Bringing Pancreas Cancer into the Lab

Modern oncologic care for many major malignancies has been revolutionized by our ability to interrogate tumor genomes quickly, accurately, deeply, and relatively inexpensively. The interplay between diagnostics and therapeutics has formed a virtuous cycle of more testing being performed in more patients, enabling personalized therapies for ever smaller molecularly identified subsets. Patients do better for longer with less toxicity as the end result. Not unexpectedly, not all cancers have benefitted equally from this revolution. Pancreatic ductal adenocarcinoma (PDA) cancer is the most deadly major human cancer. Unlike the other major cancer killers (lung and colon cancers, for example) where some form of molecular analysis is first and foremost in establishing a treatment plan, molecular testing is rarely performed in pancreatic cancer, because the treatments we use today do not require any testing. As a result, a different sort of cycle has taken hold in pancreatic cancer, where few patients are tested at all, treatments do not evolve to serve targeted subsets of patients, and we are left with nontargeted combination chemotherapy regimens selected more by toxicity avoidance than clinical efficacy against any defined molecular subset of patients.

Two articles in this issue of Cancer Discovery start to make a compelling case for more testing more often in patients with advanced pancreatic cancer. These highly collaborative projects (many authors are on both manuscripts) adopt distinct but complementary approaches to pave the way for personalized therapies for patients with pancreatic cancer.

Aguirre and colleagues performed targeted sequencing of tumor and normal genomic DNA and tumor RNA in 79 patients with PDA (1). Although genomic profiling of PDA has been performed in the past (2, 3), it has virtually always been on resected, primary tumors. The technical reasons for this focus have been to ensure sample quality, a necessary first step for next-generation sequencing (NGS) technologies. The amount of DNA and RNA required for sequencing has dropped, and the authors here focused on what we see in clinic most often: metastatic disease. The team here took great care to methodically pilot their workflow on biopsy-amenable disease, and then rolled out a protocol optimized to obtain NGS data from metastatic PDA in the first line of treatment. The quality of the data, the safety of the approach, and the speed of production are encouraging and indicate feasibility of this biopsy to data paradigm in metastatic PDA under clinically relevant timelines.

The study provides an important proof of principle, that despite challenges of anatomy, disease symptomatology, low tumoral cellularity, and the frequently small amounts of tissue obtained at biopsy, genomic profiling can be performed in PDA. The authors then turn to whether it should be.

Although this technical advance is progress in and of itself, the authors also demonstrated a surprisingly high rate of potentially clinically actionable findings emerging from comprehensive genomic profiling in PDA, which has long been thought to be a monomorphic disease of KRAS, CDKN2A, and p53 mutations, and little else of genomic interest. The results from Aguirre and colleagues argue this is not the case. Actionable examples include the high rate of druggable oncogene lesions in cases harboring wild-type (WT) KRAS genes. One example reduced to practice is the first reported therapeutic response to MEK inhibition in a BRAF-mutated PDA as well as a diversity of other alterations in BRAF (point mutation, in-frame deletion), ROS1 (fusion), and FGFR1 (amplification) in others without KRAS mutation. An important precedent is demonstrated in the acquisition of known trametinib resistance–conferring mutations in MEK2 detected in the BRAF-mutated patient's cell-free DNA obtained at the time of progression on trametinib. The emergence of these mutations proves the tumor's dependence on constitutive MEK signaling, in this case virtually certainly driven by BRAF activation mediated by the N486-P490 deletion, whereas others have shown the importance of RAF signaling in PDA and the mouse (4) and the activating nature of the BRAF N486-P490 deletion (5).

The group also found a high rate of germline mutations in genes known to be involved in cancer. Chief among these were genes integral in the DNA-damage response, including BRCA1, BRCA2, and ATM. The matched tumor exome enabled concurrent examination of somatic loss of heterozygosity as well as genomic signatures reflective of underlying mutational processes (6).

Genomic characterization provides ever more insight into the molecular wiring of human pancreatic cancers. A limitation of this approach is that the process of genomic characterization is destructive in that the living tissue being studied is no longer available for functional testing after it is consumed in the testing process.

Tiriac and colleagues also present in this issue their deep functional and molecular characterization of 66 patient-derived organoid (PDO) models derived from primary resected tumors, needle biopsies of metastatic sites, and autopsy sampling (7). The majority of patient donors were treatment-naïve. Molecular characterization of the PDOs showed reassuringly high overlap with the tumor biopsy DNA sequencing, often with improved resolution of copy-number gains and losses, due to the absence of “contaminating” diploid genomes. DNA sequencing again called attention to certain potentially important, rare subsets of PDA that begin to follow molecularly identifiable patterns. As in Aguirre and colleagues' work, those PDOs with WT KRAS genes (only 4% of the total) seemed to harbor potentially druggable mutations in the RAS pathway (MAP2K1, PIK3CA, and ERBB2).

The clinical utility of any testing platform or assay is defined by changing what doctors do, hopefully for the better. Tiriac and colleagues move beyond characterization of the PDO library and bring the technology tantalizingly close to where the community can begin to think about clinical utility–demonstrating studies. Through an elegant series of PDO-based sensitivity testing in vitro, coupled with response data from the clinically well-annotated COMPASS trial (8), the investigators were able to start to approach transcriptomic predictors of gemcitabine, 5-FU, oxaliplatin, taxane, and CPT-11 therapies used in combination in the standard-of-care treatment of advanced PDA.

These two pieces of original research are important among the first fruits of platforms constructed to better understand pancreatic cancers we see in the clinic, directly from the patients we treat, rather than the more circuitous study of cell lines derived decades ago or genetically engineered mouse models driven by a single-nucleotide change in one gene. As such, they inform the field that rapid genomic characterization of patients' tumors is feasible in real time. Additionally and importantly, functional characterization can be achieved in the majority of cases; although the timelines required here may not yet be clinically useful, the results do appear to reflect patient outcomes to some degree. These achievements both raise the bar for both basic research into topics pursued in the lab today in pancreatic cancer as well as clinical trials under consideration in unselected PDA patient populations. The profiling of Aguirre and colleagues clearly demonstrates that identifiable, potently important subsets of patients come through our clinic doors every day. Tiriac and colleagues show us that functional characterization of most biopsy specimens can be performed on biopsy samples with a high rate of success and robustly characterized and tested over in vitro culture using controlled conditions for maximally reproducible results. These results may remove one of the last standing objections to personalized trials in PDA: the scourge of feasibility in biomarker acquisition.

Questions remain. Although each report had many collaborators, most of the lab work was performed at single centers of excellence in their field. How can the pancreatic cancer community efficiently obtain high-quality data (NGS, PDOs, or both) without “sending out” tissues to many sites? When is the right time to act on genomic or functional PDO-based findings with potential predictive value (drug responsiveness)? Patients with this disease deteriorate quickly after first-line therapy fails them, and all die, often in months. This clinical scenario is not ideal for testing concepts that might change the equation for small subsets of patients, and first-line “window of opportunity” trials might be considered if biomarker turnaround time could be further accelerated to meet the clinical need. Finally, although many patients had actionable findings, the vast majority did not. Combination chemotherapy, administered with maximal symptom control, nutritional support, and pain management were and remain these patients' best shot at a decent tomorrow.

E.A. Collisson reports receiving commercial research grants from AstraZeneca, Gilead, and Daiichi Sanyko and is a consultant/advisory board member for Takeda, Guardant Health, Celgene, and OncLive.