Unique DNA Methylation Loci Distinguish Anatomic Site and HPV Status in Head and Neck Squamous Cell Carcinoma

Head and neck squamous cell carcinoma (HNSCC), an anatomically heterogeneous group arising most often from the oral cavity, oropharynx, hypopharynx, larynx, and nasopharynx, is the fifth most common malignancy in men worldwide, representing a major international health problem (1). Tumors originating from these different locations can exhibit varying behavior that is not predictable by histopathology of the primary tumor but is discernible by other methods such as gene profiling (2). Conventional treatments employ surgery, radiation therapy, and chemotherapy either alone or in combination with any, or all, of the other modalities. Any of these therapies can, and do, produce morbidities that affect speech, swallowing, and overall quality of life. Despite treatment intervention, recurrence of the disease is observed in about 50% of patients, either locally, regionally or at a distant site with high rates of associated mortality (3). As a result, the 5-year survival rate has improved only marginally over the past decade; it is estimated that over 45,000 new cases and 11,000 deaths will occur each year in the United States from HNSCC (4).

The search for biomarkers predictive of patient prognosis and treatment response has now progressed to include epigenetic changes associated with tumor initiation and progression. By far, the most common epigenetic event in the human genome is the addition of a methyl group to the carbon-5 position of cytosine nucleotides. This covalent DNA modification occurs predominantly in cytosines immediately preceding guanine nucleotides (the CpG dinucleotide). In HNSCC, promoter methylation of tumor suppressor genes seems to be a common mechanism of transcriptional silencing, as discussed by Ha and Califano (5), Shaw (6), and Pérez-Sayáns (7). In many cases, epigenetic events associated with individual genes have been of prognostic value. For example, methylation of the promoter region of MGMT is associated with increased tumor recurrence and decreased patient survival independent of other factors (8). Methylation of the DCC gene is significantly associated with bone invasion by gingival tumors, aggressive invasiveness of tumors of the tongue, and reduced survival (9). Invasion and metastasis of oral squamous cell carcinoma (SCC) cells have been shown to be dependent on methylation of the E-cadherin promoter (10). There is also evidence of a statistically significant association between DNA hypermethylation of the ADAM23 gene and progression in laryngeal cancer (11). Finally, a highly significant difference in DNA hypermethylation of the DAPK gene promoter is observed between laryngeal cancer patients with and without lymph node metastasis (12). Both MLH1 and CDKN2A are also known to play an important role in laryngeal cancer development and progression (13).

Here, we present a genome-wide view of differential DNA methylation in primary HNSCC tumors from 118 patients at Montefiore Medical Center in Bronx, NY. The goals of this study were several-fold. First, we hoped to identify novel genes affected by aberrant DNA methylation that may play a role in head and neck tumorigenesis but have not been previously identified in this disease. Second, we wanted to determine whether or not HNSCC primary tumors originating from different anatomic sites (oral cavity, larynx, and oropharynx) showed similar DNA methylation changes, or whether some DNA methylation events were specific to a single anatomic site. In each of these cases, the identification of genes specifically affected by DNA methylation represents a way forward to the identification of novel tumor suppressor genes with relevance as potential clinical biomarkers in HNSCC.

Patients recruited for this study were undergoing treatment for histologically confirmed head and neck squamous cell carcinoma at Montefiore Medical Center, Bronx, NY (Table 1). All patients consented to participation in this study under protocols approved by the Institutional Review Boards of the participating institutions. All primary tumor and histologically normal adjacent tissues were snap-frozen in liquid nitrogen within 30 minutes of surgical resection or biopsy and kept at −80°C until further processing. Adjacent mucosa was taken at a grossly unremarkable site away from the tumor identified by the surgeon (incisional biopsy) intraoperatively or the pathologist (resection specimen) when processing the specimen for frozen section or final diagnosis. In the case of tumor resections, this represented histologically normal tissue approximately 1 cm away from the primary tumor. In the case of tumor biopsies, this represented tissue taken from the contralateral side. The distributions of normal tissues for each anatomic site were as follows: oral cavity (32 adjacent, 3 contralateral), oropharynx (19 adjacent, 27 contralateral), and larynx (16 adjacent, 21 contralateral).

Histologic confirmation of tumors was conducted by a single pathologist assessing tumor grade, TNM staging based on the American Joint Committee on Cancer classification, and degree of lymphocytic infiltration. Control samples for all tumors were acquired; microscopy slides stained with hematoxylin and eosin were assessed for percentage of tumor, extent of necrosis, and degree of lymphocytic infiltration as described previously (14). The mean percentage tumor content for samples from each anatomic site, as well as the number of samples with less than 40% tumor content, are both shown in Table 1.

All patient samples were tested for HPV infection using multiple testing methods. The presence of HPV DNA and HPV 16 DNA were assayed by MY09/11-PCR (degenerate and type-specific HPV16-L1 fragments) using AmpliTaq Gold and dot-blot hybridization on matched-fresh frozen specimens as described previously (15). HPV 16 was also assayed using in situ hybridization of samples on formalin fixed, paraffin embedded tissue blocks with multiple probes (15). Finally, the presence of HPV E6 and E7 RNA transcripts was detected and quantitated using TaqMan RT-PCR. Only patients testing positive by all methods were classified as positive (Table 1). Only patients testing negative for all methods were classified as HPV negative. All others were classified as indeterminate.

Genome-wide profiling of differential DNA methylation in tumor samples was carried out as described previously (14). Briefly, genomic DNA was isolated from snap-frozen tissue using the DNeasy tissue kit (Qiagen). Quantity and purity of isolated DNA was measured spectrophotometrically using a NanoDrop ND-1000 spectrophotometer. Bisulfite conversion of tissue genomic DNA was carried out using the EZ DNA Gold methylation kit (Zymo Research). Profiling of genome-wide DNA methylation in genomic DNA (500 ng) was carried out using the Illumina Infinium assay with the HumanMethylation27 DNA Analysis BeadChip. The methylation level at each CpG loci on the beadchip was determined by measuring the methylation fraction (β), defined as the fraction of methylated signal over the total signal (unmethylated + methylated fractions) in each genomic DNA sample. This value ranged continuously from 0 (unmethylated) to 1 (fully methylated) for each CpG locus. A complete description of individual beadchip controls on the Illumina beadchip platform is available from the manufacturer (Illumina). Validation of the beadchip measurements was also carried out on a subset of identified CpG loci using matrix-assisted laser desorption/ionization—time-of-flight (MALDI-TOF) mass spectrometry using EpiTyper by massARRAY (Sequenom) on bisulfite-converted DNA to verify the accuracy of beadchip measurements (Supplementary Fig. S1).

Normalized M values were generated from primary HumanMethylation27 beadchip β values using the R package HumMeth27KQCReport function, including the X chromosome data and using an average probe P-value of 0.03 as the cutoff for sample inclusion (16). The resulting normalized M values represent a continuous variable from −6 (unmethylated) to 6 (methylated). Differences in methylation (ΔM, M_tumor − M_normal) in which both the tumor and adjacent mucosal samples passed quality control were calculated and combined with patient clinical information on nodal metastasis at diagnosis. For identifying differentially methylated CpG loci for each tumor/normal dataset, we selected CpG loci with a Storrey corrected P-value of less than 0.05 and an average ΔM of greater than 1.4 or less than −1.4 across a given dataset, as described previously for this type of data (17). A cutoff of 1.4 was established as the minimum threshold for distinguishing methylated from unmethylated DNA using this type of transformed data (17).

For each primary HNSCC and corresponding normal tissue sample pair, linear amplification and biotin labeling of total RNA (500 ng) were carried out using the Illumina TotalPrep RNA Amplification Kit (Ambion). Whole-genome expression analysis was carried out by hybridization of amplified RNA to an Illumina HumanHT-12 v3 Expression BeadChip. With this beadchip, we interrogated greater than 48,000 probes per sample, targeting genes and known alternative splice variants from the RefSeq database release 17 and UniGene build 188. Controls for each RNA sample (greater than 1,000 bead types) confirmed sample RNA quality, labeling reaction success, hybridization stringency, and signal generation. All expression data were quantile normalized and background-subtracted before analysis using BeadStudio software (Illumina).

We recruited 118 patients undergoing treatment for HNSCC at Montefiore Medical Center in the Bronx, a high-risk area of New York City with a high incidence of the disease (Table 1). All pathologic grades of tumor were represented although the majority of cases (58%) had stage IV disease at presentation. Seventy-three patients (63%) had a positive nodal status at presentation. For all patients, we also collected histologically normal adjacent mucosa in parallel with primary tumor tissue collection. In the case of tumor resections, this represented histologically normal tissue approximately 1 cm away from the primary tumor. In the case of tumor biopsies, this represented tissue taken from the contralateral side.

The goal of our first analysis was a genome-wide view of differential DNA methylation to identify novel hyper- and hypo-methylated CpG loci and their associated genes from each patient dataset (oral cavity SCC, oropharyngeal SCC, laryngeal SCC), and to examine the degree to which these tumors share common epigenetic alterations. We focused our lists for each patient set to significantly altered CpG loci (P < 0.05) where the absolute difference in the average ΔM values was set at a threshold of 1.4 as described previously (17). The results were 293 differentially methylated CpG loci (119 hypermethylated; 174 hypomethylated) in oral cavity SCC, 219 differentially methylated CpG loci (75 hypermethylated; 144 hypomethylated) in laryngeal SCC, and 460 differentially methylated CpG loci (385 hypermethylated; 75 hypomethylated) in oropharyngeal SCC (Table 2). Overall, these numbers represented less than 2% of the total CpG loci surveyed using the HumanMethylation27 beadchip.

Within each of the oral cavity, larynx, and oropharynx SCC datasets, the distribution of these CpG loci (hyper-methylated vs. hypo-methylated) coincided primarily with their chromosomal position within, or outside of, classically defined CpG islands (18). For example, of the 119 CpG loci identified as hypermethylated in our oral cavity SCC patient population, 116 loci (97%) were located within classically defined CpG island sequences in the promoters of genes. In contrast, the vast majority of the 174 hypomethylated CpG loci were located predominantly outside of CpG island sequences (137 loci, 79%). However, it should be noted that a significant number of hypomethylated CpG loci (37 loci; 21%) did reside within CpG island sequences and may represent genes regulated by DNA hypomethylation in head and neck cancer.

Overall, our analysis also showed that head and neck tumors originating from different anatomic sites do share some common epigenetic alterations. An initial overlap of the 3 datasets identified 44 CpG loci (28 hypermethylated; 16 hypomethylated) that showed significantly altered DNA methylation across all 3 anatomic sites of HNSCC (Table 3). Interestingly, only a subset of these genes had been previously studied in head and neck cancer. As with our previous study, the most overrepresented group of genes in this list were the Kruppel family ZNF genes on chromosome 19q13, most notably ZNF132, ZNF154, ZNF542, ZNF545/ZFP82, ZNF671, and ZNF781 (14). Validation of the beadchip measurements for ZNF132 and ZNF154 using MALDI-TOF mass spectrometry using EpiTyper by massARRAY (Sequenom) on bisulfite-converted DNA showed an increase in DNA methylation across multiple CpG loci associated with these genes in primary HNSCC tumors (Supplementary Fig. S6). Genes such as ATP10A, RLN3R1, RYR2, SORCS3, SPARCL1, TBX5, and the Hox genes HOXD9, HOXD10, and HOXD12 also showed hypermethylation across anatomic sites. Of the 16 hypomethylated CpG loci identified across all 3 anatomic sites of HNSCC cancer cases, only 4 loci resided within defined CpG island sequences. These genes also had not previously studied in HNSCC tumors, and several, including claudin-18 (CLDN18) and the orphan G protein–coupled receptor GPR55, have been seen as contributing to, or even promoting, tumor progression in other cancers (19, 20).

The significance of these DNA methylation events was further explored by examining their impact on the expression of proximal genes. Using microarray gene expression data from these same patients, we cross-referenced these CpG loci with global gene expression profiles, and extracted gene expression data for all of the genes associated with the differentially methylated CpG loci. We found that of the 28 hypermethylated CpG loci identified in our analysis, 14 of the associated genes also showed a significant reduction in gene expression in the primary tumor compared to adjacent mucosa when combining all HNSCC cases (Table 3). These included all of the Kruppel family ZNF genes on chromosome 19q13. Surprisingly, several genes actually showed an increase in gene expression in the primary tumor, such as ATP10A, HOXD9, HOXD10, and SLC6A2, although the reasons for this remain unclear. We used genome-wide DNA methylation data from the Cancer Genome Atlas (TCGA) to externally validate the hyper- and hypo-methylation properties of these loci (http://cancergenome.nih.gov/cancersselected/headandneck). In independent analysis, all 28 commonly hypermethylated CpG loci also showed a statistically significant increase in DNA methylation in the primary tumors compared to available normal mucosal tissue in these cohort of patients (N = 290 tumors, 45 normal, unpaired t test, P < 0.001; Supplementary Table S4). Of the 16 hypomethylated CpG loci, 11 CpGs had usable data available in the TCGA repository, and all showed a significant decrease in DNA methylation in the primary tumors compared to the available normal mucosa (N = 290 tumors, 45 normal, unpaired t test, P < 0.001; Supplementary Table S5).

Two examples of epigenetic silencing in primary HNSCC tumors were the Kruppel family zinc finger proteins ZNF132 and ZNF154 which showed both elevated DNA hypermethylation as well as reduced gene expression in head and neck primary tumors compared to adjacent mucosa from the patient (paired t test, P < 0.05; Fig. 1). In fact, our results showed that DNA hypermethylation of the Kruppel family ZNF genes on chromosome 19q13, which was showed in our previous study on oropharyngeal SCC, now extends to all anatomic sites of HNSCC, and is accompanied by a corresponding decrease in gene expression in the primary tumor. The role of these genes in head and neck carcinogenesis is still unknown, and represents a novel avenue of exploration as new tumor suppressor candidates. Complete lists of hyper- and hypo-methylated CpG loci from each anatomic site of HNSCC are available at Supplementary Tables S1 to S3.

Although some epigenetic changes were found in all anatomic sites, other CpG loci had significantly altered hyper- (or hypo-)methylation in primary tumors from only one anatomic site. This did not imply a lack of differential methylation in the other sites, only that it was not of a sufficient frequency to attain statistical significance for that specific patient dataset. The most striking example of this phenomenon was ubiquitin carboxyl-terminal esterase L1 (ubiquitin thiolesterase) (UCHL1), also known as PGP9.5, a controversial tumor suppressor gene previously identified as hypermethylated in multiple cancers, including head and neck cancer (21). Although UCHL1 was identified as hypermethylated in over 54% of oral cavity cases, we observed hypermethylation frequencies of only 16% and 4% for laryngeal and oropharyngeal SCC cases, respectively (Fig. 2A). Furthermore, hypomethylation of UCHL1 was also seen at different frequencies in all 3 anatomic sites of HNSCC, with the highest frequencies of hypomethylation seen in laryngeal (27%) and oropharyngeal (13%) SCC compared to oral cavity SCC primary tumors (6%).

Differential methylation associated with the UCHL1 gene also had a corresponding effect on UCHL1 gene expression and tumor behavior. As with the previous genes, we cross-referenced relative methylation of the CpG loci associated with the UCHL1 gene (cg08319991) with corresponding UCHL1 gene expression data extracted from microarray datasets for these same patients. From these data, we observed a significant correlation between DNA methylation status of the UCHL1 CpG loci (cg08319991) and expression of the UCHL1 gene (Fig. 2B). The results suggest that expression of UCHL1 is affected by both hyper- and hypomethylation, and that the mechanism of choice is specific to the anatomic site of HNSCC. Furthermore, these epigenetic changes manifest as a change in expression of the UCHL1 gene in HNSCC tumor cells.

Different anatomic sites of HNSCC also differed in terms of the overall number of differentially methylated CpG loci we identified using our genome-wide approach. Overall, we observed the highest number of differentially methylated CpG loci (hyper- and hypomethylated) in primary tumors originating from the oropharynx (Table 2). Given the high propensity of HPV infection in this patient population, we investigated whether or not the number of differentially methylated CpG loci might be influenced by HPV status of the patient. To eliminate anatomic site of the primary tumor as a potential confounding variable, we focused only on oropharyngeal cases, stratified these patients into 2 separate populations according to HPV status, and then repeated the earlier differential methylation analysis. From this new analysis, we identified a total of 204 differentially methylated CpG loci in the HPV− oropharyngeal cases (Table 4). In general, these differentially methylated loci shared the most similarities with tumors from the oral cavity and larynx, both in terms of the number of differentially methylated loci we identified (204 total CpG loci) and the degree of overlap with other sites (32% of loci were also seen in oral cavity cases; 25% of loci were also seen in larynx cases). However, this same analysis in HPV+ oropharyngeal cases identified more than twice as many differentially methylated loci (564 total CpG loci). These included the majority of the differentially methylated loci seen in HPV− oropharyngeal cancer cases (e.g., 74% of the hypermethylated loci); however, we also observed an additional 360 CpG loci in HPV+ oropharyngeal cancer cases not seen previously in HPV− cases. Among the 28 CpG loci whose methylation status best differentiated HPV− from HPV+ oropharyngeal cases were 3 loci within the CpG island of CDKN2A, and 2 loci located within the CpG island of the GALR1 gene (refs. 22, 23; Fig. 3). Also at the top of our list were 2 loci associated with the homeodomain transcription factor known as pancreatic and duodenal homeobox 1 (Pdx1/Ipf1), which had previously been observed to be methylated in breast and gastric cancer (24, 25). Overall, these results suggest that a higher level of aberrant DNA hypermethylation in HPV+ oropharyngeal cancer cases, and that this increased hypermethylation resulted in the targeting of genes not affected in tumors of HPV− cases.

Identification of epigenetically affected genes has become an important tool for understanding aberrant gene expression and mechanisms of tumorigenesis. Here we report a whole-genome analysis of DNA methylation profiles in fresh, frozen HNSCC tissues, and normal adjacent mucosa samples using the Illumina HumanMethylation27 Beadchip with patient genomic DNA. Overall, our findings highlighted many new targets identified across multiple anatomic sites of HNSCC, as well as loci related more specifically to the anatomic site of the tumor, and HPV status of the patient.

With a highly stringent criterion for assessing differential DNA methylation between primary HNSCC tumors and adjacent mucosal tissue, we identified hundreds of significantly altered CpG loci in 3 anatomic sites of this disease. Of the 28 genes we identified as hypermethylated across all 3 anatomic sites, many have not been extensively studied in HNSCC and thus represent possible new avenues of research. Among these examples, it has been previously reported that methylation of RLN3R1 has been associated with increased microsatellite instability in endometrial cancer (26). SPARC-like protein 1 (SPARCL1) is an extracellular matrix glycoprotein that has been recently been shown to be an independent prognostic factor in gastric adenocarcinoma patients, with loss of SPARCL1 expression being a negative event for gastric cancer progression and prognosis (27). The novel tumor suppressor gene T-box transcription factor 5 (TBX5) has been shown to be methylated and epigenetically silenced in colon cancer, and patients with TBX5 methylation have a significantly poorer overall survival than other colon cancer patients (28). All of these genes may play equally significant roles in the mechanism of head and neck carcinogenesis.

Our results also showed that epigenetic silencing of the Kruppel family ZNF genes on chromosome 19q13 which was previously identified in oropharyngeal tumors now appears to extend to all anatomic sites of HNSCC, and are among the most frequently hypermethylated genes observed in our study. The role of these proteins in HNSCC carcinogenesis remains unclear. However, we know that ZNF132 has been identified as epigenetically silenced by DNA hypermethylation in prostate cancer, and that reduced ZNF132 immunoreactivity was significantly associated with high Gleason score and advanced T stage in prostate cancer patient cohorts (29). Furthermore, hypermethylation of ZNF154 has recently been identified as part of a panel of early detection biomarkers in DNA from voided urine of bladder cancer patients, along with 2 other hypermethylated genes (HOXA9 and EOMES) identified in our HNSCC patients (30). Recently, ZNF545/ZFP82 has been identified as a tumor suppressor gene frequently methylated in multiple primary tumors of nasopharyngeal, esophageal, lung, gastric, colon, and breast, inducing tumor cell apoptosis, repressing ribosome biogenesis and inhibiting both AP-1 and NF-κB signaling (31). The exact role of these transcriptional repressors in HNSCC carcinogenesis will be an area of future study by our group.

Of the hypomethylated genes we observed across anatomic sites, claudin 18 and the G protein–coupled receptor GPR55 were of most interest. Claudin 18 expression had previously been shown to be affected by DNA methylation in pancreatic cancer cells (20). GPR55 had previously been shown to promote cancer cell proliferation and be expressed in an aggressiveness-related manner in several cancers, suggesting it also might be a target for therapeutic intervention (32, 33).

Overall, the differential methylation events observed in our study also contained significant overlap with previous studies using the Illumina HumanMethylation27 beadchip. For example, of the 28 hypermethylated CpG loci observed in our study, 27 of these were also observed in at least 1 of 3 DNA methylation “clusters” reported previously by Poage and colleagues (34). Only loci cg122238343 (1 of 2 CpG loci corresponding to the RLN3R1gene) was not previously reported. All 6 ZNF genes identified by our study were also observed in the study by Poage and colleagues, with 5 of the 6 genes (ZNF132, ZNF154, ZNF542, ZNF545/ZFP82, and ZNF781) appearing in all 3 methylation clusters. Of the 16 hypomethylated CpG loci we identified across all 3 anatomic sites, 13 of these were observed in at least 1 of 3 DNA methylation “clusters” reported previously (34). CLDN18 was included in the low methylation cluster, whereas GPR55 hypomethylation was observed across all 3 methylation clusters. When looking specifically at a previously published oral cavity SCC dataset, we observed less overlap between our identified gene lists and those described by Guerrero-Preston and colleagues (35). In this case, 48 of the 119 oral cavity hypermethylated genes (40%) and only 11 of the 174 hypomethylated genes (6%) were common to both datasets. However, this was likely due to the fact that this study incorporated gene expression changes and cross-referenced gene lists with known methylation events as part of the selection process (35). In spite of these differences in analytic methods, both HOXA9 and NID2 were also observed in our oral cavity datasets, and both ZNF132 and ZNF154 were observed to be hypermethylated in the Guerrero-Preston oral cavity SCC dataset.

The methylation of the CpG locus associated with the UCHL1/PGP9.5 gene seemed to be dependent on the anatomic site of the primary tumor. We observed both DNA hyper- and hypo-methylation of UCHL1, with most hypermethylation identified specifically in oral cavity SCC cases. Furthermore, we observed that this differential DNA methylation was directly correlated with expression of the gene. UCHL1 has been observed to be epigenetically regulated in other cancers, and is believed to induce G₀–G₁ cell-cycle arrest and apoptosis by stabilization of p53 (21, 36). Furthermore, the methylation status of UCHL1 has been shown to have prognostic significance in several cancers (37–39). The role of UCHL1 in HNSCC and the mechanisms responsible for its oral cavity SCC-specific hypermethylation pattern will be explored in future studies.

Profiles of DNA hypermethylation were greatly affected by HPV status in oropharyngeal cancer patients. We observed a significant increase in the level of differential DNA methylation (tumor vs. normal) in HPV+ oropharyngeal cases compared to HPV− cases. Much of this was due to the increased DNA hypermethylation of CpG loci in HPV+ cases, corresponding to genes such as CDKN2A and GALR1 (22, 23). Among the 28 most differentially methylated CpGs corresponding to HPV status were 2 CpG loci corresponding to the insulin promoter factor 1/homeodomain transcription factor (IPF1), a transcriptional activator that is involved in the early development of the pancreas and plays a major role in glucose-dependent regulation of insulin gene expression (40). Overexpression of PDX1 has been previously observed in gastric and esophageal carcinoma and shows a significant correlation with CDX2 expression among phenotypic classifications of gastric carcinomas, thereby suggesting a similar function to that gene (41, 42). Also included in the group of differentially methylated genes correlated to HPV status were 3 additional homeodomain-containing genes, including ALX4, CUTL2, and HOXA7, as well as previously identified cancer-related genes such as FBX039, IGSF4, PLOD2, and SLITRK3 (43–49). However, the differential methylation of PDX1 and these other genes, and their relationship to the mechanism of carcinogenesis in HPV+ cancers is not yet understood. It is somewhat surprising that of these 28 genes identified in our study, only 2 (CDKN2A and HOXA7) were also identified in a similar study that examined DNA methylation differences between HPV+ and HPV− HNSCC cell lines (50). This might be attributable to differences when comparing cell lines to whole tissues, or the relatively high levels of DNA methylation observed in many HNSCC cell lines when compared to corresponding patient samples (51). Our analysis also filtered CpG loci based on the magnitude of the observed methylation changes; removal of that magnitude filter would have expanded the overlap to include loci corresponding to DIO2, CCNA1, EREG, IL6ST, COL5A2, and C3orf14.

A study of this type involving human specimens with high-throughput profiling platforms can be susceptible to measurement bias from a variety of sources. First, we are aware of the fact that a portion of the DNA methylation and gene expression signals we collected in this study contained measurement error due to nontumor content in the processed tissue sample. Although samples containing less than 40% tumor were included in the initial identification of hyper- and hypomethylated CpG loci, these samples were excluded from subsequent analysis for the nodal metastasis signature and HPV status. Furthermore, we confirmed that: (i) there were no statistically significant associations between any of the CpG loci identified in this manuscript and the percentage tumor content, and (ii) there were no statistically significant associations between any clinical parameters (tumor stage, HPV status, patient outcome) and percentage of tumor content for these primary HNSCC samples. Second, we are aware of the potential for measurement bias due to the presence of methylation and genetic alterations in histologically normal mucosa adjacent to primary HNSCC tumors. The normal tissue samples in this study included both samples taken from sites adjacent to the primary tumor, as well as those taken from the contralateral side. We therefore confirmed that for all of the CpG loci included in this article, there were no significant associations between methylation (M value) measurements and the source of normal tissue (adjacent normal tissue vs. contralateral side).

In summary, our genome-wide survey of differential DNA methylation has revealed many new avenues of exploration as potential mechanisms promoting tumorigenesis and influencing tumor behavior in this disease, including many novel genes not previously studied in this disease. Furthermore, we have solidified the significance of HPV as a factor correlating with increased changes in DNA methylation in primary HNSCC tumors. It is hoped that further elucidation of these pathways can be useful both as potential diagnostic and prognostic biomarkers and as potential therapeutic targets for this disease in the future.

M.B. Prystowsky has ownership interest (including patents) in PalmStat Diagnostics, LLC. No potential conflicts of interest were disclosed by the other authors.

Conception and design: R.A. Lleras, N.F. Schlecht, G. Childs, M.B. Prystowsky, T.J. Belbin

Development of methodology: R.A. Lleras, N.F. Schlecht, R.D. Burk

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): R.A. Lleras, R.V. Smith, L.R. Adrien, R.D. Burk, M.B. Prystowsky, T.J. Belbin

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): R.A. Lleras, R.V. Smith, N.F. Schlecht, R.D. Burk, G. Childs, M.B. Prystowsky, T.J. Belbin

Writing, review, and/or revision of the manuscript: R.A. Lleras, R.V. Smith, N.F. Schlecht, R.D. Burk, T.M. Harris, G. Childs, M.B. Prystowsky, T.J. Belbin

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): R.A. Lleras, R.V. Smith, N.F. Schlecht, M.B. Prystowsky, T.J. Belbin

Study supervision: R.A. Lleras, T.J. Belbin

The authors thank the Albert Einstein College of Medicine Genomics Core for help with the Sequenom EpiTyper validation assays for cytosine methylation.

This work was supported by grant R21CA131648 (T.J. Belbin) from the National Cancer Institute. DNA methylation and gene expression data for all of the tissue samples used in this study have been deposited into the Gene Expression Omnibus database (http://www.ncbi.nlm.nih.gov/geo/).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.