Single-cell and bulk transcriptomic analysis of inflammation-metabolism programs and the tumor microenvironment of colorectal cancer liver metastasis

Background Colorectal cancer liver metastasis (CRLM) is a major cause of mortality in colorectal cancer (CRC) and develops within a heterogeneous tumor microenvironment (TME). How immune, stromal, and metabolic programs differ between primary CRC and matched liver metastases remains incompletely resolved. Methods We analyzed publicly available single-cell RNA sequencing (scRNA-seq) data from GSE178318 (15 specimens from six patients: six primary CRC samples, six matched liver metastases, and three peripheral-blood mononuclear cell samples) together with bulk RNA-seq data from GSE50760 (54 samples from 18 patients spanning normal colon, primary CRC, and liver metastasis). The single-cell workflow included quality control, score-based cell-type annotation, patient-level pseudobulk differential expression, pathway/module scoring, ligand-receptor co-expression analysis, and diffusion-map pseudotime. Bulk data were assessed using within-patient paired comparisons, co-expression analysis, and program-level scoring. Benjamini-Hochberg false-discovery-rate (FDR) correction was applied for multiple testing. Results After quality control, 132,915 single cells were retained. Liver metastases showed higher patient-level T/NK-cell fractions and lower malignant epithelial and B/plasma-cell fractions than primary tumors. In paired single-cell pseudobulk analysis, no gene remained significant at the prespecified FDR threshold, although nominal differences suggested extracellular-matrix and metabolic changes. Glycolysis scores were modestly higher and oxidative-phosphorylation scores modestly lower in liver metastases, whereas inflammatory and glycolytic scores were not correlated at the patient-tissue level. Ligand-receptor co-expression was strongest for MIF-CD74, VIM-CD44, and HLA-A-CD8A, with CAF/stromal and myeloid/TAM populations showing high aggregate outgoing scores. Diffusion pseudotime showed increasing EMT/invasion scores and decreasing glycolysis along the inferred transcriptomic ordering; this ordering is descriptive and not a longitudinal lineage. In the bulk cohort, 654 genes differed between liver metastasis and primary tumor, while cross-platform rank concordance was weak (Spearman rho=0.106). Conclusions Public single-cell and bulk datasets support compartment-associated differences in cellular composition and transcriptional programs in CRLM, with prominent stromal, myeloid, and metabolic features. However, limited patient numbers, cross-sectional sampling, weak cross-platform concordance, and the absence of experimental validation mean that the ligand-receptor and trajectory findings should be interpreted as hypothesis-generating rather than causal or clinically validated.