Development of a fluorescent reporter system to delineate cancer stem cells in triple-negative breast cancer

Abstract

Advanced cancers display cellular heterogeneity driven by self-renewing, tumorigenic cancer stem cells (CSCs). The use of cell lines to model CSCs is challenging due to the difficulty of identifying and isolating cell populations that possess differences in self-renewal and tumor initiation. To overcome these barriers in triple-negative breast cancer (TNBC), we developed a CSC system utilizing a green fluorescence protein (GFP) reporter for the promoter of the well-established pluripotency gene NANOG. NANOG-GFP+ cells gave rise to both GFP+ and GFP− cells, and GFP+ cells possessed increased levels of the embryonic stem cell transcription factors NANOG, SOX2 and OCT4 and elevated self-renewal and tumor initiation capacities. GFP+ cells also expressed mesenchymal markers and demonstrated increased invasion. Compared with the well-established CSC markers CD24⁻/CD44⁺, CD49f and aldehyde dehydrogenase (ALDH) activity, our NANOG-GFP reporter system demonstrated increased enrichment for CSCs. To explore the utility of this system as a screening platform, we performed a flow cytometry screen that confirmed increased CSC marker expression in the GFP+ population and identified new cell surface markers elevated in TNBC CSCs, including junctional adhesion molecule-A (JAM-A). JAM-A was highly expressed in GFP+ cells and patient-derived xenograft ALDH+ CSCs compared with the GFP− and ALDH− cells, respectively. Depletion of JAM-A compromised self-renewal, whereas JAM-A overexpression rescued self-renewal in GFP− cells. Our data indicate that we have defined and developed a robust system to monitor differences between CSCs and non-CSCs in TNBC that can be used to identify CSC-specific targets for the development of future therapeutic strategies.

Introduction

Breast cancer is the leading cause of cancer-related deaths among women worldwide [1]. Despite significant advances in the development of hormonal and systemic chemotherapy, response rates remain 30-60%, and even responsive cancers relapse and develop resistance [2]. The survival rates of metastatic breast cancer are lower than 5% [3]. The molecular genetics of breast cancer have been extensively investigated, permitting the association between distinct molecular subtypes and patient outcome. Of the different breast cancer subtypes, triple-negative breast cancer (TNBC) is the most aggressive [4, 5]. TNBC lacks expression of the estrogen and progesterone receptors and does not overexpress ERBB2 [6]. TNBC constitutes 15%-20% of all breast cancers and is characterized by poor prognosis and the lack of effective specific therapeutic options [7]. TNBC patients show higher rates of early relapse due to refractory drug-resistant local and/or metastatic disease even after an initial effective response to cytotoxic conventional chemotherapy, which remains the mainstay of TNBC treatment [8].

The hypothesis that a population of self-renewing cancer stem cells (CSCs) drives tumor recurrence and metastasis and underlies TNBC heterogeneity is well supported [9-11]. CSCs are characterized by their ability to propagate tumors and recapitulate the heterogeneity present in the original lesion [12, 13]. TNBCs are resistant to chemotherapy, and recurrence has been postulated to be a result of the chemo- and radio-resistance exhibited by CSCs [14, 15]. Due to confounding factors such as cellular heterogeneity and an evolving epigenetic state of CSCs, the mechanisms underlying their self-renewal and role in tumor progression are being actively pursued [16]. While CSCs have been postulated to be crucial for TNBC maintenance and progression, studying the characteristics of TNBC CSCs remains a challenge. A major obstacle to the identification of CSC regulatory mechanisms is a lack of experimental systems that enable the reliable enrichment of CSCs from non-CSCs for comparative analysis [17]. Many groups have isolated TNBC CSCs using CD24-negative/CD44-positive (CD24−/CD44+) cells and/or through high aldehyde dehydrogenase I activity (ALDH⁺) [18, 19]. These enrichment paradigms require refinement, as they are not universally applicable to all breast tumors [20-22]. Additionally, many CSC studies have been performed primarily in vitro, and as a result, there is limited information regarding the contribution of CSCs to tumor phenotypes in vivo. The main models of in vitro studies have used high passage TNBC cell lines that have not been well-characterized for CSC studies. Further complicating the study of CSCs in TNBC is the lack of a well-defined system to analyze these cells in real time.

To interrogate the molecular heterogeneity of TNBC cells, we developed a novel CSC reporter system using a GFP reporter driven by the promoter of the embryonic stem cell transcription factor NANOG. NANOG is a stem cell transcription factor and a master regulator of stem cell self-renewal [23, 24]. NANOG has emerged as a pro-carcinogenic factor [25], and immunostaining data show a strong correlation between NANOG and other cancer stem cell markers [25-28]. NANOG silencing in cancer cells leads to reduced proliferation, self-renewal based on tumorsphere assays, and tumor initiation in xenograft transplant studies [23, 29]. We generated two TNBC cell lines (MDA-MB-231 and HCC70) in which GFP+ and GFP− cells show differences in CSC marker expression and function [30, 31]. The cell surface signature of both GFP+ and GFP− cells was evaluated using a high-throughput screening method validated by our group, and we found that NANOG promoter-driven GFP also enriches for TNBC cells positive for CSC surface markers. The screen revealed additional receptors enriched in CSCs. Our approach has the ability to enrich for a population of CSCs, enabling interrogations to understand the key roles of CSCs in TNBC initiation and progression.

Materials and Methods

Cell culture

MDA-MB-231 and HCC70 breast cancer cells (American Type Culture Collection; Manassas, VA) were cultured in log-growth phase in modified Eagle's medium (MEM) supplemented with 1 mM sodium pyruvate (Cellgro, Kansas City, MO) and 10% heat-inactivated fetal calf serum (FCS) at 37 °C in a humidified atmosphere (5% CO₂).

Triple-negative breast cancer patient-derived xenograft tumors

Triple-negative patient-derived xenograft (PDX)-TN1 cells were procured and transduced with dTomato as previously described [32].

Immunoblotting

Cells were lysed in 20 mM Tris-HCl (pH 7.5), 150 mM NaCl, 1 mM Na₂EDTA, 1 mM EGTA, 1% NP-40, 1% sodium pyrophosphate, 1 mM β-glycerophosphate, 1 mM sodium orthovanadate, 1 ug/mL leupeptin, 20 mM NaF and 1 mM PMSF. Protein concentrations were measured using Bradford reagent (BIO-RAD, Hercules, CA). Lysates (20 μg total protein) were resolved by 10% SDS-PAGE and electrotransferred to PVDF membrane. Membranes were incubated overnight at 4°C with primary antibodies against NANOG (Cell Signaling), GFP (Zymed), SOX2 (Cell Signaling), OCT4 (Cell Signaling), VIMENTIN (Cell Signaling), N-CADHERIN (Millipore), GAPDH or β-ACTIN (Santa Cruz, CA), followed by incubation with secondary anti-mouse or anti-rabbit IgG antibodies conjugated to horseradish peroxidase (HRP) (Thermo, Rockford, IL). Immunoreactive bands were visualized using ECL plus from Pierce (Rockford, IL, USA).

Quantitative real-time PCR (qPCR)

qPCR was performed using an ABI 7900HT system with SYBR-Green MasterMix (SA Biosciences). Briefly, RNA from cells transduced with non-targeting control or JAM-A shRNA was extracted using the RNeasy kit (Qiagen), and cDNA was synthesized using the Superscript III kit (Invitrogen, Grand Island, NY). For qPCR analysis, the threshold cycle (C_T) values for each gene were normalized to expression levels of β-ACTIN. Dissociation curves were evaluated for primer fidelity. The primers used were:

β-Actin	Forward	5’-AGAAAATCTGGCACCACACC-3’
	Reverse	5’-AGAGGCGTACAGGGATAGCA-3’
NANOG	Forward	5’-CCCAAAGGCAAACAACCCACTTCT-3’
	Reverse	5’-AGCTGGGTGGAAGAGAACACAGTT-3’

Flow cytometry analysis

For flow cytometry analysis, MDA-MB-231 and HCC70 cells at a concentration of 1 million cells/mL were sorted with a BD FACSAria II and subjected to FACS analysis using the following antibodies: phycoerythrin (PE)-conjugated Integrin α6 (1:100, BD Biosciences), APC-conjugated CD24 (1:100, BD Biosciences), and PE-conjugated CD44 (1:100, BD Biosciences). Appropriate isotype control antibodies were used to set gates. Data analysis was performed using the FlowJo software (Tree Star, Inc.). Flow analysis and sorting of PDX cells was performed using APC-conjugated JAM-A antibody (1:50, BD Biosciences).

Collagen invasion assay

The collagen invasion assay was performed as previously described [33]. MDA-MB-231 and HCC70 GFP+ cells were stained with Vybrant DiO (green), and GFP− cells were stained with Vybrant Dil (red). Stained cells were then combined and co-cultured with unstained macrophages (BAC 1.2F5) in a glass-bottom tissue culture dish. Cells were then overlaid with a 3 mg/ml collagen gel, bathed in complete DMEM (10% serum), and incubated overnight. Cells were then fixed with 4% paraformaldehyde and imaged using a spinning disk confocal microscope. A 100 μm Z-stack image series was generated. Invasion was quantified as a percentage of green or red fluorescence above a 20 μm threshold distance from the top surface of the glass bottom dish.

Limiting dilution assays

For tumorsphere formation assays, cells were cultured in duplicate rows of serial dilutions per well in a 96-well plate per condition (Sarsted, Germany) with 200μl serum-free DMEM/F12 medium supplemented with 20ng/ml basic fibroblast growth factor (Invitrogen), 10ng/ml epidermal growth factor (BioSource, Grand Island, NY, USA), 2% B27 (Invitrogen), 10μg/ml insulin, and 1μg/ml hydrochloride (Sigma). Frequency of sphere formation was calculated in such a way that a well with a tumorsphere was counted as a positive well and a well with no tumorspheres was counted as a negative well. Tumorspheres were counted after 2 weeks under a phase contrast microscope. The stem cell frequencies were calculated using an extreme limiting dilution algorithm (ELDA) (http://bioinf.wehi.edu.au/software/elda/) [34].

In vivo tumor formation

NOD SCID gamma (NSG) mice were purchased from the Biological Resource Unit (BRU) at the Cleveland Clinic. All mice were maintained in microisolator units and provided free access to food and water. All mouse procedures were performed under adherence to protocols approved by the Institute Animal Care and Use Committee at the Lerner Research Institute, Cleveland Clinic. MDA-MB-231 and HCC70 NANOG-GFP cells were flow sorted for both GFP+ and GFP− cells and transduced with a luciferase lentiviral vector construct. GFP+ and GFP− cells were then transplanted in serial dilutions of 1000, 10,000 and 100,000 MDA-MB-231 cells and 1000, 10,000 and 30,000 HCC70 cells into the right subcutaneous flank of groups of female mice at 6 weeks of age. Mice were monitored every day until GFP+ tumors were palpable on day 12. Subsequent to this, biweekly bioluminescence imaging was performed on the mice by IVIS following intraperitoneal luciferin injection. Mice were euthanized, and the tumors were resected to dissociate tumor cells using papain. The cells were then sorted for GFP expression to assess tumorsphere formation as described above.

Flow cytometry screening

The BD Lyoplate Human Cell Surface Marker Screening Panel was purchased from BD Biosciences. The panel contains 242 purified monoclonal antibodies to cell surface markers and both mouse and rat isotype controls for assessing background signals. For the flow cytometry screening procedure, MDA-MB-231 and HCC70 NANOG-GFP cells were prepared in single-cell suspensions in BD Pharmingen Stain Buffer (BD Biosciences) with the addition of 5 mM EDTA. The screening was performed as previously described [30]. A total of 80 million cells of each MDA-MB-231 and HCC70 NANOG-GFP cell line was stained with DRAQ5 (eBioscience, San Diego, CA) and pacific blue dyes (Life Technologies Grand Island, NY), respectively. The cells were then pooled and plated in 96-well round-bottom plates (BD Biosciences). Reconstituted antibodies were added to the wells as per the human lyoplate screening panel. The cells were washed with stain buffer and stained with APC-labeled goat anti-mouse IgG secondary antibody (BD Biosciences). The cells were then stained with a live/dead fixable blue dead cell stain kit (Life Technologies, Grand Island, NY). Cells were washed and analyzed on an LSRII HTS system (BD Biosciences). Data were analyzed with FlowJo software. Positive immunoreactivity was based on isotype controls.

JAM-A lentiviral short hairpin RNA (shRNA) and JAM-A transducing lentiviruses were prepared as we previously reported [30, 34]. In short, using Lipofectamine 2000 (Invitrogen), 293FT cells were co-transfected with the packaging vectors psPAX2 and pCI-VSVG (Addgene) and lentiviral vectors directing expression of shRNA (Sigma) specific to JAM-A (TRCN0000061649 (KD1), TRCN0000061650 (KD2), a non-targeting control (NT) shRNA (SHC002)) and overexpression vector (Applied Biological Materials) for JAM-A or a control vector. Media of the 293FT cell cultures were changed 18 hours after transfection, and viral containing supernatants were collected 24 and 48 hours following the media change. Collected media were filtered for immediate use or concentrated with polyethylene glycol precipitation and stored at −80°C for future use.

Statistical analysis

Values reported in the results are mean values +/− standard deviation. One-way ANOVA was used to calculate statistical significance, and p-values are detailed in the text and figure legends.

Results

CSC reporter system in TNBC

A barrier to comprehensive CSCs studies in TNBC cell lines is the lack of the ability to monitor the stem cell state in real time and investigate cellular heterogeneity in vitro. To overcome this barrier, we developed reporter cell lines to track CSCs by transducing TNBC cells with a GFP reporter driven by the NANOG promoter (Fig. 1A). Cells expressing GFP represented cells with high NANOG promoter activity. GFP+ cells enriched by flow cytometry sorting gave rise to both GFP+ and GFP− cells as detected by flow cytometry analysis (Fig. 1B) and by fluorescent microscopy (Fig. 1C), demonstrating the development of cellular heterogeneity over time in vitro. The difference in GFP expression was validated by immunoblotting of GFP+ and GFP− cell lysates. The data indicated higher expression of GFP in the GFP+ cells compared with the GFP− cells (Fig. 1D). To validate that GFP expression identified NANOG-expressing cells, NANOG mRNA and protein expression were assessed in the GFP+ and GFP− cells (Fig. 1E, F). A 16-fold (MDA-MB-231) and a 2.3-fold (HCC70) increase in NANOG mRNA expression were observed in the GFP+ population compared with the GFP− population as quantified by qPCR. In accordance with mRNA data, western blot analysis detected higher levels of NANOG protein expression in GFP+ cells than in GFP− cells. These observations demonstrated that our system can be used to detect cellular heterogeneity with respect to NANOG promoter activity.

Figure 1. Development and validation of a CSC TNBC reporter system.

(A) Schematic demonstrates workflow of TNBC transduction with the NANOG-GFP reporter. (B) Histograms of MDA-MB-231 GFP cells post transduction. Cells were sorted, and GFP+ cells were cultured for 7 days, after which flow cytometry analysis was repeated. (C) Photomicrographs of NANOG-GFP reporter-transduced TNBC cell lines MDA-MB-231 and HCC70 cultured for 7 days after sorting for GFP expression. Scale bar – 100 μM. (D) Immunoblots of MDA-MB-231 and HCC70 cells sorted for GFP and probed with anti-GFP antibody. Actin was used as a loading control. (E) Quantification of NANOG mRNA expression in GFP-sorted MDA-MB-231 cells by qPCR. Actin was used as a control. (*** p < 0.001) (F) Immunoblots of GFP-sorted MDA-MB-231 and HCC70 cells probed with anti-NANOG antibody. Actin was used as a loading control.

NANOG promoter reporter enriches for cancer stem cells

The prospective identification and isolation of TNBC CSCs are based on the expression of cell surface markers including CD24−/CD44+ [20]. CD24 is a cell surface glycoprotein, and CD44 is a cell surface receptor for the extracellular matrix protein hyaluronan [12, 13]. GFP+ cells of both TNBC lines were enriched for CD24−CD44+ cells and compared with the GFP− cells (Fig. 2A). To examine whether the NANOG promoter reporter could enrich for other CSC markers, cells were also stained for the expression of CD49f/Integrin α6. CD49f plays a crucial role in cell adhesion and has also been widely shown to enrich for TNBC CSCs [35]. Compared with GFP− cells, GFP+ cells demonstrated higher expression of CD49f (Fig. 2A). The percentage of GFP+ cells expressing CD24−/CD44+ was 2.1-fold higher than the percentage of GFP− cells in the top 20% of all cells expressing CD24−/CD44+ in MDA-MB-231 cells, whereas in HCC70 cells, the percentage of GFP+ cells expressing CD24−/CD44+ was 3.5-fold higher than the percentage of GFP− cells (Fig. 2A). In both the MDA-MB-231 and HCC70 TNBC cell lines, the protein levels of the three embryonic stem cell transcription factors OCT4, NANOG and SOX2 were greatly increased in GFP+ cells compared with GFP− cells (Fig. 2B). These findings validate the hypothesis that GFP+ cells are enriched for CSC markers.

Figure 2. NANOG-GFP+ cells are enriched for CSCs.

(A) HCC70 NANOG-GFP cells were stained for CD49f/Integrin α6 and CD24^lowCD44^high expression and analyzed by flow cytometry. CD49f/integrin α6 and CD24^lowCD44^high expression were quantified using the FlowJo software (Version 10). The green and black histogram lines represent GFP+ and GFP− cells, respectively. (B) Immunoblots of GFP-sorted MDA-MB-231 and HCC70 cells probed with OCT4, NANOG, and SOX2 antibodies. Actin was used as a loading control.

GFP+ cells exhibit a mesenchymal phenotype

Vimentin and N-cadherin are markers associated with mesenchymal and invasive cellular behaviors [36]. We observed increased expression of these two markers in GFP+ cells compared with GFP− cells (Fig. 3A). To determine whether the GFP+ cells possessed increased intrinsic invasiveness, we performed a collagen-based invasion assay to measure the distance GFP+ or GFP− cells migrated into the collagen matrix (schematized in Fig. 3B). GFP+ cells displayed higher invasive and migratory potential compared with the GFP− cells at the single-cell level (Fig. 3C). The quantified average relative invasion results also showed a significant increase in the invasion of GFP+ cells compared with the GFP− cells (Fig. 3D). These results provide evidence that the GFP+ cells displayed an increased mesenchymal phenotype and invasive capacity compared with GFP− cells, which is consistent with studies suggesting increased mesenchymal and metastatic potential of CSCs [36].

Figure 3. NANOG-GFP+ cells exhibit a migratory/mesenchymal phenotype.

(A) Immunoblots of GFP-sorted MDA-MB-231 cells probed with antibodies to the mesenchymal markers N-cadherin and vimentin. Actin was used as a loading control. (B) Scheme demonstrates the workflow of collagen invasion assay. (C) Representative image of MDA-MB-231 GFP+ and GFP− cells stained with DiO (green) and Dil (red), respectively; invasive capacity was quantified as a percentage of green or red signal above a 20 μm threshold distance migration. (D) The invasion by the GFP+ and GFP− cells was quantified and plotted. (*** p < 0.001)

NANOG-GFP+ cells exhibit increased self-renewal, a hallmark of CSCs

To determine whether GFP+ cells demonstrated CSC characteristics as assessed by an enhanced capacity for self-renewal, limiting dilution analyses for tumorsphere formation were performed with GFP+ and GFP− cells. The sphere-forming cell frequencies in GFP+ MDA-MB-231 and HCC70 cells were calculated to be 1 in 5.46 and 1 in 13.3 cells, respectively, and 1 in 29.31 and 1 in 49.8 cells in the GFP− cells, respectively (Fig. 4A, B). Spheres formed from GFP+ and GFP− cells plated as single cells also showed the development of heterogeneity in tumorspheres (Supplemental Fig. 1). We next compared our reporter enrichment paradigm with established enrichment protocols. MDA-MB-231 and HCC70 parental cells were sorted for CD24−/CD44+ and CD49f expression, and limiting dilution analyses were performed. As previously reported, CD24−/CD44+ and CD49f^hi expression enriched for cells with tumorsphere formation capacity [20, 35]. The sphere-forming cell frequencies in CD24−/CD44+ MDA-MB-231 and HCC70 cells were 1 in 12.4 and 1 in 15.8 cells, respectively, and 1 in 25.1 and 1 in 32.4 cells, respectively, in the CD24−/CD44− cells (Fig. 4C, D). In CD49f^hi cells, the sphere-forming cell frequencies of MDA-MB-231 and HCC70 cells were 1 in 12.6 and 1 in 16.5 cells, respectively, whereas those of the CD49f^lo cells were 1 in 24.4 and 1 in 35.7 cells, respectively (Fig. 4E, F). ALDH⁺ cells also showed increased self-renewal compared with the ALDH⁻ cells, as reported previously [18, 19]. In ALDH⁺ cells, the stem cell frequency was 1 in 7.38 cells, whereas the stem cell frequency was 1 in 17.1 for ALDH⁻ cells (Fig. 4G, H). Although these established protocols enriched for self-renewing cells, the NANOG-GFP reporter system better enriched for self-renewing cells.

Figure 4. Increased self-renewal and sphere-forming cell frequency in GFP+ cells.

Limiting dilution assays were performed by plating MDA-MB-231 and HCC70 NANOG-GFP cells into 96-well plates with increasing cell numbers (1-20 cells/well), and tumorspheres formed were counted after 2 weeks. An extreme limiting dilution analysis algorithm was used to calculate the frequency of stem cells. (A & B) show the plots of GFP+ cells compared with the GFP− cells in both cell lines. (C & D) show the plots of CD49f^hi cells compared with the CD49f^lo cells in both cell lines. (E & F) show the plots of CD24−CD44− cells compared with the CD24−CD44+ cells in both cell lines. (G & H) show the plots of ALDH− cells compared with the ALDH+ cells in both cell lines. (* p < 0.05; ** p < 0.01; *** p < 0.001)

NANOG-GFP+ cells initiate tumor formation in vivo

The gold standard assay to functionally validate CSCs is tumor initiation. We performed in vivo limiting dilution analysis of GFP+ and GFP− cells across a range of cell numbers for transplantation (1,000 – 100,000 cells). Both cell lines were transduced with a luciferase reporter before subcutaneous injection. A significant difference was detected between GFP+ and GFP− cells with respect to tumor initiation frequency (MDA-MB-231 GFP+: 1 in 10,858; MDA-MB-231 GFP−: 1 in 225,232; p=1.5 × 10⁻⁴ ; HCC70 GFP+: 1 in 1559; HCC70 GFP−: 1 in 59747; p=1.35 × 10⁻⁶) and tumor formation latency. Tumors were visible and palpable from day 12 post-subcutaneous injection in the group of mice injected with 100,000 MDA-MB-231 GFP+ cells. Bioluminescence imaging was performed on both groups of mice (Fig. 5A) until mice reached the end point of pre-determined size (which was not achieved in the GFP− tumors even by day 40). The images were quantified, and the results were plotted to compare the group of mice injected with 100,000 GFP+ cells and those injected with 100,000 GFP− cells. A significant increase in tumor growth was observed in the group injected with GFP+ cells from day 12 until day 17 (Fig. 5B). To compare the molecular characteristics of pre- and post-transplantation xenografted NANOG-GFP cells, limiting dilution tumorsphere formation analyses were performed on cells derived from the tumors. Increased sphere-forming cell frequencies and self-renewal were observed in GFP+ cells compared with the GFP− cells isolated from tumors that developed from mice injected with either GFP+ or GFP− cells (Supplemental Fig. 2). Interestingly, the sphere-forming cell frequencies of the pre- and post-transplantation cells were similar to those determined in vitro prior to transplantation (Fig. 4A, Supplemental Fig. 2). Tumors formed from the injection of GFP+ cells showed a high percentage of GFP+ cells that ranged from 83% - 99%. Of note, tumors initiated from GFP− cells contained GFP+ cells, which may either be due to a low percentage of GFP+ cells present in the post-sorting GFP− population or the transition of GFP− to GFP+ cells in vivo. These findings validate that our system can reliably separate populations of cells with differences in tumor initiation capacity and confirm that GFP+ cells are functional CSCs.

Figure 5. Increased tumor initiation by GFP+ cells.

(A) Each group of NSG mice was injected subcutaneously with 100,000 GFP+ or GFP− cells. Representative in vivo bioluminescence images of both the GFP+ and GFP− groups on day 12 and day 19 are shown. (B) Bioluminescence images were quantified and graphed for a period of 40 days to compare the tumor growth of GFP+ and GFP− cells. Data are plotted as the mean ± SEM (* p < 0.05; ** p < 0.01).

High-throughput flow cytometry screen identified elevated junctional adhesion molecule-A (JAM-A) in GFP+ cells

To validate the utility of our reported system to identify CSC-specific molecular pathways, we performed a high-throughput flow cytometry screen. Previous screening methods for CSCs in TNBC have proven challenging due to the inability to interrogate a pure CSC population. We previously used a flow cytometry-based approach that enabled us to identify cell surface receptors in glioblastoma CSCs [30]. This screening procedure also enabled us to study intact cells and identify differentially expressed cell surface receptors in MDA-MB-231 and HCC70 GFP− and GFP+ cells (Fig. 6A). Using a commercially available panel of cell surface antibodies, we observed an increase in the expression of well-established CSC cell-surface receptors including CD29 (integrin β1), CD44, and CD49f (data not shown) in GFP+ cells in both cell lines. GFP+ cells showed increased expression of JAM-A compared with the GFP− cells in both MDAMB-231 and HCC70 cells (Fig. 6B). Expression of JAM-A has been shown to positively correlate with poor prognosis in patients with invasive breast cancer [37-39], and we previously reported elevated JAM-A in glioblastoma CSCs [30]. JAM-A expression in PDX-TN1 ALDH+ cells was observed to be 3-fold higher compared with the PDX-TN1 ALDH− cells by flow cytometry analysis (Fig. 6C) and by immunoblotting (Fig. 6D).To determine whether JAM-A is involved in CSC maintenance, we inhibited JAM-A expression via shRNA (Fig. 6E) and observed a significant decrease in self-renewal (Fig. 6G). JAM-A overexpression in GFP− cells (Fig. 6F) significantly increased the frequency of self-renewing cells compared with the control vector (Fig. 6H). Taken together, these data demonstrate that our NANOG promoter reporter system can be used for discovery approaches and identifies JAM-A as critical for self-renewal in TNBC CSCs.

Figure 6. High-throughput flow cytometry screen to identify CSC-specific surface molecules.

(A) Schematic of work flow for the high-throughput flow cytometry screen. (B) Immunoblots of GFP-sorted MDA-MB-231 and HCC70 cells probed with an antibody against JAM-A, which was identified by the screen. Data are shown for both MDA-MB-231 and HCC70 cells. Actin was used as a control. (C & D) JAM-A expression in PDX-TN1 ALDH+ and ALDH− cells was compared using flow analysis and immunoblotting. (E) Immunoblots of MDA-MB-231 unsorted cells silenced for JAM-A using two JAM-A shRNA constructs (KD1 and KD2) and a JAM-A non-targeting (NT) control probed with JAM-A antibody. (F) Overexpression of JAM-A (OE) in MDA-MB-231 GFP− cells compared with a control vector was validated by immunoblotting with JAM-A antibody (G) Limiting dilution analysis plots show JAM-A NT control compared with the JAM-A KD1 and KD2 in unsorted MDA-MD-231 cells. (H) Limiting dilution analysis plots show control vector compared with the JAM-A OE vector in MDA-MD-231 GFP− cells (* p < 0.05; ** p < 0.01; *** p < 0.001).

Discussion

We have developed a novel TNBC CSC reporter system using a GFP reporter driven by the NANOG promoter. Although CSCs have been postulated to underlie tumor initiation, progression, invasion and recurrence, the impediment to studying CSCs has been the need for an improved definition and functional characterization [40, 41]. To achieve these goals, we transduced the NANOG-GFP reporter into two established TNBC cell lines and detected CSCs in TNBC cells in real time. Breast CSCs have been shown to express the cell surface markers CD24−/CD44+ and possess high ALDH activity in breast cancer tissue, but specific TNBC CSC markers have not been identified [18-22]. In our model, CSC phenotypes were enriched in sorted GFP+ cells. GFP+ CSCs demonstrated increased expression of the embryonic stem cell transcription factors NANOG, SOX2, and OCT4, providing evidence that the GFP+ cells are stem cells. Furthermore, invasion assays demonstrated that GFP+ CSCs displayed a mesenchymal phenotype coupled with increased invasive ability. In vitro and in vivo limiting dilution analyses demonstrated that GFP+ cells possess increased self-renewal and tumor initiation capacity.

The ability of GFP+ cells to form tumorspheres demonstrated that enriching for GFP+ cells more specifically isolated self-renewing TNBC CSCs than the use of other paradigms, including enrichment by CD24−/CD44+, CD49f^hi, and ALDH⁺. We also show that our model based on the promoter activity of the embryonic stem cell transcription factor NANOG has the conceptual advantage of being able to define and functionally characterize CSCs in a more rigorous manner compared with the other established paradigms [42-44]. The reporter system that we have developed tracks NANOG promoter activity. Other reporter systems are based on tandem repeats of a composite response element or track a pseudogene, the expression pattern and functions of which are yet to be fully recognized [42, 43, 45]. While the tracking technique is similar among the reporter systems, our system is unique due to the novelty of a robust validation and characterization of CSCs in TNBC. Other promoter-reporter construct-based systems that have been used to study CSCs lack such extensive validation and application, further adding impact to our reporter system. We demonstrate that our reporter system in TNBC delineates pluripotency, self-renewal, CSC marker expression, invasiveness and in vivo tumor initiation by endogenous NANOG in two well-established TNBC cell lines. Importantly, our reporter system is more effective at enriching for self-renewing cells than conventional marker-based approaches and is amenable to screening in order to identify additional pathways important for CSC maintenance [42, 43, 45]. Additionally, our approach has identified junctional adhesion molecule-A (JAM-A) as a novel CSC regulator in TNBC which has also been validated in a triple-negative breast cancer patient-derived xenograft model. The CSC reporter model system that we developed is able to detect whether cancer cells maintain cellular heterogeneity, which remains a substantial challenge in the treatment of many advanced cancers including breast cancer. From a functional perspective, transplantation of GFP+ cells into immunodeficient mice recapitulated tumor heterogeneity. Dissociated tumors contained a heterogeneous population of GFP+ and GFP− cells that retained self-renewal abilities similar to those of the populations prior to transplantation.

Identification of novel cell surface markers enables precise detection and characterization of TNBC CSCs in primary and metastatic breast cancer tissue samples. To identify novel cell surface markers of TNBC CSCs, we utilized our NANOG promoter reporter system to perform a high-throughput flow cytometry screen that identified a novel CSC marker, JAM-A, for TNBC. JAM-A is a cell-cell adhesion protein that has been shown to influence the migration and morphology of epithelial cells [46-48]. Increased JAM-A expression in GFP+ and PDX-TN1 ALDH+ cells indicates that this surface molecule is enriched in TNBC CSC populations.

Furthermore, JAM-A silencing and overexpression studies in TNBC cells demonstrated that JAM-A is necessary and sufficient for CSC self-renewal. A strong correlation has been observed between high JAM-A protein expression and poor clinical outcome with reduced patient survival in invasive breast cancer [27, 37, 39, 49]. Furthermore, the existing literature provides evidence supporting the role of JAM-A in invasive breast cancer [37, 39, 48]. Our study establishes that JAM-A expression can be used to define and functionally characterize TNBC CSCs. JAM-A may provide clinical utility as an independent prognostic marker predicting the outcome of TNBC patients. Our study demonstrates that the NANOG promoter reporter system can be used to identify CSC biomarkers and a TNBC gene signature for prognostic outcomes. Our reporter system may prove useful for the design of new screening strategies to identify and develop CSC-specific therapeutics for therapy-refractory TNBC and other aggressive cancers. The targeting of CSCs will thereby improve patient survival outcome, especially in aggressive malignancies such as TNBC that lack effective therapies.

Conclusions

We have developed a reporter system based on NANOG promoter activity that enriches for functionally competent CSCs. The reporter system can segregate CSCs with higher fidelity compared with conventional methods and can be adapted for screening approaches to identify CSC-specific therapeutic targets and prognostic gene signatures. These studies highlight the heterogeneity present within TNBC lines and provide a system for subsequent CSC studies.