Integrated Clinical-Molecular Classification of Colorectal Liver Metastases

Abstract

ImportancePersonalized treatment approaches for patients with oligometastatic colorectal liver metastases are critically needed. We previously defined 3 biologically distinct molecular subtypes of colorectal liver metastases: (1) canonical, (2) immune, and (3) stromal.ObjectiveTo independently validate these molecular subtypes in the phase 3 New EPOC randomized clinical trial.Design, Setting, and ParticipantsThis retrospective secondary analysis of the phase 3 New EPOC randomized clinical trial included a bi-institutional discovery cohort and multi-institutional validation cohort. The discovery cohort comprised patients who underwent hepatic resection for limited colorectal liver metastases (98% received perioperative chemotherapy) from May 31, 1994, to August 14, 2012. The validation cohort comprised patients who underwent hepatic resection for liver metastases with perioperative chemotherapy (fluorouracil, oxaliplatin, and irinotecan based) with or without cetuximab from February 26, 2007, to November 1, 2012. Data were analyzed from January 18 to December 10, 2021.InterventionsResected metastases underwent RNA sequencing and microRNA (miRNA) profiling in the discovery cohort and messenger RNA and miRNA profiling with microarray in the validation cohort.Main Outcomes and MeasuresA 31-feature (24 messenger RNAs and 7 miRNAs) neural network classifier was trained to predict molecular subtypes in the discovery cohort and applied to the validation cohort. Integrated clinical-molecular risk groups were designated based on molecular subtypes and the clinical risk score. The unique biological phenotype of each molecular subtype was validated using gene set enrichment analyses and immune deconvolution. The primary clinical end points were progression-free survival (PFS) and overall survival (OS).ResultsA total of 240 patients were included (mean [range] age, 63.0 [56.3-68.0] years; 151 [63%] male), with 93 in the discovery cohort and 147 in the validation cohort. In the validation cohort, 73 (50%), 28 (19%), and 46 (31%) patients were classified as having canonical, immune, and stromal metastases, respectively. The biological phenotype of each subtype was concordant with the discovery cohort. The immune subtype (best prognosis) demonstrated 5-year PFS of 43% (95% CI, 25%-60%; hazard ratio [HR], 0.37; 95% CI, 0.20-0.68) and OS of 63% (95% CI, 40%-79%; HR, 0.38; 95% CI, 0.17-0.86), which was statistically significantly higher than the canonical subtype (worst prognosis) at 14% (95% CI, 7%-23%) and 43% (95% CI, 32%-55%), respectively. Adding molecular subtypes to the clinical risk score improved prediction (the Gönen and Heller K for discrimination) from 0.55 (95% CI, 0.49-0.61) to 0.62 (95% CI, 0.57-0.67) for PFS and 0.59 (95% CI, 0.52-0.66) to 0.63 (95% CI, 0.56-0.70) for OS. The low-risk integrated group demonstrated 5-year PFS of 44% (95% CI, 20%-66%; HR, 0.38; 95% CI, 0.19-0.76) and OS of 78% (95% CI, 44%-93%; HR, 0.26; 95% CI, 0.08-0.84), superior to the high-risk group at 16% (95% CI, 10%-24%) and 43% (95% CI, 32%-52%), respectively.Conclusions and RelevanceIn this prognostic study, biologically derived colorectal liver metastasis molecular subtypes and integrated clinical-molecular risk groups were highly prognostic. This novel molecular classification warrants further study as a possible predictive biomarker for personalized systemic treatment for colorectal liver metastases.Trial Registrationisrctn.org Identifier: ISRCTN22944367

Nontumor Cells in Tumor Samples Bias Expression-Based Models—Reply

“…In Reply We thank Liang and colleagues for their comments on our study, which validated the prognostic significance of molecular subtypes and an integrated clinical-molecular classification of liver metastases in the phase 3 New EPOC randomized clinical trial . We hope our comments provide additional context to the considerations raised by the authors.…”

“…First, careful technical measures were taken in both the discovery and validation cohorts to decrease the possibility that our model was biased by nontumor components in the samples. As described in the eMethods in Supplement 2 of the article, tumor-rich areas were identified on histopathologic review of formalin-fixed paraffin-embedded specimens . These tumor-rich sites were macrodissected for nucleic acid extraction to maximize the likelihood of true tumor being profiled (as opposed to liver parenchyma).…”

“…This underscores the importance of profiling the metastatic tumor to develop prognostic biomarkers. A metastasis-based expression model is actually a strength of our study …”

“…In response to the authors’ second point, we excluded messenger RNA/microRNA (miRNA) features that had low variability or overall expression across patients when training the model . In our discovery cohort, MIR548X and MIR8072 both exceeded the 75th percentile for expression.…”

Bioinformatics analysis reveals link between alternative complement cascade pathway and colorectal cancer liver metastasis

“…Different data processing may produce different outcomes. For example, Katipally et al [9] evaluated resected metastatic tumors by RNA sequencing and microRNA (miRNA) analyses and found an increased number of DEGs associated with DNA repair and cell cycle regulation/proliferation pathways (including E2F, G2M, and mitotic spindle pathways). Most bioinformatics analyses of CRC have screened the differential genes and molecular markers between normal colon tissue and CRC tissue or between CRC and CRC liver metastases, focusing on prognostic value.…”