If it’s one thing I’ve learned during my career it’s that success in biological research is often the product of interdisciplinary collaboration. So this is where our story begins, several years ago over lunch in the Novartis bistro between myself and a very talented immunologist, Emily Rowell. Around that time the focus of Novartis Oncology abruptly shifted from targeted chemotherapeutics to immune-modulating drugs, and I decided that I wanted to be part of this new initiative. It was completely intuitive to me (and most other cancer biologists) that manipulating the immune system to kill tumors was a highly pragmatic approach in the fight against cancer. Given the huge clinical successes of Pembrolizumab and Ipilimumab, I decided that the most straightforward approach for me to enter this new world of “IO” was via the T-cell. Enter Emily.
Jeremy To and I are nuts and bolts biologists, always having our eyes on the most practical approach to tumor modeling. I needed an expert in IO who could help me flush out the big picture concepts and act as a prism for the granularity I would need to model this very complex biology. Over the aforementioned lunch I asked Emily how could we build an ex vivo phenotypic screening model that would mimic clinical tumor immunity. If we could faithfully capture the T-cell-tumor cell interaction, incorporate clinically-relevant selection pressures and miniaturize the whole thing to fit on the head of a pin, then we could screen the model against the vast and exciting Novartis small molecule libraries.
Human T-cells (red) penetrating a 3D colorectal cancer tumor spheroid expressing EGFP (green) in a well of a 1536-well plate. The whole model fits on the head of a pin and enables large compound screening campaigns to find new drugs that augment T-cell-mediated tumor killing.
Emily’s incredible insight into this question addressed two important concepts that had been largely overlooked in conventional IO studies. First, that it is imperative to HLA-match the various cell types we would be combining in the model. Using mismatched HLA cell types invariably leads to a sort-of host versus graft effect that is not relevant for studying T-cell-tumor cell interactions. Emily’s second point was that we needed to incorporate a T-cell-mediated antigen recognition event that would mimic how a T-cell identifies a tumor cell presenting a tumor neoantigen and can distinguish self versus non-self. These two aspects of how T-cells identify tumor cells were absolutely crucial for the development of our screenable IO model.
Design and construction of a high-throughput 3D tumor spheroid screen with primary human T-cells and employing a 3-dimensional imaging-based readout required crowdsourcing the necessary expertise in high-content imaging and image analysis. Enter Doug Quackenbush and Frederick Lo who were instrumental in developing the machine learning methods we employed to analyze the highly-complex 3D tumor spheroid imaging data. Finally, we inspired Lilly Li and Connor Reed, two experts at FACS-based immune profiling, to assist in the extremely convoluted world of T-cell phenotyping and identifying the mechanisms of action of our lead hit compounds.
We ran the screen and reconfirmed the hits using time lapse high-content 3D confocal microscopy. When we stitched the images together into a movie and could actually see the T-cells penetrate and destroy the tumor spheroid we were blown away. The compounds we identified may one day be in patients augmenting their T-cells to kill tumors, but what is certain is that there is a world of immunomodulatory drug targets beyond checkpoint receptors that remains unknown or underutilized.