Endoglin Expression in Breast Tumor Cells Suppresses Invasion and Metastasis and Correlates with Improved Clinical Outcome

Oncogene (2011) 30, 1046–1058 & 2011 Macmillan Publishers Limited All rights reserved 0950-9232/11 www.nature.com/onc ORIGINAL ARTICLE Endoglin expression in breast tumor cells suppresses invasion and metastasis and correlates with improved clinical outcome

LA Henry1, DA Johnson1, D Sarrio´1, S Lee1, PR Quinlan2, T Crook1, AM Thompson2, JS Reis-Filho1 and CM Isacke1

1Breakthrough Breast Cancer Research Centre, Institute of Cancer Research, London, UK and 2Department of Surgery and Molecular Oncology, Ninewells Hospital and Medical School, Dundee, UK

Tumor growth factor-b (TGF-b) signaling in cancer has been and metastatic events in late stage carcinomas (Roberts implicated in growth suppression of early lesions and and Wakefield, 2003; Massague, 2008). TGF-b ligands act enhancing tumor cell invasion and metastasis. However, by binding to their cognate transmembrane serine/ the cellular mechanisms that determine this signaling output threonine kinase receptors (Shi and Massague, 2003; Feng in individual tumors are still largely unknown. In endothelial and Derynck, 2005). This results in the activation of the cells, TGF-b signaling is modulated by the TGF-b co- canonical TGF-b pathway mediated by the phosphoryla- receptor endoglin (CD105). Here we demonstrate that tion of Smad proteins and their subsequent translocation endoglin is expressed in a subset of invasive breast cancers to the nucleus where they promote the transcriptional and cell lines and is subject to epigenetic silencing by gene regulation of TGF-b response genes. However, TGF-b can methylation. Endoglin downregulation in non-tumorigenic also signal through a range of other pathways (Moustakas MCF10A breast cells leads to the formation of abnormal and Heldin, 2005; Zhang, 2009) and these alternate acini in 3D culture, but does not promote cell migration or pathways are particularly important in the pro-invasive transformation. In contrast, in the presence of activated responses to TGF-b in cancer (Bhowmick et al., 2001; ErbB2, endoglin downregulation in MCF10A cells leads to Ozdamar et al., 2005). enhanced invasion into a 3D matrix. Consistent with these Endoglin (CD105) is a type I integral transmembrane data, ectopic expression of endoglin in MDA-MB-231 cells glycoprotein that has predominantly been investigated blocks TGF-b-enhanced cell motility and invasion and in the vasculature where it is upregulated during reduces lung colonization in an in vivo metastasis model. angiogenesis (Bernabeu et al., 2007, 2009). In endothe- Unlike endothelial cells, endoglin does not modulate Smad- lial cells, endoglin associates at the cell surface with the mediated TGF-b signaling in breast cells but attenuates the type I and type II TGF-b receptors and has a key role in cytoskeletal remodeling to impair cell migration and modulating downstream signaling to the Smad transdu- invasion. Importantly, in a large cohort of invasive breast cer proteins (Guerrero-Esteo et al., 2002; Lebrin et al., cancers, lack of endoglin expression in the tumor cell 2004). The importance of endoglin acting as a TGF-b compartment correlates with ENG gene methylation and co-receptor in these events is evidenced by the observa- poor clinical outcome. tion that mutations in either ENG or the type I receptor Oncogene (2011) 30, 1046–1058; doi:10.1038/onc.2010.488; ALK1 give rise to the vascular disorder hereditary published online 1 November 2010 hemorrhagic telangiectasia (McAllister et al., 1994; Johnson et al., 1996). From an initial demethylation Keywords: endoglin; CD105; breast cancer; TGF-b; screen, we identified ENG as a candidate gene for MCF10A; ErbB2 epigenetic regulation in breast cancer cell lines. Together with previous reports that endoglin can inhibit the migration of other epithelial tumors (Craft et al., 2007; Perez-Gomez et al., 2007), we hypothesized that Introduction expression of endoglin might function to suppress breast cancer progression. To address this we investigated the The tumor growth factor-b (TGF-b) signaling pathway functional role of endoglin in both in vitro and in vivo has an important role in all stages of breast cancer models, its impact on the TGF-b signaling pathway and progression. It is well established that TGF-b has a dual the prognostic impact and regulation of endoglin function in acting both as a tumor suppressor in the early expression in a large cohort of invasive breast cancers. stage of malignant progression and to promote invasive Results Correspondence: Professor CM Isacke, Breakthrough Breast Cancer Research Centre, Institute of Cancer Research, 237 Fulham Road, Downregulation of endoglin expression enhances breast London SW3 6JB, UK. E-mail: [email protected] cancer cell invasion Received 30 March 2010; revised 31 August 2010; accepted 3 September A panel of cell lines was subject to immunoblotting and 2010; published online 1 November 2010 quantitative real-time PCR (qPCR). Endoglin mRNA Endoglin suppresses breast cancer invasion LA Henry et al 1047 100 Endoglin MCF10A 75 (Long exposure) Endoglin 100 75 (Short exposure) shEng1 shEng5 shEng2 shCont Parental 50 Tubulin 100 Endoglin 75

50 37 ERK1/2 Relative mRNA expression

p=0.0145

p=0.0141

p=0.1927

Parental shCont shEng2 % of abnormal acini

shEng1 shEng5 shCont shEng1 shEng2 shEng5 Parental Figure 1 Endoglin downregulation enhances abnormal MCF10A acini formation in 3D culture. (a) Endoglin expression analyzed in a series of breast cell lines by immunoblotting (upper panels) or qPCR (lower panel). Data shown are the mean±s.e.m. of RQ relative expression of two independent experiments. (b) Immunoblotting of MCF10A cells stably infected with different shRNA lentiviral constructs against endoglin (shEng) or control (shCont). Note, the shEng2 cell line showed negligible endoglin knockdown and therefore was used as an independent shRNA control. (c) MCF10A cell lines stably expressing control shRNA (shCont, shEng2) or endoglin knockdown (shEng1, shEng5) shRNA constructs were grown in 3D culture and fixed on day 20. Cells were permeabilized and stained for collagen IV and Alexa488 anti-rabbit Ig (green), the Golgi protein b-COP and Alexa555 anti-rat Ig (red) and DAPI (blue). Left panel, representative confocal images. Scale bar, 25 mm. Right panel, mean % of abnormal acini morphology for three independent experiments±s.e.m. and protein expression was readily detected in the non- endoglin lentiviral small hairpin RNAs (shRNAs) tumorigenic breast cell line MCF10A, in a subset of the (shEng1–5) targeting different regions of the ENG gene cancer cell lines (MDA-MB-468, SKBR3, T47D) and in or with control shRNA (shCont) (Figure 1b). Substan- the recently re-classified melanoma line MDA-MB-435 tial knockdown was observed with shEng1 and shEng5 (Figure 1a). This expression was 2.5–5-fold lower than but not with shEng2. The shEng2 and shCont cell lines found in the cultured endothelial cells (Supplementary expressed similar levels of endoglin as the parental line Figure 1). No endoglin expression was detected in and hence were used as two independent controls. In MCF7, Cal51, MDA-MB-231, MDA-MB-453, HCC1954 this 3D culture model, shCont and shEng2 cells were or BT474 cells. Reverse transcription–PCR analysis indistinguishable from the parental MCF10A cells and revealed that in the endoglin-positive cell lines the formed single acini that contained a hollow lumen predominant expression was of the long-endoglin isoform surrounded by a polarized layer of epithelial cells with with little or no expression of the short-endoglin isoform their Golgi orientated inwards (Figure 1c). In contrast, detected (data not shown). the endoglin-knockdown lines (shEng1 and shEng5) To explore a potential role for endoglin as a breast produced abnormally shaped acini in which the outer cancer suppressor, we initiated this study with the non- layer of cells was not fully polarized and the lumena tumorigenic MCF10A cell line. MCF10A cells when were incompletely cleared. Nonetheless, these shEng1 grown on a thick layer of Matrigel form epithelial acini and shEng5 acini, still growth arrested, did not invade (3D culture model), and this system has been used into the Matrigel and were no larger than control extensively to identify pathways that can perturb the structures indicating that the loss of endoglin expression normal breast architecture (Debnath and Brugge, 2005). alone does not result in the transformation of the non- MCF10A cells were stably infected with five individual tumorigenic MCF10A cells.

Oncogene Endoglin suppresses breast cancer invasion LA Henry et al 1048 Cell line Expression MCF10A.ErbB2 parental 100% MCF10A.ErbB2 shCont 91.1% MCF10A.ErbB2 shEng1 33.2% MCF10A.ErbB2 shEng2 71.0% MCF10A.ErbB2 shEng5 4.7%

p=0.0126 p=0.0428 Parental shEng1

p=0.0403 p=0.0087

p=0.7998

shCont shEng5 % of acini with invasive protrusions

-+++ - +++ -+++ +++ -+++- AP1510 --+--+- --+----+- -+- - TGF-β1 shEng2 ---+--+ ---+---+---- + SB431542 Parental shContshEng2 shEng1 shEng5

Figure 2 Endoglin downregulation cooperates with ErbB2 activation to promote invasion in 3D culture. (a) MCF10A.ErbB2 cells were infected with shCont or shEng lentiviral constructs. Stable cell lines were subject to FACS analysis with an isotype control or endoglin antibody. The mean level of fluorescence signal (FITC-A) relative to the parental cells is shown. Note, the shEng2 cell line has negligible endoglin knockdown and therefore was used as an independent shRNA control. (b) MCF10A.ErbB2 cell lines were grown in 3D culture for 5 days±TGF-b1±SB431542±AP1510 dimerizer added from day 4. Acini were stained for laminin V and Alexa488 anti-mouse Ig (green), Alexa546-phalloidin (red), and DAPI (blue). Left panel, representative confocal images of cultures with AP1510 dimerizer added from day 4 in the absence of TGF-b1 and SB431542. Arrowheads indicate invasion protrusions and laminin V degradation. Scale bar, 25 mm. Right panel, mean % of acini with invasive protrusions for three independent experiments±s.e.m.

We next tested whether endoglin might function to AP1510-treated cultures produced acini with signifi- modulate the effects of activated oncogenes. For this, we cantly more protrusions that invaded through the employed MCF10A cells ectopically expressing a basement membrane compared with shCont cell lines chimeric ErbB2 fusion protein (MCF10A.ErbB2). In (shEng1, P ¼ 0.0403; shEng5, P ¼ 0.0126). These endo- these MCF10A.ErbB2 cells, addition of the AP1510 glin knockdown lines also produced more invasive acini dimerizer activates ErbB2 signaling (Muthuswamy in the absence of AP1510, which probably reflects leaky et al., 2001) and can be used to delineate the molecular activation of the ErbB2 fusion protein (Seton-Rogers events that cooperate with oncogenes during tumor et al., 2004). Again, the number of invasive acini was progression (Seton-Rogers et al., 2004; Aranda et al., significantly reduced in the presence of the SB43152 2006). MCF10A.ErbB2 cells were infected with endoglin inhibitor (shEng1, P ¼ 0.0087; shEng5, P ¼ 0.0428). or shCont lentiviruses and again substantial knockdown In long-term culture (day 20), the acini from the shEng- was achieved with shEng1 and shEng5, but only knockdown lines (shEng1, shEng5) were more motile minimal knockdown was achieved with shEng2 and invasive than those from the shCont or parental (Figure 2a). In the control cell lines (parental, shCont, lines, and coalesced to form larger, more disorganized shEng2) addition of the AP1510 dimerizer for the last and multiacinar structures (Supplementary Figure 2). In 24 h of the 5-day culture resulted in an increased number both the MCF10A and MCF10A.ErbB2 lines, down- of acini invading into the Matrigel containing 3D regulation of endoglin expression had no effect on cell matrix. Despite no significant effect of exogenously proliferation in the presence or absence of TGF-b added TGF-b1, this increase was abrogated by the (Supplementary Figure 3). Together these data demon- addition of the TGF-b receptor inhibitor SB431542 strate that endoglin serves to limit the pro-invasive (Figure 2b), which is consistent with the presence of effects of ErbB2 activation in non-tumorigenic breast TGF-b in Matrigel (see Materials and methods). In the cells and suggest that in the context of an activated endoglin knockdown lines (shEng1 and shEng5), oncogene, endoglin acts as breast cancer suppressor.

Oncogene Endoglin suppresses breast cancer invasion LA Henry et al 1049 Endoglin does not modulate Smad-mediated TGF-b 2004; ten Dijke et al., 2008). To further explore the signaling interplay between endoglin and TGF-b signaling in In endothelial cells, it is well documented that endoglin breast epithelial cells Smad activation was examined in functions as a TGF-b co-receptor modulating TGF-b the MCF10A (Figures 3a and b) and MCF10A.ErbB2 signaling to the Smad transducer proteins (Lebrin et al., treated with AP1510 (Figures 3c and d) cells. It should

MCF10A MCF10A shCont shEng5

0 0.5 1 2 4 24 0 0.5 1 2 4 24 hr TGF-β1 50 p-Smad2 50 Smad2/3

50 p-Smad1/5/8 50 Smad1 37 p-ERK1/2 37 ERK1/2 50 Tubulin

SNAI2 SERPINE1 ANGPTL4 Relative expression Relative -+ -+ -+ -+ -+ -+ TGF-β1 shCont shEng5 shCont shEng5 shCont shEng5

MCF10A

MCF10A.ErbB2 MCF10A.ErbB2 shCont shEng5

0 0.5 1 2 4 24 0 0.5 1 2 4 24 hr TGF-β1 50 p-Smad2

50 Smad2/3

50 p-Smad1/5/8 37 ERK1/2

SNAI2 SERPINE1 ANGPTL4 Relative expression Relative --++ --++ --++TGF-β1 shCont shEng5 shCont shEng5 shCont shEng5

MCF10A.ErbB2 Figure 3 Endoglin does not modulate Smad-mediated TGF-b signaling in MCF10A or MCF10A.ErbB2 cells. (a, c) MCF10A or MCF10A.ErbB2 shCont and shEng5 cell lines were starved overnight and treated with 5 ng/ml TGF-b1 (MCF10A) or 5 ng/ml TGF-b1 plus 1 mM AP1510 dimerizer (MCF10A.ErbB2) for 0–24 h. Cells were lysed in reducing sample buffer, total protein lysate resolved by 10% SDS–PAGE and subject to immunoblotting with the indicated antibodies. Molecular size markers are in kDa. (b, d) MCF10A or MCF10A.ErbB2 shCont and shEng5 cell lines were starved overnight and treated±5 ng/ml TGF-b1 (MCF10A) or 1 mM AP1510±5 ng/ml TGF-b1 (MCF10A.ErbB2) for 3 h. Cells were lysed and total RNA extracted. cDNA was synthesized and subject to qPCR with Taqman probes specific for SNAI2, PAI1 and ANGPTL4. Data shown is the mean±s.e.m. of RQ relative expression from three independent experiments.

Oncogene Endoglin suppresses breast cancer invasion LA Henry et al 1050 be noted that in comparison with the data shown in with ErbB2 activation significantly increased cell moti- Figure 2b, cells were serum starved overnight to ensure lity (P ¼ 0.0255; Figure 4a, compare bars 3, 5 and 7). basal levels of TGF-b signaling at time 0. In both cell Consistent with previous reports, (Seton-Rogers et al., lines, endoglin downregulation had no effect on TGF-b 2004), we also observed that activation of ErbB2 renders induced Smad2/3 or Smad1/5/8 phosphorylation. Simi- MCF10A cells responsive to the pro-migratory effects of larly, endoglin downregulation had no effect on the exogenously added TGF-b1. Confocal microscopy transcriptional upregulation of the TGF-b responsive analysis of the wounded AP1510-treated MCF10A.Erb2 genes SNAI2, ANGPTL4 and PAI1. These data also cell monolayers revealed that shCont transfected cells demonstrate that addition of AP1510 dimerizer to the had very few focal adhesions and a thick layer of actin MCF10A.ErbB2 cells in the absence of TGF-b does not filaments parallel to the wound edge, consistent with the stimulate Smad activation or transcription of TGF-b- inability of these cells to actively migrate into the wound responsive genes. (Figure 4b). In contrast, the actively migrating shEng5 Alternative non-Smad signaling, particularly signal- cells formed numerous small vinculin-positive focal ing to the cytoskeleton, has been shown to have an adhesions and actin protrusions. Consistent with the important role in the pro-invasive effects of TGF-b cell migration data (Figure 4a), treatment of (Moustakas and Heldin, 2005; Zhang, 2009). To MCF10A.ErbB2 shCont cells with TGF-b1 resulted in investigate the effect of endoglin on cytoskeletal an increased number of focal adhesions. However, these regulation, MCF10A and MCF10A.ErbB2 cells were were notably larger in size than those found in the TGF- subject to scratch-wound migration assays (Figure 4). b-treated shEng5 knockdown cells. It is well established The MCF10A cell migration was modestly inhibited by that larger focal adhesions are more mature, longer- TGF-b and this inhibition was not relieved by endoglin lived structures resulting from reduced focal adhesion downregulation (Figure 4a). In endothelial cells and turnover and that this in turn leads to less efficient cell myoblasts, it has been reported that endoglin suppresses migration (Webb et al., 2002). Together these data TGF-b-mediated ERK1/2 activation (Rodriguez-Bar- support a role for endoglin in attenuating activated bero et al., 2006; Lee and Blobe, 2007). In these ErbB2-promoted cell migration through effects on the MCF10A epithelial cells, endoglin downregulation had focal adhesions and the actin cytoskeleton. no effect on the duration or extent of TGF-b-induced ERK1/2 activation (Figure 3a). In the MCF10A.ErbB2 cells, activation of ErbB2 with the AP1510 dimerizer Endoglin inhibits cell invasion and metastatic colonization alone did not promote cell migration (Figure 4a, bars 1 in vivo. To validate a functional role for endoglin in an and 5), but downregulation of endoglin in cooperation independent cell system and investigate directly the role

p=0.0255 - TGF-β1

Area of wound closure (mm2) + TGF-β1 -++ -+-+- TGF-β1 shCont shEng5 shCont shEng5

MCF10A MCF10A.ErbB2

shCont shEng5 MCF10A.ErbB2 Figure 4 Endoglin downregulation cooperates with ErbB2 activation to promote cell migration. MCF10A or MCF10A.ErbB2 shCont and shEng5 cell monolayers were starved overnight, scratched and treated±TGF-b1 (MCF10A) or 1 mM AP1510±TGF-b1 (MCF10A.ErbB2) for 21 h. (a) Data showing mean area of wound closure per field of view for three independent experiments±s.e.m. (b) Cell monolayers 21 h post-scratching were fixed, permeabilized and stained for vinculin and Alexa488 anti-mouse Ig (green), Alexa546-phalloidin (red) and DAPI (blue). Scale bar, 25 mm.

Oncogene Endoglin suppresses breast cancer invasion LA Henry et al 1051 of endoglin in metastasis, the endoglin-negative breast tumor cells was significantly correlated with histological cancer cell line MDA-MB-231 (Figure 1a) was trans- grade three (P ¼ 0.0028) and estrogen receptor (ER)- fected with endoglin complementary DNA (cDNA) negative (P ¼ 0.0023) tumors (Supplementary Table 1). (231-ENG) or control vector (231-VEC) (Figure 5a). This cohort of patients was uniformly treated with Ectopic endoglin expression levels in the 231-ENG cells adjuvant anthracycline-based chemotherapy plus ta- were comparable with those in the endogenously moxifen for those with ER-positive tumors. Endoglin expressing MCF10A cells (data not shown). There was expression in the tumor cells was significantly associated no difference in cell proliferation between 231-ENG and with improved metastasis-free (P ¼ 0.0241) and overall 231-VEC cells, and both were modestly growth inhibited (P ¼ 0.0245) survival (Figure 7b). Similarly, within the by TGF-b (Supplementary Figure 4). In agreement with subgroup of more aggressive ER-negative breast can- the data obtained in the MCF10A and MCF10A.ErbB2 cers, expression of endoglin in the tumor cells showed a cells, ectopic expression of endoglin did not impair significant correlation with improved metastasis-free TGF-b-mediated Smad activation as monitored by (P ¼ 0.0175) and overall (P ¼ 0.0251) survival (Supple- nuclear translocation of Smad2/3 (Figure 5b; Supple- mentary Figure 7). These observations demonstrate that mentary Figure 5) or transcription of Smad response endoglin is expressed in the tumor cells of a subset of genes (Figure 5c). As expected in this TGF-b responsive breast carcinomas and that this expression correlates cell line, TGF-b1 treatment significantly increased the with improved patient prognosis. migration (Figure 5d) and invasion (Figure 5e) of the 231-VEC cells as monitored by scratch-wound assays Regulation of endoglin expression by ENG methylation. and Matrigel-coated transwell invasion assays, respec- In our preliminary studies, we identified ENG as a tively, and this induction was blocked with the TGF-b candidate gene for epigenetic regulation in breast cancer receptor inhibitor, SB431542. Interestingly, ectopic cell lines. The ENG gene contains a small CpG island in expression of endoglin reduced both the basal level of exon-1 (Figure 8). To examine whether this island is migration and invasion, and suppressed the ability indeed methylated in breast cancer cell lines, genomic of TGF-b1 to enhance these processes in the MDA- DNA was subject to methylation-specific PCR (MSP) MB-231 cells (Figures 5d and e). (Figure 8a) and bisulfite sequencing (Figure 8b). A high To assess the relevance of these findings to cells level of methylation was detected in all endoglin- in vivo, metastatic colonization assays were undertaken negative breast cancer lines (see Figure 1a). Moreover, by inoculating the cells through the tail vein into treatment with a demethylating agent reversed this immunocompromised mice. 231-ENG cells produced transcriptional inhibition (Figure 8c). To investigate significantly less lung nodules than 231-VEC cells whether ENG methylation is a mechanism for regulating (Figure 6a). Recently, TGF-b activity has been shown expression in tumors, genomic DNA isolated from an to increase the extravasation of breast cancer cells into independent cohort of snap-frozen primary breast the lung parenchyma (Padua et al., 2008). To investigate cancers was subject to MSP analysis (Figure 8d). In this aspect of the metastatic cascade, the 231-VEC and all, 23.3% (30/129) of the tumors were endoglin-positive 231-ENG cell lines were infected with a luciferase- and again this significantly correlated with ER negativity expressing lentivirus to generate the 231-VEC(Luc) and and with high tumor grade (data not shown). MSP 231-ENG(Luc) lines. Following tail vein inoculation, analysis revealed that 17.1% (22/129) of the tumors were In Vivo Imaging Systems (IVIS) bioluminescent imaging methylated and there was a significant inverse correlation was used to monitor the short-term retention of these between the ENG methylation status and endoglin cells in the lung (Figure 6b) At 24 h, IVIS imaging expression (P ¼ 0.0250). The MSP data from the clinical demonstrated a significantly higher proportion of samples was validated by bisulfite sequencing a subset of 231-VEC(Luc) cells compared with 231-ENG(Luc) cells the tumors scored as methylated or unmethylated by MSP retained in the lung, which at later times (5 weeks) gave (Figure 8e). Finally, genomic DNA was also isolated from rise to a higher metastatic burden. fresh frozen normal breast tissue samples and from microdissected epithelial regions from paraffin-embedded Endoglin expression in breast tumors correlates with normal breast tissue samples. MSP analysis confirmed the improved clinical outcome. To relate these findings to absence of ENG methylation in all the normal breast clinical samples, endoglin expression was examined in a samples (Supplementary Figure 8). These data demon- large cohort of invasive breast cancers using the 4G11 strate that ENG methylation is a significant mechanism anti-endoglin antibody (see Supplementary Figure 6 for preventing endoglin expressionandisacquiredinsome antibody validation). In all, 20.6% (45/218) of the breast tumors during cancer progression. samples had endoglin-positive tumor cells (Figure 7a; Supplementary Table 1). As expected, endoglin was also expressed at high levels in the tumor vasculature Discussion providing an internal control for immunohistochemical scoring. Additional immunohistochemistry with an In this study we show that endoglin functions in breast alternative anti-endoglin antibody (SN6h) indepen- cancer cells to inhibit cell migration and invasion in vitro dently validated the specificity of the tumor cell staining and metastatic colonization in vivo. The clinical rele- (data not shown). Analysis of the clinicopathological vance of these findings was verified by surveying a large parameters revealed that endoglin expression in the cohort of invasive breast cancers and demonstrating

Oncogene Endoglin suppresses breast cancer invasion LA Henry et al 1052 that endoglin expression in the tumor cell compartment CpG island methylation in esophageal cell lines and two significantly correlates with improved clinical outcome. esophageal tumors (Wong et al., 2008) and indicates Further, we demonstrated by MSP analysis and bisulfite that this mechanism of ENG gene regulation may be sequencing that endoglin expression in breast cancers utilized in multiple cancer types. Of interest, reduced and cancer cell lines is epigenetically regulated by gene endoglin expression enhances malignant progression in methylation. This latter finding extends previous work a mouse model of skin carcinogenesis (Quintanilla et al., in which MSP analysis was used to demonstrate ENG 2003); however, the mode of endoglin downregulation is

− TGF-β1 SB431542

231- 231- 231-VEC VEC ENG 100 70 Endoglin

50 ERK1/2 37 231-ENG

SNAI2 PAI1 ANGPTL4

Relative expression Relative - + - + - + - + - + - + TGF-β1 VEC ENG VEC ENG VEC ENG

p=0.0095

p=0.0264

- TGF-β1 p=0.0968

+ TGF-β1

--+ + --+ + β 231-VEC 231-ENG closure (mm2) Area of wound TGF- 1 - - + + - - + + SB431542 231-VEC 231-ENG

p=0.0100

p=0.0015

- TGF-β1 p=0.3450

+ TGF-β1 Invaded cells per FOV Invaded --+ + --+ + TGF-β1 231-VEC 231-ENG - - + + - - + + SB431542 231-VEC 231-ENG

Oncogene ed fve.Dt hw ersn envle rmtreidpnetexperiments independent three from values mean represent shown Data view. of fields h ugsraewr one.Dt ssonfrtesvnaiasi ahgop e ie niaemda.( median. indicate lines Red on group. nodules each tumor in visible animals macroscopically seven of the numbers for the shown weeks, is 10 Data After counted. mice. were immunocompromised surface of lung vein the tail the into injected were b treated ahgopa 4had5wes e ieidctsmda.Gop eesgicnl ifrn yteMann–Whitney the by different significantly in were mice the Groups for median. data emission indicates bioluminescent line panels, Right Red weeks. weeks. imaging. 5 bioluminescence 5 and IVIS h and 24 by at h monitored 24 group and each at mice from immunocompromised group mouse of one each vein of imaging tail IVIS the panel, into Left injected were cells 231-ENG(Luc) or otdtaselfitr nsrmfe ei n loe omgaetwrstelwrcabrcnann 0 FCS 10% containing chamber lower the towards migrate to allowed and media serum-free in filters transwell coated ihepyvco 21VC redgi 21EG.EK/ evsa odn oto.( control. loading a as serves ERK1/2 (231-ENG). endoglin or (231-VEC) vector empty with 6h hw r ersnaieiae,1 otsrthn.Dse ie niaetesatn oiino h on.Dt represents Data wound. the of position starting the closure indicate lines wound Dashed of post-scratching. area h 16 mean images, the representative are Shown h. TGF- 16 all in of observed levels was The localization h. 75 3 bar, Smad2/3 Scale nuclear expression. endoglin Strong ectopic Ig. mouse Alexa555-anti iue5 Figure 6 Figure xrsinfo he needn xeiet.( experiments. independent three from expression 1 ± B352 hw r ersnaieiae ftasirtdcls h ubro nae el a unie ycutn three counting by quantified was cells invaded of number The cells. transmigrated of images representative are Shown SB431542. ± gm TGF- ng/ml 5 nolnihbt D-B21cl ugclnzto nan in colonization lung cell MDA-MB-231 inhibits Endoglin nolnihbt D-B21cl irto n nain ( invasion. and migration cell MDA-MB-231 inhibits Endoglin SNAI2 24 hours 5 weeks b 1or20 , PAI1 231-VEC 3-E 231-ENG 231-VEC (Luc) m and ± M ...frtreidpneteprmns ( experiments. independent three for s.e.m. B352fr2h el eefie,preblzdadsandwt md/ A olwdby followed mAb Smad2/3 a with stained and permeabilized fixed, were Cells h. 2 for SB432542 ANGPTL4 231-ENG m .( m. (Luc) c 3-E r21EGclswr tre vrih n treated and overnight starved were cells 231-ENG or 231-VEC ) rncit eeaaye yqC.Dt hw stemean the is shown Data qPCR. by analyzed were transcripts d ofletcl ooaeswr cace n treated and scratched were monolayers cell Confluent ) x10 0.2 0.4 0.6 0.8 1.0 6

Normalised photon flux Normalised photon flux Nodules per lung 6 6 100 200 300 x10 (arbitary units) x10 (arbitary units) Henry invasion LA cancer breast suppresses Endoglin 10 nvivo in 10 15 0 2 4 6 8 0 5 a muoltigo D-B21clssal transfected stably cells MDA-MB-231 of Immunoblotting ) eatssmdl ( model. metastasis tal et 231-VEC e 231-VEC el eesavdoengt ltdot Matrigel- onto plated overnight, starved were Cells ) (Luc) 231-VEC p (Luc) p =0.0262 =0.0302 ± s.e.m. p =0.0068 b -rae el n a o mardby impaired not was and cells 1-treated 231-ENG 231-ENG (Luc) b a 3-E n 3-N el were cells 231-ENG and 231-VEC ) )2 231-ENG 24 hours  (Luc) 10 5 weeks 6 3-E r21EGcells 231-ENG or 231-VEC ± b TGF- ± )2 ± ...o Qrelative RQ of s.e.m.  gm TGF- ng/ml 5 b 10 1 ± 6 231-VEC(Luc) B352for SB431542 U -test. ± b TGF- for 1 Oncogene 1053 Endoglin suppresses breast cancer invasion LA Henry et al 1054

- + + +

Metastasis-free survival Overall survival ENG + ENG +

ENG - ENG -

p=0.0241 p=0.0245 Cumulative survival Cumulative Cumulative survival Cumulative

Months Months Figure 7 Endoglin expression in breast tumor cells correlates with an improved clinical outcome. (a) A tissue microarray comprising 245 invasive breast cancers was stained with anti-endoglin mAb 4G11. Images of representative cores with no endoglin expression in the tumor cells (À) and with endoglin-positive tumor cells ( þ ). Arrows indicate positive clusters of tumor cells; arrowheads indicate endoglin expression in the vasculature. (b) Kaplan–Meier curves of metastasis-free and overall survival in the Grade three cancers scored for expression of endoglin in the tumor cells. The log-rank test was used to determine the differences in survival between groups.

by shedding of the extracellular domain (Perez-Gomez et al., 2010) and transformed mouse keratinocytes et al., 2007). As the mouse Eng gene does not contain a (Perez-Gomez et al., 2007). Consistent with our findings, CpG island this may reflect an alternative mechanism to endoglin had no effect on TGF-b-mediated Smad regulate mouse Eng expression. Finally, in our study it was activation in the prostate cancer cells. In contrast, in also noted that not all endoglin-positive breast cancers mouse keratinocytes endoglin downregulation enhanced showed ENG methylation. The reason for this is that we TGF-b-mediated Smad2/3 phosphorylation. The rea- may be underestimating the level of ENG methylation as sons for these differing effects on TGF-b signaling are our analysis was conducted on whole-tumor lysates, which not known but may reflect differing mechanisms will contain ENG-unmethylated stromal cells and normal regulating the interplay between endoglin and TGF-b epithelial cells (see Materials and methods). In addition, it in squamous carcinomas versus prostate/breast carcino- has been proposed that the methylation of CpG islands mas. Surprisingly, a pro-invasive role for endoglin has only occurs at gene promoters that have already been been reported recently in MDA-MB-231 cells in vitro transcriptionally silenced and that methylation functions (Oxmann et al., 2008), which is in direct contrast to the to maintain the silenced state (Illingworth and Bird, 2009). in vitro and in vivo findings reported here. However, it Hence, in breast tumors, methylation of the ENG CpG should be noted that these authors performed their island may be a secondary event after transcriptional scratch-wound migration and Boyden-chamber invasion silencing of the gene promoter. assays over a 3-h period and therefore may be A key aspect of our studies was to demonstrate that monitoring a short-term cellular phenotype rather than endoglin downregulation alone is insufficient to trans- cell migration/invasion. form (Figure 1c) or enhance the migration (Figure 4) of Our data clearly show that endoglin attenuates the non-tumorigenic MCF10A breast epithelial cells but is pro-migratory and pro-invasive effects of TGF-b1in required to reveal the full migratory/invasive phenotype cells containing active oncogenes (MCF10A.ErbB2, following ErbB2 oncogenic activation (Figure 2b; MDA-MB-231), but that in these cells endoglin does Figure 4). The MDA-MB-231 cell line has activating not modulate TGF-b-induced Smad signaling. In addi- K-Ras and B-Raf mutations (Ikediobi et al., 2006) and tion, it is notable that in the absence of exogenously like the activated MCF10A.ErbB2 cell line is responsive added TGF-b1 endoglin reduces the basal migration to the pro-migratory and pro-invasive effects of TGF-b and invasion of these breast cell lines. Mechanistically (Figure 5). In both the MCF10A.ErbB2 and MDA-MB- this indicates that in epithelial cells endoglin has both 231 cells, expression of endoglin attenuates the pro- TGF-b-dependent and independent functions. The migratory/invasive activities of TGF-b but has no effect current literature suggests two potential mechanisms on TGF-b-mediated Smad2/3 or Smad1/5/8 signaling. by which the TGF-b-independent effects may be Previously it has been reported that endoglin can mediated. First, endoglin shares sequence similarity inhibit TGF-b-mediated migration and invasion of with the other TGF-b co-receptor betaglycan (also prostate-cancer cell lines (Craft et al., 2007; Romero known as TGF-b receptor III). Betaglycan can also

Oncogene Endoglin suppresses breast cancer invasion LA Henry et al 1055 UMUMU MUMUMU M

MCF10 MCF12AMCF10A MCF7 T47D Cal51 -2A

UUUUMM M M UUUUUMM M M M

MDA- MDA- MDA- MDA- SKBR3HCC BT474 U M MB-231 MB-435 MB-453 MB-468 1954 Cont Cont

+190 +430 MSP F MSP R HMEC MCF10A MCF10-2A MCF12A Cal51 SKBR3 BT474

N Endoglin Endoglin p value positive negative MSP 129 p=0.025

expression U 29 78

Normalised mRNA M 1 21 -+ -+5-aza Cal51 MDA-MB-231

+190 +430 MSP F MSP R 59 78 124 137 182 289 303 307 482 Unmethylated by MSP 74 112 131 197 390 458 Methylated by MSP Figure 8 Epigenetic regulation of endoglin expression. (a) Bisulfite-modified genomic DNA from breast cancer cell lines was subject to MSP analysis. U (unmethylated), M (methylated). (b) Sequencing of bisulfite-modified DNA from a minimum of six independent clones per cell line. Individual CpG dinucleotides are displayed as circles. Shading indicates the degree of methylation: white (0%), light gray (1–25%), mid grey (26–50%), dark grey (51–75%) and black (76–100%) methylated. Crossed circles were unscorable. Positions of the MSP primers are shown. Sequence numbers relate to the transcription start according to RefSeq (NM_000118.2); the arrow indicates the ATG translation start codon. (c) Cal51 and MDA-MB-231 cells were treated±5-azacytidine for 6 days and the level of endoglin expression assessed by qPCR. Data shown represent the mean RQ values±s.e.m. of two independent experiments. (d) Bisulfite-modified genomic DNA from primary breast tumors was subject to MSP analysis and scored as U (unmethylated) or M (methylated). Scoring was based on three independent MSP reactions. Matched tumor samples were stained for endoglin and the tumor cells scored as negative (À) or positive ( þ ). P-values were calculated using Fischer’s exact test. (e) Sequencing of bisulfite- modified genomic DNA for a subset of the tumors from panel (d) that has been assigned as unmethylated (upper group) or methylated (lower group) by MSP. Sequence from a minimum of six clones per sample was scored as described in panel (b). inhibit cell migration independently of TGF-b and its endoglin has been shown to interact with zyxin-related receptors, and instead associates directly with b-arrestin protein 1 resulting in a dramatic change in the structure (Mythreye and Blobe, 2009). Here we observed that of the actin cytoskeleton and the localization of zyxin- endoglin expression modulates focal adhesion formation related protein 1 (Sanz-Rodriguez et al., 2004). Second, and turnover. This is consistent with the previous data in addition to TGF-b1, endoglin can bind, in association in endothelial cells where endoglin has been shown to with different TGF-b superfamily receptors, to BMP2, bind directly to zyxin and to sequester zyxin and its BMP7, BMP9, activin A and TGF-b3 (Cheifetz et al., binding partners p130cas and CrkII away from focal 1992; Barbara et al., 1999; Scharpfenecker et al., 2007), adhesion complexes (Conley et al., 2004). In addition, and has been reported to have a functional role in

Oncogene Endoglin suppresses breast cancer invasion LA Henry et al 1056 modulating responses to BMP2 (Ishibashi et al., 2010) on a time-lapse microscope for 16–21 h. For invasion assays, cells and BMP7 (Scherner et al., 2007) in periodontal ligament were starved overnight and 2.5  104 cells were plated onto cells and myoblasts, respectively. Although there are no Matrigel invasion chambers (BD Biosciences) with a lower well published reports of endoglin modulating the response to containing DMEM plus 10% fetal bovine serum±TGF-b1or other TGF-b family ligands in epithelial cells, it will be of SB431542. After 24 h, transmigrated cells on the underside of the interest in the future to determine whether the TGF-b1- filter were fixed in methanol and stained with 1% toluidine blue/ 1% borax. Three fields of view were taken of each insert at  5 independent functions of endoglin expressed on tumor magnification and cells counted using ImageJ software. Each cells may be mediated by binding to alternate ligands and/ condition was performed in triplicate. For 3D acini culture, or other TGF-b superfamily receptors. MCF10A and MCF10A.ErbB2 cells were cultured as previously described (Muthuswamy et al., 2001) on 100% growth factor- reduced Matrigel (BD Biosciences) or a 1:1 mixture of Matrigel Materials and methods and collagen I (Purecol, Leimuiden, The Netherlands), respec- tively. The growth factor-reduced Matrigel contains 1.7 ng/ml Cells and reagents TGF-b. For immunofluorescence staining, cells or acini were MCF10A.ErbB2 cells were obtained from Senthil Muthuswa- fixed with 4% paraformaldehyde for 40 min, permeabilized in my (Princess Margaret Hospital, Toronto, ON, Canada). All 0.1% Triton X-100 for 10 min and stained with primary antibody 1 other cell lines were obtained from ATCC (Manassas, VA, for 1 h at room temperature or overnight at 4 C, followed by AlexaFluor-conjugated secondary antibody. Cells or acini were USA). MDA-MB-231 cells were maintained in Dulbecco’s 0 Modified Eagle Medium (DMEM) with 10% fetal bovine counterstained with Alexa546-phalloidin and 4 ,6-diamidino-2- serum and, when applicable, starved overnight in serum-free phenylindole (DAPI). Fluorescent images were captured sequen- DMEM. MCF10A and MCF10A.ErbB2 cells were cultured, tially in three channels on a Leica Microsystems TCS-SP2 DMEM:F12 (1:1) with 5% donor horse serum, insulin (10 mg/ confocal microscope (see Supplementary Methods for image ml), epidermal growth factor (EGF) (20 ng/ml), hydrocorti- acquisition details). sone (0.5 mg/ml) and cholera toxin (100 ng/ml) as previously described (Muthuswamy et al., 2001) and, when applicable, qPCR starved overnight in assay media containing 1% donor horse qPCR was performed as previously described (Klingbeil et al., serum and no EGF. 2010). Briefly, qPCR primers were purchased as assays- For shRNA downregulation, MCF10A and MCF10A. on-demand from Applied Biosystems (Warrington, UK): ErbB2 cells were infected with Mission lentiviral particles Hs00164438_m1 for human ENG, Hs00161904_m1 for (Sigma, St Louis, MO, USA) according to the manufacturer’s human SNAI2, Hs01101127_m1 for human ANGPTL4, instructions. Individual infections were performed with Hs00167155_m1 for human SERPINE1 and 4310884E for five different human endoglin shRNA constructs (SHVRSC- glyceraldehyde-3-phosphate dehydrogenase. qPCR was per- TRCN0000083138 to SHVRSC-TRCN0000083142) or with a formed on 0.5 ml cDNA using Taqman gene-expression assays non-targeting shCont construct (SHVRSC-TRCN0000098116) on an ABI Prism 7900HT sequence detection system (Applied at a multiplicity of infection of 10. All five shEng constructs Biosystems). Each reaction was performed in triplicate. Data target both large- and short-endoglin isoforms. For ectopic were analyzed with the Applied Biosystems SDS 2.2.1 software endoglin expression, MDA-MB-231 cells were transfected with using glyceraldehyde-3-phosphate dehydrogenase for normal- pcDNA3.1 alone (231-VEC) or pcDNA3.1 containing human ization. Fold change is represented by relative quantification long-endoglin isoform cDNA (231-ENG). 231-ENG cells were units. FACSorted using the anti-endoglin mAb SN6h to enrich for an endoglin-positive population. The qPCR analysis demonstrated Lung colonization assay that expression of endoglin in the 231-ENG cells was equivalent A total of 2  106 cells were injected into the tail vein of 6-week to the expression in the endogenously expressing MCF10A athymic female mice (Ncr-nude). Luciferase bioluminescence and SKBR3 cells. To generate the luciferase-expressing was measured using an IVIS 100 (Caliper, Runcorn, UK). 231-VEC(Luc) and 231-ENG(Luc) lines, 231-VEC and 231- Lungs were excised from the mice at necropsy and surface ENG cells were infected with pGF1-CMV (System Biosciences, nodules counted. Mice were allowed food and water ad libitum. Mountain View, CA, USA) lentiviral particles. All procedures were in accordance with UK Home Office As indicated, cells were treated with 5 ng/ml human TGF-b1 legislation. (R&D Systems, Minneapolis, MN, USA), 20 mM TGF-b- receptor inhibitor SB431542 (Tocris), 5 ng/ml EGF and/or Clinical samples 1 mM AP1510 dimerizer (Ariad Pharmaceuticals, Cambridge, MA, USA). Antibodies were obtained as follows: human The tissue microarray consisted of replicate 0.6 mm cores of endoglin mAb (4G11, Novocastra, Newcastle Upon Tyne, 245 invasive breast carcinomas from patients diagnosed UK; and SN6h, Dako, Ely, Cambridgeshire, UK); a-tubulin and managed at the Royal Marsden Hospital. All patients (Sigma); rabbit polyclonal ERK1/2 (gift from CJ Marshall); were primarily treated with surgery, followed by anthra- collagen IV (Acris, Herford, Germany); laminin V (Chemicon, cycline-based chemotherapy and endocrine therapy for Watford, UK); b-COP (23C, ICR Hybridoma Unit); vinculin patients with ER-positive tumors. Follow-up was for (Sigma); Smad2/3 (BD Biosciences, Oxford, UK), p-Smad2, 0.5–125 months (median ¼ 67 months, mean ¼ 67 months). pSmad3 and pSmad1/5/8 (Cell Signaling, Danvers, MA, Clinicopathological features are shown in Supplementary USA); Smad1 (Chemicon); AlexaFluor secondary antibodies Table 1. The study was approved by the Royal Marsden and phalloidin conjugates (Invitrogen, Paisley, Scotland). NHS Foundation Trust Ethics Committee. An independent series of previously untreated breast cancers from the Tayside Tissue Bank were provided as a tissue microarray Cell assays and as snap-frozen tumors. Genomic DNA was extracted from For migration assays, confluent cells in a 12-well plate were the frozen material using the M48 DNA extraction robot scratched with a P1000 pipette tip and wound closure monitored (Qiagen, Crawley, West Sussex, UK) following histopatholo-

Oncogene Endoglin suppresses breast cancer invasion LA Henry et al 1057 gical review to ensure over 30% representation of neoplastic Statistical analyses cells. The study was approved by the Tayside Tissue Bank All in vitro cell experiments were analyzed with a two-tailed review committee (under delegated authority from the Local Student’s t-test with a CI of 95%. Analysis of the lung Regional Ethics Committee). metastasis experiments was performed using the two-tailed Mann–Whitney U-test. Categorical comparisons were per- formed with a two-tailed Fisher’s exact test. Metastasis-free Immunohistochemistry survival was expressed as the number of months from For immunohistochemical staining, 3 mm thick sections were diagnosis to the occurrence of distant relapse. Overall survival dewaxed in xylene, taken through ethanol (99.7–100%, v/v), (breast cancer-specific survival) was expressed as the number and subjected to high-temperature antigen retrieval (2 min of of months from diagnosis to the occurrence of an event pressure cooking) in target-retrieval citrate buffer (pH 6; (disease-related death). Cumulative survival probabilities were Dako). Slides were allowed to cool for 20 min at room calculated using the Kaplan–Meier method. Differences temperature, incubated with an avidin/biotin block (Vector between survival rates were tested with the Log-Rank test. Laboratories, Peterborough, UK) and then incubated with anti-endoglin mAb 4G11 (1:100) for 1 h at room temperature. Detection was performed with the VectaStain ABC system Conflict of interest (Vector Laboratories). Positive and negative (omission of the primary antibody and substitution with IgG-matched control) The authors declare no conflict of interest. controls were included in each slide run. Expression of endoglin was scored as positive if greater than 10% of the tumor cells were positive. Immunohistochemical analysis Acknowledgements was performed by two observers (LH and JRF) blinded to the results of other immunohistochemical markers and We thank Kay Savage and Suzanne Parry for immunohisto- patients’ outcome. chemical staining, Reshma Shah for help with the methyl- ation analysis, Caterina Marchio for microdissection, Methylation analysis Senthil Muthuswamy for the MCF10A.ErbB2 cells and Genomic DNA (gDNA) was extracted from cell pellets and Ariad Pharmaceuticals (www.ariad.com/regulationkits) for tumor samples using the DNAeasy kit (Qiagen) and subject to the AP1510 reagent. This project was funded by Break- bisulfite modification with the Zymo EZ DNA Methylation kit through Breast Cancer and from NHS funding to the NIHR (Genetix, New Milton, Hampshie, UK). Details of primers, Biomedical Research Centre, a Marie Curie Intra-European MSP conditions and sequence analysis are provided in the Fellowship (PIEF-GA-2008-221083) and Breast Cancer Supplementary Methods. Research Scotland with support from the Tayside Tissue Bank.

References

Aranda V, Haire T, Nolan ME, Calarco JP, Rosenberg AZ, Fawcett Feng XH, Derynck R. (2005). Specificity and versatility in JP et al. (2006). Par6-aPKC uncouples ErbB2 induced disruption of tgf-beta signaling through Smads. Annu Rev Cell Dev Biol 21: polarized epithelial organization from proliferation control. Nat 659–693. Cell Biol 8: 1235–1245. Guerrero-Esteo M, Sanchez-Elsner T, Letamendia A, Bernabeu C. Barbara NP, Wrana JL, Letarte M. (1999). Endoglin is an accessory (2002). Extracellular and cytoplasmic domains of endoglin interact protein that interacts with the signaling receptor complex of with the transforming growth factor-beta receptors I and II. J Biol multiple members of the transforming growth factor-beta super- Chem 277: 29197–29209. family. J Biol Chem 274: 584–594. Ikediobi ON, Davies H, Bignell G, Edkins S, Stevens C, O’Meara S Bernabeu C, Conley BA, Vary CP. (2007). Novel biochemical et al. (2006). Mutation analysis of 24 known cancer genes in the pathways of endoglin in vascular cell physiology. J Cell Biochem NCI-60 cell line set. Mol Cancer Ther 5: 2606–2612. 102: 1375–1388. Illingworth RS, Bird AP. (2009). CpG islands—‘a rough guide’. FEBS Bernabeu C, Lopez-Novoa JM, Quintanilla M. (2009). The emerging Lett 583: 1713–1720. role of TGF-beta superfamily coreceptors in cancer. Biochim Ishibashi O, Ikegame M, Takizawa F, Yoshizawa T, Moksed MA, Biophys Acta 1792: 954–973. Iizawa F et al. (2010). Endoglin is involved in BMP-2-induced Bhowmick NA, Ghiassi M, Bakin A, Aakre M, Lundquist CA, osteogenic differentiation of periodontal ligament cells through a Engel ME et al. (2001). Transforming growth factor-beta1 pathway independent of Smad-1/5/8 phosphorylation. J Cell mediates epithelial to mesenchymal transdifferentiation through a Physiol 222: 465–473. RhoA-dependent mechanism. Mol Biol Cell 12: 27–36. Johnson DW, Berg JN, Baldwin MA, Gallione CJ, Marondel I, Cheifetz S, Bellon T, Cales C, Vera S, Bernabeu C, Massague J et al. Yoon SJ et al. (1996). Mutations in the activin receptor-like kinase 1 (1992). Endoglin is a component of the transforming growth factor- gene in hereditary haemorrhagic telangiectasia type 2. Nat Genet 13: beta receptor system in human endothelial cells. J Biol Chem 267: 189–195. 19027–19030. Klingbeil P, Natrajan R, Everitt G, Vatcheva R, Marchio C, Palacios J Conley BA, Koleva R, Smith JD, Kacer D, Zhang D, Bernabeu C et al. (2010). CD44 is overexpressed in basal-like breast cancers but et al. (2004). Endoglin controls cell migration and composition of is not a driver of 11p13 amplification. Breast Cancer Res Treat 120: focal adhesions: function of the cytosolic domain. J Biol Chem 279: 95–109. 27440–27449. Lebrin F, Goumans MJ, Jonker L, Carvalho RL, Valdimarsdottir G, Craft CS, Romero D, Vary CP, Bergan RC. (2007). Endoglin inhibits Thorikay M et al. (2004). Endoglin promotes endothelial cell prostate cancer motility via activation of the ALK2-Smad1 proliferation and TGF-beta/ALK1 signal transduction. Embo J 23: pathway. Oncogene 26: 7240–7250. 4018–4028. Debnath J, Brugge JS. (2005). Modelling glandular epithelial cancers Lee NY, Blobe GC. (2007). The interaction of endoglin with beta- in three-dimensional cultures. Nat Rev Cancer 5: 675–688. arrestin2 regulates transforming growth factor-beta-mediated ERK

Oncogene Endoglin suppresses breast cancer invasion LA Henry et al 1058 activation and migration in endothelial cells. J Biol Chem 282: Rodriguez-Barbero A, Obreo J, Alvarez-Munoz P, Pandiella A, 21507–21517. Bernabeu C, Lopez-Novoa JM. (2006). Endoglin modulation of Massague J. (2008). TGFbeta in Cancer. Cell 134: 215–230. TGF-beta1-induced collagen synthesis is dependent on ERK1/2 McAllister KA, Grogg KM, Johnson DW, Gallione CJ, Baldwin MA, MAPK activation. Cell Physiol Biochem 18: 135–142. Jackson CE et al. (1994). Endoglin, a TGF-beta binding protein of Romero D, Terzic A, Conley BA, Craft CS, Jovanovic B, Bergan RC endothelial cells, is the gene for hereditary haemorrhagic telangiec- et al. (2010). Endoglin phosphorylation by ALK2 contributes to the tasia type 1. Nat Genet 8: 345–351. regulation of prostate cancer cell migration. Carcinogenesis 31: Moustakas A, Heldin CH. (2005). Non-Smad TGF-beta signals. J Cell 359–366. Sci 118: 3573–3584. Sanz-Rodriguez F, Guerrero-Esteo M, Botella LM, Banville D, Vary Muthuswamy SK, Li D, Lelievre S, Bissell MJ, Brugge JS. (2001). CP, Bernabeu C. (2004). Endoglin regulates cytoskeletal organiza- ErbB2, but not ErbB1, reinitiates proliferation and induces luminal tion through binding to ZRP-1, a member of the Lim family of repopulation in epithelial acini. Nat Cell Biol 3: 785–792. proteins. J Biol Chem 279: 32858–32868. Mythreye K, Blobe GC. (2009). The type III TGF-beta receptor Scharpfenecker M, van Dinther M, Liu Z, van Bezooijen RL, Zhao Q, regulates epithelial and cancer cell migration through beta-arrest- Pukac L et al. (2007). BMP-9 signals via ALK1 and inhibits in2-mediated activation of Cdc42. Proc Natl Acad Sci USA 106: bFGF-induced endothelial cell proliferation and VEGF-stimulated 8221–8226. angiogenesis. J Cell Sci 120: 964–972. Oxmann D, Held-Feindt J, Stark AM, Hattermann K, Yoneda T, Scherner O, Meurer SK, Tihaa L, Gressner AM, Weiskirchen R. (2007). Mentlein R. (2008). Endoglin expression in metastatic breast Endoglin differentially modulates antagonistic transforming growth cancer cells enhances their invasive phenotype. Oncogene 27: factor-beta1 and BMP-7 signaling. JBiolChem282: 13934–13943. 3567–3575. Seton-Rogers SE, Lu Y, Hines LM, Koundinya M, LaBaer J, Ozdamar B, Bose R, Barrios-Rodiles M, Wang HR, Zhang Y, Muthuswamy SK et al. (2004). Cooperation of the ErbB2 receptor Wrana JL. (2005). Regulation of the polarity protein Par6 by and transforming growth factor beta in induction of migration and TGFbeta receptors controls epithelial cell plasticity. Science 307: invasion in mammary epithelial cells. Proc Natl Acad Sci USA 101: 1603–1609. 1257–1262. Padua D, Zhang XH, Wang Q, Nadal C, Gerald WL, Gomis RR et al. Shi Y, Massague J. (2003). Mechanisms of TGF-beta signaling from (2008). TGFbeta primes breast tumors for lung metastasis seeding cell membrane to the nucleus. Cell 113: 685–700. through angiopoietin-like 4. Cell 133: 66–77. ten Dijke P, Goumans MJ, Pardali E. (2008). Endoglin in angiogenesis Perez-Gomez E, Villa-Morales M, Santos J, Fernandez-Piqueras J, and vascular diseases. Angiogenesis 11: 79–89. Gamallo C, Dotor J et al. (2007). A role for endoglin as a suppressor Webb DJ, Parsons JT, Horwitz AF. (2002). Adhesion assembly, of malignancy during mouse skin carcinogenesis. Cancer Res 67: disassembly and turnover in migrating cells — over and over and 10268–10277. over again. Nat Cell Biol 4: E97–100. Quintanilla M, Ramirez JR, Perez-Gomez E, Romero D, Velasco B, Wong VC, Chan PL, Bernabeu C, Law S, Wang LD, Li JL et al. Letarte M et al. (2003). Expression of the TGF-beta coreceptor (2008). Identification of an invasion and tumor-suppressing endoglin in epidermal keratinocytes and its dual role in multistage gene, Endoglin (ENG), silenced by both epigenetic inactivation mouse skin carcinogenesis. Oncogene 22: 5976–5985. and allelic loss in esophageal squamous cell carcinoma. Int J Cancer Roberts AB, Wakefield LM. (2003). The two faces of transforming 123: 2816–2823. growth factor beta in carcinogenesis. Proc Natl Acad Sci USA 100: Zhang YE. (2009). Non-Smad pathways in TGF-beta signaling. Cell 8621–8623. Res 19: 128–139.

Supplementary Information accompanies the paper on the Oncogene website (http://www.nature.com/onc)

Oncogene