Proteomic Landscape of Colorectal Cancer Derived Liver Metastasis Reveals Three Distinct Phenotypes With Specific Signaling and Enhanced Survival

Abstract

Colorectal carcinoma is a major global disease with the second highest mortality rate among carcinomas. The liver is the most common site for metastases which portends a poor prognosis. Nonetheless, considerable heterogeneity of colorectal cancer liver metastases (CRC-LM) exists, evidenced by varied recurrence and survival patterns in patients undergoing curative-intent resection. Our understanding of the basis for this biological heterogeneity is limited. We investigated this by proteomic analysis of 152 CRC-LM obtained from three different medical centers in Germany and Australia using mass spectrometry-based differential quantitative proteomics. The proteomics data of the individual cohorts were harmonized through batch-effect correction algorithms to build a large multicenter cohort. Applying ConsensusClusterPlus to the proteome data yielded three distinct CRC-LM phenotypes (referred to as CRLM-SD (splice-driven), CRLM-CA (complement-associated), and CRLM-OM (oxidative metabolic)). The CRLM-SD phenotype showed higher abundance of key regulators of alternative splicing as well as extracellular matrix proteins commonly associated with tumor cell growth. The CRLM-CA phenotype was characterized by a higher abundance of proteins involved in the classical pathway part of the complement system including the membrane attack complex proteins and those with antithrombotic activity. The CRLM-OM phenotype showed higher abundance of proteins involved in various metabolic pathways including amino acids and fatty acids metabolism, which correlated in the literature with advanced proliferation of metastases and increased recurrence. Patients classified as CRLM-OM had a significantly lower overall survival in comparison to CRLM-CA patients. Finally, we identified a set of prognosis-associated biomarkers for each group including EpCAM, CEACAM1, CEACAM5, and CEACAM6 for CRLM-SD, DCN, TIMP3, and OLFM4 for CRLM-CA and FMO3, CES2 and AGXT for CRLM-OM. In summary, the discovery of three proteomic subgroups associated with distinct signaling pathways and survival of the CRC-LM patients provides a novel classification for risk stratification, prognosis and potentially novel therapeutic targets in CRC-LM.

Keywords: ECM; EpCAM; alternative splicing; colorectal cancer; complement system; liver metastases; prognosis; proteomics; signaling.

Graphical abstract

Fig. 1

Proteomics workflow of international CRC-LM cohort. Workflow from CRC-LM sample retrieval in three individual studies to LC-MS/MS data integration through batch-effect reduction and subsequent bioinformatical analysis toward definition of individual molecular signatures of three proteomic phenotypes. The fresh frozen tissue slices were homogenized, lysed and the proteins were cleaved into peptides through tryptic digestion. Tryptic peptides were analyzed using LC-MS/MS and proteins were identified and quantified using database search software (DIA-NN for Sydney cohort and Proteome Discoverer for UKE and UKSH cohort). Created in https://BioRender.com. CRC-LM, colorectal cancer liver metastases; LC-MS/MS, liquid chromatography tandem mass spectrometry.

Fig. 2

Three CRC-LM phenotypes revealed through ConsensusClusterPlus. ConsensusClusterPlus algorithm revealed a statistical optimal cluster number of three, visualized through (A) heatmap visualization of hierarchal clustering of all 1667 ANOVA significant (q-value<0.05) proteins identified in 70% of all samples in the international cohort after batch-effect reduction. The (B) principal component analysis of all proteins identified in 70% of all samples is depicted as a scatter plot.

Fig. 3

Differential abundant proteins and associated signaling pathways between CRC-LM subgroups. Volcano Plots (A–C) with all proteins plotted according to their assigned fold change and q-value in the t test of (A) CRLM-SD against the rest, (B) CRLM-CA against the rest and (C) CRLM-OM against the rest. All proteins considered significant abundant (q-value <0.05) are marked in blue (fold change ≤2) and in Group color (fold change ≥2). EnrichmentMap based clustering of gene set enrichment analysis (GSEA) derived gene sets enriched in (D) CRLM-SD, (E) CRLM-CA and (F) CRLM-OM visualized using Cytoscape. For each GSEA analysis one group was tested against all other groups. Encircling of the gene sets was created with the AutoAnnotate function in Cytoscape. CRLM-SD, CRLM-splice-driven; CRLM-CA, CRLM-complement-associated, CRLM-OM, CRLM-oxidative metabolic.

Fig. 4

IPA pathways specific to each CRC-LM phenotype. Canonical pathways extracted from ingenuity pathway analysis (IPA) for (A) CRLM-SD, (B) CRLM-CA, and (C) CRLM-OM. For CRLM-SD the adapted spliceosomal cycle is depicted in (A). For CRLM-CA the complement system pathway is projected in (B) and for CRLM-OM the fatty acid beta oxidation I in (C). All red marked proteins (complexes) were found significantly higher abundant in the assigned group (q-value <0.05; 2 FC). All green marked proteins (complexes) were found significantly lower abundant in the assigned group (q-value <0.05; 2 FC). Gray proteins were identified in the dataset but not significantly abundant. White proteins were not found in the dataset. Orange and blue proteins were predicted to be increased or decreased in concentration based on linked significantly abundant proteins found in the pathway. CRLM-SD, CRLM-splice-driven; CRLM-CA, CRLM-complement-associated; CRLM-OM, CRLM-oxidative metabolic; FC, fold change.

Fig. 5

Kaplan–Meier curve and risk table of overall survival time of patients with CRC-LM assigned to phenotype CRLM-SD, CRLM-CA, or CRLM-OM. The p-value depicted in the plot referred to the survival data difference of the displayed groups. CRC-LM, colorectal cancer liver metastases; CRLM-CA, CRLM-complement-associated; CRLM-OM, CRLM-oxidative metabolic; CRLM-SD, CRLM-splice-driven.

Fig. 6

Biomarker panel specific to each CRC-LM subgroup. Box plot visualization of proteins which are significant high abundant in (A) CRLM-SD, (B) CRLM-CA and (C) CRLM-OM. The log2-transformed, column median normalized abundance values were used. Representative immunohistochemistry (IHC) staining of EpCAM in primary tumor tissue and corresponding liver metastases of three patients from the Sydney cohort. The CRC-LM from (D and E) were assigned to the proteomic subgroup CRLM-SD and (F) to CRLM-CA. EpCAM was only present in the tumor cells and not in hepatocytes. CRC-LM, colorectal cancer liver metastases; EpCAM, epithelial cell adhesion molecule; CRLM-SD, CRLM-splice-driven; CRLM-CA, CRLM-complement-associated, CRLM-OM, CRLM-oxidative metabolic.