miR-1199-5p and Zeb1 function in a double-negative feedback loop potentially coordinating EMT and tumour metastasis

Epithelial tumour cells can gain invasive and metastatic capabilities by undergoing an epithelial–mesenchymal transition. Transcriptional regulators and post-transcriptional effectors like microRNAs orchestrate this process of high cellular plasticity and its malignant consequences. Here, using microRNA sequencing in a time-resolved manner and functional validation, we have identified microRNAs that are critical for the regulation of an epithelial–mesenchymal transition and of mesenchymal tumour cell migration. We report that miR-1199-5p is downregulated in its expression during an epithelial–mesenchymal transition, while its forced expression prevents an epithelial–mesenchymal transition, tumour cell migration and invasion in vitro, and lung metastasis in vivo. Mechanistically, miR-1199-5p acts in a reciprocal double-negative feedback loop with the epithelial–mesenchymal transition transcription factor Zeb1. This function resembles the activities of miR-200 family members, guardians of an epithelial cell phenotype. However, miR-1199-5p and miR-200 family members share only six target genes, indicating that, besides regulating Zeb1 expression, they exert distinct functions during an epithelial–mesenchymal transition.

A n epithelial-to-mesenchymal transition (EMT) as well as its reversal, a mesenchymal-to-epithelial transition (MET), reflect two gradual, well-controlled processes during embryogenesis and wound healing in adults to promote tissue and organ formation and homeostasis. Both processes induce a global reorganization of a cell's constitution and allow switching back and forth between two different cell phenotypes to endow the necessity of tissue plasticity 1 . In the context of malignant tumour progression, epithelial tumour cells can undergo an EMT upon diverse extracellular stimuli and consequently gain metastatic capabilities. Among many growth factors and environmental cues, such as tissue hypoxia, transforming growth factor β (TGFβ) strongly activates the dedifferentiation process in epithelial tumour cells and, thus, induces global changes in a cell's transcriptional and post-transcriptional networks [2][3][4] . This allows tumour cells to disseminate from the primary tumour and to intravasate and survive in the blood circulation. At the distant organ, cells extravasate into the organ parenchyma and eventually grow out as lethal metastases, possibly promoted by a MET [5][6][7][8] . An EMT also provides cells with increased chemoresistance, thus impeding efficient therapy of malignant mesenchymal cancer cells 9,10 .
MicroRNAs (miRNAs) represent a class of~22 nucleotide-long non-coding RNAs that can regulate gene expression at the posttranscriptional level by either inducing target messenger RNA (mRNA) degradation or preventing mRNA translation [11][12][13] . In the context of an EMT, miRNA-200 family members (miR-200a/ b/c, miR-141 and miR-429) take a central stage: they are required to maintain an epithelial cell morphology by degrading the transcripts of the EMT-inducing transcription factors (TFs) Zeb1 and Zeb2 [14][15][16][17] . Members of the miR-200 family bind to specific seed sequences in the 3′ untranslated region (3′ UTR) of Zeb1 and 2 mRNAs and destabilize them. During an EMT, members of the miR-200 family are downregulated in their expression, which results in the increased expression of Zeb1 and 2. Conversely, Zeb1 and Zeb2 directly suppress the transcription of miR-200 family members [14][15][16][17] . Such a double-negative feedback loop is a major example for a reciprocal TF-miRNA regulation during an EMT. Similar other molecular switches also regulate EMT/MET plasticity and malignant tumour progression [18][19][20][21][22] .
Here, we report the identification of miR-1199-5p, as a repressor of EMT, tumour cell invasion and metastasis, which comparable to miR-200 family members targets Zeb1 mRNA for degradation. Conversely, Zeb1 represses the expression of miR-1199-5p and of the miR-200 family. However, miR-200 family members and miR-1199-5p seem to exert distinct functions; they share only six of their many target mRNAs, among them Zeb1.

Results
Identification of EMT-associated miRNAs. To identify regulatory miRNAs involved in the gradual process of an EMT, we performed miRNA sequencing on a detailed time course of a TGFβ-induced EMT in normal murine mammary gland cells (NMuMG subclone E9; NMuMG/E9). Analysis of the kinetics of miRNA transcript regulation during an EMT in a time-resolved manner identified 32 differentially expressed miRNAs. Unsupervised hierarchical clustering illustrated that approximately half of the differentially expressed miRNAs showed a continuous increase in their expression during an EMT, whereas the other half exhibited decreased expression (Fig. 1a). In order to identify the miRNAs functionally impacting on a TGFβ-induced EMT, we performed a microscopy-based screen in which NMuMG/E9 cells were transfected with miRNA mimics and cultured in the absence or presence of TGFβ for 4 days (Fig. 1b). Subsequently, mesenchymal cell characteristics were monitored by high-content fluorescence microscopy analysis and quantified, including the deposition of the extracellular matrix protein fibronectin and the formation of focal adhesions and of actin stress fibres 23 . Nine out of 32 differentially expressed miRNAs were able to block or at least delay the EMT process: miR-125b-5p, miR-181b-2-3p, miR-1247-3p, miR-200a-3p, miR-200b-3p, miR-429-3p, miR-1199-5p, miR-145a-3p and miR-504-5p. Conversely, the forced expression of miR-145a-5p and miR-6944-3p promoted an EMT in epithelial NMuMG/E9 cells. Changes in cell morphology, the localization of the epithelial adhesion junction protein E-cadherin as well as the mRNA levels of E-cadherin, N-cadherin, fibronectin and Zeb1 further confirmed the impact of the different miRNAs on the EMT process ( Supplementary Fig. 1a-c).
Increased cell migration is a functional output of EMT, which allows tumour cell invasion and dissemination into the surrounding tissue 1 . In a second screening step, we identified those miRNAs that were able to affect mesenchymal tumour cell migration (Fig. 1c). Mesenchymal, highly migratory and tumorigenic Py2T cells that have been treated >20 days with TGFβ 24 were transiently transfected with mimics of EMTaffecting miRNAs. Four out of the 11 miRNAs tested significantly reduced cell migration in a trans-well Boyden chamber assay: miR-200b-3p, miR-429-3p, miR-1199-5p, and to a lesser extent miR-125b-5p ( Supplementary Fig. 2a, b). As a result, our experimental strategy identified miRNAs whose transcriptional regulation and function affected both a TGFβ-induced EMT and mesenchymal breast cancer cell migration (Fig. 1d). Four miRNAs significantly fulfilled these criteria: the miRNA-200 family members miR-200b-3p and miR-429-3p, and miR-125b-5p and miR-1199-5p. Since ectopic expression of miR-1199-5p, an EMT-regulatory miRNA, seemed to affect EMT with similar potency as the well-studied miR-200 family members 15,17,18 , we further investigated the functional role and the mechanisms of action of miR-1199-5p in the regulation of an EMT.
miRNA-1199-5p inhibits EMT and tumour cell invasion. The continuous downregulation of miR-1199-5p expression during a TGFβ-induced EMT was due to transcriptional repression as determined by a miR1199-promoter/luciferase-reporter assay during a TGFβ-induced EMT in NMuMG/E9 cells and Py2T murine breast cancer cells, and in mesenchymal Py2T cells and metastatic 4T1 murine breast cancer cells treated with TGFβ for >20 days (Fig. 2a, Supplementary Fig. 3a, b). Analysis of miR-1199 expression in different human breast cancer cell lines 25 further confirmed a higher expression in epithelial than in mesenchymal breast cancer cells ( Supplementary Fig. 3c).
The forced expression of miR-1199-5p by the transfection of a construct encoding a 1199-5p miRNA mimic caused sustained epithelial cell morphology in NMuMG/E9 and in human untransformed mammary gland MCF10A cells induced to undergo an EMT by TGFβ treatment (Fig. 2b-e, Supplementary  Fig. 3d-f). The cells maintained their epithelial morphology and the cell surface localization of the adherens junction protein Ecadherin and of the tight junction protein ZO-1 at the plasma membrane, while miR-Ctr-transfected cells lost these epithelial characteristics and progressed with an EMT (Fig. 2b, Supplementary Fig. 3d). The reorganization of cortical actin to stress fibres as well as the formation of focal adhesions were also prevented by miR-1199-5p mimics (Fig. 2c). Mesenchymal markers, such as fibronectin, vimentin or N-cadherin were repressed at the protein (Fig. 2d, Supplementary Fig. 3e) and mRNA (Fig. 2e, Supplementary Fig. 3f) expression levels in miR-1199-5p-transfected cells. Furthermore, miR-1199-5p induced a significant decrease in Zeb1 mRNA, however, transcript levels of other key EMT TFs, such as Zeb2 26 or Sox4 27 remained unchanged (Fig. 2e).
A gain in cell migration and invasion can be one of the consequences of a TGFβ-induced EMT and allows tumour cells to leave the primary tumour and intravasate into the blood circulation and to extravasate at a distant organ 1,5 . Ectopic expression of miR-1199-5p displayed only minor effects on the early stages of an EMT in Py2T cells ( Supplementary Fig. 3g-i), yet it induced a significant reduction in migration and invasion of mesenchymal (>20 days TGFβ) Py2T cells, as analysed by transwell Boyden chamber assays (Fig. 2f). The inhibitory effects of miR-1199-5p on in vitro cell migration and invasion were also observed in 4T1 cells treated for >20 days with TGFβ (Fig. 2g). In summary, miR-1199-5p is sufficient to sustain an epithelial cell phenotype.
miR-1199-5p directly targets Zeb1 mRNA. The zinc-finger Ebox-binding homeobox TF Zeb1 has a well-established role as transcriptional repressor of epithelial genes and, hence, as an activator of an EMT 3, 31 . Zeb1 function is associated with increased stemness, cell survival and metastasis 32,33 , and high levels of Zeb1 have been linked to aggressive breast cancer subtypes, therapy resistance, high risk for distant metastasis and poor survival 34,35 . As expected, Zeb1 transcript levels were increased during a TGFβ-induced EMT of different murine and human cellular systems ( Supplementary Fig. 4a). Furthermore, smallinterfering RNA-mediated knockdown of Zeb1 maintained an epithelial cell morphology in NMuMG/E9 and MCF10A cells induced to undergo an EMT and significantly reduced the migratory properties of mesenchymal (>20 days TGFβ) Py2T and 4T1 cells in trans-well Boyden chamber assays (Supplementary Fig. 4b-e).
We next examined whether miR-1199-5p indeed regulated Zeb1 levels during an EMT. Ectopic expression of miR-1199-5p in NMuMG/E9 and Py2T cells induced to undergo an EMT resulted in the stabilization of cell junction protein E-cadherin, while nuclear levels of Zeb1 were reduced (Fig. 4b, Supplementary  Fig. 4f). Immunoblotting analyses for Zeb1 confirmed the repression of Zeb1 expression by miR-1199-5p in these murine cells (Fig. 4c, Supplementary Fig. 4g). Even though the seed match of human miR-1199-5p is not perfectly complementary to the seed sequence in the Zeb1 3′ UTR (mouse and human miR-1199-5p seed matches differ in one base), forced expression of hsa-miR-1199-5p still significantly reduced Zeb1 mRNA (Supplementary Fig. 3f) and protein ( Supplementary Fig. 4h) levels in human MCF10A cells. These results further imply that even imperfect binding of a miRNA to its target RNA can efficiently downregulate its expression.
To validate the direct regulation of Zeb1 by miR-1199-5p on the post-transcriptional level during an EMT we made use of two Zeb1 3′ UTR luciferase reporter constructs, one containing the species conserved, wild-type miR-1199-5p seed sequence and the other one carrying a mutated version of the seed sequence with five nucleotides exchanged (Fig. 4a, d). Transient transfection of a miR-1199-5p mimic and the Zeb1 3′ UTR wild-type reporters in NMuMG/E9 cells revealed a significant decrease in luminescence, which was not observed with the mutant version of the reporter (Fig. 4d). Together, these data identify the key EMT TF Zeb1 as a direct target of miR-1199-5p, which thus represses Zeb1 expression at the post-transcriptional level and prevents an EMT.
Zeb1 directly controls the expression of miR-1199-5p. An important mechanism that appears to be responsible for EMT cell plasticity are double-negative feedback loops between miRNAs and key EMT TFs, functioning as molecular switches for various cell differentiation states 18-20, 22, 36, 37 . Because miR-1199-5p regulates the expression of Zeb1, we assessed whether Zeb1 would in turn regulate the expression of miR-1199-5p during an EMT. siRNA-mediated ablation of Zeb1 expression in NMuMG/E9 cells cultured in the presence of TGFβ for 4 days and transfected with either a miR-1199-promoter luciferase-reporter (pGL4 miR-1199 promoter) or a control promoter luciferase-reporter (pGL4; Supplementary Fig. 3b) revealed a significant increase in miR-1199 promoter activity upon loss of Zeb1 (Fig. 5a, left). Conversely, transient overexpression of Myc-tagged Zeb1 in epithelial NMuMG/E9 cells significantly decreased miR-1199 promoter activity (Fig. 5a, right). Furthermore, the regulation of miR-1199 promoter activity by loss or gain of function experiments of Zeb1 also correlated with an increase or decrease in endogenous miR-1199-5p transcript levels, respectively (Fig. 5b).
The gene encoding for miR-1199-5p is located within the first CDS of an unknown, protein-coding gene (2210011C24Rik) on chromosome 8 in the mouse genome. Its localization is conserved in the human genome, where it is located in the first CDS of LOC113230, an orthologue of 2210011C24Rik, on chromosome 19 ( Supplementary Fig. 5a). Notably, the transcript levels of the murine host gene are significantly downregulated during a TGFβinduced EMT in different cellular models, as it was observed for a b c   Fig. 5b; Fig. 2a). Furthermore, gain and loss of function studies by siRNA-mediated ablation or transient overexpression of Zeb1 in NMuMG/E9 cells led to an increase or decrease in 2210011C24Rik expression, respectively ( Supplementary Fig. 5c, d), indicating a co-regulation of miR-1199-5p and its host gene.
The murine miR-1199/2210011C24Rik promoter region encompasses several E-box motifs (CANNTG; N = G or C) as potential Zeb1-binding sites (Fig. 5c). We next performed chromatin immunoprecipitation (ChIP) for endogenous Zeb1 in epithelial Py2T cells and in 4 days TGFβ-treated Py2T cells, a time point displaying a robust increase in Zeb1 expression ( Supplementary Fig. 4a), followed by quantitative RT-PCR analysis with various primer pairs covering different regions of the miR-1199/2210011C24Rik promoter. These experiments confirmed a direct binding of Zeb1 to a region with the closest proximity (+51/−52 bps) to the transcriptional start site (TSS) (Fig. 5d). Notably, Zeb1 binding was only observed in cells undergoing an EMT and not in their epithelial counterparts (Fig. 5d). Using the miR-1199/2210011C24Rik promoter luciferase reporter ( Supplementary Fig. 3b), we individually mutated each of four Zeb1 E-box-binding sites (CAGGTG to CATTTG). Only mutation of E-box 1 (−18 bp from TSS) significantly ablated Zeb1-mediated repression of luciferase activity in NMuMG/E9 cells (Fig. 5e).
Together, the results show that Zeb1 is a direct transcriptional repressor of the miR-1199 gene. Hence, Zeb1 and miR-1199-5p act in a reciprocal fashion to repress each other.
Comparable to miR-1199-5p, miR-200b-3p and miR-429-3p transcript levels are also strongly reduced during a TGFβ-induced EMT in NMuMG/E9 cells ( Supplementary Fig. 6a, Fig. 2a). The forced expression of miR-200b-3p or miR-429-3p in human MCF10A ( Supplementary Fig. 6b, top) and murine NMuMG/E9 cells ( Supplementary Fig. 1a-c) also prevented an EMT and  . c Schematic presentation of the genomic localization of the murine miR-1199 gene and its promoter region. Red: mmu-miR-1199; grey: 2210011C24Rik gene; green: E-boxes (CANNTG, N = G or C); blue: promoter fragments examined by ChIP-qPCR analysis. d Zeb1 directly binds to the miR-1199 promoter. Chromatin of Py2T cells cultured in the absence (green) and presence of TGFβ (red) was subjected to chromatin immunoprecipitation with antibodies against Zeb1 followed by RT-PCR analysis using primers amplifying different regions of the miR-1199 promoter illustrated in c. An intergenic region was used as negative control. Data were normalized to control IgG and are presented as mean fold enrichment above background±s.e.m. (n = 3; significance determined by an unpaired, two-sided t test; *P < 0.05, **P < 0.01, ****P < 0.0001). e Analysis of E-box-binding motifs in the miR-1199 promoter as potential Zeb1-binding sites. NMuMG/E9 cells were transfected and cultured as described in a (right). Relative luminescence (Firefly/Renilla) was calculated as described in a (mean fold changes±s.e.m.; n = 3; significance determined by an unpaired, two-sided t test; *P < 0.05, ***P < 0.001) maintained an epithelial morphology in the presence of TGFβ. However, similar to miR-1199-5p, expression of miR-200b-3p and miR-429-3p mimics in tumorigenic Py2T cells failed to completely block mesenchymal cell morphology, even though the mRNA levels of E-cadherin and Zeb1 were significantly increased or decreased, respectively ( Supplementary Fig. 6b, c). Also comparable to miR-1199-5p, ectopic expression of miR-200b-3p or miR-429-3p significantly reduced cell migration and invasion of mesenchymal Py2T and 4T1 cells ( Supplementary Fig. 2, Supplementary Fig. 6d). Notably, all three miRNAs are mechanistically embedded in a double-negative feedback regulation with Zeb1 18, 19 (Fig. 4, Fig. 5, Supplementary Fig. 6e).
Since miR-200b-3p, miR-429-3p and miR-1199-5p induced comparable functional outputs with regard to an EMT process, we assessed their effects on tumour progression in vivo. Stable expression of miR-1199, miR-200b (which also induced the expression of miR-429 and slightly miR-1199; Supplementary  Fig. 6f) and miR-429 in metastatic 4T1 cells tagged by ZsGreen led to a reduction in Boyden chamber trans-well cell migration compared to an empty vector control (Supplementary Fig. 6g) and confirmed our results for their transient overexpression (Fig. 2g, Supplementary Fig. 6d).
Upon orthotopic transplantation of 4T1 cells into the mammary fat pads of immunodeficient NSG and NMRI mice, the forced expression of miR-1199, miR-429 or miR-200b induced a significant reduction in primary tumour growth over time (Fig. 6a). Notably, the high potential of 4T1 cells to metastasize to the lungs was reduced by the expression of miR-1199 and miR-429, however not by the expression of miR-200b ( Fig. 6b-d). The failure of miR-200b to repress 4T1 metastasis was surprising, yet has been observed by others as well (G.J. Goodall, personal communication of unpublished data). Examining the metastatic outgrowth of 4T1 cells in the lung, expression of each of the three miRNAs significantly increased the size of the few metastatic nodules compared to the many nodules observed with empty vector-transduced cells (Fig. 6e). Therefore, the repressive growth effect observed for miR-1199, miR-200b and miR-429 in the primary tumour was not maintained in the outgrowth of tumour cells at the metastatic site. Of note, the number of circulating tumour cells isolated from the blood of tumour-bearing mice was decreased by the forced expression of miR-1199 and miR-429, but not by the expression of miR-200b (Fig. 6f, g).
In summary, miR-1199 and miR-429, but not miR-200b, are sufficient to reduce tumour cell intravasation into the blood circulation and the seeding of lung metastases. However, all three miRNAs seem to promote metastatic outgrowth once tumour cells have seeded in the lung parenchyma.
Common and distinct targets of miR-200s and miR-1199-5p. The functional similarities between miR-1199-5p and miR-200 family members observed during an EMT in vitro as well as during tumour progression in vivo begs the question about the shared and the distinct functions of miR-200 family members and miR-1199-5p in regulating an EMT. To reveal the transcriptomic effects of the various miRNAs during a TGFβ-induced EMT, we performed RNA sequencing of NMuMG/E9 cells transfected with miRNA mimics for miR-200b-3p, miR-429-3p and miR-1199-5p. As expected, miR-200b-3p or miR-429-3p mimics induced a block in EMT of NMuMG/E9 cells and led to an overall anticorrelative gene expression profile compared to mesenchymal, miR-Ctr-transfected cells (r = −0.505 and r = −0.51, respectively; Fig. 7a), similar to the profile observed with miR-1199-5p mimic (r = −0.324; Fig. 3a). Differential gene expression analysis (log2FC (±1); FDR <0.05) revealed 1097 and 1058 genes affected by miR-200b-3p and miR-429-3p, respectively, during an EMT, of which 982 genes were shared by the two miRNAs (Fig. 7a, b). Since they belong to the same miRNA family, such a high number of coregulated genes can be explained by an identical seed sequence on their target mRNAs. However, the difference in the effects of the two miRNAs on lung metastasis is surprising (Fig. 6b-d), and we can only speculate whether the small pool of individual target genes for each of the miRNAs is the reason for the different outcomes in vivo.
Directly compared to each other, miR-200b-3p, miR-429-3p and miR-1199-5p share 465 regulated genes during an EMT, which is more than half of the 787 genes regulated by miR-1199-5p alone (Fig. 7b, Fig. 3a). Functional annotation analysis revealed that these genes control the same biological processes and pathways as miR-1199-5p alone, that is, cell adhesion, ECMreceptor interactions and focal adhesions (Fig. 7c, Fig. 3b). A total of 267 genes are exclusively regulated by miR-1199-5p during an EMT.
The results show that miR-1199-5p, miR-200b-3p and miR-429-3p tightly control TGFβ-induced EMT plasticity with similar potency, however, they only share the regulation of six gene transcripts, one of which is the critical EMT TF Zeb1. Yet, each of the miRNAs seems to have distinct functions by repressing a larger number of unique target mRNAs.

Discussion
A cancer-associated EMT can drive early steps of the metastatic cascade by converting epithelial tumour cells into invasive metastatic cancer cells 1,5 . MiRNAs are powerful, posttranscriptional regulators of a cell's transcriptome and, therefore, they are able to coordinate changes of a cell's morphology and functional capabilities necessary for an EMT [11][12][13] . In the present study, we have set out to identify and characterize such EMT-regulatory miRNAs in normal and tumorigenic breast cancer cells. Using miRNA-sequencing analysis of the kinetics of EMT, we identified 32 strongly regulated miRNAs by a global, time-resolved miRNA profile of TGFβ-induced EMT. Subsequently, we have tested the miRNAs' functional contribution to an EMT and to mesenchymal tumour cell migration. Among a number of miRNAs, we identify miR-1199-5p as a strong regulator of an EMT. Its transcript levels as well as its promoter activity are continuously decreased during an EMT. Furthermore, we demonstrate that miR-1199-5p downregulation is required for cell dedifferentiation, for mesenchymal tumour cell migration and invasion, and for lung metastasis formation in vivo. Finally, we report that during an EMT, this exonic miRNA is mechanistically embedded in a reciprocal, direct negative feedback loop with the TF Zeb1. The double-negative feedback loop with Zeb1 during an EMT is reminiscent of the miR-200 family members 14-16, 18, 19 . Indeed, consistent with previous reports, the miR-200 family members miR-200b-3p and miR-429-3p have been identified by our EMT screen as critical mediators of an epithelial cell phenotype [14][15][16][17] .
While the gain of function approaches of miR-1199-5p and miRNA family members has clearly repressed an EMT, we have  Fig. 7). From these data, we conclude that the regulation of Zeb1 expression on the transcriptional and posttranscriptional level is complex and we suppose that the loss of one Zeb1 regulator (miR-1199-5p or miR-200b/c-3p) is not sufficient to increase Zeb1 levels and to jumpstart an EMT. The overlapping doublenegative feedback loops between miRNAs and TFs may act as a 'buffered system' to compensate for the loss of one regulator and thus ensure a system's functionality.
Our genome-wide analyses of the three miRNAs revealed a large number of regulated genes, which underscores their critical functions during an EMT. However, only six genes were identified as common direct targets of the three miRNAs, including Zeb1, Col5a1, Zfp9, Sox12, Cdon and Ncs1. The latter five are so far 'unknowns' in the context of an EMT and malignant tumour progression, most of which might be of interest considering their potential function. For instance, Sox12, together with Sox11 and   Sox4, is a member of the SoxC TF family, of which Sox4 has an established role in EMT and breast cancer progression 27 . The cell surface receptor encoded by Cdon, similar to the EMT inducer neuronal cell adhesion molecule, consists of immunoglobin/ fibronectin type III domains known to mediate cell signalling 38 . Finally, collagen type V alpha 1 (Col5a1) is mainly found in organ tissue together with collagen type I, a major ECM component of breast cancer. Furthermore, Col5a1 facilitates breast cancer formation, invasion and metastasis 39,40 .
RNA interference. miRNA mimic transfection: NMuMG/E9, MCF10A, Py2T and 4T1 cells were transfected with 20 nM or 50 nM of mouse or human pre-miR miRNA precursors (Ambion) by using Lipofectamine RNAiMax (Invitrogen) according to the manufacturer's instructions. A random sequence has been used as negative control (Ambion, AM17110 control #1). To maintain a miRNA's overexpression, the transfection was repeated every third day. All pre-miR miRNA precursors used in this study are listed in Supplementary Table 3. miR-1199/2210011C24Rik promoter reporter: Cells were plated in triplicates in a 24-well plate and treated with TGFβ. Three days before measuring luciferase activity, cells were transfected with 0.5 μg of pGL4-miR-1199/2210011C24Rik or pGL4 Firefly luciferase promoter reporter by using Lipofectamine 3000 (Invitrogen). Additionally, cells were transfected with 10 ng of pRL-CMV (Promega), which encodes a Renilla luciferase and was used for subsequent cell number normalization.
To test miR-1199/2210011C24Rik promoter activity upon Zeb1 gain and loss of function studies, NMuMG/E9 cells were plated as described before and reverse transfected with 20 nM of siRNAs against Zeb1 or a negative control (Ambion). On the next day, cells were transfected with the different reporter constructs mentioned above and treated with TGFβ for 3 days. For Zeb1 gain of function studies, NMuMG/E9 cells were forward transfected with 50 ng of pCS3-6xMyc-tag or pCS3-6xMyc-tag-Zeb1 along with the different luciferase reporters.
RNA isolation and quantitative RT-PCR. In order to quantify gene transcripts, total RNA was isolated using Tri Reagent (Sigma-Aldrich) following the manufacturer's instructions. Complementary DNA (cDNA) was generated by reverse transcription of RNA using M-MLV reverse transcriptase (Promega) and was quantified by real-time PCR using Mesa Green qPCR MasterMix plus (Eurogentec). Riboprotein L19 was used as internal normalization control. FCs were calculated using the comparative Ct method (ΔΔCt). Primers used for quantitative RT-PCR are listed in Supplementary Table 4. For quantitative miRNA analysis, total RNA was isolated using miRNeasy kit (Qiagen). Mature miRNAs were reverse-transcripted by using the miRCURY LNA Universal RT miRNA PCR kit (Exiqon). MiR-1199-5p (Exiqon; 206004) expression was measured by RT-PCR and normalized to U6 small nuclear RNA (Exiqon; 203907) expression as internal control. FCs in miRNA expression were calculated using the comparative Ct method (ΔΔCt).
Immunofluorescence microscopy. Cells were grown on uncovered glass cover slips (Menzel-Glaser), fixed with 4% paraformaldehyde/PBS for 20 min, permeabilized with 0.5% NP40 for 5 min and blocked with 3% BSA/0.01% Triton X-100/ PBS for 30 min. Afterwards, cells were incubated at room temperature with the primary antibodies (listed above) diluted in 3% BSA/0.01% Triton X-100/PBS for 1.5 h, followed by an incubation with a fluorophore-coupled secondary antibody (Alexa Fluor, Invitrogen) for 1 h. Cell nuclei were visualized with DAPI (Sigma-Aldrich). After staining, the cells were mounted in fluorescence mounting medium (Dako) on microscope slides and imaged using fluorescence microscopy (Leica DMI 4000).
EMT phenotypic microscopy-based screen. The general setup of the screen has been described recently 23 . In brief, NMuMG/E9 cells were plated as duplicates in a 96-well plate and reverse transfected with 20 nM pre-miR miRNA precursors (Ambion; Supplementary Table 4) using Lipofectamine RNAiMax (Invitrogen) 2 days prior to the addition of TGFβ for 4 days. Epithelial control cells were transfected for 3 days in the absence of TGFβ. For immunofluorescence analysis, cells were processed as described above and mesenchymal characteristics, such as formation of focal adhesions (paxillin), actin stress fibres (phalloidin) and fibronectin deposition, were stained as described in ref. 23 . Images of cells in a 96-well plate were taken with an Operetta HCS microscope (Perkin Elmers). EMT features were quantified (Columbus software, version 2.5.0) and the median of two replicates was calculated and compared to the median value of miR-Ctr-transfected cells. Standard deviations (s.d.) were estimated. MiRNAs showing at least 3× s.d. for all three EMT read-outs or 4× s.d. in two EMT read-outs aside from the median value of miR-Ctr-transfected cells were considered as hits to block or induce EMT. Some miRNAs, which turned out to be no hits in the EMT phenotypic microscopybased screen, but maintained an epithelial cell morphology of NMuMG/E9 cells cultured in the presence of TGFβ (judged by eye), were also considered hits.
Trans-well migration and invasion assays. Boyden chamber invasion assay (24well plate format): mesenchymal Py2T and 4T1 cells were reverse transfected with 50 nM of pre-miR miRNA precursors in a six-well plate. Two days later, the cells were trypsinized, washed once with PBS and counted. A total of 25,000 cells were plated as duplicates in a 24-trans-well migration or invasion insert (membrane with 8 μm pores, covered with a layer of growth factor reduced Matrigel; BD Biosciences). After 18 h, the cells were fixed with 4% paraformaldehyde/PBS, nuclei were stained with DAPI (Sigma-Aldrich) and non-migrated/invaded cells on the upper surface of the membrane were removed with a cotton swab. Migrated/ invaded cells on the bottom of the membrane were imaged with a fluorescence microscope (Leica DMI 4000) and quantified. As chemo-attractant, a gradient of 0.2-20% FCS was used.
Boyden chamber migration screen (96-well plate format): mesenchymal Py2T cells were reverse transfected with 50 nM of pre-miR miRNA precursors in a 96-well plate. After 2 days, cells were trypsinized and washed with PBS in a roundbottom 96-well plate (Corning). MiR-Ctr-transfected cells were counted and 6000 cells were plated as duplicates in the inserts of a 96-well FluoroBlok fluorescentblocking high-density positron emission tomography insert (BD Biosciences) and, in parallel, in a 96-well reference plate (Cell Carrier 96, Perkin Elmers). The same cell suspension volume was used to plate cells transfected with other pre-miR miRNA precursors. As chemo-attractant, a gradient of 0.2-20% FCS was used. Migrated cells (bottom of membrane of migration insert) as well as cells in the reference plate were stained with DAPI and imaged with an Operetta HCS microscope (Perkin Elmers), quantified (Columbus software, version 2.5.0) and normalized to each other.
Chromatin immunoprecipitation. ChIP experiments were performed as previously described 44 . Briefly, crosslinked protein-bound DNA of Py2T cells was sonicated (Bioruptor, Diagenode) to achieve chromatin fragments of an average size of 300 bps. For ChIP of endogenous Zeb1, 150 μg of chromatin was incubated with 10 μg of Zeb1 antibody (sc-25388) and immunocomplexes were precipitated with 40 μl of pre-blocked magnetic Protein G beads (Invitrogen). Immunocomplexes were eluted from the beads, de-crosslinked, and genomic DNA was purified by phenol/chloroform extraction and precipitated with sodium acetate. One out of forty of the ChIP sample and 1% of input DNA were used for quantitative RT-PCR. Fold enrichments for specific miR-1199 promoter regions were calculated by IP over input samples and normalized to isotype-specific IgG as negative control. Microarray data and analysis. In order to compare the expression of the miR-1199-5p in human cells, we downloaded raw data (CEL files) of the experiment GSE32474 25 from the Gene Expression Omnibus database. Data was analysed in R (https://www.r-project.org)/Rstudio (https://www.rstudio.com) using Bioconductor add-on packages. The background signal correction, normalization and summarization were performed by robust multiarray averaging 45 from the affy package 46 . To identify differences in gene expression between epithelial and mesenchymal samples, the linear models for microarray data (limma) 47 package was used.
miRNA sequencing and analysis. Total RNA was isolated from NMuMG/E9 cells undergoing TGFβ-induced EMT by using the miRNeasy Mini Kit (Qiagen) according to the manufacturer's instructions. Quality and quantity of total RNA was analysed by using the RNA 6000 Pico kit from Agilent. MiRNA-sequencing libraries of two biological replicates were prepared with the Illumina TruSeq Small RNA Sample Prep as described by the manufacturer. cDNA libraries were size fractionated on a 6% Tris-Borate-EDTA (TBE) gel and fragments of 140-160 nt were extracted. Libraries were loaded and sequenced on an Illumina HiSeq 2500 using protocols defined by the manufacturer.
Whole-transcriptome RNA sequencing and analysis. NMuMG/E9 cells were transfected with pre-miR miRNA precursors (Ambion) as described before prior to 4 days TGFβ treatment or not. Total RNA was isolated from cells of two independent experiments using the RNeasy Mini Kit (Qiagen) according to the manufacturer's instruction. RNA quality control was performed with a fragment analyser using the standard or high-sensitivity RNA analysis kit (DNF-471-0500 or DNF-472-0500) from Labgene and RNA concentration was measured by using the Quanti-iT TM RiboGreen RNA assay Kit (Life Technologies/Thermo Fisher Scientific). A total of 200 ng of RNA was utilized for library preparation with the TruSeq stranded total RNA LT sample prep Kit (Illumina). Poly-A + RNA was sequenced with HiSeq SBS Kit v4 (Illumina) on an Illumina HiSeq 2500 using protocols defined by the manufacturer.
Obtained single-end RNA-seq reads (51-mers) were mapped to the mouse genome assembly, version mm10, with default RNA-STAR 48 , parameters except for allowing only unique hits to genome (outFilterMultimapNmax = 1) and filtering reads without evidence in spliced junction table (outFilterType = "BySJout").
Using RefSeq mRNA coordinates from UCSC (http://www.genome.ucsc.edu, downloaded in December 2015) and the qCount function from QuasR package (version 3.12.1), we quantified gene expression as the number of reads that started within any annotated exon of a gene. The differentially expressed genes were identified using the edgeR package (version 1.10.1). Genes with FDR <0.05 and minimum log2 FC of ±1 were used for downstream analysis.
Tumour transplantation. A total of 0.5 × 10 6 4T1 cells stably expressing ZsGreenmmu-miR-1199/-200b/-429 or ZsGreen only were injected into the mammary fat pad of 7-9-week-old female NMRI nude or NSG mice. Mice were killed after 18 or 21 days post injection, respectively, and lungs were isolated. Tissue processing and haematoxylin and eosin staining were performed as previously described 24 . Lung sections were imaged using an Axio Imager scanning microscope (Zeiss), and the number of lung metastasis as well as the area per lung nodule were quantified. Mann-Whitney U test was used for statistical analysis. All studies involving mice were approved by the Swiss Federal Veterinary Office (SFVO) and the regulations of the Cantonal Veterinary Office of Basel Stadt (licence 1907).
Isolation of circulating tumour cells. Eighty microlitres of blood was drawn by heart puncture of tumour-bearing mice killed 21 days post orthotopic cell injection, mixed with normal growth medium and cultured in a six-well plate for 5 days. Adherent cells were washed with PBS daily. A MTT (3-(4,5-dimethylthiazol-2-yl)-2-5-diphenyltetrazolium bromide) staining was used to visualize tumour cell colonies. In brief, cells were incubated with 0.4 mg/ml MTT solution diluted in normal growth medium for 3 h. Afterwards, cell colonies were imaged and quantified.
Data availability. The miRNA expression data of a TGFβ time course experiment in NMuMG/E9 cells as well as RNA expression data of NMuMG/E9 cells ectopically transfected with miR-Ctr, miR-1199-5p, miR-200b-3p or miR-429-3p are deposited at Gene Expression Omnibus (GEO, accession number: GSE86026). All other remaining data are available within the article and supplementary files, or available from the authors upon request.