Nature of Tumor Control by Permanently and Transiently Modified GD2 Chimeric Antigen Receptor T Cells in Xenograft Models of Neuroblastoma

Neuroblastoma is the most common extracranial pediatric solid tumor. Derived from neuroendocrine tissue of the sympathetic nervous system, it accounts for 9% of cancer diagnoses and 15% of cancer-related deaths in children (1). The current standard of care for high-risk disease consists of chemotherapy, surgery, consolidation chemotherapy followed by stem-cell transplant, tumor-directed radiation, and antibody-based therapy. This exhaustive regimen still only yields a 3-year event-free survival rate of approximately 45% (2).

Historically, poor tumor immunogenicity and immune evasion (3, 4) have resulted in largely disappointing T-cell therapy trials. Advances in the understanding of T-cell and tumor interaction, cell production techniques, and engineering platforms, as well as the recognition of the need for host conditioning, have recently allowed for significant improvements. Chimeric antigen receptors (CAR) are molecules composed of an extracellular antigen-binding domain derived from the immunoglobulin variable chain (scFv) linked to intracellular T-cell activating and costimulatory domains (5, 6). The antigen-specific scFv allows for MHC-independent tumor antigen recognition that initiates robust T-cell stimulation, even in response to previously immunoevasive malignancies (7).

In the most successful clinical trial of CARs to date, Kalos and colleagues (8) and Porter and colleagues (9) demonstrated complete remissions lasting >3 years in 2 of 3 adults with chronic lymphocytic leukemia (CLL) using CARs directed against the B-cell antigen CD19. Translating this success to acute leukemia, our group recently reported the induction of complete remissions in 2 children with acute lymphoblastic leukemia (ALL) using the same CD19 CAR T cells (10).

The dramatic clinical responses observed in these studies have employed T cells permanently engineered to express CARs using lentiviral vectors. This process involves ex vivo exposure of harvested autologous lymphocytes to self-inactivating lentiviral vector encoding a CAR, resulting in genomic integration of the CAR transgene. While >500 patient-years of data suggest that this modification is extremely unlikely to result in insertional mutagenesis in mature lymphocytes (11), these data are from adults and the increased life-span of modified cells in children raises additional theoretical safety concerns. More importantly, when targeting solid tumor antigens, the risk of on-target off-tumor toxicity becomes a significant concern. Several adverse events have demonstrated the potential risks of uncontrolled CAR T cells (12, 13), and have highlighted the need for safer CAR T cells moving forward, especially in early clinical testing (14, 15). Given these considerations, we and other groups have previously reported the development of an mRNA electroporation–based approach to induce transient CAR expression (16–18). This strategy creates an efficient CAR expression system that ensures complete loss of CAR-driven T-cell activity in a predictable time frame without the need to administer other systemic agents to eliminate modified T cells. We have reported the efficacy of transiently modified CD19 CAR T cells in a disseminated xenograft model of systemic ALL (19), and recently demonstrated enhanced efficacy of these transiently modified cells when delivered repeatedly in an optimized dosing strategy (20). This optimized therapeutic regimen approached the antitumor responses observed with permanently modified CD19 CAR T cells and demonstrated long-term disease control, suggesting that multiple infusions of transiently modified CAR T cells may present an alternative to genome-modifying T-cell engineering techniques.

RNA CAR T cells have demonstrated in vitro activity (21) and in vivo efficacy in localized models of solid tumors, and have similarly shown enhanced efficacy using multiple cell infusions (17, 22). On the basis of these findings, as well as our own experience with RNA CAR T cells in ALL, we evaluated a CAR targeting GD2, a diasialoganglioside expressed on the surface of most neuroblastomas (1) that has already been shown to be an effective target for neuroblastoma immunotherapy (23). A single-chain antibody variable region fragment (scFv) targeting GD2 was linked to the CD3ζ and 4-1BB intracellular signaling domains and tested in localized and disseminated animal models of neuroblastoma. We demonstrate that multiple infusions of RNA GD2 CAR T cells result in effective antitumor responses to disease in vivo, and that a single low-dose infusion of permanently modified GD2 CAR T cells results in long-term control of disseminated neuroblastoma. Multiple infusions of RNA GD2 CAR T cells are less effective at controlling disseminated disease, and our data highlight the potential mechanism underlying this lack of efficacy. Together, these data clarify the necessary components for success of transiently modified CAR T cells in solid tumors.

Human T cells were isolated from healthy donors by the University of Pennsylvania Human Immunology Core. Cells were incubated with microbeads coated with CD3 and CD28 stimulatory antibodies (Life Technologies; catalog #111.32D). T cells were combined at a ratio of 1:3 (cells:beads) and cultured in T-cell culture media. Beads were removed on day 7 or 8 of stimulation. Cells were counted, and volumes were measured (Multisizer 3; Beckman Coulter) every 48 hours. When cell growth kinetics and volume suggested that cells had rested down from activation, they were cryopreserved.

CARs containing scFv domains directed against GD2 or CD19 linked to CD3ζ and 4-1BB intracellular signaling domains were produced as previously described (GD2-z construct was generously provided by Dr. Malcolm Brenner, Baylor College of Medicine, Houston, TX; refs. 24, 25). Development of constructs for RNA manufacture was performed as previously described (17). The mScript RNA System (CellScript; catalog #MSC11625) was used to generate capped in vitro transcribed RNA, which was purified using an RNeasy Mini Kit (Qiagen, Inc.; catalog #74104). T cells were expanded and frozen as described above. Before electroporation, cells were thawed, washed three times with Opti-MEM, and resuspended in Opti-MEM medium at a final concentration of 1 to 3 × 10⁸ cells/mL. T cells were then mixed with transcribed mRNA at a concentration of 10 μg mRNA/0.1 mL T cells and electroporated in a 2-mm cuvette using an ECM830 Electro Square Wave Porator (both from Harvard Apparatus BTX; catalogs #450125 and #450002) 1–2 hours before infusion into animals. Viability after transfection ranged from 50% to 80%, and in all cases, T cells demonstrated >95% CAR expression.

High-titer, replication-defective lentiviral vectors were produced using 293T human embryonic kidney (HEK) cells (26). HEK293T cells were seeded at 10⁷ cells per T150 tissue culture flask 24 hours before transfection. On the day of transfection, cells were treated with 7 μg of pMDG.1, 18 μg of pRSV.rev, 18 μg of pMDLg/p.RRE packaging plasmids, and 15 μg of transfer plasmid in the presence of either Express-In Transfection Reagent (Open Biosystems) or Lipofectamine 2000 transfection reagent (Life Technologies; catalog #11668019). Transfer plasmids containing CAR constructs were modified so that expression of the CAR was under the control of the EF-1α promoter as previously described (25). Viral supernatants were harvested 24 and 48 hours after transfection and concentrated by ultracentrifugation overnight at 10,500 ×g. Twenty-four hours after initial stimulation, T cells were exposed to lentiviral vector at a concentration of 5 to 10 infectious particles per T cell, and then cultured as described above.

Cells were washed and resuspended in FACS buffer (PBS + 1% bovine serum albumin), then incubated with biotin-labeled polyclonal goat anti-mouse F(ab)2 antibody (Jackson Immunoresearch; catalog #115-066-072) at 4°C for 25 minutes, and then washed twice. Cells were then incubated with R-phycoerythrin (PE)–conjugated streptavidin (BD Biosciences; catalog #554061) for 10 minutes at 4°C, then washed twice. Flow cytometry acquisition was performed on an Accuri C6 cytometer (BD Biosciences; catalogs #342975 and #653118). Analysis was performed using the FlowJo software (Treestar, Inc.).

Six- to 10-week-old NOD-SCID-γc^−/− (NSG) mice were obtained from The Jackson Laboratory or bred in-house under an approved Institutional Animal Care and Use Committee (IACUC) protocol and maintained in pathogen-free conditions. For flank tumor studies, animals were injected subcutaneously with 2 × 10⁷ SY5Y human neuroblastoma cells suspended in 0.2 mL Matrigel (BD Biosciences; catalog #354234). For disseminated tumor studies, animals were injected via tail vein with 2 × 10⁶ SY5Y cells in 0.1 mL sterile PBS. T cells were injected in 0.1 mL sterile PBS. Cyclophosphamide (Baxter Healthcare; catalog #1001995601) was injected i.p. at 60 mg/kg 24 hours before any cell administration following the initial infusion. Animals were monitored for signs of disease progression and overt toxicity, such as xenogeneic graft-versus-host disease (GVHD), as evidenced by >10% loss in body weight, loss of fur, diarrhea, conjunctivitis, and disease-related hind-limb paralysis.

Flank tumor measurements were made biweekly using electronic calipers (Fowler-Sylvac; catalog #54-200-777). The longest length and width measurements were recorded and tumor volume was calculated according to the formula [(width + length)/2)³]/2. Mice were sacrificed when tumors reached 3 cm³, or when tumor burden inhibited animal activity.

Disease burdens were monitored over time using a Xenogen IVIS bioluminescent imaging system, as previously described (19, 27).

Histology was performed by the Pathology Core Facility at the Children's Hospital of Philadelphia. Subcutaneous tumors and tumor-infiltrated livers were excised postmortem, and preserved in 4% paraformaldehyde. Image analysis was performed using the ImageScope software (Aperio). Staining was performed on a Bond Max automated staining system with the Bond Refine polymer staining kit (both Leica Microsystems). Standard protocol was followed, with the exception of the primary antibody incubation, which was extended to 1 hour at room temperature. Anti-human CD3 antibody (Dako; catalog #M7254) was used at 1:50 dilution and antigen retrieval was performed with the E1 retrieval solution for 20 minutes. Slides were rinsed and dehydrated through a series of ascending concentrations of ethanol and xylene. Stained slides were then digitally scanned at ×20 magnification on an Aperio OS slide scanner (Aperio).

The SH-SY5Y (SY5Y) tumor cell line was obtained originally from the American Type Culture Collection (CRL-2266; ATCC) and generously donated to our laboratory by Dr. John Maris, Children's Hospital of Philadelphia. The parent cell line was genotyped by short tandem repeat (STR) analysis (28). Cell lines were tested for Mycoplasma and verified every 6 months, or after any genetic modification, such as luciferase transduction, to ensure identity and Mycoplasma-free status.

All statistical analysis was performed with Prism 4 (Graphpad Software). For comparison among multiple groups, Kruskal–Wallis analysis was performed with Dunn multiple comparison tests to compare individual groups. Survival curves were compared using the log-rank test with the Bonferroni correction for multiple comparisons.

Subcutaneous injection of tumor cells in the mouse flank is a well-established model of solid tumor therapy in mice. To assess the antitumor efficacy of our transiently modified GD2 CAR T cells in a solid tumor model of neuroblastoma, NSG mice were injected s.c. with 2 × 10⁷ SY5Y (a GD2⁺ human neuroblastoma cell line) cells and given 2 weeks to establish large flank tumors. On day 15, 5 × 10⁶ human T cells electroporated with RNA encoding either a GD2 or CD19 CAR bearing the 4-1BB and CD3ζ signaling domains (GD2-BB/z or CD19-BB/z) were injected into the tumor. On day 20, several mice were sacrificed, and tumors were excised and probed for the presence of human T cells by immunohistochemical analysis. Intratumoral (i.t.) injection of RNA GD2 CAR T cells resulted in rapid necrosis of target neuroblastoma, with T cells diffusely present throughout the tumor environment (Fig. 1A). RNA CD19 CAR T cells, on the other hand, did not alter the tumor architecture after i.t. injection, and quickly diffused out of the tumor and localized to the periphery of the tissue (Fig. 1B). Mice not sacrificed for histology received two additional i.t. injections of 1.5 × 10⁷ cells each, and RNA GD2 CAR T cells mediated significant antitumor responses, whereas RNA CD19 CAR T cells permitted rapid tumor growth (P < 0.001; Fig. 1C). Tumor expansion observed in mice treated i.t. with RNA CD19 CAR T cells was similar to that in mice treated with saline (P = 0.7912).

Subcutaneous tumors have many anatomic and physiologic differences when compared with spontaneous intra-abdominal tumors, several of which have direct bearing on the physiology of cellular therapies. Previous studies of CAR T cells in solid tumors, such as epithelial ovarian cancer (29), have used a peritoneal tumor model in which tumor cells are injected directly into the peritoneum to establish disease along the internal peritoneal wall, mimicking natural disease. We sought to develop a model that would more closely mimic the clinical setting of disseminated neuroblastoma in which CAR T cells are likely to be given. SY5Y neuroblastoma cells were engineered to express click beetle green luciferase (SY5Y-CBG) before delivery, which allowed for longitudinal observation of a disseminated tumor burden. Tumor cells were injected i.v., and animals were observed for the establishment and progression of systemic disease. Within 2 hours of cell delivery, bioluminescent signal was observed in the lungs, and then rapidly disseminated (Supplementary Fig. S1). As early as day 7, disease was detected in the abdomen, and was subsequently observed in the bone marrow and lymph nodes. Necropsy revealed that this abdominal signal resulted from hepatic infiltration of neuroblastoma, causing diffuse nodular hepatic disease.

We first tested the efficacy of permanently modified GD2 CAR T cells in our disseminated model of neuroblastoma. CAR T cells were engineered using a well-established lentiviral vector transduction technique (9, 25). NSG mice were given 2 × 10⁶ SY5Y-CBG cells via tail vein and given 1 to 2 weeks to establish a significant disease burden (>10⁶ photons/s/cm²/sr), after which CAR T cells were delivered systemically. Animals were imaged twice weekly to monitor disease burden. A single infusion of as few as 10⁶ permanently modified GD2 CAR T cells was sufficient to suppress and control disseminated neuroblastoma over time (P < 0.001; Fig. 2A), whereas tumor growth progressed comparably after systemically delivered control CD19 CAR T cells and saline (P = 0.8872). Not surprisingly, the disease control observed had a significant impact on the survival of animals treated with GD2 CAR T cells (P = 0.0257; Fig. 2B). Interestingly, all mortality in animals treated with GD2 CAR T cells resulted from xenogeneic GVHD, with 63% of GVHD mortality occurring in disease remission and the remaining 37% with minimal disease burdens (<2 × 10⁷ photons/sec/cm²/sr).

We next used this disseminated model system to assess the efficacy of RNA GD2 CAR T cells against neuroblastoma. Disseminated disease was established as described, and transiently modified T cells were delivered intravenously. On the basis of our experience with ALL (20), we undertook a multiple infusion strategy using RNA-electroporated T cells. Mice were treated initially with 5 × 10⁶ cells on day 8, and 1.5 × 10⁷ on days 22 and 29 after tumor implantation. We observed a transient decline in disease burden after the initial infusion, and modest disease stabilization with the next two infusions (Fig. 3A). Although this strategy significantly slowed disease progression (P = 0.0032), it was unable to control the disease in the long term. This delayed progression did, however, prolong animal survival (P = 0.0128; Fig. 3B). Again, animals treated with CD19 CAR T cells were statistically similar in both disease progression (P = 0.4175) and overall survival (P = 0.9834) to those treated with saline.

Visual representations of disease burden and response to therapy highlight the variations in antitumor activity between transiently and permanently modified CAR T cells (Fig. 4). Animals treated with lentivirally modified GD2 CAR T cells showed rapid loss of bioluminescent signal with maintained disease suppression in the long term. Interestingly, these animals occasionally have focal disease recurrence, which is lost on subsequent evaluation. This likely represents the long-term, antitumor T-cell memory of permanently modified CAR T cells. As demonstrated in the quantitative bioluminescent graph (Fig. 3B), animals treated with GD2 RNA CAR T cells had an initial decline in disease burden after the first cell infusion, followed by transient stabilizations with the subsequent two infusions.

To better understand the dynamics of the T cell–tumor interaction in our disseminated solid tumor model, we examined tumor sites using immunohistochemical analysis. Animals were given the standard 1 to 2 weeks to establish disease after i.v. tumor injection, and then given either a single infusion of lentivirally modified CAR T cells or three infusions of RNA CAR T cells, each separated by 1 week, as in studies described previously. Animals were sacrificed on the day indicated, and the excised livers were sent for histopathologic examination and immunohistochemical staining for human CD3. Animals receiving permanently modified GD2 CAR T cells had rapid infiltration of CAR T cells into tumor sites by day 3, and showed a significant T-cell presence in all tumor sites by day 5 (Fig. 5A). Permanently modified control CD19 CAR T cells did not show any significant tumor infiltration (Fig. 5B). Alternatively, by day 3 after infusion of the first dose of RNA-modified GD2 CAR T cells, minimal infiltration of these CAR T cells into the tumor site was found, and this T-cell presence had largely dissipated by day 6. Twenty-four hours after the second infusion of RNA GD2 CAR T cells (on day 7), a minor accumulation of T cells was found around the periphery of the tumor, which had again largely dissipated by day 10. This pattern was also observed after the third cell infusion, with a peripheral accumulation around the edge of the tumor within 24 hours, which was quickly lost (Fig. 6A). As expected, control RNA CD19 CAR T cells did not show any significant tumor infiltration (Fig. 6B).

High-risk neuroblastoma remains a significant cause of cancer-related death in children. Although we have seen dramatic improvement in the treatment of both solid and liquid pediatric tumors, survival rates for both have plateaued since the mid-1990s (30). Patients with relapsed neuroblastoma are largely incurable, although temporary control of disease can be achieved for some patients. Thus, a new therapeutic paradigm is needed to improve survival. Not until recently has engineered T-cell therapy begun to meet its potential. Integration of the powerful cytotoxicity, significant in vivo proliferation, and surveillance mechanisms of memory T cells with the antigen specificity endowed by CARs has created a platform for the success of cell therapy. As we have suggested, the recent successes of CAR therapy (8–10) may primarily be attributed to improved cell-manufacturing processes that generate T cells armed with rapid and potent in vivo effector and proliferative capacity (31).

Previous reports have demonstrated the efficacy of GD2 CAR T cells against disseminated GD2⁺ melanoma using two infusions of 10⁷ CAR T cells each (32). In the only other study to our knowledge of CAR T cells in a model of systemic solid malignancy, 2 × 10⁷ prostate-specific membrane antigen (PSMA)–directed T cells demonstrated efficacy in a disseminated model of prostate cancer (33). In this study, we were able to demonstrate disease eradication in more than half of the animals and disease suppression in the rest of the animals treated using a single infusion of only 10⁶ cells, approximately 5% of the dose used in the previous studies of CAR T cells in solid tumors. Not only were these cells able to control disease, we also observed massive cell expansion driven by antigen-specific activation. Indeed the T-cell expansion was high enough to cause death of animals from xenogeneic GVHD. T cells that could not recognize tumor (CD19 CAR T cells) did not proliferate to the same degree, as evidenced by the lack of GVHD in these animals. Both of these observations highlight the significant proliferative capacity of our CAR T cells, an attribute endowed by a combination of cell-manufacturing processes and CAR signaling constructs. The ability to use fewer cells to achieve clinical efficacy is of particular importance in pediatric patients who undergo extensive chemotherapy, which creates significant lymphodepletion as part of standard treatment for many malignancies. Reducing the number of cells needed for treatment would enhance the number of patients eligible for this therapy.

GD2 serves as an exciting target in neuroblastoma, given its narrow expression profile. However, as is the case with most solid-tumor antigens, on-target off-tumor toxicity remains a persistent issue, as GD2 is also expressed on sensory nerves. Clinical experience has demonstrated that antibody therapy directed at GD2 can result in significant neuropathic pain (34, 35). This toxicity is managed clinically during infusion of the antibody, and when the infusion ends, the toxicity resolves. Previous clinical studies with the first-generation GD2/ζ CAR T cells found that, of the 19 patients followed for more than 5 years, only 2 patients had local pain at the site of tumor necrosis and 1 had unexplained local pain (36), with no patient experiencing pain greater than grade 3. While suggestive that the physiology of this toxicity may be distinct when using an antibody versus a cell-based therapy comprising a single-chain Fv without an Fc domain, this remains a potential on-target off-tumor toxicity of highly active GD2 CAR T cells. Our cell production techniques, structure of CAR signaling domains, in vivo expansion, and persistence all differ from those described in previous clinical trials (31). For this reason, we have been very interested in an RNA-based approach.

Having demonstrated cure of leukemic mice using multiple infusions of RNA-electroporated CAR T cells in ALL, we sought to assess the efficacy of this engineering platform in treating solid tumors. In the traditional flank tumor xenograft model, our RNA GD2 CAR T cells demonstrated rapid and potent antitumor activity when delivered locally. Although this demonstrated cytolytic activity against neuroblastoma in vivo, this model is more difficult to translate clinically, as local tumor injections may not be feasible in some patients. Moreover, late-stage cancer patients may have many sites of disease. To more closely mimic systemic disease, we developed a sensitive in vivo bioluminescent model that enabled the establishment and monitoring of disseminated disease, primarily manifested in the liver and bone marrow. We found that a single small dose of permanently modified second-generation GD2/4-1BB/ζ CAR T cells could produce long-term disease control, while multiple infusions of RNA GD2 CAR T cells were able to delay disease progression and enhance animal survival but could not control disease in the long term.

On the basis of the fundamental differences in tumor biology, we have hypothesized that the success observed using RNA CAR T cells for liquid tumors (20) might not directly translate to solid tumors. T cells targeting leukemic cells immediately traffic to sites of disease, as disease reservoirs are the natural sites of T-cell residence and surveillance. This all but removes the necessity of antigen-driven T-cell trafficking and retention. Trafficking in solid tumors is further complicated, given the disorganized nature of the tumor vasculature, another issue that is averted when targeting liquid tumors. Thus, while creating a proliferative niche and optimal dosing strategy were the primary hurdles in successful targeting of liquid tumors using RNA CAR T cells (19), solid tumors present the additional barriers of trafficking, penetration, and efficient antigen engagement within a 6-day CAR expression period.

Our histologic data support these hypotheses, elucidating the dynamics of T cell and tumor interaction in vivo. Within 72 hours of the infusion of lentiviral CAR T cells, there is a large burden of T cells within the tumor site, and by day 5 all tumor sites are heavily infiltrated with CAR T cells. The dynamics of RNA CAR T-cell activity are quite different: 72 hours after the first cell infusion, a few cells have penetrated into the tumor site, providing a potentially favorable effector:target ratio on day 3, but these cells are lost by day 6. After a second infusion, the T cells localize to site of these expanding tumors, but are unable to penetrate and are lost quickly. The third infusion results in similar kinetics of T-cell localization to the periphery of enlarging tumor masses. These observations correlate well with our bioluminescent data, which show a transient decline in tumor burden after the first cell infusion, followed by modest disease stabilization, likely reflecting a short-lived antitumor response at the periphery after the second and third T-cell infusions (Fig. 3B). The diminishing returns seen with each additional T-cell infusion reflect the fact that each infusion faces the same structural and vascular barriers as the first, in addition to an expanding tumor burden.

The rapid increase in the number of permanently modified GD2 CAR T cells present within the tumor seen over days 1 to 5 after T-cell infusion raises the question of whether this occurs as a result of continued cell recruitment, in situ expansion of recruited cells, or both. If indeed this is continued cell recruitment of circulating CAR T cells to the site of tumor antigen over days 1 to 5, it would suggest that RNA CAR T cells do not have the same antigen-driven migratory capabilities, as this is within the window of RNA CAR expression. If the increased T-cell infiltration is a result of intratumoral proliferation in response to antigen, this suggests a difference in the proliferative capacity of RNA and permanently modified CAR T cells. Although we cannot formally distinguish between these two possibilities, the latter seems likely to be the dominant mechanism, given the importance of antigen-driven proliferation to CAR T-cell efficacy.

The observation that three infusions of RNA GD2 CAR T cells were not able to control systemic disease led us to assess the efficacy of repeated infusions. In another iteration of the multiple infusion strategy, we injected a total of 10⁸ RNA CAR T cells over the course of 8 weeks and found that while nine infusions further slowed disease progression as compared with three infusions, disease continued to progress (data not shown). This observation, along with the poor tumor infiltration of RNA CAR T cells, suggests that increasing cell dose will not increase the efficacy of this therapy, as it did with ALL (20).

The mechanisms underlying this observed difference in T-cell retention and infiltration are unclear. We have evaluated CAR molecule expression after RNA electroporation in vitro and demonstrated a progressive decay over the course of 6 days, but we have not assessed expression kinetics in vivo. It is certainly possible that infusion into and circulation within a living organism alters CAR expression. It is known that upon engagement with the target antigen, the T-cell receptor is internalized as more TCRs are integrated into the cell membrane. This receptor recycling may also apply to CAR molecules, and whereas lentivirally modified CAR T cells are able to continually present new CARs on the surface, RNA-modified CARs are more limited. Another manner in which RNA CAR signal may be diluted is during in vivo expansion. As antigen engagement results in T-cell activation and proliferation, permanently modified CAR T cells produce daughter cells that also have permanent CAR expression. RNA-modified CAR T-cell proliferation, however, may produce daughter cells with a fraction of the originally delivered CAR RNA. These kinetics may play an essential role in limiting RNA CAR T-cell retention and infiltration at tumor sites.

Mechanisms independent of CAR expression kinetics may also account partially for the differences observed. Although the majority of cell production is identical for both RNA and lentivirus CAR T cells, it is unclear whether the process of electroporation has an effect on T-cell function. Data from intratumoral injection studies suggest that these RNA CAR T cells are functional and able to mount effective antitumor responses; however, subtle differences in T-cell activity may be present that are unmasked when delivering cells systemically. These differences may occur in the form of variations in cell signaling or variability in expression of surface molecules that affect T-cell trafficking, retention, and infiltration, such as chemokine receptors, adhesion molecules, and T-cell inhibitory receptors. It is possible that a combination of these factors accounts for the differences observed, and further characterization of these mechanisms will allow identification of the processes that can be targeted to enhance efficacy.

To translate this therapy, which shows greater efficacy with local delivery, into a clinically viable treatment modality, solving the problem T-cell recruitment and expansion in metastatic tumors is key. One likely option is the prolongation of RNA CAR expression. One such method under study is the biochemical stabilization of transferred RNA molecules (37). This strategy will produce CAR T cells with more sustained receptor expression that retain the self-inactivating qualities of RNA electroporation, prolonging the critical time period of T-cell infiltration.

Although the RNA CAR T-cell approach will minimize toxicities that occur over days, such as seen with permanently modified CD19 CARs (10), this may not be the case for adverse events reported in previous clinical trials (12, 13) that occur within hours of cell infusion. We can predict, however, that the toxicities observed would have been self-limited and, thus, potentially less severe, as antigen-driven cell division would result in restriction and temporally restricted expression of CAR molecules. Our recent report of cytokine release syndrome in children treated with CD19 CAR T cells (10) is an example of toxicity that may have been self-limited had these T cells been modified with RNA CARs, as opposed to permanently modified with lentiviral CARs. Similarly, long-term B-cell aplasia, which we have observed as a necessary complication of CD19 CAR T-cell persistence, would not occur with this approach. Another approach to limit CAR T-cell persistence is to use suicide gene systems that may eliminate 1 to 3 logs of cells. However, T cells have robust proliferative capacity and are capable of repopulating from a single cell, and thus a guaranteed “off” system such as RNA electroporation remains an attractive approach.

In conclusion, we demonstrate the efficacy of permanently modified GD2 CAR T cells in controlling disseminated neuroblastoma, and that a lack of longer-term control of metastatic disease by temporarily modified GD2 CAR T cells results from reduced tumor infiltration in the setting of large tumor burdens.

C.H. June, S.A. Grubb, and Y. Zhao report receiving commercial research grants from Novartis. S.A. Grubb is a consultant/advisory board member for Novartis. Y. Zhao has ownership interest in a patent from the University of Pennsylvania. No potential conflicts of interest were disclosed by the other authors.

Conception and design: N. Singh, S.A. Grupp, D.M. Barrett, Y. Zhao

Development of methodology: N. Singh, X. Liu, S.A. Grupp, D.M. Barrett, Y. Zhao

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): N. Singh, X. Liu, J. Hulitt, S.A. Grupp, Y. Zhao

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): N. Singh, S.A. Grupp, D.M. Barrett, Y. Zhao

Writing, review, and/or revision of the manuscript: N. Singh, C.H. June, S.A. Grupp, D.M. Barrett, Y. Zhao

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Jiang, S.A. Grupp

Study supervision: S.A. Grupp, D.M. Barrett, Y. Zhao

The authors thank Daniel Martinez and Dr. Bruce Pawel for their assistance with immunohistochemistry, and Dr. Curt Lind for assistance with HLA phenotyping.

This study was supported by NIH R01CA120409 (to C.H. June and Y. Zhao), P01CA066726 (to C.H. June and Y. Zhao), and grants from the Pennsylvania Department of Health, Cookies for Kids Cancer, Solving Kids Cancer, the WW Smith Charitable Trust and the Weinberg Funds (to S.A. Grupp)

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.