Epigenetic alterations, reflected in changes to the 5-methylcytosine content at specific genomic loci, are required for normal cellular function and development. Candidate gene approaches for identification of prognostic biomarkers are widely used by researchers and studies have found that DNA methylation alterations are associated with patient survival in a number of cancers (11
), including HNSCC. Head and neck malignancies are often aggressive tumors with a poor probability of survival, and there are few tools currently available to assist the oncologist in determining long-term outcome in these patients. Here we used a genome-wide array to measure DNA methylation in HNSCC, identified markers of patient survival time with a novel approach by clustering CpGs based on their association with local sequence features, and validated the identified markers in an independent set of tumors.
Recently, investigations have revealed that, in addition to aging and environment, the architecture of the genome itself may predispose certain CpG sites to DNA methylation, both in normal cells and in cancer (16
). Therefore, methylation profiling using genomically-informed methodologies is an attractive approach for identifying novel biomarkers. Here, we have developed a unique method for classifying CpG loci that allows functional sequence elements to dictate the clustering. This technique has proven useful in defining a novel association between methylation of specific locus groups and patient prognosis.
Importantly, our analysis demonstrates that HNSCC patients with tumors hypermethylated at a group of 13 CpG loci (defined by proximity to polycomb gene targets, mammalian interspersed repetitive elements, and transcription factor binding sites) have a significantly reduced survival time independent of HPV16 infection status. This suggests that methylation alterations at these sites may determine the phenotype or therapeutic response of this disease. While more work is necessary to precisely determine how the nexus of these three functional sequence element types defines an observable phenotype, it is possible that methylation of these sites, particularly PcG target gene promoters and TFBSs, may potentiate the transformation into (or represent a product of) a stem cell-like tumor. Indeed, there is accumulating evidence that DNA hypermethylation is observed in many cancers at the sites of polycomb-mediated gene repression in embryonic cells, which become relaxed during differentiation, and this methylation correlates with stem cell characteristics (39
). In addition, a number of transcription factors have been shown to be involved in the recruitment of DNA methyltransferases to PcG target sites (39
). Further, studies of aging-dependent methylation have demonstrated that PcG marking and frequency of coincident retrotransposable elements are both correlated and complementary (14
). All of these observations further support our result that the combination of these sequence elements is critical in determining tumor behavior.
Focusing on the individual gene members of the PcG
TFBS cluster, we see that many of these are highly relevant to head and neck disease. Chief among them is ALDH3A1
, which is expressed in the oral mucosa (40
) and is a member of the aldehyde dehydrogenase family of enzymes that convert the carcinogenic intermediate of ethanol metabolism, acetaldehyde, into non-toxic acetic acid. As alcohol is a major risk factor for the development of HNSCC, one might expect alterations of ALDH3A1
to play a role in the initiation and promotion of malignancy. At the same time, this gene has been shown to inhibit epithelial cell proliferation (41
) and is frequently mutated in breast cancers (42
). In addition, it is well-known that aldehyde dehydrogenase is involved in normal stem cell biology and is a functional marker for epithelial cells with enhanced tumorigenic potential. A recent publication showed enrichment for ALDH3
in the stem cell populations of mammalian oral tissues (43
). This supports our hypothesis that alterations at PCG
TFBS-associated genes define a more stem-like constitution of tumor cells that engender more aggressive HNSCCs.
Downregulation of another gene represented in our PcG
TFBS-prognostic cluster, TAP1
, allows HNSCC cells to avoid immune surveillance by cytotoxic T-lymphoctyes (44
) and a lack of protein expression has recently been shown to confer a negative prognostic risk in HNSCC (45
) as well as in many other cancers. This study, however, did not include HPV status in the survival analysis, so the question remained whether downregulation was truly independently prognostic. Another group reported that ectopic expression of TAP1
in xenograft assays significantly prolonged mouse survival time and increased immune infiltrate (46
). In addition, functional investigations have revealed that this gene is downregulated in primary HNSCCs (47
), metastasis (45
), and HNSCC cell lines (49
), although promoter methylation was not previously described in this setting. Together, these data suggest that TAP1
may be a candidate for therapy in human HNSCCs.
Another gene marked by a CpG in the PcG
TFBS cluster that was associated with HNSCC survival is the ankyrin-repeat SOCS box-containing protein 2 (ASB2
) which appears to be a modulator of notch signaling (50
) and inhibits growth of leukemic cells (51
). However, its potential status as a tumor suppressor in the head and neck had yet to be described. A prominent tumor suppressor in the PcG
TFBS cluster is CDKN1A
, encoding the p21 (WAF) protein that signals G1 cell cycle arrest or senescence. Two additional genes identified in our analysis (SPOCK2
) are known to be methylated as potential biomarkers in cancer (52
). Consistent with our observations, the genes represented by CpGs in the PcG
TFBS group are primarily tumor suppressors, and one would therefore expect that their inactivation through a combination of hypermethylation and additional somatic alterations would result in a poorer prognosis.
In summary, we have developed a novel technique to identify clinical characteristics of HNSCCs using the genomic information of CpG loci coupled with epigenetic content at those sites. In two independent populations, we show that DNA methylation profile markers may be used to identify those at greatest risk of death, irrespective of HPV status. The identification of specific DNA methylation biomarkers, such as those presented here, may assist in selecting patients who are most likely to benefit from tissue-sparing procedures and targeted therapies. It will be important to validate this in other populations in an effort to move these biomarkers into clinical practice.