DNA Methylation-Associated Transcriptional Inactivation of the 5-Methylcytosine RNA Methyltransferase NSUN7 in Liver Cancer

To characterize possible genetic and epigenetic alterations in the RNA cytosine methyltransferases NSUN2, NSUN3, NSUN6 and NSUN7 in tumors, we first studied in silico a set of 1001 human cancer cell lines in which we had obtained the exome, transcriptome, gene copy number and DNA methylation landscapes [18]. The data mining analyses did not show the occurrence of NSUN2, NSUN3, NSUN6 and NSUN7 mutations, deletions or amplifications in the cell line panel (Dataset S1). Although DNA sequence and genomic aberrations in the aforementioned genes were not observed, promoter CpG island hypermethylation-associated transcriptional silencing is another alternative to accomplish loss of gene activity in transformed cells [19, 20]. NSUN2, NSUN3 and NSUN6 5’-end associated CpG islands were unmethylated in all the cell lines of the collection (Dataset S1A,B,C). However, NSUN7 promoter CpG island was methylated among different cancer types (Dataset S1D, Fig. 1A), being the three most often targeted sites in melanoma (35 of 47, 74.5%), liver cancer (11 of 18, 61.1%) and hematological malignancies (77 of 137, 56.2%). Data-mining of the available expression profiles of the cancer cell lines [18] showed that NSUN7 methylation was associated with RNA downregulation (Fig. 1B). Due to our long-standing interest in liver cancer and many collaborative research projects in this field [21,22,23], we decided to focus our study of NSUN7 in this tumor type, a leading cause of cancer mortality worldwide, accounting for more than 700,000 deaths each year [24].

Fig. 1

Transcriptional silencing of NSUN7 by promoter CpG island hypermethylation in cancer cell lines. (A) Percentage of human cancer cell lines from the Sanger panel, classified by primary tumor site, with a hypermethylated NSUN7 promoter. Total number of cell lines is shown on top of each bar. (B) Correlation analysis between NSUN7 promoter methylation (mean β-value) and NSUN7 transcript expression (Z-score) in cancer cell lines. Spearman’s rank correlation test with its p-value and the associated rho coefficient are shown. (C) Bisulfite genomic sequencing of NSUN7-promoter CpG Island in HCC cell lines, plus a normal liver sample. CpG dinucleotides are represented as short vertical lines, NSUN7 TSS is indicated by a black arrow, and unmethylated or methylated cytosines are represented as white or black squares, respectively. Single clones are shown for each sample (n > 10). (D) Absolute methylation β-values of NSUN7 promoter-associated CpGs analyzed by the 450 K DNA methylation microarray. Green, unmethylated; red, methylated. Data from the studied HCC cell lines, and six normal liver samples are shown. (E) NSUN7 expression in the HCC cells and two normal livers at the RNA level, analyzed by real-time PCR. (F) NSUN7 transcript expression is restored in the three NSUN7 hypermethylated cell lines (SNU-423, SNU-398 and HUH-7) by treatment with the demethylating agent 5-aza-2′-deoxycytidine (5-Aza-dC). RNA expression data shown represent the mean ± S.D. of biological triplicates, and p-values were calculated by a Student’s T test. ***p-value < 0.001, ****p-value < 0.0001

Following the above described biocomputational observation of NSUN7 CpG island hypermethylation-associated silencing in cancer cell lines, we experimentally characterized this phenomenon in liver tumor cell lines. We carried out the bisulfite genomic sequencing of multiple clones in the liver cancer cell lines HEP3B2-1-7, SNU-475, SNU-387, SNU-423, SNU-398 and HUH-7 utilizing primers that encompassed the NSUN7 transcription start site located in the 5’-end CpG island. We observed dense hypermethylation of the NSUN7-associated promoter CpG island in the SNU-423, SNU-398 and HUH-7 cells, whereas for HEP3B2-1-7, SNU-475 and SNU-387 cell lines the NSUN7 5’-end was unmethylated (Fig. 1C). Normal liver was also unmethylated (Fig. 1C). These data matched the DNA methylation profiles derived from the microarray approach (Fig. 1D). The promoter CpG island of NSUN7 was found unmethylated in all the normal human liver tissues analyzed from the TCGA data (Dataset S1E). Importantly, the unmethylated liver cancer cell lines at the NSUN7 5’-end CpG island (HEP3B2-1-7, SNU-475 and SNU-387) expressed NSUN7 RNA, determined by quantitative real-time PCR, furthermore the NSUN7-hypermethylated cells from SNU-423, SNU-398 and HUH-7 showed lack or minimal expression of the transcript (Fig. 1E). An additional link between DNA methylation and transcriptional silencing was established using the DNA demethylating agent 5-aza-2′-deoxycytidine, that restored NSUN7 expression in the hypermethylated liver cancer cell lines (Fig. 1F).

NSUN7 Epigenetic Loss Induces Hypomethylation in the mRNA of the RNA-Binding Protein CCDC9B and Protein Downregulation

The targets of the 5-methylcytosine RNA methyltransferase activity of NSUN7 in human cells are unknown. For mouse cells, NSUN7 has been reported to act on the 5-methylcytosine levels of a few enhancer RNAs [25], but the sequences of these regulatory transcripts are not conserved in humans. Thus, we used a nonbiased epitranscriptomic approach to characterize likely candidate RNA transcripts modified by NSUN7. To accomplish this aim, we coupled bisulfite conversion of cellular RNA with next-generation sequencing (bsRNA-seq) [3, 6] (Methods), as previously described [16], to identify NSUN7-dependent modified cytosine loci in the human transcriptome. For a model, we used SNU-423, a liver cancer cell line, which displays hypermethylation-associated silencing of the NSUN7 promoter. Herein, we generated NSUN7-transfected SNU-423 cells and SNU-423 cells with an empty-vector (EV) to enable comparison. Using the criteria described in Methods, we identified 925 candidate m5C sites in the transcriptome-wide mapping of NSUN7-transfected cells (Table S1). Upon efficient restoration of NSUN7 expression, determined by western-blot (Fig. 2A), we observed that the cytosine site in RNA that reached highest methylation levels with the most significant p-value (multiple correction adjusted P-value = 0.00001378) following NSUN7 transfection in comparison to EV-transfected cells was the C1324 position of the mRNA of the coiled-coil domain containing 9B (CCDC9B) gene (Table S2). To validate this potential target, we performed next the bisulfite sequencing of multiple clones of the mRNA extracted from the empty vector and NSUN7-transfected SNU-423 cell line, using primers encompassing the candidate NSUN7-methylation position at the CCDC9B transcript. This targeted approach validated the bsRNA-seq data by showing an unmethylated C1324 position in empty-vector transfected cells and the restoration of a methylated C1324 site upon NSUN7 transfection (Fig. 2B). Interestingly, we also observed that the two cytosines located immediately before the C1324 site, C1323 and C1322, followed the same RNA methylation profile (Fig. 2B). Most importantly, we generated an inert NSUN7 catalytic mutant form (Fig. S1A,B) that, upon transfection in SNU-423 cells (Fig. S1C), was unable to methylate CCDC9B mRNA (Fig. S1D), further supporting the direct role of the enzyme in the methylation of this target.

Fig. 2

NSUN7 epigenetic loss abrogates m5C methylation of the CCDC9B mRNA leading to diminished RNA stability and protein downregulation. (A) Restoration of NSUN7 protein expression by stable transduction in epigenetically-silenced SNU-423 HCC cells, analyzed by western blot. (B) RNA bisulfite sequencing of multiple clones of CCDC9B transcripts from empty vector (EV) and NSUN7-transfected SNU-423 cells. Cytosines are represented as short vertical lines, the bsRNAseq target C1324 is indicated by a black arrow, and presence of a methyl group is represented as a black square. (C) RNA bisulfite sequencing of CCDC9B transcripts from the HCC cell panel plus a normal liver sample, according to the basal NSUN7-promoter methylation status in liver cancer cell lines. (D) Efficient CRISPR-Cas9 mediated knockout (KO) of NSUN7 protein expression in the hypomethylated and expressing HEP3B2-1-7 HCC cell line. (E) RNA bisulfite sequencing of the CCDC9B transcript from NSUN7 WT and NSUN7 KO HEP3B2-1-7 cells. (F) CCDC9B transcript half-life analyzed by Actinomycin D chase assay (left) and CCDC9B protein levels (right) in EV and NSUN7-transfected SNU-423 cells. (G) CCDC9B transcript half-life analyzed by Actinomycin D chase assay (left) and CCDC9B protein levels (right) in NSUN7 WT and KO HEP3B2-1-7 cells

We were able to reproduce the above described NSUN7-dependent RNA methylation patterns of the CCDC9B transcript in our panel of unmodified liver cancer cell lines. We found that whereas promoter unmethylated and NSUN7-expressing cell lines HEP3B2-1-7, SNU-475 and SNU-387 showed methylation of the three described mRNA sites of CCDC9B; the NSUN7-hypermethylated and non-expressing cells from SNU-423, SNU-398 and HUH-7 showed a lack of C1324, C1323 and C1322 CCDC9B RNA methylation (Fig. 2C). Importantly, non-tumoral liver tissue showed methylation of the three characterized cytosine sites of the CCDC9B mRNA (Fig. 2C). Finally, we demonstrated the role of NSUN7 in the methylation of these RNA loci in the reverse genetic engineering experiment by knocking-out the NSUN7 gene using the CRISPR-Cas9 approach in the NSUN7 unmethylated and expressing HEP3B2-1-7 cells. Upon efficient deletion of NSUN7 (Fig. 2D), we observed that if the wild-type cells showed C1324, C1323 and C1322 CCDC9B RNA methylation; the genetic abrogation of NSUN7 induced an unmethylated status of these three cytosine sites in the CCDC9B transcript (Fig. 2E). Overall, all these data indicate that NSUN7 catalyzes the 5-methylcytosine methylation of the CCDC9B mRNA and that NSUN7 epigenetic inactivation induces the loss of these RNA modifications in liver cancer cells.

The occurrence of 5-methylcytosine in RNAs has been proposed to be associated with transcript stability [26, 27], thus we assessed if this was the case for CCDC9B in our models. Using actinomycin D chase assays, we demonstrated that the transfection-mediated recovery of NSUN7 expression in the epigenetically silenced SNU-423 cell line was associated with an increase in CCDC9B mRNA stability (Fig. 2F) that also led to an upregulation of CCDC9B at the protein level (Fig. 2F). The transfection of the NSUN7 mutant form did not affect the stability of the CCDC9B mRNA in the actinomycin assay (Fig. S1E) or the protein expression levels of CCDC9B (Fig. S1F). On the contrary, the NSUN7 CRISPR/Cas9 deletion in the HEP3B2-1-7 cell line (unmethylated at the NSUN7 promoter and expressing the gene) induced a decrease in CCDC9B transcript stability in the actinomycin D assay (Fig. 2G) that also led to a downregulation of CCDC9B at the protein level (Fig. 2G). Thus, these m5C demethylation events at the CCDC9B mRNA, that occur upon NSUN7 epigenetic inactivation are associated with diminished expression levels of the gene targeted by the studied 5-methylcytosine RNA methyltransferase.

Loss of CCDC9B targets the MYC-Regulator IVNS1ABP protein rendering sensitivity to Bromodomain inhibitors

Very little is known about the biological function of the CCDC9B protein beyond its location in several protein and RNA remodeling complexes [28,29,30]. Due to the presence of the coiled-coil protein domain that could act in protein-protein interactions [31], we wondered if it could exhibit this role in our models. Thus, to look for candidate protein partners of CCDC9B that can provide further clues about its biological roles, we combined immunoprecipitation and mass spectrometry (MS). We compared immunoprecipitates from CCDC9B-FLAG transfected HEP3B2-1-7 cells (NSUN7 unmethylated and expressing the gene) to immunoprecipitates of HEP3B2-1-7 cells transfected with the empty vector. Using the Human SwissProt protein database and the Significance Analysis of Interactome (SAINT) algorithm, we identified four proteins that were bound by CCDC9B: IVNS1ABP, CHTOP, PDIA4 and RCN2 (Bayesian Fold Discovery Rate < 0.05) (Table S3). As an example the MS spectra corresponding to the peptide LYIVGGSDPYGQK of the IVNS1ABP protein is shown in Fig. 3A.

Fig. 3

Loss of CCDC9B is associated with higher abundance of the MYC-regulator IVNS1ABP and sensitivity to bromodomain inhibitors. (A) MS and MS/MS spectra corresponding to the peptide LYIVGGSDPYGQK (IVNS1ABP protein) derived from the protein interactome experiments. At left, extracted Ion chromatogram showing the peptide isotopic distribution found in the CCDC9B-FLAG transfected HEP3B2-1-7 cells, and its absence in EV-transfected cells (middle). The identification of the peptide is shown at right (MS/MS spectrum). The b- and y-ions are depicted in blue and red, respectively. (B) IVNS1ABP protein levels in NSUN7 WT and KO HEP3B2-1-7 cells analyzed by western blot. (C) JQ1 bromodomain inhibitor IC50 values in Sanger cell lines according to NSUN7-promoter methylation status shows enhanced sensitivity to JQ1-mediated growth inhibition in epigenetically-silenced cells. The units in the “Y” axis are the natural logarithm (Ln) of the IC50 micromolar values. Mann–Whitney–Wilcoxon test, ****p-value < 0.0001. (D) IC50 determination for three bromodomain inhibitors using the SRB assay in NSUN7 WT HEP3B2-1-7 cells compared to NSUN7 KO HEP3B2-1-7 cells. (E) IC50 values for the three bromodomain inhibitors in NSUN7-transfected SNU-423 cells in comparison with EV-transfected cells. Data shown represent IC50 values obtained in biological triplicates, and p-values were calculated by a Student’s T test. *p-value < 0.05, **p-value < 0.01, ***p-value < 0.001, ****p-value < 0.0001

Once we observed the interaction between CCDC9B and these proteins, we wondered if CCDC9B could regulate the levels of these targets by this protein-protein binding mechanism. We addressed this point, by performing global proteomics analyses by mass spectrometry in empty-vector transfected SNU-423 cells, showing NSUN7 promoter hypermethylation-associated silencing (Fig. 1C,D,E,F), in comparison with NSUN7-transfected SNU-423 cells (Fig. 2A). We identified 579 proteins that had significantly different expression upon NSUN7 transfection in SNU-423 cells, 280 were downregulated and 299 upregulated (Table S4). Most importantly, among the first set, we identified IVNS1ABP, detected above as a protein partner for CCDC9B (Table S3), as one of the top-downregulated proteins upon restoration of NSUN7 expression (Table S4). Further validating these data, the opposite results were observed in the reverse model: NSUN7 CRISPR/Cas9 deleted HEP3B2-1-7 cells showed, by western-blot, IVNS1ABP protein upregulation compared to the NSUN7 unmethylated and expressing wild-type cells (Fig. 3B). These data support that NSUN7-mediated methylation of CCDC9B mRNA stabilizes this transcript allowing the CCDC9B protein interaction with IVNS1ABP and the lower abundance of this last multifunctional factor.

Interestingly with regard to cancer biology and for therapy-related facets, a role for IVNS1ABP in MYC-associated pathways has been reported [32,33,34] and a certain degree of association between sensitivity to bromodomain inhibitors and enhanced MYC-signaling has also been described [35,36,37]. Thus, we decided to interrogate how these networks and potential treatment vulnerabilities could be mediated by NSUN7 epigenetic inactivation acting through CCDC9B loss and IVNS1ABP engagement. To assess this, we wondered if NSUN7 activity was associated with MYC expression. We observed that NSUN7 CRISPR/Cas9 deletion in HEP3B2-1-7 cells induced the overexpression of MYC (Fig. S2A). On the reverse experiment, the restoration of NSUN7 expression by transfection in the NSUN7 hypermethylated and silenced SNU-423 cells reduced MYC expression (Fig. S2B). In this last setting, the transfection of the inert NSUN7 catalytic mutant form did not affect MYC expression (Fig. S2B).

We then data mined the collection of 1001 human cancer cell lines in which we had previously obtained the DNA methylation landscapes and where sensitivity to the bromodomain inhibitor JQ1, whose activity depend on MYC activation, was also available [18]. This analysis showed that NSUN7 promoter CpG island hypermethylation was associated with increased sensitivity to JQ1 (Fig. 3C). We then moved from the in silico results to the wet data by demonstrating that the CRISPR/Cas9 mediated deletion of NSUN7 in the HEP3B2-1-7 cell line (NSUN7 unmethylated and expressing the gene) induced an increase in the sensitivity not only to JQ1 (Fig. 3D), validating the in silico data (Fig. 3C), but to two additional bromodomain inhibitors (PFI-1 and i-BET) [37], in comparison to the wild-type cells (Fig. 3D). Most importantly, the opposite phenotype was observed in the SNU-423 cells (showing NSUN7 promoter hypermethylation-associated silencing) where the restoration of NSUN7 expression by transfection decreased sensitivity to the anti-growth effect of JQ1, PFI-1 and i-BET (Fig. 3E). The transfection of the NSUN7 mutant form did not affect bromodomain inhibitor sensitivity (Fig. S2C). Thus, all this evidence suggests that NSUN7 loss, through the CCDC9B-IVNS1ABP-MYC axis described above, could render liver cancer cells more sensitive to the action of bromodomain inhibitors.

NSUN7 Epigenetic Loss Pinpoints Primary Liver Tumors with Shorter Overall Survival

Once we demonstrated in liver cancer cell lines the occurrence of tumor-specific DNA methylation-associated transcriptional silencing of NSUN7 and its downstream effects, we wondered about the presence of epigenetic inactivation of this particular 5-methylcytosine RNA methyltransferase in human primary liver tumors. Data mining of the set of primary liver tumors from The Cancer Genome Atlas (TCGA) (https://portal.gdc.cancer.gov), which were analyzed by the same DNA methylation microarray as the one used herein for our initial cancer cell line screening, showed the occurrence of NSUN7 CpG hypermethylation in 41.4% (156 of 377) of liver tumors (Fig. 4A). Analyses of the TCGA RNA-sequencing data validated the in vitro cancer cell lines results, showing that in the primary liver tumors NSUN7 hypermethylation was also associated with low-levels of the transcript (Student t-test, P < 0.001) (Fig. 4B). All normal liver tissues from the TCGA cohort were unmethylated at NSUN7, and the NSUN7 expression levels were similar to those observed in the unmethylated cases of liver carcinoma (Fig. S3A). Interestingly, as we showed in our liver cancer cell lines models how the loss of NSUN7 was associated with lower CCDC9B stability and lower levels of this mRNA (Fig. 2F,G), we were able to recapitulate these data in the primary setting where the presence of NSUN7 hypermethylation was associated with CCDC9B transcript downregulation in the TCGA cohort of liver tumors (Fig. 4C). Importantly, the association between NSUN7 methylation and low expression of CCDC9B was validated in an additional independent cohort of liver tumors [21] (Fig. S3B). Low mRNA levels of NSUN7 in both hepatocellular cancer cohorts were also associated with lower expression of CCDC9B (Fig. 4D, Fig. S3C). Interestingly, the presence of low levels of expression for CCDC9B was associated with higher levels of MYC mRNA in the two studied sets of liver tumors (Fig. S4A). For the TCGA cohort, expression data of a subset of proteins is available, and we found that low levels of NSUN7 mRNA were also associated with higher levels of the MYC protein (Fig. S4B). All these data strengthen the results observed in liver cancer cell line models (Fig. S2A,B). Related to the clinicopathological context of NSUN7 epigenetic inactivation in liver tumors, we observed that NSUN7 hypermethylation was not associated with gender, vascular tumor cell type, ISHAK fibrosis score, histological grade and TNM stage (Table 1). However, we found that NSUN7 epigenetic loss was more commonly observed in older patients (Fisher’s exact test, p < 0.001) (Table 1).

Fig. 4

NSUN7 epigenetic loss occurs in human primary HCC tumors in association with worse clinical outcome. (A) Percentage of NSUN7 methylation in the TCGA data set of primary tumors according to cancer type. (B) NSUN7 methylation is inversely correlated with NSUN7 transcript expression in TCGA HCC tumors. (C) NSUN7 methylation is associated with decreased CCDC9B transcript levels in primary TCGA HCC tumors. (D) Low expression of the NSUN7 mRNA is associated with decreased CCDC9B transcript levels in primary TCGA HCC tumors. (E) Kaplan–Meier analysis of overall survival (OS) in the TCGA liver cancer cohort with respect to NSUN7 methylation status. Significance of the log-rank test is shown. Results of the univariate Cox regression analysis are represented by the hazards ratio (HR) and 95% confidence interval (95% CI). (F) Forest plot of the multivariable Cox regression analysis for clinical outcome in the TCGA liver cohort studied for NSUN7 methylation status taking into account different prognostic factors. P-values correspond to hazard ratios (HR), with a 95% CI, associated with OS. Co-variables with associated p-value under 0.05 were considered as independent prognostic factor (*p < 0.05, **p < 0.01, ***p < 0.001)

Table 1 Clinicopathological characteristics of the TCGA LIHC cohort

We also interrogated whether the presence of NSUN7 hypermethylation exhibited any prognostic value in liver tumors. In this regard, we found that NSUN7 epigenetic inactivation was associated with reduced overall survival (OS) in the TCGA cohort (log-rank P = 0.013; hazard ratio (HR) = 1.563, 95% CI = 1.097–2.226) (Fig. 4E). NSUN7 hypermethylation was a better predictor of OS than NSUN7 expression that can be affected by many factors beyond epigenetic inactivation. The low levels of NSUN7 were only associated with shorter survival in the non-vascular tumor type [65.4% of the cases (log-rank P = 0.021; HR = 1.852, 95% CI = 1.089–3.149] (Fig. S5) among the examined clinicopathological characteristics (Table 1). Most importantly, to assess the value of NSUN7 methylation as a potential independent biomarker, we performed multivariate analysis for NSUN7 methylation and the available clinical parameters (gender, age, vascular tumor cell type, ISHAK fibrosis score, histological grade and TNM stage) in the TCGA liver cancer cohort. Remarkably, the multivariate Cox regression analysis showed that NSUN7 hypermethylation was an independent predictor of overall survival (HR = 1.784; 95% CI = 1.012–3.145; P = 0.045) (Fig. 4F).

Finally, we wondered if our subset of cases with NSUN7 hypermethylation were enriched in any particular subclass of liver tumors. Using a recently revised immunogenomic classification of liver tumors [38], we found that TCGA HCC cases hypermethylated at NSUN7 were significantly depleted in the immune active subclass (Fisher’s exact test, P = 0.018), but enriched for the immune-like signature (Fisher’s exact test, P = 0.002). Liver tumors from both subclasses present an enrichment in the PD-1 and interferon signaling pathway and high expression of checkpoint molecules (such as CTLA4, PD-1 and PD-L1) [38], all of them biomarkers associated with response to immunotherapy, but the immune active subclass is associated with increased overall survival and the immune-like signature is not [38]. Importantly using the Chiang molecular classification of HCC [39], we observed that TCGA liver tumors hypermethylated at NSUN7 were significantly enriched in the CTNNB1 subclass (Fisher’s exact test, P = 0.011) and it has been recently proposed that β-catenin activation in HCC promotes resistance to anti-PD-1 therapy [40]. Thus, the described analyses warrant further investigation into a role of NSUN7 in immunotherapy response.