Dawn of Chimeric Antigen Receptor T Cell Therapy in Non-Hodgkin Lymphoma

Abstract

Two Chimeric Antigen Receptor (CAR) T cell therapies are now approved for the treatment of relapsed and refractory large cell lymphomas, with many others under development. The dawn of CAR T cell therapy in non-Hodgkin Lymphoma (NHL) has been characterized by rapid progress and high response rates, with a subset of patients experiencing durable benefit. In this review, we describe commercially available and investigational CAR T cell therapies, including product characteristics and clinical outcomes. We review patient selection, with an emphasis on sequencing cell therapy options in the refractory setting. Finally, we discuss durability of response, highlighting mechanisms of escape and investigational approaches to prevent and treat relapse after CAR T cell therapy.

Manuscript

Chimeric Antigen Receptor (CAR) T cell constructs are composed of an extracellular domain containing an antibody fragment that recognizes a tumor-associated antigen, fused to signaling domains that activate T cells upon binding^1,2. This permits T cells, which normally recognize unique antigens in the context of Major Histocompatibility Complex (MHC), to recognize tumor-associated extracellular antigens in an MHC-independent fashion.

Although CARs can theoretically be designed to target any extracellular antigen, the target antigen CD19, which is displayed on both normal B cells and in malignant B cell lymphomas and leukemia at high levels ³, has been the most clinically successful, demonstrating the potential of cellular therapy. CAR technology has been further enhanced by the development of second generation CARs^4,5 which contain both the CD3 signaling domain and a co-stimulatory signaling domain leading to robust activation⁶.

CD19 CARs were quickly translated to human disease by a series of investigator-initiated, single-center proof-of-concept trials. The first case report of CD19 CAR activity, against a multiply relapsed follicular lymphoma, was published in 2010 by researchers from the National Cancer Institute (NCI)⁷. Over the next five years, impressive results were described for adult acute lymphoblastic leukemia (ALL)^8,9, pediatric ALL^10,11, and chronic lymphocytic leukemia (CLL)¹². Importantly, these studies demonstrated the feasibility of the complex CAR manufacture process, which involves cell apheresis, in vitro expansion, genetic manipulation and re-infusion^13,14.

Researchers at the NCI were also the first to report an academic, single-center experience in relapsed or refractory diffuse large B cell lymphoma (DLBCL)¹⁵, while subsequent reports would optimize conditioning, toxicity and activity¹⁶. Researchers at Fred Hutchinson Cancer Research Center (FHCRC) and Seattle Children’s Hospital reported response rates ranging from 50–70% across a variety of lymphoma histologies, including DLBCL, transformed follicular lymphoma (tFL), FL, and mantle cell lymphoma (MCL)¹⁷. Most recently, researchers at the University of Pennsylvania reported their single-center experience with a CD19 CAR construct, achieving similar levels of activity across aggressive and indolent histologies¹⁸.

The most common toxicity of CAR therapy is Cytokine Release Syndrome (CRS)^8,19,20, which can manifest on a spectrum from mild febrile reaction to severe multi-organ dysfunction requiring intensive care and death. CAR-T Related Encephalopathy Syndrome (CRES) is neurologic syndrome that similarly occurs on a wide spectrum, from subtle defects in orientation and speech to severe encephalopathy, seizures, cerebral edema, and death²⁰. The mainstays of treatment include intensive supportive care, the anti-IL6 agent Tocilizumab, and high dose steroids. Although CRS is associated with increased risk of CRES^21–23, they appear to occur by pathologically distinct mechanisms, with the latter hypothesized to involve cytokine disruption of the blood-brain-barrier²². The pathogenesis and treatment of CRS and CRES are extensively reviewed elsewhere^20,24, and their prevention and management is a key aspect of CAR research.

Comparison of Products

The high rates of durable activity observed in early, investigator-sponsored single-center trials attracted interest from pharmaceutical, biotechnology, and investment groups, leading to rapid translation to global multi-center trials. A key hurdle to commercialization was the development of global manufacturing infrastructure for the off-site manipulation and transport of CAR T cells^25,26, along with the administrative, regulatory, and logistic processes to safely deliver them²⁷.

In October of 2017, Kite’s Axicabtagene Ciloleucel (Yescarta) was the first CAR T product approved for adult lymphoma, based on the NCI construct and results of the ZUMA-1 study²⁸. Tisagenlecleucel (Kymriah), which was derived from work at the University of Pennsylvania and had previously been approved for pediatric ALL, was subsequently approved for adult lymphoma based on the results of Novartis’ JULIET study²⁹. Finally, Juno Therapeutics, which has licensed technology from Memorial Sloan Kettering Cancer Center, the Fred Hutchinson Cancer Center, and the Seattle Children’s Research Institute, is developing their product Lisocabtagene maraleucel as part of the TRANSCEND trial³⁰ [Table 1].

Table 1.

Characteristics of Chimeric Antigen Receptor (CAR) T Cell Products

	Kite (Gilead)	Novartis	Juno (Celgene)
Product	Axicabtagene Ciloleucel Yescarta	Tisagenlecleucel Kymriah	Lisocabtagene maraleucel
Key Academic Partnership	NCI15	University of Pennsylvania18	FHCRC, Seattle Children’s17
Construct	FMC63 Mouse anti-CD19 CD28 Hinge CD28	FMC63 Mouse anti-CD19 CD8a Hinge 4-1BB	FMC63 Mouse anti-CD19 IgG4 Hinge 4-1BB EGFR suicide gene
Manufacturing	Anti-CD3 mAb Expansion γ-retroviral vector Whole PBMCs	Dynabead Expansion Lentiviral Vector Whole PBMCs Cryopreservation	Lentiviral Vector Defined CD4:CD8 ratio Cryopreservation
Key Trial	ZUMA-128	JULIET29	TRANSCEND30
Population	DLBCL, TFL, PMBCL	DLBCL	DLBCL, MCL, FL
Cell Dose	2×106 cells/kg	5×108 cells flat dose	5×107 cells flat dose
Conditioning	fludarabine 30 mg/m2 and cyclophosphamide 500 mg/m2 for 3 days	fludarabine 25 mg/m2 and cyclophosphamide 250 mg/m2/day × 3 days or bendamustine 90 mg/m2/day × 2 days	fludarabine 30 mg/m2 and cyclophosphamide 300 mg/m2 daily for 3 days
Response Rates	ORR = 82% CR (6-month) = 36%	ORR = 53% CR (6-month) = 30%	ORR = 84% CR (6-month) = 37%
Toxicity
Grade 3/4 CRS**	12%	23%	1%
Grade 3/4 Neurotox	31%	12%	12%

Abbreviations: DLCBL: Diffuse Large B-Cell Lymphoma. MCL: Mantle Cell Lymphoma. FL: Follicular Lymphoma. PMBCL: Primary Mediastinal Large B-Cell Lymphoma. TFL: Transformed Follicular Lymphoma.

^**Tisagenlecleucel uses Penn toxicity scale (which typically results in higher toxicity grades)⁴², other two products use Lee et al. scale⁴³

The three products share many similarities in design and clinical behavior. Data from key registration trials^28–30 suggest broadly similar rates of activity. Best response rates vary from 50–80%, but may be dependent in part on conditioning regimen and kinetics of early cell expansion, which do not necessarily predict for meaningful clinical benefit³¹. A more clinically meaningful measure is 3 or 6-month CR rate, as the majority of these patients who achieve these landmarks have durable responses²⁸.

All three constructs utilize the murine FMC63 scFv to target human CD19 and an intracellular human CD3 ζ chain to initiate signaling. However, there are differences in the remainder of the construct. Each uses a different hinge spacer region to join the extracellular scFV to the intracellular signaling domains, which can have subtle effects on CAR function^32,33.

An even more critical distinction is the co-stimulatory domain used to enhance function, with Yescarta utilizing a CD28 domain, while Kymriah and Juno’s construct utilize 4–1BB. Both signaling domains enhance T cell activity and signaling^34,35. However, emerging evidence from animal models suggest CD28 CARs expand more robustly and rapidly³⁶, while 4–1BB co-stimulation may induce less T cell exhaustion³⁷, in part by activating alternative metabolic and signaling pathways involved in T cell memory and differentiation³⁸. Some groups have demonstrated greater persistence though similar overall efficacy of 4–1BB CARs³⁹, which has not been observed in all models⁴⁰.

Clinically, CD28 based CARs display more robust early in vivo expansion^18,41, whereas differences in persistence or long-term activity remain to be demonstrated. In ALL³¹ and CLL⁴², robust expansion of a 4–1BB construct correlates with CR, but this relationship may not hold in NHL¹⁸. However, robust ²² CAR T cell expansion correlates with neurologic toxicity, which may explain the higher rate of CRES observed with CD28 CARs, although this remains speculative and no studies have directly addressed this question. Notably, toxicity in trials is measured using different scales, the Lee scale⁴³ for Juno and Kite and the University of Pennsylvania scale (which typically results in higher toxicity grades) ⁴² for Novartis.

There are also differences in the manufacturing (Table 1) and T cell expansion process. For example, axicabtagene is manufactured with a γ-retroviral vector, compared to the lentiviral vector used for tisagenlecleucel. Gamma retrovirus are more likely to integrate into the 5’ untranslated region of genes^44,45, which carries a risk of insertional oncogenesis, but this has not been observed in genetically modified mature T cells and remains a theoretical risk.

Based on studies of NHL ¹⁷ and B-ALL⁴⁶ performed at FHCRC, Juno uses a defined composition with a 1:1 ratio of CD4 and CD8 T cells to enhance activity and control toxicity. Activity differs between naïve and effector subsets within CD4 and CD8 subgroups, and CD4 cells provide help to CD8 killing⁴⁷. This may in part explain the low rates of toxicity, including neurotoxicity, reported thus far in TRANSCEND. However, this remains theoretical and clinical benefit over a non-selected product has not been formally demonstrated.

Although the three highlighted products are the most commercially developed, a large number of new products are currently under investigation. At the time of writing, approximately 300 CAR T cell trials are registered with ClinicalTrials.gov⁴⁸, with the bulk occurring in PR China⁴⁹ and the United States. Similarly, global infrastructure for cell therapy is expanding rapidly⁵⁰, such as Gilead’s manufacturing facility in Amsterdam, Netherlands. New strategies and approaches for CAR T cell development are highlighted in subsequent sections.

Patient Selection

Approximately half of patients treated for DLBCL are refractory to or relapse after front-line chemoimmunotherapy⁵¹. The SCHOLAR-1 analysis described poor outcomes associated with primary refractory disease^51,52 and disease refractory to second-line salvage^51,53,54 – response in these patients is on the order of 10–40%, and 2-year survival between 10–30%.

CAR approval trials typically enrolled patients failing two prior lines of therapy, and the Yescarta and Kymriah labels carry this requirement. In Zuma-1, most patients had failed more than 3 prior lines of therapy, making the high rates of durable response particularly impressive. Performance status (PS) has emerged a critical patient selection factor. ZUMA-1 did not enroll patients with ECOG PS > 1, and Juno has also adopted this requirement for their pivotal cohort in TRANSCEND due to inferior outcomes in patients with ECOG PS > 1.

DLBCL genotypes and immunophenotypes that predict response to chemoimmunotherapy and targeted therapy do not seem to be prognostic for CAR T cell response. Neither TRANSCEND, ZUMA-1, nor the University of Pennsylvania study detected a difference in response between Germinal Center B Cell-like and Activated B Cell-DLBCL ²⁸ or “double-hit” lymphomas ^55,56. The numbers for subset analysis are still small, and the outcome measured is the less clinically relevant response rate, rather than durability of response. However, this is a potentially exciting development for lymphoma therapy, as these patients are enriched in the relapsed and refractory population. Furthermore, in sub-group analysis of Zuma-1, there was no difference by number of prior treatments or prior auto-SCT²⁸. Retrospective analyses of commercial products may be beneficial in providing more data to guide treatment selection and prognosis.

While trials have primarily enrolled patients with DLBCL, other histologies are represented, though in limited number, and show significant activity. Disease eligibility differs between trials: for example, the ZUMA-1 trial, and therefore the approved indication for Yescarta, includes primary mediastinal B cell lymphoma (PBML), but not mantle cell lymphoma (MCL) patients. MCL patients were eligible for the TRANSCEND trial, although the core cohort, which will be used for approval and therefore the expected indication, will not include MCL. Furthermore, responses in MCL have not yet been reported for this study. Finally, both labels include patients with large cell lymphomas transformed from follicular lymphoma, but not other histologies such as CLL.

Response rate appears to be similar between aggressive and indolent lymphomas, but durability of response and overall response can differ. For example, in the University of Pennsylvania study, follicular lymphoma (FL) and DLBCL had response rates of 89% and 86% respectively, but FL had long PFS (median not reached) and OS (median not reached at 28.6 months), compared to a median PFS of 3.2 months in DLBCL and and OS 22.2 months¹⁸. Both the biology of disease and mechanisms of resistance and escape may explain differences in durability of response. To date, the University of Pennsylvania is the largest published experience with indolent lymphoma (iNHL), and iNHL-specific studies such as Kite’s ZUMA-5 (NCT02926833) are ongoing.

Bridging To CARs

Unlike traditional chemotherapeutics, CAR T cells require a “bridging” period during which patients must be managed while cells are shipped, manufactured, released and returned to the treating facility. Average time from initial apheresis to return of cells to the clinical site is 17 days for axicabtagene²⁸, while tisagenlecleucel expansion appears to be somewhat slower, measured at 23 days (range, 21–37 days)⁵⁷. The T cell collection process also requires at least 2 weeks off therapy to optimize cell counts, meaning that patients can be awaiting therapy for at least one month.

Clinical management during this “bridging” period is a challenge. For example, in the UPenn study, 10 of 38 enrolled patients were not treated due to rapid clinical deterioration¹⁸. Patients who have previously shown rapid progression with chemo-refractory disease, and those with disease at sites where it could be the cause of imminent catastrophe, must be carefully monitored, and often offered additional therapy while treatment planning proceeds.

Treatment can be delayed if patients experience a transient complication or require further treatment. Tisagenlecleucel is cryopreserved after collection and has been validated for up to 9 months of storage ( furthermore, patients in JULIANA were treated up to 360 days after collection). While axicabtagene is not frozen prior to manufacture, it returns as a cryopreserved product and can be stored for several months.

The rate of manufacturing failure in the key clinical trials was higher for tisgenlecleucel²⁹ compared to axicabtagene²⁸, at 7% and 1% respectively. Furthermore, manufacturing failure and disease progression affect trial interpretation, as 101 of 111 patients were ultimately dosed in the axicabtagene study²⁸, whereas only 67% of patients were treated in the tisagenlecleucel study²⁹, and both studies reported outcomes in the infused population. Therefore, limiting the length and increasing the reliability of the manufacturing process is an ongoing focus of both commercial and academic groups⁵⁸. An alternative approach involves allogeneic CAR products⁵⁹, which can be prepared in bulk and used “off-the-shelf,” eliminating bridging necessitated by autologous manufacturing, avoiding manufacturing failures, and streamlining production to reduce manufacturing cost⁶⁰. However, the use of allogeneic cell products is limited by both graft-versus-host and host-versus-graft immune rejection⁶¹. Approaches to permit allogeneic donor tolerance include knockout of the endogenous TCR to eliminate recognition of host antigens^62,63,64, knockout of MHC expression to prevent host rejection⁶⁵, and use of T cell precursors^66–68 to leverage tolerance⁶⁹. Clinical experience with allogeneic CARs is limited, but Phase I studies have demonstrated the feasibility of allogeneic manufacturing for treatment of ALL⁷⁰ (NCT02808442).

Conditioning, which is given 2–7 days prior to T cell infusion, is a key part of the bridging strategy. The theoretical benefits of conditioning include direct cytotoxic effect against malignancy, depletion of regulatory T and other suppressive immune cells and decreased competition for homeostatic cytokines^71,72. The optimal conditioning regimen has not been rigorously examined, but clinical experience from the NCI ⁷³and FHCRC ¹⁷ are instructive. While there are slight differences in the conditioning regimens used for different products and trials (Table 1), investigators have coalesced around low-dose cyclophosphamide and fludarabine, as utilized in TRANSCEND and ZUMA-1. By contrast, investigators in the UPenn single center study allowed treating physicians to select between several conditioning regimens, including minimal conditioning in heavily pre-treated patients¹⁸. This permits the possibility of tailoring treatment, such as giving more aggressive alkylator-based therapy to more rapidly progressing patients. This flexibility has extended in a more limited form to the Novartis’ JULIET study, where a bendamustine-based conditioning regimen was also allowed²⁹.

Auto, Allo and CAR: Sequencing Cell Therapy

With the approval of CAR T cells, there are now three highly active cell-based therapies with high rates of activity and durable response: high dose chemotherapy with autologous hematopoietic stem cell transplant (auto-HSCT), allogeneic hematopoietic stem cell transplant (allo-HSCT), and CAR T therapy. Selecting the appropriate therapy for each patient will be a key area of clinical investigation over the next decade. Here, we use DLBCL as a model for sequencing of cell therapy.

For patients with DLBCL who relapse after prolonged (>12 months) response to front-line therapy, response rates to second line salvage chemotherapy are high (approximately 90%)⁷⁴. Responding (chemosensitive) patients are subsequently consolidated with high dose chemotherapy autologous stem cell transplant (auto-SCT), with 3-year EFS ranges from 40–50% after consolidation in both the pre-Rituximab⁷⁵ and post-Rituximab era⁷⁴. Patients with early relapse (<12 months) are a higher-risk group with lower response rate to salvage therapy⁵⁴, although, among chemosensitive patients, approximately 50% can have durable response to auto-SCT⁷⁶.

Auto-SCT remains therefore the standard of care for chemosensitive relapsed DLBCL and should be pursued outside of a clinical trial. However, the recently initiated ZUMA-7 (NCT03391466) study is randomizing patients in first relapse to either CAR T cell therapy with Yescarta or standard of care R-ICE salvage followed by auto-SCT. Patients with early relapse and/or high-risk disease biology may be particularly appropriate candidates for trials of alternative therapies such as CAR T cells or early allogeneic transplant^77–79.

All patients enrolled on CAR T cell clinical studies necessarily had relapsed, measurable disease. Whether CAR T cells can be used as consolidative therapy is an open question. For example, an ongoing study at Memorial Sloan Kettering is exploring whether CAR T cells can be infused to further consolidate response after auto-SCT in patients at high risk of relapse (NCT01840566). In theory, immune memory and surveillance is an attractive mechanism to control disease and prevent relapse⁸⁰. On the other hand, CAR T cell activation and expansion may be dependent on tumor burden, as evident in ALL and CLL trials with 4–1BB constructs^31,42. Therefore, heavily pre-treated patients with low disease burden may not induce sufficient expansion for CAR function and long-term persistence.

Allogeneic Transplant

Allogeneic T cell transplant (allo-SCT) has demonstrated benefit for patients relapsing after auto-SCT, particularly if patients are chemosensitive and have achieved a late relapse⁸¹. The past decade has been marked by a shift toward Reduced Intensity Conditioning (RIC) allo-SCT^81–83, with decreased non-relapse mortality and morbidity from transplant complications such as graft-versus-host disease. This leads to the paradigm that patients with chemosensitive relapse after auto-SCT should be considered for consolidative allo-SCT, whereas patients with persistent disease after auto-SCT are candidates for CAR therapy.

Patients with complete response to CAR T therapy appear to have high rates of durable response, on the order of 60–70% at 2 years²⁸. Although these data are not yet mature, we do not routinely consolidate CAR T responders with allo-HSCT.

Since CAR T cells frequently induce deep responses which could facilitate transplant, but this deep response only leads to durable responses in about half of NHL patients, there would be tremendous clinical value in predicting which patients relapse and planning early transplantation. Upcoming clinical trials and experience with commercial products must define parameters to predict early relapse and a role for consolidative allo-SCT; potential parameters may include lymphoma genotype or phenotype, depth of response, duration of prior response, and recovery of B cell aplasia. Allo-SCT remains a curative option for patients progressing after CAR therapy.

For patients who relapse after allogeneic stem cell transplants, CAR T cells can be derived from either the host or donor ^84,85. The previously described trials of autologous CAR therapy enrolled small numbers of patients progressing after allo-HSCT ^43,86, including one patient in the UPenn lymphoma trial¹⁸. Although these cells are harvested from the transplanted host, they are donor-derived cells which have been tolerized in vivo, which decreases (though does not eliminate) the risk for graft-versus-host disease (GVHD). No GVHD was reported in these studies.

There is a greater theoretical risk for graft-versus-host disease when cells are directly derived from an allogeneic donor, as CARs may simultaneously express donor-derived allo-reactive TCR. This concern is partially mitigated by animal models showing suppression of allogeneic responses under certain conditions, which include robust allogeneic TCR ⁸⁷. Donor-derived CARs have therefore been safely infused as therapy for B cell malignancies^88–90. Critically, in these studies, patients were no longer receiving active immunosuppressive therapy. Third party cells reactive against EBV and CMV are highly effective^91–96 for the treatment of viremia and EBV-associated lymphomas in highly immunosuppressed hosts, and their graft-versus-host potential is somewhat limited by their restricted viral specifity. Virus-specific T cells from an allogeneic donor could therefore theoretically be transduced with anti-CD19 CAR with minimal risk of GVHD, which is being explored in several trials including an EBV-specific CAR construct at MSK (NCT01430390).

Sequencing of auto, allo, and CAR is complex and evolving, but simple principles can be extracted from existing data. Patients with chemosensitive first relapse should proceed to auto-SCT outside of a clinical trial. In primary refractory disease, second-line relapse, or early relapse, outcomes are less robust with conventional auto-SCT, and these populations are excellent candidates for CAR trials. For patients who have CR to CAR T cell therapy, responses can be durable, and therefore we do not routinely consolidate with auto- or allo-SCT, though longer follow up and characteristics predictive of relapse are necessary.

Relapse and Persistence

While response rates across CAR T trials are high, on the order of 80%, rates of long-term durable responses are significantly lower, on the order of 30–40%. The most common and direct mechanism of tumor escape after initial response is through loss of CD19 antigen expression.

For example, in ALL patients treated with CD19 CAR T cells, 28% of patients relapsed with CD19 negative blasts ¹¹. Lymphoblasts can lose CD19 expression via alternatively spliced isoforms⁹⁷ which are present prior to therapy and selected by CD19 treatment ⁹⁸. Interestingly, escape has been observed through a lineage switch ^99,100 with emergence of a myeloid clone at relapse.

To counter CD19 negative relapse, CARs have been developed against additional B cell antigens such as CD20¹⁰¹ and CD22¹⁰². Investigators at the NCI reported promising safety data with CD22 CARs in 17 patients previously treated with CD19 CAR T cells¹⁰³. Investigational approaches to prevent CD19 negative relapse include planned sequential infusions of CD19 and CD22 CARs¹⁰⁴, and the use of dual-targeted CARs that recognize both CD19 and a second antigen, decreasing the odds of escape^104,105.

CD19 Positive Relapse

Unlike traditional small molecule or protein-based therapies, relapse to cell therapy can occur not only by development of tumor resistance, but also by loss of intrinsic activity of the ‘living drug’. B cell aplasia serves as a biomarker of CD19 CAR T cell activity, but recovery of B cell counts does not necessarily portend immediate relapse. In the NCI lymphoma experience, 5 of 6 patients in CR who recovered blood counts were able to maintain CR ¹⁶. Similarly, in the University of Pennsylvania study¹⁸, 8 of 16 patients in sustained CR > 6 months persisted in CR despite recovery of polyclonal B cells. This occurs despite the persistence of detectable CAR DNA in responders for the duration of the 24 month trial period¹⁸. By comparison, in pediatric ALL patients, loss of aplasia is associated with more rapid relapse⁸⁶, which likely speaks to the underlying biology and kinetics of disease.

If CD19 positive relapse occurs, suggesting loss of T cell function as a possible mechanism of escape, there are several options. The most straight-forward is re-treatment with CD19-targeted CARs. This has been described in the axicabtagene trial, where nine patients who had progressive disease after initial response were re-treated, with 5 responses and 2 on-going response²⁸.

There is also a concern that anti-CAR immune responses, including human anti-mouse antibodies (HAMA), can have reactivity against the foreign CAR construct and lead to CAR rejection^106–108. HAMA antibodies are present in the vast majority of CAR patients, although approximately 1/3^rd had transient increase in antibodies and the presence of such response did not correlate with Tisagenlecleucel response¹⁰⁹. To address this issue, in pediatric ALL, the University of Pennsylvania is studying re-treatment with a CD19 CAR using a humanized scFv construct¹¹⁰.

Exhaustion and Escape

Functional exhaustion is a second hypothesized mechanism of CAR T cell dysfunction. T cells exposed to chronic antigenic stimulation, as occurs in some viral infections and cancers, undergo a regulatory differentiation program defined by loss of polyfunctional cytokine production and constitutive expression of inhibitor surface receptors such as PD-1, LAG-3, TIM-3 ^111,112. In mouse models, chronic antigen exposure is necessary for CAR T cell persistence, but is also associated with upregulation of exhaustion markers^113,114. Expression of these inhibitory checkpoint molecules has been observed in patients treated with CAR T cells, particularly in progressors¹⁸.

Antibodies against inhibitory checkpoint receptors such as PD-1^115,116 may therefore be a clinically relevant strategy to reverse exhaustion in patients who progress after CAR T cell therapy despite persistent CD19 expression. Early case reports suggest checkpoint blockade can reverse CAR exhaustion^117,118. A Phase I study of the anti-PD1 drug Pembrolizumab is ongoing at UPenn in patients who progress after initial CAR-T therapy (NCT02650999), while the ZUMA-6 study is studying the upfront combination of Atezolizumab and Yescarta to prevent exhaustion and enhance activity in NHL¹¹⁹.

Investigational methods to prevent exhaustion involve engineering disruption of checkpoint pathways into the T cell genome. CAR T cells can synthesize and secrete an anti-PD-1 antibody leading to enhanced function in mouse models ^120–123. Alternatively, the PD-1 locus can be directly disrupted ^65,123,124 as can additional checkpoint proteins such as LAG-3 ¹²¹. Alternatively, constructs that incorporate both CD28 and 4–1BB⁴⁰ signaling motifs display more robust CAR signaling⁴⁰, with less tendency toward exhaustion upon repeated stimulation¹²⁵.

The Microenvironment

Infused, activated T cells traffic indiscriminately¹²⁶ across body tissues and compartments, although memory cells have a natural tendency to exchange with the bone marrow T cell pool¹²⁷; this may in part explain the efficacy of CARs against leukemias. Strategies to enhance adoptive T cell trafficking to other sites has focused on expressing chemokine receptors^128–131 that drive T cell homing to peripheral tissue¹³². Furthermore, T cells in both solid tumors^133,134 and lymphomas¹³⁵ must penetrate an organized and dense stroma. Strategies to enhance T cell trafficking within the tumor include heparanase degradation of tumor stroma¹³⁶.

Once present in a tumor, CAR T cells must overcome extrinsic suppression from a hostile tumor microenvironment. In response, “Armored CARs” can be engineered to secrete IL-12¹³⁷, a proinflammatory cytokine, which enhances CAR function^138–141 both by autocrine signaling on T cells themselves¹⁴² and through modulation of the immune microenvironment¹⁴¹, such as inhibitory macrophages^138,143. This strategy is being tested to enhance CAR T cell efficacy in CD19+ hematologic malignancies (NCT03085173) and serves as a prototype for strategies to counteract inhibitory signals from regulatory cells. CAR T cells can alternatively be modified to secrete pro-inflammatory cytokines such as IL-7 ¹⁴⁴ and IL-15¹⁴⁵ and IL-18¹⁴⁶ or express immunostimulatory receptors^147,148.

Conclusion

CAR T cell therapies have rapidly progressed to become a highly active, FDA approved therapy for relapsed and refractory lymphoma. Clinically, there remains much to be learned about patient selection and management, including the optimal sequencing of CAR T cells compared to other cell therapies. Meanwhile, we have only begun to design new and better CAR T cells in the laboratory, and the clinical relevance of these enhancements remains unknown.

We therefore expect to see a rapid proliferation in investigational products and approved indications for CAR T cells, potentially including a wide variety of lymphoma histologies. The pace of progress, while welcome, will make it challenging to thoughtfully incorporate these therapies into clinical practice. We must harness clinical experience to design well-controlled randomized trials, not only to test the safety and efficacy of new CAR designs, but also to find the best ways to use them.