Oncogenic RAS Pathway Activation Promotes Resistance to Anti-VEGF Therapy Through G-CSF–Induced Neutrophil Recruitment

Oncogenic RAS pathway activation promotes resistance to anti-VEGF therapy through G-CSF–induced neutrophil recruitment

Vernon T. Phan1, Xiumin Wu, Jason H. Cheng, Rebecca X. Sheng, Alicia S. Chung, Guanglei Zhuang, Christopher Tran, Qinghua Song, Marcin Kowanetz, Amy Sambrone, Martha Tan, Y. Gloria Meng, Erica L. Jackson, Franklin V. Peale, Melissa R. Junttila, and Napoleone Ferrara2,3

Genentech, Inc., South San Francisco, CA 94080

Contributed by Napoleone Ferrara, February 25, 2013 (sent for review November 10, 2012)

Granulocyte-colony stimulating factor (G-CSF) promotes mobiliza- RAF/MEK pathway is constitutively active in many cancer cells + + tion of CD11b Gr1 myeloid cells and has been implicated in re- that produce high G-CSF levels and that expression of the Ets2 sistance to anti-VEGF therapy in mouse models. High G-CSF transcription factor is directly correlated with high G-CSF pro- production has been associated with a poor prognosis in cancer duction. Treatment with a MEK inhibitor substantially reduced patients. Here we show that activation of the RAS/MEK/ERK path- G-CSF release and myeloid cell mobilization. We also provide way regulates G-CSF expression through the Ets transcription fac- evidence that combination treatments of MEK inhibitor and anti- fi tor. Several growth factors induced G-CSF expression by a MEK- VEGF signi cantly decreased growth and tumor angiogenesis – dependent mechanism. Inhibition of G-CSF release with a MEK in anti-VEGF resistant tumor models, including a genetically inhibitor markedly reduced G-CSF production in vitro and syner- engineered mouse model (GEMM) of Kras-driven pancreatic + + gized with anti-VEGF antibodies to reduce CD11b Ly6G neutro- ductal adenocarcinoma (PDAC). phil mobilization and tumor growth and led to increased survival Results in animal models of cancer, including a genetically engineered mouse model of pancreatic adenocarcinoma. Analysis of biopsies Ets2 Transcriptional Regulation of G-CSF in Cancer. To identify from pancreatic cancer patients revealed increased phospho-MEK, transcription factors regulating G-CSF release in cancer cells, we expressed a G-CSF promoter–driven luciferase reporter in G-CSF, and Ets expression and enhanced neutrophil recruitment the 4T1-related mouse breast cancer cell lines (14). Consistent compared with normal pancreata. These results provide insights with our previous findings (13) that the nonmetastatic 67NR and into G-CSF regulation and on the mechanism of action of MEK 168FARN cells have undetectable G-CSF levels, whereas the inhibitors and point to unique anticancer strategies. metastatic 4T07 and 4T1 cells express high G-CSF levels, we detected strong luciferase activation in 4T07 and 4T1 but not in angiogenesis | microenvironment | tyrosine kinase 67NR or 168FARN cells (Fig. S1 A and B). We next identified

conserved transcriptional binding sites at the G-CSF promoter MEDICAL SCIENCES ngiogenesis is recognized as an important aspect of tu- region upstream of the ATG initiation codon (Fig. S1D). We Amorigenesis. VEGF-A (hereafter called VEGF) is a well- then performed site-directed mutagenesis to screen for potential characterized regulator of normal and pathological angiogenesis transcriptional binding sites that might regulate G-CSF transcrip- (1). Strategies targeting VEGF signaling have been shown to tion. We identified the binding sites of two Ets transcriptional inhibit tumor angiogenesis in a variety of animal models (2). consensuses—ACCCg (−232) and TAAAc (−101)—binding sites Several VEGF pathway inhibitors have demonstrated clinical and verified that these two sites are important mediators of G-CSF efficacy and have been US Food and Drug Administration– expression. Site-directed mutagenesis of either ACCCg or TAAAc approved for treatment of several malignancies. However, like significantly decreased luciferase activity (Fig. 1A). Importantly, most cancer therapies tested to date, patients treated with such mutatingbothsites(bar4)didnotfurther reduce luciferase activity, inhibitors eventually progress (3). suggesting that the two Ets binding sites control synergistically G- The stroma can facilitate tumor growth and angiogenesis through CSF transcription (Fig. 1A). Ets transcription factors have been a variety of mechanisms, including production of cytokines and in- reported to be highly expressed in human cancers (15). We found flammatory cell recruitment (4, 5). Also, much evidence supports that Ets2 protein levels are elevated in 4T07 and 4T1 compared the notion that various bone marrow–derived cell types play im- with 67NR and 168FARN l cells (Fig. S1C). Enforced expression of portant roles in regulating tumor angiogenesis (4). A population of Ets2 in 4T1 cells further increased G-CSF expression (Fig. 1B). myeloid cells, identified in the mouse by the expression of the cell Targeting Ets2 with shRNAs directly correlated with reduction in surface markers CD11b and Gr1, has generated considerable G-CSF expression (Fig. 1C). To confirm Ets2-induced G-CSF interest because of its ability to facilitate tumor angiogenesis and expression, we coexpressed either WT Ets2 or a dominant neg- + + metastasis (6, 7). Furthermore, subsets of CD11b Gr1 cells, ative Ets2 with the G-CSF promoter-driven luciferase reporter termed myeloid-derived suppressor cells, have the ability to construct in 4T1 cells. The dominant negative Ets2 abolished suppress T-cell responses and thus they may also promote tumor progression through escape from immune surveillance (8). Previous studies have shown that tumor recruitment of + + Author contributions: V.T.P., M.R.J., and N.F. designed research; V.T.P., X.W., J.H.C., R.X.S., CD11b Gr1 cells mediates refractoriness to anti-VEGF ther- A.S.C., G.Z., C.T., Q.S., M.K., A.S., M.T., Y.G.M., F.V.P., and M.R.J. performed research; M.K. apy in several murine models (9). The hematopoietic growth and E.L.J. contributed new reagents/analytic tools; V.T.P., X.W., G.Z., Y.G.M., F.V.P., and factor granulocyte-colony stimulating factor (G-CSF) (reviewed M.R.J. analyzed data; and V.T.P. and N.F. wrote the paper. in ref. 10) was reported to be a major mediator of expansion and The authors declare no conflict of interest. + + mobilization of CD11b Gr1 cells (11, 12). G-CSF–mobilized 1Present address: Principia Biopharma, South San Francisco, CA 94080. + + CD11b Gr1 cells also produce a variety of factors that facilitate 2Present address: Department of Pathology and Moores Cancer Center, University of primary tumor growth and metastasis, including MMP9, S100A8, California San Diego, La Jolla, CA 92093. and S100A9 (13) as well as proangiogenic factors such as Bv8 (12). 3To whom correspondence should be addressed. E-mail: [email protected]. In this study, we sought to elucidate the signaling pathways that This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. control G-CSF expression in tumor cells. We show that the RAS/ 1073/pnas.1303302110/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1303302110 PNAS | April 9, 2013 | vol. 110 | no. 15 | 6079–6084 Downloaded by guest on September 27, 2021 A B C including cell proliferation, apoptosis, hematopoiesis, angiogenesis, and tumorigenesis (16). In normal and cancer cells, the * 6 600 * 6 * RAS/RAF/MEK signaling pathway increases Ets2 activity 500 rease 4 through ERK-dependent phosphorylation (16). Because the RAS 400 4

300 pathway activates Ets2 transcriptional activity, we investigated 2 200 2 whether this pathway is activated in 4T1-related cell lines. Our

100 (G-CSF/Gapdh)

(G-CSF/Gapdh) analysis indicates that the RAS pathway is active in 4T1 but not in Relative Fold Increase Relative Fold Increase Luciferase Activity 0 0 Relative Fold Inc 0 67NR shCT shEts2 67NR cells, as assessed by BRAF and ERK phosphorylation CMV Ets2 A WT 4T1 (Fig. S2 ). We then tested whether inhibiting MEK activity can AAAc AAAc ACCCg T ACCCgT suppress G-CSF release in 4T1 cells. To this end, we used the DFE MEK inhibitor (MEKi) GDC-0973/XL518. This agent is a po-

500 * tent, selective, orally active inhibitor of MEK1/2 with an IC of 700 * 2 67NR 50 * < 600 400 4T1 1 nM in vitro (17) and is currently undergoing clinical trials (18). 1.6 500 300 G-CSF production by 4T1 cells was directly correlated with ERK 400 1.2 Activity 300 200 phosphorylation levels, which could be modulated by treatment 0.8 200 B 100 with different concentrations of MEKi (Fig. S2 ). The RAS sig-

G-CSF (pg/ml) 0.4 100 (Fold Increase)

Binding naling pathway controls growth, proliferation, and survival of can- Luciferase Activity 0 0 0 GFP Ets2 Ets2DN GFP Ets2 Ets2DN IgG anti-Ets2 cer cells by activating multiple downstream effectors including the RAF/MEK/ERK and the PI3K pathways (19). Our data indicate G Ovary Bladder Head & Neck Pancreas that the RAF/MEK/ERK, but not the PI3K pathway, is responsible for G-CSF overexpression in cancer cells (Fig. S3 A and B).

IgG G-CSF Expression in Mouse and Human Cancer Cell Lines. Mutations in the RAS signaling pathway have been detected in ∼30% of all human cancers (19, 20). We examined G-CSF expression proﬁles in mouse Kras mutant cancer cell lines. Many of the cell lines tested expressed high G-CSF levels in a MEK-dependent activation Ets2 manner (Figs. S2C and S3A). In contrast, PI3Ki treatment had no effect on G-CSF expression, conﬁrming that even though PI3K pathway is downstream of RAS, it does not play a role in G-CSF expression. Interestingly, in agreement with previous studies showing that inhibiting RAF could further activate the MAPK

G-CSF pathway (21, 22), we found that a RAF inhibitor, GDC-0879, further increased ERK phosphorylation and induced G-CSF expression in mouse cancer cell lines (Figs. S2C and S3A). It remains to be established whether G-CSF increases in metastatic Fig. 1. The Ets2 transcription factor regulates G-CSF expression in human melanoma patients treated with RAF kinase inhibitors play cancer. (A) Site-directed mutagenesis of Ets2 transcriptional binding sites a role in resistance to this therapy. We next screened 31 human (−232 aa) and (−101 aa) before G-CSF ATG start codon. Luciferase activity cancer cell lines representing six different cancer types. We was measured in 4T1 cells. WT, single-mutated binding sites, ACCCg and found that 45% (14/31) of the human cell lines express high TAAAc, or double-mutated sites ACCCg/TAAAc, *P < 0.001. Error bars in- G-CSF (Table S1). Eight of 13 G-CSF–positive cell lines have dicate SD. (B) Ets2 was expressed in 4T1 cells and G-CSF transcription was mutations in KRAS (Calu-1, Calu-3, Calu-6, EBC-1, HCC-15, assessed by quantitative PCR. Relative fold increase was measured and SW1463, H2122, MDA-MB231) (23, 24). Three cell lines have compared with CMV control, *P < 0.000003. Error bars indicate SD. (C) receptor tyrosine kinase amplifications or mutations that lead to shRNAs targeting Ets2 or control were transfected into 4T1 cells and G-CSF activation of the RAS pathway, as measured by ERK phos- expression was detected by quantitative PCR, *P < 0.01. Error bars indicate phorylation. These include EGFR mutations in the H1975 lung SD. Luciferase activity (D) or G-CSF (E) detected in 4T1 cells coexpressing cancer cell line (25), epidermal growth factor receptor (EGFR) either GFP control plus Ets2 WT (Ets2) or GFP control plus dominant negative and FGF amplification in the 5637 bladder cancer cell line (26) < × −5 Ets2 (Ets2DN) where the N terminus of Ets2 was deleted, *P 1.0 10 . and eriythropoietin receptor (EpoR) amplification the H838 Error bars indicate SD. (F) ChIP analysis of Ets2 binding to the G-CSF pro- < lung carcinoma cell line. The bladder cell line BFTC-095 has an moter in 4T1 and 67NR cells, *P 0.03. Error bars indicate SD. (G) Tumor active NRAS mutation (27) and UM-UC-1 has a constitutively biopsies from patients with ovarian, bladder, head and neck, or pancreatic B cancer were immunostained for Ets2 and G-CSF. (Scale bar, 20 μm.) active RAS pathway (Fig. S3 ). Similar to the mouse cancer cell lines, MEK inhibition reduced G-CSF release in a dose-dependent fashion in all of the G-CSF–positive cell lines tested (Fig. S3B). G-CSF promoter activity (Fig. 1D) and decreased G-CSF levels RAF inhibitor GDC-0879 treatment resulted in activation of the (Fig. 1E), whereas WT Ets2 further increased G-CSF expression. MAPK pathway and G-CSF expression in a subset of human To investigate whether Ets2 directly binds to the G-CSF pro- cancers. Similar to the mouse cancer cell lines, PI3K inhibitor treatment had no effect on G-CSF expression (Fig. S3B). moter, we performed ChIP analysis in 67NR and 4T1 cells. We evidenced direct binding of Ets2 to the G-CSF promoter in 4T1 F Multiple Growth Factors Induce G-CSF Expression. We sought to but not in 67NR cells by quantitative PCR (Fig. 1 ). Next, we identify factors responsible for inducing G-CSF expression. We performed immunohistochemistry to investigate whether Ets2 and previously reported that, whereas Lewis lung carcinoma (LLC) G-CSF are overexpressed in human cancers, including ovarian, cells secrete very low amounts in vitro, LLC tumors produce high bladder, head and neck, and pancreatic adenocarcinomas. We G-CSF levels in vivo (12). We hypothesized that growth factors validated antibodies recognizing G-CSF and Ets2 by immunohis- produced within the tumor microenvironment might induce tochemistry and quantitative PCR (Fig. S1 D and E). As shown in G-CSF expression in LLC cells. To identify candidate regulatory Fig. 1G, Ets2 and G-CSF are coexpressed in biopsies of multiple factors, we incubated LLC cells in vitro with a panel of growth human cancers. factors. Several members of the platelet-derived growth factor (Fig. S4) and FGF (Fig. 2A) families stimulated G-CSF ex- Activation of the RAS Signaling Pathway Drives G-CSF Expression. pression. FGFs have been shown to activate the RAS signaling The Ets proteins are important for many cellular processes, pathway (28) through MAPK-induced Ets2 transcriptional

6080 | www.pnas.org/cgi/doi/10.1073/pnas.1303302110 Phan et al. Downloaded by guest on September 27, 2021 + that FGFs could stimulate aSMA cells to release G-CSF. We AB DMSO MEKi 450 * purified aSMA/CD105 double-positive myofibroblast-like cell * * 350 PBS EGF FGF 3FGF 5FGF 6FGF 8FGF 9 EGF FGF 5FGF 6FGF 8FGF 9 fractions (30) that are negative for CD31 to exclude endothelial bFGF PBS bFGFFGF 3 * 250 * pERK cell contamination from tumors. We confirmed that these cells * Total ERK 150 express aSMA (Fig. 2F) and CD105 (Fig. 2G) and are negative G-CSF (pg/ml) G-CSF F G + + − 50 10 DMSO for CD31 (Fig. 2 and ). Incubation of aSMA CD105 CD31 MEKi 8 -50 cells with FGFs resulted in G-CSF release in a MEK-dependent PBS EGF bFGF 6 FGF 3 FGF 5 FGF 6 FGF 8 FGF 9 manner (Fig. 2H). Growth Factors 4 * 2 * * * (G-CSF/Gapdh) * * Relative Fold Increase Relative Fold Increase 0 MEK Inhibition Markedly Reduces G-CSF Release in a Kras-Driven

PBS EGF bFGF GEMM. To determine whether targeting MEK activation could FGF 3 FGF 5 FGF 6 FGF 8 FGF 9 Growth Factors inhibit G-CSF release in Kras-driven tumors, we used the LSL-G12D fl/fl K-ras ;p16/p19 ;Pdx-Cre ductal adenocarcinoma geneti- C KPP14388 D KPP14449 8 PBS cally engineered mouse model (31, 32), previously shown to be * 1600 7 bFGF DMSO * 1400 PI3Ki * resistant to anti-VEGF monotherapy (32). PDAC tumor-bearing 6 MEKi 1200 5 1000 mice had higher G-CSF plasma levels than naive WT animals 4 * * 800 (Fig. S5B). Administration of MEKi significantly reduced G-CSF 3 600 * G-CSF (pg/ml) 2 400 ** levels in the plasma of tumor-bearing mice at both 7 h and 7 d

(G-CSF/Gapdh) * 1 200 A fi Relative Fold Increase Relative Fold Increase after treatment (Fig. S5 ). We next pro led cytokines and 0 0 CMV Ets2 FGFR4 Ets2 + growth factors released in the plasma of PDAC mice and com- CMV FGFR4 FGFR1 FGFR2 FGFR3 FGFR4 pared them with MEKi-treated or naive WT animals. In addition to G-CSF, many inflammatory growth factor and cytokine levels, E FHG α 9 including basic FGF, TNF- , GM-CSF, KC (CXCL1), and IL-17, DMSO 8 MEKi B α rease 7 were increased (Fig. S5 ). Among these factors, only TNF- and 6 fi 5 G-CSF decreased signi cantly on day 7 after MEKi treatment 4 (Fig. S5B and Table S2). Importantly, MEKi administration 3 + +

(G-CSF/Gapdh) * 2 resulted in decreased CD11b Ly6G neutrophil mobilization in Relative Fold Inc 1 * * * * * C 0 the peripheral blood of Kras-driven PDAC GEMM (Fig. S5 ). Mouse PDAC aSMA+ CD31- CD105+ CD31- PBS EGF bFGF FGF6 FGF8 FGF9 + + G-CSF Induces CD11b Ly6G Neutrophil Mobilization in Anti-VEGF– Fig. 2. FGFs regulate G-CSF release in tumor and stromal cells. (A) LLC cells + + were stimulated with various FGFs for 48 h. Conditioned media were col- Resistant Allograft Models. CD11b Gr1 myeloid cells are mixed lected and G-CSF concentrations were measured by ELISA (n = 3 per group), population of cells consisting of immature dendritic cells, early + + *P ≤ 0.001. Error bars indicate SD. Data are representative of at least two myeloid progenitors, Ly6C granulocytic monocytes and Ly6G + independent experiments. (B) Immunoblot analysis of phospho-ERK (pERK) neutrophils (8). Here we investigated which subsets of CD11b + and total ERK for LLC total lysates (Top). G-CSF levels were measured in the Gr1 myeloid population drive resistance to anti-VEGF therapy. + media of FGF-stimulated LLC cells after treatment with DMSO or MEKi We used antibodies that specifically recognize Ly6G neutrophils + (Bottom). Error bars indicate SD. (C) Different FGFRs are overexpressed in (33, 34), and Ly6C monocytes (13). In addition, we used G-CSF- −/− −/− + MEDICAL SCIENCES KPP14388 PDAC cells. G-CSF copy numbers were measured by quantitative R RAG2 mice, which exhibit reduced Ly6G neutrophil pop- ≤ −/− −/− PCR, *P 0.05. Error bars indicate SD. (D) MEKi, but not PI3K inhibitor, ulations (35). We confirmed that naive G-CSF-R RAG2 mice blocked FGFR4 and Ets2-induced G-CSF release in KPP14449 mouse pancre- + + haveasignificant reduction in CD11b Ly6G neutrophils com- atic cancer cells after 48 h incubation. G-CSF levels were measured by ELISA +/+ −/− = ≤ pared with G-CSF-R RAG2 mice (Fig. S6A), but show no (n 3 per group), *P 0.005. Error bars indicate SD. (E) Kras-driven PDAC fi + + GEMM tumors immunostained with anti-CD31 (red), anti-aSMA (red), and signi cant differences in the percentages of CD11b Ly6C μ fi fi monocytes (Fig. S6B). We next investigated the contribution of DAPI (blue). (Scale bar, 100 m.) (F) Puri ed tumor-associated myo broblast- + + like stellate cells from KPP14388 PDAC tumor immunostained with anti- CD11b Ly6G neutrophils to tumor resistance to anti-VEGF aSMA (red), anti-CD31 (green, negative staining), and DAPI (blue). (G) Puri- therapy. KPP14388 cells were s.c. implanted in immunodeficient +/+ −/− −/− −/− fied tumor-associated myofibroblast-like stellate cells from KPP14388 PDAC G-CSF-R RAG2 and G-CSF-R RAG2 animals. Four tumors immunostained with anti-CD105 (green), anti-CD31 (red, negative + + − days after implantation, mice were treated with either control staining), and DAPI (blue). (H) Purified aSMA CD105 CD31 cells were anti-Ragweed or anti-VEGF (B20-4.1.1) antibodies and tumor stimulated with FGFs in the presence or absence of MEKi, *P ≤ 0.05. Error volumes were measured. Anti-VEGF treatment had little effect +/+ −/− bars indicate SD. on tumor growth in WT G-CSF-R RAG2 mice (Fig. 3A). + + Also, CD11b Ly6G neutrophil reduction alone was not suffi- cient to reduce tumor growth. In contrast, anti-VEGF antibody activation (18). Accordingly, MAPK, as measured by ERK −/− treatment significantly reduced tumor growth in the G-CSF-R −/− phosphorylation, was activated in LLC cells upon stimulation RAG2 mice (Fig. 3A). Therefore, these data suggest that the with the different FGFs. MEKi treatment strongly inhibited ERK ability of anti-VEGF to reduce tumor growth is directly correlated + + phosphorylation and G-CSF release in the presence of FGFs with decreased CD11b Ly6G neutrophil mobilization (Fig. 3B). (Fig. 2B). Mutations or amplifications of the FGF receptors have been reported in many human cancer types (29); therefore, we MEKi Treatment Is Additive with Anti-VEGF in Inhibiting LLC Tumor fi investigated whether enforced FGF receptors expression is suf - Growth. In agreement with our previous finding (12), LLC tumors cient to induce G-CSF. Expression of all four FGF-receptors, were refractory to anti-VEGF therapy (Fig. S7A). MEKi or anti– FGFR1–4, could induce G-CSF expression (Fig. 2C). FGFR4 and G-CSF antibody single-agent treatment significantly inhibited Ets2 coexpression could induce G-CSF release in mouse pancre- G-CSF levels (Fig. S7B) and directly correlated with reduction in + + atic cancer cells in vitro. MEKi, but not PI3K inhibitor, inhibits Cd11b Ly6G neutrophil mobilization in the peripheral blood of FGFR4-enforced G-CSF expression (Fig. 2D). LLC tumor-bearing animals (Fig. S7C). Combination treatment Human PDACs have a large stromal component, including of MEKi plus anti-VEGF significantly reduced tumor growth alpha-smooth muscle actin (aSMA)-positive myofibroblast-like compared with anti-Ragweed or anti-VEGF monotherapy (Fig. stellate cells (5). Accordingly, mouse PDAC tumors are highly S7A). Importantly, combination treatment of MEKi plus anti- positive for aSMA markers (Fig. 2E). Because the stroma has VEGF or anti–G-CSF plus anti-VEGF resulted in marked re- + been proposed to be responsible for PDAC pathogenesis and duction in angiogenesis (Fig. S7 D and E) as measured by CD31 resistance to chemotherapeutic treatments, we hypothesized endothelial cell density relative to anti-Ragweed–treated animals.

Phan et al. PNAS | April 9, 2013 | vol. 110 | no. 15 | 6081 Downloaded by guest on September 27, 2021 ABaRAG aVEGF aRAG aVEGF 1000 G-CSFR+/+ RAG2-/- aRAG G-CSFR+/+ RAG2-/- aVEGF G-CSFR-/- RAG2-/- aRAG G-CSFR-/- RAG2-/- aVEGF Fig. 3. Targeting G-CSF in tumor is additive with

3 800 Ly6G anti-VEGF to reduce tumor angiogenesis and growth. (A) G-CSFR WT (G-CSFR+/+) or G-CSFR 600 − − knockout (G-CSFR / ) mice were crossed with RAG2 −/− +/+ Volume mm Volume knockout (RAG2 ) mice to generate G-CSFR r − − − − − − 400 RAG2 / and (G-CSFR / RAG2 / . Mice were trans- Tumo * Ly6C planted with KPP14388 PDAC cells (n = 8–9 per 200 group) and treated with anti-Ragweed (aRAG) control or anti-VEGF (aVEGF). Starting 3 d after cell 0 inoculation, tumor volumes were measured at sev- G-CSFR-/- RAG2-/- 4 8 12 G-CSFR+/+ RAG2-/- −11 Time (Days) eral time points, as indicated, *P < 1.0 × 10 . Error CD11b bars indicate SD. (B) Flow cytometry analysis of + + C D E peripheral blood for the presence of CD11b Ly6G aRAG (n=10) neutrophils. Data are representative of each group 2000 + 1800 1800 MEKi (n=10) 70 as indicated. Myeloid cells were gated for CD45 , aVEGF (n=10) 1600 ﬁ 1600 60 followed by analysis with an antibody that speci - aG-CSF (n=10) + 1400 MEKi + aG-CSF (n=10) 1400 cally recognizes Ly6G neutrophils. (C) KPP14388 50 * 1200 MEKi + aVEGF (n=10) 1200 * PDAC tumor growth in response to MEKi, anti- 1000 aG-CSF + aVEGF (n=10) 1000 40

olume mm3 * VEGF, anti-G-CSF, or combination treatments. Nu/

V * 800 800 30 * * * Nu mice were transplanted with KPP14388 cells (n = 600 600 * G-CSF (pg/ml) 20 10 per group). Three days after tumor cell in- 400 400 ** Tumor * * * 10 200 200 % CD11b+ Ly6G+ oculation, different treatments were initiated as ≤ 0 0 0 indicated, *P 0.001. Error bars indicate SD. (D)G- 3 5 7 11 14 18 CSF levels in the plasma of KPP14388 PDAC tumor- Time (Days) NaiveaRAGMEKi NaiveaRAGMEKi aVEGFaG-CSF aVEGFaG-CSF bearing mice; n = 10 per group, *P ≤ 0.001. Error bars indicate SD. (E) Flow cytometry analysis of MEKiaMEKi + aG-CSF + aVEGF MEKi + aG-CSF aVEGF + aG-CSF aMEKiaVEGF + aVEGF + aG-CSF peripheral blood of KPP14388 tumor-bearing mice were monitored for CD11b+Ly6G+ neutrophils (n = F G < % Microvessel Density 5 per group), *P 0.001. Error bars indicate SD. (F) Coverage KPP14388 PDAC tumor sections immunostained 0 1 2 3 4 5 6 7 with anti-CD31 (red). Mice were treated with anti- aRAG Ragweed (aRag), MEKi, anti-VEGF (aVEGF), anti–G- MEKi CSF (aG-CSF), or combination treatment as in- μ aRAG MEKi aVEGF aG-CSF aVEGF dicated. (Scale bar, 100 m.) (G) Quantitative analysis of tumor vascular surface area (microvessel aG-CSF density). Whole tumor cross-sections were stained MEKi + aG-CSF with anti-CD31 and analyzed as described in SI Ex- = ﬁ MEKi + aVEGF * perimental Procedures (n 4 per group). Signi - cance compared with aRag-treated group *P < 0.05. aVEGF + aG-CSF * MEKi + aG-CSF MEKi + aVEGF aVEGF + aG-CSF Error bars indicate SD.

+ + Combined Inhibition of G-CSF or MEK with Anti-VEGF Therapy Is the CD11b Ly6C monocyte population (Fig. 4D, bars 4 and 6), + + Efficacious in a PDAC Allograft Model. We next tested combina- suggesting that the CD11b Ly6C monocyte are not included tion therapies using MEKi and anti-VEGF or anti–G-CSF and in the G-CSF–induced myeloid cell mobilization in the PDAC anti-VEGF in PDAC mouse models. Anti–G-CSF or MEKi GEMM. To investigate survival in PDAC mice, we first stratified + + alone significantly reduced CD11b Ly6G neutrophil mobiliza- the cohorts by performing G-CSF ELISA and microultrasound tion, but had little effect on tumor growth (Fig. 3 C and E). analysis as previously described (32, 36, 37). Consistent with our Similarly, even though MEKi plus anti–G-CSF combination allograft studies, PDAC GEMM cohorts that received MEKi or + + therapy significantly reduced CD11b Ly6G neutrophil mobili- anti–G-CSF as single agent therapy had no significant survival zation, we did not observe significant reduction in tumor growth benefit relative to control (Fig. 4A), despite a marked reduction + + compared with monotherapies (Fig. 3C). However, we observed in the CD11b Ly6G neutrophil population (Fig. 4C). Consis- significant reductions in tumor growth following combination of tent with previous reports (32, 38), PDAC GEMM was resistant MEKi plus anti-VEGF or anti–G-CSF plus anti-VEGF (Fig. 3C). to anti-VEGF monotherapy (Fig. 4A). In contrast, combination Targeting G-CSF combined with anti-VEGF therapy significantly therapies significantly improved median survival compared with + reduced angiogenesis as measured by CD31 endothelial cells in control vehicle. MEKi and anti-VEGF combination treatment tumors (Fig. 3F). Indeed, quantitative analysis revealed marked resulted in increased survival (median survival 3.6 wk vs. 2.3 wk reduction in microvessel density in the combinations (Fig. 3G) for controls; P = 0.002). Similarly, anti–G-CSF and anti-VEGF compared with anti-Ragweed–treated group. combination resulted in a median survival of 3.7 wk, compared with 2.3 wk in the control group (P = 0.015) (Fig. 4A). We also Combining MEKi or Anti–G-CSF with Anti-VEGF Antibody Increases performed high-resolution microultrasound imaging to measure Survival in a Kras-Driven PDAC GEMM. We investigated whether tumor volumes in the cohorts and calculated the daily fold combination treatments with either MEKi and anti-VEGF or anti– change in the treated animals (Fig. S8). Anti–G-CSF plus anti- G-CSF and anti-VEGF could prolong overall survival in the VEGF or MEKi plus anti-VEGF combination therapy resulted LSL-G12D fl/fl K-ras ; p16/p19 ;Pdx-Cre PDAC GEMM (31). We first in slower tumor growth compared with control single-treatment examined the myeloid cell subpopulations in the PDAC GEMM arms (Fig. 4B). at day 7 after drug treatments (Fig. 4 C and D). Inhibition of G-CSF with either MEKi or anti–G-CSF significantly reduced MEK Pathway Activation and Neutrophil Recruitment in Human PDAC. + + CD11b Ly6G neutrophils (Fig. 4C) in the peripheral blood. The majority of patients diagnosed with PDAC harbor KRAS However, neutralizing G-CSF did not have a significant effect on mutations (20). We investigated whether there are any correlations

6082 | www.pnas.org/cgi/doi/10.1073/pnas.1303302110 Phan et al. Downloaded by guest on September 27, 2021 Median could abolish the majority of G-CSF expression, the inhibition AB OS (wks) Control (25) 2.3 was not complete. This could be attributed to activation of other aVEGF (20) 2.3 1.20 signaling pathways that drive G-CSF expression, including NF- aG-CSF (10) 1.7 1.18 MEKi (15) 2.1 1.16 κB (41). Interestingly, Ets proteins are phosphorylated by MAPK * * aVEGF+aG-CSF (11) 3.7 * 1.14 * aVEGF+ MEKi (17) 3.6 1.12 through the activation of the FGFR pathway (16). A recently 1.10 1.08 study has shown that Ets2 transcriptional activity in tumor- 1.06 associated ﬁbroblasts is responsible for the recruitment of macro- Fraction Survival 1.04 (Daily Fold Change) Tumor Burden 1.02 phages and for inducing tumor angiogenesis (42). We show that 1.00 enforced expression of Ets2 results in high G-CSF release in both

Weeks on Study MEKi tumor and stromal cells. Importantly, we document coexpression Control aVEGF aG-CSF of Ets2 and G-CSF in multiple human tumor types. Therefore, aVEGF + MEKi aVEGF + aG-CSF targeting Ets2 activity in tumors might lead to G-CSF down- CDregulation and favorable therapeutic outcomes for patients. 70 70 Because activation of the RAS/RAF/MEK signaling pathway 60 60

50 resulted in enhanced G-CSF expression, we hypothesized that

50 y6C+ L 40 40 growth factors and cytokines produced by cells within the tumor 1b+ 30 30 * microenvironment might activate G-CSF in tumor and stromal 20 * 20 * % CD1 *

% CD11b+Ly6G+ % CD11b+Ly6G+ * * cells. Indeed, we show that several growth factors can induce 10 10 G-CSF expression in a MEK-dependent manner. Levels of mul- 0 0 tiple inﬂammatory cytokines and growth factors were elevated in Naive aRAG MEKi Naive aRAG MEKi aVEGF aG-CSF aVEGF aG-CSF the peripheral blood of Kras-driven PDAC GEMM. Interestingly,

aVEGF + MEKi aVEGF + MEKi these factors are strong inducers of the RAS/RAF/MEK pathway aVEGF + aG-CSF aVEGF + aG-CSF and therefore could stimulate G-CSF release in both tumor and Fig. 4. Inhibition of G-CSF combined with anti-VEGF (aVEGF) increases stromal cells. Indeed, isolated PDAC tumor associated aSMA- survival in Kras-driven PDAC GEMM. (A) Kaplan-Meier plots showing overall positive myofibroblast-like stellate cells could readily express G- survival for the different treatment groups. Animals were treated as in- CSF upon FGFs stimulation by a MEK-dependent mechanism. dicated: control (anti-Ragweed and/or vehicle), MEKi (15 mg/kg), aVEGF (10 Further, in addition to cytokine-induced G-CSF release in the mg/kg), anti–G-CSF (aG-CSF) (50 μg/mouse). Overall survival was assessed by PDAC microenvironment, amplifications or mutations of FGFRs either mortality or severe morbidity. The number of animals per group is have been documented in many human cancers (43, 44). Here we shown, *P = 0.01. (B) Quantification of daily fold changes in tumor burden fi show that enforced expression of FGFRs in mouse PDAC cells by treatment regimen with approximate 95% con dence intervals. Tumor can induce G-CSF release. Consistent with our in vitro findings, growth analysis is based on serial ultrasounds taken at days 0, 7, 14, and 28; we observed constitutive activation of FGFR pathway, MEK *P < 0.01. Error bars indicate SD. (C) Flow cytometry analysis of peripheral + + phosphorylation and G-CSF overexpression in the majority of blood for the presence of CD11b Ly6G neutrophils. Total myeloid cells fi were gated for CD45+ and then quantified for CD11b+Ly6G+ neutrophils. human PDAC biopsies. Taken together, these ndings indicate Naive (n = 7), anti-Ragweed (n = 7), aVEGF (n = 10), aG-CSF (n = 10), MEKi that MEKi could target both tumor and stromal cells to reduce G- (n = 9), aVEGF+aG-CSF (n = 8), and aVEGF+MEKi (n = 5); *P = 0.0001. Error CSF expression. Because G-CSF activation is MEK-dependent, bars indicate SD. (D) Flow cytometry analysis of mouse peripheral blood for we hypothesized that targeting MEK activation could inhibit G- + + thepresenceofCD11b Ly6C monocytes. Total myeloid cells were gated for CSF expression in tumors and also increase tumor responses to MEDICAL SCIENCES + + + CD45 . Quantitative analysis of CD11b Ly6C monocytes is presented. Naive anti-VEGF therapy. MEKi and anti-VEGF combination therapy (n = 7), aRagweed (n = 7), aVEGF (n = 10), aG-CSF (n = 4), MEKi (n = 4), aVEGF+ significantly reduced tumor growth in multiple allograft models and aG-CSF (n = 3), and aVEGF+MEKi (n = 5); *P = 0.05. Error bars indicate SD. prolonged survival in a Kras-driven PDAC GEMM. Currently, MEK inhibitors are undergoing clinical development for treatment of melanomas and other malignancies with tumor cell–intrinsic between high G-CSF expression, phospho-MEK (pMEK), and activation of the RAS pathway (17). Our findings provide insights phospho-FGFR (pFGFR) in human PDAC biopsies. First, we into the mechanism of action of these agents and indicate that validated antibody-binding specificity to MEK and FGFR phos- they have the potential to have a major impact also on the tumor phorylation by performing control immunohistochemical staining microenvironment (Fig. S11). experiments (Fig. S9). In 116 patient PDAC biopsies, 83% of the We have previously reported that G-CSF is a major mediator of + + samples were positive for G-CSF (97/116), 81% were positive for CD11b Gr1 myeloid cell expansion and mobilization and is an pMEK (94/116), and 25% were positive for pFGFR (27/116) (Fig. inducer of anti-VEGF resistance through activation of proan- S10 A–D). Immunohistochemical staining revealed coexpressions giogenic pathways (11, 12). Neutralization of G-CSF resulted in + + of pMEK and G-CSF (82%) or pFGFR and G-CSF (26%) in the dramatic reduction in CD11b Gr1 cells in the plasma of tumor- human PDAC biopsies (Fig. S10F). Similar to our Kras-driven bearing mice (12, 13). We further characterized the myeloid cells PDAC GEMM, we found significant increases in neutrophil re- subpopulation that is responsible for G-CSF–induced resistance − − − − cruitment in G-CSF–positive human PDAC biopsies (Fig. S10E). to anti-VEGF therapy. We used both G-CSFR / RAG2 / mice + (35) and anti–G-CSF antibody or MEKi and found that CD11b + Discussion Ly6G neutrophil mobilization significantly contributes to anti- In humans, elevated plasma G-CSF levels have been reported in VEGF resistance therapy in multiple tumor models. + + a variety of solid tumors and may be associated with severe As already noted, a subset of CD11b Gr1 myeloid cells— leukocytosis and a poor prognosis (39). Anti–G-CSF treatment myeloid-derived suppressor cells (8)—is able to suppress of T + + results in a dramatic reduction in CD11b Gr1 myeloid cells and cell–dependent responses. Indeed, recent studies report that + + Bv8 levels in tumor and plasma of tumor-bearing mice (12, 13). GM-CSF could induce CD11b Gr1 cell mobilization and as + Although some mechanisms of G-CSF regulation had been de- a result suppress CD8 T-cell functions in a Kras-driven PDAC scribed in the literature (40), the precise signal transduction GEMM (45, 46). Interestingly, these studies reported that tar- + pathways regulating G-CSF in cancer cells have not been eluci- geting GM-CSF expression alone in PDAC could reduce CD11b + dated. In this study, we identified MEK activation as the major Gr1 mobilization and block tumor development. In contrast, our + + mechanism leading to G-CSF expression in tumor and stromal study shows that targeting the CD11b Ly6G myeloid cells with cells. Our analysis in 4T1-related mouse breast cancer cells anti–G-CSF as single agent does not significantly reduce tumor revealed that the Ets2 transcriptional factor directly binds to the growth and has little effect on survival. Similarly, MEKi treatment + + G-CSF promoter and regulates its expression. Although target- readily reduced G-CSF release and decreased CD11b Ly6G ing Ets2 transcriptional binding sites at the G-CSF promoter myeloid cell mobilization in PDAC, but MEKi treatment as single

Phan et al. PNAS | April 9, 2013 | vol. 110 | no. 15 | 6083 Downloaded by guest on September 27, 2021 agent failed to extend survival in tumor-bearing mice. In contrast, tumors (20). Our findings point to strategies for biomarker de- + + neutralizing G-CSF activity and G-CSF–induced CD11b Ly6G tection and combination therapies. neutrophils was effective when combined with anti-VEGF in re- ducing tumor growth and increasing survival in the Kras-driven Experimental Procedures fi − PDAC, both in immunocompetent and immunode cient mice. KrasLSL G12D mice were from T. Jacks (Massachusetts Institute of Technology, Also, our data suggest that T cell–dependent immune responses + + Boston, MA). p16/p19fl/fl mice were from A. Berns (Netherlands Cancer In- are not required for the protumor effects of CD11b Gr1 cells, stitute, Amsterdam, the Netherlands) and Pdx1-Cre mice from A. Lowy – − − which is consistent with a recent study that showed T cell medi- (University of Ohio, Cincinnati). G-CSF-R / mice were obtained from D. Link − − ated antitumor responses are not required for PDAC pathogen- (Washington University, St. Louis, MO). RAG2 / mice were purchased from esis, both in mouse models and in patients (47). Further studies Taconic. Female Nude/Nude BALB/c-mice were from Charles River Laboratory. are required to address these differences. Animals were housed and cared for according to guidelines from the In- Finally, PDAC remains one of the most lethal malignancies stitutional Animal Care and Use Committee at Genentech, Inc. with an average 5-y survival rate of less than 5% (48); therefore, new therapeutic strategies are urgently needed. The majority of ACKNOWLEDGMENTS. We thank the Flow Cytometry laboratory for sup- patients diagnosed with PDAC have acquired KRAS mutations port, and M. Gonzalez, H. Ngu, and A. Crow for immunohistochemistry. and enhanced activation of RAF/MAPK pathway, which result We also thank J. Kaminker for bioinformatics support and E. Choo, L. Rangell, in enhanced cell proliferation, survival, and metastasis of the C. Bais, and M. Singh for helpful insights and discussions.

1. Ferrara N, Gerber HP, LeCouter J (2003) The biology of VEGF and its receptors. Nat 27. McDermott U, et al. (2007) Identification of genotype-correlated sensitivity to selec- Med 9(6):669–676. tive kinase inhibitors by using high-throughput tumor cell line profiling. Proc Natl 2. Ferrara N, Kerbel RS (2005) Angiogenesis as a therapeutic target. Nature 438(7070): Acad Sci USA 104(50):19936–19941. 967–974. 28. Turner N, Grose R (2010) Fibroblast growth factor signalling: From development to 3. Crawford Y, Ferrara N (2009) Tumor and stromal pathways mediating refractoriness/ cancer. Nat Rev Cancer 10(2):116–129. – resistance to anti-angiogenic therapies. Trends Pharmacol Sci 30(12):624 630. 29. Motoda N, et al. (2011) Overexpression of fibroblast growth factor receptor 4 in high- 4. Coussens LM, Werb Z (2002) Inflammation and cancer. Nature 420(6917):860–867. grade pancreatic intraepithelial neoplasia and pancreatic ductal adenocarcinoma. Int 5. Erkan M, et al. (2012) The role of stroma in pancreatic cancer: Diagnostic and ther- J Oncol 38(1):133–143. apeutic implications. Nat Rev Gastroenterol Hepatol 9(8):454–467. 30. Meurer SK, Tihaa L, Lahme B, Gressner AM, Weiskirchen R (2005) Identification of 6. Yang L, et al. (2004) Expansion of myeloid immune suppressor Gr+CD11b+ cells in endoglin in rat hepatic stellate cells: New insights into transforming growth factor tumor-bearing host directly promotes tumor angiogenesis. Cancer Cell 6(4):409–421. – 7. Yang L, et al. (2008) Abrogation of TGF beta signaling in mammary carcinomas re- beta receptor signaling. J Biol Chem 280(4):3078 3087. fi cruits Gr-1+CD11b+ myeloid cells that promote metastasis. Cancer Cell 13(1):23–35. 31. Aguirre AJ, et al. (2003) Activated Kras and Ink4a/Arf de ciency cooperate to produce – 8. Gabrilovich DI, Nagaraj S (2009) Myeloid-derived suppressor cells as regulators of the metastatic pancreatic ductal adenocarcinoma. Genes Dev 17(24):3112 3126. immune system. Nat Rev Immunol 9(3):162–174. 32. Singh M, et al. (2010) Assessing therapeutic responses in Kras mutant cancers using 9. Shojaei F, et al. (2007) Tumor refractoriness to anti-VEGF treatment is mediated by genetically engineered mouse models. Nat Biotechnol 28(6):585–593. CD11b+Gr1+ myeloid cells. Nat Biotechnol 25(8):911–920. 33. Daley JM, Thomay AA, Connolly MD, Reichner JS, Albina JE (2008) Use of Ly6G-specific 10. Hamilton JA (2008) Colony-stimulating factors in inflammation and autoimmunity. monoclonal antibody to deplete neutrophils in mice. J Leukoc Biol 83(1):64–70. Nat Rev Immunol 8(7):533–544. 34. Tate MD, et al. (2009) Neutrophils ameliorate lung injury and the development of 11. Shojaei F, et al. (2007) Bv8 regulates myeloid-cell-dependent tumour angiogenesis. severe disease during influenza infection. J Immunol 183(11):7441–7450. Nature 450(7171):825–831. 35. Liu F, Wu HY, Wesselschmidt R, Kornaga T, Link DC (1996) Impaired production and 12. Shojaei F, et al. (2009) G-CSF-initiated myeloid cell mobilization and angiogenesis increased apoptosis of neutrophils in granulocyte colony-stimulating factor receptor- mediate tumor refractoriness to anti-VEGF therapy in mouse models. Proc Natl Acad deficient mice. Immunity 5(5):491–501. – Sci USA 106(16):6742 6747. 36. Wirtzfeld LA, et al. (2005) A new three-dimensional ultrasound microimaging tech- 13. Kowanetz M, et al. (2010) Granulocyte-colony stimulating factor promotes lung nology for preclinical studies using a transgenic prostate cancer mouse model. Cancer metastasis through mobilization of Ly6G+Ly6C+ granulocytes. Proc Natl Acad Sci USA Res 65(14):6337–6345. 107(50):21248–21255. 37. Graham KC, et al. (2005) Three-dimensional high-frequency ultrasound imaging for 14. Dexter DL, et al. (1978) Heterogeneity of tumor cells from a single mouse mammary longitudinal evaluation of liver metastases in preclinical models. Cancer Res 65(12): tumor. Cancer Res 38(10):3174–3181. 5231–5237. 15. Oettgen P (2010) The role of ets factors in tumor angiogenesis. J Oncol 2010:767384. 38. Singh M, Ferrara N (2012) Modeling and predicting clinical efficacy for drugs tar- 16. Sharrocks AD (2001) The ETS-domain transcription factor family. Nat Rev Mol Cell Biol – 2(11):827–837. geting the tumor milieu. Nat Biotechnol 30(7):648 657. 17. Choor EF, et al. (2012) Preclinical disposition of GDC-0973 and prospective and ret- 39. Granger JM, Kontoyiannis DP (2009) Etiology and outcome of extreme leukocytosis in rospective analysis of human dose and efficacy predictions. Drug Metab Dispos 40(5): 758 nonhematologic cancer patients: A retrospective, single-institution study. Cancer 919–927. 115(17):3919–3923. 18. Frémin C, Meloche S (2010) From basic research to clinical development of MEK1/2 40. Meixner A, et al. (2008) Epidermal JunB represses G-CSF transcription and affects inhibitors for cancer therapy. J Hematol Oncol 3:8. haematopoiesis and bone formation. Nat Cell Biol 10(8):1003–1011. 19. Harris TJ, McCormick F (2010) The molecular pathology of cancer. Nat Rev Clin Oncol 41. Dunn SM, et al. (1994) Requirement for nuclear factor (NF)-kappa B p65 and NF- 7(5):251–265. interleukin-6 binding elements in the tumor necrosis factor response region of the 20. Forbes SA, et al. (2008) The Catalogue of Somatic Mutations in Cancer (COSMIC). Curr granulocyte colony-stimulating factor promoter. Blood 83(9):2469–2479. Protoc Hum Genet Chapter 10:Unit 10.11. 42. Trimboli AJ, et al. (2009) Pten in stromal fibroblasts suppresses mammary epithelial 21. Hatzivassiliou G, et al. (2010) RAF inhibitors prime wild-type RAF to activate the tumours. Nature 461(7267):1084–1091. – MAPK pathway and enhance growth. Nature 464(7287):431 435. 43. Knights V, Cook SJ (2010) De-regulated FGF receptors as therapeutic targets in cancer. 22. Poulikakos PI, Zhang C, Bollag G, Shokat KM, Rosen N (2010) RAF inhibitors Pharmacol Ther 125(1):105–117. transactivate RAF dimers and ERK signalling in cells with wild-type BRAF. Nature 44. Kompier LC, et al. (2010) FGFR3, HRAS, KRAS, NRAS and PIK3CA mutations in bladder 464(7287):427–430. cancer and their potential as biomarkers for surveillance and therapy. PLoS ONE 5(11): 23. Hoeflich KP, et al. (2009) Antitumor efficacy of the novel RAF inhibitor GDC-0879 e13821. is predicted by BRAFV600E mutational status and sustained extracellular signal- 45. Pylayeva-Gupta Y, Lee KE, Hajdu CH, Miller G, Bar-Sagi D (2012) Oncogenic Kras- regulated kinase/mitogen-activated protein kinase pathway suppression. Cancer Res 69(7):3042–3051. induced GM-CSF production promotes the development of pancreatic neoplasia. – 24. Scholl C, et al. (2009) Synthetic lethal interaction between oncogenic KRAS de- Cancer Cell 21(6):836 847. pendency and STK33 suppression in human cancer cells. Cell 137(5):821–834. 46. Bayne LJ, et al. (2012) Tumor-derived granulocyte-macrophage colony-stimulating 25. Sordella R, Bell DW, Haber DA, Settleman J (2004) Gefitinib-sensitizing EGFR muta- factor regulates myeloid inflammation and T cell immunity in pancreatic cancer. tions in lung cancer activate anti-apoptotic pathways. Science 305(5687):1163–1167. Cancer Cell 21(6):822–835. 26. Yamamoto N, Mammadova G, Song RX, Fukami Y, Sato K (2006) Tyrosine phos- 47. Beatty GL, et al. (2011) CD40 agonists alter tumor stroma and show efficacy against phorylation of p145met mediated by EGFR and Src is required for serum-independent pancreatic carcinoma in mice and humans. Science 331(6024):1612–1616. survival of human bladder carcinoma cells. J Cell Sci 119(Pt 22):4623–4633. 48. Hidalgo M (2010) Pancreatic cancer. N Engl J Med 362(17):1605–1617.

6084 | www.pnas.org/cgi/doi/10.1073/pnas.1303302110 Phan et al. Downloaded by guest on September 27, 2021