SCOPE separates cytostatic and cytotoxic drug responses in colorectal organoids

What the work claims

This is a methods-and-validation paper that introduces SCOPE, the Systematic Classification of Organoids for Phenotypic Evaluation. The core argument is that a drug's effect on organoid growth and its effect on organoid viability are separable, and that capturing both over time produces a more informative classification than static viability alone. The authors apply SCOPE to six patient-derived colorectal cancer (CRC) organoid lines treated with standard chemotherapies and to a 166-drug FDA-approved oncology panel screened in one line, and they show that the resulting phenotypic classes align with known mechanisms and reveal patient-specific differences that CellTiter-Glo misses.¹

The central claim is not merely methodological. The authors argue that distinguishing cytostatic from cytotoxic responses, and measuring the dose gap between growth inhibition and killing, can guide dose selection and combination design in ways that static assays cannot.

How it works

SCOPE combines three elements: automated 3D imaging of live and dead organoids at multiple time points, a neural-network-based segmentation pipeline, and a Gompertz growth model that translates individual organoid size trajectories into a growth score. A separate viability score is derived from the proportion of organoids that remain alive. From these two scores the authors compute a growth-viability (GV) plot and a cytostatic-cytotoxic transition range (CCTR), defined as the log10 distance between the IC50 for growth inhibition and the IC50 for viability loss.¹

The CCTR is the conceptual hinge. A small CCTR means growth inhibition and killing occur at similar doses; a large CCTR means a drug can stop growth at low doses but needs far higher, and potentially more toxic, doses to kill cells. The authors illustrate this with SN38, the active metabolite of irinotecan, across six CRC organoid lines. In most lines the growth-score dose-response curve resembled the CellTiter-Glo curve, but the viability-score curve separated from the growth curve by varying degrees, producing patient-specific CCTRs. In the UX line, for example, many organoids remained alive even at 100 µM, producing a large or effectively infinite viability IC50 despite clear growth inhibition.¹

The GV plot then classifies each drug into one of four phenotypic groups: cytotoxic, cytostatic plus cytotoxic, late cytotoxic and cytostatic. To test whether the classification is meaningful, the authors screened 166 FDA-approved anti-cancer compounds at 10 µM in the UP organoid line and manually reviewed images for 123 drugs that had any detectable effect. Only four classification mismatches occurred, and those were edge cases that could plausibly belong to two classes. The examples are concrete: osimertinib was rapidly cytotoxic, niraparib showed cytostatic-plus-cytotoxic behavior, fludarabine killed late, and zanubrutinib was cytostatic with little killing.¹

Finally, the authors apply SCOPE to EGFR inhibitors across all six lines. Gefitinib and osimertinib produced lower growth-inhibition IC50s in KRAS wild-type lines (U7, UA and UP) than in KRAS mutant lines (US, UX and V8), consistent with clinical experience. Within the KRAS wild-type group, the UP line had a higher IC50-growth and a larger CCTR, which the authors link to KRAS amplification rather than mutation. The US line, despite carrying a KRAS mutation, showed gefitinib sensitivity comparable to some wild-type lines, a reminder that genotype alone does not fully predict phenotype.¹

Where a skeptic should push

The most important assumption is that the phenotypic classes measured in a short-term organoid assay map to clinical outcomes. Growth inhibition in a dish is not regression in a patient. A cytostatic response can look favorable in the model while allowing residual disease to persist in vivo, and a cytotoxic response can be accompanied by toxicity that makes the dose unusable. SCOPE provides a better readout of mechanism, not a validated clinical predictor.

Sample size and scope are limited. The 166-drug screen is performed in a single organoid line, and the EGFR analysis spans six lines. That is enough to demonstrate the concept, but it is not a robust benchmark against patient response data. The authors do not report prospective clinical correlations, so claims about "clinical applicability" remain aspirational.

There is also a technical question about the imaging pipeline. Neural-network segmentation of brightfield or fluorescence images can be sensitive to organoid density, matrix background and culture format. The authors show consistency within their own workflow, but transfer to another lab's microscope, plate type or staining protocol would need revalidation. If the CCTR is fragile to imaging conditions, its value as a standardized metric diminishes.

What this changes for organoid-based drug discovery

The non-obvious implication is that organoid screens may have been systematically misranking cytostatic drugs. Many targeted therapies work by arresting proliferation rather than killing outright. In a static ATP assay, a cytostatic drug can score as weak or inactive because the readout conflates growth arrest with death. SCOPE separates the two, which means compounds that looked like failures in conventional screens may deserve a second look, and compounds that looked potent may be merely cytostatic at clinically relevant doses.

The opportunity is dose and combination optimization. A large CCTR tells you that a low dose can control growth but that a different agent, or a much higher dose, will be needed to eliminate the tumor. That is exactly the kind of information needed to design rational combinations: pair a cytostatic anchor with a cytotoxic partner, or escalate dose only in patients whose CCTR is small enough to expect killing. For organoid-based precision oncology services, reporting a CCTR alongside an IC50 could make the output more actionable for oncologists.

The threat is added complexity without added validation. SCOPE requires live imaging over multiple days, segmented organoid tracking and mathematical modeling. That is more expensive and slower than a single endpoint assay. If the field adopts the richer readout before demonstrating that it predicts patient outcomes better than simpler assays, biopharma could end up paying a throughput penalty for a metric that is mechanistically elegant but clinically unproven. The authors acknowledge this by framing SCOPE as a complement to, not a replacement for, existing platforms.

The bottom line

Established: SCOPE can classify drug responses in patient-derived colorectal organoids into four phenotypic groups using multi-timepoint imaging and growth modeling, and the classification agrees with manual image review for 119 of 123 drugs with detectable effects. The CCTR captures patient-specific differences in the dose gap between growth inhibition and killing. Unestablished: whether the phenotypic classes or CCTR values predict clinical response, whether the assay is portable across labs, and whether the added cost improves decision-making over simpler endpoints. What would confirm the value is a co-clinical study correlating SCOPE outputs with patient outcomes. What would weaken it is evidence that the CCTR is dominated by imaging artifacts or that static assays perform equally well for predicting clinical benefit. For the foundry, SCOPE is a credible next step in making organoid pharmacology more mechanistic, provided it is not adopted as a clinical oracle before validation.