Head and Neck Cancers Are Not Alike When Tarred with the Same Brush: An Epigenetic Perspective from the Cancerization Field to Prognosis

Simple Summary Squamous cell carcinomas affect different head and neck subsites and, although these tumors arise from the same epithelial lining and share risk factors, they differ in terms of clinical behavior and molecular carcinogenesis mechanisms. Differences between HPV-negative and HPV-positive tumors are those most frequently explored, but further data suggest that the molecular heterogeneity observed among head and neck subsites may go beyond HPV infection. In this review, we explore how alterations of DNA methylation and microRNA expression contribute to head and neck squamous cell carcinoma (HNSCC) development and progression. The association of these epigenetic alterations with risk factor exposure, early carcinogenesis steps, transformation risk, and prognosis are described. Finally, we discuss the potential application of the use of epigenetic biomarkers in HNSCC. Abstract Head and neck squamous cell carcinomas (HNSCC) are among the ten most frequent types of cancer worldwide and, despite all efforts, are still diagnosed at late stages and show poor overall survival. Furthermore, HNSCC patients often experience relapses and the development of second primary tumors, as a consequence of the field cancerization process. Therefore, a better comprehension of the molecular mechanisms involved in HNSCC development and progression may enable diagnosis anticipation and provide valuable tools for prediction of prognosis and response to therapy. However, the different biological behavior of these tumors depending on the affected anatomical site and risk factor exposure, as well as the high genetic heterogeneity observed in HNSCC are major obstacles in this pursue. In this context, epigenetic alterations have been shown to be common in HNSCC, to discriminate the tumor anatomical subsites, to be responsive to risk factor exposure, and show promising results in biomarker development. Based on this, this review brings together the current knowledge on alterations of DNA methylation and microRNA expression in HNSCC natural history, focusing on how they contribute to each step of the process and on their applicability as biomarkers of exposure, HNSCC development, progression, and response to therapy.


Introduction
Head and neck squamous cell carcinomas (HNSCC) are among the ten most frequent and lethal cancers among men worldwide, affecting both developed and developing hol consumption were associated with global hypomethylation in HNSCC [39], while HPV-positive tumors show higher global methylation levels relative to HPV-negative tumors [40]. It is intriguing that while the expression of microRNAs can be regulated by DNA methylation [41], DNA methylation machinery can also be the target of microRNAs [42,43], creating a negative feedback. Additionally, microRNA expression can be influenced by genetic factors such as single nucleotide polymorphisms and mutations [44,45]. However, given the stochasticity of HNSCC carcinogenesis, the order of events is still far from being understood.
Furthermore, specific microRNA expression and DNA methylation signatures for tobacco [46], alcohol [47,48], and HPV infection [49,50] were established and represent not only a source of exposure biomarkers, but may also add information on carcinogenic mechanisms ( Figure 2). in cancer, and high stability in biological fluids, miRNAs are promising biomarkers in oncology.

HNSCC Etiology and Epigenetics
All etiological factors associated with HNSCC development have been shown to affect epigenetic mechanisms in different tissues. For example, both smoking and alcohol consumption were associated with global hypomethylation in HNSCC [39], while HPVpositive tumors show higher global methylation levels relative to HPV-negative tumors [40]. It is intriguing that while the expression of microRNAs can be regulated by DNA methylation [41], DNA methylation machinery can also be the target of microRNAs [42,43], creating a negative feedback. Additionally, microRNA expression can be influenced by genetic factors such as single nucleotide polymorphisms and mutations [44,45]. However, given the stochasticity of HNSCC carcinogenesis, the order of events is still far from being understood.
Furthermore, specific microRNA expression and DNA methylation signatures for tobacco [46], alcohol [47,48], and HPV infection [49,50] were established and represent not only a source of exposure biomarkers, but may also add information on carcinogenic mechanisms ( Figure 2).

Tobacco
Tobacco components have been shown to induce epigenetic alterations ( Figure 2 and Table 1). Although the molecular mechanisms are not completely elucidated, some clues have been gathered. In HeLa cells, benzo(a)pyrene (BaP) was shown to induce the deposition of activating histone marks in the promoter of long interspersed element 1 (LINE1) [51]. This was followed by the inhibition of DNMT1 binding to these genomic regions, and by the proteasome-mediated degradation of this enzyme. As a result, BaP induced LINE1 hypomethylation [51]. By binding to nicotinic cholinergic receptors, nicotine was shown to decrease Dnmt1 mRNA and protein levels and induce demethylation of specific loci [52]. Other tobacco components may induce DNA hypermethylation. The tobacco-specific nitrosamine 4-(methylnitro-samino)-1-(3-pyridyl)-1-butanone (NNK) can induce AKT activation that phosphorylates and inhibits GSK3β [53]. Once inactivated, GSK3β can no longer phosphorylate DNMT1, which is not marked for proteasomal degradation, and it is therefore stabilized. Another mechanism by which tobacco components may induce hypermethylation is via DNA damage; Dnmt1 was shown to be recruited to DNA repair sites to restore epigenetic marks [54]. In vitro experiments in HeLa and mouse embryonic stem cells corroborated these findings, showing that following double-strand breaks, half of the strands repaired by homologous recombination become hypermethylated, which is mediated by DNMT1 [55]. Cigarette smoke condensate can also induce Sp1 expression and DNA binding [56]. Since Sp1 is a transcription factor that recognizes CG-rich genomic regions [57], its binding may prevent DNA methylation in these regions [58,59]. Finally, tissue hypoxia induced by cigarette smoke carbon monoxide can induce global DNA hypomethylation. Hypoxia-inducible factor 1α (HIF-1α) induces the expression of methionine adenosyltransferase 2A (MAT2A, involved in SAM synthesis), and as a consequence, SAM levels are decreased, impairing DNA methylation reactions [59,60].   Epigenome-wide association studies (EWAS) on different tissue types were carried out to identify tobacco epigenetic signatures. Using blood samples and the HumanMethy-lation27 assay, Breitling and colleagues identified one CpG locus annotated to F2RL3 that was hypomethylated in heavy smokers relative to never smokers. Former smokers showed methylation patterns similar to never smokers [140]. A following study applying the Hu-manMethylation450 beadchip, both in lymphoblasts and alveolar macrophages, identified one differentially methylated probe in AHRR gene body when comparing smokers and nonsmokers [141]. Similar findings for AHRR methylation and additional differentially methylated CpGs in CYP1A1 and GFI1 were observed in newborns following maternal smoking [142]. Additionally, in blood-derived DNA analyzed by HumanMethylation 450 assay, a meta-analysis with 15,907 individuals from 16 cohorts identified 18,760 differentially methylated CpGs between current and never smokers, including CpGs mapped to previously identified genes, such as AHRR, RARA, F2RL3, and LRRN3 [143]. A total of 2568 CpGs were also differentially methylated between never and former smokers. The consistent association between the methylation profile of genes involved in xenobiotic metabolism and smoking adds a biological plausibility to the results obtained.
An EWAS carried out in buccal cells from 790 women showed that smoking also leaves a DNA methylation signature in epithelial cells, which can be even more prominent throughout the genome than that determined in blood cells [144]. In this study, the topranked CpG sites showed a hypomethylation profile in smokers and affected genes such as AHHR, F2RL3, and CYP1A1, previously identified in blood-based EWAS/smoking studies. While hypomethylated genes were not enriched for specific cell pathways, hypermethylated CpGs were enriched for genes bivalently marked in human embryonic stem cells, and for binding sites of transcription factors associated with stem cell differentiation. These smoking-responsive CpG sites were also differentially methylated in tobacco-associated cancers, suggesting a contribution of these alterations to tumor development.
In OPSCC patients, the blood of 106 never and 303 ever smokers were compared and revealed the differential methylation of 52 CpG sites [145]. Of these, 49 sites had lower methylation levels in smokers, including a CpG mapped to AHRR. A similar analysis was performed to predict survival, and 18 CpG sites annotated to three genes (GFI1, SPEG, and PPT2) were differentially methylated according to both survival and smoking history. Additionally, in OPSCC patients, a DNA methylation score for smoking based on five EWAS was applied on 364 blood samples and was able to explain 51.1% of phenotypic variance [146]. This score showed an accuracy of 0.70 to predict mortality, slightly improving the accuracy of smoking self-report (0.67).
In gene-specific studies, p16INK4a hypermethylation was already reported in 9.7% of oral mucosa and/or tongue of healthy smokers [147]. Similarly, higher methylation levels of p16INK4a and MYOD1 were observed in HNSCC from smokers relative to nonsmokers [61]. SFRP4 hypermethylation seems to be more common in never and former smokers than in current smokers, and it is also independently associated with HPV16 infection [62]. In a study evaluating the methylation profile of 10 genes (p16INK4a, DAPK, GSTP1, RASSF1, BRCA1, ECAD, MLH1, MINT1, MINT2, and MINT31), both the methylation index (determined by the frequency of hypermethylated promoters) and the frequency of CpG island methylator phenotype-high (CIMP-high, at least five hypermethylated genes) cases were higher among smokers relative to nonsmokers [78]. In opposition to the hypomethylation profile observed in the EWAS, CYP1A1 hypermethylation was more frequently observed in HNSCC from smokers relative to healthy tissues from controls [63]. When promoter methylation and gene deletion of RBSP3, LIMD1, and CDC25A were analyzed together in HNSCC, the frequency of alterations was higher in smokers than in nonsmokers [64].
MicroRNAs are also targets of dysregulation by different tobacco components, varying according to route of exposure and head and neck subsite. Immortalized human oral keratinocytes OKF6/TERT exposed to cigarette smoke condensate or chewing tobacco extract showed six and ten differentially expressed microRNAs relative to nontreated cells, respectively, with none in common between treatments [148]. After treatment with smoke condensate, members of the miR-200 family were downregulated. These data are consistent with their downregulation observed in OSCC patients with smoking/chewing tobacco habits [65]. The exposure of oral keratinocytes OK4 to cigarette extract further showed the upregulation of miR-101, miR-181b, miR-486, and miR-1301 [149].
The exposure of HNSCC cells to NNK induced the expression of miR-21 and miR-155, while decreased miR-422a expression. Overexpression of miR-155 was also associated with the habit of chewing tobacco [66]. MiR-21 expression was further associated with MSH2 silencing [150]. MSH2 is a member of the mismatch repair pathway, and its downregulation was associated with poor survival of HNSCC patients [151]. NNK can also induce the expression of miR-944 that targets CISH, a suppressor of the STAT pathway [152,153]. CISH inhibition by miR-944 induces STAT3 activation, modulating the inflammatory response and increasing tumor malignancy [153]. MiR-30a and miR-379 expression levels were downregulated in OSCC cells exposed to NNK. Both miRNAs are downregulated in OSCC relative to adjacent nontumor tissue and target DNMT3B. This can lead to the hypermethylation of ADHFE1 and ALDH1A2 promoters, and their consequent downregulation [154]. Since both genes are involved in alcohol metabolism, these data suggest an interesting potential mechanism for the synergy between alcohol and tobacco in HNSCC.
MiR-29c downregulation was associated with smoking and worse survival in laryngeal squamous cell carcinoma (LSCC) patients [67]. Additionally, in LSCC, smokers showed miR-202-3p upregulation and downregulation of miR-4768-3p, miR-548aa, and miR-3713 [68]. Among HNSCC patients, miR-34 family was shown to be upregulated in tumors from smokers relative to the adjacent tissues, while no differences were observed for nonsmokers [69]. Smoking or chewing tobacco was associated with miR-429 and miR-141 downregulation in OSCC patients [148]. In OPSCC, a comparison between smokers and nonsmokers HPV-positive patients revealed 38 differentially expressed miRNAs [70]. Mir-133a-3p was among the downregulated miRNAs, and its expression was further shown to be decreased by treatment with cigarette smoke condensate and associated with EGFR and HuR expression in HNSCC cell lines [70].

HNSCC Etiology and Epigenetics: Alcohol
Epigenetics and metabolism are two substations in the same system, intertwined. This is probably the simplest way to understand how alcohol may induce aberrant DNA methylation ( Figure 2, Table 1). Epigenetic writers and erasers are regulated, among other mechanisms, by the availability of their substrates and cofactors. Ten-eleven translocation (TET) enzymes are α-ketoglutarate-dependent oxygenases involved in active DNA demethylation [155]. Due to this cofactor dependency, TET activity is regulated by different factors that affect the tricarboxylic acid cycle (TCA), including hypoxia [156]. For DNMTs activity, the levels of the methyl donor SAM are determinant. SAM is produced by one-carbon metabolism, which relies on the dietary intake of folate and B vitamins. Ethanol is an inhibitor of folate absorption, besides inhibiting different steps of one-carbon metabolism, leading to a reduction of SAM levels. Therefore, alcohol consumption may impair the methylation reactions catalyzed by DNMTs. Indeed, global DNA hypomethylation was observed in the liver of ethanol-fed rats [157]. While the higher intake of folate has already been associated with a decreased risk of HNSCC development [158], the interaction between folate, alcohol, and DNA methylation has been neglected. Nevertheless, several studies have shown an association between DNA methylation patterns and alcohol consumption in different settings, and surprisingly, it is not restricted to DNA hypomethylation.
A recent EWAS, including 13,317 individuals, proposed a 144-CpG signature in whole blood able to discriminate current heavy alcohol drinkers [47]. This model was used to predict all-cause mortality among OPSCC patients, but although it explained 16.5% of phenotypic variance, it showed a similar predictive accuracy to self-reported data [146]. Additionally, in the blood of OPSCC patients, an EWAS identified three new CpG sites and 40 differentially methylated regions associated with alcohol consumption [145]. The gene with the lowest p-value for both analyses was SLC7A11, with the methylation levels of one of its CpG decreasing 0.10% per unit increase in alcohol consumption. This gene plays a role in glutathione synthesis, an antioxidant that can be depleted by excessive ROS production during alcohol metabolism [159].
Other gene-specific studies have suggested an association between alcohol consumption and DNA methylation profiles in HNSCC. P15INK4b methylation was detected in the oral and throat rinses of 48% of HNSCC patients and 68% of healthy individuals who smoked and/or drank alcohol, but in only 8% of healthy nonsmokers and nondrinker individuals [71]. FUSSEL18 and SEPT9 methylation were also associated with alcohol and tobacco consumption history in HNSCC [72]. In a meta-analysis including 18 studies with different sample sources (tumors, liquid biopsies, and buccal scraping), DAPK methylation was significantly associated with alcohol consumption (OR = 1.85, 95% CI = 1.07 ± 3.21) [73]. SFRP1, a member of the secreted frizzled receptor protein family, is hypermethylated in HNSCC and this event is more common in ever drinkers than in never drinkers [62]. Other genes, such as p16INK4a, RASSF1A, and FANCF are also more frequently hypermethylated in HNSCC associated with alcohol consumption [61,74]. On the other hand, MLH1 methylation seems to be more common in individuals who drink less than 100 mL of alcohol/day in comparison with those who drink larger amounts [75].
Recently, alcohol consumption was associated with an aberrant DNA methylation profile in the blood of HNSCC survivors [160]. This dysregulation affected genes encoding histone and mitochondrial ribosomal proteins, as well as genes involved in toll-like receptor signaling, suggesting that alcohol intake can have a systemic effect, even after treatment.
In comparison with data showing an association between alcohol and DNA methylation in HNSCC, data for such an association with microRNA expression are still limited. As observed for DNA methylation, alcohol metabolism is also related to the expression of microRNAs. ADH1B is a key controller of alcohol metabolism, affecting acetaldehyde levels in the body [161], and its expression was shown to be regulated by miR-205-5p in HNSCC [162]. Other genes related to alcohol metabolism, such as ADHFE1 and ALDH1A2, are indirectly regulated by miR-30a and miR-379-mediated DNMT3B downregulation in OSCC [154], as previously mentioned, revealing a crosstalk between the methylation machinery and microRNAs in the xenobiotic response. The dysregulation of genes involved in alcohol metabolism by microRNAs was also associated with aggressive phenotypes. ADHFE1 and ALDH1A2 higher expression, as a consequence of miR-30a and miR-379 induction, were associated with proliferation inhibition of OSCC cell lines [154]. It is important to note that ALDH1A2 does not metabolize acetaldehyde efficiently, but it is important in the metabolism of retinoic acid [163]. In addition, ALDH1A2 is a tumor suppressor candidate in HNSCC [164].

HNSCC Etiology and Epigenetics: HPV
Regarding HPV infection, viral oncoproteins were shown to modulate levels of DN-MTs ( Figure 2) [169]. E6 knockdown in HPV16 infected cervical cancer cell lines led to decreased DNMT1 expression, which was mediated by higher p53 levels [170]. Similarly, HPV16 E7 oncoprotein was also shown to stimulate Dnmt1 activity, both by promoting E2F activity through Rb downregulation [171], and by direct binding [172], with a consequent decreased E-cadherin expression [173]. Therefore, it is expected that HPV infection affects DNA methylation patterns.
Different studies addressed genome-wide DNA methylation differences between HPV-positive and HPV-negative HNSCC (reviewed in [174]). Hypermethylation was more frequent in HPV-positive HNSCC samples and cell lines relative to HPV-negative counterparts [175,176]. Furthermore, a DNA methylation signature for HPV infection composed of five CpG sites was proposed and showed high specificity to discriminate HPV-positive and HPV-negative HNSCC in different datasets, as well as a capacity to predict survival similar to HPV status, determined by viral mRNA levels [175]. In OPSCC, another DNA methylation signature composed of differentially methylated regions in the promoters of five genes (ALDH1A2, OSR2, GATA4, GRIA4, and IRX4) was shown to predict overall and progression-free survival, independently of HPV status [177].
Promoter methylation of genes recurrently reported altered in HNSCC also seem to differ according to HPV infection. For p16INK4a, results are discordant between studies, including hypermethylation [77,78], hypomethylation [79,80], and no differences found in HPV-positive relative to HPV-negative HNSCC cases [178]. This may be a consequence of both the different sample sources (tumors and saliva) and the different tumor subsites used, including oral cavity, oropharynx, both combined, and even HNSCC generally. Therefore, further studies are needed to resolve these discrepancies. RASSF1A, TIMP3, and PCQAP/MED15 were found to be hypomethylated in the saliva of HPV-positive HNSCC [79], while RXRG, CTNNA2, GHSR, and ITGA4 were reported hypermethylated in HPV-positive oropharyngeal tumors [80]. Global DNA methylation also differs according to HPV status, with LINE1 methylation levels being higher in HPV-positive relative to HPV-negative tumors [179][180][181][182][183]. DNA methylation differences in transposable elements according to HPV status were recently shown to affect gene expression, which was further associated with prognosis [183].
Interestingly, DNA methylation patterns also differ when HPV DNA is integrated or episomal, with integration-positive tumors showing a profile more similar to that of HPV-negative HNSCC [20]. BARX2 and IRX4 were hypermethylated and silenced, while SIM2 and CTSE showed hypomethylation and higher expression in episomal HPV-HNSCC. Finally, differentially methylated genes were enriched for the cAMP response elementbinding protein (CREB) pathway, suggesting different carcinogenic biological mechanisms, depending on whether HPV is integrated.
DNA methylation was further associated with miRNA dysregulation in HPV-associated HNSCC [184]. MiR-139-3p was shown to be downregulated in HPV-positive tumors and cell lines relative to their HPV-negative counterparts, and it was shown to target HPV oncoproteins. In vitro experiments with miR-139-3p mimic led to E6 and E7 downregulation, reduced p16, and increased p53 protein levels. Interestingly, miR-139-3p colocalizes with PDE2A, whose promoter is hypermethylated in HPV-positive HNSCC. Corroborating this coregulation, treatment of HPV-positive cell lines with a demethylating agent induced miR-139-3p expression. MiR-375, previously described as a tumor suppressor whose expression is associated with alcohol intake, also seems to be suppressed by DNA methylation in HPV-mediated carcinogenesis [185]. Since HPV E6 and E7 are among miR-375 targets, downregulation of this miRNA contributes to a higher expression of these oncoproteins and a lower expression of their cellular targets, p53 and Rb [186].
Differentiation-associated miRNAs might also take part in HPV-mediated carcinogenesis [187]. As previously mentioned, HPV disturbs the differentiation of infected cells, and this was shown to be mediated, at least in part, by miR-203 downregulation. This miRNA regulates differentiation by targeting p63 and its expression can be downregulated by E7 via the blockage of MAPK pathway [187].
Although the miRNA biogenesis machinery was found to be dysregulated in cervical cancer [188], this is yet to be investigated in HPV-associated HNSCC. Nonetheless, miRNA signatures according to HPV status were reported in HNSCC. By comparing HPV-positive and HPV-negative tumors with adjacent tissue, 282 and 289 differentially expressed miR-NAs were identified, respectively [189]. Although these numbers were similar, when specific survival-associated miRNAs were evaluated, the expression of 39 miRNAs was associated with prognosis among HPV-positive patients, 16 among HPV-negative patients, and no miRNA was shared. The miRNA survival signatures also differed in accuracy (0.92 for HPV-positive and 0.72 for HPV-negative) and associated signaling pathways. The poor prognosis miRNA signature for HPV-positive patients was associated with immunerelated pathways. In contrast, in HPV-negative cases, it was associated with metabolism and oncogenic pathways, such as WNT and NOTCH signaling.
More recently, the direct comparison of HPV-positive and HPV-negative HNSCC revealed the differential expression of 59 miRNAs, three of which (miR-99a-3p, miR-411-5p, miR-4746-5p) were also associated with overall survival in a multivariate Cox regression analysis [81]. The Gene Set enrichment analysis performed with their potential mRNA targets showed that different pathways were likely to be dysregulated, with genes involved in epithelial-mesenchymal transition being downregulated by miR99a-3p-high/miR-411-5p-low/miR-4746-5p-high. This profile may help to explain the favorable prognosis of these patients.
Not only previously annotated miRNAs show differential expression according to HPV status. By applying a customized in silico pipeline, Rock and colleagues identified 146 novel miRNAs in samples from HNSCC patients, with three (HNnov-miR-2, HNnov-miR-30, and HNnov-miR-125) being significantly downregulated in HPV-positive tumors [194].
Although most studies bring consistent associations between risk factor exposure and aberrant epigenetic profiles in HNSCC, causal effects were not determined. In the attempt to overcome this limitation, in vivo studies should be performed with a focus on head and neck tissues. As an example, the treatment of mice with dibenzo[def,p]chrysene (DBP), a tobacco carcinogen, was able to induce DNA methylation alterations of 30 loci in oral tissues [195]. Fgf3, frequently amplified in HNSCC, was one of the affected genes and showed hypomethylation accompanied by induced expression. Considering that tobacco smoke contains more than 60 different carcinogens [196], alcohol may contribute to carcinogenesis by different mechanisms [197], HPV oncoproteins regulate key epigenetic enzymes [169,170,172,173], and that how all these factors interact is largely unknown, much is yet to be learned considering the effects of risk factor exposure on epigenetic mechanisms in head and neck tissues.

Epigenetic Biomarkers in HNSCC
As already mentioned, epigenetic changes coordinate the cellular response to environmental exposures, such as cancer risk factors [198,199]. Thus, even before the histological transformation or acquisition of driver mutations, we can have stochastic or targeted epigenetic changes, the latter sensitizing the cell to transformation, building the pre-cancer epigenetic signatures of the cancerization field ( Figure 3) [200,201]. Several of these changes would not only favor the initial gain of a pre-cancer cell phenotype, but would also be maintained or altered after malignant transformation, tumor progression, and metastasis. Thus, epigenetic signatures are not just a consequence of risk factor exposure, or linked to the tissue of origin, but a unique product of these two factors combined [202]. In addition, unlike genetic changes, these signatures are plastic, and can vary from healthy tissue through the cancerization field and tumor progression in a qualitative or quantitative manner depending on the mechanism.  Given this scenario, it is impossible not to notice the relevance and potential of epigenetic changes as biomarkers in oncology [203,204]. In the next section, we focus on epigenetic changes as potential biomarkers in clinical use without emphasizing their contribution to gene and biological regulation. Interestingly, due to the aforementioned plastic . The natural history of HNSCC development through an epigenetic perspective. The exposure of the head and neck healthy epithelium to risk factors, such as tobacco and alcohol, may lead to the accumulation of DNA methylation alterations and miRNA aberrant expression even before the first morphological changes are observed. This characterizes a molecular cancerization field that may favor the development of pre-malignant lesions. In head and neck, the presence of leukoplakias and erythroplakias, frequently associated with dysplasia, increases the risk of tumor development, and these lesions are characterized by specific epigenetic alterations. The epigenetic aberrations observed in both the cancerization field and in the pre-malignant lesions, have the potential to be used as early markers of HNSCC carcinogenesis in screening programs. Finally, tumor onset and progression may be followed by further aberrant DNA methylation and miRNA expression. HPV-positive HNSCC epigenetic natural history is largely unknown. Given this scenario, it is impossible not to notice the relevance and potential of epigenetic changes as biomarkers in oncology [203,204]. In the next section, we focus on epigenetic changes as potential biomarkers in clinical use without emphasizing their contribution to gene and biological regulation. Interestingly, due to the aforementioned plastic characteristic of epigenetic marks, it is possible to identify common epigenetic changes, perhaps with different levels, aimed at more than one biomarker objective (e.g., diagnosis and prognosis). Here, we seek to bring a snapshot of epigenetic changes as biomarkers, technical limitations, and advantages, together with future application prospects.

Biomarkers of the Cancerization Field: From Identification to Surgical Margin and Recurrence
The molecular landscape of HNSCC carcinogenesis has been unraveled, and several candidates for biomarkers to identify pre-cancer lesions in the cancerization field are emerging (Table 1). Except for HPV-positive oropharyngeal tumors, the main genetic alterations in HNSCC are TP53 mutations, present in up to 84% of cases [23]. TP53 mutations have become the focus of many studies (perhaps the majority) aimed at genetic characterization of the cancerization field [205][206][207]. Although less frequently, these changes can already be found in the adjacent nontumoral dysplastic [208,209] and nondysplastic tissues of HNSCC patients [209], becoming a frequent event in key moments of the dysplasia-carcinoma transition [210].
However, TP53 mutations can be detected throughout the entire gene without clear hotspots [211], requiring techniques that are resourceful and time-consuming, such as nextgeneration sequencing. In any case, we would still have a high number of false negatives due to wild-type TP53 tumors and the HPV-positive group, requiring an even larger gene panel to cover the high genetic heterogeneity of this group [23]. By contrast, studies on esophageal squamous cell carcinoma (ESCC), which shares the same histology, epithelium of origin and concomitant exposure to the same risk factors as HNSCC, suggest not only that epigenetic alterations such as DNA methylation are more frequent, but could even precede the genetic changes in precursor lesions [212,213].
In HNSCC precursor lesions, p16INK4a is one of the first genes inactivated either by deletion or DNA methylation [23,214]. P16INK4a promoter hypermethylation was already extensively evaluated, but results vary widely, probably due to the divergence of the target location within the promoter region, and the methodologies applied, mostly methylation-specific PCR (MSP). Despite cost-effectiveness, it is a technique with a predominant qualitative profile, and may suffer a strong bias due to primer design, for example [215]. Interestingly, the methylation profile of p16INK4a promoter region does not seem to be associated with HPV status assessed by p16 immunohistochemistry, whereas gene body hypermethylation does [216]. However, p16INK4A gene body hypermethylation is likely to occur after its overexpression [217], being more suitable as an infection marker. In addition, deletion is the most frequent p16INK4a inactivation mechanism in HPV-negative HNSCC (57%), and practically nonexistent in HPV-positive tumors [23], in which p16 immunostaining is used as a surrogate biomarker of HPV infection [22].
Other genes with aberrant methylation in precursor lesions include MGMT, p14ARF, p15INK4b, RARB, and SOCS-3 [83,84,86,218]; p16INK4a and MGMT hypermethylation can already be detected in oral rinse [84] and blood [85] of individuals with precursor lesions or a history of OSCC, reducing the invasiveness of the procedure. Higher methylation levels of p16INK4a, p14ARF, and p15INK4b were observed in LSCC precursor lesions [218]. RARB is hypomethylated in 27% of normal adjacent HNSCC tissues and in 53% of oral precursor lesions [86]. For more detailed reviews on the evaluation of these genes as biomarkers of the cancerization field, please refer to [219,220].
ZNF282 and PAX1 hypermethylation were originally proposed as biomarkers of cervical carcinoma precursor lesions [221][222][223]. In the oral cavity, ZNF582 and PAX1 methylation increase as the epithelium goes through transformation, being able to differentiate the visually normal mucosa or with minimal histological changes (hyperplasia or mild dysplasia) from those more advanced pre-cancer changes (moderate or severe dysplasia) and the tumor itself, with sensitivities of 85% and 72% and specificities of 68% and 86%, respectively [87]. Genes involved in the WNT and the MAPK signaling pathways show a similar profile, with aberrant methylation being observed as the precursor lesions progress [224], and may be the focus of future studies.
Although, to the best of our knowledge, whole-genome bisulfite sequencing has not yet been applied to identify DNA methylation alterations throughout HNSCC natural history, next-generation sequencing (NGS) was used to assess the methylation of specific target genes in oral brushing from healthy donors, low-grade squamous intraepithelial lesions (LG-SIL), high-grade SIL (HG-SIL), and OSCC patients [88]. Among the seven targets selected, ZAP70 was hypermethylated in 100% of OSCC and HG-SIL cases, while GP1BB hypomethylation was detected in 90.9% of these individuals. Healthy donors did not show alterations of these genes, while LG-SIL showed an intermediate profile.
Although the results were encouraging, this was a preliminary study with a limited number of samples.
Besides their potential to be applied in the early diagnosis, cancerization field biomarkers can also be used to establish surgical margins, reducing the chance of recurrences. Although histopathological evaluation is widespread, 10-30% of HNSCC patients with free surgical margins will recur [226][227][228], which is one of the main causes of mortality among HNSCC patients [228,229]. Therefore, a molecular screening of cancerization field alterations may predict this phenomenon and, therefore, improve patients' prognosis [229]. Approximately 25% of HNSCC surgical margins contain genetic alterations [230], and TP53 mutations were already associated with recurrence [231]. Epigenetic alterations, however, also show great potential to predict relapse.
P16INK4A and/or MGMT promoters were found hypermethylated in 32% of HNSCC surgical margins, despite all being considered negative in the histopathological analysis [232]. When evaluated together at the tumor margin, the methylation levels of EDNRB and HOXA9 are predictors of locoregional recurrence-free survival, with their hypermethylation increasing the chance of recurrence approximately three-fold [125].
The analysis of a DNA methylation panel composed of three genes (DCC, CCNA1, and p16INK4A) in surgical margins was able to predict all recurrences in five out of 47 HNSCC patients [126]. In OSCC patients, ZNF582 and PAX1 methylation levels are reduced after treatment, and interestingly they rise again at the recurrence site a few months before the diagnosis [87]. In HNSCC surgical margins, the promoter methylation of another Homeobox gene, PAX5, increases the risk of locoregional recurrence approximately four-fold [127]. While most methylation studies focus on the promoter region of CDKN2A (which encodes both p14ARF and p16INK4A), its gene body hypermethylation and increased p14ARF expression in LSCC were associated with a lower incidence of locoregional recurrence [128]. More recently, a genome-wide DNA methylation screening identified 392 differentially methylated regions associated with recurrence when OSCC free surgical margins were analyzed. Accuracy of up to 98% was achieved to discriminate relapsed and nonrelapsed OSCC patients when a 14-CpG panel signature was applied [233].
MiRNA expression also shows recurrence predictive value. Oral precursor lesions of patients who relapsed showed miR-375 downregulation relative to those who did not relapse, presenting an accuracy of 95.7% and sensitivity of 90% to discriminate the two groups. Despite outstanding discrimination values, it is important to highlight that only six patients were included in the nonrelapsed group [129].
Although little explored in general, the oral cavity has the most well-defined and explored epigenetic cancerization field among HNSCC. This might be a consequence of the relatively easy access to the tumor site and the visual identification of precursor lesions, such as leukoplakias. In addition to the need to gather more data on other head and neck sites and to consistently validate the findings, studies on epigenetic alterations of the cancerization field should adopt more sensitive methodologies, and explore alterations in the mucosa of individuals with low and high exposure to risk factors, different types of precursor lesions, and cancer. Further, large-scale platforms may add new information and help to identify new biomarkers.

Diagnostic Biomarkers
In comparison to the few common genetic alterations found in HNSCC generally, DNA methylation shows more frequent and subsite-specific changes, an important asset for molecular diagnosis [24,212,213,234]. P16INK4A is hypermethylated in 27-44% of HNSCC [93,97,98,235]; however, it has already been reported that almost half of the samples could not be analyzed due to loss of heterozygosity [98,236]. CDH1 is hypermethylated in 35-43% [93,95,235], DAPK in 43% [73], and MGMT in one-third of HNSCC cases [98,103,113,235,237], and hypermethylation of the latter is independent of HPV status. This hypermethylation profile is not exclusive to the tumor and can be observed in the cancerization field, as already mentioned. For example, adjacent tissues show DAPK hypermethylation in 20% of samples [73]. GNAT1 and SEMA3B are hypermethylated in 75% and 77% of OSCC and 56% and 38% in adjacent tissues, respectively. These and other studies emphasize the importance of applying a sensitive methodology to define and differentiate this hypermethylation status [118,238]. Aberrant methylation in other genes, such as FAM135B and ZNF610 hypermethylation and HOXA9 and DCC hypomethylation, show accuracies of 79%, 93%, 93%, and 93% in distinguishing adjacent nontumor tissues and HNSCC, respectively [239]. Table 1 brings additional candidate diagnostic biomarkers based on aberrant DNA methylation in HNSCC.
Despite the promising discrimination values mentioned above, few studies share the same findings on miRNA dysregulation in HNSCC (see Table S1 and Figure 4) [240]. This underscores the high intertumoral heterogeneity and the need for establishing subsitespecific biomarkers.
combined accuracy of 92% to discriminate the tissues [123]. Combining the expression of miR-657 and miR-1287, an accuracy of 97% and sensitivity of 86% were achieved in early stage LSCC [124].
Despite the promising discrimination values mentioned above, few studies share the same findings on miRNA dysregulation in HNSCC (see Table S1 and Figure 4) [240]. This underscores the high intertumoral heterogeneity and the need for establishing subsitespecific biomarkers.
With the exception of the oral cavity, access to other sites of the head and neck might be challenging, and less invasive diagnostic techniques are required. The assessment of epigenetic biomarkers in body fluids, such as saliva and blood, has emerged as an alternative and may even impact the biomarker performance [258]. HNSCC patients already show a global hypomethylation in whole blood, estimated by the evaluation of LINE1 transposable elements. However, in addition to being a generic tumor biomarker [259][260][261][262][263][264], global methylation levels showed overlap between HNSCC patients and controls, jeopardizing their discrimination capacity [110]. THF methylation profile was able to differentiate with 93% accuracy (86% sensitivity) individuals without cancer from those with OSCC or OP-SCC using oral rinse [265]. The methylation profile of MED15 promoter region in the saliva of HNSCC patients showed an accuracy of 63-70% [266]. In the plasma of LSCC patients and individuals without cancer, 26 differentially expressed microRNAs were identified, 17 overexpressed, and 9 downregulated in patients [243].
In addition to tumor diagnosis, the DNA methylation profile can also help to identify the risk factors associated with tumor development. By evaluating the saliva of HPVnegative HNSCC patients, RASSF1A, p16INK4A, TIMP3, and PCQAP/MED15 genes were found hypermethylated relative to healthy individuals and showed a sensitivity of 71% to discriminate the groups [79]. The same genes are hypomethylated in the saliva of HPVpositive HNSCC patients relative to healthy individuals and distinguished the groups with a sensitivity of 80% [79]. A panel with the methylation profile of p16INK4A, RASSF1A, TIMP3, and PCQAP/MED15 genes in saliva was able to differentiate OSCC and OPSCC patients from individuals without cancer with 91% and 92% sensitivity and 99% and 92% specificity, respectively [117].
A common feature of HNSCC is the marked disparity in the overall survival of patients diagnosed in early stages compared to the more advanced ones. Screenings widely applied in oncology, such as those for breast [267] and cervical [267] cancer early diagnosis have sensitivities of approximately 96%. Studies aimed at tumor screening in the oral cavity showed that despite a good number of pre-cancer lesions were identified (~10%), the number of carcinomas found was low [268], with a sensitivity of 71-76% [269,270]. However, even with the low frequency of tumors, reduced staging and improved overall survival were observed in patients due to screening, showing that it can be worthwhile [269].

Biomarkers of Treatment Response and Prognosis
Despite advances in oncology, the average five-year survival of HNSCC patients remains the same, 13-57%, varying according to the affected primary site [271,272]. The therapeutic approach depends on the location, resectability, and strategies focused on organ preservation, including surgery, chemotherapy, and radiotherapy [273]. The gold-standard treatment for the disease is surgery, which may be combined with radiotherapy and, in locally advanced cases, chemotherapy based on taxanes and platinum [273].
In HNSCC, molecular changes and etiology can drive treatment choice. Patients with HPV-positive OPSCC respond better to therapy, and treatment deintensification was already proposed [274]. By contrast, smokers and drinkers generally show a poor response to treatment [275][276][277]. In this context, epigenetic changes have been proposed as surrogate markers of therapeutic response and epigenetic drugs were shown to resensitize cell lines to conventional therapy.
DNA methylation alterations have been recently shown to predict locoregional recurrence after surgery, for example. By evaluating the methylation levels of 13 genes by bisulfite NGS in pre-operative OSCC samples collected by oral bushing, Gissi and colleagues showed that five CpG sites in three genes (EPHX3, ITGA4, and MiR193A) were able to predict adverse outcomes with high accuracy (AUC = 0.85) [138].
HNSCC cell lines resistant to irradiation showed global hypermethylation and differential methylation and expression of up to 84 genes relative to sensitive strains [278]. Among these, CCND2, a cell cycle controller, was hypermethylated in the promoter region [278]. DAPK promoter hypermethylation was detected in cell lines resistant to EGFR inhibitors, one of the few target therapies approved by the FDA to treat HNSCC patients [279,280]. Interestingly, DAPK knockout itself is able to induce resistance against these drugs [281]. In taxol-resistant nasopharyngeal carcinoma cell lines, global hypermethylation and 48 differentially methylated genes (30 hypermethylated and 18 hypomethylated) were observed, and treatment with demethylating agents could resensitize cells to chemotherapy [282]. Treatment with demethylating agents also sensitizes HNSCC cell lines to cisplatin [283].
More recently, FDA also approved immunotherapy based on PD1 and PD-L1 immune checkpoint blockade (pembrolizumab and nivolumab) for treatment of platinum-refractory metastatic HNSCC [289]. PD-L1 and PD-L2 promoter methylation levels are correlated with their expression and associated with HPV infection in HNSCC. Therefore, their assessment might be a promising predictive biomarker for treatment [290]. Furthermore, PD-1 hypermethylation was already associated with a worse prognosis in HNSCC [291].
Epigenetic biomarkers in liquid biopsies also show promising results in predicting patient prognosis. TIMP3 promoter hypermethylation detected in HNSCC patients' saliva seems to be an independent marker of recurrence-free survival [136]. Higher SEPT9 and SHOX2 methylation levels in cell-free circulating DNA in the plasma of HNSCC patients were associated with worse overall survival and increased the chances of death approximately 5.2 and 2.3-fold, respectively [137]. When detected in the plasma of HNSCC patients, the high expression of miR-186-5p and miR-374b-5p before treatment and miR-142-3p after treatment were associated with reduced locoregional control. The expression of miR-142-3p was also associated with reduced disease-free progression [139].

The Epilogue of Epigenetic Biomarkers in HNSCC
The acquisition of epigenetic alterations due to risk factor exposure and transformation in a plastic but stable fashion, and their independence of genetic changes frequency present good potential for epigenetic biomarker development. These changes can be detected early during tumor development, enabling screening programs and increasing survival rates. Furthermore, the same marker has the potential to be used to predict more than one endpoint, such as diagnosis and recurrence, decreasing development time and cost. However, the establishment of the biomarker must be cautious, reproducible, and apply primarily sensitive methodologies.
In the case of DNA methylation, most studies cited here applied MSP or quantitative-MSP (qMSP) for biomarker identification, either directly or after selecting targets from microarray data. Although these techniques are quite user-friendly and generally available, they are subject to amplification biases and usually interrogate single CpG sites, which may impair the interpretation of results. Additionally, MSP does not provide a quantitative measure of methylation levels. This is especially relevant when we consider that pools of cells or of circulating DNA (in liquid biopsies) are usually analyzed. Since loci methylation will vary according to cell type and exposures, DNA methylation levels should be considered as a continuous and not a qualitative variable. Thus, bisulfite pyrosequencing or bisulfite amplicon-NGS could be more suitable for identifying potential biomarkers. Both techniques provide quantitative measures of DNA methylation levels and interrogate a higher number of CpG sites, but have a higher cost and require equipment that is not available in every research or clinical laboratory. Finally, all these techniques require a prior conversion of unmethylated cytosines into uracil by DNA treatment with sodium bisulfite. Despite its widespread use, this treatment might be an additional challenge for the analysis of poor-quality DNA. When these molecules are extracted from formalin-fixed paraffinembedded (FFPE) tissues or from liquid biopsies, they are usually highly degraded, and the treatment with sodium bisulfite induces further degradation. However, the alternatives, such as the use of methylation-sensitive restriction enzymes or CpG methylation immunoprecipitation suffer from lower sensitivity. Based on these considerations, the ideal combination of techniques to evaluate DNA methylation alterations as biomarkers will depend on sample type, difference range, resources available, and biomarker sensitivity, among other factors. Nevertheless, after the establishment of methylation and discrimination levels, the construction of low-cost and more qualitative tests (such as qMSP) could be valid to decrease time and cost [294].
Although generally less explored than DNA methylation, miRNA-based biomarker development in HNSCC seems more advanced in terms of methodology. While many candidates are identified by microarray, a semiquantitative technique, their validation and assessment in other cohorts and sample sources are usually accomplished by reverse transcription followed by quantitative PCR (RT-qPCR). This is a widespread quantitative approach already applied in the clinics, as recently observed in COVID-19 diagnosis. The major challenge in the development of these biomarkers, however, is probably the definition of cutoffs for distinguishing groups, although this can be overcome by reproducibility studies, necessary for the development of any biomarker.
It is important to note that some genetic factors can affect the performance of the epigenetic biomarker. For example, MTHFR polymorphisms can affect p16INK4A and MGMT methylation levels in the oral cavity [295,296]. Even unexpected factors, such as pregnancy with intrauterine growth restriction, can cause aberrant methylation, such as RASSF1A hypermethylation [297], also observed in HNSCC [298]. In this context, the development of panels can be more sensitive and specific, avoiding interference from different factors and the generation of false negatives. The panel can be applied in minimally invasive approaches such as body fluids, where epigenetic changes are highly stable, or even in tumor biopsies, when available [299,300].
Until now, most of the proposed biomarkers are based on promoter hypermethylation of tumor suppressor genes, the mechanism generally involved in their silencing. Thus, the methylation status of these genes may also be indicative of the use of demethylating agents in the treatment, aiming for their reactivation. This bias towards promoters highlights the need for further studies characterizing the methylation profile in other genomic regions, such as intergenic regions, enhancers, and gene bodies.
Although less explored than DNA methylation and miRNAs, histone modifications have also been reported in HNSCC and have been associated with patients' outcomes and cancer phenotypes. This is another potential field for the development of biomarkers, not addressed in this review (for further information, please refer to [285,[301][302][303][304]), and should be further explored. In OSCC, H3K9ac reduction was associated with the activation of epithelial-mesenchyme transition, cell proliferation, and poor prognosis [305]; also in OSCC, the activation of Wnt/β-catenin signaling and the consequent induction of cell proliferation and invasion were associated with increased H3K27ac through the activation of PLAC2, a long-noncoding RNA [306]. Finally, high H3K27me3 and low H3K4ac levels were associated with poor prognosis [307], while reduced H3K4me3 and increased H3K4me2 levels were observed in OSCC relative to normal tissues [308], suggesting that histone methylation could also be an OSCC biomarker.
As mentioned previously, most studies considered all HNSCC sites as a single entity or focused on the oral cavity, probably due to the easier access and higher tumor occurrence. This might be an additional challenge because of differences in carcinogenesis mechanisms behind each specific subsite. Thus, the epilogue of HNSCC biomarkers is still blurry, and epigenetic changes can be an eye drop.

Conclusions
As highlighted in this review, epigenetic alterations are a common feature of HNSCC, and affect all steps in their natural history. In this context, intratumor heterogeneity, a major cause of cancer treatment failure and relapse, cannot be addressed only genetically. Epigenetics confers different phenotypes to subclones and, due to its reversibility, emerges as a potential therapeutic target. Therefore, to improve clinical management, it is of utmost importance that all professionals involved in patient care come to know this somewhat unexplored field.
Thus far, most studies have focused on the methylation of promoters of tumor suppressor genes. Although this approach has already identified putative carcinogenic pathways and biomarkers, much is yet to be learned from epigenetic alterations in HNSCC. Not only the methylation of other genes and other genomic regions, but also other players, including noncoding RNAs and chromatin remodelers, should be explored. For example, global hypomethylation, associated with poor prognosis in other cancer types [189,[309][310][311], seems to differ according to HPV status and could help explain the prognosis differences observed. Furthermore, mutations in the histone methyltransferases NSD1 and NSD2 define a group of good prognosis among LSCC patients, but not in other HNSCC subtypes [312]. These data highlight the importance of painting a more detailed canvas of epigenetic landscape in HNSCC. The integration of different epigenetic mechanisms and the comprehension of both inter-and intratumoral heterogeneity may help achieve more personalized HNSCC patient care.