Altered Microenvironment Promotes Progression of Preinvasive Breast Cancer: Myoepithelial Expression of αvβ6 Integrin in DCIS Identifies High-risk Patients and Predicts Recurrence

This article is featured in Highlights of This Issue, p. 273

The majority of invasive breast cancers arise from the precursor lesion ductal carcinoma in situ (DCIS; ref. 1), whereby neoplastic cells proliferate and fill the breast ductal tree but are separated from the surrounding stroma by the intact myoepithelial–basement membrane wall of the ducts (Fig. 1A). At some point during progression, the tumor cells penetrate the myoepithelial–basement membrane interface and move into the interstitial matrix becoming an established invasive carcinoma with metastatic potential (Fig. 1A). However, DCIS does not invariably progress to invasive breast cancer: an early study of untreated low-grade DCIS indicated that 39% (11 of 28 patients) progressed to invasive breast cancer after a median follow-up of 31 years (2). High-grade DCIS is considered to progress more rapidly to invasive disease, but overall it is estimated that between one-third and half of DCIS diagnosed will progress to invasive breast cancer during a woman's lifetime (3). The frequency of DCIS has increased dramatically with the widespread use of mammographic screening, accounting now for 1 in 4 breast cancer diagnoses (3), contributing to recent concerns surrounding overdiagnosis and overtreatment within the breast screening setting (4, 5). Indeed, the need to understand more about the biologic and molecular mechanisms underlying progression of DCIS to invasive disease and to generate robust risk stratification methods for patients with DCIS was highlighted in a recent NIH State-of-the-Science conference on DCIS (3).

A number of studies have aimed to identify markers or signatures that may predict progression of DCIS, most of which have focused on comparing DCIS tumor cells with its invasive counterpart. These studies generally show no specific changes associated with transition to invasion (6–8) and suggest that DCIS is genetically as advanced as invasive breast cancer (9). Other studies have shown that DCIS and invasive breast cancer largely cluster according to intrinsic molecular subtypes (10–12), although one gene array–based study identified a subset of DCIS cases with an expression profile more similar to invasive cancer and found that the genes characterizing this subset were related to reorganization of the microenvironment (13). The microenvironment of DCIS is complex, comprising the myoepithelial population as well as the stromal compartment. Myoepithelial cells have been shown to be potent tumor suppressors (14) exerting their effects through autocrine and paracrine pathways (15–18). However, several recent studies (19, 20) have suggested that DCIS-associated myoepithelial cells may lose their tumor suppressor ability and actually promote breast cancer progression (21).

Here, we show for the first time upregulation of integrin αvβ6 in DCIS-associated myoepithelial cells. This change results in a switch in the myoepithelial cells from tumor suppressive to tumor promoting in a TGFβ- and MMP9-dependent manner. Analysis of 532 DCIS cases shows that αvβ6 is more frequently expressed in DCIS associated with invasion and in a separate cohort of DCIS with long-term follow-up αvβ6 expression significantly predicts both progression and recurrence. We suggest that upregulation of αvβ6 may be a key marker of transition of DCIS to invasive disease, is predictive of prognosis, and that TGFβ and MMP9 could be targeted in early breast cancer for therapeutic benefit.

Clinical samples were obtained from surgical specimens from patients undergoing breast surgery between 2004 and 2009 at Barts Health NHS Trust London. Further details are provided in the Supplementary Methods.

The study was conducted following patient consent and approval from the local research ethics committee (ref: 05/Q0403/199 and 09/H075/39). Further analysis was conducted on a series of cases derived from the UK-DCIS study (22, 23) to determine relationship between αvβ6 and progression/recurrence. The UK/ANZ DCIS trial was a randomized 2 × 2 factorial trial recruiting women between May 1990 and August 1998. Inclusion criteria were diagnosis of DCIS, undergoing breast conserving surgery, and with confirmed clear margins. Patients were randomized to receive radiotherapy, tamoxifen, or both (22). Clinicopathologic details are provided in Supplementary Methods.

Sections were dewaxed in xylene and antigen retrieved in pepsin solution followed by incubation with 0.5 mg/mL αvβ6 mouse monoclonal antibody (6.2G2 Stromedix; refs. 24, 25), MMP9 (My Biosource, MBS 240189), p63 (Dako, M7247), SMA (Dako, M0851), or CK5/6 (Dako, M7237) overnight. Sections subsequently were incubated with rabbit anti-mouse biotinylated F(ab′)2 for 30 minutes, developed using ABC reagent and superDAB (Dako), and then counterstained with hematoxylin. Details of scoring method are given in Supplementary Methods.

T47D, MCF-7, MDA-MB-231, and HFFF2 cells, obtained from the American Type Culture Collection, were maintained in Dulbeccos' Modified Eagle's Media (DMEM) with 10% fetal calf serum (FCS) and 2 mmol/L glutamine, AM12 cells transfected with pBABE puro β6 were maintained in DMEM with 10% FCS and 2 mmol/L glutamine with puromycin (1 μg/mL, Sigma). Mink lung epithelial cells (MLEC; a kind gift from D. Rifkin, New York University) were maintained in DMEM 10% FCS and 2 mmol/L glutamine with puromycin (1 μg/mL). Primary myoepithelial cells were isolated as described previously (26). An immortalized myoepithelial cell line (1089) (provided by M. O'Hare and P. Jat, Institute of Neurology, UCL) was generated from normal myoepithelial cells isolated from reduction mammoplasty material (27). Cells were sorted using integrin β4–labeled sheep anti-mouse magnetic beads (Invitrogen), to purify the myoepithelial cell population (1089) and maintained in Ham's/F12 media with 10% FCS, 2 mmol/L glutamine, hydrocortisone (1 μg/mL), EGF (10 ng/mL), and insulin (5 μg/mL). All cell lines were verified by STR profiling (LGC Standards, tracking number 710081047). The 1089 cell line showed no similarity to any other cell line in the ATCC database (LGC Standards, tracking number U10010392). Frozen stocks were banked, and cell lines were grown for no longer than 6 months then discarded and fresh cells were then recovered from liquid nitrogen.

Primary myoepithelial cells were recovered from liquid nitrogen and plated onto a 6-well plate precoated with collagen (10 mg/mL) at a density of 2 × 10⁴ per well. Cells were transfected 72 hours later using the standard JetPrime protocol (PolyPlus) and 2 μg of integrin β6 pcDNA1 neo (Addgene plasmid 13580; ref. 28) or pcDNA1 empty vector. After 24 hours, media were changed for serum-free Ham's/F12 and conditioned for 48 hours.

Conditioned medium containing retrovirus from AM12/pBABE-β6 or AM12/pBABE-Puro cells, was incubated with 1089 cells (29). β6-positive 1089 and puromycin control cells were selected using puromycin (1 μg/mL, Sigma). Control, puromycin-resistant 1089 cells were designated normal (N-) 1089; β6-transduced cells were further enriched using integrin αvβ6–labeled anti-mouse IgG magnetic beads (Invitrogen) and designated β6-1089.

A total of 2 × 10⁴ cells were plated onto glass coverslips and grown for 48 hours before fixation with 4% paraformaldehyde. Cells were permeabilized with 0.1% Triton X-100 in PBS before staining. All antibodies were diluted in 20% normal rabbit serum in PBS, except for the αvβ6 antibody (MAB2077Z, Clone 10D5; Millipore), which was diluted in 0.1% bovine serum albumin (BSA) in DMEM. Slides then were stained with a secondary rabbit anti-mouse 488 (Invitrogen, A11059) and counterstained with 4′,6-diamidino-2-phenylindole (DAPI). Images were viewed on a Zeiss LSM 510 Meta microscope.

A total of 3 × 10⁵ cells were incubated with primary antibody diluted in 0.1% BSA in PBS, except for the anti-αvβ6 antibody (10D5), which was diluted in 0.1% BSA in DMEM. Cells subsequently were stained with secondary rabbit anti-mouse fluorescein isothiocyanate (FITC)-conjugated antibody (Dako, F0313) before reading on a Flow Cytometer (BD LSR II).

Cells were trypsinized, centrifuged at 1,200 rpm, re-suspended in radioimmunoprecipitation (RIPA) lysis buffer (Millipore) with “Complete” protease inhibitors (Roche). Protein concentrations were determined using the Dc Protein Assay (Bio-Rad), and 50 μg of protein was loaded on to an 8% Tris/Glycine gel, electrophoresed and transferred on to nitrocellulose membranes that were blotted with αvβ6 antibody (1:1,000, SC-6632, Santa Cruz), β-actin (1:10,000, Santa Cruz, SC47778), or HSC70 (1:10,000, SC-53935, Santa Cruz,) and donkey anti-goat horseradish peroxidase (HRP)-conjugated secondary (1:2,000, Dako) or goat anti-mouse HRP-conjugated secondary (1:2,000, Dako) diluted in 5% milk in TBS/T.

MLECs transfected with a PAI-1 promoter fused to a luciferase reporter gene were used to determine the amount of active TGFβ (30). Briefly, 4 × 10⁴ MLEC's were seeded in a 96-well plate and incubated overnight then serum starved for 4 hours before 4 × 10⁴ test cells were plated on top of the MLECs and cocultured for 16 hours. Medium was removed, lysis buffer (Promega) added, and the plate incubated at −80°C for 1 hour before thawing at room temperature. Cells were scraped to ensure lysis and the luciferase substrate added according to manufacturer's instructions (Promega). Luminescence was recorded on a Victor microplate reader.

Latency-associated peptide (LAP, Sigma) was diluted to 0.5 μg/mL, aliquots were added to the wells of a 96-well plate and incubated for 1 hour at room temperature. A total of 2 × 10⁴ cells were incubated with IgG isotype control antibody or αvβ6-blocking antibody (10D5) at 10 μg/mL in serum-free media (SFM) for 20 minutes at 4°C before plating onto LAP. Cells were incubated at 37°C for 45 minutes before fixing with 30% methanol and staining with 0.1% crystal violet. Stained cells were dissolved with 30% acetic acid and absorbance read at 550 nm.

Twenty-four-well Transwells (8-μm pore, BD Bioscience) were coated on the underside with 0.5 μg/mL LAP, for 1 hour at room temperature, and then placed in a 24-well plate with DMEM, lacking serum, in the lower chamber. A total of 3 × 10⁴ cells were incubated with either an IgG isotype antibody or an αvβ6-blocking antibody (10D5) diluted in SFM (20 minutes at 4°C). Cells then were placed in the upper chamber and incubated for 24 hours. Migrating cells were quantified by applying trypsin to the underside of the Transwell and counting the detached cells on a CASY counter.

Conditioned medium was generated from N-1089 or β6-1089 by culturing in serum-free Ham's/F12 for 48 hours and was stored at −80°C. The upper surface of 24-well Transwell was coated with 70 μL growth factor–reduced Matrigel (diluted 1:3 in Ham's/F12 without serum) and incubated for 1 hour at room temperature. Five hundred microliters conditioned medium from either N-1089 or β6-1089 was then placed in the lower well. For inhibitor experiments, either dimethyl sulfoxide (DMSO) or MMP9 inhibitor (Millipore, 50 nmol/L, 444278) was added to conditioned media. Two hundred microliters of 3 × 10⁴ tumor cells was added to each well and incubated for 24 to 48 hours. Invading cells were quantified by applying trypsin to the bottom of the Transwell and counting the detached cells on a CASY counter.

The skin organotypic method developed by Fusenig and colleagues (31), was adapted by Nystrom and colleagues (32), and further modified. Matrigel:collagen gels were prepared as follows: 3.5 volumes of collagen type 1 (BD Bioscience) and 3.5 volumes of Matrigel were mixed on ice with 1 volume 10× DMEM, 1 volume FCS, and 1 volume DMEM + 10% FCS in which 5 × 10⁵ cells per mL HFFF2 fibroblasts had been suspended. Then 1 mL of this solution was placed into wells of a 24-well plate and allowed to polymerize for 30 minutes at 37°C. One milliliter of DMEM + 10% FCS and glutamine was added to each well and incubated overnight.

The next day 2.5 × 10⁴/500 μL (DMEM + 10% FCS) N-1089 or β6-1089 cells were plated on top of the gel and allowed to settle for 6 hours at 37°C; then 2.5 × 10⁴/500 μL MDA-MB-231 cells were plated on each gel. Gels were fixed in formal saline and placed into paraffin blocks, 5-μm sections were cut and analyzed by immunohistochemical staining. Invasion was determined by taking images of 3 random fields of view from triplicate gels and analyzing the number of invasive particles, the percentage area of the field the particles occupied, and the depth the particles penetrated the gel (determined by 3 measurements per field of view) using ImageJ (NIH). The product of the 3 values gave an overall “invasive index” (32).

Female 6-week-old CB17 SCID mice were purchased from Harlan and were maintained in IVC cages in a pathogen-free, temperature-controlled, 12-hour light and dark cycle environment. All animals were used in accordance with United Kingdom Home Office regulations and ethical approval by the local ethics committee.

Tumor volumes based on caliper measurements were calculated by the modified ellipsoidal formula: tumor volume = 1/2(length × width²) (33). Further information on the mouse experiments are in the Supplementary Methods.

Cells (1 × 10⁵) were preincubated with IgG antibody, β6-blocking antibody (10 μg/mL, 10D5), or TGFβRII-blocking antibody (10 μg/mL, AF-241-NA, R&D Systems) in SFM before they were plated on coverslips coated with fibronectin (1 μg/mL) or LAP (0.5 μg/mL). Media were collected after 24 hours, and MMP9 activity was analyzed using SDS-PAGE substrate gels. Gelatin (Bloom 300, Sigma) was added to a 10% acrylamide separating gel at a final concentration of 1 mg/mL. The intensity of the bands was measured by densitometric analysis. Direct comparisons between separate gels were not made, as the intensity of background staining was variable. Experiments were repeated a minimum of 3 times in triplicate. Further information on the zymogram protocol is in the Supplementary Methods.

RNA was isolated from N-1089 and β6-1089 cell pellets using a kit (Qiagen, 74104). Analysis was conducted on RNA from 3 separate experiments. TaqMan real-time PCR for MMP9 (Invitrogen, Hs00234579_m1) was carried out as previously described (34).

We determined statistical significance of the differences between experimental groups by the 2-tailed Student t test with PRISM software (GraphPad Software). P < 0.05 was considered significant. The relationship between staining patterns and DCIS grade, between MMP9 and αvβ6, and between αvβ6 and outcome was analyzed using Pearson χ² test using SPSS software (SPSS Inc.) with P < 0.05 considered as significant. Cases from UK-DCIS were matched on treatment received, age and time of follow-up and analyzed by logistical regression, adjusted and unadjusted for tumor grade and size. Animal studies were assessed for statistical significance between 2 medians by Mann–Whitney U test.

A total of 583 clinical breast samples comprising normal breast tissue, epithelial hyperplasia of usual type, “pure” DCIS and DCIS with invasion were analyzed for myoepithelial-associated expression of αvβ6. This revealed no significant staining for αvβ6 integrin in normal or hyperplastic cases (Fig. 1B), whereas 69% of high-grade and 52% of non–high-grade DCIS cases without invasion showed myoepithelial staining for αvβ6 (Table 1), with a significantly higher frequency of positivity in high-grade cases (P < 0.001). The frequency of myoepithelial cell αvβ6 positivity in DCIS with associated invasion is significantly higher than in pure DCIS (P < 0.001) with 96% of high-grade and 87% of non–high-grade DCIS exhibiting staining.

To assess the predictive value of αvβ6, we examined expression in a unique series of DCIS with long-term follow-up, derived from the UK/ANZ DCIS study, the endpoint assessed was the appearance of any new breast cancer, either DCIS or invasive cancer, as the numbers are insufficient to assess progression to invasion only. Cases and controls were matched for treatment protocol. The median follow-up was 114 months (22, 23). Staining was evaluated in 52 case–control pairs (details of cases shown in Supplementary Table S1), with cases scored using an H score to take account of the extent and intensity of myoepithelial-associated αvβ6 positivity, according to the criteria described by Allred and colleagues (35). Analysis showed a significant association between extent of αvβ6 expression (evaluated as an H score; ref. 35) and disease recurrence or progression (P = 0.006). This association was maintained following adjustment for DCIS grade and size (P = 0.02, H score; Table 2). All cases were confirmed as having clear margins as part of the inclusion criteria to the trial. We found that those cases positive for αvβ6 recurred more quickly than those lacking the integrin, with the αvβ6-positive cases showing a median time to recurrence of 2.3 years and those negative for αvβ6 having a median of 11.4 years to recurrence.

We obtained an immortalized cell line, derived from normal breast tissue (27), which was positively selected for expression of integrin α6β4 (a marker of myoepithelial cells; ref. 36). This “pure” α6β4-expressing cell population, (N-1089), was characterized by immunofluorescent staining for key myoepithelial cell markers (Fig. 1C, top) following introduction of the puromycin resistance gene. Integrin β4 (36), P-cadherin (P-Cad; ref. 37), cytokeratin 17 (CK17; ref. 38), and desmoglein 3 (Dsg3; ref. 39) expression is detected exclusively in myoepithelial cells in the normal breast and maintained in the N-1089 cells. Low levels of αvβ6 expression by N-1089 cells likely were induced because of culture of the cells on plastic (29). We then retrovirally transduced the N-1089 cell line with β6 integrin subunit cDNA, and the cell line was re-examined to confirm maintenance of myoepithelial-restricted markers in the correct subcellular localizations (Fig. 1C, middle). A considerable increase in the amount of β6 integrin by flow cytometry, consistent with successful transduction, was observed but no differences between N-1089 and β6-1089 with regard to other integrin levels or myoepithelial markers were detected (Fig. 1D and data not shown). Expression of αvβ6 was also confirmed by Western blotting (Fig. 1E). Primary myoepithelial cells were isolated from normal reduction mammaplasty material (n = 2) and purified as previously described (26). Cells were plated onto collagen-coated glass coverslips and shown by immunofluorescence to maintain expression of myoepithelial markers (Fig. 1C, bottom). αvβ6-overexpressing primary myoepithelial cells were generated (see Materials and Methods), and further characterization indicated significant increase in β6 integrin with no change in other myoepithelial markers (Supplementary Fig. S1 and data not shown).

Expressed αvβ6 in the β6-1089 cell line was functional as shown by several in vitro assays. Integrin αvβ6 binds to the LAP, which is part of the trimeric complex that forms secreted TGF-β (40). Therefore, β6-1089 and N-1089 cells were plated on to LAP, poly-d-lysine (PDL), or BSA, to determine background adhesion. Figure 2A shows β6-1089 cells adhered preferentially to LAP compared with N-1089 cells (P = 0.003); this adhesion was reduced significantly after incubation of cells with an αvβ6-blocking antibody (10D5) before plating onto matrix (P = 0.008). Figure 2B indicates β6-1089 cells migrate significantly more to LAP than do N-1089 cells (P = 0.02) in modified Boyden chamber (Transwell) assays (41). Migration of β6-1089 cells to LAP was ablated significantly (P = 0.04) in the presence of blocking antibody whereas N-1089 cells were unaffected, showing that migration to LAP was mediated exclusively by αvβ6.

Binding of αvβ6 to LAP causes TGF-β activation (40); therefore, N-1089 or β6-1089 cells were cocultured with MLECs stably transfected with a TGF-β responsive plasminogen activator inhibitor 1 (PAI-1) promoter fused to a luciferase gene. Coculture with the β6-1089 cells caused a significant increase in luciferase activity (P = 0.02), compared with N-1089 cells, which was abolished with a monoclonal antibody to αvβ6 (Fig. 2C). No TGF-β activation was detected when MLEC cells were cultured with media conditioned on β6-1089 or N-1089 cells (data not shown), indicating that direct cell–cell contact is required. Adhesion, migration, and reporter assay results were replicated using RNA interference to αvβ6 (Supplementary Fig. S2a–S2d).

Normal myoepithelial cells exert a tumor suppressor function (14, 15, 17, 19, 20, 42) but cancer-derived myoepithelial cells may lose this capacity (21). We examined whether myoepithelial cells expressing αvβ6 exhibited compromised tumor suppressor activity compared with normal (αvβ6-negative) myoepithelial cells. Breast cancer cell lines were placed in Transwells coated with Matrigel, and conditioned medium (CM) from primary myoepithelial cells transfected with pcDNA1 or pcDNA1-β6 was placed in the lower well. After varying incubation times, invading cells were quantified (see Materials and Methods). MDA-MB-231 and MCF7 tumor cells showed a significant increase in invasion (P < 0.0001) in the presence of CM from primary myoepithelial cells expressing αvβ6 (Fig. 3A). The same assays were conducted with MDA-MB-231, MCF7, and T47D cell lines, using N-1089 or β6-1089 CM in the lower well. Invasion of MCF-7, T47D, and MDA MB 231 cells was increased significantly (P = 0.0007, P = 0.02, and P = 0.006, respectively) when exposed to CM from β6-1089 cells (Fig. 3B) compared with CM from N-1089 cells.

We have modified the organotypic skin model, developed by Fusenig and colleagues (31), to establish a more physiologically relevant model of breast cancer invasion. Fibroblasts (HFFF2) were embedded in a 1:1 collagen:Matrigel mixture onto which the N-1089 or β6-1089 cells were plated with MDA-MB-231 cells layered on top. After 10 days, culture gels were processed and embedded in paraffin and sections stained with hematoxylin and eosin (H&E; Fig. 3C), p63 (to label myoepithelial cells), pan-Cytokeratin (to show myoepithelial cells and MDA-MB-231), and αvβ6 (Supplementary Fig. S3 and data not shown). The “invasion index” was determined as described in Materials and Methods. Gel analysis showed a significant increase in the invasion index of MDA-MB-231 cells when cocultured with β6-1089 compared with N-1089 cells (Fig. 3D, P < 0.0001).

The mammary fat pads of female severe combined immunodeficient mice (SCID) mice were injected with 3 million cells comprising a 50% mixture of the breast cancer cell line MDA-MB-231 and either N-1089 or β6-1089. Tumor measurements were taken 3 times a week and the subcutaneous volume determined. It was observed that the MDA-MB-231 cells mixed with β6-1089 cells formed established tumors more rapidly than did those tumor cells injected with N-1089 cells (data not shown). Those tumors produced in the presence of β6-1089 were significantly larger in volume (β6-1089 median; 174 mm³) than those produced when mixed with N-1089 (N-1089 median; 75 mm³, P = 0.018; Fig. 3E and F). These data indicate that a distinct growth advantage was conferred upon the MDA-MB-231 mixed with β6-1089; possibly due to increased MMP secretion allowing more unrestricted growth and invasion.

Previously, we showed that αvβ6 expression by keratinocytes results in upregulation of MMP9 and a concomitant increase in invasive capacity (29). Normal myoepithelial cells have been shown to downregulate MMP9 in breast cancer cell lines (17). Therefore, we investigated whether αvβ6 expression by myoepithelial cells modulates MMP9 levels. Conditioned media from N-1089, β6-1089, MDA-MB-231, and MCF-7 cells were generated over 24 hours, after which medium was collected and run on a gelatin zymogram. A clear band at 92 kDa, corresponding to MMP9, was identified in the CM from β6-1089 cells but not in CM from breast cancer cells (Fig. 4A). This was confirmed by quantitative PCR (Fig. 4B), which showed significant upregulation of MMP9 RNA in β6-1089 cells, versus their normal counterparts (P = 0.01).

The β6-1089 cells were preincubated with a β6-blocking antibody (10D5) or IgG isotype control before plating and incubating for 24 hours. The resultant CM was collected and run on a gelatin zymogram. A clear band, corresponding with the recombinant MMP9, was detected from β6-1089 treated with IgG antibody, but not with β6-blocking antibody. To determine whether αvβ6 upregulated MMP9 through TGFβ activation, the β6-1089 cells were preincubated with a TGFβRII-blocking antibody before plating. The CM was run on a gelatin zymogram and a clear band corresponding with the recombinant MMP9 was detected from the β6-1089 cells treated with IgG antibody but not TGFβRII antibody–treated cells (Fig. 4C).

Transwell invasion assays were conducted using N-1089 or β6-1089 CM in the presence of an MMP9 inhibitor (50 nmol/L, Millipore, 444278) or DMSO control (Fig. 4D). MDA-MB-231 cell invasion was decreased significantly, to control levels, in the presence of β6-1089 CM plus MMP9 inhibitor (P = 0.005), whereas the presence of the MMP9 inhibitor had no significant effect on invasion mediated by N-1089 CM. Thus, upregulation of myoepithelial cell αvβ6 expression caused a TGFβ-dependent increase in MMP9 secretion and a concomitant increase in tumor cell invasion.

The clinical relevance of these findings was verified by staining serial sections of pure DCIS and DCIS with invasion (n = 108) for MMP9 and αvβ6. Staining for MMP9 was heterogeneous, with some tumor cell cytoplasmic staining as well as stromal and inflammatory cell positivity, whereas there frequently was enhanced staining at the myoepithelial–basement membrane interface (Fig. 4E). Strong staining for αvβ6 was restricted to myoepithelial cells (Fig. 4E). Overall, 64% (n = 101) of DCIS ducts exhibited myoepithelial positivity for MMP9 whereas 59% (n = 93) showed αvβ6 staining, with a highly significant association between expression of both molecules (P < 0.0001; Fig. 4F and Table 3).

Breast cancer develops through defined clinical and pathologic stages, from hyperplasia to DCIS and finally invasive disease and metastasis. Since the introduction of mammographic screening programs, the incidence of detection of DCIS has increased 7-fold (3), and recent reviews suggest that only one-third to half of these cases will progress to invasive breast cancer during a woman's lifetime (3, 4). Current management strategies treat all DCIS lesions as potentially invasive as there are no robust markers to distinguish between those cases with the potential to progress and those unlikely to. In DCIS, neoplastic cells fill and distend breast ducts but are separated from the surrounding stroma by an intact myoepithelial–basement membrane interface. Myoepithelial cells have been shown to express several tumor suppressor proteins (e.g., p63, p73, 14-3-3-σ, Maspin; refs. 43–45), consistent with their established tumor suppressor function (14). However, it has been shown that tumor-derived myoepithelial cells differ from normal and lose the ability to polarize luminal epithelial cells (20). Further studies have shown differences in gene expression between normal and DCIS-associated myoepithelial cells such as increased lysyl oxidase (LOX; ref. 46) and neuropillin 1 (47). Allinen and colleagues examined the microenvironment of normal and tumor breast tissue and found that myoepithelial cells exhibited the most numerous changes in gene expression of all the microenvironmental cell types in DCIS (19), although the predictive and functional relevance of such changes has not been established.

In this study, we show that myoepithelial cells exhibit functionally relevant changes in their phenotype and behavior in a subset of DCIS cases, leading to acquisition of invasion-promoting properties. We have shown upregulation of integrin αvβ6 by myoepithelial cells in 52% to 69% of non–high-grade and high-grade DCIS, respectively, but it is present in almost 100% of DCIS associated with invasive disease. These data suggest that αvβ6 might be a marker of DCIS that is more likely to progress to invasive disease. It already has been established that high-grade DCIS progresses to invasion and local recurrence more quickly than low-grade (48, 49). To more directly address whether expression of αvβ6 can predict recurrence and progression, we assessed expression in a cohort of DCIS cases treated with local excision, confirmed margins free of disease and with long-term follow-up as part of the UK/ANZ DCIS trial (22), and showed a significant correlation between αvβ6 expression in myoepithelial cells and recurrence of breast cancer either as in situ or invasive disease, independent of patient age, disease grade, or extent. Furthermore, those with αvβ6 positivity developed recurrence more quickly than those lacking αvβ6, with median time to recurrence of 2.3 versus 11.4 years, respectively. This work supports the proposal that expression of αvβ6 in the DCIS-associated myoepithelial cells is a marker of DCIS cases more likely to progress to invasion and recurrence.

The integrin αvβ6 is restricted to epithelial cells and is usually expressed only in embryogenesis and wound healing in adult humans. αvβ6 primarily binds to the RGD motif of LAP of pro-TGFβ leading to activation of the protein. The downstream consequence of αvβ6 activation has so far been shown to lead to activation of MMPs (e.g., MMP3 and MMP9; refs. 50–52). Activation of MMPs plays an important role in the remodeling of the ECM and de-regulation of these proteins in cancer leads to destruction of basement membrane that would otherwise constrain tumor cells from invading.

Previous studies have identified αvβ6 expression in invasive carcinomas, and these indicate that expression is associated with aggressive tumor behavior and reduced survival (53–55). However, this is the first study to describe upregulation of αvβ6 in the nontumorigenic microenvironment of a tumor.

We used both primary human myoepithelial cells and a myoepithelial cell line (N-1089) engineered to overexpress αvβ6 as a model of DCIS-associated myoepithelial cells. The functionality of αvβ6 was confirmed by enhanced attachment and migration to the αvβ6 ligand LAP, which is blocked by an αvβ6-blocking antibody (10D5) and siRNA to αvβ6. Furthermore, we show that αvβ6 activates a PAI-1 TGFβ luciferase reporter, which is also blocked by 10D5 and siRNA.

Using our model system in Transwell invasion assays of several breast cancer cell lines, we have shown that CM from the β6-1089 cell line promotes invasion of MDA-MB-231 (basal-like), T47D [estrogen receptor (ER⁻)] and MCF-7 (ER⁺) when compared with the model of normal myoepithelial cells (N-1089), with a similar activity shown by primary myoepithelial cells. The breast cancer cell lines studied represent a cross-section of different breast cancer cell types indicating the broad relevance of the effect of αvβ6-positive myoepithelial cells in breast cancer. We further show enhanced invasion in the presence of αvβ6-positive myoepithelial cells in a modified organotypic assay used previously in our laboratory (32). In this assay, MDA-MB-231 cells exhibited significantly increased invasion into a Matrigel:collagen:fibroblast mixture when cocultured with β6-1089 compared with N-1089 cells. To show that this effect was not simply an in vitro phenomenon, we injected the mammary fat pad of 20 SCID mice with a 50:50 mixture of MDA-MB-231 and either β6-1089 or N-1089. The tumors that formed in the β6-1089 cohort grew with faster kinetics and the median tumor size was significantly increased compared with the N-1089 group after 13 weeks (Fig. 3F and e, respectively).

As αvβ6 activates TGF-β in a localized manner and does not release it from the matrix, and therefore we postulated that there must be a secondary active agent(s) mediating the effect. It has already been shown that αvβ6 is able to activate MMP9 in oral squamous cell carcinoma to drive increased invasion. In keeping with this, we show that β6-1089 cells produce more MMP9 than the N-1089 (Fig. 4A) and that this could be blocked both by αvβ6-blocking antibody 10D5 and by inhibiting TGFBRII with a blocking antibody. We further show that an MMP9 inhibitor can significantly reduce the invasion of MDA-MB-231 cells in presence of β6-1089 CM. These data show for the first time a paracrine protumor effect from αvβ6-expressing myoepithelial cells on breast cancer cells and that this effect is dependent on αvβ6-TGF-β–mediated activation of MMP9.

TGFβ is the archetypal pleiotropic cytokine, which has been shown to be both a tumor suppressor and a tumor promoter (56). The role of TGFβ as a tumor suppressor is clearly evidenced by the frequency with which various components of the pathway are lost, TGFBR2, TGFBR1 and SMAD4 are commonly mutated by LOH or allelic loss (57). However, high expression of TGFβ also correlates with tumor progression on non–small cell lung carcinoma (NSCLC), colorectal, prostate, and gastric cancer. Intense staining has been seen and positively correlated with metastasis in breast, prostate, and colorectal cancer (58). Breast cancer along with skin cancer and gliomas exhibit fewer TGFβ pathway mutations and may use TGFβ for a growth advantage. Indeed low-level expression of TGFβRII in ER⁻ breast cancer is associated with better outcome (59), whereas overexpression of TGFβ is associated with increased incidence of metastasis (60). A bioinformatic approach by Padua and colleagues to examine breast cancer for a TGFβ-responsiveness signature (TBRS) indicated that as many as 40% of breast cancers are capable of responding to TGFβ signals (61). Here, we show that TGFβ signaling can also influence earlier stage disease.

We have shown that a subset of pure DCIS and almost all DCIS with invasive disease exhibit myoepithelial cells that express αvβ6 integrin and show that this change results in a switch in the myoepithelial cells from tumor-suppressive to tumor-promoting activity via TGFβ and MMP9 signaling. Our data indicate a significant association between αvβ6 expression and disease recurrence and shows that αvβ6-positive DCIS recurs more rapidly than αvβ6-negative disease. Further studies are needed to establish whether αvβ6 could be used in the clinical setting to stratify patient care and whether TGFβ and MMP9 could be targeted in early breast cancer for therapeutic benefit. This may represent a key stage in the evolution of breast cancer that could be used in a predictive and prognostic setting, allowing more tailored management of women with DCIS and may offer opportunities for therapeutic intervention.

J.F. Marshall is a Consultant/Advisory Board member of GSK. No potential conflicts of interest were disclosed by the other authors.

Conception and design: M.D. Allen, J.F. Marshall, I.R. Hart, J.L. Jones

Development of methodology: M.D. Allen, G.J. Thomas, M.M. Dawoud, S. Vallath, S.J. Payne, S. Dickinson, D.R. Edwards, J.L. Jones

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M.D. Allen, G.J. Thomas, S.E. Clark, M.M. Dawoud, S. Vallath, S.J. Payne, J.J. Gomm, S.A. Dreger, C.J. Pennington, J. Cuzick, J.L. Jones

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M.D. Allen, M.M. Dawoud, S.A. Dreger, S. Dickinson, I. Sestak, J. Cuzick, J.L. Jones

Writing, review, and/or revision of the manuscript: M.D. Allen, S.J. Payne, S.A. Dreger, D.R. Edwards, I. Sestak, J. Cuzick, J.F. Marshall, I.R. Hart, J.L. Jones

Study supervision: J.F. Marshall, I.R. Hart, J.L. Jones

Scored IHC: M.D. Allen, J.L. Jones

Assisted with gelatin zymography: G.J. Thomas

Collected the cases: S.E. Clark, M.M. Dawoud

Carried out immunohistochemistry: S.A. Dreger, S. Vallath

Assisted with PAI-1 luciferase assays: S. Dickinson

Carried out animal experiments: S.J. Payne, I.R. Hart

Provided access to the UKDCIS tissue collection and patient data and conducted statistical analysis: I. Sestak, J. Cuzick

Carried out MMP qPCR analysis: D.R. Edwards, C.J. Pennington

Isolated and provided all the primary myoepithelial cell lines: J.J. Gomm

Critically evaluated the study: J.F. Marshall, I.R. Hart

All authors participated in discussion of the work and gave final approval.

The authors thank Dr. Paul Weinreb (Biogen IDEC) and Dr. Shelia Violette (Stromedix) for supplying the 6.2G2 antibody; Prof. Dean Sheppard for providing the integrin pcDNA1 neo via Addgene; and Prof. D. Rifkin for providing the MLEC TGF-β reporter cell line.

M.D. Allen was funded by the Breast Cancer Campaign.

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