NEW YORK – The number of chimeric antigen receptor (CAR)-based clinical trials have quadrupled from 2014 to 2018. While these types of therapies have the potential to help cancer patients achieve remission, they can also produce undesired side effects like cytokine release syndrome and cardiac arrest.
These side effects occur from 'on target, off-tumor toxicity,' when CAR targets appear on normal tissues. Aiming to avert such off-tumor effects, a team of researchers from Weill Cornell Medicine and elsewhere sought to identify more specific targets in acute myeloid leukemia. The team cross-analyzed proteins, RNAs, cells, tissues, and tumor types from more than 20,000 patients from over 500 clinical trials. Their findings, published this week in Nature Biotechnology, could be useful for researchers and drugmakers as they try to home in on the population of patients most likely to benefit from CAR therapies without serious side effects.
"We've entered this new era where we're engineering T cells and [natural killer] cells to go out into the body with very specific targets that they're hunting for without as clear of a delineation as to what else might be collateral damage in the immunotherapy," said Christopher Mason, associate professor at Weill Cornell Medicine and senior author on the study. "The two therapies that are FDA approved and extraordinarily successful, in lymphoma in particular, are the ones where there is exquisite specificity at those targets only in the immune system."
According to Mason, his team conducted the study to see what else is being targeted by therapies currently in development, could they be directed at better targets, and could new cells be built. Ultimately, the researchers wanted to present the data in a way that is useful in the development of new CAR therapies.
What evolved out of the data gathered in the study is a clinically applicable interactive database, which can potentially serve as a resource for physicians, researchers and patients to filter through all CAR trials by location, indication, and disease phenotypes.
CAR cell therapy is often described as a 'living' immunotherapy that involves taking a patient's immune cells and engineering them to include chimeric antigen receptors that, when reinfused into a patient, can recognize specific proteins on cancer cells and attack them. CAR cell therapy has evolved to employ a variety of immune cells including T cells, natural killer cells, or a mix of the two.
The US Food and Drug Administration approved its first CAR T cell therapy in 2017. Axocabtagene ciloleucel (Gilead Sciences' Yescarta) promised a one-time treatment for patients with non-Hodgkin's lymphoma and large B cell lymphoma. In early 2019, the FDA approved tisagenlecleucel (Novartis' Kymriah) for patients with B cell precursor acute lymphoblastic leukemia (ALL) and B cell lymphoma. Tisagenlecleucel has a reported response rate of around 52 percent.
Mason has seen patient responses to CAR therapy that are as high as 70 to 90 percent. This is based on responses reported for axocabtagene ciloleucel and data presented at the American Society of Hematology meeting in 2019. "But in other cases, there have been reports of toxicity and even in rare cases, death. So, that was the other challenge we've been looking at," he said. "The vast majority of [clinical trials] have not yet reported their results."
Mason and his colleagues predict in their paper that some of these ongoing CAR trials may not show a good outcome due to the lack of target specificity. "You're taking a T cell and amping it up and then sending them into the body when it likely looks like you're gonna hit five, six, or sometimes [up to] 15 cell types."
To determine which antigens are most at risk for an off-tumor effect, the team referenced CAR targets against normal-tissue protein expression in the Human Protein Atlas.
They found that most CAR targets are expressed at medium-to-high protein levels in less than 25 percent of tissues. The researchers identified the best targets for therapy are CD38, CD20, mesothelin, CD22 and CD19, since they are expressed in a low percentage of non-immune tissues. The two FDA-approved CAR T therapies are both directed at CD19.
However, this study also showed that the riskiest targets are ROR2, ERBB3, DR5, TAC1 and cMet, which are expressed across a high proportion of tissues. HER2 falls between the best and riskiest target classification in the analysis, since it is expressed at medium levels in many tissues but isn't highly expressed in any tissue type.
When the researchers analyzed CAR trials targeting HER2, they found a wide range of safety responses, which indicates that a number of factors beyond the target itself — including CAR design, therapy structure, number of infused cells per session and patient history — could influence the safety profile.
Three targets that Mason and colleagues flagged as potentially toxic — CD123, MUC1, and HER2 — have the most clinical trials. Since most of the US trials investigating these CAR targets are in early phases, this research could be useful to advance safer trials where CARs are designed to target proteins that show up only in the tumor.
General analysis has singled out candidates that only show up in specific tumors, which could in theory be safer targets. These include vascular endothelial growth factor receptor 2 (VEGFR2) for head and neck or cervical cancer, and EGFR for skin or bladder cancer. However, to confirm the clinical utility of the targets, Mason proposes that epitope and protein mapping for each patient should be combined with the CAR design to ensure the efficacy and safety, since Human Protein Atlas and Genotype-Tissue Expression data is not always representative of individual variation of RNA and protein levels across tissues.
Another interesting insight out of this research is that some CAR targets tend to show up differently in men and women. As such, the team suggested that it may be important for precision therapy to further stratify treatment groups based on sex.
"In some brain and lung cancer targets, we do see pretty big [sex-based] differences based on RNA and protein," said Mason. "This gives us evidence that certain targets will very likely work better for different sexes of patients."
For example, in the paper, the authors wrote that CARs directed at MAGE-A1 and MAGE-A4 would probably have the least on-target, off-tumor toxicity in a female trial participant, while CLD18-directed treatment targeting cells in the stomach or mesothelin may be more appropriate for males who have had their tonsils removed.
When researchers analyzed the profile for CD19, the current target of FDA-approved treatments, they detected protein levels in cells of the immune system but not in normal tissue systems. This distribution might explain the positive results seen with these agents in clinical trials.
Mason and his colleagues identified antigen CD22 as having a similar gene expression and protein profile as CD19, though they flagged lower levels of CD22 in the immune system. Among the approximately 500 trials the researchers analyzed, nearly half of studies are investigating CARs with these targets.
Conversely, the targets that show up in normal cells in various tissues and therefore carry a greater risk of off-tumor effects — CD123, FAP, CD133, MUC1, EPCAM, CD138 and HER2 — are being studied in 11.4 percent of trials.
In contrast to these targets, Mason estimated that there are around 65 targets that have a lower toxicity profile, but which are not in use. "I think it opens up the idea of switching [these targets] for other targets on the market," said Mason. "We can change either the single-target CAR cells, or have an inhibitory or co-stimulatory domains on the same CAR design. We can make them safer by using a predictive cellular logic for mapping what is on the surface of the cells."
The researchers suggested that certain combinations of different targets could optimize efficacy. To date, the field has largely designed CARs targeting one receptor, but Mason noted that by targeting two different markers on cells, it might be possible to increase the specificity of CARs towards specific tissue types, and for men and women.
"We put in here what would be the likely best candidates for combined new synthetic CARs that could be designed. This would include a chimeric antigen receptor that has logical switches," said Mason.
The most notable example they found is CD19, which is already used to target lymphoma. Yet it cannot currently distinguish between normal B cells or lymphoma cells, which can cause a side effect called B cell dysplasia. According to the researchers, ITGAM, CD86, or ITGAL could potentially serve as inhibitors that can help distinguish lymphoma from normal CD19-positive cells.
The analysis could also be used to identify new targets for CAR therapy, drawing from preclinical models. These include (TNFRSF13C)19 and CD37, as well as ADGRV1 for colon or prostate cancer.
However, this study does have its limits in the types of proteins it can predict the risk profile for. The target α-fetoprotein, for example, is a secreted protein that requires additional design factors. Because of that, researchers can't predict its toxicity based on looking at its tissue expression levels alone.
This is only the first step in Mason's work to better characterize CAR targets. "One thing that's really exciting is there are a lot more proteomics and transcriptome databases that are coming out. We're also in the process of engineering some of these cells in the lab," he said. "This is version 1.0."
He added that with more data, researchers will be able to better characterize the role of specific targets and use that to design more precise CAR therapies. "Once we start to actually create some of these hybrid cells, these logically gated cells, we will then be able to see how well they work," Mason said.