TC

KW

*DMC

Bctfiuaeeisbtcpogifioawmwgbisapc

Tmsalctmef

GASTROENTEROLOGY 2008;134:1981–1993

he Essential Role of Fibroblasts in Esophageal Squamous Cellarcinoma–Induced Angiogenesis

AZUHIRO NOMA,* KEIRAN S. M. SMALLEY,* MERCEDES LIONI,* YOSHIO NAOMOTO,‡ NORIAKI TANAKA,‡

AFIK EL–DEIRY,§ ALASTAIR J. KING,� HIROSHI NAKAGAWA,¶ and MEENHARD HERLYN*

The Wistar Institute, Philadelphia, Pennsylvania; ‡Departments of Gastroenterological Surgery, Transplant and Surgical Oncology, Graduate School of Medicine,entistry and Pharmaceutical Sciences, Okayama University, Okayama, Japan; §Hematology-Oncology Division and ¶Gastroenterology Division, Department ofedicine, Department of Genetics, Abramson Cancer Center, University of Pennsylvania School of Medicine, Philadelphia, Pennsylvania; and �GlaxoSmithKline,

ollegeville, Pennsylvania

Otfeapts

ronmososg(cput(mrvbsft

bfiEGtbf

BA

SIC–

ALI

MEN

TARY

TRA

CT

ackground & Aims: Esophageal squamous cell car-inoma (ESCC) is known to be a highly angiogenicumor. Here, we investigated the role of the stromalbroblasts in the ESCC-induced angiogenic responsesing a novel 3-dimensional model. Methods: A novelssay was developed where cocultures of ESCC andsophageal fibroblasts induced human microvascularndothelial cell (HMVEC) vascular network formationn a 3-dimensional collagen gel. Biochemical studieshowed that the ESCC-induced activation of the fibro-lasts was required to induce vascular network forma-

ion via a transforming growth factor (TGF)-� and vas-ular endothelial growth factor (VEGF)-dependentathway. Results: Conditioned media from a panelf 4 ESCC lines transdifferentiated normal esopha-eal fibroblasts into myofibroblasts via TGF-� signal-ng. The presence of fibroblasts was essential for ef-cient HMVEC network formation, and the additionf ESCC cells to these cultures greatly enhanced thengiogenic process. The role of TGF-� in this processas shown by the complete inhibition of network for-ation following TGF-� inhibitor treatment. Finally,e showed that ESCC-derived TGF-� regulates angio-enesis through the release of VEGF from the fibro-lasts and that the VEGF release was blocked follow-

ng TGF-� inhibition. Conclusions: This studyhows the essential role of fibroblasts in the ESCCngiogenic-induced response and suggests that theharmacologic targeting of the TGF-� signaling axisould be of therapeutic benefit in this deadly disease.

he tumor “organ” consists of a dynamic mixture oftumor cells, fibroblasts, endothelial cells, and im-

une cells that all work together to drive tumor progres-ion.1 Activated fibroblasts, also known as carcinoma-ssociated fibroblasts (CAFs),2 have been identified at theeading edges of many solid tumors, including breast,olon, and melanoma.3–5 The presence of CAFs withinhe tumor microenvironment is preceded by the che-

oattraction and migration of precursor cells, which canither arise from the surrounding host fibroblasts or

rom circulating mesenchymal precursors/stem cells.6,7

nce recruited, paracrine tumor-derived growth factors ac-ivate the CAFs, which undergo a myofibroblastic transdif-erentiation, defined by an elongated spindle shape, and thexpression of contractile �–smooth muscle actin (�-SMA)nd vimentin.8 CAFs are hypothesized to drive tumorrogression through the deposition of extracellular ma-rix proteins, the secretion of growth factors, and thetimulation of invasion.9

One area that has been little explored is the potentialole of CAFs in tumor angiogenesis. Much of the growthf solid tumors is dependent on the ready supply ofutrients and oxygen from a local blood supply. As tu-ors grow beyond a few millimeters in size, they readily

utstrip the local supply of nutrients available throughimple diffusion and stimulate the formation of theirwn tumor vasculature. Although it has been shown thattromal fibroblasts are an important source of the proan-iogenic factor vascular endothelial cell growth factorVEGF),10 it has been difficult to study the interaction ofarcinoma cells, fibroblasts, and endothelial cells in ahysiologically relevant model. In the present study, wesed a novel 3-dimensional (3D) in vitro model in whichhe interaction of esophageal squamous cell carcinomaESCC) cells with fibroblasts drives vascular network for-

ation in a 3D collagen gel. We show that ESCC cellsequire the presence of stromal fibroblasts to stimulateascular network formation, thereby suggesting that fi-roblasts are the critical mediators of angiogenesis in thisystem. Mechanistic studies reveal that paracrine trans-orming growth factor (TGF)-� from the ESCC cells leadso activation of the fibroblasts and that pharmacologic

Abbreviations used in this paper: CAF, carcinoma-associated fibro-last; DAPI, 4=,6-diamidino-2-phenylindole; DMEM, Dulbecco’s modi-ed Eagle medium; ELISA, enzyme-linked immunosorbent assay;SCC, esophageal squamous cell carcinoma; FBS, fetal bovine serum;FP, green fluorescent protein; HMVEC, human microvascular endo-

helial cell; MTT, 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazoliumromide; �-SMA, �–smooth muscle actin; TGF, transforming growthactor; 3D, 3-dimensional; VEGF, vascular endothelial growth factor.

© 2008 by the AGA Institute0016-5085/08/$34.00

doi:10.1053/j.gastro.2008.02.061

ia

Tsbdthiorbptotb

a(ttTaO4(nTopzwU

cwvmpwvmtw1c

ifaBwbngV(eNN

tcnAcsm

1cmfs

4f

wnwmcStTt

tmdm

BA

SIC–

ALIM

ENTA

RY

TRA

CT

1982 NOMA ET AL GASTROENTEROLOGY Vol. 134, No. 7

nhibitors of TGF-� signaling can reverse both fibroblastctivation and vascular network formation.

Materials and MethodsCell LinesEsophageal cancer cells, TE cell lines (TE1, TE8,

E10, TE11, TE12), were cultured as previously de-cribed.11 Human esophageal keratinocytes EPC2 haveeen described previously.11,12 Human microvascular en-othelial cells (HMVECs) are available commerciallyhrough Cascade Biologics, Inc (Portland, OR).13 Primaryuman esophageal fibroblasts designated as FEF3 were

solated from human fetal esophagus as described previ-usly.11 FEF3 cells were stably transduced using the Vi-aPower Lentiviral Expression System (Invitrogen, Carls-ad, CA) containing the gene for green fluorescentrotein (GFP). GFP lentivirus was raised in our labora-ory, and the FEF3 cells were transduced in the presencef 6 �g/mL polybrene. Forty-eight hours after transduc-ion, cells were selected in the presence of 10 �g/mLlasticidin for 14 days.

Antibodies and ReagentsThe following antibodies were used in this study:

nti-human CD31 (Dako, Carpinteria, CA), anti–�-SMASigma-Aldrich, St Louis, MO), anti–von Willebrand fac-or (Neomarkers, Fremont, CA), anti–fibroblast activa-ion protein (EMD Biosciences, San Diego, CA), anti-GF�RII (Santa Cruz Biotechnology, Santa Cruz, CA),nd phalloidin/Texas Red (Molecular Probes, Eugene,R). The anti-smad2, anti–phospho-smad2 (Ser465/

67), anti–phospho-smad3 (Ser423/425), and Smad1Ser463/465) antibodies were purchased from Cell Sig-aling Technology (Beverly, MA). Recombinant humanGF-�1 was purchased by R&D Systems, Inc (Minneap-lis, MN). SB-505124, GW788388, and GW654652 wererovided by GlaxoSmithKline (Collegeville, PA). Bevaci-umab (Avastin; Genentech, South San Francisco, CA)as obtained from the pharmacy of the Hospital of theniversity of Pennsylvania.

In Vitro 3D Network Formation Assay andFluorescence ImagingReconstruction of a vessel-like structure in 3D

ollagen gels and subsequent fluorescent staining of net-orks/cords in whole-mount gels were performed as pre-

iously described.13 Briefly, HMVECs were cultured asonolayers on bovine type I collagen-coated 24-well

lates at 1.5 � 105 cells/well for 24 hours and overlaidith acellular collagen mixed in 10� Medium 199 (In-

itrogen) with heparin (100 U/mL), vitamin C (50 �g/L), and fetal bovine serum (FBS; 1%). After polymeriza-

ion of the collagen gels, the cells were further overlaidith a second collagen layer containing each 1.0 –2.5 �05 cells/mL FEF3, 0.5 � 105 cells/mL TE cells, or both

ells. Wells were then filled with EBM-2 medium contain- w

ng EGM-2MV. The reconstructs were incubated at 37°Cor 7 days. To prepare for staining, medium was removednd the collagen gels were fixed in Prefer (Anatech Ltd,attle Creek, MI) for 4 hours at room temperature. Gelsere processed as whole mounts. After blocking with 1%ovine serum albumin, gels were stained with monoclo-al anti-CD31 antibody followed by Texas Red– conju-ated secondary antibody. Gels were treated withectaShield with 4=,6-diamidino-2-phenylindole (DAPI)

Vector Laboratories, Burlingame, CA), and the stainedndothelial cell networks were photographed under aikon E600 fluorescent microscope (Nikon, Melville,Y).The capillary-like networks were scored by counting

he number of CD31-stained branches. One branch wasounted to be 3 cells thick or less (to discount disorga-ized masses of HMVECs) and at least 3 whole cells long.t least 5 randomly chosen low-power fields wereounted per sample (10� magnification). Each figurehows one representative experiment. Data show the

ean of at least 3 independent experiments.

Immunofluorescence MicroscopeFor cocultures, FEF3 cells and TE cells mixed in a

:1 ratio (2.5 � 104 cells per each) were seeded onto glassoverslips in 6-well plates and cultured in Dulbecco’sodified Eagle medium (DMEM) containing 10% FBS

or 48 hours. Cells were then fixed and stained as de-cribed previously.11

Western Blotting AnalysisSubconfluent cells were lysed and separated on a

%–12% sodium dodecyl sulfate/polyacrylamide gel be-ore being blotted as described previously.11

Treatment of FEF3 Cells With ConditionedMediaFor preparation of conditioned medium, TE cells

ere cultured with DMEM containing 10% FBS over-ight. Supernatants were removed, and cells were washedith DMEM. Cells were cultured for 48 hours with freshedium DMEM containing 2% FBS. FEF3 cells were

ultured overnight with DMEM containing 10% FBS.upernatants were replaced and cultured with fresh cul-ure medium DMEM containing 2% FBS with or withoutGF-�1 or cultured with conditioned medium for each

ime.

TGF-�1/2 Enzyme-Linked ImmunosorbentAssayCells were cultured overnight with DMEM con-

aining 10% FBS. Supernatants of these cells were re-oved, and cells were washed with DMEM basal me-

ium. Cells were cultured for 48 hours with fresh cultureedium DMEM containing 2% FBS. These supernatants

ere measured as samples using each enzyme-linked im-

Ft3lvii

BA

SIC–

ALI

MEN

TARY

TRA

CT

June 2008 FIBROBLASTS DRIVE ESCC–INDUCED ANGIOGENESIS 1983

igure 1. The interaction of ESCC cells and fibroblasts drives efficient vascular network formation in 3D culture. Immunofluorescence staining shows GFP-agged FEF3 esophageal fibroblasts (green), HMVECs (CD31; red), and total nuclei (DAPI; blue). (Inset) CD31 staining. (A) Incubation of the HMVECs with each ofESCC lines (TE1, TE8, TE11) was not associated with any vascular network formation. (B) Coculture of the HMVECs with human esophageal fibroblasts (FEF3)

ed to increased vascular network formation. (C) The addition of ESCC cells to the fibroblast/endothelial cell coculture markedly increased the organization of theascular networks. Increasing the culture time to 14 days dramatically enhanced the organization of the vascular network. (D) ESCC number was also found to

ncrease the degree of vascular network formation. (E) The extent of network formation under each of the culture conditions was scored, with the addition of

ncreasing numbers of ESCC cells found to significantly increase the level of vascular network formation. Scale bar � 200 �m.

BA

SIC–

ALIM

ENTA

RY

TRA

CT

1984 NOMA ET AL GASTROENTEROLOGY Vol. 134, No. 7

mTTtwmNt

cwpDmcpchV

owSapd

mtt

noms(tm

tdivf(fatdvs1a(Hvm1aosaw

untppbTndTdcafifwb

4FG4faS

BA

SIC–

ALI

MEN

TARY

TRA

CT

June 2008 FIBROBLASTS DRIVE ESCC–INDUCED ANGIOGENESIS 1985

unosorbent assay (ELISA). ELISA kits for TGF-�1,GF-�2, and VEGF were purchased from R&D Systems.hese assays were performed according to the manufac-

urer’s instructions. For TGF-�1 and TGF-�2, samplesere activated by adding 0.1 mL of 1 mol/L HCl for 10inutes and neutralized with 100 �L of 1.2 mol/LaOH/0.5 mol/L HEPES before assay to measure the

otal amount of TGF-�.

VEGF ELISAFor 2-dimensional cocultures, fibroblasts and TE1

ells at a 1:1 ratio were seeded onto plates and culturedith or without SB-505124 compound overnight. Su-ernatants were removed, and cells were washed withMEM. Cells were cultured for 48 hours with fresh cultureedium DMEM containing 2% FBS with or without the

ompound. These supernatants were measured as sam-les using VEGF ELISA. For 3D culture, each sample wasultured with or without SB-505124 compound for 48ours, and these supernatants were measured using theEGF ELISA kit (R&D Systems).

Cell Proliferation AnalysisCells were plated into a 96-well plate at a density

f 2.5 � 104 cells/mL and left to grow overnight. Cellsere treated with increasing concentrations of TGF-�,B-505124, or GW788388 in triplicate. Proliferation wasnalyzed using the 3-(4,5-dimethylthiazol-2-yl)-2,5-di-henyltetrazolium bromide (MTT) assay as previouslyescribed.14

StatisticsUnless otherwise stated, all experiments show the

ean � SD of at least 3 independent experiments. Sta-istical significance was measured using the Student test, where P � .05 was judged to be significant.

ResultsThe Presence of Fibroblasts Is Required forESCC-Induced AngiogenesisStudying the role of fibroblasts in ESCC-induced

etwork formation in vivo is technically challenging. Tovercome some of these issues, we developed a novel 3Dodel of ESCC-induced angiogenesis, allowing us to

tudy the role of CAFs in vascular network formation13

see Supplementary Figure 1 for scheme; see supplemen-al material online at www.gastrojournal.org). In the

odel, factors derived from either fibroblasts or a mix-

™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™igure 2. Coculture of esophageal fibroblasts with ESCC cells leadsFP-tagged fibroblasts (green) expressed very little �-SMA. (B–D) Fibro8 hours stain strongly for stress fibers of �-SMA (red). (E) Western blo

rom ESCC lines for 47 hours leads to increased �-SMA expression ins a loading control. (F) Conditioned media from human esophageal k

cale bar � 100 �m.

ure of fibroblasts and ESCC cells cause HMVECs toetach from the tissue culture plate and migrate upward

nto the 3D collagen gel, where they organize and formascular networks. In an initial series of studies, it wasound that coculture of the ESCC lines with the HMVECsTE1, TE8, and TE11) did not lead to vascular networkormation (Figure 1A). However, coculture of the esoph-geal fibroblasts (FEF3) with the HMVECs led to endo-helial cell migration and formation of moderately well-ifferentiated vascular networks as shown by CD31 andon Willebrand factor staining (Figure 1B and nothown). An increase in the fibroblast concentration from� 105 cells/mL to 1.5 � 105 and 2.5 � 105 cells/mL was

ssociated with increased vascular network organizationFigure 1B). Addition of ESCC cells to the fibroblast/

MVEC cocultures had the most striking effects onascular network formation and led to the establish-ent of very organized capillary-like structures (Figure

C). Increasing the numbers of both the fibroblastsnd ESCC cells was associated with significantly morerganized capillary networks (Figure 1B, D, and E),howing that the interaction of both the ESCC cellsnd fibroblasts was critical for efficient vascular net-ork formation.

Coculture of Esophageal Fibroblasts WithESCC Cells Leads to Their Activation andTransdifferentiation Into Myofibroblasts

CAFs are typically in an activated state, havingndergone transdifferentiation to a myofibroblast phe-otype. Growing the human esophageal fibroblasts withhe ESCC line TE1 for 48 hours led to a change inhenotype associated with the increased cytoplasmic ex-ression of the myofibroblast marker �-SMA and fibro-last activation protein (Figure 2A and B and not shown).hree additional esophageal carcinoma lines were alsooted to induce a similar degree of myofibroblast trans-ifferentiation (TE11, TE8, and TE12; Figure 2C and D).he induction of the myofibroblast phenotype was in-uced through soluble factors derived from the ESCCells, as shown by the ability of conditioned media frompanel of ESCC lines to induce �-SMA expression in thebroblasts (Figure 2E). In contrast, conditioned media

rom primary human esophageal keratinocytes (EPC2)ere unable to induce �-SMA expression in the fibro-lasts (Figure 2F).

™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™eir activation and transdifferentiation into myofibroblasts. (A) Controlcocultured with the ESCC lines (TE1, TE11, TE8, TE10, and TE12) for

ysis showing that incubation of fibroblasts with conditioned media (CM)hageal fibroblasts. Blots were stripped and reprobed with anti–�-actinocytes does not induce �-SMA expression in esophageal fibroblasts.

™™™to th

blastst analesoperatin

s3f

cimTtpe

BA

SIC–

ALIM

ENTA

RY

TRA

CT

1986 NOMA ET AL GASTROENTEROLOGY Vol. 134, No. 7

Paracrine TGF-� Secreted From ESCC CellsIs Responsible for Esophageal FibroblastTransactivationELISA experiments showed that all 4 ESCC lines

ecreted high levels of both TGF-�1 and TGF-�2 (FigureA). To rule out the role of other possible growth factors,urther studies were performed showing that the ESCC

ells secreted very little platelet-derived growth factor ornsulin-like growth factor (data not shown). In a similar

anner to that of ESCC-conditioned media, exogenousGF-� was also shown to induce esophageal fibroblast

ransdifferentiation associated with increased �-SMA ex-ression (Figure 3B). Stimulation of the fibroblasts withxogenous TGF-� was also accompanied by increased

SpEsi(TnHafTgbt(mtgo

itcTTafwk(cnoe(

an

cecs�4acTiwiia3itGtit

bnHl5tn(babocc(fi

4Ft(i(r(nnseg�

BA

SIC–

ALI

MEN

TARY

TRA

CT

June 2008 FIBROBLASTS DRIVE ESCC–INDUCED ANGIOGENESIS 1987

MAD signaling, as shown by the rapid increase in phos-ho-SMAD2 (Figure 3B). Evidence for TGF-�1 being theSCC-derived fibroblast-activating factor came from theimilar induction of nuclear SMAD3 accumulation seenn both exogenous TGF-�1 and ESCC conditioned mediaFigure 3C). Next, we investigated whether exogenousGF-� would stimulate the fibroblasts to drive vascularetwork formation. Treatment of esophageal fibroblast/MVEC cocultures with 1 ng/mL TGF-� for 7 days led tostatistically significant increase in capillary network

ormation (Figure 3D) and showed the essential role ofGF-�–induced fibroblast differentiation in the angio-enic process. To show the essential role of the fibro-lasts in TGF-�–induced vascular network formation, wereated 3D monocultures of endothelial cells with TGF-�1 ng/mL) and observed very little vascular network for-

ation. Indeed, the endothelial cells remained attachedo the plates and did not migrate upward into the colla-en (Supplementary Figure 2; see supplemental materialnline at www.gastrojournal.org).

TGF-� plays a complex role in tumor progression ands known to be growth inhibitory to most epithelial cellypes. To overcome the effects of TGF-�, most carcinomaells escape by down-regulating their expression ofGF-� receptors and instead secrete autocrine/paracrineGF-� that recruits the surrounding stromal cells. Ingreement with this idea, the ESCC lines tested were notound to express any TGF-� receptor subtype II protein,hereas esophageal fibroblasts and the parent esophagealeratinocyte line EPC2 maintained receptor expressionFigure 3E). The ESCC lines were shown to have es-aped the inhibitory effects of TGF-� and proliferatedormally in the presence of increasing concentrationsf TGF-� (0.1–10 ng/mL), whereas the primary humansophageal keratinocyte line EPC2 was growth arrestedFigure 3F).

The TGF-�–Specific Inhibitor SB-505124Inhibits ESCC-Induced FibroblastTransdifferentiationPharmacologic approaches that block tumor neo-

ngiogenesis are an attractive therapeutic option. Weext tested whether the specific inhibitor of TGF-� re-

™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™igure 3. ESCC lines secrete paracrine TGF-� that induces the transd

otal TGF-�1 and TGF-�2 by esophageal cancer cell lines (TE). TE celB) Exogenous TGF-�1 induces �-SMA expression in fibroblasts and snduces �-SMA expression in esophageal fibroblasts. Exogenous rhTC) Both exogenous rhTGF-�1 and ESCC conditioned media induce SMhTGF-�1 (1 ng/mL) or conditioned medium of TE1 for 1 hour. Imagegreen), cell morphology is indicated by phalloidin (red), and nuclei are inetwork formation in the absence of ESCC cells. Fibroblasts and HMVECetwork formation was stained for CD31 (red), fibroblasts (GFP; green)ignificantly (P � .005) increased numbers of microcapillary networks pxpression in a panel of ESCC lines, fibroblasts, and primary human esorowth of human esophageal keratinocytes (EPC2) but not ESCC lines.

m. (D) Scale bar � 200 �m.

eptor kinase SB-505124 blocked fibroblast transdiffer-ntiation. Pretreating the esophageal fibroblasts with in-reasing concentrations of SB-505124 (0.01–10 �mol/L)uccessfully blocked the TGF-�–induced increases in-SMA expression in the esophageal fibroblasts (FigureA). Maximal inhibition of the TGF-�–induced fibroblastctivation was seen at 1 �mol/L SB-505124, and this con-entration was selected for all subsequent experiments.reatment of the cells with SB-505124 also blocked �-SMA

nduction and SMAD2 activation following treatmentith ESCC conditioned media (Figure 4B and C). To

nvestigate whether TGF-� inhibition was growth inhib-tory on any of the cell types used in the neoangiogenesisssay, endothelial cells (HMVEC), fibroblasts (FEF3), and

ESCC lines (TE1, TE8, and TE12) were treated withncreasing concentrations of each of 2 structurally dis-inct small molecule inhibitors of TGF-� (SB-505124 andW788388) for 72 hours. Measuring levels of prolifera-

ion using the MTT assay showed that neither TGF-�nhibitor affected the growth of any of the cell typesested (Figure 4D and E).

The TGF-� Inhibitors SB-505124 andGW788388 Inhibit ESCC-Induced VascularNetwork FormationWe next investigated whether inhibition of fi-

roblast activation blocked ESCC-induced vascularetwork formation. Coculture of fibroblasts with theMVECs led to endothelial cell migration but very

ittle differentiated vascular network formation (FigureA). Treatment of the fibroblast/HMVEC cocultures withhe TGF-� inhibitor GW788388 (1 �mol/L) did not sig-ificantly alter the level of vascular network organization

Figure 5A), showing that there was only limited fibro-last activation in the absence of ESCC cells. As before,ddition of the ESCC lines TE1 and TE8 to the fibro-last/HMVEC coculture dramatically increased the levelf vascular network organization (Figure 5B and C). Inontrast, treatment of the ESCC/fibroblast/HMVEC co-ultures with either SB-505124 (1 �mol/L) or GW7883881 �mol/L) led to a complete reversal of vascular networkormation (Figure 5B and C). These results show thatnhibition of fibroblast activation by small molecule

™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™™tiation of esophageal fibroblasts. (A) ELISA assay showing secretion ofre cultured for 48 hours, and secreted TGF was measured by ELISA.tes TGF-related signaling pathways. Exogenous rhTGF-�1 (48 hours)

1 induces SMAD2 phosphorylation in human esophageal fibroblasts.ctivation in esophageal fibroblasts. Fibroblasts were treated with eitherw p-SMAD3 up-regulation and its subsequent nuclear translocation

ed by DAPI (blue). (D) Exogenous TGF-�1 markedly enhances vascularre cocultured in the presence of TGF-�1 (1 ng/mL) for 7 days. Vascularnuclei (DAPI; blue). (Inset) CD31 staining alone. The graph shows the

ld treated by TGF-�1. (E) Western blot showing expression of TGF�RIIal keratinocytes (EPC2). (F) Exogenous TGF-�1 (72 hours) reduces the

proliferation was measured by the MTT assay. Images: scale bar � 50

™™™ifferenls wetimulaGF-�AD3 as shodicats we

, ander fiephageCell

Te

gVdcupcw(TVnwclfSnsficsT5

lEf

VctscGE(fettfswdmwEw

ii

FSceaE�Hf e eva

BA

SIC–

ALIM

ENTA

RY

TRA

CT

1988 NOMA ET AL GASTROENTEROLOGY Vol. 134, No. 7

GF-� inhibitors blocks ESCC-induced neoangiogen-sis (Figure 5D).

Pharmacologic Inhibition of TGF-� InhibitsVEGF Secretion From Both EsophagealFibroblasts and ESCC CellsVEGF is one of the most potent inducers of an-

iogenesis. To investigate the possible role of paracrineEGF release from esophageal fibroblasts in ESCC-in-uced neoangiogenesis, we tested VEGF release fromocultured and TGF-�–stimulated esophageal fibroblastssing an ELISA. Unstimulated esophageal fibroblastsroduce a basal level of VEGF under 2-dimensional cellulture conditions, and this was significantly increasedhen the fibroblasts were grown under 3D conditions

Figure 6A and B). Treatment of the fibroblasts with theGF-� inhibitor SB-505124 did not reduce basal levels ofEGF under 2-dimensional cell culture conditions (dataot shown); however, it did inhibit basal VEGF releasehen the fibroblasts were grown under 3D cell culture

onditions (Figure 6B). This showed that there was someevel of fibroblast activation when the cells were switchedrom the 2-dimensional to 3D cell culture conditions.timulation of the fibroblasts with exogenous TGF-� (1g/mL) led to a significant increase in the level of VEGFecreted (Figure 6A). In a similar manner, coculture of thebroblasts with the ESCC line TE1 also led to a signifi-ant increase in VEGF secretion under both 2-dimen-ional and 3D cell culture conditions (Figure 6A and B).reatment of the ESCC/fibroblast cocultures with SB-

igure 4. Pharmacologic inhibitors of TGF-� signaling block ESCC-indB-505124 blocks TGF-�–induced fibroblast transdifferentiation. Fibrooncentrations of SB-505124 (0.01–10 �mol/L) for 48 hours, followexpression in fibroblasts stimulated by ESCC-conditioned medium (CM;nd Methods in the absence or presence of SB-505124 (1 �mol/L) foSCC-conditioned media. Fibroblasts were cultured with TGF-�1 (1 nmol/L) for 1 hour. (D and E) TGF-� inhibitors have little effect on thMVECs, and esophageal cancer cell lines (TE) were treated with increa

or 72 hours before being subjected to the MTT assay. The results wer

05124 (1 �mol/L) led to a significant inhibition in the d

evel of VEGF, showing a possible mechanism by whichSCC-activated fibroblasts can induce vascular network

ormation.As a final test, we determined whether inhibition of

EGFR2 signaling via the selective inhibitor GW654652ould also block ESCC-induced vascular network forma-ion. Coculture of the HMVECs and fibroblasts led toome limited vascular network formation after 7 days ofulture, and this was not affected by the presence ofW654652 (1 �mol/L) (Figure 6C). Treatment of theSCC, fibroblast, and HMVEC cultures with GW654652

1 �mol/L) did not completely inhibit vascular networkormation but did lead to a reduction in the number ofndothelial cell tubes formed (Figure 6C). Analysis ofhese results showed a significant reduction (P � .005) inhe extent of ESCC-induced vascular network formationollowing GW654652 treatment (Figure 6D). Similar re-ults were also seen upon ESCC-induced vascular net-ork formation following treatment with an antibodyirected against VEGF receptor (bevacizumab) (Supple-entary Figure 3; see supplemental material online atww.gastrojournal.org). A sample scheme showing howSCC cells stimulate fibroblasts to drive vascular net-ork formation is shown in Figure 7.

DiscussionHere we show for the first time, using a novel 3D

n vitro model of the tumor microenvironment, the crit-cal role of the host fibroblasts in directing ESCC-in-

fibroblast transdifferentiation. (A) The TGF-� receptor kinase inhibitors were cultured in the presence of TGF-�1 (1 ng/mL) and increasingblotting for �-SMA expression. (B) SB-505124 inhibits �-SMA overTE1 cells). Fibroblasts were cultured with CM as described in Materials

hours. (C) SB-505124 inhibits phosphorylation of SMAD2 induced by) or conditioned medium as indicated and treated with SB-505124 (1liferation of ESCC cells, fibroblasts, or HMVECs. Fibroblasts (FEF3),concentrations of TGF�RI inhibitors (either SB-505124 or GW788388)luated as a percentage of control absorbance.

ucedblastd byfromr 48g/mLe prosing

uced vascular network formation. This model differs

Fvoivr

BA

SIC–

ALI

MEN

TARY

TRA

CT

June 2008 FIBROBLASTS DRIVE ESCC–INDUCED ANGIOGENESIS 1989

igure 5. Pharmacologic inhibitorsofTGF-� signalingblockESCC-inducedvascularnetwork formation. (A) TGF-� inhibitionhas little effectonfibroblast-inducedascularnetwork formation.Fibroblasts (FEF3;1.5�105 cells/mL)were incubatedwithHMVECs in theabsenceorpresenceof1of2TGF-� inhibitors (SB-505124r GW788388, both 1 �mol/L) for 7 days. Cultures were stained for HMVECs (CD31; red), fibroblasts (GFP; green), and nuclei (DAPI; blue). (B and C) TGF-�

nhibition completely blocks ESCC-induced vascular network formation. Addition of either SB-505124 or GW788388 (both 1 �mol/L) led to complete inhibition ofascular network formation. (D) Bar graph shows mean data for vascular network formation in the absence and presence of either SB-505124 or GW788388. All

epresentative images are shown as 3-color merges, and original monochrome CD31 (white) images are inset. Scale bar � 200 �m.

FTFSoafa

BA

SIC–

ALIM

ENTA

RY

TRA

CT

1990 NOMA ET AL GASTROENTEROLOGY Vol. 134, No. 7

igure 6. Activated fibroblasts secrete VEGF, leading to increased vascular network formation. (A) Stimulation of esophageal fibroblasts with eitherGF-� through coculture with ESCC cells leads to enhanced VEGF release in 2-dimensional adherent culture. Fibroblasts (FEF3) in monoculture orEF3 and ESCC cells (TE1) at a ratio of 1:1 in coculture were cultured in the absence or presence of SB-505124 (1 �mol/L) for 48 hours.upernatants were harvested and quantified for VEGF expression using a specific ELISA. (B) Stimulation of esophageal fibroblasts with either TGF-�r ESCC conditioned media leads to enhanced VEGF release in 3D culture. Monocultures and cocultures of esophageal fibroblasts were grown in3D collagen in the absence or presence of SB-505124 for 48 hours. (C) The VEGF inhibitor GW654652 (1 �mol/L) inhibits vascular network

ormation. (D) The bar graph shows mean data for vascular network formation in the absence and presence of GW654652. All representative images

re shown as 3-color merges; monochrome images of CD31 are inset. Scale bar � 200 �m.

fcEnteclwabeawfiwvambbib

nLcbTs

Twtwstp

wtToaetftiti

fphuwIib

FibEtTttdf

BA

SIC–

ALI

MEN

TARY

TRA

CT

June 2008 FIBROBLASTS DRIVE ESCC–INDUCED ANGIOGENESIS 1991

rom that in previous studies because it contains cancerells, fibroblasts, and endothelial cells. Coculture ofSCC cells with HMVECs did not stimulate vascularetwork formation in the absence of fibroblasts. Whereashe coculture of fibroblasts with HMVECs induced somendothelial cell migration into the collagen, it was asso-iated with large, poorly defined vascular networks withittle organized branching. Truly organized vascular net-ork formation was only seen when ESCC cells weredded to the fibroblast/HMVEC cultures. This suggestedoth that the fibroblasts were absolutely required forndothelial tube formation and that a prior fibroblastctivation step, via factor(s) released from the ESCC cells,as required for efficient network formation. Activatedbroblasts are often observed in the stroma associatedith growing tumors.15,16 In agreement with this obser-

ation, we noted that coculture of ESCC cells with esoph-geal fibroblasts led to their transdifferentiation into theyofibroblast phenotype. Myofibroblasts are known to

e a good model for CAFs because the activated fibro-lasts within the tumor milieu exhibit a phenotype that

s virtually indistinguishable from that of myofibro-lasts.17

One of the best known factors responsible for the phe-otypic switch of fibroblast to myofibroblast is TGF-�.18

oss of TGF-� receptor function is a common event inancer,18 and ESCC cells are no exception. They typicallyecome insensitive to the growth inhibitory effects ofGF-� through either the loss of TGF-� receptor expres-

igure 7. Schematic illustration show-ng the role of ESCC cells and fibro-lasts in vascular network formation.sophageal cancer cells produce TGF-�o activate stromal normal fibroblasts.umor stromal fibroblasts become

ransdifferentiated into myofibroblastshat secrete VEGF, which in turn in-uces endothelial cell migration and the

ormation of a microcapillary network.

ion or the acquisition of missense mutations in r

GF�RII.19 –21 In agreement with these published studies,e found that our panel of ESCC lines was refractory to

he growth inhibitory effects of exogenous TGF-�,hereas primary human esophageal keratinocytes were

trongly growth inhibited. Loss of TGF�RII expression inhe ESCC lines was also accompanied by the autocrine/aracrine production of both TGF-�1 and TGF-�2.Both ESCC conditioned media and exogenous TGF-�

ere similarly able to induce the myofibroblast pheno-ype and stimulate SMAD signaling, suggesting thatGF-� was the factor likely for the fibroblast activationbserved. The requirement for TGF-�–induced fibroblastctivation in neoangiogenesis was shown by the ability ofxogenous TGF-� to enhance vascular network forma-ion in the absence of the ESCC cells. Further evidenceor the role of TGF-� in ESCC-induced fibroblast activa-ion came from studies showing that a pharmacologicnhibitor of TGF-� blocked the ability of ESCC-condi-ioned media induction of the myofibroblast phenotypen esophageal fibroblasts.

Having shown that ESCC-secreted TGF-� is essentialor fibroblast-induced activation, we next tested whetherharmacologic inhibition of TGF-� signaling could in-ibit vascular network formation. For these studies wesed SB-505124, a selective TGF-� receptor antagonistith very little activity against any other kinase tested.22

ncreasing concentrations of SB-505124 were found tonhibit myofibroblast transdifferentiation induced bothy TGF-� and ESCC conditioned media. The essential

ole of TGF-�–induced fibroblast activation in vascular

n5Ggfiftd

vVsapgammstcVntwfiabttbd

euggdcBaaTtht

aG1

1

1

1

1

1

1

1

1

1

1

2

2

BA

SIC–

ALIM

ENTA

RY

TRA

CT

1992 NOMA ET AL GASTROENTEROLOGY Vol. 134, No. 7

etwork formation was indicated by the fact that SB-05124 and the structurally unrelated TGF-� inhibitorW788388 significantly inhibited ESCC-induced neoan-

iogenesis. The requirement for both ESCC cells andbroblasts in the angiogenic process was shown by the

act that neither of the TGF-� inhibitors tested blockedhe limited amount of vascular network formation in-uced by fibroblasts alone.ESCC-induced fibroblast activation was shown to drive

ascular network formation through the stimulation ofEGF release. VEGF is a potent proangiogenic factor thattimulates endothelial cell migration and proliferationnd regulates microvascular permeability.23 There is com-elling evidence that fibroblasts can be a major source ofrowth factors that may contribute to tumor progressionnd possibly angiogenesis.9,17,24 Studies using transgenicice expressing GFP under the control of VEGF pro-oter showed strong GFP staining in the stroma of

pontaneously arising mammary tumors.10 Here we showhat fibroblasts activated by exogenous TGF-� throughoculture with ESCC cells secrete significantly higherEGF levels through a mechanism involving SMAD sig-aling. The fact that a selective inhibitor of VEGF recep-or signaling blocks efficient ESCC-induced vascular net-ork formation provides the link between ESCC-inducedbroblast activation and angiogenesis in this model. Ingreement with this, others have also shown that fibro-lasts can release VEGF in response to hypoxic condi-ions25,26 and exogenous TGF-�.27,28 It is unclear whetherhe ESCC cells are releasing latent VEGF from the fibro-lasts, as has been suggested by others,29 or are insteadriving de novo protein expression.Here, we have shown that the stromal fibroblasts are

ssential for the angiogenic response in ESCC, openingp an intriguing new possibility for therapy. There is arowing body of evidence suggesting that therapies tar-eted against angiogenesis can lead to dramatic responses iniseases such as colorectal carcinoma, particularly whenombined with established chemotherapeutic regimens.30

ecause ESCC cells are typically detected at relativelydvanced stages, optimized antiangiogenic/chemother-py combinations are ideal novel therapeutic candidates.he postulated role of CAFs in tumor invasion also makes

argeting the stromal fibroblast activation through the in-ibition of TGF-� signaling an appealing therapeutic op-ion.

Supplementary Data

Note: To access the supplementary materialccompanying this article, visit the online version ofastroenterology at www.gastrojournal.org, and at doi:0.1053/j.gastro.2008.02.061.

References

1. Bissell MJ, Radisky D. Putting tumours in context. Nat Rev Cancer

2001;1:46–54.

2. Dvorak HF. Tumors: wounds that do not heal. Similarities be-tween tumor stroma generation and wound healing. N Engl J Med1986;315:1650–1659.

3. Elenbaas B, Weinberg RA. Heterotypic signaling between epithe-lial tumor cells and fibroblasts in carcinoma formation. Exp CellRes 2001;264:169–184.

4. Nakayama H, Enzan H, Miyazaki E, et al. The role of myofibro-blasts at the tumor border of invasive colorectal adenocarcino-mas. Jpn J Clin Oncol 1998;28:615–620.

5. Tsukamoto H, Mishima Y, Hayashibe K, et al. Alpha-smoothmuscle actin expression in tumor and stromal cells of benign andmalignant human pigment cell tumors. J Invest Dermatol 1992;98:116–120.

6. De Wever O, Mareel M. Role of tissue stroma in cancer cellinvasion. J Pathol 2003;200:429–447.

7. Direkze NC, Hodivala-Dilke K, Jeffery R, et al. Bone marrow con-tribution to tumor-associated myofibroblasts and fibroblasts. Can-cer Res 2004;64:8492–8495.

8. Sappino AP, Skalli O, Jackson B, et al. Smooth-muscle differen-tiation in stromal cells of malignant and non-malignant breasttissues. Int J Cancer 1988;41:707–712.

9. De Wever O, Nguyen QD, Van Hoorde L, et al. Tenascin-C andSF/HGF produced by myofibroblasts in vitro provide convergentproinvasive signals to human colon cancer cells through RhoAand Rac. FASEB J 2004;18:1016–1018.

0. Fukumura D, Xavier R, Sugiura T, et al. Tumor induction of VEGFpromoter activity in stromal cells. Cell 1998;94:715–725.

1. Lioni M, Brafford P, Andl C, et al. Dysregulation of claudin-7 leadsto loss of E-cadherin expression and the increased invasion ofesophageal squamous cell carcinoma cells. Am J Pathol 2007;170:709–721.

2. Andl CD, Mizushima T, Nakagawa H, et al. Epidermal growthfactor receptor mediates increased cell proliferation, migration,and aggregation in esophageal keratinocytes in vitro and in vivo.J Biol Chem 2003;278:1824–1830.

3. Velazquez OC, Snyder R, Liu ZJ, et al. Fibroblast-dependent dif-ferentiation of human microvascular endothelial cells into capil-lary-like, three-dimensional networks. FASEB J 2002;16:1316–1318.

4. Smalley KS, Contractor R, Haass NK, et al. Ki67 expressionlevels are a better marker of reduced melanoma growth followingMEK inhibitor treatment than phospho-ERK levels. Br J Cancer2007;96:445–449.

5. Orimo A, Weinberg RA. Stromal fibroblasts in cancer: a noveltumor-promoting cell type. Cell Cycle 2006;5:1597–1601.

6. Kalluri R, Zeisberg M. Fibroblasts in cancer. Nat Rev Cancer2006;6:392–401.

7. Orimo A, Gupta PB, Sgroi DC, et al. Stromal fibroblasts present ininvasive human breast carcinomas promote tumor growth andangiogenesis through elevated SDF-1/CXCL12 secretion. Cell2005;121:335–348.

8. Levy L, Hill CS. Alterations in components of the TGF-beta super-family signaling pathways in human cancer. Cytokine GrowthFactor Rev 2006;17:41–58.

9. Tanaka S, Mori M, Mafune K, et al. A dominant negative mutationof transforming growth factor-beta receptor type II gene in micro-satellite stable oesophageal carcinoma. Br J Cancer 2000;82:1557–1560.

0. Garrigue-Antar L, Souza RF, Vellucci VF, et al. Loss of transform-ing growth factor-beta type II receptor gene expression in primaryhuman esophageal cancer. Lab Invest 1996;75:263–272.

1. Fukai Y, Fukuchi M, Masuda N, et al. Reduced expression oftransforming growth factor-beta receptors is an unfavorable prog-nostic factor in human esophageal squamous cell carcinoma. Int

J Cancer 2003;104:161–166.

2

2

2

2

2

2

2

2

3

Sd@

M

June 2008 FIBROBLASTS DRIVE ESCC–INDUCED ANGIOGENESIS 1993

2. DaCosta Byfield S, Major C, Laping NJ, et al. SB-505124 is aselective inhibitor of transforming growth factor-beta type I recep-tors ALK4, ALK5, and ALK7. Mol Pharmacol 2004;65:744–752.

3. Dvorak HF, Brown LF, Detmar M, et al. Vascular permeabilityfactor/vascular endothelial growth factor, microvascular hyper-permeability, and angiogenesis. Am J Pathol 1995;146:1029–1039.

4. Dong J, Grunstein J, Tejada M, et al. VEGF-null cells requirePDGFR alpha signaling-mediated stromal fibroblast recruitmentfor tumorigenesis. EMBO J 2004;23:2800–2810.

5. Hlatky L, Tsionou C, Hahnfeldt P, et al. Mammary fibroblasts mayinfluence breast tumor angiogenesis via hypoxia-induced vascu-lar endothelial growth factor up-regulation and protein expres-sion. Cancer Res 1994;54:6083–6086.

6. Nakagawa T, Lan HY, Zhu HJ, et al. Differential regulation of VEGFby TGF-beta and hypoxia in rat proximal tubular cells. Am J PhysiolRenal Physiol 2004;287:F658–F664.

7. Kobayashi T, Liu X, Wen FQ, et al. Smad3 mediates TGF-beta1induction of VEGF production in lung fibroblasts. Biochem Bio-

phys Res Commun 2005;327:393–398.

8. Renner U, Lohrer P, Schaaf L, et al. Transforming growth factor-beta stimulates vascular endothelial growth factor production byfolliculostellate pituitary cells. Endocrinology 2002;143:3759–3765.

9. Ito TK, Ishii G, Chiba H, et al. The VEGF angiogenic switch offibroblasts is regulated by MMP-7 from cancer cells. Oncogene2007;26:7194–7203.

0. Jain RK. Normalization of tumor vasculature: an emerging con-cept in antiangiogenic therapy. Science 2005;307:58–62.

Received January 17, 2008. Accepted February 19, 2008.Address requests for reprints to: Meenhard Herlyn, DVM, or Keiran

. M. Smalley, PhD, The Wistar Institute, 3601 Spruce Street, Phila-elphia, Pennsylvania 19104. e-mail: herlynm@wistar.org or ksmalleywistar.orgSupported by National Cancer Institute grant P01-CA098101 (to.H.).

The authors declare no financial conflicts of interest.

BA

SIC–

ALI

MEN

TARY

TRA

CT