Whether facing down a virus-infected cell or a cancerous one, T cells fight off threats to keep the body healthy. To make these T cells even more effective, particularly in treating blood cancers, scientists created chimeric antigen receptor (CAR) T cells that recognize these cancerous cells by the CD19 marker on their cell surface. So far, the FDA has approved seven different CAR T cell therapies for hematological cancers.

However, when scientists tried to use CAR T cell therapy to treat solid tumors, they ran into multiple hurdles — most critically, unlike CD19 for liquid tumors, there is no single marker that T cells can use to recognize a solid tumor.

“Solid tumors — breast cancer, renal cancer, prostate cancer — are fundamentally different beasts to leukemias and lymphomas for reasons that we partially understand and [that] we don't fully understand,” said Nick Haining, the Chief Scientific Officer of the T cell therapy company ArsenalBio. “Experience has taught us to date that simply taking the playbook from CAR T cell therapy for hematologic malignancies and applying it to solid tumors just doesn't work.”

Instead, the ArsenalBio team has developed a multi-pronged approach to reprogram cancer patients’ T cells to recognize and attack solid tumors.

“We're using genome editing and synthetic biology to really fundamentally change how T cells behave,” said Haining. He added that their engineered T cells perform well in the tumor microenvironment, allowing them to kill “enough tumor cells that patients are going to feel the benefit.”

How is ArsenalBio addressing the challenges facing T cell therapies for solid tumors?

About six or seven years ago, our founders got together around the idea that we knew three things that individually weren't enough but together could form an arsenal of tools that we could bring to bear on solid tumors. The first was understanding why T cells don't work when they should. For example, unlike in response to an acute infection like influenza, when T cells respond to cancer or a chronic viral infection, they become exhausted and eventually drop out and disappear from circulation. Work in my lab, John Wherry's lab, and many others had started to identify what it was about those exhausted cells that wasn't right.

The second component was that we could use CRISPR to change how a T cell behaved. Obviously, many people were using CRISPR to turn off a gene, but we became increasingly interested in work from one of our collaborators, Alex Marson, who had pioneered the use of knock-in technology in primary human T cells.

The third realization was that we needed a way to change how T cells function, so we turned to another one of our collaborators, Kole Roybal. He had developed a remarkable set of tools that allow a T cell to recognize not just single antigens, but pairs of antigens. That's really important because it’s been super difficult to find the solid tumor equivalent of CD19. By requiring that T cells recognize two antigens in the tumor microenvironment, we could target antigens that might be expressed in a normal tissue individually, but both would only be present on the tumor.

What barriers did you overcome in developing this T cell engineering technology?

We had to invent a lot of technical solutions to problems that were really difficult. To begin with, we use a non-viral approach to deliver our gene editing machinery to T cells: all of the CRISPR components and the repair template. Our original system worked for a couple hundred base pairs, but that wasn't big enough to encode all of the information that we needed. So, we had to invent new ways to increase the efficiency of that process. Now, we can basically encode anything we want into the cell. That took a huge amount of deep genomic understanding and technical advances to get there.

Then, we realized that if we were engineering T cells with multiple synthetic features, such as short hairpin RNAs (shRNAs) that turn off genes and coding sequences that turn on genes, into an array on this repair template, what's the right combination? To canvas that enormous search space, we had to build a very large high-throughput screening platform. And similarly, we had to develop new manufacturing processes based on our non-viral CRISPR knock-in capabilities, and we do that with astonishingly high efficiency. Greater than 95 percent of the products that we make are ready to give to patients. What's been particularly impressive is the speed with which the team has been able to do that. We went from basically opening up our doors in the equivalent of a garage in south San Francisco to dosing our first patient in four years — an incredibly fast turnaround time — and we have continued at what we like to call “Arsenal speed” ever since.

Your drug AB-2100 is currently in a Phase 1/2 trial. How does that cell therapy work?

We're very excited about progress on that trial. AB-2100 targets two antigens that are expressed in the tumor microenvironment: carbonic anhydrase 9 and — in an innovative twist, we're using a tumor endothelial marker as the second of the two antigens — prostate-specific membrane antigen (PSMA) (1). We usually think of PSMA as a prostate-directed antigen, but for reasons nobody fully understands, PSMA is also selectively expressed on the tumor neovasculature. So, we've taken advantage of that to focus the activity of AB-2100 in the tumor microenvironment.

The second feature that we’ve encoded is a gain-of-function capability, which we call a signal pathway activator (SPA). It's essentially a synthetic, ligand-independent cytokine receptor. That gives the T cells a tremendous boost in potency and persistence, and it also blocks that process of T cell exhaustion that I was talking about.

The third component is a pair of shRNAs that knock down inhibitory T cell receptors: the Fas ligand and TGF beta receptor two (TGFβR2). Fas is a death ligand that is a surefire off switch that T cells express to prevent excessive activation under normal circumstances. Tumors co-opt that by expressing the Fas ligand on the endothelium of the blood vessels supplying the tumors, which can serve as a strategy to prevent incoming T cells from arriving. TGFβR2 is a receptor that sees signaling from TGFβ, an inhibitory cytokine that's present at high levels in the milieu of many tumors, including renal cell carcinoma for which AB-2100 is being developed. Knocking both of those down provides a robust armor against the poisonous environment that exists around the tumor.

We are in the early stages of this trial, so we're investigating what the right dose is and what the right supportive care strategies are to manage patients. So far, we've been incredibly encouraged by the early signs of efficacy that we're seeing, which suggest that — whilst we have a ways to go — we are definitely heading in the right direction, in a way that makes me really excited to continue to develop this drug.

What’s next for ArsenalBio?

What makes me so optimistic about what we're doing is the fact that we have such exquisite control over cell programming. The limitation on our ability to design increasingly more effective drugs is our ability to understand T cell biology and how to combinatorially introduce new changes into the cell. One of the exciting developments that we've started telling people about is our application of artificial intelligence (AI) to really understand how T cells work and how to program them better. The investment that we've made in generating massive amounts of single cell RNA sequencing data as a part of our drug development pipeline has allowed us to create a second kind of product that's essentially a large language model of the cell that we've termed GX1.

We're using it for our own internal discovery, but I think the applications for GX1 go far beyond our own internal purposes. So, we are developing GX1 as a service, a partnering modality. We think it can help accelerate drug discovery in immunology and inflammation, autoimmunity, transplant, and in inflammatory diseases like fibrosis as well as immuno-oncology. We think there are also opportunities to use a large language model for patient segmentation, for instance, to predict who will respond to a cell therapy or not. A large language model of the cell could look at biopsies or blood samples from patients and learn the meaning of what a drug response looks like in a way that would be invisible to simpler biomarker discovery efforts. Similarly, we've used GX1 to simulate phenotypic target discovery — to predict what would happen if we turned a target gene on or off in a particular cell — far faster and with much more scope than could ever be done in a physical experiment. We're very excited to be closing in on the release of our first version of GX1, which we think is the new frontier of AI models for drug discovery.

This interview has been condensed and edited for clarity.

Reference

1. Mohanty, S. et al. Abstract 38: AB-2100, a PSMA-inducible CA9-specific CAR T cell product for the treatment of ccRCC provides long-term tumor responses in preclinical mouse model. Cancer Res 84, 38 (2024).