E74-Like Factor 3 Is a Key Regulator of Epithelial Integrity and Immune Response Genes in Biliary Tract Cancer

E74-like factor 3 (ELF3) belongs to the E26 transformation-specific (ETS) family of transcription factors, which share a conserved ETS DNA-binding domain (1, 2), and either activate or repress transcription by binding to the core consensus sequence GGAA/T within the promoter and enhancer regions of target genes (3). ELF3 is an epithelium-specific ETS transcription factor (4), and is strongly expressed in the digestive tract, urinary bladder, uterus, tonsil, and bronchus (5). It plays crucial roles in several biological processes, including the regulation of physiological development, differentiation, and homeostasis in epithelial tissues (6–8).

We previously identified ELF3 with loss-of-function mutations (frame-shift, indel, or nonsense) as a novel driver gene in 8/242 (3.3%) of biliary tract cancers (BTC) and in 19/172 (11.0%) of ampullary carcinoma, implying that it has an inherent tumor-suppressor role (9, 10). These ELF3 mutations occur with high allele frequencies, and some have been found in duodenal adenoma with intraepithelial neoplasia, suggesting that they are likely founder mutations (10–12). More recently, inactivating ELF3 mutations have been reported in other cancer types, particularly cancers of the bladder, cervix, and stomach (13–15). Conversely, the amplification and overexpression of ELF3 have been identified in breast, prostate, colorectal, and lung cancers (16–20). Hence, further studies are needed to resolve this apparent contradiction and to clarify the tumor-suppressive or oncogenic roles of ELF3 in different organ systems.

In the present study, we generated ELF3-knockout and ELF3 conditional knockdown bile duct epithelial cells to determine the contribution of the loss of ELF3 function to tumorigenesis. We then performed a genome-wide expression profiling analysis of these cells, as well as ELF3-overexpressing cells, and identified several previously unknown direct target genes of ELF3 that appear to be implicated in biliary duct carcinogenesis. We further investigated the role of ELF3 and these putative ELF3 target genes in tumor formation in a BTC xenograft model.

This study was approved by the institutional review board of each participating institute (National Cancer Center, 2002–69, 2012–071, 2015–108; Osaka University, 755).

An immortalized normal epithelial cell line of common bile duct origin (HBDEC2–3H10) was established and maintained as described previously (10, 21). Briefly, a normal common bile duct was obtained from a 70-year-old male patient undergoing surgery for a disease other than bile duct cancer at National Cancer Center Hospital. The Ethics Committees of the National Cancer Center approved this study (2002–69 and 2015–108) and the subject gave his written informed consent for participation. The bile duct tissue, cut into small pieces, was frozen in TC-Protector Cell Freezing Medium (DS Pharma Biomedical) and digested with a Tumor Dissociation Kit, Human MACS (Miltenyi Biotech) according to the manufacturer's instruction. The details of the establishment of HBDEC2–3H10 cells are in the Supplementary Experimental Procedures. An ELF3 gene knockout clone was obtained using the CRISPR/Cas9-system, with support from Takara Bio, Inc. HBDEC2–3H10 cells were electroporated with oligonucleotides carrying gRNA silencing ELF3 cloned in a pRGEN-Cas9_CMV plasmid (Takara Bio, Inc.). After electroporation, single cells were seeded in 96-well plates by dilution and expanded. For inducible ELF3 gene knockdown, cells were first transduced with a retrovirus, pCLXIN-TetOnADV, and a lentivirus, CSII-CMV-tTS, and then transduced with the lentiviruses CSII(ins)-TRE-Tight-puro-ELF3miR1 and -ELF3miR2, expressing microRNA-155-based conditional RNA interference targeting the human ELF3 gene (miR#1; 5′-CCTGTGAGTAGCAACATTT-3′ and miR#2; 5′-AGGTTCTGCTGGATCAGCT-3′) and EGFPmiR, targeting EGFP as a control (EGFP-miR; 5′-ACAAGCTGGTACAACTACA-3′). An inducible ELF3-expressing cell line (HBDEC2 ELF3 Tet-ON cells) was generated by infection with the retroviral vector pRetroX-TetOne-Puro-ELF3-flag, expressing ELF3 fused to a flag-tag. Conditional tumorigenic transformed HBDEC2 ELF3-knockout cells, using defined oncogenes [human papillomavirus type 16 E6 and E7 genes, MYC ^T58A, and HRAS ^G12V (EMR); HBDEC2^ELF3−/− EMR cells], were generated under the control of the doxycycline (DOX)-responsive promoter (HBDEC2^ELF3−/− EMR; tetOFF) as described previously (21). HBDEC2^ELF3−/− EMR cells were infected with a retroviral vector, PQCXIP–ELF3–ERT2, expressing ELF3 fused with a mutant estrogen receptor ligand-binding domain (ERT2) to generate HBDEC2^ELF3−/− EMR ELF3–ERT2 cells, which can be functionally activated by tamoxifen. All cells were used for no more than 10 passages after electroporation or infection and routinely tested for Mycoplasma contamination by PCR (the latest test date: February 18, 2019).

Cell lysates were prepared and analyzed as described previously (10). The antibodies used are listed in the Supplementary Experimental Procedures. Total RNA was extracted using the RNeasy Mini kit (74104, Qiagen). Quantitative RT-PCR was performed with FS Essential DNA Green MMx (06402712001, Roche Diagnostics), using primers for ZEB1, ZEB2, TWIST1, VIM, KRT19, OCLN, CGN, DSC2, CXADR, ALOX5, CXCL16, or GAPDH (see Supplementary Experimental Procedures). The values obtained from quantitative RT-PCR were normalized to GAPDH as an internal control. Immunofluorescence was performed using rabbit anti-vimentin primary antibody (ab92547, 1:500, Abcam) with goat anti-rabbit Alexa Fluor 488 secondary antibodies (A11034, Thermo Fisher Scientific). Stained sections were viewed and photographed using a microscope (BZ-9000, KEYENCE).

Animal studies were carried out according to the Guideline for Animal Experiments in National Cancer Center and Osaka university, which meet the ethical standards required by the law and the guidelines for use of experimental animals in Japan, and approved by the Committee for Ethics in Animal Experimentation of the National Cancer Center (T15–004) and Osaka university (30–084–003). A 100 μL volume of 1 × 10⁷ cells (ELF3-knockout or wild-type cells) in a 1:1 mixture of Matrigel (354234, BD Biosciences) was subcutaneously injected into 6 weeks old female BALB/c nude mice (Charles River Laboratories). Three months after inoculation, tissue sections were fixed with formalin and embedded in paraffin. Five mice used in each group. A 50 μL volume of 1 × 10⁶ HBDEC2^ELF3−/− EMR ELF3–ERT2 cells in a 1:1 mixture of Matrigel (BD Biosciences) was subcutaneously injected into 7-week-old female BALB/c nude mice (Charles River Laboratories). Mice were housed with drinking water supplemented with or without tamoxifen (AstraZeneca; 0.1 mg/mL in 5% sucrose). Twenty-one days after inoculation, tissue sections were fixed with formalin and embedded in paraffin. Three mice were used in each group. IHC was performed using rabbit anti-vimentin primary antibody (1:200, Abcam) with EnVision+ System-HRP labeled polymer anti-Rabbit (K4002, Dako-Agilent).

Gene expression microarray analyses were carried out using the SurePrint G3 Human GE 8 × 60K v3 with support from Takara Bio. Functional enrichment analyses were performed using DAVID Bioinformatics Resources 6.8, GSEA and Kyoto Encyclopedia of Genes and Genomes (KEGG). The original data are available in the Gene Expression Omnibus (GEO) database (GSE148108).

Cell invasion/migration assays were performed using a BD BioCoat Matrigel invasion chamber and control inserts (24-well, BD Biosciences) and analyzed as previously described (10).

Chromatin immunoprecipitation sequencing (ChIP-Seq) was carried out using a previously described protocol (22). Briefly, 5 × 10⁷ fixed cells were lysed to prepare nuclear extracts. Chromatin was sheared by sonication, and the lysates were incubated overnight at 4°C with Dynabeads protein G (10004D, Thermo Fisher Scientific) coupled with 100 μg anti-ELF3 antibody (HPA003316, Sigma-Aldrich). After immunoprecipitation, beads were recovered using a magnet and washed; next, chromatin was eluted, and cross-links were reverted overnight at 65°C. DNA was purified and quantified with the Agilent Bioanalyzer (Agilent Technologies). Sequencing was performed on the HiSeq 3000 platform (Illumina) as single-end 36-bp reads. Sequence reads were mapped to the human genome (hg19) using BWA software (23) with the default parameters. Peak calling for ELF3 was performed using MACS1.4 (24) with the following parameters: -g hs -q 0.01 -c control.bam. The input DNA sequence was used as a control. The original data are available in the GEO database (GSE156165).

Fixed cells were lysed to prepare nuclear extracts according to the manufacturer's instructions (SimpleChIP, #9005, Cell Signaling Technology). Chromatin immunoprecipitation (ChiP) was performed with anti-ELF3 antibody (HPA003316, Sigma-Aldrich) and IgG control antibody. Samples were analyzed by real-time PCR using primers for ZEB2, CGN, ALOX5, or CXCL16 (see Supplementary Experimental Procedures).

Human peripheral blood primary natural killer (NK) cells and CD8⁺ T cells were purchased from STEMCELL Technologies (NK cells: 70036, CD8⁺ T cells: 70027). For Transwell assays, NK cells (1 × 10⁵) and CD8⁺ T cells (1 × 10⁵) were added to 150 μL serum-free medium in the upper chambers (5.0-μm pore size, 3421, Corning Inc.). The lower chambers contained 500 μL cell-free culture supernatant from HBDEC2 ELF3 Tet-ON cells treated with doxycycline (DOX, 1 μg/mL for 48 hours, 631311, Clontech Laboratories, Inc.). After incubation for 2 hours at 37°C in a humidified environment containing 5% CO₂, migrated NK cells and CD8⁺ T cells in the lower chamber were determined by counting with a hematocytometer.

Statistical analyses were carried out using GraphPad Prism version 5.0a (GraphPad). The statistical significance of differences was assessed with one-way ANOVA followed by the Bonferroni test, with the student t test or the Mann–Whitney test. Data from at least two or three independent experiments are presented as the mean ± SD or mean ± SEM. A P value of <0.05 was considered statistically significant.

To investigate the role of ELF3 in the normal bile duct epithelial cell line HBDEC2, we generated knockout clones using CRISPR/Cas9-mediated genome editing in HBDEC2 cells. To select a proper control for the CRISPR experiment, we equalized the passage number of wild-type cells and ELF3-knockout cells based on the estimated population doublings during cloning for 32 days. Furthermore, we selected a clone with an epithelial-like morphology (KO#1) from six ELF3-knockout clones (KO#1–6), some of which showed mesenchymal morphology (Supplementary Figs. S1 and S2A). At first, we confirmed that the selected clone lacked ELF3 expression (Fig. 1A). Gene expression profiling, with wild-type cells as a control, identified 2,666 genes whose expression was changed more than 2-fold in the knockout cells. Gene ontology (GO) analysis showed the altered genes in ELF3-knockout #1 cells were predominantly related to cell migration, along with leukocyte migration, extracellular matrix (ECM) organization, inflammatory response, and cell adhesion (Fig. 1B). The top-ranking categories in KEGG pathway analysis of the genes downregulated in ELF3-knockout #1 cells were leukocyte transendothelial migration, cell adhesion molecules, and tight junctions (Supplementary Table S1). Furthermore, gene set enrichment analysis (GSEA) showed that the epithelial–mesenchymal transition (EMT)-related gene signature was significantly enriched in cells with ELF3 deletion (Fig. 1C). Quantitative RT-PCR (qPCR) analysis and western blotting confirmed that the expression levels of EMT-related transcription factors, including ZEB1 (encoding ZEB1), ZEB2 (encoding ZEB2), and TWIST1 (encoding TWIST1), were higher in ELF3-knockout #1 cells than in wild-type cells (Fig. 1D and E). In addition, ELF3 knockout led to significant increases in the expression of the mesenchymal marker VIM (encoding vimentin; Fig. 1D–F), as well as a reduction in the expression of the epithelial marker KRT19 (encoding cytokeratin 19; Fig. 1D and E). More of these mesenchymal phenotypes were observed ELF3-knockout clones than in ELF3 wild-type clones (Supplementary Fig. S2B). Consistent with these findings, the motility and invasive activity of ELF3-knockout (#1 and #2) cells were significantly enhanced compared with those of wild-type cells (Fig. 1G; Supplementary Fig. S3). We assessed the effects of ELF3 deletion on in vivo tumorigenicity of HBDEC2 cells. ELF3-knockout #1 and wild-type cells were implanted subcutaneously in nude mice. After 3 months, ELF3-knockout #1 cells generate glands with histological atypia, such as nuclear pseudostratification, variations in nuclear size, and focal loss of cellular polarity, without forming tumors (Fig. 1H). In IHC, the glands of ELF3-knockout#1 HBDEC2 cells were partially positive for anti-Ki67 and anti-vimentin antibodies, which are markers of proliferation and mesenchymal cells, respectively (Supplementary Fig. S4).

To rule out CRISPR/Cas9-mediated off-target effects, we next generated DOX-dependent inducible ELF3-knockdown cells using microRNA-based conditional RNA interference (miR). The inducible expression of miR targeting the human ELF3 gene (miR#1 and miR#2) was performed in HBDEC2 cells. Western blot analyses demonstrated that miR#1 and miR#2, but not the negative control (miR-EGFP), clearly suppressed the expression of ELF3 in DOX-treated cells (Fig. 2A). Functionally, the conditional knockdown of ELF3 significantly increased cell migration and invasion without affecting cell proliferation compared with those observed in DOX-untreated cells (Fig. 2B; Supplementary Fig. S5A). The expression levels of EMT-related transcription factors and vimentin were upregulated in miR#2 ELF3-knockdown cells, with a marked effect on cell motility (Fig. 2C and D). Gene expression profiling revealed that cell adhesion and ECM were the predominant deregulated pathways in ELF3 conditional knockdown cells (Fig. 2E).

qPCR analysis revealed that the expression levels of OCLN (encoding occludin), CGN (encoding cingulin) and CXADR (encoding coxsackie virus and adenovirus receptor), which are related to tight-junction proteins, and DSC2 (encoding desmocollin 2), a desmosome component, were significantly decreased in ELF3-knockout #1 cells compared with those in wild-type cells (Fig. 3A). Ultrastructural examination by transmission electron microscopy showed an increased abundance of microvilli in monolayer cultures of ELF3-deficient cells, with fewer cell–cell junctions compared with wild-type cells (Fig. 3B). These data suggest that the loss of ELF3 may cause cytoskeletal changes that affect cell shape, intracellular trafficking, and cell migration. We next applied a three-dimensional approach to ELF3-knockout #1 cells using Matrigel, which contains basement membrane ECM proteins. In this three-dimensional culture, ELF3-knockout #1 cells exhibited few cell–cell junctions compared with wild-type cells, as assessed by electron microscopy (Fig. 3C).

To examine whether the overexpression of ELF3 could reverse the migratory mesenchymal properties of these cells, we generated DOX-inducible ELF3-overexpressing cells from ELF3-knockout #1 HBDEC2 cells. The overexpression of ELF3 was confirmed by western blotting (Fig. 4A). We observed a significant decrease in cellular migration/invasion and a slight decrease in cell proliferation compared with those observed in control DOX-untreated cells (Fig. 4B; Supplementary Fig. S5B). Western blot and qPCR analysis showed that the expression of EMT-related transcription factors in ELF3-overexpressing cells was clearly decreased compared with that in control cells (Fig. 4A and C). We further found that ELF3 overexpression induced the upregulation of cell adhesion-related genes (Fig. 4C).

Given that ELF3 is a transcription factor, we explored the target genes of ELF3 using ELF3 ChIP-Seq with ELF3-overexpressing cells. The number of uniquely mapped reads identified by ChIP-Seq was 109.9 mol/L for the ELF3-binding site and 59.7 mol/L for the control input DNA. These read data were evaluated with MACS 1.4 (24), and 13,960 peaks were identified throughout the genome with an FDR of <0.1. First, we focused on genes related to EMT or cell adhesion. By investigating the ELF3 peak distribution among these genes and the flanking regions, we found peaks on the ZEB2 gene at the 3′-UTR (untranslated region), and on the CGN gene at the first intron (Fig. 4D). These ELF3-specific peaks on ZEB2 and CGN were verified by ChIP coupled with qPCR analysis, confirming that ZEB2 and CGN were direct targets of ELF3 in our experiment (Fig. 4D). Through the combined use of ChIP-Seq datasets and microarray datasets, we next tried to identify genes that were directly regulated by ELF3 in a genome-wide manner. ELF3-binding peaks were annotated using RefSeq, and a peak was considered to be associated with transcriptional regulation when it was found on the gene body or in the flanking 50-kb region. Analysis with these parameters identified 17,925 genes as potential candidates. Next, to identify genes with expression levels that were intrinsically regulated by ELF3, the candidate set was compared with a set of differentially expressed genes generated from the microarray data. Venn diagrams showed the overlap between the genes identified in ELF3 ChIP-Seq and genes with altered expression in the microarray (ELF3-knockout #1 vs. wild-type: 1,467 genes, P < 0.05, fold change ≧ 2; Supplementary Fig. S6A). Similarly, Supplementary Fig. S6B and S6C show the overlap between genes from ELF3 ChIP-Seq and genes from microarray analyses of miR#2 ELF3-knockdown cells (3,013 genes, P < 0.05, fold change ≧ 1.5) and ELF3-overexpressing cells (1,770 genes, P < 0.05, fold change ≧ 1.5). Finally, consistent ELF3-dependent changes in expression in all three experiments were observed for 53 genes (Supplementary Fig. S6D, group A), of which 43 were upregulated and 10 were downregulated (Supplementary Table S2). Among these 53 genes, GO analysis revealed that defense response, inflammatory response, and positive regulation of cell migration were the predominant pathways (Supplementary Fig. S6D; Supplementary Table S3).

The genes upregulated by ELF3 included the defense response-related genes ALOX5 (encoding 5-lipoxygenase) and CXCL16 (encoding C–X–C motif chemokine ligand 16; Supplementary Tables S2 and S3), which were validated by qPCR and western blotting (Fig. 5A and B). We also found through ELISA that the abundance of CXCL16 in the supernatant was significantly increased in ELF3-overexpressing cells (Fig. 5C). Conversely, the expression levels of ALOX5 and CXCL16 were downregulated under ELF3 deficiency (Supplementary Figs. S7A, S7B, and S8). ELF3-specific peaks on the ALOX5 and CXCL16 genes were verified by ChIP, suggesting that ALOX5 and CXCL16 are directly regulated by ELF3 (Fig. 5D). We also performed ChIP-Seq with wild-type HBDEC2 cells to investigate where endogenous ELF3 binds. An ELF3-specific peak on CXCL16 was observed in wild-type HBDEC2 cells, whereas no clear peak on the ALOX5 gene was detected (Supplementary Fig. S9). CXCL16 and its receptor CXCR6 induce an antitumor immune response by promoting immune cell migration (25, 26). In chemotaxis assays, the migration of NK cells and CD8⁺ T cells toward conditioned medium from DOX-treated ELF3 Tet-ON cells was clearly enhanced compared with that with DOX-untreated cells (Fig. 5E), suggesting that ELF3 may modulate the expression of 5-lipoxygenase and CXCL16, which are involved in the recruitment of NK and CD8⁺ T cells in bile duct epithelial cells. Notably, the overexpression of ELF3 led to a robust increase in VTCN1 (encoding B7-H4), which serves as an immune checkpoint molecule (ICM; Supplementary Table S2; Supplementary Fig. S10A–S10C). Several lines of evidence have indicated that the overexpression of B7-H4 in cancer cells may be associated with low immunogenicity in the developing tumor (27).

To examine the expression level of ELF3 in biliary cancer cell lines, we performed western blot analyses using two biliary cancer cell lines, BD1 cells with an ELF3 heterozygous frame-shift mutation and TFK1 cells without an ELF3 mutation. The expression level of ELF3 in BD1 cells was lower than that in TFK1 and HBDEC2 cells (Supplementary Fig. S11A). We generated DOX-inducible ELF3-overexpressing cells from BD1 cells (hereafter referred to as BD1 Tet-ON cells). The overexpression of ELF3 in BD1 Tet-ON cells was confirmed by western blotting (Supplementary Fig. S11A). The elevated expression of ELF3 target genes, ALOX5 and VTCN1, was observed in DOX-treated BD1 Tet-ON cells compared with DOX-untreated cells, whereas the expression of CXCL16 and CGN was not changed (Supplementary Fig. S11B). To investigate the role of ELF3 in tumor formation, HBDEC2^ELF3−/− EMR ELF3–ERT2 cells were implanted subcutaneously in nude mice. After 21 days, the increased expression of ELF3–ERT2 fusion protein was observed in tamoxifen-treated HBDEC2^ELF3−/− EMR ELF3–ERT2 xenograft tumors (Fig. 6A), whereas the activation of ELF3 did not affect the tumor size. qPCR analysis revealed that the expression levels of ALOX5, CXCL16, CGN, and tight-junction proteins (OCLN and CXADR) in xenograft tumors were increased by tamoxifen-induced ELF3 activation (Fig. 6B). Consistent with these findings, the tubular structure with loss of vimentin was apparently increased in ELF3-activating xenograft tumors (Fig. 6C). In contrast, the expression of the EMT-related transcription factor TWIST2 was decreased by tamoxifen treatment (Fig. 6B).

To verify the relevance of the ELF3-related genes identified in HBDEC2 cells to human BTCs, we conducted GSEA using the list of genes upregulated together with ELF3 in HBDEC2 cells (Supplementary Table S4) and the BTC dataset previously evaluated by transcriptome sequencing (9). In BTCs, the expression level of ELF3 is variable. We arbitrary defined the top 30 cases (ELF3 RPKM: 152.73–744.67) as ELF3 ^high and the bottom 30 cases (ELF3 RPKM: 0.14–35.99) as ELF3 ^low (Supplementary Fig. S12). GSEA showed significant enrichment of these genes in ELF3^high tumors compared with that in ELF3^low tumors (Fig. 7A). A total of 146 leading-edge genes, termed core enrichment genes, were identified, including CXCL16, ALOX5, and VTCN1 (Supplementary Table S5). Figure 7B shows a heat map panel of the expression values for the subset of leading-edge genes in both ELF3^high and ELF3^Low BTCs. In accordance with these findings, our analysis of the gene expression data of esophageal carcinoma, stomach adenocarcinoma, and pancreatic adenocarcinoma obtained from The Cancer Genome Atlas (TCGA) database showed that the expression of CXCL16, ALOX5 and VTCN1 was higher in the ELF3^high tumor subgroup than in the ELF3^low tumor subgroup (Supplementary Fig. S13A–S13C) for these cancers. Although the expression of VTCN1 is high in the ELF3^high BTCs, ELF3^low BTCs demonstrated elevated expression of other ICMs, CD274 (encoding PD-L1), and PDICD1LG2 (encoding PD-L2; Supplementary Fig. S14).

Recent reports have suggested both oncogenic and tumor-suppressive roles for ELF3 in various human cancers (4, 5, 16–20, 28). In the present study, we investigated the tumor-suppressive role of ELF3 in normal bile duct epithelial cells, and identified novel direct ELF3 targets through a combination of ChIP-Seq and comprehensive transcriptome analyses. Our microarray data in two types of ELF3-deficient cells demonstrated that cell migration and cell adhesion were the predominant pathways deregulated by ELF3 depletion. We showed that ELF3 regulates these cellular programs through the direct transcriptional repression of ZEB2 and activation of CGN. In accordance with these findings, gene expression profiling for ELF3^low BTCs presented increased expressions of ZEB2, and decreased expressions of CGN (Supplementary Fig. S15). ZEB2 is a transcription factor that plays an important role in EMT and cell invasion (29–31). Cingulin is a tight-junction scaffolding protein that interacts with several tight-junction proteins and the cytoskeleton to regulate epithelial morphogenesis (32–35). Inactivation of ELF3 results in dysmorphogenesis and the altered differentiation of small intestinal epithelium (8). These findings suggest that the loss of ELF3 induces the disassembly of cell–cell junctions, resulting in enhanced cell motility and the perturbation of epithelial morphogenesis through the transcriptional regulation of ZEB2 and cingulin.

It has been reported that ELF5 (also known as ESE2) regulates EMT in mammary epithelial cells and breast cancer cells through the direct negative regulation of SNAI2 (36). EHF (also known as ESE3) directly represses the expression of TWIST1 and ZEB2, and its loss induces EMT and promotes tumor-initiation in prostate epithelial cells (37). These findings suggest that epithelium-specific ETS transcription factors, including ELF3, might maintain epithelial homeostasis by directly repressing key EMT transcription factors. Although ELF3-deficient cells cannot form tumors, they were found to generate glands with histological atypia when injected subcutaneously in nude mice. Furthermore, knockout cells did not grow in soft agar. This result supports that ELF3 deletion alone is not sufficient to induce anchorage-independent growth.

We also identified two novel ELF3 target genes, ALOX5 and CXCL16, that are implicated in host immune response. Accordingly, we found the significant enrichment of genes identified as upregulated together with ELF3 in ELF3^high BTCs, confirming that ELF3^high tumors show high expression of ALOX5 and CXCL16. This was also observed in esophageal, stomach, and pancreatic cancers based on transcriptome sequencing data obtained from TCGA. 5-Lipoxygenase controls arachidonic acid metabolism and mediates synthesis of the potent inflammatory mediator leukotriene (38). Depletion of 5-lipoxygenase promotes cancer progression and metastasis by recruiting CD8⁺ T cells expressing LTB₄ receptor BLT1 (39, 40). The soluble form of CXCL16 induces the chemotactic migration of cells with high expression of CXCR6, including CD8⁺ T cells, NK cells, and natural killer T (NKT) cells (41–44), and enhances the effectiveness of immune-mediated tumor suppression (25, 26). In the current study, the expression of 5-lipoxygenase and CXCL16 was increased concomitantly with the overexpression of ELF3. As expected from these changes, the recruitment of NK cells and CD8⁺ T cells toward conditioned medium obtained from ELF3-overexpressing cells was clearly enhanced. The binding of ELF3 to the CXCL16 gene was confirmed in wild-type HBDEC2 cells, whereas clear peaks on ZEB2, CGN, and ALOX5 genes were not detected, possibly due to the lower expression of ELF3 and/or sensitivity of the ChIP-Seq assay (Supplementary Fig. S9). Interestingly, exposure to primary bile acids increased CXCL16 expression in both liver sinusoidal endothelial cells and cholangiocarcinoma cells, facilitating hepatic NKT cell accumulation and decreasing liver tumor growth (45). We also found that the primary bile acid chenodeoxycholic acid (CDCA) elicited a robust increase in CXCL16 expression in both DOX-untreated and DOX-treated ELF3 Tet-ON cells (Supplementary Fig. S16). The common bile duct drains the bile into the duodenum through the ampulla of Vater and is in open communication with the gastrointestinal tract. Duodenal microorganisms are the major source of reflux hepatobiliary infections. To guard against reflux-induced infection of pathogens, the bile duct epithelium is equipped with a variety of defense mechanisms (46). Thus, ELF3 expressed in the bile duct epithelium might play a role in regulating the accumulation of cytotoxic immune cells.

Our results indicate that ELF3 is required for the maintenance of epithelial morphology via the direct negative regulation of ZEB2 and the positive regulation of cingulin in bile duct cells. ELF3 also directly upregulates the expression of 5-lipoxygenase and CXCL16, and ELF3-deficient cells may escape the attention of immune cells. EMT and cancer metastasis have been reported to be associated with tumor immune microenvironment (47, 48). Taken together, the loss of ELF3 is implicated as a driving force for early-onset and progression of disease with EMT through dampening of host immune defenses in the initial stages of tumor development. Importantly, the expression of ELF3 is variable in BTCs. Thus, ELF3-induced immune-modulation, by a 5-lipoxygenase inhibitor, CXCL16/CCR6 blocker, or anti-B7H4 antibody, could be useful in tumors expressing high levels of ELF3. Furthermore, ELF3^low BTCs had a higher expression of PD-L1 and PD-L2 than ELF3^high BTCs. Therefore, PD-1–blockade immunotherapy might be a promising therapeutic strategy in tumors with deleterious ELF3 aberrations. These issues should be investigated further.

D. Maeda reports grants from Takara Bio, Inc. outside the submitted work. T. Kiyono reports grants, personal fees, and non-financial support from National Cancer Center, as well as grants from Ministry of Education, Science and Culture of Japan during the conduct of the study. S. Yachida reports grants from Ministry of Education, Science and Culture of Japan, Integrated Frontier Research for Medical Science Division, Institute for Open and Transdisciplinary Research Initiatives, Osaka University, Japan Agency for Medical Research and Development, Joint Research Project of the Institute of Medical Science, University of Tokyo, and Yasuda Medical Foundation during the conduct of the study. No disclosures were reported by the other authors.

M. Suzuki: Conceptualization, data curation, formal analysis, validation, investigation, writing-original draft. M. Saito-Adachi: Data curation, formal analysis, writing-original draft. Y. Arai: Resources, supervision, methodology. Y. Fujiwara: Formal analysis, investigation, methodology. E. Takai: Supervision, investigation. S. Shibata: Investigation, methodology, writing-original draft. M. Seki: Formal analysis, methodology. H. Rokutan: Investigation, methodology. D. Maeda: Supervision, methodology. M. Horie: Supervision, methodology. Y. Suzuki: Methodology. T. Shibata: Supervision, methodology. T. Kiyono: Conceptualization, resources, supervision, funding acquisition, methodology, project administration, writing-review and editing. S. Yachida: Conceptualization, supervision, funding acquisition, methodology, project administration, writing-review and editing.

We are grateful to Risa Usui, Chiho Kohno, Takako Ishiyama, Drs. Mamoru Kato, Hiromi Nakamura, and Yasushi Totoki (National Cancer Center Research Institute) for technical assistance. The authors thank Dr. Hiroyuki Miyoshi (RIKEN, BioResource Center) for providing lentiviral constructs. The authors thank Dr. Shohei Koyama (National Cancer Center) and for helpful discussions. This work was supported by the following grants: grants-in-aid for Scientific Research from the Ministry of Education, Science and Culture of Japan (16H04701 to T. Kiyono; A17H04275 to S. Yachida); Integrated Frontier Research for Medical Science Division, Institute for Open and Transdisciplinary Research Initiatives, Osaka University; Japan Agency for Medical Research and Development (JP20ck0106558); Joint Research Project of the Institute of Medical Science, University of Tokyo (2020); Yasuda Medical Foundation; Yakult Bio-Science Foundation; Princess Takamatsu Cancer Research Fund; and Takeda Science Foundation (to S. Yachida).

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