ZEB1‐regulated inflammatory phenotype in breast cancer cells

Zinc finger E‐box binding protein 1 (ZEB1) and ZEB2 induce epithelial‐mesenchymal transition (EMT) and enhance cancer progression. However, the global view of transcriptional regulation by ZEB1 and ZEB2 is yet to be elucidated. Here, we identified a ZEB1‐regulated inflammatory phenotype in breast cancer cells using chromatin immunoprecipitation sequencing and RNA sequencing, followed by gene set enrichment analysis (GSEA) of ZEB1‐bound genes. Knockdown of ZEB1 and/or ZEB2 resulted in the downregulation of genes encoding inflammatory cytokines related to poor prognosis in patients with cancer, including IL6 and IL8, therefore suggesting that ZEB1 and ZEB2 have similar functions in terms of the regulation of production of inflammatory cytokines. Antibody array and ELISA experiments confirmed that ZEB1 controlled the production of the IL‐6 and IL‐8 proteins. The secretory proteins regulated by ZEB1 enhanced breast cancer cell proliferation and tumor growth. ZEB1 expression in breast cancer cells also affected the growth of fibroblasts in cell culture, and the accumulation of myeloid‐derived suppressor cells in tumors in vivo. These findings provide insight into the role of ZEB1 in the progression of cancer, mediated by inflammatory cytokines, along with the initiation of EMT.


Introduction
Breast cancer is a leading cause of cancer death in female patients. It is a heterogeneous disease, similar to cancers of other organs (Badve et al., 2011), and it is an especially troubling disease because healthy young women without any previous history of disease are often affected and show poor prognosis (Anders et al., 2008). In particular, triple-negative breast cancers that do not express the estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor 2 (HER2) are associated with very poor prognosis because they have a strong ability to metastasize, a high risk for relapse, and are refractory to chemotherapy (Anders and Carey, 2009). Among several subtypes of breast cancer, basal-type breast cancer accounts for most triple-negative breast cancers and shows a mesenchymal phenotype that is accompanied by high expression of mesenchymal genes including vimentin, and genes involved in the induction of epithelial-mesenchymal transition (EMT; Neve et al., 2006;Sarrio et al., 2008).
Zinc finger E-box binding protein 1 (ZEB1) and ZEB2 (also known as dEF-1 and SIP-1, respectively) are well-known transcriptional regulators that induce EMT, which plays important roles in both normal physiological and pathological processes (Eger et al., 2005;Vandewalle et al., 2005;Zhang et al., 2015). While ZEB1 and ZEB2 have many similar properties in transcriptional regulation, they are different in their expression profiles, some molecular and biological functions, including regulation of cell differentiation and disease progression (Postigo and Dean, 2000;Wakamatsu et al., 2001). Epithelial cells lose their adhesive property and become migratory and invasive as they become mesenchymal cells during the EMT process (Nieto et al., 2016;Thiery et al., 2009). Transforming growth factor-b (TGF-b) is one of the main cytokines that promotes the EMT (Heldin et al., 2012;Lamouille et al., 2014). Specifically, TGF-b binds to type I and type II receptors and transduces signals through Smad and non-Smad signaling pathways (Derynck and Zhang, 2003;Heldin et al., 1997;Massague, 2012). The TGF-b type I receptor is activated by ligand stimulation and induces the phosphorylation of the receptor-regulated Smads (R-Smads), Smad2 and Smad3, which form trimeric complexes with the common-partner Smad, Smad4. These Smad complexes translocate into the nucleus where they regulate the transcription of various target genes in cooperation with other transcription factors. While expression of ZEB1 and ZEB2 is suppressed by epithelial miR-200 family of miRNA (Gregory et al., 2008), TGF-b induces their expression in addition to some other EMT-related transcription factors, including Snail, and Slug, in certain types of normal and cancer cells (Gregory et al., 2011;Heldin et al., 2012;Miyazono et al., 2012;Xu et al., 2009).
We reported previously that TGF-b decreases the expression of E-cadherin through the induction of ZEB1 and ZEB2 in mouse mammary epithelial cells (Shirakihara et al., 2007). We also reported that it induces isoform switching of fibroblast growth factor receptors by alternative splicing, which occurs through downregulation of the expression of epithelial splicing regulatory proteins (ESRPs) by ZEB1 and ZEB2 (Horiguchi et al., 2012). Furthermore, the expression profiles of ZEB1 and ZEB2 are inversely correlated with those of ESRPs in human breast cancer cell lines and tumor specimens (Horiguchi et al., 2012). Consistent with the relationships between the EMT and cancer malignancy, high expression of ZEB1 is associated with poor prognosis of many types of cancer, including breast cancer (Chu et al., 2013;Jang et al., 2015). ZEB2 expression is also reported to be associated with poor prognosis of several types of cancer, although less frequently than ZEB1 (Fang et al., 2013;Prislei et al., 2015). It was recently reported that the EMT is involved in cancer malignancy by contributing not only to metastasis but also to the acquisition of cancer stem cell properties and chemoresistance (Fischer et al., 2015;Zheng et al., 2015). A recent genome-wide analysis of EMT-related transcription factor binding regions in pancreatic cancer cells suggested that ZEB1 plays a role in inducing the mesenchymal phenotype by suppressing enhancers that regulate the expression of epithelial genes (Diaferia et al., 2016). However, that analysis focused only on epithelial gene expression that was related to the EMT. Thus, the detailed mechanisms by which ZEB1 and ZEB2 contribute to poor prognosis in cancer remain to be elucidated.
Here, we employed chromatin immunoprecipitation sequencing (ChIP-seq) and RNA sequencing (RNAseq) to investigate the transcriptional program that is regulated by ZEB1 in several basal-type breast cancer cell lines. We found that ZEB1 directly upregulated the production of inflammatory cytokines in the basaltype breast cancer cell lines MDA-231-D (a highly metastatic clone of MDA-MB-231; Ehata et al., 2007) and Hs578T. ZEB2 partially showed similar function, including the induction of interleukin (IL)-6 and IL-8 production. Our findings suggested that ZEB1 promotes the proliferation of cancer cells and contributes to the formation of the tumor microenvironment by regulating the expression of inflammatory cytokines. The ZEB1-regulated inflammatory phenotype identified in this study provides insights into a mechanism that is critical for cancer progression and helps explain the poor prognosis of basal-type breast cancer.

Cell culture
The MDA-231-D human basal-type breast cancer cell line, a highly bone metastatic clone of MDA-MB-231, was described previously (Ehata et al., 2007) and cultured in Dulbecco's modified Eagle's medium (DMEM; #11965, Thermo Fisher Scientific, Waltham, MA, USA). Hs578T human basal-type breast cancer cells were obtained for use in this study from the American Type Culture Collection (ATCC, Manassas, VA, USA) and cultured in RPMI1640 (#11875, Thermo Fisher Scientific) with 0.01 mgÁmL À1 of insulin (12585-014, Thermo Fisher Scientific). HCC1954-Luc human basal-type breast cancer cells were obtained for use in this study from the Japanese Collection of Research Bioresources (JCRB) Cell Bank (Ibaragi, Osaka, Japan) and cultured in RPMI1640. MCF7 human luminal-type breast cancer cells were obtained from JCRB and cultured in DMEM with 0.01 mgÁmL À1 of insulin. 4T1 mouse breast cancer cells were obtained from the ATCC and cultured in high-glucose DMEM. WI-38 and IMR-90 human lung fibroblast cells were obtained from the ATCC for use in this study and cultured in high-glucose DMEM. Hs578T, HCC1954-Luc, MCF7, WI-38, and IMR-90 cells were used within 6 months of passage after purchase. All culture media included 10% fetal bovine serum (FBS), 100 unitsÁmL À1 penicillin G, and 100 lgÁmL À1 streptomycin. All cells were maintained in a 5% CO 2 atmosphere at 37°C.

RNA-seq and data analysis
RNA-seq was performed as described previously (Isogaya et al., 2014;Mizutani et al., 2016). cDNA libraries were prepared using the RNeasy Mini Kit with the On-Column DNase Digestion Set (QIAGEN, Venlo, The Netherlands), Dynabeads mRNA DIRECT Purification Kit, and the Ion Total RNA-Seq Kit v2 (Thermo Fisher Scientific).
2.5. Chromatin immunoprecipitation, ChIP-seq, and data analysis MDA-231-D, Hs578T, and MCF7 cells were cultured in 10-cm plates, and ChIP and ChIP-seq were performed as described previously (Koinuma et al., 2009;Murai et al., 2015). Data were obtained using the Ion Proton sequencer (Thermo Fisher Scientific). Unfiltered 50-bp sequence reads were aligned against the human reference genome (NCBI Build 36, hg19). Public anti-ZEB1 ChIP-seq data were obtained from GEO (GSM1574278, GSM1010809, GSM803411; ENCODE Project Consortium, 2012). Peaks were called using MACS2 (Zhang et al., 2008). CISGENOME software was used to assign a binding site to the nearest gene within 50 kb of a peak (Ji et al., 2011).

RNA isolation and quantitative RT-PCR (qRT-PCR)
Total RNA was extracted using the RNeasy Mini Kit (QIAGEN). First-strand cDNA synthesis was performed using PrimeScript2 reverse transcriptase and oligo dT primers (TaKaRaBio, Shiga, Japan) according to the manufacturer's instructions. qRT-PCR was performed using the ABI PRISM7500 Fast Real-Time PCR System or the StepOnePlus Real-Time PCR system (Thermo Fisher Scientific) and the Fast Start Universal SYBR Green Master Mix with ROX (Roche Diagnostics, Basel, Switzerland). Mouse and human GAPDH were used for normalization. The primer sequences are shown in Table S1. Data are reported as the means of two technical replicates unless otherwise indicated in the figure legends.

Preparation of conditioned medium and enzyme-linked immunosorbent assay (ELISA)
MDA-231-D cells and Hs578T cells were seeded (2 9 10 5 per well in six-well plates for IL-6 experiments and 1 9 10 5 per well in 12-well plates for IL-8 experiments). After overnight incubation, siRNA was transfected as described previously, followed by TGF-b incubation (1 ngÁmL À1 ), LY364947 treatment (1 lM), or a medium change (2 mL per well for six-well plates and 1 mL per well for 12-well plates) on the next day of transfection. The supernatant was collected after incubation for 48 h. To prepare the supernatant from HCC1954-Luc cells, the cells were seeded on a six-well plate (1 9 10 5 per well), followed by TGF-b stimulation, LY364947 (3 lM) treatment, or a medium change (2 mL) the next day. After 48 h of incubation, the supernatant was collected. The concentrations of IL-6 and IL-8 were measured using the human IL-6 Quantikine ELISA Kit and the human CXCL8/IL-8 Quantikine ELISA Kit (R&D systems), respectively, according to the manufacturer's instructions. Data are reported as the means of two biological replicates.

Lentiviral vector preparation and infection, and construction of plasmids
Lentiviral expression vectors were obtained from Hiroyuki Miyoshi (RIKEN BioResource Center; present address: Keio University, Tokyo, Japan). Lentiviral vectors were prepared by cotransfection of 293FT cells with pCSII-EF-mZEB1 or pCS-CDF-CG-PRE (for EGFP expression) and packaging vectors (pCAG-HIVgp and pCMV-VSV-G-RSV-Rev). The medium was changed after 24 h of transfection, and the culture media containing virus particles were collected after incubation for an additional 48 h. cDNAs encoding mouse ZEB1 and human ZEB2 were cloned into lentiviral expression vector or pcDEF3 expression vector. These plasmids were introduced into cells using Lipofectamine 2000 or Lipofectamine 3000 (Thermo Fisher Scientific) according to the recommended protocols.

Antibody array
The Human Cytokine Antibody Array C2000 (Ray Biotech, Norcross, GA, USA) was used according to the manufacturer's instructions. The LAS-4000 lumino-image analyzer (GE Healthcare, Buckinghamshire, UK) was used for chemiluminescence detection, and the strength of each spot was measured using the line profile function of MultiGauge software (FUJIFILM, Tokyo, Japan) and analyzed using the Analysis Tool for AAH-CYT-2000 (Ray Biotech).

Tissue array
A tissue array of multiple organ tumors (MC6163) was purchased from US Biomax (Rockville, MD, USA). The paraffin-embedded array was deparaffinized and rehydrated followed by antigen retrieval using 10 mM sodium citrate buffer (pH 6.0). Endogenous peroxidase activity was blocked by 3.0% hydrogen peroxide. The array was then blocked with Blocking One reagent (Nacalai Tesque) and incubated with anti-ZEB1 antibody (NBP1-05987; Novus Biologicals) and human IL-6 antibody (R&D systems). Anti-rabbit Alexa488 and anti-goat Alexa594 were used as the secondary antibodies. The array was mounted with DAPI-containing mounting medium. Tile scanning was performed using the Leica DMI6000 B inverted microscope with adaptive focus control (Leica Microsystems, Wetzlar, Germany). The signal intensity of each array spot was scored by two researchers (A.K. and Y.T.).

Tumor model
All animal experiments were performed in accordance with the policies of the animal ethics committee of the University of Tokyo. HCC1954-Luc cells (4 9 10 5 ) were injected into the mammary fat pads of 6-weekold female BALB/c nude mice, and 4T1 cells (5 9 10 5 ) were injected subcutaneously into BALB/c mice. The lengths and widths of the resulting tumors were measured using calipers, and the tumor volume was calculated as follows: 0.5 9 (major axis) 9 (minor axis) 2 .
2.13. Senescence-associated b-galactosidase (SA-bGal) staining Senescent cells were detected using the Senescence b-Galactosidase Staining Kit (Cell Signaling Technology) according to the manufacturer's instructions.

Statistical analysis
Student's t-test was used for two-sample analyses, and the Tukey-Kramer test was used for multisample analyses. Mann-Whitney U-test was used for in vivo data.

Identification of ZEB1 target genes in breast cancer cells
To determine the genome-wide distribution of ZEB1binding regions in MDA-231-D and Hs578T basal-type breast cancer cells, we performed ChIP-seq analysis using a validated ZEB1 antibody that did not crossreact with ZEB2 ( Fig. S1A; Horiguchi et al., 2012). We also obtained ChIP-seq data in MCF7 luminal-type breast cancer cells. Data were obtained from TGFb-treated cells based on the known functional interaction between Smad signaling and ZEB1 and ZEB2 (Postigo, 2003;Postigo et al., 2003;Verschueren et al., 1999). We identified 32 907 binding regions in MDA-231-D cells, 13 514 regions in Hs578T cells, and 281 regions in MCF7 cells that had q-values < 0.05. Using a more stringent threshold for peak calling, 14 811, 3131, and 108 binding regions were identified in MDA-231-D, Hs578T, and MCF7 cells, respectively (q < 10 À4 ). Significant peaks were found at known binding sites (ESRP2 and CDH1 gene loci) in the basal-type breast cancer cells (Horiguchi et al., 2012), but not at the HBB gene locus, which served as a negative control ( Fig. 1A and data not shown). No peaks were found at the ESRP2 and CDH1 gene loci in MCF7 cells, which likely reflected the low expression of ZEB1 in luminal-type breast cancer cells (Horiguchi et al., 2012). One of the de novo predicted common motifs in the ZEB1-binding regions in MDA-231-D cells matched the known ZEB1binding motif, which contained 'CACCT' (q = 0.0289) and was found in 38% of the 14 811 binding regions (Fig. 1B). The identified motif was also enriched toward the peak summit positions of ZEB1-binding regions, supporting the validity of the obtained ChIP-seq data (Fig. 1C).
Our data were then compared to all of the publically available human ZEB1-binding data for human pancreatic carcinoma Panc-1, human hepatoblastoma HepG2, and the Gm12878 B-lymphoblastic cell lines using the same read aligner, peak caller, and parameters (Diaferia et al., 2016;Hensen et al., 2014). We found that MDA-231-D cells and HepG2 cells had distinct profiles in terms of their ZEB1-binding regions and that Hs578T and Panc-1 cells, but not Gm12878 cells, shared ZEB1-binding regions with MDA-231-D cells (Fig. 1D). Many of the ZEB1-binding regions in Hs578T and Panc-1 cells overlapped with those in MDA-231-D cells (69.8% and 71.7%, respectively). In contrast, Gm12878 cells shared only 29.3% of their ZEB1-binding regions with MDA-231-D cells, and most of the remaining regions were shared with HepG2 cells, indicating that ZEB1-binding regions were relatively conserved in basal-type breast cancer cells, but not in other types of cancer cells.
ZEB1 acts as both a transcriptional activator and repressor (Lehmann et al., 2016;Sanchez-Tillo et al., 2011). To clarify the global view of ZEB1-regulated transcription, we performed RNA-seq using MDA-231-D cells. To examine the function of ZEB1 in the context of the EMT, MDA-231-D cells were transfected with a combination of siRNA targeting ZEB1 and ZEB2 and were then left untreated or were stimulated with TGF-b. The amounts of ZEB1 and ZEB2 proteins were efficiently decreased by the siRNA in MDA-231-D cells (Fig. S1B) and the specificity of the siRNA was confirmed (Fig. S1C). The ZEB1-binding regions were present within 50 kb of the transcriptional start site of 14 900 genes, which accounted for 64.0% of the total number of genes analyzed. Of the ZEB1-bound genes, 5975 were expressed at 10 or more FPKM (fragments per kilobase of exon per million mapped reads), and the number of genes that were either upregulated or downregulated by ZEB1/2 knockdown was similar (Fig. 1E).
We then evaluated the effect of ZEB1/2 siRNA on TGF-b-induced changes in the expression of ZEB1bound genes. Focusing on the ZEB1-target genes that were either up-or downregulated more than 2-fold by TGF-b, we found that knockdown of ZEB1/2 largely upregulated the expression of genes that were downregulated by TGF-b (Fig. 1F). This tendency was not observed for genes that were upregulated by TGF-b.

ZEB1 regulates the expression of inflammatory response genes
To identify the functional signatures of ZEB1-bound genes that were enriched in MDA-231-D cells, gene set enrichment analysis (GSEA) was conducted using MDA-231-D cell gene expression data ( Fig. 2A). The most notable finding was that the expression of inflammatory response genes was downregulated by the knockdown of ZEB1 and ZEB2 in the TGF-buntreated condition (Fig. 2B). The downregulated genes that were categorized as inflammatory response genes are listed in Table S2, along with positional information about ZEB1 binding. A similar tendency was observed in the TGF-b-treated condition; remarkably, the IL1A, CCL20, IL6, SERPINE1, IL8, and IL1B genes were commonly and strongly downregulated in both conditions (Fig. 2C). Focusing on the ZEB1 ChIP-seq data, peaks were found at the promoter or enhancer regions of the identified genes, including IL6, IL8, and CSF2, in MDA-231-D and Hs578T cells, but not in MCF7 cells (Fig. 2D), suggesting that ZEB1 binds and regulates the expression of inflammatory response genes in basal-type breast cancer cells.
Based on these findings, we focused on ZEB1-regulated secretory proteins. An antibody array detected 174 human cytokines in the conditioned culture media of MDA-231-D cells that were transfected with ZEB1/ 2 siRNA. ZEB1/2 siRNA downregulated secreted The y-axis scales are not equal as they were adjusted to show the binding signals clearly for each cell line. The kb sizes denote the ranges shown in the panels. The bars below the binding signals indicate the significant binding regions (q < 0.05). (B) De novo motif prediction was performed from the 500bp sequences flanking the summit position of each ZEB1-binding region using the Gibbs Motif Sampler provided by CisGenome. Default parameters were used except for the mean motif length (8), maximum motif length (15), and K (15). The identified E-box-like motif was compared to known motifs using the Tomtom motif comparison tool in MEME suite version 4.11.2. The alignment of the calculated motif with a top-ranked motif (MA0103.2) that was deposited as a ZEB1-binding site is shown. Y-axis scale: bits. (C) Motif centrality analysis was performed using the motif calculated in (B) and CentriMo of the MEME suite. The x-axis indicates the relative position of the best site from the peak summit of each ZEB1-binding region. (D) A heatmap representation of the ZEB1-binding regions in the genomes of the indicated cell lines. The y-axis indicates the union of the ZEB1-binding regions that were obtained from five cell lines as sorted by the signal intensities of MDA-231-D and HepG2. X-axis: the position of each binding region relative to the peak summits of the ZEB1-binding regions shown on the y-axis. (E) Frequencies of the ZEB1-bound genes showing expression that was up-or downregulated more than 1.5-or 2-fold by siRNA targeting ZEB1 and ZEB2 (siZEB1-1 and siZEB2-1) in the absence of TGF-b. Of the 15 175 ZEB1-bound genes (ZEB1-binding regions within 50 kb from the transcription start site, TSS), the genes encoding small RNA or showing expression that was < 10 FPKM (fragments per kilobase of exon per million mapped reads) were excluded from the evaluation. In panels (E) and (F), MDA-231-D cells were transfected with siRNA and treated with or without 1 ngÁmL À1 of TGF-b for 24 h. RNA-seq was then performed. siNC, control siRNA. (F) A scatter plot showing the relationship between the effect of ZEB1/2 siRNA (siZEB1-1 and siZEB2-1) and TGF-b as determined by RNA-seq. ZEB1-bound, TGF-b-regulated genes showing expression that was induced or suppressed ≥ 2-fold were selected (n = 5975). siZEB1/2-1, siZEB1-1 + siZEB2-1. Each dot represents a single gene. IL-1b, IL-8, and IL-1a proteins in MDA-231-D cells. The expression levels of the CCL20, PLAUR (plasminogen activator, urokinase receptor), CSF3 (granulocyte colony-stimulating factor, G-CSF), and CSF2 (granulocyte-macrophage colony-stimulating factor, GM-CSF) proteins were also regulated by ZEB1 (Fig. 2E). Of note, reduction in secreted IL-6 protein by ZEB1/2 siRNA was not observed because of the saturated signals in both control siRNA-and ZEB1/2 siRNA-treated conditions. We then used two sets of ZEB1/2 siRNA to quantitatively evaluate the effects of ZEB1/2 siRNA on IL-6 and IL-8 secretion by MDA-231-D cells and Hs578T cells using ELISA and found a similar tendency as in the RNA-seq analysis ( Fig. 2F  and G). Notably, the amounts of proteins secreted by MDA-231-D cells were comparable to the levels secreted by inflammatory cells, suggesting that IL-6 and IL-8 were produced at functional levels (Nastasi et al., 2015;Pazmandi et al., 2012).

ZEB1 and ZEB2 have similar functions in the regulation of inflammatory cytokine expression
While both ZEB1 and ZEB2 suppress the expression of CDH1 and induce EMT, previous reports have also revealed some different functions between them. We thus evaluated whether ZEB1 and ZEB2 have similar functions in the production of inflammatory cytokines. We knocked down the expression of ZEB1 or ZEB2 and obtained RNA-seq data from MDA-231-D cells. Knockdown of ZEB1 or ZEB2 did not strongly affect each other's protein expression (Fig. S1C). GSEA analysis suggested that siRNA against either ZEB1 or ZEB2 downregulated the expression of inflammatory response genes, including IL6 and IL8, although normalized enrichment score (NES) of inflammatory response genes in one of the ZEB2 siRNA was not more than 1.5 (NES = 1.34; Fig. 3A and Fig. S1D). qRT-PCR analysis also revealed that expression of IL6 and IL8 was decreased by knockdown of either ZEB1 or ZEB2 (Fig. 3B). Therefore, ZEB1 and ZEB2 have essentially similar functions in the regulation of inflammatory response gene expression, especially IL6 and IL8, in the basal-type breast cancer cells.
3.4. TGF-b affects the production of cytokines and chemokines that are regulated by ZEB1 ZEB1 functions as a transcriptional activator by binding to Smad1, Smad2, and Smad3 and supports their transcriptional activities (Postigo, 2003;Postigo et al., 2003), while ZEB2 inhibits TGF-b signaling (Verschueren et al., 1999). Based on the similar functions of ZEB1 and ZEB2 on the expression of inflammatory cytokines, we then investigated whether ZEB1 and ZEB2 regulate the expression of inflammatory response genes induced by TGF-b stimulation. Of the basal-type breast cancer cell lines, MDA-231-D and Hs578T expressed ZEB1 and ZEB2, while HCC1954-Luc cells showed only small amounts of ZEB1 and ZEB2 proteins ( Fig. S2A; Horiguchi et al., 2012). In agreement with the immunoblot data, quantitative evaluation using RNA-seq data in MDA-231-D cells suggested that the expression of ZEB2 was lower than that of ZEB1 (Fig. 4A). Of note, ZEB1 and ZEB2 expression was not upregulated by TGF-b in MDA-231-D cells (Fig. S2B).
We found that TGF-b increased the expression of IL6 and IL8, while LY364947, which is a TGF-b type I receptor kinase inhibitor, decreased their baseline  Fig. 1E and F. ZEB1-bound genes were selected for evaluation. (A) Gene sets in MDA-231-D cells that were differentially expressed after treatment with ZEB1/2 siRNA (siZEB1-1 and siZEB2-1). The most enriched MSigDB hallmark signatures that had normalized enrichment scores (NES) > 1.5 or < À1.5 are listed. (B) Enrichment plot of the HALLMARK_INFLAMMATORY_RESPONSE gene set. ZEB1-bound genes (vertical black bars in the bottom panel) were enriched toward the left; ZEB1 and ZEB2 functioned as transcriptional activators. siZEB1/2-1, siZEB1-1 + siZEB2-1. (C) Venn diagram showing the overlap of genes that were downregulated by ZEB1/2-1 siRNA in the presence and absence of TGF-b. Genes with maximum FPKM values ≥ 10 in the four samples were selected for further evaluation, and those with fold changes (FPKM (siZEB1/2) + 0.01)/(FPKM(siNC) + 0.01) ≤ 0.5 were defined as downregulated genes. (D) Visualization of the ZEB1 ChIP-seq data tracks at the IL6, IL8, and CSF2 gene loci. Data are shown as in Fig. 1A. (E) Antibody array analysis of proteins in conditioned medium obtained from serum-starved MDA-231-D cells. ZEB1-bound genes and genes regulated by ZEB1/2 siRNA according to ChIP-seq and RNAseq analysis are shown in a heat map in the order of downregulation by siRNA (left panel). Conditioned medium was collected from MDA-231-D cells treated with ZEB1/2-1 siRNA and analyzed using an antibody array. The effects of ZEB1/2 siRNA on cytokine secretion are shown in the right panel. Genes were selected according to the following criteria: Maximal expression values (FPKM) were ≥ 10, and there was a fold change ≥ 1.5 in response to ZEB1/2 siRNA. Note that the amount of IL-6 protein in this cell line was very high and could not be quantitatively evaluated in this array. n.d.: not detected. (F) The effect of ZEB1/2 siRNA on the secretion of IL-6 and IL-8 by MDA-231-D cells as determined by ELISA. Error bars indicate the S.D. *P < 0.05. (G) The effect of ZEB1/2 siRNA on the secretion of IL-6 and IL-8 by Hs578T cells as determined by ELISA. Error bars indicate the S.D. *P < 0.05. In (F) and (G), siZEB1/2-1, siZEB1-1 + siZEB2-1; siZEB1/2-2, siZEB1-2 + siZEB2-2. expression levels (Fig. 4B). The effect of TGF-b on the expression of IL6 and IL8 was not observed in the absence of ZEB1 and ZEB2. A similar tendency was observed regarding the TGF-b-induced expression of CSF2 and CXCL5 (Fig. S2B). Significant induction of IL1B by TGF-b could not be seen; however, the effect of LY364947 on IL1B expression in the control siRNA-transfected cells was inhibited by ZEB1/2 siRNA. Consistent results were not obtained regarding the effect of ZEB1/2 siRNA on TGF-b-induced CXCL1 expression.
We then exogenously expressed ZEB1 and ZEB2 in HCC1954-Luc cells by transfection of expression plasmids (Fig. S2C), and found that ZEB1, and to a lesser extent ZEB2, increased the IL6 and IL8 expression. We also established HCC-1954-Luc cells stably  expressing ZEB1 (HCC-ZEB1; Fig. S2D). Analysis of secreted proteins from HCC-ZEB1 cells using an antibody array revealed that ZEB1 induced the production of IL-6 and CSF2, but not IL-8, in this cell line (Fig. 4C). IL6 expression in HCC1954-Luc cells was significantly increased by exogenous ZEB1 expression, similar to its effect in MDA-231-D cells (Fig. 4D). In addition, TGF-b increased the expression of IL6 in the presence of ZEB1 in HCC-1954-Luc cells (Fig. 4D). These results suggested that ZEB1 and possibly ZEB2 played a central role in IL6 and IL8 transcription independent of TGF-b signaling, and simultaneously, they were required for the induction of IL6 and IL8 by TGF-b. These results also suggested that production of inflammatory cytokines by ZEB1 and ZEB2 is induced in a context-dependent manner and that expression of ZEB1 and ZEB2 is not sufficient to induce some of the target proteins.

Correlation of the expression of IL-6 and IL-8 with ZEB1 expression in various types of cancer
To evaluate the regulatory functions of ZEB1 in other types of breast cancer cells and cancer tissues, we obtained expression microarray data from the Cancer Cell Line Encyclopedia (CCLE; Barretina et al., 2012) and analyzed the relationship of the expression of ZEB1 with that of inflammatory cytokines, that is, IL6 and IL8, in breast cancer cells. We found that a small subset of the cell lines was 'double-positive' for the expression of ZEB1 and IL6 and/or IL8 (Fig. 5A). In contrast, the expression of ZEB2 was generally low in breast cancer cells compared to THP-1 cells, an acute myelogenous leukemia cell line with known ZEB2 function (Li et al., 2017), and showed weaker correlation with IL6 and IL8 expression than that of ZEB1 (Fig. S3A). We also examined the relationship between ZEB1 and IL-6 in other types of cancers using a tissue array of 448 cancer tissues from multiple organs and found other double-positive cancers ( Fig. 5B and Fig. S3B,C). These findings suggested that the transcriptional activation of IL6 by ZEB1 is  Fig. S3. The top panel shows a case that was positive for both nuclear ZEB1 staining (green) and whole-cell IL-6 staining (red); scores = 4. The bottom panel shows a case that was negative for both ZEB1 and IL-6; scores = 1. The samples were counterstained with DAPI to show cell density in the spot. Original magnification: 20 9. (C) Kaplan-Meier survival curves of breast and lung cancer patients obtained from a public meta-analysis database and Kaplan-Meier plotter (Gyorffy et al., 2010(Gyorffy et al., , 2013. The probability of overall survival of patients as split by median IL6 and IL8 expression is shown. Red: IL6 and IL8 high expression group; black: IL6 and IL8 low expression group. not restricted to breast cancer cells and tissues and found in a subset of other types of ZEB1-expressing tumors. We then examined the prognostic importance of IL6 and IL8, which are regulated by ZEB1 using publically available meta-analysis database (Kaplan-Meier plotter, http://kmplot.com/analysis/; Fig. 5C). High expression of IL6 and IL8 was significantly correlated with poor survival in breast cancer and lung cancer, which is consistent with a previous report (Hartman et al., 2013).

ZEB1 promotes HCC1954-Luc cell proliferation and tumor growth
Next, we used HCC-GFP and HCC-ZEB1 cells to investigate the biological importance of ZEB1-induced cytokines. Phosphorylation of STAT3 was observed in the absence of IL-6 expression, suggesting the activation of STAT3 by other pathways, and was moderately enhanced in ZEB1-expressing HCC1954-Luc cells (Fig. 6A). ZEB1 promoted the proliferation of HCC1954-Luc cells (Fig. 6B), and this effect was attenuated by IL6 siRNA (Fig. 6C). Conditioned medium from HCC-ZEB1 cell culture enhanced the proliferation of the parental HCC1954-Luc cells (Fig. 6D). These results suggested that secreted factors, especially IL-6, contributed to the growth of ZEB1-expressing HCC-ZEB1 cells in an autocrine manner. We also found that ZEB1 enhanced tumor growth when HCC1954-Luc cells were grafted into nude mice (Fig. 6E).
IL-6, IL-8, and some other ZEB1-regulated secretory proteins are related to the senescence-associated secretory phenotype (Perez-Mancera et al., 2014). However, the induction of cellular senescence in MDA-231-D cells by irradiation or doxorubicin treatment, followed by SA-bGal staining, failed to show a significant correlation between ZEB1 expression and cellular senescence (Fig. 6F and G).

Conditioned medium from ZEB1-expressing MDA-231-D cells increases fibroblast growth in a paracrine manner
We further studied the functional relationship between cancer cells and fibroblasts and found that conditioned media from HCC-ZEB1 and MDA-231-D cells upregulated phosphorylated STAT3 in IMR90 and WI38 human fibroblasts (Fig. 7A). Although conditioned medium from HCC-ZEB1 cells did not increase the growth of IMR90 and WI38 cells (Fig. 7B), conditioned medium from ZEB1/2-silenced MDA-231-D cells significantly decreased the growth of IMR90 and WI38 cells (Fig. 7C). Taken together, these results suggested that ZEB1 expressed in cancer cells contributes to the growth of fibroblasts in paracrine and contextdependent manners.

ZEB1 induces myeloid cells that express markers characteristic of polymorphonuclear myeloid-derived suppressor cells (MDSCs) within tumors
Because ZEB1 regulated the production of inflammatory cytokines, including IL-6, IL-8, IL-1b, CXCL1, and CXCL5, we evaluated the paracrine effect of ZEB1-regulated secretory proteins in terms of their antitumor immune functions. MDSCs comprise heterogeneous cell populations that have the potential for immunosuppressive activity. MDSCs induce immune tolerance in the tumor microenvironment by suppressing cytotoxic T-lymphocyte activity (Gabrilovich et al., 2012;Ugel et al., 2015). They are defined as CD11b + Gr-1 + cells in mouse and are categorized as polymorphonuclear (PMN)-MDSCs and monocytic (Mo)-MDSCs. PMN-MDSCs are defined as CD11b + Ly6C low Ly6G + cells that express high levels of arginase 1. In contrast, Mo-MDSCs are defined as CD11b + Ly6C high Ly6G À cells that express high levels of Nos2 (iNOS; Talmadge and Gabrilovich, 2013). The number of MDSCs in tumor tissue is associated with disease stage in patients with breast cancer (Markowitz et al., 2013). To examine the effect of ZEB1-regulated secretory proteins on the accumulation and maturation of MDSCs in vivo, we established murine breast cancer 4T1 cells that overexpressed mouse ZEB1 (4T1-ZEB1). We confirmed the increased expression of Il6, Il8, and Il1b mRNA in 4T1-ZEB1 cells versus 4T1-GFP cells (Fig. 8A).
STAT3 has been reported to be related to the accumulation and expansion of MDSCs (Condamine and Gabrilovich, 2011). To determine whether ZEB1-regulated secretory proteins affect MDSC development, we focused on STAT3 phosphorylation in mouse bone marrow (BM) cells using conditioned medium from HCC-ZEB1 cells. We found that STAT3 phosphorylation was enhanced by conditioned medium from HCC-ZEB1 cells (Fig. 8B).
4T1-ZEB1 cells were then used in a syngeneic tumor model in immunocompetent BALB/c mice. Tumor size and lung metastasis were not significantly different between 4T1-ZEB1 cells and 4T1-GFP cells, possibly because of the highly aggressive nature of 4T1 cells. We then evaluated the accumulation and maturation of MDSCs in the tumor sites by flow cytometry and found that the percentage of Gr-1-positive cells was increased in the 4T1-ZEB1 tumors compared to 4T1-GFP tumors (Fig. 8C). To characterize the increased MDSCs in more detail, we also evaluated the expression of Ly6C and Ly6G on the cells obtained from tumors to identify PMN-MDSCs (CD11b + Ly6C low (G) MDA-231-D cells were transfected with siRNA as indicated and cellular senescence was induced by 16-Gy irradiation or 75 lM doxorubicin treatment. In terms of irradiation conditions, siRNA was transfected again 2 days after senescence induction. Cells were stained for SA-bGal 6 days after treatment. Three images were taken for each condition except for one of the controls (two images), and the percentage of SA-bGal cells was calculated.
Ly6G + ) and Mo-MDSCs (CD11b + Ly6C high Ly6G À ; Fig. 8D). We found that the percentage of PMN-MDSCs in the CD11b + FSC + CD45 + population of the cells was significantly increased in 4T1-ZEB1 tumors, while the percentage of Mo-MDSCs was not (Fig. 8E). Together with the findings from a report showing the immunosuppressive activity of Ly6G + PMN-MDSCs from 4T1 tumor-bearing mice (Kim et al., 2014), these findings suggested that ZEB1 expression in tumor cells increases the PMN-MDSC population in CD11b + cells.

Discussion
Expanding and differential roles of ZEB1 and ZEB2 have been identified in various types of cancers in recent years (Chaffer et al., 2013;Krebs et al., 2017;Mejlvang et al., 2007;Morel et al., 2017;Si et al., 2015;Spaderna et al., 2008). ZEB1 is expressed primarily in nonepithelial cells, and a genome-wide analysis of ZEB1-binding regions performed in an adipogenesis model clearly shows its physiological importance (Gubelmann et al., 2014). Therefore, it is not surprising that the main consequence of the ectopic expression of ZEB1 differs greatly between cancer cell types. This is because transcription factor binding regions are affected by the epigenetic environment in the cells, which is dynamically regulated by TGF-b and other extracellular stimuli (Ostuni et al., 2013). The observation that ZEB1 has differential effects on the expression of genes that are either downregulated or upregulated by TGF-b in MDA-231-D cells also supports the importance of cellular context. In addition, Hs578T and Panc-1 cells showed very similar fibroblasts after the addition of conditioned media. Cells were collected 20 min after stimulation, and the cell lysates were subjected to immunoblot analysis as indicated. CM, conditioned medium; rhIL-6, recombinant human IL-6 (100 ngÁmL À1 ). (B, C) The effect of conditioned medium from cultured breast cancer cells on fibroblast cell growth. The number of IMR90 and WI38 cells was counted after 72 h of incubation using the trypan blue exclusion assay. Conditioned media were obtained from ZEB1-overexpressing HCC1954-Luc cells and control cells (HCC-GFP) (B), or from ZEB1/2-silenced MDA-231-D cells and control cells (C). siZEB1/2-1, siZEB1-1 + siZEB2-1; siZEB1/2-2, siZEB1-2 + siZEB2-2. Conditioned medium was replenished every 24 h. In panel (C), data are shown as the means of three (IMR90) or four (WI38) biological replicates. CM, conditioned medium; n.s., not significant; *P < 0.05. ZEB1 binding profile, which likely reflected both similar binding strength and regulatory mechanism of ZEB1 binding.
Thus, apart from its general role as an EMT inducer, ZEB1 and ZEB2 may have multiple functions that will be elucidated by analyses in specific cancer types in the future.
Inflammatory cytokines play crucial roles in various aspects of cancer, including cancer development, progression, treatment resistance, and prognosis. IL-1b promotes colon cancer cell stemness and invasiveness (Li et al., 2012). IL-8 is also associated with cancer stem cell-like properties, and its expression correlates with poor prognosis in human pancreatic cancer (Chen et al., 2014). Of these inflammatory cytokines, IL-6 is reported to play especially important roles in the development of lung and breast cancers (Gao et al., 2007;Sansone et al., 2007). With regard to the effect of these cytokines on the EMT, previous reports focused mainly on the regulation of EMT-related downstream factors, with ZEB1 reported as a target gene of certain inflammatory cytokines Peinado et al., 2007). A recent report suggested that inflammation induces disseminated, dormant cancer cells to form metastatic tumors through functions of ZEB1 (De Cock et al., 2016). It is also reported that ZEB2 is induced by inflammation (Katoh and Katoh, 2009). However, the induction of IL-6 and IL-8 by ZEB1 and ZEB2, an inverse relationship that was revealed in this study, has not been investigated in detail.
It is widely accepted that cancer cells show EMTlike phenotypes due to the production of various cytokines. For example, elevated expression of IL6 and other chemoresistance-related genes accompanies the EMT in a mouse breast cancer model (Fischer et al., 2015). Our finding that ZEB1 and ZEB2 are inducers of inflammatory cytokines is supported by a report by Suarez-Carmona et al. (2015) that Slug, Snail, and other EMT-related transcription factors regulate the production of soluble factors, such as IL-8, IL-6, sICAM-1, PAI-1, and GM-CSF/CSF2. Taken together, these data indicate that inflammatory cytokines induce EMT-related transcription factors and vice versa in certain cancer cells to enhance tumor progression. These processes can be targeted by molecular therapies.
Our findings indicated that ZEB1 and partly ZEB2 regulated the characteristic inflammatory phenotype of breast cancer cells, in part through IL-6 and IL-8. The ZEB1-regulated inflammatory phenotype identified in this study was characterized by enhanced breast cancer cell growth, fibroblast proliferation, and PMN-MDSC accumulation, although each of these was observed only in certain cell types and was context dependent. The correlation between IL-6 production and cancer proliferation has been reported in various types of cancers, including lung, prostate, and breast cancers (Gao et al., 2007;Giri et al., 2001;Sansone et al., 2007;Yamaji et al., 2004). IL-6 secreted by cancer cells activates STAT3, and induces downstream events, including cancer cell proliferation and apoptosis inhibition. Our results support these reports and revealed that IL-6, as a central soluble factor that was positively regulated by ZEB1, constantly activated STAT3 in HCC1954-Luc cells and exhibited cell-proliferating potency both in cell culture and in vivo. IL-6, which was regulated by ZEB1 and ZEB2 in breast cancer cells, also induced the proliferation of fibroblasts in a context-dependent manner, suggesting that ZEB1 and ZEB2 regulate the production and function of cancerassociated fibroblasts, which are the main constituents of tumor microenvironments, thereby enhancing tumor growth (Kalluri and Zeisberg, 2006;Xing et al., 2010).
In terms of MDSCs, IL-6 strongly induces their accumulation and maturation. IL-6 is reported to restore the impaired accumulation of MDSCs and tumor progression in tumor-bearing mice lacking IL-1b or indoleamine 2,3-dioxygenase (IDO; Bunt et al., 2007;Smith et al., 2012). Activated STAT3 is the main regulator of IL-6 in MDSCs, inducing cell survival, cell proliferation (Xin et al., 2009), and immunosuppressive activity (Kujawski et al., 2008). IL-8 also enhances PMN-MDSC infiltration into tumor tissues (Kumar et al., 2016;Sandhu et al., 2000). Furthermore, IL-1b is reported to induce the accumulation of MDSCs (Bunt et al., 2007;Elkabets et al., 2010). In addition to IL-6, IL-8, and IL-1b, another ZEB1-regulated cytokine, G-CSF/CSF3, also enhances the accumulation of MDSCs (Talmadge and Gabrilovich, 2013). ZEB1 also regulates the expression of the chemokines CXCL1 and CXCL5, which are the ligands for the CXCR2 receptor, and increases the infiltration of PMN-MDSCs (Acharyya et al., 2012;Katoh et al., 2013;Toh et al., 2011). Although the present study showed no additional effect of ZEB1 on tumor progression or on the metastasis of 4T1 cells, which may be due to the aggressive nature of the parental 4T1 cells, these observations suggest that inflammatory cytokines induced by ZEB1 play critical roles in the progression of cancer in a context-dependent manner.
In basal-type breast cancer cells, high ZEB1 expression was observed even in the absence of TGF-b (Horiguchi et al., 2012). Of note, in mammary carcinomas that do not express the TGF-b type II receptor, CXCL5 expression is increased, resulting in the recruitment of MDSCs (Yang et al., 2008). Thus, although TGF-b induces the expression of ZEB1 and ZEB2, which function as key transcription factors in the induction of the EMT in various types of epithelial cells, TGF-b signaling and ZEB1 and ZEB2 act in opposite ways in cancer cells in some contexts, including in the regulation of MDSCs, and they play distinct roles in cancer progression.
In conclusion, ZEB1 and ZEB2, through the induction of various cytokines, including IL-6 and IL-8, facilitate tumor growth both in autocrine and in paracrine manners in basal-type breast cancer cells. Future studies will focus on evaluating these extracellular proteins as potential anticancer targets to inhibit the progression of cancer.
This work was also supported by the Princess Takamatsu Cancer Research Fund (14-24609 to DK). YT is supported by the Yoshida Scholarship Foundation and by the Graduate Program for Leaders in Life Innovation from the Japan Society for the Promotion of Science.

Availability of data and materials
The raw ChIP-seq and RNA-seq data are available at GEO (GSE89206). Epithelial-to-mesenchymal transition is dispensable for metastasis but induces chemoresistance in pancreatic cancer. Nature 527, 525-530.

Supporting information
Additional Supporting Information may be found online in the supporting information tab for this article: Fig. S1. Specificity of ZEB1 and ZEB2 antibodies and the result of GSEA analysis showing the effect of ZEB1 or ZEB2 siRNA in MDA-231-D cells. Fig. S2. The effect of ZEB1/2 siRNA on the expression of inflammatory response genes. Fig. S3. Tissue array analysis of ZEB1 and IL-6 expression using fluorescent immunohistochemistry. Fig. S4. Efficiency of IL6 siRNA and the amount of ZEB1 protein in 4T1 breast cancer cells. Table S1. Sequences of the primers used for RT-PCR. Table S2. A list of ZEB1-bound inflammatory response genes.