Targeting glioblastoma with NK cells and mAb against NG2/CSPG4 prolongs animal survival.

Glioblastoma (GBM) is the most malignant brain tumor where patients' survival is only 14.6 months, despite multimodal therapy with debulking surgery, concurrent chemotherapy and radiotherapy. There is an urgent, unmet need for novel, effective therapeutic strategies for this devastating disease. Although several immunotherapies are under development for the treatment of GBM patients, the use of natural killer (NK) cells is still marginal despite this being a promising approach to treat cancer. In regard of our knowledge on the role of NG2/CSPG4 in promoting GBM aggressiveness we investigated the potential of an innovative immunotherapeutic strategy combining mAb9.2.27 against NG2/CSPG4 and NK cells in preclinical animal models of GBM. Multiple immune escape mechanisms maintain the tumor microenvironment in an anti-inflammatory state to promote tumor growth, however, the distinct roles of resident microglia versus recruited macrophages is not elucidated. We hypothesized that exploiting the cytokine release capabilities of activated (NK) cells to reverse the anti-inflammatory axis combined with mAb9.2.27 targeting the NG2/CSPG4 may favor tumor destruction by editing pro-GBM immune responses. Combination treatment with NK+mAb9.2.27 diminished tumor growth that was associated with reduced tumor proliferation, increased cellular apoptosis and prolonged survival compared to vehicle and monotherapy controls. The therapeutic efficacy was mediated by recruitment of CCR2low macrophages into the tumor microenvironment, increased ED1 and MHC class II expression on microglia that might render them competent for GBM antigen presentation, as well as elevated IFN-γ and TNF-α levels in the cerebrospinal fluid compared to controls. Depletion of systemic macrophages by liposome-encapsulated clodronate decreased the CCR2low macrophages recruited to the brain and abolished the beneficial outcomes. Moreover, mAb9.2.27 reversed tumor-promoting effects of patient-derived tumor-associated macrophage/microglia(TAM) ex vivo.Taken together, these findings indicate thatNK+mAb9.2.27 treatment may be an amenable therapeutic strategy to treat NG2/CSPG4 expressing GBMs. We provide a novel conceptual approach of combination immunotherapy for glioblastoma. The results traverse beyond the elucidation of NG2/CSPG4 as a therapeutic target, but demonstrate a proof of concept that this antibody may hold potential for the treatment of GBM by activation of tumor infiltrated microglia/macrophages.


INTRODUCTION
Glioblastoma (GBM) is the most common and malignant brain tumor in adults, classified as grade IV astrocytoma by the World Health Organization (WHO) [1]. Despite the multimodal treatment consisting of debulking surgery, radiotherapy and chemotherapy, the prognosis remains dismal, with a median survival of 14.6 months [2]. The main challenge for successfully treating GBM is its diffuse invasion of the brain parenchyma that renders the tumor cells refractory to surgery, chemo-radiotherapy and immune surveillance, leading invariably to recurrence. Moreover, the molecular and cellular heterogeneity of GBM underlies their inherent resistance to radio-and chemotherapy. Thus, there is an urgent need for identifying cancer biomarkers that drive malignant progression and that may be amenable for effective combination immunotherapeutic approaches. Indeed there is a growing interest in establishing novel immunotherapeutic approaches to the management of GBM patients. This is boosted partly by the Federal Drug Administrations (FDA) approval of Sipuleucel-T for prostate cancer and Ipilimumab for metastatic melanoma treatment. Moreover, the continual discovery of novel tumor antigens that are abundantly and specifically expressed in GBM tissue compared to normal brain [3,4]may spur further interest in vaccine based immunotherapies.
Potential targets may also be the interactions between tumor cells and constituents of the microenvironment that regulate the bioavailability of ligands for signaling receptors and promotion of malignant progression. Amongst these, proteoglycans could impact on malignant progression by binding growth factors, sequestration of chemokines and proteases. Recently the knockdown of the SULF2 heparin sulfatase in astrocytoma was demonstrated to the diminish activity of several receptor tyrosine kinase (RTK) signaling pathways known to be active in GBM [5]. Our team previously demonstrated the existence of a subpopulation of GBM cells expressing Neuroglial-2 (NG2/CSPG4) transmembrane chondroitin sulfate proteoglycan, with an immature phenotype denoted by nestin and vimentin positivity, but not CD133 [6]. The gene, CSPG4, encoding the NG2 or human melanoma proteoglycan, is turned off upon terminal differentiation, but is aberrantly re-expressed by several tumor types [7-10]. Recently, we described that 50 % of GBM patients' biopsies show high levels of NG2/CSPG4 expression and that this was an independent prognostic factor for shorter patient survival [6]. These NG2/CSPG4 positive GBMs also corresponded to the proliferative and mesenchymal molecular phenotypes that are associated with poor prognosis [11]. In addition, we demonstrated that NG2/ CSPG4 expression by GBM cells promotes angiogenesis [8], cellular proliferation [12], and chemo-resistance [13]. In this latter study, we demonstrated that NG2/CSPG4 expressing GBM cells were highly resistance to tumor necrosis factor alpha (TNF-α) mediated apoptosis due to elevated PI3K/Akt survival signaling. NG2/CSPG4´s impact on the key hallmarks of GBM and its cell surface expression on both tumor and neovasculature render it an amenable target for immunotherapy using monoclonal antibodies that aid the recognition and destruction of malignant cells via the immune system. The adoptive cell transfer (ACT) of autologous or allogeneic lymphocytes to treat cancer has demonstrated great therapeutic potential in early clinical trials of several solid malignancies, including melanoma and neuroblastoma [14,15]. A number of phase I/II clinical trials have been conducted using ACT of lymphokine activated killer (LAK) cells or cytotoxic T lymphocytes (CTLs) to treat GBM. To date the efficacy of adoptive cellular therapies has been moderate, partly due to the need to generate tumor specific lymphocytes for each individual patient, which poses technical and economic hurdles for fast tracking the treatments to the clinic. To bypass these limitations Marcus et al., [14] pioneered chimeric antigen receptor (CAR) redirected T cells that are composed of single chain antibody fragments fused to T cell activating signaling motifs. This endows the allogeneic T cells major histocompatibility complex (MHC) unrestricted tumor specificity and thus, could be used as " off the shelf, universal effector cells" against cancer in all patients. However, ACT of allogeneic T cells poses the inherent risk of host-versus-graft response (HvG) and the danger of graft-versus-host disease (HvHD).
Among cytotoxic lymphocytes, NK cells are the most efficient effectors against tumors, capable of direct killing without prior immunization [16]. So far only two studies investigating purified NK cells for GBM treatment have been conducted, and they demonstrated a preferential killing of GBM stem-like cells [17,18] purported to contribute to therapy resistance and recurrence. Several studies have demonstrated that NK cells can modulate the development of tumor-specific cytotoxic T lymphocytes and induce a T-helper 1 (Th1) cytokine profile in cancer [19][20][21]. Moreover, the immunomodulation capacity of NK cells through cytokine secretion could skew tumorassociated macrophages (TAM) from anti-inflammatory to pro-inflammatory phenotypes [22] and was shown to enhance tumor immunoediting [23]. These last properties are essential for the success of an immunotherapy against GBM where up to 70 % of the tumor mass represent TAM polarized to the M2 phenotype that promote tumor cell proliferation [24], [25]. While NK cell therapies were shown to be effective in hematological malignancies, there is a paucity of studies evaluating their efficacy in brain tumors in vivo [26]. Whereas most studies utilize NK cells for their direct cytotoxicity capabilities, in the present study we investigated a novel approach to exploit the potential of NK cells to revert the immune contexture from anti-inflammatory to pro-inflammatory through cytokine release. We further investigated the therapeutic potential of NK cells to induce antibody dependent cellular www.impactjournals.com/oncotarget cytotoxicity (ADCC) in the brain through ligation of the mAb9.2.27 directed against NG2/CSPG4. The mAb9.2.27 is one of the first monoclonals ever to be produced against NG2/CSPG4, as well as one of the antibodies having the longest track-record of being exploited to target NG2/ CSPG4-positive cancer cells in vitro and in vivo. All attempts to use mAb9.2.27 to abrogate tumors in vivo have focused upon the use of antibody complexes carrying cytotoxic agents, whereas no published study has clearly demonstrated a direct anti-neoplastic effect of the naked antibody. Other anti-NG2/CSPG4 antibodies have been reported to display an anti-tumor potential in vivo, but none of these have proven to be effective on glioblastoma and none have been extensively tested in combination therapy with immune cells. Thus, the present study provides a novel conceptual approach to the combination immunotherapeutic treatment of glioblastoma. Our findings have greater implications beyond the elucidation of NG2/CSPG4 as a therapeutic target, but demonstrate a proof of concept that mAb9.2,27 could activate cytotoxic functions of glioma infiltrated microglia/macrophages that may further hold therapeutic potential.
The principal aims of the present paper were to investigate the therapeutic efficacy of combining adoptively transferred, purified, activated NK cells with passive immunotherapy using mAb9.2.27 in GBMbearing rats and to identify the mechanisms and cellular subsets mediating the anti-tumor effects. We demonstrated that the combination treatment with activated NK cells and mAb9.2.27 eradicated the tumor more efficiently compared to monotherapies with mAb9.2.27 or NK cells and vehicle-treated controls. The mechanism involved the recruitment of macrophages/microglia with a proinflammatory phenotype into the tumor. In addition, activated macrophages/microglia became highly cytotoxic against tumor cells ex vivo in presence of mAb9.2.27

Combination treatment with mAb9.2.27 and adoptively transferred NK cells diminishes GBM cell proliferation and increases survival
We demonstrated previously that elevated levels of the NG2/CSPG4 proteoglycan on GBM cells and angiogenic vasculature is associated with a more aggressive disease course [6,8,12,13]. We therefore hypothesized that perturbation of NG2/CSPG4 signaling with mAb9.2.27 alone or in combination with adoptively transferred NK cells might have therapeutic benefits for tumor-bearing rats. First we investigated the efficacy of the combination treatment in eradicating U87MG gliomas that are 99.2±0.2 % (n=3) NG2/CSPG4 positive, as recognized by mAb9.2.27 ( Supplementary Fig. 1A). Four weeks after treatment, control untreated U87MG tumors were strongly contrast enhancing on T1-weighted MR images indicating increased angiogenesis and rapid growth compared to monotherapy and combination treated animals (Fig.  1A). However, while the monotherapy groups exhibited initial radiological responses of reduced tumor sizes on T1 weighted MRI with contrast, (Fig. 1A), after 5 weeks both monotherapy and control tumors progressed and killed their hosts. The NK+mAb9.2.27 combination treated tumors regressed as indicated by dramatically diminished contrast enhancement in MR images 3 months post-treatment (Fig. 1A). Tumor cell proliferation was significantly attenuated in the combination treatment compared to all other groups (One way ANOVA F 7.4, NK p=0.006, n=6; mAb9.2.27 and control p=0.001, n=5), (Fig.  1B). The tumors treated with combined NK+mAb9.2.27 contained significantly larger areas with apoptotic and necrotic tissue compared to all other treatments (One way ANOVA F 20 , df=3, p=0.0001, n=32), (Fig. 1C). Correspondingly, the combined treatment significantly prolonged the survival of the animals with a median survival time of 91 days compared to 52, 44, and 39 days in the mAb9.2.27, NK cell and control groups, respectively (Log Rank 9.3 , df=3, p=0.026, n=7), (Fig. 1D). In NK+mAb9.2.27 group, 60 % of the animals sacrificed for autopsy displayed necrosis with no visible tumor mass.
Next, we investigated the requirement for NG2/ CSPG4 recognition in NK+mAb9.2.27 therapeutic efficacy in U251-NG2 GBM tumors that stably overexpress rat NG2/CSPG4 but have a lower affinity for the human mAb9.2.27 as indicated by 34.5±4.2 % (n=3) positivity, (Supplementary Fig. 1B). T1-weighted MR images showed contrast agent leakage in all tumors one week prior to treatment. Treatment with mAb9.2.27 alone or in combination with NK cells diminished contrast agent leakage at 10 days post treatment compared to control tumors. However, all the tumors continued to grow as seen on T1-weighted MRI after 3 weeks (Supplementary Fig. 2A). Optical imaging demonstrated the presence of mAb9.2.27 retained in the tumor bed 5 days postinfusion while the IgG2a isotype control antibodies could not be visualized ( Supplementary Fig. 2B Fig. 2C). However, the combination treated tumors also contained greater areas with necrotic/apoptotic cells compared to control and monotherapy groups (ANOVA F 2.8 , df=3, p=0.0001, n=5), (Supplementary Fig. 2D). The latter finding was reflected in the survival curves, as the NK+mAb9.2.27 combination treatment marginally prolonged the median survival to 34.5 days compared to 24 days in NK cell  Histological analyses of U87MG tumors revealed that while untreated control and NK cell monotherapy treated tumors were angiogenic with no discernible   (Fig. 1E). The necrotic areas showed an intense myeloperoxidase (MPO) activity in the combination treated tumors, whereas it was markedly attenuated in NK cell monotherapy and control groups (One-way ANOVA 14.97 , df=3, p=0.0018, n=5), (Fig. 1E and 1F). Granzyme B was also abundantly expressed after NK+mAb9.2.27 treatment in areas adjacent to focal necrosis/apoptosis compared to the control tumors (One way ANOVA 13.21 , df=3, p=0.0042, n=4), ( Fig. 1E Fig. 3A). Collectively, these data indicate a tropism for activated immune cells with the capacity to secrete cytotoxic granules within tumor in the mAb9.2.27 and NK+mAb9.2.27 treated rats.
Next, we aimed to identify the cell populations implicated and delineate the mechanisms mediating the therapeutic effect of the combination treatment.
Histological analysis revealed striking clusters of large foamy cells at the periphery and within the core of the mAb9.2.27 and NK+mAb9.2.27 treated U251-NG2 tumors that were markedly diminished in the control and NK cell treated tumors ( Fig. 2A). At perivascular cuffs and choroid plexus, CD8 + cells were most pronounced in the mAb9.2.27 monotherapy and NK+mAb9.2.27 treated tumors compared to the control (One-Way ANOVA F 11.0, df=3, p=0.0011, n=5) and NK cell monotherapy tumors (One-Way ANOVA F11.0 , p=0.0001, n=5), where the CD8 + cells were restricted to the tumor/brain periphery ( Fig. 2B and 2C). No significant difference in the abundance of phagocytic ED1 + cells across the different treatment groups was found in the U251-NG2 tumors (Fig. 2B). However, the majority of CD8 + cells in the NK cell monotherapy and combination therapy tumors co-expressed ED1. Moreover, while CD8 + ED1 + cells deeply infiltrated the combination treated tumors, these cells remained at the tumor/brain border in the NK cell monotherapy (Fig. 2D). In contrast, CD8 -ED1+ cells densely infiltrated the mAb9.2.27 treated tumors, while the control tumors contained predominantly CD8 + ED1cells (Fig. 2D). The tropism for CD8 + ED1 + cells in the combination treated tumors was also confirmed in the U87MG tumors ( Supplementary Fig. 6A-D). CD8+ cells were more abundant in the NK+mAb9.2.27 treated tumors compared to monotherapy and controls (One-Way ANOVA 19.24 , p=0.0002, n=5, (Supplementary Fig 4C). Likewise, ED1 + cells were most abundant after NK+mAb9.2.27 treatment (Two-Way ANOVA t 3.928 , p=0.0003, n=5), (Supplementary Fig 4D). As CD8 and ED1 markers can be expressed by both microglia and macrophages [27] that are difficult to distinguish by immunohistochemistry, we systemically depleted macrophages by intraperitoneal injection of clodronate once a week for 4 weeks, starting at the day of the treatment.

Macrophage
depletion by clodronate treatment abolishes the therapeutic response in NK+mAb9.

combination therapy.
To investigate further the mechanism leading to tumor regression, we utilized a unique GBM animal model developed in our laboratory where patient GBM biopsy spheroids are serially propagated in vivo over several generations. The resultant xenografts retain the genetic background, cellular heterogeneity and biological features of the original patient tumor [28]. P3-30 GBM cells express elevated levels of NG2/CSPG4 (93.3±3.2 %, n=3), (Supplementary Fig. 1C). As observed with the U87MG tumors, the combination therapy diminished P3-30 lesions volumes including vasogenic edema on T2-and solid tumor sizes on T1-weighted MR imaging, respectively, compared to the vehicle treated control group. The macrophage depletion abrogated this therapeutic effect as indicated by increased lesion volumes including tumor and vasogenic oedema on T2-, as well as solid tumor volume on T1-weighted MR imaging respectively (Fig. 3A). The NK+mAb9.2.27 treated tumors had larger regions with tissue necrosis compared to the control and clodronate treated group that exhibited typical pseudopalisading necrosis and angiogenic vasculature (Fig. 3B). Moreover, the clodronate NK+mAb9.2.27 treated tumors had strikingly numerous mitotic figures and significantly increased tumor proliferation indicated by elevated Ki67 labeling index compared to NK+mAb9.2.27 treated animals (One-Way ANOVA F 11.19, df=2, p=0.0011, n=5), (Fig. 3C-D). The macrophage depletion also significantly reduced the fraction of apoptotic/necrotic cells compared to vehicle treated controls, and NK+mAb9.2.27 tumors (One-Way ANOVA F 7.75, p=0.0079, n=4), (Fig. 3B and  3E). The NK+mAb9.2.27 treatment prolonged animal survival compared to control, with median survival of 46 days versus 38.5 days respectively (Log Rank 10.06 , df=1, p=0.0015, n=7), (Fig. 3F). However, the macrophage depletion abrogated this increase of animal survival, with a median survival of 40.5 days (Log Rank 9.8, df=1, p=0.0017, n=8), (Fig. 3F) . 4A and 4B). Iba1, the calcium binding protein expressed by activated and phagocytic macrophage/microglia was also elevated in the core of the combination treated tumors compared to vehicle treated control (Two-Way ANOVA t 4.633 , p=0.01, n=4), and macrophage depleted tumors (Two-Way ANOVA t 3.235 , p=0.0243, n=4), (Fig. 4B). Significant numbers of CD8 + cells were recruited into the tumor microenvironment after the NK+mAb9.2.27 treatment (One-Way ANOVA t 7.482, p=0.0001, n=5), (Fig. 4C)

mAb9.2.27 and IFN-γ-activated microglia are more cytotoxic to GBM cells than NK cells.
In order to investigate the mechanisms of the synergistic mAb9.2.27 and NK cell cross-talk with microglia, the cytotoxicity of resting microglia and NK cells with and without mAb9.2.27 was tested in NG2/CSPG4 positive GBM cells in vitro. First, we observed that activated NK cells in contact with GBM cells produced IFN-γ, as revealed by intracellular flow cytometry (Two tailed T-test, NK cells alone vs. NK cells+U87MG p=0.0043, n=5 and vs. HF66 p=0.0095, n=4), (Fig. 5A). This was corroborated by increased IFN-γ and TNF-α released into the supernatant culture medium from NK cells in contact with GBM cells compared to supernatants from tumor cells only, (IFN-γ: One-Way ANOVA 6.54 , p=0.03; n=11; and TNFα: One-Way ANOVA 9.59 , p=0.0083; n=11). Correspondingly, the killing capacity of resting microglia against U87MG GBM cells was increased by overnight activation with IFN-γ and against target pre-incubated with mAb9.2.27 compared to the cytotoxicity of resting microglia with IgG2a isotype control (IC) (One-Way ANOVA, p<0.05, n=5 and p<0.05, n=4 respectively), (Fig. 5B). Interestingly, there was no cumulative effect of IFN-γ and the mAb9.2.27 on the cytotoxicity of microglia. Indeed the cytotoxic capacity of microglia following activation with IFN-γ in absence of mAb9.2.27 was not significantly different from their killing capacity in presence of this mAb, with respectively 37.8 ±4.2 % and 43.2±5.9 % U87MG lysis (Fig. 5B). In contrast, there was no significant difference between cytotoxicity of IL-2 activated NK cells in the presence or absence of mAb9.2.27 against all GBM cell lines tested (Fig. 5C). The cytotoxicity of microglia and activated NK cells was also tested against the HF66 GBM cells lines, recognized by mAb9.2.27 at almost 100 %, and the previous results with U87MG were confirmed. Activation of microglia by IFN-γ , overnight or during 5 days rendered them highly cytotoxic against GBM, whereas culture with GBM conditioned medium promoted the survival of HF66 GBM cells, consistent with microglial M1-and M2-like physiology and phenotype, (Supplementary Fig. 5A and B respectively). As was the case for U87MG, the mAb9.2.27 abrogated the survival of HF66 tumor cells, mediated by M2-like differentiated microglia (Supplemental Fig. 5B). Moreover, NK cells preferentially killed the differentiated M2-like microglia, whereas M1-like microglia was more resistant to NK cell lysis (Supplementary Fig. 5C). Collectively, these data indicate that mAb9.2.27 could mediate ADCC by

mAb9.2.27 and IFN-γ abrogate tumor-promoting effect of tumor-associated macrophage/microglia ex-vivo.
Since our in vivo results demonstrated that mAb9.2.27 monotherapy slowed down tumor growth in the initial phases, we investigated whether mAb9.2.27 could impact on macrophage/microglia function directly. These macrophage/microglia were purified from the brain of P3-30 tumor bearing rats and examined for their cytotoxic capacity ex vivo against P3-30 cells preincubated or not with mAb9.2.27, and after 18 h or 96 h of culture in the presence of IFN-γ. The TAM promoted P3-30 tumor survival ex vivo, indicated by a diminution of its spontaneous death in culture, presented as a negative percentage of lysis (Fig. 5D). P3-30 tumor cells became susceptible to rat TAMs cytotoxicity in the presence of mAb9.2.27 (One-Way ANOVA, p<0.01, n=6), (Fig. 5D).
Remarkably, this effect was reproduced using patient GBM derived TAM (Fig. 5E) defined as CCR2 high (Fig. 5E histogram) indicating the robust capability of mAb9.2.27 to inhibit tumor promoting effect of TAMs. Finally, although rat TAM failed to inhibit the tumor survival after 18 h in culture with IFN-γ, a prolonged activation to 96 h did significantly inhibit tumor survival (One-Way ANOVA, p<0.01, n=4), (Fig. 5 F). The conclusions from the in vivo and in vitro studies are summarized in the schematic figure 5G.

DISCUSSION
It is established that gliomas, like many cancers, develop multiple strategies to evade detection and destruction by the immune system. One of these mechanisms involves the recruitment of macrophages/ microglia and their maintenance in an anti-inflammatory (M2-like) state, through the secretion of tumor-associated factors likeTGF-β or prostaglandin E 2 , which promote tumor growth, migration and angiogenesis [29]. Many new therapeutic strategies have the ambition to skew M2-like TAM to M1-like phenotypes with anti-tumoral properties [30,31]. We investigated a novel combination immunotherapeutic strategy to treat GBM using intratumoral administration of mAb9.2.27 against human NG2/CSPG4 with activated NK cells.
We treated human U87MG and U251-NG2 GBM cell lines xenografted in athymic nude rats. While the treatment significantly increased survival in both models, the U251-NG2 tumors exhibited only partial responses. As mAb9.2.27 labeled almost 3-fold less U251-NG2 cells than U87MG cells, the modest effect of the combination treatment on the U251-NG2 tumors might be expected and reinforces the specificity of the mAb9.2.27 in the therapeutic response. Moreover it emphasizes the importance of the presence of an interaction between mAb9.2.27 and NG2/CSPG4 antigen in future clinical applications. In addition, we used a more relevant model based on rats bearing patient derived GBM biopsy established to be more therapy resistant due to their high cellular heterogeneity. Although several mouse models of brain tumors have been engineered by altering signaling pathways that are disrupted in human gliomas (reviewed in [32]), our patient derived xenografts retain the cellular heterogeneity, invasive, angiogenic and comparative genomic hybridisation (cGH) gene profiles as the original tumor [33]. Our results demonstrated that NK+mAb9.2.27 treatment led to increased animal survival through the recruitment of ED1 positive cells via the choroid plexus into the tumor, in addition to increased numbers of cells with heightened capacity to secrete cytotoxic components such as myeloperoxidase, granzyme B and IFN-γ. The difference in efficacy between the U87 and P3-30 combination treated xenografts might be expected when treating homogenous cell line versus heterogenous, patient-derived GBM xenografts. However, confirming therapeutic efficacy using different model systems adds robustness to our findings. The combination treatment prolonged survival of P3-30 xenografts from 39 to 46 days, representing an increase in survival of 18 %. However, the median survival after GBM treatment in contemporary patient studies is 14.6 months. Thus, this 18 % increase amounts to 2.5 months if extrapolated to a clinical setting. When we consider the very few and small improvements in prognosis for GBM patients over the last 5 decades, this 18 % longer survival might indeed have clinical significance and realistically highlight the challenges of treating human cancers with high degree of heterogeneity.
Furthermore, we could demonstrate that the combination treated animals had elevated concentrations of the pro-inflammatory cytokines IFN-γ and TNF-α in their CSF and diminished levels of anti-inflammatory cytokines such as IL-10 compared with control animals. These results indicated that the intra-tumoral treatment of NK+mAb9.2.27 modified the tumor microenvironment from anti-inflammatory to pro-inflammatory one resulting in tumor regression. This massive recruitment of inflammatory cells into the tumor is reflected by the MR images at 3 and 4 weeks in the U87MG tumors in vivo, where increased lesion size could be demonstrated by T1-weighted MRI. Smith et al. [34] previously reported this phenomenon as the flare effect following intralesional immunotherapy with activated lymphocytes and IL-2, in patients with recurrent high grade astrocytomas [34]. Apparent worsening MRI imaging directly after therapy indicated by increased nodular enhancement, edema, and mass effect [7, 34] characterized the flare effect. This phenomena was also reported following gene therapy [35], convection enhanced delivery of cytokines [36] or after placement of the GliaSite radiation therapy system [37] to treat brain tumors. The flare effect highlights the challenges of using the Macdonald ´s radiological response criteria for assessing therapeutic efficacy after immunotherapy [38]. The immune cell recruitment characterized by hyper-intense lesions on T1weighted MRI might easily be misinterpreted as tumor progression whereas in reality, it represents a pseudoprogression. Other measures of immune responses have been demonstrated as independent predictive factors for favorable immune responses after immunotherapy [39]. Paradoxically, vitiligo is still reported as an adverse event in clinical trials evaluations, however, its presence strongly portends enhanced survival in melanoma patients. More research is required to develop clinically reliable methods for evaluating response to successful immunotherapy.
In contrast to untreated animals, microglia from the NK+mAb9.2.27 combination treated animals up-regulated MHC class II molecules and ED1 expression, rendering them competent for presentation of GBM antigens [40]. In addition, microglia expressing ED2 scavenger receptor were diminished by the combination NK+mAb9.2.27 treatment, indicating a reduction of perivascular microglia postulated to facilitate GBM progression [41]. In parallel the NK+mAb9.2.27 treatment decreased the proportion of macrophages expressing CCR2, a subpopulation already proven to promote GBM invasiveness [42]. These characteristic features of activated cytotoxic cells, as well as the increased survival were abrogated by depletion of circulating macrophages with clodronate treatment. This indicated that the beneficial outcomes were mediated by the recruitment of pro-inflammatory macrophages from the periphery to the tumor. Nevertheless, we cannot exclude the possibility that dendritic cells play a role in the therapeutic efficacy, as liposome-encapsulated clodronate was shown to selectively induce apoptosis of macrophages [43] and phagocytic dendritic cells [44]. Indeed, both cell types share the expression of CD45 high CD11b/c + and only OX-62 can distinguish dendritic cells from macrophages [45][46][47]. Unfortunately, we did not use anti-OX-62 (CD103) mAb in our flow cytometry staining to be able to differentiate MHC class II positive activated macrophages from dendritic cells. Taken together our results indicate that the combination treatment permits the infiltration of ED1 + CCR2 low macrophages, which then differentiate into or promote ED1 + ED2 low MHC II + microglia that favor tumor destruction by creating pro-inflammatory tumor microenvironment.
We propose that implantation of activated NK cells was the decisive element required for the development of the pro-inflammatory environment. Neither NK cells nor mAb9.2.27 monotherapy significantly prolonged survival, but since combination treatment did, we suggest that a synergistic action of NK cells and mAb9.2.27 facilitated tumor regression. Recently, NK cells were demonstrated to mediate the therapeutic efficacy by ADCC of a chimeric antibody directed against an intracellular antigen, phosphatase of regenerating liver-3 (PRL-3), in nude mice. Depletion of NK cells with asialo GM-1 antibodies abolished the therapeutic efficacy [48] underscoring the therapeutic impact of NK cell induced ADCC. Using T1-weighted sequences we established in the present study that the mAb9.2.27 treated tumors exhibited reduced leakage of contrast agent from the tumor vasculature. This effect of mAb9.2.27 is likely due to blocking the angiogenic potential of NG2/CSPG4 expressing pericytes on the tumor vasculature. Although some provocative studies reported that immature GBM stem-like cells are capable of differentiating into functional vascular endothelium and might contribute to tumor angiogenesis [49,50]recent systematic studies demonstrated using chromogenic in situ hybridization on intact tissue that neoplastic cells are a rare component in human GBM microvasculature, do not incorporate into the vessel wall or express CD34 [51], Thus, blocking NG2/ CSPG4 signaling may have decreased the angiogenic capacity of the tumors mediated by pericytes. We have previously shown that NG2/CSPG4 was located on the tumor cells as well as their angiogenic vasculature on pericytes [8,52]. Others have demonstrated that targeting NG2/CSPG4 on these pericytes diminished mouse pathological retinal and corneal angiogenesis [53]. Recently, Wang et al. demonstrated in a xenograft model of breast cancer that the administration of the anti-NG2/CSPG4 mAb225.28 diminished tumor growth, associated with reduced proliferation and vascular density, and increased apoptosis [13]. Moreover, we proved that purified GBM associated macrophage/microglia, from rat and acutely dissociated patient biopsy, were no longer able to promote tumor survival in presence of mAb9.2.27 ex vivo. It would be interesting to study further this capacity of mAb9.2.27 to modulate macrophage polarization and to examine the mechanisms involved. Numerous studies have reported that ligation of the Fc receptor induces the transcription of several genes in macrophages that contribute to subsequent inflammatory and immune responses [54]. The mechanisms of macrophage/microglia ADCC against brain tumors are not well documented, nonetheless a study from the 90's demonstrated microglia ADCC using an anti-EGFR mAb [55]. Furthermore, using an FcγR deficient mouse model of brain metastatic B16 tumor expressing EGFRvIII, Sampson et al. demonstrated that the survival advantage of anti-EGFRvIII mAb treated animals was FcγR dependent [56]. As resident macrophage/microglia are known to express FcγR [57], these findings support our data implicating a subpopulation of macrophage/microglia in the tumor destruction. Since NK cells alone were not able to induce potent ADCC, but were required for the synergistic effects of the combination therapy, we hypothesized that NK cells could be involved in generation of tumor immunoediting, as recently described by O´Sullivan et al. [23]. We demonstrated that activated NK cells were able to secrete TNF-α and IFN-γ in contact with GBM. However, the direct effect of TNF-α expressing GBM cells is unclear since we previously demonstrated that these cells were resistant to TNF-α mediated apoptosis due to augmented PI3K/Akt and NFκB survival signaling [13]. Indeed, the Fernandez-Luna group showed that blockade of NFκB in glioma initiating cells (GICs) with genetic strategies or small molecule inhibitors induced cell cycle arrest, cellular differentiation and senescence [58]. However as these effects were described for these so-called GICs, the implications of their findings remain to be proven since the ontogeny of GBM from GICs and the extent to which GBM cells are capable of undergoing senescence remain controversial. Interferon-γ was previously shown to switch human immunosuppressive TAMs into immunostimulatory cells [59], to upregulate their expression of MHC class II [60] and co-stimulatory molecules [61]. The macrophages were subsequently more phagocytic [62] and released more cytotoxic pro-inflammatory cytokines [59,63], nitric oxide [28] and MPO [64]. In this study we found that IFN-γ may also inhibit the survival promoting effect of TAMs on P3-30 GBM in vitro. However it is difficult to translate this finding in vivo as we may have used a high concentration of IFN-γ in vitro (100 ng/ml) compared to the concentration we found in the CSF (20 pg/ml). Indeed, lower levels of IFN-γ (1 to 20 ng/ml) were shown to promote microglia neurogenesis while high concentrations could be neurotoxic (50 ng/ml) [65,66]. Future studies should address this question. In addition, we showed that NK cells preferentially killed the differentiated M2-like microglia, whereas M1-like microglia were more resistant to NK cell lysis. These results are similar to those from a recent work of Bellora et al. showing that mouse activated NK cells preferentially kill M2 macrophages rather than M1 macrophages [67]. To summarize, our findings demonstrate that activated NK cells could modulated the anti-inflammatory tumor microenvironment to a pro-inflammatory contexture facilitating the recruitment and the differentiation of cytotoxic and inflammatory macrophage/microglia first by secretion of pro-inflammatory cytokines and then by the destruction of anti-inflammatory tumor associated macrophage/microglia. As we utilized athymic nude rats, the possible influence of regulatory T cells could be underestimated. Nonetheless, to develop effective immunotherapy strategies for aggressive GBM tumors necessitates use of preclinical models that exhibit similar cellular heterogeneity, invasive behavior, and express appropriate tumor antigens. Our patient derived P3-30 GBM model fulfills these requirements, however requires propagation in athymic rats. Apart from diminished T cells, the other immune cells are representative and fully functional [68]. The alternative to xenotransplantation would have been to use syngeneic models, where the entire immune system is intact. However, as we are targeting human NG2/CSPG4 with an anti-human mAb9.2.27, this model would not be appropriate due to poor recognition of NG2/CSPG4 from rat origin by mAb9.2.27. Since pro-inflammatory microglia are implicated in several brain diseases [69,70], future work will have to verify the safety and tolerability of our combination treatment prior to clinical application. In a potential future clinical trial, we envision that our combination NK+mAb9.2.27 immunotherapy would be administered locally after surgical debulking and concurrent temozolomide and radiotherapy treatment. In conclusion, administering NK cells as adjuvant treatment might be a useful strategy for increasing the therapeutic efficacy of several mAb-based passive immunotherapies and thereby positively impact the survival of GBM patient.

Tumor cell lines and their culture, Patient Biopsy
The human GBM cell line U251-NG2 overexpressing NG2/CSPG4 [8] was originally derived from U251. The U251, U87MG (American Type Culture Collection, Rockville, MA) and HF66 (Ford Cancer Centre, Detroit, MI) GBM cell lines were propagated as previously described [8]. The GBM biopsies were obtained from surgical resections performed at Haukeland University Hospitals. The local ethical board (REK Vest) and the Data Protection Agency in Norway approved the collection of tumor tissue. Patients gave their informed consent to specimen collection for research purposes and their samples were analyzed anonymously.

Antibody PEGylation
The azide free mAb9.2.27 (10mg/ml) against the cell surface chondroitin sulphate proteoglycan NG2 was a generous gift from Professor Reisfeld (Scripps Research Institute, La Jolla, San Diego, CA). Since the mAb9.2.27 is an anti-human mouse IgG2a isotype known to bind weakly to Fc receptors on effector cells [71], polyethylene glycol (PEG) was conjugated to the mAb in order to increase its interaction with the Fc receptor (FcγRIII) on the effector cells [72]. An aliquot of this antibody was PEGylated by 10 times dilution with a 24 % PEG 20,000 (Sigma Aldrich, Bornem, Belgium) solution (DMEM supplemented with 10 % FBS) for at least 30 min at 4 °C. The pegylated reaction was mixed with a 4 X Tris-HCl 0.1 M Buffer (pH 7) containing 4 % SDS, 40 % Glycerol, 0.001 % Bromophenol Blue and 1 % Beta-Mercaptoethanol and then charged in a 7.5 % acrylamide gel and electrophoresis was performed at 60 mV/gel in SDS-PAGE system. We demonstrated that the mAb9.2.27 was successfully PEGylated by approximately 2 molecules of the 20 kDa PEG, as revealed by Comassie blue staining (Supplementary Fig. 6A) Fig. 6B). We also investigated the ability of the mAb9.2.27 to crosslink the FcγR on purified microglia, the major brain effectors. The PEGylated mAb9.2.27 had greater affinity for microglia compared to NK cells, and other controls (Supplemental Fig. 6C).

Animals and intracranial implantation
8-10 week old immunodeficient athymic nude rats were bred from Hsd:RH-Foxn1 rnu/+ female and Hsd:RH-Foxn1 rnu male breeding pairs. Rats of both sexes were used for tumor implantation and were maintained as previously described [8]. This athymic nude rat strain has autosomal recessive mutation on rnu locus of chromosome 10 [73]. It is deficient for some T-cell subpopulations, but has normal B-cell function [74], increased NK cell number [75] and macrophage cell populations [74]. 7-11 week old immunocompetent LEWIS (LEW/Han TM Hsd; Harlan, Horst, The Netherlands) male rats were also used for in vitro assays. All animal procedures were performed in accordance with protocols approved by The National Animal Research Authority (Oslo, Norway). A burr-hole was made 1 mm posterior to the bregma and 3 mm to the right of the sagittal suture using a micro-drill with a bit diameter of 2.9 mm. A Hamilton syringe with inner diameter of 810 μm was introduced to a depth of 2.5 mm below the brain surface, and the spheroids were slowly injected and the syringe left in place for 3 min before withdrawal.15 tumor spheroids (each containing 30,000 cells) were selected under a stereo light microscope. 25 animals were xenografted with U251-NG2, 28 animals with U87MG and 28 animals with patient GBM biopsy spheroids from patient 3 (P3-30) that had been serially in vivo passaged as previously described [76]. The skin was closed with an Ethilon 3-0 suture. The tumors were allowed to grow for 3 weeks prior to intra-lesional treatment with mAb9.2.27, NK cells from littermates (2 million cells) or combined NK+mAb9.2.27. Antibodies and NK cells were administered using a 26-gauge cannula connected to an infusion catheter of an osmotic mini pump (Alzet Inc.,) or a syringe, respectively, to the same coordinates as were used for the tumor implantation procedure. The animals were sacrificed at the onset of neurological symptoms of lethargy and paralysis in order to obtain overall survival information. An additional cohort of 31 animals xenografted with P3-30 spheroids and treated with NK+mAb9.2.27 with or without liposome-encapsulated clodronate was sacrificed at set time points as controls by CO 2 inhalation and decapitation. Their brains were removed, dissociated into single cell suspensions or fixed in 4 % formalin for the characterization of immune cells infiltrating the brain by flow cytometry and/or immunohistochemistry.

Macrophage depletion by clodronate treatment
Depletion of macrophages was done by intraperitoneal injection (IP) of 1 ml/100 g body weight using 5 mg/ml of liposome-encapsulated clodronate (Encapsula NanoScience). The injections were repeated weekly for 4 weeks after treatment or until development of neurological symptoms. Control rats were injected with same volume of PBS or control liposome at the same time. Depletion of macrophages was confirmed on splenocytes for P3-30 xenografted rats by flow cytometric determination of ED1 and ED2 positive cells (supplementary Fig. 7A and 7B). It should be noted that the IP injection of liposome-encapsulated clodronate could not deplete resident microglia ( Supplementary  Fig. 7C). It was demonstrated by several groups that the microglia depletion could only be achieved by intracranial administration of liposome-encapsulated clodronate [77,78]. Indeed, we drastically decreased the CD11b/ c + CD45 high macrophages recruited to the brain in context of brain tumor, without major modification of CD11b/ c + CD45 low microglia population (Supplementary Fig 7C).

Convection enhanced delivery of mAb9.2.27
Azide free mAb9.2.27 (4 mg/ml in 24 % PEG solution) and IgG2a isotype controls were administered through a 26-gauge cannula connected to an osmotic mini pump (AD01; Alzet Inc., Mountainview, CA). The pumps were installed into the same coordinates as the tumor using a stereotactic frame and infused at a CED rate of 8 µl/h over 24 h.

Histology, immunohistochemistry and quantification.
Brains were formalin fixed, paraffin embedded, and every 20 th , 3-5μm thick tissue section was stained with haematoxylin and eosin. Consecutive sections were immunohistochemically stained with various mAb: anti-IFNγ (AbDSerotec, Oxford, UK) anti-MIB-1 (Ki67, Dako, Glostrup, Denmark), rat specific anti-CD8 (OX-8, AbDSerotec,), anti-CD68 (ED1, AbDSerotec), anti-granzyme B, polyclonal anti-myeloperoxidase, anti-CD3 (Dako), using the ABC method with 3, 3"-Diaminobenzidine (DAB) and H 2 O 2 (DCS, Hamburg, Germany). Percentages of CD8, Granzyme B, and MPO+ immune cells against the negative, haematoxylin counterstained nuclei were quantified in at least 10 fields of view/per section for all the animals in each group. Immunohistochemically stained sections were segmented and quantified with the use of CellProfiler software package [80] and custom made image analysis pipelines. Hematoxylin and DAB staining were separated by the use of the UnmixColors module. Thereafter the CorrectIlluimination module applied an illumination correction in order to adjust uneven light illumination of the image. Hematoxyllin stained nuclei and positive DAB stained cells were identified using the IdentifyPrimaryObjects module with the Otsu Global threshold method. For assessing the precision of the segmentation, the images were saved with an overlay of the segmentation masks. For further details, the specific custom-made pipelines used with CellProfilerare attached in the supplementary files.
ED1 positivity was assessed using morphometry with NIS Elements v 4.0 software and Nikon Eclipse 600 microscope (Nikon). Semi-quantification of DAB staining by morphometry on immunohistochemistry was based on signal intensity of ED1 positivity relative to negative control section that was set as threshold. The threshold was stored and subsequently applied with identical microscope settings for scoring of all tumor sections. Immune positive elements (pixels above threshold) were measured and expressed as area fraction of the visual field for each tumor section. For delimitation of the tumor regions core vs. periphery, 100 mm from the normal brain boundary was defined as periphery for all samples using a grid tool. The fraction of Ki67 positive against Ki67 negative tumor cells (Ki67 labeling index) was quantified in 5 microscopic high power fields (400 X magnification) in all animals in the study group. Detection of apoptotic cells was performed with the terminal deoxynucleotidyltransferase mediated nick endlabeling (TUNEL) assay according to the manufacturer's instructions (Roche Applied Bioscience, Mannheim, Germany) and apoptotic cells were quantified as previously described [79].

NK cell purification and culture
Rat NK cells were purified by negative selection and cultured for 4-5 days with 1000 U/ml of IL-2 as previously described [17].

TAMs and naïve microglia purification and culture
Human biopsy tissue and rat brain were dissected from the meninges and mechanically dissociated as previously described [6]. Rat homogenates were centrifuged and the pellet was suspended in 50 % isotonic Percoll (GE Healthcare, Brabant, Belgium), and then layered on the top of 70 % isotonic Percoll, and then 1XPBS was gently layered on the top. For human TAM isolation we used 40 % and 60 % isotonic Percoll gradient. After centrifugation at 1,200 × g for 35 min, microglia were collected at the interface between the 70 and 50 % for rat or 60 % and 40 % for human, isotonic Percoll. The purity of TAMs was always superior to 90 %. TAMs and naïve microglia were used directly for cytotoxicity assays against U87MG, HF66 and P3-30 or adjusted at 2.10 6 cells/ml and cultured for 18 h or 96 h in complete RPMI-1640, or complete medium supplemented 100 ng/ml of rat IFN-γ (R&D Systems) (M1-like microglia) or GBM supernatant to get M2-like microglia.

Cytotoxicity assays
The cytotoxicity assays were performed as previously described [17]. Briefly, NK cell cytotoxicity was determined after 4 days in culture with 1,000 U/ml of IL-2, and microglia cytotoxicity was performed with resting cells, differentiated microglia after activation with IFN-γ or as TAMs against the U87MG, HF66 and P3-30 target cells labeled with 5µM of CFSE (Sigma Aldrich), pre-incubated or not with 1mg/ml of PEGylated mAb9.2.27. Effector cells (E) were mixed with target cells (T) at E/T ratios of 2/1. After 18 h incubation at 37 °C, cells were analyzed on a FACSCanto flow cytometer. To identify dead cells, 15 µM of the dead cell marker TO-PRO-3 was added. At least 2,500 target cells per sample were examined.

Cytokine measurements
The levels of IL-1α, IL-6, IL-10, IFN-γ, and TNF-α in CSF of untreated control rats, athymic rats treated with mAb9.2.27 alone or NK+mAb9.2.27 and cytotoxicity assay culture supernatants from NK, NK+GBM cells or GBM cells only were measured by CBA (BD Biosciences) according to the manufacturer's protocol. The levels of MCP-1 and MCP-3 in plasma were performed with ELISA kit from eBioscience and USCN Life Science according to manufacturer ´s protocol.

Statistical analysis
All results were expressed as mean± SEM from at least three independent experiments. A probability level of ≤0.05 was considered significant. We used One-Way (with Tukey's Multiple Comparison Test) or Two-Way (with Bonferroni a priori Post-hoc test) analysis of variance (ANOVA) with post-hoc analyses for comparisons of more than 2 groups or variables. Survival was analyzed using the Kaplan-Meier and the log rank test using Graphpad Prism 5.0 (Graphpad Software, La Jolla, CA).