To the Editor,

Nucleotide methylation, notably in the forms of 5-methylcytosine (5mC) in DNA and N6-methyladenosine (m6A) in mRNA, carries important information for gene regulation [1]. Recent research advances highlight the biological importance of m6A methylation as a dynamic and reversible post-transcriptional modification [2]. 5mC DNA methylation, a conserved epigenetic modification along with m6A RNA modification, also plays critical roles in fundamental biological processes [3, 4]. In addition, recent studies have identified 5mC methylation as a modulator of alternative mRNA splicing at the post-transcriptional level [5, 6]. Although Zhou and colleagues [7] established a molecular link between 5mC DNA methylation and m6A mRNA methylation during fruit ripening, the potential cross-talk still remains uncharacterized in human cancers.

To address this issue, we curated a catalog of 20 and 21 genes that function mainly as regulators of RNA and DNA methylation, respectively (Fig. 1a). The genome-wide omics data comprising of 11,080 human samples across 33 cancer types from the The Cancer Genome Atlas (TCGA) were obtained for analyses (please see Methods and Table S1). First, most of the m6A and 5mC regulators were found to exhibit comparable expression levels across the 33 cancer types (Supplementary Fig. S1). Basing on the Gene Set Cancer Analysis (GSCA) web server [8], we further assessed the gene set differential expression profiles among 14 cancer types with available paired tumor-normal tissue expression data. Across multiple cancer types, the differentially expressed genes (upregulated or downregulated) included both m6A and 5mC regulators (Supplementary Fig. S2). Then, we investigated the mutation frequencies of the m6A and 5mC regulators. Intriguingly, m6A and 5mC regulators exhibited comparable levels of mutation frequency, and significant co-occurrences of genetic alterations were observed between the two regulators (Fig. 1b). Our results showed correlated expression patterns for genes within the same regulator class and even high correlations between the expression of m6A and 5mC regulators (Fig. 1b). Moreover, these m6A and 5mC regulators interacted with one another frequently in protein-protein interaction networks (Fig. 1d).

Fig. 1
figure 1

Cross-talk identified among the m6A and 5mC regulators. a m6A and 5mC regulator genes and a diagram of their potential cross-talk. b Correlations between the expression of m6A and 5mC regulators. The scatter plot shows the strong positive correlation between YTHDC1 and TET2. The Pearson correlation coefficients (R) are shown. c Co-occurrence of genetic alterations in m6A and 5mC regulators. The log2 (odds ratio) is colored as a heat map. The Nightingale rose diagram shows the mutation frequency distribution of m6A and 5mC regulators across different cancer types. d Protein-protein interactions among the m6A and 5mC regulators based on the GeneMANIA database

Full size image

To identify the hub regulators involved in RNA and DNA methylation, we then applied weighted gene coexpression network analysis (WGCNA) to determine the hub genes in m6A and 5mC regulators (Fig. 2a). Strikingly, the number of hub m6A regulators was highly correlated with that of hub 5mC regulators in different cancer types (R = 0.84; Fig. 2b), which may be explained by the cross-talk. We then combined the hub m6A/5mC genes to develop an epigenetic module eigengene (EME), which may reflect both the pre- and post-transcriptional modification statuses. Next, we examined the correlation between EMEs and the activity of hallmark oncogenic pathways (Fig. 2c). Interestingly, our results indicate that high expression of the EME may reflect a highly proliferative and aggressive status in the majority of tumors. In addition, we applied GSCA [8] to analyze the effect (activation or inhibition) of m6A/5mC regulators on cancer-related pathways and confirmed that the m6A and 5mC regulators may be functionally related (Supplementary Fig. S3).

Fig. 2
figure 2

Development and characterization of the m6A/5mC epigenetic module eigengenes (EMEs). a Module membership-based hub m6A and 5mC regulators across 33 cancer types. The lower panel shows the number of hub m6A and 5mC regulators in each cancer type. b Correlations between the number of hub m6A regulators and the number of hub 5mC regulators. The Pearson correlation coefficients (R) are shown. c Gene set enrichment analysis (GSEA) results (normalized enrichment scores [NES] and q values) regarding the hallmark oncogenic pathways for EMEhigh versus EMElow subgroups across 33 cancer types. Enrichment score terms with an FDR < 0.05 are shown. d Heatmap showing the Pearson correlation coefficients between the EMEs and immuno-stromal signatures across 30 cancer types. Diffuse large B cell lymphoma (DLBC), acute myeloid leukemia (LAML), and thymoma (THYM) were excluded, as they mainly consist of immune cells

Full size image

In addition to the tumor compartments, we further investigated the associations between the EME and immuno-stromal signatures representing different statuses of the immune and stromal cells (Table S2) across cancer types. In general, relatively low expression of inflammatory markers and low infiltration of immune cells were observed in the EMEhigh versus EMElow subgroups across cancer types (Fig. 2d). Interestingly, the high enrichment of stromal-related signatures was observed in the EMEhigh subgroups in almost all cancer types, indicating that hub m6A/5mC regulators may generally be involved in stroma activation (Fig. 2d).

Finally, we assessed the prognostic value of the EME in various types of cancers. We found that the EME showed oncogenic features in most cancer types, with overall survival (OS) hazard ratios larger than one (Supplementary Fig. S4a–c). Of these, high expression of the EME was significantly associated with unfavorable OS in cancer types such as KICH, ACC, and LGG (Supplementary Fig. S4a, b). Among HNSC, KIRC, and READ, improved survival was observed in the EMEhigh versus EMElow groups (Supplementary Fig. S4a, c).

In summary, to our best knowledge, this is the first study suggesting potential cross-talk between m6A and 5mC regulators in human cancers. This study provides essential insights into epigenetic regulation in cancer and paves new ways for related therapeutic targets.