Propagation of oestrogen receptor-positive and oestrogen-responsive normal human breast cells in culture

Investigating the susceptibility of oestrogen receptor-positive (ERpos) normal human breast epithelial cells (HBECs) for clinical purposes or basic research awaits a proficient cell-based assay. Here we set out to identify markers for isolating ERpos cells and to expand what appear to be post-mitotic primary cells into exponentially growing cultures. We report a robust technique for isolating ERpos HBECs from reduction mammoplasties by FACS using two cell surface markers, CD166 and CD117, and an intracellular cytokeratin marker, Ks20.8, for further tracking single cells in culture. We show that ERpos HBECs are released from growth restraint by small molecule inhibitors of TGFβ signalling, and that growth is augmented further in response to oestrogen. Importantly, ER signalling is functionally active in ERpos cells in extended culture. These findings open a new avenue of experimentation with normal ERpos HBECs and provide a basis for understanding the evolution of human breast cancer.

U nderstanding the taxonomy and evolution of breast cancer has always relied heavily on the use of normal cell types as reference (for review see ref. 1). Nevertheless, ever since the first protocol for cultivation of normal human breast epithelial cells appeared three decades ago 2 , it has become increasingly clear that there are no protocols that support propagation of oestrogen receptor-positive (ER pos ) cells. Thus, along with the appreciation of epithelial cell lineages in the human breast, primarily the luminal lineage and the basal/ myoepithelial lineage, it became evident that the fastest growing cells in culture are of basal origin 3,4 . Moreover, when it was revealed that ER pos cells in situ accounted for an average of about 7% (mean 6.6%, ranging from 1.2 to 19.1% in a series of 15 normal breast samples) of the cells within the luminal epithelial lineage 5 , the chances of recovering these cells in culture without prospective isolation would in many cases be elusive. Thus, in culture medium that allowed luminal cells to be maintained after passaging, endogenous ER expression disappeared 6,7 . Likewise, even when employing freshly isolated small pieces of breast tissue, including the surrounding stroma thus preserving tissue architecture, steroid receptor expression is eventually lost 8,9 . As a consequence of this, the comparison of cancer with 'normal', for example, the HMT-3522, MCF10A and 184B5 cell lines 1,6,10 , in cell-based assays has relied on normal cells lacking ER expression.
In an attempt to overcome the loss of receptor expression, ER has been ectopically introduced into such cell lines. This approach, however, has had a number of shortcomings, for example, instead of responding to oestrogen by increased proliferation as expected, the ER-transfected cells under standard culture conditions show growth inhibition 11,12 . Accordingly, most of our current knowledge of ER expression, regulation and action comes from breast carcinoma cell lines, whose relation to ER pos normal breast cells at best remains speculative.
Here we first identify the ER pos cells in situ and search for markers that allow their subsequent tracking in culture. We then describe culture conditions for primary ER pos cells in the presence of small-molecule inhibitors of transforming growth factor-beta (TGFb) signalling. Importantly, these conditions also yield ER pos cells from luminal ER neg progenitors but not from basal cells. We envision that the present protocol will serve to bridge the existing gap of knowledge between normal human breast, which contain a small pool of ER pos cells and the overwhelming ER expression found in the majority of breast cancers.

Results
Identification and isolation of normal ER pos HBECs. To answer the long-standing question of whether loss of hormone receptors in culture is due to the loss of cells or the loss of receptor protein expression, it was necessary first to provide tools for cell tracking and sorting of the relevant cells. To unequivocally track ER pos HBECs at the single-cell level we screened our antibody library for surrogate markers with a long half-life, for example, cytokeratins 13 , in culture. In situ staining of more than 30 reduction mammoplasties revealed a surprising pattern with a monoclonal antibody (clone Ks.20.8) originally raised against cytokeratin 20, a simple epithelial cytokeratin with a very restricted expression pattern and not expressed in normal human breast 14 . The lack of true cytokeratin 20 expression in normal breast was here confirmed with two specific antibodies (listed in Table 1). Instead, Ks20.8 stained a subpopulation of luminal cells in a unique scattered pattern (Fig. 1a). Ks20.8 antibodies from four different suppliers (Table 1) revealed similar staining patterns ( Supplementary Fig. 1a). The characteristic staining pattern led us to speculate that it indeed represented ER pos cells. While ER and Ks20.8 apparently co-localized in acini as well as in ducts  Supplementary Fig. 2). For cell sorting purposes we found that CD166 and CD117 made good candidate surface markers of potential ER pos and ER neg cells, respectively-again as revealed by enhanced multicolour immunofluorescence (Fig. 1c). The mutual exclusivity of CD117 and ER (Fig. 1c) confirms what has been reported by others 21 . Accordingly, we designed a fluorescence-activated cell sorting (FACS) protocol to first separate the basal cell population from the luminal cell population based on EpCAM (CD326) and NGFR (CD271) followed by sorting with CD166 and CD117 to further dissect the luminal compartment ( Fig. 2a and Supplementary Fig. 3). This protocol yields three populations, the purity of which was assessed by staining smears with lineage and progenitor markers K14, K18 and K15 (ref. 22), ER-PR as well as the novel ER pos cell surrogate marker Ks20.8 (Fig. 2a). As expected, we found that ER pos /PR pos HBECs were highly enriched in the CD166 high /CD117 low gate (Fig. 2b). While immunofluorescence staining for ER alone evoked a cytoplasmic background staining, which prevented reliable quantification, the smears turned out to be ideal for accurate assessment of the level of co-localization of ER-PR and Ks20.8. Indeed, up to 90% of Ks20.8 pos cells were also ER-PR pos ( Supplementary Fig. 4) and up to 87% of ER-PR pos cells were Ks20.8 pos . The separation of the three subpopulations was further validated by quantitative reverse transcription-PCR (qRT-PCR), which confirmed a high ER expression (ESR1) in the CD166 high cells compared with the other subpopulations (Fig. 2c). An additional panel of markers further distinguished the two luminal subpopulations from the basal cell population ( Supplementary  Fig. 5). Importantly, we found that known ER signalling-related genes such as trefoil factor family-1 (TFF1; ref. 23) and growth regulation by oestrogen in breast cancer 1 (GREB1; ref. 24) were highly expressed in CD166 high cells as compared with other HBECs ( Supplementary Fig. 5). The degree of separation in the CD166/CD117 FACS analysis was, however, somewhat biopsy dependent. In a series of six biopsies originating from women between 19 and 44 years old, five exhibited a similar separation with 11-49% of the cells being CD166 high /CD117 low , while one biopsy apparently did not contain a CD117 high population ( Supplementary Fig. 6). As an alternative, CD117 could be replaced with the laminin receptor 67LR in the CD166 FACS to obtain enriched ER pos HBECs (80% increase in Ks20.8-positive cells in the 67LR high gate versus the 67LR low gate; Supplementary  Fig. 7). In conclusion, the combination of Ks20.8, CD166 and CD117 is a promising marker for ER pos HBEC tracking and sorting.
ER protein is lost in culture. With the surrogate markers in hand we were now able to examine the behaviour of sorted ER pos HBECs in culture, potentially irrespective of ER expression. From the point of view that favourable conditions for luminal epithelial cells also apply to ER pos HBECs, we refined a protocol to permit ample colony formation of luminal cells at clonal density. As cell culture plastic we used Primaria, a substrate with a high content of nitrogen previously shown to promote adhesion of breast cells 25 . The growth medium 'FAD2' was modified from previously described media for culturing keratinocytes or basal breast epithelium on mouse fibroblast feeders 26 ARTICLE but with less serum, that is, 5%, as in Liu et al. 26 Under these conditions, we gauged for colony formation among the three FACS gated populations described above. When plated at a clonal density of 400 cells per cm 2 , indeed colony forming luminal cells from the CD117 high gate were highly favoured over basal cells (Fig. 3a). However, it was also clear that the ER pos HBECs from the CD166 high gate entirely failed to form colonies under similar conditions. However, a scrutiny of cultures at a higher magnification revealed that CD166 high cells in fact did not disappear from the culture. Rather, they plated, survived and also stained with Ks20.8 as a testimony of their original identity, but they entirely refrained from growth and rapidly lost the ER protein (Fig. 3b). By comparison, other HBEC culture media, that is, M87A (ref. 28), MEGM or WIT-P-NC did not support plating of CD166 high cells. Therefore, we conclude that failure of culturing ER pos HBECs is caused by both lack of growth and loss of ER expression under conditions otherwise favouring propagation of luminal epithelial cells.
TGFb inhibitors induce ER expression and release ER pos HBECs. We noted that earlier in vivo studies had implicated TGFb1 signalling in the restraint of ER pos mammary epithelial cells 29,30 . Here we therefore examined three small inhibitor molecules of TGFb signalling, RepSox, SB431542 and SD208, and combinations hereof for their ability to relieve a potentially negative regulation of ER pos HBEC growth in culture. We found that dual inhibition with SB431542 and RepSox, hereafter collectively termed TGFbR2i, recapitulated ER expression and stimulated ER pos HBEC colony formation in four out of four biopsies. Moreover, the cells maintained Ks20.8 reactivity, albeit more widespread than ER expression compared with what is seen   in situ (Fig. 4a). ER pos cells from three different biopsies in lowdensity primary cultures in TGFbR2i (4,000 cells per cm 2 ) exhibited a clonal capacity of 0.54%, 0.33% and 0.43%, respectively. By comparison, control conditions resulted in small, mostly abortive ER neg clones (Fig. 4b). In general, ER expression was particularly evident in the dense centre of proliferating colonies or at confluency. Two different sources of SP1 antibody gave similar results (Table 1). Moreover, staining was confirmed with the less-sensitive ER antibody 1D5 (ref. 31). The response to TGFbR inhibitors was reproducibly observed in 10 out of 10 biopsies. In addition to ER, TGFbR2i induced the expression of keratin K8 and the luminal cell transcription factors forkhead box protein A1, FOXA1 and E74-like factor 5, ELF5 (for reviews, see refs 32,33) as revealed at the transcriptional level (Fig. 4c). This expression pattern was confirmed using another biopsy. TGFbR2i also upregulated the transcription of a number of genes known to be downstream targets of ER or modulators of ER activity, such as TFF1 (ref. 23) and insulin growth factor-binding protein 5 (IGFBP5; ref. 34; Supplementary Fig. 8a). Moreover, induction of ER protein expression to a significant degree was specific to RepSox as twice the concentration of SB431542 or replacement of RepSox with another ALK5 kinase inhibitor SD208 was insufficient to induce K8 and ER (data not shown). RepSox alone was not as effective in inducing ER or PR as in combination with SB431542 ( Supplementary Fig. 8b). Likewise, while SB431542 alone was capable of inducing increased expression of ER at the mRNA level (Fig. 4c), this translated to the protein level to a significant degree only in the presence of RepSox ( Fig. 4d and Supplementary Fig. 9). These findings suggest that ER expression is controlled through specific inhibition of TGFb signalling. The presence of TGFbR in HBECs as well as the inhibition of phosphorylated SMAD2 concurrent with inhibitor-induced ER expression supported this (Supplementary Fig. 10 and Fig. 4d). Indeed, TGFbR2i culture was a key to sustained ER protein expression as removal of TGFbR2i led to complete loss of ER protein expression as observed in four independent experiments. Thus, in a representative experiment the frequency of ER pos cells   was reduced from 33.7% (s.d.±5.0%; 3 Â 100 cells) in continuous TGFbR2i culture to 10.7% (s.d. ± 0.5%) after 3 days and to 0% after 5 days without TGFbR2i. We next tested whether TGFbR2i would induce de novo ER expression in other subpopulations of luminal cells, for example, in the much more frequent CD117 high -derived HBECs. Consistent with the presumed progenitor status of CD117 high cells in the human breast 35 , clones emerged from CD117 high cells that gained ER and stained positively for Ks20.8. This was, however, the only additional source we found of ER pos HBECs, since TGFbR2i failed to induce ER in basal cells, in breast fibroblasts or in the established normal cell breast lines MCF10A or HMT-3522. Thus, both from preexisting ER pos HBECs and from ER neg progenitors, TGFbR2i readily provides growing colonies of ER pos HBECs.
TGFbR2i allows growth of ER pos HBECs in serial subculture. TGFbR2i appeared to support growth of ER pos HBECs also after passaging. To test this systematically, we plated CD166 highderived cells at a density of 6,400 cells per cm 2 in primary culture, and the cultures were subsequently passaged at a density of 4,000 cells per cm 2 . On passaging, ER expression was particularly evident in the dense centre of proliferating colonies. Continuous proliferation under these conditions was maintained for up to six passages, corresponding to 15 population doublings (exemplified in Fig. 5a). Importantly, however, in the absence of RepSox, the cells could not be expanded beyond fourth passage (Fig. 5a). The lifespan and ER expression (ranging from 21 to 77% in second-to fourth-passage cultures, Table 2) were somewhat biopsy dependent, and in general proliferation slowed between fourth and sixth passages, and at the same time downregulation of ER expression was observed. While this narrowed the window of experimentation to up to fourth passage, the cells could easily and reproducibly be replaced with new cultures with similar luminal characteristics (Table 2). However, passage number could be increased by increasing seeding density. Thus, ER pos cells derived from three biopsies could be cultured for another two to three passages, that is, up to passage six to nine when passaged at a high density of 8,000-20,000 cells per cm 2 . Moreover, the lifespan of ER pos HBECs passaged at low seeding density could be further extended by initial plating on 3T3 feeders, which lead to proliferation for more than 10 passages, corresponding to more than 25 population doublings (Fig. 5a). Cells from a different sorting exhibited similar characteristics (Fig. 5a). We subsequently addressed whether TGFbR2i culture would allow for alternative approaches to extend the lifespan of normal breast-derived ER pos cells. ER pos HBECs were successfully transduced with pBABE-neo-hTERT and pLenti X2 hygro/shp16 constructs and have now been cultured for 4 months with weekly passages at 6,000 cells per cm 2 , exceeding 12 passages (Fig. 5a). Transduction of ER pos cells from a different biopsy and subsequent splitting at up to 1:4 supported the robustness of this approach. Of note, these long-term cultured cells exhibited a phenotype essentially similar to cultures with definitive lifespan, including a relatively high level of ER expression (Supplementary Fig. 11 and Table 2). Thus, while the proliferation of ER pos cells with definite lifespan is somewhat slow as it took more than 100 days to generate more than 25 population doublings, the present protocol nevertheless allows a considerable expansion of the ER pos cell population extending from low-passage cultures with low seeding density through medium-passage cultures with high seeding density to high-passage cultures of hTERT/shp16-tranduced cells, which provides a relatively wide window of experimentation. Of note, until senescence the cells maintain their expression of ER, Ks20.8 reactivity as well as-mainly in densely packed colonies-expression of PR (Fig. 5b). In conclusion, TGFbR2i readily supports serial subculture of a population of ER pos HBECs, that is, ER pos progenitors.
ER pos progenitors respond to oestrogen. As an ultimate test for a physiologically relevant ER expression we decided to assess the effects of added cognate ligand, that is, oestrogen (b-oestradiol). Accordingly, ER pos cells were exposed to 10 À 8 M of b-oestradiol or vehicle. The effect of oestrogen was tested in the absence of epidermal growth factor (EGF) since the transcriptional activation of ER can be induced by EGF-induced MAP kinase activity 36 . As a first indication of a functional ER, a higher focal expression of PR protein, a downstream target of ER signalling, was seen in the presence of oestrogen (Fig. 6a). Low-passage If RepSox is omitted, the cells cannot be expanded beyond fourth passage (cross). Initial plating on 3T3 feeders with quantification starting in passage three extends proliferation to more than 10 passages, corresponding to more than 25 population doublings (open circle). Cells from a different sorting (albeit followed for a shorter period) exhibit similar extended proliferative capacity (closed circle). hTERT/shp16-transduced CD166 high /CD117 low cells subsequently passaged at a fixed number of 6,000 cells per cm 2 at each split extended the proliferative capacity even further (closed triangle), and the cells have now been growing for more than 12 passages. hTERT/shp16-transduced CD166 high /CD117 low cells derived from a different biopsy, split at a ratio of up to 1:4, has so far been growing up to passage 9 (open square). (b) Even beyond 20 population doublings (passage 9), ER pos cells with definitive lifespan maintain ER and PR expression as shown by immunoperoxidase and haematoxylin staining (cells in ninth passage seeded at 4,000 cells per cm 2 and stained at day 5 with SP1 prediluted and SAN27, respectively). Scale bar, 50 mm.
cultures (passage 2-4) were exposed to oestrogen or vehicle, and the cell number was quantified. As seen in Fig. 6b, significantly higher cell numbers were recorded in cultures with oestrogen (Fig. 6b). Medium-passage cultures (passage 5-6) reached by high seeding density exhibited a similar response, albeit in general exhibiting a slower growth (Fig. 6b). Likewise, a proliferative response was observed in high-passage culture of hTERT/shp16transduced ER pos cells (passage 7-12; Fig. 6b). That the observed increase in cell number was indeed due to the specific effect of signalling through ER was verified on addition of the ER antagonist ICI182,780, which could abolish the effect completely ( Supplementary Fig. 12). Long-term exposure to oestrogen further augmented the oestrogen response at the transcriptional level, as indicated by increased expression of PGR and GREB1 (Fig. 6c), as well as at the translational level as observed in 10 preparations from three biopsies tested in passage 2-6 ( Supplementary Fig. 13a) with a significant overlap in ER and PR protein expression ( Supplementary Fig. 13b). Thus, the cultured ER pos cells retain their ability to respond to oestrogen in a physiologically relevant manner. As an alternative approach we plated ER pos cells on fibroblast feeders, which have been used previously to reveal an oestrogenic response in mouse mammary cells 37 . ER pos cells were plated on confluent fibroblast feeders in TGFbR2i with or without oestrogen. At day 8, there was a significant increase in cell number with oestrogen as compared with the control (Fig. 6b). In the presence of fibroblasts, ER pos cells responded to oestrogen irrespective of the presence of EGF (Fig. 6b). These data imply that the stromal microenvironment beyond the influence of TGFb is important in the regulation of the growth of ER pos cells.
In conclusion, we have developed a method to isolate, track and subculture ER pos and oestrogen-responsive cells from normal human breast.

Discussion
In spite of the fact that there are multiple protocols for enrichment, long-term cultivation and clonal growth of HBECs, none of them are able to isolate, track or support the growth of ER pos HBEC. This led us to screen our antibody repository for antibodies against human breast antigens eligible for flow cytometry and cell tracking, and to test whether the reluctant growth of ER pos HBEC in culture is due to inhibitory TGFb signalling. Our present findings demonstrate that ER pos HBECs identified in situ by a panel of markers, including PR, AP2b, GATA3, Bcl2, CDw75, N-cadherin and 67LR, are enriched for in a CD326 high /CD271 low /CD166 high / CD117 low gate, and these cells express an antigen, unexpectedly decorated by an antibody, clone Ks20.8, originally raised against cytokeratin 20 (ref. 14). The distinction between ER pos cells, other luminal cells and basal cells is further supported by their relative expression of a panel of transcripts.
The prospective isolation and tracking of ER pos single cells from normal breast tissue hold promises for the future comparisons between normal, benign and malignant ER pos cells, which will hopefully shed some light on the evolution and pathogenesis of the most frequent form of human breast cancer. Being able to isolate and track the cells, however, would be of limited value if the ER expression was lost on culture. It has been anticipated that ER pos normal cells cannot divide and that this is why HBECs rapidly lose ER expression in culture 38 . We show here that ER pos cells can be released from growth restraint and sustained by TGFbR2i and that ER pos cells under these conditions can be expanded considerably. These findings may represent a paradigm shift in studying ER expression and function in the breast, in the future no longer relying exclusively on in vivo rodent models and human breast cancer cell lines. In addition, we believe the results answer the long-standing question of whether ER pos cells can self-renew, and further establishes that ER pos cells can be generated from ER neg progenitors [39][40][41][42] , here represented by CD117 high luminal cells. Importantly, the response to TGFbR2i is specific to luminal cells only, as neither basal cells, fibroblasts nor ER neg normal breast cell lines are able to switch on ER in response to TGFbR2i. The fact that ER pos HBECs are here shown to be proliferating is in favour of the existence of ER pos progenitors. In general, however, most ER pos HBECs are considered to be close to the base in the hierarchy 35 . Thus, as has been hypothesized for breast cancer hierarchies 43 , our data could be interpreted in favour of the existence of bidirectionality also in the normal breast hierarchy, that is, on appropriate stimuli a subpopulation of apparently differentiated cells turn out in reality to be facultative progenitors.

ARTICLE
The release of ER pos HBECs in culture by TGFbR2i is clinically relevant because it may help explain an enigmatic difference between the normal human breast and breast cancers. Thus, in the normal breast in vivo there is a strict dissociation between steroid receptor expression and proliferation 15 . In breast cancer and to a varying degree in precancerous breast lesions this negative association is lost 44 . It has been speculated that this may represent an important early change in the genesis of breast cancer either reflecting a failure to downregulate ER as cells enter the cell cycle or a failure to suppress division of ER pos cancer cells 44 . Our data indeed are in favour of the latter possibility because ER pos HBECs do divide if TGFb signalling is perturbed, a very likely scenario in cancer (for review see ref. 45). While in primary breast cancer a normal-like TGFb signalling is still in operation to restrain growth, in metastatic breast cancer TGFb signalling has shifted to that of an epithelial-mesenchymal transition response 45 .   h T E R T / s h p 1 6 P 9 5 7 P 9 5 8 P 9 5 9 h T E R T / s h p 1 6 a b c Figure 6 | ER pos cells in culture respond to oestrogen. (a) Staining of PR (SAN27) in second-passage CD166 high cells deprived of EGF and exposed to vehicle (EtOH) or oestrogen (E2) for 10 days. ER pos cells respond to oestrogen by increased expression of PR. (b) Quantification of cell number in four sets of pairwise triplicate cultures of EGF-deprived, oestrogen-stimulated CD166 high low-passage cells grown for 4 days in second (P959) or third passage (P958), or 13 days in fourth passage (P959); medium-passage cells grown at high density before growth for 6 days (P958), 15 days (P957) and 14 days (P957 cultured without EGF since second passage) in fifth passage or cells with extended lifespan grown for 14 days in sixth passage; or high-passage cells (hTERT/shp16) in seventh passage grown for 13 days or in tenth to twelfth passage grown for 11, 15 and 11 days, respectively, in the presence of oestrogen (dark grey) or vehicle (light grey); or pairwise triplicate cultures of CD166 high cells plus feeder grown for 8 days in the presence of oestrogen (dark grey) or vehicle (light grey). Second-passage cultures derived from two different biopsies (first and fourth set of bars) do not respond differently from third-passage cultures (second set of bars), irrespective of omission of EGF from TGFbR2i before co-culture (second set of bars)). Omission of EGF throughout the entire experimental period reduced the total cell number, but did not augment the response to oestrogen (third set of bars Another implication of the TGFbR2i culture protocol is that it represents a much-in-demand cell-based assay for oestrogen action on normal cells. It is already well established that oestrogen is a mitogen for ER pos breast cancer. However, its role in relation to ER pos HBECs has remained a mystery due to lack of ER pos cell culture models, and because ectopic expression of ER in basal cell lines under standard culture conditions has provided the paradoxical result of growth inhibition 11,12 . As a proof of principle, we here show that the TGFbR2i protocol serves as a physiologically relevant cell-based assay for oestrogen action, and moreover, in ER pos cells, several oestrogen-responsive genes downstream of ER are upregulated. Thus, the TGFbR2i protocol represents a cell-based assay for gauging oestrogen action reminiscent of its action in vivo. Intriguingly, in the presence of fibroblasts, growth of ER pos HBECs is stimulated by oestrogen irrespective of the presence of EGF. This implies that stromal factors modulate ER activity. With the relevant representatives of ER pos cells in hand, the complexity of this interaction can now be elucidated.
Our data further suggest that TGFb signalling is key to the CD166 high phenotype, and the exact mechanisms by which TGFbR2i generate ER pos cells clearly await further scrutiny. It cannot be excluded, however, that what we observe is part of a more general association between CD166 expression and TGFb signalling. Thus, very recently others have found that TGFb signalling is a main driver of the behaviour of CD166-expressing prostate cells 46 .
The present protocol is the result of several years of experimentation including many reduction mammoplasties from different donors until the present conditions were established. We cannot exclude that further improvements are possible. The data illustrate, however, that apparent post-mitotic cells in vivo, given the right conditions may multiply considerably, yet still exhibiting a definitive lifespan, and the approach moreover serves as a platform for extending the lifespan of ER pos cells into serial subculture.
While we have reproducibly recovered and propagated ER pos cells from all biopsies tested, we nevertheless wish to emphasize on a number of technicalities, which should be carefully observed when the protocol is applied. Biopsies inherently vary with respect to the frequency of ER pos cells 5 , which may eventually influence the definite number of ER pos cells that can be derived from a given biopsy. Biopsies also vary with respect to growth potential, but in our hands they all perform well up to at least passage 4 ( Table 2). It should be kept in mind, however, that the protocol releases only a minority of the total number of initially quiescent ER pos cells, that is, up to 0.54% when plated at 4,000 cells per cm 2 on sorting. Also, we strongly recommend that any attempts to validate Ks20.8 as well as co-expression of ER and PR by multicolour immunofluorescence are accompanied by immunoperoxidase staining to unravel the true overlap in staining. This is particularly important because we found that tracking of ER pos HBECs in smears must rely on either the ER/PR antibody mixture or the Ks20.8 surrogate marker specifically combined with the ER fixation protocol. Last, we would like to draw the attention to the fact that while the cultured CD166 high / CD117 low cells for the major part retain an elaborate profile of ER pos cells in situ, they may-again somewhat dependent of the biopsy of origin-display focal traits of basal or myoepithelial cells, such as p63, concurrently with luminal traits (Table 2).
That said, there are multiple implications of the present protocol. First, knowledge about how to turn on and off the ER expression in non-malignant breast epithelial cells may offer an alternative to selective oestrogen receptor modulators in prevention of breast cancer in women with elevated risk of disease 47 . Second, a reproducible source of normal human ER pos HBECs will represent a first step towards a physiological, cell-based screen for environmental oestrogenic activity and susceptibility of normal cells to breast cancer 48 . And third, the protocol may serve to test the functional role of recently identified single-nucleotide polymorphisms associated with increased risk of breast cancer 49 .
In conclusion, we have provided a cell-based assay that allows normal breast ER pos cells to present themselves for investigation. These findings may fuel future advances in breast cancer prevention.

Methods
Tissue. Normal breast biopsies were collected with consent from women undergoing reduction mammoplasty for cosmetic reasons. The use of human material has been reviewed by the Regional Scientific Ethical Committees (Region Hovedstaden) and approved with reference to H-2-2011-052. Normal breast tissue was prepared as described 25 . In brief, normal breast specimens were minced with opposing scalpels and treated overnight at 37°C with rotation in collagenase (Type 3, Worthington Biochemical Corporation, 900 U ml À 1 in DMEM/F-12 (Life Technologies, catalogue no. 21041) with 2 mM glutamine (Sigma, G7513) and 50 mg ml À 1 gentamycin (Biological Industries)) before differential centrifugation to aspirate dissolved fat and to isolate epithelial organoids and fibroblasts. These were then either used directly or frozen in liquid nitrogen for later use. To detect TGFb receptors on the cell surface, third-passage CD166 high / CD117 low cells were incubated with monoclonal antibody TGFbRII (MM0056-4F14, 1:20, Abcam) followed by secondary antibody AF488 (IgG1, 1:500) and analysed by FACS. Overlay histograms were produced using Flowing Software 2.5.1 (University of Turku, Finland).
Cell culture. On sorting, the primary cell populations were plated in Primaria T25 flasks (#3813, Becton Dickenson) in the presence of 'FAD2' ((DMEM, high glucose, no calcium, Life Technologies):Ham's F12 Nutrient Mixture (F12, Life Technologies) 3:1 v/v) with 2 mM glutamine, 0.5 mg ml À 1 hydrocortisone, 5 mg ml À 1 insulin, 10 ng ml À 1 cholera toxin (Sigma-Aldrich), 10 ng ml À 1 epidermal growth factor (Peprotech), 1.8 Â 10 -4 M adenine, 10 mM Y-27632 (Y0503, Sigma-Aldrich or 1683, Axon Medchem) and 5% fetal bovine serum (Sigma-Aldrich), modified from refs 26,27). On plating, which could take up to 2 days for the luminal ER neg (CD166 low /CD117 high ) cells and thus determined the time point for addition of TGFbR2i, a combination of the selective inhibitor of TGFb type I receptor activin receptor-like kinase ALK5, ALK4 and ALK7, SB431542 (ref. 50) (10 mM, S4317, Sigma-Aldrich or 1661, Axon Medchem) and an inhibitor of autophosphorylation of ALK5, RepSox 51,52 (25 or 50 mM, R0158, Sigma-Aldrich) was added. To test the specificity of the effect of TGFbR2i, in some experiments SB431542 alone or the double concentration of SB431542 was used instead of TGFbR2i, or RepSox was replaced by another TGFb type I receptor activin receptor-like kinase ALK5 inhibitor, SD208 (1 mM, Tocris Bioscience). The vehicle, dimethyl sulfoxide (Sigma-Aldrich), was included in all experiments in appropriate concentrations for control cultures. Gentamycin (50 mg ml À 1 , Biological Industries) was added to the cultures throughout the first week after sorting, otherwise antibiotics were not included. To address whether propagation of ER pos cells was restricted to the FAD2 medium, CD166 high /CD117 low cells sorted from one of the biopsies were plated at a density of 4,000 cells per cm 2 on NATURE COMMUNICATIONS | DOI: 10.1038/ncomms9786 ARTICLE Primaria six-well in either FAD2, M87A 28 , MEGM (LONZA) or WIT-P-NC (STEMGENT) medium including cholera toxin (100 ng ml À 1 ) (ref. 53) and the cultures were observed for plating for 3 days. The possible induction of ER by TGFbR2i in other breast cells was further tested in MCF-10A 54 , HMT-3522 (ref. 55), fibroblasts purified from normal breast tissue 25 , CD117 high cells purified from three different biopsies, as well as in basal cells isolated by FACS as described above and cultured on irradiated NIH-3T3 feeders 26 before exposure to TGFbR2i for up to 7 days followed by staining for ER and Ks.20.8. Fibroblasts were routinely grown to confluency on collagen-coated T25 flasks (Nunc, 8 mg collagen per cm 2 , PureColl, CellSystems) in DMEM/F12, with 2 mM glutamine and 5% FBS before co-culture with CD166 high /CD117 low luminal cells. For phase contrast microscopy a Nikon Diaphot 300 microscope was used.
Cloning efficiency at low density in FAD2 was observed by plating the sorted basal cells, and the luminal populations CD166 low /CD117 high and CD166 high / CD117 low , at 400 cells per cm 2 and culturing for 14 days, followed by fixation in methanol for 5 min at À 20°C and counterstaining of nuclei with haematoxylin. To assess the Ks20.8 and ER expression of the sorted populations and to further assess whether starting the culture at a higher density would change the expression, cells were seeded at 3,000 cells per cm 2 and cultured for 9 days before immunocytochemical staining.
To quantify the frequency of Ks20.8 high /ER pos colony-forming units CD166 high / CD117 low luminal cells from three different biopsies were plated at a density of 4,000 cells per cm 2 with or without TGFbR2i and grown for 13 days before staining for ER and Ks20.8 followed by quantification using an ocular grid. The number of stained colonies defined as presence of either marker in three areas of each culture relative to the initial number of seeded cells was calculated.
Low-density split cultures were initially seeded at 6,400 cells per cm 2 and grown for up to 15 days in primary culture in TGFbR2i or with SB431542 alone. Next, the cultures were trypsinized and seeded at 4,000 cells per cm 2 in triplicate cultures, and subsequently passaged at the same density before the cultures reached confluency. Parallel cultures were stained to assess Ks20.8, ER and PR expression status. The number of cells was quantified manually using a counting chamber. Medium-passage cultures were passaged at high seeding density, that is, 8,000-20,000 cells per cm 2 . For extended cultivation of low-density cultures sorted luminal Ks20.8 pos /ER pos (CD166 high /CD117 low ) cells were first allowed to form colonies on irradiated NIH-3T3 feeders in modified breastoid base medium (BBM) without HEPES 56 (DMEM/F12 supplemented with glutamine, 100 mM ethanolamine (Sigma , E0135), 1 mg ml À 1 hydrocortisone (Sigma, H0888), 9 mg ml À 1 insulin (Sigma, I6634), 5 mg ml À 1 transferrin (Sigma, T1147), 5.2 ng ml À 1 selenous acid (sodium salt, BD Biosciences, 354201), 20 ng ml À 1 basic fibroblast growth factor (Peprotech, 100-18B) and 5nM amphiregulin (R&D Systems, 262-AR-100/CF)) with the addition of Y-27632, adenine and the serum replacement B27 (20 ml ml À 1 , Life Technologies), and subsequently sorted by FACS on incubation with CD326 and CD271 as described above to isolate CD326 high /CD271 high cells before plating and passaging in TGFbR2i culture. Population doublings were calculated as n ¼ 3.32(log UCY À log I) þ X, where n ¼ population doubling, UCY ¼ cell yield, I ¼ inoculum number and X ¼ population doubling rate of inoculum.
To determine whether continuous TGFbR2i culture was needed to sustain ER and Ks20.8 expression, cells in the sixth passage were switched to FAD2 and the number of ER pos cells (n ¼ 3 Â 100) was compared with parallel TGFbR2i cultures at day 3 and 5. Three similar experiments were performed with cloned hTERT/ shp16-transduced cells (please see below).
As an alternative approach to extend the lifespan of Ks20.8 pos /ER pos cells, the method by Kiyono et al. 57 was adapted by introducing human telomerase (hTERT) and shRNA p16 (shp16) to second-passage CD166 high /CD117 low cells. The viral constructs, pLENTi X2 Hygro/shp16 (w192-1, AddGene #22264, a gift from Eric Campeau) and pBABE-neo-hTERT 58 (AddGene #1774, a gift from Robert Weinberg) were prepared as follows: lentiviral particles containing the shp16 construct were generated by transient co-transfection into HEK-293T cells using the calcium phosphate method with the vesicular stomatitis virus glycoprotein expressing construct pCMV-vesicular stomatitis virus glycoprotein as well as polgag construct pLP1 and rev construct pLP2. Retroviral hTERT particles were generated by transient co-transfection of the vector construct into a pantropic retroviral packaging cell line GP2-293 (Clonetech), which stably expresses the retroviral gag-pol genes. Medium was changed B16 h after transfection and viral media was collected B48 h later and stored at 4°C for further purification. Hightitre stocks of the virus were purified using a 20% sucrose gradient during ultracentrifugation with a Beckman SW32Ti rotor at 25,000 r.p.m. for 1.5 h. Purified viral solution was stored at À 80°C.
Before transduction, the cells were first treated with 200 mU ml À 1 neuraminidase (N7885, Sigma) for 2 h at 4°C, and then transduced using a highviral titre 59 containing the pBABE-neo-hTERT construct in TGFbR2i without FBS and incubated at 37°C in 5% CO 2 overnight. The culture medium was changed and the transduced cells underwent antibiotic selection for 10 days with 500 mg ml À 1 G418 (Life Technologies). On confluency, the cells were passaged and underwent an identical transduction procedure with viral particles containing the pLenti X2 Hygro/shp16 construct and subsequent antibiotic selection with 100 mg ml À 1 hygromycin (Sigma) for more than 2 weeks. The efficiency of the protocol was confirmed by hTERT/shp16 transducing second-passage CD166 high /CD117 low /Ks20.8 pos /ER pos cells derived from a different biopsy.
To test the proliferative response to oestrogen in low-passage (p2-4) cultures, ER pos cells in TGFbR2i culture were seeded with 4,000 cells per cm 2 in TGFbR2i with 25 mM RepSox and without EGF, supplemented with oestrogen (10 -8 M, b-oestradiol, E2758, Sigma-Aldrich) and cultured for up to 13 days with medium change every other day. A primary culture from another biopsy was cultured 27 days in the presence of oestrogen and without EGF before splitting. The cells were passaged at 4,000 cells per cm 2 in triplicate cultures with oestrogen or vehicle (96% ethanol). At day 4, the cultures were trypsinized and quantified using a counting chamber. Medium-passage (p5-6) cultures were passaged at 10,000-20,000 cells per cm 2 before seeding at 4,000 cells per cm 2 , and the high-passage (p7-12) cultures including two hTERT/shp16-transduced populations derived from separate biopsies were passaged at 6,000 cells per cm 2 or split at up to 1:4 before the growth experiment with a seeding density of 6,000 cells per cm 2 . The effect of an oestrogen antagonist (ICI182,780, Sigma-Aldrich), was tested in sixteenth passage, single-cell-cloned hTERT/ shp16-transduced cells seeded at 4,000 cells per cm 2 in TGFbR2i exposed to vehicle or oestrogen (10 -10 M) and increasing concentrations of ICI182,780 (10 -10 , 10 -9 and 10 -8 M). At day 6, cultures were trypsinized and quantified in a cell counter (Roche Innovatis).
To assess whether the response of ER pos cells to oestrogen was modulated by the presence of fibroblasts, second-or third-passage ER pos cells, with or without the omission of EGF in the previous passage, were plated at a density of 5,600 cells per cm 2 on confluent fibroblast feeders in triplicate culture. The following day, the culture medium was switched to TGFbR2i with oestrogen or vehicle. Under these conditions, fibroblasts did not grow. At day 8, the cultures were trypsinized and the total cell number was quantified. Whether a continuous lack of EGF influenced the response to oestrogen was tested in a similar set-up, where EGF was omitted from the medium during the entire experimental period.
The long-term effect of oestrogen on ER and PR expression in cultures without EGF was tested in 10 preparations of pairwise cultures in passage 2-6 derived from three different biopsies. The cultures were grown up to 26 days with oestrogen and vehicle, respectively, and subsequently stained by immunoperoxidase or immunofluorescence (see below).
RNA extraction and qRT-PCR. Total RNAs from sorted normal primary cells from eight biopsies were extracted using Trizol (Invitrogen) and were reverse transcribed using the High Capacity RNA-to-cDNA Kit (Applied Biosystems). For general gene expression profiling, 2 ng of total cDNA was used to perform quantitative real-time PCR using the TaqMan Gene Expression Assays (Applied Biosystems) on CFX384 Touch Real-Time PCR Detection System (Bio-Rad). The primers are listed in Supplementary Table 1, and real-time PCR conditions were the following: 50°C for 2 min and 95°C for 10 min, followed by 40 cycles at 95°C for 10 s and 60°C for 30 s. To quantify cDNA concentration represented by cycle threshold (Ct) values, Cts were determined at the initial period of exponential amplification in triplicates and the different PCR runs were adjusted by inter-run calibrators (Bio-Rad CFX manager 3.0). Each gene expression level was to the mean of four reference gene expressions (GAPDH, HPRT1, TBP, and TFRC). The qRT-PCR data were then visualized as a heatmap using the statistical programme R (ref. 60) or as a bar graph using the values calculated by the 2 À DDCt method 61 . For qRT-PCR of ER, K8, FOXA1, ELF5 and K18, second-passage cultures of CD166 high cells at a density of 8,000 cells per cm 2 in FAD2, SB431542 or SB431542 switched to TGFbR2i culture conditions for 5 days before RNA extraction were used.
The expression of ER signalling and its downstream target genes were analysed in CD166 high cells with or without TGFbR2i by using a RT 2 Profiler PCR array (human oestrogen signalling, Qiagen) according to the manufacturer's instructions. Total 4 mg of RNA was used per array, which was performed in duplicates from two different biopsies. Three sets of E2-or ethanol (vehicle)-treated second-or thirdpassage ER pos cells or hTERT/shp16-transduced cells were further analysed in triplicate for PGR and GREB1 expression with the TaqMan Gene Expression Assays in the same condition mentioned above, using 20 ng of cDNA in each PCR reaction.
Antibodies are listed in Table 1. Many antibodies stain independently of fixation procedure, but of note, to stain for ER and PR, cultures were rinsed in PBS, pH 7.4, or sections were air dried before fixation, for 5 min at room temperature, in 3.7% formaldehyde, two rinses in PBS, fixation in methanol:acetone (1:1) for 5 min at 20°C, two rinses in PBS, permeabilization in 0.1% Triton X-100 in PBS, twice for 7 min, rinse in PBS and kept wet before application of UltraV Block (Thermo Scientific). ER and PR expression in cultures with long-term oestrogen exposure was assessed by immunoperoxidase for ER (SP1, prediluted) and PR (SAN27 or PgR636) or by double-labelling immunofluorescence for ER (SP1, 1:10) and PR (PgR636, 1:10) with Alexa Fluor 488 Goat anti-Rabbit IgG (1:500) and Alexa Fluor 568 Goat anti-Mouse IgG1 (1:500) as secondary antibodies.
To stain for K15, UltraV Block was substituted for 10% normal goat serum in PBS. Immunofluorescence and peroxidase stainings were evaluated, quantified and photographed using a laser-scanning microscope (LSM 510 or LSM700; Carl Zeiss MicroImaging, Inc.) and bright-field microscopes (Laborlux S or DM5500B, Leica), respectively. For quantification of immunoperoxidase staining of ER, K8, K19 and p63, nuclei were counterstained with haematoxylin and counted in randomly selected fields using an ocular grid and given as the percentage of stained cells of a total of 1,000 cells evaluated with a Â 25 objective.