TRH site-specific methylation in oral and oropharyngeal squamous cell carcinoma

The research methodology employed in this study was approved by The Institutional Review Board of the Faculty of Medicine, Chulalongkorn University, Bangkok, Thailand (IRB No. 426/58 and 135/59). All study subjects provided written informed consent.

The bioinformatics pipeline used in this study is shown in Fig. 1a. Methylation microarray data were retrieved from the Gene Expression Omnibus (GEO) repository of NCBI (http://​www.​ncbi.​nlm.​nih.​gov) [7]. These data contained the Series as GSExxxx, Platform as GPLxxxx, and Sample accession numbers as GSMxxxx. The methylation profiling array Platform, GPL8490 (Illumina® HumanMethylation27 BeadChip Kit, Illumina Inc., San Diego, CA, USA), was used to search for GSE data related with the following keywords: head and neck cancer, HNSCC, OSCC, and normal oral epithelial cell. The GPL8490 methylation profiling array Platform was selected because it has been widely used and provides more data than other microarrays [8, 9]. Samples from each GSE were analyzed according to the inclusion criteria (normal epithelium, precancerous epithelium, and cancerous epithelium of head and neck), and exclusion criteria (cell line, stem cell, non-human tissue, non-head and non-neck tissue sample, inflammation, congenital disease, and blood cells). Based on these criteria, data were narrowed down to seven GSEs (by the end of July 31, 2015) as shown in Fig. 1b [6, 10‐13]. The Connection Up- and Down-Regulation Expression Analysis of Microarrays program [14] was used to calculate the average methylation percentage of differentially methylated CpG sites of each sample from the seven GSEs; these data were recorded in Microsoft Office Excel (Additional file 1). The average methylation percentages of normal, precancerous, and cancerous cells were plotted for each CpG site using R.

To validate the data obtained using bioinformatics, a total of 18 subjects, including 9 healthy controls (four males and five females) and 9 OSCC subjects (five males and four females) were recruited. The samples were supported by the Department of Pathology, Faculty of Medicine, Chulalongkorn University, Bangkok, Thailand. Tissues of gum were collected from healthy controls, and those of tongue, lips, and gingiva were collected from cancer subjects (histological grade; six well-differentiated, one moderate-differentiated, two poorly differentiated). Healthy and cancerous tissue samples were stored as formalin-fixed paraffin-embedded (FFPE) tissues [15, 16]. All FFPE samples were sliced into five mirror-image 3 μm thick sections. One section was stained with hematoxylin and eosin stain and visualized under a microscope, and the four remaining sections were labeled with a permanent marker and microdissected to isolate healthy and cancerous cells. The histological grade of cancer was identified by a pathologist (NK).

Samples recruited from Department of Otolaryngology, Faculty of Medicine, Chulalongkorn University, Bangkok, Thailand. There are 2 cohorts as followed:

Cohort 1. respresented as a discovery set, derived from patients during 2015–2016. This composed of oral rinse from 33 healthy controls and oral rinse as well as oral swab from 23 oral cancer subjects.

Cohort 2. represented as a validation set, derived from patients during 2016–2017. This composed of oral rinse from 54 healthy controls and oral rinse from 42 oral as well as 24 oropharyngeal cancer subjects.

Volunteers for this experiment were examined for their oral health. Subjects with no oral mucosal lesion, history of malignancy, or smoking behavior were included as healthy controls. All cancer subjects with oral or oropharyngeal cancer confirmed on the basis of incisional biopsy results of squamous cell carcinoma (SCC) were included in this experiment for comparisons. Oral mucosal cells were collected using oral rinse and oral swab techniques. Oral rinse was collected from oral and oropharyngeal SCC subjects and healthy controls, whereas oral swab was collected only from OSCC subjects. All demographic data, cancer histological grade and clinical stage were demonstrated in Additional file 2. For the oral rinse technique, the study subjects were required to gargle 10 mL of sterile 0.9% saline solution for 15 s. For collecting oral swab, a foam-tipped applicator (Puritan Medical Products, ME, USA) was applied to the OSCC lesion for 5–10 s. The oral rinse solutions and swab foams were stored in sterile tubes at 4 °C until needed for DNA extraction.

DNA was extracted from healthy and cancerous cells using QIAamp DNA FFPE Tissue Kit (Qiagen, Valencia, CA, USA). The absorbance of extracted DNA was measured using NanoDrop (ND-1000 Spectrophotometer, NanoDrop Technologies, Wilmington, DE, USA) at 260 and 280 nm to verify its purity and concentration. Subsequently, DNA was subjected to bisulfite conversion using EZ DNA Methylation-Gold™ Kit (Zymo Research Corporation, Irvine, CA, USA) [17, 18] following the manufacturer’s instructions. Treatment of DNA with bisulfite converts unmethylated CG dinucleotide to TG but leaves the methylated CG dinucleotide (CpG sites) unaffected.

Primers for polymerase chain reaction (PCR) amplification of DNAs and sequencing of PCR products were designed using Pyromark Assay Design software (Qiagen, Valencia, CA, USA). DNAs were amplified using the forward (5′-GGG GTT TTT AGA GTT GTA GAT TTT TGA-3′) and reverse (5′-CCA AAA ATA AAC TCC ACA AAA TAA ATC-3′) primers and sequenced using the sequencing primer (5′-TTT AGA GTT GTA GAT TTT TGA TTT G-3′). Pyrosequencing was then performed using the Pyromark Q96 ID platform (Qiagen, Valencia, CA, USA) to determine the methylation level according to the manufacturer’s instructions.

For screening purposes, real-time PCR was established, as it is easier than pyrosequencing and suitable for a large sample size. We designed the real-time methylation-specific PCR, which was performed using the Sensi-FastTM SYBR® Low-Rox Kit (Bioline, Alexandria, NSW, Australia). PCR reactions were prepared in a volume of 20 μL containing 10 μL of 2X Sensi-FastTM SYBR® Low-Rox reagent, 0.8 μL of 500 nmoles of each gene-specific forward and reverse primers and 1 μL of bisulfite-treated DNA template; the remaining volume was adjusted by adding milliQ DNase-free sterile water. Two gene-specific primer pairs were used in this experiment: one pair was specific to the target gene in its methylated state (met forward primer, 5′-ATT CGG GGA TTC GGG ATT C-3′ and met reverse primer, 5′-GAC GAC CCA TCT AAA AAA AAC TCG-3′) and the other pair was specific to the target gene in its unmethylated state (unmet forward primer, 5′-GGA TTT GGG GAT TTG GGA TTT-3′ and unmet reverse primer, 5′-CAC TCA AAC CAC CAC CTA ACA-3′). Real-time PCRs were performed in duplicates using an Applied Biosystem® 7500 Real-Time PCR System (Thermo Scientific™, Waltham, MA, USA). No template control and positive control were included in each PCR to observe for possible contamination. The PCR conditions were as follows: 45 cycles of denaturation at 95 °C for 2 min followed by annealing at 60 °C for 30 s. Fluorescence signals from the amplified product were detected at the end of the annealing step. A melting curve was generated to determine the specificity of the primers. The threshold cycle (C_t) of the amplified methylation products was detected. Non-methylation-specific PCRs were used as an internal control to adjust C_t values. C_t values of all samples were used to calculate DNA concentrations on the basis of the standard curve prepared using 10-fold serial dilutions of universal methylated DNA standard (Zymo Research Corporation, Irvine, CA, USA) (10, 1, 0.1, 0.01, and 0.001 ng/μL).

All statistical analyses were conducted using SPSS software for Windows version 22 (SPSS Inc., Chicago, IL, USA). Independent sample t-test was performed to determine differences among all groups except matched cases of oral rinse and swab samples, for which the paired t-test was used. p-values < 0.05 were considered statistically significant. A receiver operating characteristic (ROC) curve was generated to determine the diagnostic ability of gene methylation level to differentiate between epithelial cells from oral rinse and swab samples and those from healthy controls.

Following the bioinformatics approach (Fig. 1a), 27,578 CpG sites were identified from the GEO repository that showed differential methylation between healthy and cancerous cells. The average methylation percentage was plotted for each differentially methylated CpG site. From a collection of 27,578 graphs, one graph showing the highest difference in methylation levels between healthy and cancerous cells was selected. This graph represented the CpG site, cg01009664 in the thyrotropin-releasing hormone (TRH) gene sequence (Fig. 1c).

DNA isolated from microdissected oral epithelial cells of nine healthy and nine OSCC FFPE samples were subjected to pyrosequencing for the validation of the CpG site, cg01009664. Pyrosequencing revealed five differentially methylated CpG sites surrounding cg01009664 in the TRH gene (Fig. 2a and b). The average methylation percentage of the nine samples at the second CpG site of TRH (cg01009664) in healthy cells (7% ± 1.14%) was significantly lower than that in cancerous cells (58.44% ± 5.68%) (p < 0.001; Fig. 2c). Similarly, the average of methylation percentage of nine samples at the five CpG sites surrounding cg01009664 in healthy cells (5.7% ± 0.85%) was also significantly lower than that in cancerous cells (52.96% ± 5.36%) (p < 0.001; Fig. 2d).

Next, we evaluated TRH methylation levels as a screening marker using real-time PCR. Oral rinse samples from healthy controls showed an amplification signal only for unmethylated TRH, whereas the amplification of methylated TRH was barely detectable (Fig. 3a). By contrast, oral rinse samples from OSCC subjects showed amplification of both methylated and unmethylated TRH (Fig. 3b). The standard curve was prepared using 10-fold serial dilutions (10–0.001 ng/μL) of control DNA samples (Fig. 3c) and used to calculate the concentration of methylated TRH samples based on their C_t values (Fig. 3d).

In cohort 1, the average TRH methylation level of oral rinse samples collected from healthy controls was significantly lower than that of oral rinse samples collected from OSCC subjects (2.66 ± 0.76 ng/μL vs. 3.77 ± 0.60 ng/μL, respectively; p < 0.001) (Fig. 4a, Additional file 2). Additionally, the average TRH methylation levels of oral swabs were significantly higher than that of oral rinses in the same case significantly (4.17 ± 0.58 ng/μL vs. 3.77 ± 0.60 ng/μL, respectively; p = 0.0012) (Fig. 4b, Additional file 2).

In cohort 2, the average TRH methylation level of oral rinse samples collected from healthy controls was significantly lower than that of oral rinse samples collected from OSCC subjects same as cohort 1 (2.86 ± 0.64 ng/μL vs. 3.78 ± 0.48 ng/μL, respectively; p < 0.001) (Fig. 4c, Additional file 2). Moreover, the average TRH methylation levels of oral rinse samples collected from oropharyngeal SCC subjects was significantly higher than that of healthy controls same as OSCC subjects (3.54 ± 0.37 ng/μL vs. 2.86 ± 0.64 ng/μL, respectively; p < 0.001) (Fig. 4c, Additional file 2). The average TRH methylation levels of oral rinse samples collected from oropharyngeal SCC subjects was a little bit lower than OSCC subjects but not significantly different (3.54 ± 0.37 ng/μL vs. 3.78 ± 0.48 ng/μL, respectively) (Fig. 4c, Additional file 2).

All cancer samples, there were no significant differences in TRH methylation levels between different genders, ages, stages, and tumor grades (data not shown).

To evaluate the potential use of TRH methylation as a biomarker for detecting OSCC and oropharyngeal SCC, we performed ROC curve analysis. This method is an effective statistic for assessing the diagnostic test. It compromises the value of sensitivity and specificity, determining the best cut-off points, which receives the highest true positive rate along with lowest false positive rate. The area under the curve (AUC), cutoff value, sensitivity, and specificity for cohort 1, cohort 2 and cohort 1 + 2 can be seen in Table 1. Combination of oral rinses collected from OSCC subjects in cohort 1 and 2, ROC analyses using the cutoff TRH methylation value 3.31 ng/μl revealed a sensitivity of 86.15%, a specificity of 89.66% and an AUC of 0.93. At the same cutoff value oral rinse collected from oropharyngeal SCC, showed a sensitivity of 82.61%, a specificity of 92.59% and an AUC of 0.88. The ROC curves also indicated that oral swab samples (sensitivity of 91.30%, a specificity of 84.85% and an AUC of 0.97) were superior to oral rinse samples for detecting OSCC.