A Role for Endoglin as a Suppressor of Malignancy during Mouse Skin Carcinogenesis

Endoglin is a membrane glycoprotein that acts as a coreceptor for transforming growth factor-B. We and others have previously suggested a function of endoglin as a tumor suppressor in epithelial cancer. Here, we study the expression of endoglin during chemical mouse skin carcinogenesis. We find that shedding of membrane endoglin, allowing the secretion of a soluble endoglin form, is a late event associated with progression from squamous to spindle cell carcinomas. Knock-down of endoglin in transformed keratinocytes activates the Smad2/3 signaling pathway resulting in cell growth arrest, delayed tumor latencies, and a squamous to spindle pheno-typic conversion. Forced expression of the long endoglin iso-form in spindle carcinoma cells blocks transforming growth factor-B1 stimulation of Smad2/3 signaling and prevents tumor formation. In contrast, expression of the short endoglin isoform has no effect on spindle cell growth in vitro or in vivo. Our results show that endoglin behaves as a suppressor of malignancy during the late stages of carcinogenesis. Therefore , disruption of membrane endoglin emerges as a crucial event for progression to spindle cell carcinomas.


Introduction
Endoglin (CD105) is a homodimeric cell membrane glycoprotein that is considered as a transforming growth factor-h (TGF-h) auxiliary receptor (1).Several data support an important role for endoglin in vascular development and physiology.Endoglin-null (Eng À/À ) mouse embryos die at midgestation due to cardiac and vascular defects (2)(3)(4).In addition, the endoglin gene is a target for the disorder hereditary hemorrhagic telangiectasia type I (HHT-1), a vascular dysplasia characterized by mucocutaneous telangiectasias and arteriovenous malformations (5).A second target gene for hereditary hemorrhagic telangiectasia (HHT-2) is the TGF-h type I receptor (ThRI) activin receptor-like kinase I (ALK1; ref. 6), which points to a close relationship between endoglin and TGF-h signaling in endothelial cells (7).Endoglin is overexpressed in endothelial cells during tumor neoangiogenesis, and several studies have defined this glycoprotein as a powerful marker to quantify intratumoral microvessel density in solid and hematopoietic malignan-cies (8,9).Current evidence suggests that endoglin acts as a molecular switch in endothelial cells by promoting TGF-h signaling through ALK1, which phosphorylates Smad1/5 and leads to the stimulation of cell migration and proliferation.In the presence of low levels of membrane-bound endoglin, TGF-h signals through a distinct ThRI, ALK5, which activates Smad2/3 and inhibits cell proliferation and migration (10,11).
In addition to the predominant long (L) isoform of endoglin, we have previously reported the expression of an alternative shorter (S) splice variant, called S-endoglin, in human and mouse tissues (12,13).These two endoglin isoforms share the extracellular and transmembrane domains and only differ from each other in their cytoplasmic tails.Studies with a transgenic mouse model that targets S-endoglin expression to the endothelium suggest an antagonistic antiangiogenic role for this variant with respect to the L-endoglin form (13).
Endoglin is highly expressed in endothelial cells of the tumor vasculature and at much lower levels in tumor cells.Besides the proangiogenic role of endoglin in endothelial cells (7)(8)(9)(10), there is also evidence supporting its involvement in malignant progression by its direct function on the tumor cells themselves.Thus, loss of endoglin expression in cultured human prostate cancer cells enhances cell migration and invasiveness (14).Also, we have reported that endoglin-heterozygous (Eng +/À ) mice exhibit accelerated malignant progression during chemical mouse skin carcinogenesis in vivo (13,15).
The two-stage chemical protocol (16) involves treatment of mice with a single dose of a carcinogen; i.e., 7,12-dimethylbenz(a)anthracene (DMBA), followed by repeated applications of the tumor promoter 12-O-tetradecanoylphorbol-13-acetate (TPA).This treatment gives rise to benign papillomas, some of which progress to malignant squamous cell carcinomas (SCC).SCCs can be classified into three different groups according to the degree of squamous differentiation (17): well differentiated (grade 1), moderately differentiated (grade 2), and poorly differentiated (grades 3/4).Some pathologists include spindle cell carcinomas (SpCC), a highly anaplastic tumor predominantly formed by elongated or spindle-shaped cells, into the group of poorly differentiated SCCs.The physiologic inducer that pushes SCC towards SpCC seems to be TGF-h1 (16).Several data support a causal relationship between TGF-h1 and the SCC-SpCC transition.First, transformed keratinocytes cultured in the presence of TGF-h1 undergo a reversible epithelial-mesenchymal transition (EMT) associated with the SCC-SpCC transition in vivo (18)(19)(20).Second, targeted expression of TGF-h1 in the epidermis of transgenic mice has a dual effect during two-stage carcinogenesis, by inhibiting the early appearance of benign papillomas, but increasing malignant progression to SpCC (21).These and other studies support the notion of a double role for TGF-h1 in carcinogenesis, by acting as a suppressor at early stages of tumorigenesis, but also as a stimulator of malignancy at later stages (16,22).The fact that Eng +/À mice exhibit an identical double phenotype to that of mice overexpressing TGF-h1 during carcinogenesis led us to postulate a role for endoglin in skin tumorigenesis by attenuating TGF-h1 signaling in keratinocytes (15).More importantly, this result suggested that endoglin could act as a suppressor of malignancy during multistage carcinogenesis.These hypotheses have been addressed here by analyzing the expression of endoglin mRNA and protein during mouse skin carcinogenesis in vivo and in vitro.In addition, we have investigated the functional effects of endoglin knockdown in transformed keratinocytes and endoglin overexpression in spindle carcinoma cells.

Materials and Methods
Cell culture conditions and treatments.Epidermal cell lines (Supplementary Table S1) were grown as described elsewhere (15).CarED was derived by explanting a poorly differentiated carcinoma induced in an Eng +/À mouse by DMBA/TPA (15).CarED cells are spindle-shaped in vitro and produce SpCC with latency periods of f4 weeks when injected into immunodeficient mice.
For TGF-h1 treatments, recombinant human TGF-h1 (PreproTech EC) was added to the culture medium at 10 ng/mL.The ThRI kinase inhibitor SB431542 (Sigma-Aldrich) was added to the cells at 10 Amol/L, and the antagonists of TGF-h binding P144 and P17 (DIGNA Biotech) were used at 200 Ag/mL.
Conditioned medium was obtained by growing cells in serum-free conditions (f10 6 cells/mL) for 48 h.Proteins were then concentrated by cold acetone at À20jC, and the presence of soluble endoglin determined by Western blotting.
Chemical carcinogenesis, tumor characterization, and immunohistochemistry.All animal experiments were approved and done according to institutional guidelines for animal care.Induction of tumors by initiation with DMBA and promotion with TPA for 15 weeks has been described (15).A total of 20 Swiss albino mice were used, from which 2 mice were left untreated to obtain normal epidermis.Groups of two to four animals bearing tumors were sacrificed at 15, 30, 38, and 43 weeks post-DMBA initiation.Tumors were excised, divided into two regions, and immediately frozen in liquid nitrogen or fixed in 10% formaldehyde and embedded in paraffin.Tumors were histologically typed by H&E staining of paraffin sections, and by reverse transcription-PCR (RT-PCR) analysis of the following differentiation/progression markers: K1 and K10, which are lost in SCCs (23); the extracellular matrix protein SPARC, which is induced during progression from papillomas to SCC (24); and the E-cadherin transcriptional repressor Snail, which accumulates in poorly differentiated SCCs (25).
Stable transfections and reporter assays.For short interfering RNA (siRNA)-mediated knockdown of endoglin, double-stranded oligonucleotides encoding siRNA that silence the mouse endoglin gene (10) were inserted into the pSUPER-GFP vector.The pSUPER-GFP Thermotoga maritima vector containing an endoglin-unrelated sequence (Qiagen) was used as a control.Transfections were carried out using LipofectAMINE plus (Invitrogen) according to the manufacturer's instructions.Individual clones were isolated using a FACS Vantage cell sorter (Becton Dickinson).Stable transfections with human L-and S-endoglin cDNAs subcloned into the pcDNA3 vector were carried out as previously described (13).
Reporter assays with TGF-h-responsive promoters were done as described elsewhere (11), using p(CAGA) 12 -luc (26) and pARE-Fast-luc (27,28) constructs, in the presence or absence of TGF-h1, as indicated.The pRL-tk Renilla normalizing luciferase vector (Promega) served as an internal control to correct for transfection efficiency and for normalization.
Quantitative and semiquantitative RT-PCR analysis.Semiquantitative and quantitative RT-PCR analyses were carried out as described (13,29).The oligonucleotides and PCR conditions used in these studies are described in Supplementary Table S2.
Western blot and flow cytometry analysis.Western blot analyses in lysates obtained from tumors and cell lines were done as reported (15).The primary antibodies used in this study are described in Supplementary Table S3.For flow cytometry analysis, cells (5 Â 10 5 ) were incubated with the anti-human endoglin monoclonal antibody (mAb) P4A4 (30) for 30 min at 4jC.Alexa Fluor 488-labeled anti-mouse IgG (Molecular Probes) was used as secondary antibody.Fluorescence was estimated with a BD FACScan using logarithmic amplifiers.
Cell proliferation, migration, and invasion assays.For cell growth assays, cells left at confluence for 24 h were plated in triplicate onto glass coverslips at a density of 5 Â 10 4 cells/60 mm dish with or without TGF-h 1 .Seventy-two hours later, cells were trypsinized and counted.Incorporation of 5-bromo-2 ¶-deoxyuridine (BrdUrd) into DNA was also determined by using the Cell Proliferation ELISA BrdUrd kit (Roche Diagnostic).In vitro wound healing and Matrigel invasion assays were done as described (31).
Tumorigenicity assays.For tumorigenicity assays, f10 6 viable cells were i.d.injected into the two flanks of 10-week-old nonobese diabeticsevere combined immunodeficiency mice, as detailed in Supplementary Table S4.The size of tumors was calculated from caliper measurements of two orthogonal diameters at different times.Tumors were excised in two regions and immediately frozen in liquid nitrogen ( for RT-PCR and Western blot analyses) or fixed in 10% formaldehyde ( for histologic and immunohistochemical characterization).RNA and protein was extracted from the frozen tumors using Tripure reagent (Roche Diagnostic).

Results
Membrane-bound endoglin is shed during the squamous to SpCC transition in vivo.The levels of endoglin transcripts present in the normal epidermis and chemically induced skin tumors were determined by quantitative RT-PCR (Fig. 1A).Two primer sets were used to amplify sequences either common to L-and S-endoglin mRNAs or specific for S-endoglin mRNA.The levels of S-endoglin transcripts in all samples were negligible with respect to the values obtained using primers that simultaneously amplify both endoglin isoforms (data not shown), confirming the absence of S-endoglin in the epidermis (13).Endoglin mRNA expression was reduced f5to 10-fold in papillomas relative to normal epidermis.These levels remained relatively low in early papillomas (15 weeks postinitiation), late papillomas (38 weeks), and well/moderately differentiated SCCs, but increased during progression to poorly differentiated carcinomas to almost equalize with those of the normal epidermis (Fig. 1A).
Endoglin expression was also studied by Western blotting using the specific mAb MJ7/18, which recognizes the NH 2 -terminal region of the mouse endoglin ectodomain (32).A specific band of f130 kDa corresponding to the membrane monomer (mEng) was detected in the normal epidermis (Supplementary Fig. S1).The expression profile of mEng during carcinogenesis was rather sinuous.The amount of mEng decreased in early papillomas paralleling the mRNA data, but rose to normal values in late papillomas to diminish again in well and moderately differentiated carcinomas.Interestingly, mEng was undetectable in poorly differentiated carcinomas (Fig. 1A), despite the relatively high levels of endoglin transcripts in these tumors.We also observed the presence of endoglin polypeptides of f50 to 65 kDa in the tumors (Supplementary Fig. S1).These smaller endoglin forms correspond to a soluble form of endoglin (sEng; ref. 33).sEng was absent from normal epidermis and the levels of sEng found in papillomas and early carcinomas were relatively low.However, the production of sEng increased significantly in poorly differentiated carcinomas, and this enhancement occurred concomitantly with the disappearance of mEng (Fig. 1A).Immunostaining in paraffin-embedded sections revealed the presence of endoglin at the plasma membrane of papillomas, but we were unable to detect membranebound endoglin in poorly differentiated carcinomas.Instead, in these tumors, intense endoglin staining was seen dispersed throughout the stroma and filling the lumen of blood vessels (Fig. 1B), indicating the presence of a cleaved endoglin ectodomain that contained the epitope recognized by the MJ7/18 mAb.Altogether, these results suggest that mEng is shed from the surface of skin tumor cells to produce a circulating sEng form, and that shedding occurs, i.e., during the progression from SCC to SpCC.
Spindle carcinoma cells produce soluble endoglin in vitro.In order to confirm whether endoglin shedding was associated with the spindle phenotype, we analyzed the expression of endoglin in a panel of cell lines corresponding to different stages of carcinogenesis; i.e., papilloma, SCC and SpCC (Supplementary Table S1).Overall, endoglin mRNA expression in these epidermal cell lines was highly reduced with respect to the skin in vivo, with levels f1% of those found in normal epidermis.The level of endoglin transcripts varied among the different cell lines regardless of its origin or malignant potential (Fig. 1C).The SpCC cell line CarED had the lowest levels of endoglin transcripts, as it was derived from a DMBA/TPA-induced skin carcinoma in an Eng +/À mouse.Nevertheless, the rest of the SpCC cell lines (CarC, CarB, and MSC11A5) expressed substantial amounts of endoglin mRNA comparable to those expressed in papilloma and SCC cell lines.Western blot analysis revealed the presence of the 130-kDa mEng monomer in premalignant keratinocytes (MCA3D, PB, and MSC11P6) and SCC cell lines (MSC11B9, PDV, HaCa4, Pam212, and CarC-R), but we were unable to detect mEng in the four spindle cell lines analyzed (Fig. 1C), as occurs in vivo.Among these cell lines, each pair formed by MSC11B9/MSC11A5 and CarC-R/ CarC represents a homogeneous cell system for malignant progression.Both MSC11B9 and MSC11A5 cell lines were derived from the same carcinoma, B9 from the squamous component and A5 from the anaplastic spindle region (34).Similarly, CarC-R represents a minor, less aggressive epithelial cell subpopulation isolated from the spindle carcinoma cell line CarC (35).Because mEng was present in MSC11B9 and CarC-R but not in MSC11A5 and CarC (Fig. 1C), these results suggest that ''loss'' of the 130-kDa membrane endoglin form is a specific event associated with the squamous to spindle transition.In addition, we were able to detect sEng in the conditioned medium of CarC, but not in the conditioned medium of MCA3D keratinocytes that express mEng (Fig. 1D).These data indicate that loss of mEng in spindle cells is due to shedding.We could not assess whether truncated mEng lacking part or all the ectodomain was present at the surface of CarC cells because of the lack of antibodies reacting with the endoglin cytoplasmic tail.
siRNA-mediated knockdown of endoglin in transformed keratinocytes promotes EMT.In order to examine whether loss of endoglin expression has any effect on the phenotype of squamous carcinoma cells, we chose the PDV cell line that exhibits an epithelial phenotype in vitro and produces well to moderately differentiated SCCs upon injection in mice (Supplementary Table S1).PDV cells were stably transfected with a vector encoding siRNA sequences that specifically suppress endoglin expression (10).As a control, an irrelevant siRNA was used (siCont).Two clones (siEng2 and siEng4) that showed reduced mRNA expression (Fig. 2A) and synthesized about a half and a third of the endogenous endoglin protein levels were obtained (Fig. 2B).In contrast to control cells, siEng cells had an elongated shape, grew dispersed, and were unable to form cohesive islands at low density (Fig. 2B, bottom).Consistent with these results, siEng2 and siEng4 clones showed diminished levels of epithelial proteins, such as E-cadherin, and induced the expression of mesenchymal/progression markers (Fig. 2B; Supplementary Fig. S2 and Supplementary Information).Down-regulation of E-cadherin protein did not involve a transcriptional mechanism, as the levels of E-cadherin mRNA expressed in siEng clones were similar to that of control cells (Fig. 2A).
PDV cells elicit a complete EMT when grown in the presence of TGF-h1, but it requires the continuous stimulus of the growth factor for >2 weeks (18).However, siEng cells were more sensitive to TGF-h1-induced complete EMT than control cells, as monitored by Western blot analysis of epithelial (E-cadherin) and spindle (vimentin) markers (Fig. 2C).We also analyzed whether TGF-h1-mediated EMT involved down-regulation of mEng.As shown in Fig. 2D, TGF-h1 promoted the progressive ''disappearance'' of mEng in siCont cells and the concomitant appearance of sEng in the conditioned medium.At 7 days of treatment with TGF-h1, loss of mEng was almost complete, indicating that the growth factor stimulates endoglin shedding prior to TGF-h1-induced EMT.
siRNA-mediated knockdown of endoglin in transformed keratinocytes activates Smad2/3 signaling and decreases cell growth in vitro and in vivo.In order to examine whether endoglin modulates TGF-h signaling in transformed keratinocytes, we analyzed the effect of endoglin knockdown on TGF-h1-induced Smad phosphorylation.In the absence of TGF-h1, siCont and siEng cell transfectants showed low amounts of phosphorylated Smad2 (P-Smad2) and P-Smad3 (Fig. 3A).However, the levels of P-Smad2 and P-Smad3 (normalized by the total amounts of protein) were slightly higher in siEng clones with respect to control cells, as seen by a longer exposure of the blot (data not shown).TGF-h1 stimulated Smad2 and Smad3 phosphorylation in both siCont and siEng cells, but P-Smad2 and P-Smad3 levels were higher in endoglin Figure 3. Endoglin down-regulation in PDV keratinocytes leads to enhanced Smad2/3 signaling, cell growth inhibition, and a SCC-SpCC conversion.A, Western blot analysis of Smad2/3 phosphorylation (P-Smad ) relative to the total expression levels at different times of treatment with TGF-h1.Representative of two independent experiments.B, TGF-h1-induced transcriptional activation of pARE-Fast and pCAGA luciferase reporter genes.Cells were transiently transfected with the abovementioned reporters and treated or not with TGF-h1 for 24 h.Transcriptional activity was measured using the luciferase reporter assay.A representative experiment using triplicate samples corrected for transfection efficiency out of three.C, cell growth assays in the absence or presence of TGF-h1.Synchronized cells were seeded in culture dishes and grown in medium plus serum in the absence or presence of TGF-h1.After 72 h, cells were trypsinized and counted.Columns, means of triplicate incubations; bars, SD.D, tumorigenic potential of siCont and siEng cells.Cells were injected into the two flanks of nonobese diabetic-severe combined immunodeficiency mice, and the sizes of the tumors induced by the cell lines were measured every 2 to 3 d.Right, average values from six tumors for each cell line.Left, the expression of E-cadherin and vimentin was analyzed by Western blot on samples isolated from individual tumors.Tumors were excised at 38 d post-injection.Detection of immunoglobulins present in the tumors served as a control for protein loading.knockdown cells.Consistent with these results, we found that both basal and TGF-h1-stimulated activities of the pARE-Fast and pCAGA reporter constructs, which are specific for Smad2 and Smad3 activation, respectively, were significantly higher in siEng cells (Fig. 3B).Basal pARE-Fast and pCAGA luciferase activities were f2to 3-fold higher in siEng clones compared with siCont cells, whereas TGF-h1 was able to stimulate pARE-Fast and pCAGA transactivation by 9-and 150-fold, respectively.These results indicate that reduced endoglin expression leads to increased basal and TGF-h1-stimulated Smad2/3 signaling in transformed keratinocytes.
TGF-h1 is a potent inducer of keratinocyte growth arrest (36), a response that seems to be mediated by the Smad2/3 pathway (22,37).In this regard, increased Smad2/3 signaling in siEng cell clones is compatible with their slow growth and increased sensitivity (f3-fold) to TGF-h1-induced growth inhibition compared with control cells (Fig. 3C).Similar results were obtained by measuring the incorporation of BrdUrd into DNA (data not shown).Reduced growth of siEng cells was also observed in vivo after injection of the cells into immunodeficient mice.siEng tumors grew slower and had longer latency periods than tumors induced by control cells (Fig. 3D; Supplementary Table S4).Nevertheless, they were vimentin-positive and E-cadherin-negative (Fig. 3D) and were typed as SpCCs, whereas siCont tumors were well/moderately differentiated SCCs, the same phenotype as tumors induced by the parental cell line (18).These results suggest that down-regulation of endoglin expression in transformed keratinocytes simultaneously leads to decreased cell growth and a SCC-SpCC conversion in vivo.
siRNA-mediated knockdown of endoglin in transformed keratinocytes stimulates cell migration and invasiveness.The phenotypic changes observed in siEng cells were associated with increased basal and TGF-h1-stimulated migratory and invasive abilities (Fig. 4A and B).To test whether the enhanced basal migratory potential of siEng cells involved the activation of TGF-h signaling, two distinct TGF-h antagonists, SB431542 and P144, were used.SB431542 is a potent inhibitor of the kinase activity of ThRI and blocks Smad2/3 signaling mediated by ALK5 (38).P144 is a betaglycan-derived soluble peptide that inhibits the binding of TGF-h1 to its receptors (39).Interestingly, treatment with SB431542, but not with P144, reduced the basal cell migration and invasiveness of siEng cells to the levels exhibited by siCont cells.In contrast, both antagonists blocked TGF-h1-stimulated cell migration and invasiveness of siCont cells (Fig. 4A and B).Identical results to those raised with P144 were obtained with P17, another antagonist peptide of similar characteristics as P144 (data not shown).These data show that activated Smad2/3 signaling in endoglin-down-regulated keratinocytes results in enhanced migration/invasiveness.Furthermore, activation of the Smad2/3 pathway is not triggered by an autocrine mechanism involving elevated secretion of TGF-h1, neither by up-regulation of ALK5 expression (see below).Instead, down-regulation of endoglin expression leads to the intrinsic activation of ALK5 at the cell surface, allowing enhanced Smad2/3 signaling and EMT.As a matter of fact, siEng cells reverted to the epithelial phenotype by blocking ThRI activity with SB431542, but not by preventing the binding of TGF-h1 with P144 (Fig. 4C).A, in vitro wound-healing assay.Wounded confluent cell cultures were cultured for 24 h with or without TGF-h1, in the absence or presence of either the ThRI kinase inhibitor SB431542 (10 Amol/L) or the TGF-h antagonist peptide P144 (200 Ag/mL).Quantification of migrated cells was done by measuring the healed area (n = 5).Data are representative of three independent experiments.B, Matrigel invasion assay.Serum-starved cells were seeded on the upper compartments of Matrigel invasion chambers.Serum (10%) and TGF-h1 (where indicated) were used as chemoattractants.SB431542 and P144 inhibitors were added to both the upper and lower compartments of the chamber at the concentrations described in A .Cells that went through the Matrigel-coated filter after 24 h were counted by submerging the underside of the filter in a 4 ¶,6-diamino-2-phenilindole solution (10 Ag/mL) in order to stain nuclei.Columns, means of four fields from duplicated assays; bars, SD.Represents one out of three independent experiments.C, treatment with SB431542, but not with P144, promotes a reversion to the epithelial phenotype.Phase contrast micrographs of siCont and siEng4 cells incubated for 24 h in the presence (or absence) of either SB431542 or P144 (top and middle ).The expression of E-cadherin in siEng4 cells was detected by immunofluorescence analysis (bottom ).siEng4 cells treated with SB431542 shows the typical keratinocyte E-cadherin staining at cell-cell contacts, whereas untreated cells or cells treated with P144 exhibit an aberrant E-cadherin cytoplasmic localization.
Expression of endoglin in spindle carcinoma cells inhibits Smad2/3 signaling and suppresses tumorigenicity.Next, we studied whether forced expression of endoglin affected the phenotype of spindle carcinoma cells.To this aim, CarC cells were transfected with cDNAs encoding the human L and S endoglin isoforms.Clones that stably expressed L-and S-endoglin (L-Eng and S-Eng, respectively) were obtained by cell sorting.A pool of CarC cells transfected with the empty vector was used as a control.Interestingly, flow cytometry analysis using the specific mAb P4A4 directed against the human endoglin ectodomain (30) showed the presence of cell surface L-endoglin in only 53% to 55% of L-Eng cells, whereas membrane S-endoglin was detected in the vast majority of S-Eng cells (Fig. 5A).Similar results were obtained by Western blot analysis of mEng expression (Supplementary Fig. S3A).Because small sEng forms were detected in the conditioned media of L-Eng but not of S-Eng transfectants (data not shown), the lower surface expression of L-endoglin relative to S-endoglin in the transfectants seems to be due to shedding.In addition, these data suggest that the short isoform is refractory to shedding.Another compelling observation is derived from the comparative analyses of the size of endoglin protein expressed in distinct cell types (see Supplementary Information).
Although expression of L-or S-endoglin did not significantly alter the fibroblastic morphology of CarC cells (Supplementary Fig. S3B), studies with specific reporter genes showed that L-endoglin blocked TGF-h1 stimulation of Smad2/3 signaling (Fig. 5B).Interestingly, S-endoglin did not only fail to inhibit transactivation of the reporter genes but seemed to enhance f2-fold the TGF-h1 stimulation of pCAGA luciferase activity (Fig. 5B), indicating that the presence of S-endoglin in CarC cells favors TGF-h1/Smad3 signaling.CarC cells are not growth-inhibited by TGF-h1 (36), and the expression of L-or S-endoglin did not significantly affect the growth properties of CarC cells in vitro (data not shown).However, L-endoglin had a remarkable effect on the growth of spindle CarC cells in vivo.Tumors induced by L-Eng clones had latency periods 3-to 4-fold longer than those induced by S-Eng and control cells (Fig. 5C; Supplementary Table S4).This difference was confirmed by injecting an additional third transfectant clone for each isoform (L-Eng3 and S-Eng3).All neo, L-Eng, and S-Eng tumors were histologically typed as SpCCs (Supplementary Table S4; Supplementary Fig. S3C).When the expression of endoglin was analyzed in vivo, significant levels of L-endoglin transcripts were observed in tumors induced by L-Eng transfectants, but we were unable to detect L-endoglin protein by Amplification of glyceraldehyde-3-phosphate dehydrogenase (GAPDH ) was used as a control for the amount of cDNA present in each sample, whereas detection of immunoglobulins present in the tumors served as a control for protein loading.Bottom, the production of sEng in the conditioned media of cell lines (passage = 2) derived from explantation of neo and L-Eng tumors was analyzed by Western blot.*, a nonspecific band unrelated to endoglin that partially masks sEng.
Western blot (Fig. 5D) or immunohistochemistry (Supplementary Fig. S3C).In contrast, S-endoglin protein was easily detected in the tumors induced by S-Eng cells.Furthermore, no L-endoglin protein was detected in the cell lysates of explanted L-Eng tumors, whereas smaller sEng forms were found in the conditioned media of these cells (Fig. 5D, bottom).In the case of L-Eng tumors, these results indicate that only those CarC cells that have lost surface expression of L-endoglin allow their growth in vivo, suggesting that those cells which actively shed mEng are selected for tumor formation.
ALK5 expression is reduced in endoglin-down-regulated cells.As the expression of ThRI and ThRII have been reported to change in Eng À/À and Eng +/À endothelial cells (10,40,41), we analyzed whether similar changes occur in our PDV and CarC cell transfectants.ALK5 mRNA expression was highly reduced in endoglin-down-regulated PDV keratinocytes, whereas the expression of ThRII was not affected (Fig. 6A).Both ALK5 and ThRII levels were down-regulated in CarC compared with PDV cells, and neither L-endoglin nor S-endoglin modified the expression of both receptors (Fig. 6B).Reduced levels of ALK5 and ThRII might explain the weaker response of CarC cells to pARE-Fast and pCAGA reporter genes compared with PDV keratinocytes (2.5-and 7-fold stimulation in CarC versus 9-and 150-fold in PDV, respectively).No expression of the endothelial-specific ThRI ALK1 was found in the cell lines.

Discussion
We show in this work that the TGF-h coreceptor endoglin plays a crucial role in skin carcinogenesis by modulating tumor cell growth and invasiveness.This role of endoglin most likely derives from its ability to interfere with TGF-h signaling.From the two known endoglin isoforms, it is the L form which is relevant for epidermal carcinogenesis; S-endoglin is barely expressed during carcinogenesis in vivo and in vitro, and it does not seem to have any influence on the growth or invasiveness of tumor cells.
An important finding in the present study is that endoglin shedding from the membrane, which disrupts mEng function and releases soluble sEng into the stroma, is an event associated with progression to the highly invasive spindle tumor phenotype in vivo.The proteolytic cleavage of the ectodomain is a phenomenon that occurs in many transmembrane proteins.For example, shedding of betaglycan, another TGF-h coreceptor, disrupts the enhancer activity of TGF-h binding exhibited by the membrane-bound form and produces a soluble protein that inhibits TGF-h binding to its surface receptors (42).Similarly, sEng has been involved in the pathogenesis of the pregnancy-specific hypertensive syndrome preeclampsia by preventing binding of TGF-h1 to its receptors and impairing downstream signaling activity (33,43).sEng is also found in the sera and/or plasma of patients with different neoplasias (9).More importantly, elevated levels of serum sEng (in other words, enhanced endoglin shedding) correlates with metastasis in patients with colorectal, breast, and other solid malignancies (44,45).These observations are in line with our finding that endoglin shedding is a late event in carcinogenesis associated with the development of poorly differentiated carcinomas, and suggest that the tumor suppressor function of mEng, as determined in this work, might also be relevant for human cancer.The enzyme(s) (sheddases) involved in the release of the endoglin ectodomain are presently unknown.Nevertheless, a clue for the identification of this sheddase(s) comes from the fact that TGF-h1 stimulates endoglin shedding in cultured keratinocytes, and therefore, the enzyme(s) involved might be regulated by TGF-h1.Although our in vivo and in vitro data point to neoplastic cells as a key contributor to sEng, the role of stromal cells in sEng production cannot be ruled out.Further studies are necessary to determine the relative contribution of the distinct tumor cell components to sEng and whether sEng itself has any influence on tumor development.
To study the functional implications of down-regulating mEng function in tumor cells, we used siRNA technology.Our data indicate that endoglin attenuates TGF-h1/Smad2/3 signaling in keratinocytes.Knockdown of endoglin in transformed keratinocytes stimulated basal and TGF-h1-mediated ALK5/Smad2/3 signaling activity.Interestingly, basal ALK5 activation was innate and did not involve enhanced binding of TGF-h or increased expression of the receptors.ALK5 activation promoted EMT and the in vivo development of a spindle phenotype associated with enhanced cell migration/invasiveness.These results are in accordance with the fact that progression to poorly differentiated carcinomas was vastly accelerated during carcinogenesis in Eng +/À mice (13,15), and strongly indicate that mEng behaves as a suppressor of malignancy at late stages of carcinogenesis.Consequently, disruption of mEng by shedding emerges as a crucial event for TGF-h1-mediated SCC-SpCC transition in vivo.Interestingly, a recent report (46) shows that endoglin also regulates EMT in the developing heart during embryogenesis.
Down-regulation of mEng in transformed keratinocytes leads to cell growth arrest in vitro and in vivo, likely because of increased ALK5/Smad2/3 signaling.Keratinocytes with reduced mEng levels down-regulate ALK5 expression probably as an adaptation mechanism in order to proliferate, as reported in endothelial cells lacking endoglin (10).The fact that growth inhibition occurs concomitantly with enhanced cell migration and invasiveness indicates that the tumor suppressor and oncogenic functions of TGF-h overlap.Therefore, loss of the TGF-h growth-inhibitory response seems to be a prerequisite for SCC to progress to SpCC.Most cells derived from SCCs remain responsive to the TGF-h growth-inhibitory response, and resistance to the inhibitory effects of the growth factor seems to occur late during carcinogenesis, which is associated with the spindle phenotype (36,47).Several mechanisms might account for the lack of an antiproliferative response in tumor cells harboring no mutations in the TGF-h receptors or Smads (22,37).Thus, unresponsive spindle CarC cells have deleted the Ink4b locus encoding the cyclin-dependent kinase inhibitor p15 (48), a key mediator of the TGF-h antiproliferative response (37).CarC cells, however, respond to the TGF-h1 stimulation of Smad2/3 signaling and, therefore, have an intact TGF-h/ Smad pathway.Nevertheless, the magnitude of stimulation of the Smad2/3 pathway by TGF-h1 in CarC is lower than in PDV cells, likely because CarC cells have down-regulated both ALK5 and ThRII expression.We postulate that loss-of-function of mEng must exert a strong selective pressure during tumor evolution to prime cells that have missed or inactivated components of the TGF-h antiproliferative response.These selected cells will be prone to TGF-h1-mediated EMT giving rise to SpCCs.This could be the reason why loss of the TGF-h1 antiproliferative response cosegregates with the highly invasive spindle phenotype.As downregulation of endoglin expression leads to the inhibition of keratinocyte cell growth, the significance of the reduction of mEng expression observed in early papillomas could be to counteract the chronic proliferative response of keratinocytes to TPA (16).Thus, in Eng +/À mice, or in mice overexpressing TGF-h1 in the epidermis, chemical carcinogenesis produces a much reduced papilloma load (15,21).
Interestingly, forced expression of L-endoglin was shown to suppress the growth of CarC spindle cells in vivo but not in vitro.In contrast, S-endoglin was innocuous for CarC cell growth.The molecular basis for this effect remains to be determined, but seems to reside in the cytoplasmic tail of the molecule, as it is the sole domain which is different between L-and S-endoglin.On the other hand, S-endoglin specifically stimulated TGF-h1/Smad3 signaling, an opposite action with respect to L-endoglin, reinforcing our suggestion that L-and S-endoglin isoforms exert antagonistic roles (13).

Figure 1 .
Figure 1.Endoglin expression during mouse skin carcinogenesis.A, transcriptional expression of endoglin in normal epidermis and tumors induced by DMBA/ TPA (top ).Normalized endoglin transcripts levels were measured by quantitative RT-PCR relative to glyceraldehyde-3-phosphate dehydrogenase transcript levels.Columns, means from three independent experiments; bars, SD.Quantification of membrane endoglin (mEng ) and soluble endoglin (sEng ) protein expression (middle and bottom ).The intensities of the 130-kDa (mEng ) and 50 to 65 kDa (sEng ) bands of the blot in Supplementary Fig. S1 (at the top of the corresponding graphs) were quantified by densitometric analysis and normalized to h-actin, used as a control for protein loading.Ep, normal epidermis; Pap, papilloma; SCC, squamous cell carcinoma.All SCCs were obtained at 43 wk post-initiation, except the tumor on lane 16 which was excised at 38 wk.B, immunohistochemical detection of endoglin in skin tumors.A membrane-bound endoglin form (mEng ) was detected in a 15-week-old papilloma, which corresponds to that of lane 3 in the blots in A. A circulating soluble form of endoglin (sEng ) was detected in the stroma (s ) and inside of blood vessels (*) in a SCC 3/4 corresponding to that of lane 22 in the blots in A .t, tumor mass; e, hyperplastic epidermis adjacent to the tumor.C, endoglin expression in mouse epidermal cell lines.The normalized endoglin transcript levels (arbitrary value = 1 for MCA3D), as measured by quantitative RT-PCR (top ).Representative Western blot of mEng in whole cell lysates.No sEng form was found in the blot (bottom ).a-Tubulin was used as a control for protein loading.D, sEng is produced in spindle carcinoma cells.Western blot analysis of endoglin in cell lysates and conditioned media of MCA3D and CarC cells.

Figure 2 .
Figure 2. Endoglin down-regulation in transformed PDV keratinocytes promotes an EMT.A, expression of endoglin and E-cadherin transcripts by RT-PCR in PDV cells stably transfected with siRNA endoglin (siEng ) or siRNA control (siCont ).Glyceraldehyde-3-phosphate dehydrogenase (GAPDH ) was amplified as a control for the amount of cDNA present in each sample.B, Western blot analysis of mEng and of differentiation/progression markers in siCont and siEng transfectants.The extent of mEng protein reduction in siEng cells is indicated.a-Tubulin was used as a control for protein loading.Phase contrast micrographs of siCont and siEng cells (bottom ).C, siEng are more sensitive than siCont cells to TGF-h1-induced complete EMT.Western blot analyses of E-cadherin and vimentin expression at different times of treatment with TGF-h1.D, TGF-h1 stimulates mEng shedding in siCont cells.Western blot analysis of endoglin in cell lysates (top ) and conditioned media (bottom ) of siCont cells at different times of treatment with TGF-h1.*, nonspecific band corresponding to an abundant secreted polypeptide unrelated to endoglin.

Figure 4 .
Figure 4. Endoglin down-regulation in PDV keratinocytes stimulates cell migration and invasiveness.A, in vitro wound-healing assay.Wounded confluent cell cultures were cultured for 24 h with or without TGF-h1, in the absence or presence of either the ThRI kinase inhibitor SB431542 (10 Amol/L) or the TGF-h antagonist peptide P144 (200 Ag/mL).Quantification of migrated cells was done by measuring the healed area (n = 5).Data are representative of three independent experiments.B, Matrigel invasion assay.Serum-starved cells were seeded on the upper compartments of Matrigel invasion chambers.Serum (10%) and TGF-h1 (where indicated) were used as chemoattractants.SB431542 and P144 inhibitors were added to both the upper and lower compartments of the chamber at the concentrations described in A .Cells that went through the Matrigel-coated filter after 24 h were counted by submerging the underside of the filter in a 4 ¶,6-diamino-2-phenilindole solution (10 Ag/mL) in order to stain nuclei.Columns, means of four fields from duplicated assays; bars, SD.Represents one out of three independent experiments.C, treatment with SB431542, but not with P144, promotes a reversion to the epithelial phenotype.Phase contrast micrographs of siCont and siEng4 cells incubated for 24 h in the presence (or absence) of either SB431542 or P144 (top and middle ).The expression of E-cadherin in siEng4 cells was detected by immunofluorescence analysis (bottom ).siEng4 cells treated with SB431542 shows the typical keratinocyte E-cadherin staining at cell-cell contacts, whereas untreated cells or cells treated with P144 exhibit an aberrant E-cadherin cytoplasmic localization.

Figure 5 .
Figure 5.Effect of the ectopic expression of human L-and S-endoglin on the behavior of spindle CarC cells.A, the expression of exogenous L-and S-endoglin in the transfectants was analyzed by flow cytometry.Top right, percentages of cells expressing endoglin at the cell surface.B, TGF-h1-induced transcriptional activation of pARE-Fast and pCAGA luciferase reporter genes.Experimental conditions were as in the legend of Fig. 3B .C, tumorigenic potential of CarC cell transfectants.Points, means of all tumors induced by the three clones corresponding to each transfected isoform, as detailed in Supplementary Table S4; bars, SD.D , characterization of exogenous endoglin expression in tumors induced by the cell transfectants.Tumors induced by the S-Eng and neo cell transfectants were excised at 21 d post-injection, whereas those induced by L-Eng transfectants were excised at 31 d post-injection.The expression of either L-endoglin or S-endoglin was analyzed by RT-PCR (top ) and Western blot (middle ).Amplification of glyceraldehyde-3-phosphate dehydrogenase (GAPDH ) was used as a control for the amount of cDNA present in each sample, whereas detection of immunoglobulins present in the tumors served as a control for protein loading.Bottom, the production of sEng in the conditioned media of cell lines (passage = 2) derived from explantation of neo and L-Eng tumors was analyzed by Western blot.*, a nonspecific band unrelated to endoglin that partially masks sEng.

Figure 6 .
Figure 6.Expression of ALK5, ALK1, and ThRII transcripts in siRNA endoglin PDV cell transfectants (A) and human L-and S-endoglin CarC cells transfectants (B) by RT-PCR.Amplification of glyceraldehyde-3-phosphate dehydrogenase (GAPDH ) was used as a control for the amount of cDNA present in each sample.ÀRT lane, the results of amplification in the absence of cDNA.RNA isolated from murine lung was used as a positive control for ALK1 expression.