Background
Disruption of the epigenomic landscape is recognized as a widespread feature inherent in tumor development and progression [
1,
2]. In particular, aberrant patterns of histone modifications and DNA methylation have been extensively studied because of their relevance in altering the chromatin structure and thereby also gene transcription. Specifically, research on DNA methylation changes in neoplasia has generated a multitude of biomarkers for diagnosis, prognosis, and response to treatment with application in the clinical management of several types of cancer [
3].
DNA methylation changes in cancer include a wave of global DNA hypomethylation along with loci-specific hypermethylation predominantly affecting CpG islands in gene regulatory regions. Downstream transcriptional alterations have been described at all stages of tumor progression, affecting virtually all signaling pathways and unleashing a profound transformation of the cellular phenotype.
Cutaneous melanoma is the most life-threatening form of skin cancer, and its incidence and mortality keeps on rising, with the highest increase among men aged older than 55 years and women of all ages [
4]. Nonetheless, clinical staging of patients with primary tumors relies entirely on classical histological biomarkers such as tumor thickness and ulceration [
5]. This particular neoplasm exhibits a phenotypic plasticity that accounts for the high degree of intrinsic and acquired resistance to antineoplastic, targeted therapies, and immunotherapies [
6‐
10]. Large-scale studies of transcriptomic alterations, along with the development of new molecular tools and in vivo models, have helped elucidate molecular cues contributing to metastasis, allowing a better understanding of melanoma biology and setting the basis for new treatment strategies [
7,
11‐
14]. On the epigenomic side, several studies have reported DNA methylation changes in melanoma associated with inactivation of candidate tumor suppressor genes (e.g.,
MAPK13) or abnormal re-expression of oncogenes during tumor progression (e.g., TBC1D16), when examining pre-selected promoter regions for the presence of DNA methylation, or by genome-wide based approaches [
15‐
23]. Importantly, however, the vast majority of these studies are limited to melanoma metastases and lack primary melanomas, making it problematic to identify early events during melanoma development and progression. In addition, the absence of primary tumors makes it impossible to determine DNA methylation biomarkers associated with prognosis of the patient.
Here, we present a comprehensive analysis of DNA methylation patterns during all progression stages of cutaneous melanoma. By using Infinium HumanMethylation450 BeadChips (Illumina) [
24] and integrating the DNA methylome of benign nevi (n = 14) and malignant melanoma from both primary (n = 33) and metastatic (n = 28) sites with gene expression data, we identify, as well as validate in independent patient cohorts, biomarkers for melanoma development (e.g.,
HOXA9 DNA methylation), tumor progression (e.g.,
TBC1D16 DNA methylation), and patient prognosis (e.g.,
PON3 DNA methylation and OVOL1 protein expression).
Discussion
To enable the discovery of novel biomarkers and the development of more efficient therapies for melanoma, our understanding of the molecular features underlying its aggressive phenotype, and how these traits are regulated by constant modifications of its transcriptome, need to be enhanced. In this study, we aimed to profile, in an unbiased manner, DNA methylation changes occurring along the evolution of melanoma development and progression. Moreover, DNA methylation biomarkers represent a valuable tool for the clinical management of several cancer types [
3]. Despite several DNA methylation changes identified in melanoma [
21‐
23,
47], there is a lack of unbiased comprehensive analysis of clinical specimens that describes the molecular pathways targeted by epigenomic changes, and provide biomarkers that can be readily used as markers for the diagnosis and evaluation of melanoma aggressiveness. To overcome this, our study represents the most comprehensive epigenomic profiling assessment of well-annotated human melanomas. In more detail, we (1) performed genome-wide DNA methylation profiling of clinical specimens covering various stages of development and progression of SSMM; (2) integrated the observed changes with gene expression data, in order to gain insights of potential functional relevance; (3) proved the robustness of our findings through extensive validation in multiple independent cohorts; and (4) finally translated our results to potentially valuable protein biomarkers.
The present study illustrates the DNA methylation dynamics during melanoma development and progression. Aberrant DNA hypermethylation occurs predominantly in CpG island-associated promoters in melanoma cells, as compared with benign nevi. This has been described for several tumor types, and represents a common hallmark of neoplastic transformation. DNA hypomethylation, by contrast, was more frequently found at later stages of progression and predominantly associated with gene bodies, although some loci-specific changes were observed. A previous study suggested that DNA methylation alterations in melanoma could be partly attributable to the dramatic loss of 5-hydroxymethylcytosine observed during malignant progression, caused by mutation of the TET2 enzyme coding gene [
48]. Altogether, a large number of DNA methylation changes were identified in relation to different stages of the disease. We were able to confirm several hypermethylated genes (see Additional file
1: Tables S4–S6 for gene lists) reported in previous studies, including transcription factor AP2 (TFAP2) genes [
49], which play essential roles in the development of the epidermis and migratory cells of the neural crest, HLA-class I members [
50],
SOCS-1 and -
2, and members of the tumor necrosis factor receptor superfamily (TNFRSF)
TNFRSF10C and
TNFRSF10D [
18], as well as
MAPK13 and
PLEKHG6 [
21], and HOX family genes such as
HOXD9 [
22]. We did not detect DNA methylation differences in any of the
MAGE genes, but observed frequent hypomethylation in
TBCD1D16 [
47] and in several members of the
SERPINB gene cluster also involved in tumorigenesis (see Additional file
1: Tables S7–S9 for gene lists) [
51].
By crossing our dataset with available gene expression databases, we gained insight into the potential functional relevance of DNA methylation in altering the phenotype of melanoma cells. Promoter hypermethylation of genes involved in cell adhesion, such as
ANXA9,
CLDN5,
GJA1,
GJB2, or
LAMA3, was enriched as determined by gene ontology and GSEA analysis (Additional file
1: Tables S19 and S21), in line with previous reports (see Additional file
1: Table S18 for gene list) [
52,
53]. The deregulation of cell adhesion has been recognized in other neoplasms as a characteristic event facilitating escape of the primary niche, and has been confirmed in our study by comparison with available methylation and expression databases. Loss of terminal differentiation traits, as observed by inactivation of
ESR1,
PTPRS, or the metastasis suppressor gene
GATA3, may reflect the intrinsic capacity of melanoma cells to gain plasticity, and to progressively acquire changes that trigger metastatic dissemination [
54,
55]. In line with this, GSEA indicated considerable and significant overlap between genes with downregulated expression in melanoma metastases compared to the primary tumor [
36] and our set of differentially hypermethylated genes, and between genes with upregulated expression in invasive breast cancer compared to non-invasive tumors [
40] and our differentially hypomethylated genes. The regulation of gene expression patterns by DNA methylation changes at different stages seems to reflect the phenotype switch concept that emerged from transcriptomic studies of melanomas [
56‐
58]. Moreover, a series of studies have observed a stem-cell phenotype increasing during melanoma progression, which was strongly sustained by a tumor-promoting microenvironment [
59‐
62]. Pathways activated by DNA hypomethylation were mostly linked to inflammation and innate or adaptive immunity processes (Additional file
1: Tables S20 and S22). Of note, although the effect of tumor-associated immune and stromal cells was minimized (by only including lesions with at least 75% of tumor cells; see Methods), some of the observed changes in DNA methylation are likely to originate from both tumor cells and normal cells. It has been hypothesized that expression of these immune and inflammatory factors in advanced melanomas interacts with the tumor microenvironment and creates a milieu supportive of tumor progression [
63]. Specifically, overexpression of
TLR4 and
CCR7 in advanced melanomas as a result of loss of promoter DNA methylation fosters tumor progression by hijacking immune responses (see Additional file
1: Table S18 for gene list) [
64,
65]. Further, DNA repair processes are also empowered by hypomethylation of
PARP1 (Additional file
1: Tables S8 and S18), a chromatin-associated enzyme involved in base-excision repair [
66,
67]. In agreement with our data, upregulation of DNA repair pathways concomitant with a loss of cell-cell adhesion has also been reported in vertical-growth phase and metastatic melanomas in relation to regulation of NF-kappaB signaling and inhibition of apoptosis [
13,
67‐
69].
Overall, our data support a central role for DNA methylation in modulating the transcriptome of melanoma cells, thereby changing their phenotype to promote tumor progression. At initial steps, prominent epigenomic inactivation induces loss of cell-cell contacts and truncates differentiation programs, increasing plasticity of tumor cells to acquire invasive capacities. In this line, epigenomic regulation underlies previous observations reporting downregulation of cell adhesion molecules in the most aggressive vertical-growth phase melanomas [
13,
70]. Subsequently, as melanoma gains depth and invades the dermis, a transcriptional switch occurs through modulation of DNA methylation patterns leading to the epigenome displayed in the metastatic sites. DNA hypomethylation seems to be predominant at this point, and reactivation of immune and inflammation processes is evident. Upregulation of inflammation and immune response pathways in tumor cells seem to co-opt to turn the microenvironment into a tumor-promoting milieu [
71,
72], and has been associated with shortened relapse-free survival [
73].
Within the large panel of genes that were identified to be transcriptionally altered during melanoma progression, we selected a series of markers (
AKT3,
EPHX3,
GJB2,
HOXA9,
MEOX2,
PON3,
RBP1,
SERPINE2,
TBC1D16,
TFAP2B, and
TWIST1) for further validation. The robustness of our findings was confirmed following pyrosequencing of the genes in an independent patient cohort, pointing at these alterations as widespread attributes of melanoma progression and worth further characterization. In support of this, one of the members of our gene signature, TBC1D16, has recently been shown to be involved in the metastatic cascade of melanoma [
47].
A melanoma survival signature could also be inferred from this integrative study. Through a supervised correlation of the DNA methylation profiles with clinical parameters, we were able to refine a DNA methylation panel predictive of melanoma-specific survival. In line with this, significant overlap was observed, by GSEA, between our differentially hypermethylated genes and downregulated genes in melanoma patients with a reported distant metastasis within 4 years [
11], and our differentially hypomethylated genes and upregulated genes in high versus low risk uveal melanomas [
41]. Nowadays, prognosis for patients with clinically localized primary cutaneous melanoma relies mostly on histological parameters as tumor thickness, ulceration, and mitotic rate in the invasive component. Here, we identified, and validated in an independent validation cohort, three genes (
MEOX2,
OLIG3, and
PON3) for which the degree of DNA methylation can predict the prognosis of melanoma patients. Importantly,
PON3 DNA methylation was independent of classical prognostic parameters and could, therefore, be of added value when implemented in the pathological staging procedure. In addition, we validated by IHC the prognostic usefulness of protein biomarkers (AKT3, OVOL1, and TFAP2B) that were discovered by our DNA methylation analyses, thereby verifying DNA methylomics as a valid screening tool to identify potential protein biomarkers. Furthermore, in the current era of “liquid biopsies”, the observed changes in methylation might be targets for the study of cell-free DNA in the serum of melanoma patients. Once these findings are corroborated, it could be of great utility for its clinical implementation to improve the management of melanoma patients.
Acknowledgements
We thank Direna Alonso-Curbelo for fruitful discussions; members of Translational Cell and Tissue Research (KU Leuven) for helpful criticism; and Catia Moutinho, Sebastián Moran, and Bozena Fender for help with sample processing.