Infusing a living cell therapy into patients to fight disease may have sounded far-fetched decades ago. Today, the results of this potentially paradigm-shifting approach to “cure” cancer have shown that it works, at least in some treatment recipients.
But everyone who receives the treatment does not have a positive long-lasting response. Even though some responses indicate that chimeric antigen receptor (CAR) T-cell therapies definitely work, scientists and oncologists still struggle to understand the exact underlying mechanisms, optimize the therapies for all recipients, and reduce side effects such as cytokine release syndrome and neurotoxicity.
What influences the therapy’s results? Possibilities include the distinctive characteristics of the individual patient’s disease, the quality of the T cells taken from the patient and used to manufacture the product, the quality of the assays used to predict individual efficacy and toxicity responses, and the manufacturing process.
GEN talked to five leading researchers to get their input on the status of the field, the outstanding challenges, and the tools and techniques needed to help complete the circle of understanding to and advance the field.
GEN: Preclinically, how does the development of cell therapies differ from traditional antibody therapeutics in terms of characterization of the therapy type, potency, and overall performance? What advanced analytics specific to cell therapy development are needed?
Dr. Adusumilli: Antibody dosage and kinetics can be correlated with response and toxicity; thereby an optimal regimen can be derived. Cell therapies are a living drug. You can give a certain dose of cells, and depending on activation, the cells increase in number. This expansion cannot be predicted uniformly across a group of patients as it depends on multiple factors, including the tumor microenvironment, target antigen expression intensity, distribution, and organ site. Characterization of dose kinetics, response, and toxicity is challenging, and assessments are of different magnitude between cell therapies and antibodies.
As we progress with immunotherapy, we are using combinations, for example, cell therapy combined with antibodies, or cell therapy combined with chemotherapy. To do this in an optimal fashion, we need advanced models, and the analysis has to be much more sophisticated than the analysis for antibodies.
Dr. Davila: Preclinical development is much more complicated. Cell therapy is a living therapeutic versus a cellular protein product. In animal models, you look for tumor reduction and enhanced survival. Safety cannot be addressed because the target is a human target.
The goal of cell therapy is not only to get tumor reduction, but also to activate an immune response against the tumor, to get the T cells or other cells to release cytokines that activate other cells. Efficacy is more complicated than just enhanced survival. You want to know if there is evidence of epitope spreading and if targets other than the CAR are being eradicated because of activation of other T cells.
There are traditional analytics, immunophenotyping, in vitro cytotoxicity, and in vitro cytokine production. Advanced analytics have been developed only in the last few years and will continue to evolve. The two biggest new analytics are the Polyfunctional Strength Index (PSI), which has been correlated with outcomes, and metabolomics.
Dr. DiPersio: Antibody production is a very complicated process due to its large scale, and it is typically limited to biopharma companies, whereas cell therapies can be developed by institutions on a smaller scale.
Unlike an antibody, or a drug, large animal toxicology studies are not usually needed for gene therapy because the vectors have been characterized and the toxicities are known. The process of retroviral or lentiviral transduction is relatively standardized and can be performed to obtain 1–3 integrations per cell with a multiplicity of infection (MOI) of 5–10. In clinical trials, transduction efficiencies and replication-competent virus titers are measured. The potency of the product is determined in preclinical in vitro and in vivo studies. In patients, the performance is measured by clinical benefits and expansion of the genetically modified T cells. Off-target and on-target toxicities are also measured.
Cell therapy works, but we have very little knowledge on how to optimize it. The quality of the T-cell product is being described only in the most general terms. We know how many cells are present, but we may not know what percentage of actual CAR T cells there are in any specific product. There is no way to know how good the CAR T cells are and whether they will expand normally. Sometimes laboratories measure CD4 and CD8, but the association between efficacy and toxicity is unknown.
Dr. Fan: Cell therapy uses living cells that may be persistent and potent to the extent that they expand post delivery in vivo, amplifying the therapeutic efficacy and preventing the disease from recurring. It is a new type of therapy and could be paradigm shifting.
With these new constructs, which are living and changing, we need better characterization methods and techniques. Unlike a small molecule or antibody that always has the same structure, every time you develop an autologous cell therapy, the starting material comes from a different patient. There is no way to get an identical “drug” every time. Even if you were to start with similar cells, you are most likely going to end up with different potencies and activities.
Rather than try to standardize, tools are needed to perform quality control to characterize the cells and define the range of potency and toxicity. Currently, there is not a standard methodology defining how to characterize these cells in terms of live-cell potency, live-cell activity, and live-cell function. Better characterization and quality control can help ensure delivery of a quality cell therapy that maximizes efficacy and minimizes adverse effects.
Dr. Fraietta: A living, replicating cell therapy must persist to detect and respond to the tumor cells of a recurrent cancer. The cells can go anywhere and do anything, and they can make decisions. That is what differentiates a cellular drug from a traditional antibody-based therapy.
There are defined release criteria in terms of the transduction efficiency, basic compositional qualities, and confirmation that the cell products are safe for administration and free from contamination. The jury is still out on what additional metrics should be. Currently lacking is how to predict potency. Not all cell therapies are the same because autologous cells from patients are engineered and infused, and unlike a traditional antibody therapy, there can be differences in the composition and consistency that could influence clinical responses. Our group has done a lot of work on using some of the qualities that you see when T cells are collected, or after they are manufactured, to determine if they are predictive of CAR T-cell potency and toxicity.
The traditional potency assay is based on the production of interferon gamma after stimulation. But how predictive that is of clinical efficacy of the cell product is unknown. We need better potency assays for T cells. Advanced analytics using biomarkers to predict potency would make a huge difference.
GEN: What is lacking in the current standard methods of evaluating the quality, potency, and durability of cell therapy CAR T-cell products to better define in vivo clinical indications, such as neurotoxicity, cytokine release syndrome, and other toxicity implications?
- Good mouse models are lacking. For cell therapies, we need to measure long-term outcomes and toxicity.
- Even when we have good mouse models, with translational cell therapies, we want to test human cells, such as T cells. When you test human cells in a mouse model, you do not get the full information. In vivo systems that have human components are needed.
- In cancer treatment, we look at the initial cytotoxicity, the number of tumor cells that are getting killed. In cell therapy, the cells keep dividing in the right environment, so we need long-term assays. Available assays are short-term potency assays. Lacking are long-term assays to look at outcomes and toxicities in a human culture system.
Dr. Davila: The current standard methods to evaluate quality and potency are inadequate to select products that provide better patient outcomes. Quality assays are mainly descriptive with some objective parameters like cytokine production, cytotoxicity, and proliferation. Only recently, with PSI and metabolomics, can we start to differentiate products.
Immunophenotyping is used to look at durability in animal models. But we cannot say which subsets are critical for certain outcomes. Although still in its infancy, metabolomics is coming into play. It has been reported that cells that use a lot of mitochondria for their metabolism or oxidative phosphorylation seem to have more memory-like qualities.
In patients, durability is mainly evaluated using blood cells as a biomarker to assay for expansion and persistence because of ease of access. However, this is not where the tumor targets reside. There may be no detectable CAR T cells in the blood but plenty in the lymph nodes or bone marrow.
More real-time toxicity analyses for cytokine-related abnormalities are also needed. We have good biomarkers, such as lactate dehydrogenase, ferritin, and C-reactive protein, to monitor toxicity. Point-of-care instruments are being developed to monitor patient cytokine levels in patients with a turnaround for results of 1–2 hours. Neurotoxicity point-of-care testing lags because cerebrospinal fluid samples are needed but rarely obtained.
Dr. DiPersio: The quality of the T-cell product right now is being described only in the most general terms. Generally speaking, except for standard release criteria, we are somewhat in the dark about what qualities of a CAR T-cell product make the therapy optimal. The production of a CAR T-cell product using one approach versus another has never been compared in a group of patients. Although expensive and time consuming, these experiments need to be done.
There is no consensus about how to determine what the in vivo effects of cell therapy products will be in a patient. For example, with non-Hodgkin lymphoma and acute lymphoblastic leukemia, most of the patients that fail therapy do so within the first six months. The assumption is that there is something about their tumors that result in these early failures, but it could be the patient’s immune response to the CAR T-cell product, the quality and potency of the CAR T-cell product, or the manufacturing process.
Dr. Fan: Ensuring that CAR T cells are standardized in terms of level of CAR expression does not tell you the whole story. Product expression levels can range, and patients can respond even with a lower CAR expression level in the infusion mixture.
CAR T cells need functions other than cytotoxicity, for example, to migrate or to respond to a chemotaxis signal, because eventually the cells need to get into the bone marrow or lymph organs and kill the tumor progenitor cells. Treatments are expensive. Being able to predict response prior to treatment is important.
There are no currently well-defined methods to predict or characterize adverse effects, such as severe cytokine release syndrome (CRS) or neurotoxicity. Although circulating cytokine levels in blood are monitored, often that is too late, and drugs, such as IL-6, cannot prevent the patient’s death. In addition, the CAR T-cell infusion product may contain some cells with bizarre functions. Characterizing potential immunotoxicity earlier could help guide treatment and patient monitoring.
IsoPlexis’ technology can comprehensively evaluate the cells with almost no bias, so you get a full portrait of the whole infusion product, even small subsets of the cells with bizarre functions. Cytotoxicity or cytokine screening does not give the full picture. And since the patient’s tumor microenvironment is equally important, an understanding of it, along with a comprehensive characterization of the CAR T-cell product, could be used to better predict outcomes.
Dr. Fraietta: Durability of responses is very hard to predict. Some work we did in chronic lymphocytic leukemia found that long-term durable remissions were associated with a certain frequency of a particular memory T-cell subset. These early memory cells constitute the ideal seed population for CAR T-cell manufacturing. They are potent, can self-renew, and can differentiate at the tumor site. In nonresponding patients, we found low numbers of these cells and also that there might be something intrinsically deficient in terms of their quality.
Retrospective RNA-sequencing data on some of these cell products were concordant with the aforementioned memory phenotype. We saw early memory gene signatures enriched from completely responding patients. In contrast, CAR T-cell infusion products from nonresponding patients were enriched in pathways of terminal differentiation and T-cell exhaustion. This demonstrates that response is very much driven by the up-front quality of the T cells.
Managing toxicity along with the potency of the CAR T cells in vivo has big challenges. Most of the cytokines produced that mediate CRS are produced by macrophages. It is very difficult to recapitulate CRS immunotoxicity in vitro or in current animal model systems. The ability to manage CRS effectively without hampering the therapeutic efficacy of the CAR T cells is also lacking. Blocking the IL-6 receptor reverses CRS to some extent, but it could affect CAR T-cell function. If steroids are given when blockade of IL-6 signaling fails in patients, the transferred T cells are wiped out.
GEN: How does the increasing evidence that CAR T cells are heterogeneous in performance, despite being relatively homogeneous phenotypically, impact evaluation of CAR T-cell therapies preclinically and during bioprocessing?
Dr. Adusumilli: With CAR T cells, the percentage of transduction matters. Even when we get high transduction, the number of vector copies in a particular T cell can influence both antitumor efficacy and toxicity. The phenotype of the T cell that we can subject to CAR transduction can affect the outcome. More important, these are not independent factors. Multiple permutations of these factors can happen in CAR T cells that ultimately influences the outcome.
Laboratory investigations on CAR T cells use healthy donors to develop data. Patients who will be undergoing CAR T-cell therapy are not healthy. They may have gone through rounds of chemotherapy or radiation therapy. Even if they have not gone through treatment, their T cells are influenced by their cancer and body immunity. This can influence the outcome as well. In a broad perspective, all these factors matter.
Dr. Davila: The infusion mixture is a heterogeneous population of cells, immune subsets, CAR-positive and CAR-negative cells, and non-T cells. For example, in a clinical trial that enrolls 100 patients, essentially 100 different products are made. Understanding the quality of the products and how they correlate with clinical outcomes is challenging.
Looking at these products on a single-cell level could drill down to see if we can phenotypically identify the optimal cell types that correlate to better outcomes as it is unlikely that the vast majority of cells provide a benefit. This is what PSI gets to. Many people think a smaller subset of cells is the critical subset for activity. The sooner we can identify that subset, the sooner we can tailor production schemes to specifically make those cells. This would allow us to infuse fewer cells and reduce the chance for toxicity.
Dr. DiPersio: The assumption is that these products are phenomenally heterogeneous even though they share the ability to kill appropriate target-positive tumor cells. We do not know how this impacts clinical outcomes. Trying to understand some of the heterogeneities and controlling for them may mean we would make a significantly better product. We need to spend time with preclinical models to identify what we might be able to modify to make the CAR T cells more optimal and homogenous. Once these concepts can be explored in preclinical models, one can move on to test these in well-defined clinical trials in humans. In theory, each product will be different since every patient is different; their T cells are different; and the methods for T-cell transduction, expansion, and processing may be different as well.
There are different ways of generating CAR constructs using various “endodomains.” All commercial products currently use either CD28 or 4-1BB. In animal models, there are some nuanced studies that have used NSG mice and looked the relative potency, expansion, and persistence in vivo of CAR T cells possessing either of these two endodomains. We need to do a lot more to understand the differences between the endodomains, the impact of varying methods of CAR T-cell transduction and expansion, the T-cell subsets chosen for transduction, and the impact of CD4/CD8 ratios prior to infusion on clinical efficacy and toxicity in humans.
Dr. Fan: Every cell is different in an infusion product. Cells may have different CAR expression levels and different differentiating stages, such as T-cell polarization stage, type 1 versus type 2, etc. My laboratory has found that sometimes a heterogeneous population is needed to ensure the cell therapy product is going to work, both immediately and in the long term. For example, the effector cells can do their job right away, but after they become exhausted, you need other cells, memory or stem-like T cells, to regenerate effectors to become active functioning cells. The science is not yet fully understood, but it seems to indicate that heterogeneity is necessary. Cells occupy different states. Whole populations of cells may collectively work better.
Helper cell composition is also important (for example, type 1 versus type 2, in case CAR T cells are out of control) so there is a response mechanism. In my opinion, the mixture has to be a population of different cells that work collectively to produce the best efficacy and to minimize toxicity.
Characterizing at the population level cannot define functions of all the individual cells. You have to use single-cell analysis tools, which are very expensive and time consuming. Although these tools are useful in mechanistic studies, they are not readily integrated with quality control. Better tools that could serve both purposes include the IsoPlexis platform, which can directly measure single cells for all possible functions and also operate as quality control. Ideally, you would want to measure genome wide, but now it is better to concentrate on the important CAR T-cell functions.
Dr. Fraietta: When we do basic assessments of T cells at the time of collection, or of CAR T cells after manufacture, we are often biased by what markers we think are important. You could use markers to indicate that CAR T cells look homogeneous in composition; then, when you infuse them, the performance could be different. The difference could be disease specific, or there could be other causes for treatment failures. Disease biology can be very different, complicating prediction.
Some of our biomarker work has shown that this issue can be circumvented by trying to make more consistently defined cell products. Some of the biomarkers could be used to stratify poor- vs. high-quality CAR T-cell products as a patient selection strategy to help predict a complete and durable remission with a certain degree of accuracy.
Based on infusing large numbers of effector cells in early CAR T-cell trials, we thought they would be very potent mediators of antitumor activity. But if you start with effector cells, put them through a long manufacturing process, bank them, then put them into patients, they will have very short-lived antitumor activity, if any, and certainly little proliferative capacity. If you start with the right seed population for CAR T-cell manufacturing, like early memory cells that have the capacity to proliferate robustly, that may improve the depth and durability of function as well as persistence. In some of our patients treated over 8 years ago, we still see CAR T cells persisting, and some of these individuals remain cancer free.
GEN: Where do you see the largest opportunity for future cell therapy advancements, and what cellular characterization is needed to advance these therapeutic opportunities?
Dr. Adusumilli: Going forward, we need to understand how these cells work and how they influence the tumor microenvironment. When we monitor serial biopsies of the tumor or peripheral blood, we need to be able to quantify each cell type, and within that, we need to be able to subquantify each phenotype, and more important, we need to have some idea about their function. These cells will have multiple functions, so we need to know the functional ability and persistence of each cell.
The combination of this data; the quantity, phenotype, and functionality of each cell; and the overall composition of multiple different types of cells all point out that we need better methods to analyze these over time to be able to understand how these cell therapies work and what improvements we should look at. Any assay systems that can help us to get this data from limited samples in a cost-efficient manner is what will advance cell therapy, in addition to basic science research.
Dr. Davila: We need fast production to manufacture high-quality product with a failure rate as close to zero as possible. This is where the field is struggling. Once you have an FDA-approved process, it becomes harder to modify that process. Hopefully, our new insights will help inform the next generation of products.
There is also a huge potential for defining what quality is and how it impacts clinical outcomes. We have started using metabolomics and PSI, but there are other ways of looking at these products, such as genomic accessibility, the nature of the CAR integration sites, or the ability to secrete cytokine, in addition to the standard gene transfer rate, cytotoxicity, and sterility. Although important, the standard characterization measurements are not correlated with positive clinical outcomes.
We also need to improve in vivo monitoring. Tools such as flow cytometry or qPCR give some information, like activation, but are insufficient. To understand a living therapy, we need tools to confirm that there are copies of genes in the cell and that the cells are working.
Dr. DiPersio: When things get commercialized, the incentive to dig in disappears. Several areas need to be expanded upon:
- Investigating heterogeneity and functionality and optimizing the role of the CAR T cell using immunocompetent mouse models, mouse T cells, and mouse CARs.
- Adopting a more scientific approach to assessing the quality of the infused product. If we could determine a way to save a portion of all CAR T-cell products infused into patients so that specific analyses (such as single-cell RNA-sequencing, immunophenotyping, and CyTOF) could be performed, then we might be able to make some important insights in the future.
- Using off-the-shelf or allogeneic CAR T cells. The relative efficacy and toxicities of autologous vs. allogeneic CAR T cells remain unknown but could be investigated using immunocompetent mouse models in which CAR T-cell trafficking, efficacy, expansion, and persistence can be studied and compared.
- Looking at the roles of alternative effector cells such as macrophages, natural killer cells, and iNKT (invariant NKT) cells. These cells have advantages and disadvantages. They can be genetically manipulated and useful in killing tumors.
Dr. Fan: We should be more open to implementing new characterization tools. Assays in cell therapy characterization for preclinical or clinical monitoring are needed to improve our understanding of the activation states. The T-cell activation assay has been around for decades and dominated by one technology where you flow antigen and evaluate how many T cells become active and produce interferon gamma. For cell-based therapies, you need a more comprehensive characterization of T-cell activation, not just a single parameter. The gold standard eventually has to be direct evidence of what engineered T cells can do—not indirect correlations.
We need all kinds of tools to characterize the cytotoxicity of cell therapy products. One day, single-cell tools, including transcriptomic and genomic sequencing, will become more accessible. I am optimistic that these tools can advance from research to clinical tools, and we should seriously consider how we can move forward with those technologies. One option is to reduce the panel since we know which genes are most important.
Some of the implemented methods are no longer the best methods. I hope that this will change and that new technologies and approaches will be standardized at all levels.
Dr. Fraietta: Universal CAR T cells with consistent quality could resolve intrinsic T-cell differences or variability in patient-to-patient T-cell quality. But we do not know what type of T-cell donor would be ideal. That is where advanced analytical tools that serve to elucidate determinants of CAR T-cell potency or other biomarkers would make a big difference.
The field is also awaiting a clear demonstration that CAR T cells could be just as effective in solid tumors as they are in B-cell malignancies. Solid tumors present different issues; the CAR T cells have to traffic to the tumor site in the face, sometimes, of antigen heterogeneity, and there can be antigen escape. Once the CAR T cells get to the tumor bed, there is a profoundly immunosuppressive microenvironment composed of inhibitory immune cells and soluble immunosuppressive factors, and some cytokines and other soluble factors are present that can hamper CAR T-cell function. T cells can upregulate inhibitory receptors in this environment, and tumor or immunosuppressive cells can upregulate cognate ligands, adding many additional barriers that solid tumor CAR T-cell therapy needs to address. We could make a huge difference in this setting, and the question is whether the CAR T-cell strategy is going to be broadly disruptive. I think the latter is true, but there is so much we still do not know.