Introduction
The classification of breast cancer into several distinct molecular and histological subtypes can provide information to help guide patient therapy and predict outcome [
1,
2]. Tumors that retain histological and molecular attributes of normal tissue are considered well differentiated and are generally less aggressive and correlate with better patient prognosis. In contrast, the loss of normal tissue structure and the dysregulation of genes involved in modulating growth and differentiation indicate transition of the disease into a more advanced stage [
3]. A better understanding of tumor etiology and processes that control the transition between early and advanced states of breast cancer may improve strategies for detection, treatment, and prevention of the disease.
How a particular cell responds to a transforming event, its susceptibility for malignant progression, and its role in establishing a tumor’s histological and molecular fate are poorly understood. The mammary gland is a complex tissue composed of two distinct cell lineages, the luminal epithelium and myoepithelium, with each lineage encompassing a hierarchy of cells at various states of differentiation [
4,
5]. When a normal cell is transformed, preexisting signaling networks intrinsic to that particular cell type may become dysregulated and contribute to tumor growth and progression. For example, tumors classified as a hormone-receptor positive subtype express estrogen receptor (ER or
Esr1) and are generally dependent on estrogen for growth, recapitulating characteristics of a subset of normal, luminal epithelial cells found in the breast [
6]. Molecular similarities are also observed between other normal cell populations and cancer subtypes. Mammary stem cells have a gene expression signature similar to spindloid and claudin-low tumors [
7‐
9], whereas the molecular signature of normal luminal progenitors is associated with basal-like breast cancer [
8]. Even differentiated mammary epithelial cells (MECs) share molecular features with a cancer subtype. Tumors histologically classified as lipid-rich carcinoma of the breast express metabolic and differentiation markers observed in alveolar cells, the milk-producing cells of the mammary gland [
10‐
12]. These molecular associations suggest that tumors can arise from different cell populations and maintain signaling networks of their cell of origin.
Although transforming events within a cell may initiate and drive the process of tumor progression, these events do not necessarily establish a tumor’s histological or molecular fate. Data derived from The Cancer Genome Atlas demonstrate that major oncogenic drivers and tumor-associated mutations are broadly represented among breast cancer subtypes, suggesting that diverse mechanisms contribute to a tumor’s phenotype [
13]. Such mechanisms may include cell-intrinsic networks acquired from the cell of origin. In support of this, several studies have demonstrated that normal cell types can respond to the same oncogenic pathway in unique ways. Ince et al. used different cell culture conditions to enrich for either luminal-like or myoepithelial-like human breast cell populations [
14]. Both populations were subsequently transformed with a common set of oncogenic drivers. When transplanted, each precursor cell population generated a distinct tumor phenotype, with notable differences in tumor histology and metastasis. The myoepithelial-like cells established tumors similar to squamous carcinoma of the breast, whereas the luminal-like cells generated papillary adenocarcinomas [
14]. Other studies with mouse mammary tumor models have also supported a role for the cell of origin in controlling tumor fate. Differences in tumor phenotype were observed when a
Brca1 mutation was induced in different normal cell populations. Conditional
Brca1 loss of function targeted to the basal compartment using
Keratin-14-Cre Brca1
fl/fl
p53
+/−
mice established tumors that were primarily adenosquamous carcinomas and adenomyoepitheliomas. In contrast, disruption of
Brca1 in luminal progenitors using
Blg-Cre Brca1
fl/fl
p53
+/−
mice resulted in ductal carcinomas with a basal-like molecular subtype [
15]. These studies suggest that intrinsic differences between cell populations may influence the histopathology of the tumors they generate.
The polyomavirus middle T antigen (
PyMT) oncogene has been used extensively in mice to model breast cancer [
16,
17]. In these models,
PyMT drives transformation of MECs by signaling through several pathways, including
Src,
Ras, and phosphoinositide 3-kinase [
18‐
21], resulting in a phenotype similar to ErbB2/Neu-induced tumors [
7,
22,
23]. Transgenic mice that express
PyMT under control of the promoter derived from the long terminal repeat (LTR) of the mouse mammary tumor virus (MMTV) develop mammary tumors that undergo progressive transition from precancerous lesions to late-stage malignant tumors and exhibit a high frequency of metastasis [
16,
24,
25]. Tumor progression is marked by a loss of both myoepithelial cells and ER+ luminal cells [
24], and a concomitant expansion of cells expressing the luminal progenitor marker CD61 [
26]. Through intrinsic gene set analysis and hierarchical clustering of gene expression profiles, MMTV-
PyMT tumors have been classified within the luminal subgroup [
7,
9]. In addition, a close association between the molecular signature derived from luminal progenitors and MMTV-
PyMT tumors has been described [
9]. Similar to other mouse models that function through ErbB/Ras signaling proteins, the predominant tumor histology in the MMTV-
PyMT model is solid adenocarcinoma. However, varied histopathology is observed, with approximately 30 % of tumors having papillary, glandular, or acinar features and 10 % exhibiting either adenosquamous, pilar, or type P histology [
22].
These data demonstrate that the MMTV-
PyMT model can establish tumors with both a luminal phenotype and diverse histopathology. It is unknown whether these characteristics are a result of the activity of the MMTV LTR within a particular cell type or through unique transforming activity of the PyMT oncogene. Because the MMTV LTR is widely active in mammary epithelium and drives expression at early stages of postnatal development [
27‐
29], it is difficult to identify the cell of origin for tumors in this model. In order to assess how cellular context affects tumor progression, the
PyMT oncogene has been targeted to various MEC populations by uncoupling expression of the oncogene from the MMTV LTR. For example, virus-based approaches have been used to express
PyMT either ubiquitously in all mammary cell populations or specifically within distinct cell types, and these studies have shown that tumor histology and molecular subtype can vary as a result of the targeting approach [
12,
30,
31]. The restricted expression of
PyMT in the keratin 6 (K6) mammary cell population resulted in tumors with predominantly papillary and cystic histology and a distinct normal-like molecular subtype, representing a phenotype dissimilar to what has been observed in the MMTV-
PyMT mouse model [
27]. In contrast, non-cell-type–specific expression of
PyMT in MECs using the ubiquitous elongation factor 1 alpha (EF1α) promoter generated tumors with diverse histology, including a high frequency of adenosquamous carcinomas and the occurrence of a unique lipid-rich carcinoma [
12]. In addition, this model produced tumors that classified within both luminal and basal molecular subgroups [
12]. In the present study, we extended upon these studies by evaluating the latency, histopathology, molecular subgroup, and metastatic potential of tumors derived from four different fluorescence-activated cell sorting (FACS)-enriched MEC populations. The results suggest that the originating cell population influences several tumor characteristics and further implicates a cell residing predominantly in the basal and stem cell compartment as the cellular origin for squamous metaplasia.
Methods
Mice
FVB/NJ mice were obtained from The Jackson Laboratory (Bar Harbor, ME, USA) and maintained in a pathogen-free facility. The University of Utah Institutional Animal Care and Use Committee approved mouse handling and procedures.
Generation of mouse mammary tumors
MECs were collected from 8–10-week-old FVB/NJ mice as described previously [
12]. Then, freshly isolated MECs were infected with EF1α-
PyMT-ZsGreen lentivirus overnight at 37 °C as described previously [
12]. Following infection, cells were washed five times with Hanks’ balanced salt solution (HBSS; Gibco/Thermo Fisher Scientific, Grand Island, NY, USA) and incubated with 0.05 % trypsin-ethylenediaminetetraacetic acid (EDTA; Gibco/Thermo Fisher Scientific, Grand Island, NY, USA) to isolate single cells. Trypsin was inactivated with MEC media [
12], and cell clumps were removed by straining MECs through a 40-μm cell strainer (Falcon; Fisher Scientific, Pittsburgh, PA, USA). MECs were then resuspended in wash buffer (HBSS + 2 % fetal bovine serum; HyClone Laboratories/GE Healthcare, Logan, UT, USA) and kept on ice for antibody staining and FACS.
Staining consisted of six tubes: (1) no antibody control, (2) CD24-V450 control, (3) CD49f-phycoerythrin (CD49f-PE) control, (4) CD133-allophycocyanin (CD133-APC) control, (5) 7-amino-actinomycin D (7-AAD) control, and (6) CD24-V450/CD49f-PE/CD133-APC/7-AAD sample. During antibody staining, control tubes contained 5 × 10
4 cells and the sample tube contained 20 × 10
6 cells resuspended in 200 μl of wash buffer. All antibodies were obtained from BD Pharmingen (San Diego, CA, USA) and were used at a 1:100 dilution. After the primary antibodies were added, cells were incubated on ice for 15 minutes. Following incubation, cells were washed with 1 ml of wash buffer and centrifuged at 1000 ×
g for 2 minutes. Stained MECs were then resuspended in wash buffer and sorted into luminal CD133+, luminal CD133−, basal, and stem cell populations as described previously [
32‐
35] on a BD FACSAria Cell Sorter using FACSDiva version 6.1.3 software for analysis (BD Biosciences, San Jose, CA, USA). Isolated MEC populations were kept on ice until transplantation.
For each transplantation, 1 × 10
5 untransduced and unsorted MECs were mixed with 2 × 10
4 transduced luminal CD133+, luminal CD133−, basal, or 5 × 10
3 stem enriched MECs. MECs were then resuspended in 10 μl of Matrigel (BD Biosciences, San Jose, CA, USA) by transplantation, and the Matrigel cell mixture was injected into the fourth cleared inguinal mammary fat pad of 3-week-old FVB/NJ mice. Only a single fat pad was injected per mouse. Tumor growth was monitored, and tumors were collected upon reaching 2 cm in diameter. Once tumors were harvested, viable cells were collected using the same protocol for MEC isolation and then frozen in freeze media as described previously [
12]. Portions of the tumors were also flash-frozen for RNA isolation using a Qiagen RNeasy kit (Qiagen, Valencia, CA, USA), and additional tumor fragments were processed for paraffin embedding.
Infection, cell sorting, and transplantation experiments were performed over two rounds. Each time, 10 transplants were performed per sorted MEC population, for a total 20 transplants per group.
Antibody staining and histology
Portions of transduced and FACS-sorted MECs were used to quantify basal and luminal cell enrichment. For each isolated population, 1 × 104 MECs resuspended in 200 μl of wash buffer were centrifuged (Cytospin 4 Cytocentrifuge; Thermo Scientific, Hudson, NH, USA) onto slides (Shandon Cytoslide; Thermo Scientific, Hudson, NH, USA) at 900 rpm for 10 minutes. Cytospun cells were then incubated with fixative solution (4 % paraformaldehyde in phosphate-buffered saline [PBS]) for 15 minutes and washed five times with PBS for 5 minutes each. Fixed cells were permeabilized for 10 minutes with 0.2 % Triton X-100 (Sigma-Aldrich, St. Louis, MO, USA) in PBS, washed with 1 % bovine serum albumin (BSA; EMD Millipore, Billerica, MA, USA) in PBS, and blocked with 1 % BSA in PBS for 10 minutes. Cells were then incubated with primary antibodies against keratin 14 (K14, 1:400 dilution, rabbit, PRB-P-100; Covance, Princeton, NJ, USA) and keratin 8 (K8, 1:50 dilution, rat, TROMA-I; Developmental Studies Hybridoma Bank, Iowa City, IA, USA) for 1 h at room temperature. Following incubation, slides were washed with 1 % BSA in PBS and stained with 4′,6-diamidino-2-phenylindole dihydrochloride and secondary antibodies Alexa Fluor 594 chicken anti-rat immunoglobulin G (IgG, 1:1000 dilution; Invitrogen/Life Technologies, Carlsbad, CA, USA) and Alexa Fluor 488 goat anti-rabbit IgG (1:1000 dilution; Invitrogen/Life Technologies, Carlsbad, CA, USA).
Paraffin-embedded tumor samples were processed and subjected to hematoxylin and eosin (H&E) staining, ESR1 immunohistochemistry (IHC), and cytokeratin staining as described previously [
12]. The following primary antibodies were used: K14 (1:400 dilution, rabbit, PRB-P-100; Covance), K8 (1:50 dilution, rat, TROMA-I; Developmental Studies Hybridoma Bank, Iowa City, IA, USA), and ESR1 (1:200 dilution, MC-20, sc-542; Santa Cruz Biotechnology, Santa Cruz, CA, USA). Secondary antibodies included Alexa Fluor 594 chicken anti-rat IgG (1:1000 dilution; Invitrogen/Life Technologies, Carlsbad, CA, USA), Alexa Fluor 488 goat anti-rabbit IgG (1:1000 dilution; Invitrogen/Life Technologies, Carlsbad, CA, USA), and biotin-SP-conjugated protein (1:1000 dilution; Jackson ImmunoResearch Laboratories, West Grove, PA, USA).
All immunofluorescence imaging was performed on an Olympus IX81 microscope (Olympus, Tokyo, Japan) using a Hamamatsu Photonics ORCA-ER camera (Hamamatsu Photonics, Hamamatsu City, Japan). Fluorescence image recording and processing were performed using SlideBook 64 version 5.0.0.24 software (Intelligent Imaging Innovations, Denver, CO, USA). Slides processed for IHC and H&E staining were imaged on an Olympus BX50 microscope with a Canon EOS Rebel XSI camera using EOS imaging software (Canon, Melville, NY, USA). Any changes in contrast and brightness were performed using Photoshop CS4 software (Adobe Systems, San Jose, CA, USA) on entire images to enhance appearance without altering image content.
In vitro and in vivo assessment of tumor growth with estrogen receptor inhibition
For in vitro dose–response studies, tumor cells and MECs were isolated using the same tissue dissociation protocol as described above. Single cells were suspended in Matrigel (BD Biosciences, San Jose, CA, USA) and 10 μl of a cell/Matrigel mixture was plated per well in a 96-well plate (Costar; Corning Life Sciences, Oneonta, NY, USA). 4-Hydroxytamoxifen (Sigma-Aldrich, St. Louis, MO, USA) was dissolved in ethanol and serially diluted in MEC media. Cells were dosed with 100 μl of media containing drug or vehicle control for 48 h. Media with drug or vehicle control was refreshed every 24 h. Each drug concentration was tested in triplicate. Cell viability was measured using the ATPlite assay (PerkinElmer, Waltham, MA, USA) according to the manufacturer’s instructions and normalized to vehicle control.
Tumor growth dependence on estrogen was tested in vivo by conducting an additional primary cell infection, FACS, and transplantation experiments as described above. Ovaries were removed from all mice that received cell transplants, as described previously [
36]. Ten surgeries were performed per infected and sorted population.
Microarray analysis
Flash-frozen tumors were randomly selected from each of the tumor groups for RNA extraction and microarray analysis. Total RNA was isolated using the Qiagen RNeasy kit.
To perform supervised hierarchical clustering, all steps of microarray processing, data filtering and normalization, and analysis were performed as described previously [
12]. Batch adjustment was performed in two batches. A dataset generated by Herschkowitz et al. was treated as one batch (Gene Expression Omnibus [GEO] accession number [GEO:GSE3165]) [
7], and data generated at the Huntsman Cancer Institute (HCI) was treated as a second batch. The HCI microarray dataset has been deposited in the National Center for Biotechnology Information GEO database under accession number [GEO:GSE64453].
To perform unsupervised hierarchical clustering, quantile-normalized, log-scaled microarray intensity data were hierarchically clustered in R using Ward’s method. The resulting dendrogram revealed a marked batch effect related to the date on which the arrays were processed. Each class of samples (luminal CD133−, luminal CD133+, basal, and stem) were adjusted for batch effect separately using the ComBat procedure [
37] and then recombined. Differential expression analysis comparing the luminal CD133− samples with the remaining samples identified 1111 microarray probes representing 1046 unique genes showing at least twofold differential expression at an adjusted
p value <0.05 (
p value by
t test with adjustment using the Benjamini-Hochberg method).
As described above, transplantation experiments were performed twice for each transduced and enriched MEC population. Lung metastasis for the first round was assessed by H&E staining. Lung tissue processing and staining were performed as described above. Paraffin-embedded lungs were serially sectioned at 10 μm, and every fifth slide was stained and examined for metastasis. For the second round of transplants, lung metastases were analyzed by fluorescence imaging after lungs were flattened between two glass slides. Slides were imaged as described above, and numbers of unique metastatic sites and tumor areas were quantified using ImageJ software (National Institutes of Health, Bethesda, MD, USA). Prevalence of lung metastases and numbers of metastatic foci were consistent over two rounds of transplants.
To quantify circulating tumor cell (CTC) numbers, fresh whole blood was collected by cardiac puncture immediately after mice were killed according to University of Utah–approved Institutional Animal Care and Use Committee procedures. CTCs were isolated for FACS analysis as described previously from mice bearing primary tumors, as well as from no-tumor control mice [
38]. CTCs expressing ZsGreen were detected by analyzing cells using a FACScan cytometer (BD Biosciences, San Jose, CA, USA), and results were quantified using FlowJo software (Tree Star, Ashland, OR, USA). Owing to low numbers of CTCs present within isolated whole blood, the ZsGreen-positive threshold was set at 0.05 % of no-tumor control background signal. This threshold was then used as a baseline for detecting CTCs in tumor-bearing mice. All CTC values were then normalized to no-tumor control background signal.
Tail vein injections were performed to assess the ability of tumor cells to colonize the lungs after introduction into the bloodstream. Single cells were isolated from primary tumors using the same procedure employed for MEC isolation. Cells were then resuspended in HBSS at 10 × 106 cells/ml. A quantity of the HBSS/cell mixture (250 μl; 2.5 × 105 cells) was injected into the lateral tail veins of 8–12-week-old FVB/NJ mice. Cells isolated from individual tumors were injected into five mice each. Twenty days postinjection, mice were killed and tumor lung foci numbers were quantified by fluorescence imaging with ImageJ software.
Analysis of PyMT expression in transduced mammary epithelial cells
Freshly isolated MECs were infected in suspension overnight and then transferred to adherent culture in MEC media the following day. Separate plates of cells were analyzed for ZsGreen and PyMT expression each day for 5 days. Protein lysates were prepared for simple Western blot analysis to detect PyMT expression, and cells were analyzed for ZsGreen expression by flow cytometry. For cytometry, cells were washed with PBS, lifted with 0.25 % trypsin-EDTA (Gibco/Thermo Fisher Scientific), fixed in 2 % paraformaldehyde, and resuspended in wash buffer on ice. Single cells were isolated by straining them through a 40-μm cell strainer (Falcon; Fisher Scientific) and immediately analyzed with a BD LSR II flow cytometer (BD Biosciences, San Jose, CA, USA) for ZsGreen expression.
Real-time quantitative PCR analysis
RNA was isolated from freshly sorted MECs or flash-frozen tumors using the Qiagen RNeasy Mini Kit. For cDNA synthesis from tumor RNA, 1 μg of total RNA was reverse-transcribed using the iScript Reverse Transcription Supermix for real-time quantitative PCR (RT-qPCR) synthesis kit (Bio-Rad Laboratories, Hercules, CA, USA) according to the manufacturer’s instructions. For sorted MECs, cDNA was synthesized from 20 ng of total RNA via reverse transcription, followed by preamplification with pooled PrimeTime assays (Integrated DNA Technologies, Coralville, IA, USA) for genes of interest using the Qiagen RT
2 PreAMP cDNA Synthesis Kit. Genes investigated were
K8 (assay identification: Mm.PT.58.6862465),
K14 (Mm.PT.58.43652691),
progesterone receptor (
Pgr; Mm.PT.58.10254276), Esr1 (Mm.PT.58.8025728),
p63 (Mm.PT.58.13970687),
Slug (Mm.PT.58.43645779),
c-Kit (Mm.PT.56a.33701407), and
glyceraldehyde 3-phosphate dehydrogenase (
Gapdh; Mm.PT.39a.1) for reference. RT-qPCRs were performed in 20-μl volumes using 2× SYBR Green PCR Master Mix (Applied Biosystems, Foster City, CA, USA) in a Roche LightCycler96 system (Roche Applied Science, Indianapolis, IN, USA). The primers for
PyMT and the reference gene,
Rplp0, have been described previously [
12]. Each sample was normalized to the reference gene, sorted MEC cycle threshold change (ΔC
T) values were compared with a control population of unsorted MECs, and relative fold induction level for each sorted population/tumor was calculated using the 2
−ΔΔCt method [
39].
Simple Western blot analysis
Using a PowerGen 125 sawtooth homogenizer (Fisher Scientific), protein lysates were prepared from flash-frozen tumors using Pierce IP Lysis Buffer (25 mM Tris∙HCl, pH 7.4, 150 mM NaCl, 1 % Nonidet P-40, 1 mM EDTA, 5 % glycerol; Pierce Biotechnology/Thermo Scientific, Rockford, IL, USA) with 10 mM sodium pyrophosphate tetrabasic (Sigma-Aldrich, St. Louis, MO, USA), 1 mM sodium orthovanadate (Sigma-Aldrich, St. Louis, MO, USA), Phosphatase Inhibitor Cocktails 2 and 3 (Sigma-Aldrich, St. Louis, MO, USA), and Mammalian ProteaseArrest cocktail (G-Biosciences, St. Louis, MO, USA). Lysates from cultured cells were prepared without pyrophosphate tetrabasic and sodium orthovanadate. Protein was quantified using a DC Protein Assay Kit (Bio-Rad Laboratories, Hercules, CA, USA) and a DU 730 UV/VIS spectrophotometer (Beckman Coulter, Fullerton, CA, USA).
Simple Western blot analyses were performed as instructed by the ProteinSimple user manual. Briefly, cell lysates were mixed with a master mix (ProteinSimple, San Jose, CA, USA) to a final concentration of 1× sample buffer, 1× fluorescence molecular weight marker, and 40 nM dithiothreitol. Following 5-minute denaturation at 95 °C, the samples, blocking agent, primary antibodies, horseradish peroxidase–conjugated secondary antibody, and chemiluminescence substrate were dispensed into the designated wells of the 384-well plate. Automated separation electrophoresis and immunodetection were performed using a ProteinSimple WES instrument. All antibodies were diluted in the antibody diluent II (ProteinSimple, San Jose, CA, USA) and incubated with the protein for 10–15 minutes. A luminol-peroxide mixture (ProteinSimple, San Jose, CA, USA) was used to generate chemiluminescence, which was captured with a charge-coupled device camera. The resulting digital image was analyzed with Compass software (ProteinSimple, San Jose, CA, USA), and quantified data were reported as molecular weight, signal and peak intensity, and area under the curve.
The following antibodies used were obtained from Cell Signaling Technology (Danvers, MA, USA): β-actin (13E5, 1:50 dilution, catalog number 4970P), AKT (C67E7, 1:50 dilution, catalog number 4691P), extracellular signal-regulated kinase 1/2 (ERK1/2; 137 F5, 1:50 dilution, catalog number 4695P), SRC (36D10, 1:100 dilution, catalog number 2109S), phospho-AKT (Thr308, C31E5E, 1:25 dilution, catalog number 2965S), phospho-ERK1/2 (Thr202/Tyr204, dilution 1:50, catalog number 4370P), and phospho-SRC (Tyr527, 1:50 dilution, catalog number 2105P). The following antibodies were obtained from Santa Cruz Biotechnology(Santa Cruz, CA, USA): hemagglutinin probe (Y-11, 1:12.5 dilution, catalog number sc-805) and GAPDH (FL-335, 1:1500 dilution, catalog number sc-25778).
Statistical analysis
All data analyses were performed in using GraphPad Prism 6.0d software (GraphPad Software, La Jolla, CA, USA). For each analysis, specific statistical tests are indicated in the figure legends.
Discussion
The
PyMT oncogene has a broad capability to establish a variety of tumor histologies and subtypes. Members of the Li laboratory developed a model wherein a modified avian leukosis sarcoma virus expressing
PyMT (RCAS-
PyMT) was used to infect transgenic mice engineered to express
tva, the RCAS (replication competent avian sarcoma leukosis virus LTR splice acceptor) receptor, on different mammary cell types [
53]. Intraductal infection with RCAS-
PyMT was performed in mice expressing
tva under the control of several promoters, including the MMTV LTR and K6 promoter [
27,
31]. When tumors generated by the different models were compared, dissimilar histological and molecular phenotypes were observed. MMTV-
tva/RCAS-
PyMT tumors were acinar and composed of luminal and myoepithelial cells, a phenotype that contrasted with the papillary tumors generated by K6-
tva/RCAS-
PyMT. These researchers also showed that the molecular profiles of tumors generated by MMTV-
tva/RCAS-
PyMT, K6-
tva/RCAS-
PyMT, and MMTV-
PyMT were not similar, with the K6-
tva tumors having a unique profile similar to that of a normal-like breast cancer subtype [
27]. In an expanded analysis of molecular profiles from the RCAS-
PyMT and MMTV-
PyMT models, Hollern and Andrechek further demonstrated significant diversity, particularly among tumors derived from different mouse strains [
40]. Taken together, these data demonstrate that the
PyMT oncogene has the capacity to generate several tumor subtypes and that differences in the cellular context of the oncogene can affect tumor heterogeneity.
We have similarly observed a variety of histological and molecular subtypes in tumors induced by the
PyMT oncogene. Previously, we showed that broad expression of the oncogene in mammary epithelium, using the EF1α-
PyMT-ZsGreen lentivirus, generated late-stage tumors consisting of both luminal and myoepithelial cells, with a majority of tumors having acinar and solid histology [
12], which are similar features of tumors established by the MMTV-
tva/RCAS-
PyMT model [
53]. However, a notable difference between these models was the appearance of squamous metaplasia, which was observed in 30 % of the EF1α-
PyMT-ZsGreen tumors, but not in MMTV-
tva/RCAS-
PyMT tumors [
12,
27]. This tumor subtype is rare in the MMTV-
PyMT model, with a reported frequency of only 4–8 % [
22,
40]. The data in our present study expand upon those previous studies by demonstrating that squamous metaplasia was observed more frequently when basal and stem cell populations were infected with the EF1α-
PyMT-ZsGreen lentivirus. Similar data were obtained by Keller et al. from studies performed with normal human breast cells. They showed that transduction of CD10-enriched human basal cells with oncogenic lentiviruses resulted in ER− and metaplastic tumors with squamous differentiation. In contrast, epithelial cell adhesion molecule–positive (Epcam+) luminal cells generated ductal carcinomas with luminal features [
54]. In addition, Ince et al. showed that transformation of myoepithelial-like human breast cells generated squamous carcinomas of the breast, whereas luminal-like cells gave rise to adenocarcinomas [
14]. Collectively, these studies provide strong support that a cell within the basal lineage, either a myoepithelial cell or a stem cell, is the origin of squamous metaplasia in mammary tumors.
The ability to fractionate primary mouse mammary cells into luminal, basal, and stem cell populations, and their receptiveness to ex vivo manipulation and outgrowth following transplantation, provides a unique method to study how cellular context affects oncogenesis [
12]. In the present study, we investigated whether tumor latency, histology, metastasis, and molecular subtype are altered when the
PyMT oncogene is targeted to distinct FACS-enriched mammary cell populations. The data demonstrate that each MEC population is able to generate tumors at similar latency and with broad pathology. However, some cell populations preferentially established distinct pathologies. Most striking was that luminal CD133+ cells gave rise to a higher proportion of tumors with papillary histology and ESR1 expression and the lowest proportion of tumors with squamous metaplasia. Consistent with their well-differentiated pathology, tumors from luminal CD133+ cells also produced fewer CTCs and metastases. An opposing phenotype was observed in tumors generated by enriched stem cells. Notably, these cells generated tumors that were ER−, exhibited squamous metaplasia, and produced more CTCs and metastases than luminal CD133+ cells. Thus, luminal CD133+ cells had a propensity to produce well-differentiated tumors, whereas poorly differentiated and metaplastic tumors were more commonly observed from enriched stem cells.
Although the MMTV LTR is expressed primarily in luminal cells, it is also active broadly in basal, stem, and luminal cell populations of the mammary gland [
27,
29]. In the MMTV-
PyMT model, the oncogene may target a multipotent progenitor because clonal cell lines derived from precancerous lesions can establish cell types expressing markers for both luminal and myoepithelial populations [
25,
55]. Histological and cellular heterogeneity are also observed in vivo in the MMTV-
PyMT model, particularly during precancerous development. Early lesions are composed of both hormone receptor–positive and hormone receptor–negative cells, with sporadic myoepithelial coverage [
24]. As the tumors develop to malignant lesions, they exhibit a loss of both hormone receptor–expressing cells and myoepithelial cells [
24] and an expansion of cells double-positive for CD61 and CD29 [
26], markers that are coexpressed on normal luminal progenitors. Some similarities between MMTV-
PyMT tumors and those derived from the luminal CD133− cell population are evident. In particular, the histopathology of tumors from this population was most similar to that of tumors in the MMTV-
PyMT mouse model. Solid adenocarcinomas are seen in 60 % of MMTV-
PyMT tumors [
22], and this was the dominant histology in approximately 40 % of tumors derived from the luminal CD133− cell population. However, tumors established by luminal CD133− cells were classified as a basal subgroup, whereas MMTV-
PyMT tumors clustered within the luminal subgroup [
7]. The reason for this discrepancy is unclear, but it may be a result of differences in the developmental stage of cells targeted in each model. The MMTV LTR is active at prepubertal stages of mammary gland development [
29], which suggests that immature mammary epithelium is the origin for MMTV-
PyMT tumors. In contrast, the primary MECs used for FACS enrichment in our study were derived from postpubertal mice. These cells would have had been exposed to the maturation effects of systemic hormones, which may alter their lineage potential. Recent studies have shown that oncogenic stress can significantly increase the plasticity and multilineage potential of differentiated MECs [
56‐
58]. Both the MMTV-
tva/RCAS-
PyMT and EF1α-
PyMT-ZsGreen models generate heterogeneous tumors consisting of myoepithelial and luminal cell lineages, which contrasts with the predominantly luminal cell type that is observed in late-stage MMTV-
PyMT tumors [
59]. These data suggest that tumor cells generated by the MMTV-
PyMT and virus-based models have differences in their lineage plasticity, particularly at more advanced stages of progression, which may influence the molecular classification of tumors. Further studies are necessary to assess the effect of oncogenic stress on the lineage plasticity of MECs at pre- and postpubertal stages of development.
Using the method described by Herschkowitz et al., we categorized tumors derived from the sorted cell populations and 12 different mouse models of breast cancer into 2 general molecular subgroups: basal and luminal [
7,
60]. These data demonstrate that both luminal and basal tumor subgroups can arise from enriched luminal CD133+, stem, and basal cell populations. However, we also show that transformation of CD133− luminal cells, which are enriched for luminal progenitors [
34], generated tumors of the basal subgroup. This finding suggests luminal progenitors preferentially establish basal rather than luminal tumors. Consistent with this, several recent observations attribute basal-like breast cancer to a luminal population. Lim et al. demonstrated that the molecular profile of untransformed luminal progenitors most closely resembles basal-like breast cancers [
8]. In addition, transformation of human EpCAM+/CD10−/CD49f + luminal progenitors derived from reduction mammoplasties established tumors with features similar to basal-like breast cancer, including reduced ESR1 and greater K14 expression than tumors derived from differentiated luminal cells [
54]. Furthermore, targeting
Brca1 loss of function to luminal cells in mice generated tumors with basal-like features that closely resembled those observed in patients carrying the
BRCA1 mutation. However, the same loss of function in basal cells generated adenomyoepitheliomas [
10,
15]. Taken together, these data support luminal progenitors as a cellular origin of basal-like breast cancer.
There are a number of challenges to studying how the cell of origin influences tumor progression in breast cancer. These include the complexity of normal mammary cell types from which a tumor may arise [
60] and technical difficulties in precisely targeting oncogenes to specific cell types. As such, the data from this study should be considered in the context of limitations of the approach used to target expression of
PyMT to distinct cell populations. For example, the cells that were transplanted represent enriched populations but not purified cell types. Thus, a tumor phenotype cannot be credited to a specific cell type. In addition, specific cell populations may have a transduction bias, as has been observed between cultured luminal and myoepithelial cells, which may limit the diversity of cell types targeted by the lentivirus [
61]. Transduction of primary MECs with lentiviruses and the subsequent expression of
PyMT and
ZsGreen may also alter the sorting process. However, maximum expression of lentiviral transgenes generally takes several days. ZsGreen expression was observed in only about 2 % of cells 24 h after transduction, and PyMT protein expression was low at this time. Because cell sorting and transplantation were performed within 24 h after the cells were exposed to virus, it is unlikely that expression of
PyMT would have any significant influence on the populations before the sort. In addition, the study was designed to ensure each sorted population was exposed to similar conditions during the ex vivo procedures. Primary cells were collected, transduced, and sorted by FACS from a pool, and each population was cotransplanted with 5–20-fold more uninfected, unsorted MECs, which would provide comparable in vivo environments for tumor development. Thus, the main variable was the cell population used to generate the tumors.
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
DD participated in study design and performed experiments, analyzed data, and drafted the manuscript. BAS participated in study design, experiments, data analysis, and edited the manuscript. MM and KPG designed and performed RT-qPCR and protein analysis experiments and edited the manuscript. HAE performed tail vein injections and lung metastasis quantification and contributed to editing of the manuscript. BEW participated in study design and coordination and helped draft and edit the manuscript. All authors read and approved the final manuscript.