Construction and validation of fluorescent EMT sensors
To establish inducible models that alter the EMT state of carcinoma cells, we selected three mesenchymal-like, claudin-low breast cancer models: the human MDA-MB-231 cell line, the mouse T11 cell line [21], and the human BLSL12 breast cancer cell line derived from the WHIM12 patient-derived xenograft (PDX) [22]. To induce MET in these cells, we transduced each cell line with the pINDUCER lentivirus [23] containing the doxycycline-inducible human miR-200c/141 cluster (miR-200c), followed by selection for provirus-positive cells. We confirmed that the mesenchymal-like claudin-low cells switch to an epithelial-like (MET) morphology upon miR-200c induction as compared to non-induced cells (Fig. 1a). Induction of miR-200c (+DOX) was confirmed by qRT-PCR in each cell line (Fig. 1b). MET was further confirmed by reduced ZEB1 expression and increased E-cadherin expression in each cell line by qRT-PCR and western blot analysis (Fig. 1b–c).
To generate stable fluorescent sensors that can identify carcinoma cells with EMT or MET properties, we used the lentiviral expression vector FUGW, which expresses eGFP from the constitutively active ubiquitin C (UbC) promoter [24]. The human ZEB1 3′ UTR, a direct target of miR-200 family members containing eight miR-200 target sequences [25], or a 3′ UTR containing five miR-200 target sequences was inserted downstream of GFP (Fig. 2a and Additional file 1: Figure S1A). It is important to note that the ZEB1 3′ UTR sensor does not report transcriptional activity of the ZEB1 promoter, but instead reports post-transcriptional regulation of ZEB1 via its 3′ UTR. The eGFP fluorescent protein has a stability of >24 hours [26], which prevents rapid detection of decreasing GFP protein expression. Because we were interested in detecting rapid changes in GFP in response to changes in miR-200 family member activity (e.g., GFPhi to GFPlow/neg), we replaced eGFP with a destabilized GFP (d2GFP), which has a half-life of about 2 hours [27]. We designated the sensor using the human ZEB1 3′ UTR as d2GFP-Z1 3′ UTR and the 3′ UTR containing five miR-200 target sequences as d2GFP-200. Use of these sensors in mesenchymal-like cells and cells undergoing an EMT, which express low levels of miR-200 family members, should result in high GFP expression. Conversely, their use in epithelial-like cells or cells undergoing an MET, which have increased miR-200 expression, should result in low GFP expression by attenuation of GFP translation (Fig. 2a). Therefore, these GFP-based sensors should identify EMT (or mesenchymal cells) versus MET (or epithelial cells) states as a function of changing miR-200 levels.
Although our GFP-based sensors report a single aspect of EMT regulation (namely miR-200 expression and other regulators of ZEB1 via its 3′ UTR), many other factors can determine an epithelial versus mesenchymal-like state. The use of multiple detectors of EMT should provide greater resolution in identifying changes in cellular state. The human E-cadherin (CDH1) promoter contains three proximal E-boxes that serve as binding sites for E-cadherin transcriptional repressors including ZEB1, ZEB2, Snail, and Slug [28]. E-cadherin is not expressed in the mesenchymal state, and its loss promotes EMT and CSC properties [29]. Therefore, to detect the presence of an addition regulator of an epithelial versus mesenchymal state, we replaced the cytomegalovirus (CMV) promoter in a pHAGE lentiviral vector that expresses dsRed [30] with the E-cadherin promoter. We have designated this as the Ecad-RFP sensor (Fig. 2a). Simultaneous introduction of the GFP- and RFP-based sensors, therefore, should allow robust identification of both epithelial- and mesenchymal-like cells. Furthermore, this dual sensor approach should provide an additional level of specificity by helping to distinguish general effects on transcription and microRNA biogenesis.
To test the function of these sensors, each claudin-low cell line containing inducible miR-200c was co-transduced with the d2GFP-Z1 3′ UTR and the Ecad-RFP sensors, which, when used in combination, we have designated as the Z-cad dual sensor. If the sensors function properly in cells containing the Z-cad dual sensor, miR-200c should directly suppress GFP (and endogenous ZEB1) expression via the ZEB1 3′ UTR. Reduction of endogenous ZEB1, in turn, should elicit de-repression of RFP regulated by the E-cadherin promoter (Fig. 2a). We induced miR-200c expression and performed flow cytometry analysis for GFP and RFP. The non-induced cells expressed GFP and little to no RFP, the expected fluorescent characteristics of EMT cells harboring the Z-cad dual sensor, in MDA-MB-231, T11, and BLSL12 cells (Fig. 2b, –DOX). However, in each cell line tested, miR-200c induction inhibited GFP expression and induced RFP expression from the Z-cad dual sensor, the expected fluorescent characteristics of MET cells, validating the ability of these sensors to successfully report MET (Fig. 2b, +DOX). We saw a similar effect in each cell line with combinatorial use of d2GFP-200/Ecad-RFP (Additional file 1: Figure S1C–D). Control GFP expression was not reduced upon miR-200c induction as shown in Additional file 1: Figure S1C–D.
To demonstrate the ability of the Z-cad dual sensor to detect dynamic changes in response to stimuli affecting the mesenchymal state, we treated MDA-MB-231 cells containing the Z-cad dual sensor with DOX to induce miR-200c and imaged the cells over the course of 72 hours. We first observed loss of GFP expression between 4 and 6 hours (Additional file 2: Figure S2B), and by 12 hours, almost all GFP expression was non-detectable, whereas no changes in GFP fluorescence levels were observed in the control (–DOX) cells (Fig. 2c and Additional file 2: Figure S2A). RFP was first observed to increase beginning at 32 hours (Additional file 2: Figure S2B) and continued to increase in intensity throughout the 72 hours of DOX induction of miR-200c (Fig. 2c and Additional file 2: Figure S2B). These results demonstrated that in response to miR-200c expression, GFP expression was rapidly lost and there was a delayed but significant increase in RFP expression, as would be expected because E-cadherin is an indirect downstream target of miR-200c. The ability of the Z-cad sensor to detect these dynamic changes is a significant improvement over analysis of cells by immunostaining or via flow cytometry at a single time point. Importantly, the changes in GFP and RFP reflected the changes in ZEB1 and E-cadherin mRNA and protein levels (Fig. 1), respectively, demonstrating that expression of these fluorescent sensors reflected the expression of their endogenous counterparts. We additionally confirmed the changes in Z-cad fluorescence by microscopy in the mouse T11 and human PDX-derived BLSL12 cells 4 days after DOX treatment (Additional file 1: Figure S1B and E). Therefore, using the Z-cad dual sensor system, we were able to detect changes in the mesenchymal state upon MET induction in both human and mouse breast cancer cells.
The Z-cad dual sensor identifies EMT induced by TGFβ1 in HMLER cells
Having validated the use of the Z-cad dual sensor system to detect EMT/MET in claudin-low breast cancer cells expressing inducible miR-200c, we next asked if we could identify cells with EMT/MET properties using the Z-cad dual sensor in response to TGFβ1. For this, we used the epithelial-like, experimentally transformed human mammary epithelial cell line, HMLER [31]. HMLER cells containing the Z-cad sensor were treated with TGFβ1 (5 ng/mL) to induce EMT or vehicle over the course of several days, as it took this time for the cells to undergo an epithelial to mesenchymal switch. At the end of the treatment, cells in the TGFβ1-treated group had a high incidence of F-actin stress fibers, compared with few stress fibers in the vehicle-treated cells, consistent with cells exhibiting an EMT phenotype (Fig. 3a). We harvested TGFβ1- versus vehicle-treated cells at several time points throughout a 24-day treatment period and performed qRT-PCR for EMT-associated genes. Analysis of ZEB1, vimentin, E-cadherin, miR-200b, and miR-200c expression levels demonstrated an early induction of the mesenchymal-associated genes ZEB1 and vimentin (by 6 days) followed by a reduction of the epithelial-associated genes E-cadherin and miR-200b (Fig. 3b). We did not observe a significant reduction in miR-200c RNA during the time course of the experiment (Fig. 3b).
Next, we asked whether the observed changes in gene expression were reflected by Z-cad dual sensor expression. We noted that HMLER cells not treated with TGFβ1 displayed a wide range of RFP by flow cytometry analysis, although the majority did express RFP (90 ± 3.0 % RFP-positive cells; n = 8), suggesting that there is heterogeneity in HMLER cells with respect to E-cadherin expression (see Fig. 5). At day 14 of treatment, TGFβ1- or vehicle-treated cells were assessed for differences in fluorescence by microscopy. The majority of vehicle-treated cells showed weak GFP expression and a wide range of RFP levels (from weak to strong), whereas the TGFβ1-treated cells displayed brighter GFP expression with fewer RFP-expressing cells (Fig. 3c). In addition, the high GFP-expressing cells in the TGFβ1-treated group frequently appeared mesenchymal in morphology (Fig. 3c, arrows). Flow cytometry analysis of the cells throughout the 24-day treatment period showed a gradual increase in the proportion GFPhi/RFPlow/neg (EMT signature) cells in the TGFβ1- compared to vehicle-treated cells (Fig. 3d–e). The observed changes in gene expression (Fig. 3b) reflected the increasing EMT population observed by flow cytometry, demonstrating that the Z-cad dual sensor was able to identify cells that have undergone EMT in response to TGFβ1.
The Z-cad dual sensor identifies differential plasticity in EMT subpopulations of HMLER cells upon TGFβ1 exposure
To assess the plasticity of cells following TGFβ1 treatment, we used fluorescent activated cell sorting (FACS) to separate TGFβ1-treated cells into two populations, the EMT signature cells and the bulk population depleted of the EMT signature cells (Bulk-EMT) at 12 (data not shown) or 24 days of treatment, and then, each population was cultured for 13 days in the absence of TGFβ1. Intriguingly, following 13 days of TGFβ1 removal, the EMT signature cells were mesenchymal in appearance and maintained their EMT fluorescence signature (GFPhi/RFPlow/neg), whereas the Bulk-EMT cells appeared only partially mesenchymal and displayed a heterogeneous fluorescence signature (Fig. 4a). These data suggest that only a subset of the parental HMLER cells undergo complete EMT during TGFβ1 treatment, whereas the majority of cells, even at late time points, are still transitioning and display a partial mesenchymal phenotype. The EMT signature sorted cells treated with TGFβ1 displayed no capacity to revert to their parental phenotype, both morphologically and by Z-cad expression, upon subsequent passaging after 13 days (data not shown). However, FACS-isolated parental HMLER cells that displayed the EMT signature and were never exposed to TGFβ1 eventually re-expressed RFP, lost GFP expression, and lacked the full mesenchymal morphology (Fig. 4b). This suggested that parental HMLER cells that display the EMT signature retain plasticity, as they have the capacity to undergo MET. Together, these data indicate that TGFβ1 treatment reprograms a subset of HMLER cells and inhibits their capacity for EMT/MET plasticity.
HMLE cells, from which the HMLER cells are derived, undergo spontaneous EMT with concomitant methylation of miR-200c, re-expression of which induces MET [32]. Therefore, we hypothesized that prolonged exposure of cancer cells to TGFβ1 potentially induces DNA methylation and fixes cells in the mesenchymal state. To test this hypothesis, we treated fully transitioned HMLER EMT cells with the DNA methyltransferase inhibitor decitabine for 5 days. Decitabine treatment reduced the number of cells displaying the Z-cad EMT signature compared to vehicle-treated cells (Fig. 4c). Quantitation of RFP expression showed a significant increase in RFP+ cells from 1.09 % in vehicle-treated cells to 2.84 % (p = 0.006) and 5.33 % (p = 0.003) for 100 nM and 500 nM decitabine-treated cells, respectively (Additional file 3: Figure S3). Moreover, we observed a significant increase in GFPlow expressing cells in vehicle-treated cells from 4.63 % to 8.07 % (p = 0.007) and 15.7 % (p = 0.0000002) for 100 nM and 500 nM decitabine treated cells, respectively (Additional file 3: Figure S3). Decitabine-treated cells acquired the expression of E-cadherin, miR-200b, and miR-200c, and reduced ZEB1 expression (Fig. 4d). Thus, the changes in the Z-cad sensor reflect the changes in the endogenous expression of these genes. These data further suggest that the permanent EMT induced by TGFβ1 treatment may be due, at least in part, to epigenetic silencing of epithelial genes by DNA methylation, and this can be reversed, in at least a subset of cells, by DNA methyltransferase inhibition.
Combinatorial use of EMT sensors detects rare mesenchymal-like cells among a heterogeneous population largely comprising epithelial-like cells
It is apparent that considerable heterogeneity exists in most established epithelial cell lines. Parental HMLER cells cluster into the basal-like breast cancer subtype and contain a small population of CD24low/neg/CD44hi CSCs that also display EMT properties [2, 33]. In our studies using HMLER cells, TGFβ1 treatment slowly induced EMT, and the transition only appeared to be complete in a subset of cells. Notably, the Z-cad dual sensor allowed the resolution of cells displaying a partial versus complete EMT during TGFβ1 treatment, suggesting that there is significant heterogeneity within the HMLER population during TGFβ1-induced EMT. Therefore, we next asked whether parental HMLER cells display epithelial/mesenchymal heterogeneity similar to that observed during TGFβ1 induction of EMT, and if the Z-cad dual sensor can resolve distinct cellular populations prior to experimental induction of EMT. Using fluorescent microscopy, we observed differences in the intensity of fluorescence among individual cells, and we identified a small subpopulation of GFP+RFPneg cells that appeared mesenchymal in morphology (Fig. 5a). Flow cytometry analysis (Fig. 5b) of parental HMLER cells also showed a group of GFPhiRFPlow/neg cells, similar to the EMT signature cells that expanded during TGFβ1 treatment (Fig. 3). We asked whether cells falling within this EMT signature group in parental HMLER cells display characteristics of mesenchymal cells even without experimental EMT induction. Using FACS we separated cells displaying the EMT signature (GFPhiRFPlow/neg) from the Bulk-EMT cells. We cytospun the cells immediately post sorting and performed co-immunofluorescence for pan-cytokeratin (CK), an epithelial marker, and vimentin, a mesenchymal marker. Although all cells expressed basal levels of both CK and vimentin, the immunofluorescent staining of vimentin was more intense in the EMT signature associated cells relative to the Bulk–EMT population (Fig. 5c). Moreover, RNA analysis showed reduced expression of E-cadherin, miR-200b, and miR-200c and increased expression of ZEB1 in the EMT signature population relative to the Bulk–EMT population (Fig. 5d). We also used the Z-cad dual sensor system in MCF10A cells, which have been shown to also be heterogeneous and contain a more mesenchymal subpopulation [34], and observed a similar RNA expression pattern when sorted into EMT and Bulk–EMT populations to that observed in the HMLER cells (Additional file 4: Figure S4). Finally, sorted HMLER cells were cultured and their morphology was assessed 48 hours post sorting. Bulk–EMT cells appeared cobblestone-like and formed tightly organized colonies, similar to the unsorted, parental HMLER cells (Fig. 5e). Conversely, the EMT signature cells displayed a loosely packed organization and were more mesenchymal in morphology than the parental cells (Fig. 5e), but similar to the sorted parental EMT signature cell morphology observed 13 days after sorting (Fig. 4b). Taken together, these data indicate that the Z-cad dual sensor can be used to identify and isolate cellular subpopulations that display mesenchymal-like morphology and gene expression from a heterogeneous population of cells.
During TGFβ1 treatment of HMLER cells (Fig. 3), we observed a gradual induction of a complete EMT via formation of a GFPhiRFPlow/neg population, and these cells were more mesenchymal in morphology than the Bulk-EMT population (Fig. 4a). While the entire TGFβ1-treated cell population gained expression of mesenchymal genes early on compared to the vehicle-treated cells, a decrease in epithelial gene expression was delayed (Fig. 3b). However, the separation of cells with the EMT signature from the Bulk–EMT population resolved differences in RNA and protein expression with respect to EMT regulating factors during TGFβ1 treatment (Additional file 5: Figure S5). This validated the ability of the Z-cad dual sensor to identify changes in gene expression among a heterogeneous population even when these changes were not apparent among the entire TGFβ1-treated population as compared to the vehicle-treated population until later time points. Taken together, these data demonstrate that the Z-cad dual sensor can be used to isolate and identify different cellular populations within a heterogeneous population displaying more mesenchymal gene expression patterns and properties.
Z-cad dual fluorescence better identifies CSC-like cells compared with either sensor alone
Previous studies have demonstrated that HMLE cells contain a subpopulation of CSCs as demonstrated by their CD24low/negCD44hi expression and mammosphere-forming capacity [2, 32]. Because EMT has been linked to CSC properties, we asked how well the HMLER cells displaying an EMT signature, which were isolated using the Z-cad dual sensor system from the parental HMLER cells, correlated with HMLER cells displaying a CSC signature (CD24low/neg/CD44hi). Moreover, we wanted to test whether the Z-cad sensor system could better identify CSCs than either sensor alone. We used flow cytometry to distinguish cells displaying the GFPhiRFPlow/neg EMT signature to those displaying GFPhi, RFPlow/neg, Bulk-EMT, and the total cellular population (Bulk), and then quantified the percentage of these populations that fell within the CSC-enriched CD24low/CD44hi population (Fig. 6a–b; Additional file 6: Figure S6A–B). The GFPhi only and the RFPlow only populations contained more cells in the CD24low/neg/CD44hi, CSC-enriched group than the Bulk population (13 % versus 2.5 %, p = 0.0000037) and (6.1 % versus 2.5 %, p = 0.00052), respectively, demonstrating that either sensor alone can enrich for CSCs. Importantly, the EMT signature population was the most highly enriched in the CD24low/CD44hi CSC signature group compared with the Bulk population (32 % versus 2.5 %, p = 0.00018) (Fig. 6b; Additional file 6: Figure S6B), and compared with any of the other populations, including GFPhi (p = 0.00074), RFPlow (p = 0.00017), and Bulk–EMT (p = 0.00017) (Fig. 6b and Additional file 6: Figure S6B). These data suggest that the Z-cad sensor better enriches for CSCs than either sensor alone.
To test the stem cell function of the EMT signature cells isolated using the Z-cad sensor, we performed tumorsphere assays. Importantly, the EMT signature cells formed tumorspheres (>75 μm) at a frequency of about 3 %, much higher than the approximately 0.5 % tumorsphere-forming capacity of the Bulk-EMT population (Fig. 6c–d). In addition, cells falling into the CD24low/neg/CD44hi CSC-enriched population formed tumorspheres at a higher rate than the non-CSC (NCSC) population (Fig. 6d). Confirming the ability of the Z-cad sensor to identify cells with CSC properties, cells with the EMT signature formed spheres at a similar efficiency as that of the CSC-enriched population (Fig. 6d). These data, therefore, demonstrated that the Z-cad dual sensor facilitated the isolation of cells with EMT properties from a mixed population, and furthermore, it was able to enrich for cells with CSC properties to a greater extent than either sensor alone. Therefore, the Z-cad dual sensor should provide a stable and facile method, in the absence of immunostaining, for identifying and isolating EMT/CSC-like cells by flow cytometry in this model as well as for their detection via microscopy.