Matched Colorectal and Liver-Metastasis Models Link Drug Responses with Molecular Profiles
Five patient-matched model sets combined sequencing, exosomal miRNA and 21-drug screening. The associations nominated metastasis-related features, but were not validated as predictors of patient treatment response.
Comparing Primary and Metastatic Models Within Patients
Seoul National University researchers assembled matched colorectal cancer and liver-metastasis models to study how molecular features relate to drug response. Their Translational Oncology paper combined whole-exome sequencing, RNA sequencing, exosomal microRNA measurements and pharmacology, then used supervised analyses to connect the data layers.
The collection represented five patients and 11 organoid cultures. Four matched sets were generated from early-passage cell lines, while one was derived directly from tissue. One patient had received neoadjuvant chemoradiation before sampling; the others had not received neoadjuvant treatment.
The source tumors were not preserved for a direct genomic comparison with the resulting models. The investigators could therefore compare primary-site and metastatic organoids with one another, but could not confirm their genomic fidelity to the original patient tissues.
Shared Drivers Coexisted with Patient-Specific Expression Patterns
Primary and liver-metastasis models from the same patient generally shared similar somatic mutation profiles, including familiar colorectal cancer drivers. Transcriptomic profiles also showed strong within-patient similarity. Morphology did not clearly separate primary and metastatic cultures.
Differential-expression analysis identified 21 genes upregulated in liver-metastasis models under the study’s thresholds. Four were also upregulated in a public dataset with a similar comparison, and CENPI was statistically significant in both cohorts. Pathway patterns were heterogeneous across patients: epithelial–mesenchymal transition and angiogenesis signatures were often higher in primary-site cultures, rather than uniformly enriched in metastatic models.
Exosomal microRNAs were measured in conditioned culture medium. The team checked CD63 and calnexin to assess the preparation and used quality-filtered array probes for analysis. Predicted microRNA target genes were compared with observed RNA-expression changes, producing patient-dependent candidate pathways. In the directly tissue-derived case, too few target genes met the criterion to yield a significant biological-process enrichment.
Drug Screening Supplied a Third Layer for Integration
The pharmacology panel contained 21 compounds, including conventional colorectal chemotherapy agents and targeted drugs. Screens used six serial concentrations and were repeated at least three times. The area under each dose-response curve summarized drug sensitivity for the subsequent integration.
The team used DIABLO, a supervised multi-omics integration method in the mixOmics R package, to combine 319 messenger-RNA features, 296 microRNA features and 21 drug-response features, with primary versus metastatic origin supplied as class labels. It identified 15 multi-layer signatures that separated those classes within the collection. Examples linked miR-3613-5p and eEF2K with TAS-102 response, and miR-663b and HTD2 with apitolisib response.
These are statistical connections among measured features. The study did not directly demonstrate that the candidate microRNAs regulate the proposed targets or cause the drug-response changes. The authors identify gain- and loss-of-function experiments as necessary to resolve those mechanisms.
Classification Features Remained Exploratory
Additional automated modeling used 102 predictors drawn from drug responses, pathway scores and microRNA expression. A KRAS-related expression signature and regorafenib response contributed to a decision-tree separation of primary and liver-metastasis models. TGF-β signaling was a leading discriminator, with apitolisib response and miR-3613-5p also among the selected variables.
The KRAS signature was the HALLMARK_KRAS_SIGNALING_DN expression set, whose lower enrichment the authors interpreted as relatively greater KRAS activity. Its relationship to regorafenib response varied with molecular context, and a direct causal mechanism was not tested.
With five patients, many molecular predictors and no independent patient-response cohort, these analyses cannot establish a generalizable clinical prediction rule. The collection instead provides matched experimental resources and candidate molecular–pharmacological relationships for testing. Its mixed derivation histories and absent parental-tissue comparison remain relevant when those hypotheses are carried into another model system.