Multiple Metastatic Colorectal Cancer Organoids Reveal Within-Patient Drug Heterogeneity
A collection of 213 organoids from 102 patients compared liver metastases, molecular features and drug response. In two selected clinical cases, chemotherapy rechallenge produced a response, while metastases progressed during the targeted treatment.
Sampling More Than One Metastasis
A Norwegian academic and hospital collaboration established a living organoid collection from resected colorectal liver metastases. Of 346 lesions sampled from 132 patients, 213 lesions from 102 patients produced organoids. Multiple spatially separated metastases were represented for 65 patients, allowing drug-response differences within a patient to be compared with differences between patients.
The median time from sampling to a completed drug screen and cryopreservation was 8.4 weeks, with a 10th–90th percentile range of 4–21 weeks. Models came from microsatellite-stable tumors in a population eligible for liver resection. This clinical setting defined the collection's biological and treatment context.
A Pharmacological and Molecular Map
The screening used two custom libraries, containing 41 and 47 drugs, with 24 shared agents supporting cross-library comparisons. CellTiter-Glo 3D measurements after 96 hours were used to fit concentration-response curves and calculate drug-sensitivity scores. Technical replicates and library-specific quality filters supported the downstream comparisons.
Histology divided the organoids into cystic, solid and mixed patterns. Cystic models were generally more drug-sensitive and enriched for a cancer-stem-cell-associated expression pattern. Solid models more often carried RAS or BRAF V600E alterations and showed different transcriptional programs. The study connected these morphological and molecular features with pharmacology rather than treating all organoids as an interchangeable format.
How Much Did Metastases from One Patient Differ?
Within-patient drug-response differences were smaller overall than differences between patients. However, they remained substantial in some individuals. Patients with organoids of different morphological phenotypes had greater pharmacological heterogeneity, and solid organoids were generally less sensitive than cystic organoids within the same patient.
Morphology did not explain every case. The highest within-patient heterogeneity score occurred in a patient with three cystic organoids and heterogeneous APC mutation status. Despite their differences across several drugs, all three models were highly sensitive to the EGFR inhibitor afatinib and showed similar responses to an irinotecan-based combination. The corresponding lesions had similar responses to the patient's earlier treatment.
Two Prospective Treatment Examples Had Different Outcomes
For one patient, organoids showed activity of the EGFR–ERBB2 inhibitor lapatinib and tissue analysis showed ERBB2 amplification. The patient subsequently received trastuzumab plus pertuzumab, but the metastases progressed. Heterogeneous ERBB2 expression and an ERBB2 S310F mutation were potential explanations.
A second patient's organoids were highly sensitive to SN-38 and irinotecan-related combinations after several treatment lines. Reintroducing FOLFIRI produced a partial response in the largest liver lesions and a reduction in most lung metastases, with an approximately 50% reduction in total disease burden at the first evaluation. Treatment continued for a further 60 days before lung progression.
Culture Selection Remains Part of the Result
Organoids that grew successfully came from patients with poorer subsequent survival than those whose samples failed to grow. Lesions that did not propagate also tended to have responded more strongly to neoadjuvant treatment. Moreover, drug sensitivity in the established cultures did not reflect the lesions' responses to the corresponding preceding neoadjuvant regimens.
These findings make culture selection a substantive limitation of the resource. The two selected treatment examples do not supply a cohort-wide predictive-accuracy estimate, and systematic co-clinical outcome data were unavailable. The collection nevertheless provides a detailed way to study how metastatic sampling, tumor phenotype and assay conditions influence the drug responses obtained from patient-derived models.