How Model Context Changed What a PDAC Drug Discovery Study Could Reveal
In pancreatic cancer, response is shaped by more than tumor cells alone. In our June 2026 webinar, Martyna Solecka of Bristol Myers Squibb described how a more context-rich 3D model revealed biology and drug-response insights that simpler systems missed.
Confidence in drug discovery is often discussed at the back end of the process, once data is in and teams are deciding what to advance. In reality, that confidence starts much earlier. It starts with the model.
That is especially true in pancreatic ductal adenocarcinoma, or PDAC, where tumor behavior is driven by far more than the epithelial compartment alone. Matrix composition, tissue stiffness, stromal signaling, and tumor-stroma interactions all influence how disease progresses and how therapies perform. Remove too much of that context and the model becomes easier to run, but less likely to reflect the biology discovery teams actually need to understand.
That challenge sat at the center of a recent webinar featuring Martyna Solecka of Bristol Myers Squibb, who shared how her team developed a more representative 3D PDAC model using patient-derived organoids, cancer-associated fibroblasts, and RASTRUM bioprinting technology from Inventia Life Science. The work asked a practical question with broad relevance for discovery teams: what changes when more of the biology shaping response is built into the model from the start?
The data presented showed that adding stromal and mechanical context changed morphology, altered signaling, preserved fibroblast heterogeneity, and exposed a drug-combination signal that did not appear in the simpler systems tested alongside it.
PDAC exposes the limits of reductionist models
PDAC remains one of the most difficult solid tumors to treat. Prognosis is poor, chemotherapy still carries much of the burden of care, and progress has been slowed not only by clinical complexity but by the challenge of modeling the disease in a way that reflects patient biology.
A large part of that challenge sits in the tumor microenvironment. PDAC is highly stromal, with fibrotic stroma making up most of the tumor mass and CAFs contributing to oncogenesis, signaling, and treatment resistance. For a disease like this, biology does not happen in isolation. Response is shaped by the structure around the tumor cells, the extracellular matrix beneath them, and the signaling exchanged across the surrounding tissue.
That immediately creates a problem for standard preclinical systems. Traditional 2D models are limited in obvious ways. They do not capture spatial organization, extracellular context, or the cell-cell and cell-matrix interactions that shape tumor behavior. Matrigel organoid systems get closer, but PDAC still pushes past what they can represent. They typically lack a meaningful stromal compartment, they do not reproduce pathological tumor stiffness, and they miss extracellular matrix components that matter in fibrotic disease.
Those omissions matter in PDAC. As Solecka put it during the webinar, “Unfortunately, our current preclinical models don't capture the complexity of tumor microenvironment and don't model stroma very well. So we often miss key factors that influence treatment response and this is ultimately why progress in PDAC has been so slow, because we simply don't have good enough models to get us there.”
Building a model with more of the relevant biology in place
The Bristol Myers Squibb team started with patient-derived PDAC organoid material representing a range of derivation sites, treatment histories, transcriptional subtypes, and driver mutations. For some lines, they also had access to matched CAFs from the same patient. That gave them a biologically diverse starting point. The next step was to build an assay environment capable of supporting both compartments in a more defined and disease-relevant way.
Using RASTRUM, the team moved beyond standard Matrigel culture by tuning both matrix composition and stiffness. The resulting model incorporated PDAC-relevant extracellular matrix components, including collagen I, fibronectin, hyaluronic acid, and laminin, while reaching stiffness levels more representative of the disease state. The system also supported CAF growth, which made it possible to study tumor-stroma interactions directly rather than treating the epithelial compartment as the whole disease.
The coculture conditions were then optimized across cell density, PDAC-to-CAF ratio, matrix formulation, and media composition. The resulting model provided a biological setting for asking drug discovery questions in PDAC.
The added context influenced the biology they could measure
Once the coculture was established, the differences were visible almost immediately.
Organoids grown on their own appeared larger and more spherical. In coculture with CAFs, they became more irregular, more elongated, and more physically associated with stromal cells. Those changes were backed by transcriptomic and qPCR data showing increased extracellular matrix remodeling, fibroblast activation, EMT-associated signaling, TGF-β pathway activity, and tumor-stroma crosstalk.
The model also preserved CAF heterogeneity, including markers associated with both inflammatory and myofibroblastic CAF states. That matters in PDAC, where fibroblasts are not a single functional population and where stromal composition can influence disease behavior in different ways.
These findings suggest that when more of the native context is built into the model, the biological and pharmacological responses observed may differ.
Drug-response observations
The webinar also described drug efficacy studies.
They tested a combination of paclitaxel and a BMS compound. Solecka framed the combination study this way: “Our hypothesis was that these two compounds may act synergistically and this was based on the clinical data.” The team then tested the combination across Matrigel monoculture, Matrigel coculture, and RASTRUM coculture.
The results differed across model systems. As Solecka reported, “We found that the synergy between these two compounds appears only in the RASTRUM co-culture condition in all three lines.” In the Matrigel monoculture and Matrigel coculture conditions, that signal did not emerge.
That finding illustrates how model context can influence whether a drug-response effect is observed. A simpler system does not always fail dramatically. Sometimes it just stays quiet where a more representative model produces a signal worth following.
For discovery teams, these findings highlight the importance of considering model context when interpreting study results If a system is too reductionist, it can narrow the biology enough to make screening easier while also narrowing the set of vulnerabilities the screen can reveal. In a disease like PDAC, where stromal influence is part of the disease rather than a secondary feature, context can shape whether a response looks weak, absent, or actionable.
Model performance at discovery scale
A stronger model becomes useful only if it can fit into real workflows.
That practical side came through clearly in the webinar as well. The BMS team showed that the RASTRUM structures remained stable over extended culture periods, unlike softer Matrigel domes that degraded more quickly at high cell density. They also demonstrated that the system could separate drug effects on PDAC organoids and CAFs through imaging-based readouts, opening the door to more detailed analysis of tumor-intrinsic and tumor-extrinsic responses.
With the higher-throughput RASTRUM Allegro platform, the workflow could also move into 384-well screening formats with acceptable intra- and inter-plate variability and reproducible responses across plates. These findings suggest that the approach can be adapted to larger-scale screening workflows.
Biologically relevant models need more than complexity. They need consistency, repeatability, and enough throughput to support real discovery programs. In that setting, biological relevance and operational feasibility are both important considerations.
What this means for drug discovery teams
The findings presented in the webinar may have implications beyond pancreatic cancer.
Drug discovery teams do not make decisions from biology in the abstract. They make decisions from the responses their models produce. Those responses become datasets, those datasets shape confidence, and that confidence influences what moves forward.
That is why model context matters. A system carrying more of the biology that shapes response can change what becomes visible in the first place. In this case, it changed morphology, signaling, fibroblast behavior, and therapeutic response. It also identified a combination-response signal that was not observed in the comparator model systems. Solecka’s conclusion during the webinar was direct: “These advanced models help us better predict how the drug will perform in the clinic.”
The webinar highlighted how a context-rich model can reveal biology and drug-response behavior that may influence downstream decisions.
Model context can influence what becomes visible in the first place.
Learn more
Explore how RASTRUM helps discovery and translational teams build 3D cell models with more of the biological context that shapes response: