2.1 Study participants and sample collection

This observational cross-sectional study included six consecutive patients with early-stage OTSCC diagnosed between September 2020 and January 2021 at the Department of Oral and Maxillofacial Surgery from Complexo Hospitalario Universitario of A Coruña (CHUAC, SERGAS) in Galicia, Spain. Inclusion criteria included histologically confirmed OTSCC at early-stage pT1-pT2N0M0 according to the 8th edition of the Tumor-Node-Metastasis (TNM)-staging system promulgated by the American Joint Commission on Cancer (AJCC) and the International Union Against Cancer (UICC). Regarding family history, no relevant data related to cancer were identified. With respect to past medical history, no patient received surgery, chemotherapy, or radiotherapy before sample collection. This study was approved by the Galician Ethics Committee of Clinical Research (Ref. No. 2018/003) and carried out in accordance with principles outlined in the Declaration of Helsinki (World Medical Association, 2013). Written informed consent was obtained from all participants before enrollment in the study and sample collection.

Tumor and epithelium without dysplasia adjacent to tumoral tissue were obtained from tissue biopsy (Figure S1). The histologic sections were subsequently confirmed by an expert pathologist (R.A.R). Saliva was also collected from OTSCC patients at the time of cancer diagnosis. Additionally, two salivary samples from healthy individuals were obtained. Unstimulated saliva samples were collected using Danasaliva sample collection kit (Danagene) according to the manufacturer's instructions. The participants were asked to refrain from eating, drinking, smoking, and using oral hygiene products for at least 1 h prior to sampling collection. After collection, saliva was centrifuged at 2600 g for 15 min at room temperature, separating the cellular pellet from the cell-free salivary supernatant. Then, the cellular pellet was resuspended with phosphate-buffered saline (PBS) and preserved at −20°C prior to assay. A description of the clinicopathological characteristics of patients is shown in Table S1.

2.2 DNA extraction and bisulphite modification

DNA extraction from formalin-fixed paraffin-embedded (FFPE) tissue samples was performed using the AllPrep® DNA/RNA FFPE kit (Qiagen) according to the manufacturer's recommendations. Genomic DNA from tumor samples was eluted in 100 μl elution buffer AE and stored at −20 °C for further use. The QIAmp DNA Blood Mini Kit (Qiagen) was used to extract DNA from cell pellets of saliva samples according to the manufacturer's instructions. Genomic DNA from cell pellets was eluted in 100 μl elution buffer AE and stored at −20 °C for further use. DNA concentration and quality were measured by Nanodrop Spectrophotometer (Thermo Fisher Scientific) and by Qubit 4.0 Fluorometer (Thermo Fisher Scientific USA). Agilent's TapeStation 4200 (Agilent Technologies) was used to assess the integrity of the extracted DNA. Before bisulphite conversion, the DNA from FFPE samples was undergone to Illumina FFPE QC kit for quality control (QC) as per the manufacturer's protocol. Then, 400 ng of extracted DNA from each sample was used for bisulfite conversion using the EZ DNA Methylation Kit (Zymo Research) according to the manufacturer's instructions. After bisulphite conversion, only FFPE-isolated DNA was further subjected to restoring with the Illumina Infinium HD FFPE Restoration kit as per the protocol. Genome-wide DNA methylation profiling was carried out using the Illumina Infinium MethylationEPIC BeadChip (EPIC array; Illumina), which overspreads the DNA methylation profile across approximately 850,000 CpGs. A total of 4 μl of the bisulphite converted DNA was used as template for target preparation to hybridize on the BeadChip and was processed as per the manufacturer's instructions. To avoid the batch effect, all samples were processed together. The Bead Chip was imaged on HiScan System and intensity values (IDAT files) were extracted.

2.3 Genome-wide DNA methylation analysis

Methylation array data were processed in the R 4.1.1 statistical environment (https://www.r-project.org/). The R package (R Core Team, 2021) RnBeads was used for quality control and preprocessing (Assenov et al., 2014). Firstly, a greedycut algorithm was used to filter out probes and/or samples. Next, probes overlapping with single nucleotide polymorphisms, probes in sex chromosomes, and probes whose sequence maps to multiple genomic locations (cross-reactive) were removed. Raw intensities obtained in the array were normalized using the BMIQ method, which is a model-based normalization approach to correct β values of type II probes according to the beta distribution of β values of type I probes. For each CpG site, a specific β value was obtained which is defined as the ratio of fluorescent signal between methylated (M) probe relative to the sum of the M and unmethylated (UM) probes (β = M/(M + UM)). The β values range from 0 (no methylation) to 1 (100% methylation of both alleles). Finally, hierarchical linear models were used with limma package to obtain the differences between groups (Ritchie et al., 2015). p-Values were corrected for multiple testing (false discovery rate, FDR) using the Benjamini–Hochberg method, and a threshold of p < 0.05 was selected for significance. A similar bioinformatic pipeline was applied to all the comparisons performed.

2.4 TCGA methylation analysis

Methylation data were obtained from The Cancer Genome Atlas database (TCGA, 2013). Infinium Human Methylation 450 K BeadChip (HM450K) IDAT files from 36 early-stage OTSCC based on pathological diagnosis and four healthy controls were downloaded from TCGA database using the TCGAbiolinks R package. The IDAT files were subjected to preprocessing, normalization, and QC steps similar to our EPIC array files.

2.5 TCGA gene expression analysis

To identify the functional relevance of the differential methylated genes in OTSCC, we used normalized RNA-sequencing (RNA-seq) expression data from 36 early-stage OTSCC and four healthy controls from TCGA database (TCGA, 2013) using the TCGAbiolinks R package. Differentially expressed genes were obtained using generalized linear models with a function included in the package. A FDR < 0.05 and |log2 fold change (FC)| ≥1 was set as the criteria for screening differentially expressed genes. The genes differentially expressed between OTSCC and normal tissue TCGA samples were linked with the differentially methylated genes in our methylation analysis by Venn diagram using the VennDiagram R package.

2.6 Saliva DNA methylation analysis

To identify salivary DNA methylation changes, we obtained salivary DNA methylation data of four healthy controls (HM450K) from the NCBI Gene Expression Omnibus (GEO) under accession number GSE55734. In addition to the IDAT files from these four healthy controls, we also retrieved for salivary methylation analysis the EPIC array data from two saliva samples previously analyzed in our research group. A FDR < 0.05 and p-value < 0.05 were set as the criteria for screening DMCpG sites.

2.7 Gene ontology and functional pathway analysis

Gene Ontology (GO) enrichment analysis was performed using GeneCodis 4 (Tabas-Madrid et al., 2012) to classify the differentially methylated genes into categories of cellular component, biological process, and molecular function (Ashburner et al., 2000). In addition, the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis was performed to detect the potential pathways of the differentially methylated genes (Kanehisa & Goto, 2000).

2.8 Receiver operating characteristic analysis

To assess the diagnostic value of candidate DNA methylation markers, receiver operating characteristic (ROC) analysis was performed using the pROC package in R (Robin et al., 2011). The area under the curve (AUC) and accuracy were calculated with the aforementioned package to assess the performance of each DNA methylation marker. Principal component analysis (PCA) and ROC curve analysis were assessed to obtain the best models for OTSCC diagnosis in saliva.

3.1 Identification of tumor-specific differentially methylated CpG sites

After preprocessing the EPIC array data, a total of 785,613 methylation CpG sites provided a high-quality methylation signal. The comparison of the mean methylation values of all CpGs sites analyzed between tumor and adjacent non-tumoral tissue samples showed a high correlation across all CpGs (Spearman, r² = 0.99, p < 2.2 × 10⁻¹⁶), indicating a similar global DNA methylation pattern in both groups (Figure S2). Of them, a total of 25,890 tumor-specific DMCpGs spanning 7210 genes with FDR < 0.05 were identified between OTSCC and adjacent non-tumoral tissue samples. The circos representation of DNA methylation locations in the genome showed that these 25,890 DMCpGs were widely distributed throughout all the chromosomes (Figure 1a). In addition, among these DMCpGs, we found a higher number of CpGs hypomethylated (20,505 CpGs; 79% of all DMCpGs) than hypermethylated (5385 CpGs; 21% of all DMCpGs) in OTSCC respect to the adjacent non-tumoral tissue, as indicated in Figure 1b.

In terms of the genomic distribution, DMCpGs were mostly distributed in the gene body (17,152 hypomethylated and 3875 hypermethylated) followed by the 5´UTR (2875 hypomethylated and 1881 hypermethylated), TSS1500 (2569 hypomethylated and 1314 hypermethylated), TSS200 (533 hypomethylated and 1218 hypermethylated), 1st exon (374 hypomethylated and 987 hypermethylated), 3´UTR (644 hypomethylated and 184 hypermethylated), and ExonBnd (344 hypomethylated and 49 hypermethylated). In terms of the CpG context, DMCpGs were mostly distributed in the opensea (17,592 hypomethylated and 1701 hypermethylated) followed by the island (250 hypomethylated and 2565 hypermethylated), shore (1371 hypomethylated and 940 hypermethylated), and shelf (1292 hypomethylated and 179 hypermethylated) (Figure 2). Unsupervised hierarchical clustering and a heat-map using the methylation values of random 1000 DMCpGs, of which 44 were hypermethylated and 956 hypomethylated, were performed. The heat-map showed two DNA methylation clusters indicating that DNA methylation profiling of OTSCC was significantly different from adjacent non-tumoral tissue (Figure S3).

A genome-wide differential methylation analysis in specific promoter regions including CpGIs and CpG shore was performed between tumor and adjacent non-tumoral tissue. A total of 2700 tumor-specific DMCpGs spanning 1597 genes were identified of which 2216 were hypermethylated (82%) and 484 hypomethylated (18%) (Figure 3a). Unsupervised hierarchical clustering analysis and a heat-map using the methylation values of random 1000 DMCpGs within/near to CpGIs, of which 714 were hypermethylated and 286 hypomethylated, was produced. The heat-map showed two DNA methylation clusters indicating that DNA methylation profiling of OTSCC was significantly different from adjacent non-tumoral tissue at specific promoter regions (Figure 3b).

Enrichment analysis of the 807 genes containing DMCpGs revealed several important functional pathways and ontologies (Figures S4 and S5).

3.2 Evaluation of the impact of differential methylation on associated gene expression

The expression of 19,947 genes was accessible from TCGA data, of which 2494 genes were differentially expressed between OTSCC and normal tissue samples (FDR < 0.05 and |logFC| ≥1). Specifically, the expression of 73 genes from the 807 methylated genes identified in our methylation assay was also available in the TCGA data. A total of 40 genes hypermethylated in our assay showed a downregulated expression in tissue samples from TCGA cohort. By the contrary, 33 genes hypomethylated showed an upregulated expression (Figure 4).

3.3 Validation of tumor-specific methylated CpGs using TCGA OTSCC data

Since promoter methylation can modulate tumor-specific gene expression leading to its transcriptional silencing, the methylation levels of 11 genes, whose hypermethylation in their promoter region was associated with a downregulated gene expression, were further validated using TCGA OTSCC methylation data as an independent sample cohort. We evaluated the overall overlap of the DMCpGs located at the promoter region of 40 hypermethylated and 33 hypomethylated genes selected in our discovery assay with the tumor-specific DMCpGs identified from 36 OTSCC and four normal tissue samples which were available in the TCGA database. Although HM450K array used in the TCGA data was limited in genome coverage with almost half the number of CpGs respect to HM850K array, we identified overlapping in 68/88 (77.27%) hypermethylated and 11/34 (32.35%) hypomethylated CpGs. Focused on the hypermethylation, 45 CpGs spinning 22 genes were differentially methylated (FDR < 0.05 and p-value < 0.05) between tumor and normal tissue of which 11 genes (A2BP1, ACSS3, ALDH1A2, ANK1, CA3, GABRB3, GFRA1, HS3ST1, NDRG2, TTYH1, and PDE4B) were methylated in at least two CpGs. As shown in Figure S6 (Table S2), the differential methylation profiles of each CpG methylation loci of these 11 genes found in our OTSCC cohort were also observed in the TCGA OTSCC cohort, thus validating the findings of our discovery assay. Overall, these results suggest that these tumor-specific hypermethylated CpGs are frequent alterations in OTSCC and could be an early event with a crucial role in carcinogenesis.

3.4 Accuracy of tumor-specific DNA methylation markers

In addition, we performed ROC analysis, estimated AUC, and diagnostic accuracy values of the 11 tumor-specific promoter hypermethylated genes (Figure S7). Overall, these DNA methylation markers showed high sensitivity and specificity values (Table S3) indicating the potential of methylation profiling of these gene promoters for early-OTSCC detection.

3.5 Identification of methylated biomarkers in saliva from OTSCC patients in saliva

To test the potential of tumor-specific CpG methylation markers in the A2BP1, ACSS3, ALDH1A2, ANK1, CA3, GABRB3, GFRA1, HS3ST1, NDRG2, TTYH1, and PDE4B genes as salivary biomarkers for early detection of OTSCC, we analyzed salivary cell pellets of four OTSCC patients and six healthy individuals. Differential methylation levels between saliva from OTSCC patients and healthy individuals were observed for specific-site methylated CpGs in A2BP1 (p = 0.005), ALDH1A2 (p = 0.033), ANK1 (p = 0.033), GFRA1 (p = 0.033), PDE4B (p = 0.012), and TTYH1 (p = 0.009) genes (Figure 5). The sensitivity observed for the individual markers ranged from 75% to 100%, and the specificity ranged from 83% to 100% (Figure 6 and Table S4). In addition, using PCA and ROC curve analysis, we were able to obtain the best combination of salivary biomarkers for the detection of OTSCC patients (Figure S8 and Table S5).