International Journal of Molecular Sciences · 12 authors, 6 centres
This summary was generated by AI from a single paper. It has not been reviewed by a clinician and is not clinical advice. Verify against the source before acting on it.
This study found that untargeted metabolomics of peripheral blood can distinguish uveal melanoma (UM) patients from controls with near-perfect accuracy (AUC up to 0.99), suggesting a potential noninvasive diagnostic tool. However, metabolite patterns could not differentiate prognostically relevant UM subtypes (BAP1, SF3B1, EIF1AX) or predict metastasis. These findings support further research into blood-based metabolomics for earlier UM detection and referral prioritization.
**Background:** Uveal melanoma (UM) is a rare intraocular malignancy with an incidence of 2–9 per million persons yearly. Prognosis is strongly linked to secondary driver mutations in BAP1 (poor), SF3B1 (intermediate), and EIF1AX (favorable). Current diagnosis relies on specialized ophthalmologic examination, and referral delays are common. Blood-based biomarkers could offer a noninvasive diagnostic alternative, but circulating tumor DNA and cells are often scarce. Metabolomics, which captures the downstream products of dysregulated metabolic pathways, has shown promise for cancer detection. This pilot study aimed to evaluate whether untargeted metabolomics of peripheral blood can discriminate UM patients from controls and distinguish prognostically relevant UM subgroups.
**Methods:** The study included 110 UM patients (56 BAP1, 33 SF3B1, 24 EIF1AX) and 46 non-UM controls from two Dutch biobanks (ROMS and CORRBI). Samples were split into a discovery cohort (n=114: 37 BAP1, 16 SF3B1, 15 EIF1AX, 46 controls) and a replication cohort (n=45: 19 BAP1, 17 SF3B1, 9 EIF1AX). Plasma was obtained from lithium-heparin blood collected within 6 hours, centrifuged, and stored at −80°C. Untargeted metabolomics was performed using ultra high-pressure liquid chromatography-orbitrap mass spectrometry (UHPLC-MS) in both positive and negative ion modes. After normalization (linear regression on internal standards for individual cohorts; Metchalizer for merged dataset), feature abundances were log- and Z-transformed. Supervised and unsupervised analyses included PCA, t-SNE, PLS-DA, and random forest classifiers (RFC) with leave-one-out cross-validation (LOOCV). Performance was assessed by AUC, precision, recall, and F1-score. Pathway analysis was performed using Ingenuity Pathway Analysis on annotated metabolites.
**Key Results:** UM patients and controls showed distinct metabolite patterns. In the discovery cohort, PLS-DA clearly separated the two groups. RFC with LOOCV achieved a precision of 0.96 and recall of 0.79 (positive mode) and precision 1.00, recall 0.94 (negative mode). AUCs were 0.99 (positive) and 0.97 (negative) in the discovery cohort, and 0.99 for both modes in the merged dataset. In silico repeated experiments (10 bootstraps) yielded AUCs ranging from 0.953 to 0.998. In contrast, RFCs could not distinguish UM molecular subtypes: F1-scores for BAP1, SF3B1, and EIF1AX in the discovery cohort were 0.60, 0.00, and 0.00 (negative mode) and 0.53, 0.00, and 0.00 (positive mode). Performance was similarly poor in the replication cohort and merged dataset. Classifiers for primary driver mutations (GNAQ vs. GNA11) and for future metastasis also failed (F1-scores ≤0.64). Pathway analysis revealed upregulation of tRNA charging and glycine usage for creatine biosynthesis, and downregulation of purine ribonucleoside degradation and citrulline metabolism. Several amino acids (arginine, asparagine, cysteine, leucine, lysine, methionine, phenylalanine, proline, tryptophan, tyrosine, valine) were lower in UM patients, while glycine was higher. Adenine, guanine, and D-ribose-5-phosphate were lower, suggesting altered purine metabolism.
**Clinical Implications:** This pilot study demonstrates that untargeted metabolomics of peripheral blood can differentiate UM patients from controls with near-perfect accuracy, potentially enabling earlier detection and reducing referral delays. However, the inability to distinguish prognostic subgroups (BAP1, SF3B1, EIF1AX) or predict metastasis limits its use for prognostication at diagnosis. The identified metabolic pathways (tRNA charging, purine salvage, arginine metabolism) align with known cancer-associated processes and may offer insights into UM biology. Larger prospective studies with well-matched controls (e.g., patients' partners) are needed to validate these findings and address potential confounders such as storage time and comorbidities. If confirmed, this approach could support noninvasive screening and prioritization of referrals for suspicious choroidal lesions.