A study utilized multi-omics machine learning to investigate host-microbiome interactions in early-onset colorectal cancer, revealing distinct molecular profiles and potential biomarkers compared with average-onset disease.
In the study, published in npj Precision Oncology, researchers from the Cleveland Clinic and collaborating institutions analyzed plasma metabolomics and tumor tissue microbiome data from 64 patients with colorectal cancer—20 of whom had early-onset disease, age ≤ 50 years; and 44 of whom had average-onset disease, age ≥ 60 years—using machine learning techniques.
In the early-onset colorectal cancer group, 60% of the patients were male and 100% were White; whereas in the average-onset colorectal cancer group, 59% were male and 91% were White. Tumor characteristics differed between the groups:
- Left-sided tumors: 85% in early-onset colorectal cancer vs 68% in average-onset colorectal cancer
- Rectal primaries: 65% in early-onset colorectal cancer vs 39% in average-onset colorectal cancer
- Stage IV disease: 45% in early-onset colorectal cancer vs 20% in average-onset colorectal cancer.
Although these differences were noted, they were not statistically significant (P > .05 for all comparisons).
The researchers utilized gas chromatography time-of-flight mass spectrometry (GC-TOF-MS) for untargeted plasma metabolomics and 16S rRNA gene amplicon sequencing for tumor tissue microbiome analysis. Machine learning models were constructed using Data Integration Analysis for Biomarker Discovery using Latent variable approaches for Omics studies (DIABLO).
For microbiome analysis, raw 16S amplicon sequences were processed using the Divisive Amplicon Denoising Algorithm (DADA) pipeline. Alpha and beta diversity analyses were performed using the phyloseq and microbiomeSeq packages in R. Permutational Multivariate Analysis of Variance with 999 permutations was conducted on principal coordinates derived from Canonical Correspondence Analysis (CCA).
Differential abundance analysis employed the random-forest algorithm, with lefseq and Anova selected as preferred methods based on DAtest benchmarking results. Statistical significance was set at P < .05, with adjusted P values for multiple comparisons using the Benjamini and Hochberg method to control the False Discovery Rate.
Among the key findings were:
- Distinct clustering patterns in multi-omic dimension reduction analysis for early-onset vs average-onset colorectal cancer
- Identification of 25 metabolites and 10 microbial taxa contributing to the classification model
- Differential correlations between metabolites and microbial genera in early-onset vs average-onset colorectal cancer
- Network analysis revealed distinct clustering of urea cycle metabolites with the microbiome in early-onset colorectal cancer compared with average-onset disease.
The researchers found that plasma metabolomic features more effectively distinguished early-onset colorectal cancer from average-onset colorectal cancer compared with tumor microbiome features, with the metabolomics-based classifier achieving an area under the curve (AUC) of 0.98, while the microbiome-based classifier achieved an AUC of 0.61. The combined multi-omics approach yielded an intermediate AUC of 0.83.
Circular correlation analysis highlighted key associations between metabolites and microbial genera. For instance:
- Glycerol and pseudouridine (higher in average-onset colorectal cancer) negatively correlated with Parasutterella and Ruminococcaceae UCG 002 (higher in early-onset colorectal cancer)
- Cholesterol and xylitol negatively correlated with Erysipelatoclostridium and Eubacterium, but positively correlated with Acidovorax (higher in early-onset colorectal cancer)
- Erythritol, lyxitol, myoinositol, uric acid, and arachidonic acid negatively correlated with Parasutterella, Ruminococcaceae UCG002, and Acidaminococcus.
Network analysis revealed distinct clustering patterns for urea cycle metabolites in early-onset vs average-onset colorectal cancer:
- Urea and ornithine displayed higher centrality in early-onset colorectal cancer
- Citric acid exhibited greater centrality in average-onset colorectal cancer
- Akkermansia showed different centrality between groups, with stronger negative associations with serine and glutamate in early-onset colorectal cancer.
A control group of 49 patients without colorectal cancer was included in the study. The metabolomics-based classifier for non–colorectal cancer controls had a lower AUC of 0.78 in distinguishing young vs old patients compared with the AUC of 0.98 for patients with colorectal cancer. This suggested that the alterations associated with colorectal cancer contributed to the distinction between early-onset and average-onset disease.
The findings suggested that multi-omics analysis could help identify unique host-microbiome interactions associated with early-onset and average-onset colorectal cancer. The robust performance of the metabolomics-based classifier indicated its potential as a biomarker for precise risk assessment and therapeutic intervention development.
The pronounced differences in host-microbiome interactions between early-onset and average-onset colorectal cancer provided insights into potential pathogenic mechanisms that may warrant further investigation. The distinct clustering of urea cycle-related metabolites and microbes in early-onset vs average-onset colorectal cancer may offer opportunities for targeted therapeutic interventions.
Limitations of the study included the lack of an external validation cohort and the inability to determine causality in the observed correlations. The researchers noted that longitudinal analysis and mechanistic studies are crucial to elucidate the relationships from a pathogenesis standpoint and understand the biological implications.
Declaration of interests can be found in the study.