Introduction

Esophageal squamous cell carcinoma (ESCC), the major type of esophageal cancers in China accounting for over 477,900 new cases and 375,000 deaths recorded annually, remains one of the most lethal of malignancies and a major health burden1. Although chemotherapy and radiotherapy can improve the disease outcome, relapse is frequent and treatment options and molecular markers are limited, leading to a poor prognosis with 20% of 5-year survival rates in China1,2,3.

Recent studies have profiled ESCC genomic alterations and have identified significantly mutated genes (SMGs) including TP53, ZNF750, NOTCH1, FAT1, NFE2L2, copy number amplifications occurring in SOX2, TERT, FGFR1, MDM1, and common deletions of RB1 etc4,5,6,7. Of these genes, ZNF750, a nuclear factor that plays a critical role in control of terminal epidermal differentiation gene program8,9, was frequently disrupted by somatic inactivating mutations on 17q25.34,6. ZNF750 has been reported to directly regulate FGF14 promoting cell apoptosis thus inhibited nasopharyngeal carcinoma10. Meanwhile, ZNF750 inhibited the malignant progression of oral squamous cell carcinoma by regulating tumor vascular microenvironment11. TINCR lncRNA, one of the downstream targets of ZNF750, participated in control of cellular proliferation, migration, and differentiation of SCC cells. ZNF750 also suppressed migration of SCC cells by directly inhibiting transactivation of LAMC212. Recent attempts to decipher the significance of ZNF750 for tumorigenesis in ESCC have revealed that ZNF750 might act as a tumor suppressor gene4,6. However, the molecular mechanism underlying ZNF750 contributes to tumorigenesis and the clinically relevant of genetic changes of ZNF750 in ESCC remain largely unresolved.

In this study, we reveal the associations ZNF750 mutations and/or deletions with clinical variables using genomic sequencing data of 612 pairs of ESCC tumor and normal samples from China and explore the molecular mechanism through which ZNF750 plays a critical role in driving the formation of metastatic ESCC. We found loss-function of ZNF750 evokes DANCR expression and prevents miR-4707-3p to interact with FOXC2 mRNA in a ceRNA manner, leading to enhanced FOXC2 signaling and angiogenesis phenotype in ESCC. Importantly, we performed RNA-sequencing of paired fresh tumor tissues and matched adjacent non-cancerous specimens from 97 ESCC subjects; together with available TCGA data, we validated the associations among ZNF750 and the identified downstream targets in ESCC and other squamous carcinomas. Our findings reveal an underlying mechanism by which loss-function of ZNF750 contributes to ESCC progression, provide a potential metastatic and prognostic biomarker and several therapeutic targets for ESCC patients harboring ZNF750 mutations and/or deletions.

Materials and methods

Samples and clinical data

Tumor samples and adjacent normal tissues with good quality and sufficient quantity for in-depth pathological and molecular investigation were obtained from 508 ESCC patients recruited from Shanxi and ** ratio. If alignment result passes QC, we proceeded the analyses including gene and isoform expression, deep analysis based on gene expression (PCA/correlation/screening differently expressed genes and so on), exon expression, gene structure refinement, alternative splicing, novel transcript prediction and annotation, SNV detection, indel detection. Based on differentially expressed genes (DEGs), we also performed deep analyses including gene ontology (GO) enrichment, pathway enrichment, cluster, protein-protein interaction network, and identification of transcription factor. The TargetScan (targetscan.org) and miRTarBase (mirtarbase.mbc.nctu.edu.tw) software packages were used to predict the targets of miRNAs.

Chromatin immunoprecipitation (ChIP)

ChIP was performed using a Chip assay kit as recommended by the manufacturer (Millipore, USA). Approximately 1 × 107 cells transfected with pcDNA3.1-HA ZNF750-wt were cross-linked in 1% formaldehyde at 37 °C for 10 min, and then 1.25 M glycine was added to quench the excess formaldehyde at room temperature for 5 min. Cells were washed twice in ice-cold PBS containing protease inhibitors, lysed in 200 μl of SDS lysis buffer for 10 min on ice, sonicated to break DNA into fragments of less than 1 kb and centrifuged at 12,000g for 15 min at 4 °C. The sonicated cell supernatants were subjected to immunoprecipitation (IP) with the following antibodies respectively, anti-HA (Abcam, USA, ab9110) or normal IgG (proteintech, USA, B900610). 1 mg of antibodies was used for each IP. After washing, immunoprecipitated DNA was eluted with elution buffer containing 1% SDS and 0.1 M NaHCO3, and the cross-links were reversed by incubation at 65 °C for 4 h in the presence of 200 mM NaCl and RNase A. After deproteinization with proteinase K, DNA was recovered by phenol-chloroform extraction and ethanol precipitation. Ordinarily PCR was performed to amplify the target DNA fragments which were visualized by agarose gel electrophoresis. The primers are listed in Table S1.

Dual luciferase reporter assay

The luciferase activities were measured using a luciferase reporter assay system according to the manufacturer’s instructions (Transgene, Bei**g, China). Briefly, cells were washed with ice-cold PBS and harvested in reporter lysis buffer. After centrifugation, 10 μl of the supernatants were mixed with 50 μl of solution1 and measured for firefly luciferase activity by using a TransDetect double-luciferase reporter assay kit (Transgene, China). Then mixed with 50 μl solution 2 and measured for Renilla luciferase activity.

RNA-binding protein immunoprecipitation (RIP)

RIP was performed to assess the interaction between DANCR and miR-4707-3p using a Magna RIP kit (Millipore, USA) according to the manufacturer’s instruction. Briefly, cells were lysed in RIP lysis buffer, and suitable proportion was stored as input. Cell lysate was incubated with RIP buffer containing magnetic beads conjugated with human anti-Ago2 antibody (Sigma, USA, HPA058075) or anti-IgG (proteintech, USA, 10283-1-AP), the negative control, at room temperature for 30 min, then washed twice with wash buffer. Cell lysates were incubated with the magnetic beads overnight at 4 °C, then the beads were collected and washed several times. During all the procedures, RNase inhibitor cocktail was added to prevent RNA degradation. RNA on the beads was extracted by the phenol-chloroform method according to the instruction and analyzed by qRT-PCR to demonstrate the presence of the binding targets.

Fluorescent in situ hybridization (FISH)

DANCR FISH staining was performed according to the manufacturer’s instruction. Briefly, cells were washed with 1× PBS for 5 min, fixed with 4% paraformldehyde at room temperature for 10 min, washed and permeabilization in 1 ml ice-cold 1×PBS with 0.5% Triton X-100, then subjected to pre-hybridization solution blocking the samples at 37 °C for 30 min. The 5 μM of anti-DANCR probe and anti-18s probe (RiboBio, China) were incubated in 100 μl hybridization solution at 37 °C overnight in a dark place. Samples were then washed three times for 15 min (5 min/each time) at 42 °C in a 4 × SSC solution with 0.1% Tween-20, once for 5 min in 2 × SSC, once for 5 min in 1 × SSC in a dark place, and once for 5 min at room temperature in PBS. DAPI in 50% glycerol was applied before imaging.

Statistical analysis

Sample size was chosen based on the need for statistical power. All statistical evaluations were performed using the SPSS 22 software package (IBS SPSS, Armonk, NY, USA) and GraphPad Prism 6 (GraphPad Software, San Diego, California, USA). Rank sum and Chi square (χ2) tests were used to analyze the association of ZNF750 mutations with clinical and pathological features. The follow-up period for progression-free survival was defined as the interval between the date of diagnosis of metastatic disease and that of progression disease. Survival curves were constructed using the Kaplan-Meier method and differences in survival were evaluated using the log-rank test. Overall survival (OS) was evaluated from the time of diagnosis to death or the final follow-up. Censored cases were defined as patients who lost contact during the follow-up and who were still alive at the end of the study. Univariate and multivariate survival analyses were performed by a Cox proportional hazards regression model.

All experiments were done in triplicates and data were presented as mean ± SD. Student’s t test was used for statistical analysis, and data from more than two groups were analyzed by one-way analysis of variance (ANOVA) followed by Dunnett’s test. The variance between the groups that are being statistically compared is similar. The correlation between ZNF750 and FOXC2 expression level was analyzed using nonparametric correlation (Spearman) in GraphPad Prism 6. A two-side P value < 0.05 was considered statistically significant.

Results

Recurrent losses of ZNF750 is associated with metastasis and poor prognosis in ESCC patients

In parallel, we profiled WGS on 508 pairs and RNA-sequencing on 97 pairs of ESCC, respectively. All FASTQ files are going to be uploaded to Genome Sequence Archive (GSA) in Bei**g Institute of Genomics (BIG) Data Center, the accession number is HRA000021, that will be publicly accessible at http://bigd.big.ac.cn/gsa. Notably, ZNF750 was identified as one of the most significantly mutated genes (SMGs, q value = 0) in this enlarged cohort, confirming that ZNF750 may play critical roles in the development of ESCC. A total of 30 somatic mutations and 19 indels were detected in CDS region of ZNF750 gene, that totally occurred in 48 out of 612 (7.84%) tumors (Fig. 1a). Of these mutations, 85.72% were inactivating mutations that include non-silent mutations (44.9%), indels (38.78%) and splice sites (2.04%), followed by missense mutations (14.29%) (Fig. S1a). Moreover, we observed recurrent mutations in noncoding regions of ZNF750, being mutated in 48 out of 508 (9.5%) tumors that have WGS data available. Of which, promoter mutations, that were found in 33 out of 508 patients (6.5%), were the most common somatic mutations (34.86% of total mutations), followed by intron (17.43% of total mutations) and untranslated regions (UTRs, 2.75% of total mutations) (Fig. S1b). Furthermore, analysis of focal copy number alterations (FCNA < 100 kb) across the 508 ESCC genomes revealed that ZNF750 was present in a significantly deleted focal peak around 17q25.3 (q value = 8.72E–05) and ZNF750 locus was to be deleted in 71 out of 508 tumors (14%, Fig. 1b, c), that was further validated by a real-time quantitative PCR (qRT-PCR) copy number assay (Fig. 1d, e).

Fig. 1: Genetic alterations of ZNF750 across 612 ESCC genomes.
figure 1

a The mutation sites and types in CDS region of ZNF750 across 612 ESCC genomes. b The significant focal SCNA filtered by GISTCI across 508 ESCC genomes. c Heatmap of CNV log2 ratio of read coverage across 71 ESCC individuals in 17q25.3 and ZNF750 regions (upper) and detected significant deletion of ZNF750 (bottom). d, e ZNF750 copy number was assayed by qRT-PCR in BDESCC0122, BDESCC0481, BDESCC0177 and BDESCC0189. RPPH1 was used as a normal reference, and TMC8 (located near this region) was used as a focal CNA control. f–h Kaplan–Meier survival analysis between ESCC patients with ZNF750-mut and ZNF750-wt, ESCC patients with ZNF750-wt and ZNF750-del, and ESCC patients with ZNF750-wt and ZNF750-Val (ZNF750-mut + ZNF750-del). The horizontal axis is the survival time and the vertical axis is the percentage of survival cases in total cases. Every cross on the survival curves stands for censored data. Log-rank test P-value, Cox regression P-value, Hazard Ratio and 95% CI are displayed on the graph. i The mRNA of ZNF750 was significantly decreased in ESCC tissue compared with paired normal tissue in 97 cases. Data represent the mean ± SD. All assays were performed in triplicate. Statistical analysis was performed with a two-side paired-t test. **p < 0.01, ***p < 0.0001.

Full size image

Previously, we performed 104 pairs of WGS/WES from ESCC patients4, and then combined this cohort with the 508 pairs of WGS cohort (n = 612). To further estimate the associations between ZNF750 mutations and/or deletions with clinicopathological features, rank sum and Chi-square (χ2) tests were applied in this enlarged cohort (n = 612). Although most features were not statistically significant, this analysis identified the following statistically significant patterns: (i) ZNF750 mutations (n = 48 tumors) were significantly associated with late pathological stage (P = 0.000078), (ii) patients harboring ZNF750 mutations exhibited more lymph node metastasis (P = 0.00011) and more distant metastasis (P = 0.001) compared with patients harboring no mutations in ZNF750 (Table 1), and (iii) patients in subgroup of “ZNF750-wt” showed significantly better survival as compared to patients in subgroup of “mutated” (Kaplan–Meier analysis, P = 0.026; Cox regression, P = 0.029, hazard ratio (HR): 1.541, 95% confidence interval (CI): 1.044–2.261) (Fig. 1f). Strikingly, analysis of copy number alterations in 508 pairs of WGS revealed that ZNF750 copy number loss (n = 71) was statistically associated with more lymph node metastasis (P = 0.046) compared with patients harboring no deletions in ZNF750. Subsequently, Kaplan–Meier’s analysis revealed that patients with ZNF750 copy number loss developed more-frequent recurrence and had poorer survival (P = 0.048, Fig. 1g and Table S2). Cox’s proportional hazards regression analysis indicated that ZNF750 deletion was an independent predictor for the ESCC prognosis. When ZNF750 mutations and deletions were combined, the associations with late stage (P = 0.006), more lymph node metastasis (P = 0.000453), and poorer survival (P = 0.016) were highly significant (Fig. 1h and Table S3). Additionally, across 97 pairs of RNA-seq dataset, ZNF750 expression level was significantly decreased in tumor samples in comparison with matched normal samples (Fig. 1i). Together, these findings suggest that there are clear associations between ZNF750 mutations/deletions and clinical characteristics of Chinese ESCC patients, supporting the clinical potential of ZNF750 mutations/deletions with its ability to estimate the metastasis and survival of patients with ESCC.

Decreased ZNF750 promoted tumor angiogenesis in ESCC

The pattern and significance of ZNF750 genetic alterations and expression promote us to investigate the correlation of ZNF750 with tumor metastasis in ESCC. Our previous study had showed ZNF750 inhibited the proliferation and invasion of ESCC cells4. Angiogenesis plays a critical role in the growth and spread of cancer. We then explored the functional relevance of ZNF750 with tumor angiogenesis in ESCC. We expressed ZNF750-sh in KYSE140 (Fig. S2a, b), and detected its effect on EA.hy926 cells, one human umbilical vein endothelial cell (HUVEC) line. We found that the proliferation rates of HUVEC were not significantly increased when treated with conditioned medium from KYSE140 cells with ZNF750 knockdown. However, ZNF750-knockdown KYSE140 cell strongly promoted their ability of migration, invasion and tube formation in HUVEC cells (Fig. 2a–e). Similar results were observed in KYSE180 cells and KYSE450 cells (Fig. S2). In contrast, forced expression of ZNF750-wt in KYSE150 cells dramatically reverted the angiogenesis phenotypes of migration, invasion and tube formation of HUVEC cells in vitro (Fig. 2f–j). Consistent patterns were observed in HAEC, another human arterial endothelial cell line (Fig. S3).When transplanted ZNF750 overexpression ESCC cells into subcutaneous, the tumor had a lower microvessel density compared with the control cells (Fig. 2k, l) and decreased expression of CD31 and FLT1, two markers of angiogenesis phenotypes (Fig. 2m, n). Consistently, based on the TCGA data, a negative correlation between ZNF750 and the expression of CD31 and FLT1 was observed in LUSC, CESC and HNSC, respectively (Fig. 2o). Collectively, these results indicate that ZNF750 may be a critical suppressor of tumor angiogenesis to regulate metastasis of ESCC.

Fig. 2: Decreased ZNF750 promoted tumor angiogenesis in ESCC.
figure 2

a Representative images of conditioned medium of ZNF750 knockdown cells promoted the tube formation (1st column), migration (2nd column), and invasion (3rd column) of HUVEC. b–e The conditioned medium of ZNF750 knockdown cells promoted cell angiogenesis (c), migration (d) and cell invasion (e) but had no effect on cell proliferation (b) of HUVEC. f Representative images of conditioned medium of ZNF750 overexpression cells inhibited the tube formation (1st column), migration (2nd column) and invasion (3rd column) of HUVEC. g–j The conditioned medium of ZNF750 overexpression cells inhibited the tube formation (h), migration (i) and invasion (j) but had no effect on cell proliferation (g) of HUVEC. Data represent the mean ± SD. All assays were performed in triplicate. Statistical analysis was performed with a two-sided t test. k, l Representative immunostaining images show the microvessel density of tumor tissues from xenograft mouse injected with KYSE150NC more than that with ZNF750-wt cells (bar = 100 μm). m, n FLT1 were significantly reduced in ZNF750-wt animals. P values were obtained using non-paired t test. o Correlation analysis between ZNF750 and CD31,FLT1 in LUSC, CESC, HNSCC from TCGA. The horizontal axis is the expression of ZNF750 mRNA and the vertical axis is the expression of CD31 and FLT1. Statistical analysis was performed with pearson correlation. *p < 0.05, **p < 0.01, *** p < 0.001.

Full size image

FOXC2 as a mediator of tumor angiogenesis induced by decreased ZNF750

Angiogenesis has been implicated as an essential component of tumor metastasis. To explore the potential molecular mechanism underlying decreased ZNF750 induced angiogenesis and metastasis, we performed PCR array in ZNF750 knockdown cells using the kit of Cancer Pathway Finder PCR Array and validated by qRT-PCR. The results showed that angiogenesis pathway was the top altered pathway, and FOXC2, a critical transcription factor regulating tumor angiogenesis and inducing epithelial-mesenchymal transition in various cancers, was significantly affected (Fig. 3a). qRT-PCR confirmed the increased expression of FOXC2 in ZNF750 knockdown cells and the decreased expression in ZNF750-wt forced expression KYSE150 cells (Fig. 3b, c). Furtherly, we found FOXC2 knockdown reversed the effect on angiogenesis in ZNF750 knockdown cells. Importantly, knockdown FOXC2 in ZNF750 knockdown KYSE140 cells decreased the ability of migration, invasion and tube formation in HUVEC cells treated with conditioned medium (CM) from ZNF750 knockdown KYSE140 cells and KYSE450 cells, respectively (Fig. 3d, e and Fig. S4a, b). Similar patterns were observed in HAEC (Fig. S4c–f). As we know, FOXC2 has a paracrine effect on angiogenesis via regulation of its target gene angiopoietin 2 (Ang-2) and VEGFs13. We then detected the effect of ZNF750 and FOXC2 on Ang-2 and VEGFs, and found ZNF750 knockdown induced up-regulation of Ang-2 and VEGFs in ESCC cells whereas downregulated FOXC2 reversed the phenomenon (Fig. 3f). Thus, FOXC2 may act as a mediator of tumor angiogenesis induced by ZNF750 knockdown in ESCC.

Fig. 3: FOXC2 as a mediator of tumor angiogenesis induced by decreased ZNF750.
figure 3

a PCR array in ZNF750 knockdown cells using the kit of Cancer Pathway Finder PCR Array. b, c qRT-PCR showed the expression of FOXC2 in ZNF750 knockdown cells (b) and ZNF750-wt cells (c). GAPDH was performed as a loading control. Statistical analysis was performed with a two-sided t test. d Representative images of tube formation (upper), migration (middle) and invasion(lower) of HUVEC. e The conditioned medium from FOXC2 knockdown in ZNF750 knockdown cells reversed the effect on tube formation (left), migration (middle) and invasion(right) of HUVEC. f Angiogenic biomarkers ANGPT2, FLT1, CCL2 and FGF2 were detected by qRT-PCR in ZNF750 knockdown cells and FOXC2 knockdown in ZNF750 knockdown cells. Statistical analysis was performed with one-way ANOVA. **p < 0.01, ***p < 0.0001.

Full size image

A long noncoding RNA (lncRNA) DANCR may be a downstream target of ZNF750

Although there was a putative binding site of ZNF750 on FOXC2 promoter, Chromatin immunoprecipitation (ChIP) result showed ZNF750 could not bind its promoter region directly (Data not shown). To explore how ZNF750 contributes to regulation of FOXC2 mRNA expression in ESCC, we performed RNA-sequencing to detect global alterations at mRNA level upon ZNF750 overexpression in KYSE150 cells that have relatively low endogenous ZNF750 level. Pathway enrichment analysis displayed that the differentially expressed genes (Fisher-exact test, P < 0.05) were enriched in pathways such as PI3K/Akt signaling, Hippo signaling, regulation of actin cytoskeleton and proteoglycans in cancer (Fig. 4a). Next, we integrated our RNA-seq data with Boxer’s ChIP-sequencing data that explored the potential direct targets of ZNF750 in keratinocytes14 and found 64 genes occurring simultaneously (Fig. 4b). Of these genes, 19 were identified containing putative ZNF750 binding sites within 2000 bp regions upstream of the transcription start site (TSS) (Fig. S5a). Six genes were downregulated upon ZNF750 overexpression whereas up-regulated in ZNF750 knockdown cells as demonstrated by qRT-PCR, that was consistent with RNA-seq data. Moreover, five genes (HERC6, SPOCK2, CARD14, SEMA6D, and G6PD) were positively correlated with ZNF750 expression level (Fig. S5b, c) whereas the lncRNA DANCR showed a negatively association with ZNF750 (Fig. 4c). In particular, DANCR has been implicated as an anti-differentiation ncRNA and was required for dedifferentiation of epidermal cells15,16,17. We then examined the direct interaction between ZNF750 and DANCR via ChIP and dual luciferase reporter assays in KYSE140 cells that harbor ZNF750-wt genotype. ChIP result demonstrated that ZNF750 bound to the promoter region from -342 to -334 upstream of TSS of DANCR (Fig. 4d). The dual luciferase assay showed that ZNF750 overexpression significantly decreased the luciferase activity of DANCR promoter containing ZNF750 binding site, while the inhibitory effect was attenuated when the site was mutated or deleted (Fig. 4e), suggesting that ZNF750 directly bound to the promoter of DANCR, leading to an inhibition of DANCR expression.

Fig. 4: DANCR is a critical downstream target of ZNF750 to regulate FOXC2 expression in a ceRNA manner.
figure 4

a Pathway enrichment analysis displayed that the differentially expressed genes (Fisher-exact test, p < 0.05) were enriched in pathways such as PI3K/Akt signaling, Hippo signaling, regulation of actin cytoskeleton and proteoglycans in cancer. b 64 genes as potential direct targets of ZNF750 occurring simultaneously in RNA-seq data and Boxer’s ChIP-sequencing data. c qRT-PCR showed the expression of DANCR in ZNF750 knockdown cells (left) and ZNF750-wt cells (right). GAPDH was performed as a loading control. Statistical analysis was performed with a two-sided t test. d Upper: A putative ZNF750 binding site on the -342 to -334 region before transcriptional start site of DANCR. Lower: ChIP-PCR showed that ZNF750 directly bound to the promoter of DANCR. e Dual luciferase assay showed that ZNF750 overexpression decreases the luciferase activity of DANCR reporter. This effect was abrogated when the ZNF750 binding site was mutated or deleted. P values were obtained using ANOVA. f DANCR localization in both cytoplasm and nuclear of KYSE150 cells as measured by RNA-FISH. DAPI labels the nucleus. Scale bars represent 100 μm. g The putative binding site of DANCR was found to be similar to the binding site of miR-4707-3p on 3’UTR of FOXC2. h Dual luciferase report assay showed miR-4707-3p mimics significantly reduced the luciferase activity of pSicheck2-DANCR-wt rather than pSicheck2-DANCR-mut. Statistical analysis was performed with a two-sided t test. (i) qRT-PCR showed the expression of DANCR in miR-4707-3p inhibition cells. j RIP showed DANCR and miR-4707-3p were presented as fold enrichment in Ago2 relative to IgG immunoprecipitates. Statistical analysis was performed with a two-sided t test. k Dual luciferase assay showed that miR-4707-3p mimics decreased the luciferase activity of FOXC2-3’UTR reporter rather than mutant FOXC2-3′UTR reporter. P values were obtained using ANOVA. l, m Augmentation of miR-4707-3p led to a decrease of FOXC2 expression level whereas silence of miR-4707-3p caused an increase as measured by qRT-PCR (l) and western blot (m). Statistical analysis was performed with a two-sided t test. *p < 0.05, **p < 0.01,***p < 0.0001.

Full size image

DANCR may be a critical downstream target of ZNF750 to regulate FOXC2 expression in a ceRNA manner

The predominant cytoplasm localization of DANCR in KYSE150 cells (Fig. 4f) suggests that DANCR tends to participate in post-transcriptional regulation by interacting with miRNAs. Interestingly, we found DANCR and the 3′-UTR of FOXC2 mRNA had the similar binding sites (or called microRNA response element, MRE) of miR-4707-3p using miRTarBase and TargetScan (Fig. 4g). We co-transfected pSicheck2-DANCR-wt or -mut plasmid with miRNA-expression plasmids and subjected to luciferase report assay. Notably, we observed a 67% reduction of luciferase activity in miR-4707-3p expressed cells compared to that of empty vector control, meanwhile the suppression of luciferase activity was dramatically abolished by DANCR-mut expression(Fig. 4h), indicating that miR-4707-3p probably binds to the predicted MRE region of DANCR, thus downregulates DANCR expression (Fig. 4i). Moreover, RNA-Binding Protein Immunoprecipitation Assay (RIP) experiment using antibody against Ago2 in KYSE150 cell extracts showed that DANCR and miR-4707-3p were preferentially enriched in Ago2-containing miRNPs relative to the control IgG immunoprecipitates (Fig. 4j), indicating that DANCR, likely through interacting with miR-4707-3p, was present in Ago2-containing miRNPs. Similarly, as expect, FOXC2 expression was negatively regulated by miR-4707-3p at the transcriptional level and post-transcriptional level. Silence of miR-4707-3p led to an increase of FOXC2 mRNA and protein level in KYSE-150 cells whereas augmentation of miR-4707-3p caused a decrease of FOXC2 in Eca109 cells (Fig. 4l, m). Dual luciferase reporter assay showed that miR-4707-3p inhibited the luciferase activity of FOXC2-3′UTR but had no significant influence on the luciferase activity of FOXC2-3′UTR-mut (Fig. 4k).

Considering that both DANCR and FOXC2 mRNA interact with miR-4707-3p, we then hypothesized that DANCR might function as a ceRNA for FOXC2 to be modulated by miR-4707-3p in ESCC. To further explore this possibility, we detected the association between DANCR and FOXC2 expression and found that DANCR knockdown resulted in a decrease of FOXC2 expression whereas DANCR overexpression led to an increased expression of FOXC2 in vitro (Fig. 5a). Additionally, we also found FOXC2 expression was inhibited by miR-4707-3p, that was reversed when exogenous DANCR was expressed (Fig. 5b), suggesting that DANCR may compete with FOXC2 to be targeted by miR-4707-3p in ESCC cells.

Fig. 5: ZNF750 may act as a tumor suppressor via DANCR/miR-4707-3p/FOXC2 axis that works in a ceRNA manner in ESCC.
figure 5

a, b DANCR silence led to a decrease of FOXC2 protein level whereas augmentation of DANCR caused an increase in vitro (a), and the inhibition of FOXC2 protein was partially abolished by DANCR-wt overexpression in KYSE150-miR-4707-3p mimics cells (b). β-Actin was used as control. Statistical analysis was performed with a two-sided t test and one-way ANOVA. c Dual luciferase assay showed that the inhibitory effect of miR-4707-3 on FOXC2-3’-UTR reporter was partially abolished by DANCR-wt overexpression. P values were obtained using ANOVA. d–f Immunohistochemistry (d, e) and qRT-PCR (f) showed the expression of FOXC2 in DANCR knockdown or overexpression xenograft mouse samples. GAPDH was used as loading control. Statistical analysis was performed with a two-sided t test. g–j Western blot (g, h) and qRT-PCR (i, j) showed the expression of DANCR, miR-4707-3p, FOXC2 in ZNF750 knockdown and overexpression cells. β-Actin, GAPDH and U6 were used as loading control respectively. Statistical analysis was performed with a two-sided t test. (k–m) Immunohistochemistry (k, l) and qRT-PCR (m) were used to detect the expression of DANCR, FOXC2 in ZNF750 overexpression xenograft mouse samples. GAPDH was used as loading control. Statistical analysis was performed with a two-sided t test. *p < 0.05, **p < 0.01, *** p < 0.001.

Full size image

To further confirm the ceRNA regulatory mechanism among DANCR, miR-4707-3p and FOXC2, dual luciferase reporter assay was applied. As shown in Fig. 5c, the luciferase activity of FOXC2-3′UTR was inhibited by miR-4707-3p, this effect was attenuated by DANCR-wt expression. Moreover, the luciferase activity of FOXC2-3′UTR was impaired in DANCR-wt and miR-4707-3p mimics co-transfection group but not in DANCR-mut and miR-4707-3p mimics co-transfection group, indicating that the direct interaction between DANCR-wt and miR4707-3p is critical to abolish the negative regulation of miR4707-3p on FOXC2. We also found DANCR was positively with the expression of FOXC2 in xenograft mouse models at protein level (Fig. 5d, e) and mRNA level (Fig. 5f). Together, this observation demonstrates that DANCR may act as a ceRNA to compete binding to miR-4707-3p, thus attenuates the inhibition of miR-4707-3p on FOXC2 mRNA expression in ESCC cells.

ZNF750 may act as a tumor suppressor via DANCR/miR-4707-3p/FOXC2 axis in ESCC

Given that the DANCR/miR-4707-3p/FOXC2 axis in ESCC, we hypothesize ZNF750 may directly downregulate DANCR expression, strengthened the interaction of miR-4707-3p with FOXC2-3′UTR in a ceRNA manner, leading to degradation of FOXC2 mRNA, thus playing tumor suppressive role in ESCC. To further validate this hypothesis, western blot and qPCR were used to detect the expression of DANCR, miR-4707-3p, FOXC2 in ZNF750 knockdown or overexpression cells and xenograft mouse samples. Our results showed that ZNF750 overexpression caused a remarkable decrease of DANCR and FOXC2 with a significant increase of miR-4707-3p whereas ZNF750 knockdown led to a remarkable increase of DANCR and FOXC2 accompanied by a significant reduction of miR-4707-3p in vitro (Fig. 5g–j). The IHC results also showed FOXC2 was negatively with the expression of ZNF750 in xenograft mouse models at protein level (Fig. 5k, l) and mRNA level (Fig. 5m). In addition, we performed IHC in 508 ESCC tissues and paired normal esophagus tissues. As Fig. S7 shown, ZNF750 was markedly decreased in tumors compared to that of normal esophagus tissues, while FOXC2 was markedly increased in tumors compared to that of normal esophagus tissues. The results were consistent with the pattern observed in xenograft mouse models. Together, our results suggest that ZNF750 plays a tumor suppressive role via DANCR/miR-4707-3p/FOXC2 axis that works in a ceRNA manner in ESCC.

ZNF750 was negative correlated with FOXC2 in multiple types of human squamous cell carcinoma

Finally, we explored the correlation of ZNF750 with FOXC2 expression level across various human squamous cell carcinomas. We analyzed their transcriptional level in ESCC based on our 95 pairs of RNA-seq data and found that ZNF750 mRNA expression was negatively correlated with FOXC2 mRNA level (P = 1.2931E-20, Fig. 6a, b). Additionally, we also conducted the correlation analysis of ZNF750 with FOXC2 transcriptional level in ESCC and other squamous cell carcinomas based on the TCGA database. Consistent with our paired ESCC samples, we found the negatively correlation of ZNF750 mRNA with FOXC2 mRNA in 92 ESCC cases (r = −0.2151, P = 0.0395), 513 cases of head and neck squamous cell carcinoma (r = −0.2211, P < 0.0001), 493 cases of squamous cell carcinoma of lung (r = −0.1327, P = 0.0031), and 300 cases of cervical squamous cell carcinoma (r = −0.1518, P = 0.0085), suggesting a possibility of ZNF750 deregulates FOXC2 expression in human squamous cell carcinoma (Fig. 6c–e).These datas indicate a possibility of ZNF750 deregulates FOXC2 expression in human squamous cell carcinoma. Together with our observations in vitro and in vivo, our findings suggest that ZNF750, a significantly mutated driver gene in ESCC, may regulate tumor angiogenesis via DANCR/miR-4707-3p/FOXC2 axis that works in a ceRNA manner (Fig. 6f).

Fig. 6: ZNF750 is negative correlated with FOXC2 in squamous cell carcinoma.
figure 6

a The log2 ratio heatmap of differences expression of ZNF750 and FOXC2 between ESCC tissues and paired normal tissues. Red represents up-regulation and blue represents down-regulation. b Correlation of differences expression between ZNF750 mRNA and FOXC2 mRNA. The horizontal axis is the log2 ratio of FOXC2 and the vertical axis is the log2 ratio of ZNF750. Red dot showed the same differences expression trend between ZNF750 and FOXC2 mRNA. Blue dot showed the opposite trend between them. c–e Correlation analysis between ZNF750 mRNA and FOXC2 in 92 ESCC samples, 513 HNSC samples, 494 LUSC and 300 CESC samples from TCGA. The horizontal axis is the expression of ZNF750 mRNA and the vertical axis is the mRNA expression of FOXC2. f Diagram showing how loss-function of ZNF750 contributes to tumorigenesis of ESCC via DANCR/miR-4707-3p/FOXC2 axis that works in a ceRNA manner.

Full size image

Discussion

In this study, we profiled mutation and copy number alterations of ZNF750 and uncovered its potential prognostic value for ESCC patients in an enlarged ESCC cohort. We applied genome sequencing approaches, in vitro and in vivo methods to reveal the tumor suppressor role of ZNF750 and a possible novel mechanism underlying malignant phenotypes caused by loss-function of ZNF750 in ESCC. Our data indicate that, for the first time, ZNF750 copy number losses may be an attractive biomarker for risk metastasis and prognostication and ZNF750 loss-of-function promotes tumorigenesis of ESCC via DANCR/miR-4707-3p/FOXC2 axis in a ceRNA manner.

In view of previous whole-genome or -exome sequencing studies4,5,6,7, including our own, interrogated only a small number of samples, the global molecular portrait of ESCC remains incomplete. Profiling larger numbers of samples can help to identify cancer driver genes by providing the additional statistical power. Hence, our 612 WGS data could help us to understand the drivers of clinical phenotypes better. Consistent with previous reports4,6,7,18,19, ZNF750 was identified as one of most SMGs in this enlarged cohort. Specially, more than 10% of 508 ESCC cases had focal copy number deletion, representing the most frequent molecular event reported on ZNF750 gene. Together with several lines of functional evidences, ZNF750 may work in a loss-of-function manner and its dysregulation may be crucial for tumor formation and progression in ESCC.

Clinically, we observed a significant correlation of patient ZNF750 genotype with tumor phenotype. We found that individuals with mutations in ZNF750 had a more later stage, more lymph node metastasis and much worse prognosis than individuals without mutations. And the results of ZNF750 CNV were similar. Therefore, the correlation of ZNF750 mutations and deletion with advanced disease stage and poor clinical outcome may explain the aggressive behavior of the ZNF750 mutated ESCC. Although previous studies reported several potential biomarkers associated with clinical features especially patients’ outcome5,7, the statistics power of these biomarkers became weaker in further larger validate cohort, resulting in the clinical significance of genetic abnormalities of ESCC remains poorly defined and no reliable molecular markers for diagnosing or predicting patient outcomes have been identified to date20. Given that the greatly increased statistical power with our 612-case population, our finding highlights the clinical potential of ZNF750 for Chinese ESCC patients and warrant further clinical investigation through prospective randomized clinical trials to confirm the application.

Currently, the treatment of ESCC relies on surgery, chemotherapy, radiotherapy, or combinations of these2,21, but limited on effective molecularly targeted therapies that may attribute to the precise molecular events underlining ESCC formation and metastasis remain only partially understood. In this study, we found that loss-function of ZNF750 significant promoted tumor angiogenesis, which indicates anti-angiogenesis might be an efficient method in the inhibition of the growth and metastasis of ESCC with ZNF750 mutation or deletion. Furtherly, we found the effect of ZNF750 is via DANCR/miR-4707-3p/FOXC2 axis in a ceRNA manner in ESCC. Recent studies have implicated that DANCR might act as an oncogenic lncRNA in tumor progression such as hepatocellular carcinoma18,22, prostate cancer23, colorectal cancer24, glioma25, gastric cancer26, osteosarcoma27, lung adenocarcinoma28. In agreement with previous reports on various types of human cancers, DANCR may also act as an oncogene in ESCC progression, supporting that DANCR may offer a potential therapeutic target for those ESCC patients harboring mutations of ZNF750.

It is well known that lncRNAs can affect cellular behavior by diverse mechanisms29,30,31, for example, direct interaction with microRNAs. In this study, the interaction between DANCR and miR-4707-3p was found to regulate FOXC2 expression in a ceRNA manner. ZNF750 loss-of-function led to increased DANCR expression level that competed to bind to miR-4707-3p, attenuated the degradation of oncogene FOXC2 by miR-4707-3p, thus promoted angiogenesis in ESCC cells. MiR-4707-3p was previously rarely reported in human cancers. Functionally, knockdown of miR-4707-3p promoted tumor angiogenesis probably by releasing the oncogenic role of FOXC2. FOXC2 has been implicated as an oncogene promoting tumor invasion and metastasis in various of human cancers, for example, it mediated epithelial-mesenchymal transition32, multidrug resistance33, or regulator of vascular development in basal-like breast cancer, non-small cell lung cancer34, colon cancer35. FOXC2 also has a paracrine effect on angiogenesis via regulation of its’ target gene Angiopoietin 2 (Ang-2) and VEGFs13. Our functional results also support the oncogenic roles of FOXC2 in ESCC. Hence, besides the DANCR, our study provides evidence to support additional potential therapeutic target, FOXC2, for those ESCC patients with ZNF750 mutations.

In summary, our study, for the first time, explore a novel DANCR/miR-4707-3p/FOXC2 regulatory pathway through which ZNF750, a significant mutated driver gene, may deregulate angiogenesis involving in ESCC tumorigenesis and progress. We also highlight its potential to impact clinical outcomes of patients and offer potential therapeutic targets for ESCC treatment. Further efforts are required to exploit this information to develop prognostic method and to identify therapeutic targets that could be used to treat biomarker-selected groups of patients with ESCC.