Metabolomics signature as a survival predictor in patients with resectable colorectal liver metastasis

Abstract

Metabolomics signature as a survival predictor in patients with resectable colorectal liver metastasis Dear Editor, Colorectal cancer (CRC) is the third most common cancer diagnosed worldwide, with over 1.9 million new cases per year (.9 million deaths) in 2020. 1 In Spain, CRC was the most frequent cancer in 2022, with colorectal liver metas-tasis (CRLM) representing the leading cause of death. 2 The treatment of choice for metastatic patients with potential survival benefit is surgery, but more than 50% relapse. 3 Therefore, there is an urgent need to anticipate disease progression and prolong survival by defining predictive and prognostic biomarkers in CRLM patients after hepatic resection. Metabolomics has been previously used to detect CRC biomarkers, 4 but this is the first report that identifies specific metabolome alterations related to survival expectancy in a metastatic setting. 5 In this pilot study (Figure 1), we analysed paired plasma samples of 39 patients with CRLM from the University Hospital of Jaén (Sections 1.1 and 2.1 in Supporting Information and Table S1) according to pre-and post-hepatic resection using untargeted metabolomics (Figure S1). Our research aims to determine metabolomics differences after surgery, when metastatic disease is still present versus successfully eliminated. This will shed light on the specific metabolic changes associated with relapse and survival, enabling the creation of a risk metabolomics score. The risk score could help to define which patients need close monitoring or more intensive treatment, even before the manifestation of recurrence symptoms. Once we filtered metabolomics data matrices, the clustering of quality control samples in unsupervised principal component analysis confirmed the analytical stability of our methodology (Section 1.2 in Supporting Information and Figure S1). The ability to discriminate between presence or absence of metastasis in CRLM patients was determined by supervised partial least square-discriminant analysis (PLS-DA) (Figure 2A,B). Dysregulated metabo-lites between paired samples with a p-value lower than .05 by Student's t-test with Benjamini-Hochberg false discovery rate correction were selected. Metabolites with a fold change (FC) > 1.3 were identified between pre-and post-This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. surgery samples (Sections 1.4-1.6 and 2.2 in Supporting Information). Statistical analyses showed that 346 metabo-lites were differentially expressed pre-and post-surgery, which could be used to discriminate the metastatic status. Besides the paired analysis, we hypothesised that in patients where the disease is still present after resection of metastasis, the detection of onco-metabolites could predict CRC prognosis. For this analysis, we considered only post-surgery samples and metabolites with a FC > 1.3 and variable important of projection >1 between the groups of recurrence. Interestingly, we found that PLS-DA could discriminate between recurrent and non-recurrent patients (Figure 1C,D). The discrimination was based on 57 differentially expressed metabolites between these groups. Among the 57 metabolites, 20 candidates were identified (Table 1 and Section 1.5 in Supporting Information) and the rest remained unidentified (Table S6). The differences involved molecular changes in the metabolism of taurine and hypotaurine, the biosynthesis of primary bile acids (BAs), and the biosynthesis of phenylalanine, tyro-sine and tryptophan (Figure 2E and Sections 1.7 and 2.2 in Supporting Information). The results demonstrate that the early detection of onco-metabolites could help in predicting the risk of disease recurrence after surgery and guide treatment decisions for optimal clinical management in a metastatic scenario. We built a metabolomics model with the most predictive markers identified according to the value of the multivariate area under the curve (AUC-ROC). A precise model including 13 compounds showed the highest prediction ability (AUC = .793, 95% confidence interval [CI]: .585-.974, p = .023; Section 1.7 in Supporting Information and Figure S2). Finally, to assess the prognostic value of these candidate metabolites, we used a univariate Cox-regression analysis and Kaplan-Meier curves (Sections 1.8 and 2.3 in Supporting Information and Table S7). The stratifi-cation of patients based on a potential metabolomics risk score (mRS) revealed that patients with an mRS of more than six candidate metabolites (high risk to Clin. Transl. Med. 2024;14:e1541. wileyonlinelibrary.com/journal/ctm2 1 of 6 https://doi.