SorafenibR HCC cells display improved viability, enhanced migration and reduced apoptosis after transient sorafenib treatment
To gain insights into the mechanisms of sorafenib resistance, we initially modeled the response of HCC Hep3b and Huh7 cells to sorafenib by continuously culturing them with a step-wise increase of drug dosages for 8 weeks. The final concentration of sorafenib was 2 µM, which exerted sufficient inhibitory action and was in the range of clinically achievable levels.23,24 To characterize these sorafenib-selected cells, we first measured the proliferation of parental and resistant (sorafenibR) cells upon transient exposure to 2 µM sorafenib. Although all parental controls displayed significant decreases of cell proliferation, the sorafenibR cells did not show obvious difference in cell proliferate (Fig. 1a and Fig. S1a), upon exposure to sorafenib. This was further confirmed by the enhanced wound-healing capability in sorafenibR versus parental Hep3b and Huh7 cells (Fig. 1b and Fig. S1b). Second, flow cytometry analysis revealed that transient sorafenib treatment does not have obvious alterations in cell apoptosis and cell viability in sorafenibR Hep3b and Huh7 cells, but significantly promotes cell apoptosis and impairs cell viability in parental controls (Fig. 1c and 1d). We did not see obvious changes in cell cycles in both parental and sorafenibR Hep3b and Huh7 cells when exposed to 2 µM sorafenib (Fig. S1c). Third, we observed that, as compared with parental cells, the oncosphere formation is enhanced in sorafenibR with less affected by sorafenib treatment (Fig. 1e; Fig. S1d), and resistant cells exhibit a higher rate of cell proliferation in drug free medium (Fig. 1f; Fig. S1e). These findings are in line with our previous data showing that lung cancer cells resistant to tyrosine kinase inhibitors (i.e., midostaurin) migrate and grow at a faster pace than parental cells in vitro, and possess higher tumorigenic potential in vivo.9
As sorafenib discontinuation is a frequent event in HCC patients, we mimicked sorafenib “holiday” by culturing sorafenibR cells in drug-free medium for 14 days (released cells). Interestingly, the released cells displayed proliferation impairment, which is supported by a reduction of cell proliferation, and increase of cell apoptosis in the presence of sorafenib, when compared with parental and sorafenibR cells (Fig. 1g; Fig. S1f). These results indicate that partial re-acquisition of sensitivity to sorafenib leads to a transient proliferation arrest upon drug withdrawal, phenocopying the drug holiday effect seen in the clinic. This is consistent with our recent findings in leukemia that resistance to TKIs (i.e., nilotinib, imatinib) displays reversible features or a transient proliferation arrest upon exposure to the same drug.21 Finally, we performed sequencing for epigenetic regulators, for example, DNMT3a and TET2 that are frequently mutated in cancers, in sorafenibR Hep3b and Huh7 cells. We did not find any acquired mutations in these genes. Taken together, these results suggest that non-genetic mechanisms could be essential in sorafenib resistance, and although generated in vitro, our drug-resistant HCC cell line derivatives faithfully recapitulate clinical drug resistance.
HCC cells with sorafenib-acquired resistance have activation of pathways involved in DNA methylation
To identify critical molecules that promote sorafenib resistance, first we examined the expression of DNA methyltransferases (DNMTs) in parental and sorafenibR Hep3b and Huh7 cells, because epigenetic aberration (i.e., DNA hypermethylation) is becoming increasingly important in the development of TKI resistance.9,21 The results from qPCR and Western blotting disclosed that expression of DNMT3a and DNMT3b, two de novo DNA methyltransferases,7 is increased at both RNA and protein levels, but expression of DNMT1, a maintenance DNMT,25 as well as HDAC2 and HDAC3, histone protein modifiers, is not obviously changed (Fig. 2a and 2b). As DNA methylation is installed by DNMTs via adding a methyl group to C-5 position of cytosine residues yielding 5-methylcytosine (5mC), we proceeded to determine whether 5mC amount is changed in sorafenibR cells. The DNA dotblotting by anti-5mC antibody10,26 revealed that 5mC production is significantly elevated in sorafenibR Hep3b and Huh7 cells compared with parental counterparts (Fig. 2c and 2d). These results extend previous findings showing that 5mC amount is markedly increased in leukemia cells resistant to nilotinib10 and in lung cancer cells resistant to midostaurin.9
It has been shown that the TET methylcytosine dioxygenases catalyze the conversion of 5mC to 5hmC, resulting in DNA demethylation.8,27-29 To further understand the role of DNA methylation in sorafenib resistance, we performed qPCR and Western blot for changes of TETs, and found that expression of TET1, TET2 and TET3 is significantly upregulated at both RNA and protein levels in sorafenibR compared with parental cells (Fig. 2e and 2f). Dotblotting analysis using anti-5hmC antibody revealed that 5hmC level is much higher in sorafenibR Hep3b and Huh7 cells than that in parental counterparts (Fig. 2g and 2h), in agreement with the methylcytosine dioxygenase activities of TET1, TET2 and TET3. Collectively, these findings suggest that long-term exposure to sorafenib alters the functions of DNA methylation machinery in HCC cells.
Knockdown of DNMT3a and TET2 impairs sorafenibR HCC cell growth
Having demonstrated the upregulation of DNMT3a, a de novo DNA methyltransferase,7 in sorafenibR HCC cells, next we sought to determine the biological functions of DNMT3a aberrations. SorafenibR Hep3b and Huh7 cells were infected with DNMT3a shRNA or scrambled vectors for 24 hours and further selected by 2 µM puromycin for additional 96 hours. First, the efficacy of virus infection was verified by the high rate of GFP/fluorescent cells (Fig. S2a). The shRNA-3, which showed the most DNMT3a reduction, was used for further investigations. As shown in Fig. 3a and 3b, both RNA and protein expression of DNMT3a was markedly decreased in DNMT3a shRNA compared with control group. The specificity of DNMT3a knockdown was supported by the unchanged expression of DNMT1, DNMT3b and TET2 in DNMT3a shRNA versus control cells. Second, the results from dotblotting showed that 5mC production is much lower in DNMT3a-depleted cells than that in control groups (Fig. 3c and 3d). Interestingly, DNMT3a knockdown led to a decrease of 5hmC abundance even when TET2 protein levels are not changed. This may result from the reduction of 5mC amount, the original substrate of TET2. Third, these DNMT3a depleted cells were subjected to proliferation and wound-healing assays. We observed that cells with DNMT3a knockdown could proliferate and migrate at a much lower rate, in a time-dependent manner, than that in control counterparts (Fig. 3e and 3f; Fig. S2a and Fig. S2b). These findings support an important contribution of DNMT3a to sorafenibR cell growth.
Having shown that TET2 is upregulated in sorafenibR compared with parental cells, next we sought to examine whether TET2 expression is essential for sorafenibR cell growth. To this end, sorafenibR Hep3b and Huh7 cells were infected with TET2 shRNA or control vectors for 24 hours followed by puromycin selection for additional 96 hours. The high rate of GFP/fluorescent cells indicated the high efficacy of virus infection (Fig. S2c). The results from qPCR and Western blot demonstrated the efficient knockdown of TET2 gene, and the specificity of TET2 knockdown was supported by the unchanged levels of TET1, TET3 and DNMT3a in TET2 shRNA-transfected cells (Fig. 3g and 3h). As TET2 is a methylcytosine dioxygenase, we speculated that TET2 ablation may decrease 5hmC production. Indeed, the results from dotblotting using 5hmC antibody support that sorafenibR Hep3b and Huh7 cells have much lower levels of 5hmC than parental cells (Fig. 3i and 3j). Notably, TET2 knockdown led to an increase of 5mC abundance even when DNMT3a protein expression are not changed. This may be attributed to the downregulation of TET2, leading to less conversion of 5mC to 5hmC.
To investigate the biological outcomes of TET2 knockdown, the sorafenibR Hep3b and Huh7 cells with TET2 depletion were subjected to proliferation and wound-healing assays. We found that, compared to those with scrambled controls, the cells with TET2 knockdown proliferate at a much slower rate (Fig. 3k, Fig. S2c) and migrate at a much shorter distance (Fig. 3l, Fig. S2d), which occurs in a time-dependent manner. These findings suggest that TET2 is required for the aggressive proliferation and migration of sorafenibR HCC cells.
CSCs with upregulation of DNMT3a and TET2 are more tolerant to sorafenib-induced cell death
Given the highly heterogeneous of HCC cells and the role of oncospheres in developing drug resistance, we performed oncosphere-forming assays in HCC Hep3b and Huh7 cells, and found that about 6% and 5% cells (ratio: Onco/total) can form oncospheres, respectively (Fig. 4a). To examine the drug sensitivity of oncospheres, we replated oncospheres and treated them with different doses of sorafenib. As shown in Fig. 4b, the replated oncosphere cells displayed IC50 values to sorafenib much larger than those exhibited by their parental counterparts, suggesting that oncospheres may mediate, at least partially, the development and maintenance of sorafenib resistance. In line with this, sorafenibR Hep3b and Huh7 cells had a greater tendency to form oncospheres than parental cells (Fig. 4c). To identify essential regulators for oncosphere formation, we examined the levels of DNMTs and TETs. We found that expression of DNMT3a and TET2 at RNA and protein levels is highly elevated in oncosphere compared to parental cells (Fig. 4d and 4e). To assess the involvement of DNMT3a and TET2 in oncosphere formation, we used shRNA lentiviruses to knock down them, and found that DNMT3a or TET2 knockdown inhibits the oncosphere growth supported by reduction of oncosphere number without obvious changes of oncosphere sizes (Fig. 4f). Further, we found that oncosphere cells express elevated levels of stem cell makers like CD133, CD25, CD44 and c-KIT (Fig. 4g). Therefore, we propose that upregulation of DNMT3a and TET2 in CSCs is essential for the development of sorafenib resistance.
Dysregulation of DNMTs and TETs is linked to sorafenib responses and survivals in HCC patients
To explore the clinical implications of DNMT upregulation in sorafenibR HCC cells, we sought to determine if DNMT3a expression can predict HCC patient responses receiving sorafenib therapy. We first analyzed DNMT3a mRNA levels in public datasets GDS4887, GSE45267 GSE54236 and GSE109211, and found that DNMT3a is significantly upregulated in HCC tumors compared to non-tumor tissues (Fig. 5a). Importantly, sorafenib non-responders (n = 21) had higher levels of DNMT3a expression than responders (n = 40) (Fig. 5b). Notably, we also found that DNMT1 and DNMT3b expression is higher in HCC tumors than non-tumor tissues as well as in sorafenib non-responders than responders (Fig. S3a). To validate prognostic implications for DNMT3a overexpression, we analyzed the survival of HCC patients from GSE54236, GSE109211 or used online tools (protein atlas, Gepia2) to examine the association between patient outcomes and DNMT3a levels. We observed that HCC patients with higher DNMT3a expression have significantly shorter survival time than patients with lower levels (Fig. 5c and 5d). Importantly, sorafenib non-responding patients, who have higher DNMT3a expression, had a shorter survival time than responders carrying lower DNMT3a expression (Fig. 5e). We also assessed the association of patient outcomes with the expression of DNMT1 and DNMT3b, and found that upregulation of DNMT1 and DNMT3b corresponds with short overall survival in HCC patients as well as in sorafenib non-responders (Fig. S3b and 3c). Given that DNMT3a is a de novo DNA methyltransferase, these findings suggest that DNMT3a upregulation followed by 5mC increase is more essential in sustaining sorafenib resistance.
To demonstrate the clinical significance of TET2 upregulation in HCC patients, we analyzed public datasets GSE45267, GSE54236 and GSE109211 for TET2 mRNA levels. We found that, in general, TET2 expression has a trend toward upregulation in HCC tumors compared to normal tissues (Fig. 5f), which is required to be validated in a larger cohort of patients. Importantly, sorafenib nonresponders (n = 21) had significantly higher levels of TET2 (p < 0.05) than responders (n = 40) (Fig. 5g). Notably, we also found that TET1 is higher in HCC tumors than non-tumor tissues and in sorafenib non-responders than responders (Fig. S3d). To explore the prognostic implications of TET2 overexpression, we analyzed the survival of HCC patients from GSE54236, GSE109211 or used online tools (protein atlas, Gepia2) to examine the association between patient outcomes and TET2 levels. We observed that HCC patients with higher TET2 have significantly shorter survival time than those with lower TET2 (Fig. 5h). Importantly, sorafenib non-responding patients, who have higher TET2 expression, had a small survival probability than responders carrying lower TET2 expression (Fig. 5i). We also assessed the association of patient outcomes with the expression of TET1 and TET3, and found that upregulation of TET1 and TET3 corresponds with short overall survival in HCC patients (Fig. S3e and S3f). These findings extended the previous studies showing that TET2 is upregulated in HCC cells and may be important for HCC cell growth.14 Collectively, our data suggest that TET2 is an essential regulator of HCC sorafenibresistance.
Pharmacological targeting of TET2 impairs sorafenibR cell growth
The findings, which TET2 is upregulated in sorafenibR cells, and TET2 gene knockdown inhibits resistant cell growth, imply that TET2 could be a pharmacological target. To test this, we treated sorafenibR Hep3b and Huh7 cells with bobcat339, a selective cytosine-based TET enzyme inhibitor.30 Western blotting revealed that exposure to bobcat339 does not change protein expression of either TET2 nor TET1 and TET3 (Fig. 6a), and dotblotting analysis showed that bobcat339 treatment reduces DNA 5-hmC abundance (Fig.6b and 6c) without obvious changes of 5mC amount (Fig.6d and 6e), in line with the concept that bobcat339 does not inhibit de novo methyltransferase DNMT3a.30
To assess if exposure to bobcat339 inhibits sorafenibR cell growth, we performed flow cytometry and observed that cell apoptosis is significantly increased in the presence versus absence of bobcat339 (Fig. 6f and 6g). Further, proliferation and wound-healing assays found that sorafenibR cells treated with bobcat339 proliferate and migrate at a much lower rate than untreated cells (Fig. 6h and 6i; Fig. S4a and Fig. S4b). Furthermore, combination of sorafenib with bobcat339 resulted in more pronounced inhibition of sorafenibR cell proliferation than the treatment with single reagents (Fig. 6j; Fig. S4c). Collectively, these results support that TET2 inhibitors may be promising therapeutic reagents in overcoming sorafenib resistance, which merits thorough investigations in vivo.
DNMT3a and TET2 regulate sorafenib sensitivity and coordinately modulate sorafenibR cell proliferation
Given that both DNMT3a and TET2 are upregulated in resistant cells, and as knockdown of DNMT3a or TET2 suppresses resistant cell growth, we speculated that DNMT3a or TET2 expression may be associated with sorafenib sensitivity. To this end, DNMT3a or TET2 was knocked down in sorafenibR Hep3b and Huh7 cells having their overexpression. These cells were treated with 2 µM of sorafenib for indicated time points. Cell proliferation assays revealed that depletion of DNMT3a or TET2 sensitizes sorafenibR cells to sorafenib-inhibited cell proliferation (Fig. 7a and 7b), and wound-healing studies uncovered that DNMT3a or TET2 knockdown enhances sorafenib-impaired cell migration (Fig. 7c and 7d). These findings suggest that DNMT3a or TET2 expression is required to maintain sorafenib-resistant phenotypes of HCC cells.
To establish the role of DNMT3a and TET2 interaction on sorafenibR cell growth, we knocked down either DNMT3a and TET2 alone or both. The cell proliferation assays revealed that concurrent knockdown of DNMT3a and TET2 leads to more robust inhibition of cell proliferation and migration than that of a single gene change (Fig. 7e and 7f), suggesting that DNMT3a and TET2 have functional cooperation in sustaining proliferation and migration of sorafenibR HCC cells.
DNMT3a physically interacts with TET2, but does not form a regulatory loop, to enhance sorafenibR resistant phenotypes
Because DNMT3a and TET2 are upregulated simultaneously in sorafenibR HCC cells, we first determined if DNMT3a regulates TET2 transcription or vice versa. By using online tools (GEPIA 2) and analyzing public data GDS4887, GSE45267 and GSE54236, we did not find an obvious correlation between DNMT3a and TET2 (Fig. S5a and S5b). These findings, together with the observations that DNMT3a knockdown does not change TET2 protein expression or vice versa (Ref. Fig. 3b and 3h), suggest that DNMT3a and TET2 don’t transcriptionally regulate each other. We then examine if DNMT3a and TET2 have protein interactions in HCC cells. Co-IP assays revealed that DNMT3a forms a complex with TET2 and HDAC2 in both sorafenibR and parental HCC cells with a slight increase of their interaction potential in sorafenibR cells (Fig. S5c and S5d). The protein interaction among DNMT3a, TET2 and HDAC2 was further confirmed by the results from string-interaction assays (Fig. S5e; https://string-db.org), which was consistent with previous studies showing that HDAC2 and TET2 are co-transcription factors of DNMT3a.31-33 Given the dual function of TET2 in cancers,33 these findings suggest that a transcriptional complex of DNMT3a, TET2 and HDAC2 exists in sorafenibR HCC cells.
Tumor suppressor genes are further silenced in sorafenibR HCC cells by DNMT3a- mediated promoter DNA hypermethylation
To define the mechanisms that DNA methylation aberrations contribute to faster growth of sorafenibR cells, we examined the changes of epigenetically silenced tumor suppressor genes (TSGs), for example, p15, p16, p18, FHIT, SOCS1 and SOCS2, which are frequently silenced by promoter DNA methylation and their downregulation predicts worse prognosis in HCC patients. The results from qPCR showed that expression of p15 and SOCS2 is largely decreased in sorafenibR compared with parental Hep3b and Huh7 cells, but no obvious changes were detected in the expression of p16, p18, FHIT, and SOCS1 (Fig. 8a). These results suggest that further TSG silencing is required for the survival and faster proliferation of sorafenibR cells.
As epigenetic regulators TET2 and DNMT3a are upregulated in sorafenibR cells, we reasoned that overexpression of TET2 and DNMT3a may account for TSG silencing. To test this, we knocked down DNMT3a or disrupted TET2 activity via bobcat339 in sorafenibR Hep3b and Huh7 cells, and assessed expression of these TSGs. As expected, DNMT3a ablation significantly increased the levels of p15 and SOCS2, but genetic or pharmacological inactivation of TET2 did not obviously change their expression. Notably, co-knockdown of TET2 and DNMT3a led to more pronounced upregulation of p15 and SOCS2 (Fig. 8b), suggesting that TET2 and DNMT3a coordinately silence TSGs in sorafenibR cells.
To investigate the methylation status of p15 and SOCS2 promoters, genomic DNA from sorafenibR and parental cells was digested with the restriction enzyme HpaII cutting only nonmethylated sites or BstUI cutting only methylated sites. Decitabine-treated cells were used as positive controls. Following digestion, DNA was analyzed by qPCR using primers specific to p15 or SOCS2 gene promoter containing CpG islands. Efficient digestion by HpaII or BstuI leads to stronger amplification in hypomethylated or hypermethylated p15 and SOCS2 promoters, respectively, in sorafenibR cells compared to parental controls. Consistent with the silencing data (see Fig. 8a) and opposite to decitabine-treated controls, these data provide the evidence that the p15 and SOCS2 promoters are more hypermethylated in sorafenibR cells than that in parental controls (Fig. 8c). To precisely measure the alterations of promoter methylation, we performed MeDIP assays, in which methylated DNA was enriched by 5mC antibody and quantified by qPCR. Consistent with the findings from enzyme digestion (see Fig. 8c), the levels of 5mC in the promoters of p15 and SOCS2 were significantly increased in sorafenibR cells compared to parental controls (Fig. 8d). To address how TSG promoters become hypermethylated, we performed ChIP assays, and found that both DNMT3a and TET2 bind the promoters of p15 and SOCS2 (Fig. 8e), which are in line with the physical protein interactions between DNMT3a and TET2, leading to DNA hypermethylation that interacts with HDAC2 in the promoter to further silence TSGs.
DNMT3a and TET2 activate cell proliferation genes through promoter binding in a DNA methylation-independent manner
To further understand how sorafenibR cells more aggressively proliferate and migrate, we first examined the expression of certain oncogenes, like CDK1, CCNA2 and RASEF. We found that these oncogenes are highly elevated in sorafenibR compared to parental cells (Fig. 8f). These results suggest that, in addition to TSG silencing, oncogenic upregulation may contribute to the survival and faster proliferation of sorafenibR cells. Second, inactivation of either TET2 or DNMT3a led to a decrease of oncogene expression, but we did not see a synergistic effect on oncogenic downregulation upon knockdown of both TET2 and DNMT3a (Fig. 8g). Third, the assays of enzymatic digestion from bulk DNA did not see obvious changes of DNA methylation in the promoters of oncogenes CDK1, CCNA2 and RASEF (Fig. S6a). Instead, MeDIP assays revealed a slight decrease of 5mC in the promoters of CDK1 and CCNA2, but not RASEF (Fig. S6b). Fourth, ChIP analysis disclosed that DNMT3a, TET2 and HDAC2 are enriched in the promoters of oncogenes CDK1, CCNA2 and RASEF (Fig. S6c), although such binding did not increase the levels of DNA methylation in sorafenibR cells. These findings strongly support that both DNMT3a and TET2 have DNA methylation-independent functions.34,35