Human esophageal squamous cell carcinoma cell lines (KYSE30, EC109 and KYSE150) and 293 T cells were were purchased from the China Center for Type Culture Collection (Wuhan, China), which were maintained in RPMI/1640 medium (Gibco) supplemented with 10% fetal bovine serum (Hyclone). Cells were cultured in an atmosphere of 5% CO₂ at 37°C.

Cell proliferation rates were evaluated by CCK-8 assay and colony formation assay. Briefly, we seeded the cells into 96-well culture plates (Corning Life Science). During the timepoint of 24, 48, 72, 96 and 120 h post seeding, the 10 μl CCK-8 reagent was added to the wells and incubated for 1 h. The SpectraMax 190 Microplate Reader (Molecular Devices) was used to assess the absorbance at 450 nm. For the colony formation assay, 1000–2000 indicated cells were seeded into the six-well plates. After two weeks, these wells were fixed with 1.5% crystal violet/methanol solution and counted. Lastly, for the soft-agar colony formation assay, 2000 evaluated KYSE30 cells were mixed into 0.36% agarose culture medium as the upper layer, whereas the bottom layer is mixed with 0.75% agarose and DMEM culture medium. The colony were stained via crystal violet and the clone numbers were counted.

ESAC cells (KYSE30 and EC109) were seeded on the coated filter in 100 μl of serum-free medium, and the bottom chamber was filled with 600 μl complete culture medium. Then, the invasion assays were conducted in the 24-well cell chamber coated with the 30 μl of Matrigel (Corning Incorporated). The invasive ESCA cells were then stained with the crystal violent after 24 h incubation at 37°C under the 5% CO₂. The migration assays were performed in a similar process without coating with Matrigel.

The pX459 plasmid was used to clone guide oligos targeting KDM4D. KYSE30 and EC109 cells were plated and transfected with pX459 constructs overnight. After 24h transfection, 1μg/ml puromycin was added into the culture to screen cells for 72 hours. Afterwards, living cells were seeded into the 96 well plate by limited dilution to isolate monoclonal cell line. The KDM4D knock out cell clones are screened by western blot and confirmed by sanger sequencing. Sequences of specific sgRNAs for targeting KDM4D are listed as the following: sgKDM4D#1: F: 5’-CACCGCCACGCCGATCTGCAGTTAG-3’; R: 5’-AAACCTAACTGCAGATCGGCGTGGC-3’. sgKDM4D#2: F: 5’-CACCGCGACCACGTCTGCCATTTGA-3’; R: 5’-AAACTCAAATGGCAGACGTGGTCGC-3’.

A total of 150 pairs primary ESCA tumor tissues and matched normal tissues were collected from patients who received surgical treatment at department of Thoracic Surgery, Henan Provincial People’s Hospital between July 2015 and February 2019. All patients were diagnosed by original histopathological detection. Samples of the collected tissues were preserved in liquid nitrogen and clinical information of patients were also collected. The collection of clinical specimens was approved by the ethics review committee of the Henan Provincial People’s Hospital, and every patient was informed of their consent.

To investigate the presence of KDM4D, HMGB1 and Ki67 proteins in tumor tissues or mice, the IHC assay was performed. The following antibodies were used in this assay: KDM4D with 1:4000 dilution; HMGB1 with 1:100 dilution. The human ESCA tissues and matched adjacent tissues used in this study were obtained from Henan Provincial People’s Hospital. These individuals did not receive any radiotherapy, chemotherapy or immunotherapy before surgery and were informed consent.

Flag-HMGB1 together with HA-Ub and Myc-SYVN1 plasmids were transfected into HEK293T, respectively. Ubiquitinated HMGB1 proteins were immunoprecipitated by anti-Flag affinity gel with an incubation of 10 μM MG132 for 6 h. Purified HMGB1 were obtained after elution with FLAG or MYC peptides and dialysis. They were incubated in a deubiquitination reaction buffer at 30°C for 2 h and analyzed by immunoblotting as previously described.

Eighteen 6 weeks old male BALB/c-nude mice were randomly divided into three groups, and subcutaneously injected with 6×10⁵ cancer cells into the right flank region of each mouse. After visible tumors had developed, tumor volume was measured and calculated as length × width2/2 every 2 days. The Institutional Animal Care and Use Committee of Chinese Academy of Henan Provincial People’s Hospital has approved these experiment procedures.

The two-tailed Student’s t-test was used to compare results from two groups. The Pearson correlation analysis was conducted to determine the correlation between two groups. Results from in vitro assays exhibited in the study are representative of three independent experiments. All quantitative data are indicated as Mean ± SD. GraphPad Prism Version 6.01 was utilized for statistical analysis. A P-value < 0.05 was considered to be statistically significant.

Given the essential roles of epigenetic remodeling in ESCC tumorigenesis, we intended to screen potential chromatin regulators that are indispensable for tumor progression. Firstly, we extracted the expression data of 562 epigenetic genes from the TCGA-ESCC cohort. The corresponding clinical characteristics of patients, like age, gender, pathological stages, TNM stages, were all collected and summarized in Table S1 . The univariate Cox regression analysis was conducted to screen 20 prognostic candidates with P < 0.01 as the cutoff ( Figures 1A, B and Table S2 ). Intriguingly, we found several oncogenic targets associated with ESCC, like UBE2B, TCF20 and RBBP7. Meanwhile, in line with previous studies, KDM4D, SMARCA4 and RAI1 were shown to be suppressors in ESCC. Besides, the underlying associations among the 12 essential epigenetic candidates were calculated and illustrated in the correlation heatmap ( Figure 1C ). To further identify the hub targets, we selected the top 12 candidates and designed specific siRNAs to perform a validation screen. As shown by MTT assay in KYSE150, KDM4D was the most potent hit ( Figure 1D ). Though targeting RAI1, SMARCA4 or KDM4D indeed resulted in elevated cell growth, KDM4D inhibition led to the most increase of growth rates compared with other candidates. In addition, we further used 2 different shRNA constructs to deplete KDM4D expression and found that reduced KDM4D expression significantly enhanced the growth of ESCC cells, including KYSE30 and EC109 ( Figure 1E ). The knockdown efficiency of KDM4D in two cell lines were confirmed by qPCR assays ( Figure 1F ). We collected the expression data of KDM4D and related survival data of patients. Kaplan-Meier analysis revealed that ESCC patients with high KDM4D had worse overall survival (OS) outcomes relative to those with low KDM4D levels ( Figure 1G ).

Accordingly, we collected tumors from totally 150 patients from the Henan Provincial People’s Hospital to determine the expression levels of KDM4D in ESCC. First of all, we found KDM4D expressed lowly in tumor samples relative to normal tissues ( Figure 2A ). Besides, low KDM4D correlated with advanced tumor grades, as evidenced by the IHC graphs ( Figure 2A ). In addition, we quantified the H-scores of KDM4D staining and calculated the ΔKDM4D across samples ( Figure 2B ). In line with these findings, the protein levels of KDM4D were further found to be significantly lower in 7/10 (70%) human ESCC tissues than in their paired normal tissues through western blot assays ( Figure 2C ). Lastly, we integrated the KDM4D expression data with clinical information and confirmed that KDM4D correlated negatively with hazard factors, including T stages, N stages and tumor grades ( Figure 2D ). Collectively, these data implicated that KDM4D expressed lowly in ESCC and correlated with hazard clinical parameters.

To further elucidate the functional roles of KDM4D in ESCC, we utilized the CRISRP/Cas9 technology to delete KDM4D in ESCC cells (KYSE30 and EC109), which was validated via western blot ( Figure 3A ). Besides, we utilized the lentivirus-mediated transfection to construct the KMD4D-overexpressing ESCC cells (KYSE30 and EC109), which was confirmed via western blot ( Figure 3B ). The up-regulation of KDM4D notably suppressed the clonogenic ability of ESCC cells (KYSE30 and EC109), as indicated by clone numbers ( Figure 3C ). Meanwhile, CCK-8 assays revealed that KDM4D overexpression attenuated the growth effect of cells (KYSE150) ( Figure 3D ), whereas KDM4D knockout significantly enhanced the growth of ESCC cells (KYSE30 and EC109) ( Figure 3E ). To determine the roles of KDM4D in ESCC metastasis, we observed that inhibition of KDM4D promoted the migration ability of KYSE30 and EC109 cells ( Figure 3F ). Interestingly, an increase in sphere numbers and sizes was observed in KDM4D-deficient KYSE30 and EC109 cells compared with the corresponding control cells, indicating that down-regulation of KDM4D could enhance ESCC self-renewal ability ( Figure 3G ). Considering that tumor stemness correlated tightly with drug resistance, we treated the ESCA cells with cisplatin and observed that KDM4D loss rendered cells more resistant to chemotherapy-induced apoptosis ( Figure 3H ). Taken together, our data indicated that KDM4D is a tumor suppressor and KDM4D inhibition could promote ESCC proliferation, migration, stemness features and resistance to chemotherapy.

Our previous data implicated that HMGB1 proteins were up-regulated in KDM4D-knockdown cells (KYSE30 and EC109) ( Figure 4A ). As reported, HMGB1 is an onco-protein that plays essential roles in tumorigenesis and stemness processes across multiple tumors, including ESCC. We designed specific shRNAs to knockdown HMGB1 in ESCC cells, which was validated by western blot and qPCR assays ( Figure 4B ). As expected, targeting HMGB1 indeed remarkably suppressed cell growth, as evidenced by the CCK-8 assay ( Figure 4C ). Besides, we inhibited HMGB1 by specific shRNAs in KDM4D-deficienct ESCC to conduct the CCK-8 assays. In line with our speculations, KDM4D deficiency could elevated cell growth rates (KYSE30, EC109 and KYSE150), which could be significantly suppressed when HMGB1 was down-regulated ( Figures 4D–F ). The clone formation efficiency of KDM4D-deficient ESCC could also be notably impaired with HMGB1 inhibition ( Figure 4G ). Similarly, we also conducted the migration assay and confirmed that HMGB1 knockdown could also suppressed the migration ability induced by KDM4D deficiency ( Figure 4H ). Lastly, we also found that KDM4D deficiency depended on HMGB1 to enhance stemness potentiality of EC109 cells. ( Figure 4I ). We further collected the ESCC samples and found the negative expression levels between KDM4D and HMGB1 via IHC assays ( Figure 4J ). Taken together, our findings implicated that low KDM4D depended on elevated HMGB1 to drive ESCC malignant features, including proliferation, migration or self-renewal abilities.

To elucidated the underlying mechanisms that KDM4D relies on to modulate HMGB1, we first detected the mRNA levels of HMGB1 in ESCC cells. Intriguingly, we found that KDM4D deficiency promoted the elevated HMGB1 protein levels. However, no alterations of mRNA levels of HMGB1 were detected in KDM4D-KO and control cells ( Figure 5A ). Of note, MG132, one well-known proteasome inhibitor, could significantly restored the protein levels of HMGB1 in KDM4D-deficient cells, but not the mRNA levels ( Figure 5B ). Briefly, we speculated that KDM4D may regulate HMGB1 proteins at the post-transcriptional level, especially through the ubiquitin-proteasome pathway. Given that KDM4D is a chromatin regulator, not the direct ubquitin ligase, we thus considered there might be other ligases that linked KDM4D and HMGB1. We queried the UbiBrowser website (http://ubibrowser.ncpsb.org.cn/ubibrowser/) to predict the essential E3 ubiquitin ligases that target HMGB1 ( Figure 5C ). After validations, we found that only SYVN1 inhibition could increase HMGB1 levels ( Figure 5D ). The endogenous interactions between HMGB1 and SYVN1 were confirmed by the Co-immunoprecipitation (Co-IP) analysis using the anti-HMGB1 antibody ( Figure 5E ). Importantly, SYVN1 mediated the polyubiquitination process of HMGB1, indicating that HMGB1 is the SYVN1 downstream substrate ( Figure 5F ). SYVN1 inhibition remarkably prolonged the half-life of HMGB1 protein in EC109 cells ( Figure S1A ). Meanwhile, positive associations between KDM4D and SYVN1 were confirmed in TCGA-ESCC cohort (r=0.36, P<0.0001, Figure 5G ). KDM4D loss could notably decrease SYVN1 mRNA levels, and KDM4D overexpression could elevate its mRNA levels relative to control cells ( Figure 5H ). Previous data implicated that KDM4D transcriptionally activates downstream targets via H3K9me3 demethylation at the promoter region (29, 30). ChIP-qPCR assay was performed to confirm that KDM4D deficiency could indeed increase the H3K9me3 binding abilities at the promoter of SYVN1 ( Figure 5I ). Additionally, KDM4D overexpression indeed enhanced the polyubiquitination levels of HMGB1, which could be notably impaired upon knockdown of SYVN1 ( Figure 5J ). Lastly, KDM4D overexpression decreased the HMGB1 protein levels, but SYVN1 inhibition restored the corresponding proteins ( Figure 5K ). Collectively, our findings indicated that KDM4D restricts HMGB1 protein levels via activating SYVN1 expressions.

Considering that KDM4D is an essential molecular biomarker in ESCC and no selective inhibitors are available, we thus wondered whether HMGB1 inhibitor could be effective to treat ESCC, especially the KDM4D^low tumors. Glycyrrhizin, a major constituent of licorice root, is a highly selective drug that targets HMGB1. The half maximal inhibitory concentration (IC50) values of Glycyrrhizin were first evaluated and calculated in three ESCC cell lines ( Figures 6A–C ). Then, we found that the cell growth ability of the ESCC cells (KYSE30, EC109 and KYSE150) were all remarkably impaired with Glycyrrhizin treatment in a dose‐dependent manner ( Figures 6D–F ). In line with the in vitro results, we also observed that Glycyrrhizin could effectively suppressed the elevated in vivo tumor growth induced by KDM4D deficiency, as indicated by the tumor volumes and weight ( Figures 6G–I ). Cumulatively, these results established that targeting HMGB1 (Glycyrrhizin) is proved to be an effective strategy to inhibit ESCC tumor growth, especially for the KDM4D^low cases.