Engineering the TGFβ Receptor to Enhance the Therapeutic Potential of Natural Killer Cells as an Immunotherapy for Neuroblastoma Free

Infusions of allogeneic effector cells are used with increasing success in cancer immunotherapy (1–4). Natural killer (NK) cells have innate immunity, rapidly lysing target cells without prior priming (5, 6), and are often used as agents of cell-based immunotherapy. Cell lysis is permitted when killer immunoglobulin-like receptors on the NK-cell fail to engage with their cognate HLA molecule on the target cell. Antitumor cytolysis is triggered through the “missing self” mechanism when NK cells engage with tumor cells lacking surface HLA molecules. Although NK cells are critical for successful antitumor activity in early oncogenesis (7), their effect is reduced in advanced disease, in tumors that upregulate HLA molecules to evade NK-based surveillance, and in those which produce immunosuppressive cytokines such as IL6, IL10, and TGFβ (8). As such, there is a clear need to improve the therapeutic activity of NK cells against solid tumors.

Many solid tumors, including neuroblastoma, evade immune control by generating a suppressive tumor microenvironment, dominated largely by the cytokine TGFβ (8–10). Secretion of TGFβ is a potent immunosuppressive strategy employed by neuroblastoma, which inhibits immune effector cell–mediated cytotoxicity and promotes a protumorigenic environment (8, 9, 11, 12). This highly tumor-protective environment can only be overcome by significant immune modulation (13), which is why outcomes for advanced relapsed/refractory neuroblastoma is dismal and the development of immunotherapy strategies to overcome tumor evasion is urgently needed.

Adoptive cell therapy using NK cells has clear but limited clinical efficacy in the treatment of neuroblastoma (14), and the critical role of TGFβ provides compelling rationale for neutralizing this cytokine as a means to improving the efficacy of NK-cell–based therapeutics. Soluble TGFβ interacts with the TGFβRII and TGFβRI heterodimer complex on the surface of NK cells. This interaction leads to the phosphorylation of Smads 2 and 3 (15), which recruits other soluble proteins and triggers a regulatory cascade leading to decreased cytolytic function and impaired expression of activating receptors (16). Specifically, TGFβ has a detrimental effect on the innate expression of NK-cell–activating receptors NKG2D (8, 17) and DNAM1 (18). Therefore, we hypothesized that modifying the TGFβ receptor in NK cells to abrogate or subvert immunosuppressive signaling could allow NK cells to maintain therapeutic efficacy in the presence of this suppressive cytokine.

Several groups, including our own, have demonstrated that targeting the TGFβ pathway by arming immune effectors with a dominant negative receptor (DNR) can enhance effector function in a TGFβ-rich environment, with clear superiority over unmodified cells (19–23). In this report, we extended this approach by developing two novel variant TGFβ receptors that couple the TGFβ dominant negative receptor to intracellular signaling domains, mediating NK-cell activation. One construct, “NKA,” contains the truncated TGFβRII domain fused to the DNAX-activation protein 12 (DAP12) NK activation motif, which initiates signaling through its single immunoreceptor tyrosine-based activation motif (ITAM), with the aim of enhancing NK-cell activation (24, 25). The other construct, “NKCT” contains the truncated TGFβRII domain fused to a synthetic Notch-like receptor (“synNotch”; refs. 26, 27) coupled to RELA, to initiate NK-cell activation directly at a transcriptional level. These innovative approaches to hijack the TGFβ receptor and target TGFβ in the tumor microenvironment allows for NK cells to simultaneously (1) resist the immune suppression in the microenvironment, and (2) initiate activation to increase their ability to kill target tumor cells (Fig. 1).

Initial clinical efforts with NK cells as agents of adoptive immunotherapy have used NK-cell lines (28, 29), autologous ex vivo expansion (30), or allogeneic peripheral blood (31, 32) as a cell source. In vitro studies suggest that umbilical cord blood (UCB)-derived NK cells may be more advantageous (33). With over 500,000 validated banked UCB units worldwide (34), in addition to a constant supply of fresh cells, UCB represents a practical and readily available source for generating banks of “off-the-shelf” cell products. The ability to select optimally mismatched donor–recipient pairs to enhance cytotoxicity contributes to the practical and functional appeal of UCB as a source of cells for adoptive NK-cell immunotherapy (33, 35, 36).

Here, we demonstrate robust generation of gene-modified NK cells from UCB, which resisted the suppressive effects of tumor-associated TGFβ and exhibited enhanced antitumor effects in vitro and in vivo. This strategy using allogeneic UCB-derived NK cells genetically modified to resist suppression in the tumor microenvironment is therefore a potentially new treatment modality for patients with neuroblastoma and other malignancies that are amenable to NK-cell attack and utilize TGFβ secretion as a potent immune evasion mechanism.

Umbilical cord blood mononuclear cells were harvested from fresh cord blood units obtained from MD Anderson Cancer Center (Houston, TX) under approved Institutional review board–approved protocols (Pro00003896) by density gradient separation, and NK cells were isolated by negative selection with the EasySep Human NK Cell Isolation Kit (StemCell Technologies). Cord blood units were obtained under informed written consent and in accordance to the Declaration of Helsinki and the guidelines of the Institutional Review Board at MDACC (Houston, TX). After 24 hours of activation with 10 ng/mL of human IL15 (R&D Systems), NK cells were stimulated with K562 feeder cells, modified to express membrane-bound IL15 and 41BBL (refs. 31, 37; generously obtained from Baylor College of Medicine, Houston, TX; Pro00003869), irradiated at 200 Gy and cultured with NK cells at a 2:1 K562:NK-cell ratio. NK cells were expanded in Stem Cell Growth Medium (CellGenix) supplemented with 200 IU/mL human IL2, 15 ng/mL human IL15, 10% FBS (Gibco, Thermo Fisher Scientific), and 1% Glutamax (Gibco, Thermo Fisher Scientific). NK cells were isolated from 30 total cord blood donors for downstream use, and untransduced and transduced cells were generated from each individual donor line. Sample size (number of donor-derived lines) used for each experiment is specified in each figure legend. Modified and unmodified K562 cell lines were cultured with Iscove's modified Dulbecco's medium (Thermo Fisher Scientific) supplemented with 10% FBS (Gibco, Thermo Fisher Scientific), 1% penicillin–streptomycin, and 1% Glutamax (Gibco, Thermo Fisher Scientific). Neuroblastoma line SHSY5Y was purchased from ATCC and grown in a 1:1 medium of DMEM and F12K medium supplemented with 10% FBS (Gibco, Thermo Fisher Scientific), and 1% Glutamax (Gibco, Thermo Fisher Scientific). We performed HLA and short tandem repeat profiling to verify the identity and type of the SHSY5Y tumor line (Genetica Cell Line Testing). We also verified that the SHSY5Y neuroblastoma line produces high levels of TGFβ in vivo from SHSY5Y-inoculated NSG mice, and expresses low levels of MHC class I molecules (Supplementary Fig. S1). For generating the bioluminescent neuroblastoma line used in vivo, SHSY5Y was transduced with 2.5 × 10⁶ CFU of CMV-Firefly-luciferase-puro-resistant (Cellomics Technology) as per manufacturer's protocol. Bioluminescence was assessed with the Pierce Luciferase Dual Assay Kit (Thermo Fisher Scientific) and positive clones isolated by puromycin resistance and expanded for use, and the cell line was identified as SHSY5Y-luc. Identical in vitro experiments were performed with the neuroblastoma line HTLA230, purchased from ATCC.

Three modified plasmids were constructed as follows (Fig. 2A): (I) RBDNR: human type II TGFβ receptor cDNA was truncated at nt597 as described previously (38) and coupled to a truncated CD19 tag and pac puromycin-resistant gene via T2A sequences. (ii) NKA: human type II TGFβ receptor cDNA was truncated at nt597 as described previously (38), containing extracellular and transmembrane moieties, and coupled to the transmembrane and intracellular coding region of DAP12 as derived from full-length DAP12 cDNA (39), a truncated CD19 tag and a pac puromycin-resistant gene via T2A sequences. (iii) NKCT: human type II TGFβ receptor cDNA was truncated at nt597 as described previously (38) and coupled to a “SynNotch” receptor (26) composed of the Notch1 minimal regulatory region fused to the DNA binding domain for RELA (p65) and a VP64 effector domain (40), coupled to a truncated CD19 tag and a pac puromycin-resistant gene via T2A sequences. The RBDNR, NKA, and NKCT constructs were then individually integrated at the BamHI and NcoI sites of the retroviral vector SFG to generate plasmids of the same name. A control GFP-containing plasmid was generated elsewhere (41). Phoenix-ecotropic cells (ATCCs) were transfected with SFG:RBDNR, SFG:NKA, and SFG:NKCT, with Lipofectamine 2000 (Thermo Fisher Scientific) reagents used as per manufacturer's protocol. Transient retroviral supernatant was collected 48 and 72 hours after transfection and was used to transduce the PG13-stable packaging cell line (ATCC). Transduced PG13 cells were evaluated for transduction efficiency as described below, and single-cell FACS sorting was performed to isolate single clonally–derived producer lines for RBDNR, NKA, and NKCT constructs. For FACS sorting, single cells that expressed high levels of CD19 and TGFβRII expression were isolated with the Becton Dickinson Influx Cell Sorter (BD Biosciences) and selectively expanded in puromycin-containing DMEM with 10% FBS (Gibco, Thermo Fisher Scientific) and 1% Glutamax (Gibco, Thermo Fisher Scientific). Retroviral supernatants containing RBDNR, NKA, and NKCT constructs were harvested from subconfluent PG13 cells, passed through a 0.45-μm filter, and stored at −80°C until needed for transduction.

Activated NK cells were plated on retronectin-coated nontissue culture–treated plates (Takara) and transduced with RBDNR, NKA, or NKCT-containing retroviral supernatant in the presence of IL2 (200 IU/mL). After transductions, NK cells were assessed for transduction efficiency by staining with antibodies against CD19 conjugated to allophycocyanin (BD Biosciences) and TGFβRII conjugated to phycoerythrin (R&D Systems). After transduction, NK cells were expanded with additional stimulations with irradiated modified K562s, as described above, and exogenous IL2 and IL15. To enrich for phenotypic, functional, and in vivo assays, transduced NK cells were stained with CD19 microbeads (Miltenyi Biotec), and enriched by positive immunomagnetic bead selection according to the manufacturer's protocol.

NK cells were harvested from 21- or 28-day cultures, washed with FACS buffer, and incubated with human FcR Blocking Reagent for 10 minutes (Miltenyi Biotec). 21-day cultures were used for analysis of NK-cell molecular signaling, whereas 28-day cultures were used for all other endpoint NK-cell assays including phenotype, cytotoxicity, and in vivo applications, to allow for maximal cell expansion. Unmodified and modified NK cells, or cell lines, were stained with antibodies specific for NKp30, NKG2D, NKp44, CD16, PD1, CD56, CD3, DNAM1, CD19, TGFβRII (R&D Systems), HLA-ABC, or MICA/B. Antibodies were conjugated to FITC, PE, PerCP, APC, APC-Cy7, Pe-Cy7, or PerCP-Cy5.5 (BD Biosciences, unless otherwise identified). Samples were run on the Accuri C6 (BD Biosciences) or CytoFLEX S (Beckman Coulter) flow cytometers and analysis conducted using Flow Jo 7.6.5 (FlowJo LLC). To assess the cytokine profile of transduced and untransduced NK cells, cell supernatant was harvested from 21/28-day NK cultures and used in the Bio-Plex Human Cytokine 17-plex Assay according to the manufacturer's instructions (Bio-Rad Laboratories). For examination of cellular proliferation at endpoint, NK cells were labeled with carboxyfluorescein succinimidyl ester (CFSE) as per manufacturer's protocol (Thermo Fisher Scientific) and cocultured with modified K562 cells for 72 hours after assay establishment. To determine the cytolytic properties of unmodified and modified NK cells in various conditions, standard ⁵¹Cr release cytotoxicity assays were performed as described elsewhere (22). NK cells were incubated with ⁵¹Cr-labeled target cells (unmodified K562s, SHSY5Y cell lines–loaded with 10 μCi ⁵¹Cr per 10,000 cells) at 40:1, 20:1, 10:1, and 5:1 ratios for 5 hours in triplicate, and percent killing was determined by the following formula: (experimental count – spontaneous count)/(maximum count – spontaneous count) × 100%. For phenotypic and functional assessment of NK cells after exposure to TGFβ, NK cells were cultured with 10 ng/mL TGFβ (activated with 4 mmol/L HCl) added every other day. Five days after assay establishment, NK cells were isolated and examined by flow cytometry, multiplex assays, or cytotoxicity assays, as described above. Further details of NK-cell culture can be found in the Supplementary Data.

To examine the molecular effects of TGFβ, unmodified and modified NK cells (from 21-day cultures) were cultured with 10 ng/mL TGFβ (activated with 4 mmol/L HCl) at 37°C. At 30 minutes, 1, 3, 24, 48, and 72 hours post-TGFβ addition protein was isolated for molecular assessment. Briefly, unmodified or modified NK cells were pelleted and resuspended in RIPA lysis buffer (Thermo Fisher Scientific) containing protease inhibitor and phosphatase inhibitor cocktails (Roche Diagnostics). After 10-minute incubation at 4°C, protein was isolated and particulate matter removed by filtration with Ultrafree-CL centrifugal filter units (EMD Millipore). Protein was quantified with a Pierce BCA Protein Assay Kit (Thermo Fisher Scientific) and 30 μg of protein lysate was isolated and used in the TGFβ Signaling Pathway Magnetic Bead 6-plex Cell Signaling Multiplex Assay (EMD Millipore) as per manufacturer's instructions. Protein expression of phospho-Akt (Ser473), phospho-ERK (Thr185/Tyr187), phospho-Smad2 (Ser465/467), and phospho-Smad3 (Ser423/425) was quantitated with Luminex xMap detection, based on positive and negative quantified protein controls.

Male and female NSG (NOD.Cg-Prkdc^scidIl2rg^tm1Wjl/SzJ) mice were purchased from Jackson Laboratories and bred in-house in accordance with approved protocols with the Institutional Animal Care and Use Committee at Children's National Health System (Washington, D.C.). For in vivo neuroblastoma treatment experiments, 6- to 10-week-old male and female mice were preconditioned with sublethal irradiation (300 cGy) and inoculated with 2.5 × 10⁶ SHSY5Y-luc cells, administered subcutaneously in the dorsal flank of animals. This sublethal irradiation was performed at doses similar to that reported by other groups, which has verified successful immune depletion and immune engraftment in these models (42–45).

Animals were treated immediately following inoculation, a model commonly used in the field (43), with systemic administration of 1.5 × 10⁷ unmodified or modified NK cells via tail veins. For long-term studies, animals received weekly doses of 5–10 × 10⁶ unmodified or modified NK cells, administered systemically (five doses in total). All mice were treated with 0.2-μg human IL2, administered intraperitoneally every other day over the course of their cell therapy doses. The SHSY5Y neuroblastoma line was specifically chosen over the HTLA230 neuroblastoma line due its superior production of TGFβ both in vitro and in vivo in preliminary xenograft experiments (Supplementary Fig. S2). In addition, The SHSY5Y neuroblastoma line derives from the SK-N-SH line originating from a 4-year-old neuroblastoma patient and is a well-established neuroblastoma line used in the field and published in other immunotherapy studies (46–50). For examination of tumor progression, animals were imaged every other day with the IVIS Lumina 100 (PerkinElmer), and images were scaled to the same minimum and maximum photon distribution prior to analysis. Animals were injected with 150 mg/kg Xeno-Light d-Luciferin (PerkinElmer) 10 minutes prior to imaging with the IVIS, during which time animals were anesthetized with 2% isoflurane. Bioluminescent images were captured with 15-second exposure, with small binning and f-stop 2, and total bioluminescence was quantified by photon counts under individual murine regions of interest (photon counts). For analysis of NK-cell persistence, blood was collected at designated time points from submandibular veins with Goldenrod Animal Lancets (Braintree Scientific Inc.) and stored in K2EDTA-containing Microtainer tubes (BD Biosciences) at −80°C.

Transduced NK cells were detected and quantified in the peripheral blood using digital droplet PCR (ddPCR) methods. RNA was extracted from collected blood using the Whole Blood Quick-RNA Kit according to the manufacturer's instructions (Zymo Research). cDNA was prepared from 2,000 ng of isolated RNA by performing PCR amplification with RT buffer, dNTP Mix, MultiScribe RT, RNAse inhibitor, random primers, and nuclease-free water according to the High Capacity RT cDNA Kit (Thermo Fisher Scientific) and samples were run with the BioRad QC200 Droplet system according to manufacturer's protocols (Bio-Rad Laboratories Inc.). For identification of NK cells, primers specific to GFP, RBDNR, NKA, and NKCT construct were used, as described in the Supplementary Data and Methods.

All experiments were performed in duplicate or triplicate, with sample sizes indicated in each corresponding figure legend. Data were analyzed using GraphPad Prism software (GraphPad), and across all figures the solid color bars indicate non-TGFβ–treated groups, whereas striped bars indicate TGFβ-treated groups. Comparisons between untransduced, RBDNR, NKA, and NKCT data were performed using Student t test or χ² tests, with P < 0.05 considered as significant and denoted with an asterisk (*) and P < 0.0001 denoted with a two asterisks (**), unless otherwise noted. For in vivo experiments, we performed the log-rank (Mantel–Cox) test for Kaplan–Meier–generated survival data, with P < 0.05 considered as significant. Schematic signaling diagrams were generated using Biorender.

To examine NK-cell phenotype and function following genetic modification of the TGFβ receptor, cord blood–derived NK cells (33, 34, 36) were isolated and stimulated with irradiated feeder cells and supplemented with recombinant human IL2 and IL15 (31, 37). Four days after stimulation, NK cells were divided in to four groups: untransduced (UT), RBDNR-transduced, NKA-transduced, and NKCT-transduced NK cells as described (Fig. 2A). Cord blood–derived NK cells were successfully transduced with RBDNR, NKA, or NKCT variant TGFβ receptors, as indicated by surface staining of TGFβRII and truncated CD19, which was included in receptor design for identification and selection (TGFβRII+CD19+: UT 1.92% ± 2.64% vs. RBDNR 43.9% ± 24.1% vs. NKA 43.2% ± 27.1% vs. NKCT 39.1% ± 26.3%, CD19+: UT 1.86% ± 3.57% vs. RBDNR 42.6% ± 27.6% vs. NKA 43.9% ± 30.2% vs. NKCT 36.9% ± 29.4%, n > 30; Fig. 2B). Transduced NK cells could be enriched by performing immunomagnetic sorting with CD19 microbeads to achieve >90% enrichment (Supplementary Fig. S3). Staining for natural cytotoxicity receptors NKp44 and NKp30 showed no significant difference in expression on transduced NK cells compared with their untransduced counterparts (NKp44: UT 27.4% ± 15.6% vs. RBDNR 25.1% ± 18.0% vs. NKA 31.9% ± 14.9% vs. NKCT 26.4% ± 18.2% P > 0.05, NKp30: UT 41.1% ± 27.7% vs. RBDNR 44.2% ± 28.9% vs. NKA 41.7% ± 26.5% vs. NKCT 41.9% ± 31.4% P > 0.05, n > 5; Fig. 2C; Supplementary Fig. S4). Similarly, no impairment in the expression of other NK-cell surface markers NKG2D, CD69, CD16, or PD1 was found (P > 0.05, n > 5; Fig. 2C; Supplementary Fig. S4). NK cells were labeled with CFSE and cocultured with unlabeled modified K562s. Flow cytometric analysis of CFSE dilution over three days demonstrated no changes in NK-cell proliferation after transduction with RBDNR, NKA, or NKCT receptors (fold change compared with unstimulated; UT 75.3-fold vs. RBDNR 88.5-fold vs. NKA 41.3-fold vs. NKCT 64.2-fold, P > 0.05, n > 5; Fig. 2D; Supplementary Fig. S4). ⁵¹Cr-based cytotoxicity assays with untransduced and transduced NK cells showed maintained cytolysis of K562 target cells in all conditions (UT vs. RBDNR vs. NKA vs. NKCT P > 0.05, n > 5; Fig. 2E). Additional cytotoxicity assays with untransduced and transduced cells showed maintained cytolysis of HTLA230 neuroblastoma target cells in all conditions (UT vs. RBDNR vs. NKA vs. NKCT P > 0.05, n > 5; Supplementary Fig. S2). These results showed that introducing an engineered TGFβ receptor for any of the RBDNR, NKA, or NKCT constructs did not affect NK-cell phenotype and function.

TGFβ binding initiates the phosphorylation of intracellular Smad2 and Smad3 proteins (15). To investigate the ability of RBDNR, NKA, and NKCT constructs to prevent TGFβ-mediated signaling, we cocultured untransduced, RBDNR, NKA, and NKCT-transduced NK cells with TGFβ. Cells were harvested 0.5, 1, or 3 hours after TGFβ exposure, and either assayed by flow cytometry or lysed to isolate and characterize intracellular proteins. Flow cytometry demonstrated rapid phosphorylation (Ser465/467) of Smad2/3 when untransduced NK cells were exposed to TGFβ (pSmad2/3: UT 1.36 ± 0.95% vs. UT+TGFβ UT 73.9 ± 20.5%, P = 0.04 at 1 hour, n > 3; Fig. 3A), but not in NK cells transduced with either RBDNR, NKA, or NKCT receptors following TGFβ exposure (P > 0.05 at 1 hour, P > 0.05 at 3 hours, n > 3; Fig. 3A). Similarly, evaluation of Smad2 (Ser465/467) and Smad3 (Ser423/425) phosphorylation from protein lysate isolated from untransduced and transduced cells after 1 hour of TGFβ exposure further demonstrated the protective effect of TGFβ receptor modifications conferred to NK cells. Protein lysate results are shown from one representative NK line (Fig. 3B) as well as from pooled NK donor lines (pSmad2 UT+TGFβ vs. RBDNR+TGFβ P = 0.025, UT+TGFβ vs. NKCT+TGFβ P = 0.031; pSmad3 UT+TGFβ vs. RBDNR+TGFβ P = 0.037, n > 5; Fig. 3C). These results demonstrated that Smad2 was only phosphorylated in UT NK cells exposed to TGFβ, while expression of the RBDNR, NKA, or NKCT receptors protected from Smad2 phosphorylation.

To assess whether the protection from the molecular changes occurring after TGFβ exposure translated to a phenotypic or functional advantage, untransduced, and RBDNR, NKA, and NKCT-transduced NK cells were examined after 5 days in culture with TGFβ. Flow cytometry showed decreased expression of DNAX Accessory Molecule-1 (DNAM1 fold change from non-TGFβ exposed: UT 0.39-fold, P = 0.0163, n > 5; Fig. 4A) and in NKG2D (fold-change from non-TGFβ exposed: UT 0.58-fold, P = 0.04, n > 5; Fig. 4A) in untransduced NK cells following exposure to TGFβ. Surface marker downregulation was not observed in RBDNR, NKA, or NKCT-transduced NK cells, which all exhibited protection from these TGFβ-mediated phenotype impairments (P > 0.05, n > 5; Fig. 4A). In addition, expression of CD16 was not impaired in transduced cells following TGFβ exposure, alluding to their potential to successfully mediate an antitumor effect via ADCC as well as cytolysis (Supplementary Fig. S5). Indeed, whereas untransduced NK cells showed dose-dependent cytotoxicity against SHSY5Y neuroblastoma cells (38.2% ± 4.69% killing at E:T ratio 40:1), they exhibited impaired cytolytic activity (24.6% ± 4.58% killing at E:T ratio 40:1) following preculture with TGFβ (Fig. 4B and C). Impaired cytolytic activity was not demonstrated when NK cells transduced to express the variant TGFβ receptors (RBDNR, NKA, or NKCT) were evaluated following pretreatment with TGFβ (Fig. 4B and C), suggesting their functional superiority at killing target cells in a TGFβ-rich environment. As such, we found that not only did expression of the modified TGFβ receptors protect from the molecular signaling occurring in endogenous NK cells following TGFβ exposure, but this protection translated to a protection from altered phenotype and decreased antitumor activity occurring in untransduced cells exposed to TGFβ.

To examine the induction of NK-cell activation, we cocultured untransduced, RBDNR, NKA, and NKCT-transduced NK cells with TGFβ. Cells were harvested 0.5, 1, or 3 hours after TGFβ exposure and either lysed to isolate protein or assayed by flow cytometry. Using flow cytometry, we demonstrated decreasing levels of RELA (p65) in untransduced NK cells at 1 and 3 hours post-TGFβ exposure (UT 42.3% ± 13.7% vs. UT+TGFβ UT 2.02% ± 1.08%, P = 0.02 at 1 hour UT 21.5% ± 11.5% vs. UT+TGFβ UT 0.47% ± 0.46%, P = 0.18 at 3 hours, n > 3; Fig. 5A and B). Similar trends in RELA were seen in RBDNR-transduced NK cells at 1-hour post-TGFβ exposure (P = 0.31 at 1 hour, P = 0.18 at 3 hours, n > 3; Fig. 5A and B) NK cells transduced with either NKA or NKCT-variant TGFβ receptors demonstrated unaltered p65 expression following exposure to TGFβ (NKA P = 0.92 at 1 hour, P = 0.61 and 3 hours, n > 3; NKCT P = 0.96 at 1 hour, P = 0.75 at 3 hours, n > 3), suggesting that NFκB-mediated signaling persisted in these cells. Evaluation of ERK1/2 (Thr185/Tyr187) and Akt (Ser473) phosphorylation occurring in protein lysate isolated from untransduced and transduced cells after 1 hour of TGFβ exposure further showed activation in NKA and NKCT-transduced NK cells. While untransduced and RBDNR-transduced NK cells exhibited decreased or unchanged levels of Akt phosphorylation (UT vs. UT+TGFβ P = 0.0075, RBDNR vs. RBDNR+TGFβ P = 0.282, n > 5; Fig. 5C), NK cells equipped with the activation-inducing TGFβ variants had increased Akt phosphorylation (NKA vs. NKA+TGFβ P = 0.0013, NKCT vs. NKCT+TGFβ P = 0.0037, n > 5; Fig. 5C). In an examination of supernatant isolated from cell cultures after 12 hours of exposure to TGFβ, we found significantly increased TNFα production in NKA-transduced NK cells after cytokine exposure, as compared with either untransduced or other variant transduced NK-cell groups (NKA+TGFβ vs. UT+TGFβ P = 0.039, NKA+TGFβ vs. RBDNR+TGFβ P = 0.006, NKA+TGFβ vs. NKCT+TGFβ P = 0.041; Fig. 5D). Taken together, these results suggest that NK cells transduced to express the TGFβ receptor variants, in particular the NKA-modified receptor, demonstrated heightened NK activation, consistent with our observed molecular changes occurring along the NFκB and PI3K signaling pathways.

We established a xenograft model of human neuroblastoma using SHSY5Y human neuroblastoma cells (51), inoculated subcutaneously in preconditioned immunodeficient animals. Animals were randomly assigned to six treatment groups: untreated, untransduced NK cells (UT), mock GFP-transduced NK cells (Mock-Tdx), RBDNR-transduced NK cells (RBDNR), NKA-transduced NK cells (NKA), and NKCT-transduced NK cells (NKCT). After inoculation, animals were treated systemically (43) with 1.5 × 10⁷ NK cells, and monitored during alternate day intraperitoneal IL2 administration for the duration of the study. Repeated doses of untransduced or transduced NK cells were subsequently given on days 0, 7, 14, 21, and 28 following tumor inoculation (Fig. 6A), which mirrors desired clinical dosing regimens. Tumor growth was monitored every other day by quantifying bioluminescence (total photon counts) of animals imaged with the IVIS system, using a normalized photon scale (52, 53). Rapid tumor progression was seen in untreated animals, who had a median survival of 31 days (Fig. 6B and C; Supplementary Fig. S6). Animals infused with untransduced or mock-transduced NK cells showed delayed tumor progression compared with untreated animals; however, these animals eventually succumbed to tumor progression (UT median survival = 43 days, Mock-Tdx median survival = 48.5 days; Fig. 6B and C). In contrast, infusion of RBDNR or NKCT-transduced NK cells led to improved tumor control and prolonged survival (RBDNR median survival = 88 days, NKCT median survival = 65 days; survival untreated vs. RBDNR P < 0.0001, untreated vs. NKCT P < 0.0001; Fig. 6D). Animals treated with NKA-transduced NK cells exhibited superior protection from tumor progression (Fig. 6B and C; Supplementary Fig. S6) and significantly enhanced survival (progression-free survival = 72.9%, survival untreated vs. NKA P < 0.0001, UT vs. NKA P = 0.0001, RBDNR vs. NKA P = 0.0333, NKCT vs. NKA P = 0.0313; Fig. 6D). In an assessment to determine the immune populations in the peripheral blood of mice using flow cytometry, we showed that NK cells represented a very minor (<1%) population of the total lymphoid compartment (Supplementary Fig. S7), and as such, the more sensitive ddPCR assay was used to identify the presence of unmodified or modified NK cells peripherally. Therefore, peripheral blood was isolated weekly following the final therapeutic dose of NK cells on day 28, and RNA was extracted from the blood to evaluate the presence of the NK-cell transgene (GFP or TGFβ variant receptor) by quantitative ddPCR assay. At 5 and 9 days after the final infusion, modified NK cells were identified in circulation. Over the next 6 weeks, there was some evidence of RBDNR and NKCT-transduced NK cells persisting, although in progressively dwindling numbers as time continued and tumors progressed (Fig. 6E; Supplementary Table S1). NKA-transduced NK cells, however, persisted in higher frequencies than either RBDNR or NKCT-transduced NK cells (Fig. 6E; Supplementary Table S1). Analysis of the TBP transgene in all samples ensured a sufficient quantity and quality of DNA, and was used to normalize all results.

Taken together, these data indicate that NK cells modified to express novel variants of a TGFβ receptor protect cells from the inhibitory effects of neuroblastoma-associated TGFβ and demonstrate superior antitumor efficacy in vivo. Furthermore, the enhanced persistence of NKA-transduced NK cells and the significant improvement in progression-free survival in mice administered NKA-transduced NK cells over the RBDNR- and NKCT-transduced NK-cell products suggest that coupling the TGFβ receptor modification to the NK-specific signaling motif DAP12 confers additional therapeutic advantages and prolonged NK-cell persistence in vivo (Supplementary Fig. S6).

In this study, we genetically engineered NK cells with novel TGFβ receptors to counter any suppressive TGFβ-mediated signaling and investigated whether we could switch the negative TGFβ signal into an activating signal. We demonstrated that phosphorylation of Smad2 and Smad3 occurred as early as 30 minutes after TGFβ exposure in unmodified NK cells, but was blocked in RBDNR-, NKA-, and NKCT-transduced NK cells. The signaling cascade initiated by the phosphorylation of Smad2/3 led to impaired expression of surface receptors (54) and consequent impairment of antitumor cytolytic function. We found that not only were cord blood–modified NK cells resistant to the inhibitory effects of tumor-associated TGFβ–they also showed superior antitumor efficacy in a TGFβ-rich tumor setting, specifically when transduced with the NKA receptor. The strategy of rendering cell therapy products resistant to inhibitory TGFβ has been explored in a number of malignancies (19, 23). However, by fully inactivating the negative TGFβ pathway and converting the inhibitory signal to an ancillary signal, we created a novel and potent NK-cell–specific therapeutic which could be used as an allogeneic “off-the-shelf” cellular therapy for the treatment of patients with neuroblastoma.

Use of the synthetic Notch receptor into the NKCT receptor is a strategy conceptualized and first applied in the setting of chimeric antigen receptor generation for T cells (26, 27). This strategy employs logic gating, requiring the cell to receive a primary signal to trigger a secondary signal through a “SynNotch” receptor. The “SynNotch” receptor contains a core regulatory Notch domain, coupled to an intracellular transcriptional domain that cleaves and engages with nuclear promoters to initiate a given transcriptional change. The NKCT receptor used here contains the extracellular TGFβ dominant-negative receptor coupled to a Notch and RELA-linked domain; engagement of TGFβ with this receptor would trigger cleavage of the “SynNotch” motif leading to increased transcription of RELA (p65) and consequent increase in NK-cell activation. Our in vitro experiments with the NKCT construct validated this strategy for activating NK cells. However, the potential advantage of this construct was not borne out in vivo, as systemic treatment with NKCT-modified NK cells achieved antitumor efficacy and progression-free survival no better than achieved by RBDNR-modified NK cells that only block TGFβ-mediated signaling. The size of the construct might have been a limiting factor, impairing cleavage and translocation of the large intracellular signaling portion of this receptor. In addition, because the construct bypassed a natural signaling cascade instead of leading directly to transcriptional activation, it is possible that in the TGFβ-rich environment NKCT-transduced NK cells could be chronically activated causing NK-cell dysfunction and apoptosis. Alternatively, chronic activation could have generated a negative feedback loop from inhibitory cytokines (55).

In contrast, the NKA receptor (containing DAP12 fused to the dominant-negative receptor facilitating NK-specific intracellular signaling) led to improved activity in vivo. In unmodified NK cells, DAP12 associates with natural activating and cytotoxicity receptors, such as NKG2C and NKp44. Once dimerized, the ITAM-containing cytoplasmic domain can readily dock with Zap70 and Syk proteins. Global cell activation is the resultant effect of DAP12 activation, which signals through the PI3K/ERK and Akt pathways (25, 56–58). By incorporating the transmembrane and ITAM-containing domains of DAP12 in the NKA construct, TGFβ binding with the engineered receptor triggered activation of DAP12 signaling and enhanced the NK-cell activity. The antitumor efficacy of the NKA construct was superior to that obtained with NK cells engineered only to block TGFβ signaling, as in the RBDNR-engineered cells. Furthermore, this additional modification conferred a distinct survival advantage, with NKA-transduced cells persisting up to 7 weeks following their final infusion in treated animals. This in turn led to a superior antitumor effect and a survival advantage in these mice. Further assessment of activation markers expressed by NK cells isolated ex vivo from treated animals would allow a greater depth of understanding into the in vivo mechanism, and will be an important component of larger scale efforts as this approach is translated to the clinic. Although the enhanced PI3K/Akt signaling found in vitro indicates successful propagation of DAP12-mediated activation, it does not specifically address the mechanism through which the TGFβRII-DAP12–linked receptor is forming a dimer or tetramer, and the resultant signaling cascade. As such, it would be essential for future studies to further elucidate this signaling mechanism as well as examine other downstream molecular targets to ensure that enhanced Akt activity would not lead to artificially enhanced NK-cell exhaustion.

Topfer and colleagues have also incorporated DAP12 signaling into a prostate stem cell antigen (PSCA)-specific CAR construct. Preliminary results confirmed the benefit of the DAP12 construct over non-DAP12–containing CAR cells (39); however, this effort was conducted with the NK-cell line YTS, which, although similar to endogenous NK cells in phenotype, lacks the KIR expression resident to primary NK cells. Our efforts genetically modifying primary NK cells derived from cord blood sources represents a clinically relevant application, where interaction between inhibitory KIRs on NK cells with MHC I variants can have a large influence on the resultant activity (cytotoxicity or suppression) of NK cells used for cell therapy. Our modification of NK cells with a combination of enhanced cell activity through DAP12 and ameliorated TGFβ blockade represents a novel and promising cell therapy approach for neuroblastoma and other malignancies.

One drawback of using CD19 expression to identify transduced cells is that selective downregulation of either the TGFβ-modified receptor or the CD19 tag could occur. By using immunomagnetic beads to selectively enrich our cell populations, we minimized the likelihood of this happening (Supplementary Fig. S3). While engineered NK cells might downregulate the modified TGFβ receptor over time, our in vivo studies identified gene-modified NK cells with biological activity beyond four weeks suggesting that the cell constructs were stable and could exert long-term antitumor effects.

This report demonstrates preclinical efficacy of a novel mechanism to convert a customarily inhibitory signal, TGFβ, into an activating pathway for NK cells—by doing so, the TGFβ-rich tumor microenvironment is transformed to enhance NK-cell–mediated cytotoxicity of tumors. By generating NK-cell products from over 30 umbilical cord blood units, and through in vitro and in vivo testing in a human xenograft model of neuroblastoma, this report supports translation to clinical applications. Further preclinical work is being pursued to identify the potential mechanisms of escape that could be faced clinically. For example, examining the function of these variant TGFβ receptors in a humanized model would be of considerable future interest because humanized neuroblastoma models would provide the opportunity to examine interactions with other immune components (e.g., myeloid-derived suppressor cells) that may also play a role in promoting NK-cell dysfunction in the neuroblastoma setting. Furthermore, ex vivo profiling of immune subsets over time would allow for further in depth analysis of the interactions between NK cells and other immune effectors, and could help determine whether the gene engineered NK cells are capable of eliciting enhanced cytotoxicity through supporting ADCC in addition to tumor-targeted cytotoxicity. Although many neuroblastomas have decreased or absent levels of MHC I, rendering them attractive targets for NK-mediated cytolysis, it would also be of considerable interest in further studies to examine the efficacy of this NK-based immunotherapy in a tumor that has upregulated MHC I expression as a method of tumor escape. In such a setting, however, combining NK-cell therapy with other immunomodulatory agents (small molecule or epigenetic) may represent an attractive therapeutic avenue. Finally, another priority in further preclinical testing and in initial clinical readouts would be to determine the extent of NK-cell migration to tumor-draining lymph nodes and other biological niches following repeat NK-cell dosing. While preliminary efforts revealed that CCR2 expression is impaired in NK cells following exposure to TGFβ, and modification with the dominant negative receptor (and variants) may protect from this decline, further probing of the complete effect on NK cells migration is the subject of future study.

In summary, cord blood–derived NK cells modified to avoid the inhibitory effects of TGFβ represent an efficient way to harness fast-acting innate immune cells for therapy. Furthermore, our development of novel variant TGFβ receptors, composed of the dominant-negative receptor coupled to intracellular signaling domains initiating NK-cell activation, represents a unique approach to transform a classical tumor-inhibitory mechanism into a therapeutic weapon. Our preclinical results support translational research to establish allogeneic, cord blood–derived, gene-modified NK cells to treat patients with neuroblastoma and other malignancies that use TGFβ secretion as a potent immune evasion mechanism.

C.R. Cruz is an employee of and holds ownership interest (including patents) in Mana Therapeutics, reports receiving speakers bureau honoraria from Georgetown University, and other remuneration via institutional support from Torque Pharmaceuticals. C. Bollard holds ownership interest (including patents) in Mana Therapeutics, Torque Therapeutics, Neximmune, and Cabaletta Bio, and is a consultant/advisory board member for Cellectis. No potential conflicts of interest were disclosed by the other authors.

Conception and design: R.A. Burga, R. Fernandes, C.R. Cruz, C.M. Bollard

Development of methodology: R.A. Burga, R. Fernandes, C.R. Cruz, C.M. Bollard

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): R.A. Burga, E. Chorvinsky, R. Fernandes, C.M. Bollard

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): R.A. Burga, E. Yvon, R. Fernandes, C.R. Cruz, C.M. Bollard

Writing, review, and/or revision of the manuscript: R.A. Burga, E. Yvon, R. Fernandes, C.R. Cruz, C.M. Bollard

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): R.A. Burga, E. Chorvinsky

Study supervision: C.M. Bollard

The authors would like to gratefully acknowledge the Institute for Biomedical Sciences at The George Washington University, where R.A. Burga is a doctoral candidate. This work is supported by funding to Dr. Catherine Bollard from the Department of Defense (award W81XWH-15-1-0334).

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.