Eradication of Neuroblastoma by T Cells Redirected with an Optimized GD2-Specific Chimeric Antigen Receptor and Interleukin-15 Free

Neuroblastoma is the most common extracranial tumor of childhood (1). Because children with high-risk neuroblastoma continue to have poor outcome despite intensive chemotherapy regimens, new treatment strategies are required (2, 3). The disialoganglioside GD2, which is consistently expressed in neuroblastoma tumor cells with restricted expression in normal tissues, represents the target of choice for immunotherapy in neuroblastoma, through either a specific mAb (Ch14.18, dinutuximab) or, more recently, chimeric antigen receptor engineered T cells (GD2.CAR-Ts) (3–6).

In the first clinical study with GD2.CAR-Ts, the specific scFv was derived from the murine 14.G2a mAb coupled with the ζ-chain endodomain (first-generation CAR), and engrafted in Epstein–Barr virus–specific CTLs (5, 6). In the second clinical trial, the same CAR was modified to encode two costimulatory endodomains in tandem (CD28 and OX40) (third-generation CAR) and expressed in polyclonal T lymphocytes (4). Both studies showed that infusions of GD2.CAR-Ts were well tolerated, even when combined with conditioning regimens to lymphodeplete the host, and promoted some objective clinical responses (4–6).

However, response rates in patients with neuroblastoma receiving GD2.CAR-Ts remain significantly inferior to those observed in patients with acute lymphoblastic leukemia treated with CD19-specific CAR-Ts (7). It is also evident that, although the inclusion of costimulatory endodomains like CD28 or 4-1BB in the CD19-specific CAR is necessary and sufficient to achieve clinical responses (7–11), CAR costimulation is not equally effective in neuroblastoma (4). Furthermore, although the expansion of CD19-specific CAR-Ts in the peripheral blood correlates with clinical responses (7), increases of GD2.CAR-Ts in the peripheral blood, even when these cells are infused in lymphodepleted patients, do not promote better antitumor responses (4). We speculated that, despite homeostatic expansion in peripheral blood, in contrast to CD19-specific CAR-Ts, GD2.CAR-Ts do not promptly receive the costimulation required for full activation while in the circulation, as they do not engage their cognate antigen. Furthermore, when GD2.CAR-Ts reach the tumor site, the costimulation provided through the signaling moieties of the CAR remains suboptimal to achieve complete elimination of tumor cells due to local inhibitory/suppressive factors. We observed here that the inclusion of the IL15 cytokine within the CAR cassette is beneficial for GD2.CAR-Ts by promoting both enhanced survival in the absence of the cognate antigen, and prolonged antitumor activity.

The neuroblastoma cell lines IMR-32, LAN-1, and SKNLP were purchased from ATCC, while the CHLA255 cell line was described previously (12). Cells were maintained in RPMI1640 supplemented with 10% FBS and 2 mmol/L l-glutamine (Invitrogen). Tumor cell lines were transduced with a gamma retroviral vector encoding the enhanced GFP (eGFP) or the Firefly-luciferase (FFLuc) genes as described previously (13). Cells were routinely (monthly) screened for the presence of Mycoplasma and expression of the GD2 antigen and kept in culture for less than 3 consecutive months, after which aliquots from the original expanded vial were used.

The cassette encoding the GD2-specific scFv (scFv14G2a), the CD8α stalk and transmembrane domain, the CD28 intracellular domain and CD3ζ chain was cloned into the SFG backbone (GD2.CAR). The scFv14G2a was previously identified (14) and used in clinical studies (4–6). We modified this scFv14G2a by removing 16 amino acids from the COOH-region of the V_H (AKTTPPSVYGRVYVSS). Indeed, when we aligned the previously used V_H sequence into the IGBLAST NCBI database, these 16 amino acids, which belong to the CDR3-IMGT (germline) of the V_H, exceeded the alignment of the V_H sequence reported in the dataset. Furthermore, two versions of the scFv14G2a were produced by conserving the previously used 9-aminoacid linker (SL) between the V_L and V_H fragments or including a 20-aminoacid linker (LL). We then generated vector cassettes encoding the CAR in combination with the IL-15 cdna (GD2.CAR.15) using a 2A-sequence peptide (15). In a third vector, we also cloned the inducible caspase-9 suicide gene (16) into the GD2.CAR.15 using a second 2A-sequence peptide (iC9.GD2.CAR.15). Finally, for selected experiments, the scFv14G2a in the iC9.GD2.CAR.15 vector was swapped with the CD19-specific scFv (17) to obtain the iC9.CD19.CAR.IL-15 vector. Vectors encoding the Firefly-luciferase (FFluc) and the fusion protein GFP have been previously described (13). The full-length human PD-L1 (accession_NM_014143.4) was PCR amplified, cloned into the retroviral vector SFG and used to transduce the CHLA255 tumor cell line to generate tumor cells constitutively expressing high levels of PD-L1 (CHLA255-PD-L1⁺). Transient retroviral supernatants were produced as previously described (13). For GMP translation and validation studies a clinical grade packaging cell line was generated by using the PG13 packaging cell line (gibbon ape leukemia virus pseudotyping packaging cell line; CRL-10686, ATCC) as previously described (5). We used the highest titer clone to establish the master cell bank as recommended by FDA guidelines.

Peripheral blood mononuclear cells (PBMC) were isolated from buffy coats (Gulf Coast Regional Blood Center) using Lymphoprep (Accurate Chemical and Scientific Corporation). PBMCs were then activated using 1 μg/mL immobilized CD3 (Miltenyi Biotec) and CD28 (BD Biosciences) mAbs. After 3 days, cells were transduced with retroviral supernatants and expanded in complete medium (45% RPMI1640 and 45% Click's medium, 10% FBS, and 2 mmol/L GlutaMAX) supplemented with IL7 (10 ng/mL, PeproTech) and IL15 (5 ng/mL, PeproTech) and expanded by feeding them every 3 to 4 days with fresh complete media and cytokines, as described previously (18–20).

T-cell lysates were resuspended in 2× Laemelli Buffer (Bio-Rad) in reducing (with β-mercaptoethanol) or nonreducing (without β-mercaptoethanol) conditions. All lysates were separated in 4% to 15% SDS-PAGE gels and transferred to polyvinylidene diflouride membranes (all Bio-Rad). Blots were probed for human CD3ζ Ab (Santa Cruz Biotechnology) diluted 1:1,000 in TBS-Tween/5% skim milk. Membranes were then incubated with HRP-conjugated goat α-mouse or goat α-rabbit IgG (both Santa Cruz Biotechnology) at a dilution of 1:3,000 and imaged using the ECL Substrate Kit on a ChemiDoc MP System (both BioRad) according to the manufacturer's instructions.

eGFP-transduced tumor cells were plated in 24-well plates at a concentration of 0.25 or 0.5 × 10⁶ cells/well the day before the addition of T cells, in the absence of cytokines. Different effector cells to target cells (E:T) ratios were used (1:1, 1:5 and 1:10). For the repetitive coculture experiments, T cells and eGFP-tagged neuroblastoma cells were cocultured at the E:T ratio of 1:1 or 1:5. At day 3, all cells were transferred into a new well and seeded with 0.25 × 10⁶ neuroblastoma cells. T cells and neuroblastoma cells were quantified by flow cytometry after 4 cycles based on CD3 and GFP expression, respectively. Supernatant were also collected for cytokine measurements after each cycle. For each experiment, nontransduced T cells were used as a negative control (Ctrl-Ts).

CAR-Ts and control T cells were labeled with carboxyfluorescein succinimidyl ester (CFSE; Invitrogen) and plated in the wells precoated with or without the 1A7 antibody (0.5 μg/well). On day 5, cells were collected and analyzed for CFSE dilution by flow cytometry.

Cells were incubated in Hank's Balanced Salt Solution (Life Technologies) containing 2% FBS, 10 mmol/L HEPES (Life Technologies), and either 10 μg/mL Hoechst 33342 (Sigma-Aldrich) for 90 minutes at 37°C. Cells were then washed with ice-cold HBSS and labeled with CD3-APC for flow cytometry analysis. To detect the side population (SP) cells, Hoechst 33342 samples were excited using a 355 nm UV laser on a LSRII/Fortessa flow cytometer (BD Biosciences; refs. 21, 22).

Culture supernatants were collected after 24 hours of coculture to measure the release of IL2, IFNγ, and TNFα, and after 72 hours for IL15, using specific ELISA (R&D Systems, Inc). Plasma obtained from mice was analyzed for the presence of human cytokines using the Bio-Plex Pro Human Cytokine 8-plex Assay (Bio-Rad) or Human High Sensitivity Cytokine B Premixed Mag Luminex Performance Assay (R&D Systems) following the manufacturer's instructions. Data were collected and analyzed using the Lumina-200 System and the Bio-Plex Manager 6.1 software (Bio-Rad).

The B/B homodimerizer AP20187, a chemical inducer of dimerization drug (CID; Clontech Laboratories), was added at 50 nmol/L to T-cell cultures and induction of apoptosis evaluated 24 hours later using Annexin-V/7AAD staining (BD Biosciences) and FACS analysis (16).

Cells were acquired using the BD FACS Fortessa (BD Biosciences) and analyzed using the FlowJo Software (Tree Star). mAbs specific for GD2, CD45, CD3, CD8, CD45RA, CD45RO, PD-1, CCR7, CD45, LAG-3, TIM-3, TIGIT, CTLA-4, CD25, CD69, active caspase-3, and granzyme B were obtained from BD Biosciences. To detect the GD2.CAR expression in transduced T cells, we used the 1A7 anti-idiotype mAb specific for the 14G2a.scFv and PE or APC-conjugated rat anti-mouse secondary mAbs (BD Biosciences; ref. 5). In selected experiments, counting beads (Count Bright, Invitrogen) were added following the manufacturer's instructions. For the intracellular staining, cells were stained with Zombie dye and surface antibodies, including CD3 and GD2, washed, fixed, and permeabilized with Cytofix/Cytoperm (BD Biosciences), and then stained with active caspase-3 in 1× permeabilization buffer, following the manufacturer's instructions.

Mouse experiments were performed in accordance with the University of North Carolina (UNC, Chapel Hill, NC) animal husbandry guidelines according to protocols approved by the UNC institutional animal care and use committee. For these studies, 6- to 8-week-old female and male NSG mice (NOD.Cg-Prkdc^scid IL2rg^tm1Wj1/SzJ, UNC Animal Studies Core Facility) received intravenously via tail injection 2 × 10⁶ CHLA255 cells labeled with FFLuc. Tumor engraftment was measured as bioluminescence signal intensity (BLI) and expressed as total flux (p/s) using the in vivo IVIS imaging (Caliper Life Science; ref. 23). Two weeks after tumor engraftment, mice received intravenously either 2 × 10⁶, 5 × 10⁶, or 10 × 10⁶ control-Ts (Ctrl) or CAR-Ts. No exogenous cytokines were administered to the mice. Mice were bled at specific intervals (10–15 days, as per UNC-IACUC guidelines) to measure cytokines in the plasma and frequency of CAR-Ts. Mice were followed after CAR-Ts treatment and euthanized when tumor growth caused discomfort as per veterinarian's recommendation. At the time of euthanasia, blood, spleen, liver, and bone marrow were isolated and analyzed to detect CAR-Ts. To assess differences in response to continuous exposure to tumor cells, we performed tumor rechallenge experiments, where tumor-bearing mice were further injected intravenously with 3 × 10⁶ CHLA255 cells labeled with FFLuc at day 7 and 14 or at day 28 after CAR-T treatment. In selected experiments, tumor-bearing mice received intraperitoneally the PD-1–blocking antibody (clone EH12.2H7, BioLegend; 200 μg/mouse every 4 days) starting from the day before the CAR-Ts infusion. In some experiments, mice were engrafted with the FFLuc CHLA255 tumor cell line genetically engineered to express high levels of PD-L1 (CHLA255-PD-L1⁺ cells) in the attempt to further promote T-cell exhaustion in vivo. To ensure the function of the suicide gene, 10 × 10⁶ FFLuc-tagged CAR-Ts were injected intravenously into NSG mice. T-cell expansion was then tracked by in vivo imaging measuring T-cell BLI. Mice were randomized to receive vehicle or CID (AP20187, 50 μg/mouse), and then imaging was performed to evaluate the elimination of T cells. In selected experiments, mice were challenged with 3 × 10⁶ CHLA255 cells, 4 days after CID/vehicle treatment.

ANOVA with Bonferroni correction and the two-sided unpaired t test were used for comparison of 3 or more groups, or 2 groups, respectively, as stated in the figure legends. Mixed model ANOVA was applied to compare tumor growth in different groups of mice. Survival curves were plotted using the Kaplan–Meier method, and the differences in the survival between groups were assessed by log-rank test. Data are presented as mean ± SD or SEM as stated in the figure legends. Statistical significance was defined at P < 0.05. Statistical analysis was performed with Prism 5 (GraphPad Software).

We first reevaluated the structure of the GD2.CAR generated using the scFv derived from the 14G2a mAb. Specifically, we compared the previously constructed GD2.CARs encoding the CD28 endodomain (24) with a modified GD2.CAR, characterized by a shortened V_H sequence and containing the hinge and transmembrane domain of human CD8α instead of the Ig-derived hinge and CD28 transmembrane domain, and found that the new optimized GD2.CAR had superior activity in in vitro experiments (Supplementary Fig. S1A–S1F), and reduced self-dimerization (Supplementary Fig. S1G). We then generated two cassettes, encoding either the optimized GD2.CAR alone or the GD2.CAR in combination with IL15 (GD2.CAR.15) using a 2A-sequence peptide (Fig. 1A). CAR expression (Fig. 1B) and expansion (Fig. 1C) of T cells transduced with either GD2.CAR or GD2.CAR.15 vectors were similar. Phenotypically both GD2.CAR-Ts and GD2.CAR.15-Ts contained similar amounts of CD4⁺ and CD8⁺ T cells (Supplementary Fig. S2A). In contrast, GD2.CAR.15-Ts contained higher frequency of central memory T cells (CD45RO⁺CCR7⁺; 15% ± 7% vs. 7% ± 3%, P < 0.01) and stem cell–like T cells (CD45RA⁺CCR7⁺; 25% ± 11% vs. 6% ± 2%, P < 0.001) than GD2.CAR-Ts (Fig. 1D). Furthermore, GD2.CAR.15-Ts showed lower constitutive expression of PD-1 as compared with GD2.CAR-Ts in both central memory (25% ± 20% vs. 39% ± 17%, P < 0.001) and stem cell–like T cells (8% ± 11% vs. 17% ± 14%, P < 0.01; Fig. 1E; Supplementary Fig. S2B). Downregulation of PD-1 in GD2.CAR.IL15-Ts was confirmed at the transcription level (Supplementary Fig. S2C). In GD2.CAR.IL15-Ts, we also observed low expression of another inhibitory receptor, namely LAG-3 (Supplementary Fig. S2D), while the activation markers CD25 and CD69 were equally expressed in GD2.CAR-Ts and GD2.CAR.IL15-Ts (Supplementary Fig. S2E). Susceptibility to activation induced cell death (AICD) was also comparable between GD2.CAR-Ts and GD2.CAR.15-Ts (Supplementary Fig. S2F). Overall these data suggest that the low expression of PD-1 and LAG-3 in GD2.CAR.IL15-Ts does not reflect lower activation but rather a superior fitness. These data were corroborated in experiments showing that GD2.CAR.15-Ts proliferated and persisted longer in the absence of exogenous cytokines or antigen support in vitro (Supplementary Fig. S2G) and in vivo (Supplementary Fig. S3A and S3B), mimicking the advantage they will have in the circulation before engaging tumor cells.

We cocultured GD2.CAR-Ts and GD2.CAR.15-Ts with four neuroblastoma cell lines (including MYCN-amplified neuroblastoma cells) at an E:T ratio of 1:1 (Supplementary Fig. S4A). Both GD2.CAR-Ts and GD2.CAR.15-Ts efficiently eliminated tumor cells (Fig. 2A and Supplementary Fig. S4B), including putative tumor stem cells identified as SP cells (Supplementary Fig. S4C). However, GD2.CAR.15-Ts proliferated significantly better in response to all neuroblastoma cell line tested (Fig. 2B). IL15 was detectable in the supernatant of cocultures with GD2.CAR.15-Ts, while no significant differences were observed for other cytokines, including IL2, TNFα, and IFNγ or granzymeB release as compared with GD2.CAR-Ts (Fig. 2C; Supplementary Fig. S4D and S4E). Moreover, as observed during the in vitro expansion with cytokines, without antigen stimulation, GD2.CAR.15-Ts showed lower expression of PD-1 (Fig. 2D) and LAG-3 (Supplementary Fig. S4F) upon coculture with neuroblastoma cells as compared with GD2.CAR-Ts. Target cells showed the same level of cleaved caspase-3, which is another marker of apoptosis, 24 hours post coculture (Supplementary Fig. S4G), indicating similar kinetic of cytotoxic activity of GD2.CAR-Ts and GD2.CAR.IL15-Ts. Remarkably, GD2.CAR.15-Ts showed sustained antitumor activity in stress coculture experiments (25) in which T cells were repetitively exposed to antigen engagement using CHLA255 and LAN-1 cells (Supplementary Fig. S5A). After 4 cycles of coculture, GD2.CAR.15-Ts, but not GD2.CAR-Ts, significantly controlled tumor growth of both CHLA255 and LAN-1 cells (P < 0.05 and P < 0.01, respectively; Fig. 2E; Supplementary Fig. S5B) and proliferated in response to repetitive exposure to each respective tumor cell line (P < 0.01 and P < 0.05, respectively; Fig. 2F). GD2.CAR.15-Ts continued to produce IL15 after each cycle of stimulation in response to both CHLA255 (Fig. 2G) and LAN-1 (Supplementary Fig. S5C) cells. IFNγ was released by CAR-Ts independently of the IL15 expression after the second round of stimulation with CHLA255 cells (Supplementary Fig. S5D), and upon repetitive antigen exposure, GD2.CAR.15-Ts continued to show lower PD-1 expression as compared with GD2.CAR-Ts (Fig. 2H). As observed after repetitive exposures to the tumor cells, CD25 and CD69 expression were similar between GD2.CAR-Ts and GD2.CAR.IL15-Ts (Supplementary Fig. S5E), and similar AICD was observed (Supplementary Fig. S5F). Overall, transgenic expression of IL15 harnessed significantly superior antitumor activity in an antigen-dependent fashion by promoting sustained T-cell proliferation after multiple exposures to tumor cells.

GD2.CAR.15-Ts are protective in an in vivo tumor rechallenge model. We compared the antitumor activity of GD2.CAR-Ts and GD2.CAR.15-Ts in a metastatic neuroblastoma model using NSG mice engrafted with the CHLA255 neuroblastoma tumor cell line (Fig. 3A). At the time of T-cell infusion mice were randomized based on the tumor BLI to ensure similar tumor burden between the experimental groups (Supplementary Fig. S6A). Tumor grew rapidly in mice treated with Ctrl cells, while GD2.CAR-Ts and GD2.CAR.15-Ts promoted equal control of tumor growth at day 50 when 1 × 10⁷ T cells/mouse were infused (Fig. 3B). In contrast, only GD2.CAR.15-Ts succeeded in controlling tumor growth at day 50 when 5 × 10⁶ T cells/mouse or 2 × 10⁶ T cells/mouse were infused (Fig. 3C and D). The protective effect of GD2.CAR.15-Ts was even more prominent in tumor rechallenge experiments, when mice, in addition to the initial tumor burden, received weekly infusions of neuroblastoma tumor cells (Fig. 4A). GD2.CAR.15-Ts–treated mice were tumor free >50 days after T-cell infusion, while tumor growth was evident 2 weeks after the first tumor rechallenge in mice treated with GD2.CAR-Ts (Fig. 4B–D). GD2.CAR.15-Ts were detectable in the peripheral blood, and, consistently with the in vitro data, showed lower PD-1 expression (Fig. 4E). GD2.CAR.15-Ts were also detected in bone marrow, liver, and spleen at the time of euthanasia (Fig. 4F) and showed low PD-1 expression, although in these later stages they were mostly effector cells (CD45RO⁺CD45RA^–CCR7^–; Supplementary Fig. S6B–S6D). In addition to IFNγ, IL15 produced by GD2.CAR.15-Ts was detectable in the plasma up to day 23 postinfusion (Fig. 4G), while IL2 and TNFα were below the limit of detection (Supplementary Fig. S6E). To investigate whether the decreased PD-1 expression in GD2.CAR.15-Ts protects these cells in vivo from tumor induced exhaustion, tumor-bearing mice were treated with GD2.CAR-Ts or GD2.CAR.15-Ts in combination with the PD-1–blocking Ab (Supplementary Fig. S7A). PD-1 blockade did not provide any advantage in mice treated with GD2.CAR.15-Ts, because these cells show low PD-1 expression (Supplementary Fig. S7B). However, we also found that PD-1 blockade did not significantly improve persistence, cytokine release (Supplementary Fig. S7C and S7D), and antitumor activity (Supplementary Fig. S7B) in mice treated with CAR.GD2-Ts. When we used CHLA255 cells constitutively expressing PD-L1 (Supplementary Fig. S8A), GD2.CAR.15-Ts cells continued to control tumor growth with or without PD-1 blockade, while a modest advantage was observed with PD-1 blockade in mice treated with GD2.CAR-Ts (Supplementary Fig. S8B–S8D). These data indicate that GD2.CAR.15-Ts persist better after tumor clearance and after tumor rechallenge, but these effects are not directly correlated with an improved resistance to PD-L1 inhibition.

Because of the potential for aberrant T-cell proliferation or toxicities in the presence of sustained exposure to IL15, we included the iC9 safety switch gene into our construct, to promptly remove GD2.CAR.15-Ts. Upon inclusion of the iC9 cdna into the GD2.CAR.15 cassette, we unexpectedly observed a progressive loss of CAR expression when cells were maintained in culture, with a consequent reduced antitumor activity in vitro (Supplementary Fig. S9A). This phenomenon was a unique feature of the GD2.CAR, as a similar cassette constructed to express the CD19-specific CAR did not show any significant reduction of the CAR expression (Supplementary Fig. S9B). Because the only difference in the two scFvs (CD19-specific and GD2-specific) was the size of the linker (20 versus 9 amino acids, respectively), we replaced the short-linker in the GD2-specific scFv with a long-linker (Supplementary Fig. S9C). Although the size of the linker did not significantly affect the expression and function of the GD2.CAR in the absence of the iC9 (Fig. 9D–F), the long-linker stabilized the GD2.CAR expression when the iC9 cdna was incorporated within the cassette (iC9.GD2.CAR.15; Supplementary Fig. S10A and S10B), allowing proper T-cell expansion in vitro (Supplementary Fig. S10C), and preservation of the same cell subset composition (Supplementary Fig. S10D). Expanded iC9.GD2.CAR.15-Ts showed functional activity of all three genes incorporated within the cassette. Specifically, the inclusion of the iC9 caused cell death when iC9.GD2.CAR.15-Ts were incubated with 50 nmol/L of the CID/AP20187 (Supplementary Fig. S10E). Moreover, iC9.GD2.CAR.15-Ts showed in vitro antitumor activity, proliferative capacity against CHLA255 or LAN-1 cells comparable with that of GD2.CAR.15-Ts (Supplementary Fig. S10F and Supplementary Fig. S11A and S11B), similar pattern of PD-1 expression (Supplementary Fig. S10G and Supplementary Fig. S11C), and similar release of cytokines (Supplementary Fig. S10H and Supplementary Fig. S11D). Overall, an optimal design of the GD2.CAR allowed the functional accommodation of both the IL-15 and iC9 transgenes.

Using our rechallenge tumor model, we finally assessed in vivo the antitumor effects of iC9.GD2.CAR.15-Ts (Fig. 5A). Although tumor BLI rapidly increased in the control group, mice treated with GD2.CAR.15-Ts and iC9.GD2.CAR.15-Ts remained tumor free by day 40 after rechallenge (Fig. 5B and C). This translated in a significantly better tumor free survival (P < 0.0001; Fig. 5D). GD2.CAR.15-Ts and iC9.GD2.CAR.15-Ts showed similar persistence (Fig. 5E) in mice after infusion, without significant differences in T-cell composition (Supplementary Fig. S12A). Together, our data show that the incorporation of the iC9 safety switch into an optimized GD2.CAR.15 construct remains beneficial in vivo.

To validate the functionality of the iC9 suicide gene, NSG mice infused intravenously with eGFP-FFLuc labeled iC9.GD2.CAR.15-Ts were treated with either AP20187 (CID; 50 μg/mouse) or diluent (vehicle; Fig. 6A). While progressively increasing in mice treated with vehicle, BLI decreased promptly and remained low after CID administration (Fig. 6B). Similarly, phenotypic analysis showed that circulating human T cells almost completely disappeared in CID-treated mice, while increased in vehicle-treated mice (2% ± 0.5% vs. 25% ± 2%, P < 0.001; Fig. 6C). When we analyzed tissues collected at the end of the observation period, CAR-Ts were reduced in CID-treated mice as compared with vehicle-treated mice (Fig. 6D). Finally, in selected experiments when mice were rechallenged with tumor cells after CID or vehicle treatment to reactivate residual T cells, we observed no increase in BLI in any of the mice that previously received CID (Fig. 6E), which suggests that the residual CAR-Ts after CID treatment are not reactivated and thus cannot cause toxicity in vivo.

To further demonstrate the clinical applicability of the proposed approach, we have generated a clinical grade PG13 retroviral packaging cell line (20) and produced clinical grade supernatant. As shown in Supplementary Fig. S13, iC9.GD2.CAR.15-Ts generated under GMP conditions for clinical use maintained the same characteristic of those generated in preclinical experiments ensuring the clinical application of the proposed strategy in patients with neuroblastoma. The clinical protocol was recently approved by FDA (IND 18500).

GD2.CAR-Ts expressing costimulatory endodomains have been proven safe in patients with neuroblastoma, but their antitumor activity remains limited (4), suggesting that additional preclinical assessment is needed to improve their clinical activity. Here, we show that modifications of the GD2.CAR structure and the incorporation of IL15 proved to be key requirements to enhance antitumor activity without causing on target toxicity.

Our rationale for redesigning GD2.CAR-Ts stems from both laboratory and clinical observations. Laboratory observations have highlighted how structural characteristics of CAR molecules, such as the framework regions of the scFv, and the length and origin of the hinge, and of the transmembrane endodomain, can impact the functionality of CAR-Ts (26–28). Specifically for the scFv of the GD2.CAR obtained from the 14g2a Ab, high tendency for self-aggregation secondary to interactions within the scFv framework regions has been described (27). This phenomenon was associated to an overstimulation of GD2.CAR-Ts, resulting in the premature exhaustion and reduced antitumor activity of these cells when the CAR contained the CD28 costimulatory signal (27). Furthermore, recent evidences indicated that a mutated version of the 14g2a.scFv (E101K), which enhances its binding affinity, although effective in controlling tumor growth, caused severe neurotoxicity in one study (29, 30), but not in another study (31). On the other hand, in patients with B-cell–derived malignancies the expansion of CD19-specific CAR-Ts in the peripheral blood correlates with objective antitumor activity (7). In contrast GD2-specific CAR-Ts, although expanding in the peripheral blood when infused upon lymphodepletion, did not promote significant clinical benefits (4). On the basis of these observations for the 14g2a-derived CAR, we sought to modify structural parts of our construct to address intrinsic limitations of the GD2.CAR cassette. In addition, we optioned for the inclusion of the IL15 cytokine as a survival factor for GD2.CAR-Ts rather than relying simply on PIK3 and TRAF mediated signaling delivered by the combination of CD28 or OX40/4-1BB endodomains upon antigen encounter (24, 32–35). We reasoned this would provide a cytokine signaling that is independent from the costimulation received, and from IL2 that is released by CAR-Ts upon antigen engagement (32, 35). We found that structural modifications of the GD2.CAR cassette reduced its spontaneous self-aggregation. Furthermore, upon incorporation of IL15, GD2.CAR-Ts retained specificity, but showed enhanced expansion and antitumor activities in vitro, after repeated tumor exposures, and in vivo in our innovative xenograft solid tumor model, which includes multiple tumor cell rechallenges to recapitulate the high tumor burden in patients with solid tumors. Importantly, these effects were obtained without evidences of “on target off tumor” toxicities.

Our study highlights that the inclusion of IL15 affords GD2.CAR-Ts with other important properties for translation into an effective therapy. First, the incorporation of the IL15 within the GD2.CAR produced an enrichment of cells with “memory” and “stem-cell” like phenotypes. This occurred during the expansion culture period, was independent from the specific antigen encounter by CAR-Ts, and was significantly superior to that already provided by the addition of exogenous IL15 and lL7 cytokines in the culture (19, 36). Although the mechanistic explanation of this phenomenon requires further studies, the immediate implication of this observation is that IL15 secreted by GD2.CAR-Ts promotes additional effects that are superior to those produced by soluble IL15. This may also be critical in patients, as the IL15 available in the plasma upon lymphodepletion, despite promoting T-cell expansion, may be inferior to the IL15 directly produced and released by GD2.CAR-Ts (4). Exogenous administration of IL15 is accompanied by systemic toxicities (37, 38), and thus we did not formally compare exogenous administration versus the endogenous production, as the former would not likely be clinically relevant. A membrane-bound form of IL15 expressed by CAR-Ts may be also beneficial (39). However, the incorporation of this gene would compromise the expression and function of the CAR, as the construct exceeds the cargo capacity of one single vector.

As second advantage, we observed is that GD2.CAR.15-Ts had significantly reduced expression of PD-1 and LAG-3 with and without antigen stimulation, and thus may be less susceptible to PD-L1–mediated inhibition, because neuroblastoma cells can express PD-L1, either constitutively or upon exposure to IFNγ (Supplementary Fig. S8A). The observed low expression of PD-1 and LAG-3 in GD2.CAR.15-Ts is not related to a different activation status of these cells, but rather suggests their potential superior fitness upon multiple exposures to the antigen. The low expression of PD-1 in GD2.CAR.15-Ts correlated in vivo with lack of additive effects with PD-1 blockade in controlling the tumor growth. Of note, in our xenograft model of neuroblastoma, we also observed that PD-1 blockade did not improve the antitumor effects of GD2.CAR-Ts that express PD-1, and should be reinvigorated by PD-1 blockade. This outcome may be partially related to the murine model that does not allow sufficient time to establish T-cell exhaustion. However, our data resemble the lack of activity we have previously reported in a pilot clinical trial in patients with neuroblastoma (4). Additional studies are needed to clearly assess if PD-1/PD-L1 blockade can have a positive effect when combined with GD2.CAR-Ts in neuroblastoma. Nevertheless, our data indicate that GD2.CAR-Ts expressing IL15 remain effective even in a model in which we caused constitutive expression of PD-L1 in neuroblastoma cells.

Because IL15 is a T-cell growth cytokine, and GD2.CAR.15-Ts have more “stem-ness” properties, we incorporated a safety switch for regulation of potential unwanted and uncontrolled expansion (16, 17, 40, 41). Our data demonstrate that further CAR construction was required to functionally accommodate these three genes. However, upon optimization, the final vector cassette showed functionality in vivo, and we were able to scale up the production of the CAR-Ts for clinical use. This final step is important because complex CAR-T engineering strategies often fail to reach clinical applications due to difficulties in scaling up at GMP level the results obtained in preclinical experiments.

In conclusion, we demonstrate that GD2.CAR.15-Ts have functional properties that should provide enhanced antigen-independent persistence and improved antitumor activity. The clinical trial in patients with neuroblastoma based on the proposed strategy will be performed in our institution (IND 18500) and if proven safe and effective, it can have broader application for targeting solid tumors.

G. Dotti reports receiving commercial research grants from Cell Medica and Bellicum Pharmaceutical and is a consultant/advisory board member for MolMed. No potential conflicts of interest were disclosed by the other authors.

Conception and design: Y. Chen, G. Dotti, B. Savoldo

Development of methodology: Y. Chen, L. Metelitsa, G. Dotti, B. Savoldo

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Chen, C. Sun, E. Landoni, G. Dotti

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y. Chen, C. Sun, G. Dotti, B. Savoldo

Writing, review, and/or revision of the manuscript: Y. Chen, L. Metelitsa, G. Dotti, B. Savoldo

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y. Chen, G. Dotti

Study supervision: Y. Chen, G. Dotti, B. Savoldo

This work was supported by UNC UCRF funds and in part by the Barnhill Family Foundation. Dr. Savoldo is supported by a NHLBI grant (R01HL114564) and a Hyundai Hope on Wheels Foundation grant.

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.