1264 VOLUME 19 | NUMBER 10 | OCTOBER 2013 NATURE MEDICINE

A R T I C L E S

GBM, the most aggressive form of glioma, has an invariably terminal prognosis, as patients respond minimally to currently used thera-pies, including surgery, radiation and chemotherapy1. One challenge in treating GBM is substantial tumor-cell and genetic heterogene-ity, leading to aberrant activation of multiple signaling pathways2,3. Underscoring this heterogeneity was the identification of several GBM molecular subtypes: proneural, neural, mesenchymal and classical4. In contrast, noncancerous stromal cells in the tumor microenviron-ment are genetically stable therapeutic targets5,6. TAMs in particular are associated with high tumor grade and poor prognosis in many cancers, including gliomas79. A paracrine CSF-1epidermal growth factor (EGF) signaling loop has also been implicated in breast cancer and glioma invasion10,11.

Several approaches have been used to ablate TAMs or inhibit their tumor-promoting functions in mouse models of cancer12. One strat-egy is CSF-1R inhibition, which depletes macrophages and reduces tumor volume in several xenograft models13,14. Here we used a potent, selective CSF-1R inhibitor in multiple preclinical GBM models. These included RCAS-hPDGFB (the human platelet-derived growth factor B gene)/nestinTv-a; Cdkn2a/ transgenic mice15,16, referred to here as PDGF-Bdriven glioma (PDG) mice, and intracranial xenografts of either proneural GBM cell lines or patient-derived tumor spheres. We found that CSF-1R inhibition blocks glioma growth and progression

through a mechanism in which TAMs are not depleted but are instead re-educated within the glioma microenvironment.

RESULTSMacrophages are targets of CSF-1R inhibition in PDG GBMsIn PDG mice, nestin+ glial progenitors express the RCAS virus recep-tor, Tv-a, which enables specific delivery of the initiating oncogene, PDGFB. PDG tumorigenesis recapitulates pathological and molecular features of human proneural GBM4,1719. We first investigated whether macrophages accumulate in PDG GBMs as in human gliomas8,9. Indeed, the numbers of CD11b+ myeloid cells and macrophages were markedly increased in PDG GBMs compared to normal brain, and the mRNA expression of Cd68, Csf1 and Csf1r was elevated in the PDGs (Supplementary Fig. 1a,b). Glioma cells and TAMs express Csf1, whereas Csf1r is only expressed in macrophages (Fig. 1a,b and Supplementary Fig. 1ce).

BLZ945 is a brain-penetrant inhibitor of CSF-1R (Supplementary Fig. 2a and Supplementary Methods). The biochemical half-maximum inhibitory concentration (IC50) for CSF-1R is 1 nM, which is >3,200-fold higher than its affinity for other kinases (Supplementary Table 1). Treatment of wild-type (WT) bone marrowderived macrophages (BMDMs) with BLZ945 inhibited CSF-1dependent prolifera-tion (half-maximum effective concentration (EC50) = 67 nM) and

1Cancer Biology and Genetics, Memorial Sloan-Kettering Cancer Center (MSKCC), New York, New York, USA. 2Human Oncology and Pathogenesis Program, MSKCC, New York, New York, USA. 3Brain Tumor Center, MSKCC, New York, New York, USA. 4Computational Biology, MSKCC, New York, New York, USA. 5Novartis Institutes for Biomedical Research, Emeryville, California, USA. 6These authors contributed equally to this work. Correspondence should be addressed to J.A.J. (joycej@mskcc.org).

Received 31 December 2012; accepted 14 August 2013; published online 22 September 2013; doi:10.1038/nm.3337

CSF-1R inhibition alters macrophage polarization and blocks glioma progressionStephanie M Pyonteck1,6, Leila Akkari1,6, Alberto J Schuhmacher1,6, Robert L Bowman1, Lisa Sevenich1, Daniela F Quail1, Oakley C Olson1, Marsha L Quick1, Jason T Huse2,3, Virginia Teijeiro1, Manu Setty4, Christina S Leslie4, Yoko Oei5, Alicia Pedraza2, Jianan Zhang2,3, Cameron W Brennan2,3, James C Sutton5, Eric C Holland1,3, Dylan Daniel5 & Johanna A Joyce1,3

Glioblastoma multiforme (GBM) comprises several molecular subtypes, including proneural GBM. Most therapeutic approaches targeting glioma cells have failed. An alternative strategy is to target cells in the glioma microenvironment, such as tumor-associated macrophages and microglia (TAMs). Macrophages depend on colony stimulating factor-1 (CSF-1) for differentiation and survival. We used an inhibitor of the CSF-1 receptor (CSF-1R) to target TAMs in a mouse proneural GBM model, which significantly increased survival and regressed established tumors. CSF-1R blockade additionally slowed intracranial growth of patient-derived glioma xenografts. Surprisingly, TAMs were not depleted in treated mice. Instead, glioma-secreted factors, including granulocyte-macrophage CSF (GM-CSF) and interferon-g (IFN-g), facilitated TAM survival in the context of CSF-1R inhibition. Expression of alternatively activated M2 markers decreased in surviving TAMs, which is consistent with impaired tumor-promoting functions. These gene signatures were associated with enhanced survival in patients with proneural GBM. Our results identify TAMs as a promising therapeutic target for proneural gliomas and establish the translational potential of CSF-1R inhibition for GBM.

npg

20

13 N

atur

e Am

eric

a, In

c. A

ll rig

hts

rese

rved

.

A R T I C L E S

NATURE MEDICINE VOLUME 19 | NUMBER 10 | OCTOBER 2013 1265

decreased CSF-1R phosphorylation, similarly to blockade with CSF-1Rspecific antibody (Fig. 1c and Supplementary Fig. 2bd). BLZ945 also reduced the viability of CRL-2467 microglia, Cdkn2a/ BMDMs (from the PDG genetic background) and BMDMs from non-obese diabetic severe combined immunodeficient (NOD-SCID) mice (Supplementary Fig. 3ac). Notably, BLZ945 treatment in culture did not affect the proliferation of any PDG-derived tumor-cell lines (all of which were Csf1r negative) or U-87 human malignant glioma (MG) cells, and PDG cell tumor-sphere formation was unaffected (Fig. 1d and Supplementary Fig. 3dg). Thus, BLZ945 has no direct effects on glioma cells and perturbs macrophage survival through CSF-1R inhibition.

CSF-1R inhibition blocks glioma progression and improves survivalWe next assessed BLZ945 in preclinical trials using PDG mice. At 2.5 weeks after glioma initiation, 30% of the animals had small tumors (grade II or III) and 70% had no tumors by histology or magnetic resonance imaging (MRI) (Supplementary Fig. 4a,b). We treated mice with BLZ945 or vehicle and evaluated them for symptom-free survival (Fig. 1e). The median survival time in the

vehicle-treated cohort was 5.7 weeks. In contrast, BLZ945 treatment significantly improved long-term survival, with 64.3% of the treated mice surviving to the trial endpoint at 26 weeks (Fig. 1f). We chose this endpoint because Cdkn2a/ mice develop spontaneous tumors, including lymphomas and sarcomas, beginning at ~30 weeks20. BLZ945 was well tolerated over the long-term treatment with no visible side effects (Supplementary Fig. 4e), which is consistent with results from previous histopathological studies21. Histological grad-ing revealed high-grade, invasive gliomas in all vehicle-treated mice. In contrast, BLZ945-treated animals had tumors that were signifi-cantly less malignant, and there were no detectable lesions in 55.6% of the mice that were asymptomatic at the trial endpoint (Fig. 1g and Supplementary Fig. 4c,d,f).

We then monitored the effects of BLZ945 on established high-grade PDGs, which are histologically similar to human GBMs at time of diagnosis17. We performed a 7-d trial incorporating MRI to assess ini-tial tumor volume and subsequent growth (Fig. 2a). We randomized PDG mice with detectable tumors (4.540 mm3) into BLZ945 or vehi-cle treatment cohorts. In the vehicle treatment group, tumor growth increased substantially. In contrast, BLZ945 halted glioma growth, and the majority of tumors decreased in size in the mice treated with

Figure 1 CSF-1R inhibition specifically targets macrophages, improves survival and decreases glioma malignancy in the transgenic PDG model. (a) Expression of Csf1 and Csf1r in different cell populations from GFP-expressing PDGs: a mixed population of live cells (DAPI), purified glioma cells (GFP+) and macrophages (CD11b+Gr-1). Cd11b and Tv-a were used as cell typespecific control genes for macrophages and glioma cells, respectively. Expression is shown relative to that in the live cell fraction and was normalized to Ubc for each sample (n = 3 mice per group). (b) Representative immunofluorescence images of normal brain or GBM PDGs co-stained with CSF-1R, CD68 (for macrophages and microglia) and DAPI. Scale bars, 50 Mm. (c) Graph showing that the CSF-1R inhibitor BLZ945 blocked BMDM survival, with an effect comparable to that of CSF-1 deprivation, assessed by MTT assays. n = 13 independent replicates. (d) Graph showing MTT assays for BLZ945 treatment of independent PDGF-driven glioma cell (PDGC) primary lines derived from PDG mice (Supplementary Figs. 3d and 7c). Concentrations up to 6,700 nM BLZ945 (100 times the dose required to effectively kill BMDMs in vitro) had no effect on glioma cell survival or proliferation. n = 3 independent replicates. (e) Experimental design for the long-term survival trial: PDG mice were injected with RCAShPDGFBhemagglutinin (HA) between 5 and 6 weeks of age to induce glioma formation and were assigned randomly to vehicle (20% Captisol, n = 22) or BLZ945 (200 mg per kg body weight (mg kg1), n = 14) treatment groups at 2.5 weeks after injection. Mice were dosed once daily until they developed symptoms or reached the trial endpoint. (f) Symptom-free survival curves for the mice described in e. (g) Histological grading of the vehicle (n = 14) and BLZ945 (n = 13) treatment groups. The data in a, c and d are shown as the mean o s.e.m. Statistical significance was calculated by unpaired two-tailed Students t test (c,d), log-rank (Mantel-Cox) test (f) or Fishers exact test (g). *P < 0.05, ***P < 0.001.

No CSF-1Vehicle67 nM BLZ945670 nM BLZ9456,700 nM BLZ945

Initiate tumors withRCAS-hPDGFB

(56 weeks of age, t = 0)

Vehicle or 200 mg kg1 BLZ945treatment started

(t = 2.5 weeks after tumor initiation)

Dose once per day

Kill when symptomaticor at defined endpoint

(t = 26 weeks)

PDGC-21

Pop

ulat

ion

doub

ling

0 1 2Time (d)

3 4

4.0

3.0

2.0

1.0

0

PDGC-31

Pop

ulat

ion

doub

ling

0 1 2Time (d)

3 4

4.0

3.0

2.0

1.0

0

100

Treatmentstarted

Definedendpoint

Per

cent

age

sym

ptom

-fre

e su

rviv

al

0

75

50

25

02 4 6 8 10 12

Time after RCAS-hPDGFB-HA injection (weeks)

BLZ945P = 1 107

14 16 18 20 22 24 26 28

Vehicle

5

Rel

ativ

e m

RN

A le

vel

4

3

2

1

0

Cd11b Tv-a Csf1 Csf1r

Live cellsGlioma cellsMacrophages

2.0BMDM

Pop

ulat

ion

doub

ling

1.5

1.0

0.5

00 1 2

Time (d)3 4

**** *** ***

a

c d

e

fNo gliomaGrade IIGrade IIIGrade IV/GBM

100 ***

80

60

40

Per

cent

age

of m

ice

20

0Vehicle BLZ945

g

DAPI

GB

MN

orm

al b

rain

CSF-1R DAPI CSF-1R CD68b

npg

20

13 N

atur

e Am

eric

a, In

c. A

ll rig

hts

rese

rved

.

A R T I C L E S

1266 VOLUME 19 | NUMBER 10 | OCTOBER 2013 NATURE MEDICINE

BLZ945 (Fig. 2b,d,e and Supplementary Fig. 5a,b). We then studied a third cohort of mice with large gliomas (>40 mm3) that we treated with BLZ945, and the vast majority of these mice showed regression (Fig. 2ce and Supplementary Fig. 5c). Comparison with a size-matched vehicle-treated cohort was not possible, as most of those mice would not have survived to the trial endpoint. We confirmed inhibition of CSF-1R phosphorylation in all BLZ945-treated tumors, and hPDGF-B production by glioma cells was unaffected by treatment with BLZ945 (Supplementary Fig. 6).

Because we used a PDGF-driven model for these initial trials, we wanted to exclude possible off-target effects of BLZ945 on glioma PDGF receptor (PDGFR) signaling. Glioma cells express PDGFR-A, and peri-cytes express PDGFR-B (Supplementary Fig. 7a,b). PDGFR inhibition reduces U-87 MG cell viability22,23. PDG tumors were modestly respon-sive to PDGFR inhibition in vivo24 and in culture (Supplementary Fig. 7c). The affinity of BLZ945 for PDGFR-A is approximately 10,000-fold lower than for CSF-1R (Supplementary Table 1). Unlike PDGFR inhibitors, treatment with BLZ945 or an antibody

to CSF-1R did not affect the viability of the U-87 MG or PDG cell lines (Fig. 1d and Supplementary Figs. 3de and 7c). Collectively these data establish that the therapeutic effects of BLZ945 in vivo are through CSF-1R inhibition and not off-target inhibition of PDGFR.

Multiple hallmarks of cancer are altered by CSF-1R blockadeTo determine the mechanisms underlying the marked response to CSF-1R inhibition in established gliomas, we investigated multiple tumorigenic processes, including tumor grade, proliferation, angio-genesis and resistance to apoptosis. (Supplementary Table 2). We performed analyses on tissues from the 7-d BLZ945 trial, including tissues taken at the midpoint (3 d) and endpoint (7 d) to capture the dynamic effects of CSF-1R inhibition. Whereas all vehicle-treated mice had high-grade gliomas, BLZ945-treated animals exhibited a pronounced histological response that was char-acterized by tumor-cell depopulation (Fig. 3a and Supplementary Fig. 8a). Correspondingly, glioma cell numbers within the regions of the original tumors decreased markedly (Fig. 3bd).

Initiate tumors withRCAS-hPDGFB

(56 weeks of age)

MRI for tumor volume(45 weeks after

tumor initiation; t = 1 d)

Treatmentstarted(t = 0 d)

aDose once per day

Euthanize animals attrial midpoint or endpoint

(t = 3 d, 7 d)

Follow-up MRI(t = 23 d, 6 d)

e 900800700600500400300200100

0Mouse number

Per

cent

age

diffe

renc

e in

tum

or v

olum

e (

1 d

to 6

d)

1 2 3 4 5 6 7 8 9 10 11

Vehicle

d 1 d 6 d

BLZ

945

larg

eB

LZ94

5V

ehic

le

c Treatmentstarted125

100

75

50

25

*****

Tum

or v

olum

e (m

m3 )

02 1 0 1 2

Treatment day (d)3 4 5 6 7

Trialendpoint

Treatmentstarted

125

b

100

75

50

25 **

***

Tum

or v

olum

e (m

m3 )

02 1 0 1 2

Treatment day (d)3 4 5 6 7

Trialendpoint

VehicleBLZ945BLZ945 large

1251007550250

255075

Per

cent

age

diffe

renc

e in

tum

or v

olum

e (

1 d

to 6

d)

Mouse number 1 2 3 4 5 6 7 8 9 10 11

BLZ94525

0

25

50

75Per

cent

age

diffe

renc

e in

tum

or v

olum

e (

1 d

to 6

d)

Mouse number

BLZ945 large

1 2 3 4 5 6 7 8 9 1011 18171615141312

Figure 2 CSF-1R inhibition blocks tumor growth and effectively regresses established gliomas. (a) Experimental design for the short-term 7 d trial: PDG mice bearing detectable tumors by MRI were randomly assigned to the vehicle or BLZ945 treatment group, with follow-up MRI as shown. (b,c) Mean tumor volume over the time course of the experiment for mice whose starting tumor volume was 4.540 mm3 (vehicle or BLZ945, n = 11 per group; b) or >40 mm3 (BLZ945 large, n = 18; c). (d) Representative images of T2-weighted MRI scans from the start and endpoint of the trial. The dashed lines indicate the regions of interest used to calculate tumor volume. (e) Waterfall plots showing the change in tumor volume at the experimental endpoint relative to the starting tumor volume for each individual mouse. Horizontal dashed lines indicate a 30% decrease in tumor volume. In the vehicle-treated group, there was a progressive, substantial increase in tumor growth ranging from 195% to 879%. In contrast, in this short treatment period, 6 of 18 mice in the BLZ945 large group had >30% reduction in tumor volume, qualifying as a partial response according to the response evaluation criteria in solid tumors (RECIST). The data in b and c are shown as the mean o s.e.m. Statistical significance was calculated by unpaired two-tailed Students t test. **P < 0.01, ***P < 0.001.

npg

20

13 N

atur

e Am

eric

a, In

c. A

ll rig

hts

rese

rved

.

A R T I C L E S

NATURE MEDICINE VOLUME 19 | NUMBER 10 | OCTOBER 2013 1267

Proliferation analyses revealed a 6798% reduction in glioma-cell proliferation after BLZ945 treatment (Fig. 3b,e). In addition, apopto-sis increased by 9- to 17-fold after 3 d (Fig. 3b,f), and tumor vascular-ity decreased. This reduction was probably not due to direct effects of BLZ945 on endothelial cells, as their viability was unaffected in cul-ture (Supplementary Fig. 8bf). Thus, CSF-1R inhibition effectively reduces the growth and malignancy of gliomas through the combined effects of blocking cell proliferation and increasing apoptosis.

CSF-1R inhibition slows intracranial growth of proneural GBM xenograftsWe next asked whether CSF-1R inhibition is effective in experimental models of human proneural GBM, including primary tumor spheres (TS573 and TS1137; Supplementary Table 3) and established cell lines (U251 and LN229) (Supplementary Methods). The tumor spheres and cell lines were either negative for CSF1R expression or showed negligible CSF1R mRNA levels compared to the human macrophage line THP-1, and none of these tumor spheres or GBM cell lines was sensitive to BLZ945 in monoculture (Fig. 4a,b and Supplementary Fig. 9a,b).

We intracranially injected the glioma tumor sphere or cell lines into NOD-SCID mice and treated established tumors with either vehicle or BLZ945. All four human lines responded to BLZ945 in vivo, showing a clear histological response and significantly reduced tumor growth and invasion (Fig. 4cf and Supplementary Fig. 9ce). Collectively these results demonstrate the preclinical efficacy of CSF-1R inhibi-tion across multiple proneural GBM models, including genetically engineered mice and human-derived xenografts.

Glioma-supplied survival factors protect TAMs from BLZ945-induced depletionMacrophage survival depends on CSF-1R signaling25, and we thus expected CSF-1R inhibition to deplete macrophage numbers. Indeed,

BLZ945 treatment reduced the number of CD11b+Ly6G cells in the blood of WT mice. Microglia numbers in normal brains were sub-stantially decreased by BLZ945 treatment; surprisingly, however, CSF-1R inhibition did not affect TAM numbers in PDGs compared to vehicle treatment (Fig. 5a,b and Supplementary Figs. 10 and 11). Similarly, TAMs were not depleted in human tumorsphere and cell-line xenografts (Supplementary Fig. 12a,b).

BLZ945-treated TAMs retained CSF-1R expression in vivo, and there were no significant differences in expression compared to vehicle-treated PDGs (Supplementary Fig. 13a). We examined whether the relative proportions of resident microglia and recruited BMDMs were altered in the PDG model. We performed BM trans-plantation experiments using GFP+ BM and initiated gliomas after confirming BM reconstitution (Supplementary Fig. 13b,c). The proportion of GFP+CD68+ macrophages in gliomas, determined by immunofluorescence analysis, was unchanged after BLZ945 treat-ment (Supplementary Fig. 13d), indicating no evident alteration in peripheral recruitment. We evaluated additional strategies to deplete macrophages in other tumor types or the brain and found that although the numbers of macrophages in peripheral tissues or normal brain decreased, there was no effect on TAM numbers in PDGs (Supplementary Fig. 14). We analyzed other immune cell types in gliomas after BLZ945 treatment and found no significant differences in cell numbers (Supplementary Fig. 13e). In addition, given varying reports of CSF-1R expression by neurons26,27, we exam-ined neuron numbers in gliomas and normal brain and found no differences after treatment with BLZ945 (Supplementary Fig. 13f), which is consistent with our detection of CSF-1R in macrophages only (Supplementary Fig. 1ce).

To explore the mechanisms by which glioma TAMs are spe-cifically protected from BLZ945-induced death, we investigated whether glioma-supplied factors support macrophage survival in the

aGrade IV/GBM

(vehicle)

Vehicle

Histological response(BLZ945)

BLZ945

3 d

DA

PI

Brd

u O

lig2

DA

PI

CC

3

7 d 3 d 7 d

BLZ945 largeb

*****

***

*** ***P

erce

ntag

e of

Olig

2+ c

ells

Vehicle BLZ9453 d 7 d 3 d 7 d

BLZ945 large

120d100806040200

***** *

*

*

Per

cent

age

ofC

C3+

cel

ls

Vehicle BLZ9453 d 7 d 3 d 7 d

BLZ945 large

f 109876543210

********

****

Per

cent

age

ofB

rdU

+ Olig

2+ c

ells

Vehicle BLZ9453 d 7 d

NS

3 d 7 dBLZ945 large

25e20151050

Vehicle BLZ9453 d

**

******

*

Tot

al n

umbe

rof

DA

PI+

cel

ls

7 d 3 d 7 dBLZ945 large

c 4 1053 105

2 105

1 105

0

Figure 3 Short-term BLZ945 treatment results in reduced tumor grade and proliferation and increased apoptosis. (a) Representative H&E images from the 7-d trial in the PDG model showing a GBM (vehicle treatment) and the histological response to BLZ945. (b) Representative images from the 3-d and 7-d time points stained for Olig2 (glioma cells), BrdU, CC3 and DAPI. White arrows indicate rare BrdU+Olig2+ cells in the BLZ945-treated groups. (cf) Quantification of the total number of DAPI+ cells per tumor (c), the percentage of Olig2+ cells relative to the total number of DAPI+ cells (d), the percentage of proliferating BrdU+Olig2+ cells relative to the total number of DAPI+ cells (e) and the percentage of apoptotic CC3+ cells relative to the total number of DAPI+ cells (f) (n = 6 for vehicle, n = 5 for 3 d BLZ945, n = 6 for 7 d BLZ945, n = 5 for 3 d BLZ945 large, n = 5 for 7 d BLZ945 large). These analyses revealed a progressive reduction in cell number, and by 7 d, the average glioma cell density was reduced to a20% of the total cells in both BLZ945 treatment groups. Analysis of glioma cell proliferation revealed a 6798% reduction after BLZ945 treatment. The circles in cf represent individual mice, and horizontal lines represent the means. Scale bars (a,b), 50 Mm. Statistical significance was calculated by unpaired two-tailed Students t test. NS, not significant. *P < 0.05, **P < 0.01, ***P < 0.001.

npg

20

13 N

atur

e Am

eric

a, In

c. A

ll rig

hts

rese

rved

.

A R T I C L E S

1268 VOLUME 19 | NUMBER 10 | OCTOBER 2013 NATURE MEDICINE

presence of CSF-1R inhibition. Indeed, glioma cellconditioned medium (GCM) from PDGs or human cell lines protected BMDMs from death induced by BLZ945 (Fig. 5c and Supplementary Figs. 12c and 15a,b). This was analogous to the TAM mainte-nance that we observed for all of our in vivo models (Fig. 5a and Supplementary Figs. 11, 12a and 13d) and also indicated that the survival factors were secreted. We screened GCM from a panel of PDG cell lines and identified multiple protective lines and one nonprotective line (Supplementary Fig. 16a). By comparing the GCM from the protective lines to that from the nonprotective line using antibody arrays, we identified differentially secreted pro-teins (Supplementary Fig. 16b,c), which we tested in parallel with known macrophage survival factors28. From this screen, we identi-fied three cytokines that promoted the survival, proliferation or both of BMDMs in the presence of BLZ945: GM-CSF (also called CSF-2), IFN-G and chemokine (C-X-C motif) ligand 10 (CXCL10) (Fig. 5d,e and Supplementary Figs. 17 and 18ad). Notably, when we added these factors in combination to GCM from the nonprotec-tive line, it conferred protection from BLZ945-induced killing to the BMDMs (Supplementary Fig. 18e) in association with elevated Akt activation in macrophages, similarly to treatment with GCM from protective lines (Fig. 5f,g). Consistent with their protective effect, expression of these cytokines and their receptors was substantially elevated in PDGs compared to normal brain and were also pro-duced by TAMs themselves (Supplementary Fig. 19). These results establish that glioma-supplied factors promote TAM survival in the presence of a CSF-1R inhibitor.

CSF-1R inhibition reduces M2 macrophage polarization in treated gliomasWe next asked how CSF-1R inhibition elicited such a potent antitumor response in vivo despite no evident TAM depletion. We hypothesized that TAMs from CSF-1R inhibitortreated animals were functionally altered, and we used microarray expression profiling of TAMs isolated from BLZ945- and vehicle-treated gliomas to identify 257 differentially expressed genes. 52 genes were significantly upregulated and 205 were downregulated; together this is referred to as the total gene signature (Fig. 6a, Supplementary Fig. 20a,b and Supplementary Table 4). CSF-1R inhibition in BLZ945-treated TAMs was corroborated by downregu-lation of targets of early growth receptor 2 (Egr2), a transcription factor that is located downstream of CSF-1R29 (Supplementary Fig. 20c). To determine the minimal set of genes that discriminated the treat-ment groups, we employed a lasso logistic regression model30 (Online Methods). This method identified a five-gene signature for BLZ945 treatment: Adm (adrenomedullin), Arg1 (arginase 1), F13a1 (encoding a clotting factor), Mrc1 (mannose receptor C type 1, also called Cd206), Serpinb2 (encoding a protease inhibitor and also called Pai2) (Fig. 6b); this is referred to as the minimal gene signature. Of these genes, four were downregulated after BLZ945 treatment. Quantitative PCR analy-ses on whole gliomas and sorted cell populations confirmed these find-ings (Supplementary Fig. 20d,e). Serpinb2, the only upregulated gene in the minimal signature, correlates positively with increased survival in breast cancer31. Interestingly, each of the downregulated genes has been associated previously with alternatively activated/M2 macrophage polarization3237 (Supplementary Table 4).

Figure 4 BLZ945 inhibits orthotopic tumor growth of patient-derived proneural tumor spheres and cell lines in vivo. (a) CSF1R mRNA expression in human proneural tumor sphere cells (TS573) and in the human proneural cell line U251 compared to the human macrophage line THP-1 (positive control). Expression is normalized to B2M (also called B2M) for each sample. n = 3 independent replicates. (b) MTT assays of BLZ945 treatment tested in TS573 and U251 human glioma cells demonstrating no effect of BLZ945 at concentrations up to 6,700 nM. n = 3 independent replicates. (c,d) Relative tumor growth determined by bioluminescence imaging (BLI) output in NOD-SCID mice injected intracranially with 5 104 TS573 cells (c) or 2.5 105 U251 cells (d). Treatment with BLZ945 was initiated when tumors were in a positive growth phase determined by BLI, which corresponded to day 14 for TS573 cells and day 7 for U251 cells. Mice were randomly assigned to vehicle (n = 12 for TS573, n = 16 for U251) or BLZ945 (n = 11 for TS573, n = 17 for U251) treatment groups. Tumor growth was evaluated every 5 d for 15 d, at which point vehicle-treated mice became symptomatic and both cohorts were euthanized for further analyses. (e,f) Representative images of the vehicle-treated and BLZ945-treated TS573 (e) and U251 (f) xenograft tumors described in c and d stained for GFP (tumor cells) and DAPI. Representative H&E images of both treatment groups for the U251 xenografts are shown in the bottom row in f. Scale bars (e,f), 50 Mm. The data in ad are shown as the mean o s.e.m. Statistical significance was calculated by nonparametric two-tailed Mann-Whitney test. *P < 0.05, **P < 0.01, ***P < 0.001.

a

c e

d

bCSF1R expression

1.00.5

0.01

Rel

ativ

e m

RN

A le

vel

500 TS573

Vehicle BLZ945

******Re

lativ

e tu

mor

gro

wth

4

DA

PI

GF

P

f Vehicle BLZ945

DA

PI

H&

EG

FP

0

THP-

1

TS57

3U2

51

3

2

1

0

4

Pop

ulat

ion

doub

ling

3

2

1

00 1 2 3

Time (d)

TS573 U251

4 0 1 2 3Time (d)

4

250

50

25

14 0 5 10 15Time (d)

Tumor cellinoculation

Treatmentstarted

0

250U251

*******Rela

tive

tum

or g

row

th

200

50

100

150

7 0 5 10 15Time (d)

Tumor cellinoculation

Treatmentstarted

0

Pop

ulat

ion

doub

ling

Vehicle 67 nM BLZ945670 nM BLZ945 6,700 nM BLZ945

VehicleBLZ945

VehicleBLZ945

npg

20

13 N

atur

e Am

eric

a, In

c. A

ll rig

hts

rese

rved

.

A R T I C L E S

NATURE MEDICINE VOLUME 19 | NUMBER 10 | OCTOBER 2013 1269

In many tumors, TAMs are M2 polarized, which is associated with protumorigenic functions38. Further, macrophages in human gliomas exhibit an M2-like phenotype, which is correlated with higher tumor grade9. Given the striking M2 gene enrichment in the minimal sig-nature, we examined the total gene list to determine whether other M2-associated markers were similarly downregulated by BLZ945, which revealed five additional genes (Supplementary Fig. 20f and Supplementary Table 4). Apart from interkeukin-1B (IL-1B), clas-sically activated/M1 genes were not correspondingly upregulated (Supplementary Fig. 20g). Interestingly, CSF-1 favors M2 macro-phage polarization in culture39,40, which is consistent with our finding that BLZ945 reduces M2 gene expression in vivo.

To explore how glioma cells and CSF-1R inhibition regulate M2 macrophage polarization, we used cell-based assays in which we exposed BMDMs to GCM to model microenvironmental inter-actions. Expression of all nine differentially expressed M2 genes (Supplementary Fig. 20f) significantly increased, and these increases were reversed by BLZ945 (Fig. 6c). Analysis of a subset of these genes in two microglia cell lines, BV-2 and CRL-2467, showed similar results (Supplementary Fig. 21a,b). We next examined Mrc1 expression in cocultures and freshly isolated mouse glioma microenvironment cultures containing TAMs. We selected Mrc1

because it is a well-established M2 flow cytometry marker, which facilitated the determination of its macrophage-specific expression in mixed cultures. Mrc1 expression also decreased in response to BLZ945 (Fig. 6d and Supplementary Fig. 21c,d), paralleling its downregulation in vivo.

Interestingly, human GCM suppressed macrophage phagocytosis in vitro in conjunction with M2 polarization41. To determine whether BLZ945 altered macrophage phagocytosis in vivo, we co-stained tis-sues for the macrophage cell-surface marker CD11b, the glioma cell marker Olig2 (ref. 42) and cleaved caspase 3 (CC3). Phagocytic capac-ity (number of CD11b+ macrophages that had engulfed Olig2+CC3+ glioma cells), increased 7.1-fold in the group of PDG mice with large gliomas treated with BLZ945 compared to mice treated with vehicle (Supplementary Fig. 22 and Supplementary Table 2). To control for enhanced glioma cell death, we analyzed phagocytic capacity after irradiation and found that it increased fourfold. Therefore, macro-phages in BLZ945-treated gliomas undergo an increase in phagocytic function that is apparently not dependent solely on enhanced tumor apoptosis. Together these data suggest that after CSF-1R inhibition, glioma TAMs lose M2 polarization and show enhanced phago-cytosis, providing a molecular corollary for their impaired tumor- promoting functions.

Figure 5 CSF-1R inhibition depletes normal microglia but not TAMs in treated PDGs as a result of the production of glioma-supplied survival factors. (a) Representative images of PDGs (top) and adjacent normal brain from the contralateral hemisphere of tumor-bearing mice (bottom) from the 7 d BLZ945 trial stained for CD68 and DAPI. White arrows indicate CD68+ macrophages in adjacent normal brain from both the vehicle- and BLZ945-treated groups. Scale bars, 50 Mm. (b) Quantification of the mean number of CD11b+ macrophages per 63 field of view (FOV) within each mouse tumor of the different treatment groups. (c) Graph showing MTT assays of BMDMs demonstrating that GCM induced BMDM proliferation and protected BMDMs from BLZ945-induced apoptosis. n = 19 independent replicates. For comparison, BMDMs were cultured in nonconditioned medium supplemented with CSF-1. (d,e) Results compiled from MTT assays demonstrating that GM-CSF and IFN-G individually protected BMDMs against BLZ945-induced death (n = 9 independent replicates; d), whereas CXCL10 promoted proliferation (n = 6 independent replicates; e). These effects were not reproduced by other candidate factors, for example, CXCL1 and vascular endothelial growth factor A (VEGF-A), which are shown here and in Supplementary Figure 17. (f) Western blots showing activation of Akt (Ser473 phosphorylation site) in BMDMs. BMDMs were stimulated with freshly prepared GCM from protective (PDGC-23, PDGC-17 and PDGC-02) or nonprotective (PDGC-55) cell lines with or without BLZ945. In addition, GCM from nonprotective PDGC-55 cells was supplemented with survival factors (GM-CSF, IFN-G and CXCL10) with or without BLZ945. (g) Quantification of pAkt normalized to total Akt demonstrating significant changes in pAkt levels between BMDMs stimulated with the protective compared to the nonprotective GCM with or without BLZ945. n = 4 independent replicates. RU, relative units. BLZ945 was used at 670 nM in all cell culture assays unless otherwise specified. The circles in b represent individual mice (n = 7 for vehicle, n = 5 for 3 d BLZ945, n = 5 for 7 d BLZ945, n = 5 for 3 d BLZ945 large, n = 5 for 7 d BLZ945 large), and horizontal lines represent the means. The data in c, d, e and g are shown as the mean o s.e.m. Statistical significance was calculated by unpaired two-tailed Students t test. NS, not significant. *P < 0.05, **P < 0.01, ***P < 0.001. There were no significant differences for any of the comparisons performed in b.

a Vehicle

Glio

ma

DA

PI

CD

68C

D68

DA

PI

Adj

acen

t bra

in

BLZ945

3 d 7 d

Vehicle BLZ945

3 d 7 d

BLZ945 large

3 d 7 d

18

Mea

n nu

mbe

r of

CD

11b+

cells

per

63

FO

V p

er m

ouse

b

1614121086420

f

BLZ945

CSF-

1

PDGC

-23 G

CM

PDGC

-17 G

CM

PDGC

-02 G

CM

PDGC

-55 G

CM

PDGC

-55 G

CM

CSF-

1

PDGC

-23 G

CM

PDGC

-17 G

CM

PDGC

-02 G

CM

PDGC

-55 G

CM

+ co

mbine

d fac

tors

PDGC

-55 G

CM

+ co

mbine

d fac

tors

BLZ945

pAkt(Ser473)Total Akt

c4

GCM + vehicle

GCM + 67 nM BLZ945GCM + 670 nM BLZ945

DMEM + BLZ945

4

******

***

3

3

2

2Time (d)

Pop

ulat

ion

doub

ling

1

10

0

CSF-

1

PDGC

-23 C

M

PDGC

-17 C

M

PDGC

-02 C

M

PDGC

-55 C

M

PDGC

-55 C

M

+ co

mbine

d fac

torsCS

F-1

PDGC

-23 C

M

PDGC

-17 C

M

PDGC

-02 C

M

PDGC

-55 C

M

PDGC

-55 C

M

+ co

mbine

d fac

tors

g1.5

*

**

**

**

**

******

pAkt

nor

mal

ized

to to

tal A

kt (

RU

)

BLZ945

1.0

0.5

0

NSNS

NSNS

NS

Vehicle

**

e

Vehicle BLZ945

4

3

2

Pop

ulat

ion

doub

ling

1

0

CXCL10CXCL1VEGF-A

Vehicle BLZ945

***NS

Vehicle

******

d4

3

2

Pop

ulat

ion

doub

ling

1

0

GM-CSFIFN-

npg

20

13 N

atur

e Am

eric

a, In

c. A

ll rig

hts

rese

rved

.

A R T I C L E S

1270 VOLUME 19 | NUMBER 10 | OCTOBER 2013 NATURE MEDICINE

CSF-1R inhibition abrogates macrophage gliomacell heterotypic signalingAs BLZ945 blocked PDG proliferation in vivo, we examined whether glioma-cell proliferation was modulated by exposure to macrophages in cell culture and whether it was reduced by CSF-1R inhibition. We stimulated WT BMDMs with GCM to mimic their education in the glioma microenvironment (referred to hereafter as stimulated BMDMs) and subsequently added them to glioma cells in coculture (Supplementary Fig. 15a). Cell-cycle analysis showed that coculture of stimulated BMDMs with PDG cells markedly increased their pro-liferation. Notably, this induction was abrogated by the addition of BLZ945 or a CSF-1Rspecific antibody (Fig. 6e and Supplementary Fig. 15c), suggesting the M2-dependent nature of this effect.

We investigated whether macrophage-secreted factors influence the activation of known signaling effectors of proliferation and survival within glioma cells, including Akt and extracellular signalingrelated kinase (Erk). Levels of phosphorylated Akt (pAkt) in glioma cells increased in response to CM from the stimulated

BMDMs compared to CM from naive BMDMs. This induction was abrogated when we collected the CM from stimulated BMDMs in the presence of BLZ945 (Fig. 6f and Supplementary Fig. 23a,b). pErk levels in glioma cells did not change during this time course (Fig. 6f). Addition of the Akt inhibitor MK2206 markedly impaired the macro-phage CMinduced proliferation of glioma cells (Supplementary Fig. 23c), which implicates this pathway as an important driver of signaling in the GBM microenvironment. We also investigated whether macrophages promoted glioma-cell proliferation in vivo by orthotopically co-injecting BMDMs with PDG cells or inject-ing PDG cells alone. Notably, co-injection of macrophages with glioma cells enhanced tumor growth by 62% compared to injection of glioma cells alone, and this induction was blocked by treatment with BLZ945 (Supplementary Fig. 23d; P < 0.05). Collectively these data demonstrate that macrophages and glioma cells have recipro-cal effects on the survival, proliferation and/or polarization of each other to promote tumorigenesis, and this heterotypic signaling can be blocked by CSF-1R inhibition.

Figure 6 CSF-1R inhibition impairs heterotypic signaling between macrophages and glioma cells and is predictive of improved survival in patients with proneural GBM. (a) Volcano plot showing 257 differentially expressed genes in TAMs from BLZ945-treated mice compared to TAMs from vehicle-treated controls (7-d treatment, n = 8 mice per group), which is termed the total gene signature. 205 downregulated and 52 upregulated genes were differentially expressed in TAMs in the BLZ945-treated group. (b) Heat map of the minimal gene signature determined using a lasso logistic regression model trained on the expression data in a. This analysis identified five genes, which when considered together accurately differentiated between the BLZ945 and vehicle treatment groups. (c,d) Graph of the expression of M2-associated genes in cultured BMDMs demonstrating an increase in expression with exposure to GCM that was blocked in the presence of BLZ945 as determined by quantitative PCR for a panel of M2-associated genes identified in the total gene signature (Supplementary Fig. 20f) (c) or flow cytometry for Mrc1 (d). n = 7 independent replicates. (e) Glioma cell-cycle analyses in which glioma cells were cocultured with either naive-BMDMs or stimulated (stim) BMDMs (treated with GCM before coculture, as shown in Supplementary Fig. 15a). Cocultures were treated with BLZ945 or vehicle and subjected to flow cytometric analyses. n = 8 independent replicates. GC, glioma cells. (f) Western blots showing pAkt (Ser473) and pErk (Thr202/Tyr204) levels in glioma cells at the time points indicated. Glioma cells were exposed to CM produced by naive or stimulated BMDMs with or without BLZ945 (schematic in Supplementary Fig. 23a). BLZ945 was used at 670 nM in all cell culture assays unless otherwise specified. (g) Patients with proneural GBM in TCGA were classified into BLZ945-like and vehicle-like classes using the minimal gene signature from b (Supplementary Methods). Patients classified as BLZ945 like showed an increased median survival of 10 months. (h) Hazard ratios and 95% confidence intervals for the minimal gene signature determined for each subtype of the TCGA and combined data sets. Hazard ratios with a confidence interval that did not cross 1.0 were considered significant. The data in ce are shown as the mean o s.e.m. Statistical significance was calculated by unpaired two-tailed Students t test (ce), log-rank test (g) or Walds test (h). *P < 0.05, **P < 0.01, ***P < 0.001.

10

8

10

log

10 P

6

4

2

012 8 6 4 2 2

Fold change (vehicle/BLZ945)4 6 8

F13a1

AdmMrc1

Arginase 1 Serpinb2

205 52

a b Color key2 1 0 1 2

Z-score

BLZ9

45 6

BLZ9

45 1

BLZ9

45 7

BLZ9

45 5

BLZ9

45 8

BLZ9

45 4

BLZ9

45 2

BLZ9

45 3

Vehic

le 7

Vehic

le 5

Vehic

le 6

Vehic

le 2

Vehic

le 1

Vehic

le 8

Vehic

le 3

Vehic

le 4

Mrc1

Adm

F13a1

Arg1

Serpinb2

c

***

****

*****

*****

***

****

*****

****

**

302520158

6

4

2

0Adm Arg1 Cd163 Cdh1 F13a1Hmox1 Il1r2 Mrc1 Stab1

Rel

ativ

e m

RN

A le

vel

DMEMGCMGCM + BLZ945

d

*** ******

**1,500

1,000

500

0

Vehic

le

BLZ945Ve

hicle

Mea

n in

tens

ity o

f Mrc

1

IL-4

67 nM

670 n

M

GCMe

*** ***30

20

10

0

BLZ9

45

BLZ9

45

+ naiveBMDMs

+ stimBMDMs

GC

alone

Per

cent

age

of g

liom

ace

lls in

S p

hase

Vehic

le

Vehic

le

f GCMNaive BMDM CMStim BMDM CM

Stim BMDM + BLZ945 CM

pAktpErk

GAPDH

1 h 2 h 12 h 24 h 48 h

+ + + + + + + + + +

+ + + + + + + + + +

g BLZ945-like classVehicle-like class

100

P = 5.4 10650

00 20 40 60 80 100 120

Time (months)

Per

cent

age

surv

ival h Proneural

0.1 1 10Hazard ratio

Classical

Classical

Mesenchymal

Mesenchymal

TCGAdata set

Combineddata sets

Neural

Neural

Proneural

***

**

npg

20

13 N

atur

e Am

eric

a, In

c. A

ll rig

hts

rese

rved

.

A R T I C L E S

NATURE MEDICINE VOLUME 19 | NUMBER 10 | OCTOBER 2013 1271

Gene signatures of CSF-1R inhibition predict survival advantage in proneural GBMWe next examined whether BLZ945 gene signatures generated from mouse TAMs might have translational value through differential sur-vival associations in individuals with GBM. We used the total gene signature to classify subjects in The Cancer Genome Atlas (TCGA)43 and a second combined series4446 into BLZ945-like and vehicle-like classes. Notably, there was a proneural-specific increase in median survival for subjects in the BLZ945-like class that ranged from 7.5 to 31.5 months (Supplementary Fig. 24a,b). We also used the minimal gene signature and observed that this markedly smaller gene set reca-pitulated the BLZ945-like survival advantage in subjects with prone-ural GBM (Fig. 6g,h, Supplementary Fig. 24ad and Supplementary Tables 5 and 6). Notably, this survival increase was not evident in other GBM subtypes and was independent of glioma CpG island methylator phenotype (G-CIMP) status47 (Supplementary Fig. 24e and Supplementary Table 7).

To determine whether the minimal gene signature predicted sur-vival independently of macrophage number, we used expression of macrophage-specific genes (AIF1, CD68 and ITGAM) as surro-gates for macrophage number in a Cox proportional hazards model (Supplementary Tables 8 and 9). None of the macrophage-specific genes correlated with worse prognosis, indicating that the minimal gene signature of BLZ945 treatment is a subtype-specific, prognostic predictor of the survival of patients with GBM that is independ-ent of total macrophage number. This also suggests that the TAM phenotype within a tumor may predict overall survival better than TAM number per se. The subtype specificity for this survival differ-ence is consistent with the minimal and total gene signatures having been generated from the PDG model, which most closely represents proneural GBM19.

DISCUSSIONWe show here that a new CSF-1R inhibitor blocks malignant progres-sion, regresses established gliomas and markedly enhances survival in a transgenic mouse model of proneural GBM. Moreover, multiple proneural GBM human xenografts responded similarly to CSF-1R inhibition, underscoring the therapeutic relevance of these findings. Surprisingly, we found that TAMs in all models were specifically protected from CSF-1R inhibitorinduced death. This contrasted with the observed depletion of microglia in the normal brain and the depletion of macrophages in other tissues, which is consistent with previous reports13,14. These results suggest that glioma-supplied factors facilitate TAM survival in the presence of BLZ945, whereas neighboring microglia outside of the tumor mass are not exposed to these protective signals and are thus depleted. Modeling these microenvironment-mediated effects in culture allowed us to screen and identify GM-CSF and IFN-G as glioma-supplied factors that facilitate macrophage survival in the context of CSF-1R inhibition. Notably, expression of these factors (or their receptors) was enriched in gliomas compared to normal brain, which is consistent with their prosurvival functions. This led us to ask whether TAM functions were impaired after CSF-1R inhibition, given that their numbers were not altered. Indeed, expression analyses of surviving TAMs in vivo revealed a decrease in alternatively activated M2 macrophage mark-ers, which is consistent with blunted tumor-promoting functions. We also modeled this phenomenon in culture and demonstrated that induction of M2 polarization by glioma-secreted factors and its reversal by CSF-1R inhibition are both direct effects on macrophages. Together these data indicate that the presence of survival factors in

the glioma microenvironment protects TAMs from CSF-1R inhibitor mediated killing, enabling their re-education and an associated marked antitumor response.

There are several potential clinical implications of these findings. First, macrophage accumulation correlates with malignancy in human gliomas8,9, which supports therapeutic targeting of TAMs. Second, we found that alteration of TAM tumor-promoting functions can signifi-cantly abrogate malignancy and that depletion is not strictly necessary for effective macrophage-targeted therapy. Third, in light of our pre-clinical data in proneural human xenografts and mouse models and the prognostic advantage associated with BLZ945-like signatures in subjects with proneural GBM, it is possible that proneural GBMs are particularly dependent on the tumor-promoting functions of TAMs. As such, it will be interesting to determine in future studies whether other GBM subtypes respond similarly to CSF-1R inhibition. Fourth, myeloid cells, including macrophages, blunt the chemotherapeutic response in breast cancer models48,49 and promote revascularization after irradiation in GBM xenografts50. Thus, it may be instructive to consider CSF-1R inhibitors in combination with glioma celldirected therapies, opening the possibility for synergistic effects.

In sum, we uncover a new therapeutic strategy for targeting cells in the glioma microenvironment. Rather than depleting stromal cells, as has been the goal with many microenvironment-targeted therapies thus far, re-educating these cells has the potential to not only abolish their protumorigenic functions but also actively enlist them as suppressors of tumorigenesis.

METHODSMethods and any associated references are available in the online version of the paper.

Note: Any Supplementary Information and Source Data files are available in the online version of the paper.

ACKNOWLEDGMENTSWe thank K. Simpson and X. Chen for excellent technical support, members of the Joyce and Holland labs, particularly T. Ozawa, K. Pitter and M. Squatrito, for technical advice and reagents and D. Chakravarti and the MSKCC Brain Tumor Center for assistance with patient and tumor sphere line information. We thank R. Benezra, K. Hunter and H.-W. Wang for reading the manuscript. We thank E. Pamer and T. Hohl (MSKCC) for providing CCR2-DTR mice and R. Lang (Cincinnati Childrens Hospital Medical Center) for providing CD11b-DTR mice. We are grateful to the MSKCC Core Facilities of Genomics, Flow Cytometry and Small Animal Imaging, Geoffrey Beene Translational Oncology and the Novartis Institutes for BioMedical Research Emeryville Analytical Sciences group for technical assistance. This research was supported by US National Cancer Institute program grants of the Integrative Cancer Biology Program (CA148967; J.A.J. and C.S.L.) and Tumor Microenvironment Network (CA126518; J.A.J. and E.C.H.), Cycle for Survival (J.A.J.), the Geoffrey Beene Foundation (J.A.J., R.L.B. and O.C.O.), the MSKCC Brain Tumor Center (J.A.J. and L.A.), the Fundacin Ramn Areces and Ibercaja (A.J.S.), Deutsche Forschungsgemeinschaft (L.S.), Canadian Institutes of Health Research (D.F.Q.), US National Institutes of Health T32 Institutional Research training grant (5T32GM008539, S.M.P.), US National Cancer Institute F31 fellowships (F31CA167863, R.L.B.; F31CA171384, O.C.O.) and Cornell and Gerstner Sloan Kettering graduate programs (S.M.P., R.L.B., O.C.O. and V.T.).

AUTHOR CONTRIBUTIONSS.M.P., L.A., A.J.S., L.S., D.F.Q. and O.C.O. performed and analyzed experiments. R.L.B., M.S. and C.S.L. performed computational analyses. M.L.Q., V.T., Y.O., A.P. and J.Z. provided technical assistance or derived patient tumor sphere lines. J.T.H. performed histopathological analyses. C.W.B., J.C.S., E.C.H. and D.D. provided reagents. J.A.J. conceived, designed and supervised the study and wrote the manuscript. All authors edited or commented on the manuscript.

COMPETING FINANCIAL INTERESTSThe authors declare no competing financial interests.

npg

20

13 N

atur

e Am

eric

a, In

c. A

ll rig

hts

rese

rved

.

A R T I C L E S

1272 VOLUME 19 | NUMBER 10 | OCTOBER 2013 NATURE MEDICINE

Reprints and permissions information is available online at http://www.nature.com/reprints/index.html.

1. Stupp, R. et al. Radiotherapy plus concomitant and adjuvant temozolomide for glioblastoma. N. Engl. J. Med. 352, 987996 (2005).

2. Wen, P.Y. & Kesari, S. Malignant gliomas in adults. N. Engl. J. Med. 359, 492507 (2008).

3. Dunn, G.P. et al. Emerging insights into the molecular and cellular basis of glioblastoma. Genes Dev. 26, 756784 (2012).

4. Verhaak, R.G. et al. Integrated genomic analysis identifies clinically relevant subtypes of glioblastoma characterized by abnormalities in PDGFRA, IDH1, EGFR, and NF1. Cancer Cell 17, 98110 (2010).

5. Balkwill, F.R. & Mantovani, A. Cancer-related inflammation: common themes and therapeutic opportunities. Semin. Cancer Biol. 22, 3340 (2012).

6. Joyce, J.A. & Pollard, J.W. Microenvironmental regulation of metastasis. Nat. Rev. Cancer 9, 239252 (2009).

7. Bingle, L., Brown, N.J. & Lewis, C.E. The role of tumour-associated macrophages in tumour progression: implications for new anticancer therapies. J. Pathol. 196, 254265 (2002).

8. Hussain, S.F. et al. The role of human glioma-infiltrating microglia/macrophages in mediating antitumor immune responses. Neuro-oncol. 8, 261279 (2006).

9. Komohara, Y., Ohnishi, K., Kuratsu, J. & Takeya, M. Possible involvement of the M2 anti-inflammatory macrophage phenotype in growth of human gliomas. J. Pathol. 216, 1524 (2008).

10. Wyckoff, J. et al. A paracrine loop between tumor cells and macrophages is required for tumor cell migration in mammary tumors. Cancer Res. 64, 70227029 (2004).

11. Coniglio, S.J. et al. Microglial stimulation of glioblastoma invasion involves EGFR and CSF-1R signaling. Mol. Med. 18, 519527 (2012).

12. Ruffell, B., Affara, N.I. & Coussens, L.M. Differential macrophage programming in the tumor microenvironment. Trends Immunol. 33, 119126 (2012).

13. Manthey, C.L. et al. JNJ-28312141, a novel orally active colony-stimulating factor-1 receptor/FMS-related receptor tyrosine kinase-3 receptor tyrosine kinase inhibitor with potential utility in solid tumors, bone metastases, and acute myeloid leukemia. Mol. Cancer Ther. 8, 31513161 (2009).

14. Patel, S. & Player, M.R. Colony-stimulating factor-1 receptor inhibitors for the treatment of cancer and inflammatory disease. Curr. Top. Med. Chem. 9, 599610 (2009).

15. Fomchenko, E.I. et al. Recruited cells can become transformed and overtake PDGF-induced murine gliomas in vivo during tumor progression. PLoS ONE 6, e20605 (2011).

16. Hambardzumyan, D., Amankulor, N.M., Helmy, K.Y., Becher, O.J. & Holland, E.C. Modeling adult gliomas using RCAS/t-va technology. Transl. Oncol. 2, 8995 (2009).

17. Huse, J.T. & Holland, E.C. Genetically engineered mouse models of brain cancer and the promise of preclinical testing. Brain Pathol. 19, 132143 (2009).

18. Kennedy, B.C. et al. Dynamics of central and peripheral immunomodulation in a murine glioma model. BMC Immunol. 10, 11 (2009).

19. Hambardzumyan, D., Parada, L.F., Holland, E.C. & Charest, A. Genetic modeling of gliomas in mice: new tools to tackle old problems. Glia 59, 11551168 (2011).

20. Serrano, M. et al. Role of the INK4a locus in tumor suppression and cell mortality. Cell 85, 2737 (1996).

21. Wang, T. et al. Investigation of correlation among safety biomarkers in serum, histopathological examination, and toxicogenomics. Int. J. Toxicol. 30, 300312 (2011).

22. Kilic, T. et al. Intracranial inhibition of platelet-derived growth factor mediated glioblastoma cell growth by an orally active kinase inhibitor of the 2-phenylaminopyrimidine class. Cancer Res. 60, 51435150 (2000).

23. Takeuchi, H., Kanzawa, T., Kondo, Y. & Kondo, S. Inhibition of platelet-derived growth factor signalling induces autophagy in malignant glioma cells. Br. J. Cancer 90, 10691075 (2004).

24. Shih, A.H. et al. Dose-dependent effects of platelet-derived growth factor-B on glial tumorigenesis. Cancer Res. 64, 47834789 (2004).

25. Chitu, V. & Stanley, E.R. Colony-stimulating factor-1 in immunity and inflammation. Curr. Opin. Immunol. 18, 3948 (2006).

26. Luo, J. et al. Colony-stimulating factor 1 receptor (CSF1R) signaling in injured neurons facilitates protection and survival. J. Exp. Med. 210, 157172 (2013).

27. Erblich, B., Zhu, L., Etgen, A.M., Dobrenis, K. & Pollard, J.W. Absence of colony stimulation factor-1 receptor results in loss of microglia, disrupted brain development and olfactory deficits. PLoS ONE 6, e26317 (2011).

28. Lin, H. et al. Discovery of a cytokine and its receptor by functional screening of the extracellular proteome. Science 320, 807811 (2008).

29. Bradley, E.W., Ruan, M.M. & Oursler, M.J. Novel pro-survival functions of the Kruppel-like transcription factor Egr2 in promotion of macrophage colony-stimulating factormediated osteoclast survival downstream of the MEK/ERK pathway. J. Biol. Chem. 283, 80558064 (2008).

30. Friedman, J., Hastie, T. & Tibshirani, R. Regularization paths for generalized linear models via coordinate descent. J. Stat. Softw. 33, 122 (2010).

31. Croucher, D.R., Saunders, D.N., Lobov, S. & Ranson, M. Revisiting the biological roles of PAI2 (SERPINB2) in cancer. Nat. Rev. Cancer 8, 535545 (2008).

32. Biswas, S.K. & Mantovani, A. Macrophage plasticity and interaction with lymphocyte subsets: cancer as a paradigm. Nat. Immunol. 11, 889896 (2010).

33. Trcsik, D. et al. Factor XIII-A is involved in the regulation of gene expression in alternatively activated human macrophages. Thromb. Haemost. 104, 709717 (2010).

34. Probst-Cousin, S., Rickert, C.H. & Gullotta, F. Factor XIIIa-immunoreactivity in tumors of the central nervous system. Clin. Neuropathol. 17, 7984 (1998).

35. Chen, P. et al. Tumor-associated macrophages promote angiogenesis and melanoma growth via adrenomedullin in a paracrine and autocrine manner. Clin. Cancer Res. 17, 72307239 (2011).

36. Schroder, W.A., Major, L. & Suhrbier, A. The role of SerpinB2 in immunity. Crit. Rev. Immunol. 31, 1530 (2011).

37. Solinas, G. et al. Tumor-conditioned macrophages secrete migration-stimulating factor: a new marker for M2-polarization, influencing tumor cell motility. J. Immunol. 185, 642652 (2010).

38. Sica, A., Schioppa, T., Mantovani, A. & Allavena, P. Tumour-associated macrophages are a distinct M2 polarised population promoting tumour progression: potential targets of anti-cancer therapy. Eur. J. Cancer 42, 717727 (2006).

39. Fleetwood, A.J., Lawrence, T., Hamilton, J.A. & Cook, A.D. Granulocyte-macrophage colony-stimulating factor (CSF) and macrophage CSF-dependent macrophage phenotypes display differences in cytokine profiles and transcription factor activities: implications for CSF blockade in inflammation. J. Immunol. 178, 52455252 (2007).

40. Sierra-Filardi, E. et al. Activin A skews macrophage polarization by promoting a proinflammatory phenotype and inhibiting the acquisition of anti-inflammatory macrophage markers. Blood 117, 50925101 (2011).

41. Wu, A. et al. Glioma cancer stem cells induce immunosuppressive macrophages/microglia. Neuro-oncol. 12, 11131125 (2010).

42. Squatrito, M. et al. Loss of ATM/Chk2/p53 pathway components accelerates tumor development and contributes to radiation resistance in gliomas. Cancer Cell 18, 619629 (2010).

43. Cancer Genome Atlas Research Network. Comprehensive genomic characterization defines human glioblastoma genes and core pathways. Nature 455, 10611068 (2008).

44. Freije, W.A. et al. Gene expression profiling of gliomas strongly predicts survival. Cancer Res. 64, 65036510 (2004).

45. Phillips, H.S. et al. Molecular subclasses of high-grade glioma predict prognosis, delineate a pattern of disease progression, and resemble stages in neurogenesis. Cancer Cell 9, 157173 (2006).

46. Murat, A. et al. Stem cellrelated self-renewal signature and high epidermal growth factor receptor expression associated with resistance to concomitant chemoradiotherapy in glioblastoma. J. Clin. Oncol. 26, 30153024 (2008).

47. Noushmehr, H. et al. Identification of a CpG island methylator phenotype that defines a distinct subgroup of glioma. Cancer Cell 17, 510522 (2010).

48. DeNardo, D.G. et al. Leukocyte complexity predicts breast cancer survival and functionally regulates response to chemotherapy. Cancer Discovery 1, 5467 (2011).

49. Shree, T. et al. Macrophages and cathepsin proteases blunt chemotherapeutic response in breast cancer. Genes Dev. 25, 24652479 (2011).

50. Kioi, M. et al. Inhibition of vasculogenesis, but not angiogenesis, prevents the recurrence of glioblastoma after irradiation in mice. J. Clin. Invest. 120, 694705 (2010).

npg

20

13 N

atur

e Am

eric

a, In

c. A

ll rig

hts

rese

rved

.

NATURE MEDICINEdoi:10.1038/nm.3337

ONLINE METHODSMice. All animal studies were approved by the Institutional Animal Care and Use Committee of MSKCC under protocol 04-08-022. The nestinTv-a; Cdkn2a (Ink4a/Arf)/ mouse model51,52 (mixed strain background), trans-genic CCR2diphtheria toxin receptor (DTR)-CFP mice (C57BL/6)53 (referred to here as CCR2-DTR) and transgenic CD11b-DTR mice (FVB/N)5456 have been described previously. WT C57BL/6 and NOD-SCID mice were purchased from Charles River Laboratories, B-actinGFP (C57BL/6) mice57 were from Jackson Laboratories, and athymic nude mice were from the National Cancer Institute Frederick. All lines were bred within the MSKCC animal facility.

Culture of established cell lines. The U-87 MG (HTB-14)58 human glioma and CRL-2467 (ref. 59) mouse microglia cell lines were purchased from ATCC, and human umbilical vein endothelial cells (HUVECs) and human brain microvascular endothelial cells (HBMECs) were from Sciencell. The BV-2 (ref. 60) mouse micro-glia cell line and the U251 and LN229 human glioma cell lines were kind gifts from E.C.H. All cells were grown at 37 C with 5% CO2, passaged with trypsin and main-tained in DMEM, 10% FBS (Gibco) and 1% penicillin-streptomycin (Gibco) unless otherwise noted. The CRL-2467 cell line was cultured in DMEM and 10% FBS with 30 ng ml1 recombinant mouse CSF-1 (R&D Systems). HUVECs and HBMECs were cultured on gelatin-coated cell culture dishes in endothelial cell medium (Sciencell) and 10% FBS supplemented with endothelial cell growth factors.

Culture and characterization of human primary glioma tumor spheres. Human gliomaderived tumor sphere lines (TS1137 and TS573) were derived from consenting patients under Institutional Review Board (IRB)-approved protocols for the banking of excess tumor tissue during routine surgical resec-tion (MSKCC IRB# 99-125A(2) and 06-107), as previously described61. Tumor spheres were maintained in the human NeuroCult NS-A Proliferation Kit (Stem Cell Technologies) and were dissociated with Accutase cell detachment solu-tion (Millipore). Characterization and molecular subtyping were performed as described in the Supplementary Methods.

Thymidine kinaseGFP-luciferase cell labeling. The indicated cell lines (TS573, TS1137, U251, LN229 and PDGC-23) were labeled with a triple-imaging vector (thymidine kinase (TK)-GFP-luciferase (Luc) (TGL))62 to allow for noninvasive in vivo imaging of tumor growth. A standard protocol for retroviral transduction was used. Briefly, GP2-293T cells were transfected with the TGL construct and pCL-Ampho at a 1:1 ratio using Fugene (Promega) and OptiMEM (Gibco). 12 h later, the medium was replaced with complete antibiotic-free DMEM and was collected for 3 d for the transduction of target cells.

Intracranial injection for tumor initiation in the mouse PDG and human glioma models. Adult mice were fully anesthetized with ketamine and xylazine and the local anesthetic bupivacaine at the surgical site before intracranial injec-tion to the right frontal cortex (approximate coordinates: 1.5 mm lateral, 1 mm caudal from bregma, depth of 23 mm) using a fixed stereotactic apparatus (Stoelting) as previously described16,24. Additional details are available in the Supplementary Methods. TGL-labeled human cell lines were resuspended in antibiotic-free serum-free medium, and 5 104 (TS1137, TS573 and LN229) or 2.5 105 (U251) cells in a volume of 2 Ml were injected into 6-week-old female NOD-SCID mice. Athymic nude mice (6-week-old females) were injected with 2 Ml of either 5 104 TGL-labeled PDGC-23 cells alone or PDGC-23 cells mixed with 30% RFP+ BMDMs (1.66 104) in antibiotic-free serum-free medium.

BLZ945 treatment in vitro and in vivo. The CSF-1R inhibitor BLZ945 (supplied by Novartis) was synthesized as described in the Supplementary Methods. For cell culture studies, a 10 mM stock of BLZ945 was formulated in DMSO; DMSO was used as the vehicle control. For administration to mice, BLZ945 was for-mulated in 20% Captisol at a concentration of 12.5 mg ml1. For in vivo dosing, mice received 200 mg per kg body weight BLZ945 or vehicle (20% Captisol) by oral gavage once daily. For long-term survival studies in PDG mice, dosing was begun at 17 d after injection of RCAShPDGFBHA. For studies at a fixed time point, PDG mice underwent MRI scans at 45 weeks after injection of RCAShPDGFBHA, as previously described15. Additional information can be found in the Supplementary Methods.

Immunodeficient (NOD-SCID or athymic nude) animals were orthotopi-cally transplanted with TGL-labeled PDGC-23 cells (in athymic nude mice), TS573, TS1137, U251 or LN229 cells (in NOD-SCID mice). Mice were randomly assigned to BLZ945- or vehicle-treatment cohorts starting at the indicated time points after intracranial injection. Mice were dosed daily for 15 d, and tumor growth was monitored every 5 d by bioluminescence imaging using a Xenogen IVIS-200 Optical In vivo Imaging System.

Radiation treatment. Tumor-bearing PDG mice were sedated with ketamine and xylazine before a single radiation dose of 10 Gy (115 cGy min1)42 using the X-RAD 320 from Precision X-Ray. Mice were euthanized 24 h after irra-diation, and tissues were collected for histological analysis of macrophage phagocytic capacity.

Bone marrow transplant (BMT). For BMT experiments, nestinTv-a; Cdkn2a/ mice were bred to WT C57BL/6 mice to generate nestinTv-a; Cdkn2a+/ transplant recipients. NestinTv-a; Cdkn2a/ mice were also bred to C57BL/6 B-actinGFP transgenic mice to generate Cdkn2a+/ bone marrow donors that ubiquitously expressed GFP. For BMT, recipient mice were lethally irradiated (10 Gy) at 44.5 weeks of age. The next day, donor-mouse bone marrow was flushed from femurs and tibiae with X-Vivo 20 Media (Cambrex) and resuspended in sterile PBS at a concentration of 107 cells per ml. Recipients received 1 106 donor BM nucleated cells in 100 Ml of PBS by intravenous tail vein injection. After 4 weeks, successful engraftment was assessed by the percent-age of GFP+ cells in the blood of BMT recipients by flow cytometry. If engraftment was less than 60%, mice were excluded from all subsequent experiments.

DT/DTR and clodronate liposome-mediated macrophage depletion. A detailed description of the experimental procedures can be found in the Supplementary Methods.

Mouse euthanasia and tissue harvest. Mice were euthanized at the time points described in the figure legends or when they became symptomatic from their tumors, which included signs of poor grooming, lethargy, weight loss, hunching, macrocephaly or seizures. To isolate tissues for snap freezing in liquid nitrogen or BMDM isolation, mice were euthanized 1 h after the final treatment dose by CO2 asphyxiation or fully anesthetized with Avertin (2,2,2-tribromoethanol; Sigma) and cervically dislocated. For flow cytometry, mice were fully anesthe-tized with Avertin and transcardially perfused with 20 ml of PBS. The brain was then isolated, and the tumor was macrodissected from the surrounding normal tissue. For proliferation analysis, mice were injected intraperitoneally with 100 mg per g body weight of BrdU (Sigma) 2 h before euthanasia. To isolate tissues for frozen histology, mice were fully anesthetized with Avertin, transcar-dially perfused with 10 ml of PBS and 10 ml of 4% paraformaldehyde in PBS (PFA). The brain was post-fixed in PFA overnight at 4 C, and other tissues were cryopreserved in 30% sucrose at 4 C. After post-fixation, the brain was transferred to 30% sucrose at 4 C for at least 2 d. All tissues were then embed-ded in optimal cutting temperature compound (OCT) (Tissue-Tek), and 10-Mm cryostat tissue sections were used for all analyses.

Preparation of glioma single-cell suspensions. For the derivation of primary glioma cell cultures or investigation of brain macrophage populations by flow cytometry, macrodissected PDGs were digested to a single-cell suspension by 812 min of incubation at 37 C with 5 ml of papain digestion solution (0.94 mg ml1 papain (Worthington), 0.48 mM EDTA, 0.18 mg ml1 N-acetyl-l-cysteine (Sigma) and 0.06 mg ml1 DNase I (Sigma) diluted in Earls Balanced Salt Solution (EBSS) and allowed to activate at room temperature for at least 30 min). After digestion, the enzyme was inactivated by the addition of 2 ml of 0.71 mg ml1 ovomucoid (Worthington). The cell suspension was passed through a 40-Mm mesh filter to remove undigested tissue and washed with Neural Stem Cell (NSC) Basal Medium (Stem Cell Technologies) for subsequent culture or FACS buffer (1% IgG-free BSA in PBS (Jackson Immunoresearch)) for flow cytometry and centrifuged at a low speed, 750 r.p.m. (Sorvall Legend RT), to remove debris and obtain the cell pellet.

As many immune-cell epitopes are papain sensitive, for investigation of immune-cell infiltrates by flow cytometric analysis, tumors were digested to a

npg

20

13 N

atur

e Am

eric

a, In

c. A

ll rig

hts

rese

rved

.

NATURE MEDICINE doi:10.1038/nm.3337

single-cell suspension by incubation for 10 min at 37 C with 5 ml of 1.5 mg ml1 collagenase III (Worthington) and 0.06 mg ml1 DNase I in 1 Hanks Balanced Salt Solution (HBSS) with calcium and magnesium. The cell suspension was then washed with PBS and passed through a 40-Mm mesh filter to remove undigested tissue. To remove myelin debris, the cell pellet was resuspended in 15 ml of 25% Percoll (room temperature) prepared from stock isotonic Percoll (90% Percoll (Sigma) and 10% 10 HBSS) and then spun for 15 min at 1,500 r.p.m. (Sorvall Legend RT) with the accelerator and brake set to 1. The cell pellet was then washed with 1 HBSS before being resuspended in FACS buffer.

Derivation of mouse primary glioma cultures, tumor spheres and glioma cell lines. PDG single-cell suspensions were resuspended in DMEM containing 10% FBS to derive mouse primary glioma cultures, which were used at early passage (23) and contained a mixture of different cell types, including tumor cells, macrophages and astrocytes (Supplementary Fig. 21c). For immunostaining, primary glioma cultures were grown for 24 h on poly-l-lysinecoated coverslips (BD Biocoat), fixed with PFA overnight at 4 C, permeabilized with 0.1% Triton-X for 5 min and blocked with 0.5% PNB (phosphate-NaCl buffer) for at least 1 h before antibody staining for CD11b+ macrophages (1:200), nestin+ glioma cells (1:500) and GFAP+ astrocytes (1:1,000) (Supplementary Table 10).

For tumor sphere cultures, the PDG cell pellet was resuspended in mouse NSC Basal Medium with NSC proliferation supplement, 1 mg ml1 heparin (Stem Cell Technologies), 10 ng ml1 recombinant human EGF (Invitrogen) and 20 ng ml1 recombinant human basic fibroblast growth factor (FGF) (Sigma). Fresh medium was added every 72 h for 2 weeks. Primary tumor spheres were collected, mechanically disaggregated to a single-cell suspension and propagated by serial passaging. For the secondary tumor sphereformation assay, 5 103 cells were plated in a six-well plate in neurosphere medium in the presence of BLZ945 or DMSO as a vehicle. The medium was changed every 48 h, and the number of tumor spheres present after 2 weeks was counted.

To generate PDGC lines, secondary tumor spheres were dissociated to a sin-gle-cell suspension and cultivated in DMEM and 10% FBS as a monolayer63. Multiple glioma cell lines were derived from independent mice, denoted as PDGC lines. The PDGC-23 line was infected with a pBabe-H2B-mCherry construct as described previously64.

Isolation of BMDMs. For BM isolation followed by macrophage derivation, femurs and tibiae were harvested under sterile conditions and flushed. For all experiments, unless otherwise specified, BM was isolated from WT C57BL/6 mice. The marrow was passed through a 40-Mm strainer and cultured in 30-ml Teflon bags (PermaLife PL-30) in DMEM and 10% FBS supplemented with 10 ng ml1 recombinant mouse CSF-1 (R&D Systems). BM cells were cultured in Teflon bags for 7 d with fresh CSF-1containing medium changes every other day to induce macrophage differentiation, after which they were referred to as BMDMs.

Histology, immunohistochemistry, flow cytometry and staining analysis. A detailed description of the experimental procedures can be found in the Supplementary Methods and in Supplementary Tables 10 and 11.

Preparation of glioma-cell and BMDM CM and heterotypic glioma cell and BMDM stimulation experiments. CM from PDGC lines or human glioma cell lines was generated by incubating cells in serum-free DMEM or NeuroCult medium (for human tumor spheres) for 24 h (Supplementary Fig. 15a). Collected CM was passed through 0.22-Mm filters to remove cellular debris and is referred to here as GCM. Undiluted GCM was used to stimulate differenti-ated C57BL/6 WT or B-actinGFP+ BMDMs without further addition of CSF-1 or other media. Control macrophages received fresh DMEM, 10% FBS and 10 ng ml1 recombinant mouse CSF-1. When indicated, differentiated BMDMs were cultivated in GCM containing either DMSO as vehicle, 67 nM BLZ945, 670 nM BLZ945 or 8 Mg ml1 of the CSF-1Rneutralizing antibody or in DMEM, 10 ng ml1 mouse recombinant CSF-1 and 10 ng ml1 IL-4 (R&D Systems) for 24 h or 48 h before experimental analysis.

To generate the BMDM CM, three experimental conditions were established (Supplementary Fig. 23a): (i) serum-free medium and 10 ng ml1 CSF-1 (naive), (ii) GCM and vehicle and (iii) GCM and 670 nM BLZ945, with (ii) and

(iii) referred to as the stimulated conditions. The CM from each of these three treatment groups was then added to serum-starved PDGC-23 glioma cells in the presence or absence of 200 nM of the Akt inhibitor MK-2206 (ref. 65) before western blotting or cell-cycle analysis.

Protein array analysis of mouse cytokines and angiogenic factors. Membranes immobilized with captured antibodies to cytokines and angiogenic factors were purchased from R&D Systems (from the Cytokine and Angiogenesis Arrays, respectively) and processed according to the manufacturers instructions. Briefly, after membranes were blocked for 1 h at room temperature, they were incubated with 1 ml of each PDGC-derived GCM overnight at 4 C together with the respective antibody detection cocktails. After extensive washing (three times, 10 min each) to remove unbound factors, membranes were then incubated indi-vidually with horseradish peroxidase (HRP)-conjugated streptavidin provided by the manufacturer for 30 min at room temperature. Unbound materials were washed out (three times, 10 min each wash). The signal was detected by the enhanced chemiluminescence (ECL) system provided by the manufacturer.

Cell-cycle analysis. Control BMDMs or BMDMs derived from B-actinGFP+ mice that were prestimulated with GCM were cocultured in a 1:1 ratio with 1 105 serum-starved mCherry-positive PDGC-23 cells for 48 h in the presence of 670 nM BLZ945, 8 Mg ml1 of the CSF-1Rneutralizing antibody or DMSO as a vehicle. After the collection of trypsinized cocultured cells, wells were rinsed in additional medium, and this volume was collected to ensure recovery of all adherent BMDMs. Samples were washed once with FACS buffer and incubated for 10 min at room temperature in permeabilizing buffer (10 mM piperazine-N,N2-bis(2-ethanesulfonic acid) (PIPES), 0.1 M NaCl, 2 mM MgCl2 and 0.1% Triton X-100, pH 6.8) containing 0.1 mg ml1 DAPI (Invitrogen). After acquisi-tion on an LSR II flow cytometer (BD) using an ultraviolet laser (350360 nm), the cell-cycle status of the glioma cells was analyzed using the Flow Jo Dean-Jett-Fox program for cell-cycle analysis. The same procedure was used to analyze the cell-cycle entry of PDGC-23 cells after exposure to BMDM CM generated from the three experimental conditions detailed above.

Proliferation assay, protein and RNA isolation, western blotting and quan-titative real-time PCR. A detailed description of the experimental procedures can be found in the Supplementary Methods.

Microarrays and gene expression profiling. All samples were prepared and processed by the genomics core facility at MSKCC. RNA was isolated using TRIzol, and the quality was assessed by running on an Agilent Bioanalyzer. 75 ng of total RNA was reverse transcribed and labeled using the Genechip 3` IVT Express Kit (Affymetrix). The resulting cRNA was hybridized to Affymetrix MOE 430A 2.0 chips. Raw expression data were analyzed using GCOS 1.4 (Affymetrix). Data were normalized to a target intensity of 500 to account for differences in global chip intensity. The microarray data are deposited in NCBI Gene Expression Omnibus (GEO) under the accession number GSE37475.

Microarray analysis. All bioinformatics analyses were completed in R using the Bioconductor suite of packages. Expression values were computed using the robust multiarray average (RMA) method and then quantile normalized in the affy package66,67. The limma package68 was used to identify differentially expressed genes between the vehicle- and BLZ945-treated samples. Differential expression was considered significant at a fold change of o2 with a false discovery rate of 10%. Gene set enrichment analysis (GSEA) was used as described previously69. For subsequent analyses and comparison to human data sets, mouse expression values were mean centered across all samples.

Lasso regression method for gene signature identification. Mouse expres-sion data were normalized and mean centered as described above. Differentially expressed genes were used for further analysis. We trained a logistic regression model with lasso constraints to differentiate between vehicle- and BLZ945-treated samples using the glmnet package30 by setting the family parameter to binomial in the glmnet function. The regularization parameter for lasso regression was chosen by fourfold cross validation.

npg

20

13 N

atur

e Am

eric

a, In

c. A

ll rig

hts

rese

rved

.

NATURE MEDICINEdoi:10.1038/nm.3337

Patient glioma data sets, subtyping of non-TCGA patients, human glioma classification and stratification of patients by G-CIMP status. A detailed description of the analyses can be found in the Supplementary Methods.

Survival analyses. Survival analyses were completed using the survival pack-age in R70. Hazard ratios were determined using the coxph function from the survival package. Patients were stratified on the basis of the probability of the lasso logistic regression classification, G-CIMP status or both, as indicated. P values were generated using Walds test.

Plots for patient analyses. All Kaplan-Meier survival curves, heat maps and volcano plots were generated in R v 2.14.1 using the gplots and ggplot2 pack-ages71,72. Hazard ratio forest plots were generated in GraphPad Prism Pro5.

Data presentation and statistical analyses. Data are presented as the means with their respective standard errors (s.e.m.) or as statistical scatter plots gener-ated using GraphPad Prism Pro5. Within each experiment, independent biologi-cal replicates represent a minimum of three technical replicates. Numerical data were analyzed by unpaired two-tailed Students t test unless otherwise noted. For survival curves, P values were obtained using the log-rank (Mantel-Cox) test or the two-tailed Mann-Whitney test when indicated, and Fishers exact test was used for histological tumor grading. P < 0.05 was considered statisti-cally significant.

55. Duffield, J.S. et al. Selective depletion of macrophages reveals distinct, opposing roles during liver injury and repair. J. Clin. Invest. 115, 5665 (2005).

56. Ueno, M. et al. Layer V cortical neurons require microglial support for survival during postnatal development. Nat. Neurosci. 16, 543551 (2013).

57. Okabe, M., Ikawa, M., Kominami, K., Nakanishi, T. & Nishimune, Y. Green mice as a source of ubiquitous green cells. FEBS Lett. 407, 313319 (1997).

58. Pontn, J. & Macintyre, E.H. Long term culture of normal and neoplastic human glia. Acta Pathol. Microbiol. Scand. 74, 465486 (1968).

59. Walker, W.S. Establishment of mononuclear phagocyte cell lines. J. Immunol. Methods 174, 2531 (1994).

60. Bocchini, V. et al. An immortalized cell line expresses properties of activated microglial cells. J. Neurosci. Res. 31, 616621 (1992).

61. Szerlip, N.J. et al. Intratumoral heterogeneity of receptor tyrosine kinases EGFR and PDGFRA amplification in glioblastoma defines subpopulations with distinct growth factor response. Proc. Natl. Acad. Sci. USA 109, 30413046 (2012).

62. Ponomarev, V. et al. A novel triple-modality reporter gene for whole-body fluorescent, bioluminescent, and nuclear noninvasive imaging. Eur. J. Nucl. Med. Mol. Imaging 31, 740751 (2004).

63. Jensen, J.B. & Parmar, M. Strengths and limitations of the neurosphere culture system. Mol. Neurobiol. 34, 153161 (2006).

64. Florey, O., Kim, S.E., Sandoval, C.P., Haynes, C.M. & Overholtzer, M. Autophagy machinery mediates macroendocytic processing and entotic cell death by targeting single membranes. Nat. Cell Biol. 13, 13351343 (2011).

65. Hirai, H. et al. MK-2206, an allosteric Akt inhibitor, enhances antitumor efficacy by standard chemotherapeutic agents or molecular targeted drugs in vitro and in vivo. Mol. Cancer Ther. 9, 19561967 (2010).

66. Irizarry, R.A. et al. Summaries of Affymetrix GeneChip probe level data. Nucleic Acids Res. 31, e15 (2003).

67. Gautier, L., Cope, L., Bolstad, B.M. & Irizarry, R.A. affyanalysis of Affymetrix GeneChip data at the probe level. Bioinformatics 20, 307315 (2004).

68. Smyth, G.K. Linear models and empirical bayes methods for assessing differential expression in microarray experiments. Stat. Appl. Genet. Mol. Biol. 3, Article3 (2004).

69. Subramanian, A. et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc. Natl. Acad. Sci. USA 102, 1554515550 (2005).

70. Therneau, T. A package for survival analysis in S. (R package version 2.3612, 2012).

71. Warnes, G.R. et al. gplots: various R programming tools for plotting data (R package version 2.10.1, 2011).

72. Wickham, H. ggplot2: Elegant Graphics for Data Analysis (Springer, New York, 2009).

51. Dai, C. et al. PDGF autocrine stimulation dedifferentiates cultured astrocytes and induces oligodendrogliomas and oligoastrocytomas from neural progenitors and astrocytes in vivo. Genes Dev. 15, 19131925 (2001).

52. Tchougounova, E. et al. Loss of Arf causes tumor progression of PDGFB-induced oligodendroglioma. Oncogene 26, 62896296 (2007).

53. Hohl, T.M. et al. Inflammatory monocytes facilitate adaptive CD4 T cell responses during respiratory fungal infection. Cell Host Microbe 6, 470481 (2009).

54. Cailhier, J.F. et al. Conditional macrophage ablation demonstrates that resident macrophages initiate acute peritoneal inflammation. J. Immunol. 174, 23362342 (2005).

npg

20

13 N

atur

e Am

eric

a, In

c. A

ll rig

hts

rese

rved

.

CSF-1R inhibition alters macrophage polarization and blocks glioma progressionRESULTSMacrophages are targets of CSF-1R inhibition in PDG GBMsCSF-1R inhibition blocks glioma progression and improves survivalMultiple hallmarks of cancer are altered by CSF-1R blockadeCSF-1R inhibition slows intracranial growth of proneural GBM xenograftsGlioma-supplied survival factors protect TAMs from BLZ945-induced depletionCSF-1R inhibition reduces M2 macrophage polarization in treated gliomasCSF-1R inhibition abrogates macrophage gliomacell heterotypic signalingGene signatures of CSF-1R inhibition predict survival advantage in proneural GBM

DISCUSSIONMethodsONLINE METHODSMice.Culture of established cell lines.Culture and characterization of human primary glioma tumor spheres.Thymidine kinaseGFP-luciferase cell labeling.Intracranial injection for tumor initiation in the mouse PDG and human glioma models.BLZ945 treatment in vitro and in vivo.Radiation treatment.Bone marrow transplant (BMT).DT/DTR and clodronate liposome-mediated macrophage depletion.Mouse euthanasia and tissue harvest.Preparation of glioma single-cell suspensions.Derivation of mouse primary glioma cultures, tumor spheres and glioma cell lines.Isolation of BMDMs.Histology, immunohistochemistry, flow cytometry and staining analysis.Preparation of glioma-cell and BMDM CM and heterotypic glioma cell and BMDM stimulation experiments.Protein array analysis of mouse cytokines and angiogenic factors.Cell-cycle analysis.Proliferation assay, protein and RNA isolation, western blotting and quantitative real-time PCR.Microarrays and gene expression profiling.Microarray analysis.Lasso regression method for gene signature identification.Patient glioma data sets, subtyping of non-TCGA patients, human glioma classification and stratification of patients by G-CIMP status.Survival analyses.Plots for patient analyses.Data presentation and statistical analyses.

AcknowledgmentsAUTHOR CONTRIBUTIONSCOMPETING FINANCIAL INTERESTSReferencesFigure 1 CSF-1R inhibition specifically targets macrophages, improves survival and decreases glioma malignancy in the transgenic PDG model.Figure 2 CSF-1R inhibition blocks tumor growth and effectively regresses established gliomas.Figure 3 Short-term BLZ945 treatment results in reduced tumor grade and proliferation and increased apoptosis.Figure 4 BLZ945 inhibits orthotopic tumor growth of patient-derived proneural tumor spheres and cell lines in vivo.Figure 5 CSF-1R inhibition depletes normal microglia but not TAMs in treated PDGs as a result of the production of glioma-supplied survival factors.Figure 6 CSF-1R inhibition impairs heterotypic signaling between macrophages and glioma cells and is predictive of improved survival in patients with proneural GBM.