TWIST1-Induced miR-424 Reversibly Drives Mesenchymal Programming while Inhibiting Tumor Initiation

Over the last decade, there has been a growing emphasis on understanding the metastatic cascade of carcinomas by examining the epithelial-to-mesenchymal transition (EMT; ref. 1). An oncogenic EMT, similar to a developmental EMT, causes epithelial cells to lose tight cell–cell contacts and gain a more motile and invasive phenotype, traits postulated to facilitate dissemination from a primary tumor (2). Carcinoma-associated experimental studies and patient tissue analyses have further demonstrated that acquisition of a more mesenchymal phenotype is linked with other prometastatic traits such as tumor-initiating cell (TIC) characteristics (e.g., self-renewal, multipotency) and resistance to conventional therapeutics (3–5). Many of these informative studies have demonstrated that EMT-inducing transcription factors such as SIX1, TWIST1/2, and SNAI1/2 can mediate these prometastatic and TIC phenotypes (1).

MicroRNAs (miRNA) have relatively recently been identified as a class of factors that also regulate oncogenic EMTs. Such mechanisms include, but are not limited to, ability of miRNAs to target the 3′ untranslated region (UTR) of CDH1 (E-cadherin; ref. 1) or to target the 3′ UTR of the CDH1 repressors ZEB1/2 (6). Furthermore, miRNAs have been reported to alter signaling pathways implicated in EMT, such as the EGFR pathway (1). To date, no miRNAs have been identified that solely upregulate the mesenchymal arm of an oncogenic EMT without repressing epithelial programming.

Identification of potential mesenchymal-specific regulators has likely been hindered by the ability of EMT-inducing factors to influence one another (7). The ubiquitous loss of functional E-cadherin during an EMT has further confounded analysis of mesenchymal programming in isolation from epithelial programming, as E-cadherin loss can be sufficient to drive a full EMT (8). Studies manipulating the mesenchymal half of the EMT program are increasingly relevant, as human carcinoma cells are documented to coexpress epithelial and mesenchymal markers (9–11), thus having not fully lost their epithelial identity. Such an intermediate EMT state, as opposed to a fully mesenchymal state, may more easily undergo a reverse mesenchymal-to-epithelial transition (MET), which has been postulated to facilitate metastatic colonization at secondary sites (2), and is compatible with the epithelial nature of carcinoma metastases and the similarity of a metastatic lesion to its originating primary tumor.

Studies now experimentally support that an MET, following an EMT, facilitates metastatic colonization. Such studies focused on the EMT inducers TWIST1 and PRRX1, which promoted invasive phenotypes in vitro and in vivo, but needed to be downregulated, concurrent with an MET, for metastatic outgrowth to occur (12, 13). Contrary to previous studies that positively link EMT and TIC attributes (3, 4), an MET link to stemness appears just as relevant to metastatic outgrowth, as metastases are epithelial in nature. Oncogenic EMT inducers may thus be regulating a more plastic state, rather than a full EMT, to enable reversion to an epithelial phenotype along with self-renewal and proliferative capability at the secondary site. To better understand mediators of this plastic state, we set out to identify novel regulators of the mesenchymal arm of EMT and to ask whether such regulators impinge on TIC phenotypes, thereby influencing various stages of the metastatic cascade. In doing so, we identify miR-424, acting through TGFBR3, as the first miRNA regulator of mesenchymal programming that simultaneously maintains epithelial morphology and molecular characteristics. Although miR-424 enhances early prometastatic functions, our data demonstrate that this microRNA inhibits TIC phenotypes, mediated by ERK signaling, and ultimately becomes downregulated in experimental and clinical macrometastases.

HMLE derivatives and SUM149PT lines were generous gifts from Dr. Robert Weinberg (Massachusetts Institute of Technology; 2009) and Dr. Stephen Ethier (Medical University of South Carolina; 2012), respectively. MCF12A cells were obtained from The University of Colorado Comprehensive Cancer Center (UCCC) Tissue Culture Shared Resource and cultured according to the ATCC recommendations. Cell lines were authenticated on April 3, 2013, by short tandem repeat profiling or by methods previously described (14). TWIST1 and SNAI1 induction were performed as previously described (4). MEK inhibitors, 5 μmol/L PD98059 (Sigma) or 5 μmol/L AZD6244 (Otava Chemicals), were used for 24 hours. Doxycycline (Sigma) inductions were done at 0.5 μg/mL in complete SUM149PT media for the indicated amount of time for each assay. Inductions were always performed on cells that had never previously been exposed to doxycycline.

Attached cells were serum-starved, then given complete medium supplemented with 10 μmol/L bromodeoxyuridine (BrdUrd; Sigma) for 2 hours. Cells were fixed with ice-cold 70% ethanol, permeabilized with 0.5% Tween-20/PBS, and denatured with 2N HCl. Cells were incubated with anti-BrdUrd antibody (Roche; 1:100), FITC-anti–mouse IgG (Sigma; 1:20), and propidium iodide (PI)/RNaseA. Cells were analyzed as in flow-cytometry analyses.

A 10 μg total RNA was isolated using the miRNeasy Mini Kit (Qiagen) and was submitted in triplicate to LC Sciences for microRNA microarray services and data analysis using miRBase v15.0. Results are deposited under GSE58560.

A 1 μg total RNA isolated using the miRNeasy Mini Kit (Qiagen) was submitted in triplicate to the University of Colorado AMC Genomics and Microarray Core for construction and sequencing of the RNA-seq library on a HiSeq-2000 (Illumina). On average, we obtained approximately 54 million (range, 44–60 million) single-end 50 bp sequencing reads per sample. RNA-seq analysis was performed as previously described (15, 16). Results are deposited under GSE54505.

Motility was determined by gap closure assay, where a culture insert (Ibidi) created a 500-μm gap between 4 and 6 × 10⁴ cells per chamber. Inserts were removed the next day, and distance migrated was measured using DP2-BSW software (v2.2; Olympus). Adhesion assays were performed as previously described (17).

Whole-cell extracts were collected in RIPA buffer as previously described (17) and run on 7%, 8%, or 10% SDS-polyacrylamide gels. Antibodies used for immunoblots: α-catenin, E-cadherin, fibronectin, N-cadherin, plakoglobin (BD Biosciences), actin, β-tubulin, vimentin (Sigma), total ERK1/2, p-ERK1/2 (Cell Signaling Technology), TGFBR3 (Santa Cruz Biotechnology), ST3GAL5 (Abgent), and TMPRSS4 (Proteintech).

ALDEFLUOR assays (Stemcell Technologies) to detect ALDH1 activity were performed according to the manufacturer's recommendation. Diethylaminobenzaldehyde (DEAB) was used as an ALDH1 inhibitor to set ALDH1 gates. Surface TGFBR3 levels were determined with a polyclonal goat primary antibody (R&D Systems) and AlexaFluor-594 secondary antibody (Life Technologies). All samples were analyzed on a Gallios Flow Cytometer (Beckman Coulter) at the UCCC Flow Cytometry Shared Resource.

Immunocytochemistry was performed as previously described (17). Antibodies used: N-cadherin (BD Biosciences; 1:50), E-cadherin (Cell Signaling Technology; 1:200), AlexaFluor-488 (Life Technologies; 1:800), AlexaFluor-594 (Life Technologies; 1:800). Coverslips were mounted with Vectashield (Vector Laboratories) and DAPI (1 μg/mL).

Primary and secondary mammosphere formation were performed as previously described (18). Mammosphere media: MEBM basal media (500 mL), 1X B27, hydrocortisone (1 μg/mL), insulin (5 μg/mL), β-mercaptoethanol (5 μg/mL), and EGF (10 ng/mL).

Reporter constructs and 5 nmol/L miR-424 or negative control mimics (Thermo Scientific) were cotransfected into cells with X-tremeGENE siRNA Transfection Reagent (Roche). Lysates were collected and analyzed as previously described (18). See Supplementary Materials and Methods for construct design.

Prism software (v5.0; GraphPad) was used to conduct appropriate statistical procedures, as noted in the individual figure legends. P value <0.05 or FDR <0.25 were considered significant unless noted otherwise. The frequency of stem cells in the tumor-initiation experiment was estimated with 95% confidence using the ELDA analysis tool (19).

Tumor-initiation experiments and experimental metastasis assays are detailed in Supplementary Materials and Methods. All animal studies were performed according to protocols reviewed and approved by the Institutional Animal Care and Use Committee at the University of Colorado AMC.

To examine the temporal regulation of mesenchymal and epithelial genes during an EMT, we used inducible models of EMT previously established in the nontumorigenic, human mammary epithelial HMLE cell line (4). In these models, TWIST1 or SNAI1 are fused to the estrogen receptor ligand-binding domain, allowing for posttranslational activation by 4-hydroxytamoxifen (4-OHT). After 12 days of 4-OHT treatment, we observed the expected morphologic and molecular changes (Fig. 1A and C; Supplementary Fig. S1A and S1C) characteristic of an EMT (4). Interestingly, 4 days of 4-OHT treatment was not sufficient to alter the epithelial morphology (Fig. 1A; Supplementary Fig. S1A, middle) or to decrease levels of the epithelial markers E-cadherin or miR-200c, but was sufficient time for TWIST1 and SNAI1 to induce expression of the mesenchymal markers fibronectin, N-cadherin and vimentin (Fig. 1B; Supplementary Fig. S1B). These findings suggested the existence of a mesenchymal regulator(s) that operates upstream of E-cadherin repression during an EMT.

To identify a potential mechanism by which mesenchymal markers are upregulated early during an EMT, we screened RNA from different time points during a TWIST1-induced EMT using an miRNA microarray. As members of the miR-8 family (e.g., miR-200a/b/c) have been well documented to promote an epithelial phenotype (20), we reasoned that another miRNA may serve as a mesenchymal-promoting counterpart. 41 and 77 miRNAs were differentially expressed after 4 (early EMT) and 12 (late EMT) days of TWIST1 induction, respectively (Fig. 1D; Supplementary Tables S1 and S2). Notable microarray changes included members of the miR-8 family, miR-146a, miR-31 and the previously identified TWIST1-regulated miR-10b, each of which have been implicated in the metastatic process (20–23). Because the miR-8 family was not downregulated until day 12 (Fig. 1D), yet mesenchymal markers were upregulated earlier in the EMT process, additional mechanisms downstream of TWIST1 that mediate the mesenchymal phenotype likely exist.

We used quantitative PCR (qPCR) with independent TWIST1 inductions to validate the changes of miR-200a/b/c, as well as miR-429, miR-10a, and miR-152 (data not shown). Additional validation of the screen led to the confirmation that miR-424 was upregulated in HMLE cells early after TWIST1 or SNAI1 induction (Fig. 1E; Supplementary Fig. S1D), which we further pursued, as miR-424 had not previously been directly implicated as an exclusive regulator of mesenchymal programming. Primary miR-424 (pri–miR-424), the first precursor during miR-424 biogenesis, was also upregulated by TWIST1 and SNAI1 induction (Fig. 1F; Supplementary Fig. S1E). Together, these data demonstrate that TWIST1 and SNAI1 can regulate miR-424 expression.

We next asked whether miR-424 could mediate a change in expression of mesenchymal markers by creating three human breast cell lines that stably overexpressed miR-424. The rationale for cell line choice was as follows: HMLE, cell line used for the original miRNA screen; MCF12A, nontransformed human mammary cell line to verify HMLE findings; SUM149PT, human breast cancer cell line with low endogenous miR-424 useful for in vivo study of exogenous miR-424. In SUM149PT cells, miR-424 (Fig. 2A) induced a dramatic upregulation of N-cadherin with more modest increases in vimentin and fibronectin (Fig. 2B). In contrast, miR-424 had little to no effect on the levels of the epithelial markers E-cadherin, α-catenin, plakoglobin, and miR-200c, nor did miR-424 alter cellular morphology (Fig. 2B; Supplementary Fig. S2A and S2B). Similar results were observed in miR-424–expressing MCF12A and HMLE cells (Supplementary Fig. S2C–S2F, S2G–S2J, respectively). Notably, immunocytochemistry revealed miR-424 expression caused a mixed/intermediate phenotype, where 52% of SUM149PT cells simultaneously expressed N-cadherin and E-cadherin at cell membranes, as compared with 10% of empty vector (EV) control cells (Fig. 2C and D); immunocytochemistry also demonstrated localization of E-cadherin was unaffected by miR-424. Thus, miR-424 is sufficient to regulate the mesenchymal arm of an EMT in multiple cell lines, independent of epithelial arm repression, mimicking an early TWIST1 or SNAI1-induced EMT.

Gains in mesenchymal EMT markers, particularly N-cadherin, have been associated with increased motility (24). We thus tested whether miR-424 affects cellular migration, focusing on the SUM149PTs and MCF12As to characterize the tumorigenic cell line chosen for in vivo studies and an additional nontransformed cell line that had the more robust intermediate EMT profile (Supplementary Fig. S2), respectively. Using gap closure assays, we found that miR-424 increased migration of SUM149PT cells (Fig. 2E). In addition, miR-424 shifted cells into the G₁ phase, akin to an EMT-associated growth arrest (Fig. 2F). Finally, we examined cell-matrix adhesion, finding that miR-424 decreased cell adhesion to fibronectin-coated plates (Fig. 2G). Similar effects of miR-424 on motility, growth, and adhesion were seen in MCF12A cells (Supplementary Fig. S3A–S3C). Together, these data demonstrate that miR-424 induces EMT-associated functions while promoting an intermediate EMT.

As recent evidence has demonstrated the importance of an MET following an oncogenic EMT, we asked whether the intermediate EMT state was reversible by developing a doxycycline-inducible miR-424 system (i-miR-424) in SUM149PT cells. We characterized clones from the EV control (i-EV 1 and i-EV 2) and miR-424 (i-miR-424 1 and i-miR-424 2) expressing cells. Of note, 3′ UTR luciferase reporters with a perfect miR-424 binding site demonstrated that doxycycline led to repression of luciferase in the i-miR-424 clones, which was alleviated by mutating the reporter's miR-binding site (Fig. 2H). Removal of doxycycline also alleviated luciferase repression, demonstrating that i-miR-424 activity is functionally reversible (Supplementary Fig. S4A).

Importantly, induction of miR-424 led to an increase in N-cadherin and fibronectin levels without affecting levels of E-cadherin (Fig. 2I), and did so in a reversible manner (Fig. 2J). This proof-of-principle experiment demonstrates that reduction of miR-424, after it has already driven an intermediate EMT, is sufficient to reverse mesenchymal phenotypes similar to an MET.

We also validated that i-miR-424 recapitulated functions observed with constitutive miR-424 expression. Doxycycline treatment shifted i-miR-424 cells into the G₁-phase in a reversible manner (Supplementary Fig. S4B, white, gray bars, respectively). In addition, although doxycycline treatment led to an increase in migration in the control i-EV 2 clone, doxycycline increased migration to a significantly greater extent in the i-miR-424 1 clone (Supplementary Fig. S4C).

To address the temporal role of miR-424 in EMT–MET and metastasis, we decided to characterize our inducible model in vivo. By using doxycycline to manipulate miR-424 levels before or after injections (i.e., high to low miR-424, low to high miR-424, always low miR-424, always high miR-424), we aimed to determine which metastatic phenotypes miR-424 regulates and whether manipulation of miR-424 to allow completion of an EMT–MET axis was required. However, neither orthotopic nor tail vein injections resulted in metastatic burden after 70 and 90 days, respectively, with either the i-EV or i-miR-424 clones (data not shown). We were also unable to detect circulating tumor cells (CTC) after orthotopic injection. As SUM149PT cells have been recently documented to consist of varying subpopulations (25), interclonal cooperativity may be necessary for SUM149PT cells to metastasize, or the clones selected may lack the innate ability to metastasize. Regardless, in vitro reversibility of the intermediate EMT and its associated functions, by manipulating miR-424 levels, is consistent with an EMT, followed by an MET, potentially facilitating the various steps of metastasis.

Although we could not evaluate inducible miR-424 effects in vivo, the role of an intermediate EMT in vivo remained unknown. This question is important, as tumor cells have been found to exist in an intermediate state both in primary tumors and in circulation (11). Thus, we characterized our constitutive miR-424 system in an experimental metastasis mouse model. EV control and miR-424–overexpressing SUM149PT cells were tagged with firefly luciferase to allow detection of metastatic burden by in vivo imaging, and were verified to express exogenous primary and mature miR-424 before injection (Fig. 3A). Cells were subsequently injected into the left ventricle of nude female mouse hearts. Following injection, each mouse was imaged to ensure systemic luciferase signal indicative of dissemination. Over 15 weeks, a trend toward decreased metastasis incidence was observed in the miR-424 group, though this difference was not significant (Fig. 3B).

To verify that ectopic expression of miR-424 was maintained in metastases that occurred in the miR-424 group, metastatic tissues were collected using ex vivo luciferase imaging to confirm the presence of a metastatic lesion and to discard as much normal tissue as possible, before RNA extraction. Remarkably, levels of mature miR-424 were relatively similar in metastases from mice injected with either EV or miR-424 SUM149PT cells (Fig. 3C). This similarity contrasted with mature miR-424 expression levels in cells preinjection, where there was an approximately 15-fold difference in mature miR-424 levels (Fig. 3A, left). However, we found that ectopic pri–miR-424 was still expressed at relatively high levels within the miR-424 metastases (Fig. 3D). This expression of pri–miR-424 was similar to preinjection cells (Fig. 3A, right), demonstrating that miR-424 was posttranscriptionally regulated in vivo in the miR-424 metastases. Global miRNA processing did not appear affected, as multiple other mature miRNAs were not downregulated in the same metastatic tissue (Supplementary Fig. S5).

A recent study showed that, although ectopic TWIST1 increased tumor cell dissemination, subsequent downregulation of TWIST1 enhanced metastatic colonization (13). These data suggest a requirement for biphasic TWIST1 expression at different stages of the metastatic cascade, or for biphasic expression of TWIST1-induced genes that regulate various aspects of the metastatic process. Thus, we examined how miR-424 levels are regulated upon long-term TWIST1 and SNAI1 expression, especially in light of miR-424 posttranscriptional downregulation in metastases. We found that constitutive TWIST1 or SNAI1 expression in HMLE cells resulted in downregulation of mature miR-424 (Fig. 3E and Supplementary Fig. S6A), as opposed to the upregulation observed early after TWIST1 or SNAI1 induction (Fig. 1E). This downregulation is controlled posttranscriptionally, as pri–miR-424 is not repressed by either constitutive TWIST1 or SNAI1 (Fig. 3E and Supplementary Fig. S6A). Together, these data are consistent with tumor cells downregulating mature miR-424 to reverse the intermediate EMT to allow outgrowth at a secondary site.

Increasing evidence suggests that the EMT state does not always increase TIC characteristics (12, 26, 27); therefore, we asked whether miR-424 influences TIC characteristics, as this may help explain why miR-424 trends toward decreasing metastasis incidence. We first determined ALDH1 activity in response to miR-424, as ALDH1-active populations are enriched for TICs (28). Overexpression of miR-424 in SUM149PT and MCF12A cells decreased ALDH1 activity (Fig. 4A and B; Supplementary Fig. S3D), whereas alternative TIC-enriching markers [i.e., CD24, CD44 and EPCAM (ESA)] were not consistently changed. In addition, inducible miR-424 also decreased ALDH1 activity, which was reversible by doxycycline removal (Supplementary Fig. S4D, white, gray bars, respectively). The lack of CD24/CD44 changes was unexpected given the ALDH1 changes and the fact that both sets of markers enrich for a self-renewal phenotype. However, CD24⁻/CD24⁺ populations are known to have relatively little overlap with ALDH1⁺ populations (28, 29), thus, miR-424 seems to specifically affect ALDH1⁺ cells.

We then examined mammosphere formation as a functional TIC assay (30). Consistent with the effects on ALDH1, miR-424 decreased numbers of secondary mammospheres formed (Fig. 4C). Thus, miR-424 is sufficient to impart EMT-related functions on cells, while simultaneously repressing TIC characteristics, traits previously thought to positively associate with one another (1).

Our in vitro TIC data prompted us to address whether miR-424 can affect tumor initiation by performing a limiting-dilution analysis in vivo. Between 20 and 300 SUM149PT cells overexpressing miR-424 or control EV were injected into the mammary fat pads of female NOD/SCID mice, which were then palpated weekly for tumor formation. Although no differences in tumor-initiating capability were observed when larger numbers of cells were injected, injection of lower numbers of cells (20 cells) led to a significant reduction in TIC frequency due to miR-424, as assessed (19) up to 3 months (Fig. 4D), supporting miR-424 as an inhibitor of tumor initiation.

Interestingly, an oncogenic EMT has been shown to increase tumor initiation in RAS-transformed HMLE (HMLER) cells constitutively expressing TWIST1 or SNAI1 (4). Because we initially observed TWIST1 and SNAI1 increasing miR-424 levels, which is sufficient to decrease TIC properties, we attempted to reconcile the conflicting TIC data by examining miR-424 levels in HMLER cells, finding HMLER cells posttranscriptionally downregulated miR-424 (Supplementary Fig. S6B). This additional example of miR-424 posttranscriptional downregulation is consistent with decreased miR-424 as a mechanism to allow tumor initiation or metastatic outgrowth, possibly by removing signals that inhibit tumor initiation and cell proliferation.

To investigate how miR-424 controls TIC and EMT-like phenotypes, we used Gene Set Enrichment Analysis (GSEA) after performing next-generation RNA sequencing (RNA-seq) on MCF12A cells overexpressing miR-424. The nontumorigenic MCF12A cells were used to reduce potentially masking effects of broader genomic alterations in the transformed SUM149PT cells. GSEA revealed that miR-424 represses several genes that are normally downregulated in an EMT signature (7), revealing hypothetical targets of miR-424 involved in mesenchymal programming (Supplementary Fig. S7A). miR-424 also repressed genes typically upregulated in a breast cancer stem cell gene signature (3), giving insight into how miR-424 represses TIC phenotypes (Supplementary Fig. S7B). TargetScan (31) analysis of the 3′ UTRs of these downregulated genes revealed that a significant number (20 of 60, P < 10⁻⁵; 8 of 23, P = 0.002, respectively) of the most negatively associated core genes in the two signatures had predicted miR-424–binding sites (Supplementary Fig. S7, asterisks). Further GSEA analysis, using genome-wide predicted miR-424 targets (TargetScan) as a gene signature, revealed that our RNA-seq results were enriched with repressed predicted target genes (P < 0.001), demonstrating utility in coupling functional data and in silico predictions. These data suggest that miR-424 may drive mesenchymal EMT programming and inhibit TIC phenotypes by targeting multiple genes simultaneously.

To narrow down a mechanism downstream of miR-424, we validated several RNA-seq hits. Immunoblot analyses of SUM149PT lysates confirmed that miR-424 downregulates TGFBR3, likely observed as a number of bands due to known extensive posttranslational modifications (32), ST3GAL5 and TMPRSS4 (Fig. 5A). TGFBR3 was identified directly from RNA-seq differences, whereas ST3GAL5 and TMPRSS4 came through GSEA of the breast cancer stem cell and EMT signatures, respectively. Each gene also contains a predicted miR-424–binding site (TargetScan) and has not previously been identified as a target of miR-424.

Because TGFβ signaling has been implicated in both EMT and TIC phenotypes (32–35), we focused on TGFBR3 downstream of miR-424. QPCR confirmed that miR-424 decreases TGFBR3 mRNA in SUM149PT, HMLE, and MCF12A cells (Fig. 5B; Supplementary Fig. S7C and S7D), whereas a 3′ UTR reporter assay demonstrated that miR-424 can repress the TGFBR3 3′ UTR (Fig. 5C). Mutating the putative miR-424 seed-binding site in the TGFBR3 3′ UTR reporter construct significantly alleviated miR-424 repression of the reporter, though to a small degree (Fig. 5C).

To determine whether TGFBR3 inhibition, mimicking miR-424 expression, is sufficient to recapitulate an intermediate EMT, we used two shRNAs to knockdown (KD) TGFBR3 in parental SUM149PT cells (Fig. 5D). TGFBR3 KD increased vimentin and fibronectin levels, by qPCR and immunoblot, and N-cadherin, by qPCR, without affecting E-cadherin levels (Fig. 5D and E). We also determined whether TGFBR3 loss is necessary for an miR-424–induced intermediate EMT by restoring the coding sequence of TGFBR3, which cannot be repressed by miR-424 as it lacks a 3′ UTR, to miR-424–overexpressing cells (Supplementary Fig. S7E). TGFBR3 restoration reversed the mesenchymal marker levels originally increased by miR-424 while not affecting E-cadherin levels (Fig. 5F). Together, decreased TGFBR3 is necessary in mediating miR-424–dependent EMT changes and sufficient to regulate, at least in part, mesenchymal EMT programming.

Because TGFBR3 is responsible for the majority of mesenchymal changes induced by miR-424, we asked whether TGFBR3 also plays a role in repressing TIC phenotypes. We found that TGFBR3 KD, phenocopying miR-424 expression, did not lower ALDH1 activity (Fig. 6A), nor was TGFBR3 restoration able to rescue decreased ALDH1 activity downstream of miR-424 (Fig. 6B).

Subsequently, we shifted our focus to ERK signaling, as we observed a marked increase in this noncanonical TGFβ signaling arm downstream of miR-424 (Fig. 5F), and as ERK signaling is known to negatively regulate self-renewal of embryonic stem cells (36), though in other contexts ERK has been found to positively regulate self-renewal (37). Interestingly, although TGFBR3 KD increased phospho-ERK1/2 (Fig. 5E), revealing TGFBR3 can inhibit ERK, restoring TGFBR3 did not reverse miR-424′s ability to increase phospho-ERK1/2 levels (Fig. 5F), indicating that miR-424 activates the ERK pathway through multiple mechanisms.

To examine ERK signaling functionally, we used two small-molecule MEK inhibitors, AZD6244 (38) and PD98059 (39), to prevent MEK from phosphorylating ERK1/2 (Fig. 6C). ERK signaling inhibition reversed miR-424′s ability to decrease ALDH1 activity (Fig. 6D). Extending the PD98059 MEK inhibitor to functional mammosphere formation assays, we observed a partial reversal of mammosphere repression due to miR-424 (Fig. 6E). Interestingly, although MEK inhibition downstream of miR-424 increased the number of mammospheres, they were smaller in size (Fig. 6F), suggesting divergent effects of the ERK pathway on self-renewal and proliferation in mammosphere-initiating cells and the progeny that comprise the bulk of the mammosphere, respectively. We thus decided against using MEK inhibitors to examine tumor initiation, as effects on TIC properties would likely be masked by the antiproliferative properties. In light of TGFBR3 KD not altering ALDH1 activity, yet being sufficient to increase ERK signaling, these data together suggest that ERK signaling is necessary, but not sufficient, for repressing in vitro tumor-initiating phenotypes downstream of miR-424.

To determine the clinical relevance of miR-424, we assessed the expression of miR-424 and EMT-inducing transcription factors in human breast cancer datasets containing patient-matched miRNA and mRNA expression data. In two datasets (40, 41), a positive relationship existed between miR-424 and TWIST1 mRNA (Fig. 7A), in addition to TWIST2 mRNA (Fig. 7B), which has similar EMT- and TIC-associated properties as TWIST1 (42).

Recently, CTCs from triple-negative breast cancers, which largely overlap with basal tumors, and from HER2-enriched breast cancers have been shown to consist largely of mesenchymal-like and intermediate EMT-like cells (11). We thus examined The Cancer Genome Atlas (TCGA) data (43) to determine whether miR-424 levels were associated with a particular breast cancer subtype. Basal tumors had significantly higher miR-424 levels than luminal A and B tumors, whereas HER2-enriched tumors expressed significantly more miR-424 than luminal A tumors (Fig. 7C). These subtype differences are consistent with miR-424 inducing an intermediate EMT phenotype.

We then examined patient mRNA profiles from high and low miR-424 groups (40). GSEA identified a significant enrichment of three different EMT-associated gene signatures (7, 8, 44) in patient tumors that expressed high miR-424 (Fig. 7D). These data corroborate our experimental findings that miR-424 induces an intermediate EMT and suggest that miR-424 may regulate an EMT-like program in human breast cancers.

Finally, to assess miR-424 levels during cancer progression, we investigated our own independent cohort of triple-negative breast cancer patients (45) where high expression of miR-424 was expected. We found that miR-424 was upregulated in primary tumors compared with matched normal tissue (Fig. 7F), and in these primary tumors, miR-424 inversely correlated with TGFBR3 (Fig. 7E). Furthermore, miR-424 was downregulated in metastases compared with matched primary tumors (Fig. 7G), similar to what occurred in our experimental metastases. Thus, examination of multiple patient cohorts revealed that miR-424 was upregulated in primary tumors in association with EMT drivers and EMT gene signatures, but was downregulated in matched metastatic lesions. These data are consistent with our hypothesis that miR-424 expression may promote early stages of metastasis associated with an EMT, but that downregulated miR-424 facilitates metastatic outgrowth.

The oncogenic EMT has been implicated as a critical mediator of metastasis. Previous controversy in the field may have been due to the challenge of identifying human tumor cells that are histologically mesenchymal and definitively originate from a carcinoma (46). However, recent studies using experimental models and human tumors indicate that disseminating cells may exist in a more plastic state where cells transition between epithelial and mesenchymal phenotypes, or exist in an intermediate state between the two (11, 13). We thus examined an inducible EMT model to characterize a dynamic, intermediate EMT state. In doing so, we demonstrated that EMT-inducing factors TWIST1 and SNAI1 can increase expression of mesenchymal genes before repression of epithelial genes. Furthermore, we found that miR-424, downstream of TWIST1 and SNAI1, is sufficient to induce multiple mesenchymal EMT characteristics, without altering epithelial attributes, in a reversible manner. To our knowledge, this is the first demonstration of an miRNA that can control specifically the mesenchymal arm of an EMT independent of epithelial changes.

Although miR-424 induced functions associated with a full oncogenic EMT, miR-424 decreased TIC characteristics in vitro and in vivo, in contrast with other EMT-inducing factors (4, 42). The negative growth effects of miR-424, observed by others (47) and us, could be partially responsible for the TIC repression; however, over the last month of the experiment, no new miR-424 tumors were formed, whereas new control tumors formed each week. In addition, although miR-424 overexpression led to smaller mammosphere formation, possibly due to growth effects, there were importantly fewer mammospheres, possibly indicative of TIC inhibition. These data, coupled with ALDH1 repression, suggest that miR-424 may decrease tumor initiation through TIC effects and not just through growth inhibition.

Interestingly, although miR-424 did not significantly affect overt metastasis incidence, we observed posttranscriptional regulation of miR-424 in metastatic outgrowths. HMLE and HMLER cells constitutively expressing TWIST1 and SNAIL1 over time demonstrated posttranscriptional reduction of miR-424. Although posttranscriptional regulation of miR-424 has also been observed in endothelial cells (48), the mechanism remains unknown. Posttranscriptional regulation of miR-424 in the constitutive TWIST1 and SNAI1 HMLER cells, which have increased TIC frequency (4), may have allowed those cells to overcome the inhibitory tumor-initiation effects of miR-424, while maintaining other protumorigenic phenotypes mediated by TWIST1 or SNAI1.

Examination of how miR-424 mediates an intermediate EMT led to the identification of TGFBR3. Previous studies demonstrated that TGFBR3 decreases during breast cancer progression (49), as well as during an EMT in prostate cancer before E-cadherin downregulation (33); similarly during an EMT, miR-424 increases before E-cadherin downregulation. Importantly, TGFBR3 KD, while increasing vimentin and fibronectin protein levels, did not increase N-cadherin protein. These data suggest that, although TGFBR3 functions downstream of miR-424 for induction of the mesenchymal phenotype, it does not act alone, and may act in combination with other genes identified by GSEA.

To repress TIC attributes, we found that miR-424 requires ERK signaling. A similar negative TIC role for ERK signaling has been shown in normal and tumorigenic settings (36, 50), but ERK signaling can also increase TIC properties (37), implicating important contextual differences. In addition, because ERK signaling is necessary, but not sufficient, for repression of TIC phenotypes, we speculate that numerous effectors of miR-424 are responsible for mediating TIC phenotypes, particularly as miR-424 represses part of a breast cancer stem cell signature.

Clinically, we found multiple associations with miR-424 that parallel our experimental data. Analysis of primary breast tumors from four independent cohorts demonstrated a positive link between miR-424 and EMT-associated molecular changes. Our own cohort revealed increased miR-424 in primary tumors compared with matched normal tissue, with an inverse correlation between TGFBR3 and miR-424 in the same tumors. Critically, miR-424 was decreased in metastases compared with matched primary tumors. These human data are consistent with our experimental findings that miR-424 induced a mesenchymal-like phenotype, but that mature miR-424 was downregulated in metastases in vivo, supporting the possibility that decreased miR-424 at a distant site may facilitate metastatic outgrowth (Supplementary Fig. S8).

Recently, miR-424 was shown to induce a full oncogenic EMT (51), though we have not observed alterations in the epithelial EMT arm in four different models. Comparing techniques may help reconcile differing results, as ectopic miR-424 expression was achieved through very different approaches. Nonetheless, our findings contribute to other studies dissociating mesenchymal and TIC phenotypes (12, 26, 27). These studies demonstrate that an EMT may only be relevant to the metastatic cascade in the context of an EMT–MET spectrum. Critically, the intermediate EMT state has been observed in human tumor cells (9–11). Our findings that miR-424 drives an intermediate EMT, where the mesenchymal N-cadherin protein is coexpressed with the epithelial E-cadherin protein, suggest that miR-424 may poise cells for movement along the epithelial–mesenchymal spectrum as a means to mediate plasticity that facilitates an MET and metastatic outgrowth.

As more experimental studies demonstrate roles for both EMT and MET in metastatic progression, we must reevaluate whether clinically targeting EMT-inducing factors will be uniformly efficacious. The position of a tumor cell along the EMT–MET axis, as well as physical location in a patient, may dictate different responses to novel therapeutics. Thus, obtaining a better understanding of the role of plasticity within the metastatic cascade will help identify which patients will benefit from potential EMT-related therapies.

No potential conflicts of interest were disclosed.

Conception and design: D.J. Drasin, H.L. Ford

Development of methodology: D.J. Drasin, J.H. Cabrera

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): D.J. Drasin, A.L. Guarnieri, D. Neelakantan, J.H. Cabrera, V. Zaberezhnyy, P. Gasparini, H.L. Ford

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): D.J. Drasin, A.L. Guarnieri, J. Kim, P. Gasparini, L. Cascione, K. Huebner, A.-C. Tan, H.L. Ford

Writing, review, and/or revision of the manuscript: D.J. Drasin, J. Kim, K. Huebner, A.-C. Tan, H.L. Ford

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): D.J. Drasin, J.H. Cabrera, C.-A. Wang

Study supervision: H.L. Ford

The authors thank Paul Jedlicka for examining tumor sections, as well as the UCCC Functional Genomics Facility, the Tissue Culture Shared Resource and Lester Acosta, Christine Childs, and Karen Helm of the Flow Cytometry Shared Resource for technical expertise and reagents (P30CA046934). Robert Weinberg (Massachusetts Institute of Technology), Stephen Ethier (Medical University of South Carolina), and Gerard Blobe (Duke University) all generously provided cell lines or reagents. The authors thank members of the Ford lab for critical reading of this article.

This work was supported by grants from the National Cancer Institute (R01-CA095277) and from the Metavivor Foundation (H.L. Ford). D.J. Drasin was supported by an NCI predoctoral fellowship (F31-CA165617), and A.L. Guarnieri was supported by a DOD Breast Cancer Research Program Predoctoral Fellowship (W81XWH-10-1-0296). P. Gasparini, L. Cascione, and K. Huebner were supported by a grant from the National Cancer Institute (U01-154200).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.