Research ArticleCancer

Dynamic changes in glioma macrophage populations after radiotherapy reveal CSF-1R inhibition as a strategy to overcome resistance

See allHide authors and affiliations

Science Translational Medicine  15 Jul 2020:
Vol. 12, Issue 552, eaaw7843
DOI: 10.1126/scitranslmed.aaw7843

Helping macrophages battle glioblastoma

Although most cancer therapies target the tumor cells themselves, the tumor microenvironment also plays a key role in patients’ outcomes. Akkari et al. examined the interaction between radiation therapy, a mainstay of treatment for the brain tumor glioblastoma, and innate immune cells in the tumor microenvironment. The authors focused on two immune cell types, brain-resident microglia and monocyte-derived macrophages, and identified radiation-induced changes in protein abundance and gene expression in both cell types. On the basis of these findings, the authors selected a pharmacological treatment that enhanced the efficacy of radiotherapy in mouse models of glioblastoma.

Abstract

Tumor-associated macrophages (TAMs) and microglia (MG) are potent regulators of glioma development and progression. However, the dynamic alterations of distinct TAM populations during the course of therapeutic intervention, response, and recurrence have not yet been fully explored. Here, we investigated how radiotherapy changes the relative abundance and phenotypes of brain-resident MG and peripherally recruited monocyte-derived macrophages (MDMs) in glioblastoma. We identified radiation-specific, stage-dependent MG and MDM gene expression signatures in murine gliomas and confirmed altered expression of several genes and proteins in recurrent human glioblastoma. We found that targeting these TAM populations using a colony-stimulating factor–1 receptor (CSF-1R) inhibitor combined with radiotherapy substantially enhanced survival in preclinical models. Our findings reveal the dynamics and plasticity of distinct macrophage populations in the irradiated tumor microenvironment, which has translational relevance for enhancing the efficacy of standard-of-care treatment in gliomas.

INTRODUCTION

Reciprocal interactions between cancer cells and noncancerous immune and stromal cells in the tumor microenvironment (TME) not only support tumor development and progression but can also contribute to intrinsic and acquired resistance to anticancer targeted therapies (1, 2). Tumor-associated macrophages (TAMs) are one of the most abundant TME cell types across multiple cancers, and their accumulation correlates with poor patient prognosis in many tumors including gliomas (3–6).

The most common and aggressive primary brain tumor in adults is glioblastoma. Responses to postoperative standard-of-care therapy, which include ionizing radiation (IR) often in combination with temozolomide chemotherapy, are transient, and tumors inevitably recur. Median survival is just over 14 months after diagnosis, with 5-year survival rates of ≤5% (7, 8). Radiotherapy efficacy can be impaired by several tumor cell–intrinsic resistance pathways (9). Although not as intensively investigated to date, tumor cell–extrinsic mechanisms of resistance may also be important in blunting the effects of radiotherapy (10). On a population-wide level, TAMs modulate the efficacy of different therapeutic agents in multiple cancers (11), including tyrosine kinase inhibitors in preclinical glioma models (12). Consequently, developing therapeutic strategies to target TAMs in combination with standard-of-care treatment or emerging molecular targeted therapies is important for potential clinical translation (11, 13). Macrophage-targeting agents, including several small-molecule inhibitors of colony-stimulating factor–1 receptor (CSF-1R), have been used to inhibit TAMs in mouse glioma models (12, 14–16) and are under clinical evaluation in patients with glioblastoma, in combination with either standard-of-care therapy (NCT01790503) or immune checkpoint inhibitors (NCT02829723).

In recent years, the heterogeneity in TAM populations with regard to their developmental origin/ontogeny and diverse education patterns has been increasingly appreciated (13). Several preclinical studies identified distinct phenotypes and functions for macrophages originating from the tissue-resident population compared to those recruited via the blood in multiple cancer types, including brain tumors (17–21). In murine and human gliomas, the pool of TAMs is composed of brain-resident microglia (MG), which develop from fetal yolk sac progenitors, and peripherally recruited monocyte-derived macrophages (MDMs), which originate from adult hematopoietic progenitors (6, 17, 22–24). The ontogeny of MG and MDMs remains clearly distinguishable in high-grade glioblastoma (6, 17). We previously showed that because of their ontogenically poised ability to integrate glioma-derived signals, MG adopt a pro-inflammatory signature, whereas MDMs acquire an education signature dominated by wound-healing genes (17). These analyses led to the definition of ontogeny-specific and glioma education–specific signatures in MG and MDMs (6, 17).

In this study, we have specifically examined the relative contributions and dynamics of TAM populations during the time course of glioblastoma IR response, emergence of resistance, and tumor recurrence. Using transgenic mouse glioma models, we found that MG and MDMs display both stage- and subtype-dependent phenotypic heterogeneity during the initial response to IR and in subsequent recurrent tumors. In addition, we identified a convergence of MG and MDM transcriptional signatures specifically at recurrence after IR. Together, our results support the notion that therapeutic strategies which alter acquired IR-dependent characteristics may be more efficient than simple depletion of total TAMs or TAM subpopulations.

RESULTS

Radiotherapy elicits a transient antitumor response and a progressive accumulation of TAMs in gliomas

We first analyzed the initial response to fractionated radiotherapy (2 Gy/day for 5 days; 10 Gy in total) in two genetically engineered mouse models (GEMMs) of glioma. Both involve expression of platelet-derived growth factor–β in nestin-positive progenitor cells (RCAS-hPDGF-B; Nestin-Tv-a), concomitant with either short hairpin–mediated knockdown of p53 (termed PDG-p53 KD herein) or in an Ink4a/Arf-deficient background (termed PDG-Ink4a/Arf KO) (Fig. 1A) (14, 16, 17, 25). We additionally used the Flt3:Cre;Rosa26:mTmG myeloid cell–lineage tracing model in which peripherally recruited MDMs, which develop from Flt3+ hematopoietic progenitors, are green fluorescent protein–positive (GFP+) and resident MG, which originate from yolk sac precursors, are GFP−, to generate Fcre;PDG-p53 KD glioma-bearing mice (17).

Fig. 1 Radiotherapy elicits a transient antitumor response associated with a progressive accumulation of TAMs in gliomas.

(A) Experimental design: Gliomas were initiated as described in Supplementary Materials and Methods to generate the PDG-Ink4a/Arf KO and PDG-p53 KD models. Four and a half to 6 weeks after glioma initiation, animals underwent MRI to assess tumor volume and were randomly assigned to control (treatment-naïve) or fractionated IR (5 × 2 Gy; 5d-IR) groups. Mice were euthanized 24 hours after the fifth and final dose of 2 Gy IR. (B) Representative images of T2-weighted MRI scans of control and 5d-IR–treated high-grade PDG-Ink4a/Arf KO gliomas at trial initiation (day 0) and endpoint (day 5). Dashed lines indicate regions of interest used to calculate tumor volume. (C) Change in tumor volume, calculated by MRI, after 5 days in control (n = 7) and 5d-IR–treated (n = 16) PDG-Ink4a/Arf KO mice. (D) Percentage of proliferating Ki67+ cells and apoptotic cleaved caspase 3–positive (CC3+) cells and (E) Iba1+ macrophages relative to total DAPI+ (4′,6-diamidino-2-phenylindole–positive) cells during the course of the 5-day fractionated IR regimen in PDG-Ink4a/Arf KO mice (n = 4 to 6 per group). Samples were collected 6 and 24 hours after 1 × 2 Gy IR dose, 24 hours after 2 × 2 Gy IR dose (48 hours), 24 hours after 3 × 2 Gy IR dose (3 days), and 24 hours after 5 × 2 Gy IR dose (5 days). (F) Experimental design: PDG-Ink4a/Arf KO or PDG-p53 KD glioma-bearing animals were randomly assigned to control or IR (5 × 2 Gy) groups, with biweekly follow-up MRI as depicted (with no further treatment). Mice were euthanized when symptomatic. (G) Kaplan-Meier survival curve of mice bearing high-grade PDG-Ink4a/Arf KO tumors, comparing 5 × 2 Gy IR treatment versus control (log-rank Mantel-Cox test, P = 0.0034). (H) Kaplan-Meier survival curve of mice bearing high-grade PDG-p53 KD tumors, comparing 5 × 2 Gy IR treatment versus control (log-rank Mantel-Cox test, P = 0.0001). Graphs show means + SEM. P values were obtained using unpaired two-tailed Student’s t test (C to E) and log-rank Mantel-Cox test (G and H). **P < 0.01, ***P < 0.001, and ****P < 0.0001.

Consistent with previous reports (26, 27), tumor growth was only transiently slowed by the 5-day IR (5d-IR) time course in both PDG-p53 KD and Fcre;PDG-p53 KD GEMMs (fig. S1, A and B), whereas in PDG-Ink4a/Arf KO mice, gliomas regressed as determined by magnetic resonance imaging (MRI) (Fig. 1, B and C). IR led to reduced Ki67+ glioma cell proliferation during the 5d-IR time course, which was most pronounced at 48 hours of treatment, and the number of cleaved caspase 3–positive (CC3+) cells gradually increased over time (Fig. 1D and fig. S1D). Accumulation of tumor-associated recruited macrophages and resident MG (collectively referred to here as “total TAMs,” identified by the pan-macrophage marker Iba1) was observed at later time points during 5d-IR treatment (Fig. 1E and fig. S1E).

We investigated the long-term effect of fractionated radiotherapy in the PDG-Ink4a/Arf KO and PDG-p53 KD glioma GEMMs to assess the emergence of recurrence (Fig. 1F). After the 5d-IR–responsive phase (fig. S1, F and G), PDG-Ink4a/Arf KO and PDG-p53 KD tumors entered a quiescent, controlled period of variable length, during which the tumor volume was stabilized and no sign of regrowth was detected by MRI (fig. S1, H and I). Consequently, IR treatment resulted in increased survival compared to nonirradiated controls: median survival of 10.42, 8.95, and 8.57 weeks for PDG-Ink4a/Arf KO, PDG-p53 KD, and Fcre;PDG-p53 KD models, respectively, compared to 7.85, 6.57, and 5.85 weeks for the corresponding treatment-naïve controls (Fig. 1, G and H, and fig. S1J). However, tumors eventually regrew in 100% of animals, highlighting the modest effect of radiotherapy in prolonging the life span of glioma-bearing animals, independent of genetic background.

Relative proportions of MG and MDMs are altered in IR-recurrent gliomas

Immune cells, including TAMs, can interfere with the efficacy of different therapies such as chemotherapy, radiation, and kinase inhibitors (1, 2, 12). To address the potential role of myeloid cells in promoting tumor regrowth after IR, we first analyzed TAMs in IR-treated gliomas for up to 5 days (5d-IR), at 21 days after IR (21d-IR; a time point showing no tumor regrowth by MRI; fig. S1, H and I) and in recurrent tumors (IR-Rec) (Fig. 1, G and H, and fig. S1, H and I). Although we observed an increase in total TAM content at several time points during IR treatment (Figs. 1E and 2A and fig. S1E; 3 and 5 days), this was not the case in 21d-IR or recurrent tumors (Fig. 2A).

Fig. 2 Relative proportions of MG and MDMs are altered in recurrent gliomas after IR.

(A) Quantitation of Iba1+ total TAMs relative to DAPI+ cells in stained tissue sections from untreated control, 5d-IR (collected after 5 × 2 Gy IR), 21d-IR (collected 21 days after start of fractionated IR), and recurrent (IR-Rec) PDG-Ink4a/Arf KO tumors. (B and C) Flow cytometry quantitation of CD49d− MG and CD49d+ MDMs in control, 5d-IR, and IR-Rec gliomas in the (B) PDG-Ink4a/Arf KO or (C) PDG-p53 KD model. (D) Flow cytometry quantitation of GFP− TdTom+ MG and GFP+ TdTom− MDMs in control and IR-Rec PDG-p53 KD gliomas generated in Flt3:Cre;Rosa26:mTmG;Nestin-Tv-a mice. (E) Experimental design: PDG-Ink4a/Arf KO or PDG-p53 KD mice bearing high-grade tumors were assigned to either IR or IR + α-CD49d treatment groups. Kaplan-Meier survival curves for (F) the PDG-Ink4a/Arf KO model (P = 0.0025) and (G) PDG-p53 KD mice (P = 0.0004). (H) Quantitation of Ki67+ glioma cells (Iba1−) relative to DAPI+ cells in stained tissue sections from control or from 21d-IR and recurrent tumors emerging in IR- or IR + α-CD49d–treated PDG-Ink4a/Arf KO mice. (I and J) Flow cytometry quantitation of CD69+ T cells in control and recurrent tumors emerging in IR-, α-CD49d–, or IR + α-CD49d–treated (I) PDG-Ink4a/Arf KO and (J) PDG-p53 KD glioma-bearing mice. (K) Flow cytometry quantitation of granzyme B–positive (GrzB+) T cells in control and recurrent tumors emerging in IR-, α-CD49d–, or IR + α-CD49d–treated PDG-p53 KD glioma-bearing mice. Graphs show means + SEM. P values were obtained using unpaired two-tailed Student’s t test (A to C and H to K) and log-rank Mantel-Cox test (F and G). *P < 0.05, **P < 0.01, and ***P < 0.001. ns, nonsignificant.

We then examined whether the ratio of tissue-resident MG and tumor-infiltrating MDMs was altered during the course of IR response and recurrence. We assessed the number of Iba1+GFP+ MDMs and Iba1+TdTomato+ MG in FCre; PDG-p53 KD tumors treated with fractionated radiotherapy (Fig. 1F and fig. S1J). We detected abundant GFP+ MDMs in recurrent tumors (fig. S2A), which were positive for integrins α4/CD49d and αL/CD11a (fig. S2B), two MDM-specific markers not expressed in resident MG in treatment-naïve glioma (17).

We found that the baseline ratio of MDMs to MG in glioblastoma varied according to the GEMM analyzed. MDMs comprised 30 to 40% of total TAMs in control PDG-p53 KD tumors, but only 10 to 20% of TAMs in control PDG-Ink4a/Arf KO gliomas (Fig. 2, B and C). Analysis of 5d-IR and 21d-IR tumors showed that MDM:MG ratios remained unchanged in comparison to control gliomas (Fig. 2, B and C, and fig. S2C), despite increased TAM numbers in 5d-IR tumors (Fig. 2A). In IR-Rec gliomas, however, MDMs were increased compared to untreated controls or 5d-IR tumors, constituting ≥50% of total TAMs, and MG abundance decreased (Fig. 2, B and C). By immunofluorescence staining of total TAM (Iba1+) and MG (Iba1+P2ry12+) in tissue sections, we confirmed that this change was specific to recurrent gliomas and not observed in 21d-IR tumors (fig. S2C). We further validated these results in Fcre;p53-KD gliomas (Fig. 2D). Alteration of the MDM:MG ratio in recurrent PDG-Ink4a/Arf KO tumors was associated with increased numbers of neutrophils (Ly6CintLy6Ghigh) and inflammatory monocytes (Ly6ChighLy6G−) (fig. S2D). Total CD3+ lymphocytes remained unaltered by IR treatment (fig. S2D), and the CD4/CD8 T cell content and activation phenotype showed no changes in IR-Rec tumors compared to control gliomas (fig. S2E).

Inhibition of MDM infiltration delays glioma recurrence after IR

Infiltration of monocyte-derived cells from the circulation can promote neovascularization in recurrent tumors after IR (10, 28–30). Therefore, to address the potential contribution of MDM accumulation to glioma recurrence in the GEMMs used herein, we sought to block the ability of MDM precursors to enter the brain. Neutralization of CD49d, a member of the leukocyte very late antigen-4 (VLA-4) integrin complex (31, 32), limits immune cell entry into the brain and ameliorates symptoms in different neuropathies (33, 34). We therefore opted to use a CD49d-neutralizing antibody (α-CD49d) to interfere with peripherally derived MDM recruitment during IR response and recurrence in glioma GEMMs while leaving the resident MG population unperturbed (Fig. 2E and fig. S2F). Circulating monocytes (CD45+CD11b+Ly6Chigh) were not affected by α-CD49d treatment (fig. S2G), whereas MDM numbers were reduced in recurrent tumors after IR [assessed by the independent MDM-specific marker CD11a (17)], though with differential efficiency in the two glioma GEMMs (fig. S2, H and I). For comparison, total lymphocyte numbers, the CD4/CD8 ratio, total neutrophil, and monocyte content were not altered in IR + α-CD49d–recurrent tumors (fig. S2, J and K).

As a monotherapy, α-CD49d treatment had no effect on prolonging animal life span. However, in combination with IR, α-CD49d treatment resulted in increased survival in PDG-Ink4a/Arf KO mice [median survival of 11 weeks versus 9.92 weeks for IR + immunoglobulin G (IgG) control; Fig. 2F] and in the PDG-p53 KD GEMM (median survival of 12.92 weeks versus 9.2 weeks for IR + IgG control; Fig. 2G). Whereas tumor cell proliferation was unchanged in recurrent PDG-Ink4a/Arf KO tumors treated with IR + α-CD49d, Ki67+ cells were reduced at the 21-day time point in this treatment group compared to 21d-IR tumors (Fig. 2H), consistent with previous reports using distinct approaches to inhibit monocyte recruitment to tumors after IR (28, 29).

In both models, IR + α-CD49d treatment increased the expression of the differentiation marker CD69 in tumor-infiltrating CD4+ and CD8+ T cells (Fig. 2, I and J), and reduced the neutrophil immunosuppressive phenotype (fig. S2L). A potential explanation for the extended life span of PDG-p53 KD mice treated with IR + α-CD49d is that α-CD49d reduces MDMs more effectively in this model compared to PDG-Ink4a/Arf KO gliomas (fig. S2, H and I). In addition, the number of CD8+ T cells expressing granzyme B (GrzB) increased in PDG-p53 KD recurrent gliomas emerging after IR + α-CD49d treatment compared to IR alone (Fig. 2K). Together, these results indicate that blocking MDM infiltration in the context of IR partially relieves immunosuppressive properties of the glioma TME. However, these cumulative changes in the immune landscape only moderately delayed the emergence of recurrent tumors after IR, particularly in the PDG-Ink4a/Arf KO model.

CSF-1R inhibition enhances the initial response to radiotherapy

Given the modest effect of blocking MDM infiltration alone, we evaluated whether targeting both MDMs and MG would further improve the treatment efficacy of IR. We used BLZ945, a blood-brain barrier–permeable CSF-1R tyrosine kinase inhibitor (14, 16), to specifically target these TAM populations together. We first performed short-term combination trials with IR [2 Gy/day for 5 days, concurrent with BLZ945 (200 mg/kg per day) for 5 days; fig. S3A] and assessed tumor response by MRI (Fig. 3, A and B, and fig. S3, B and C). In PDG-Ink4a/Arf KO and PDG-p53 KD glioma-bearing mice, combination treatments were more effective than either therapy alone (Fig. 3B and fig. S3C). Tumor regression was associated with increased apoptosis and decreased glioma cell proliferation in the combination treatment group compared to each monotherapy (Fig. 3C). Monocyte and lymphocyte numbers were unchanged in short-term 5d-IR + BLZ945–treated gliomas compared to either monotherapy, whereas neutrophil numbers were increased (fig. S3, D to F). No changes in total TAMs were observed between IR- and IR + BLZ945–treated animals, with both displaying higher TAM numbers compared to control gliomas (fig. S3G), and MG remained the dominant population in the early response phase of IR + BLZ945 (fig. S3, H and I).

Fig. 3 TAM targeting by CSF-1R inhibition enhances initial response to radiotherapy.

(A) Representative images of T2-weighted MRI scans of PDG-Ink4a/Arf KO gliomas from the different treatment groups at trial initiation (day 0) and endpoint (day 5), as presented in fig. S3A. Dashed lines indicate regions of interest used to calculate tumor volume. (B) Change in tumor volume in control (n = 7), IR (n = 16), BLZ945 (n = 9), and IR + BLZ945 (n = 18) 5 day–treated PDG-Ink4a/Arf KO tumors. (C) Percentage of proliferating Ki67+ cells and apoptotic CC3+ cells relative to total DAPI+ cells determined by immunofluorescence staining of tissue sections. Samples were collected from control tumor-bearing mice or after 5 days of treatment with IR, BLZ945, or IR + BLZ945. (D) Representative immunofluorescence images of γH2aX staining in control gliomas or 5 days after treatment with IR, BLZ945, or IR + BLZ945. Scale bar, 50 μm. (E) Quantitation of number of γH2aX+ cells from (D) and average number of γH2aX+ foci per tumor cell in the different treatment groups. Graphs show means + SEM. P values were obtained using unpaired two-tailed Student’s t test (B, C, and E). *P < 0.05, **P < 0.01, and ***P < 0.001.

Given the superior efficacy of combination treatment, we investigated whether the effects of IR on glioma cells were altered by TAM targeting. We assessed the DNA damage response and found increased γH2aX+ tumor cells and γH2aX+ foci in the IR + BLZ945 group compared to either monotherapy (Fig. 3, D and E). We used cell-based assays incorporating control tumor cells (derived from treatment-naïve PDG-Ink4a/Arf KO gliomas) and macrophages (fig. S3J) to explore how BLZ945 affected DNA damage–sensing signaling pathways in IR-treated tumor cells. In response to a single 10-Gy radiation dose, γH2aX and p53 phosphorylation increased in tumor cells (fig. S3, K and L). By contrast, in cocultures, macrophages blunted the activation of these DNA damage sensors in tumor cells, whereas CSF-1R inhibition restored their sensitivity to IR (fig. S3, K and L). Similar effects were observed when tumor cells were incubated with macrophage-conditioned medium (CM) for 24 hours before irradiation. Macrophage CM inhibited the induction of γH2aX, phospho-Ataxia Telangiectasia and Rad3-related protein (ATR), and phospho-p53 in tumor cells, without altering Ataxia Telangiectasia Mutated (ATM) activation, and these effects were maintained 6 hours after IR treatment (fig. S3M), indicating that macrophage-secreted factors can dampen the DNA damage response after IR.

We assessed the abilities of MG and MDM to modulate proliferation of either control or recurrent tumor cells by measuring S-phase entry in different coculture conditions. In monoculture, proliferation of tumor cells isolated from PDG-Ink4a/Arf KO control or IR-Rec gliomas was comparable, and only control tumor cells responded to 10 Gy IR by a decrease in proliferation (fig. S3N). MG promoted control tumor cell proliferation in a more pronounced manner compared to MDM in coculture. In control cells treated with IR and in IR-recurrent tumor cells, both TAM populations promoted resistance and enhanced proliferation (fig. S3, O and P). When BLZ945 was added to these cocultures, the pro-proliferative effect of MG on control tumor cells and of both TAM populations on recurrent cells was diminished (fig. S3, O and P). Together, these results support our in vivo findings that both MG and MDMs limit the efficacy of radiotherapy and support glioma recurrence after IR.

Combined radiation and acute CSF-1R inhibition delays glioma relapse

Given the substantial effect of short-term (5 days) BLZ945 in improving the initial response to radiation (Fig. 3, A and B, and fig. S3, B and C), we next examined the effects on long-term survival. We designed preclinical trials in which BLZ945 was administered concurrently with IR (2 Gy/day for 5 days), followed by an additional 7 days of BLZ945 alone (12 days of BLZ945 in total) (Fig. 4A), to ensure that CSF-1R inhibition is sustained in the context of IR-induced effects on the TME. Twelve days of BLZ945 alone (which we term “acute” treatment herein) did not extend the life span of PDG-Ink4a/Arf KO animals despite the initial tumor regression observed (Fig. 4, B and C, and fig. S4A). This is in contrast to continuous long-term treatment with BLZ945 as a monotherapy, which resulted in a pronounced survival advantage (14, 16), emphasizing the necessity of daily BLZ945 dosing to maintain monotherapy efficacy. Nonetheless, combined acute BLZ945 treatment with IR enhanced tumor regression and led to increased survival compared to either monotherapy (median survival of 13.86 weeks versus 10.2 or 9.07 weeks; Fig. 4, B and C, and fig. S4, A to C). Analyses of proliferation at 21 days after treatment showed that acute IR + BLZ945 decreased Ki67+ glioma cells compared to IR alone (Fig. 4D).

Fig. 4 Acute CSF-1R inhibition combined with irradiation delays glioma recurrence.

(A) Experimental design of combination trials in PDG-Ink4a/Arf KO model. Mice were treated with vehicle (n = 8), BLZ945 (n = 15), IR (n = 21), or IR + BLZ945 (n = 15). BLZ945 was administered for 12 days in total (acute treatment). Mice were monitored by regular MRI. (B) Representative plots for longitudinal assessment of long-term individual tumor volume progression by MRI in PDG-Ink4a/Arf KO mice treated as in (A) with vehicle (n = 4), IR (5 × 2 Gy, n = 10), BLZ945 (12 days, n = 7), or IR + BLZ945 (n = 8). (C) Kaplan-Meier survival curves of PDG-Ink4a/Arf KO mice treated with vehicle, IR, BLZ945, or IR + BLZ945. (D) Quantitation of Ki67+ glioma cells (Iba1−) by immunofluorescence staining relative to total DAPI+ cells in control, 5d-, 21d-, and recurrent tumors after acute treatment with IR, BLZ945, and IR + BLZ945 in the PDG-Ink4a/Arf KO glioma model. (E) Flow cytometry quantitation of CD49d− MG and CD49d+ MDMs in vehicle-treated gliomas and at endpoint for recurrent tumors emerging after acute treatment with IR, BLZ945, and IR + BLZ945. Graphs show means + SEM. P values were obtained using log-rank Mantel-Cox test (C) or unpaired two-tailed Student’s t test (D and E). *P < 0.05, **P < 0.01, and ***P < 0.001.

Although the number of TAMs was comparable in recurrent tumors from the different treatment groups (fig. S4, D and E), a decrease in MDMs and an increase in MG were observed in IR + BLZ945–treated gliomas compared to IR alone (Fig. 4E and fig. S4F). Analysis of additional immune cell populations showed that monocyte numbers decreased in recurrent gliomas from the IR + BLZ945 group compared to IR, whereas neutrophils and lymphocytes were unchanged (fig. S4, G to I). Together, these results indicate that even transient targeting of MG and MDMs with CSF-1R inhibition delays recurrence after IR, resulting in improved efficacy compared to simply blocking MDM infiltration as in the combination of α-CD49d with IR (Fig. 2F). These results prompted us to specifically explore the transcriptional changes induced by IR in MG and MDMs, which were mitigated by CSF-1R inhibition, thereby limiting tumor recurrence.

Alterations in MG and MDM transcriptional programs after IR treatment are reversed by CSF-1R inhibition

Profiling of MDM and MG transcriptional networks has shed light on how their distinct origins affect their subsequent education and activation profiles in treatment-naïve gliomas (17, 18). To identify changes in different TAM populations that potentially mediate their protumorigenic effects in the initial phase after IR, we first analyzed MG (CD49d− CD11a−) and MDMs (CD49d+ CD11a+) sorted from treatment-naïve, age-matched tumors (control) and 5d-IR–treated PDG-Ink4a/Arf KO tumors and performed RNA sequencing (RNA-seq; Fig. 5A). We identified a set of 64 genes up-regulated in both MG and MDMs in 5d-IR PDG-Ink4a/Arf KO tumors compared to control tumors, herein termed the “TAM-IR” transcriptional signature (fig. S5, A and B, and data file S1A). Hypergeometric optimization of motif enrichment (HOMER) transcription factor landscape analysis (35) and gene ontology analysis revealed an enrichment in p53 motifs and p53 transcriptional targets, associated with a decrease in cell cycle–regulated genes (fig. S5A and data file S1B). MDMs showed more transcriptional plasticity than resident MG in 5d-IR–treated tumors, with 710 genes versus 125 genes specifically up-regulated compared to control tumors, respectively (fig. S5B). Although BLZ945 alone did not alter the expression of TAM-IR genes, this signature was diminished in MG and MDMs isolated from 5 day–treated IR + BLZ945 tumors (fig. S5C and data file S1C). These results indicate that the initial IR-mediated expression changes in TAM populations can be blocked by CSF-1R inhibition.

Fig. 5 MG and MDM transcriptional programs are altered during IR treatment response and recurrence.

(A) Experimental design of time course for MG and MDM isolation in the PDG-Ink4a/Arf KO model. (B) Heatmap depicting row-normalized log2 gene expression values for the indicated genes in MG and MDMs from control, 5d-IR–treated, BLZ945-treated, and IR + BLZ945 5 day–treated PDG-Ink4a/Arf KO tumors (data file S1C). (C) qRT-PCR validation analyses of a subset of M2-like genes in CD49d− MG (top) and CD49d+ MDMs (bottom) sorted from PDG-Ink4a/Arf KO tumors treated with 5d-IR (n = 5), BLZ945 (n = 4), and IR + BLZ945 (n = 4), as depicted in fig. S3A. The results are shown relative to the expression in control tumors set to 1 (dashed lines). (D) t-SNE plot depicting global gene expression differences between MG and MDMs in control (Cont), 5x2 Gy IR-treated (5d-IR), and recurrent (IR-Rec) PDG-Ink4a/Arf KO tumors (see data file S2 for gene lists). (E) Tukey box plots depicting single-sample gene set enrichment analysis scores for MG and MDMs in the indicated treatment groups in PDG-Ink4a/Arf KO gliomas. “Ontogeny” and “education” signatures were derived from (17). (F) Volcano plot analyses depicting log2 fold change (x axis) versus significance [−log10(P value)] of up-regulated genes in control (blue circle) and IR-Rec tumors (purple circle) in MG and MDMs. Graphs show means + SEM. P values were obtained using unpaired two-tailed Student’s t test (C). *P < 0.05, **P < 0.01, and ***P < 0.001. Student’s t test was performed to assess significant differences between groups in (E). No pairwise comparisons other than the ones depicted were found to be significantly different. *P < 0.05.

Genes associated with DNA damage response and cell cycle, together with several genes previously identified in glioma TAMs and associated with alternative macrophage activation (14, 16), were enriched in the TAM-IR signature and down-regulated in the IR + BLZ945 combination (Fig. 5B and fig. S5A). Conversely, the effect of IR on the wounding-associated macrophage response (including Ccl22, Ccl17, Chil3l3, and Retnla) was largely unchanged (fig. S5D). Quantitative reverse transcription polymerase chain reaction (qRT-PCR) analyses of independent sets of MG and MDMs collected 5 days after IR treatment confirmed that BLZ945 reversed the IR-induced up-regulation of a subset of TAM-IR alternative activation genes in both populations (Fig. 5C).

Collectively, these results show that IR treatment rapidly induces a heterogeneous protumorigenic phenotype in MG and MDMs. CSF-1R targeting blocked the IR-specific signature in both cell types, including their alternative activation, which correlated with increased glioma cell apoptosis, DNA damage sensitization, and tumor debulking (Fig. 3). Together, our findings indicate that the radiation response can be enhanced by macrophage targeting early in the treatment course.

MG and MDM gene signatures are dynamically altered in recurrent tumors while maintaining their ontogeny-based identities

Given the changes in TAM education in the initial phase of IR (5 days) and consequently the enhanced response to radiation observed by targeting total TAMs, we sought to examine whether MG and MDM transcriptional programs were also altered in IR-recurrent (IR-Rec) PDG-Ink4a/Arf KO tumors (Fig. 5, A and D, and data file S2). We examined the “core” ontogeny signatures of MG and MDMs in glioma, characteristic of their distinct developmental origins (17), in IR-Rec and 5d-IR tumors. At all time points, we observed ontogeny-associated gene expression differences between MG and MDMs (fig. S5, E and F). Several cell identity markers, including Itga4 (CD49d) and Itgal (CD11a), the cell surface proteins used to distinguish MDM from MG by flow cytometry, were among the genes enriched in MDMs. Conversely, MG were enriched in the established ontogeny markers Siglech, P2ry12, Jam2, Sall1, and Tmem119 (fig. S5G). These results show that the core MG and MDM ontogeny signatures, which delineate their different origins, are maintained throughout IR response and recurrence. We thus aimed to investigate changes in TAM subsets relating to alteration of their tumor-specific education without affecting their underlying ontogeny.

Global clustering analyses revealed that whereas MG and MDM isolated from control and 5d-IR gliomas clustered in an ontogeny-specific manner, MG and MDM from IR-Rec tumors formed their own cluster (Fig. 5D). These findings suggest that in recurrent gliomas, MG and MDMs undergo considerable expression changes resulting in a convergence of their transcriptional education programs at recurrence while retaining the fingerprints of their distinct ontogenies (Fig. 5D and fig. S5, E to G). Together with the alterations in MDM:MG ratios in recurrent tumors (Fig. 2B and fig. S2C), these results prompted us to examine how transcriptional networks of MG and MDMs may change dynamically throughout radiation response and recurrence.

We queried expression signatures previously identified as being associated with MG-specific or MDM-specific glioma education (17) to interrogate the phenotype of MG and MDMs after IR (Fig. 5E). These analyses revealed that even though MDMs maintained their ontogeny signature throughout, they displayed a lower enrichment of the glioma-specific education signature at recurrence and acquired an “MG-like” signature (Fig. 5E). We examined which genes were up-regulated specifically at recurrence in MG and MDMs (Fig. 5F). Most genes up-regulated in IR-Rec MDMs were already highly expressed in MG isolated from treatment-naïve gliomas (Cont MG), and reciprocally, genes up-regulated in IR-Rec MG were highly expressed in MDM isolated from treatment-naïve gliomas (Cont MDM) (fig. S5, H to J, and data file S3A). Furthermore, expression of Vcam1, an important mediator of leukocyte extravasation through the blood-brain barrier, was up-regulated in both IR-Rec MG and IR-Rec MDMs (fig. S5K). Thus, changes in glioma-specific education signatures are reflective of plasticity in both TAM populations at recurrence.

Together, our results reveal that core ontogeny signatures are maintained in MG and MDMs throughout the IR response, in agreement with studies highlighting the preservation of macrophage developmental traits in other pathological conditions (36). MG and MDM phenotypes in recurrent tumors are nonetheless distinct from their respective baseline states and display substantial modulation of the subtype-specific glioma education signatures, without this being a simple carryover of changes resulting from the initial exposure to radiation.

MG and MDM transcriptional programs converge upon a common phenotype in recurrent tumors

To identify the transcriptional regulators responsible for convergence of MG and MDM signatures at recurrence (Fig. 5D), we focused on the 693 up-regulated genes common to both populations (Fig. 6A). We refined our analyses by incorporating stage-specific comparisons between MG and MDMs, as well as cross-comparison to 5d-IR samples (Fig. 6A). This identified a set of 417 genes only up-regulated in recurrent tumors in both MG and MDMs (Fig. 6A). Genome-wide transcription factor activity analyses identified SMAD and RBPJ enrichment in recurrent tumors, with HOMER analyses confirming enrichment for SMAD binding motifs (Fig. 6, A and B, and data file S3, B and C). Both SMAD and RBPJ upstream signaling pathways, transforming growth factor–β (TGF-β) and Notch, respectively, have been implicated in MG or MDM homeostatic maintenance or immunosuppressive phenotypes (20, 37–39). Together, 251 and 118 target genes of SMAD and RBPJ, respectively, were part of the 417-gene signature common to MG and MDMs in recurrent gliomas. Seventy-eight of these genes are predicted to be regulated by both transcription factors (data file S3D), suggesting an intersection of TGF-β and Notch signaling pathways in TAMs at recurrence.

Fig. 6 Acquisition of a distinct transcriptional signature underlies MG and MDM convergence at recurrence.

(A) Venn diagram depicting MG and MDM gene expression differences between control and IR-Rec tumors. The intersect yielded 693 shared genes up-regulated in MG and MDMs at recurrence. These genes were then filtered to remove genes found to be altered in the 5d-IR treatment group. The resulting recurrence signature shared by MG and MDMs specifically in IR-Rec tumors consists of 417 genes (line plot, data file S2). (B) Motif enrichment analysis (HOMER) in the IR-Rec 417-gene signature (data file S3B). (C) Box plots depicting normalized transcription factor activity scores for the indicated motifs in MG and MDMs from control, 5d-IR, and IR-Rec PDG-Ink4a/Arf KO gliomas (complete list in data file S3C). (D) Box plots depicting normalized log2 gene expression values for CXCR4, VLDLR, and NRXN2 genes (SMAD/RBPJ1 target genes) in CD49D− MG and CD49D+ MDMs sorted from treatment-naïve (n = 3) and recurrent high-grade gliomas that emerged in patients after standard-of-care treatment (n = 2). (E) Quantitation of CD68+ total TAMs relative to total DAPI+ cells and of CD68+P2RY12+ MG relative to total DAPI+ cells in matched treatment-naïve and recurrent patient gliomas (n = 7). (F) Box plots depicting normalized log2 gene expression values for the pan-macrophage genes CD68 and ITGAM (CD11B) and the MG-specific genes P2RY12, OLFML3, SLC22A5, SPARC, and GPR34 in 74 matched initial and recurrent high-grade patient gliomas [gene expression data derived from (40)]. (G) Representative immunofluorescence images of P2RY12+, CD68+, NOTCH4+ cells and DAPI+CD68+P2RY12+ NOTCH4+ costaining in matched treatment-naïve and recurrent patient gliomas. Scale bars, 50 μm. (H) Quantitation of the number of CD68+NOTCH4+ cells relative to total CD68+ TAMs and of NOTCH4+ cells relative to total P2RY12+ MG in matched treatment-naïve and recurrent patient gliomas (n = 7). (I) Quantitation of the number of CD68+ CXCR4+ cells relative to total CD68+ TAMs and of CXCR4+ cells relative to total P2RY12+ MG in matched treatment-naïve and recurrent patient gliomas (n = 7). Graphs show means + SEM. P values were obtained using a negative binomial test (D) and unpaired (E, H, and I) or paired (F) two-tailed Student’s t test: *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001.

We investigated whether acquisition of a recurrence-specific signature in murine MG and MDMs after IR resembled changes occurring in treatment-naïve and recurrent human glioblastoma. We sorted MG and MDMs from three treatment-naïve and two recurrent samples from unmatched patients and performed RNA-seq. Several RBPJ and SMAD target genes that were identified in the glioma GEMMs (fig. S6A) also showed increased expression in MG and MDMs sorted from patients with recurrent disease (fig. S6B and data file S3E), including CXCR4, VLDLR, and NRXN2 (Fig. 6D). We analyzed tissue sections from seven matched pairs of treatment-naïve and recurrent glioblastoma (after fractionated 30 × 2 Gy IR, combined with temozolomide chemotherapy). Although the number of CD68+ TAMs was comparable in matched patient samples, P2RY12+ MG numbers decreased in recurrent tumors (Fig. 6E), as in glioma GEMMs (Fig. 2, A to C). Analysis of a publicly available RNA-seq dataset of 74 matched initial and recurrent human glioblastoma tissues [whole tumor samples (40)] confirmed decreased expression of MG-specific genes in recurrent tumors compared to the initially diagnosed glioblastoma, whereas expression of pan-TAM genes CD68 and ITGAM remained comparable (Fig. 6F), suggestive of a reduction in MG. Staining of matched treatment-naïve and recurrent patient glioblastoma tissue sections for CD68, P2RY12, and a subset of IR-Rec–associated markers (NOTCH4, CXCR4, and VCAM-1) corroborated the alterations of TAM populations observed in recurrent GEMM tumors after IR (Fig. 6, G to I, and fig. S6, C to E). Collectively, these analyses highlight the transcriptional changes in murine and human MG and MDMs at recurrence, which distinguish them from their treatment-naïve counterparts.

CSF-1R inhibition impairs MG/MDM transcriptional education signatures at recurrence

To determine whether TAM targeting via CSF-1R inhibition induces alterations to MG and MDM recurrence phenotypes, contributing to decreased glioma relapse in preclinical models, we queried RNA-seq data from MG and MDMs sorted from acute BLZ945 and IR + BLZ945 recurrent mouse tumors (Fig. 4, A to C). Expression of SMAD, RBPJ, and SMAD/RBPJ intersect target genes was down-regulated in the combination treatment group at recurrence (Fig. 7, A and B; fig. S7, A and B; and data file S4). Furthermore, a subset of IR-Rec genes (Cxcr4, Nes, and Vcam1) was already induced in MG and MDMs at 21d-IR, before evident tumor recurrence, and BLZ945 blocked up-regulation of these genes at this time point (fig. S7C). In vitro, cocultures of MG or MDMs with recurrent tumor cells, but not control tumor cells, resulted in Vcam1 and Nes up-regulation in both TAM populations, which was impaired by BLZ945 addition (fig. S7D). qRT-PCR experiments confirmed that MG and MDMs maintain the gene expression underlying their distinct ontogenies in vivo, with P2ry12 and Runx3 expressed exclusively in MG and MDMs, respectively (fig. S7E). MG and MDMs isolated from recurrent tumors after 12 days of BLZ945 showed expression of IR-Rec genes to some extent (Fig. 7A and fig. S7A). These results suggest that alteration of the acquired IR-Rec signature in IR + BLZ945 TAMs does not rely on the intrinsic effects of BLZ945 on MDMs and MG but occurs subsequent to IR treatment combined with CSF-1R inhibition. These analyses support the hypothesis that blocking the IR-acquired phenotype of MG and MDM limits recurrence, thereby enhancing survival in glioma-bearing mice.

Fig. 7 CSF-1R inhibition alters MG and MDM recurrence-associated signatures and delays glioma recurrence.

(A) Box plots depicting normalized gene expression values for SMAD, RBPJ1, and SMAD/RBPJ1 target genes in MDMs and MG from control tumors or recurrent tumors emerging from IR-treated (5 × 2 Gy), BLZ945-treated (200 mg/kg for 12 days), and IR + BLZ945–treated PDG-Ink4a/Arf KO mice, as depicted in Fig. 4A (see data file S4 for statistically significant differences). (B) qRT-PCR expression analyses of the recurrence-associated genes Notch4 and Cxcr4 in CD49d− MG and CD49d+ MDMs sorted from control tumors (n = 3) or from recurrent tumors emerging after treatment with IR (n = 4), BLZ945 (n = 4), and IR + BLZ945 (n = 4) in the PDG-Ink4a/Arf KO model. (C) Experimental design for long-term trial with IR and BLZ945 combination therapy in PDG-Ink4a/Arf KO tumors. Mice bearing high-grade PDG-Ink4a/Arf KO tumors were treated with 5 × 2 Gy IR (n = 21) or IR + BLZ945 (BLZ945 was dosed daily for the entire duration of the trial; n = 27 mice), and (D) symptom-free Kaplan-Meier survival curves were generated. (E) Tissues from endpoint control (n = 7), IR (n = 14)–treated, and IR + BLZ945 (n = 20)–treated PDG-Ink4a/Arf KO mice were graded histologically in a blinded manner. (F and G) Quantification of bioluminescent imaging (BLI) from intracranial transplantation of (F) patient-derived tumorspheres (TS573, 5 × 104 cells) and (G) human glioblastoma cell line (U251, 2.5 × 105 cells) in nude mice subjected to treatment with vehicle, IR (5 × 2 Gy), BLZ945 (200 mg/kg daily; 28 days), or combined IR + BLZ945. Graphs show means + SEM. P values were obtained using unpaired two-tailed Student’s t test (B, F, ad G) and log-rank Mantel-Cox test (D). *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001.

Long-term CSF-1R inhibition with IR inhibits glioma recurrence in preclinical trials

To determine whether long-term, daily BLZ945 treatment would result in improved survival compared to the 12-day acute treatment (Fig. 4), we performed another series of combination trials in PDG-Ink4a/Arf KO mice [IR (2 Gy/day) for 5 days, with daily BLZ945 until trial endpoint; Fig. 7C]. After the initial regression, the vast majority of tumors remained in a quiescent state until the trial endpoint (fig. S7, F to H), and we observed a marked extension in the life span of mice that underwent combined treatment compared to animals treated with IR alone (median survival of 25.85 weeks versus 10.42 weeks, respectively; Fig. 7D). By comparison, using the same preclinical trial design, we previously showed that BLZ945 monotherapy resulted in median survival of 20.85 weeks (16), thus less than the 25.85 weeks conferred by IR + BLZ945. Endpoint histological grading analyses indicated that 80% of IR + BLZ945–treated mice showed no evidence of residual tumor (Fig. 7E and fig. S7H). By comparison, treatment of PDG-Ink4a/Arf KO glioma-bearing mice with the current standard-of-care therapy of IR + temozolomide led to only a modest survival effect (median survival of 9.5 weeks; fig. S7, I and J). To further evaluate the therapeutic potential of macrophage targeting combined with IR, we performed orthotopic xenograft experiments using patient-derived tumorspheres (TS573) or a human glioma cell line (U251). Mice bearing established tumors were treated with fractionated IR + BLZ945, and combined therapeutic efficacy was observed in both models, with no tumor regrowth occurring during the 28-day preclinical trial in TS573 tumor-bearing animals (Fig. 7, F and G). Collectively, these preclinical results demonstrate the superior efficacy of IR + BLZ945 versus the current standard-of-care treatment.

DISCUSSION

Genetic and epigenetic adaptation of cancer cells over the course of radiotherapy has been extensively reported as major mechanisms promoting resistance and recurrence (41). However, there has been a limited appreciation to date of the dynamic and longitudinal alterations elicited in infiltrating immune and stromal cells within the TME during radiation response and recurrence. Recent studies have revealed the presence of both monocyte-derived and long-lived, prenatally seeded macrophages in various normal tissues and in tumors, including in the brain (17, 22, 23, 42). Despite the fact that TAMs are the most abundant immune cell type in glioblastoma (6), insight into how these heterogeneous cells dynamically change over the course of disease progression and in response to therapeutic intervention has been minimal and generally focused on examining the total TAM pool.

By investigating preclinical glioma models arising from different tumor suppressor gene perturbations, we found that p53-deficient tumors displayed a higher ratio of MDM to MG at baseline, compared to Ink4a/Arf-null (p53-proficient) glioblastoma. These results support the emerging viewpoint that the underlying cancer cell genetics can sculpt the TME in distinct ways, including by altering the abundance and education of resident versus peripherally recruited immune cells (43). A marked increase in the MDM:MG ratio was a common feature of recurrent gliomas after IR, independent of their genetic makeup. These results were corroborated by analyses of tissue samples from recurrent human gliomas after chemoradiation, where MG abundance decreased compared to patient-matched treatment-naïve gliomas. However, selective targeting of MDMs in combination with IR in preclinical models was not sufficient to substantially enhance survival, despite affecting several immunosuppressive phenotypes in the TME. Even though it could have it could have also been informative to specifically target MG alone in our study, it is widely recognized that these cells are challenging to stably deplete over the long term without altering brain physiology (44). As such, a more promising translational strategy would be to combine IR with continuous CSF-1R inhibition, which targets both TAM populations, resulting in a reversion of the MG/MDM IR-associated phenotype and thereby enhancing the initial response to IR to delay or prevent the emergence of recurrent disease.

The effects of BLZ945 monotherapy on macrophage reprogramming and tumor growth control were lost when acute treatment (12 days in total) was ceased, resulting in no overall survival advantage. By comparison, in an irradiated TME, short-term CSF-1R inhibition blocked induction of SMAD and RBPJ signature genes in MG and MDM and extended animal survival. In addition, a subset of these recurrence-associated genes was up-regulated in MG and MDM samples from patients who relapsed after standard of care, underscoring the potential clinical relevance of dual MG/MDM targeting combined with chemoradiation treatment for glioblastoma.

The clinical efficacy of immunomodulatory monotherapies in glioblastoma, such as pan-macrophage targeting or immune checkpoint blockade, has shown limited beneficial outcome to date in phase 2/3 trials in recurrent glioblastoma (45, 46). Although careful examination of the shortcomings of these single therapies is underway, combination treatments using these strategies are nonetheless the preferred multimodality approach in ongoing clinical trials in newly diagnosed glioblastoma. The efficacy of long-term combination therapy that we report here in the preclinical setting supports the therapeutic consideration of CSF-1R inhibitors in this context. Although we have not implemented an exhaustive panel of different timings or dosings of BLZ945 in combination with IR herein, we did find that daily administration of the CSF-1R inhibitor was essential to sustain the efficacy of combination therapy. Thus, appropriate dosing strategies will need to be carefully considered in ongoing and future clinical trials, as short-term or intermittent administration of CSF-1R inhibitors may not be optimal.

In sum, we propose that CSF-1R inhibition can enhance the initial glioma-debulking effects of radiotherapy by blocking IR-induced alternative activation in MG and MDM. This prevents the acquisition of recurrence-specific phenotypes in MG and MDM, thereby hindering the protumorigenic functions of these cells in supporting glioma proliferation and regrowth. Our findings provide insights into therapy-induced dynamic changes in different TAM populations, which may have important implications for strategies to combine TAM targeting with cytotoxic therapy and immunotherapy in cancer.

MATERIALS AND METHODS

Study design

The aim of this study was to analyze the dynamic changes associated with irradiation response and recurrence in glioma macrophage populations and to assess the impact of different therapeutic strategies combining macrophage targeting and radiotherapy in gliomas. The effect of radiotherapy and pan-TAM or TAM subpopulation targeting was assessed as single or combined therapies in glioblastoma-bearing mice (PDG-Ink4a/Arf KO or PDG-p53 KD). Sample size for animal experiments (n = 5 to 24) was based on previous experiments using these models (12, 14, 16, 17) and discussed with the Memorial Sloan Kettering Cancer Center (MSKCC) and NKI (Netherlands Cancer Institute) Biostatistics departments to ensure sufficient statistical power to compare the efficiency of combined treatments. Mouse survival and tumor burden assessment were the primary outcomes of these experiments. These data were supported by flow cytometry, immunofluorescence staining, and gene expression analyses of immune cell compartments (with a particular emphasis on TAM populations) to examine immune landscape changes associated with therapeutic response. The choice of time point analyses was based on longitudinal assessment of therapeutic response. Age-matched mice were assigned to experimental cohorts based on matching tumor volumes, and data presented include all outliers. Investigators were not blinded when monitoring animal survival. Blinding was applied during flow cytometric analyses of immune cell content and activation and all tissue immunostaining experiments and analyses. Biological replicates are indicated in the figure legends by n. Three or more independent trials were performed. All animal studies were reviewed and approved by the University of Lausanne (UNIL), MSKCC, and NKI Institutional Animal Care and Use Committees.

Tumor model generation and treatment

Mouse models of gliomagenesis and lineage-tracing models have been previously reported (17, 47, 48) and are described in full in Supplementary Materials and Methods. Glioblastomas were induced in 5- to 6-week-old GEMM animals as previously described (14). Drug administration, radiation treatment, and tissue processing can be found in full in Supplementary Materials and Methods.

IRB approval and patient information

Human specimens were collected from several institutes: NKI—patients consented to NKI Biobank CFMPB541; Cambridge—tissue collection complied with the UK Human Tissue Act 2004 and was approved by the Local Regional Ethics Committee (LREC ref. 04/Q0108/60); MSKCC—patients consented to MSKCC institutional review board (IRB) protocol #06-107 and MSKCC IRB protocol #14-230, as previously described (17). Patients diagnosed with confirmed grade IV glioma and either no previous history of brain malignancy or recurrent disease after surgical resection, fractionated radiotherapy, and temozolomide were included. Tumor specimens were collected from the operating room and processed immediately. Tissue processing protocols can be found in full in Supplementary Materials and Methods.

Cells

DF1 chicken fibroblasts and U251 cells were obtained from the American Type Culture Collection (ATCC). RCAS vectors expressing Platelet-Derived Growth Factor β-hemagglutinin (PDGFB-HA), and a short hairpin against mouse TP53 (shP53) were provided by T. Ozawa and E. Holland (Fred Hutchinson Cancer Research Center) (49). Derivation of glioma cell lines from PDG-Ink4a/Arf KO mice was as described previously (14). Culture of cell lines and coculture experimental procedures can be found in full in Supplementary Materials and Methods.

RNA-seq and bioinformatics

RNA was isolated following the TRIzol LS manufacturer’s instructions involving a chloroform extraction and isopropanol precipitation with glycogen carrier. The SMART-Seq library preparation kit (Clontech) and further analyses were performed as previously described (17). The 417-gene signature in Fig. 6A was identified by selecting genes that were up-regulated in both MDM and MG in recurrent tumor samples. Further refinement of the signature eliminated genes that showed any alteration in 5d-IR samples. The “Rtsne” package (50) was used to generate t-SNE (t-distributed stochastic neighbor embedding) probability distributions, which were visualized with ggplot2 (51). Gene Ontology analysis was performed using HOMER with default parameters (35). The “pheatmap” package was used to draw heatmaps, and Venn diagrams were drawn using the “venneuler” package (52). Line charts, box plots, and volcano plots were drawn using “ggplot2.”

Transcription factor activity analysis was performed as described previously (16, 17) and can be found in Supplementary Materials and Methods. Results presented in Fig. 6C are the z-scored single-sample gene set enrichment scores, termed “transcription factor activity score.”

Statistical analyses

Statistical analyses were performed using GraphPad Prism (GraphPad 8.0 software). For mouse survival studies, P values were calculated using the log-rank test. For tumor burden, immunostaining, and multicolor flow cytometry, only two groups were compared at any given time, allowing a t test to be used to calculate the statistical significance of differences between groups. Each specific test used is reported in the figure legends. For all statistical tests, significance was defined by P < 0.05. Data are presented as means ± SEM. Original data are provided in data file S5.

SUPPLEMENTARY MATERIALS

stm.sciencemag.org/cgi/content/full/12/552/eaaw7843/DC1

Materials and Methods

Fig. S1. Tumor regression and recurrence after radiotherapy in mouse glioma models.

Fig. S2. Changes in immune cell content in IR-recurrent gliomas and after MDM targeting.

Fig. S3. CSF-1R inhibition alters MG/MDM transcriptional programming after IR and modulates glioma cell DNA damage response.

Fig. S4. Analyses of recurrent gliomas from the short-term BLZ945 combination trial.

Fig. S5. MG and MDM ontogeny-specific signatures are unchanged in response to IR, whereas their tumor-specific education is altered.

Fig. S6. Analysis of recurrence-associated markers in patient samples.

Fig. S7. Analyses of PDG-Ink4a/Arf KO tumors treated with radiotherapy and long-term CSF-1R inhibition.

Table S1. Table of antibodies used in the study.

Data file S1. Analyses of TAM-IR 64-gene signature in MG and MDM isolated from 5d-IR PDG-Ink4a/Arf KO gliomas.

Data file S2. Differential gene expression in MG and MDMs isolated from control and IR-recurrent PDG-Ink4a/Arf KO gliomas.

Data file S3. Gene expression analyses of the 417-gene signature identified in MG or MDM sorted from IR-Rec PDG-Ink4a/Arf KO gliomas compared to control and in primary and recurrent human gliomas.

Data file S4. Significant changes in SMAD, RBPJ, and SMAD/RBPJ target genes between groups depicted in Fig. 7A.

Data file S5. Raw data.

References (53, 54)

References and Notes

  1. ↵
    1. F. Klemm,
    2. J. A. Joyce
    , Microenvironmental regulation of therapeutic response in cancer. Trends Cell Biol. 25, 198–213 (2015).
    OpenUrlCrossRefPubMed
  2. ↵
    1. C. Engblom,
    2. C. Pfirschke,
    3. M. J. Pittet
    , The role of myeloid cells in cancer therapies. Nat. Rev. Cancer 16, 447–462 (2016).
    OpenUrlCrossRefPubMed
  3. ↵
    1. L. Bingle,
    2. N. J. Brown,
    3. C. E. Lewis
    , The role of tumour-associated macrophages in tumour progression: Implications for new anticancer therapies. J. Pathol. 196, 254–265 (2002).
    OpenUrlCrossRefPubMedWeb of Science
    1. Q. W. Zhang,
    2. L. Liu,
    3. C. Y. Gong,
    4. H. S. Shi,
    5. Y. H. Zeng,
    6. X. Z. Wang,
    7. Y. W. Zhao,
    8. Y. Q. Wei
    , Prognostic significance of tumor-associated macrophages in solid tumor: A meta-analysis of the literature. PLOS ONE 7, e50946 (2012).
    OpenUrlCrossRefPubMed
    1. S. F. Hussain,
    2. D. Yang,
    3. D. Suki,
    4. K. Aldape,
    5. E. Grimm,
    6. A. B. Heimberger
    , The role of human glioma-infiltrating microglia/macrophages in mediating antitumor immune responses. Neuro Oncol. 8, 261–279 (2006).
    OpenUrlCrossRefPubMedWeb of Science
  6. ↵
    1. F. Klemm,
    2. R. R. Maas,
    3. R. L. Bowman,
    4. M. Kornete,
    5. K. Soukup,
    6. S. Nassiri,
    7. J. P. Brouland,
    8. C. A. Iacobuzio-Donahue,
    9. C. Brennan,
    10. V. Tabar,
    11. P. H. Gutin,
    12. R. T. Daniel,
    13. M. E. Hegi,
    14. J. A. Joyce
    , Interrogation of the microenvironmental landscape in brain tumors reveals disease-specific alterations of immune cells. Cell 181, 1643–1660.e17 (2020).
    OpenUrl
  7. ↵
    1. R. Stupp,
    2. M. E. Hegi,
    3. W. P. Mason,
    4. M. J. van den Bent,
    5. M. J. Taphoorn,
    6. R. C. Janzer,
    7. S. K. Ludwin,
    8. A. Allgeier,
    9. B. Fisher,
    10. K. Belanger,
    11. P. Hau,
    12. A. A. Brandes,
    13. J. Gijtenbeek,
    14. C. Marosi,
    15. C. J. Vecht,
    16. K. Mokhtari,
    17. P. Wesseling,
    18. S. Villa,
    19. E. Eisenhauer,
    20. T. Gorlia,
    21. M. Weller,
    22. D. Lacombe,
    23. J. G. Cairncross,
    24. R. O. Mirimanoff
    , Effects of radiotherapy with concomitant and adjuvant temozolomide versus radiotherapy alone on survival in glioblastoma in a randomised phase III study: 5-year analysis of the EORTC-NCIC trial. Lancet Oncol. 10, 459–466 (2009).
    OpenUrlCrossRefPubMedWeb of Science
  8. ↵
    1. R. Stupp,
    2. W. P. Mason,
    3. M. J. van den Bent,
    4. M. Weller,
    5. B. Fisher,
    6. M. J. Taphoorn,
    7. K. Belanger,
    8. A. A. Brandes,
    9. C. Marosi,
    10. U. Bogdahn,
    11. J. Curschmann,
    12. R. C. Janzer,
    13. S. K. Ludwin,
    14. T. Gorlia,
    15. A. Allgeier,
    16. D. Lacombe,
    17. J. G. Cairncross,
    18. E. Eisenhauer,
    19. R. O. Mirimanoff
    , Radiotherapy plus concomitant and adjuvant temozolomide for glioblastoma. N. Engl. J. Med. 352, 987–996 (2005).
    OpenUrlCrossRefPubMedWeb of Science
  9. ↵
    1. M. Squatrito,
    2. E. C. Holland
    , DNA damage response and growth factor signaling pathways in gliomagenesis and therapeutic resistance. Cancer Res. 71, 5945–5949 (2011).
    OpenUrlAbstract/FREE Full Text
  10. ↵
    1. H. E. Barker,
    2. J. T. Paget,
    3. A. A. Khan,
    4. K. J. Harrington
    , The tumour microenvironment after radiotherapy: Mechanisms of resistance and recurrence. Nat. Rev. Cancer 15, 409–425 (2015).
    OpenUrlCrossRefPubMed
  11. ↵
    1. B. Ruffell,
    2. L. M. Coussens
    , Macrophages and therapeutic resistance in cancer. Cancer Cell 27, 462–472 (2015).
    OpenUrlCrossRefPubMed
  12. ↵
    1. D. Yan,
    2. J. Kowal,
    3. L. Akkari,
    4. A. J. Schuhmacher,
    5. J. T. Huse,
    6. B. L. West,
    7. J. A. Joyce
    , Inhibition of colony stimulating factor-1 receptor abrogates microenvironment-mediated therapeutic resistance in gliomas. Oncogene 36, 6049–6058 (2017).
    OpenUrl
  13. ↵
    1. L. Cassetta,
    2. J. W. Pollard
    , Targeting macrophages: Therapeutic approaches in cancer. Nat. Rev. Drug Discov. 17, 887–904 (2018).
    OpenUrlCrossRefPubMed
  14. ↵
    1. S. M. Pyonteck,
    2. L. Akkari,
    3. A. J. Schuhmacher,
    4. R. L. Bowman,
    5. L. Sevenich,
    6. D. F. Quail,
    7. O. C. Olson,
    8. M. L. Quick,
    9. J. T. Huse,
    10. V. Teijeiro,
    11. M. Setty,
    12. C. S. Leslie,
    13. Y. Oei,
    14. A. Pedraza,
    15. J. Zhang,
    16. C. W. Brennan,
    17. J. C. Sutton,
    18. E. C. Holland,
    19. D. Daniel,
    20. J. A. Joyce
    , CSF-1R inhibition alters macrophage polarization and blocks glioma progression. Nat. Med. 19, 1264–1272 (2013).
    OpenUrlCrossRefPubMed
    1. S. J. Coniglio,
    2. E. Eugenin,
    3. K. Dobrenis,
    4. E. R. Stanley,
    5. B. L. West,
    6. M. H. Symons,
    7. J. E. Segall
    , Microglial stimulation of glioblastoma invasion involves epidermal growth factor receptor (EGFR) and colony stimulating factor 1 receptor (CSF-1R) signaling. Mol. Med. 18, 519–527 (2012).
    OpenUrlCrossRefPubMed
  16. ↵
    1. D. F. Quail,
    2. R. L. Bowman,
    3. L. Akkari,
    4. M. L. Quick,
    5. A. J. Schuhmacher,
    6. J. T. Huse,
    7. E. C. Holland,
    8. J. C. Sutton,
    9. J. A. Joyce
    , The tumor microenvironment underlies acquired resistance to CSF-1R inhibition in gliomas. Science 352, aad3018 (2016).
    OpenUrlAbstract/FREE Full Text
  17. ↵
    1. R. L. Bowman,
    2. F. Klemm,
    3. L. Akkari,
    4. S. M. Pyonteck,
    5. L. Sevenich,
    6. D. F. Quail,
    7. S. Dhara,
    8. K. Simpson,
    9. E. E. Gardner,
    10. C. A. Iacobuzio-Donahue,
    11. C. W. Brennan,
    12. V. Tabar,
    13. P. H. Gutin,
    14. J. A. Joyce
    , Macrophage ontogeny underlies differences in tumor-specific education in brain malignancies. Cell Rep. 17, 2445–2459 (2016).
    OpenUrl
  18. ↵
    1. Z. Chen,
    2. X. Feng,
    3. C. J. Herting,
    4. V. A. Garcia,
    5. K. Nie,
    6. W. W. Pong,
    7. R. Rasmussen,
    8. B. Dwivedi,
    9. S. Seby,
    10. S. A. Wolf,
    11. D. H. Gutmann,
    12. D. Hambardzumyan
    , Cellular and molecular identity of tumor-associated macrophages in glioblastoma. Cancer Res. 77, 2266–2278 (2017).
    OpenUrlAbstract/FREE Full Text
    1. Y. Zhu,
    2. J. M. Herndon,
    3. D. K. Sojka,
    4. K. W. Kim,
    5. B. L. Knolhoff,
    6. C. Zuo,
    7. D. R. Cullinan,
    8. J. Luo,
    9. A. R. Bearden,
    10. K. J. Lavine,
    11. W. M. Yokoyama,
    12. W. G. Hawkins,
    13. R. C. Fields,
    14. G. J. Randolph,
    15. D. G. DeNardo
    , Tissue-resident macrophages in pancreatic ductal adenocarcinoma originate from embryonic hematopoiesis and promote tumor progression. Immunity 47, 323–338.e6 (2017).
    OpenUrl
  20. ↵
    1. R. A. Franklin,
    2. W. Liao,
    3. A. Sarkar,
    4. M. V. Kim,
    5. M. R. Bivona,
    6. K. Liu,
    7. E. G. Pamer,
    8. M. O. Li
    , The cellular and molecular origin of tumor-associated macrophages. Science 344, 921–925 (2014).
    OpenUrlAbstract/FREE Full Text
  21. ↵
    1. J. G. Price,
    2. J. Idoyaga,
    3. H. Salmon,
    4. B. Hogstad,
    5. C. L. Bigarella,
    6. S. Ghaffari,
    7. M. Leboeuf,
    8. M. Merad
    , CDKN1A regulates Langerhans cell survival and promotes Treg cell generation upon exposure to ionizing irradiation. Nat. Immunol. 16, 1060–1068 (2015).
    OpenUrlCrossRefPubMed
  22. ↵
    1. F. Ginhoux,
    2. M. Greter,
    3. M. Leboeuf,
    4. S. Nandi,
    5. P. See,
    6. S. Gokhan,
    7. M. F. Mehler,
    8. S. J. Conway,
    9. L. G. Ng,
    10. E. R. Stanley,
    11. I. M. Samokhvalov,
    12. M. Merad
    , Fate mapping analysis reveals that adult microglia derive from primitive macrophages. Science 330, 841–845 (2010).
    OpenUrlAbstract/FREE Full Text
  23. ↵
    1. E. Gomez Perdiguero,
    2. K. Klapproth,
    3. C. Schulz,
    4. K. Busch,
    5. E. Azzoni,
    6. L. Crozet,
    7. H. Garner,
    8. C. Trouillet,
    9. M. F. de Bruijn,
    10. F. Geissmann,
    11. H. R. Rodewald
    , Tissue-resident macrophages originate from yolk-sac-derived erythro-myeloid progenitors. Nature 518, 547–551 (2015).
    OpenUrlCrossRefPubMed
  24. ↵
    1. D. H. Gutmann,
    2. H. Kettenmann
    , Microglia/brain macrophages as central drivers of brain tumor pathobiology. Neuron 104, 442–449 (2019).
    OpenUrl
  25. ↵
    1. D. Hambardzumyan,
    2. N. M. Amankulor,
    3. K. Y. Helmy,
    4. O. J. Becher,
    5. E. C. Holland
    , Modeling adult gliomas using RCAS/t-va technology. Transl. Oncol. 2, 89–95 (2009).
    OpenUrlCrossRefPubMed
  26. ↵
    1. J. Halliday,
    2. K. Helmy,
    3. S. S. Pattwell,
    4. K. L. Pitter,
    5. Q. LaPlant,
    6. T. Ozawa,
    7. E. C. Holland
    , In vivo radiation response of proneural glioma characterized by protective p53 transcriptional program and proneural-mesenchymal shift. Proc. Natl. Acad. Sci. U.S.A. 111, 5248–5253 (2014).
    OpenUrlAbstract/FREE Full Text
  27. ↵
    1. K. Leder,
    2. K. Pitter,
    3. Q. Laplant,
    4. D. Hambardzumyan,
    5. B. D. Ross,
    6. T. A. Chan,
    7. E. C. Holland,
    8. F. Michor
    , Mathematical modeling of PDGF-driven glioblastoma reveals optimized radiation dosing schedules. Cell 156, 603–616 (2014).
    OpenUrlCrossRefPubMedWeb of Science
  28. ↵
    1. M. Kioi,
    2. H. Vogel,
    3. G. Schultz,
    4. R. M. Hoffman,
    5. G. R. Harsh,
    6. J. M. Brown
    , Inhibition of vasculogenesis, but not angiogenesis, prevents the recurrence of glioblastoma after irradiation in mice. J. Clin. Invest. 120, 694–705 (2010).
    OpenUrlCrossRefPubMedWeb of Science
  29. ↵
    1. J. H. Stafford,
    2. T. Hirai,
    3. L. Deng,
    4. S. B. Chernikova,
    5. K. Urata,
    6. B. L. West,
    7. J. M. Brown
    , Colony stimulating factor 1 receptor inhibition delays recurrence of glioblastoma after radiation by altering myeloid cell recruitment and polarization. Neuro Oncol. 18, 797–806 (2016).
    OpenUrlCrossRefPubMed
  30. ↵
    1. J. Xu,
    2. J. Escamilla,
    3. S. Mok,
    4. J. David,
    5. S. Priceman,
    6. B. West,
    7. G. Bollag,
    8. W. McBride,
    9. L. Wu
    , CSF1R signaling blockade stanches tumor-infiltrating myeloid cells and improves the efficacy of radiotherapy in prostate cancer. Cancer Res. 73, 2782–2794 (2013).
    OpenUrlAbstract/FREE Full Text
  31. ↵
    1. J. L. Baron,
    2. J. A. Madri,
    3. N. H. Ruddle,
    4. G. Hashim,
    5. C. A. Janeway Jr..
    , Surface expression of alpha 4 integrin by CD4 T cells is required for their entry into brain parenchyma. J. Exp. Med. 177, 57–68 (1993).
    OpenUrlAbstract/FREE Full Text
  32. ↵
    1. P. Kubes,
    2. P. A. Ward
    , Leukocyte recruitment and the acute inflammatory response. Brain Pathol. 10, 127–135 (2000).
    OpenUrlPubMedWeb of Science
  33. ↵
    1. K. J. Myers,
    2. D. R. Witchell,
    3. M. J. Graham,
    4. S. Koo,
    5. M. Butler,
    6. T. P. Condon
    , Antisense oligonucleotide blockade of alpha 4 integrin prevents and reverses clinical symptoms in murine experimental autoimmune encephalomyelitis. J. Neuroimmunol. 160, 12–24 (2005).
    OpenUrlPubMed
  34. ↵
    1. J. R. Lakritz,
    2. D. M. Thibault,
    3. J. A. Robinson,
    4. J. H. Campbell,
    5. A. D. Miller,
    6. K. C. Williams,
    7. T. H. Burdo
    , α4-integrin antibody treatment blocks monocyte/macrophage traffic to, vascular cell adhesion molecule-1 expression in, and pathology of the dorsal root ganglia in an SIV macaque model of HIV-peripheral neuropathy. Am. J. Pathol. 186, 1754–1761 (2016).
    OpenUrlCrossRef
  35. ↵
    1. S. Heinz,
    2. C. Benner,
    3. N. Spann,
    4. E. Bertolino,
    5. Y. C. Lin,
    6. P. Laslo,
    7. J. X. Cheng,
    8. C. Murre,
    9. H. Singh,
    10. C. K. Glass
    , Simple combinations of lineage-determining transcription factors prime cis-regulatory elements required for macrophage and B cell identities. Mol. Cell 38, 576–589 (2010).
    OpenUrlCrossRefPubMedWeb of Science
  36. ↵
    1. M. Guilliams,
    2. A. Mildner,
    3. S. Yona
    , Developmental and functional heterogeneity of monocytes. Immunity 49, 595–613 (2018).
    OpenUrl
  37. ↵
    1. A. Buttgereit,
    2. I. Lelios,
    3. X. Yu,
    4. M. Vrohlings,
    5. N. R. Krakoski,
    6. E. L. Gautier,
    7. R. Nishinakamura,
    8. B. Becher,
    9. M. Greter
    , Sall1 is a transcriptional regulator defining microglia identity and function. Nat. Immunol. 17, 1397–1406 (2016).
    OpenUrlCrossRefPubMed
    1. D. Gosselin,
    2. V. M. Link,
    3. C. E. Romanoski,
    4. G. J. Fonseca,
    5. D. Z. Eichenfield,
    6. N. J. Spann,
    7. J. D. Stender,
    8. H. B. Chun,
    9. H. Garner,
    10. F. Geissmann,
    11. C. K. Glass
    , Environment drives selection and function of enhancers controlling tissue-specific macrophage identities. Cell 159, 1327–1340 (2014).
    OpenUrlCrossRefPubMedWeb of Science
  39. ↵
    1. L. Grandbarbe,
    2. A. Michelucci,
    3. T. Heurtaux,
    4. K. Hemmer,
    5. E. Morga,
    6. P. Heuschling
    , Notch signaling modulates the activation of microglial cells. Glia 55, 1519–1530 (2007).
    OpenUrlCrossRefPubMedWeb of Science
  40. ↵
    1. J. Wang,
    2. E. Cazzato,
    3. E. Ladewig,
    4. V. Frattini,
    5. D. I. Rosenbloom,
    6. S. Zairis,
    7. F. Abate,
    8. Z. Liu,
    9. O. Elliott,
    10. Y. J. Shin,
    11. J. K. Lee,
    12. I. H. Lee,
    13. W. Y. Park,
    14. M. Eoli,
    15. A. J. Blumberg,
    16. A. Lasorella,
    17. D. H. Nam,
    18. G. Finocchiaro,
    19. A. Iavarone,
    20. R. Rabadan
    , Clonal evolution of glioblastoma under therapy. Nat. Genet. 48, 768–776 (2016).
    OpenUrlCrossRefPubMed
  41. ↵
    1. D. Schaue,
    2. W. H. McBride
    , Opportunities and challenges of radiotherapy for treating cancer. Nat. Rev. Clin. Oncol. 12, 527–540 (2015).
    OpenUrlCrossRefPubMed
  42. ↵
    1. M. Prinz,
    2. J. Priller
    , The role of peripheral immune cells in the CNS in steady state and disease. Nat. Neurosci. 20, 136–144 (2017).
    OpenUrlCrossRefPubMed
  43. ↵
    1. M. D. Wellenstein,
    2. K. E. de Visser
    , Cancer-cell-intrinsic mechanisms shaping the tumor immune landscape. Immunity 48, 399–416 (2018).
    OpenUrlCrossRefPubMed
  44. ↵
    1. P. Wieghofer,
    2. K. P. Knobeloch,
    3. M. Prinz
    , Genetic targeting of microglia. Glia 63, 1–22 (2015).
    OpenUrlCrossRefPubMed
  45. ↵
    1. N. Butowski,
    2. H. Colman,
    3. J. F. De Groot,
    4. A. M. Omuro,
    5. L. Nayak,
    6. P. Y. Wen,
    7. T. F. Cloughesy,
    8. A. Marimuthu,
    9. S. Haidar,
    10. A. Perry,
    11. J. Huse,
    12. J. Phillips,
    13. B. L. West,
    14. K. B. Nolop,
    15. H. H. Hsu,
    16. K. L. Ligon,
    17. A. M. Molinaro,
    18. M. Prados
    , Orally administered colony stimulating factor 1 receptor inhibitor PLX3397 in recurrent glioblastoma: An Ivy Foundation Early Phase Clinical Trials Consortium phase II study. Neuro Oncol. 18, 557–564 (2016).
    OpenUrlCrossRefPubMed
  46. ↵
    1. T. R. Hodges,
    2. M. Ott,
    3. J. Xiu,
    4. Z. Gatalica,
    5. J. Swensen,
    6. S. Zhou,
    7. J. T. Huse,
    8. J. de Groot,
    9. S. Li,
    10. W. W. Overwijk,
    11. D. Spetzler,
    12. A. B. Heimberger
    , Mutational burden, immune checkpoint expression, and mismatch repair in glioma: Implications for immune checkpoint immunotherapy. Neuro Oncol. 19, 1047–1057 (2017).
    OpenUrlCrossRefPubMed
  47. ↵
    1. C. Dai,
    2. J. C. Celestino,
    3. Y. Okada,
    4. D. N. Louis,
    5. G. N. Fuller,
    6. E. C. Holland
    , PDGF autocrine stimulation dedifferentiates cultured astrocytes and induces oligodendrogliomas and oligoastrocytomas from neural progenitors and astrocytes in vivo. Genes Dev. 15, 1913–1925 (2001).
    OpenUrlAbstract/FREE Full Text
  48. ↵
    1. E. Tchougounova,
    2. M. Kastemar,
    3. D. Brasater,
    4. E. C. Holland,
    5. B. Westermark,
    6. L. Uhrbom
    , Loss of Arf causes tumor progression of PDGFB-induced oligodendroglioma. Oncogene 26, 6289–6296 (2007).
    OpenUrlCrossRefPubMedWeb of Science
  49. ↵
    1. T. Ozawa,
    2. M. Riester,
    3. Y. K. Cheng,
    4. J. T. Huse,
    5. M. Squatrito,
    6. K. Helmy,
    7. N. Charles,
    8. F. Michor,
    9. E. C. Holland
    , Most human non-GCIMP glioblastoma subtypes evolve from a common proneural-like precursor glioma. Cancer Cell 26, 288–300 (2014).
    OpenUrlCrossRefPubMedWeb of Science
  50. ↵
    1. L. van der Maaten,
    2. G. E. Hinton
    , Visualizing high-dimensional data using t-SNE. J. Mach. Learn. Res. 9, 2579–2605 (2008).
    OpenUrlCrossRefPubMedWeb of Science
  51. ↵
    H. Wickham, ggplot2: Elegant Graphics for Data Analysis (Springer, 2009).
  52. ↵
    1. R. Kolde,
    2. J. Vilo
    , GOsummaries: An R package for visual functional annotation of experimental data. F1000Res 4, 574 (2015).
    OpenUrl
  53. ↵
    1. M. E. Ritchie,
    2. B. Phipson,
    3. D. Wu,
    4. Y. Hu,
    5. C. W. Law,
    6. W. Shi,
    7. G. K. Smyth
    , Limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res. 43, e47 (2015).
    OpenUrlCrossRefPubMed
  54. ↵
    1. S. Hanzelmann,
    2. R. Castelo,
    3. J. Guinney
    , GSVA: Gene set variation analysis for microarray and RNA-seq data. BMC Bioinformatics 14, 7 (2013).
    OpenUrlCrossRefPubMed
Acknowledgments: We thank members of the Joyce and Akkari laboratories for insightful discussion. We are grateful to Novartis for providing BLZ945 and M. Wiesmann, Novartis, for critical input and discussion. We thank E. Holland and T. Ozawa for providing RCAS vectors and Nestin-Tv-a mice, C. Brennan for the TS573 patient tumor sphere line, C. Forsberg for Flt3:Cre;Rosa26:mTmG animals, and K. Allinson for assistance with human tissue sample collection. We thank the animal facilities, imaging cores, and flow cytometry cores at UNIL, NKI, and MSKCC for excellent technical assistance; M. Quick for technical support; and J. Kowal for critically reading the manuscript. Funding: This research was supported by the Swiss Cancer League (KFS 3990-08-2016 to J.A.J.); Ludwig Institute for Cancer Research (to J.A.J.); Dutch Cancer Society (KWF 10658 to L.A.); Dutch Research Council (NWO 91719355 to L.A.); Brain Tumour Charity GN 10/136 and GN-000429 (to C.W.); Brain Tumor Funders Collaborative (to L.A. and D.F.Q.); and fellowships from the American Brain Tumor Association (to L.A.), NCI F31CA167863 (to R.L.B.), Fondation Medic (F.K.), and Canadian Institutes of Health Research (to D.F.Q.). Author contributions: L.A. and J.A.J. conceived the study, designed experiments, interpreted data, and wrote the manuscript. L.A., J.T., S.M.H., M.d.G., D.F.Q., L.T., and J.G. performed and analyzed experiments. R.L.B. and F.K. performed all computational analyses. J.T.H. performed histopathological analyses. C.W., D.B., and J.W. provided patient samples. All authors edited or commented on the manuscript. Competing interests: The authors declare that they have no competing interests. Data and materials availability: The raw RNA-seq data generated in this study are accessible in the GEO under number GSE99537. All data associated with this study are present in the paper or the Supplementary Materials.

Article Tools

My saved folders

Stay Connected to Science Translational Medicine

Related Content

Similar Articles in:

Citing Articles in:

Navigate This Article