TGFβR1 Blockade with Galunisertib (LY2157299) Enhances Anti-Neuroblastoma Activity of the Anti-GD2 Antibody Dinutuximab (ch14.18) with Natural Killer Cells.

PURPOSE
Immunotherapy of high-risk neuroblastoma using the anti-GD2 antibody dinutuximab induces antibody-dependent cell-mediated cytotoxicity (ADCC). Galunisertib, an inhibitor of TGFβR1, was examined for its ability to enhance the efficacy of dinutuximab in combination with human ex vivo activated NK (aNK) cells against neuroblastoma.


EXPERIMENTAL DESIGN
TGFB1 and TGFBR1 mRNA expression was determined for 249 primary neuroblastoma tumors by microarray analysis. The ability of galunisertib to inhibit SMAD activity induced by neuroblastoma patient blood and bone marrow plasmas in neuroblastoma cells was tested. The impact of galunisertib on TGFβ1-induced inhibition of aNK cytotoxicity and ADCC in vitro and on anti-neuroblastoma activity in NOD-scid gamma (NSG) mice was determined.


RESULTS
Neuroblastomas express TGFB1 and TGFBR1 mRNA. Galunisertib suppressed SMAD activation in neuroblastoma cells induced by exogenous TGFβ1 or by patient blood and bone marrow plasma, and suppressed SMAD2 phosphorylation in human neuroblastoma cells growing in NSG mice. In NK cells treated in vitro with exogenous TGFβ1, galunisertib suppressed SMAD2 phosphorylation and restored the expression of DNAM-1, NKp30, and NKG2D cytotoxicity receptors and the TRAIL death ligand, the release of perforin and granzyme A, and the direct cytotoxicity and ADCC of aNK cells against neuroblastoma cells. Addition of galunisertib to adoptive cell therapy with aNK cells plus dinutuximab reduced tumor growth and increased survival of mice injected with two neuroblastoma cell lines or a patient-derived xenograft.


CONCLUSIONS
Galunisertib suppresses activation of SMAD2 in neuroblastomas and aNK cells, restores NK cytotoxic mechanisms, and increases the efficacy of dinutuximab with aNK cells against neuroblastoma tumors. Clin Cancer Res; 23(3); 804-13. ©2016 AACRSee related commentary by Zenarruzabeitia et al., p. 615.


Introduction
High-risk neuroblastoma accounts for a disproportionate burden of childhood cancer morbidity and mortality, representing 7% of childhood malignancies but accounting for 15% of all childhood cancer-related deaths (1). Although event-free survival for patients with high-risk neuroblastoma has improved with use of the anti-disialoganglioside (anti-GD2) chimeric mAb dinutuximab (ch14.18) plus IL2 and GM-CSF immunotherapy, 40% of patients still relapse during or after immunotherapy (2). The reasons for failure of this immunotherapy are not known.
Increasing evidence indicates that the tumor microenvironment (TME) supports tumor growth and survival and regulates immune responses (3). Within the TME, the TGFb family has an important role in tumor immune evasion, leading to tumor progression and metastasis (4,5). The clinical importance of TGFb1 in suppressing NK cells is indicated by two studies of patients with breast or squamous cell carcinoma (6,7). Among the TGFb family, TGFb1 is the most commonly upregulated in tumor cells (5,8). This ligand binds to TGFb receptor type I (TGFbR1), which results in its dimerization to TGFb receptor type II (TGFbR2). This heterodimer then phosphorylates SMAD2 and SMAD3, which complex with SMAD4 to modulate transcription of downstream genes (9,10). TGFb1 is known to inhibit the IFNg production, proliferation, and function of natural killer (NK) cells, an important type of immune effector cell expressing the antibody receptor FcgRIIIa (CD16) and mediating ADCC against neuroblastoma cells (11)(12)(13). The neuroblastoma TME can include TGFb1, and higher than median TGFb1 in MYCN nonamplified neuroblastoma patient tumors correlates with worse event-free survival (14).
Approaches for inhibiting TGFb-induced signaling include targeting ligand-receptor interactions and intracellular signaling (15). Galunisertib (LY2157299 monohydrate) is a recently developed small-molecule inhibitor of TGFbR1. Galunisertib binds antagonistically to TGFbR1, preventing the intracellular phosphorylation of SMAD2 and SMAD3 (16)(17)(18). This agent has demonstrated antitumor activity in combination with paclitaxel or sorafenib in xenograft models of breast or hepatocellular carcinoma (17)(18)(19). Phase I studies have demonstrated that galunisertib is safe in adult patients with advanced solid tumors (20,21). However, it is unknown whether galunisertib can augment anti-GD2 antibody therapy or the antitumor cytotoxicity of NK cells propagated and activated ex vivo with K562.mbIL21 artificial antigen-presenting cells (22)(23)(24), which we and others are using to generate activated NK (aNK) cells for evaluation in clinical trials of adoptive cell therapy (ClinicalTrials.gov identifiers: NCT01787474 and NCT02573896).
We demonstrate that galunisertib significantly restores the cytotoxicity of aNK cells following their inhibition by TGFb1 in vitro and enhances the combination of dinutuximab and aNK cell immunotherapy against neuroblastoma cell lines and a patientderived xenograft (PDX) growing in kidneys of NOD-scid gamma (NSG) mice. These findings support the clinical testing of galunisertib in combination with dinutuximab for the immunotherapy of neuroblastoma.

Materials and Methods
Neuroblastoma cells, patient specimens, aNK cells, and reagents CHLA-255 and CHLA-136 neuroblastoma cell lines were maintained in Iscove's modified Dulbecco's medium (IMDM) with 10% FBS. CHLA-255-Fluc and CHLA-136-Fluc cells were transduced with the firefly luciferase (Fluc) gene using a lentivirus vector (24). COG-N-415x PDX neuroblastoma cells expressing mutated ALK (F1174L) and amplified MYCN gene were kindly provided by Dr. C. Patrick Reynolds (Texas Tech University, Lubbock, Texas). The correct identity of cells was authenticated using the AmpFLSTR Identifiler PCR Amplification Kit (Applied Biosystems). Primary neuroblasto-ma tumors were obtained from patients enrolled and consented for Children's Oncology Group (COG) biology and therapeutic protocols. Plasma from whole blood and bone marrow aspirates were obtained from patients with relapsed and refractory neuroblastoma enrolled on the New Approaches to Neuroblastoma Therapy (NANT) Biology Study N2004-05.
Anti-GD2 chimeric monoclonal antibody (mAb) ch14.18/dinutuximab was provided by the National Cancer Institute (Frederick, MD). Human TGFb1 (R&D Systems) was reconstituted at 10 mg/mL in sterile 4 mmol/L HCl containing 0.1% BSA. Aliquots were kept at À80 C and discarded after 3 months. Galunisertib was provided by Eli Lilly and Company. For in vivo experiments, galunisertib was freshly suspended in a formulated vehicle (1% carboxymethylcellulose sodium salt, 0.5% SDS, 0.085% povidone, and 0.05% antifoam Y-30 emulsion) and kept at 4 C for up to 1 week. Galunisertib was dissolved in DMSO at 10 mmol/L and kept at À20 C as a stock solution for in vitro experiments.

Gene expression analysis
Affymetrix Human Exon Array data (manuscript in preparation; see https://ocg.cancer.gov/programs/target/research) of 249 primary neuroblastoma tumor specimens obtained at diagnosis was normalized by quantile normalization and summarized using robust multichip average (Affymetrix Power Tools software package version 1.12). This dataset includes samples from 219 patients with highrisk (68 with amplified and 151 with nonamplified MYCN) and 30 with low-risk primary tumors. The transcript level data of core probe sets for each sample were averaged on the basis of gene symbol annotations provided by the manufacturer (17,422 unique genes). To identify relative expression of genes in neuroblastomas, the  percentile values of TGFBR1, TGFB1, TGFBR2, TGFB2, TBX21, IFNG,  NTRK1, and MYCN were computed from the cumulative distribution function for each sample's gene profile. As an independent dataset, Agilent single-color expression profiles of 478 samples were downloaded from the GEO GSE16716 dataset. Patients with stage 4S disease in this latter dataset (n ¼ 62) were excluded from analysis to allow comparison with our Human Exon Array data. Expression profiles from the resulting cohort of 416 tumors from patients with high-risk (n ¼ 135), intermediate-risk (n ¼ 34), or low-risk (n ¼ 247) neuroblastoma were used to assess expression of TGFBR1, TGFB1, TGFBR2, TGFB2, TBX21, IFNG, as well as NTRK1 and MYCN as internal controls. Concordant results were obtained between our neuroblastoma dataset and the GEO GSE16716 dataset.

Neuroblastoma patient plasma SMAD activity assay
The Cignal Lenti SMAD Reporter, purchased from SA Biosciences, is comprised of lentiviral particles containing the firefly

Translational Relevance
Anticancer functions of natural killer (NK) cells may be suppressed in the tumor microenvironment. The cytokine TGFb1 can be a major effector of this suppression. Galunisertib, a small-molecule inhibitor of TGFbR1, enhances antitumor activity when combined with paclitaxel or sorafenib in xenograft models of breast or hepatocellular carcinoma and is being tested clinically. However, the effect of galunisertib on TGFb suppression of NK cell function has not been investigated. We demonstrate that galunisertib reverses TGFb1induced suppression of direct cytotoxicity and anti-GD2 antibody-dependent cell-mediated cytotoxicity of human ex vivo propagated and activated NK (aNK) cells against neuroblastoma cells in vitro. Furthermore, galunisertib enhances aNK adoptive cell therapy with the anti-GD2 monoclonal antibody dinutuximab against neuroblastoma cell lines and a neuroblastoma patient-derived xenograft growing in NSG mice. These data suggest that galunisertib may improve anti-GD2 antibody-based immunotherapy of neuroblastoma.
luciferase gene under the control of SMAD transcription response elements. CHLA-255 cells expressing a high level of Renilla luciferase (CHLA-255-hRL) were transduced by this SMAD reporter lentivirus, and the stable CHLA255hRL-SmadFluc cell line was established by puromycin selection. To validate detection of SMAD activity, 5 Â 10 4 CHLA255hRL-SmadFluc cells were seeded into each well of a 96-well plate overnight and then cultured with various doses of galunisertib with and without recombinant human TGFb1 (10 ng/mL). Beetle luciferin (7 mL, 5 mg/mL; Promega) was added to each well for 5 minutes in the dark. Activation of SMAD transcription response elements by TGFb1 or patient plasma resulted in expression of firefly luciferase, which when activated by luciferin, was measured as luminescence using the GloMax Multi-Detection System (Promega; model #E8032). To determine whether neuroblastoma patient blood or bone marrow plasma could activate SMAD2, samples from 17 patients with high-risk neuroblastoma (1:10 dilution with 1% FBS-IMDM) were added into wells with or without galunisertib for 18 or 36 hours.

Immunohistochemical staining of phospho-SMAD2 in human neuroblastoma tumors formed in NSG mice
Tumor tissues from kidneys of NSG mice in different treatment groups were placed into formalin (Fisher Scientific Company LLC) for 2 days and then into 75% ethanol at 4 C. Formalinfixed, paraffin-embedded sections in Leica BOND-MAX (Leica Microsystems) were heated for 30 minutes in Bond Epitope Retrieval Solution 2 (No. AR9640; Leica Biosystems Newcastle Ltd.). Sections then were incubated for 2 hours at room temperature with anti-phospho-SMAD2 antibody (Ser465/467; rabbit polyclonal antibody; EMD Millipore) at a dilution of 1:500, followed by a poly-horseradish peroxidase-conjugated goat anti-rabbit antibody (Leica). Kidney tissue of a normal NSG mouse was stained as a negative control.

Flow cytometry
Cell surface staining was performed, as previously described (24,25). Briefly, cells were washed twice in FACS buffer (PBS with 0.1% NaN3 and 0.1% BSA) and centrifuged for 10 minutes at 400 Â g. Antibodies listed in Supplementary Table S1 were added in the dark at 4 C using concentrations previously determined by titration. Isotype-matched irrelevant mAbs were used to define nonspecific staining. Cells were incubated at 4 C for 90 minutes and washed twice in FACS buffer. Dead cells were excluded according to positivity for DAPI. Cell aggregates were excluded from analysis by gating out events exhibiting high forward and side light scatter. Flow cytometry analysis was performed using a BD LSR II flow cytometer with DIVA software (BD Biosciences) and FCS Express software (DeNovo Software). The Stain Index was calculated using values of median fluorescence intensity (MFI) and robust SD (rSD) as follows: (MFI of viable aNK cells stained with specific antibody À MFI of viable aNK cells stained with an isotype-matched irrelevant antibody)/(2 Â rSD of the isotype control).

NK cytotoxicity assay
Frozen K562.mbIL21-expanded aNK cells (24) were cultured in 10% FBS-RPMI1640 with 100 U/mL of IL2 for 24 hours and then were pretreated in individual wells of 6-well plates for 48 hours with (i) 5 mmol/L galunisertib, (ii) 10 ng/mL human TGFb1, (iii) galunisertib for 30 minutes followed by TGFb1, or (iv) 24 hours with TGFb1 followed by treatment with galunisertib for an additional 24 hours. For the aNK cytotoxicity assay, neuroblastoma cell lines (CHLA-255-Fluc and CHLA-136-Fluc) were labeled with calcein-AM for 30 minutes and washed once (26). A total of 1 Â 10 4 -labeled CHLA-255-Fluc or CHLA-136-Fluc were seeded into individual wells of a 96-well plate, and 5 Â 10 3 aNK cells pretreated as described above were added to neuroblastoma cells at a ratio of 1:2. Dinutuximab was added to the indicated wells at a concentration of 1 mg/mL for the ADCC assay. The plate was incubated at 37 C in 5% CO 2 for 6 hours, and surviving tumor cells were quantified as calcein-containing cells using a digital imaging microscopy system (DimScan; ref. 26).

Luminex assay
Granzyme A and perforin were measured using a custom-plex bead array from EMD Millipore following the manufacturer's instructions with a Luminex-200 instrument (Luminex Corporation), as described previously (24).

Intrarenal neuroblastoma model and treatment of NSG mice
All in vivo experiments utilized 4-to 6-week-old male and female NSG (NOD.Cg-Prkdc scid Il2rg tm1Wjl /SzJ) mice, which were bred in-house, genotyped for colony maintenance, and housed in a pathogen-free environment. Mice were implanted using our previously described intrarenal xenograft model of neuroblastoma (27). Briefly, 1 Â 10 6 cells from neuroblastoma cell lines or PDX cells in 100-mL PBS were injected in the left kidney of mice. Tumor growth in mice injected with luciferase-expressing cell lines was assessed by bioluminescence imaging using a Xenogen IVIS 100 instrument (IVIS Lumina XR System; Caliper Life Sciences). All animal experiments were performed in accordance with a protocol approved by the Institutional Animal Care and Usage Committee of Children's Hospital Los Angeles.
Mice in groups receiving aNK cells and dinutuximab were intravenously injected with 1 Â 10 7 aNK cells (immediately after thawing) plus 15 mg mAb twice per week per mouse for 4 weeks, as described previously (24). IL2 (2 mg/mouse) and IL15 (4.9 mg/mouse) were injected intraperitoneally at the same time as dinutuximab and aNK cells. Mice in galunisertib treatment groups were gavaged twice a day with 75 mg/kg of galunisertib (16) suspended in formulated vehicle, as described above. Mice not receiving galunisertib were gavaged twice a day with the formulated vehicle solution alone.

Statistical analysis
Data were analyzed using Stata statistical software (version 11.2) and are represented as means AE standard deviation (SD) unless otherwise stated. ANOVA was performed to determine the significance of observed differences. Analysis of tumor bioluminescence data transformed the photon flux for each mouse using the log (flux þ 1) transformation and then area under the growth curve (AUC) was calculated. The AUC values were used in the analysis to compare differences in tumor photon flux between treatment groups. Mouse survival time was defined as the length of time (in days) from the tumor injection date until the end of the study or time of sacrifice due to disease progression. Censored normal regression was used to examine whether any difference in survival time existed due to treatments. The censored Wilcoxon test was used to examine the difference in the survival curves among the different treatment groups. A P value of <0.05 was considered statistically significant.

Results
TGFB1 and TGFBR1 genes are expressed by high-risk human neuroblastoma tumors, and TGFb activity is present in blood and bone marrow plasma from neuroblastoma patients Exon gene expression in 249 primary neuroblastoma tumors was analyzed using Affymetrix human exon arrays (17,422 genes). TGFBR1 and TGFBR2 genes were expressed at high levels relative to all genes analyzed, and the median expression of TGFB1 was above the 30th percentile (Fig. 1). MYCN expression was at the 100th percentile in the MYCN-amplified group of tumors, and NTRK1 expression was high in the low-risk group, as expected. mRNA expression was at less than the 5th percentile for IFNg (IFNG) and above the 30th percentile for its controlling transcriptional regulator TBX21 (the human ortholog of the mouse Tbet/Tbx21 gene). Median percentile expression of TGFB1 and TGFB2 and high percentile expression of their receptors with low expression of IFNG suggest activity of the TGFb signaling pathway in neuroblastomas.
Next, we used the neuroblastoma SMAD reporter cell line CHLA255hRL-SmadFluc to determine whether blood and bone marrow plasmas from neuroblastoma patients activate the SMAD pathway and whether this activation could be suppressed by the TGFbR1 inhibitor galunisertib. First, we performed serial 5-fold dilutions of galunisertib and found that a clinically achievable dose of 5 mmol/L (1.8 mg/mL; refs. 21, 28) gave maximum inhibition of SMAD activity induced by 10 ng/mL of TGFb1 without cell toxicity (Fig. 1B). Then, using 5 mmol/L galunisertib and the CHLA255hRL-SmadFluc reporter cell line, we observed that SMAD activity was induced by blood and bone marrow plasma obtained from 17 neuroblastoma patients (Fig. 1C) and that this activity could be inhibited by galunisertib. These results suggest that TGFb1 and/or TGFb2 are present in blood and bone marrow from neuroblastoma patients.

Galunisertib decreases suppression of aNK cells by TGFb1
Having established that galunisertib reduces SMAD activity in neuroblastoma cells, we examined whether it might reduce the effects of TGFb1 on aNK cells. Ex vivo-propagated aNK cells were cultured with galunisertib (5 mmol/L), TGFb1 (15 ng/mL), or the combination of galunisertib with TGFb1 for 18 hours, and lysates were subjected to immunoblot assay to detect phospho-SMAD2, total SMAD2, and b-actin. Figure 2A shows strong induction of

SMAD2 phosphorylation in aNK cells by TGFb1 and strong inhibition of this phosphorylation by galunisertib.
TGFb1 also decreased aNK cell expression of the cytotoxicity receptors DNAM-1, NKp30, and NKG2D (Fig. 2B), as reported by others for human NK cells cultured for 2 to 7 days with IL2 or IL15 (3,29). TGFb1 also decreased aNK cell expression of the membrane-bound form of the death ligand TRAIL (Fig. 2B), which we have previously shown to supplement the cytotoxicity of aNK cells against neuroblastoma cells (30). Our experiments did not demonstrate that TGFb1 affected expression of NKp46, NKG2A, Effect of galunisertib on TGFb1-induced SMAD2 phosphorylation, downregulation of cytotoxicity receptors and TRAIL, and inhibition of release of perforin and granzyme A in aNK cells. A, aNK cells cultured with IL2 (10 ng/mL) were treated with galunisertib alone (5 mmol/L), TGFb1 alone (15 ng/mL), or TGFb1 and galunisertib for 18 hours. Whole lysates from aNK cells were subjected to pSMAD2, total SMAD2, and b-actin immunoblot assay. B, Downregulation by TGFb1 of DNAM-1, NKp30, NKG2D, and TRAIL and reversal by galunisertib. aNK cells were pretreated for 48 hours with TGFb1 (10 ng/mL) alone, with galunisertib (5 mmol/L) alone, or with TGFb1 and galunisertib. An additional group received TGFb1 alone for 24 hours, followed by addition of galunisertib for another 24 hours. Using 9-color flow cytometry, viable aNK cells were identified according to CD56 and CD16 expression and absence of CD3, CD14, and CD19 expression, and then their levels of cytotoxicity receptors and of TRAIL were determined. Stain Index values are given in each histogram overlay. The presented results are similar to those for aNK cells propagated from three additional human donors. Red lines, specific antibody; black lines, isotype control. C, Expression of ligands for DNAM-1 (CD112 and CD155), NKp30 (B7-H6), and NKG2D (MICA, MICB, ULBP1, ULBP2/5/6, and ULBP3) and of TRAIL-R2 on neuroblastoma cell lines and on PDX cells (COG-N-415x). D, suppression of perforin and granzyme A secretion from aNK cells by TGFb1 is inhibited by galunisertib. After 6 hours of coculturing aNK cells with neuroblastoma cells, perforin and granzyme A were quantified in the supernatant of each treatment condition using a Luminex multiplexed microbead assay ( Ã , P < 0.05).  (Fig. 2B). Among the receptors for TRAIL, TRAIL receptor 2 (TRAIL-R2) was expressed by the neuroblastoma cell lines CHLA-255-Fluc and CHLA-136-Fluc and by PDX cells COG-N-415x (Fig. 2C). There was little or no expression of TRAIL-R1, TRAIL-R3, or TRAIL-R4 (data not shown). Ligands for DNAM-1 (CD112 and CD155) were also expressed by CHLA-255-Fluc and CHLA-136-Fluc cells and by PDX cells COG-N-415x (Fig. 2C).  Fig. S1).
Because perforin and granzymes are important for NK cell cytotoxicity (31,32), we next examined the effect of TGFb1 and galunisertib on aNK cell release of perforin and granzyme A. As expected, TGFb1 (10 ng/mL) reduced, by 30% or more (P 0.02), the release of both perforin and granzyme A from aNK cells cultured alone or from aNK cells cultured in direct contact with neuroblastoma cells (Fig. 2D). TGFb1-induced inhibition of perforin and granzyme A secretion was prevented by 5 mmol/L of galunisertib ( Fig. 2D; P < 0.05). Intracellular immunostaining for perforin and granzyme A expression indicated no effect of TGFb1 or of galunisertib ( Supplementary Fig. S2), suggesting that TGFb1 inhibits the release of perforin and granzyme A from aNK cells rather than inhibiting their intracellular expression. Taken together, these results indicate that galunisertib inhibits multiple suppressive effects of TGFb1 on the cytotoxic mechanisms of aNK cells.
Having established that galunisertib significantly inhibits effects of TGFb1 on aNK cells, we examined whether it also prevents inhibition of their cytotoxicity by TGFb1. aNK cells that have been propagated using K562.mbIL21 cells are known to be highly cytotoxic, and this cytotoxicity against multidrug-sensitive and -resistant neuroblastoma cell lines in vitro is increased by dinutuximab (24). We found that aNK cytotoxicity against CHLA-255-Fluc (moderately multidrug sensitive) and CHLA-136-Fluc neuroblastoma cells (multidrug resistant; ref. 33) was inhibited by TGFb1, but this inhibitory effect was significantly reversed by addition of 5 mmol/L galunisertib 30 minutes before (Fig. 3). Importantly, addition of galunisertib 24 hours after TGFb1 also inhibited the reduction in aNK cytotoxicity by TGFb1. These findings demonstrate that galunisertib inhibits the suppressive effects of TGFb1 on aNK cell-mediated direct cytotoxicity and ADCC against neuroblastoma cells.

Galunisertib decreases phosphorylation of SMAD2 in neuroblastoma xenografts in NSG mice
To evaluate the inhibitory effect of galunisertib on the TGFb pathway in neuroblastoma tumors, intrarenal tumors were examined for phospho-SMAD2 on day 36, 21, or 27 from mice injected in their left kidneys with CHLA-255-Fluc, CHLA-136-Fluc, or patient-derived xenograft COG-N-415x cells, respectively. These mice were injected intravenously twice a week with dinutuximab mixed with K562.mbIL21-propagated aNK cells, starting three days after intrarenal injection of neuroblastoma cells. IHC demonstrated phospho-SMAD2 in nuclei of tumors that were untreated or treated with dinutuximab and aNK cells ( Fig. 4A-C, brown color). Compared with untreated tumors, phospho-SMAD2 was decreased in neuroblastomas treated with galunisertib and was decreased the greatest in tumors treated with the combination of galunisertib, dinutuximab, and aNK cells. These results indicate that galunisertib is able to penetrate human neuroblastoma xenografts where it reduces SMAD2 pathway activation.

Galunisertib enhances anti-neuroblastoma activity of dinutuximab plus human aNK cells in NSG mice
The ability of galunisertib to inhibit growth of human neuroblastoma tumors and improve survival of NSG mice treated with dinutuximab plus human aNK cells was tested in mice bearing tumors formed by intrarenal injection of CHLA-255-Fluc, CHLA-136-Fluc, and COG-N-415x cells. K562.mbIL21-propagated aNK cells were mixed with dinutuximab and injected intravenously twice weekly, starting 3 days after intrarenal injection of neuroblastoma cells. Bioluminescence imaging was performed for mice injected with luciferase-expressing CHLA-255-Fluc or CHLA-136-Fluc cells, and decreases in bioluminescence were observed in mice treated with the combination of dinutuximab, aNK cells, and galunisertib (Fig. 5A). In contrast, galunisertib alone had no Effect of TGFb1 and galunisertib on human aNK cytotoxicity and ADCC against neuroblastoma cells. aNK cells sustained with IL2 (10 ng/mL) were pretreated 48 hours with TGFb1 alone (10 ng/mL), galunisertib alone (5 mmol/L), TGFb1 and galunisertib, or TGFb1 for 24 hours followed by addition of galunisertib for an additional 24 hours. aNK cells were then cultured with calcein-AM-labeled neuroblastoma cell lines CHLA-255-Fluc or CHLA-136-Fluc at a 1:2 E:T ratio for 6 hours. aNK cell-mediated killing of neuroblastoma cells was assayed by digital imaging microscopy. Galunisertib (5 mmol/L) by itself had no cytotoxic effect (data not shown); Ã , P < 0.05; ÃÃ , P < 0.01; ÃÃÃ , P < 0.001. discernable effect on luciferase signals. In a meta-analysis of AUC for these experiments involving a total of 52 mice and both of the neuroblastoma cell lines, the combination of galunisertib with dinutuximab and aNK cells significantly reduced tumor growth compared with the untreated control group (Supplementary Fig.  S3, P ¼ 0.0003 in the meta-analysis, and P ¼ 0.003 and 0.02 for CHLA-255-Fluc and CHLA-136-Fluc cells, respectively). Treatment with dinutuximab plus aNK cells without galunisertib exhibited a trend toward tumor growth reduction (P value of meta-analysis ¼ 0.08), and treatment with galunisertib alone did not significantly decrease tumor growth (P ¼ 0.41). Importantly, the combination of galunisertib with dinutuximab and aNK cells significantly extended survival of mice injected with CHLA-255-Fluc, CHLA-136-Fluc, or COG-N-415x neuroblastoma cells compared with either untreated mice or mice treated with dinutuximab and NK cells (Fig. 5B-D). Taken together, these data indicate that the addition of galunisertib significantly enhances the antitumor effect of dinutuximab and aNK cells in NSG mice implanted with neuroblastoma cell lines or a PDX.

Discussion
Using microarray gene expression profiling of 249 untreated primary neuroblastomas from patients, we show TGFBR1, TGFBR2, TGFB1, and TGFB2 expression both in high-risk tumors that have either amplified or nonamplified MYCN and in low-risk neuroblastomas. We also show very low expression of IFNG in these tumors, consistent with TGFb-mediated suppression of NK cells (13). A previous study of 61 neuroblastomas of all clinical stages using conventional RT-PCR and electrophoresis also showed expression of TGFBR1, TGFBR2, and TGFB1 (34). In addition, we show for the first time that bone marrow and blood plasmas from patients induce SMAD signaling in a reporter neuroblastoma cell line and that galunisertib, a small-molecule inhibitor of the TGFbR1 signaling pathway that is in phase I and II clinical trials (15), blocks this activity. In agreement with these findings, we show that phospho-SMAD2, which accumulates in cell nuclei downstream of TGFbR1 signaling (35), can be detected in untreated neuroblastomas growing in NSG mice. Treatment of mice with galunisertib inhibited phosphorylation of SMAD2 in these tumors, which indicates its activity in the tumor microenvironment.
We and others have previously demonstrated that aNK cells can be propagated using K562.mbIL21 cells to numbers that should be sufficient for adoptive cell therapy in humans (22)(23)(24). Here, we demonstrate for the first time that TGFb1 inhibits both direct killing and ADCC of neuroblastoma cells by such aNK cells in vitro and that galunisertib can substantially reverse this inhibition. Restoration of ADCC by galunisertib did not involve modulation of CD16 expression. Instead, our findings show that galunisertib treatment of aNK cells in vitro inhibits TGFb1-induced SMAD2 phosphorylation and significantly restores expression of DNAM-1, NKp30,  were injected into the left kidney of each NSG mouse on day 0. Formulated galunisertib (75 mg/kg) was gavaged twice a day from days 3-10 and then 5 days per week from days 13-31. aNK cells (1 Â 10 7 ) plus dinutuximab (15 mg/mouse) were injected intravenously twice a week from day 3, along with IL2 (2 mg) and IL15 (4.9 mg/mouse) intraperitoneally. Paraffin-embedded tumor sections (5 mm) were prepared and immunostained with rabbit polyclonal anti-pSMAD2 as described in Materials and Methods. Tumors from the four treatment groups (untreated, aNK þ dinutuximab, galunisertib alone, or aNK þ dinutuximab þ galunisertib) were harvested at 36 and 21 days after injection of CHLA-255-Fluc and CHLA-136-Fluc, respectively, and 28 days after injection of COG-N-415x PDX cells. Brown, phosho-SMAD2; blue, hematoxylin; galun, galunisertib; dinutux, dinutuximab. NKG2D, and TRAIL and release of perforin and granzyme A, which could contribute to the observed reversal of TGFb1mediated inhibition of cytotoxicity. The ability of a specific TGFBR1 inhibitor to inhibit SMAD2 phosphorylation, an indicator of TGFBR1 activity, and to reverse multiple downstream effects of TGFb1 suggests that TGFBR1-mediated signals are key regulators of multiple components of the cytotoxicity of aNK cells and that this regulation may be reversible upon specific treatment.

CHLA-255-Fluc tumors
Importantly, addition of galunisertib to dinutuximab and aNK not only prevented in vivo phosphorylation of SMAD2 but enhanced the immunotherapy and survival of NSG mice bearing tumors formed by two human neuroblastoma cell lines or a PDX. Although several small-molecule agents that bind and inhibit TGFbR1 have been shown to inhibit TGFbinduced signaling (4,16,(36)(37)(38), our study is the first to demonstrate that a TGFBR1 inhibitor in clinical trials (15), galunisertib, is able to prevent and reverse the suppressive effects of TGFb1 on activated, nontransformed NK cells prop-agated in a manner nearly identical to that being tested in adoptive cell therapy trials of acute myeloid leukemia (Clin-icalTrials.gov identifier: NCT01787474) and of neuroblastoma (NCT02573896).
In preclinical studies of cancer, galunisertib has been reported to enhance the activity of paclitaxel or sorafenib but to have limited activity by itself (17,18). Treatment with galunisertib alone inhibited the growth of only 2 of 13 lung and prostate carcinoma cell lines grown in nude mice (39) and had little or no effect against triple-negative breast cancer xenografts, even though pSMAD2 was decreased (17). Galunisertib enhanced paclitaxel treatment of breast cancer stem cells by blocking paclitaxel-induced interleukin-8 transcription and associated cell proliferation, and treatment of breast cancer xenografts with both drugs prevented reestablishment of tumors (17). Galunisertib decreased phosphorylation of SMAD2 in hepatocellular carcinoma cells, limiting their invasive properties ex vivo and potentiating sorafenib-mediated apoptosis and a decrease in proliferation in vitro (18). Our  study, which did not show galunisertib alone to be active against neuroblastomas growing in NSG mice, is in agreement with these previous reports. Taken together, these findings indicate that combining galunisertib with other therapeutic modalities is necessary for realizing its anticancer potential.
Our study is the first to show that galunisertib can act as a combinatorial agent with immunotherapy, enhancing mAb and NK cell-based treatment of human tumor xenografts. We show that galunisertib administered to NSG mice according to a previously established pharmacokinetic/pharmacodynamic protocol (16) enhances the anti-neuroblastoma effect of dinutuximab with adoptively transferred aNK cells. These findings provide preclinical support for testing galunisertib in combination with dinutuximab in clinical trials for neuroblastoma.

Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.