Background
Early diagnosis of colorectal cancer (CRC) requires reliable non-invasive biomarkers. Plasma metabolomic approaches have shown promise in accurately distinguishing CRC patients from controls and individuals with precancerous lesions. Furthermore, integrative multi-omics analysis methods are now available to explore molecular features that may be functionally related to observed metabolomic alterations, including gut microbiome dysbiosis and altered transcriptional profiles.
 
Objectives
Identification of plasma metabolomic features characteristic of CRC across two independent European cohorts and integrative analysis of their levels relative to patient tissue transcriptomics and gut microbiota profiles.
 
Methods
We performed untargeted plasma metabolomics in samples from 444 individuals (158 patients with CRC, 64 with precancerous lesions, 96 with other gastrointestinal diseases, and 126 healthy controls) from Italy and the Czech Republic. Data were integrated with CRC and adenoma tissue transcriptomics (RNA-Seq and small RNA-Seq), mutational profiles (TruSight Oncology 500 target sequencing), and fecal shotgun metagenomic sequencing.
 
Results
Ninety-nine metabolite clusters differed significantly (adj. p < 0.05) in CRC patients compared with healthy controls, as identified in age-, sex-, BMI-, and smoking-adjusted generalized linear models. In addition, 18 were identified in subjects with colorectal adenomas, including decreases in Indole-3-propionic acid and Phenylacetylglutamine, and two metabolites shared with the CRC patients subset.
Among the distinct metabolites identified in CRC patients, 82 showed a consistent pattern across both cohorts (ρ = 0.47, p < 0.001), with coherent alterations in patients with severe gastrointestinal inflammation (ρ = 0.85, p < 0.001) or precancerous lesions (ρ = 0.48, p < 0.001). CRC patient clustering based on these coherent metabolite profiles revealed two main clusters that differed significantly in lesion localization (p < 0.05). Conversely, 17 CRC-associated metabolites showed an opposite trend across the two cohorts, including Cortisol, Creatinine, and Leucine, which increased in Italian but decreased in Czech CRC patients.
Feature selection and supervised machine learning analyses identified candidate signatures of metabolomic and metagenomic features with high classification performance (AUC > 0.85), including Hypoxanthine and Bromotryptophan.
Finally, twenty-four significant correlations (adj. p < 0.05) were identified between the coherent metabolites and the eighteen differentially abundant microbial species (adj. p < 0.05) from the same CRC patients and healthy controls, including Hypoxanthine with Solobacterium (SGB6833) (ρ = 0.28, adj. p < 0.05).
 
Conclusions/Implications
Our results confirm the efficacy of plasma metabolomics for identifying precancerous and cancer-specific signals for early disease diagnosis and, possibly, for dissecting the metabolic alterations occurring in patient tissues.
The identified metabolic features are undergoing in-depth characterization, and their dysregulation patterns are evaluated across healthy controls and different disease stages. In addition, integration with other -omics profiles is ongoing to infer the origins of circulating metabolites and their interplay in host-microbiome interactions in CRC patients.