Papillary thyroid cancer is the most common cancer derived from follicular cells. It is estimated that approximately 23,500 cases of differentiated thyroid cancer occur per year in the United States (Jemal et al., 2005), and 19,000 papillary thyroid cancer cases per year in the European Union (http://globocan.iarc.fr/). In Japan, about 8000 patients suffer from thyroid cancer every year, 80% of which are papillary cancer. While prognosis for papillary thyroid cancer is generally good, with a 10-year survival rate above 90%, some patients die of distant metastases and/or repeated recurrence (Ezaki et al., 1992; Yamashita et al., 1998).

RET/PTC (Rearranged in Transformation/Papillary Thyroid Carcinoma) re-arrangement, BRAF (V-Raf murine sarcoma viral oncogene homolog B) and RAS (Rat Sarcoma viral oncogene homolog) point mutations are frequently observed in papillary thyroid cancer (Mitsutake et al., 2006; Knauf and Fagin, 2009). Mutation of T to A at 1799 in the exon 15 of the BRAF gene has been reported in 28–69% of papillary thyroid cancer cases, while point mutations of RAS genes are detected in 5–20% (Cohen et al., 2003; Kimura et al., 2003; Namba et al., 2003; Kondo et al., 2007). Papillary thyroid cancer with poor prognosis is associated with BRAF mutation (Xing et al., 2005; Lee et al., 2012), whereas concomitantly, lengthy disease-free interval is not (Ulisse et al., 2012).

Patients with papillary thyroid cancer are generally treated by surgery. But it is difficult to decide whether total thyroidectomy, hemithyroidectomy, prophylactic central neck dissection or no dissection, should be performed in patients without preoperative or intraoperative evidence of metastatic lymph nodes (Xing et al., 2013). Association of BRAF mutation with occult central neck lymph node metastases (Joo et al., 2012) might support use of BRAF mutation as an indication for prophylactic central neck dissection for patients with conventionally low- to intermediate-risk papillary thyroid cancer. But precise diagnosis to define prognosis and suitable therapy is currently impossible. Molecular biomarkers would therefore simplify disease management (McLeod et al., 2013).

Along with genetic alterations, accumulation of epigenetic alterations is known to affect cancer development (Baylin and Ohm, 2006; Feinberg et al., 2006; Esteller, 2007). Aberrant DNA methylation at promoter regions is a major epigenetic alteration to silence tumor suppressor genes in many cancer types. RASSF1A (Ras association (RalGDS/AF-6) domain family member 1) is methylated in 20% of papillary thyroid cancer, leading to activation of the RAS-MAPK (Mitogen-Activated Protein Kinase) signal (Xing et al., 2004). Papillary thyroid cancer is also reported to involve methylation of other genes, including RARB (Retinoic Acid Receptor, Beta), p16^INK4A (CDKN2A, Cyclin-Dependent Kinase Inhibitor 2A), TSHR (Thyroid Stimulating Hormone Receptor), CDH1 (Cadherin 1, type 1, E-cadherin), DAPK (Death-Associated Protein Kinase 1), and MLH1 (mutL Homolog 1) (Hoque et al., 2005; Guan et al., 2008; Mohammadi-asl et al., 2011). While a few genes known to be aberrantly methylated in other cancers were analyzed in these studies, methylation frequencies ranged from 15 to 33%. Involvement of genes in aberrant DNA methylation, however, has not been well-clarified in papillary thyroid cancer. Whether any subset of papillary thyroid cancer shows preferential aberrant methylation, and whether such methylation and other clinicopathological factors are associated are also unclear.

We here analyzed DNA methylation status of promoter regions on a genome-wide scale, using the Illumina Infinium HumanMethylation27 BeadChip technique on 14 clinical papillary thyroid cancer samples and 10 normal thyroid samples. For genes frequently hypermethylated in cancer, methylation status was validated quantitatively by pyrosequencing, using 20 additional clinical cancer samples and 10 additional normal samples. Methylation-associated gene silencing was confirmed by gene re-expression following 5-aza-2′-deoxycytidine and trichostatin A treatment, and by quantitative reverse transcription-polymerase chain reaction (RT-PCR) on thyroid cancer cell lines. We found a number of genes with frequent aberrant methylation and silencing in papillary thyroid cancer, and a subset of cancer with preferential aberrant methylation.

In this study, we performed genome-wide DNA methylation analysis in 14 human papillary thyroid cancer samples and 10 normal samples. Although papillary thyroid cancer apparently involves fewer aberrantly methylated genes than other types of cancers, we detected 25 genes frequently hypermethylated in papillary thyroid cancer. Methylation status was quantitatively validated in six out of the 25 genes by pyrosequencing, using the genome-widely analyzed samples and additional samples. Gene silencing in papillary thyroid cancer cell lines was confirmed by real-time RT-PCR. While a subset of cancer cases had no aberrant methylation at all, cancer with preferential methylation tended to have oncogene mutation and to be larger tumor.

Papillary thyroid cancer displayed fewer aberrantly methylated genes, compared with other cancer types (Figure 2). For genes previously reported to be methylated in thyroid cancer, such as TSHR (Xing et al., 2003), or in other cancer types, such as RASSF1A, RAR-β2, p16, CDH1, DAPK, and MLH1, the methylation frequency in papillary thyroid cancer ranges from 15 to 33% (Hoque et al., 2005; Guan et al., 2008; Mohammadi-asl et al., 2011). In these reports, no or few normal samples were analyzed (Guan et al., 2008; Mohammadi-asl et al., 2011), methylation was also detected in normal samples (Hoque et al., 2005), or a non-quantitative method, i.e., standard methylation-specific PCR, was used (Guan et al., 2008). Standard methylation-specific PCR (Herman et al., 1996) can amplify and detect minor fraction of methylated alleles, but its high sensitivity can lead to overestimation of methylation frequency. Our analysis did not select these genes as frequently methylated ones in the first 14 cancer samples, because normal thyroid tissues also showed high methylation levels or because methylation frequencies in papillary thyroid cancer samples were low (≤2 of 14 cancer samples). Instead, we detected 25 novel genes that were frequently aberrantly methylated (β > 0.25) in at least three of the 14 cancer samples, and not methylated in any of the 10 normal thyroid samples (β < 0.2).

Interestingly, three of the 14 papillary thyroid cancer samples showed no aberrant methylation in the 25 genes, but the other 11 cancer samples showed hypermethylation in at least three of the 25 genes. No cancer sample showed aberrant methylation in just one or two genes. This unusual distribution of aberrant methylation is similar to the CpG island methylator phenotype (CIMP), which was first proposed in colorectal cancer (Toyota et al., 1999). As Toyota et al. demonstrated in colorectal cancer (Toyota et al., 1999), we calculated probability of methylation distribution in papillary thyroid cancer using these 25 genes. The fraction of methylated tumors in each gene was 3/14 for ATP8A2, 3/14 for C19orf4, …, 5/14 for ANKRD45, …, 6/14 for MYLK, …, 4/14 for NRN1, …, 5/14 for SHOX2, and 3/14 for SST (Figure 3). Assuming that methylation of these genes is random, the probability that none of the 25 genes would be methylated in three cancer samples is P = 1.2 × 10⁻⁸. This was calculated using the following formula: p(x)=(Nx)∏g(N−xf(g))∏g(Nf(g)) where x indicates number of samples which do not have methylated genes (x = 3 in the present case), N indicates number of cancer samples (N = 14 here), g indicates one of the 25 genes, and f(g) indicates the number of samples in which g is methylated. Similarly, the probability that at least three of the 25 genes are methylated in 11 cancer samples would be P = 0.0028. In each case, a random event is highly unlikely. We rather observe that the associated methylation or un-methylation events both occur at relatively high frequencies. Our data thus suggest that there are two distinct classes of papillary thyroid cancer. One is a subset of hardly methylated cancer. The second one is a subset of preferentially methylated cancer, prone to transcriptional silencing and with the potential to inactivate several genes simultaneously, as CIMP has been proposed in colorectal cancer and other cancers (Toyota et al., 1999; Kaneda et al., 2002; Noushmehr et al., 2010).

Although the number of analyzed samples was not large, preferentially methylated papillary thyroid cancer showed mutation of BRAF/RAS oncogenes more frequently than methylation(−) cancer (P = 0.04, Fisher's exact test). In previous studies of papillary thyroid cancer, although methylation of RASSF1A and BRAF mutation were detected in a mutually exclusive manner (Xing et al., 2004; Hoque et al., 2005), methylation of RAR-β2 or MLH1 significantly correlated to BRAF mutation (Hoque et al., 2005; Guan et al., 2008). Correlation of aberrant methylation and oncogene mutation are also reported in colorectal cancer; high-methylation and intermediate-methylation epigenotypes strongly correlated to BRAF mutation and KRAS mutation, respectively, and low-methylation epigenotype strongly correlated to lack of oncogene mutation (Shen et al., 2007; Yagi et al., 2010; Hinoue et al., 2012). The mechanism of these correlations is still unknown, but oncogene mutation may somehow induce aberrant methylation, or aberrant methylation may help escape from senescence by disrupting factors critical in RAF/RAS-induced senescence (Kaneda and Yagi, 2011).

Preferentially methylated cancer also tended to have larger tumors and higher thyroglobulin levels, which might relate to cancer progression (Piccardo et al., 2013). Although 90% of papillary thyroid cancers are considered to be at low risk with a mortality rate of 1–2%, the mortality rate of the high risk group is 50–75% (Hay et al., 1993; Noguchi et al., 1994; Shaha et al., 1996; Dean and Hay, 2000). The tumor-node-metastasis (TNM) classification is a tool for cancer prognosis; each variable used in TNM staging (age, tumor size, extent of primary tumor, and presence of nodal or distant metastases) shows significant association with observed end points of cancer recurrence or death. Cancer recurrence and mortality ratios are significantly lower in stage I (15.4% and 1.7%, respectively) compared with more advanced tumors (22% and 15.8% in stage II, 46.4% and 30% in stage III, and 66.7% and 60.9% in stage IV, respectively) (Loh et al., 1997). Molecular diagnostic markers are still not used, although their development is anticipated (McLeod et al., 2013). Although aberrant methylation was not significantly associated with lymph node metastasis, distant metastasis, or recurrence in analysis of the 34 cancer samples in this study, further study should be performed using larger set of clinical samples for comparison of aberrant methylation, gene mutation status, and prognosis.

As for detected genes, TLX3 (HOX11L2) is a transcription factor highly expressed in T-cell leukemia (Baak et al., 2008), and its aberrant methylation was observed in cisplatin-resistant bladder cancer (Tada et al., 2011). SHOX2 is a member of the homeobox gene family, and is reported to relate to a short-stature phenotype of Turner syndrome (Clement-Jones et al., 2000). DNA methylation of SHOX2 was suggested to be a biomarker for diagnosis of lung cancer based on bronchial aspirates (Schmidt et al., 2010). HOXA7 is also a transcription factor belonging to the homeobox gene family that regulates gene expression, morphogenesis, and differentiation (La Celle and Polakowska, 2001). POU4F2 is one of POU family genes with Pit-Oct-Unc domain, and is a transcription factor with a role in cell identity and regulation of nerve cell or retinal development (Weishaupt et al., 2005). PHKG2 is the gamma subunit of phosphorylase kinase, containing the active site of the enzyme. Phosphorylase kinase-deficient liver glycogenosis can be caused by mutations of phosphorylase kinase subunits, PHKA2, PHKB, or PHKG2, but PHKG2 mutation was reported to cause a severe phenotype of this disease (Burwinkel et al., 2003). HIST1H3J encodes a member of histone H3 family, and is found in the small histone gene cluster on chromosome 6p22-p21.3 (NCBI gene data bank). If the role of histone modifications is known to affect the regulation of gene expression, less is known about the possible direct involvement of histones, an H3 variant in the present case, in thyroid tumorigenesis. Further investigation is necessary to clarify tumorigenic roles of these genes and their methylation, in papillary thyroid cancer and other types of cancer (Schmidt et al., 2010; Tada et al., 2011).

In summary, 25 new genes were found to be frequently methylated in papillary thyroid cancer. There might be subsets of papillary thyroid cancer hardly methylated and preferentially methylated, and aberrant methylation of these genes correlates a priori to BRAF/RAS oncogene mutation in papillary thyroid cancer.

Yasuko Kikuchi, Koichi Yagi, and Keisuke Matsusaka performed the experiments. Eiichi Tsuji and Toshihisa Ogawa prepared clinical samples and information. Junichi Kurebayashi established and supplied cell lines. Yasuko Kikuchi, Shingo Tsuji and Atsushi Kaneda analyzed and interpreted the data. Yasuko Kikuchi and Atsushi Kaneda wrote the manuscript. Toshihisa Ogawa, Hiroyuki Aburatani, and Atsushi Kaneda supervised the study.