Genome–wide binding of transcription factor ZEB1 in triple‐negative breast cancer cells

Zinc finger E‐box binding homeobox 1 (ZEB1) is a transcriptional regulator involved in embryonic development and cancer progression. ZEB1 induces epithelial‐mesenchymal transition (EMT). Triple–negative human breast cancers express high ZEB1 mRNA levels and exhibit features of EMT. In the human triple–negative breast cancer cell model Hs578T, ZEB1 associates with almost 2,000 genes, representing many cellular functions, including cell polarity regulation (DLG2 and FAT3). By introducing a CRISPR‐Cas9‐mediated 30 bp deletion into the ZEB1 second exon, we observed reduced migratory and anchorage‐independent growth capacity of these tumor cells. Transcriptomic analysis of control and ZEB1 knockout cells, revealed 1,372 differentially expressed genes. The TIMP metallopeptidase inhibitor 3 and the teneurin transmembrane protein 2 genes showed increased expression upon loss of ZEB1, possibly mediating pro‐tumorigenic actions of ZEB1. This work provides a resource for regulators of cancer progression that function under the transcriptional control of ZEB1. The data confirm that removing a single EMT transcription factor, such as ZEB1, is not sufficient for reverting the triple–negative mesenchymal breast cancer cells into more differentiated, epithelial‐like clones, but can reduce tumorigenic potential, suggesting that not all pro‐tumorigenic actions of ZEB1 are linked to the EMT.

In carcinomas, but also during embryogenesis, EMT is guided by extracellular growth factors, such as transforming growth factor β (TGFβ), hepatocyte growth factor, fibroblast growth factor (FGF), and the Notch receptor system (Nieto et al., 2016). The transmembrane TGFβ receptors type II and type I, members of the receptor serine/ threonine kinase family, that also exhibit weak tyrosine kinase activity, signal via Smad proteins, lipid, and protein kinases and control gene expression via specific transcription factors (Moustakas & Heldin, 2012). TGFβ contributes to metastatic progression of carcinomas, by promoting EMT, suppressing anti-tumoral immune responses, and by enhancing the differentiation of cancer-associated fibroblasts and the growth of the tumor vasculature (Bierie & Moses, 2006).
Thus, ZEB1 is best known as a transcriptional repressor of CDH1 and inducer of EMT in breast and other carcinomas (Eger et al., 2005).
During embryogenesis, ZEB1 controls several mesenchymal cell lineages giving birth to cranial, limb, thoracic, and vertebral bones and cartilage (Takagi, Moribe, Kondoh, & Higashi, 1998). For this reason, mice lacking ZEB1 die early after birth due to skeletal and thymic defects (Takagi et al., 1998). In mediating EMT, ZEB1 represses epithelial polarity genes, such as Crumbs3 and Lgl2 (Aigner et al., 2007;Spaderna et al., 2008). Repression of laminin-332 (LAMC2) and integrin-β4 (ITGB4) genes contributes to the invasiveness associated with EMT in prostate carcinoma cells (Drake et al., 2010). Extensive alternative splicing occurs during EMT, in part mediated by ZEB1, which transcriptionally represses the epithelial splicing regulatory protein genes, favoring expression of spliced isoforms of the FGF receptors that help maintain EMT in breast cancer cells in response to TGFβ (Horiguchi et al., 2012). The micro-RNA 200 (miR-200) gene family is actively repressed by ZEB1 in response to TGFβ signaling; miR-200 pairs with the ZEB1 and TGFβ2 mRNAs and inhibits their translation, thus forming a double-negative feedback loop that is critical for breast carcinoma EMT (Burk et al., 2008). Epithelial miR-200 expression is maintained by the transcription factor c-Myb, which is transcriptionally repressed by ZEB1 (Hugo et al., 2013;Pieraccioli, Imbastari, Antonov, Melino, & Raschella, 2013). Thus, ZEB1 represses several genes in carcinomas, but also activates transcription, when pairing with the co-activator YAP of the Hippo pathway, inducing mesenchymal gene expression (Lehmann et al., 2016). ZEB1 promotes metastasis in breast and pancreatic carcinomas (Krebs et al., 2017;Spaderna et al., 2008). For example, ZEB1 facilitates bonespecific metastasis of breast carcinomas by inducing expression of noggin, follistatin and chordin-like 1, extracellular antagonists that inactivate ligands of the activin, and bone morphogenetic protein branches of the TGFβ family (Mock et al., 2015). ZEB1 contributes to the resistance to anticancer therapy by establishing a repressive chromatin state (Meidhof et al., 2015). Resistance also extends to radiotherapy, as radiation stabilizes ZEB1 and promotes signaling by the CHK1 protein kinase, stimulating homologous DNA recombination (Zhang et al., 2014).
Overall, the transcription factor ZEB1 mediates functions that link cancer EMT to TGFβ signaling, metastatic dissemination, stemness, and resistance to therapy. This generates a strong interest in deciphering the complete regulatory network downstream of ZEB1 in carcinomas. Based on this premise, we analyzed the genome-wide association of ZEB1 and evaluated the loss of function mutation in ZEB1 in breast carcinomas.
Solidified plates were then incubated at 37°C for 16 days, cells were viewed under a phase-contrast microscope, and colonies were counted. Each experiment was performed three times and each condition included triplicates.

| T-Scratch assay
Cells were seeded in a six-well plate such that they were 90% confluent the following day. A " + " scratch was made on the cell layer using a pipetman tip. Cells were washed with PBS twice and left in DMEM. The scratched area was photographed under a phase contrast microscope. The culture was left at 37°C overnight and then photographed under the same microscope. Cell images on day 1 and 2 were analyzed using the T-Scratch software (http://www.cse-lab. | ethz.ch/index.php?&option = com_content&view = article&id = 363) to quantify cell migration. Each experiment was performed three times and each condition included triplicates.
Lysates were heated at 95°C for 5 min and subjected to SDS-PAGE.

| Immunofluorescence microscopy
Cells were cultured on glass slides in six-well plates with DMEM/10% FBS, fixed in 3.7% paraformaldehyde for 10 min at room temperature, washed twice with PBS and permeabilized in 1% Triton X-100 in PBS for 15 min at room temperature. Cells were washed in PBS-T (0.2% Triton X-100 in PBS) and left for blocking in 20% FBS in PBS-T for 1 hr at room temperature. Cells were incubated overnight with primary antibodies

| Chromatin immunoprecipitation
Cells were cultured and fixed with 2% formaldehyde for 10 min at 37°C, washed in ice-cold PBS twice, scraped in PBS and spun down at 4,000 rpm for 5 min. Cell pellets were lysed in 1% SDS, 10 mM EDTA, 50 mM Tris, pH 8.1, with protease inhibitors for 20 min on ice, then sonicated to an average DNA fragment size of 250 bp. Input chromatin aliquots (10%) were frozen at −20°C. The remaining lysate was diluted 10 times in 0.01% SDS, 1.0% Triton X-100, 1.2 mM EDTA, 16.7 mM Tris-HCl pH 8.1, 167 mM NaCl, with protease inhibitors, and proceeded for immunoprecipitation using anti-ZEB1 or control rabbit antiserum, overnight at 4°C.

| DNA library preparation and sequencing protocols
ChIP DNA was obtained with four biological and technical replicates and pooled for sequencing. ChIP DNA quality was analyzed with a Bioanalyzer. Input and ChIP DNA was sheared with a Covaris S2 sonicator (Covaris, Inc., Woburn, MA). DNA libraries were constructed using the AB Library Builder System (Life Technologies, Carlsbad, CA), followed by amplification and wildfire conversion according to the manufacturer's protocols. Library preparation was performed using the library kit (5,500 SOLiD Library Builder Fragment Core Kit + 5500W Conversion Primer Kit), after which sequencing was performed at 75 bp read length on the SOLiD 5,500W system (Life Technologies) at sequencing unit (SOLiD 5,500W FlowChip). Raw sequences were aligned to the human genome hg19 using maximum stringency with default settings via LifeScope (version 2.1 Thermo Fisher Scientific), retaining only uniquely mapped reads and unique sequences were retained in the . BAM file format.

| Real time RT-PCR analysis
RNA from Hs578T wild-type and ZEB1 knockout clones was extracted using the TRIzol reagent protocol (Ambion, LifeTechnologies, Thermo

| Transcriptomic analysis in the GOBO database
The gene expression data sets for ZEB1, ZEB2, and TIMP3 in human breast cancer cells were obtained by utilizing the cell line module of the webbased tool gene expression-based Outcome for breast cancer Online (GOBO) (Ringner, Fredlund, Hakkinen, Borg, & Staaf, 2011). Data were obtained using default settings and protocols prescribed by the authors.

| ChIP-Seq analysis
Aligned reads were filtered on mapping quality using samtools (Li et al., 2009) with "-q 20." Peaks were called from ChIPs with the Input as background using MACS software (version 1.4.2) (Zhang et al., 2008) with the following changes to default settings: "-nomodel−shiftsize = 125-keep-dup = 1." Identified peaks were annotated to the closest gene using BEDTools (Quinlan & Hall, 2010) and protein coding genes from the refseq-databases (hg19). FASTA-sequence (Human Genome version GRCh37.57) +/ − 125 bp from the center of each peak-summit as predicted by MACS was extracted and enriched DNAmotifs were identified using TOMTOM motif identification suit (Bailey et al., 2009). Data was obtained using default settings and prescribed protocols by the authors.

| Visualization of peaks
ChIP-Seq peaks were uploaded into the UCSC genome browser (Kent et al., 2002) using custom bigwig tracks and overlayed using publicly available H3K27AC data.

| GeneOntology
Differentially expressed genes and binding targets of ZEB1 were classified into groups of molecular, biological, and cellular components using GO Panther online tool (Mi, Muruganujan, & Thomas, 2013). Genes were grouped into functional groups using default settings and protocols.

| DMFS plot
Distant metastasis-free survival (DMFS) plots were analyzed using the . ZEB1 expression positively correlates with breast cancer aggressiveness and metastatic potential (Aigner et al., 2007;Spaderna et al., 2008). We confirmed this knowledge by querying the expression of ZEB1 (and its related transcription factor ZEB2) in several human breast cancer cell lines based on data available in the GOBO database (Ringner et al., 2011). Cells with high ZEB1 (and ZEB2) expression classified as basal-B breast cancer cells (Figure 1a). Basal-A and luminal epithelial breast cancer cells expressed low or undetectable levels of ZEB1 (and ZEB2, Figure 1a). On the other hand, not all basal-B cells exhibited robust mRNA levels for ZEB1 (or ZEB2, Figure 1a). Basal-B breast cancer cells are frequently reported as being EMT-like tumor cells (Hennessy et al., 2009;Taube et al., 2010).
However, mRNA expression profiles, although useful and widely used, not always correlate with protein expression. We screened for

| Genome-wide association of ZEB1 in triple-negative breast cancer cells
We then performed high yield ChIP using the Hs578T cells and the same ZEB1 antibody used in Figure 1, followed by sequencing analysis (ChIP-Seq, Figure 2a, Supplementary Table S1) suggesting less functional relevance to ZEB1 (Figure 2d). In the superclass of cellular components, ECM genes had a higher relevance, and fold enrichment (Figure 2d).

| Novel ZEB1 targets in breast cancer cells
As part of a strict validation of peaks mapping between −10,000 to +2,500 bp from the TSS, we selected genes from two different classes of the GO analysis, nuclear binding (DLG2, discs large MAGUK scaffold protein 2) and embryo development (FAT3, fat atypical cadherin 3) (Figure 3). ChIP followed by quantitative qPCR, using as primers the DNA sequences corresponding to the coordinates of peaks obtained from the ChIP-Seq experiment, revealed significant ZEB1 binding to DLG2 and FAT3 (Figure 3b-d).
Visualization of the ChIP-seq peaks on the UCSC genome browser MATURI ET AL.

| ZEB1 knockout suppresses cell migration and anchorage-independent growth
As the germline knockout mouse of ZEB1 is lethal due to skeletal defects and severe T cell deficiency in the thymus (Takagi et al., 1998), and in order to exclude the chance of insufficient transient knockdown, a complete knockout was made in Hs578T cells using CRISPR-CAS9

| Transcriptomic analysis of genes regulated by ZEB1
The availability of a clean knockout system in the tumor cells, led us to perform a whole genome transcriptomic analysis, using the AmpliSeq assay in order to analyze the expression levels of all RefSeq genes in the C3 and CZ1 cells ( Figure 5, Table S2). The assay uses targeted enrichment of over 21,000 genes and resulted in 10.5-37.8M reads out of which over 99% aligned to the human reference genome and 230 (red dots) up-regulated and down-regulated genes respectively, with many genes expressed but having unchanged levels (gray dots) (Figure 5a-d). To our surprise, we found a substantial set of genes that were down-regulated in ZEB1 knockout cells, which suggests that ZEB1 could act as a direct positive regulator of transcription; this class included DNA binding, chromatin organization, and organelle organization genes (Figure 4b). Gene Enrichment analysis showed that the positive transcriptional action of ZEB1 is compatible with previous analyses (Lehmann et al., 2016). On the other hand, and as expected, ZEB1 showed a negative correlation with a larger set of genes belonging to classes like lipid binding, protein binding, enzymatic activity, small molecule transportation, and calcium-dependent phospholipid binding, which scored positive, with a 3.83-fold enrichment relative to all RefSeq genes (Figure 5c). ZEB1 also exhibited a strong impact on the expression of genes linked to the regulation of cell adhesion, system development, and sensory development (Figure 5c).
To measure the direct impact of ZEB1 on gene expression, both ChIP-Seq and transcriptomic array data were over-laid (Figure 5d, " + "), which resulted in a total of 154, 21 down-regulated and 133 upregulated genes (Figure 5d, Supplementary Table S3). Gene ontology analysis on this subset of 154 genes did not show significant enrichment for any specific functional class of genes (data not shown), because of the relatively low number of genes analyzed. These data suggest that ZEB1 regulates several hundreds of genes, but only a subset (around 11% of these) are direct target genes to which ZEB1 binds in triple-negative breast cancer cells.

| ZEB1 regulates the metalloproteinase inhibitor TIMP3
One of the 133 up-regulated genes in the ZEB1 knockout cells which  (Carvalho et al., 2008). The H3K27Ac enhancer marker profile showed that ZEB1 bound to a region of tandemly clustered enhancer elements (Figure 6b), which was validated by independent ChIP-qPCR analysis (Figure 6c). In agreement with the AmpliSeq results, RT-PCR analysis confirmed that TIMP3 mRNA levels increased dramatically after ZEB1 knockout (Figure 6d). These data also support the notion that ZEB1 knockout cells drift toward an epithelial phenotype.

| The transmembrane protein TENM2 is a novel target gene of ZEB1
Another of the 133 up-regulated genes in the ZEB1 knockout cells, which exhibited binding of ZEB1 by ChIP-Seq analysis (Figure 5d), was teneurin-2 (TENM2) a cell adhesion transmembrane protein studied in neuronal and adipocyte progenitor cells (Tews et al., 2017). Gene ontology classifies TENM2 in the calcium ion binding and cell adhesion category, which was negatively regulated by ZEB1 (Figure 5c). On a global scale, transcriptomics and antibody-based proteomics showed that TENM2 expression is higher in the heart and brain (Fagerberg et al., 2014), whereas breast tissue expression was not reported earlier.

| DISCUSSION
Previous investigations on the transcription factor ZEB1 analyzed its contribution to the process of EMT and identified specific genes, whose expression is regulated by ZEB1 (Gheldof et al., 2012). By analyzing triple-negative breast cancer cells we identified a genome-wide binding profile of ZEB1 on a few 1,000 genes. CRISPR-Cas9-mediated knockout of ZEB1 in the same cells resulted in the misregulation of a similarly large number of genes. This experimental approach has revealed: a) previously unrecognized cohorts of genes, whose expression is regulated by ZEB1 in breast cancer cells and b) the unexpected finding that ZEB1 contributes to the oncogenic potential of breast cancer cells, a function that may not necessarily link to the process of EMT.
Transcriptomic analyses across various carcinomas proposed association of ZEB1 mRNA expression and tumor aggressiveness, including metastatic potential (Aigner et al., 2007;Spaderna et al., 2008;Taube et al., 2010). This is classically associated with the contribution ZEB1 makes to EMT (Gheldof et al., 2012). or of ubiquitin ligases, such as Siah (Chen et al., 2015) that keep ZEB1 protein levels low. Despite this difficulty, we analyzed Hs578T cells that express robust endogenous ZEB1 protein levels.
Hs578T cells belong to the triple-negative breast cancer group and classify under the basal-B group and the claudin-low subgroup that exhibits mesenchymal features (Hennessy et al., 2009;Taube et al., 2010). We confirmed this fact, and identified significant constitutive activity of TGFβ signaling in these cells. Reverting these cells to a more epithelial phenotype using a TGFβ receptor type I inhibitor (GW6604) was impossible. We conclude that Hs578T cells are mesenchymal and do not exhibit so-called hybrid epithelial/mesenchymal (E/M) features, possibly reflecting a tumor cell type more enriched in cancer stem cells (Lambert et al., 2017).
ChIP-Seq analysis identified ∼3,900 genomic sites where ZEB1 binds and a large proportion of these sites mapped on (the TSS or within 10,000 bp or less) well characterized gene bodies. ZEB1 binds to the E-box of CDH1 and a few other genes (Gheldof et al., 2012). DNA binding motif identification analysis revealed that ZEB1 can additionally associate to DNA sequences, previously defined by other transcription factors, including RREB1, RUNX2, and Smad. ZEB1 binding to RUNX2 and Smad motifs in breast cancer cells appears functionally likely, as RUNX2 regulates mesenchymal progenitor to osteoblast differentiation, and mutations in RUNX2 perturb the RUNX2-Smad association, linking to syndromes of craniofacial dysplasia (Zhang et al., 2000). Mice lacking ZEB1 also exhibit craniofacial malformations (Takagi et al., 1998), suggesting that ZEB1 and RUNX2 may act in synergy by regulating common genes during embryogenesis. The physical association between ZEB1 and Smad (Remacle et al., 1999) or between Smad and RUNX2 (Zhang et al., 2000), also underscore the importance of the motifs identified in this screen. The importance of a transcriptional crosstalk between ZEB1-RUNX2 and Smad in breast cancer progression and EMT remains to be analyzed; such crosstalk most probably links to the mesenchymal program induced by these three transcriptional regulators.
The functional approach we used in order to validate ZEB1 target genes, ZEB1 inactivation by CRISPR-Cas9 targeting, generated some important conclusions. The combined ChIP-Seq and transcriptomic analysis generated a list of 154 genes whose levels changed after ZEB1 knockout and which exhibited direct binding of ZEB1 (Supplementary   Table S3). These genes provide new clues to the function of ZEB1 in binding to the TIMP3 gene; ChIP-Seq peaks (marked in red box) and tracks of H3K27Ac ChIP-Seq, which is used as a marker of enhancer activity, using the UCSC genome browser. (c) ChIP-qPCR showing significant enrichment of the TIMP3 promoter region in a ChIP experiment using the ZEB1 antibody, relative to enrichment by non-specific IgG. (d) Relative amount of TIMP3 mRNA expressed in C3 and CZ1 Hs578T cells after normalization with the HPRT1 house-keeping mRNA. Statistical significance p-value <0.01; n = 3 breast cancer. For example, CDH1 and polarity genes (Crumbs3, Lgl2) are known direct target genes repressed by ZEB1 in carcinoma EMT (Aigner et al., 2007;Eger et al., 2005;Spaderna et al., 2008). We provide evidence for additional members of the extended cadherin family such as FAT3 (Zhang et al., 2016), and polarity-linked regulators such as DLG2 (Roberts, Delury, & Marsh, 2012 (Lehmann et al., 2016).
The second unexpected finding with the ZEB1 knockout Hs578T cells was the suppression of oncogenicity observed in vitro in terms of anchorage-independent growth. Some of the newly identified target genes of ZEB1 may thus be involved in survival and proliferation pathways, mechanisms possibly not directly linked to the EMT. This is compatible with ZEB1 being a pleiotropic transcription factor, and is supported by recent evidence that explains how ZEB1 contributes to tumor cell radioresistance by regulating homologous DNA recombination (Zhang et al., 2014). EMT is thought to be linked to a relatively slow cell proliferation, cell cycle arrest, and long-term survival (Nieto et al., 2016). Accordingly, knockout of ZEB1 should have generated cells that However, the result obtained was the opposite.
The third lesson derived from the ZEB1 knockout analysis was the relative inability to revert the mesenchymal Hs578T breast cancer cells to epithelial cells. A trend toward epithelialization was obvious morphologically, but was not supported by molecular analysis of epithelial proteins, such as the tight junction protein CAR or E-cadherin (data not shown). We conclude that removal of a single EMT-TF, ZEB1, from an established mesenchymal breast cancer cell, is not sufficient to revert the tumor cells into a more epithelial phenotype, as we also demonstrated for Snail1 and Twist1 in an independent breast cancer model (Tan et al., 2015).
In summary, this work provides a useful resource for the exploration of novel functions of ZEB1 in the context of breast cancer. It opens the possibility of identifying ZEB1 target genes, which may allow interference with their function and generate, together with ZEB1 loss of function approaches, synthetic lethality that eliminates breast cancer cells.

CONFLICTS OF INTEREST
The authors declare no conflicts of interest.