Abstract

   Enhancer RNAs (eRNAs) are a pivotal class of enhancer-derived
   non-coding RNAs that drive gene expression. Here we identify the SNAI1
   enhancer RNA (SNAI1e; SCREEM2) as a key activator of SNAI1 expression
   and a potent enforcer of transforming growth factor-β (TGF-β)/SMAD
   signaling in cancer cells. SNAI1e depletion impairs TGF-β-induced
   epithelial-mesenchymal transition (EMT), migration, in vivo
   extravasation, stemness, and chemotherapy resistance in breast cancer
   cells. SNAI1e functions as an eRNA to cis-regulate SNAI1 enhancer
   activity by binding to and strengthening the enrichment of the
   transcriptional co-activator bromodomain containing protein 4 (BRD4) at
   the local enhancer. SNAI1e selectively promotes the expression of
   SNAI1, which encodes the EMT transcription factor SNAI1. Furthermore,
   we reveal that SNAI1 interacts with and anchors the inhibitory SMAD7 in
   the nucleus, and thereby prevents TGF-β type I receptor (TβRI)
   polyubiquitination and proteasomal degradation. Our findings establish
   SNAI1e as a critical driver of SNAI1 expression and TGF-β-induced cell
   plasticity.

   Subject terms: Epithelial-mesenchymal transition, Long non-coding RNAs,
   Cancer epigenetics
     __________________________________________________________________

   Enhancer RNAs (eRNAs) regulate gene expression in cancer. Here, the
   authors identify the eRNA SNAI1e as a key activator of SNAI1
   expression, promoting TGF-β-induced EMT, migration, and chemotherapy
   resistance in breast cancer by enhancing BRD4 enrichment at the SNAI1
   enhancer.

Introduction

   Epithelial-to-mesenchymal transition (EMT) is a biological process in
   which epithelial cells lose their cell junctions and apicobasal
   polarity, undergo cytoskeletal reorganization, secrete extracellular
   matrix (ECM) components, and consequently transdifferentiate into
   migratory mesenchymal cells^[44]1. EMT is essential for proper
   embryonic development, as well as for tissue remodeling and wound
   healing. The abnormal reactivation of EMT plays a role in the
   progression of various human diseases, especially those linked to
   tissue fibrosis, cancer cell invasion, and metastasis^[45]1–[46]3. EMT
   is mainly induced by the EMT transcription factors (EMT-TFs), including
   SNAI1, SLUG, TWIST1 and ZEB1/2, which repress the promoter activity of
   genes encoding epithelial markers, such as E-cadherin, and induce the
   expression of mesenchymal markers, including N-cadherin, Vimentin and
   Fibronectin^[47]1,[48]3. The cytokine transforming growth factor-β
   (TGF-β) is a main driver of EMT in cancer^[49]4–[50]7.

   TGF-β signaling is initiated upon TGF-β ligand binding to the cell
   surface complexes of type I and type II serine/threonine kinase
   receptors (TβRI and TβRII, respectively). TβRI recruits and
   phosphorylates regulated (R)-SMAD2/3, which form heteromeric complexes
   with SMAD4 that accumulate in the nucleus. Together with other
   transcription factors, the SMAD2/3/4 complex contacts to DNA and
   stimulates the transcription of target genes such as PAI-1, CTGF and
   EMT-TFs, including SNAI1^[51]8,[52]9. TGF‐β signaling is fine-tuned at
   multiple levels including the receptor level^[53]10,[54]11. By
   recruiting the SMAD‐Specific E3 Ubiquitin Protein Ligase 1/2
   (SMURF1/2), the inhibitory protein SMAD7 binds and targets activated
   TβRI for polyubiquitination and degradation^[55]12–[56]14.

   Long non-coding RNAs (lncRNAs) are transcripts exceeding 200
   nucleotides in length and lack protein-coding potential^[57]15. LncRNAs
   can function as guides, scaffolds, and decoys to influence interactions
   between biological macromolecules, such as protein-protein and
   protein-DNA interactions, and thereby modulate gene expression at
   various levels^[58]16,[59]17. In particular, enhancer RNAs (eRNAs), a
   subclass of nuclear lncRNAs transcribed from active enhancer loci, play
   a critical role in facilitating local enhancer activities and
   enhancer-promoter looping to influence gene
   transcription^[60]18,[61]19. For example, SCREEM (SNAI1 cis-regulatory
   eRNA expressed in monocytes) eRNAs, produced from a SNAI1 enhancer
   region, enhance SNAI1 expression in a cis-regulatory manner, thereby
   promoting EMT in human primary bronchial epithelial cells^[62]20.
   Although several studies have implicated lncRNAs in the modulation of
   TGF‐β signaling^[63]21–[64]23, the interplay between eRNAs and TGF‐β
   signaling in EMT is not well-characterized.

   In this work, we perform a CRISPR activation (CRISPRa)^[65]24-based
   screen in breast cancer cells and identify TGF‐β-induced SNAI1 enhancer
   RNA (SNAI1e; SCREEM2) that strongly promotes TGF-β/SMAD signaling by
   attenuating TβRI degradation. Mechanistically, SNAI1e binds
   bromodomain containing protein 4 (BRD4) to enforce the local enhancer
   activity and stimulate SNAI1 transcription. SNAI1 interacts with and
   retains SMAD7 in the nucleus, thereby antagonizing TβRI
   polyubiquitination and degradation. We characterize SNAI1e as an eRNA
   that promotes TGF‐β signaling and EMT-associated hallmarks, including
   increased motility, invasion, stemness, and therapy resistance in
   cancer cells.

Results

TGF-β-induced SNAI1e is an enforcer of TGF-β signaling

   We sought to identify key lncRNAs that potentiate TGF-β signaling in
   breast cancer cells. Toward this end, we conducted a functional screen
   on TGF-β-induced lncRNAs, as TGF-β-induced gene products frequently act
   as negative or positive feedback regulators of TGF-β
   signaling^[66]23,[67]25. Reanalysis of our previous transcriptome data
   enriched 107 lncRNAs, whose expression was induced upon TGF-β
   stimulation at early (2 h), intermediate (8 h) and late (24 h) time
   points in MDA-MB-231 triple-negative breast cancer (TNBC) cells^[68]21
   (Supplementary Fig. [69]1a, Supplementary Table [70]1). We performed a
   CRISPRa-based gain-of-function screen in MDA-MB-231 cells that stably
   express a selective synthetic SMAD3/4‐driven transcriptional reporter,
   CAGA[12]-dynGFP^[71]26, to monitor the TGF-β/SMAD-induced
   transcriptional output (Fig. [72]1a). For each lncRNA hit, we designed
   5 gRNAs, which fell within 500 bp upstream and 200 bp downstream of the
   transcripts’ annotated transcription start sites (TSSs)^[73]27,[74]28.
   MDA-MB-231 cells were transduced with a gRNA library targeting 107
   lncRNAs at a low multiplicity of infection (MOI) to obtain only one
   gRNA per cell. Upon TGF-β treatment for 48 h, cells were sorted into
   two groups (GFP-high and -low; Fig. [75]1a and “Methods” section) by
   fluorescence-activated cell sorting (FACS). Screen hits were ranked
   based on the comparison of gRNA distributions between the GFP-high and
   GFP-low populations in two biological replicates (Fig. [76]1b,
   Supplementary Data [77]1). gRNAs targeting positive control genes, i.e.
   two CAGA minimal promoter (MLP) regions, TBRII and RNF11 (encoding an
   E3 ubiquitin ligase that targets inhibitory SMAD7), potentiated
   TGF-β/SMAD signaling, which validated the technical robustness
   (Fig. [78]1b). Of note, multiple lncRNAs enforced the TGF-β-induced
   transcriptional response with similar potency as the positive controls
   (Fig. [79]1b). We focused on one of the top hits, SCREEM2^[80]20, and
   renamed it as SNAI1 enhancer RNA (SNAI1e) due to its proximity to SNAI1
   in the human genome (Supplementary Fig. [81]1b) and its regulatory role
   in SNAI1 gene expression (as detailed below).

Fig. 1. TGF-β-induced SNAI1e is an enforcer of TGF-β signaling.

   [82]Fig. 1
   [83]Open in a new tab

   a Schematic overview of CRISPRa-mediated lncRNA screen. b Diagram of
   lncRNA screening results. c RT-qPCR analysis of SNAI1e expression in
   MDA‐MB‐231 cells upon TGF-β stimulation. The results are presented as
   mean ± SD from three biological replicates. d RT-qPCR analysis of
   SNAI1e expression in MDA‐MB‐231 cells upon shRNA-mediated SMAD4
   knockdown. Co.sh, empty vector for shRNA expression. The results are
   expressed as the mean ± SD from three biological replicates. e
   Luciferase reporter assay to determine the effects of SMAD3 and caTβRI
   on SNAI1e promoter 1 (P1) activity. The data are plotted as mean ± SD
   from three biological replicates. Co.vec, empty vector control. f
   ChIP-qPCR analysis of the SNAI1e transcription start site (TSS) in
   MDA-MB-231 cells. The results are expressed as mean ± SD from three
   biological replicates. g Luciferase reporter assay to determine the
   effects of TGF-β on SNAI1e promoter 7 (P7) activity. The data are
   presented as mean ± SD from three biological replicates. h ChIP-qPCR
   analysis of the SNAI1e promoter region (-2095 - -1987) in MDA-MB-231
   cells. The results are expressed as mean ± SD from three biological
   replicates. i Coding probability prediction of SNAI1e with the CPAT
   software. Protein‐coding mRNAs (GAPDH and ACTB2) and well‐annotated
   lncRNAs (XIST and NKILA) serve as positive and negative controls,
   respectively. j Subcellular localization analysis of SNAI1e in
   MDA‐MB‐231 cells by RT-qPCR. NEAT1 serves as positive control for the
   nuclear fraction, whereas H19 and GAPDH serve as positive controls for
   the cytoplasmic fraction. The data are presented as mean ± SEM from
   three biological replicates. k RNA fluorescence in situ hybridization
   was performed to evaluate SNAI1e subcellular localization in MDA-MB-231
   cells. Representative images from two independent experiments are
   shown. Scale bar = 10 μm. The results are quantified as a box plot with
   min to max Whiskers from 19 (−TGF-β) and 9 (+TGF-β) technical
   replicates (cells) and the fold change is shown. The boundaries of the
   box indicate the 25th percentile and the 75th percentile, and the
   center indicates the median. Significance was calculated using
   two-tailed unpaired Student’s t-test (e), and one-way ANOVA followed by
   Dunnett’s (c) and Tukey’s (d, f, g, h) multiple comparisons test. gEV
   gRNA expression vector. Co.vec empty control vector. Ab antibody.

   The 5′ and 3′ rapid amplification of cDNA ends (RACE) assays
   demonstrated that SNAI1e is a 4055 nt one‐exon transcript nearly
   identical to the annotated lncRNA lnc-SNAI1-5:2 in the LNCipedia
   database^[84]29, or NONHSAG110604.1 in the NONCODE database^[85]30,
   differing only by the absence of a single adenine at the 5′ end
   (Supplementary Fig. [86]1c, d and Supplementary Table [87]2). Assessing
   the kinetic SNAI1e expression pattern upon TGF-β stimulation by RT-qPCR
   showed that TGF-β induced an early increase of SNAI1e expression, and
   that the induction was sustained until 72 h in MDA-MB-231 cells
   (Fig. [88]1c). Similar results were obtained in MCF10A-M2 pre-malignant
   breast and A549 non-small cell lung adenocarcinoma cells (Supplementary
   Fig. [89]1e, f). To verify SNAI1e as a target gene of TGF-β/SMAD
   signaling, we depleted SMAD4 with a short hairpin (sh)RNA and found a
   dramatic suppression of both basal and TGF-β-induced SNAI1e expression
   in MDA-MB-231 cells (Fig. [90]1d, Supplementary Fig. [91]1g). Next, we
   characterized the SNAI1e promoter by cloning a 2640 bp DNA fragment
   (Promoter 1 (P1); −2472 to +168; chromosome 20: 50,027,744-50,030,383,
   GRCh38.p14) upstream the SNAI1e TSS and placed it at the 5′ end of a
   luciferase gene (Supplementary Fig. [92]1h). TGF-β treatment, along
   with the ectopic expression of SMAD3 or constitutively active (ca)TβRI,
   enhanced the reporter activity of SNAI1e P1 in HepG2 cells
   (Fig. [93]1e). Chromatin immunoprecipitation (ChIP)-qPCR demonstrated
   that RNA polymerase (Pol) II bound to the SNAI1e TSS (−99 - +53;
   chromosome 20: 50,030,117- 50,030,268,GRCh38.p14) and the binding was
   facilitated by TGF-β stimulation in MDA-MB-231 cells (Fig. [94]1f).
   Analysis of the promoter truncation mutants revealed a minimal DNA
   fragment (P7; −2072 to -2026; chromosome 20: 50,028,143 to 50,028,190
   (GRCh38.p14)) that was required for TGF-β/SMAD-triggered SNAI1e
   promoter activity (Fig. [95]1g, Supplementary Fig. [96]1h and i).
   Mutation of the putative SMAD3 binding element abrogated both basal and
   TGF-β-induced SNAI1e promoter activity, suggesting that activated SMAD3
   directly contacts the SNAI1e promotor (Fig. [97]1g). ChIP-qPCR
   consolidated the direct binding of SMAD3 to the SNAI1e promoter, which
   was increased upon TGF-β treatment in MDA-MB-231 cells (Fig. [98]1h).
   Bioinformatic analysis with Coding Potential Assessment Tool
   (CPAT)^[99]31 predicted a lack of coding potential of SNAI1e
   (Fig. [100]1i). Subcellular fractionation followed by RT-qPCR was
   performed to explore SNAI1e localization in MDA-MB-231 cells, as this
   information may aid in elucidating its mechanism of action. SNAI1e was
   mainly localized in the nucleus (Fig. [101]1j), which was further
   validated by fluorescence in situ hybridization (FISH) in MDA-MB-231
   cells (Fig. [102]1k). TGF-β stimulation did not change SNAI1e
   cytoplasmic and nuclear distribution (Fig. [103]1k). Taken together,
   these results demonstrate that SNAI1e is a TGF-β/SMAD-induced nuclear
   lncRNA that promotes TGF-β/SMAD signaling.

SNAI1e promotes TGF-β-induced EMT and cancer cell migration

   To evaluate the correlation between SNAI1e expression and breast cancer
   progression, we analyzed several TCGA datasets^[104]32 and observed
   that SNAI1e expression was significantly higher in tumor samples from
   breast cancer patients in comparison to that of normal breast tissues
   (Fig. [105]2a). In line with this result, SNAI1e was more highly
   expressed in TNBC cell lines than in MCF10A-M1 normal breast cells,
   MCF10A-M2 pre-malignant breast cells, and the less aggressive MCF7
   luminal breast cancer cells (Fig. [106]2b). To explore the effect of
   SNAI1e on EMT, we activated endogenous SNAI1e expression using two
   individual CRISPRa guide RNAs (gRNAs) in epithelial MCF10A-M2 and A549
   cells (Supplementary Fig. [107]2a, b). CRISPRa-guided increased
   expression of SNAI1e potentiated TGF-β-induced downregulation of the
   expression of epithelial marker E-cadherin, and upregulation of the
   expression of mesenchymal markers, including N-cadherin, Fibronectin,
   Vimentin and SNAI1 (Fig. [108]2c, d). On the contrary, GapmeR-mediated
   SNAI1e knockdown mitigated the changes in EMT marker expression
   triggered by TGF-β in MCF10A-M2 cells (Fig. [109]2e and Supplementary
   Fig. [110]2c). In agreement with these results, SNAI1e overexpression
   augmented, but GapmeR-mediated SNAI1e knockdown attenuated,
   TGF-β-induced filamentous actin (F-actin) stress fiber formation in
   A549 cells (Fig. [111]2f, g, Supplementary Fig. [112]2d). Additionally,
   SNAI1e overexpression promoted, whereas SNAI1e depletion inhibited,
   TGF-β-induced cell migration in MDA-MB-231 cells (Fig. [113]2h, i,
   Supplementary Fig. [114]2e, f). To validate our in vitro findings, an
   in vivo zebrafish embryo breast cancer xenograft model^[115]33 was
   applied, where SNAI1e promoted breast cancer cell extravasation
   (Fig. [116]2j, k). Moreover, to consolidate the results from
   GapmeR-directed loss-of-function experiments, an orthogonal approach
   CRISPR interference (CRISPRi)^[117]34 was used to diminish SNAI1e
   expression (Supplementary Fig. [118]2g–m). Since EMT is correlated with
   the gain of stemness and chemotherapy resistance in cancer
   cells^[119]7,[120]35,[121]36, we assessed the mammosphere formation
   abilities of MCF10A-M2 cells upon SNAI1e overexpression. As expected,
   SNAI1e enhanced mammosphere formation in MCF10A-M2 cells
   (Fig. [122]2l), which was accompanied by an increase of CD44^+CD24^−
   breast cancer stem cell population^[123]37 (Supplementary
   Fig. [124]2n). Consistent with these observations, SNAI1e
   overexpression conferred greater resistance to the chemotherapeutic
   drugs Doxorubicin (Doxo) and Paclitaxel (PTX) in MCF10A-M2 cells
   (Fig. [125]2m, n). Collectively, our results reveal that TGF-β-induced
   SNAI1e strongly promotes TGF-β-induced EMT, migration, stemness and/or
   drug resistance in breast and lung cancer cells (Supplementary
   Fig. [126]2o).

Fig. 2. SNAI1e promotes TGF-β-induced EMT and migration.

   [127]Fig. 2
   [128]Open in a new tab

   a Comparison of SNAI1e expression between normal and tumor breast
   samples from the TCGA dataset. The results are expressed as mean ± SEM
   from 101 normal and 1002 tumor breast samples, respectively. b RT-qPCR
   analysis of SNAI1e expression in multiple cell lines. The data are
   plotted as mean ± SD from three biological replicates. c–e Effect of
   SNAI1e on TGF-β-induced EMT marker expression in MCF10A-M2 (c, e) or
   A549 (d) cells upon CRISPRa-mediated SNAI1e overexpression (c, d) or
   GapmeR‐mediated SNAI1e knockdown (e). f, g Immunofluorescence analysis
   of F‐actin expression in A549 cells upon CRISPRa-mediated SNAI1e
   overexpression (f) or GapmeR‐mediated SNAI1e knockdown (g). Scale
   bar = 36.8 μm. h, i Transwell assay to evaluate the effect of SNAI1e on
   TGF-β-induced migration in MDA-MB-231 cells. SNAI1e overexpression and
   knockdown were achieved by CRISPRa (h) and GapmeR (i), respectively.
   The results are expressed as mean ± SD from eight (h) and four (i)
   biological replicates, respectively. j, k In vivo zebrafish xenograft
   experiments with MDA-MB-231 cells upon CRISPRa-mediated SNAI1e
   overexpression (j) or GapmeR‐mediated SNAI1e knockdown (k).
   Extravasated breast cancer cell clusters are indicated with yellow
   arrows. Whole zebrafish image, bar = 618.8 μm; zoomed image,
   bar = 154.7 μm. The results are expressed as mean from 27 (gEV), 25
   (g1), and 23 (g2) biological replicates in (j), and from 20 (Scramble
   GapmeR) and 22 (SNAI1e GapmeR) biological replicates in (k),
   respectively. l Effect of CRISPRa-mediated SNAI1e overexpression on
   mammosphere formation in MCF10A-M2 cells. The numbers of mammospheres
   are presented as mean ± SD from eight biological replicates. Scale
   bar = 100 μm. m, n Dose-response curves for doxorubicin (Doxo; m) or
   paclitaxel (PTX; n) in MCF10A-M2 cells upon CRISPRa-mediated SNAI1e
   overexpression The results are expressed as mean ± SD from three
   biological replicates. Significance was assessed using two-tailed
   unpaired Student’s t-test (a, k), two-way ANOVA followed by Dunnett’s
   (h, i) and Tukey’s (m, n) multiple comparisons test, and one-way ANOVA
   followed by Dunnett’s multiple comparisons test (j, l). Data are
   representative of two (f, g) and at least three (c–e) independent
   experiments with similar results. gEV gRNA expression vector. Scr
   scramble.

SNAI1e inhibits TβRI polyubiquitination and degradation

   Next, we validated the effects of SNAI1e on promoting TGF-β signaling.
   The dynamic activity of the CAGA[12]-dynGFP reporter revealed that
   CRISPRa-mediated SNAI1e overexpression facilitated the TGF-β-induced
   transcriptional response in MDA-MB-231 cells (Fig. [129]3a). In
   contrast, SNAI1e knockdown by GapmeR, CRISPR/Cas13d^[130]38 or CRISPRi
   reduced the TGF-β-induced reporter activity (Fig. [131]3b,
   Supplementary Fig. [132]3a–c). Consistently, transcriptome analysis
   enriched TGF-β target genes, including IL11, SNAI1, PMEPA1 and ANGPTL4,
   as top genes upregulated by SNAI1e in MDA-MB-231 cells (Fig. [133]3c).
   We conducted pathway enrichment analysis on the 84 SNAI1e-upregulated
   genes and identified TGF-β signaling and EMT as the most significantly
   impacted cellular processes by SNAI1e overexpression (Fig. [134]3d,
   Supplementary Fig. [135]3d). Moreover, gene set enrichment analysis
   (GSEA) confirmed a strong positive correlation between manipulated
   SNAI1e expression and the TGF-β response gene signature or the EMT gene
   signature (Fig. [136]3e, Supplementary Fig. [137]3e). However, no
   correlation was found between SNAI1e expression and the response gene
   signatures of BMP, WNT, and YAP (Supplementary Fig. [138]3f–h). In
   addition, SNAI1e overexpression promoted the expression of TGF-β target
   genes (i.e., PAI-1, CTGF, and SMAD7), and SNAI1e depletion showed the
   opposite effect in MDA-MB-231 cells (Supplementary Fig. [139]3i–k).
   Interestingly, SNAI1e overexpression facilitated TGF-β-induced SMAD2
   phosphorylation (p-SMAD2) in MDA-MB-231 cells (Fig. [140]3f), while
   this effect was suppressed upon SNAI1e depletion (Fig. [141]3g,
   Supplementary Fig. [142]3l). This observation prompted us to
   investigate the impact of SNAI1e on TGF-β receptor expression, as
   p-SMAD2 is a direct indicator of TGF-β receptor activity. SNAI1e
   overexpression elevated TβRI, but not TβRII, protein expression in
   MDA-MB-231 cells (Fig. [143]3h), whereas CRISPRi-mediated SNAI1e
   knockdown suppressed TβRI protein expression (Supplementary
   Fig. [144]3m). However, TBRI and TBRII mRNA levels were not altered by
   SNAI1e in MDA-MB-231 cells (Supplementary Fig. [145]3n), suggesting
   that SNAI1e may modulate TβRI protein stability. Consistent with this
   notion, cycloheximide (CHX)-directed time-course experiments revealed
   that SNAI1e protected TβRI protein from degradation in MDA-MB-231 cells
   (Fig. [146]3i). To determine whether TβRI degradation involves the
   proteasome or lysosome, MDA-MB-231 cells with SNAI1e knockdown were
   challenged with proteasomal or lysosomal inhibitors. We found that the
   proteasomal inhibitor MG132, but not the lysosomal inhibitor
   Bafilomycin (BafA1), restored TβRI expression in SNAI1e depletion cells
   (Fig. [147]3j). In agreement with this result, SNAI1e elicited an
   inhibitory effect on TβRI polyubiquitination in MDA-MB-231 cells
   (Fig. [148]3k and Supplementary Fig. [149]3o). Blocking TβRI kinase
   activity with the selective small molecule inhibitor SB431542
   (SB)^[150]39 alleviated the changes of EMT marker expression and cell
   migration by CRISPRa-directed SNAI1e overexpression (Fig. [151]3l, m).
   On the contrary, TβRI ectopic expression counteracted the suppressive
   effect of SNAI1e knockdown on migration in MDA-MB-231 cells
   (Fig. [152]3n, o), suggesting that TβRI is a crucial target of SNAI1e
   in promoting EMT and migration.

Fig. 3. SNAI1e inhibits TβRI polyubiquitination and degradation.

   [153]Fig. 3
   [154]Open in a new tab

   a, b Effect of SNAI1e overexpression (a) and knockdown (b) on
   CAGA[12]-dynGFP reporter in MDA-MB-231 cells. The data are presented as
   mean ± SD from four (a) and twelve (b) biological replicates. c Volcano
   plots showing the differentially expressed genes upon SNAI1e
   overexpression. d Plot exhibiting the top 15 cellular processes
   affected by SNAI1e. e GSEA of correlations between SNAI1e and the TGF-β
   response gene signature. f, g Effect of SNAI1e overexpression (f) and
   knockdown (g) on TGF-β-induced p-SMAD2. The relative abundance of
   p-SMAD2 is shown as mean ± SD from five (f) and three (g) independent
   experiments. h Effect of SNAI1e overexpression on TβRI and TβRII
   protein expression. The relative abundance of TβRI and TβRII is shown
   as mean ± SD from three independent experiments. i Western blotting
   analysis of TβRI protein stability upon SNAI1e overexpression.
   Quantitative data show the relative TβRI abundance as mean ± SEM from
   four independent experiments. j Western blotting analysis of TβRI
   expression upon SNAI1e knockdown with the lysosome or proteasome
   inhibitors. Quantitative data show the relative TβRI abundance as
   mean ± SD from three independent experiments. k Effect of SNAI1e
   overexpression on TβRI polyubiquitination. Relative ubiquitination
   abundance is shown as mean ± SD from three independent experiments. l
   Effect of SB431542 (SB) on TGF-β-induced EMT marker expression in
   MCF10A-M2 cells upon SNAI1e overexpression. m Effect of SB431542 (SB)
   on MDA-MB-231 cell migration upon SNAI1e overexpression. The results
   are expressed as mean ± SEM from six biological replicates. n Western
   blotting results showing TβRI ectopic expression upon SNAI1e knockdown.
   o Effect of TβRI ectopic expression on MDA-MB-231 cell migration upon
   SNAI1e knockdown. The results are expressed as mean ± SEM from six
   biological replicates. Significance was assessed using non-parametric
   permutation test (e), two-tailed paired Student’s t-test (g), two-way
   ANOVA followed by Tukey’s (a, m, o) and Dunnett’s (b, i) multiple
   comparisons test, and one-way ANOVA followed by Tukey’s (j) and
   Dunnett’s (f, h, k) multiple comparisons test. Data are representative
   of at least three (l, n) independent experiments with similar results.
   gEV gRNA expression vector. Scr scramble. Co.vec empty control
   vector. a.u. arbitrary units.

SNAI1e selectively induces SNAI1 expression

   We sought to decipher the mechanism by which SNAI1e potentiates TβRI
   stability and TGF-β/SMAD signaling. SNAI1 has been implicated in
   reinforcing TGF-β/SMAD signaling in breast cancer
   cells^[155]40,[156]41. Considering that SNAI1e is localized in the
   nucleus (Fig. [157]1k), and the close proximity between SNAI1e and
   SNAI1 in the human genome (Supplementary Fig. [158]1b), we hypothesized
   that SNAI1e directly targets SNAI1 for transcriptional activation to
   promote TGF-β/SMAD signaling. Both basal and TGF-β-induced SNAI1 mRNA
   and protein levels were dramatically increased upon SNAI1e
   overexpression in MDA-MB-231 cells (Fig. [159]4a, b), supporting the
   RNA-seq results mentioned above (Fig. [160]3c). In contrast, SNAI1e
   depletion by multiple approaches (i.e. GapmeR, siRNA, Cad13d, and
   CRISPRi) greatly reduced SNAI1 expression in MDA-MB-231 cells
   (Fig. [161]4c, d and Supplementary Fig. [162]4a–d). Importantly, the
   expression of the other neighboring genes was unaffected upon
   CRISPRa-mediated SNAI1e overexpression in MDA-MB-231 cells
   (Supplementary Figs. [163]1b, [164]4e).

Fig. 4. SNAI1e induces SNAI1 expression.

   [165]Fig. 4
   [166]Open in a new tab

   a RT-qPCR analysis of SNAI1 expression in MDA‐MB‐231 cells upon SNAI1e
   overexpression. The results are expressed as the mean ± SD from three
   biological replicates. b Effect of SNAI1e overexpression on SNAI1
   protein expression. c RT-qPCR analysis of SNAI1 expression upon SNAI1e
   knockdown. The data are presented as mean ± SD from three biological
   replicates. d Effect of SNAI1e knockdown on SNAI1 protein expression.
   Quantitative data show the relative abundance of SNAI1 as mean ± SD
   from three independent experiments. e GSEA of correlations between
   SNAI1e expression and SNAI1-induced gene signature. NES, normalized
   enrichment score. f Effect of SNAI1 ectopic expression on TGF-β-induced
   p-SMAD2 in MDA‐MB‐231 cells. Quantitative data show the abundance of
   p-SMAD2 as mean ± SD from three independent experiments. g Effect of
   SNAI1 ectopic expression on TβRI polyubiquitination in MDA-MB-231 cells
   expressing HA-ubiquitin (HA-Ub). Relative ubiquitination is shown as
   mean ± SEM from three independent experiments. h Effect of SNAI1e
   overexpression and shRNA-mediated SNAI1 knockdown on TGF-β-induced
   p-SMAD2. Quantitative data show the abundance of p-SMAD2 as mean ± SD
   from three independent experiments. i Interactions between SNAI1 and
   SMAD4 or SMAD7 were analyzed by co-immunoprecipitation assays in
   HEK293T cells. j The endogenous SNAI1-SMAD7 interaction as evaluated by
   PLA. The red and blue dots indicate the SNAI1‐SMAD7 interaction and the
   staining of nuclei by DAPI, respectively. Scale bar = 20 μm (left
   image) and 40 μm (right image). k Immunofluorescence analysis of SMAD7
   localization upon SNAI1 ectopic expression. Scale bar = 58 μm. l
   Western blotting analysis of SMAD7 localization upon SNAI1 ectopic
   expression. The levels of the cytoplasmic marker GAPDH and the nuclear
   marker BRD4 serve as positive controls. Relative SMAD7 abundance is
   shown as mean ± SD from three independent experiments. m Schematic
   working model. SNAI1e-induced SNAI1 interacts with and potentiates
   SMAD7 nuclear retention, resulting in the decrease of TβRI
   polyubiquitination and proteasomal degradation. Significance was
   assessed using two-way ANOVA followed by Šidák’s multiple comparisons
   test (a), one-way ANOVA followed by Tukey’s multiple comparisons test
   (c, d, h), non-parametric permutation test (e), and two-tailed paired
   Student’s t-test (f, g, l). Data are representative of at least three
   (b, i–k) independent experiments with similar results. gEV gRNA
   expression vector. Scr scramble. a.u. arbitrary units. Nuc nucleus.
   Cyto cytoplasm.

   GSEA showed a positive correlation between SNAI1e expression and
   SNAI1-induced gene signature, indicating that SNAI1 is a key downstream
   effector of SNAI1e (Fig. [167]4e). To validate the effect of SNAI1 on
   enhancing TGF-β/SMAD signaling, a Tet-ON inducible system was used for
   SNAI1 ectopic expression. Doxycycline (Dox) treatment showed a
   dose-dependent induction of SNAI1 expression in MDA-MB-231 cells
   (Supplementary Fig. [168]4f). We selected 0.1 µg/mL of Dox to stimulate
   SNAI1 expression in the following experiments (Supplementary
   Fig. [169]4f), as this concentration induced ectopic SNAI1 expression
   at levels comparable to those induced by SNAI1e in MDA-MB-231 cells
   (Fig. [170]4a). As expected, ectopic SNAI1 expression promoted
   the TGF-β-induced transcriptional response and p-SMAD2, as well as TβRI
   protein expression in MDA-MB-231 cells (Fig. [171]4f, Supplementary
   Fig. [172]4g, h). Moreover, TβRI polyubiquitination was suppressed upon
   SNAI1 overexpression in MDA-MB-231 cells (Fig. [173]4g). To further
   evaluate whether SNAI1 is essential for SNAI1e to promote TGF-β/SMAD
   signaling, we used an shRNA to deplete SNAI1 (Fig. [174]4h). We found
   that SNAI1 knockdown diminished the increase of TGF-β-induced p-SMAD2
   and downstream reporter activity upon SNAI1e overexpression in
   MDA-MB-231 cells (Fig. [175]4h and Supplementary Fig. [176]4i).

   SNAI1 has been reported to enhance TβRI expression in MDA-MB-231
   cells^[177]40, but the underlying mechanism remains unclear. Since
   SMAD7 and SNAI1 are both localized in the nucleus, we assumed that
   SNAI1 may bind SMAD7. Indeed, we observed a SMAD7-SNAI1 interaction
   that was even stronger than the SMAD4-SNAI1 interaction, which served
   as a positive control^[178]42, in HEK293T cells (Fig. [179]4i).
   Proximity ligation assays (PLA) confirmed the endogenous SNAI1-SMAD7
   interaction in the nucleus of MDA-MB-231 cells (Fig. [180]4j). Next, we
   tested the effect of SNAI1 overexpression on SMAD7 localization. SNAI1
   retained SMAD7 protein in the nucleus of MDA-MB-231 cells as shown by
   immunofluorescence staining (Fig. [181]4k) and subcellular
   fractionation followed by western blotting (Fig. [182]4l). Taken
   together, our data demonstrate that SNAI1e induces SNAI1 expression at
   the transcriptional level, and SNAI1 protein binds and sequesters SMAD7
   protein in the nucleus to decrease TβRI polyubiquitination and
   proteasomal degradation (Fig. [183]4m).

SNAI1e functions as an eRNA to facilitate SNAI1 transcription

   Given the proximity of SNAI1e gene body to SNAI1 in the human genome
   (Supplementary Fig. [184]1b), we hypothesized that SNAI1e transcribes
   from an enhancer region and functions as an eRNA^[185]18,[186]19 to
   activate SNAI1 transcription. Firstly, we performed Circular Chromosome
   Conformation Capture (4C)-seq^[187]43 to analyze the chromatin
   interactions between SNAI1e gene body, which was used as a bait
   (viewpoint; VP), and its surrounding DNA regions in MDA-MB-231 cells
   (Fig. [188]5a). We observed that the highest peak, representing the
   most potent SNAI1e interaction site, was in the genomic region of SNAI1
   (Fig. [189]5a). Interestingly, TGF-β stimulation facilitated the
   SNAI1e-SNAI1 genomic interaction (Fig. [190]5a). Next, we quantified
   the abundance of two specific histone modifications associated with
   active enhancers (i.e. H3 lysine 27 acetylation (H3K27ac) and H3 lysine
   4 monomethylation (H3K4me1))^[191]44–[192]46 by performing ChIP-qPCR
   with multiple primer pairs targeting the SNAI1e gene body region
   (Fig. [193]5b). We found that these two histone markers were enriched
   in the SNAI1e gene body region, with their abundance further increased
   by TGF-β treatment in MDA-MB-231 cells (Fig. [194]5c). Next, we
   employed CRISPR/Cas9 with paired gRNAs to deplete the SNAI1e gene body
   in MDA-MB-231 cells (Supplementary Fig. [195]5a). As expected, SNAI1
   expression was reduced in two independent knockout (KO) clones compared
   to wild-type (WT) MDA-MB-231 cells (Supplementary Fig. [196]5b, c).
   These results suggest that the gene body region of SNAI1e functions as
   an enhancer that targets SNAI1 in the human genome.

Fig. 5. SNAI1e functions as an eRNA to facilitate SNAI1 transcription.

   [197]Fig. 5
   [198]Open in a new tab

   a Circular chromosome conformation capture (4C)-seq profile overlays
   displaying the in cis chromatin contacts of the SNAI1e gene body. The
   increased interactions between SNAI1e viewpoint (VP) and other regions
   in −TGF-β group or +TGF-β group are labeled in green or yellow,
   respectively. Common interactions are labeled in gray. Coverage
   represents an average of three biological replicates. y axis, 4C
   coverage per 1 million normalized reads. b Schematic of the amplified
   regions by the indicated primer pairs against the SNAI1e gene body. c
   ChIP-qPCR analysis of the H3K4me1 and H3K27ac enrichment within the
   SNAI1e gene body region in MDA-MB-231 cells. The data are shown as
   mean ± SD from three biological replicates. Cumulative analysis of the
   histone marker enrichment fold is shown as a box plot with min to max
   Whiskers from eight technical replicates (eight genomic regions). The
   boundaries of the box indicate the 25th percentile and the 75th
   percentile, and the center indicates the median. d ChIP-qPCR analysis
   of RNA Pol II and H3K4me3 enrichment at the SNAI1 TSS upon
   CRISPRa-mediated SNAI1e overexpression. RT-qPCR results are shown as
   mean ± SD from three biological replicates. e ChIP-qPCR analysis of RNA
   Pol II enrichment at the SNAI1 TSS upon GapmeR-mediated SNAI1e
   depletion. RT-qPCR results are shown as mean ± SD from three biological
   replicates. f RT-qPCR analysis of SNAI1 expression in MDA-MB-231 cells
   upon CRISPR-Display-mediated in cis overexpression of SNAI1e or
   SNAI1e-V1. The data are presented as the mean ± SD from three
   biological replicates. g Schematic working model. TGF-β-induced SNAI1e
   stimulates local enhancer activity, marked by increased H3K27ac and
   H3K4me1 levels, and thereby triggers SNAI1 transcription by promoting
   recruitment of RNA Pol II and H3K4me3 to SNAI1 TSS. Significance was
   assessed by using two-way ANOVA followed by Dunnett’s (c, d) Šídák’s
   (e) multiple comparisons test, and one-way ANOVA followed by Tukey’s
   multiple comparisons test (f). gEV gRNA expression vector. Co.vec
   control empty vector. Scr scramble. TSS transcription start site.

   To investigate the role of SNAI1e as an eRNA in targeting SNAI1 for
   transcriptional activation, we checked the recruitment of RNA Pol II
   and the trimethylation of histone H3 lysine 4 (H3K4me3), a marker of
   transcription initiation^[199]47,[200]48, at the SNAI1 TSS (−38 to +83,
   chromosome 20 49,982,942–49,983,062, GRCh38.p14). As expected,
   ChIP-qPCR results showed that CRISPRa-mediated SNAI1e overexpression
   promoted the enrichment of both RNA Pol II and H3K4me3 at the TSS of
   SNAI1 in MDA-MB-231 cells (Fig. [201]5d). To eliminate the potential
   confounding effect of CRISPRa on altering local enhancer activity, and
   to separately assess the role of SNAI1e from that of the enhancer
   itself, we depleted SNAI1e using GapmeR. SNAI1e knockdown inhibited RNA
   Pol II recruitment at the SNAI1 TSS (Fig. [202]5e). To further confirm
   this result, we employed the CRISPR-Display system^[203]49 to tether
   SNAI1e to the SNAI1e gene body. SNAI1e was directed to two genomic loci
   of the SNAI1e gene body using two independent gRNAs and the deactivated
   Cas9 (dCas9; Supplementary Fig. [204]6a, b). A short variant,
   SNAI1e-variant1 (SNAI1e-V1; lnc-SNAI1-5:1), derived from the same
   enhancer region, was included for analysis (Supplementary
   Fig. [205]6a). In this setup, in cis overexpression of SNAI1e (but not
   SNAI1e-V1) using two gRNAs promoted SNAI1 expression (Fig. [206]5f,
   Supplementary Fig. [207]6c). Consistent with its role as an eRNA
   functioning in cis, ectopic SNAI1e expression (uncoupled from gRNAs)
   did not affect SNAI1 expression (Fig. [208]5f, Supplementary
   Fig. [209]6c). Consistently, ectopic SNAI1e expression using a
   lentiviral vector did not affect SNAI1 expression or TGF-β/SMAD
   signaling (Supplementary Fig. [210]7a–e). Taken together, these results
   suggest that TGF-β-induced SNAI1e stimulates its local enhancer
   activity, marked by increased H3K27ac and H3K4me1 levels, and thereby
   triggers SNAI1 transcription by promoting the recruitment of RNA Pol II
   and H3K4me3 to the SNAI1 TSS (Fig. [211]5g).

SNAI1e directly binds to BRD4

   BRD4 interacts with acetylated histones and eRNAs at enhancer regions
   through its bromodomains (BDs) to stimulate local enhancer activities
   and facilitate transcriptional elongation^[212]50,[213]51. We
   speculated that SNAI1e may bind BRD4 to alter the local chromatin
   landscape for enhancer activation. RNA immunoprecipitation (RIP)-qPCR
   demonstrated that SNAI1e co-immunoprecipitated with BRD4, but not with
   the negative control nuclear protein YTH N6-methyladenosine RNA-binding
   protein C1 (YTHDC1), from MDA-MB-231 cell lysates (Fig. [214]6a).
   Consistently, biotinylated SNAI1e, but not its antisense counterpart
   (SNAI1e-AS) or the negative control nuclear lncRNA LETS1^[215]22,
   pulled down ectopically expressed BRD4 protein in HEK293T cells
   (Fig. [216]6b). To map the binding region of BRD4 on SNAI1e, we divided
   SNAI1e into 4 fragments (T1–T4), each representing a quarter of the
   SNAI1e sequence (Supplementary Fig. [217]8). RNA pull-down analysis
   demonstrated that only the 5′ fragment (T1; 1–1023) was able to
   interact with BRD4, although its binding capacity was dampened compared
   to the full-length SNAI1e (Fig. [218]6c). Furthermore, we investigated
   the BRD4 domain(s) responsible for SNAI1e binding by analyzing BRD4
   truncation mutants (Fig. [219]6d). Depletion of either of the two
   bromodomains (BD1 or BD2) abolished the interaction between BRD4 and
   SNAI1e, indicating that the tandem bromodomains are essential for BRD4
   binding to SNAI1e (Fig. [220]6e). In vitro RIP-qPCR was performed to
   assess the direct interaction between BRD4-BD1/2 and SNAI1e. Compared
   to the mock control and the negative control recombinant FLAG-tagged
   SMURF2 protein, recombinant FLAG-BRD4-BD1/2 exhibited direct binding to
   SNAI1e (Fig. [221]6f). Consistent with this finding, in vitro RNA
   pull-down further validated the direct association between BRD4-BD1/2
   and SNAI1e (Fig. [222]6g).

Fig. 6. SNAI1e directly interacts with BRD4.

   [223]Fig. 6
   [224]Open in a new tab

   a The interaction between SNAI1e and BRD4 in MDA‐MB‐231 cells was
   analyzed by RNA immunoprecipitation (RIP). YTHDC1 served as a negative
   control. RT-qPCR was performed to detect SNAI1e expression in
   immunoprecipitants from MDA‐MB‐231 cells. The results are expressed as
   mean ± SD from three independent experiments. b The interaction between
   SNAI1e and BRD4 in MDA-MB-231 cells was analyzed by RNA pull-down.
   Western blotting analysis was performed to detect FLAG expression in
   whole-cell lysates (Input) and immunoprecipitants (IP). LETS1 and
   SNAI1e-AS served as negative controls. The RNA amounts used for
   pull-down were evaluated by agarose gel electrophoresis. c The
   interaction between SNAI1e truncation mutants and BRD4 in MDA-MB-231
   cells was analyzed by RNA pull-down. Western blotting analysis was
   performed to detect FLAG expression in whole-cell lysates (Input) and
   immunoprecipitants (IP). The RNA amounts used for pull-down were
   evaluated by agarose gel electrophoresis. d Schematic representation of
   full-length (FL) BRD4 and the truncation mutants tested. e The
   interactions between SNAI1e and BRD4 FL or the truncation mutants in
   MDA-MB-231 cells were analyzed by RNA pull-down. SNAI1e-AS, antisense
   SNAI1e; SNAI1e-S, sense SNAI1e. Western blotting analysis was performed
   to detect FLAG expression in whole-cell lysates (Input) and
   immunoprecipitants (IP). The RNA amounts used for pull-down were
   evaluated by agarose gel electrophoresis. f The direct interaction
   between SNAI1e and the FLAG-BRD4 BD1/2 recombinant protein was analyzed
   by in vitro RIP. The results are expressed as mean ± SD from three
   independent experiments. The FLAG-tagged proteins in immunoprecipitants
   were evaluated by western blotting. g The direct interaction between
   SNAI1e and the recombinant FLAG-BRD4 BD1/2 protein was analyzed by in
   vitro RNA pull-down. Western blotting analysis was performed to detect
   FLAG expression in whole-cell lysates (Input) and immunoprecipitants
   (IP). The RNA amounts used for pull-down were evaluated by agarose gel
   electrophoresis. Significance was calculated by using one-way ANOVA
   followed by Dunnett’s (a) and Tukey’s (f) multiple comparisons test.
   Data are representative of at least three (b, c, e, g) independent
   experiments with similar results. Co.vec empty control vector.

BRD4 is required for SNAI1e to induce SNAI1 expression and EMT

   Since eRNAs have been implicated in facilitating BRD4 binding to local
   enhancer regions^[225]50, we checked whether BRD4 binds to the SNAI1e
   gene body, and if any such interaction would be impacted by SNAI1e.
   Given that BRD4 binds to highly acetylated chromatin^[226]51,[227]52,
   we selected two H3K27ac-enriched regions (Region 2 and Region 9) within
   the SNAI1e gene body, as identified in Fig. [228]5c. ChIP-qPCR results
   demonstrated that SNAI1e depletion by GapmeR or CRISPRi suppressed BRD4
   association with the two SNAI1e gene body regions in MDA-MB-231 cells
   (Fig. [229]7a, Supplementary Fig. [230]9a). On the contrary,
   CRISPR-Display-mediated in cis tethering of SNAI1e transcript (but not
   SNAI1e-V1 transcript) by two gRNAs enhanced BRD4 recruitment to the two
   SNAI1e gene body regions, respectively (Fig. [231]7b). To investigate
   whether BRD4 is required for SNAI1e to elicit its biological effects,
   we utilized two independent shRNAs to knockdown BRD4 in MDA-MB-231
   cells (Fig. [232]7c). We observed that SNAI1e-induced SNAI1 mRNA
   expression was reduced upon BRD4 knockdown in MDA-MB-231 cells
   (Fig. [233]7d). In addition, masking of histone acetyl-lysine-binding
   pockets in BRD4 with the small molecule inhibitor JQ-1^[234]53
   confirmed the role of BRD4 in promoting SNAI1 expression upon both
   CRISPRa- and CRISPR-Display-mediated SNAI1e in cis overexpression
   (Fig. [235]7e, f, Supplementary Fig. [236]9b). Moreover, under both
   TGF-β-stimulated and unstimulated conditions, JQ-1 treatment reduced
   SNAI1e-induced migration in MDA-MB-231 (Fig. [237]7g) and decreased
   SNAI1e-induced changes in EMT marker expression in MCF10A-M2 cells
   (Fig. [238]7h). Additionally, JQ-1 suppressed SNAI1e-induced in vivo
   extravasation of MDA-MB-231 cells in zebrafish embryos (Fig. [239]7i)
   and reduced mammosphere formation by MCF10A-M2 cells (Fig. [240]7j).
   Taken together, our results suggest that blocking BRD4 with JQ-1
   inhibits SNAI1e-induced SNAI1 expression, EMT, and migration in breast
   cancer cells (Fig. [241]7k).

Fig. 7. BRD4 is required for SNAI1e to induce SNAI1 expression and promote
EMT and migration.

   [242]Fig. 7
   [243]Open in a new tab

   a, b ChIP-qPCR analysis of BRD4 enrichment at two regions in SNAI1e
   gene body upon GapmeR-mediated SNAI1e knockdown (a) and
   CRISPR-Display-mediated in cis overexpression of SNAI1e or SNAI1e-V1
   (b). RT-qPCR results are shown as mean ± SD from three biological
   replicates. c Validation of the knockdown efficiency of BRD4 by western
   blotting. d RT-qPCR analysis of SNAI1e-induced SNAI1 expression upon
   BRD4 knockdown. The results are shown as mean ± SD from three
   biological replicates. e, f Effect of JQ-1 on SNAI1 expression induced
   by SNAI1e overexpression using CRISPRa (e) and CRISPR-Display (f).
   RT-qPCR results are shown as mean ± SD (e) and mean ± SEM (f) from
   three biological replicates. g A transwell migration assay was
   performed to evaluate the effect of JQ-1 on TGF-β/SNAI1e-induced
   migration in MDA-MB-231 cells. The results are expressed as mean ± SD
   from seven biological replicates. h Effect of JQ-1 on
   TGF-β/SNAI1e-induced EMT marker expression in MCF10A-M2 cells. i In
   vivo zebrafish extravasation experiments with MDA-MB-231 cells upon
   CRISPRa-mediated SNAI1e overexpression and JQ-1 treatment. Extravasated
   breast cancer cell clusters are indicated with yellow arrows. Whole
   zebrafish image, bar = 618.8 μm; zoomed image, bar = 77.3 μm. The
   results are expressed as mean from 31 (gEV-JQ-1), 36 (gEV+JQ-1), 27
   (g1-JQ-1), and 41 (g1+JQ-1) biological replicates. j Mammosphere
   formation assays to check effect of JQ-1 and CRISPRa-mediated SNAI1e
   overexpression on MCF10A-M2 cell stemness. The numbers of mammospheres
   are expressed as mean ± SD from 24 biological replicates in the right
   panel. Scale bar = 100 μm. k Schematic working model. SNAI1e interacts
   with BRD4 to facilitate its binding to H3K27ac on the local enhancer,
   and thereby triggers SNAI1 transcription, TGF-β-induced EMT, migration,
   and stemness. JQ-1 treatment disrupts BRD4 binding to the local
   enhancer and inhibits SNAI1 transcription, TGF-β-induced EMT,
   migration, and stemness. Significance was assessed using one-way ANOVA
   followed by Dunnett’s (a, b, i) and Tukey’s (d–g, j) multiple
   comparisons test. Data are representative of at least three (c, h)
   independent experiments with similar results. gEV gRNA expression
   vector. Co.vec empty control vector. Scr scramble.

Discussion

   Our study proposes a mechanism by which a cis-regulatory eRNA
   potentiates TGF-β signaling. TGF-β induces SNAI1e expression from a
   SNAI1 enhancer region. SNAI1e interacts with and facilitates BRD4
   enrichment on the local enhancer, resulting in the promotion of the
   enhancer activity and upregulation of SNAI1 expression. SNAI1 interacts
   with and sequesters SMAD7 in the nucleus, leading to the suppression of
   TβRI polyubiquitination and degradation. As a consequence, TGF-β
   signaling, TGF-β-induced EMT and migration are strengthened by SNAI1e
   in breast cancer cells. We demonstrated that SNAI1e was more highly
   expressed in TNBC cell lines compared to normal breast cells and
   less-malignant cancer breast cell lines. The differences in basal
   SNAI1e expression across these four TNCB cell lines may be attributed
   to intrinsic genetic mutations or epigenetic changes, which require
   further investigation.

   SNAI1 expression is tightly controlled at multiple levels, from
   epigenetic modifications to post-translational regulation^[244]54. In
   response to TGF-β, SMAD3 directly binds to the SNAI1 promoter to
   stimulate transcriptional activation^[245]55,[246]56. Moreover, SNAI1
   can interact with SMAD3/4 and form a transcriptional repressor
   complex^[247]42. We discovered that SNAI1e is a direct target gene of
   TGF-β/SMAD signaling, and that SNAI1e depletion greatly mitigated
   TGF-β-induced SNAI1 expression, suggesting that TGF-β induces SNAI1e to
   promote SNAI1 expression in a feedforward loop. These results add
   another layer of regulation to TGF-β-induced SNAI1 expression. A recent
   study demonstrates that RAS-responsive element-binding protein 1
   (RREB1) binds to and primes SNAI1 enhancer regions for TGF-β-induced
   activation, which is required for transcriptional activation of SNAI1
   by TGF-β in mouse lung adenocarcinoma cells^[248]57. This aligns with
   our findings that TGF-β induces the enrichment of H3K27ac and H3K4me1
   at the SNAI1e gene body region in MDA-MB-231 cells. Further
   investigation is needed to determine whether SNAI1e interacts with
   RREB1 to enhance SNAI1 enhancer activity. In addition to serving as a
   TGF-β/SMAD signaling effector to drive EMT^[249]42, SNAI1 has been
   reported to reinforce TGF-β/SMAD signaling by enhancing TβRI protein
   expression in breast cancer cells^[250]40. Our results support this
   notion and further elucidated that SNAI1 interacts with and sequesters
   SMAD7 in the nucleus to alleviate TβRI polyubiquitination and
   proteasomal degradation. However, we have not excluded the possibility
   that SMAD7 binding may affect the transcriptional repressor activity of
   SNAI1.

   We characterized SNAI1e gene body as an active enhancer and validated a
   direct and specific contact between the gene bodies of SNAI1e and SNAI1
   by 4C-seq analysis. As a subclass of nuclear lncRNAs transcribed from
   enhancer regions, eRNAs contribute to alterations in their local
   enhancer landscape^[251]58,[252]59. We showed that SNAI1 expression was
   downregulated upon CRISPR/Cas9-mediated knockout of the entire enhancer
   region, but this method cannot separate the specific role of SNAI1e
   from that of the enhancer itself. To address this issue, we employed
   two complementary methods, GapmeR and CRISPR-Display, which
   specifically target the SNAI1e transcript for manipulation while
   preserving the integrity of the enhancer. Results from CRISPR-Display
   confirmed that SNAI1e induces SNAI1 expression in an in cis manner,
   highlighting the function of SNAI1e as a cis-regulatory eRNA. Moreover,
   these approaches excluded the potential effects of CRISPRa and CRISPRi
   on enhancer activity. A recent study identified the SCREEM locus as an
   enhancer region that induces SNAI1 expression, and thereby promotes EMT
   in human primary bronchial epithelial cells and regulates monocyte
   differentiation^[253]20. The authors demonstrated that SNAI1e, also
   referred to as SCREEM2, is one of three eRNAs (SCREEM1-3) produced from
   this locus^[254]20. A strong positive correlation between the
   expression of SNAI1e and SNAI1 was observed in monocytes^[255]20,
   supporting our findings that SNAI1 is one of the most upregulated genes
   upon SNAI1e overexpression in breast cancer cells. However, the
   previous study was unable to discriminate whether SNAI1 activation was
   driven by the SCREEM locus itself or by its transcripts^[256]20. The
   authors suggested that SCREEM transcripts might be indispensable or
   function redundantly, as shRNA-mediated depletion of individual SCREEM
   eRNAs did not reduce SNAI1 expression^[257]20. In contrast, our study
   employed GapmeR, which outperforms shRNA or siRNA for nuclear
   transcript knockdown^[258]60, to deplete SNAI1e and observed a
   significant reduction in SNAI1 expression across all the tested cell
   lines. Moreover, we confirmed this result using the orthogonal method
   of Cas13d-mediated depletion of the SNAI1e transcript. Furthermore, our
   CRISPR-Display results suggest that, despite being part of an eRNA
   cluster, in cis overexpression of SNAI1e alone is sufficient to
   activate SNAI1 expression. However, eRNAs generally exhibit a
   tissue-specific expression pattern and therefore play a role in
   activating their neighboring target gene expression in confined types
   of tissues^[259]19,[260]61,[261]62. Further investigation is needed to
   determine whether the effects of SNAI1e on promoting SNAI1 expression
   and TGF-β signaling can be expanded to other disease models.

   We uncovered that locally transcribed SNAI1e maintains enhancer
   activity by interacting with BRD4. BRD4 has been shown to promote GLI
   family zinc finger 1 (Gli1)-induced SNAI1 transcription in breast
   cancer cells^[262]63 and to prevent SNAI1 protein degradation by
   competing with the E3 ligases FBXL14 and β-TRCP1 in gastric cancer
   cells^[263]64. Our findings provide further insight into BRD4-mediated
   multi-level regulation of SNAI1 expression. Moreover, we found that
   SNAI1e binds to BRD4 through direct interaction with its two BD domains
   to reinforce BRD4 association with the histone modification H3K27ac.
   This is consistent with a previous study in which BRD4 tandem BD
   domains were found to be required for interaction with a broad spectrum
   of eRNAs in colon cancer cells^[264]50. BD domains are known to
   recognize histone acetyl-lysine residues in promoters and
   enhancers^[265]65. It is likely that the BD region binding to SNAI1e is
   distinct from the one recognizing histone acetyl-lysine. This is
   supported by previous studies indicating that BD point mutations
   disrupt DNA binding without altering acetyl-lysine association
   affinity^[266]66,[267]67. Hence, SNAI1e binding may change the BRD4
   conformation toward a state for enhanced interactions with
   acetyl-lysines. However, given that BD domains are essential for BRD4
   binding to both acetylated histones and SNAI1e, a comparison between
   full-length BRD4 and a BRD4 BD domain truncation mutant cannot clearly
   discriminate the effects of BRD4-SNAI1e interaction from those of
   BRD4-histone interaction in our study.

   We demonstrated that the 1000 nt fragment at the 5′ end of SNAI1e
   (SNAI1e-T1) is required for SNAI1e to bind BRD4. However, BRD4 could
   not interact as potently with SNAI1e-T1 as full-length SNAI1e. Given
   that the folding structure of lncRNAs determines their interactions
   with proteins^[268]68,[269]69, it is possible that SNAI1e truncation
   may dampen its original folding structure that is needed for BRD4
   binding. Further investigation may focus on mapping the minimal BRD4
   binding region on SNAI1e by other approaches such as cross‐linking and
   immunoprecipitation (CLIP) coupled with RNA footprinting^[270]70 in
   live cells. Our ChIP-qPCR results, conducted upon
   CRISPR-Display-directed SNAI1e in cis tethering, demonstrated that
   SNAI1e can facilitate BRD4 enrichment at the local enhancer. However,
   we cannot exclude the possibility that this effect may result from the
   alteration of H3K27ac at the local enhancer due to changes in the
   overall level of gene transcription. To further discriminate the
   SNAI1e-dependent BRD4 recruitment from acetylated histone-mediated BRD4
   enrichment, future studies could focus on disrupting the minimal core
   region required for BRD4 binding in SNAI1e, while preserving its
   structural integrity. This could be achieved using advanced RNA editing
   systems^[271]71,[272]72. Besides functioning as a scaffold protein,
   BRD4 acetylates histone H3 at Lys122 (H3K122) through its intrinsic
   histone acetyltransferase domain^[273]73. Whether SNAI1e binding
   affects BRD4-histone modification activities warrants further
   investigation.

   Our findings underscore a significant role of BRD4 in SNAI1e-driven
   EMT, migration, and stemness, as demonstrated by both BRD4 genetic and
   pharmacological inhibition. Given that BRD4 is widely expressed across
   tissues and is critical in normal physiological processes, such as
   hematopoiesis and host immune response^[274]74,[275]75, systemic
   targeting BRD4 with selective inhibitors can cause undesirable
   on-target effects, including digestive system dysfunctions and
   metabolic disorders in (pre)clinical trials^[276]52,[277]76, which
   greatly limit the therapeutic application of BRD4 inhibitors in cancer
   treatment. Moreover, cancer cells gain resistance to BRD4 inhibition
   after prolonged treatment in preclinical studies^[278]77,[279]78. Since
   EMT inhibition can suppress both cancer cell invasion and drug
   resistance^[280]2, targeting SNAI1e to suppress EMT in cancer cells may
   overcome the toxicity and drug resistance associated with BRD4
   inhibitors.

Methods

Lentiviral CRISPRa screen

   MDA-MB-231-CAGA[12]-GFP cells were created by lentiviral, single-copy
   integration of CAGA[12]-GFP and isolating a single cell clone that
   showed good induction in response to TGF-β. pLenti-dCas9-VP64-BFP
   (Addgene; 196711) was packaged into lentivirus in HEK293T (ATCC) using
   plasmids psPAX2 (Addgene; 12260, a gift from Dr. D. Trono, Swiss
   Federal Technology Institute of Lausanne, Lausanne, Switzerland) and
   pCMV-VSV-G (Addgene; 8454, a gift from Dr. R.A. Weinberg, Whitehead
   Institute for Biomedical Research, Cambridge, MA, USA). The
   virus-containing supernatant was concentrated 40-fold with Lenti-X
   concentrator (Takara), aliquoted and stored in liquid nitrogen.
   MDA-MB-231-CAGA[12]-GFP cells were then transduced with
   pLenti-dCas9-VP64-BFP, and a stable BFP positive population was
   isolated by repeated cell sorting (Sony SH800). gRNA sequences were
   designed against the 694 CAGE peaks from the FANTOM5 (GRCh38 v9 DPI
   clustering) dataset^[281]27,[282]28 which fell within 500 bp upstream
   and 200 bp downstream of the transcripts’ annotated start sites, using
   a −300 to 0 window relative to the maximum of the CAGE peak.
   CRISPick^[283]79,[284]80 was used to design and rank gRNAs, and the top
   five ranked gRNAs were included in the library. 400 non-targeting
   control gRNAs as well as positive control gRNAs against known
   TGF-β-pathway genes and the promoter region of the CAGA[12]-MLP were
   also included. The guide library was synthesized (Genscript), including
   an anchor sequence enabling synthesis of several libraries on the same
   array^[285]80, double stranded using an ultramer containing random
   sequence labels^[286]81 and amplified with primers CRISPRa_fw and
   CRISPRa_rev. The anchor sequence was removed by digestion with EcoRV,
   and the resulting final insert was cloned by Gibson assembly into
   pXPR_502 digested with BsmBI/EcoRI (Addgene; 96923, a gift from
   Dr. J.G. Doench and Dr. D. Root, Broad Institute of MIT and
   Harvard, Cambridge, MA, USA). The library plasmid was sequenced by NGS
   to confirm gRNA representation and packaged into lentivirus as
   described above. The functional titer of the library virus was
   estimated from the fraction of surviving cells after transduction with
   serial dilutions of virus followed by puromycin selection. dCas9-VP64
   expressing MDA-MB-231-CAGA[12]-GFP cells were then transduced with the
   library virus in duplicate at an approximate MOI of 0.3 and a coverage
   of 1000 cells per gRNA in the presence of 2 µg/ml polybrene. Transduced
   cells were selected with 2 µg/ml puromycin. A control sample was
   harvested on day 5 post transduction (p.t., early time point). Cell
   numbers per replicate were kept at 1000× coverage throughout. On day 7
   p.t., cells were treated with TGF-β (5 µg/ml) for 48 h. An unsorted
   control sample was harvested, and cells were sorted based on GFP
   expression, collecting the top and bottom 10% GFP. Genomic DNA was
   isolated using the QIAamp DNA Blood Mini (Qiagen), and gRNA and RSL
   sequences were amplified by PCR as described^[287]81, but with modified
   primers PCR2_fw and PCR3_fw. The amplicon was sequenced (Illumina)
   reading 20 cycles Read 1 with custom primer CRISPRSeq; 10 cycles index
   read i7 to read the RSL (not used in data analysis), and six cycles
   index read i5 for the sample barcode. NGS data was analyzed with the
   MaGeCK software, v.0.5.6^[288]82. The data is presented in
   Fig. [289]1a. x and y axes represent the −log[10] transformation of the
   positive robust ranking aggregation (RRA) scores comparing the GFP-high
   population to the GFP-low population. The dashed line indicates the
   cutoff of −log[10] (RRA score) at 2. The primers used for library
   construction and NGS sequencing are listed in Supplementary
   Table [290]3.

Cell culture and reagents

   HEK293T (CRL-1573), HepG2 (HB-8065), A549 (CRM-CCL-185), MDA-MB-231
   (CRM-HTB-26), MDA-MB-436 (HTB-130), HCC38 (CRL-2314), BT549 (HTB-122),
   and MCF7 (HTB-22) cells were purchased from the American Type Culture
   Collection (ATCC). SUM149-PT is a gift from Dr. S.E. Le Dévédec (Leiden
   Academic Centre for Drug Research, Leiden, The Netherlands). The cells
   mentioned above were cultured in Dulbecco’s modified Eagle medium
   (DMEM; Thermo Fisher Scientific; 41965062) supplemented with 10% fetal
   bovine serum (FBS; Thermo Fisher Scientific; 16000044) and 100 U/mL
   penicillin/streptomycin (Thermo Fisher Scientific; 15140163). MCF10A-M1
   and MCF10A-M2 cells were kindly provided by Dr. F. Miller (Barbara Ann
   Karmanos Cancer Institute, Detroit, USA) and cultured in DMEM/F12
   (GlutaMAX™ Supplement; Thermo Fisher Scientific; 31331028) containing
   5% horse serum (Thermo Fisher Scientific; 26050088), 0.1 μg/mL cholera
   toxin (Sigma‒Aldrich; C8052), 0.02 μg/mL Epidermal Growth Factor (EGF;
   Sigma‒Aldrich; 01-107), 0.5 μg/mL hydrocortisone (Sigma‒Aldrich;
   H0135), 10 μg/mL insulin (Sigma‒Aldrich; I6634) and 100 U/mL
   penicillin/streptomycin. All cell lines were maintained in a 5% CO[2],
   37 °C humidified incubator, tested monthly for mycoplasma contamination
   and checked for authenticity by short tandem repeat (STR) profiling.
   Recombinant TGF-β3 is a kind gift from Dr. A. Hinck (University of
   Pittsburgh, Pittsburgh, USA). CHX (50 μg/mL; Sigma‒Aldrich; C1988),
   BafA1 (5 μM for 6 h; Sigma‒Aldrich; B1793) and MG132 (20 nM for 6 h;
   Sigma‒Aldrich; 474787), Doxorubicin (Sigma‒Aldrich; D5220), Paclitaxel
   (Sigma‒Aldrich; T7191), SB431542 (1 μM) and JQ-1 (1 μM for 24 h;
   MedChemExpress; HY-13030) were used in the cell culture experiments.

Plasmid construction

   Full-length SNAI1e was amplified by PCR from MDA-MB-231 cell-derived
   cDNA and inserted into the lentiviral vector pCDH-EF1α-MCS-polyA-PURO.
   The inducible vector for SNAI1 ectopic expression was generated using
   Gateway cloning into the pLIX-403 vector (Addgene; 41395). BRD4
   truncation mutants were amplified by PCR from pcDNA5-Flag-BRD4-WT
   (Addgene; 90331) and inserted into the pCDNA3.1-FLAG vector. CRISPRa,
   CRISPRi, and Cas13d gRNAs were inserted into the lentiviral vectors
   lenti sgRNA(MS2)_puro optimized backbone (Addgene; 73797),
   pLKO.1-U6-PURO-AA19 (kindly provided by Dr. M.A.F.V. Gonçalves, LUMC,
   Leiden, The Netherlands), and pHR_hU6-crScaffold_EF1a-PURO (modified
   from pSLQ5465_pHR_hU6-crScaffold_EF1a-PuroR-T2A-BFP; Addgene; 155307),
   respectively. CRISPR-Display gRNAs were cloned to a modified pCDNA.3.1
   vector with a gRNA scaffold, followed by the insertion of SNAI1e or
   SNAI1e-V1. SNAI1e promoter fragments were amplified from MDA-MB-231
   genomic DNA and subcloned into the pGL4-luc backbone (Promega). All
   plasmids were verified by Sanger sequencing, and the primers used for
   plasmid construction are listed in Supplementary Data [291]2.

Lentiviral transduction

   Packaging plasmids (VSV, gag, and Rev) and expression constructs were
   cotransfected into HEK293T cells. At 48 h post‐transfection,
   supernatants were collected and added to target cells. After 48 h of
   infection, selection antibiotics were added to the medium to ensure the
   selection of stable cells. We used shRNA constructs from Sigma‒Aldrich
   for gene knockdown: TRCN0000040031 for SMAD4 knockdown, TRCN0000199427
   (#1) and TRCN0000318771 (#2) for BRD4 knockdown, and TRCN0000063819 for
   SNAI1 knockdown. lenti-dCAS-VP64_Blast (Addgene; 61425) and
   lenti-MS2-P65-HSF1_Hygro (Addgene; 61426) were used for the generation
   of CRISPRa stable cells. pHR-SFFV-dCas9-BFP-KRAB (Addgene; 46911) and
   pXR001: EF1a-CasRx-2A-EGFP (Addgene; 109049) were used for the
   generation of CRISPRi and CRISPR/Cas13d stable cells, respectively.

siRNA and GapmeR transfection

   A final concentration of 25 nM scramble GapmeR (UGGGCGUAUAGACGUGUUACAC)
   or GapmeR targeting SNAI1e (UGCAUCUGGACAGGGGUCUU) was transfected using
   lipofectamine 3000 (Thermo Fisher Scientific; L3000015). A final
   concentration of 25 nM non-targeting siRNA (siNT, Horizon) or siRNA
   targeting SNAI1e (UGCAUCUGGACAGGGGUCUU, Horizon) was also transfected
   using lipofectamine 3000.

RT‒qPCR

   To detect TGF-β target gene (PAI-1 and CTGF) and SNAI1 expression,
   cells were treated with TGF-β (1 ng/ml) or vehicle control for 8 h. To
   check TGF-β–induced SNAI1e expression, TGF-β (5 ng/ml) or vehicle was
   added for indicated time points or 8 h, if the treatment duration is
   not specified. MDA‐MB‐231 cells with a Tet-ON system were treated with
   different doses of Doxycycline (Dox; 0.1 μg/mL) for 2 days to induce
   SNAI1 expression. A NucleoSpin RNA kit (Macherey Nagel; 740955) was
   used to isolate total RNA from cells. Reverse transcription was carried
   out with a RevertAid RT Reverse Transcription Kit (Thermo Fisher
   Scientific; K1691). The indicated genes were amplified using the
   synthesized cDNA with specific primer pairs, and signals were
   visualized with a CFX Connect Real-Time PCR Detection System (Bio-Rad).
   GAPDH was used as the reference gene for normalization by the 2^−ΔΔCt
   method. The primer sequences used for RT‒qPCR are listed in
   Supplementary Data [292]3. All experiments were performed at least
   three times, and representative results are shown.

Western blotting

   To detect EMT marker expression, A549 or MCF10A-M2 cells were treated
   with TGF-β (1 ng/ml for A549 and 2.5 ng/ml for MCF10A-M2, respectively)
   or vehicle control for 1 day (A549) or 3 days (MCF10A-M2). To test JQ-1
   effect on EMT marker expression, MCF10A cells were pre-treated without
   or with JQ-1 (1 μM) for 24 h before seeding. To check TGF-β–induced
   p-SMAD2, TGF-β (1 ng/ml) or vehicle control was added for indicated
   time points or 1 h, if the treatment duration is not specified. To
   check TGF-β–induced SNAI1 expression, TGF-β (5 ng/ml) or vehicle
   control was used to treat cells for 8 h. MDA‐MB‐231 cells with a Tet-ON
   system were treated with Doxycycline (Dox; 0.1 μg/mL) for 2 days to
   include SNAI1 expression. RIPA buffer (150 mM sodium chloride, 1.0%
   Triton‐X‐100, 0.5% sodium deoxycholate, 0.1% sodium dodecyl sulfate
   (SDS), and 50 mM Tris–HCl (pH 8.0)) supplemented with complete protease
   inhibitor cocktail (Roche; 11836153001) was applied to lyse cells.
   Subsequently, protein concentrations were evaluated with a DC™ protein
   assay kit (Bio‐Rad; 5000111). Next, SDS–polyacrylamide gel
   electrophoresis (PAGE) was performed, and proteins were then
   transferred onto a 0.45‐μm polyvinylidene difluoride (PVDF) membrane
   (Merck Millipore; IPVH00010). Subsequently, the membrane was blocked
   with 5% nonfat dry milk in Tris‐buffered saline (TBS) with 0.1% Tween
   20 detergent (TBST) for 1 h at room temperature. After probing the
   membranes with the corresponding primary and secondary antibodies,
   images were acquired with a ChemiDoc Imaging System (Bio‐Rad). The
   primary antibodies are listed in Supplementary Table [293]4. All
   experiments were performed at least three times, and representative
   results are shown. ImageJ (National Institutes of Health, United
   States) was used to quantify relative protein expression levels by
   densitometry.

Chromatin immunoprecipitation (ChIP) assays

   MDA-MB-231 cells were treated without or with TGF-β (5 ng/mL) for 24 h.
   1 × 10^7 MDA-MB-231 cells were collected, cross-linked with 1%
   formaldehyde and sonicated into 200–700 bp fragments using sonication
   beads (Diagenode; C01020031). The collected supernatant was diluted
   five times and incubated with 5 μg of IgG (Cell Signaling Technology;
   2729 and 5415), anti-RNA polymerase II antibody (Active Motif; 39097),
   anti-SMAD3 antibody (Abcam; ab208182), anti-H3K4me1 antibody (Active
   Motif; 39635), anti-H3K27ac antibody (Active Motif; 39685),
   anti-H3K4me3 antibody (Active Motif; 61379), or anti-BRD4 antibody
   (Cell Signaling Technology; 13440) overnight at 4 °C. The next day, the
   antibody-chromatin complex was captured by 30 μL of Protein A/G
   Sepharose beads. After five washes, followed by RNase A (Thermo Fisher
   Scientific; R1253) and Proteinase K (Thermo Fisher Scientific;
   25530049) treatment, the DNA was extracted by isopropanol. RT-qPCR was
   performed to quantify the enrichment of protein at the regions of
   interest. Representative results from three independent experiments are
   shown.

Race

   RACE was performed on MDA-MB-231 cells using a 5′/3′ RACE Kit, 2nd
   generation (Roche; 03353621001). The resulting 5′/3′ RACE products were
   cloned to a pU.CAG.MSC2.rBGpA vector (kindly provided by Dr. M.A.F.V.
   Gonçalves, LUMC, Leiden, The Netherlands) and transformed into
   competent cells. Representative Sanger sequencing results from three
   independent colonies are presented.

CRISPR-Display

   MDA-MB-231 cells were transfected with indicated constructs expressing
   gRNA fused to SNAI1e or SNAI1e-V1, along with a dCas9 expression
   construct (kindly provided by Dr. M.A.F.V. Gonçalves, LUMC, Leiden, The
   Netherlands), using Lipofectamine 3000 (Thermo Fisher Scientific;
   L3000015). 6 h post-transfection, medium was replaced with fresh DMEM
   medium containing JQ-1 (1 μM) or a vehicle control (DMSO). RNA and DNA
   samples were collected after 48 h post-transfection. ChIP-qPCR and
   RT-qPCR were performed as described above. All experiments were
   performed at least three times, and representative results are shown.

SNAI1e KO

   MDA-MB-231 cells stably expressing Cas9 (Addgene; 98290) were
   transduced with gRNA expression constructs by lentivirus infection.
   Following puromycin selection, single cells were seeded into 96-well
   plates using a dilution method. Genomic DNA was isolated from multiple
   single clones using the Phire Tissue Direct PCR Master Mix kit (Thermo
   Fisher Scientific; F170L). The SNAI1e gene body region was
   characterized by PCR-based genotyping with LongAmp® Taq DNA Polymerase
   (NEB; M0323S). Two single clones exhibiting SNAI1e gene body depletion
   (SNAI1e KO) were selected for RT-qPCR analysis. A pool of
   MDA-MB-231-Cas9 cells transduced with an empty gRNA expression vector
   served as the control (WT).

Transcriptional reporter assays

   Cells transfected with indicated constructs were serum-starved for 6 h
   and stimulated with TGF-β (5 ng/ml) or vehicle control for 16 h.
   Luciferase activity was measured with the substrate D-luciferin
   (Promega) and a luminometer (PerkinElmer) and normalized to
   β-galactosidase activity. MDA-MB-231 cell with stable expression of the
   CAGA[12]-dynGFP reporter^[294]26 was used to monitor the TGF-β-induced
   transcriptional response in an IncuCyte live cell imaging system (Essen
   BioScience). Cells were serum-starved for 16 h and stimulated without
   or with TGF-β (0.5 ng/mL). GFP intensity was quantified as total green
   integrated intensity normalized with cell confluence. All experiments
   were performed three times, and representative results are shown.

Immunofluorescent staining

   To evaluate the expression and localization of SMAD7, MDA-MB-231 cells
   with a Tet-ON system for SNAI1 ectopic expression were treated without
   or with Doxycycline (Dox; 0.1 μg/mL) for 2 days. Cells were washed with
   phosphate-buffered saline (PBS), fixed, permeabilized by
   phosphate-buffered saline (PBS) supplemented with 0.1% Triton X-100 for
   10 min. Afterward, non-specific binding was blocked with 3% bovine
   serum albumin (BSA) dissolved in PBS for 1 h at room temperature. Cells
   were incubated with a primary antibody against SMAD7 (1:100 dilution;
   R&D; MAB2029) for 1 h at room temperature. After three times of washing
   with PBS, the specimens were probed with secondary antibodies in a
   dilution of 1:1000 for 1 h at room temperature. The specimens were then
   subjected to three washes with PBS and mounted with VECTASHIELD
   antifade mounting medium with DAPI (Vector Laboratories; H-1200).
   Experiments were performed three times, and representative results are
   shown. For F-actin staining, A549 cells were treated without or with
   TGF-β (2.5 ng/mL) for 48 h and incubated with Phalloidin conjugated
   with Alexa Fluor 488 (1:500 dilution; Thermo Fisher Scientific;
   A12379). Experiments were performed twice, and representative results
   are shown. Images were acquired with a Leica SP8 confocal microscope
   (Leica Microsystems).

Ubiquitination assay

   MDA-MB-231-HA-Ub cells were treated with MG132 (5 µM) for 5 h prior to
   harvesting. Cells were washed with cold PBS twice containing 10 mM
   N-ethylmaleimide (NEM; Sigma-Aldrich; E3876), cells were lysed in 1%
   SDS-RIPA buffer (25 mM Tris–HCl, pH 7.4, 150 mM NaCl, 1% NP40, 0.5%
   sodium deoxycholate and 1% SDS) consisting of protease inhibitors and
   10 mM NEM. Lysates were subsequently boiled for 5 min and diluted to
   0.1% SDS at a final concentration in RIPA buffer. Then protein
   concentrations were measured and the same amount of proteins were
   incubated with 2 µL TβRI antibody (Abcam; ab235578) for 16 h at 4 °C.
   The mixture was then incubated with 20 µL of Protein A Sepharose (GE
   Healthcare; 17-0963-03) for 2 h at 4 °C. After five washes, the beads
   were boiled in 2× sample buffer and analyzed by western blotting.
   Experiments were performed three times, and representative images are
   shown.

Co-immunoprecipitation (Co-IP)

   HEK293T transfected with indicated plasmids were lysed with TNE lysis
   buffer (50 mM Tris–HCl, pH 7.4, 1 mM EDTA, 150 mM NaCl, and 1% NP40)
   containing freshly added complete protease inhibitor cocktail and kept
   on ice for 15 min. The lysates were centrifuged at 1.4 × 10^4 g for
   10 min at 4 °C. Equal amounts of protein were incubated with anti-FLAG
   agarose beads (Sigma‒Aldrich; A2220) for 30 min at 4 °C with rotation.
   Pierce™ Protein G Agarose (Thermo Fisher Scientific; 20397) was added
   and kept for 2 h at 4 °C with rotation. Beads were washed five times
   with the TNE buffer for 5 min at 4 °C with rotation. Afterward, samples
   were boiled with 2× sample buffer for 5 min and subjected to SDS-PAGE
   analysis.

IncuCyte migration assays

   40 μl MDA-MB-231 cells suspended in medium supplemented with 0.5% serum
   were seeded at a density of 1 × 10^3 in the inserts of an Incucyte
   clearview 96-well plate (Essen BioScience; 4582). 20 μl medium
   supplemented with 0.5% serum-containing indicated compounds or
   corresponding vehicle controls was added into the inserts. Afterward,
   cells were allowed to settle at ambient temperature for 20 min. In
   parallel, 200 μl medium supplemented with 10% serum was added to the
   reservoir plates. Then the inserts containing cells were placed into a
   pre-filled plate. Cells on the top and bottom of inserts were imaged
   and analyzed using the IncuCyte live cell imaging system (Essen
   BioScience). Cells were treated without or with TGF-β (5 ng/mL) during
   the assays. To test the effect of JQ-1 on cell migration, cells were
   treated without or with JQ-1 (1 μM) for 24 h before seeding. Cells in
   the top and bottom chambers were imaged and quantified with the
   IncuCyte system. Experiments were performed three times, and
   representative images are shown.

Subcellular fractionation

   Cytoplasmic and nuclear fractions were collected from MDA-MB-231 cells
   (challenged with or without TGF-β (5 ng/mL) for 1 day or Dox
   (0.1 μg/mL) for 2 days). Cells were collected and lysed in 250 μl of
   buffer A (50 mM Tris–HCl (pH 7.4), 150 mM NaCl, 1% NP40, and 0.25%
   sodium deoxycholate) for 15 min on ice. After centrifugation at
   3000 × g for 5 min, the supernatant was collected and saved as the
   cytoplasmic fraction. The pellet was washed with PBS twice and
   resuspended in 150 μl of buffer B (50 mM Tris–HCl (pH 7.4), 400 mM
   NaCl, 1% NP40, 0.5% sodium deoxycholate, and 1% SDS). After 20 min of
   incubation on ice and centrifugation at 12,000 × g for 15 min, the
   supernatant was collected and saved as the nuclear fraction. The
   isolated cytoplasmic and nuclear fractions were used to quantify the
   expression of RNA transcripts by RT-qPCR. Experiments were performed
   three times, and representative results are shown.

RNA immunoprecipitation (RIP)

   To identify interactions between SNAI1e and BRD4, RIP was performed
   with a Magna RIP™ RNA-Binding Protein Immunoprecipitation Kit (Merck
   Millipore; 17-700). Anti-BRD4 antibody (Cell Signaling Technology;
   13440), anti-YTHDC1 antibody (Cell Signaling Technology; 77422s), or
   normal rabbit IgG was incubated with MDA-MB-231 cell lysates for 16 h
   at 4 °C. For in vitro RIP, 10 pmol of in vitro-transcribed SNAI1e was
   incubated with 2 µg recombinant FLAG-SMURF2 protein (Sigma‒Aldrich;
   SRP0228) or FLAG-BRD4 BD1/2 recombinant protein (R&D systems; SP-600)
   for 16 h at 4 °C as described previously^[295]21. RNA was extracted
   from the beads, and RT‒qPCR was performed as mentioned above.
   Experiments were performed three times, and representative results are
   shown.

RNA pull-down assay

   A MEGAscript Kit (Thermo Fisher Scientific; AM1334) was used for in
   vitro RNA synthesis. 50 pmol of RNA was biotinylated with an RNA 3′ End
   Desthiobiotinylation Kit (Thermo Fisher Scientific; 20160). The
   tertiary structure of each lncRNA was recovered by 10 min of incubation
   at 70 °C followed by gradual cooling to room temperature. HEK293T cell
   lysates or FLAG-BRD4 BD1/2 recombinant protein (R&D systems; SP-600)
   was incubated with biotinylated LETS1^[296]22, SNAI1e-AS or SNAI1e for
   16 h at 4 °C. Proteins were eluted from the beads and analyzed by
   western blotting. Experiments were performed three times, and
   representative images are shown.

Proximity ligation analysis (PLA)

   PLA was performed to analyze the endogenous interaction between SMAD7
   and SNAI1. MDA-MB-231 cells were fixed and incubated with primary
   antibodies against SMAD7 (R&D; MAB2029) and SNAI1 (Cell Signaling
   Technology; 3879S) at a 1:500 dilution for 16 h at 4 °C. PLUS and MINUS
   PLA probes conjugated to secondary antibodies (Sigma‒Aldrich; DUO92001
   and DUO92005) were used to incubate cells for 1 h at 37 °C. Ligase
   (Sigma‒Aldrich; DUO92008) was added to the cells for a 30-min
   incubation before Duolink® Polymerase (Sigma‒Aldrich; DUO82028)
   incubation for 90 min at 37 °C. Images were acquired with a Leica SP8
   confocal microscope (Leica Microsystems). Experiments were performed
   three times, and representative images are shown.

Flow cytometry

   MCF10A-M2 cells were incubated with fluorescein isothiocyanate
   (FITC)-conjugated anti-human CD44 (BD Biosciences; 347943) and
   R-phycoerythrin (PE)-conjugated anti-human CD24 (BD Biosciences;
   555428) antibodies (1:500 dilution) for 30 min at 37 °C. IgG isotypes
   (560952/560951; BD Biosciences) were used as a control. At least 10,000
   cells were acquired with a BD LSR II flow cytometer (BD Biosciences),
   and the results were analyzed with FlowJo 10.5.0 software (for gating
   strategy, see Supplementary Fig. [297]10). Experiments were performed
   three times, and representative results are shown.

Tumor sphere formation assays

   MCF10A-M2 cells were resuspended in DMEM/F12 medium (no phenol red;
   Thermo Fisher Scientific; 21041033) containing 1× B27 (Thermo Fisher
   Scientific; 17504044) and 20 ng/ml EGF (Sigma‒Aldrich; 01-107). 2000
   cells were seeded into wells of an ultralow attachment 24-well plate
   (Corning; CLS3473-24EA). To test the effect of JQ-1 on cell stemness,
   cells were pre-treated without or with JQ-1 (1 μM) for 24 h before
   seeding. After 14 days of standard incubation, the numbers of spheres
   (>50 mm diameter) were counted using an inverted microscope (DMi8;
   Leica Microsystems). Experiments were performed three times, and
   representative results are shown.

RNA-seq-based transcriptional profiling and GSEA

   To screen for mRNAs affected by SNAI1e, the DNBSeq platform (BGI, Hong
   Kong) was used to perform RNA-seq in MDA-MB-231 cells upon
   CRISPRa-mediated SNAI1e overexpression. RNA-seq reads were processed
   using the opensource BIOWDL RNA-seq pipeline v5.0.0
   ([298]https://eur03.safelinks.protection.outlook.com/?url=https%3A%2F%2
   Fzenodo.org%2Frecord%2F5109461%23.Ya2yLFPMJhE&data=05%7C02%7Cc.fan%40lu
   mc.nl%7C745ef3b26cca4179ad3108dcb5216766%7Cc4048c4fdd544cbd80495457aacd
   2fb8%7C0%7C0%7C638584402282389605%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wL
   jAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C0%7C%7C%7C&sdat
   a=f%2FKAhFLm%2BzgqqZEAhf0O1ns2ygC5sFlRJAgCXN5qgPs%3D&reserved=0)
   developed at the LUMC. This pipeline performs FASTQ preprocessing
   (including quality control, quality trimming, and adapter clipping),
   alignment, read quantification, and optionally transcript assembly.
   FastQC (v0.11.9) was used for checking raw read QC. Adapter clipping
   was performed using Cutadapt (v2.10) with the default settings. RNA-seq
   reads’ alignment was performed using STAR (v2.7.5a) on human reference
   genome GRCh38. The gene read quantification was performed using
   HTSeq-count (v0. 12.4) with the Ensembl gene annotation version 111.
   |Log2(fold change)| >1, false discovery rate <0.05 was set as a cutoff.
   The upregulated genes upon SNAI1e overexpression are listed in
   Supplementary Data [299]4. The TGF-β^[300]83, BMP^[301]84, WNT^[302]85
   and YAP^[303]86 response gene signature, SNAI1-induced gene
   signature^[304]87, and EMT (GO: 0001837) gene signature were used to
   perform GSEA with the GSEA software^[305]85.

Fluorescence in situ hybridization (FISH)

   MDA‐MB‐231 cells were treated without or with TGF-β (5 ng/mL) for 8 h.
   An RNAScope® Multiplex Fluorescent Kit (Advanced Cell Diagnostics;
   323100) was utilized to evaluate the expression and localization of
   SNAI1e in the cells. Images were acquired with a Leica SP8 confocal
   microscope (Leica Microsystems). Representative results from two
   independent experiments are shown.

Embryonic zebrafish extravasation assay

   The experiments were conducted in a licensed establishment for the
   breeding and use of experimental animals (LU) and subject to internal
   regulations and guidelines, stating that advice is taken from the
   animal welfare body to minimize suffering for all experimental animals
   housed at the facility. The zebrafish assays described are not
   considered an animal experiment under the Experiments on Animals Act
   (Wod, effective 2014), the applicable legislation in the Netherlands in
   accordance with the European guidelines (EU directive no. 2010/63/EU)
   regarding the protection of animals used for scientific purposes,
   because non-self-eating larvae where used. Therefore, no license
   specific for these assays on zebrafish larvae (<5 days post
   fertilization) was required. MDA-MB-231 cells labeled with mCherry were
   injected into the duct of Cuvier of embryos from transgenic zebrafish
   (fli; EGFP) as previously described^[306]33. Approximately 200
   MDA-MB-231 cells were injected in Fig. [307]2j and Supplementary
   Fig. [308]2m, while around 400 MDA-MB-231 cells were injected in
   Figs. [309]2k, [310]7i. Vehicle control (DMSO) or JQ-1 (1 μM) was added
   to the egg water from the first day post injection. An inverted SP5
   STED confocal microscope (Leica) was used to visualize the injected
   cancer cells and zebrafish embryos. Two independent experiments were
   performed, and representative results are shown., zebrafish

Circular chromosome conformation capture (4C) sequencing

   4C experiments were performed strictly following a detailed protocol as
   described elsewhere^[311]43. Briefly, MDA-MB-231 cells were
   serum-starved for 16 h and stimulated without or with TGF-β (5 ng/mL)
   for 24 h. 5 × 10^6 cells were collected and fixed with 2% formaldehyde
   for 10 min, followed by quenching with 0.13 M glycine. After cell lysis
   and nuclei isolation, the samples were digested with 100 U DpnII (New
   England Biolabs, R0543M) and ligated using 50 U T4 DNA ligase (Thermo
   Fisher Scientific; EL0012). Next DNA samples were reverse cross-linked
   and purified by Nucleomag PCR beads (Macherey Nagel; 744100.34). 50 U
   Csp6I (Thermo Fisher Scientific; ER0211) was added for second
   digestion, after which the second ligation was performed using 50 U T4
   DNA ligase. A 2-step 4C PCR was carried out using viewpoint specific
   primers for SNAI1e (TACACGACGCTCTTCCGATCTTCTGGGCAACAGAGGGAGAT,
   ACTGGAGTTCAGACGTGTGCTCTTCCGATCTGCTTCAGTCCTCGCTAAGAA) using Expand™ Long
   Template PCR System (Roche; 11681834001) to prepare the library and add
   Illumina adapters for sequencing. 4C libraries were sequenced on an
   Illumina NextSeq 2000 platform. FastQ files were demultiplexed and
   processed as previously described^[312]41
   ([313]https://github.com/deLaatLab/pipe4C). Reads were mapped against
   hg38. Read counts were normalized to one million mapped
   intra-chromosomal reads excluding the two highest covered fragments and
   21 fragment end rolling mean scores were calculated for every fragment
   end. Biological replicates (n = 3) were averaged, and profile overlays
   were produced using R 4.2.0 ([314]https://www.R-project.org).

Statistics & reproducibility

   Statistical analyses were performed using a two-tailed unpaired
   Student’s t-test, or one/two-way ANOVA with GraphPad Prism 10.2.3 as
   indicated in the figure legends. Each experiment was repeated
   independently for two or three times (as indicated in the figure
   legends and “Methods” section) with similar results. The exact value of
   n, which represents the number of technical or biological replicates
   used in the experiments, is indicated in the figure legends. P < 0.05
   was considered statistically significant. For the number of zebrafish
   used in the study, sample sizes were determined according to experience
   with previous experiments^[315]33,[316]88. Zebrafish isolated from eggs
   were mixed in one plate containing egg water to randomize them before
   injection of cells with genetic manipulation. No data points were
   excluded from the analyses. The investigators were not blinded to
   allocation during experiments and outcome assessment. All measurements
   in this study were taken from distinct samples.

Reporting summary

   Further information on research design is available in the [317]Nature
   Portfolio Reporting Summary linked to this article.

Supplementary information

   [318]Supplementary Information^ (2.7MB, pdf)
   [319]41467_2025_58032_MOESM2_ESM.pdf^ (81.3KB, pdf)

   Description of Additional Supplementary Files
   [320]Supplementary Data 1^ (907.5KB, xlsx)
   [321]Supplementary Data 2^ (17.6KB, xlsx)
   [322]Supplementary Data 3^ (12.3KB, xlsx)
   [323]Supplementary Data 4^ (12.5KB, xlsx)
   [324]Reporting Summary^ (851KB, pdf)
   [325]Transparent Peer Review file^ (290KB, pdf)

Source data

   [326]Source Data^ (4.4MB, xlsx)

Acknowledgements