Abstract
Glioblastoma (GBM) is the most common and malignant intracranial tumor in adults and currently incurable. To specifically target natural killer (NK) cell activity to GBM, we employed NK-92/5.28.z cells that are continuously expanding human NK cells expressing an ErbB2-specific chimeric antigen receptor (CAR).
ErbB2 expression in 56 primary tumors, four primary cell cultures, and seven established cell lines was assessed by immunohistochemistry and flow cytometry. Cell killing activity of NK-92/5.28.z cells was analyzed in in vitro cytotoxicity assays. In vivo antitumor activity was evaluated in NOD-SCID IL2Rγ null (NSG) mice carrying orthotopic human GBM xenografts (6 to 11 mice per group) and C57BL/6 mice carrying subcutaneous and orthotopic ErbB2-expressing murine GBM tumors (5 to 8 mice per group). Statistical tests were two-sided.
We found elevated ErbB2 protein expression in 41% of primary GBM samples and in the majority of GBM cell lines investigated. In in vitro assays, NK-92/5.28.z in contrast to untargeted NK-92 cells lysed all ErbB2-positive established and primary GBM cells analyzed. Potent in vivo antitumor activity of NK-92/5.28.z was observed in orthotopic GBM xenograft models in NSG mice, leading to a marked extension of symptom-free survival upon repeated stereotactic injection of CAR NK cells into the tumor area (median survival of 200.5 days upon treatment with NK-92/5.28.z vs 73 days upon treatment with parental NK-92 cells, P < .001). In immunocompetent mice, local therapy with NK-92/5.28.z cells resulted in cures of transplanted syngeneic GBM in four of five mice carrying subcutaneous tumors and five of eight mice carrying intracranial tumors, induction of endogenous antitumor immunity, and long-term protection against tumor rechallenge at distant sites.
Our data demonstrate the potential of ErbB2-specific NK-92/5.28.z cells for adoptive immunotherapy of glioblastoma, justifying evaluation of this approach for the treatment of ErbB2-positive GBM in clinical studies.
Glioblastoma (GBM) is the most common and aggressive primary brain tumor in adults and currently incurable. Present standard of care includes surgical resection followed by radiotherapy and chemotherapy. Despite this aggressive treatment, median survival of GBM patients is still only about 15 months, and recurrence remains almost inevitable ( 1–4 ). Escape from immune surveillance is thought to contribute to development and progression of GBM ( 5 ). While GBM tumors are frequently infiltrated by natural killer (NK) cells ( 6 ), these are actively suppressed by GBM cells through expression of ligands for inhibitory NK cell receptors and factors such as TGF-β ( 7–9 ). GBM cells also inhibit NK cell activity indirectly through myeloid cells that induce downregulation of the activating NK cell receptor NKG2D ( 10 ). Nevertheless, ex vivo activation with IL-2 or IL-15 restores cytolytic activity of NK cells against GBM ( 11 , 12 ), indicating that NK cells have potential for adoptive immunotherapy of GBM if potent cytotoxicity can be maintained in vivo. Indeed, antitumor activity of lymphokine-activated killer (LAK) cells directly applied into the resection cavity in GBM patients has been reported in some clinical trials ( 13 , 14 ).
Donor-derived allogeneic NK cells are typically being used for adoptive cancer immunotherapy because they do not recognize tumor cells as ‘self’ and may bypass inhibitory signals ( 15 ). Similarly, in clinical trials the continuously expanding human NK cell line NK-92 has been safely applied as an allogeneic cell therapeutic with clinical responses observed in some of the cancer patients treated ( 16–18 ). Like T cells, NK cells can be genetically modified to express chimeric antigen receptors (CARs) that recognize tumor-associated cell surface antigens and mediate specific recognition and lysis of cancer cells. Thereby CAR-engineered NK cells can overcome endogenous resistance mechanisms in tumor cells as demonstrated for NK-92 as well as primary NK cells ( 19–24 ).
The growth factor receptor tyrosine kinase ErbB2 (HER2) has previously been suggested as a suitable target for CAR T cells in glioblastoma ( 25 ). Elevated ErbB2 protein levels have been reported in GBM and were correlated with impaired survival ( 26–29 ). To investigate the potential of ErbB2-specific CAR NK cells for glioblastoma therapy, we employed NK-92/5.28.z cells, a continuously expanding single cell clone currently being developed for clinical applications. The cell clone was established under GMP conditions and carries a codon-optimized CAR (CAR 5.28.z) based on ErbB2-specific antibody FRP5 and CD28 and CD3ζ signaling domains ( 30 ). Here, to assess the frequency of ErbB2 expression we first analyzed ErbB2 protein levels in primary human glioblastoma. We then evaluated the antitumor activity of ErbB2-specific NK-92/5.28.z cells against established GBM cell lines and primary GBM cell cultures in vitro and orthotopic GBM xenografts in immunodeficient mice. To investigate the effects of the CAR NK cells on endogenous antitumor immunity, we treated immunocompetent mice carrying syngeneic ErbB2-positive GBM tumors with NK-92/5.28.z cells and analyzed treatment-induced tumor rejection and subsequent protection of cured mice against repeated tumor challenge.
Methods
Analysis of ErbB2 and CAR Expression
Immunohistochemical analysis of sections of primary and recurrent glioblastoma was performed as described in the Supplementary Methods (available online). ErbB2 expression on the surface of cultured tumor cells was determined by flow cytometry using Alexa Fluor 647- or PE-conjugated antihuman ErbB2 antibody 24D2 (Biolegend, Fell, Germany) or purified monoclonal antihuman ErbB2 antibody FRP5 ( 31 ) followed by APC-conjugated goat antimouse antibody (Dianova, Hamburg, Germany). Flow cytometric analysis was performed with FACSCalibur or FACSCanto II flow cytometers (BD Biosciences, Heidelberg, Germany), and data were analyzed using CellQuest Pro Version 0.0 (BD Biosciences) or FlowJo Version 10.0.7 (FlowJo, Ashland, OR) software. Immunoblot analysis of cell lysates was performed as described in the Supplementary Methods (available online). CAR expression by NK-92/5.28.z cells was analyzed by flow cytometry using ErbB2-Fc fusion protein (R&D Systems, Wiesbaden-Nordenstadt, Germany) followed by APC-coupled goat antihuman antibody (Dianova).
Cytotoxicity Assays
Cytotoxicity of NK cells was analyzed in fluorescence-activated cell sorting (FACS)–based assays as described previously ( 21 ). Briefly, target cells were labeled with calcein violet AM (Life Technologies), washed, and co-incubated with effector cells at different effector to target (E/T) ratios for two or 16 hours at 37°C. After coculture, cells were centrifuged, supernatant was removed, and cells were resuspended in 200 µL of a 1 µg/mL propidium iodide (PI) solution before flow cytometric analysis. Dead target cells were determined as calcein violet AM and PI double positive. Spontaneous target cell lysis in the absence of effector cells was determined in samples only containing labeled target cells and subtracted to calculate specific cytotoxicity. Data were analyzed using FACSDiva Version 6.1.3 software (BD Biosciences). For cytotoxicity assays under hypoxic conditions, effector and labeled target cells were cocultured at low E/T ratios for 16 hours using the GasPak EZ Anaerobe Pouch System (BD Biosciences) before analysis of specific cytotoxicity as described above.
In Vivo Glioblastoma Models
Six- to eight-week-old female NOD-SCID IL2R γ null (NSG) and C57BL/6 mice were used for LN-319 human glioblastoma xenograft and syngeneic GL261/ErbB2 murine glioblastoma models, respectively. The in vivo experiments were approved by the responsible government committee (Regierungspräsidium Darmstadt, Darmstadt, Germany) and were conducted according to the applicable guidelines and regulations. Initial analysis of the in vivo activity of CAR NK cells was performed in subcutaneous glioblastoma models as described in the Supplementary Methods (available online). For orthotopic glioblastoma models, mice were anesthetized with ketamine and xylazine, immobilized in a stereotaxic fixation device (Stoelting, Wood Dale, IL) and injected through a burr hole in the skull with 1 x 10 5 (LN-319 tumors in NSG) or 5 x 10 3 cells (GL261/ErbB2 tumors in C57BL/6) in 2 µL PBS using a 10 µL Hamilton syringe (Hamilton, Bonaduz, Switzerland) and a Quintessential Stereotaxic Injector (Stoelting). Cells were injected at a speed of 0.5 µL per minute into the right striatum with a depth of 3mm to the skull, referring to the tip of the syringe. To avoid cell extrusion, the needle was left in place for two minutes before withdrawal at a speed of 1mm per minute. Seven days after tumor cell inoculation, mice were treated by stereotactic intratumoral injection of 2 x 10 6 (GL261/ErbB2 tumors), or 2 x 10 6 or 4 x 10 6 (LN-319 tumors) NK-92/5.28.z (n = 10 for LN-319 tumors, n = 8 for GL261/ErbB2 tumors) or parental NK-92 cells (n = 11 for LN-319 tumors, n = 6 for GL261/ErbB2 tumors) in 3 µL or 6 µL of injection medium. Treatment was repeated weekly for 11 (LN-319 tumors) or three weeks (GL261/ErbB2 tumors). The mice were observed daily and sacrificed when they developed neurological symptoms or lost more than 20% body weight. Tumor growth was monitored by magnetic resonance imaging (MRI) as described in the Supplementary Methods (available online).
Statistical Analysis
Results were analyzed by two-tailed Student’s t test. Symptom-free survival was analyzed by Kaplan-Meier plot and log-rank (Mantel-Cox) test. P values of less than .05 were considered statistically significant. Statistical calculations were performed using Prism Version 5.0 software (GraphPad Software, La Jolla, CA).
All other methods are described in detail in the Supplementary Methods (available online).
Results
Expression of ErbB2 in Primary Glioblastomas
To confirm the relevance of ErbB2 as a therapeutic target in glioblastoma, we investigated ErbB2 expression by immunohistochemistry (IHC) in tissue samples from 56 primary human glioblastomas (see the Supplementary Methods , available online). For quantification, we applied a scoring system with immune-reactivity scores (IRS) of 0 to 12 that takes intensity of membrane staining and staining frequency into account ( 32 ) ( Figure 1A ). We found high ErbB2 expression in 21.5% of samples (IRS 8–12; n = 12), more moderate expression in another 19.6% of samples (IRS 4–6; n = 11), and no or low expression in 58.9% of samples (IRS 0–3; n = 33) ( Figure 1B ). Where available, we also analyzed corresponding recurrent tumors. Thereby, for the majority of samples ErbB2 expression in the relapsed tumors was comparable with the primary tumor or higher ( Figure 1A , lower right panel; Supplementary Figure 1 , available online).
Expression of ErbB2 in primary glioblastoma. Tumor sections were stained with ErbB2-specific primary antibody and HRP-coupled antirabbit secondary antibodies. The scoring system applied takes staining intensity (0 to 3: no, low, moderate, strong staining) and staining frequency (0 to 4: 0%, 1%-10%, 11%-25%, 25%-50%, >50% of cells) into account. The two scores were multiplied resulting in immune-reactivity score (IRS). A) Examples of primary glioblastomas (IRS 0, 4, 8) and recurrent glioblastoma (IRS 12). Scale bars : 50 µm. B) Percentages of primary glioblastomas with an IRS of 0–3, 4–6, 8, 9–12 (n = 56). HRP = horseradish peroxidase.
Sensitivity of Glioblastoma Cells to NK-92/5.28.z
NK-92/5.28.z is a single-cell clone derived from NK-92 cells by transduction with a lentiviral vector encoding a second-generation CAR that contains ErbB2-specific single-chain antibody scFv (FRP5) and CD28 and CD3ζ signaling domains ( 30 ) ( Figure 2 , A and B). After two hours of co-incubation with established human GBM cells, NK-92/5.28.z cells efficiently and selectively killed ErbB2-positive LN-319, LNT-229, and LN-428 cells, which were resistant to untargeted NK-92 ( Figure 2 , C-E). In contrast, NK-92/5.28.z cells showed only marginal activity against ErbB2-negative G55T2 cells. Analysis of additional GBM cell lines revealed similar selectivity of NK-92/5.28.z cells for ErbB2-positive targets ( Supplementary Table 1 , available online). When co-incubated with tumor cells for five hours at an effector to target (E/T) ratio of 1:1, NK-92/5.28.z–sensitive LN-319 cells readily triggered NK cell degranulation, while this was not the case for antigen-negative G55T2 cells ( Figure 3A ). Ectopic expression of ErbB2 in G55T2 cells restored high cytotoxicity of NK-92/5.28.z, establishing specific CAR triggering as the decisive factor for cell killing ( Figure 3 , B and C; Supplementary Figure 2 , available online). High and selective CAR-mediated cytotoxicity was also observed at E/T ratios of 1:1 or lower if exposure time was extended ( Figure 3D ). Specific cytotoxicity of NK-92/5.28.z cells was thereby not affected by hypoxia, which is usually present in large areas of GBM tumors. Likewise, depending on E/T ratio and exposure time, the addition of exogenous TGF-β had little effect on CAR-mediated cytotoxicity ( Supplementary Figure 3 , available online). Next we analyzed primary GBM stem cell cultures kept under conditions that maintain most of the characteristics of the original tumors ( 33 ). All primary GBM cell lines investigated displayed elevated ErbB2 levels ( Figure 4A ) and, similar to established GBM cells, were sensitive to NK-92/5.28.z cytotoxicity while showing resistance towards parental NK-92 cells ( Figure 4B ). Also, NK-92/5.28.z more rapidly than parental NK-92 cells infiltrated neurospheres of primary GBM cells, which depend on the tumor cells’ stem cell characteristics. Over an observation period of 18 days, cytotoxicity of the CAR NK cells resulted in complete disintegration of the neurospheres and elimination of all tumor cells ( Supplementary Figure 4 , available online).
Specific cytotoxicity of chimeric antigen receptor (CAR)–engineered NK-92 cells against established glioblastoma cells. A) Schematic representation of CAR 5.28.z expressed by NK-92/5.28.z cells, which consists of an immunoglobulin heavy chain signal peptide, ErbB2-specific scFv (FRP5) antibody fragment, a CD8α hinge region, followed by transmembrane and intracellular domains of CD28 and the intracellular domain of CD3ζ. CAR expression is driven by the spleen focus–forming virus promoter. B) CAR expression by clonal NK-92/5.28.z cells was determined by flow cytometry with ErbB2-Fc fusion protein ( black area ). Parental NK-92 cells served as control ( gray area ). C) Expression of ErbB2 on the surface of established LN-319, LNT-229, LN-428, and G55T2 glioblastoma (GBM) cells was determined by flow cytometry with ErbB2-specific antibody ( open areas ). Cells treated with isotype antibody served as controls ( filled areas ). D) Total ErbB2 protein in cell lysates of GBM cells was analyzed by immunoblotting with ErbB2-specific antibody. γ-tubulin served as loading control. E) Cell killing by ErbB2-specific NK-92/5.28.z ( filled circles ) and parental NK-92 cells ( open circles ) was investigated after co-incubation with the indicated target cell lines for two hours at different effector to target ratios. Mean values ± SD are shown (n = 3). CAR = chimeric antigen receptor; CD8α = CD8α hinge region; E/T = effector to target ratio; scFv = ErbB2-specific scFv antibody fragment; SFFV = spleen focus–forming virus; SP = signal peptide.
Dependence of NK-92/5.28.z cytotoxicity on ErbB2 expression. A) Degranulation of NK-92/5.28.z cells ( red lines ) was analyzed by flow cytometry determining CD107a surface expression after five hours of coculture with the indicated glioblastoma target cells at an effector to target ratio of 1:1. Parental NK-92 cells were included for comparison ( blue lines ). B) Expression of ErbB2 on the surface of G55T2/ErbB2 glioblastoma cells derived by stable transfection with an ErbB2 encoding construct was determined by flow cytometry with ErbB2-specific antibody FRP5 followed by APC-conjugated secondary antibody ( red line ). ErbB2 expression by cells transfected with empty vector is shown for comparison ( black line ). Cells treated only with secondary antibody ( filled areas ) served as controls. C) Cytotoxicity of NK-92/5.28.z cells against G55T2/ErbB2 cells was investigated in fluorescence-activated cell sorting–based cytotoxicity assays after co-incubation of effector and target cells for two hours at different E/T ratios. Cells transfected with empty vector were included as control ( neo ). D) Specific cytotoxicity of NK-92 and NK-92/5.28.z cells towards LN-319 cells at low E/T ratios was investigated after co-incubation of effector and target cells for 16 hours. To assess the effect of hypoxia on NK cell activity, coculture was performed in parallel using an anaerobic pouch system. APC = allophycocyanin; E/T = effector to target ratio.
Specific cytotoxicity of NK-92/5.28.z against primary glioblastoma cells. A) Expression of ErbB2 on the surface of primary MNOF168, GBM22, RAV19, and MNOF76 glioblastoma cells was determined by flow cytometry with ErbB2-specific antibody ( open areas ). Cells treated with isotype antibody served as controls ( filled areas ). B) Cell killing by ErbB2-specific NK-92/5.28.z ( filled circles ) and parental NK-92 cells ( open circles ) was investigated after co-incubation with primary glioblastoma cells as targets for two hours at different effector to target ratios. E/T = effector to target ratio.
Activity of NK-92/5.28.z Against Orthotopic Glioblastoma Xenografts
In vivo antitumor activity of NK-92/5.28.z was first investigated in a subcutaneous LN-319 glioblastoma xenograft model in NSG mice. Three days after tumor cell inoculation, mice were treated by peritumoral injection of 2 x 10 7 NK-92/5.28.z or parental NK-92 cells, repeated weekly for six weeks. While unmodified NK-92 had no effect and tumors continued to grow under therapy, tumors in NK-92/5.28.z–treated mice regressed quickly, with five of six mice in this group being tumor-free at termination of the experiment on day 52 ( Supplementary Figure 5 , available online).
Next we evaluated NK-92/5.28.z in an orthotopic xenograft model more closely resembling a clinical situation. LN-319 cells were stereotactically injected into the right brain hemisphere of NSG mice. Seven days later, mice received intratumoral injections of 2 x 10 6 NK-92/5.28.z or parental NK-92 cells once per week for 11 weeks. Thereby, therapy with NK-92/5.28.z prevented tumor outgrowth as assessed by MRI ( Figure 5A ), while treatment with parental NK-92 cells similar to injection medium alone could not suppress tumor growth. This resulted in a marked extension of symptom-free survival of NK-92/5.28.z–treated mice (survival between 149 and 303 days, median survival of 200.5 days) ( Figure 5B ) in comparison with mice treated with untargeted NK-92 or medium (survival between 62 and 122 days, median survival of 73 days, and survival between 64 and 97 days, median survival of 74.5 days, respectively; NK-92/5.28.z vs controls: P < .001). NK-92/5.28.z therapy did not result in the selection of ErbB2-negative tumor cells, indicated by similar target antigen expression in tumors developing in mice treated with NK-92/5.28.z, parental NK-92, or injection medium ( Figure 5C ). We did not observe weight loss, distress, or other abnormalities that would indicate treatment-related adverse reactions.
In vivo antitumor activity of NK-92/5.28.z cells against orthotopic glioblastoma xenografts. A) LN-319 cells (1 x 10 5 ) were stereotactically injected into the right striatum of NOD-SCID IL2Rγ null mice. Seven days later, mice were treated by intratumoral injection of 2 x 10 6 parental NK-92 (n = 11) or ErbB2-specific NK-92/5.28.z cells (n = 10) once per week for 11 weeks. Control mice were treated with injection medium (n = 6). Tumor growth was monitored by MRI. Tumor development in representative mice from each group at days 49 and 84 is shown. B) Symptom-free survival of the mice from the experiment described in (A) analyzed by Kaplan-Meier plot and two-sided log-rank test. C) Immunohistochemical analysis of ErbB2 expression in intracranial LN-319 tumors from mice treated with injection medium, parental NK-92 cells, or ErbB2-specific NK-92/5.28.z cells as indicated. Infiltration zones with ErbB2-positive tumor cells (indicated by arrowheads ) and adjacent normal brain ( upper panels ; 10x magnification) and tumor tissues exhibiting ErbB2 membrane staining ( lower panels ; 40x magnification) are shown. Scoring of ErbB2 expression revealed an immune-reactivity score of 8. Scale bars : 200 µm ( upper panels ) and 50 µm ( lower panels ).
In clinical trials with untargeted NK-92, irradiation of cells with 10 Gy prior to infusion was included as a precaution to prevent permanent engraftment ( 16–18 ). While γ-irradiation of NK-92/5.28.z resulted in a progressive loss of viable cells over time, those cells that were still viable at a specific time point retained cytotoxicity against GBM cells in vitro ( Supplementary Figure 6, A and B , available online). To assess the effect of γ-irradiation on in vivo antitumor activity of the CAR NK cells, NSG mice carrying orthotopic LN-319 xenografts were treated by intratumoral injection of 4 x 10 6 NK-92 or NK-92/5.28.z cells, or NK-92/5.28.z irradiated with 10 Gy prior to injection once per week for 11 weeks. As observed before, treatment with parental NK-92 did not suppress tumor growth, while both nonirradiated and irradiated NK-92/5.28.z cells effectively controlled tumor development ( Supplementary Figure 6, C and D , available online). This resulted in a median symptom-free survival of 307 days for mice treated with nonirradiated NK-92/5.28.z (range of survival between 93 and 323 days). In the group treated with irradiated NK-92/5.28.z, four of six mice were still alive at day 330. In contrast, the NK-92–treated mice had become symptomatic and were killed latest by day 189 (median survival 119.5 days, range of survival between 86 and 189 days).
Activity of NK-92/5.28.z Against Subcutaneous Glioblastoma Tumors in Immunocompetent Mice
CAR-mediated activation of NK-92/5.28.z cells induces production of high levels of pro-inflammatory cytokines, such as IFN-γ and MIP-1α, and lower levels of TNF-α ( Supplementary Figure 7 , available online) ( 30 ). In addition to enhancing direct antitumor activity, this may result in activation of bystander-immune cells and endogenous antitumor immunity. Human MIP-1α and TNF-α can activate their respective murine receptors. Hence, to investigate potential immunostimulatory effects of NK-92/5.28.z cells, we employed C57BL/6-derived GL261 cells, which are widely used as an immunocompetent murine glioblastoma model ( 34 , 35 ). In in vitro cytotoxicity assays, GL261/ErbB2 cells ectopically expressing the human target antigen were readily killed by NK-92/5.28.z cells, while unmodified GL261 showed only limited sensitivity ( Figure 6 , A and B).
In vivo antitumor activity of NK-92/5.28.z cells against subcutaneously growing murine glioblastoma tumors. A) Expression of ErbB2 on the surface of GL261/ErbB2 cells derived by stable transfection of the murine glioblastoma cells with an ErbB2 encoding construct was determined by flow cytometry with ErbB2-specific antibody ( open areas ) ( right panel ). Cells treated with isotype antibody served as controls ( filled areas ). Parental GL261 cells were included for comparison ( left panel ). B) Cytotoxicity of ErbB2-specific NK-92/5.28.z against GL261/ErbB2 cells was investigated in fluorescence-activated cell sorting–based cytotoxicity assays after co-incubation of effector and target cells for two hours at different E/T ratios. Cells transfected with empty vector were included as control ( neo ). C) GL261/ErbB2 cells (1 x 10 6 ) were subcutaneously injected into the right flank of immunocompetent C57BL/6 mice. Seven days later, mice were treated by peritumoral injection of 1 x 10 7 ErbB2-specific NK-92/5.28.z (n = 5) or parental NK-92 cells (n = 5) twice per week for four weeks. Control mice received injection medium (n = 5). D) Tumor growth in the individual mice was followed by caliper measurements. E) Mice that had rejected initial tumor challenge upon treatment were rechallenged at day 100 by subcutaneous injection of 1 x 10 6 GL261/ErbB2 cells into the left flank, and tumor growth was followed. Naïve C57BL/6 mice injected at day 100 with GL261/ErbB2 cells served as a control (n = 4). E/T = effector to target ratio.
Next, GL261/ErbB2 cells were injected subcutaneously into immunocompetent C57BL/6 mice and allowed to form tumors for one week ( Figure 6C ). Then mice were treated by peritumoral injection of 1 x 10 7 NK-92/5.28.z or parental NK-92 cells twice per week for four weeks. While tumors grew rapidly in control mice injected with medium, in four of five mice treated with NK-92/5.28.z and one of five mice treated with parental NK-92 cells, tumors initially developed but completely regressed under therapy, with the mice being tumor-free latest by day 49 (NK-92/5.28.z) or 56 (NK-92) ( Figure 6D ). To assess treatment-induced endogenous antitumor immunity, mice that had rejected the initial tumor were rechallenged at day 100 by injection of GL261/ErbB2 cells on the contralateral side without any subsequent treatment. While tumors readily developed in three of four naïve mice included as a control, none of the mice that had rejected the glioblastoma cells under NK cell therapy showed durable tumor development ( Figure 6E ).
Effect of NK-92/5.28.z Treatment on the Growth of Intracranial Glioblastoma and Endogenous Antitumor Immunity
To evaluate therapeutic activity of NK-92/5.28.z cells against intracranial GL261/ErbB2 tumors, the GBM cells were stereotactically injected into the right brain hemisphere of C57BL/6 mice. Seven days later, mice were treated by intratumoral injection of 2 x 10 6 NK-92/5.28.z or parental NK-92 cells once per week for three weeks ( Figure 7A ). While the NK-92–treated mice rapidly developed intracranial tumors and had to be killed latest by day 42 (median survival 33 days), five of eight NK-92/5.28.z–treated mice rejected the tumor cells and did not display any symptoms indicating cure of the disease (NK-92/5.28.z vs NK-92: P < .001) ( Figure 7 , B and C). At day 126, the five surviving mice were rechallenged by stereotactic injection of a second dose of GL261/ErbB2 cells into the contralateral brain hemisphere and left untreated. The mice also rejected the second tumor challenge and survived without symptoms until day 194 when the experiment was terminated ( Figure 7D ). In contrast, naïve control mice injected at day 126 with GL261/ErbB2 cells all developed disease and had to be killed latest 39 days after tumor inoculation (day 165 of the experiment; median survival 29 days; NK-92/5.28.z–pretreated vs naïve control: P = .002). Sera from the cured mice all contained IgG antibodies reactive with GL261/ErbB2 cells indicating induction of endogenous tumor-specific immunity following NK-92/5.28.z treatment ( P = .005) ( Figure 7E ).
In vivo antitumor activity of NK-92/5.28.z cells against intracranial glioblastoma in immunocompetent mice. A) GL261/ErbB2 cells (5 x 10 3 ) were stereotactically injected into the right striatum of C57BL/6 mice. Seven days later, mice were treated by intratumoral injection of 2 x 10 6 parental NK-92 (n = 6) or ErbB2-specific NK-92/5.28.z cells (n = 8) once per week for three weeks. B) Symptom-free survival of the mice analyzed by Kaplan-Meier plot and two-sided log-rank test. C) Tumor development was assessed by MRI at day 28 and is shown for representative mice from each group. D) Mice that were cured upon NK-92/5.28.z treatment (n = 5) were rechallenged at day 126 by stereotactic injection of 5 x 10 3 GL261/ErbB2 cells into the left brain hemisphere, and symptom-free survival was followed without any further therapy and analyzed by Kaplan-Meier plot and two-sided log-rank test. Naïve C57BL/6 mice stereotactically injected into the brain with GL261/ErbB2 cells at day 126 served as a control (n = 5). E) Induction of IgG serum antibodies against glioblastoma cells in NK-92/5.28.z–treated mice (n = 4) from the experiment shown in (D) was investigated by flow cytometry. Sera from naïve C57BL/6 mice (n = 2) served as controls. Results were analyzed by two-tailed Student’s t test. MFI = mean fluorescence intensity (geometric mean).
Discussion
In this study, we demonstrate potent and specific activity of NK-92/5.28.z cells against ErbB2-positive glioblastoma in preclinical models and show induction of protective endogenous antitumor immunity following therapy with these CAR NK cells. NK-92/5.28.z is a continuously expanding NK cell clone carrying an ErbB2-specific second generation CAR (CAR 5.28.z) that was derived from the established human NK-92 cell line under GMP conditions by transduction with a lentiviral CAR vector, enabling CAR-triggered release of cytotoxic granules and induction of target cell apoptosis ( 30 ). The tumor-associated cell surface antigen ErbB2 is an important therapeutic target, and ErbB2 overexpression has been reported in many human cancers including breast, ovarian, and lung carcinomas ( 36 , 37 ). While absent in the adult central nervous system ( 38 ), ErbB2 expression is often associated with high-grade gliomas ( 29 ) and has been correlated with early mortality ( 27 ). In immunohistochemical analysis of 56 primary human GBM samples, we found elevated ErbB2 expression in 21.5%, with another 19.6% of tumors displaying more moderate ErbB2 staining. We applied a scoring system that takes into account intensity of membrane staining and staining frequency ( 32 ) and appears well-suited to identify GBM patients that may benefit from ErbB2-targeted immunotherapy. While ErbB2 mRNA levels were consistent with immunohistochemistry scores (data not shown), analysis of mRNA levels does not allow assessment of ErbB2 membrane expression, which is crucial for recognition of tumor cells by NK-92/5.28.z. For in vitro functional analysis, we employed established human GBM cell lines and primary GBM isolates, including cells with stem cell–like characteristics and tumor-initiating potential. Unlike breast cancer cells that often harbor amplified copies of the ERBB2 gene ( 39 ), ERBB2 gene amplification has not been found in glioblastoma and ErbB2 levels are more moderately enhanced ( 26 ). Nevertheless, NK-92/5.28.z in contrast to parental NK-92 lysed all ErbB2-positive GBM cells tested. Furthermore, ectopic expression of ErbB2 in ErbB2-negative and therefore NK-92/5.28.z–resistant GBM cells allowed rapid and efficient killing by the CAR NK cells, confirming CAR-dependent triggering of the NK cells’ cytolytic activity.
Earlier studies with LAK and NK cells suggest a beneficial role for these cells in adoptive immunotherapy of glioblastoma ( 11–14 ). In clinical trials, the immune cells were directly placed into the resection cavity through a Rickham catheter, with the number of injected cells positively correlating with patient survival ( 14 ). A similar strategy may be followed for NK-92 cells, which in rodent models did not cross the blood-brain barrier unless local access was enhanced by focused ultrasound ( 40 ). Because in more than 80% of GBM patients recurrence is local ( 41 ), an intratumoral treatment strategy appears feasible and may prevent early relapse. To model this clinical strategy for local therapy, we applied weekly stereotactic injection of CAR-expressing NK-92/5.28.z cells directly into the tumor area in orthotopic GBM xenografts in NSG mice. None of the mice carrying LN-319 xenografts and treated with NK-92/5.28.z showed any visible tumor progression during therapy as assessed by MRI. This resulted in long-term therapeutic benefit with an almost three times longer symptom-free survival when compared with treatment with untargeted NK-92 cells, which had no effect on the course of the disease, resulting in the death of the majority of mice while still under therapy. An increase in the therapeutic dose further enhanced efficacy ( Supplementary Figure 6 , available online). In vivo antitumor activity of NK-92/5.28.z was thereby retained upon irradiation of the cells with 10 Gy. This will be relevant for clinical application of NK-92/5.28.z, where irradiation may be included as a safety measure as previously applied in clinical trials with untargeted NK-92 cells ( 16–18 ). Immunohistochemical analysis of LN-319 xenografts from tumor-bearing mice applying the same scoring system used for primary GBM samples revealed an ErbB2 immune-reactivity score of 8, confirming that with respect to ErbB2 expression our model corresponds well to the clinical situation. Tumors developing in NK-92/5.28.z–treated mice that developed after completion of therapy did not show a reduction in ErbB2 expression, suggesting that in this model no immune escape was induced.
Successful application of CD19 CAR-modified T cells in patients with B-cell malignancies has demonstrated the potency of this approach for adoptive cancer immunotherapy ( 42–45 ). Nevertheless, treatment of solid cancers with CAR-engineered effectors remains a challenge, in part because of immunosuppressive effects of the tumor microenvironment. Upon CAR-mediated activation, NK-92/5.28.z cells secrete pro-inflammatory cytokines such as IFN-γ, TNF-α, and MIP-1α ( Supplementary Figure 7 , available online) ( 30 ), which are of human origin but in part are also active in mice. This may contribute to direct antitumor effects but could also enhance endogenous antitumor immunity by activation of bystander immune cells. We investigated this possibility in glioblastoma models in immunocompetent C57BL/6 mice employing syngeneic ErbB2-expressing GL261 cells ( 35 ). Both in subcutaneous and intracranial GL261/ErbB2 GBM models, locoregional treatment with NK-92/5.28.z cells resulted in complete tumor regression and cures in the majority of mice, which were subsequently protected against rechallenge with a lethal dose of GL261/ErbB2 cells at a distant site. This suggests induction or enhancement of endogenous antitumor immunity by NK-92/5.28.z therapy resulting in long-lasting systemic immunological memory. The moderate immunogenicity of GL261 cells can be increased by immunotherapy with cancer vaccines or cytokines ( 46 , 47 ). Likewise, direct tumor cell lysis and cytokine release by NK-92/5.28.z cells facilitated a strong antitumor immune response indicated by the appearance of IgG serum antibodies reactive with GL261/ErbB2 cells in mice treated with NK-92/5.28.z. Activation of endogenous antitumor immunity by NK cell therapy will likely also be important in a clinical setting. NK cells can induce dendritic cell (DC) activation and eliminate tolerogenic immature DCs ( 48 , 49 ), which together with direct NK-mediated tumor cell lysis can enhance antigen presentation and overcome immunosuppressive effects in the tumor microenvironment. In turn, endogenous immune responses against mutated GBM antigens may be intensified and further delay tumor recurrence ( 50–52 ).
Phase I clinical trials in cancer patients demonstrated safety of infusion of irradiated NK-92 cells, with clinical responses achieved in a subset of patients ( 16–18 ). Here, we showed that CAR-engineered NK-92 cells can be employed in a similar fashion as a clonal, molecularly and functionally well-defined, and continuously expandable off-the-shelf cell therapeutic agent, suggesting these cells as a readily available and cost-effective alternative to CAR-modified patient T cells ( 25 ). Our data demonstrate potent and selective antitumor activity of ErbB2-specific NK-92/5.28.z against GBM cells in vitro and in orthotopic GBM xenograft models in vivo, as well as cures and induction of endogenous antitumor immunity following NK-92/5.28.z treatment in immunocompetent mice.
Nevertheless, our study is not without limitations. The ErbB2-specific CAR expressed by NK-92/5.28.z cells does not recognize murine ErbB2, and potential on-target/off-tumor activity of the CAR NK cells could not be evaluated in the mouse models employed. Also, effects of NK-92/5.28.z cells on endogenous antitumor immunity may be different in GBM patients because the human cytokines produced by the effector cells are only partially active in mice. Safety and efficacy of adoptive immunotherapy with the CAR NK cells need to be further investigated in clinical trials. Preparations for a phase I clinical trial in recurrent ErbB2-positive glioblastoma are underway, which will be based on local application of NK-92/5.28.z cells directly into the resection cavity.
Funding
This work was supported in part by grants from the German Federal Ministry of Education and Research (BMBF) (Cluster für individualisierte Immunintervention, Ci3; FKZ 131A009A, 131A009B, 131A009C), Deutsche Forschungsgemeinschaft (DFG) (GRK1172), LOEWE Center for Cell and Gene Therapy Frankfurt (CGT), and institutional funds of the Georg-Speyer-Haus. The Georg-Speyer-Haus is funded jointly by the German Federal Ministry of Health (BMG) and the Ministry of Higher Education, Research and the Arts of the State of Hessen (HMWK). The LOEWE Center for Cell and Gene Therapy Frankfurt is funded by HMWK, reference number: III L 5–518/17.004 (2013).
The study funders had no role in the design of the study; the collection, analysis, or interpretation of the data; the writing of the manuscript; or the decision to submit the manuscript for publication.
We thank Manuel Grez for helpful discussions, Stefan Stein and Tefik Merovci for help with flow cytometric cell sorting, Maurice Harth for help with magnetic resonance imaging, Sarah Oelsner for help with cytokine analysis, Barbara Uherek and Thorsten Geyer for technical assistance, and the staff at the animal facility of the Georg-Speyer-Haus for their support.
References
