.73 (IV) and 0.69 (EV). Adding the MRP to these risk models did not change the C-indices in both cohorts.

Conclusions: Genetic and environmental risk information provided strong predictive accuracy for CRC risk, with no improvements from adding metabolomics data. These findings suggest that metabolomics data may have limited impact on enhancing established CRC risk models in clinical practice.

Keywords: Biomarkers; Colorectal cancer; Metabolomics; Risk stratification.

Fig. 1

Flow diagram showing selection of study participants from the UK Biobank and the ESTHER cohorts. CRC, colorectal cancer

Fig. 2

Data processing and analyses flow diagram. The figure summarizes the development and validation of the metabolomics risk panel (MRP). Participants from the UK Biobank were split into a training set (70%) and a testing set (30%). In the training set, LASSO Cox regression identified predictive metabolites, and the MRP was validated in the testing set and the ESTHER cohort. Predictive performance was assessed using C-index and compared to the genetic risk panel (GRP) and environmental risk panel (ERP), individually and in combination. CRC, colorectal cancer

Fig. 3

Hazard ratios (95% CI) for the selected metabolites, by cohort. Cause-specific Cox proportional hazards models were used to estimate hazard ratios (HR) and 95% confidence intervals (CI) for the association between selected metabolites and CRC risk. Age at recruitment was defined as study entry, and exit time was determined by CRC diagnosis, death, or end of follow-up. HRs were reported per 1-SD increase of the log1p-transformed value for each metabolite. Multiple testing correction was applied using the Benjamini–Hochberg method. Full dots indicate metabolites that remained significant after correction, while hollow dots indicate non-significant associations. BCAA branched-chain amino acids, C cholesterol, CE cholesteryl esters, CI confidence interval, FA fatty acids, FB fluid balance, FC free cholesterol, Glycolysis glycolysis-related metabolites, HDL high-density lipoproteins, HR hazard ratio, Ketone ketone bodies, L large, LA linoleic acid, LDL low-density lipoproteins, M medium, PL phospholipids, S small, SD standard deviation, TG triglycerides, UKB UK Biobank, VLDL very low-density lipoproteins, XL very large, XS very small, XXL extremely large, % ratio