Background
The ducts of the human breast are composed primarily of two cellular elements in a bilayer structure: luminal epithelial cells, which form a polarised layer around the central ductal cavity, and myoepithelial cells that are positioned between the basement membrane and the luminal epithelial layer. These myoepithelial cells secrete extracellular matrix components required for the correct polarity of the luminal cells and also contract during lactation in order to propel milk through the ductal tree [
1,
2].
An intriguing relationship between these two cell types is observed in ductal carcinoma in situ (DCIS). DCIS is characterised by a proliferation of neoplastic luminal cells into the luminal space of the breast duct, whereas the outer ring of myoepithelial cells remains intact. Accordingly, many have proposed that DCIS is a precursor to invasive breast cancer [
3,
4]. However, as many as 50% of DCIS cases will not develop into invasive breast cancer [
5,
6]. Combined with earlier detection of DCIS, there has been a rise in potential overdiagnosis of breast cancer and, as a consequence, potentially unnecessary treatment [
7]. Novel prognostic markers are therefore needed to identify which cases of DCIS will progress to invasive cancer and which will remain benign. Putative markers are likely to reflect either a loss of myoepithelial integrity, which facilitates subsequent cancer cell invasion, or an alteration in the myoepithelial phenotype [
8]. Moreover, deeper understanding of how the myoepithelial cells maintain the polarised luminal surface, as well as how the relationship between luminal and myoepithelial cells alters with tumour evolution, is key to understanding early mechanisms of cancerous transformation in the breast.
2D cell culture is the primary tool for cancer researchers because this technique provides a standardised high-throughput system whereby the characteristics of specific cell types can be dissected and compared with those of other researchers across the world. However, this is largely reliant on cancer cell lines that do not adequately recapitulate the complexity of the cancer environment [
9]. 3D culture systems are more advantageous because they better reflect the environment in vivo and allow the impact of the extracellular matrix to be assessed. For instance, a number of breast epithelial cell lines, such as MCF-10A cells, are able to form spheroids of polarised epithelial cells with luminal centres, similar to breast duct morphology, when placed in extracellular matrix gels [
10‐
12]. However, a caveat to these systems is that there is no true bilayer structure, thus preventing dissection of the myoepithelial-luminal cell relationship.
Tissue banks for cancer research are becoming increasingly accessible (e.g.,
http://breastcancertissuebank.org), providing researchers with access to normal as well as tumour-derived patient cells. These invaluable resources require new methodologies and models to maximise their potential benefit for research and patients. Using cells derived from normal human breast tissue, we have developed a novel 3D model of the human breast duct bilayer for research applications. We demonstrate that our model is more reflective of the human breast duct than current models and highlight its translational utility by adapting a lentiviral engineering approach to allow objective evaluation of early-stage breast cancer.
Discussion
3D models are becoming increasingly commonplace as the technologies and associated imaging techniques become more accessible to researchers [
22]. Complex models such as these require building blocks that best resemble the tissue they seek to mimic. Cancer cell lines in 3D environments are very useful for answering basic biological questions, but they are of little use when modelling the complex tumour architecture because many cell lines do not adequately reflect the tumour they purport to represent [
9,
12]. Furthermore, they typically represent late-stage disease and not early disease and its interaction with components of the microenvironment. Primary cells derived from patient samples are the perfect resource for complex 3D models that seek to represent the normal and progressive tumour environment. We have demonstrated that normal myoepithelial and luminal cells can be isolated from reduction mammoplasty specimens and maintain their differentiated state in traditional 2D cultures. Recombined into a 3D environment, these cells reorganise into a native bilayer structure. This is a unique strength of our system compared with those that use common cell lines such as MCF10A cells, which can form spheroids with myoepithelial and luminal characteristics but lack a bilayer arrangement of cells.
The 3D environment in which the cellular components reside is a critical component of any model and modifies the resulting structure and behaviour. Our model highlights the importance of using an appropriate matrix where collagen-based gels promoted the re-formation of a bilayer from isolated myoepithelial and luminal cells, whereas Matrigel did not. Matrigel is rich in laminin, which is required for correct luminal orientation in the breast [
23]. Myoepithelial cells are a major source of laminin in the breast [
24]; therefore, in a Matrigel environment, there is no incentive for the two cell types to come together. However, in collagen, the luminal cells require myoepithelially expressed laminin, thus promoting their coalescence into co-units. It is intriguing that the cells in Matrigel cultures adopted characteristics of both myoepithelial and luminal cells. This would suggest that one of the many components of Matrigel is sufficient to drive the plasticity of these cells, which may have implications for the use of Matrigel in primary cell-based work.
Overexpression of HER2 in the luminal component of our bilayer model resulted in the destabilisation of the bilayer and filling of the luminal centre. The breakdown in epithelial orientation in response to overexpression of HER2 has been documented in cancer cell lines [
25]. Strikingly, this loss of lumen could be blocked with HER2-targeted therapy. This proof of concept establishes the use of this model as a screen for novel therapies, not only for HER2-driven cancer but also potentially for any breast cancer subtype, given appropriate manipulation of the cellular components. Intriguingly, the outer myoepithelial layer of our model remained intact following HER2 overexpression in the luminal compartment, recapitulating the development of DCIS. Furthermore, this model can be modified to recapitulate the heterogeneity present in breast tumours. Heterogeneity within breast tumours accounts for variation in treatment response and the emergence of therapy-resistant tumours. Current animal- and cell-based models of breast cancer inherently model a homogeneous tumour and therefore are inadequate to study this crucial component of breast cancer [
26]. The ability to objectively quantify luminal filling is a significant strength of this model because it demonstrates the reproducibility of our model between patient samples and confirms the appropriateness of this model for semi-high-throughput applications.
Methods
Cell isolation and culture
Ductal organoids from reduction mammoplasty specimens were obtained from the Breast Cancer Now Tissue Bank (REC 15/EE/0192). Ductal organoids were digested to a single-cell suspension through digestion in a 0.05% trypsin, 0.4 mg/ml DNase solution at 37 °C for 15 minutes as described previously [
30]. Pure populations of myoepithelial and luminal cells were then isolated through either magnetic bead or FACS separation. Briefly, a single-cell suspension of cells derived from organoids was incubated at 4 °C for 20 minutes with a mouse anti-human CD10 antibody (catalogue number mca1556; Bio-Rad Laboratories, Oxford, UK) conjugated to sheep anti-mouse immunoglobulin G Dynabeads (Thermo Fisher Scientific, Paisley, UK) at a ratio of 2:1 (cells/beads) to label myoepithelial cells. Tagged cells were then pulled out through magnetic separation, and the remaining cells were incubated with Epithelial Enrich Dynabeads (Thermo Fisher Scientific) for 20 minutes at 4 °C to label luminal cells.
Alternatively, for isolation by FACS, single cells derived from organoids were resuspended at 20 × 106 cells/ml and incubated with 0.25 μg/ml allophycocyanin (APC)-conjugated mouse anti-human CD10 (catalogue number 332777; BD Biosciences, Oxford, UK) and 0.06 μg/ml phycoerythrin (PE)-conjugated mouse anti-human EpCAM (catalogue number 347198; BD Biosciences) antibodies for 45 minutes at 4 °C. Cells were then incubated with 0.1 μg/ml 4′,6-diamidino-2-phenylindole (DAPI) to label dead cells prior to the separation of myoepithelial and luminal cells based on APC and PE fluorescence, respectively. FACS separation was performed on a BD FACSAria II cell sorter (BD Biosciences).
Isolated luminal cells were cultured in DMEM/F-12 medium (Sigma-Aldrich, Poole, UK) supplemented with 10% FBS, 0.5 μg/ml hydrocortisone (Sigma-Aldrich), 10 μg/ml apo-transferrin (Sigma-Aldrich), 5 μg/ml insulin (Sigma-Aldrich) and 10 ng/ml epidermal growth factor (EGF; Sigma-Aldrich). Isolated myoepithelial cells were cultured in HuMEC medium (Thermo Fisher Scientific) supplemented with 0.5 μg/ml hydrocortisone, 5 μg/ml insulin, 10 ng/ml EGF and 50 μg/ml bovine pituitary extract (Thermo Fisher Scientific).
3D ductal culture
Primary myoepithelial and luminal cells were combined in a 1:1 ratio and placed in collagen gels, consisting of 2 mg/ml collagen type I (Corning Life Sciences, Flintshire, UK), and 25 mM HEPES, prepared in luminal culture medium adjusted to neutral pH with NaOH. Gels were allowed to set at 37 °C before being overlaid with luminal culture medium. Culture media and indicated drug treatments were changed every 2–3 days. Alternatively, equal proportions of myoepithelial and luminal cells were placed on a pre-set bed of Matrigel Growth Factor Reduced (Corning Life Sciences) and maintained in luminal culture medium containing 5% Matrigel. Doxycycline was sourced from Sigma-Aldrich. Trastuzumab was a kind gift from Roche Pharma (Basel, Switzerland).
Immunofluorescence
Cells cultured on coverslips were fixed in 10% neutral buffered formalin, permeabilised with 0.05% saponin and blocked with 5% bovine serum albumin (BSA) prior to incubation with primary antibody diluted in 5% BSA. Samples were incubated subsequently in a species-appropriate fluorescent secondary antibody before being mounted.
Collagen gels were first treated with 1 mg/ml collagenase (Sigma-Aldrich) for 10 minutes at 37 °C prior to fixation in 10% neutral buffered formalin. Gels were then permeabilised overnight with 1% Triton X-100 and blocked in 10% FBS/2% BSA. Gels were then incubated in primary antibodies for 48 h, followed by a 2-h incubation with species-appropriate fluorescent secondary antibody prior to mounting. Where indicated, gels were incubated in 1 μg/ml DAPI prior to mounting to label cell nuclei.
Paraffin-embedded sections of normal and DCIS breast tissue samples were obtained from the Breast Cancer Now Tissue Bank (REC 15/EE/0192). Sections were de-waxed, and antigen retrieval was performed though boiling sections with 10 mM sodium citrate buffer, pH 6.0. Sections were subsequently permeabilised in 0.1% Triton X-100 and blocked with 10% FBS/2% BSA. Sections were then incubated in primary antibodies as indicated above prior to secondary incubation in a species-appropriate fluorescent antibody. Prior to mounting, sections were incubated with 1 μg/ml DAPI to label cell nuclei.
Primary antibodies used were CK19 (catalogue number RB-9021; NeoMarkers/Thermo Fisher Scientific, Fremont, CA, USA), EpCAM (catalogue number MA5-12436; Thermo Fisher Scientific), p63 (catalogue number SC-8431; Santa Cruz Biotechnology, Dallas, TX, USA), vimentin (catalogue number HPA001762; Atlas Antibodies, Bromma, Sweden), P-cadherin (catalogue number 2198 s; Cell Signaling Technology, Danvers, MA, USA), HER2 (catalogue number 2165 s; Cell Signaling Technology), CK14 (catalogue number ab51054; Abcam, Cambridge, UK), CK8 (catalogue number C5301; Sigma-Aldrich), calponin (catalogue number ABT129; EMD Millipore, Watford, UK) and vimentin (catalogue number M0725; Dako, Carpinteria, CA, USA). Secondary antibodies used were Alexa Fluor 488 donkey anti-mouse (catalogue number A21202; Life Technologies) and Alexa Fluor 555 donkey anti-rabbit (catalogue number A31572; Life Technologies). Fluorescent images were acquired using a Carl Zeiss LSM710 confocal microscope (Carl Zeiss Microscopy, Thornwood, NY, USA).
Immunoblotting
Cells lysates were prepared in 50 mM Tris-HCl, 150 mM NaCl, 1% Nonidet P40 buffer supplemented with protease (EMD Millipore) and phosphatase (EMD Millipore) inhibitor cocktails. Proteins were separated on a 10% SDS-PAGE gel, transferred onto a nitrocellulose membrane and blocked in 5% milk before being incubated in primary antibody diluted 1:1000 in 5% BSA. Membranes were then incubated with a species-appropriate HRP-conjugated secondary antibody (Dako) before bands were visualised using an enhanced chemiluminescence detection kit (GE Healthcare Life Sciences, Piscataway, NJ, USA). In addition to the primary antibodies listed above, the antibody HSC70 (catalogue number sc-7298; Santa Cruz Biotechnology) was used. HRP-linked secondary antibodies used were goat anti-mouse (catalogue number P0447; Dako) and goat anti-rabbit (catalogue number P0448; Dako).
Lentiviral cloning and production
An inducible HER2 expression vector was constructed by subcloning the ERBB2 open reading frame from pDONR223-ERBB2 (23888; Addgene, Cambridge, MA, USA) into pINDUCER21 (46948; Addgene) using the Gateway LR Clonase kit (Thermo Fisher Scientific) following the manufacturer’s guidelines. Lentiviral particles were generated by co-transfecting HEK293T cells with the packaging plasmids pMD2.G (12259; Addgene) and pCMVR8.2 (12263; Addgene) and either pLV-GFP (36083; Addgene), pLV-Azurite (36086; Addgene) or pINDUCER21-ERBB2 using FuGENE HD transfection reagent (Promega, Madison, WI, USA) following the manufacturer’s guidelines. Virus-containing supernatant was collected 48 h post-transfection.
When primary myoepithelial and luminal cells were infected with lentiviral particles, particles were treated with 20 mU/ml neuraminidase (Sigma-Aldrich) at 37 °C for 30 minutes prior to their addition to cells. Particles provided with SMARTchoice promoter selection plates (GE Dharmacon, Lafayette, CO, USA) were also treated with 20 mU/ml neuraminidase prior to their application to cells; fluorescence intensity values were acquired using a FLUOstar Omega fluorescent plate reader (BMG Labtech, Cary, NC, USA).
Image analysis
To objectively and systemically quantify the spheroid volume, a customised eight-step image analysis process was developed. Z-sections of DAPI-labelled cells were converted into greyscale and used to build the greyscale distribution profile for the set of images. A threshold indicating whether a pixel in the image relates to a cell is then calculated; using this, z-sections are further processed into binary images that indicate cell presence. The pixels indicating cells are then extracted and translated into a geometrically accurate point cloud using original image resolution values.
The generated point cloud contains the spheroid volume of interest, cells that exist outside the spheroid and potential noise from the data capture. To determine the points that represent the spheroid volume of interest, density-based spatial clustering of applications with noise was employed. This detects clusters of points within the 3D space with the largest cluster in terms of the number of points being the spheroid volume. These points are extracted to form the final point cloud. The alpha-shape algorithm is then applied to form triangulated bodies that represent the cell and spheroid volumes. The algorithm requires selection of the alpha radius to compute the bodies, and this parameter was calculated as a function of the image resolution. Using the triangulated bodies, the cell and spheroid volumes are calculated alongside the resultant cell/body ratio.
Initial image analysis was performed using the Python programming language and the Python Imaging Library, SkLearn, NumPy and Matplotlib libraries. The alpha-shape algorithm was performed using MATLAB software and the Computational Geometry toolbox (MathWorks, Natick, MA, USA).
Statistical analysis
Change in volume ratio between indicated treatments was compared by analysis of variance followed by Tukey’s post hoc test using Prism 5 software (GraphPad Software, La Jolla, CA, USA).
Acknowledgements
We are grateful to all the donors who supplied tissue to the Breast Cancer Now Tissue Bank cell culture program, from which all normal and cancer material was derived.