Hedgehog signaling is restricted to the stromalcompartment during pancreatic carcinogenesisHua Tiana, Christopher A. Callahanb, Kelly J. DuPreec, Walter C. Darbonnec, Christina P. Ahna, Suzie J. Scalesa,and Frederic J. de Sauvagea,1

Departments of aMolecular Biology, bPathology, and cPD Biomarkers, Genentech, Incorporated, 1 DNA Way, South San Francisco, CA 94080

Communicated by Napoleone Ferrara, Genentech, Inc., South San Francisco, CA, January 13, 2009 (received for review October 17, 2008)

The Hedgehog (Hh) pathway has been implicated in pancreatic cancerbut its role remains controversial. To delineate the cell populationsable to respond to Hh ligand stimulation, we expressed an oncogenicallele of Smoothened (SmoM2) to cell autonomously activate Hhsignaling in the mouse pancreas. Surprisingly, we found that expres-sion of SmoM2 in epithelial cells was not able to activate the pathwayand had no impact on pancreatic development or neoplasia. Incontrast, activation of Smo in the mesenchyme led to Hh pathwayactivation, indicating that only the tumor stroma is competent totransduce the Hh signal. Using a Ptc-LacZ reporter mouse, we showthat Hh signaling is active in stromal cells surrounding Hh-expressingtumor epithelium in various mouse pancreatic cancer models. Acti-vation of the Hh pathway in the tumor stroma of human pancreaticand metastatic cancer specimens was confirmed by quantitativeRT-PCR of microdissected tissue samples. These data support a para-crine model of Hh-mediated tumorigenesis, in which tumor cellssecrete Hh ligand to induce tumor-promoting Hh target genes inadjacent stroma.

pancreatic cancer � paracrine � tumor stroma � SmoM2 � Kras

Pancreatic ductal adenocarcinoma (PDA) is one of the most aggres-sive forms of cancer in the world, with a 5-year survival rate of less

than 5%. Potential precursors of PDA include exocrine neoplasticchanges, such as pancreatic intraepithelial neoplasms (PanINs), whichdemonstrate more severe epithelial atypia as they progress towardmalignancy. Genetic analyses have linked mutations in human KRASto PDA (1), and the functional role of oncogenic KRAS in both PDAinitiation and progression were subsequently confirmed using geneti-cally engineered animal models of pancreatic cancer (2–9). In additionto KRAS, a number of signaling pathways, including Wnt, TGF�,TGF�, Notch, EGF, and most recently, Hedgehog (Hh) have beenimplicated in PDA (10–15).

Paracrine Hh signaling plays a critical role during the develop-ment and maintenance of many epithelial appendages, includingderivatives of the gut (16). This involves Hh ligand being secretedfrom the epithelium and binding to the Hh receptor Patched1(Ptch1) in the adjacent mesenchymal receiving cells, which leads torelocalization of Smoothened (Smo) to primary cilia and activationof the Gli family of transcription factors (17). While there is littleevidence for Hh pathway activity within the normal adult humanexocrine pancreas (13), the expression of Hh ligands and essentialcomponents of the pathway, including PTCH and SMO, have beenreported in up to 70% of human PDA specimens (13). In addition,several studies have proposed that Hh ligands produced by tumorcells activate Hh signaling and tumor cell growth in an autocrine/juxtacrine manner (13–14). Recently, using xenograft tumor mod-els, our laboratory demonstrated a paracrine paradigm for Hh-mediated tumorigenesis, where Hh ligands produced by tumor cellsact on the stromal compartment to support tumor growth indirectly(18). These data are consistent with the absence of mutations incanonical Hh pathway components that are commonly found incell-autonomous, Hh-pathway-driven, tumor types, including me-dulloblastoma, basal cell carcinoma, and rhabdomyosarcoma (17).

In this study, we set out to further define a potential stromalversus epithelial role for Hh signaling in autochthonous pancreaticmouse and human cancers. Although several mouse models have

been established previously to investigate a causative role of Hhsignaling in pancreatic tumorigenesis, none of these distinguishedparacrine versus autocrine canonical Hh signaling in PDA. Amongthese models, ectopic expression of Sonic hedgehog (Shh) in thepancreas (Pdx-1-Shh) led to intestinal metaplasia with cellularatypia (13, 19) and a mouse with pan-epithelial expression of adominant active Gli2 allele (CLEG2) developed invasive, undiffer-entiated, pancreatic tumors (14). For the purposes of identifying thecell populations within the pancreas that are activated by Hhligands, we chose to express in the pancreas an oncogenic allele ofSmo, SmoM2, because this mutation activates the canonical Hhsignaling pathway in a cell-autonomous manner.

Using this model in combination with a Ptc-lacZ reporter allele,we show here that the pancreatic epithelium is unable to transducethe canonical Hh signal. In contrast, we demonstrate that Hhsignaling is restricted principally to smooth muscle actin (SMA)-expressing spindle cells adjacent to Hh-producing tumor epithe-lium. Quantitative RT-PCR on laser capture microdissected(LCM) tumor samples from both mice and humans confirm Hhsignaling in the tumor stroma. Taken together, these results dem-onstrate that the pancreatic epithelium is not receptive of tumorcell-derived Hh ligands, but instead, Hh ligands promote PDA viaa paracrine signaling mechanism received by tumor stromal cells.

ResultsEpithelial Expression of SmoM2 Is not Sufficient to Induce NeoplasticTransformation in Mouse Pancreatic Epithelium. The activating mu-tation W535L in human SMO (SMOM2) was initially identified inhuman basal cell carcinoma, and its ectopic expression causes consti-tutive, ligand-independent activation of the Hh pathway (20). Toinvestigate whether Hh signaling in pancreatic epithelium is sufficient toinduce pancreatic cancer, we used a previously described mouse model(21), R26-SmoM2, in which a cDNA fragment encoding the SmoM2-YFPfusionproteinwas targeted into theubiquitouslyexpressed Rosa26locus (R26) behind a stop cassette flanked by Loxp sites. YFP fused atthe C terminus of SmoM2 does not interfere with its activity in neuralepithelium and provides a tool to trace the lineage of all Cre-expressingcells and their progeny (22).

We crossed R26-SmoM2 with PdxCre transgenic mice (2) toactivate Hh signaling within the pancreatic epithelium (Fig. 1A).PdxCre;SmoM2 mice were born at Mendelian frequencies andshowed no gross abnormalities up to 18 months of age. Microscopicexamination of 12-week-old mice showed normal lobular pancreaticarchitecture with expression of both exocrine (cytokeratin 19 orCK19) and endocrine (insulin) markers (Fig. S1). Significantly,none of the PdxCre;SmoM2 mice developed pancreatic neoplasms,

Author contributions: H.T., C.A.C., and F.J.d.S. designed research; H.T., K.J.D., W.C.D., andC.P.A. performed research; S.J.S. contributed new reagents/analytic tools; H.T., C.A.C.,K.J.D., and F.J.d.S. analyzed data; and H.T., C.A.C., and F.J.d.S. wrote the paper.

Conflict of interest statement: The Sponsor declares a conflict of interest (such as definedby PNAS policy). Both the Sponsor and the authors are employees of Genentech, Inc. Theauthors declare no conflict of interest (such as defined by PNAS policy).

Freely available online through the PNAS open access option.

1To whom correspondence should be addressed. E-mail: sauvage@gene.com.

This article contains supporting information online at www.pnas.org/cgi/content/full/0813203106/DCSupplemental.

4254–4259 � PNAS � March 17, 2009 � vol. 106 � no. 11 www.pnas.org�cgi�doi�10.1073�pnas.0813203106

indicating that ectopic expression of the SmoM2 oncogene in thepancreatic epithelium is not sufficient to initiate tumorigenesis.

To determine if SmoM2 expression in the pancreas indeed leadsto activation of the Hh signaling pathway, we crossed PdxCre;S-moM2 mice with Ptc-LacZ reporter mice, in which �-galactosidaseis expressed under the control of a Ptch1 promoter element (23, 24).This reporter line provides a faithful and sensitive measure of

endogenous canonical Hh pathway target gene Ptch1 expressionduring both development and tumorigenesis (23, 25, 26). Interest-ingly, while the SmoM2 transgene was expressed by the pancreaticepithelium of ducts (marked by CK19 in Fig. 1 B and F), acini, andislets, no exocrine epithelial Hh signaling was identified in eithercontrol (Ptc-LacZ; Fig. 1C, and PdxCre;Ptc-LacZ; data not shown)or PdxCre;SmoM2;Ptc-lacZ mice (Fig. 1E). The lack of Hh pathway

aciniacini

ductduct

STOPRosa26 SmoM2-YFPR26-SmoM2:

Pdx-Cre

SmoM2-YFPSmoM2-YFP:Rosa26

B

E

isletacini

aciniacini

A

C D

F

duct duct

duct

0

2

4

6

8

10

12

14

GFP SmoM20

10

20

30

40

50

60

DMSO Shh Hh agonist

Gli1 mRNA expGlires

ductfibroblast

Gli1

mR

NA

expr

essi

on

Gli1

mR

NA

expr

essi

on

ductfibroblast

G H

Fig. 1. Pancreatic epithelium is not competent to transduce Hh signal downstream of SmoM2. (A) Schematic of the R26-SmoM2 targeted allele and PdxCre transgene.PdxCremediatesLoxprecombination, removingthestopcassettetoresult inexpressionofSmoM2-YFPwithinthepancreaticepithelium.FusionofYFPtotheCterminusof SmoM2 allows immunofluorescence detection of the transgene and thus Cre activity. (B) Stochastic expression of SmoM2-YFP (red) within the pancreatic epithelium,including ducts (CK19, green), acini, and islet. (C) �-galactosidase immunoreactivity (red) in periductal mesenchyme of Ptc-LacZ transgenic animal (ductal cells expressCK19, green). (D) �-galactosidase immunoreactivity (red) in periductal mesenchyme (vimentin, green) of Ptc-LacZ mice (adjacent, serial section to C). (E) �-galactosidaseimmunoreactivity (red) in periductal mesenchyme of PdxCre;SmoM2;PtcLacZ animal (ductal cells express CK19, green). (F) SmoM2-YFP (red) is expressed by ductal cells(CK19,green)andacinarepithelium(adjacent, serial sectiontoE) .NotethatSmoM2accumulates intheprimarycilia (arrow)of interlobularducts.NoSmoM2isdetectedin the periductal mesenchyme. These images are representative of analysis from 3 animals per genotype. (Scale bars, 50 �m.) (G) Both Shh (rShh, 200 ng/ml) and Hhagonist (1 �g/ml) fail to induce the Hh target gene Gli1 in PDECs, but strongly activate the pathway in pancreatic fibroblasts. (H) SmoM2 over-expression up-regulatesGli1 mRNA levels in pancreatic fibroblasts, but not in PDEC.

Tian et al. PNAS � March 17, 2009 � vol. 106 � no. 11 � 4255

CELL

BIO

LOG

Y

activation is not because of the low expression level of SmoM2transgene, as we can readily detect accumulation of SmoM2-YFPfusion protein in both the exocrine (acini and ducts) and endocrine(islets) portions of the pancreas by immunofluoresence (see Fig.1B). Interestingly, SmoM2 was also detected in primary cilia of bothduct (see Fig. 1F) and islet cells.

In contrast, �-galactosidase expression was seen in mesenchymalcells, marked by vimentin staining, adjacent to pancreatic ducts inboth control and SmoM2 transgenic mice (Fig. 1D). Because thesecells did not express SmoM2-YFP (see Fig. 1F), the �-galactosidaseactivity likely reflects the reception of Hh signals from ligandsexpressed by the ductal epithelium that are not detectable byimmunohistochemistry (IHC) (13). This mesenchymal Hh signalsurrounding the ducts was not affected by SmoM2 oncogeneexpression in the pancreatic epithelium, as comparable levels of�-galactosidase staining were observed in both PdxCre;Ptc-LacZand PdxCre;SmoM2;Ptc-LacZ mice.

Pancreatic Ductal Epithelial Cells Are Incompetent to Transduce HhSignaling in Vitro. To further validate the inability of ductal epithe-lial cells to transduce the Hh signal, we treated a pancreatic ductalepithelial cell (PDEC) line, established from PdxCre;KrasG12D mice,with recombinant Shh (rShh) or with a small molecule agonist ofSmo (27). Similar to the observations in vivo, Gli1 mRNA levelswere not up-regulated in PDECs in response to ligand stimulation.In contrast, rShh or agonist treatment led to a 10- to 50-foldGli1mRNA increase in primary pancreatic fibroblasts (Fig. 1G).This was not a result of the lack of native Smo expression in ductalcells because we detected similar levels of Smo mRNA in bothductal and fibroblast cells by RT-PCR (data not shown). Consistentwith these results, electroporation of SMOM2-GFP in both fibro-blasts and PDECs led to Gli1 up-regulation only in fibroblasts (Fig.1H). Therefore, failure to increase Hh signaling is most likelyattributable to a block downstream of Smo in PDECs.

SmoM2 Does not Potentiate KrasG12D-Driven PDA. While expressionof SmoM2 alone in pancreatic epithelium was unable to activate theHh pathway and did not initiate tumorigenesis, we asked whetherSmoM2 could stimulate Hh signaling in pancreatic tumor epithelialcells and potentially accelerate the course of PDA tumor progres-sion driven by oncogenic KrasG12D. The expression of oncogenicKrasG12D in murine pancreatic epithelial cells (PdxCre;KrasG12D)drives the development of PanINs as early as 4 weeks of age, butmost lesions remain low grade and progression to malignant PDAoccurs slowly during the following year (2). PdxCre;KrasG12D;SmoM2 mice appeared normal at birth and did not display highergrade PanINs when compared with PdxCre;KrasG12D littermates.However, PdxCre;KrasG12D;SmoM2 mice developed gastrointesti-nal obstructions and had to be killed by 24 weeks of age. Thesegastrointestinal obstructions were secondary to mesenchymal tu-mors, arising within both the pancreas and stomach (with 100% and80% penetrance, respectively, n � 30), reflecting previously doc-umented extra-pancreatic PdxCre transgene activity (2, 9). Thesemesenchymal tumors stained positively for SMA (Fig. 2A), vimen-tin (Fig. 2B), and the SmoM2 transgene (Fig. 2I). They did notexpress CK19 (a cytokeratin expressed in the ductal cell lineage),

*

*

*

MT

MT

duct

islet

acini

vessel

MT

vessel vessel

A B

C D

E F

G H

I J

K L

Fig. 2. SmoM2 and KrasG12D cooperate to induce pancreatic mesenchymaltumors in PdxCre;KrasG12D;SmoM2 mice. (A) SMA expression by mesenchymaltumor in a PdxCre;KrasG12D;SmoM2 mouse. (B) Vimentin expression by mesen-chymal tumor in a PdxCre;KrasG12D;SmoM2 mouse. (C) Muscular blood vessel inthe pancreas of a control (PdxCre;KrasG12D) mouse at 4 weeks. (D) Small, mesen-chymal tumor arising from a muscular blood vessel in the pancreas of a 4-week-old PdxCre;KrasG12D;SmoM2 mouse. Arrow highlights tumor extension into sur-rounding soft tissue. (E) Arteries in control PdxCre;KrasG12D animal at 8 weeks. (F)Artery and mesenchymal tumor in 8-week-old PdxCre;KrasG12D;SmoM2 mouse(MT* highlights tumor). (G) X-gal staining highlights the stochastic activity ofPdxCre recombinase in pancreatic epithelium (duct, acini, and islet) when com-bined with the R26-LSL-LacZ reporter. (H) PdxCre activity is occasionally detectedin adventitia surrounding blood vessels in PdxCre;R26R mice. (I) SmoM2-YFP

expression (anti-YFP, red) in mesenchymal tumor (MT) arising from a muscularblood vessel (CD31, green) in a 12-week-old PdxCre;KrasG12D;SmoM2 mouse.(J) Hh signaling is active in mesenchymal tumor, as indicated by nuclear�-galactosidase enzymatic activity (blue) around blood vessel (CD31, brown).(K) SmoM2-YFP (red) expression in a higher-grade mPanIN epithelium(marked by CK19, green). (L) Hh signaling is not active in CK19-positive (green)neoplastic epithelium, as demonstrated by a lack of �-galactosidase immuno-reactivity (red). �-galactosidase expression (red) in only observed in stromalcells adjacent to a higher-grade mPanIN (CK19, green; adjacent serial sectionto K). These images are representative of analysis of three animals per geno-type. (Scale bars, 50 �m.)

4256 � www.pnas.org�cgi�doi�10.1073�pnas.0813203106 Tian et al.

E-cadherin, S100, myogenin, BCL-2, or Hh ligands (data notshown). This immunophenotype is indicative of a mesenchymaltumor demonstrating smooth muscle differentiation, as opposed topancreatic adenocarcinomas undergoing epithelial to mesenchymaltransition. Consistent with this phenotype, pancreatic mesenchymaltumors first appeared within muscular blood vessels in 4-week-oldPdxCre;KrasG12D;SmoM2 animals (Fig. 2 C–F), and these changespreceded metaplastic changes of the pancreatic epithelium andPDA precursor lesions, such as PanINs. To confirm that cells withinthe wall of blood vessels expressed PdxCre before tumor formation,we crossed PdxCre mice with R26-LSL-LacZ mice to lineage-tracethe PdxCre-positive cells. Consistent with previous reports, PdxCretransgene activity was detected in pancreatic (Fig. 2G), gastric, andduodenum epithelium (data not shown) (2, 4). In addition, �-ga-lactosidase activity was also detected in rare, scattered, spindle-shaped cells within the adventitia of blood vessels, demonstratingstochastic activity in this cell compartment (Fig. 2H). In agreementwith these results, adventitial cells have been shown to respond toHh signals (28).

Because oncogenic Smo expression within pancreatic epitheliumwas unable to potentiate KrasG12D-mediated neoplasia, we testedwhether Hh signaling was active in KrasG12D -driven tumor epithe-lium by crossing PdxCre;KrasG12D;SmoM2 mice with the Ptc-LacZreporter line. Active Hh signaling, as revealed by nuclear �-galac-tosidase staining, was not identified in the epithelium ofPdxCre;KrasG12D;SmoM2;Ptc-lacZ mice (Fig. 2 K and L). In con-trast, we found robust expression of nuclear �-galactosidase activityin mesenchymal tumors expressing the transgene (Fig. 2 I and J),indicating that the SmoM2 transgene was functional and able toactivate the Hh pathway in these cells. The fact that Ptc-LacZexpression was absent from neoplastic epithelium (driven byKrasG12D), despite high SmoM2 expression levels, indicates that, incontrast to the mesenchyme, neoplastic pancreatic epithelium is notcompetent to transduce the Hh signal downstream of Smo.

Hh Activity Is Restricted to the Stromal Compartment in AdditionalKrasG12D-Driven Mouse Models of PDA. While PdxCre;KrasG12D ani-mals develop early stage PanIN at a young age (2–4 weeks), highergrade PanINs do not arise until 7 to 10 months of age (2). However,when combined with the tumor-suppressor gene Ink4a/Arf deletionor with an oncogenic mutation in p53R270H, PanIN progresses toPDA within 7 to 20 weeks (4–6). To test whether Hh signaling isrestricted to the stroma in more advanced stages of PDA, weintroduced the Ptc-lacZ reporter in these models. Similar toPdxCre;KrasG12D;SmoM2;Ptc-LacZ mice, Hh signaling was not de-tected in epithelial cells of PanINs in PdxCre;KrasG12D;Ptc-LacZ(Fig. 3A) or PDA of PdxCre;KrasG12D;Ink4a/Arfflox/flox;Ptc-lacZ mice(Fig. 3D). Instead, Ptc-LacZ staining was restricted to spindle-shaped, stromal cells that were in close proximity to CK19-

positive, Hh-expressing tumor epithelium (Fig. 3 A, B, D, and E),further corroborating a model in which ligand-dependent paracrineHh signaling occurs between the tumor and its stroma.Interestingly, similar to the mesenchymal tumor cells ofPdxCre;KrasG12D;SmoM2 mice, the responsive tumor stromalcells in PdxCre;KrasG12D;Ptc-LacZ and PdxCre;KrasG12D;Ink4a/Arfflox/flox;Ptc-lacZ mice also expressed SMA (Fig. 3 C and F) andvimentin (data not shown). The shape, location, and expressionprofiles of these Hh-responsive stromal cells are most consistentwith myofibroblasts (activated fibroblasts).

Hh Target-Gene Expression Within the Stromal Compartment of BothMouse and Human PDAs. To confirm the results obtained in ge-netic models of PDA using the Ptc-LacZ reporter allele, we laser-capture microdissected PDA tumors from PdxCre;KrasG12D,PdxCre;KrasG12D;p53R270H, and PdxCre;KrasG12D;SmoM2 mice. Follow-ing the separation of tumor epithelium from the surrounding stroma,we analyzed Gli1 levels by quantitative RT-PCR (qRT-PCR). In all 3PDA models examined, stromal Gli1 levels were between 13- and150-fold higher than that expressed by PDA tumor cells (Fig. 4A). Totest whether these findings in mouse models could be extended tohuman PDA, we examined GLI1 transcripts in human primary PDAs.Similar to mouse PDAs, we found significantly higher (between 40- and120-fold)GLI1mRNAlevels in tumorstromathan in tumorepithelium(Fig. 4B). In contrast, Hh ligand expression (SHH and IHH) wasenriched in the tumor epithelium (see Fig. 4 A and B).

Hh Target-Gene Expression Within the Stromal Compartment of Met-astatic Human Carcinomas. Using colon xenograft models, we haverecently shown that stromal Hh signaling may also potentiate colorectalcarcinogenesis (18). To examine whether Hh signaling was up-regulatedin the stromal compartment of ligand-expressing metastatic humancarcinomas, we examined human liver metastases for GLI1, SHH, andIHH transcript levels by LCM and RT-PCR. As observed in primaryhuman and mouse PDAs, the stroma of metastatic colorectal carcino-mas demonstrated elevated GLI1 levels, while Hh ligand expressionwas greater in metastatic tumor cells (Fig. 4C).

DiscussionIn contrast to early reports of Hh pathway activity in pancreatictumor cells (13), we have recently shown in xenograft models aparacrine requirement for the Hh pathway, where Hh ligand isproduced by the tumor cells and the pathway is activated in tumorstroma (18). To address whether a paracrine Hh signal is present inautochthonous mouse pancreatic tumors, and to test if tumorepithelium is competent to transduce the Hh signal, we used anoncogenic form of Smo to activate the pathway cell autonomously.This approach was successfully used to study cell-autonomous Hhpathway activation in adult neural stem cells, cerebellum granulecell precursors, and neural crest progenitor differentiation (21, 22,

βgal(Ptc-LacZ)/CK19 ShhPdxCre;KrasG12D;

Ptc-LacZ;Ink4a/Arflox/lox

PdxCre;KrasG

12D;

Ptc-LacZ

A B

D

βgal(Ptc-LacZ)/SMAC

E F

Fig. 3. Hh pathway activity is restricted to the stromalcompartment. (A) �-galactosidase expression (red) is re-stricted to stromal cells surrounding a low-grade mPa-nIN lesion in PdxCre;KrasG12D;Ptc-LacZ mice. (B) Hh ex-pression (brown) in the low-grade mPanIN lesion(adjacent serial section to A). (C) �-galactosidase expres-sion (red, arrows) within SMA-expressing stromal cells(green cytoplasmic staining; adjacent serial section to B).(D) �-galactosidase expression (red) is restricted to tu-mor stromal cells within PDA of a PdxCre;KrasG12D;Ink4a/Arfflox/[supi]flox;Ptc-LacZ mouse. (E) Hh expression (brown)in PDA tumor epithelium (adjacent serial section to D).(F) �-galactosidase expression (red, arrows) within SMA-expressing tumor stromal cells (green cytoplasmic stain-ing; adjacent serial section to E). These images are rep-resentative of analysis of two animals per genotype.(Scale bars, 50 �m.)

Tian et al. PNAS � March 17, 2009 � vol. 106 � no. 11 � 4257

CELL

BIO

LOG

Y

29). Expression of SmoM2 alone within pancreatic epithelium didnot lead to significant pathology by 18 months of age. In addition,SmoM2 in combination with KrasG12D does not accelerateKrasG12D-driven pancreatic adenocarcinoma progression. The find-ing that SmoM2 could not activate Hh signaling in normal, meta-plastic, premalignant, or malignant exocrine pancreatic epithelia invivo, and in pancreatic ductal epithelial cells in vitro, indicates thatepithelia of the exocrine pancreas are not competent to respond toHh signals. Interestingly, correct localization of SmoM2 was de-tected in primary cilia of both ductal and islet epithelium, which isrequired for pathway activation (30). These results indicate afunctional block of signaling downstream of Smo within the pan-creatic epithelium in vivo.

The expression of Hh ligands by the tumor epithelium and therestriction of Ptc-LacZ in several mouse models of PDA to sur-

rounding SMA-positive stromal cells support a paracrine mecha-nism of Hh activation. Our finding that Hh signaling does notoccur in tumor epithelium is also consistent with recent resultsshowing that genetic ablation of Smo in the pancreatic epitheliumof p48-Cre/�;LSL-KrasG12D;Trp-53F/� mice does not affect PDAtumor progression (31). The mechanism by which Hh pathwayactivation in stromal cells supports tumor growth remains to beestablished. The spindle-shape and SMA expression of Ptc-lacZ-expressing cells within mouse PDA suggests that many Hh-receptive cells in the tumor stroma are myofibroblasts. Consistentwith this, cultured primary mouse pancreatic fibroblasts up-regulate the expression of secreted growth factors, such as Igf andPdgf, following Hh stimulation (data not shown). These factors aresimilar to those regulated in subcutaneous xenograft models ofPDA upon treatment with Hh pathway inhibitors (18).

The expression profiles of both the stromal and epithelial com-partments of human pancreatic and metastatic colorectal tumorsreported here support a model where activation of the Hh pathwayis restricted to the stroma. The absence of mutations in Hh pathwaycomponents, such as PTCH, SMO, or SUFU in PDA is alsoconsistent with the absence of canonical Hh pathway activation intumor epithelium. Our findings that Hh signaling is restricted to thetumor stroma contradicts previous reports supporting a role forligand-driven epithelial Hh signaling in tumor cell growth. Whilethese conclusions were mostly derived from studies carried out invitro, using tumor cell lines and cyclopamine at concentrations invast excess to what is needed to inhibit Hh signaling (18), a recentmouse model has argued in favor of a role for Hh signaling in tumorepithelium. By expressing a dominant active Gli2 (CLEG2), inpancreatic epithelium using a Pdx promoter, both KrasG12D-drivenPDA progression and the development of undifferentiated carci-noma was seen (14). However, this strategy does not reflectligand-mediated activation of the canonical Hh pathway, as doesSmoM2. CLEG2-expressing mice may develop pancreatic tumorsbecause this dominant active Gli2 allele is downstream or resistantto key posttranslational regulators of canonical Hh signaling. Post-translational regulators downstream of Smo have been shownpreviously to be key determinants in the temporal and spatialcontrol of Hh target-gene expression in tumors (32). In anotherstudy, mutations in the GLI transcription factors were identified inhuman PDA tumor epithelium (33). While this leaves open thepossibility that pancreatic epithelial GLI activity contributes toPDA tumorigenesis, the functional significance of the reported GLImutations remains to be determined. Our study does not rule outthe possibility that GLI activation plays a role in some PDA tumors,but shows that this activation is not the result of canonical Hhpathway signaling induced by Hh ligand over-expression. Interest-ingly, Gli activation in response to TGF� signaling has recently beenreported in PDA (31), suggesting that other signaling pathways maybe using the Gli transcription factors.

While the Hh signaling pathway is not active in tumor cells, its effectin tumor stroma may offer new avenues for the treatment of PDA, adisease often associated with abundant fibrosis and desmoplasticstroma (34, 35). This new paracrine model of Hh-mediated tumori-genesis may have important therapeutic consequences. Inhibition of apathway active in tumor stroma may target cells and tumor-promotingmechanisms that are currently not affected by other cytotoxic agentsthat target the tumor epithelium, and may therefore be effectively usedin combination with such agents.

Materials and Methods

Mice. R26-SmoM2 (referred to as SmoM2 allele in the presentstudy) was kindly provided by Andrew McMahon before publica-tion. PdxCre transgenic mice and LSL-KrasG12D mice (2) (referredas KrasG12D allele) were obtained from Andrew Lowy, Universityof Cincinnati, OH. Ink4a/Arfflox/flox was provided by Anton Bernsfrom the Netherlands Cancer Institute. P53R270H and LSL-KrasG12D

0.0001

0.001

0.01

0.1

1

10

100

1000

Gli Shh Ihh CK19

stromal

epithelial

PdxCre;KrasG12D;P53R270H

PdxCre;KrasG12D

PdxCre;KrasG12D;SmoM2

Rel

ativ

e ex

pres

sion

:stro

ma

vs.

epith

eliu

m,2

-∆∆C

t

0.01

0.1

1

10

100

1000

GLI1 SHH IHH CK19

stromal

epithelial

HF218761HF21886HF24320

Rel

ativ

e ex

pres

sion

:stro

ma

vs.

epith

eliu

m,2

-∆∆C

tA

B

0.001

0.01

0.1

1

10

100

1000

10000

GLI1 SHHIHH

Rel

ativ

e ex

pres

sion

:stro

ma

vs.e

pith

eliu

m,2

-∆∆C

t

stromal

epithelial

C

Fig. 4. Quantitative RT-PCR analysis of Hh pathway components in PDA andcolorectal tumors. (A) RT-PCR analysis of Hh pathway genes in the stromal vs.epithelial compartments following LCM of pancreatic samples from variousmouse models. Relative expression of Hh pathway genes is displayed as 2-��Ct inthe stroma vs. epithelium. CK19 shows the expected epithelium marker enrich-ment within the epithelial compartment. (B) RT-PCR analysis of Hh pathwaygenes in the stromal vs. epithelial compartment of human pancreatic carcinomasamples. GLI1 is up-regulated up to 120-fold in the stroma compared to theepithelium. Both SHH and IHH are enriched in the epithelium (up to 10-fold)when compared with the stroma. (C) RT-PCR analysis of Hh pathway genes in thestromal vs. epithelial compartments of liver metastases from human colorectalcancers.

4258 � www.pnas.org�cgi�doi�10.1073�pnas.0813203106 Tian et al.

mice (2) (referred as KrasG12D) were provided by Taylor Jacks(Massachusetts Institute of Technology). The Ptc-LacZ reporterline, also known as Ptch1D11 (24), was obtained from Matt Scott(Stanford University). These strains were intercrossed to producethe experimental cohorts. Mice were genotyped by PCR. Animalexperiments were approved by the Institutional Animal Care andUse Committee of Genentech.

Histology and IHC. Pancreatic tissue was fixed in 4% paraformalde-hyde in PBS for 1 h before processing for paraffin or OCTembedding. For histology and IHC, 3- to 6-�m sections of thepancreas were pressure cooked in target retrieval reagent (DAKO)for 15 min, then incubated with peroxidase blocking solution(DAKO) for 5 min before incubation with primary antibodies.Primary antibodies were: rat anti-CK19 (TROMAIII, 1:200, pro-vided by Dr. R. Kemler, Max Planck Institute, DevelopmentalStudies Hybridoma Bank), guinea pig anti-insulin (DAKO; 1:500),Rabbit anti-GFP (1:1,000, Torrey Pines), rabbit anti-p-ERK (1:500,Cell Signaling), Chicken anti vimentin (1:5,000, chemicon), rabbitanti-�-galactosidase (1:10,000, Cappel) (36) and rabbit anti-Shh(Genentech Inc.). Bright field IHC was done using the Envision�

system (DAKO). Immunofluorescence staining was carried out on7-�m cryosections and visualized using secondary antibodies fromJackson Immunoresearch.

�-Galactosidase Staining of the Pancreas. Pancreatic tissue was fixedin 4% paraformaldehyde for 1 h before processing for cryosection-ing. Seven-micrometer cryosections were stained with an Xgalstaining assay kit (Genlantis) at 37 °C for 2 h, according to themanufacturer’s instructions.

Cell-Based Assays. PDEC were obtained from Dave Tuveson’slaboratory. This ductal cell line was isolated from PDA of aPdxCre;KrasG12D animal. Pancreatic fibroblasts were isolated ac-cording to published methods. Briefly, wild-type pancreas wasdigested in PBS with 5 mg/ml of collagenase II and collagenase IVat 37 °C for 30 min before passage through a 100-�m cell strainer.Collagenase activity was stopped by washing the cells 2 times with

mouse embryonic fibroblast medium. The cells obtained from onepancreas were then seeded onto two 15-cm plates. The platebecame confluent in 7 days and the experiments were carried outwithin the next 2 passages.

Cells were cultured in 6-well plates in duplicate in 10% serum-containing medium until confluence before switching to 0.5% serumcontaining rShh (200 mg/ml) or Hh agonist (1 �g/ml). After 24-hincubation, RNA was isolated for standard qRT-PCR analysis (18).

Electroporation of ductal and fibroblast cells was performed onAmaxa electroporation apparatus, Solution V and program T-020was used for both cell types at 1 million cells per 3 �g DNA. Afterelectroporation, cells were allowed to recover in 10% serumcontaining MEF medium for 6 h before switching to 0.5% serummedium for 24 h. SmoM2-GFP-positive cells were FACS sorted andcollected for qRT-PCR analysis.

LCM and RT-PCR Analysis of Hh Pathway Genes. Fresh frozen, human,pancreatic adenocarcinomas were sectioned at 5 �m and mountedonto membrane slides for LCM. Up to 1 million �m2 of tumorepithelium and stroma was collected using a MMI automatedmicroscope. RNA was isolated by Pico pure RNA isolation kit(Arcturus), cDNA was prepared using the high capacity CDNAtranscription kit (Applied Biosystems). To analyze Hh pathwaygenes, cDNA from LCM was preamplifed using a TaqManPreAmp Master Mix Kit (Applied Biosystems) before qPCR. Theprimer sets for PreAmp and qPCR are as follows (AppliedBiosystems): Mm01197698-m1 (mGusb), Mm00439613�m1-(mIhh), Mm00436528�m1(mShh), Mm00494645�m1(mGli1),Hs00171790�m1(hGli1), Hs00179843�m1(hShh), Hs01081800�m1-(hIhh), Hs99999908�m1(hGUSb).ACKNOWLEDGMENTS. We thank Junhao Mao, Andrew P. McMahon, MattScott, Anton Berns, Andrew Lowy, and David Tuveson for mice and reagents. Wethank Anne Clermont, Jennifer Cox, Mallika Singh, and Monica Kong-Beltran formice, reagents, and valuable advice and discussions. Anti-Hh antibodies wereestablished with the assistance of Kurt Schroeder, Jo-Anne Hongo, Peggy Wen,Navneet Pal, Sheila Bheddah, Linda Rangell, Josman Labs LLC, and Epitomics. Wethank the Genentech microarray laboratory as well as the animal facility and theproduction assistance from Ryan Ybarra.

1. Hruban RH, Iacobuzio-Donahue C, Wilentz RE, Goggins M, Kern SE (2001) Molecularpathology of pancreatic cancer. Cancer J 7(4):251–258.

2. Hingorani SR, et al. (2003) Preinvasive and invasive ductal pancreatic cancer and its earlydetection in the mouse. Cancer Cell 4(6):437–450.

3. Guerra C, et al. (2007) Chronic pancreatitis is essential for induction of pancreatic ductaladenocarcinoma by K-Ras oncogenes in adult mice. Cancer Cell 11(3):291–302.

4. HingoraniSR,etal. (2005)Trp53R172HandKrasG12Dcooperatetopromotechromosomalinstability and widely metastatic pancreatic ductal adenocarcinoma in mice. Cancer Cell7(5):469–483.

5. Aguirre AJ, et al. (2003) Activated Kras and Ink4a/Arf deficiency cooperate to producemetastatic pancreatic ductal adenocarcinoma. Genes Dev 17(24):3112–3126.

6. Bardeesy N, et al. (2006) Both p16(Ink4a) and the p19(Arf)-p53 pathway constrain pro-gression of pancreatic adenocarcinoma in the mouse. Proc Natl Acad Sci USA103(15):5947–5952.

7. BardeesyN,etal. (2006)Smad4isdispensablefornormalpancreasdevelopmentyetcriticalin progression and tumor biology of pancreas cancer. Genes Dev 20(22):3130–3146.

8. Ijichi H, et al. (2006) Aggressive pancreatic ductal adenocarcinoma in mice caused bypancreas-specific blockade of transforming growth factor-beta signaling in cooperationwith active Kras expression. Genes Dev 20(22):3147–3160.

9. Izeradjene K, et al. (2007) Kras(G12D) and Smad4/Dpc4 haploinsufficiency cooperate toinduce mucinous cystic neoplasms and invasive adenocarcinoma of the pancreas. CancerCell 11(3):229–243.

10. Pasca di Magliano M, et al. (2007) Common activation of canonical wnt signaling inpancreatic adenocarcinoma. PLoS ONE 2(11):e1155.

11. Wagner M, et al. (2001) A murine tumor progression model for pancreatic cancer reca-pitulating the genetic alterations of the human disease. Genes Dev 15(3):286–293.

12. Miyamoto Y, et al. (2003) Notch mediates TGF alpha-induced changes in epithelial differ-entiation during pancreatic tumorigenesis. Cancer Cell 3(6):565–576.

13. Thayer SP, et al. (2003) Hedgehog is an early and late mediator of pancreatic cancertumorigenesis. Nature 425(6960):851–856.

14. Pasca di Magliano M, et al. (2006) Hedgehog/Ras interactions regulate early stages ofpancreatic cancer. Genes Dev 20(22):3161–3173.

15. Morton JP, et al. (2007) Sonic hedgehog acts at multiple stages during pancreatic tumor-igenesis. Proc Natl Acad Sci USA 104(12):5103–5108.

16. van den Brink GR (2007) Hedgehog signaling in development and homeostasis of thegastrointestinal tract. Physiol Rev 87(4):1343–1375.

17. Rubin LL, de Sauvage FJ (2006) Targeting the Hedgehog pathway in cancer. Nat Rev DrugDiscov 5(12):1026–1033.

18. Yauch RL, et al. (2008) A paracrine requirement for hedgehog signalling in cancer. Nature455(7211):406–410.

19. Apelqvist A, Ahlgren U, Edlund H (1997) Sonic hedgehog directs specialised mesodermdifferentiation in the intestine and pancreas. Curr Biol 7(10):801–804.

20. Xie J, et al. (1998) Activating Smoothened mutations in sporadic basal-cell carcinoma.Nature 391(6662):90–92.

21. Mao J, et al. (2006) A novel somatic mouse model to survey tumorigenic potential appliedto the Hedgehog pathway. Cancer Res 66(20):10171–10178.

22. Jeong J, Mao J, Tenzen T, Kottmann AH, McMahon AP (2004) Hedgehog signaling in theneural crest cells regulates the patterning and growth of facial primordia. Genes Dev18(8):937–951.

23. MerchantM,etal. (2005)Lossof theserine/threoninekinasefusedresults inpostnatalgrowthdefects and lethality due to progressive hydrocephalus. Mol Cell Biol 25(16):7054–7068.

24. Oro AE, Higgins K (2003) Hair cycle regulation of Hedgehog signal reception. Dev Biol255(2):238–248.

25. Goodrich LV, Milenkovic L, Higgins KM, Scott MP (1997) Altered neural cell fates andmedulloblastoma in mouse patched mutants. Science 277(5329):1109–1113.

26. Berman DM, et al. (2002) Medulloblastoma growth inhibition by hedgehog pathwayblockade. Science 297(5586):1559–1561.

27. Frank-Kamenetsky M, et al. (2002) Small-molecule modulators of Hedgehog signaling: iden-tification and characterization of Smoothened agonists and antagonists. J Biol 1(2):10.1–10.19.

28. Passman JN, et al. (2008) A sonic hedgehog signaling domain in the arterial adventitiasupports resident ScaI� smooth muscle progenitor cells. Proc Natl Acad Sci USA105(27):9349–9354.

39. Han YG, et al. (2008) Hedgehog signaling and primary cilia are required for the formationof adult neural stem cells. Nat Neurosci 11(3):277–284.

30. Corbit KC, et al. (2005) Vertebrate Smoothened functions at the primary cilium. Nature437(7061):1018–1021.

31. Nolan-Stevaux O, et al. (2009) GLI1 is regulated through Smoothened-independent mech-anisms in neoplastic pancreatic ducts and mediates PDAC cell survival and transformation.Genes Dev 23(1):24–36.

32. Huntzicker EG, et al. (2006) Dual degradation signals control Gli protein stability andtumor formation. Genes Dev 20(3):276–281.

33. Jones S, et al. (2008) Core signaling pathways in human pancreatic cancers revealed byglobal genomic analyses. Science 321(5897):1801–1806.

34. Mahadevan D, Von Hoff DD (2007) Tumor-stroma interactions in pancreatic ductal ade-nocarcinoma. Mol Cancer Ther 6(4):1186–1197.

35. Korc M (2007) Pancreatic cancer-associated stroma production. Am J Surg 194(4 Sup-pl):S84–S86.

36. Allen BL, Tenzen T, McMahon AP (2007) The Hedgehog-binding proteins Gas1 and Cdocooperate to positively regulate Shh signaling during mouse development. Genes Dev21(10):1244–1257.

Tian et al. PNAS � March 17, 2009 � vol. 106 � no. 11 � 4259

CELL

BIO

LOG

Y