Introduction

Nasopharyngeal carcinoma (NPC) is one of the predominant malignancies in South Asian continents, primarily in southern China, Indonesia, and India1. According to the global cancer repository report published by GLOBOCAN in 2020 (https://gco.iarc.fr/), NPC ranked 22nd based on occurrence. Among the in-vogue treatment regimens, radiotherapy and parallel chemotherapy increase NPC patients’ survival. To date, identifying NPC at an initial stage is challenging because the symptoms of this cancer resemble other, more common conditions, which creates confusion for clinicians to diagnose. It is also hard to examine the nasopharynx. Sometimes, the submucosal lesions spread silently with a standard appearance during clinical tests. Thus, many NPC patients appear in clinics with advanced stages, resulting in a poor prognosis2.

So far, the TNM (tumor–node–metastasis) staging technique is a decisive parameter in providing analytical information and directing treatment choices for cancers, including NPC. To date, this model is not enough to determine the best personalized treatment. About 20–30% of NPC patients with the same stages receiving similar treatment show local relapse or distal metastasis3,4,5, pointing out that molecular subclassification may be clinically appropriate, but more accurate molecular tools are a prerequisite that can stratify patients concerning prognosis and response to therapy. Since NPC is diversely categorized into four subtypes based on histological character and ethnicity-specific occurrence, there is very little knowledge available for NPC’s prognostic biomarkers to predict and improve patient outcomes.

The development of malignancy or its distribution is a complex phenomenon comprising many factors, which may differ uniquely from one cancer to another. Cumulative evidence suggests that the tumors’ molecular characteristics add predictive strength in developing molecular biomarkers. DNA methylation at the 5ʹ cytosine (5mc) was revealed with capabilities to define a better prognosis of various malignancies6,7,8,9. DNA methylation is one of the most extensive genomic aberrations occurring during carcinogenesis10. These abnormalities can broadly be categorized as focal hypermethylation and global hypomethylation11. Hypermethylation can promote the silencing the regulatory region of tumor suppressor genes, contributing to cell growth deregulation12,13.

On the other side, hypomethylation mainly predominates in the repetitive regions and sometimes unwantedly activates transposable elements within the genome, resulting in further genetic damage14. Hypomethylation also drives genomic instability, causing missegregation of chromosomes during cell division15. Thus, studying DNA methylation in cancer is a fundamental requirement for comprehending that cancer’s outcome and prognostic behaviour and developing a better treatment regime.

Genome-wide DNA methylation between tissue and whole blood has been investigated in several tumors and displayed a dissimilar portrait of methylation signature, which indicates that blood DNA methylation has an additional role in tumor progression and metastasis over tissue DNA methylation. Moreover, the Study of DNA methylation from blood samples has some advantages over other sampling methods as it is commonly available in epidemiological studies, marginally aggressive, and more suitable to the screening population. Although DNA methylation can be detected in cell-free DNA (cfDNA) from serum or plasma, its low yield and variation in the processing method of isolation indicated significant problems16.

However, the underlying relationship between DNA methylation and NPC’s development is poorly understood. Besides two genome-wide DNA methylation studies of primary NPC, only a few biomarkers with diagnostic and prognostic values have been reported5,17,18. The only existing report of global blood DNA methylation of head and neck cancer (HNSCC) suggested that environmental risk factors could modify the DNA methylation and are associated with the prognosis of HNSCC. Although this study did not imply the advancement of HNSCC, it led us to hypothesize whether the blood DNA methylation signature differs from tissue DNA methylation in advanced NPC.

Therefore, we aimed to determine the DNA methylation signature from the whole blood of advanced NPC patients and examined the differences in DNA methylation patterns of blood and tissue of NPC patients to identify unique methylation biomarkers. The predictive role of biomarkers in NPC metastasis was investigated further.

Results

Blood DNA methyl array data revealed a differential methylation pattern in metastatic NPC

The methylation analysis results are displayed in Figs. 1, 2, Tables S2 and S3. After normalization, probes with Δβ between ≤ 0.2 and ≥ 0.2 were removed, and 4093 probes were identified as DMPs, which covered approximately 2875 genes throughout the genome (Fig. 1 and Table S2). The distribution of DMPs in different chromosomes showed that Chr.1 and 2 were highly methylated (Fig. 2a). Analysis of methylation related to the gene’s location indicated that the highest number of methylations occurred at the gene body (47.50%), followed by the intragenic region or IGR (29.76%) (Fig. 2c left). The open sea region showed comparatively higher DMPs (8.0.92%) than any other locations related to CpG (Fig. 2c right). The majority of the probes were hypermethylated (97.39%) in NPC patients (Fig. 2e). The hypermethylation frequency was highest in the gene body (46.43%), followed by IGR (29.05%). The open sea region was more hypermethylated (80.07%) than other CpG locations.

Figure 1
figure 1

Heatmap of differentially methylated probes (DMPs).

Full size image

Methylation in the untranslated region (UTR) is essential in controlling gene expression. In our study, a total of 12.05% hypermethylation was observed in the untranslated region, including 5ʹ-UTR (9.92%) and 3ʹ-UTR (2.13%). Promoter methylation of about 9.28% was noted, in which TSS200 and TSS1500 had 1.93% and 7.35%, respectively (Fig. 2f). The distribution of hypomethylation concerning genes and CpGs followed a similar trend (Fig. 2g).

Unlike 5mc, hydroxymethylation (5hmc) is mainly distributed to the active transcriptional region, accompanied by open chromatin, allowing histone modification. The expression of 5hmc varies from tissue to tissue, and its expression is correspondingly low in cancer stage 4 with metastasis. Patients were shown to live longer with a higher 5hmc level than one with a lower level19. Thus, using 5hmc as the ideal prognostic biomarker could be one of the most excellent possibilities for understanding cancer susceptibility. We further conducted the analysis and distribution of 5hmc to understand the ratio of 5hmc to 5mc in advance NPC and found a total of 1232 5hmc probes, which comprised almost one-third of the 5mc (Fig. 2b and Table S3). The highest amount of 5hmc was noticed in the open sea region (82.06%), followed by the shore (18.91%), shelf, and island. The distribution of 5hmc related to gene location exposed the gene body as the highest 5hmc covering region compared to other sites. The UTR contained 14.78% of the 5hmc, in which 5ʹ-UTR and 3ʹ-UTR had 10.23% and 4.55%, respectively (Fig. 2d). A total of 14.01% of 5hmc was noted in the promoter region, including TSS200 (10.63%) and TSS1500 (3.41%).

Figure 2
figure 2

Differential methylation analysis. Genomic regions were annotated as (i) locations related to gene: TSS200 (covers − 200 nucleotide upstream of TSS, generally considered as a proximal promoter); TSS1500 (covers − 200 to − 1500 nucleotide upstream of the transcription start site or TSS, considered as a distal promoter); 5ʹ-UTR and 3ʹUTR (covers the entire untranslated regions); 1st exon; body (Gene Body); ExonBnd (exon boundaries) and IGR (intergenic region). (ii) Locations related to CpG: Island (CpG island, usually extends for 300–3000 base pairs); Shore (up to 2 kb upstream or downstream from CpG island); Shelf (2–4 kb upstream or downstream from CpG island); Open sea (a region that is not associated with CpG island and has not determined yet). (a) DMP in different chromosomes; (b) the methylation vs hydroxymethylation; (c) distribution of DMPs in regions related to genes and CpGs; (d) distribution of DhMPs in regions related to genes and CpGs; (e) hyper vs hypomethylation; (f) distribution of hypermethylation in regions related to genes and CpGs; (g) distribution of hypomethylation in regions related to genes and CpGs.

Full size image

Comparative analysis exhibited uniquely methylated probes

The comparative analysis of the present blood DNA methyl array data with the previously existing tissue DNA methylation dataset represented that despite few, most of the genes and probes were uniquely methylated (Fig. 3a,b). The distribution of differential methylation in different chromosomes indicated a discriminative picture. The highest methylation was observed in chromosome 1, above 10%, followed by chromosomes 2 and 19, showing the lowest methylation (< 5%) compared to other datasets (Fig. 3c). The distinct methylation pattern indicated that blood DNA methylation might have an additional role in NPC progression.

Figure 3
figure 3

Comparative methylation analysis of the present methyl array data with the existing NPC methyl array GEO dataset. In the first panel, (a) shows the volcano plot of two NPC GEO dataset, and (b) indicates the comparative analysis through Venn diagram; (c) demonstrates the comparative distribution of DMPs in different chromosomes.

Full size image

Early evidence pointed out that CpG methylation in untranslated regions, promoters, and gene bodies could alter gene expression in many cancers. Recent studies have demonstrated the role of DNA Methylation in the untranslated regions. Methylation in the 5ʹ-UTR regions often upregulates the gene expression. DNA methylation in 3ʹ-UTR is also functionally involved in gene regulation in the pan-cancer network20,21,22. A correlation between CpG methylation in the gene body and cancer prognosis is now established23. For further continuation of our study, probes in the CpG island containing hypermethylation and hypomethylation were selected. With this criterion, we found a total of 25 probes that correspond to 22 genes; of these, 21 probes were hypermethylated. The most significant number of hypermethylation and hypomethylation was noted in PLCB3 (Δβ: 0.30, adj. p 0.015) and HLA-DRB5 (Δβ: -0.52, adj. p 0.043) genes, respectively. Among 22 genes, PLCB3, C18orf1, ZNF516, PRKCZ, KDM4B, HLX, MGRN1, UHRF1, SPI1, PLEC1, MPO, FLNB, MLLT1, and HLA-DRB5 resided in the gene body; promoter containing six genes (FGR, COL11A2, SMTN, KCNT1, and APEH); 5ʹ UTR, 3ʹ-UTR and 1st exon were having MBP, ADRBK1 and FUT4 genes respectively. The overall result is depicted in Table 1. The identified probes were cross-validated through TCGA dataset analysis. Data showed that the distribution of DMPs varied according to cancer types, among which cancer BRCA, HNSC, KIRC, KIRP, and UCEC were enriched with statistically significant methylation (p < 0.05).

Table 1 Differentially methylated probes in the CpG island in NPC.
Full size table

Selected gene characteristics such as coding sequence length, transcript, 5ʹ-UTR, 3ʹ-UTR, genome span, GC content, distribution of gene types, and number of exons were compared with genomes using chi-squared and Student’s t-tests. Data indicated that the identified genes had a statistically significant (p < 0.05) amount of 5ʹ-UTR, an increased percentage of GC content, and chromosome distribution towards the coding region (Fig. S3).

Functional analysis identified a cancer-specific correlation between DNA methylation and gene expression

All 22 genes were validated for gene expression by nalysing the RNA sequencing data from the GEO dataset. Fold change was determined by dividing the mean of control FPKM with that of RPKM of cancer. The Data explained that genes SMTN, KCNT1, APEH, ZNF516, PRKCZ, KDM4B, MGRN1, MPO, MLLT1, and FLNB were upregulated. FGR, COL11A2, MBP, PLCB3, HLX, UHRF1, SPI1, FUT4, and HLA-DRB5 were downregulated (Fig. 4a).

Figure 4
figure 4

Function analysis of the uniquely methylated probes. (a) The RNA-seq. analysis results from GEO dataset and (b) shows the result of TCGA methylation analysis and correlation detection between DNA methylation and gene expression of identified probes.

Full size image

We further investigated the correlation between DNA methylation and gene expression by determining the correlation coefficient (r), where r with a positive, negative, and zero value symbolized the positive, negative, and no correlation, respectively. P value ≤ 0.05 was considered significant. Results demonstrated that the methylation of the genes FGR, COLIIA2, APEH, HLX, SPI1, and FUT4 were negatively correlated with gene expression. Similarly, a reverse correlation was observed from the probe of SMTN. These two data were consistent with the RNA-seq result. Conversely, the probe of MGRN1 and MPO did not follow the trend of the RNA-seq data. Additionally, methylation in PLCB3, MBP, ZNF156, PRKCZ, KDM4B, UHRF1, MLLT1, FLNB, and HLA-DRB5 showed both positive and negative associations with gene expression in different tissue-specific carcinomas, suggesting that the methylation and gene expression could be tissue specific. Probes that did not show significant methylation were not included in the correlation analysis. The result of the correlation analysis is depicted in Fig. 4b and Table S4.

Validation of genome-wide methylation and gene expression data

To check the accuracy of the methyl array and RNA sequence data, MSP-RT and qRT-PCR for gene expression were employed. MSP has been widely used in clinical settings for the diagnosis and prognosis of diseases such as cancer due to its high specificity, sensitivity, cost, and labor-saving method24. Using SYBR-Green technology in real-time PCR enables the objective evaluation of MSP data, providing an easy way to express the methylation status semi-quantitatively without having to perform laborious gel electrophoresis, as shown by existing research25,26,27. Methylation status is quantified by evaluating the difference between the CT values (ΔCT = uCT − mCT) as described by Yoshioka et.al28, where u and m indicate unmethylation and methylation, respectively. Results of the MSP RT-PCR study revealed that while ZNF516, PLCB3, MBP, and FGR were found to be hypermethylated in NPC (ΔCTZNF516 = 0.5, ΔCTPLCB3 = 0.639, ΔCTMBP = 1.138 and ΔCTFGR = 0.733), HLA-DRB5 was hypomethylated. The result of the gene-wise methylation difference and melting curve of each gene is depicted in Figs. 5a and S5, respectively. Gene expression study showed a significantly lower expression of RNA in PLCB3, MBP, and FGR, HLA-DRB5 (p ≤ 0.05) and a higher expression in ZNF516 (Fig. 5b).

Figure 5
figure 5

The methyl array and RNA sequence data validation by MSP-RT PCR and quantitative RT-PCR. (a) The result of MSP-RT PCR. (b) Gene expression by qRT-PCR. P value ≤ 0.05 was considered significant.

Full size image

DNA Methylation in CpG harbors several transcription factor binding sites

Transcription factor takes the overall control of gene regulation, thus maintaining normal cellular functions. They are commonly dysregulated in many human cancers and are associated with approximately 20% of oncogenesis29. DNA methylation, mainly in the CpG region, potentially alters TFs binding to DNA and promotes gene regulation modification30. We predicted TFBSs present in the CpG island of designated probes. TFs were selected according to the PWM relative score ≥ 0.8 and a p-value < 0.05. The shading of the boxes indicated the p-value of the profile’s match to that position (scaled between 0 and 1000 scores, where 0 corresponds to a p-value of 1 and 1000 to a p-value ≤ 10–10); thus, the darker the shade, the lower (better) the p-value. The predicted score was adjusted to 600, and TFs with a score more than 600 were obtained. Data indicated that the identified genomic region of FUT4 and APEH were enriched with several TF binding sites that scored ranging from 963 to 611. Moreover, binding sites for Zinc finger proteins (ZBTB18, ZNF680, ZNF460, ZNF135, ZNF384, ZNF354A, Zic3, Zic1) and FOX family TFs (FOXD3 and FOXO1) were mostly observed. TFs in the ZNFs and FOX family modulates major regulatory pathways and involve in the onset of cancers. The result of TFBS prediction is depicted in Fig. 6 and Table S7.

Figure 6
figure 6

Prediction for transcription factor binding site in the region of methylation.

Full size image

Methylation in CpG could modulate immunological functions in various cellular pathways

Pathway enrichment study is a statistical method of exploring pathways enriched in the input gene list relative to what is expected by chance31. The enrichment analysis results are displayed in Fig. 7 and Table S8. After the analysis, pathways were filtered based on a user-specified FDR cut-off, followed by shorting the significant pathway by FDR, Fold Enrichment, or other metrics. Results indicated that genes were connected with different immune-modulatory functions, such as T-cell differentiation, cell–cell adhesion, neutrophil, and leucocyte-mediated immunity etc. (Fig. 7a). A hierarchical clustering tree was created to comprehend the correlation among significant pathways listed in the enrichment tab, in which bigger dots indicate more significant P-values. Data showed that T-cell differentiation, cell–cell adhesion, and leucocyte activation were potentially augmented (Fig. 7b). We further generated the interactive plot to establish the intricate relationship between genes and enriched pathways (Fig. 8a). The Gene interaction network was constructed to check the protein–protein interaction among the selected genes; a bigger node indicates a significant interaction (Fig. 8b).

Figure 7
figure 7

Gene enrichment analysis. (a) The enrichment plot, and (b) indicates the hierarchical clustering tree summarizes the correlation among significant pathways.

Full size image
Figure 8
figure 8

Network analysis: (a) indicates the interaction of identified genes with biological pathways. (b) Describes the interconnection among the genes.

Full size image

Discussion

So far, the study of DNA methylation in NPC has mainly focused on tumor suppressor genes’ methylation patterns17. Existing genome-wide DNA methylation reports deciphered high methylation among a Chinese population. The report suggested that this could affect the current therapy in NPC patients5. Another methylome study explained increased methylation at chromosome 6p in the same ethnic population18. Recently, peripheral blood methylation signatures in various solid tumors exhibited a distinct epigenetic signature, indicating cancer’s unique prognostic and diagnostic biomarkers and their resistance to therapy16,32,33,34. Blood DNA methylation in head and neck cancer (HNSCC) was first reported in 2006. It explored an independent association between hypomethylation and HNSCC prognosis and displayed a complex relationship with the known risk factors associated with this cancer35. This growing evidence indicated the additional role of blood DNA methylation on cancer progression, metastasis, and recurrence.

Our current study exhibited a different pattern of methylation in NPC. Chromosomes 1 and 2 have the highest methylation, indicating a distinctive signature from NPC tissue DNA methylation. According to previous reports, the gene body’s methylation represses intragenic transcription and allows efficient transcriptional elongation23,36. Our data suggested that the distribution of DMPs in the location relative to the gene and CpG showed increased methylation in the gene body and the open sea region. A similar pattern was also found in other cancers. A higher amount of hypermethylation in the NPC patients’ samples was observed in this study. This finding was consistent with the previous reports published in China, signifying that hypermethylation is comparatively higher in NPC than in other cancers18.

CpG methylation is a primary concern of cancer risk. The Discovery of hypermethylation in the CpG island of certain tumor suppressor genes interrupts many cellular pathways like DNA repair, apoptosis, cell cycle control, cell–cell adhesion, etc.37. Unlike hypermethylation, hypomethylation in CpG comes later in cancers and contributes to metastatic tumor heterogeneity38. We found twenty-two unique genes (corresponding to twenty-five probes) that were differentially methylated in their CpG islands. The functional study analyzing RNA sequence revealed the differential expression of those genes in NPC, and TCGA analysis further reinforced the study. The results of the genome-wise methyl array and RNA sequence were further cross-validated by performing MSP-RT PCR and quantitative RT-PCR for gene expression; both exhibited consistency with the previous data, suggesting the accuracy and robustness of the methylation analysis.

Moreover, a correlation was detected between methylation of the specific probe and gene expression in cancers other than NPC. This study also implied that the methylation within the particular region of CpG and gene expression could vary according to cancer. Further, the analysis of TFBS of CpG methylated promoter and gene bodies illustrated binding sites for several essential TFs such as ZNFs and FOXs. Early reports demonstrated that ZNF and FOX family TFs are crucial in cell migration and proliferation39,40. A differential expression of other TFs investigated in many cancers is associated with tumors’ development, spread, invasiveness, and lack of proper immune responses41,42,43,44,45. It can be proposed that CpG methylation could alter the binding of TFs, which is thought to modify the gene expression of the corresponding proteins, resulting in NPC advancement. However, further research is needed to imply the hypothesis.

Additionally, the gene enrichment analysis of the identified genes was connected to various cellular pathways and immune activations. Previous research projected the contribution of these genes in different cancers like breast, pancreatic, and ovarian. Genes PRKCZ (Protein Kinase C Zeta), FGR (Src family protein), KDM4B, MPO, COL11A1, FUT4 (Fucosyltransferase IV), and APEH (acylpeptide hydrolase) are directly associated with tumor aggressiveness, growth, and migration, and invasiveness, resulting in poor clinical outcome in cancer patients41,46,47,48,49,50. Similarly, HLX (H2.0-like homeobox) controls early hematopoiesis and promotes acute myeloid leukemia51, and HLA-DRB5 (HLA major histocompatibility complex, class II, DR beta 5) acts as a receptor for T cell activation. Its expression is shown to be upregulated in breast cancer52.

The present study elucidated a lucid picture of genome-wide blood DNA methylation of metastatic NPC, which could set a goal and a future direction of developing more precise molecular tools based on methylation biomarkers for detecting and diagnosing primary and advanced NPC. Though the study was statistically robust and comprehensive, a few limitations exist. Limited sample size could affect statistical interpretation, and a lack of downstream experiments might create a lag for further validation.

Methods

Study selection

Blood samples from four NPC patients were collected from the Eden Medical Centre, Dimapur, Nagaland, India, between 2018 and 2019. For the clinical examination, detailed medical history, physical check-up, Nasopharyngioscopy, serum biochemistry, blood count, chest X-ray, CT or MRI examination of the nasopharynges, skull base, and any suspect metastatic sites (paranasal sinuses) were tested. All the patients suffered from non-keratinizing undifferentiated carcinoma (NKUC). NPC types were determined following the guidelines of WHO53. TNM classification was confirmed according to the recommendation of the American Joint Committee on Cancer (AJCC)54,55. All the samples were obtained before the start of the diagnosis. Four control samples were obtained separately from matched disease-free healthy individuals’ age, sex, and ethnicity. Both cases and controls tested negative for Epstein Bar Virus (EBV). Written informed consent was obtained from all the patients. Ethical approval was obtained from the ethics committee of Visva-Bharati University, West Bengal and Eden Medical centre, Dimapur, Nagaland, as the participating institute. All methods were performed in accordance with the relevant guidelines and regulations. Demographic and Clinical characteristics of NPC cases and controls are shown in Table S1.

Methylation profiling and data processing

Genomic DNA was extracted from the whole blood of eight samples, followed by the bisulfite conversion using the EZ DNA methylation Gold Kit (Zymo Research, USA). The whole-genome methylation array was carried out using the Infinium Methylation EPIC BeadChip Kit (Illumina Inc, USA). The array data (IDAT files) were nalysed using the ChAMP Bioconductor package in R studio (Version 1.2.5042)56. Based on the methylation intensities, the methylation level was measured by the beta-value (β) method, ranging from 1 to 0 (where the value of 1 indicates a fully methylated probe and 0 indicates an unmethylated probe)57. Β value is calculated as the ratio of the methylated probe intensity to the overall intensity (sum of methylated and unmethylated probe intensities)11.

From the methylation EPIC BeadChip, 865,918 methylated probes were detected initially. Probes having bad samples (p > 0.01), a bead count of < 3, having non-CpGs, containing SNPs, probes aligned to multiple locations, and those located at X and Y chromosomes were removed. Eventually, 731,042 methylated probes were recorded. The result of the quality control (QC), beta value distribution plot, and Frequency polygon is shown in Fig. S1. The beta-mixture quantile normalization (BMIQ) method was used to adjust the probe type or color bias, subtract background signals, and eliminate systematic errors (Fig. S2). It is a well-established normalization method that decomposes the β profiles of Type I and Type II probes into two mixtures of 3 methylation states. Then quantile normalized the three distributions of Type II profile corresponding to those of Type I profile58. Wilcoxon signed-rank test was performed for each locus of the sample group (Cancer vs. Control) to identify a significantly differential methylated locus. Multiple testing correction was performed over p values using the Benjamini-Hochberg (BH) method. The cut-off for differential methylation was restricted to BH-adjusted p-values < 0.05 and absolute beta-value (mean beta-value difference between cancer and control) or Δβ ≥ 0.2. CpG was considered hypermethylated or hypomethylated if the Δβ was ≥ 0.2 or ≤ 0.2, respectively59,60. The outline of the overall process is depicted in Fig. 9.

Figure 9
figure 9

Schematic diagram of methylation data processing.

Full size image

Comparative analysis of methyl array data with global NPC methylation dataset

The present methyl array data were compared with available global NPC methyl array data (GSE52068 and GSE62336) deposited in the NCBI-GEO database (https://www.ncbi.nlm.nih.gov/gds). Data from GEO datasets were retrieved and re-analyzed by ‘GEOquery’ and ‘Limma’ R Bioconductor packages. Genes and CpG probes from all the datasets were compared separately by a Venn diagram to view common and unique genes or probes.

RNA-Seq and TCGA data analysis

RNA-sequencing data of two NPC and one para-cancerous tissue was retrieved from the GEO database (GSE134886) and re-analyzed to inspect the gene expression. Comprehensive methylation analysis was performed between the differentially methylated probes (DMPs) of NPC and other carcinomas existed in the Cancer Genome Atlas Program (TCGA): bladder urothelial carcinoma (BLCA, n = 424), breast invasive carcinoma (BRCA, n = 853), cholangiocarcinoma (CHOL, n = 45), colon adenocarcinoma (COAD, n = 288), esophageal carcinoma (ESCA, n = 190), head and neck squamous cell carcinoma (HNSC, n = 535), kidney renal clear cell carcinoma (KIRC, n = 333), kidney renal papillary cell carcinoma (KIRP, n = 292), liver hepatocellular carcinoma (LIHC, n = 409), lung squamous cell carcinoma (LUSC, n = 370), thyroid carcinoma (THCA, n = 558), and uterine corpus endometrial carcinoma (UCEC, n = 187). Further, the Pearson method was set to calculate the corresponding correlation coefficient to conduct a pair-wise correlation analysis between methylation and gene expression.

Bisulfite conversion, methylation-specific real-time PCR and quantitative RT-PCR

Bisulfite conversion was performed by EpiJET Bisulfite Conversion Kit (Thermo Scientific, USA) following the manufacturer’s instruction. During the conversion process, methylated cytosine remains unchanged, while unmethylated cytosine is converted into uracil. Two types of primers and the SYBR Green qPCR Master Mix were utilized for MSP-RT. CpG island was identified for each gene of interest (Fig. S4), followed by designing primers specifically for the site of the methylation-favoured regions of individual genes. One primer set was designed for fully methylated sequences that recognize unconverted cytosine during bisulfite treatment. In contrast, other primer sets recognize fully unmethylated sequences and bind to uracil instead of cytosine (Table S5). RT-PCR was performed under the following conditions: An initial denaturation at 95 °C for 5 min, then 35 cycles of 95 °C for 30 s, 50–55 °C (varied for different genes primer) for 30 s, 72 °C for 1 min, and a final step at 72° for 5 min.

RNA was extracted from NPC and control samples using QIAamp RNA Blood Mini Kit (Qiagen, USA) following the manufacturer’s protocol to validate the RNA sequence data for gene expression. After cDNA synthesis, RT-PCR was carried out at 95 °C for 3 min, followed by 40 cycles of denaturation at 95 °C for 30 s, annealing at 55–60 °C (varied for different genes primer) for 30 s, and amplification at 72 °C for 1 min. Comparing CT values of NPC and control, fold change was calculated. Primers for the qRT-PCR are listed in Table S6.

Transcription factor binding site prediction

For the prediction of transcription factor binding sites (TFBS) of the methylation-specific probes, the JASPAR 2022 module of the UCSC Genome Browser was employed61. JASPAR CORE TF-binding profiles read each taxon independently using position weight matrix (PWM) Scanning and generate a score for each TF.

Gene ontology and interaction network analysis

Gene ontology analysis (GO) was performed with an FDR cut-off < 0.05 using the open-source server ShinyGO 0.7662. It accesses the Ensembl and pathway database from many other sources and uses many R packages to visualize and analyze a gene’s relative functions. FDR was calculated based on the nominal P-value from the hypergeometric test. Depending on the FDR-specified cut-off, Pathways were filtered. Significant pathways were sorted by FDR, Fold Enrichment, or other metrics. The preferred GO ontology functions were selected, viz., biological process, molecular function, and cellular component.

Using the integrated CHAMP-FEM package, we identified the connected subnetwork of the protein interaction of a differentially methylated promoter region with a large average edge-weight density. Weight edges were constructed from the statistical association of DNA methylation with the phenotype of interest (case vs. control), where uniquely methylated genes were considered as seeds for the network construction. Queries were uploaded onto NetworkAnalyst (version 3.0), a comprehensive statistical network visual analytics platform63. IMEx Interactome database was selected, which uses literature-curated complete data from InnateDB64.

Ethics declarations

Obtained from Visva-Bharati and the participating institute.

Consent to participate

Informed consent was taken from each patient.