Inhibition of BACE1 Facilitates Macrophage-based Immunotherapy to Suppress Malignant Growth of Glioblastoma

Malignant tumors, including glioblastoma (GBM), contain abundant tumor-associated macrophages (TAMs) that mainly promote tumor growth and therapeutic resistance. Reprograming tumor-promoting TAMs (pTAMs) into tumor-suppressive TAMs (sTAMs) represents an attractive therapeutic strategy. We discovered that inhibition of the β-site amyloid precursor protein cleaving enzyme 1 (BACE1) by MK-8931 potently redirects pTAMs into sTAMs and promotes macrophage phagocytosis of glioma cells to suppress the malignant growth of GBM. Moreover, low doses of radiation markedly enhance TAM inltration and synergize with MK-8931 treatment. BACE1 is preferentially expressed by pTAMs in human GBMs and is required for maintaining pTAM polarization through trans-IL-6/sIL-6R/STAT3 signaling. As several BACE1 inhibitors, including MK-8931, previously developed for Alzheimer's disease, were shown in clinical trials to be safe for humans, repurposing these inhibitors for cancer therapy should be straightforward. Collectively, this study offers a promising therapeutic approach, through inhibition of BACE1, to facilitate the macrophage-based tumor immunotherapy. signaling is required for the maintenance of pTAMs. We demonstrate that targeting BACE1 with the specic inhibitor, MK-8931, effectively converts pTAMs into sTAMs and promotes TAM phagocytosis against glioma cells in vitro and in vivo. Importantly, our preclinical studies demonstrate that BACE1 inhibition by MK-8931 potently suppresses tumor growth of human GBM in xenograft models, suggesting that pharmacological targeting of BACE1 can be used to activate TAMs to exert anti-tumor activity. Furthermore, we found that low doses of radiation markedly induce the inltration of TAMs and synergize with MK-8931 to suppress tumor growth. Thus, we used low-dose radiation to enhance the inltration of TAMs into tumors that contain relatively few TAMs and then treat the tumors with MK-8931 to activate TAM phagocytosis to improve the anti-tumor ecacy. Our studies indicate that targeting BACE1 with MK-8931, alone or in combination with low-dose radiation, is a promising therapeutic approach for treating malignant cancers, including glioblastoma. and after MK-8931 treatment. were maintained until the development of neurological signs the effect of MK-8931 treatment on and bearing were treatment signicantly reduced STAT3 activating phosphorylation in GBM xenografts. Quantication IBA1+/pSTAT3+ double shows that treatment signicantly reduced STAT3 activating in M2 the activated STAT3 (STAT3-C-Flag) with shBACE1 or shNT the transduction, the U937-derived pTAM-like macrophages immunoblot analyses of CD163, ARG1, iNOS, HLA-DR, IBA1, BACE1, STAT3, Flag (STAT3-C) and GAPDH (loading expression of STAT3-C attenuated the pTAM to sTAM switch induced by knockdown. Immunoblot analyses of the pTAM markers (CD163 and ARG1) and sTAM markers (iNOS and HLA-DR) in U937-derived pTAM-like M2 macrophages transduced with the constitutively activated STAT3 (STAT3-C-Flag) or vector (control) in combination with or without MK-8931 treatment. U937-derived pTAM-like macrophages were transduced with STAT3-C-Flag or the vector control and treated with MK-8931 (50 μg/mL) or the vehicle control for three days, and then harvested for immunoblot analyses of CD163, ARG1, iNOS, HLA-DR, IBA1, BACE1, STAT3, Flag (STAT3-C) and GAPDH (loading control). Ectopic expression of STAT3-C Tumor-bearing Mice a, A treatment schedule showing the preclinical trial of MK-8931 treatment in combination with a low dose of irradiation (IR) in GBM xenografts models. Human GSCs CCF-3264 expressing luciferase were transplanted into mouse brains through intracranial injection to establish orthotopic GBM xenografts. Irradiation (2 Gy) was performed on Days 8 and 16. Day 9, were treated with MK-8931 (30 mg/kg) or the vehicle daily oral the was twice

Recent studies indicate that pTAMs play immunosuppressive roles in the TME 14 and have a negative impact on therapy with immune checkpoint inhibitors 15 . Therefore, reprograming pTAMs into sTAMs may not only directly suppress malignant growth but may also offer the opportunity to improve current T cellbased immunotherapy. The goal of our study is to develop macrophage-based immunotherapy for treating malignant tumors, including GBMs, that contain abundant TAMs 16 .
GBM is the most common and fatal brain cancer 17 highly resistant to therapies including current anti-PD1 immunotherapy 18,19 . GBM inevitably recurs after surgical resection and radio-chemotherapy 20,21 . We and the others have demonstrated that pTAMs, glioma stem cells (GSCs), and the interplay between them play vital roles in promoting tumor progression and therapeutic resistance in GBMs 7,22,23 . TAMs and GSCs are often located in the perivascular niche in GBM 10,24 and actively interact at cellular and molecular levels to support malignant growth [10][11][12]25 . As the majority of TAMs in a GBM are pTAMs, which usually lose the ability to phagocytize tumor cells 26 , we hypothesized that redirecting pTAMs into sTAMs to activate macrophage phagocytosis against glioma cells, including GSCs, might suppress malignant tumor growth to effectively improve GBM treatment. To discover small molecules that can potently activate TAM phagocytosis against glioma cells, we designed a cell-based screen, using human iPS cells (iPSC)-derived macrophages and glioma cells including GSCs to identify drug candidates and their potential molecular targets on TAMs. We identi ed MK-8931, a speci c inhibitor of BACE1 (β-site amyloid precursor protein cleaving enzyme 1), as a top candidate to promote macrophage phagocytosis to eliminate cancer cells, and thus de ned BACE1 as a potential therapeutic target to reprogram pTAMs into sTAMs for developing the macrophage-based tumor immunotherapy.
BACE1 is a trans-membrane aspartyl protease that is responsible for the production of amyloid beta peptide (Aβ) in the brains of patients with Alzheimer's disease (AD) [27][28][29][30] . Since its discovery, BACE1 has been widely investigated as a therapeutic target for AD 31 , and several BACE1 inhibitors, including MK-8931, have been tested in clinical trials for AD treatment 32 . In this study, we found that BACE1 is preferentially expressed by pTAMs and that BACE1-mediated trans-IL-6/sIL-6R/STAT3 signaling is required for the maintenance of pTAMs. We demonstrate that targeting BACE1 with the speci c inhibitor, MK-8931, effectively converts pTAMs into sTAMs and promotes TAM phagocytosis against glioma cells in vitro and in vivo. Importantly, our preclinical studies demonstrate that BACE1 inhibition by MK-8931 potently suppresses tumor growth of human GBM in xenograft models, suggesting that pharmacological targeting of BACE1 can be used to activate TAMs to exert anti-tumor activity. Furthermore, we found that low doses of radiation markedly induce the in ltration of TAMs and synergize with MK-8931 to suppress tumor growth. Thus, we used low-dose radiation to enhance the in ltration of TAMs into tumors that contain relatively few TAMs and then treat the tumors with MK-8931 to activate TAM phagocytosis to improve the anti-tumor e cacy. Our studies indicate that targeting BACE1 with MK-8931, alone or in combination with low-dose radiation, is a promising therapeutic approach for treating malignant cancers, including glioblastoma.
MK-8931 (Verubecestat), a non-peptidic class of BACE1 inhibitor originally developed for AD treatment 33 , penetrates the blood-brain barrier (BBB) very well and blocks BACE1 activity e ciently in the brain 34 . MK-8931 is the rst BACE1 inhibitor that has proceeded to a phase III clinical trial for AD patients 31 . Recent clinical studies demonstrated that MK-8931 is generally safe and well tolerated in healthy adults 35,36 and AD patients 34,37,38 . Although targeting BACE1 by MK-8931 is not effective for treating AD patients 37,38 , our study strongly indicates that this drug can be repurposed for tumor immunotherapy as it can effectively redirect pTAMs into sTAMs to eliminate cancer cells in GBM tumors. Thus, this study offers a macrophage-based immunotherapy through BACE1 inhibition with a small molecule that promotes TAM phagocytosis of cancer cells to suppress malignant growth, which may have great potential to improve the treatment of lethal cancers.

Results
Identi cation of the BACE1 Inhibitor MK-8931 as a Potent Activator of Macrophage Phagocytosis of Glioma Cells To identify small molecular modulators that promote phagocytosis of cancer cells by macrophages, we developed a uorescent phagocytosis assay, using GFP-labeled human iPSC-derived macrophages (in green) and tdTomato-expressing human glioma stem cells (GSCs, in red) (Fig. 1a). This assay detects phagocytosis as uorescent inclusion bodies derived from GSCs (tdTomato + ) within macrophages (GFP + ) 39 . To obtain GFP-expressing macrophages, we transduced iPSCs with a GFP expression construct, then derived M2-like macrophages from GFP + iPSCs according to established protocols 11,12,40 (Supplementary Fig. 1a,b). We con rmed that GFP was constitutively expressed during macrophage differentiation and that the iPSC-derived macrophages expressed total macrophage markers, including CD11b and IBA1, and M2 markers, including ARG1 and FIZZ1 ( Supplementary Fig. 1c). Thus, small molecules that activate macrophages (GFP + ) to engulf GSCs (tdTomato + ) can be detected with a uorescent microscope (Fig. 1a). We initially screened an inhibitor library (SelleckChem) and some known drugs that displayed excellent BBB permeability and low toxicity in phase II/III clinical trials for other diseases, including AD. We obtained seven "hits" and further identi ed ve BACE1 inhibitors, including MK-8931 as the most promising candidates. The screening assay showed that MK-8931 treatment promoted phagocytosis of the iPSC-derived macrophages against GSCs (Fig. 1b,c). Moreover, MK-8931 treatment also augmented phagocytosis of bone marrow-derived macrophages (BMDMs) against GSCs (Fig. 1d,e), suggesting that MK-8931 is a potent activator of macrophage phagocytosis of cancer cells. As MK-8931 was originally developed as a BACE1-speci c inhibitor for AD patients 33,41 , we then con rmed that BACE1 was expressed by the iPSC-derived macrophages but not by the iPSCs (Fig. 1f). These data indicate that BACE1 inhibition by MK-8931 promotes macrophage phagocytosis of glioma cells in vitro.
We next sought to determine whether BACE1 inhibition by MK-8931 promotes TAM phagocytosis in vivo in GSC-derived GBM xenografts that contain abundant TAMs 10-12 . We initially examined the expression of BACE1 in TAMs in GBM xenografts derived from human GSCs (CCF-3264 or CCF-DI315) by double immuno uorescent staining, nding that BACE1 was co-expressed usually with the total TAM markers CD11b and IBA1 and the pTAM markers CD163 and FIZZ1, but rarely with the sTAM markers HLA-DR and CD11c ( Supplementary Fig. 2), suggesting that BACE1 is mainly expressed by pTAMs in the GBM xenografts. We then treated mice bearing intracranial GBM xenografts with MK-8931 (30 mg/kg) or the vehicle control once daily by oral gavage for two weeks and harvested the tumors to analyze the effect of MK-8931 treatment on macrophage phagocytosis. To detect TAM phagocytosis in tumor sections of GBM xenografts, we labeled the TAMs with anti-IBA1 antibody (in green) and the human glioma cells with antibody against the human-speci c antigen TRA-1-85 (in red), nding that MK-8931 treatment resulted in a signi cant increase of inclusion bodies derived from the glioma cells (red) within TAMs (green) (Fig. 1g,h), indicating that BACE1 inhibition by MK-8931 activates TAMs to engulf glioma cells in GBM tumors. Collectively, these data indicate that BACE1 inhibition by MK-8931 promotes macrophage phagocytosis of glioma cells in vitro and in vivo.

MK-8931 Treatment Potently Inhibits Tumor Growth and Extends the Survival of Mice Bearing GBM Xenografts
We next investigated whether MK-8931 treatment could suppress GBM growth and impact survival. To address this important point, we treated mice bearing GBM xenografts derived from the luciferaseexpressing human GSCs (CCF-3264 or CCF-DI315) with MK-8931 (30 mg/kg/daily) or a vehicle control by oral gavage and then monitored tumor growth by using the In Vivo Imaging System (IVIS) (Fig. 2a). Bioluminescent imaging demonstrated that MK-8931 treatment potently inhibited GBM tumor growth  Fig. 3g,h). These cellular alterations induced by MK-8931 treatment in the GBM xenografts were unlikely to be the direct effects of MK-8931 on glioma cells, including GSCs, because MK-8931 in cell culture did not affect proliferation, sphere formation, cell viability, or apoptosis of GSCs or non-stem glioma cells in vitro ( Supplementary Fig. 4). Collectively, our preclinical data demonstrate that targeting BACE1 by MK-8931 potently suppresses malignant growth of GBMs. Moreover, these data suggest that BACE1 inhibition by MK-8931 not only promotes TAM phagocytosis to eliminate glioma cells but also modulates the tumor microenvironment and thus potently inhibits tumor growth.

BACE1 Is Required for the pTAM Maintenance and Controls the TAM Phenotype Switch
Because the cellular alterations induced by MK-8931 treatment are similar to those de ned in tumors when pTAMs are inhibited [10][11][12]42 , we were prompted to interrogate whether BACE1 plays a functional role in the maintenance of pTAMs. We generated pTAM-like M2 macrophages from the PMA-primed U973 cells (called M0 macrophages) by using the cytokines IL4, IL10 and TGF-β according to an established protocol 11,12 (Supplementary Fig. 5a), and found that the pTAM-like M2 macrophages derived from U937 monocytes expressed high levels of BACE1 at both the mRNA and protein levels, as measured by RT-PCR and immunoblot analyses ( Supplementary Fig. 5b,c). We then examined the effects of BACE1 disruption by shRNA on pTAM-like macrophages, showing that BACE1 knockdown markedly reduced the expression of the pTAM markers CD163 and ARG1, while expression of the total TAM marker IBA1 was not affected ( Supplementary Fig. 5d). This result was further con rmed by using the BACE1 inhibitor MK-8931 ( Supplementary Fig. 5e). As BACE1 disruption or inhibition did not impact the expression of the total TAM marker IBA1 ( Supplementary Fig. 5d,e), we speculated that targeting BACE1 might promote the phenotypic transition of pTAMs into sTAMs. To address this possibility, we knocked BACE1 down in pTAM-like M2 macrophages derived from U937 monocytes, nding that BACE1 disruption indeed induced the expression of sTAM markers, including HLA-DR and iNOS (Fig. 3a), while expression of the pTAM markers CD163 and ARG1 was dramatically reduced after BACE1 disruption ( Fig. 3a and Supplementary   Fig. 5d). Moreover, functional inhibition of BACE1 by MK-8931 in pTAM-like macrophages effectively induced expression of the sTAM markers HLA-DR and iNOS, but inhibited expression of the pTAM markers CD163 and ARG1 in a dose-and time-dependent manner (Fig. 3b,c). These data indicate that BACE1 disruption or inhibition redirects pTAMs into sTAMs in vitro.
Next we studied the role of BACE1 in the TAM phenotype switch in vivo by examining pTAM and sTAM populations in control or MK-8931-treated GBM xenografts. MK-8931 treatment strongly reduced the density of pTAMs, as marked by CD163 + /IBA1 + or FIZZ1 + /IBA1 + cells (Fig. 3d,e and Supplementary   Fig. 6a,b). In contrast, the density of sTAMs, marked by HLA-DR + /IBA1 + or CD11c + /IBA1 + cells, was strikingly increased in tumors treated with MK-8931, relative to controls (Fig. 3f,g and Supplementary   Fig. 6c,d). These data suggest that BACE1 inhibition can reprogram pTAMs into sTAMs in vivo. To further address whether this effect of BACE1 disruption functionally impacts tumor growth, we established a GBM xenograft model by co-transplanting monocyte-derived pTAM-like M2 macrophages expressing shBACE1 (BACE1 shRNA) or shNT (non-targeting shRNA) with GSCs expressing luciferase, as illustrated in Supplementary Fig. 7a. Co-transplantation of GSCs with monocyte-derived pTAM-like macrophages expressing shNT substantially promoted GBM tumor growth ( Supplementary Fig. 7b,c) and reduced the survival of mice relative to control mice implanted with GSCs alone ( Supplementary Fig. 7d), consistent with a tumor-promoting role of pTAMs 11 . In contrast, co-transplantation of GSCs with monocyte-derived pTAM-like macrophages expressing shBACE1 (disrupting BACE1) did not augment GBM tumor growth, but rather suppressed malignant growth relative to GSCs alone ( Supplementary Fig. 7b,c). Consequently, the survival of mice implanted with GSCs and the pTAM-like macrophages expressing shBACE1 was extended relative to mice implanted with GSCs alone or GSCs plus monocyte-derived pTAM-like macrophages expressing shNT ( Supplementary Fig. 7d), indicating that BACE1 disruption in TAMs functionally reprograms pTAMs into sTAMs, resulting in anti-tumor activity. Collectively, these data indicate that BACE1 is required for maintaining pTAMs, and that BACE1 inhibition results in phenotypic switch of pTAMs into sTAMs, leading to suppression of tumor growth, and highlighting the role of BACE1 as a critical modulator in regulating TAM function in malignant tumors. BACE1 Maintains pTAMs by Catalyzing the Shedding of IL-6R and Activating Trans-IL-6/sIL-6R/STAT3 Signaling Since BACE1 is required for maintaining pTAMs, we next sought to understand the molecular mechanisms by which it drives pTAM polarization. Because STAT3 and STAT6 are key transcriptional regulators in M2 macrophage polarization 43,44 , we interrogated their potential role in the BACE1-mediated maintenance of pTAMs. Surprisingly, we found that the activating phosphorylation of STAT3 (pSTAT3-Y705) but not STAT6 (pSTAT6-Y641) was substantially down-regulated after BACE1 disruption by shBACE1 (Fig. 4a) or inhibition by MK-8931 (Fig. 4b). Consistently, MK-8931 treatment in vivo profoundly reduced the number of pSTAT3 + TAMs (identi ed as pSTAT3 + /IBA1 + ) and total pSTAT3 + cells in GBM xenografts ( Fig. 4c-e). To further determine whether BACE1 functions through STAT3 signaling to maintain pTAMs, we examined whether ectopic expression of a constitutively active STAT3 (STAT3-C) could rescue the attenuated pTAM phenotype induced by BACE1 disruption or inhibition. Indeed, ectopic expression of STAT3-C restored the expression of the pTAM markers CD163 and ARG1 in monocytederived macrophages (Fig. 4f,g). Importantly, ectopic expression of STAT3-C abolished the increased expression of the sTAM markers iNOS and HLA-DR that was induced by BACE1 disruption or inhibition in the macrophages (Fig. 4f,g). Thus, ectopic expression of STAT3-C restored the pTAM phenotype impaired by BACE1 disruption or inhibition. Collectively, these data demonstrate that BACE1 maintains the pTAMs mainly through STAT3 activation, indicating that BACE1-mediated STAT3 activation is required for the pTAM maintenance.
To understand how BACE1 regulates STAT3 activation in pTAMs, we analyzed potential substrates of BACE1 ( Supplementary Fig. 8a), nding that ten of them are involved in the regulation of STAT3 activity and eight are reported to play a role in macrophage polarization ( Supplementary Fig. 8b). We further investigated four substrates, including the IL-6 receptor (IL-6R) that functionally overlap in regulating STAT3 activity and macrophage polarization. To this end, we found that BACE1 mediates the shedding of IL-6R by functioning as a transmembrane protease, as BACE1 disruption or inhibition reduced not only the amount of the N-terminal fragment of IL-6R (called the soluble IL-6R, or sIL-6R) in the conditioned media of macrophages ( Fig. 4h,i, top panels) but also reduced the membrane-bound C-terminal fragments of IL-6R (CTF) in macrophage lysates (Fig. 4h,i), while the amount of full-length IL-6R (FL) was increased in lysates of macrophages upon BACE1 disruption or inhibition (Fig. 4h,i). These data indicate that the sIL-6R released into the extracellular space is generated by the BACE1-mediated cleavage of membrane-bound full length IL-6R. It is well known that sIL-6R binds to IL-6 and forms an IL-6/sIL-6R complex that retains the capacity to bind to glycoprotein 130 (gp130) and thus to activate STAT3 signaling 45 . This signaling mechanism, termed the trans-IL-6/sIL-6R/STAT3 pathway, plays a key role in the tumor microenvironment, driving tumor growth [46][47][48] . Importantly, IL-6/STAT3 activation induces the polarization of pTAMs and promotes cancer progression 49,50 . In fact, approximately 70% of the secreted IL-6 binds to sIL-6R in the circulation 51 . Thus, sIL-6R, functioning as a carrier of IL-6, is able to prolong the half-life of IL-6 in vivo and to stabilize IL-6 signaling 52 , resulting in a dramatic increase of STAT3 activation 53 . Consistently, we demonstrate that sIL-6R enhances IL-6-induced STAT3 activation in macrophages (Fig. 4j). Collectively, our data indicate that BACE1-mediated shedding of full length IL-6R generates sIL-6R, which promotes STAT3 activation induced by IL-6 to maintain the polarization of pTAMs (Fig. 4k).

Low Doses of Radiation Enhance TAM In ltration and Effectively Synergize with MK-8931 Treatment to Inhibit Malignant Growth
Our preclinical study has demonstrated that targeting BACE1 can effectively reprogram pTAMs into sTAMs to suppress GBM tumor growth. To improve the e cacy of macrophage-based immunotherapy through BACE1 inhibition for some tumors that may contain relatively few TAMs, we sought to nd an effective way to enhance the in ltration of macrophages into tumors. Because low-dose irradiation (IR) promotes macrophage in ltration into tumors 54 , we speculated that IR-enhanced TAM in ltration could synergize with MK-8931 treatment to activate macrophage phagocytosis and convert pTAMs into sTAMs more effectively. To test this hypothesis, we irradiated mouse brains bearing GSC-derived xenografts with a low dose of IR (2 × 2 Gy) to allow in ltration of more TAMs into the tumors and then treated the mice with MK-8931 (Fig. 5a). The combined treatment achieved the strongest inhibition of GBM tumor growth, relative to treatment with IR or MK-8931 alone (Fig. 5b,c). As a consequence, the combined treatment conferred the longest survival extension among four groups of mice (Fig. 5d). To determine the properties and phenotypes of TAMs in the GBM xenografts treated with IR, MK-8931, or IR plus MK-8931, we examined TAM density and subtypes in tumors from four groups of mice with different treatments. Total TAM density (IBA1 + cells) was remarkably increased by treatment with a low dose of IR ( Supplementary   Fig. 9a-c), consistent with previous reports that IR enhances TAM in ltration into tumors 54,55 . Surprisingly, the majority of TAMs induced by the low dose of IR were pTAMs but not sTAMs, as shown by the expression of CD163 but not HLA-DR ( Supplementary Fig. 9a-c). However, MK-8931 treatment effectively converted these pTAMs (CD163 + ) into sTAMs (HLA-DR + ) in the GBM xenografts treated with IR plus MK-8931 ( Supplementary Fig. 9a-c), suggesting that BACE1 inhibition by MK-8931 is able to redirect IRenhanced pTAMs into tumor-suppressive macrophages. Moreover, immunostaining analyses of the apoptotic marker cleaved caspase 3 in the GBM tumors demonstrated that the combined treatment resulted in signi cantly more apoptotic cell death than treatments with MK-8931 or IR alone ( Supplementary Fig. 9d,e). Importantly, the apoptosis induced by the combined treatment (25.9 ± 3.1%) was notably greater than the sum of apoptosis induced by MK-8931 alone (10.4 ± 0.9%) or IR alone (3.1 ± 0.7%) ( Supplementary Fig. 9e). Collectively, these data demonstrate that the enhanced TAM in ltration stimulated by a low dose of IR effectively synergizes with MK-8931 to suppress malignant growth and thus increases the survival of the tumor-bearing animals, highlighting the promising therapeutic potential of the macrophage-based tumor therapy using MK-8931 in combination with a low dose of IR.

BACE1 Is Highly Expressed by pTAMs in human GBMs and Predicts Poor Prognosis
To interrogate the clinical signi cance of BACE1 in human GBMs, we examined its expression in human GBM surgical specimens by double immuno uorescent staining, nding that BACE1 was detected in the fraction of total TAMs marked by IBA1 or CD11b (Fig. 6a). Quantitative analysis indicated that approximately 60% of the IBA1 + TAMs and about 67% of the CD11b + TAMs express BACE1 (Fig. 6b).
Further examination demonstrated that BACE1 was mainly co-expressed with the pTAM marker CD163 or FIZZ1 (Fig. 6c). Quanti cation showed that the majority of pTAMs express BACE1, as approximately 83% of the CD163 + TAMs and 86% of the FIZZ1 + TAMs are BACE1-positive (Fig. 6d). In contrast, BACE1 was rarely co-expressed with the sTAM marker HLA-DR or CD11c (Fig. 6e,f). The preferential expression of BACE1 in pTAMs was consistently found in all human GBM specimens (12 cases) examined. To further investigate the clinical relevance of BACE1 expression in pTAMs in human GBM tumors, we analyzed the relationship between BACE1 expression and the survival of GBM patients in several databases, including the Cancer Genome Atlas (TCGA), Rembrandt, Gravendeel and LeeY, nding an inverse correlation between BACE1 expression and patient survival in all these databases ( Supplementary Fig. 10a-d). GBM patients with higher BACE1 expression levels in their tumors clearly had a worse survival ( Supplementary   Fig. 10a-d), indicating that BACE1 expression predicts poor prognosis. The inverse correlation between BACE1 expression in pTAMs and the prognosis is consistent with the fact that pTAMs support tumor growth and malignant progression in human GBMs. Collectively, these data demonstrate that BACE1 is preferentially expressed by pTAMs in most human GBM tumors and predicts a poor prognosis for these patients, validating the therapeutic promise of targeting BACE1 by an inhibitor such as MK-8931. In addition, CD47 inhibition alone modestly inhibits GBM progression 61 but leads to therapeutic resistance in other cancer models 62 . In this preclinical study, we demonstrate that reprograming pTAMs into sTAMs by inhibiting BACE1 with the speci c inhibitor MK-8931 potently promotes TAM phagocytosis against glioma cells and effectively inhibits tumor growth, extending the survival of mice bearing GBM tumor (Fig. 7, graphic abstract). As BACE1 inhibition by MK-8931 has been demonstrated to be safe in healthy adults 35,36 and to be well-tolerated by AD patients 37,38 , and since MK-8931 can penetrate the BBB very well 34 , redirecting pTAMs into sTAMs by BACE1 inhibition with MK-8931 may overcome the shortcomings of targeting CSF-1R and offer a macrophage-based promising immunotherapy to improve treatment of malignant tumors, including GBM. BACE1 is a type I transmembrane β-secretase that plays important roles in physiological and pathophysiological processes 63,64 . BACE1 cleaves amyloid precursor protein to cause accumulation of Aβ production in the brains of AD patients [27][28][29][30] . However, BACE1 de ciency is well-tolerated in knockout mice, without obvious effects on development, behavior, or fertility 63 , indicating that targeting BACE1 should not result in side effects. Interestingly, microglia in BACE1-de cient mice also increased phagocytosis toward cellular debris after nerve damage 65,66 . However, to the best of our knowledge, the functional signi cance of BACE1 in regulating TAMs has not been reported. In this study, we demonstrated that BACE1 plays a critical role in maintaining pTAMs in tumors. BACE1 inhibition potently converts pTAMs into sTAMs to promote TAM phagocytosis against cancer cells in vivo. Importantly, BACE1 inhibition did not cause the death of pTAMs but redirected them into sTAMs that have tumorsuppressive roles. Thus, targeting BACE1 represents an attractive therapeutic strategy to improve tumor immunotherapy.

Discussion
Our preclinical studies indicate that MK-8931 is a promising drug to modulate TAM function for reclaiming anti-tumor activities of macrophages in tumors. To enhance the therapeutic e cacy of MK-8931, we found that a low dose of irradiation (IR) remarkably augmented TAM in ltration and effectively synergized with MK-8931. Radiotherapy is a standard treatment for GBM 17 , but this tumor is highly resistant to IR, partially due to the population of GSCs that resist to conventional treatments and contribute to tumor recurrence 22,23 . In addition, IR often triggers an in ammation response and remodulate the tumor microenvironment to induce therapeutic resistance 67,68 . Our study demonstrates that a low dose of IR markedly enhances TAM in ltration into GBM tumors. Surprisingly, the increased TAMs induced by the IR in tumors were mainly pTAMs, although a previous in vitro study showed that IR can induce the polarization of human and murine monocytes toward M1-like macrophages 69 . It is possible that IR affects macrophages in vitro and in vivo in a different manner and that low and high doses of IR differentially impact macrophage polarization. Interestingly, another study showed that IR could induce BACE1 expression 70 , which should promote the maintenance of pTAMs. Although treatment with a low dose of IR alone does not signi cantly impact tumor growth, it provides a powerful tool to enhance TAM in ltration into tumors, which allows MK-8931 treatment to redirect the increased pTAMs into more sTAMs to better suppress tumor growth. This therapeutic strategy is particularly important for those solid tumors that contain relatively few TAMs. Most malignant tumors contain abundant TAMs and treatment with MK-8931 alone should effectively promote TAM phagocytosis to engulf cancer cells, but certain types of tumors in some patients may have fewer TAMs. In either situation, the combination of MK-8931 with a low dose of IR should enhance therapeutic e cacy. Because some tumors may not contain enough in ltrating T cells to facilitate current immunotherapy with immune checkpoint inhibitors such as anti-PD1 antibody, the combination of low dose of IR with MK-8931 may provide an alternative strategy to overcome the poor response of some malignant tumors to current immunotherapy.
The molecular mechanisms underlying the polarization and maintenance of TAMs were poorly understood. In this study, we found that BACE1 mediates the critical shedding of the membrane-bound full length IL-6R by functioning as a transmembrane protease to generate the soluble form of IL-6R (sIL-6R), which activates the trans-IL-6/sIL-6R/STAT3 signaling to maintain the polarization of pTAMs (Fig. 4k). Our in vivo study further demonstrated that inhibiting BACE1 with MK-8931 suppressed STAT3 activation in TAMs, resulting in reduced number of pTAMs and increased density of sTAMs, as well as enhanced macrophage phagocytosis, to effectively inhibit tumor growth. STAT3 is a critical transcription factor that plays multiple roles in tumor development and malignant progression 71 . Our previous studies demonstrated that STAT3 hyper-activation mediated by the bone marrow X-linked kinase (BMX) is required for maintaining the self-renewal and tumorigenic potential of GSCs in GBM 72,73 . Interestingly, GSCs secrete periostin to recruit monocyte-derived TAMs into GBM tumors 10 . In turn, in ltrating TAMs support the maintenance of GSCs through PTN-PTPRZ1 signaling in GBM 11 . Because GSCs play crucial roles in tumor growth and malignant progression, including invasion 74,75 , angiogenesis 76,77 , pericyte generation 76 , blood-tumor barrier (BTB) formation 78 , and therapeutic resistance 22,79 , the reduction of GSCs caused by the decreased pTAMs induced by BACE1 inhibition with MK-8931 may also partially contribute to the suppression of tumor growth in vivo.
Immunotherapy is a promising therapeutic option for cancer treatment, but the majority of solid tumors, including GBM, respond poorly to current immune checkpoint blockade, partially due to the insu cient T cell in ltration, poor delivery of checkpoint inhibitors such as anti-PD1 to tumors, and the development of resistance to the immune stimulation 80 . However, because most malignant tumors contain abundant TAMs 1,2 , and we have identi ed an effective approach to enhance macrophage in ltration into tumors by using a low dose of radiation, macrophage-based immunotherapy through BACE1 inhibition with a small molecular modulator such as MK-8931 has several obvious advantages: (1) This therapy reprograming pTAMs into sTAMs not only inhibits TAM tumor-supportive roles but also promote macrophage phagocytosis to engulf cancer cells. The double hit effectively suppresses tumor growth and malignant progression.
(2) As current immunotherapy such as CAR-T or anti-PD1 is too expensive for most patients, using a small molecular modulator such as MK-8931 to facilitate TAM-based immunotherapy will provide a much more economic but effective approach to improve the survival of cancer patients; (3) Redirecting pTAMs into sTAMs by MK-8931 should effectively re-modulate the tumor immune microenvironment to overcome resistance to other therapies; (4) The combined low-dose IR and MK-8931 treatment can be broadly used for most malignant tumors containing abundant or relatively fewer TAMs; (5) Because all BACE1 inhibitors, including MK-9831, developed for AD clinical trials display great ability to penetrate the BBB, immunotherapy with a BACE1 inhibitor will overcome the BBB issue, as the BBB negatively impacts other therapies including immune checkpoint inhibitors in malignant brain tumors; and (6) As BACE1 inhibition by MK-8931, AZD3293, E2609, or CNP520 has been shown to be very safe for patients in AD clinical trials 32 , repurposing these BACE1 inhibitors for macrophage-based immunotherapy should be straightforward. We predict that this alternative immunotherapy through BACE1 inhibition has tremendous potential to effectively improve the survival of patients with malignant cancers, including glioblastoma and brain metastases.

Cells
Cells were cultured in a humidi ed incubator at 37ºC with 5% CO 2 and atmospheric oxygen. All cells used in this study were consistently con rmed to be free from mycoplasma by using a MycoFluor™ Mycoplasma Detection Kit (ThermoFisher, M7006). 293FT cells from Clontech (632180) were maintained in DMEM medium supplemented with 10% (v/v) fetal bovine serum (FBS, ThermoFisher, 10437-036). Human iPSCs from ALSTEM (iPS11) were grown in mTeSR1 medium (StemCell Technologies, 85850).
Human iPSC-derived monocytes and macrophages were maintained in X-VIVO TM 15 medium (Lonza, 04-418Q). Human U937 cells from ATCC (CRL-1593.2 TM ) were maintained in RPMI 1640 medium with 10% (v/v) FBS. Bone-marrow derived macrophages (BMDMs) were derived from mice according to an established protocol 81 and cultured in RPMI 1640 medium with 10% (v/v) FBS. Human GSCs (CCF-3264 and CCF-DI315) were derived from human primary GBMs as previously described 72,82,83  Recombinant Human EGF was from GoldBio (1150-04-100). According to the manufacturer's instructions, all the recombinant proteins were prepared at a 1,000× concentration as stock solutions and stored at -80ºC until use. All other chemicals and reagents were purchased from Sigma-Aldrich.

Preparation of Bone Marrow-derived Monocytes (BMDMs)
BMDMs were isolated and cultured according to an established protocol 81 . In brief, mouse bone marrow cells were collected by ushing the femurs and tibias with sterile PBS and then treated by red blood cell lysis buffer to remove red blood cells. The cells were re-suspended and cultured in RPMI 1640 medium with 10% FBS and M-CSF (100 ng/mL) for seven days to differentiate into BMDMs.
Identi cation of Potential Drugs to Activate Macrophage Phagocytosis To screen for potential small molecules that activate phagocytosis of iPSC-derived macrophages against cancer cells, GFP + iPSC-derived macrophages (5 × 10 4 cells) were seeded in each well of 24-well plates and treated with small molecules from an inhibitor library (SellckChem) or existing drugs that had been used in clinical trials for other diseases for two days. After washing, the cells were maintained in RPMI 1640 medium for two hours. Next, tdTomato-expressing glioma stem cells (GSCs: CCF-3264) were added to each well and co-incubated with the macrophages in RPMI 1640 medium with 10% FBS for another two hours. After incubation, the co-cultures were washed three times with warm RPMI 1640 medium to remove free cancer cells, and uorescent and phase images were captured with a uorescent microscope.
Phagocytosis was detected and measured as inclusion bodies of cancer cells (in red) within macrophages (in green) according to a published study 39 .
To detect the MK-8931-activated phagocytosis of bone marrow-derived macrophages (BMDMs) against cancer cells, BMDMs were treated with DMSO (Control) or MK-8931 (50 μg/mL) for two days and prestained with CellTracker™ Green CMFDA Dye (1 μM, ThermoFisher, C2925). After washing, the BMDMs were maintained in RPMI 1640 medium for two hours. Next, tdTomato-expressing CCF-3264 GSCs (2 × 10 5 cells) were added to each well and co-incubated with the labelled BMDMs in RPMI 1640 medium with 10% FBS for another two hours. After incubation, the co-cultures were washed several times with warm RPMI 1640 medium to remove free cancer cells, and the uorescent images were captured with a uorescent microscope.
For infection, cells were treated with lentivirus at a multiplicity of infection (MOI) of 1.

Generation of Stable Cell Lines
To generate tdTomato-expressing stable glioma cells, GSCs (CCF-3264) were transduced with tdTomato by means of lentiviral infection for 12 hours. Two days post-infection, the cells were treated with neomycin (500 μg/mL, Santa Cruz, sc-29065A) for seven days to select stable clones. The expression of tdTomato was con rmed by uorescence microscopy.
To generate luciferase-expressing stable glioma cells, GSCs (CCF-3264 or CCF-DI315) were transduced with re y luciferase by means of lentiviral infection for 12 hours. Two days post-infection, the cells were treated with puromycin (2 μg/mL, Fisher Scienti c, BP2956100) for seven days to select stable clones.
The luciferase activity was con rmed by using the Luciferase Assay System (Promega, E1500).
To establish stable U937 cells expressing shBACE1 or shNT, the cells were transduced with shBACE1 or shNT by means of lentiviral infection. Two days post-infection, the cells were treated with puromycin (2 μg/mL) for seven days to select stable clones. Immunoblotting was used to test the knockdown e cacy of BACE1 in the U937 stable cells.
Derivation of pTAM-like M2 Macrophages from U937 Cells U937-derived pTAM-like M2 macrophages were prepared according to an established protocol 11,12 . In brief, cells grown in a 10-cm tissue culture dish were primed with PMA (5 nM) for two days to produce M0

Establishment of Glioblastoma Xenografts (PDXs) and Drug Treatment in vivo
To establish xenografts for in vivo studies, intracranial transplantation of GSCs into the brains of NSG mice was performed as described previously [10][11][12]22,83 . Brie y, GSCs (CCF-3264 or CCF-DI315) expressing luciferase were injected into the right cerebral cortex at a depth of 3.5 mm. For the co-transplantation experiments, human glioma stem cells (GSCs, CCF-DI315) expressing luciferase in combination with or without the monocyte-derived macrophages expressing shBACE1 (shBACE1: #1 or #2) or non-targeting control (shNT) were co-injected into NSG mouse brains through intracranial injection (GSCs:Macrophages = 1:2). In vivo bioluminescent imaging (IVIS) was performed twice per week to monitor the tumor growth, using the Spectrum CT Imaging System (PerkinElmer), before and after treatment. For drug treatment, a stock solution of MK-8931 at 100 mg/mL in DMSO was diluted in 0.5% (w/v) methylcellulose (Sigma-Aldrich, M0512) to 6 mg/mL 34 . Mice bearing the xenografts were treated with MK-8931 (30 mg/kg) or the control (DMSO) once daily by oral gavage for two weeks or until the humane endpoint was reached. To collect mouse brains bearing tumors, cardiac perfusion with PBS and 4% PFA (Electron Microscopy Sciences, 15714) was performed. The brains were xed and sectioned for further immuno uorescent, histochemical and histological analyses.
Irradiation on Intracranial GBM Xenografts (PDXs) Irradiation (IR) was performed with a Pantek X-ray irradiator (once per week) at a low dose (2 Gy). To protect the mice and limit the side effects of irradiation, anesthetized mice were covered with a lead plate and only the tumor implantation sites were exposed to the fractioned radiation. Mice were sacri ced at the indicated time points or upon the appearance of neurological signs. Mouse brains bearing the tumors were collected for further analyses through cardiac perfusion with PBS and 4% PFA (Electron Microscopy Sciences, 15714). Only animals who suffered accidental death (for example, due to infection or intracranial injection) were excluded from the data analysis.

Immuno uorescent Analysis
Immuno uorescent staining of tumor tissues or cells was performed as described in our previous publications [10][11][12]22,83 . In brief, tumor sections or cells were xed with 4% PFA for ten minutes, washed three times with cold PBS for ve minutes each, permeabilized by 0.5 % (v/v) triton X-100 (Bio-Rad, 1610407) for ten minutes, and blocked with 3% (w/v) BSA (Sigma-Aldrich, A7906) in PBS for one hour at room temperature. For Ki67 and pSTAT3 staining, antigen retrieval was performed by incubating the sections in boiled antigen retrieval buffer (Vector Laboratories, H-3300) for 15 minutes before permeabilization. Primary antibodies were added to the sections or cells and incubated overnight at 4ºC. Primary antibodies used for immuno uorescence in this study were diluted as described below: anti- Fluor ® 568 Goat Anti-Armenian Hamster (Abcam, ab175716, 1:200), and Alexa Fluor ® 488 Goat Anti-Rabbit (Invitrogen, A-11008, 1:200). After washing three times with cold PBS for ve minutes each, the sections or cells were counterstained by DAPI (Cell Signaling, 4083, 1:5000) and sealed with mounting medium (Sigma-Aldrich, F4680). Finally, images were captured by a uorescence microscopy (Leica DM4000) and further analyzed with ImageJ software (https://imagej.nih.gov/).

Immunoblot Analysis
Immunoblot analysis was performed as previously described [10][11][12]22,83,85 . Brie y, cells were lysed with RIPA buffer [50 mM TrisHCl (pH7.4), 150 mM NaCl, 2 mM EDTA, 1% (v/v) NP-40, 0.1% (w/v) SDS, protease inhibitor (one tablet per 10 mL of RIPA buffer, Roche)] for 20 minutes on ice. For the blots of phosphorylated protein, phosphatase inhibitor (one tablet per 10 mL RIPA buffer, Roche) was used. The cell lysates or conditioned medium (some experiments) were collected and subjected to SDS-PAGE and blotted onto PVDF membranes (ASI, XR730). After blockade with 5% (w/v) non-fat milk (RPI, M17200) in TBST, the membranes were incubated with primary antibodies overnight at 4°C. After the incubation of the rst antibodies, the membranes were washed three times with TBST for ten minutes each. Then, the membranes were incubated with the second HRP-linked antibodies in the 5% milk for one hour at room temperature. The second HRP-linked antibodies were anti-mouse IgG (Cell signaling, 7076, 1:5000), antirabbit IgG (Cell signaling, 7074, 1:5000), and anti-goat IgG (Santa Cruz, sc-2354, 1:5000). After washing three times with TBST for ten minutes each, signals on the membranes were developed in the ECL HRP substrates (Advansta, K-12045) and images were acquired by a molecular imager (Bio-Rad, Universal Hood II) and analyzed by the Image Lab software (Bio-Rad).

Statistical Analysis
All bar graphs represent mean ± SEM unless otherwise indicated. For the survival analysis and correlation between gene expressions in glioblastoma patients, the data were provide by TCGA and downloaded from GlioVis. For the survival analysis of glioblastoma patients, the patients were divided into Bace1 high and Bace1 low groups with GlioVis (http://gliovis.bioinfo.cnio.es/) and Kaplan-Meier survival curves were generated. The log-rank survival analysis was performed with GraphPad Prism 5 software (https://www.graphpad.com/) to compare signi cance among different groups. All quantitative data presented were mean ± SEM from at least 3 repeats or samples per data point. Experimental details such as number of animals or cells and experimental replication were provided in the gure legends. Data inclusion/exclusion criteria was not applied in this study. Signi cant differences were determined between two groups using the Student's t test or among multiple groups using one-way ANOVA and statistical signi cance was set at p < 0.05. Figure 1 Identi cation of the BACE1 Inhibitor MK-8931 as a Potent Activator of Macrophage Phagocytosis of Glioma Cells a, An illustration of the cell-based uorescent screening assay. Brie y, human iPS cells (iPSC)-derived macrophages (GFP+, in green) were seeded in wells of 24-well plates and treated with different small molecules (in triplicate) for two days. Then, tdTomato+ glioma stem cells (GSCs, in red) were added to each well and co-incubated with the macrophages for two hours. After washing away free   Signaling a, Immunoblot analyses of STAT3 and STAT6 activation (pSTAT3-Y705 and pSTAT6-Y641) in U937-derived pTAM-like M2 macrophages with or without BACE1 disruption. U937-derived pTAM-like macrophages were transduced with shNT (non-targeting control shRNA) or shBACE1 through lentiviral infection for three days and then harvested for immunoblot analyses of pSTAT3-Y705, STAT3, pSTAT6-6-mediated STAT3 activation in the U937 macrophages. k, A schematic illustration shows that BACE1 maintains pTAMs through the trans-IL-6/sIL-6R/STAT3 signaling. The membrane IL-6R is cleaved by BACE1 to generate the soluble IL-6R (sIL-6R) in pTAMs. Then the sIL-6R captures the IL-6 to form an IL-6/sIL-6R complex that binds to gp130 to induce the dimerization of gp130, resulting in increased STAT3 activation to maintain pTAMs.