Introduction
A hallmark of most solid tumors, hypoxic regions are associated with resistance to radiation and chemotherapy [
1,
2]. O
2 tension in advanced breast cancers can be as low as 0.1% to 1% O
2 [
3], a range commonly used to model tissue hypoxia
in vitro. The oxygen-responsive hypoxia-inducible factor 1α (HIF-1α) protein, a master regulator of the hypoxic response [
4], is overexpressed in a variety of carcinomas and their metastases, including breast cancer [
5]. A majority of ductal carcinoma
in situ and almost all poorly differentiated, invasive breast carcinomas overexpress HIF-1α [
6]. Moreover, HIF-1α promotes multiple steps of the metastasis program [
7]. Either the overexpression of HIF-1α protein or the enrichment of a hypoxic gene signature in the primary tumor correlates with poor prognosis and decreased survival in breast cancer patients [
8,
9].
Breast cancer cells that exhibit properties of cancer stem cells (CSCs), also referred to as tumor-initiating cells (TICs), are more resistant than bulk tumor cells to therapeutic intervention, including radiation and DNA-damaging drugs [
10‐
12]. Hypoxic culture promotes self-renewal of several cell types, including neurospheres, hematopoietic stem cells (HSCs) and embryonic stem (ES) cells [
13]. HIF-1α is also necessary
in vivo for HSCs because its deletion causes stem cell exhaustion [
14].
Accumulating evidence supports the hypothesis that stem cells and TICs exist in a hypoxic niche microenvironment [
13]. The direct relationship of the hypoxic response to TIC activity has been demonstrated in adult glioma, human acute myeloid leukemia (AML) and murine lymphoma [
15,
16]. In gliomas, expression of
HIF2A mRNA was enriched, and knockdown of
HIF2A, but not
HIF1A, reduced TIC activity in patient xenografts [
15]. In contrast, HIF-1α was found to be essential for maintaining TIC activity in a syngeneic rodent transplant model of lymphoma and in AML patient xenografts through regulation of the Notch pathway [
16]. Interestingly, in gliomas, the TIC population was enriched via cell sorting based on the expression of a single cell surface marker, CD133 (PROM1). PROM1, a transmembrane protein without a known ligand, is a hypoxia-responsive protein regulated by HIF-1α [
17,
18].
Several studies have shown that the population of breast tumor cells with the ability to self-renew is enriched with the ability to initiate tumorigenesis
in vivo [
19‐
22]. Furthermore, TICs may drive metastasis [
23,
24]. The mouse mammary tumor virus polyoma virus middle T (MMTV-PyMT) mouse model is one of the most commonly utilized preclinical mouse models in breast cancer research because tumor latency is short and there is a high frequency of metastasis to the lung [
25]. Whole-genome array profiling indicates that PyMT tumors most closely resemble the luminal B subtype of human breast cancer [
26], although end-stage PyMT tumors are estrogen and progesterone receptor (ER and PR)-negative [
25].
Moreover, the specific contribution of HIF-1α in regulating TICs in breast cancer remains undefined, particularly in the context of syngeneic rodent models that recapitulate the breast cancer microenvironment. It has previously been shown that conditional deletion of
Hif1a in PyMT+ tumors by crosses to mice expressing MMTV-Cre delayed the onset of palpable tumors, delayed the progression of hyperplasias to carcinomas and reduced lung metastases [
27]. We have now created an improved model system in which
Hif1a is efficiently deleted in the mammary tumor epithelium via
ex vivo viral transduction with Cre recombinase prior to the injection of mammary tumor cells to the cleared fat pads of recipient mice. Validating the use of a transplantation paradigm in lieu of intact transgenic mice, previous studies have shown that when isolated PyMT tumor cells are serially regenerated as tumors in recipient mice the morphology and gene expression profiles are similar to the tumors found in the transgenic mouse [
28].
Using these novel HIF-1α wild-type (WT) and HIF-1α-null (KO) mammary tumor epithelial cells (MTECs), we demonstrate that HIF-1α prominently augments primary tumor growth and lung metastasis and accelerates relapse due to metastasis. Furthermore, we show for the first time that HIF-1α promotes mammary tumorsphere formation and enhances TIC frequency in vivo, in part by regulation of the expression of markers associated with the basal lineage, the Notch pathway and CD133. Together these data indicate that suppressing the hypoxic response may be beneficial not only by reducing primary tumor mass but also by suppressing the breast TIC subpopulation that may ultimately be responsible for patient relapse.
Materials and methods
Animals
Mice harboring two alleles of exon 2 of
Hif1a flanked by
loxP sites (double-
floxed, DF) were provided by Dr Randall Johnson (University of California San Diego) on a mixed genetic background (129Sv-C57BL/6 [
29]).
Hif1a stock mice were first backcrossed to the FVB/Nj strain (The Jackson Laboratory, Bar Harbor, ME, USA) for 11 generations prior to being bred to MMTV-PyMT transgenic mice obtained from Dr Kent Hunter (National Cancer Institute, Frederick, MD, USA), which had previously been backcrossed to the FVB/Nj strain. Lung metastasis induced by the PyMT transgene is highly penetrant in the FVB/Nj background [
30]. All procedures were approved by Institutional Animal Care and Use Committee at the University of Tennessee Health Science Center.
Establishing Hif1a wild-type and knockout mammary tumor epithelial cells
Several mammary tumors (> 500 mm
3) were isolated from
Hif1a DF, PyMT+ bigenic female mice. Tumors were chopped with scalpels and then with razor blades, and the paste was digested with 1 mg/ml collagenase type III (Worthington Biochemical Corp, Lakewood, NJ, USA) in RPMI media containing 5% fetal bovine serum (FBS) (5 ml/g tissue) for 2 hours at 37°C. Organoids were pelleted at 1,100 rpm, washed four times with digestion buffer and then plated into standard tissue culture plates in plating medium as described previously [
31]. After 48 to 72 hours, the plating medium was switched to complete mammary epithelial cell growth medium (GIBCO DMEM: Nutrient Mixture F-12 (DMEM/F-12; Invitrogen, Carlsbad, CA, USA), 5% FBS, 5 μg/ml insulin (Sigma-Aldrich, St Louis, MO, USA), 10 ng/ml recombinant murine epidermal growth factor (EGF) (Invitrogen)). At passage 6, MTECs were transduced with either adenovirus β-galactosidase (adeno-β-gal) or adeno-Cre at a multiplicity of infection (moi) of 80 plaque-forming units (pfu)/cell to generate WT and KO MTECs, respectively. Adenoviral transduction was repeated, and the deletion efficiency between WT and KO MTECs was confirmed by both quantitative RT-PCR (qRT-PCR) and Western blot analysis using the reagents presented in Additional file
1 Tables S1 and S2. After adenoviral transduction, MTECs were weaned to medium containing only 2% FBS (DMEM/F-12 + 2% FBS).
For subcultivation, cells were rinsed twice with Puck's A saline, then incubated for up to 60 minutes at 37°C in a 3:1 solution of dispase II/0.25% trypsin reconstituted in Puck's A. No ethylenediaminetetraacetic acid (EDTA) was utilized to subcultivate cells because treating cells with trypsin-EDTA changed tumor cell morphology from an epithelial (cuboidal) to a mesenchymal-like (spindle) appearance. All cells were passaged less than 30 times before use in tumorsphere or in vivo assays. Spent media were routinely tested for mycoplasma using the MycoAlert Kit (Lonza, Basel, Switzerland).
All cells were grown either at normoxia in an air-jacketed CO2 incubator (5% CO2; SANYO, Wood Dale, IL, USA) or at hypoxia (0.5% O2, 5% CO2) in a multigas incubator (SANYO) in which N2 gas displaces O2. Cells were exposed acutely (≤6 hours) or chronically (> 6 hours to several days) to hypoxia and were removed from chronic hypoxic exposure only for brief periods to change the media.
For immunostaining of cultured cells, WT or KO cells passaged with a dispase-trypsin mixture were plated into tissue culture-treated chamber well slides (BD Biosciences, Franklin Lakes, NJ, USA), grown to 60% to 80% confluence and then postfixed for 20 minutes at room temperature with 4% paraformaldehyde (PFA)-PBS, followed by permeabilization with 0.5% Triton X-100 for 5 minutes. Cells stained with CD133 were not permeabilized. Primary antibodies to CD133, Troma-I (K8), cytokeratin 14 (K14) and cytokeratin 5 (K5) were incubated overnight at 4°C using the dilutions listed in Additional file
1 Table S2. All cells were stained with 4',6-diamidino-2-phenylindole (DAPI) prior to being mounted with VECTASHIELD Mounting Medium (Vector Laboratories, Burlingame, CA, USA).
Western blot analysis
Insoluble material remaining after preparation of whole-cell extract (WCE) was reextracted in high-salt (HS) solution (400 mM NaCl) buffer as described previously [
32], except that the deubiquitinase inhibitor
N-ethylmaleimide (NEM) was added to a final concentration of 0.5 μM. HS-WCE was resolved on 3% to 8% Tris-acetate gels (1 to 10 μg/lane; Invitrogen) and transferred onto polyvinylidene fluoride membranes prior to blocking with 5% milk and enhanced chemiluminescence-based detection of antibody complexes. All antibodies and dilution factors are listed in Additional file
1 Table S2.
Gene expression
Total RNA was prepared using RNA-Bee RNA isolation reagent (amsbio, Lake Forest, CA, USA), and RNA quality was confirmed by the Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA) assay. RNA with an RNA integrity number > 9.0 was used to prepare cDNA using the High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems, Foster City, CA, USA). qRT-PCR was performed using optimized primer and 6-carboxyfluorescein-labeled probe sets designed using Universal ProbeLibrary Assay Design Center software (Roche Applied Science, Indianapolis, IN, USA) as described previously [
33]. To control for cDNA input (40 to 80 ng/reaction) when using cultured cells as the source of RNA, the crossing point (
Cp) values were normalized based on the expression of the integrator complex subunit 3 (
Ints3) gene, which is expressed at moderate levels in the mammary gland. This gene changes < 20% between WT and KO cells cultured at normoxia or hypoxia and in WT and KO tumors by Illumina whole-genome expression arrays (Illumina Inc, San Diego, CA, USA) (TNS, personal observations). To compensate for any changes in epithelial content in whole tumors between genotypes, because only the tumor epithelium is deleted for
Hif1a, Ints3-normalized
Cp values were also normalized to
Krt18 (cytokeratin 18).
Cell growth and invasion assays
MTECs were grown in normoxic culture (ambient air; 5% CO2) or hypoxic culture (0.5% O2, 5% CO2) in medium buffered with 25 mM 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES). On the day before enumeration, either 350,000 cells/well in 6-well format (complete growth medium) or 100,000 cells/well in 12-well format (2% FBS medium) were seeded into multiwell plates. In all cases, the medium was changed postplating, but was not replenished for the duration of the experiment. Cells were harvested after culture for 0, 24, 48, 72 or 96 hours of culture at normoxia and hypoxia. All cells were plated in triplicate or quadruplicate per genotype/oxygen tension/time point. Each replicate was counted by hemacytometer, and counts were verified using an Accuri personal flow cytometer after gating against cell debris (BD Biosciences).
For invasion assays, WT and KO MTECs that had been gradually weaned to medium supplemented with 0.5% FBS were cultured overnight in serum-free DMEM/F-12 medium. The next day 25,000 cells were plated onto control inserts or Matrigel-coated transwell inserts (BD Biosciences) and attractedto wells containing complete growth medium with 5% FBS. Cells were plated in triplicate per genotype/oxygen tension. The mean cell invasion index corrected for random migration was calculated after 48 hours according to the manufacturer's instructions. Changes in invasion are expressed as a fold changes relative to the invasion index observed for WT cells cultured at normoxia (fold change = 1.0).
Mammary tumor epithelial cell transplant into FVB recipients
MTECs dissociated into single cells with 0.05% trypsin-EDTA were counted using a hemacytometer and diluted into HBSS. When transplanted into recipients at relatively low density (≤500 cells/gland), cells were diluted 1:1 (vol:vol) with growth-factor reduced Matrigel-Hank's balanced salt solution (HBSS). At higher densities, cells were resuspended in HBSS alone. Cells were kept on ice until injection into the right inguinal mammary fat pads (10 μl) of 3-week old female FVB/Nj recipients (The Jackson Laboratory) using a 26-gauge PT2 needle mounted on a Hamilton syringe, followed by clearing of the endogenous epithelium. Recipients were palpated one or two times per week, and outgrowths were measured with digital calipers to calculate tumor volume as described previously [
27].
Tissue histology and immunostaining
Tumors were harvested from anesthetized mice and flash-frozen in liquid nitrogen for preparation of RNA or protein, or they were fixed in 10% neutral buffered formalin (NBF) for 6 hours at room temperature for histological staining (H & E) and immunostaining. Paraffin-embedded sections (5 to 7 μm) were immunostained after antigen retrieval (1 × citrate buffer) using the primary antibodies listed in Additional file
1 Table S2, followed by development with a VECTASTAIN Elite ABC Kit and ImmPACT DAB (diaminobenzidine) substrate (Vector Laboratories). Alternatively, to prepare frozen sections suitable for CD133, Troma I, K14 and K5 immunofluorescent staining, anesthetized mice were perfused intracardially with 10% NBF. Tumor tissue was postfixed for 10 minutes in NBF at room temperature prior to cryoprotection overnight at 4°C in 30% sucrose-PBS, embedding in OCT medium and preparation of 10 μm sections by cryostat. Sections used for visualizing the keratins were postfixed for an additional 10 minutes in 10% NBF prior to 0.5% Triton X-100 permeabilization. All antibodies and staining conditions are listed in Additional file
1 Table S2.
Lungs were either harvested from recipients at the same time as primary tumors (at a volume of approximately 1,000 mm3), 8 weeks after primary tumor resection (500 to 750 mm3) or when recipients subjected to tumor resection were moribund as indicated by body weight loss > 15% and panting. Lungs were inflated with 10% NBF and postfixed in NBF overnight. Paraffin-embedded sections (7 μm) representing every 100 μm of lung tissue were obtained for each paraffin block, and all sections were stained with H & E. The individual H & E-stained section containing the highest number of metastases per recipient, as evaluated by counting each slide under a light microscope at × 50 original magnification, was used to determine the grand mean of metastases per genotype.
Tumorsphere culture and immunostaining
WT or KO MTECs were briefly trypsinized, washed and strained (40 μm filter) to obtain single cells. The presence and viability of single cells were verified by viewing trypan blue-stained cells using a hemacytometer, and only cells with > 90% viability were used in sphere assays. Single cells were resuspended in serum-free DMEM-F-12 mammosphere media containing 20 ng/ml mouse recombinant EGF, 20 ng/ml basic fibroblast growth factor, 1 × B27 (all from Invitrogen) and 4 μg/ml heparin (Sigma-Aldrich, St Louis, MO, USA) as described previously [
34]. Primary tumorspheres were derived by plating 30,000 single cells/well into six-well ultra-low-adhesion dishes. Secondary and tertiary tumorspheres were plated at 5,000 to 10,000 cells/well and 2,000 cells/well, respectively. Dishes were cultivated at normoxia or hypoxia (0.5% O
2; 5% CO
2) for 10 to 14 days prior to enumeration of spheres. Individual spheres ≥ 100 μm from each replicate well (
n ≥ 12 wells/genotype/oxygen tension) were counted under a dissecting microscope. The percentage of cells capable of forming spheres, termed the "sphere formation efficiency" (SFE), was calculated as follows: [(number of spheres formed/number of single cells plated) × 100]. End point tumorspheres were collected from ultra-low-adhesion dishes, washed with PBS, then flash-frozen for preparation of RNA or dried onto slides for 10 to 15 minutes at 37°C. Slides were postfixed with 4% PFA-PBS for 15 minutes and immunostained with the antibodies indicated in Additional file
1 Table S2. All slides were stained with DAPI prior to being mounted in either ProLong Gold antifade reagent (Invitrogen) or VECTASHIELD Mounting Medium (Vector Laboratories).
Tumor digestion and flow sorting
All mammary gland tumors used for flow-sorting experiments were harvested from intact Hif1a DF, MMTV-PyMT+ (equivalent to HIF-1α WT) transgenic female mice at a size of > 300 mm3 but < 1,500 mm3. Necrotic areas were removed from solid tumor tissue, and solid tissue was cut into small fragments with scissors. Tumor tissue was then weighed and chopped for 5 minutes with scalpels, followed by 5 minutes of chopping to a fine paste with razor blades. The tissue paste was digested in digestion buffer (DMEM/F-12, 1 × gentamicin, 1 × antibiotic-antimycotic (Sigma-Aldrich), 300 U/ml collagenase type III (Worthington Biochemical Corp) and 100 U/ml hyaluronidase (Sigma-Aldrich)) for 1 to 2 hours at 37°C at 125 rpm (10 ml/g tissue). Digested tissue was pelleted by centrifugation, and the red blood cells were lysed with a solution of 0.8% NH4Cl-HBSS. To obtain single cells, organoids were further digested with 0.25% trypsin-EDTA for up to 10 minutes at 37°C and washed with serum-containing medium to inactivate trypsin, followed by a final digestion for 10 minutes at 37°C in 5 mg/ml dispase II (Roche Applied Science)/Puck's A solution containing 1 mg/ml DNase I (Roche Applied Science). The cell suspension was passed through a 40 μm filter insert, and this flow-through was then passed through a flow cytometer tube with a 35 μm filter cap insert to further enrich for single cells. Viability postdigestion was routinely assessed by trypan blue staining. Only cells with viability > 85% were subjected to antibody staining.
Isolated single cells were resuspended to a density of up to 5 million cells/ml in flow buffer (HBSS, 2% FBS, 10 mM HEPES) and incubated for 1 hour on ice with the biotin- or fluorophore-conjugated antibodies listed in Additional file
1 Table S2. These antibodies included a biotin-conjugated anti-mouse hematopoietic lineage panel supplemented with anti-mouse CD31-biotin, anti-mouse CD133-phycoerythrin (CD-133-PE) and anti-mouse CD24-fluorescein isothiocyanate (CD24-FITC), each at a 1:100 dilution. Secondary antibody incubation (streptavidin-allophycocyanin) in flow buffer was performed for 30 minutes on ice. After being stained, cells were strained through a clean flow cytometer tube with a 35 μm filter cap. To determine cell viability, either SYTOX Blue (1 μM; Invitrogen) or 7-AAD (1 μg/ml; BD Biosciences) was added to each flow tube sample 10 to 20 minutes before flow cytometry analysis.
Cells were subjected to cytometric profiling using a 100-μm nozzle at a sheath pressure of 20 psi on a BD Biosciences FACSAria flow cytometer (maintained by the University of Tennessee Health Science Center Flow Cytometry and Sorting Core) equipped with violet (404 nm), blue (488 nm), green (532 nm) and red (635 nm) lasers and applying FACSDiva software. All sample and collection tubes were maintained on ice. BD Biosciences CompBeads stained with either CD133-PE or CD24-FITC were used to set the compensation gates, and cells stained with isotype-matched secondary antibodies were used as controls for staining specificity. Cells heat-killed for 30 minutes at 65°C were used to gate against dead cells positive for SYTOX Blue or 7-AAD. Lineage-negative (Lin
neg) cells were identified by gating against cells positive for the mouse lineage panel + CD31. Because almost 100% of cells isolated from late-stage carcinomas of the PyMT-transgenic mouse were positive for CD24-FITC as previously described [
19], viable, Lin
neg mammary tumor cells were gated in a two-way sort for CD133
hi versus CD133
neg cells.
Sorted cell samples were collected directly on ice into 15 ml conical tubes precoated with 20% FBS, and a subset of collected cells was subjected to postsorting analysis to verify the purity and viability of the sorted populations. Sorted cell populations were immediately pelleted, washed to remove FBS and counted using a hemacytometer prior to reconstitution in mammosphere medium and plated in six-well ultra-low-adhesion dishes (Corning Life Sciences, Corning, NY, USA) at the indicated densities. FACSDiva plots were imported into FlowJo version 7.0 software (Tree Star Inc, Ashland, OR, USA) for data analysis and creation of histogram plots.
Limiting dilution transplantation
WT or KO MTECs were briefly trypsinized, washed with HBSS and serially diluted into 1:1 HBSS/Matrigel. Six dilutions per genotype (500, 200, 100, 50, 25 or 10 cells/10 μl) were introduced into the right inguinal cleared fat pads of FVB/Nj recipients. Tumor-initiating potential was defined as the ability to form a palpable tumor mass > 5 mm diameter (approximately pea-sized). The estimated TIC frequency was calculated using Extreme Limiting Dilution Analysis (ELDA) software [
35]. Fisher's exact (χ
2) test was used as a complementary approach to compare TIC activity between WT and KO cells at each cell density evaluated (GraphPad Prism version 4.0 software; GraphPad Software, Inc, San Diego, CA, USA). All animals were palpated twice weekly until surgical resection of the primary tumor or euthanasia.
Image acquisition
Western blots were scanned at 600 dpi and imported into Photoshop software (Adobe Systems, Inc, San Jose, CA, USA). Digital images of immunostained or H & E-stained tissue sections and cells cultured on chamber well slides were captured using a Leica DM6000 upright fluorescence microscope (Leica Microsystems, Inc, Buffalo Grove, IL, USA) mounted with a SPOT Insight cooled charge-coupled device camera (SPOT Imaging Solutions, Inc, Sterling Heights, MI, USA) and imported into Photoshop using SimplePCI software (Hamamatsu Photonics, Sewickley, PA, USA). All images between genotypes and conditions per experiment were digitally captured for the same exposure time. Immunostaining images were not digitally altered to reduce background or to adjust brightness or gain. Confocal images of spheres were captured every 0.75 μm with a Zeiss LSM210 microscope system (Carl Zeiss, Thornwood, NY, USA). Appropriate gain and black level settings were determined based on spheres incubated with secondary antibody alone. All images were captured using the same gain and exposure settings, and upper and lower thresholds were set using the range indicator function. No additional background correction algorithms were applied. Digital slices were merged into a single, composite image using Zeiss Zen software.
Discussion
Although the role of hypoxia and the HIF-1α transcriptional response in promoting tumor progression and metastasis is well-established, the direct contribution of the HIF family to the regulation of TICs in breast cancer is unknown. HIF-1α rather than HIF-2α is the predominant regulator of the hypoxic response in breast cancer [
47]; therefore, we sought to determine the effect of
Hif1a deletion in the MMTV-PyMT model of breast cancer. In this study, by using an
ex vivo genetic deletion approach, we generated constitutive HIF-1α-KO and control (WT) MTECs isolated from a pool of MMTV-PyMT tumors. One distinct advantage of this approach is that the MTECs are transplantable to immunocompetent hosts, preserving any host-derived effects on TIC potential following limiting dilution transplantation, in contrast to the xenograft of human cells into immunocompromised mice. Herein we show for the first time that HIF-1α positively regulates TIC activity in breast cancer as suggested by sphere formation assays
in vitro and validated through limiting dilution transplantation of WT and KO cells.
Additional advantages of the exogenous transduction and transplantation approach versus crosses to MMTV-Cre-transgenic mice include avoiding the mosaic nature of MMTV-Cre expression, because not all epithelial cells harboring
floxed alleles of
Hif1a undergo recombination [
48], and avoiding the use of a mixed genetic strain background by utilizing mouse models backcrossed to a single strain (FVB/Nj). Because
Hif1a deletion impairs cell proliferation [
39], an effect we also observed in cultured KO MTECs (Figure
1B), incomplete recombination could permit nontargeted cells to outgrow the recombined (KO) cells. Additionally, it had not previously been assayed whether the lung metastases originating in the MMTV-Cre-derived conditional KO females were derived from the recombined tumor cells or were generated from cells that had escaped recombination [
27]. In addition, the prior use of a mixed strain background may have introduced genetic modifier effects that could influence tumor incidence and lung metastasis. In particular, both tumor burden and lung metastasis in the MMTV-PyMT model are enhanced on the FVB/N background as compared to C57BL/6 [
49].
In agreement with the findings of previous studies, our study results confirm that HIF-1α promotes the growth of MTECs cultured at hypoxia and enhances lung metastasis
in vivo [
27]. However, there were some differences observed between the two model systems. For example, we found that deletion of
Hif1a repressed the growth of cells at normoxia as well as at hypoxia (by 48 hours of culture). In addition, the magnitude of the decrease in KO cell invasion was similar, regardless of whether cells were cultured at normoxia or hypoxia, whereas Liao
et al. reported a difference only at hypoxia [
27]. No statistically significant increase in invasion potential between WT cells cultured at normoxia and hypoxia was observed, as reported previously [
27]. In contrast to observations from either PyMT model system in which
Hif1a was deleted, short hairpin RNA-mediated knockdown of
HIF1A in MDA-MB-231 breast cancer cells does not significantly change cell number at either normoxia or hypoxia (1% O
2) [
50]. It is possible that the effects of loss of HIF-1α activity on PyMT cell growth in monolayer culture would be attenuated if cells were cultured at higher serum levels since MDA-MB-231 cells were cultured in medium containing 10% FBS whereas PyMT cell lines were cultured in medium containing 2% or 5% FBS.
We further demonstrate that primary tumor growth and survival from distant metastases are dependent upon epithelial cell intrinsic HIF-1α expression. In stark contrast to results obtained previously [
27], we found that HIF-1α plays a significant role in the control of primary PyMT-induced mammary tumor growth under either standard (50,000-cell input) or limiting dilution cell transplantation conditions (50- to 500-cell input). The cause of the control of net mammary tumor growth by HIF-1α is unclear, although our data suggest that HIF-1α-dependent control of TIC activity may be a primary mechanism driving tumorigenesis. The significant changes in
Vegf mRNA expression and microvessel density observed in early-stage KO tumors may also contribute to the phenotype by furthering restricting tumor growth once tumors are initiated. In agreement with our observations, a HIF-1α-dependent effect on tumor growth was demonstrated recently in both MDA-MB-231 and MDA-MB-435 breast cancer xenografts [
50,
51].
Characterization of the expression of luminal (K8/K18) and myoepithelial (K14/K5) lineage markers in WT and KO PyMT cells and tumors revealed that fewer KO cells than WT cells coexpressed K8/K14, primarily due to loss of K14-positive cells. These results suggest that loss of HIF-1α activity corresponds with a reduction in bipotent, stemlike cells in PyMT tumors. The most dramatic phenotype was the absence of K5-positive cells in KO cultures and the loss of both K14- and K5-positive cells in KO tumors. These data are of interest, given the stratification of triple-negative breast cancers (TNBCs) into non-basal-like or basal-like subtypes, based on the expression of both K5 and epidermal growth factor receptors (EGFRs) in tumors within the basal-like subtype [
52].
TNBC patients with a basal-like classification have shorter disease-free or overall survival [
53,
54] and tumors from TNBC patients with metastatic disease exhibit higher levels of K5 and EFGR [
55]. A potential influence of HIF-1α on the ability of the K5 and EGFR biomarkers to predict disease-free survival in basal-like TNBC is intriguing in light of our observations that, in cultured cells, EGF stabilizes HIF-1α expression at normoxia and further potentiates hypoxia-inducible expression of HIF-1α. Although the PyMT model is classified as a luminal-like cancer [
26], late-stage tumors such as the ones utilized to generate HIF-1α WT and KO cells are ER- [
25]. Subpopulations of both WT cells and WT tumors were positive for K5.
It has also been observed in luminal breast cancer cell lines (T47D and MCF7), which contain subpopulations of ER-/PR-/K5+ cells, that the K5+ cells are enriched for TIC activity and are resistant to conventional chemotherapies compared to ER+/PR+/K5- cells [
56]. Of additional relevance to our results regarding the decreased expression of K5, K14 and markers of EMT, including
Slug, by KO tumors is that hypoxia-dependent elevation in K5 mRNA levels occurs in a SLUG-dependent manner in MCF7 cells [
40]. Furthermore, a statistically significant correlation was found to exist between tumors with high
SLUG expression and
PROM1 (CD133) expression. These tumors also expressed high levels of carbonic anhydrase IX (
CAR9), which is a known HIF-1 target gene [
40].
Recent whole-genome expression profiling of breast cancers has revealed that the hypoxic response (predominantly through HIF-1α), the EGFR and signal transducer and activator of transcription 3 (STAT3) pathways are positively correlated together in TNBCs as compared to luminal cancers [
57]. In support of a functional association between the HIF-1α and EGFR pathways in TNBCs, MDA-MB-231 cells treated with gefitinib were found to exhibit downregulation of HIF-1α transcriptional activity that corresponded with decreases in cell viability and migration, whereas resistance to cetuximab or lapatinib therapy was hypothesized to be due to the inability of either drug to downregulate HIF-1α activity [
56]. Taken together, these observations suggest that targeting the HIF pathway may be beneficial to TNBC patients, particularly those patients diagnosed with basal-like TNBC.
Based upon the HIF-1α-dependent control of CD133 expression in MTECs and tumorspheres, as well as the enrichment of sphere formation in the CD133
hi versus CD133
neg populations, we have identified CD133 as a cell surface marker that may enrich for TICs in the PyMT model. Antibodies to epitope 2 of CD133 (CD133/2) have been utilized extensively to enrich for TICs in other solid tumors, particularly in human colon cancer and gliomas. In contrast, in the normal mammary gland, CD133 is expressed by differentiated ER+ luminal cells, and CD133+ cells exhibit lower regenerative capacity than CD133- cells [
58]. Notably, deletion of
Prom1 in a knockout mouse model did not impair the regenerative capacity of the normal mammary gland, but did reduce ductal branching during morphogenesis by increasing the ratio of luminal to basal cells [
59]. Yet, in the NKI 295 data set,
PROM1 expression levels were found to be lower in ERα+ tumors than in ERα- tumors [
59]. The association of HIF-1α and
PROM1 expression in ER- breast cancers is not surprising, given that hypoxia is a potent stimulator of ERα degradation [
60,
61].
In contrast to observations in the normal mammary gland, Meyer and colleagues have recently shown that, in the context of ER- breast cancers, CD133 enriches for TICs when used in conjunction with CD49f (integrin α6) and CD44 [
62]. Specifically, the CD49f+/CD44+/CD133
hi population identified tumor cells with enriched sphere-forming and xenografting potential. Likewise, CD133 has also been shown to enrich for TICs in the
Brca1 conditional mouse [
21]. That a marker of a differentiated normal mammary epithelial cell could enrich for cells with TIC activity in the context of breast cancer is also supported by recent evidence from multiple laboratories that, in
BRCA1 basal-like tumors, the TIC population arises from the luminal lineage [
63‐
65]. Interestingly, basal-like
BRCA1 tumors have previously been shown to overexpress HIF-1α and higher HIF-1α levels have been found to be correlated with decreased disease-free survival [
66,
67]. Moreover, Proia
et al. found that SLUG promoted a basal-like phenotype before and after transformation in
BRCA1 tumors [
65], which is consistent with our observation that expression of
Slug decreased more than threefold in HIF-1α-KO PyMT tumors that did not express the basal markers K5 and K14.
Furthermore, accumulating evidence suggests that differentiated cells (lineage-restricted progeny) may reacquire stem cell-like potential and tumor-initiating capacity rather than follow a strict linear hierarchy as originally proposed for the normal mammary gland. The plasticity involved in breast stem cell biology is emerging. Two independent groups have observed the spontaneous conversion of non-stem cells into stem cells [
68,
69]. Likewise, the results of recent lineage-tracing experiments by researchers in the Blanpain laboratory have challenged the requirement for a bipotent stem cell in the postnatal normal mammary gland [
70]. Under the limiting cell conditions routinely used to document the regenerative capacity of a given cell population, these authors found that the disruption of the normal luminal-to-myoepithelial cell ratio is sufficient to stimulate a unipotent myoepithelial progenitor cell to reacquire a bipotent progenitor activity that is normally restricted to the embryonic gland [
70]. Yet, how, or if, this model derived from lineage-tracing experiments can be applied to TICs during breast tumorigenesis remains unknown.
One mechanism of HIF-1α-dependent control of TIC may be through regulation of the Notch pathway. Interactions of HIF-1α with the Notch intracellular domain enhances the regulation of Notch transcriptional targets, such as the
HEY genes, and promotes EMT in breast cancer [
16,
71,
72]. In addition, in breast cancer, NOTCH1 and NOTCH4 have been positively correlated with stemness, and blocking antibodies to NOTCH4 reduces mammosphere formation [
72]. In KO tumorspheres and early-stage KO tumors, decreased expression of several members of the Notch pathway, particularly
Notch4 and
Hey1, was observed. Changes in
Notch4 are of particular interest because previous studies have shown that blocking NOTCH4 receptor activity inhibits tumor formation of xenografted breast cancer cells, whereas blocking NOTCH1 has less of an effect [
72]. Although a positive correlation between hypoxia and
NOTCH3 expression was previously described in breast cancer [
73], no HIF-1α-dependent changes in
Notch3 were observed in our studies.
HIF-1α-dependent effects on sphere formation efficiency
in vitro and TIC activity
in vivo were observed using parental tumor cells without first enriching for a putative CSC subpopulation based on cell surface markers. One rationale for this approach is the lack of comprehensive information on the markers that define TICs in the PyMT model. In a similar PyMT tumor cell transplant paradigm to ours, the population of CD24 (heat stable antigen)
hi/CD29(β
1-integrin)+/CD61(β
3-integrin)+ cells was found to significantly increase during tumor progression, specifically at the transition from hyperplasia to carcinoma. Greater than 90% of cells were characterized as CD24
hi/CD29+/CD61+ in late-stage carcinomas, and this population also had enhanced invasive potential
in vitro [
19]. More recently, researchers in the Visvader laboratory have shown that, in cells derived from PyMT adenomas (early-stage lesions), CD14 and c-kit, along with CD49f/CD24, enriches for cells with colony-forming potential [
74].
In addition, the physical stress of flow sorting decreases cell viability, which therefore directly influences the estimated TIC frequency as determined through limiting dilution transplantation. As observed for the MMTV-Neu model, the TIC frequency of unsorted cells was 1 in 61, decreasing to 1 in 177 for cells enriched by sorting [
20]. Likewise, for cells isolated from PyMT adenomas, the TIC frequency was estimated to be 1 in 556 for unsorted cells, whereas the TIC frequency in Lin
neg cells following sorting for CD24 was 1 in 648 [
74]. Using unsorted cells isolated from late-stage carcinomas of the PyMT model, we observed a TIC frequency of 1 in 82. It is possible that the TIC frequency in the PyMT model may vary based upon both the stage of progression and subtle differences in technical procedures among laboratories.
The specific cell surface markers that enrich for breast tumorsphere or TIC activity are also likely to vary in each mouse model. In the Balb/C p53
-/- model, mammary tumor cells double-positive for CD24 and CD29 were found to exhibit TIC activity [
22]. In contrast, in the MMTV-Neu model Sca-1+ cells were found to correlate with sphere formation [
75], and CD61 was also described as a marker that enriched for TICs [
20]. How HIF-1α directly affects various tumor cell subpopulations defined through flow cytometric profiling for the known murine mammary stem cell markers, or whether HIF-1α is preferentially expressed in a given subpopulation requires further extensive investigation.
Authors' contributions
LPS participated in the study design, created the MTECs, carried out the animal studies, performed the tumorsphere assays using MTECs and did the Western blot analysis and qRT-PCR. She also performed the statistical analyses, assisted with creating the figures and wrote the manuscript. DLP carried out the animal studies, flow cytometry sorting, immunostaining, Western blot analysis, qRT-PCR assays and statistical analyses and assisted with creating the figures and editing the manuscript. LPS and DLP contributed equally to the study. DM performed the tumorsphere assays using freshly digested tumors, performed qRT-PCR and analyzed the data. KDS optimized immunostaining conditions and assisted with drafting figures. JFI performed and analyzed the cell invasion assays, genotyped the animals and cultured cell lines derived from the animals. RCC and LCJ performed tumor resection surgery, assisted with limiting dilution transplantation surgeries, prepared lung sections, counted metastases and analyzed metastatic data. TNS conceived the study, created the study design, assisted with animal surgery and data analysis, created the figures and drafted the manuscript. All authors read and approved the final manuscript.