The Unfolded Protein Response Induces the Angiogenic Switch in Human Tumor Cells through the PERK/ATF4 Pathway

Dividing tumors can rapidly outgrow their blood supply. This results in a toxic tumor microenvironment (TME) characterized with hypoxia, acidic pH, glucose deprivation (GD), and amino acid deficiency. Increasing evidence suggests that the TME contains stressors that promote accumulation of misfolded proteins in the lumen of the endoplasmic reticulum (ER). This in turn activates intracellular signaling pathways termed as the unfolded protein response (UPR; refs. 1–3). Initially, the UPR is cytoprotective aimed at restoring normal ER function (4). However, in the presence of severe or prolonged ER stress, cell death programs are activated (3, 4).

Mammalian UPR is controlled by 3 ER-resident transmembrane proteins that serve as proximal sensors of ER stress and activate downstream signaling effectors, such as protein kinase RNA–like ER kinase (PERK), inositol requiring 1 (IRE1), and activating transcription factor 6 (ATF6; ref. 5). These effectors are maintained in an inactive state through association with the molecular chaperone, glucose-regulated protein 78 kDa (Grp78; ref. 2). Upon ER stress Grp78 dissociates from these sensors, which activates UPR signaling (6). After dimerization and transautophosphorylation, activated PERK relays signal by phosphorylating the alpha subunit of eukaryotic initiation factor-2 (eIF-2α), which in turn inhibits general translation initiation and selectively translates several mRNAs including ATF4 (7). ATF4 transactivates expression of several genes, such as C/EBP homologous protein (CHOP), a transcription factor implicated in apoptosis (8, 9), and Grp78 (10). Upon UPR activation, IRE1 splices an unconventional intron from the X-box-binding protein 1 (XBP1) mRNA, producing an active transcription factor XBP1-s (11, 12), which translocates into the nucleus and modulates the expression of several proteins involved in folding or clearance of aberrant proteins (2, 13). Finally, during ER stress, ATF6 is processed into an active transcription factor, moves into the nucleus, and upregulates ER chaperones and folding enzymes (2, 14).

Studies have shown that GD induces the expression of VEGF in different tumor cell lines (15–17), suggesting that besides hypoxia (18, 19), low concentration of nutrients play a role in triggering angiogenesis (20). It is also known that the pathologic stimulus GD causes ER stress and alters gene expression through UPR signaling (20). However, mechanistic studies of UPR-mediated angiogenesis have been conducted for the most part using nonphysiologic agents, such as thapsigargin (TG), focusing mainly on VEGF expression (21, 22). The mechanism underlying GD-induced UPR in tumor angiogenesis has not been fully elucidated.

In this study, we report that activation of the UPR by GD plays a pivotal role in tumor angiogenesis by activating the angiogenic switch and that the PERK/ATF4 pathway of the UPR is involved in this process. These results suggest that targeting the proangiogenic arm of the UPR may reveal new strategies for cancer treatment.

The head and neck squamous cell carcinoma (HNSCC) cell line UM-SCC-81B (from Dr. Thomas E. Carey, Departments of Otorhinolaryngology and Pharmacology, University of Michigan Medical School, Ann Arbor, MI), the breast cancer cell line MCF7 (American Type Culture Collection), the glioma cell line U87 (from Dr. Yi Sun, Department of Radiation Oncology, University of Michigan, Ann Arbor, MI), and mouse embryonic fibroblast (MEF) cell lines (MEF–PERK+/+ and MEF–PERK−/−, from Dr. Andrew Fribley, Department of Pediatrics, Division of Hematology/Oncology, Wayne State University, Detroit, MI) were maintained in Dulbecco's Modified Eagle's Medium (DMEM) with high glucose (Invitrogen) with 10% FBS. All tumor cell lines were authenticated recently by DNA fingerprinting with short tandem repeat profiling. Primary human dermal microvascular endothelial cells (HDMEC; Cambrex) were cultured in endothelial cell growth medium-2 (EGM2; Cambrex). TG and tunicamycin (TM) were purchased from Sigma.

Immunofluorescence and immunohistochemistry (IHC) were conducted as described previously (23) with primary antibodies against Grp78, CHOP (Santa Cruz), Ki67, and CD31 (BD Pharmingen). Goat antirabbit Alexa Flour 594 (Molecular Probes) and 4′,6-diamidino-2-phenylindole (DAPI) counterstaining were used for immunofluorescence staining. The Polink-2 horseradish peroxidase broad kit with 3,3′-diaminobenzidine (DAB) chromogen was used for IHC staining. Normal human oral mucosa (NHM, from Dr. Hector Rios, Department of Periodontics and Oral Medicine, University of Michigan School of Dentistry, Ann Arbor, MI) was used as control. All imaging was done using a Leica DM5000 microscope.

Laser capture microdissection (LCM) was conducted as previously described (24). Approximately 500,000 epithelial cells from either HNSCC or NHM were collected using a pulsed 337-nm UV laser. RNA from at least 15 independent tumors and 10 NHM tissues were pooled respectively and analyzed using real-time PCR (qPCR).

Total RNA was isolated from UM-SCC-81B cells treated with or without GD (glucose, 0.1 mmol/L) using the RNeasy Plus Mini Kit (Qiagen). A Human gene chip U133 Plus 2.0 (Affymetrix) was used to analyze gene expression. Details and results can be accessed in Gene Expression Omnibus (GEO) repository (GSE38583).

Whole-cell lysates were resolved by SDS–PAGE and transferred onto polyvinylidene difluoride membrane (Pierce) and probed with the antibodies: Grp78, CHOP, ATF4, β-actin (Santa Cruz), HIF1α (BD Pharmingen), PERK (Cell Signaling), and spliced XBP1 (Biolegend). Horse radish peroxidase-conjugated secondary antibodies were from Santa Cruz. SuperSignal West Pico Chemiluminescent Substrate (Pierce) was used to visualize immunoreactive bands.

ELISA was conducted according to the manufacturer's instructions. Briefly, cell culture supernatants were diluted 1:10, applied to each well (100 μL), incubated at room temperature for 2 hours, and washed 3 times. The secondary antibody reaction was conducted at room temperature (1 hour). Stabilized chromogen was used for colorimetric reactions. Optical density was measured at 450 nm using a plate reader (Spectra Max M2).

GFP-expressing lentiviral constructs expressing short hairpin RNA (shRNA) against PERK, ATF4, and HIF1α were from Open Biosystems. For infection, 1 × 10⁵ cells were plated in 6-cm plates, infected with the lentivirus and sorted with flow cytometry to ensure 100% positivity. Established stable cell lines were cultured with 2 μg/mL puromycin.

Chromatin immunoprecipitation (ChIP) analysis was conducted using an Agarose-Chip Kit (Pierce) according to the manufacturer's instruction. Cells were treated for 18 hours with complete DMEM containing normal glucose (25 mmol/L) or low glucose (2 mmol/L), respectively. The chromatin solution was incubated overnight with ATF4 antibody (sc-200, Santa Cruz), nonimmune rabbit IgG, and anti-RNA polymerase II antibody. Purified complexes and input DNAs were analyzed by PCR. The ATF4-binding sites of VEGF gene (AsnSyn site) and asparagine synthetase gene (NSRE-1, nutrient-sensing response element, used as a positive control) have been described previously (25, 26). Amplicons were resolved on 1% agarose gels and visualized by SYBR Safe DNA gel stain (Invitrogen).

Total mRNA was extracted from cultured cells with RNeasy plus minikit (Qiagen) following manufacturer's instructions. cDNA was synthesized using the Verso cDNA Kit (Thermo Scientific). Real-time PCR was conducted in 384-well plate with the ABI PRISM 7900HT Sequence Detection System. Primers used for qPCR (Grp78, CHOP, 18S, ATF4, VEGF, IL-6, FGF-2, CXCL10, and CXCL14) were from Applied Biosystems. RT-PCR was also used to check gene expression (IL-6, VEGFA, FGF-2, CTGF, Grp78, and 18s; details in Supplementary Data).

HDMECs (1.5 × 10⁵ cells) were seeded in 6-well plates containing 1.5 mL layer of gelled type I collagen (Cohesion) for 24 hours, then cultured in EBM2 (2% FBS) containing concentrated GDCM (conditioned medium collected from cells cultured with GD) or NGCM (conditioned medium collected from cells cultured with regular medium) for 3 days. Number of sprouts in 12 random microscope fields per well was counted on day 3 in triplicate per condition. Images were taken with a Leica DMI 3000B.

Tumor cells (UM-SCC-81B-scshRNA and UM-SCC-81B-shPERK, 5 × 10⁵) were injected subcutaneously in the flanks of severe combined immunodeficient mice (SCID) mice (Harlan Laboratories). Tumor volume was measured every 3 days from day 17 postinjection with a digital caliper. Tumor volumes were calculated using the formula volume (mm³) = length × width²/2. At the endpoint, tumors were surgically removed and analyzed for tumor angiogenesis.

Data are expressed as mean ± SD and analyzed using unpaired 2-tailed Student t test. A value of P < 0.05 was considered to be significant.

To investigate the role of UPR in HNSCC, we examined expression of UPR markers Grp78 and CHOP in patients with HNSCC (10 patients) and NHM (5 controls). Strong expression of both Grp78 (85%) and CHOP (88%) was detected in HNSCC as compared with NHM (11% and 30%, respectively; Fig. 1A–C and Supplementary Fig. S1), suggesting that the UPR was activated in tumor samples. With LCM, epithelial cells from both HNSCC (15 patients) and NHM (10 controls; Fig. 1D) were collected, and qPCR was carried out to determine relative expression levels of target genes. Significant Grp78 increase was observed in HNSCC (Fig. 1D), confirming activation of the UPR in cancer cells. Simultaneously, mRNA levels of IL-6 and VEGF showed a 2.9- and 3.5-fold increase, respectively, and the expression level of the antiangiogenic chemokine CXCL14 showed significant decrease (Fig. 1D). These results suggest that in human tumor samples, UPR is associated with a shift in the balance between pro- and antiangiogenic mediators in favor of angiogenesis.

To explain results obtained from human tumors, we investigated the role of GD-induced UPR in modulating angiogenesis related gene expression in vitro. Upon GD treatment (2–25 mmol/L), UPR was activated as shown by phosphorylation of PERK (upward shift in the bands) and increased expression of ATF4, Grp78, and spliced XBP1 (Fig. 2A). The strength of UPR (shown by dose-dependent increase of Grp78) correlates with upregulation of VEGF at both protein and mRNA levels (Fig. 2A and Supplementary Fig. S2A and S2B). Furthermore, mRNA levels of FGF-2 and IL-6 displayed the same trend as VEGF in response to gradient glucose treatment (Supplementary Fig. S2B–S2D). Results of RT-PCR analysis also showed that transcription of the UPR marker Grp78 as well as the proangiogenic factors VEGF, IL-6, CTGF, and FGF-2 were increased with GD treatment (Fig. 2B). Increased secretion of IL-6 and FGF-2 in response to GD was also observed (Fig. 2C).

Interestingly, the expression of CXCL10, an antiangiogenic factor was reduced with GD treatment at both mRNA and protein levels (supernatant; 5- and 2-fold, respectively; Fig. 2D). The other 2 chemical UPR inducers, TM and TG also potently suppressed its expression at mRNA levels (Fig. 2D).

To further understand how angiogenesis related genes respond to GD, UM-SCC-81B cells were subjected to GD for 4 hours and 24 hours, and genes of interest were analyzed by cDNA microarray. A panel of proangiogenic mediators, such as VEGF, IL-6, FGF-2, TGFB2, NRG1, and NGF, were upregulated and several antiangiogenic mediators, such as CXCL10, CXCL14, and THBS1 were inhibited (Table 1). Collectively, these data suggest that UPR modulates the angiogenic switch by adjusting the balance between pro- and antiangiogenic mediators in favor of angiogenesis.

To investigate whether the tumor angiogenic response was not limited to GD only, UM-SCC-81B cells were treated with chemical ER stressors, TM (1 μg/mL) and TG (1 μmol/L). Like GD, TM and TG were able to activate the UPR (Fig. 3A) and significantly increase VEGF secretion (Fig. 3B).

To eliminate the possibility of a cell-specific response, the breast cancer cell line MCF7 and the glioma cell line U87 were both subjected to GD (2 mmol/L), TG (1 μmol/L), and TM (1 μg/mL) treatment with subsequent assessment of UPR activation and VEGF secretion. We observed consistent UPR activation in both MCF7 and U87 (Fig. 3A) and increased expression of VEGF (Fig. 3B) with all 3 stressors. In addition, 2 other HNSCC cell lines, UM-SCC-11B and UM-SCC-17B showed upregulation of the proangiogenic mediators VEGF, FGF2, and IL-6 upon GD treatment (Supplementary Fig. S3A–S3D). Collectively, these data indicate that the UPR plays an important role in regulating production of angiogenic mediators in tumor cells regardless of their origin.

Hypoxia promotes the production of multiple proangiogenic mediators, including VEGF, platelet-derived growth factor (27), IL-6 (28), FGF-2 (29), and placental growth factor (30). Stein and colleagues reported that hypoxia transcriptionally increases VEGF expression and stabilizes its mRNA through HIF1α (17), and HIF1α has been reported to be involved in GD-induced VEGF expression in mouse embryonic cells (31). To address whether HIF1α is involved in GD-induced VEGF expression through UPR activation in tumor cells used, we knocked down the expression of HIF1α with shRNA. As seen from Fig. 4A, HIF1α knockdown was neither able to inhibit GD-induced upregulation of Grp78, XBP1-s, and phosphorylation of PERK nor inhibit GD-induced expression of VEGF, FGF-2, and IL-6 (Fig. 4B and C), However, knockdown of HIF1α inhibited CoCl₂ (a mimetic of hypoxia)–induced VEGF and IL-6 expression at mRNA levels but not FGF2 (Fig. 4C). This may be because IL-6 (28) and VEGF (32) are transcriptionally controlled by hypoxia, whereas FGF-2 is not (29). Collectively, the results indicate that HIF1α is not involved in UPR-mediated production of angiogenic mediators.

Upon UPR activation, approximately one third of UPR-responsive gene transcription requires phosphorylation of eIF2α, suggesting signaling from PERK regulates the UPR at transcriptional level (33) and is important for survival of tumor cells (34). We therefore, focused on the role of PERK/ATF4 pathway in the production of angiogenic mediators. During UPR, PERK activation leads to eIF2α phosphorylation, an increase of ATF4 expression, and subsequent upregulation of CHOP and Grp78 (35, 36). To examine the role of PERK/ATF4 pathway in angiogenic mediator production, lentiviral vectors containing shRNA against PERK and ATF4 were used to infect UM-SCC-81B cells, generating stable cell lines with specific knockdown of PERK and ATF4, respectively. As shown in Fig. 5A, shRNAs inhibit more than 70% of PERK expression and lead to reduced expression of its downstream genes (Grp78 and ATF4) upon GD treatment. In addition, PERK knockdown in UM-SCC-81B cell line decreased VEGF expression at both protein and mRNA levels (P < 0.05; Fig. 5B and C). Expression levels of FGF-2 and IL-6 were also significantly suppressed (Fig. 5C). To further corroborate the results above, PERK knockout (PERK −/−) MEFs were treated with GD (2 mmol/L). In PERK −/− cells, expression of VEGF was significantly suppressed as compared with PERK +/+ cells (Supplementary Fig. S4A–S4C).

As a downstream effector of PERK, it is conceivable that ATF4 is also involved in GD-induced angiogenic factor production. Studies have shown that homocysteine, an ER stressor, and arsenite increase VEGF transcription in an ATF4-dependent manner (25, 37). By knocking down ATF4, we observed decreased expression of Grp78 (Fig. 6A) and proangiogenic factors (VEGF, IL-6, and FGF-2) were significantly suppressed (P < 0.05; Fig. 6B and C).

The ChIP assay is widely used to show the interaction between transcription factor activation and cis-acting elements. The “AsnSyn” site in the VEGF gene is an important ATF4 binding site (25, 26). To further verify the role of ATF4 in regulating VEGF expression, we examined the binding of ATF4 to the “AsnSyn” site using ChIP assay. As shown in Fig. 6D, specific PCR product was present in GD-treated samples immunoprecipitated with antibody against ATF4. In addition, GD also stimulated formation of a complex between ATF4 and a functionally important NSRE-1 site in the promoter of asparagine synthetase (an ATF4 target gene) as mentioned earlier (26, 38). PCR bands for GAPDH were observed in the samples immunoprecipitated with anti-RNA polymerase II with or without GD treatment. No PCR products were observed in rabbit IgG immunoprecipited samples (Fig. 6D, left). The quantification from ChIP assay clearly shows that GD promotes strong interaction between the VEGF “AsnSyn” site and ATF4 (Fig. 6D, right). Collectively, these results show that the PERK/ATF4 pathway plays a pivotal role in GD-induced VEGF production.

To determine how the UPR stimulates angiogenesis, we used conditioned medium (CM) collected from GD-treated tumor cells (GDCM) to induce HDMEC proliferation and sprout formation on collagen gels. When exposed to angiogenic stimuli, HDMECs proliferate, migrate, and organize into capillary-like structures (39, 40). To assess sprout formation, HDMECs were grown in the presence of CM collected from UM-SCC-81B-scshRNA and UM-SCC-81B-shPERK7 with or without GD treatment. GDCM showed strong ability in promoting capillary-like sprout formation on collagen as compared with CM collected from cells grown in normal glucose (NGCM). PERK knockdown, however, decreased the ability of GDCM in inducing sprout formation (Fig. 7A).

During sprout formation assay, we noted consistently higher confluence rates in cells cultured in the presence of GDCM, despite equal seeding of cells, suggesting GDCM might be able to promote cell proliferation. MTT assay verified this observation. As shown in Fig. 7A, GDCM promoted stronger cell proliferation as compared with NGCM (P < 0.05). NGCM supplemented with FGF2 showed the strongest ability in promoting HDMEC proliferation (Fig. 7A).

The effect of PERK knockdown on tumor proliferation and blood vessel formation was also evaluated in vivo. As seen, PERK knockdown significantly reduced the tumor volume and slowed down tumor progression (Fig. 7B). Knockdown of PERK reduced Ki67-positive cells from 44.6% in controls to 12% in PERK knockdown tumors, suggesting that PERK plays an important role in tumor growth (Fig. 7C). As expected, suppressing PERK expression in tumor cells also reduced the blood vessel density from 8 to 4 per field, which can reduce tumor volume due to a decreased blood supply (Fig. 7D). To confirm these results, a rapid growing tumor cell line (UM-SCC-74B) was used in vivo. In UM-SCC-74B, PERK knockdown significantly reduced tumor proliferation as shown by tumor volume, tumor weight, and Ki67 staining (Supplementary Fig. S5A and S5B). And blood vessel density in UM-SCC-74B tumors decreased from 17 to 8 per field (Supplementary Fig. S5C).

TME is often hypoxic and nutrient deficient due to an inadequate blood supply. This in turn leads to activation of the UPR. The main function of the UPR is to maintain or restore ER homeostasis in response to environmental stress. It is therefore not surprising that one mechanism for relieving this stress is by increasing the blood supply to tumors. It has been shown that in prostate and ovarian carcinoma, tumor cells subjected to GD show elevated levels of VEGF production (41, 42). However, due to the complex nature of the angiogenic process, it is likely that multiple proangiogenic mediators produced at the tumor site accumulate in sufficient quantities to offset antiangiogenic mediators. With immunostaining, we found strong expression of the UPR markers Grp78 and CHOP in tumors from patients with HNSCC, whereas in normal human mucosa there was only weak expression of Grp78 and CHOP. Meanwhile, LCM analysis revealed expression levels of proangiogenic factors were elevated in tumor tissues, whereas the antiangiogenic mediator CXCL14 was significantly reduced. This supports the notion that UPR activation coordinates the expression of pro- and antiangiogenic mediators to facilitate blood vessel formation, which subsequently reduces stress.

Similar to tumor samples, tumor cell lines treated with GD showed UPR activation along with simultaneous increase of proangiogenic genes, such as VEGF, FGF-2, IL-6, CTGF, etc., and decrease in a number of antiangiogenic genes (CXCL14, CXCL10, and THBS1), with CXCL14 showing the most significant reduction (28.21-fold). These results confirm that UPR plays a pivotal role in the angiogenic switch in tumors.

Pereira and colleagues recently reported that treatment of the human medulloblastoma cell line (Daoy) with the chemical ER stressor TG, induced expression of proangiogenic factors including IL-8, angiogenin, angiopoietin-2, etc., and reduced expression of the antiangiogenic protein vasohibin (21). Interestingly, our studies revealed major differences in the spectrum of genes involved in angiogenesis as compared with that reported by Pereira and colleagues, which suggests different stressors induce a distinct pattern of angiogenic mediator expression. Indeed, we found that GD (2 mmol/L) and TM (1 μg/mL) can induce strong VEGF expression but not IL-8 expression. TG, at 100 nmol/L, cannot induce strong VEGF expression but can induce significant IL-8 expression in UM-SCC-81B cells (Supplementary Fig. S6A and S6B). Higher TG concentration (1 μmol/L) can, however, induce VEGF (Fig. 3A). This apparent discrepancy may be because TG releases calcium from the ER into the cytosol and IL-8 expression is dependent on the change in cytosolic calcium (43). We contend that activation of the UPR and subsequent activation of the angiogenic switch by GD more closely resembles the events occurring in the tumor microenvironment. Nevertheless, these observations corroborate our hypothesis that UPR activation in tumors induces the angiogenic switch. Indeed, HDMEC treated with GDCM produce more capillary-like sprouts on collagen gel and suppressing PERK/ATF4 pathway of the UPR reduces tube formation ability of GDCM in inducing tube formation.

It has been reported that both GD and hypoxia activate the UPR (44). The UPR component PERK, has been linked to translation inhibition under hypoxic stress (45) and XBP1 has been shown to be essential to the survival of transformed cells in response to hypoxia (46). The best-characterized stress-induced regulator of tumor angiogenesis is HIF1α. Upregulation of VEGF in response to GD was found to be HIF1α-dependent in mouse embryonic stem cells (31), suggesting the TME-related stimuli, GD, and hypoxia, share similar signaling pathways. However, others and we showed that while GD could activate UPR, it could not promote accumulation of HIF1α (21, 47). Furthermore, although the hypoxia mimetic, CoCl₂ promoted the accumulation of HIF1α, it was unable to promote overexpression of UPR markers Grp78, XBP1-s, and phosphorylation of PERK. Knockdown of HIF1α had no effect on GD-induced VEGF, IL-6, and FGF-2 expression. These results suggest that the pathways activated following GD and CoCl₂ treatment are different and HIF1α is not involved in angiogenic mediator production that is associated with GD-induced UPR. Our studies confirm earlier observations that HIF1α was not involved in UPR-mediated VEGF expression (21, 22).

The PERK/ATF4 pathway has been reported to mediate VEGF mRNA expression in mouse cell lines (MEF, neruo2A), human umbilical vein endothelial cells (HUVEC), human retinal pigment epithelial cell (ARPE-19), and hepatic carcinoma cell line (HepG2) following activation of the UPR with TG or TM (21, 22, 25, 26). However, its role in GD-induced UPR-mediated angiogenesis in human tumors has not been reported. We show here that the PERK/ATF4 pathway of the UPR is involved not only in VEGF expression but also FGF-2 and IL-6 expression in human tumor cells treated with GD. An approximately 70% knockdown of PERK is sufficient to reduce expression of these proangiogenic mediators. In vivo results confirmed that PERK plays an important role in tumor proliferation and neovascularization. Direct knockdown of the downstream target of PERK, ATF4, displayed an even stronger inhibitory effect in reducing VEGF, IL-6, and FGF-2 expression. This suggests a central role of ATF4 in the production of proangiogenic mediators. A possible explanation is that as an ATF, ATF4 regulates the expression of these genes directly.

In summary, our studies show that GD-induced UPR activation initiates an angiogenic switch that alters the balance of pro- and antiangiogenic mediators. The resulting proangiogenic environment could function to relieve the stress by increasing the blood supply to tumors. We also found that the PERK/ATF4 arm of UPR signaling is a pivotal pathway responsible for upregulating the production of multiple proangiogenic mediators. In conclusion, these results suggest that the role of UPR-mediated stress response must be taken into consideration as a potential target in the design of new cancer therapies.

No potential conflicts of interest were disclosed.

Conception and design: Y. Wang, F. Visioli, J.E. Nör, P.J. Polverini

Development of methodology: Y. Wang, F. Visioli, Z. Dong, J.E. Nör

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Wang, G.N. Alam, Y. Ning, F. Visioli, Z. Dong

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y. Wang, G.N. Alam, J.E. Nör, P.J. Polverini

Writing, review, and/or revision of the manuscript: Y. Wang, G.N. Alam, J.E. Nör, P.J. Polverini

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y. Wang, Y. Ning, Z. Dong

Study supervision: Y. Wang, P.J. Polverini

The authors thank members from Jacques E. Nör's laboratory, Dr. Andrew Fribley, and members from Randal J. Kaufman's Laboratory for helpful discussions and assistance. The authors also thank University of Michigan core facilities for their technical assistance.

This work is supported by The Sharon and Lauren Daniels Cancer Research Fund and University of Michigan, Office of the Provost, grant P50-CA97248, from the NIH/NCI, and grant R01-DE21139 from the NIH/NIDCR.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.