IFN-g-induced Jmjd3-Zeb1 axis contributes to an aggressive phenotype in lung adenocarcinoma

Active IFN-γ signaling is a common feature of tumors responding to PD-1 checkpoint blockade. IFN-γ exhibits both anti- and pro-tumor activities. Therefore, identifying the pro-tumor effects of IFN-γ and their underlying molecular mechanisms could be a critical step for developing therapeutic strategies to maximize the anti-tumor ecacy of immunotherapies. Western blot, immunouorescence, and quantitative real-time PCR assays were used to evaluate the expression of ZEB1 and EMT associated biomarkers. Trans-well assay was used to examine the role of IFN-γ on cancer cell migration in vitro. Murine tumor xenograft models were performed to examine the effect of IFN-γ on cancer cell metastasis in vivo. Colony formation assay was performed to detect the role of ZEB1 in cell proliferation. RNA-seq was performed to analyze the EMT-associated gene expression patterns in response to IFN-g treatment. Loss-of-function analysis and chromatin immunoprecipitation were used to reveal the mechanism underlying ZEB1 induction by IFN-γ. IFN-(cid:0) induces EMT in adenocarcinoma cells. a Cells were cultured or IFN-(cid:0) (100 IU/ml) for 12, 48, and 72 h, harvested, and lysed. The protein levels of E-cadherin and Vimentin were analyzed by immunoblotting. The bar graphs show densitometry analyses of changes in E-cadherin and Vimentin levels as normalized to β-actin. Data shown are representative images of four independent experiments. b Cells were treated with or without IFN-(cid:0) (100 IU/ml) for 72 h, and immunostained with anti-E-cadherin (1:200) or anti-Vimentin (1:200). Blank indicates untreated cells (magnication, 200×). c Cells were treated with or without IFN-(cid:0) (100 IU/ml) for 24 and 48 h. CDH1 and VIM mRNA levels were quantied by RT- PCR. Data are presented as mean ± SD (n = 3) and were analyzed by the two-sided Student’s t-test, *P < 0.05, **P < 0.01, *** P <0.001, ns, not signicant. d A549 and HCC827 cells were treated with or without IFN-(cid:0) (25 IU/ml) for 2 weeks. Changes in cell morphology were observed by phase-contrast light microscopy (magnication, 200×). e The effects of IFN-(cid:0) on cell migration were examined by Transwell assay. Cells (7×104 cells in 100 (cid:0) l RPMI-1640) were suspended in serum-free medium with or without IFN-(cid:0) (100 IU/ml). After 24 h, the number of cells on the bottom side of the Transwell inserts was counted in three random elds under a light microscope (magnication, 200×). Data are presented as mean ± SD and were analyzed with the two-sided Student t-test, **P < 0.01, *** IFN-(cid:0)-induced ZEB1 expression is responsible for EMT. Changes in the expression of EMT-associated genes in A549 and HCC827 cells are shown. a Comparison of expression proles of EMT-associated genes between A549 and HCC827. Red dots represent EMT-associated genes upregulated in A549 cells, and blue indicate downregulated genes in A549 cells, and blue indicate genes downregulated in A549 cells, compared with HCC827 cells. Changes in the expression of EMT-associated genes A549 and (c) HCC827 cells in response to are upregulated genes genes in response to 24 with cells. A549 and HCC827 cells were treated with IFN-(cid:0) (100 IU/ml) for 24 h. The mRNA levels of CTGF, FGF2, MAP1B, and ABLIM1 were quantied by RT- PCR. Data are presented as mean ± SD (n = 3) and were analyzed with the two-sided Student’s t-test, ***P <0.001. e A549 and HCC827 cells were transfected with siRNA-ZEB1 or siRNA-control for 48 h. Subsequently, the cells were suspended in medium containing 10% FBS with or without IFN-(cid:0) (100 IU/ml) for 24 h. The mRNA levels of VIM were quantied by RT- PCR. Data are presented as mean ± SD (n=3) and were analyzed with one-way ANOVA, ***P < 0.001). f A549 and HCC827 cells were transfected with siRNA-ZEB1 or siRNA-control for 24 h, and the cells were suspended in medium containing 10% FBS with or without IFN-(cid:0) (100 IU/ml) for c At the end of (seven days i.v. injection of A549 cells), lung tissues were harvested and stained with hematoxylin eosin (HE). Representative images of histological inspection of mouse lungs from each group are shown (magnication, 50×; insert, 200×). The number of tumor nodules and lung metastasis index (ratio of tumor area to the total tumor and lung area) were evaluated and analyzed between IFN-(cid:0)(cid:0) treated and untreated groups. Data are presented as mean ± SD (n = 3) and were analyzed with the two-sided Student’s t-test, * P < 0.05. Scale bar: 200 (cid:0) m (d). A549 cells were transfected with shRNA-ZEB1 or shRNA-control, and a stable ZEB1 knockdown cell line was generated. Then the cells (7×104 cells in 100ul RPMI-1640) were suspended in serum-free medium with or without IFN-(cid:0)(cid:0)(cid:0)100 IU/ml) for 24 h, the cells in the lower chamber were counted in three random elds under a light microscope (magnication, 200×). Data are presented as mean ± SD and were analyzed with one-way ANOVA, * P < 0.05, **P < 0.01, ns, not signicant. e A549 cells transfected with shRNA-ZEB1 or shRNA-control were injected into NCG mice via tail vain, the animal model in (b) was used. Representative images of histological inspection of mouse lungs from each group are shown (magnication, 50×; insert, 200×). The number of tumor nodules and lung metastasis index were calculated. The tumor nodules were counted and area ratio in the lung parenchyma was calculated. Data are presented as mean ± SD (n = 3) and were analyzed one-way ANOVA, *P < 0.05.

Cell proliferation was assessed using the CCK-8 Kit (Dojindo Molecular Laboratories, Kumamoto, Japan). Cells were seeded (3,000-5,000 in 100μl/well) and cultured overnight before exposure to the indicated stimuli. Absorbance was measured at 450 nm using a microplate spectrophotometer (TECAN).

Colony-forming assay
Cells were seeded in 6-well plates at a density of 800 cells /well and maintained for 14 days with or without IFN-g (100 IU/ml). The cells were xed with 4% (w/v) paraformaldehyde for 20 min, stained with crystal violet for 15 min and washed with PBS three times. The stained cell colonies (with ≥ cells 50) were counted.
Transwell migration assay The cell migration assay was carried out using 8 μm pore size Transwell chambers (Corning, NY, USA). In brief, cells were suspended in serum-free medium and plated into the upper chambers (7×10 4 cells in 100ul/well). In the lower chambers, medium supplemented with 10% FBS was used as a chemoattractant. IFN-g and other indicated reagents were added to both chambers at equal concentrations. After 24 h of incubation, the cells that migrated through the membrane to the bottom surface were xed with 4% (w/v) paraformaldehyde, and stained with 0.1% (w/v) crystal violet. The migratory cells were examined under an optical microscope at 200× magni cation. The average numbers of migratory cells were obtained from three randomly chosen elds.

Chromatin immunoprecipitation
The ChIP assay was performed using ChIP following the manufacturer's instructions Kit (Cell Signaling Technology, Inc.). Immunoprecipitation was performed overnight at 4°C with anti-H3K27me3 and anti-H3K4me3 antibodies, and normal rabbit IgG control. The fragments of the human ZEB1 promoter in immunoprecipitates were identi ed by qPCR. Detailed information on primers is provided in Supplemental Table 2.
Animal model NCG mice were purchased from the Experimental Animal Center (Hubei, China) and maintained in an environment with a standardized barrier system (System Barrier Environment No.00021082). A549 cells, A549-shRNA-control cells, and A549-shRNA-ZEB1 cells pretreated with IFN-g (400 IU/ml) for 4 days or without IFN-g treatment were resuspended in 100 ml PBS and then injected into the tail vein of NCG mice (2×10 6 cells/mouse), followed by i.p. injection of IFN-g (2000 IU in 100ul per mouse) every other day for total three times. Histologically detectable lung metastatic foci were observed microscopically 7 days post-injection [18]. The lungs were excised, xed in 10% formalin, para n-embedded and stained with hematoxylin and eosin (H&E) for pathological identi cation of tumor nodules in the lung parenchyma. The photomicrographs of the lungs were taken using a light microscope (Axio Observer 3, Zeiss, Germany). The metastasis area and lung area were quanti ed using ImageJ software.
Ex vivo culture of patient-derived lung cancer explants Fresh lung cancer tissues were obtained from patients undergoing pulmonary resection prior to radiation or chemotherapy in the Department of Thoracic Surgery, Tongji Hospital. The ex vivo culture was performed as previously described [19]. Brie y, fresh human lung cancer tissue was dissected into 1 mm 3 cubes, placed on a Gelatin sponge (HuSHiDa, Jiangxi, China), and bathed in RPMI-1640 medium supplemented with 10% heat-inactivated FBS and 100 IU/ml penicillin-streptomycin. In addition, indicated amounts of IFN-g were added to the media. Tissues were cultured at 37°C for 24-36 h and collected for RNA extraction. The use of human tissue samples was approved by the Institutional Ethics Committee of the Huazhong University of Science and Technology.

Statistical analysis
Data in bar graphs are displayed as mean ± SD. Data between two groups were compared with the twotailed Student t-test (* P<0.05, **P<0.01, ***P<0.001), and one-way ANOVA test was used to compare data between three or more groups. Statistical analysis was performed with GraphPad Prism software v. 8.0. For RNA-seq data, the P-value signi cance threshold in multiple tests was set by the false discovery rate (FDR). Fold-changes were also estimated according to the FPKM in each sample. The differentially expressed genes were selected using the following criteria: FDR ≤0.05 and log 2 (fold-change) ≥ 0.5.
Previously, we performed a global transcriptome study (microarray analysis) and compared tumor tissues with high versus low IFNG expression level. ZEB1 mRNA levels were signi cantly higher in tumors with high IFNG expression (GSE99995). To determine whether IFN-γ has the ability to induce ZEB1 transcription in lung cancer cells, ve lung adenocarcinoma cell lines and the lung carcinoid cell line UMC-11 were cultured with IFN-γ, and ZEB1 mRNA levels were evaluated. Signi cantly higher ZEB1 mRNA levels were observed in all ve lung adenocarcinoma cell lines cocultured with IFN-γ, whereas IFN-γ appeared to have no effect on ZEB1 expression in lung carcinoid cells (Fig. 1a). Immunoblotting analysis revealed that ZEB1 protein levels increased within 12 h upon IFN-γ stimulation and sustained for at least 72 h after a single treatment (Fig. 1b). Strikingly, prolonged exposure of A549 and H1975 cells to IFN-γ led to sustained ZEB1 expression even after removal of IFN-γ (Fig. 1c).
Furthermore, we cultured human-derived lung adenocarcinoma cells with IFN-γ ex vivo. IFN-γ was able to induce ZEB1 transcription (Fig. 1d). Interestingly, exogenous IFN-γ did not signi cantly increase ZEB1 mRNA levels in distant non-tumor lung tissues from patients with lung cancer (Fig. 1e). Our data demonstrate that IFN-γ induces ZEB1 expression at both mRNA and protein levels in lung adenocarcinoma cells.
IFN-γ induces EMT in lung adenocarcinoma cells. E-cadherin and Vimentin are the most commonly used EMT markers, as their expression patterns undergo dramatic changes during EMT. In A549, HCC827, and H2228 lung cancer cells, IFN-γ stimulation downregulated E-cadherin expression and upregulated Vimentin expression, as determined by immunoblot analysis (Fig. 2a). The altered expression patterns of E-cadherin and Vimentin in lung cancer cells upon IFN-γ treatment were further validated by immuno uorescence analysis (Fig. 2b). Real-time PCT (RT-PCR) analysis revealed that CDH1 transcription was decreased at 24 h and maintained at a low level in A549 and H2228 cells, whereas VIM transcription was signi cantly increased to various degrees in the three cell lines in response to IFN-γ stimulation (Fig. 2c).
Treatment with low amounts of IFN-γ (10 IU/ml) for 3 days did not signi cantly affect E-cadherin levels in A549 and HCC827 ( Supplementary Fig. 1a). While treatment with IFN-γ at 25 IU/ml signi cantly downregulated E-cadherin levels in A549 and HCC827 cells ( Supplementary Fig. 1b). Interestingly, prolonged exposure to IFN-γ induced morphological changes in A549 and HCC827 cells, which acquired a broblast-like appearance (Fig. 2c). The broblast-like appearance of A549 cells became even more evident after treatment with IFN-γ for 21 days ( Supplementary Fig. 1c), indicating that IFN-γ−treated cells enter a stable mesenchymal-associated state. Substantial evidence has shown that EMT is associated with increased cell migration in vitro. As shown in Fig. 2e, the migratory capability of A549 and HCC827 cells was indeed signi cantly increased after IFN-γ treatment. IFN-γ treatment led to increases in migration of 57% and 45% in A549 and HCC827 cells, respectively. Collectively, these data demonstrate that IFN-γ induces EMT in lung adenocarcinoma cells and promotes cell migration in vitro.
EMT involves a robust reprogramming of gene expression. We analyzed the transcriptome alternations by RNA-seq analysis during EMT following IFN-γ treatment. It has been reported that A549 cells have mesenchymal characteristics, whereas HCC827 cells have epithelial features [20]. For these reasons, we selected A549 and HCC827 cells to investigate the reprogramming of gene transcription by IFN-γ. An EMT signature consisting of 130 genes was analyzed, including 67 upregulated mesenchymal-associated genes and 63 downregulated epithelial-associated genes (Supplementary Table 3) [21]. Of the 63 epithelial-associated genes, 35 genes including CDH1, were highly expressed in HCC827 cells compared with A549 cells; of the 67 mesenchymal-associated genes, 35 genes, including VIM, were expressed at lower levels in HCC827 cells than in A549 cells (Fig. 3a). These results indicate that relative to A549 cells, HCC827 cells have more epithelial-associated features, which was consistent with the reports by others [20]. EMT is a dynamic process with intermediary states that is not easily identi ed in cultured cells. We performed transcriptome analysis after the cells were treated with IFN-γ for 8 or 24 h. The number of differentially expressed genes was dramatically increased after 24 h of stimulation with IFN-γ compared with 8 h of stimulation (Supplementary Fig. 2A).
Among these differentially expressed genes, of 67 genes that are upregulated in EMT, 13 genes were upregulated after IFN-γ treatment, and of 63 genes that are downregulated in EMT, 23 genes were downregulated after IFN-γ treatment. In HCC827 cells, 58 genes were differentially expressed after 24 h of IFN-γ treatment, including 32 upregulated and 26 downregulated genes (Log 2 FC ≥ 0.5, P < 0.05) ( Fig. 3c and Supplementary Table 5); among these differentially expressed genes, of 67 genes that upregulated in EMT, 21 genes were upregulated after IFN-γ treatment, and of 63 genes that are downregulated in EMT, 12 genes were downregulated after IFN-γ treatment. Detailed analysis revealed that IFN-γ stimulation of the mesenchymal-like A549 cells led to downregulation of more epithelial-associated genes (genes downregulated during EMT) whereas IFN-γ stimulation of the epithelial-like HCC827 cells led to upregulation of more mesenchymal-associated genes (genes upregulated during EMT). Although IFN-γ altered the expression patterns of E-cadherin and Vimentin in both A549 and HCC827 cells, transcriptome analysis of EMT-associated genes in IFN-γ-treated cells revealed that IFN-γ induced EMT in lung cancer cells is not a uni ed state. The characteristics of IFN-γ-induced EMT could be associated with the intrinsic state of untreated lung cancer cells.
Although IFN-γ differentially alters the expression of EMT-associated genes in A549 and HCC827 cells, RNA-seq analysis showed that the expression of 15 genes was equally affected in both cell lines upon IFN-γ treatment, including 7 upregulated and 8 downregulated genes. Among the upregulated genes, CTGF, which encodes connective tissue growth factor (CTGF), is able to induce EMT and its expression levels are highly correlated with EMT markers [22]. FGF2, which encodes basic broblast growth factor, and MAP1B, which encodes a protein belonging to the microtubule-associated protein family, are both upregulated during EMT [21]. FGF2 promotes EMT and metastasis through the FGFR1-ERK1/2-SOX2 axis in FGFR1-ampli ed lung cancer [23]. Actin binding LIM protein 1, encoded by ABL1M1 gene, plays multiple roles in establishing and maintaining cellular structure through mediating interactions between actin laments and cytoplasmic LIM binding partners [24]. ABLIM1 is downregulated during EMT [21]. These differentially expressed genes in response to IFN-γ stimulation in both A549 and HCC827 cells were con rmed by quantitative RT-PCR (Fig. 3d).
RNA-seq analysis revealed that ZEB1 but not Snail and Slug was signi cantly upregulated in both A549 and HCC827 cells as early as 8 h upon IFN-γ stimulation (Supplementary Fig. 2A). The effect of IFN-g on the expression of Snail and Slug was con rmed by quantitative RT-PCR ( Supplementary Fig. 2B). To determine whether ZEB1 is involved in IFN-γ induced EMT, we knocked down ZEB1 using small interfering RNA (siRNA) in lung cancer cells ( Supplementary Fig. 3a). Knockdown of ZEB1 abrogated the IFN-γinduced upregulation of VIM transcription (Fig. 3e). IFN-γ-mediated alterations in the expression patterns of E-cadherin and Vimentin in A549 and HCC827 cells were reversed upon ZEB1 knockdown (Fig. 3f).
ZEB1 is required for IFN-γ-promoted cell migration and metastasis.
To determine whether ZEB1 is required for IFN-γ-promoted cell migration, we used siRNA-ZEB1 to knock down ZEB1 (Supplementary Fig. 3a). The migratory capability of A549 cells promoted by IFN-γ was signi cantly compromised by the downregulation of ZEB1 expression (Fig. 4a). To examine the in vivo effects of IFN-γ on lung cancer cell metastasis, we established an in vivo metastasis model by intravenous injection of A549 cells that had been treated with IFN-γ in vitro for 4 days into NCG mice (2 × 10 6 cells/mouse) (Fig. 4b). The mice were given recombinant human IFN-γ (2000 IU/mouse) intraperitoneally (i.p.) every other day for total three times. Seven days after injection of A549 cells, lung tissues were collected and the presence of metastatic foci and the size of metastases were analyzed microscopically. Control mice were given untreated A549 cells and the mice were not given IFN-γ. As shown in Fig. 4c, IFN-γ treatment signi cantly increased the number and the size of metastatic nodules in lung tissues.
To determine the role of ZEB1 in this event, we transfected A549 cells with short hairpin RNA against ZEB1 (shRNA-ZEB1) and obtained a stable ZEB1-depleted cell line ( Supplementary Fig. 3b). The IFN-γpromoted increase of A549 cell migration in vitro was diminished in shRNA-ZEB1 A549 cells (Fig. 4d). Our in vivo metastasis model showed that the number and the size of metastatic nodules were reduced in mice injected with IFN-γ-treated shRNA-ZEB1 A549 cells compared with mice injected with IFN-γ-treated control A549 cells, indicating that loss of ZEB1 dramatically reduces IFN-γ-promoted metastasis of A549 cells (Fig. 4e). Our data demonstrate that ZEB1 is responsible for IFN-γ-induced cell migration in vitro and metastasis in vivo.
We next evaluated whether IFN-γ-induced activation of the JAK2-STAT1 pathway is involved in regulation of ZEB1 expression. IFN-γ-treated JAK2-de cient H1573 cells did not exhibit altered expression levels of E-cadherin or ZEB1 (Supplementary Figs. 3c and d), suggesting that IFN-γ-induced EMT and upregulation of ZEB1 might be dependent on JAK2-mediated signaling. To con rm this, we knocked down JAK2 by siRNA-JAK2 in A549 and HCC827 cells ( Supplementary Fig. 3e). These cells were subsequently stimulated with IFN-γ. Upregulation of ZEB1 by IFN-γ was no longer observed (Fig. 5a). We used the same approach to determine whether STAT1 is required for IFN-γ-induced ZEB1 expression ( Supplementary Fig. 3f). As shown in Fig. 5b, IFN-γ-induced ZEB1 expression was diminished in cells transfected with siRNA-STAT1, indicating that IFN-γ-induced ZEB1 expression requires JAK2-STAT1. ZEB1 transcription is regulated by the modulation of the chromatin environment at gene regulatory elements [25,26]. H3K27me3 is often associated with transcriptional repression. The relative absence of H3K27me3 in the chromatin at the ZEB1 promoter signals active transcription [25]. Interestingly, the expression of H3K27 trimethylation and the ratio of H3K27me3 to H3 were rapidly reduced in A549 cells after exposure to IFN-γ (Figs. 5c and d). JMJD3, a direct transcriptional target of STAT1, catalyzes the demethylation of H3K27me3 [27]. For these reasons, we examined whether IFN-γ induces JMJD3 expression. JMJD3 transcription was rapidly upregulated in A549 and HCC827 cells upon exposure to IFN-γ (Fig. 5e). Immunoblot analysis revealed increased JMJD3 expression after 3 h of IFN-γ treatment (Fig. 5f). To determine whether JMJD3 is associated with IFN-γ-induced ZEB1 expression, the JMJD3 speci c inhibitor GSK-J4 was used [28]. GSK-J4 signi cantly reduced IFN-γ stimulated increasing in ZEB1 mRNA levels (Fig. 5g) and protein levels (Fig. 5h). Knockdown experiments with siRNA-JMJD3 further con rmed that IFN-γ-induced ZEB1 expression requires JMJD3 ( Supplementary Fig. 3g and Fig. 5i). Downregulation of STAT1 expression led to abrogation of IFN-γ-induced JMJD3 expression (Fig. 5j), con rming that IFN-γ-induced JMJD3 expression is STAT1-dependent.
Since H3K27 demethylation is linked to transcriptional activation, we examined whether H3K27 demethylation is associated with the transcription of ZEB1. The ZEB1 promoter exhibits a bivalent chromatin con guration [28]. H3K4me3 is associated with transcriptional initiation [29], whereas the H3K27me3 is associated with transcriptional repression. We performed a chromatin immunoprecipitation (ChIP) assay at the ZEB1 promoter to compare the levels of histone modi cations in control versus IFNγ-treated cells. IFN-γ treatment led to a signi cant reduction in H3K27me3 levels at the ZEB1 promoter in A549 cells, whereas IFN-γ did not signi cantly affect H3K4me3 levels at the ZEB1 promoter (Fig. 5k). These data demonstrate that IFN-γ enables the ZEB1 promoter to transition from the bivalent to the active chromatin state, at least in part through the demethylation of H3K27me3.
ZEB1 transcription is tightly regulated by microRNA. Recent studies have revealed that IFN-γ promotes ZEB1 expression through IFIT5-mediated suppression of miR-363 in prostate cancer and renal cancer [13]. We did not observe downregulation of miR-363 expression by IFN-γ in lung cancer cells ( Supplementary Fig. 4). It is well known that the miR-200 family inhibits ZEB1 expression and plays a major role in preventing ZEB1 from triggering EMT. In turn, ZEB1 can directly repress the transcription of miR-200 loci [30]. As shown in Fig. 6a, miR-200c expression was signi cantly reduced in A549 and HCC827 cells after 12 h of IFN-γ treatment, and the reduction of miR-200c expression was even more dramatic after 24 and 48 h of IFN-γ treatment. We wondered whether IFN-γ-promoted ZEB1 expression in lung cancer cells is related to the downregulation of miR-200 expression. IFN-γ-induced upregulation of ZEB1 transcription was observed after 4 h in both A549 and HCC827 cells, whereas miR-200c expression was not affected even after 6 h of IFN-γ stimulation (Figs. 6b and c). Our data suggest that IFN-γ rst promotes ZEB1 transcription, and subsequently suppresses miR-200c expression.
PD-L1 expression has been reported to be directly regulated by miR-200 family members [31]. We examined whether IFN-γ-induced PD-L1 expression can be regulated by ZEB1. Knockdown of ZEB1 by siRNA did indeed lead to a signi cant reduction of IFN-γ-induced PD-L1 expression in A549 and HCC827 cells (Fig. 6d). Further analysis revealed that downregulation of ZEB1 enhanced miR-200 expression (Fig. 6e). Our data indicate that IFN-γ stimulation rapidly induces ZEB1 expression and consequently downregulates miR-200c, which at least partially contributes to the upregulation of PD-L1 expression.
IFN-γ-mediated anti-proliferation and induction of CXCL9 and CXCL10 expression are not affected by ZEB1 knockdown.
Previous studies by us and others have shown that IFN-γ suppresses the proliferation of lung cancer cells [2,32]. We wondered whether ZEB1 is involved in IFN-γ-mediated suppression of cell proliferation. The JMJID3 inhibitor GSK-J4 revered IFN-γ-induced ZEB1 expression (Figs. 5g and h). However, GSK-J4 did not affect IFN-γ-mediated suppression of A549 cell proliferation (Fig. 7a). Knockdown of ZEB1 by siRNA did not alter the anti-proliferative effects of IFN-γ in both A549 and HCC827 cells (Fig. 7b). ZEB1 knockdown had no signi cant effect on IFN-γ-mediated suppression of colony formation (Fig. 7c). Cyclin E1, which is encoded by the CCNE1 gene, plays a critical role in the control of cell cycle progression by allowing G1 to S phase transition [33]. IFN-γ-treated A549 cells exhibited signi cantly lower CCNE1 mRNA levels than untreated A549 cells (Fig. 7d). Knockdown of ZEB1 had no effect on the IFN-γ-mediated reduction of CCNE1 expression (Fig. 7d). A previous study by us has shown that IFN-γ-mediated suppression of cell proliferation requires STAT1 and IRF1 [2]. Downregulation of ZEB1 had no effect on IFN-γ-induced STAT1 and IRF1 expression at mRNA and protein levels (Figs. 7e and f). ZEB1 knockdown did not affect IFN-γ-induced phosphorylation of STAT1 (Fig. 7g). Additionally, CXCL9 and CXCL10, which encodes two chemokines that are important for the recruitment of activated lymphocytes to tumors, are also target genes of STAT1-IRF1. As shown in Fig. 7h, knockdown of ZEB1 did not alter the IFN-γ-induced expression pattern of CXCL9 and CXCL10.
Based on our results, we proposed a functional mechanism as shown in Fig. 7i. Our results showed that IFN-γ stimulation induced a dramatic change in the expression pattern of Ecadherin and Vimentin in lung cancer cells. It is widely recognized that experimental models using only a small selection of epithelial and mesenchymal biomarkers, including E-cadherin, N-cadherin and Vimentin, to de ne or con rm EMT sketch an oversimpli ed view of this complex process. EMT is not one clearly de ned tumor state but a set of multiple dynamic transitional states between epithelial and mesenchymal phenotypes [16,34]. Due to the complexity of the EMT process, reliable biomarkers are still lacking and a comprehensive method to identify and/or measure EMT, particular in vivo, does not exist. Nevertheless, numerous gene expression studies have been conducted to obtain transcriptome signatures and marker genes associated with EMT [20,35]. In our study, we not only analyzed CDH1 and VIM expression levels, but we also performed a transcriptome analysis to determine whether IFN-γ alters the expression of other EMT-associated genes. We obtained the EMT core gene signatures, which consists of 130 genes, through meta-analysis of 18 independent and published gene expression studies of EMT [21].
Comparing the expression levels of these 130 genes, we found that IFN-γ stimulation altered the transcription of almost 50% of EMT-associated genes in both A549 and HCC827 cells. Among these differentially expressed genes, the EMT-associated transcription factor ZEB1 was rapidly upregulated in response to IFN-γ stimulation. In addition to these genetic biomarkers, several in vitro criteria have been used to determine EMT, including a spindle-shape morphology and increased migratory capability [36,37]. We found that exposure of lung cancer cells to IFN-γ induced morphologic changes, and also promoted cell migration in vitro and metastasis in vivo. Our ndings demonstrate that IFN-γ is capable of inducing EMT in lung adenocarcinoma cells.
In prostate cancer and renal cancer, IFN-γ induces EMT through the IFIT5-XRN1 complex, which regulates the turnover of speci c tumor-suppressive microRNAs, such as miR-101, miR-128, and miR-363 [13]. In our study, IFN-γ treatment even upregulated miR-363 levels in lung cancer cells. Our ndings suggest that the mechanism by which IFN-γ induces EMT is cancer type-dependent and context-speci c.
In lung cancer cells, IFN-γ stimulation led to a rapid increase in mRNA and protein levels of the STAT1target gene JMJD3. In mammary epithelial cells, JMJD3 mediates TGF-β induced EMT through upregulation of SNAIL expression, leading to breast cancer invasion [38]. JMJD3 upregulates Slug and promotes cell migration, invasion, and transition towards a stem-like phenotype in hepatocellular carcinoma [39]. JMJD3 could be a key regulator of cancer aggressiveness. We found that targeting JMJD3 with the inhibitor GSK-J4 or JMJD3 knockdown using siRNA abrogated IFN-γ-induced ZEB1 expression.
Recently obtained evidence has indicated that EMT-associated transcription factors regulate large set of cancer cell features, extending beyond tumor migration, invasion, and metastasis. Recent studies have demonstrated a robust correlation between EMT score, ZEB1/miR-200 levels and PD-L1 expression in multiple cancer datasets [31]. In the present study we showed that IFN-γ-induced ZEB1 expression is involved in the upregulation of PD-L1 expression through its suppressive effects on miR-200 expression. Moreover, Lou and colleagues have reported that an EMT-related mRNA signature is in fact associated with increased expression of diverse immune inhibitory ligands and receptors in lung adenocarcinoma, including PD-L1, TIM-3, LAG3 and CTLA-4 [40]. As illustrated in our working model, IFN-γ is able to simultaneously induce EMT-like features and PD-L1 expression in lung cancer cells via the upregulation of ZEB1 expression. However, whether these two events are independent remains to be elucidated. Our ndings suggest that strong anti-tumor immune properties might be accompanied by increased tumor progression through multiple means.
In this study, we found that ZEB1 knockdown diminished IFN-γ-induced PD-L1 expression, cell migration, and metastasis. Interestingly, downregulation of ZEB1 did not affect IFN-γ-mediated suppression of cell proliferation and increased expression of CXCL9 and CXCL10. JMJD3 inhibitor GSK-J4 suppressed IFN-γinduced ZEB1 expression. GSK-J4 has been applied in the treatment of several cancers, such as acute myeloid leukemia and prostate cancer [41,42]. Our study sheds light on the functional mechanism by which targeting ZEB1 might limit the pro-tumor effects of IFN-γ. As IFN-γ has been proven to play a critical role in PD1/PD-L1 blockage-based immunotherapies, targeting ZEB1 combined with immune checkpoint blockade might improve the anti-tumor e cacy of immunotherapies.

Consent for publication
Not applicable.

Availability of data and materials
The analyzed datasets generated during the current study are available from the corresponding author on reasonable request.

Competing Interests
The authors declare no competing interests. presented as mean ± SD (n = 3) and were analyzed by the two-sided Student's t-test, *P < 0.05, **P < 0.01, *** P <0.001. b ZEB1 protein levels were analyzed by immunoblotting. Data shown are representative images of four independent experiments. c Cells were cultured with IFN-(25 IU/ml) for indicated time intervals. IFN-was removed for 3 or 7 days, and the cells were harvested and lysed. ZEB1 expression was analyzed by immunoblotting. d Fresh human lung cancer tissue (n = 6) and (e) distant non-tumor lung tissues (> 3cm away from the edge of the tumors) (n = 4) were treated with indicated concentrations of IFN-for 36 h, and the tissues were collected for RNA extraction. ZEB1 mRNA levels were quanti ed by RT-PCR. Data are presented as mean ± SD (n = 3) and were analyzed by the two-sided Student's t-test, *P < 0.05, **P < 0.01, *** P <0.001. Figure 2 IFN-induces EMT in lung adenocarcinoma cells. a Cells were cultured with or without IFN-(100 IU/ml) for 12, 48, and 72 h, harvested, and lysed.   with IFN-for indicated time intervals, harvested, and lysed, and total histones were extracted following the Histone Extract protocol. c H3K27me3 levels were analyzed by immunoblotting. d The graph presents the quanti cation of H3K27me3 levels normalized to histone H3 levels. Data are presented as mean ± SD (n = 3) and were analyzed one-way ANOVA, *P < 0.05, **P < 0.01. e A549 and HCC827 cells were cultured with IFN-(100 IU/ml) for indicated time intervals. JMJD3 mRNA levels were quanti ed by RT-PCR.
Untreated cells served as control. Data are presented as mean ± SD (n = 3) and were analyzed with the two-sided Student's t-test, ***P <0.001. f JMJD3 protein levels were analyzed by immunoblotting with anti-JMJD3. g Cells were cultured in medium with or without the JMJD3 inhibitor GSK-J4 (10 M) for 1 h.
using a CCK8 kit (n = 3). Data were analyzed with one-way ANOVA. b A549 and HCC827 cells were transfected with siRNA-control or siRNA-ZEB1 for 24 h. Subsequently, the cells (5-10×103cells/well) were treated with or without IFN-for 48 h. Proliferation was assessed using a CCK8 kit (n = 3). Data were analyzed with one-way ANOVA. c A549-shCtrol and A549-shZEB1 cells were seeded in 6-well plates at a density of 800 cells per well and maintained for 14 days with or without IFN-. The medium was changed every 3 days. A549-shControl and A549-shZEB1 cells were seeded in 6-well plates and cultured for 24 h. Then the cells were stimulated with or without IFN-(100 IU/ml) for 48 h. The mRNA levels of (d) Cyclin E and (e) STAT1 and IRF1 were quanti ed by RT-PCR. Data are presented as mean ± SD (n = 3) and were analyzed with one-way ANOVA. f A549-shControl and A549-shZEB1 cells were stimulated with or without IFN-for 24 h and harvested. STAT1 and IRF1 protein levels were analyzed by immunoblotting. g A549-shControl and A549-shZEB1 cells were stimulated with or without IFN-for indicated time intervals.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.