FOXC1 promotes HCC proliferation and metastasis by Upregulating DNMT3B to induce DNA Hypermethylation of CTH promoter

This study was approved by the Ethics Committee of Tongji Medical College. All patients provided full consent for the study. Cohort I included 280 adult patients with HCC who underwent curative resection between 2003 and 2005 at the Tongji Hospital of Tongji Medical College (Wuhan, China). Cohort II included 210 adult patients with HCC who underwent curative resection between 2006 and 2008 at the Tongji Hospital of Tongji Medical College (Wuhan, China). A preoperative clinical diagnosis of HCC was based on the diagnostic criteria of the American Association for the Study of Liver Diseases. The inclusion criteria were as follows: (a) distinctive pathologic diagnosis; (b) no preoperative anticancer treatment or distant metastases; (c) curative liver resection; and (d) complete clinicopathologic and follow-up data. The differentiation statuses were graded according to the method of Edmondson and Steine. The pTNM classification for HCC was based on The American Joint Committee on Cancer/International Union Against Cancer staging system (6th edition, 2002). Follow-up data were summarized at the end of December 2013 (Cohort I) and December 2016 (Cohort II, range 4–96 months) respectively. The patients were evaluated every 2–3 months during the first 2 years and every 3–6 months thereafter. All follow-up examinations were performed by physicians who were blinded to the study. During each check-up, the patients were monitored for tumor recurrence by measuring the serum AFP levels and by performing abdominal ultrasound examinations. A computed tomography and/or magnetic resonance imaging examination was performed every 3–6 months, together with a chest radiographic examination. The diagnostic criteria for HCC recurrence were the same as the preoperative criteria. The time to recurrence and overall survival were the primary endpoints. The time to recurrence was calculated from the date of resection to the date of a diagnosis with tumor recurrence. The overall survival was calculated from the date of resection to the date of death or of the last follow-up.

BALB/C nude mice (5 weeks old) were housed under standard conditions and cared for according to the institutional guidelines for animal care. All animal experiments were approved by the Committee on the Use of Live Animals in Teaching and Research (CULATR), Huazhong University of Science and Technology. For in vivo metastasis assay, human luciferase labeled HCC cells (4.0 × 10⁶) in the 100 μl of phosphate-buffered saline that were mixed with 100 μl Matrigel were injected into the right lobes of livers of the nude mice under anesthesia (10 for each group). The in vivo tumor formation and metastases were monitored using the bioluminescence. For in vivo signal detection, D-luciferin (Perkin-Elmer) at 100 mg/kg was injected intraperitonially into the nude mice. Bioluminescent images were captured using an IVIS 100 Imaging System (Xenogeny). At the 9 weeks, the mice were sacrificed and the livers and lungs were collected and underwent histological examination.

Plasmid construction was performed according to standard procedures as outlined in our previous study. The primers are presented in Supplementary Table S7. For instance, the DNMT3B promoter construct, (− 1303/+ 109) DNMT3B, was generated from human genomic DNA. This construct corresponds to the sequence from − 1303 to + 109 (relative to the transcriptional start site) of the 5′-flanking regions of the human DNMT3B gene. It was generated with forward and reverse primers incorporating KpnI and HindIII sites at the 5′ and 3′-ends, respectively. The polymerase chain reaction (PCR) product was cloned into the KpnI and HindIII sites of the pGL3-Basic vector (Promega). The 5′-flanking deletion constructs of the DNMT3B promoter, (− 924/+ 109) DNMT3B, (− 808/+ 109) DNMT3B, (− 195/+ 109) DNMT3B, were similarly generated using the (− 1303/+ 109) DNMT3B construct as the template. The FOXC1 binding sites in the DNMT3B promoter were mutated using the QuikChange II Site-Directed Mutagenesis Kit (Stratagene). The constructs were confirmed by DNA sequencing. Other promoter constructs were cloned in the same manner.

Genomic DNA was isolated from cells using Genomic DNA Purification kit following the manufacturer’s instructions (Promega). Bisulfite modification of genomic DNA was carried out using the EZDNA methylation Kit (Zymo Research). Briefly, 1 μg of genomic DNA was denatured by NaOH (final concentration, 0.2 mol/L) for 10 min at 37 °C. Hydroquinone (10 mmol/L, 30 μl) and 520 μl of 3 mol/L sodium hydroxide (pH 5) were added, and samples were incubated at 50 °C for 16 h. Modified DNA was purified using Wizard DNA Clean-Up System following the manufacturer’s instructions (Promega) and eluted into 50 μl water. DNA was treated with NaOH (final concentration, 0.3 mol/L) for 5 min at room temperature, ethanol precipitated, and resuspended in 20 μl water. Modified DNA was used immediately or stored at − 20 °C. Primer sequences specific to unmethylated and methylated promoter sequences are listed in Table S6. Each methylation-specific PCR reaction incorporated 100 ng of bisulfite-treated DNA as template, 10 pmol/L of each primer, 100 pmol/L deoxynucleoside triphosphate, 10 PCR buffer, and 1 unit of JumpStart Red Taq Polymerase (Sigma-Aldrich, St. Louis, MO) in a final reaction volume of 25 μl. Cycle conditions were as follows: 95 °C 5 min; 35 cycles (95 °C 30 s, 60 °C 30 s, and 72 °C 30 s); and 72 °C 5 min. Methylation-specific PCR products were analyzed with nondenaturing 6% polyacrylamide gel electrophoresis and stained with ethidium bromide.

For bisulfite cloning and sequencing, the amplification was performed using primers designed by the online program (http://​www.​urogene.​org/​cgi-bin/​methprimer/​methprimer.​cgi). Bisulfite genomic sequencing was performed to characterize the methylation density in the promoter of CTH using the BigDye Terminator Cycle Sequencing kit version 1.0 (Applied Biosystems). Thirteen CpG sites spanning − 260 to − 20 of the CTH gene were evaluated. Sequences were analyzed by using SeqScape software (Applied Biosystems) and Bioedit (http://​www.​mbio.​ncsu.​edu/​BioEdit/​bioedit.​html). The nucleotide sequences of the primers used for BGS are provided in Supplementary Table S7.

The 2′,7′- dichlorofluorescein diacetate (DCFH-DA) (Beyotime, China) was used to detect the ROS levels. Firstly, added DCFH-DA to DMEM (without FBS) in a ratio of 1:1000, and then a volume of 2 mL mixed solution was added to the HCC cells, which were washed 2 times by PBS. Then we put the cells into incubator for 30 min at 37 °C in dark. Then HCC cells were washed again with DMEM (without FBS) for three times, we measured the ROS levels by fluorescence microscope and flow cytometry.

The GSH/GSSG ratio, GSH levels and cysteine levels were detected by Reduced glutathione (GSH) assay kit (Nanjing jiancheng, China), Total glutathione / Oxidized glutathione assay kit (Nanjing jiancheng, China) and Cysteine content test kit (Nanjing jiancheng, China) on the basis of the manufacturer’s instructions. The transfected cells were lysed in culture dishes containing a lysis buffer, and 0.5 ml supernatant was taken from the resulting lysates which were centrifuged. Two ml of the application solution was added and mixed evenly, and then centrifuged at 3000 g for 10 min. Lastly 1 ml of the supernatant was taken for color reaction. The measurements were conducted with a UV–visible spectrophotometer.

A detailed description of the materials and methods used in this study can be found in the online supplementary material.

To explore whether FOXC1 participates in the regulation of amino acid metabolism in HCC cells, we first constructed four stable cell lines, Huh7-FOXC1 and MHCC97H-shFOXC1 and their control groups. Then, we performed an amino acid metabolism RT2 Profiler PCR array to examine transcriptome changes mediated by FOXC1 overexpression in Huh7 cells (Supplementary Table S1) and FOXC1 down-expression in MHCC97H cells (Supplementary Table S2). Using twofold as a cut-off to designate differentially expressed genes, 21 out of the 158 amino acid metabolism genes were down-regulated in Huh7 cells which overexpressed FOXC1, while 30 genes were up-regulated in MHCC97H cells which down-expressed FOXC1. Among the overlap of down-regulated genes in Huh7-FOXC1 compared with Huh7-Control cells and up-regulated genes in MHCC97H-shFOXC1 compared with MHCC97H-shControl cells (Fig. 1a), there are 10 genes listed. CTH attracted our attention, which is strongly inhibited by FOXC1 overexpression. We used the western blot analysis to confirm the relationship between CTH and FOXC1, the results demonstrated that up-regulation of FOXC1 expression meaningfully inhibited the expression of CTH, whereas down-regulation of FOXC1 increased the CTH levels (Fig. 1b and Supplementary Fig. S2B). According to the PCR array result, FOXC1 could regulate the mRNA level of cystathionine-β-synthase (CBS), which converted homocysteine to cystathionine [24]. But we found the protein level of CBS have no significant changes with the different expression of FOXC1 by western blot analysis (Supplementary Fig. S2C). Moreover, homocysteine is a part of methionine cycle, which contain methionine, S-adenosylmethionine (SAM) and S-adenosylhomocysteine (SAH). SAM donates its methyl group to DNA, RNA, proteins, or other cellular metabolites, generating SAH [25]. The results of LC/MS showed that SAM levels slightly increased with the upregulation of FOXC1 and decreased by downregulation of FOXC1 (Supplementary Fig. S2D). CTH is the key enzyme for cysteine synthesis (Supplementary Fig. S2A), acted on cystathionine to generate cysteine, and cysteine is the precursor of ROS scavenger [26]. We found that the levels of ROS increased in HCC cells with FOXC1 overexpression and decreased when FOXC1 is down-regulated (Fig. 1c and Supplementary Fig. S2E). Moreover, when FOXC1 overexpressed, the ratio of glutathione (GSH)/ Glutathione (Oxidized) (GSSG) and the level of cysteine and GSH decreased. Correspondingly, the ratio of GSH/GSSG and the level of cysteine and GSH increased when FOXC1 is down-regulated (Fig. 1d).

According to the data from The Cancer Genome Atlas dataset (TCGA), we found that compared with mRNA level of the CTH in normal liver tissues, the mRNA levels of CTH was markedly lessened in HCC specimens. Furthermore, based on TCGA data, the mRNA expression of CTH decreased in cholangiocarcinoma (CHOL), kidney renal clear cell carcinoma (KIRC) and thyroid carcinoma (THCA) tissues (Supplementary Fig. S1A). Human Protein Atlas program database showed high or medium CTH staining intensity in normal liver samples, whereas liver cancer samples showed low staining of CTH by immunohistochemistry (IHC) tissue microarray data (Supplementary Fig. S1B). Kaplan-Meier analysis based on TCGA data showed that HCC patients who had low CTH mRNA level, their overall survival time of were significantly shorter than that with high CTH mRNA level (Supplementary Fig. S1C). These studies suggested that low levels of CTH indicated poor prognosis, and CTH may act as an important tumor suppressor gene (TSG) in human HCC.

We upregulated the expression of CTH in Huh7-FOXC1 cells and knocked down CTH in MHCC97H-shFOXC1 cells with lentivirus transfection (Fig. 1e and Supplementary Fig. S2F). The results of CCK8 assays (Fig. 1f), plate clone formation assays (Fig. 1h) and soft agar clone formation assay (Supplementary Fig. S2G) showed that up-regulation of CTH abolished FOXC1-facilitated cell proliferation. Cell cycle analysis showed that overexpressed CTH decreased the cell numbers at S phase which increased by FOXC1 (Supplementary Fig. S2H, upper panel). Meanwhile, CTH attenuated the FOXC1-mediated cell migration and invasion (Fig. 1j). Conversely, knockdown of CTH recovered the proliferative ability inhibited by overexpression of FOXC1 (Fig. 1g and i), increased the cell numbers at the S phase (Supplementary Fig. S2H, lower panel), and reversed the migration and invasion capacities (Fig. 1k).

Up-regulation of CTH significantly inhibited tumor growth induced by Huh7-FOXC1 cells, whereas down-regulated CTH rescued the decreased tumor growth mediated by FOXC1 knockdown by In vivo tumorigenicity assays (Fig. 1l-m). The tumors were detected by Ki67 staining, it also indicated that over-expressed CTH inhibited the proliferation of HCC, and knockdown of CTH had an opposite result (Supplementary Fig. S2I). Consistently, in vivo metastasis assay indicated that overexpression of CTH reduced the HCC metastasis with Huh7-FOXC1 cells (Fig. 1n, p-q, Supplementary Fig. S2J), which prolonged overall survival time (Supplementary Fig. S2K). In contrast, the down-regulation of CTH recused the inhibition of HCC metastasis in the MHCC97H-shFOXC1 group (Fig. 1o-q, Supplementary Fig. S2J). These studies showed that FOXC1 facilitated HCC proliferation and metastasis via inhibiting CTH expression.

Epigenetic modifications such as DNA methylation of TSG promoters contribute to the progression and metastasis of HCC [27]. Recent studies indicated that the inhibition of CTH gene transcription resulted from DNA hypermethylation of CpG rich region in the CTH promoter [28‐30]. To determine whether promoter DNA hypermethylation results in CTH gene silencing in FOXC-overexpressing HCC cells, the methylation status of CpG island in the CTH promoter was analyzed by bisulfite genome sequencing (BGS) analysis, which covered 13 CpG sites from − 206 to − 20 of the CTH promoter (Fig. 2a). Hypermethylation at CpG sites in the CTH promoter region was detected in Huh7-FOXC1 cells, whereas less methylation was detected in MHCC97H-shFOXC1 (Fig. 2a and b). Compared with adjacent nontumor tissues, HCC tissues showed much higher methylation levels at CpG sites in the CTH promoter region (Fig. 2c). In addition, treatment with a demethylation agent, the DNA methyltransferase inhibitor 5-aza-2′-deoxycytidine (5-Aza), restored CTH expression in Huh7-FOXC1 cells, indicating that promoter DNA hypermethylation contributes to the transcriptional silencing of CTH gene (Fig. 2d and Supplementary Fig. S3A).

Besides DNA methylation, the metabolism-related genes could be regulated by histone acetylation [31]. We assumed that CTH may also regulated by histone acetylation. To verify the assumption, we use Trichostatin A (TSA), the inhibitor of histone deacetylase, to detect the influence of histone acetylation on expression of CTH. We found that TSA treatment can increase the expression of CTH in Huh7-Control cells, but have no significant changes in Huh7-FOXC1 cells (Supplementary Fig. S3B). It indicated that histone acetylation didn’t contribute to the expression of CTH mediated by FOXC1.

As the important DNA methyltransferases, DNMT3A and DNMT3B are involved in de novo methylation patterns, which are maintained by DNMT1 [32]. Therefore, the DNA hypermethylation of CTH promoter may be induced by DNMTs. We found that the expression of DNMT3B is increased in Huh7-FOXC1 cells, whereas DNMT3B expression is downregulated in MHCC97H-shFOXC1 cells. However, the expression of DNMT1 and DNMT3A had no significant change in Huh7-FOXC1 and MHCC97H-shFOXC1 cells compared to control cells (Fig. 2e and Supplementary Fig. S3C). Moreover, ectopically overexpression of DNMT3B decreased the expression of CTH in MHCC97H-shFOXC1 cells, whereas knockdown of DNMT3B increased CTH expression in Huh7-FOXC1 cells (Fig. 2f and Supplementary Fig. S3D). Furthermore, the BSP results showed that knockdown DNMT3B expression could decrease the DNA methylation levels of CTH mediated by upregulation of FOXC1, whereas overexpression of DNMT3B had the opposite results (Supplementary Fig. S3E). These studies suggested that FOXC1 upregulated DNMT3B expression, which resulted in CTH gene silencing in HCC cells.

The mechanism by which FOXC1 upregulated DNMT3B expression is still unclear. We hypothesized that FOXC1 may directly transactivate DNMT3B. A dual-luciferase reporter assay showed that overexpression of FOXC1 enhanced the transcription of the luciferase reporter in the DNMT3B plasmid constructs compared to that of the controls (Fig. 2g). In order to clarify the regulation mechanism of DNMT3B, the promoter sequence of DNMT3B was analyzed and we found 3 putative FOXC1 binding motifs. To further confirm whether FOXC1 targets these potential binding sites to regulate DNMT3B, we designed a series of reporter plasmids containing truncated or mutated DNMT3B promoter sequences and used them in luciferase reporter assay. Huh7 cells were transfected with these plasmids to assess their reaction to FOXC1 overexpression. The results suggested that FOXC1-induced luciferase reporter expression was significantly eliminated by deletion in the − 195 ~ + 109 bp region compared to that of the controls. Consistently, compared with the control group, mutations at the putative binding sites 2&1 of the DNMT3B promoter significantly reduced the activity of the FOXC1 overexpressed luciferase reporter gene, suggesting that these binding sites were essential for FOXC1 transactivation (Fig. 2h). A chromatin immunoprecipitation (ChIP) assay demonstrated the direct binding of FOXC1 to the putative site in the DNMT3B promoter in Huh7-FOXC1 cells (Fig. 2i upper). Moreover, we investigated whether FOXC1 binds to the same regions in patient samples, and the results revealed that compared to healthy controls, FOXC1 binding sites in the HCC samples were indeed enriched in these regions. (Fig. 2i, lower).

Then we examined the protein levels of CTH and promoter methylation levels of CTH in two independent cohorts of human HCC tissues. Microvascular invasion, poor tumor differentiation, and higher TNM stage were positively associated with the depletion of CTH (Table 1). Loss of CTH expression was an independent and meaningful risk factor for recurrence and reduced survival according to the multivariate analysis (Table 2). The expression of CTH was inversely correlated with promoter methylation levels of CTH in both cohorts (Fig. 2j). In addition, promoter methylation of CTH gene was positively correlated with aggressive tumor behaviors (Supplementary Table S3). HCC patients with promoter methylation of CTH had higher recurrence rates and shorter overall survival time than HCC patients without promoter methylation of CTH (Fig. 2k).

To explore whether DNMT3B was involved in FOXC1-mediated HCC proliferation and metastasis, we knocked down DNMT3B in Huh7-FOXC1 cells and ectopically upregulated DNMT3B expression in MHCC97H-shFOXC1 cells with lentivirus transfection (Fig. 2f). Down-regulation of DNMT3B inhibited FOXC1-facilitated HCC proliferation, migration, and invasion abilities, while up-regulation of DNMT3B had the opposite result (Fig. 3a-d and Supplementary Fig. S4A-F). In vivo tumorigenicity assays suggested that down-regulated DNMT3B significantly inhibited tumor growth induced by Huh7-FOXC1 cells (Fig. 3e), whereas upregulation of DNMT3B salvaged the suppression tumor growth induced by FOXC1 knockdown (Fig. 3f). Representative Ki67-stained images are shown (Supplementary Fig. S4G). In vivo metastatic assay suggested that down-regulated DNMT3B eliminated HCC metastasis with Huh7-FOXC1 cells, which prolonged overall survival time. In contrast, overexpression of DNMT3B rescued the inhibition of HCC metastasis in the MHCC97H-shFOXC1 groups (Fig. 3g-i and Supplementary Fig. S4H-I). Representative H&E-stained images are shown (Fig. 3j). These results identified that FOXC1 facilitated HCC proliferation and metastasis through upregulating DNMT3B expression.

Immunohistochemical (IHC) analysis was used to verify the clinical relevance of FOXC1 and DNMT3B or CTH in human HCC specimens from two independent cohorts. The representative images of FOXC1, DNMT3B, and CTH expression were showed by IHC staining (Fig. 4a). Compared to adjacent nontumor tissues, DNMT3B expression was observably increased in HCC tissues, whereas CTH expression was observably decreased in HCC tissues. FOXC1 expression was positively associated with DNMT3B expression but negatively associated with CTH expression in both cohorts (Fig. 4b). Up-expression of DNMT3B was positively correlation with poor prognosis (Fig. 4c and e, upper panel) and aggressive tumor behavior (Supplementary Table S4). Both TCGA database and Human Protein Atlas program database showed that compared to the normal liver tissues, the mRNA and protein levels of DNMT3B in liver cancer tissues were much higher (Supplementary Fig. S1D-E). Kaplan–Meier analysis based on TCGA data displayed that compared to HCC patients with low DNMT3B mRNA levels, patients who have high DNMT3B mRNA levels had shorter overall survival time and disease-free survival time (Supplementary Fig. S1F). In addition, the reduced expression of CTH indicated poor prognosis (Fig. 4d and f upper panel). Patients who had both of high expression of FOXC1 and DNMT3B endured the highest recurrence rates and lowest overall survival time (Fig. 4c and e, lower panel). Consistently, patients with the FOXC1(+)/CTH (−) expression pattern had the highest recurrence rates and lowest overall survival time (Fig. 4d and f, lower panel).

To explore whether ROS regulate FOXC1-mediated HCC proliferation, migration and invasion, N-acetylcysteine (NAC) which is an antioxidant [33] and L-Buthionine-sulfoximine (BSO) which decreases GSH levels [34], were used to treat Huh7 -FOXC1 cells and MHCC97H-shFOXC1 cells, respectively. NAC (0.2 mM) treatment abolished high levels of ROS induced by FOXC1 overexpression, whereas BSO (30 μM) treatment rescued the decreased ROS levels induced by FOXC1 knockdown (Fig. 5a and Supplementary Fig. S5A-B). Meanwhile, low levels of ROS induced by NAC inhibited the cell proliferation, migration, and invasion abilities mediated by FOXC1 overexpression and high levels of ROS induced by BSO rescued the suppression results mediated by FOXC1 knockdown (Fig. 5b-c).

We have elucidated that overexpression of FOXC1 increased ROS levels by inhibiting cysteine metabolism and ROS promotes HCC progression and metastasis. Meanwhile, we previously reported that FOXC1 is over-expressed in human HCC tissues and promoted HCC progression and metastasis [22, 23]. Therefore, we determined whether high levels of ROS regulate FOXC1 expression in HCC cells. Treatment with BSO and NAC could increase and decrease ROS levels respectively (Fig. 5d and Supplementary Fig. S5C). BSO treatment significantly increased FOXC1 expression and NAC treatment decreased FOXC1 expression at protein level in HCC cells (Fig. 5e and Supplementary Fig. S5D). Notably, with the stimulation of BSO, FOXC1 promoter activity was significantly increased, indicating that high levels ROS transactivated FOXC1 promoter to elevated the expression of FOXC1 (Fig. 5f).

To verify the accurate location of cis-regulatory elements in the FOXC1 promoter sequence that reacted to ROS, we generated a series of truncated mutants of the FOXC1 promoter (Fig. 5g). Dramatic suppression in FOXC1 promoter activity were found in mutants with two deletions from nt-1058 to nt-422 and nt-422 to nt-66, suggesting that these sequences are important in allowing ROS-enhanced FOXC1 activation. The prior region contains one ELK1 binding site, one specificity protein 1 (SP1) binding site and one ATF2 binding site. And the later region contains one SP1 binding site and ELK1 binding site. Notably, site-directed mutagenesis at the ELK1 binding sites inhibited the ROS-promoted FOXC1 activity, while no effect was found for mutations at the SP1 binding site and ATF2 binding site (Fig. 5g). Meanwhile, knockdown of ELK1 abolished FOXC1 promoter activity which is increased by BSO (Fig. 5h). We found that BSO didn’t increase the expression of ELK1 but activate phosphorylation of ELK1 when activated FOXC1 expression (Fig. 5i and Supplementary Fig. S5E).

ROS activates Nuclear factor kB (NF-kB), c-Jun N-terminal kinase (JNK), ERK1/2, p38 kinases and Phosphoinositide 3-kinase (PI3K)/ protein kinase B (Akt) pathways to promote cancer progression [35]. To verify that ROS regulate FOXC1 expression via which pathway, we treated cells with PI3K, p38 kinases, JNK, NF-κB and ERK1/2 inhibitors. Preconditioning cells with ERK1/2 inhibitor decreased ROS-induced FOXC1 expression (Fig. 5j and Supplementary Fig. S5F). Nevertheless, there was on effect on ROS regulating FOXC1 expression when cells were pretreated with other inhibitors. Further, ChIP assays showed that the ERK1/2 inhibitor dramatically weakened ELK1 binding to the FOXC1 promoter, while there were on significant changes on the binding of ELK1 to the FOXC1 promoter by other inhibitors (Fig. 5k). These results demonstrated that ROS induced FOXC1 over-expression via the ERK1/2-p-ELK1 signaling pathway.

8-OHdG is the oxidative damage marker [36]. IHC was used to detect 8-OHdG, phospho-ELK1 (activated ELK1) expression in two independent cohorts of human HCC tissue arrays. Compared to adjacent nontumor tissues, both 8-OHdG and p-ELK1 levels was markedly elevated in HCC tissues. 8-OHdG and p-ELK1 levels were localized in the nucleus (Fig. 6a). FOXC1 expression was positively associated with both 8-OHdG level and p-ELK1 expression in two cohorts (Fig. 6b). patients with elevated 8-OHdG formation (Fig. 6c and e, upper panel) and high expression of p-ELK1(Fig. 6d and f, upper panel), compared to patients with low level of 8-OHdG and p-ELK1, had shorter overall survival and higher recurrence rates. Elevated level of both 8-OHdG and p-ELK1 were positively correlated with loss of tumor encapsulation, microvascular invasion, poor tumor differentiation, and a higher TNM stage (Supplementary Table S5, S6). Furthermore, patients with positive co-expression of 8-OHdG (Fig. 6c and e, lower panel) and FOXC1 or co-expression of p-ELK1 ((Fig. 6d and f, upper panel)) and FOXC1 had the highest recurrence rates and lowest overall survival times.