If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Department of Dermatology, Leiden University Medical Center, Leiden, The NetherlandsThe first two and last three authors contributed equally to this work
Department of Pathology, GROW-School for Oncology and Developmental Biology, Maastricht University Medical Center, Maastricht, The NetherlandsThe first two and last three authors contributed equally to this work
Department of Medical Statistics and Bioinformatics, Leiden University Medical Center, Leiden, The NetherlandsCurrent address: Department for Health Evidence, Radboud University Medical Center, Nijmegen, The Netherlands
Department of Pathology, GROW-School for Oncology and Developmental Biology, Maastricht University Medical Center, Maastricht, The NetherlandsThe first two and last three authors contributed equally to this work
Department of Pathology, GROW-School for Oncology and Developmental Biology, Maastricht University Medical Center, Maastricht, The NetherlandsThe first two and last three authors contributed equally to this work
Department of Dermatology, Leiden University Medical Center, Leiden, The NetherlandsThe first two and last three authors contributed equally to this work
Dysplastic nevi are melanocytic lesions that represent an intermediate stage between common nevus and melanoma. Histopathological distinction of dysplastic nevus from melanoma can be challenging and there is a requirement for molecular diagnostic markers. In this study, we examined promoter CpG island methylation of a selected panel of genes, identified in a genome-wide methylation screen, across a spectrum of 405 melanocytic neoplasms. Promoter methylation analysis in common nevi, dysplastic nevi, primary melanomas, and metastatic melanomas demonstrated progressive epigenetic deregulation. Dysplastic nevi were affected by promoter methylation of genes that are frequently methylated in melanoma but not in common nevi. We assessed the diagnostic value of the methylation status of five genes in distinguishing primary melanoma from dysplastic nevus. In particular, CLDN11 promoter methylation was specific for melanoma, as it occurred in 50% of primary melanomas but in only 3% of dysplastic nevi. A diagnostic algorithm that incorporates methylation of the CLDN11, CDH11, PPP1R3C, MAPK13, and GNMT genes was validated in an independent sample set and helped distinguish melanoma from dysplastic nevus (area under the curve 0.81). Melanoma-specific methylation of these genes supports the utility as epigenetic biomarkers and could point to their significance in melanoma development.
Abbreviations
BMCA
bisulfite melting curve analysis
CDH11
cadherin 11 type 2
CLDN11
claudin11
FFPE
formalin-fixed paraffin-embedded
GNMT
glycine N-methyltransferase
MAPK13
mitogen-activated protein kinase 13
MSP
methylation-specific PCR
PPP1R3C
protein phosphatase-1 regulatory subunit 3C
Introduction
Cutaneous melanoma is a malignant tumor that arises from melanocytes residing in the skin. The lifetime risk of developing melanoma is approximately 2% in the United States and Europe and both incidence and mortality rates continue to rise (
). Early diagnosis is important to prevent the formation of lethal metastasis. However, the clinical diagnosis of melanoma is challenging in a proportion of cases. This applies to visual assessment of pigmented skin lesions and to histopathological examination of biopsied tissue, the standard for melanoma diagnosis. A substantial interobserver discordance rate of 14% in the pathological diagnosis of melanoma has been reported (
). Histologically, these lesions demonstrate random cytological atypia, architectural disorder, and stromal changes. The relevance of dysplastic nevus to melanoma progression is underscored by observations that dyplastic nevi are found in contiguity with melanoma in a significant subset of cases (
). On the basis of morphological and biological characteristics, dysplastic nevi have been proposed to represent an intermediate lesion between common nevi and malignant melanoma in the multistep tumor progression model of melanocytic neoplasia (
In the melanoma progression model proposed by Clark, melanocytic cells acquire malignant traits in discrete steps, a process driven by accumulation of genetic and epigenetic alterations (
). Although there is limited information about genetic and epigenetic alterations in dysplastic nevi, patterns of intragenic mutations, chromosomal aberrations, and DNA methylation alterations are assumed to differ from those in melanoma. Therefore, detection of these molecular differences could aid in the correct classification of those cases in which morphological diagnosis fails to discriminate.
Systematic characterization of molecular alterations in melanoma has provided a wealth of information on acquired DNA alterations in melanoma cells, which could be used in the molecular diagnosis of this malignant disease. Recently, we performed a genome-wide promoter methylation analysis of 14 495 genes in melanoma and common nevus samples and found widespread aberrant promoter methylation in melanoma (
). Among the hundreds of gene promoters that exhibited methylation in melanomas but not in common nevi, we identified several tumor suppressor genes, causally implicating epigenetic mechanisms in melanoma development.
The objectives of this study were to gain insight into epigenetic deregulation in the different stages of melanocytic neoplasia and to assess the potential diagnostic value of genes differentially methylated between melanoma and dysplastic nevus. To this end, we examined the methylation status of genes that were previously identified in a genome-wide methylation screen in a cohort of 251 melanocytic neoplasms and subsequently validated a diagnostic algorithm incorporating different gene methylation features in a second independent series of 154 dysplastic nevus and primary melanoma samples. The combined analysis of promoter CpG island methylation of the five genes proposed in this study could be of help in the histopathological distinction of melanoma from dysplastic nevus.
Results
Selection of genes for promoter CpG island methylation analysis in common nevus, dysplastic nevus, primary melanoma, and metastatic melanoma
In a previously performed genome-wide methylation analysis using Infinium 27-k beadchips, we identified 106 genes that were significantly and frequently more methylated in primary cutaneous melanomas than in common nevi (
). Here, we set out to analyze the methylation status of the 12 most differentially methylated genes in dysplastic nevi, next to an independent set of common nevi and primary melanomas (Figure 1a and b). Primers for bisulfite melting curve analysis (BMCA) were designed to encompass or to be in close proximity of the corresponding 50-mer probe sequence on the beadchip. C4orf8 and HIST1H3E were excluded for further methylation analyses owing to suboptimal primer design; in their place, PPP1R3C and CLDN11 were included, as we had observed that they exhibited notable differential methylation between primary melanoma and common nevus samples (
). In the independent set of 10 common nevus, 20 dysplastic nevus, and 15 primary melanoma biopsy samples, BMCA showed that C1orf106, MAPK13, CDH11, GNMT, PPP1R3C, and CLDN11 methylation was absent in common and dysplastic nevi, whereas frequent methylation (20–67%) was observed in primary melanomas (Figure 1c). The promoters of HOXA9 and CNTN1 were nonprogressively methylated in 10–80% of common nevi, dysplastic nevi, and primary melanomas. Interestingly, PLEKHG6 showed progressively higher levels of methylation with tumor progression, that is, in 0% of common nevi, 35% of dysplastic nevi, and 60% of primary melanomas, suggesting that transition from a benign melanocytic lesion to a malignant tumor can be accompanied by a gradual increase in methylation of certain genes (Figure 1c). LEP showed mosaic methylation in melanoma and nevi, with only subtle differences between these sample groups, whereas the ABCA3 promoter region appeared to be unmethylated in melanoma, as well as in nevi (Supplementary Figure S1 online). Genes showing frequent methylation in melanoma, but not, or scarcely, in common and dysplastic nevi, were prioritized for further analyses in a large series of samples.
Figure 1Differential promoter methylation of HOXA9, C1orf106, MAPK13, CDH11, EFCAB1, CNTN1, GNMT, PLEKHG6, PPP1R3C, and CLDN11 in common nevi, dysplastic nevi, and primary melanomas. (a) Schematic depiction of the workflow used to select candidate genes for methylation analyses in large series of melanocytic biopsy samples. (b) The 12 most frequently methylated genes identified by comparative analysis of genome-wide methylation data from 24 primary melanomas and five common nevi. (c) Methylation frequency of 10 genes in an independent set of 10 common nevi, 20 dysplastic nevi, and 15 primary melanomas as assessed by bisulfite melting curve analysis (BMCA). Black triangles indicate the position of the melting curve peak for the respective positive (fully methylated) and negative (fully unmethylated) control.
Differential promoter methylation of CLDN11, CDH11, PPP1R3C, MAPK13, and GNMT in common nevus, dysplastic nevus, and melanoma
Promoter methylation of CLDN11, CDH11, PPP1R3C, MAPK13, and GNMT was assessed in a large series of 251 formalin-fixed, paraffin-embedded (FFPE) biopsy samples consisting of 62 common nevi, 72 dysplastic nevi, 101 primary melanomas, and 16 melanoma metastases, designated series 1. We applied nested methylation-specific PCR (MSP) because this method is better suited for analysis of FFPE samples than BMCA (
). To compare the results of BMCA and MSP, we subjected 31 samples to methylation analysis using BMCA and MSP. This revealed a high concordance rate of 84–97% between both techniques, with higher sensitivity for detecting methylation of MSP (Supplementary Table S1 online). C1orf106 was excluded at this stage owing to suboptimal MSP primer design, and methylation analysis by MSP was performed for CLDN11, CDH11, PPP1R3C, MAPK13, and GNMT. The characteristics of the five genes are detailed in Table 1. The promoter CpG island regions of these genes as studied by MSP and BMCA are depicted in Figure 2a. MSP was successfully performed in 87–98% of the samples in series 1 (Figure 2b, Supplementary Table S2 online).
Table 1Characteristics of five genes that were selected for methylation analysis in common nevus, dysplastic nevus, primary melanoma, and metastatic melanoma biopsy samples
Described in literature as methylated in:
Gene
Function
Location
Tumor suppressor
Methylation-associated gene silencing
Cancer
Melanoma
MAPK13
p38 MAP kinase involved in relaying intracellular signals for a variety of cellular processes
The human cadherin 11 is a pro-apoptotic tumor suppressor modulating cell stemness through Wnt/beta-catenin signaling and silenced in common carcinomas.
Figure 2Methylation analysis of CLDN11, CDH11, PPP1R3C, MAPK13, and GNMT in large series of common nevi, dysplastic nevi, primary and metastatic melanomas. (a) CpG island promoter region of the five genes, with the location of the primers used for methylation-specific PCR (MSP) and bisulfite melting curve analysis (BMCA) in this study. (b) Electrophoretic analysis of MSP amplification products of CLDN11, CDH11, PPP1R3C, MAPK13, and GNMT. N, common nevus; DN, dysplastic nevus; M, melanoma; u, unmethylated; m, methylated; IVD, positive control for methylated alleles (lymphocyte DNA treated with Sss1 methyltransferase); HUV, negative control for unmethylated alleles (DNA from human umbilical vein endothelial cells); H2Oo, no template control for first amplification with flanking primers; H2Oi, no template control for second amplification with primers specific for methylated and unmethylated DNA.
Methylation frequencies of CLDN11, CDH11, PPP1R3C, MAPK13, and GNMT in common nevi, dysplastic nevi, and primary and metastatic melanomas are presented in Table 2. For further analyses and generation of a diagnostic algorithm, only methylation data of samples for which MSP was performed successfully for all five genes (55 common nevi, 57 dysplastic nevi, 79 primary melanomas, and 15 metastatic melanomas) are included in Table 2 and Figure 3. (Results for all analyzed samples of series 1 are given in Supplementary Table S3 online.) Remarkably, CLDN11 displayed the absence of methylation in both common and dysplastic nevi, whereas it was methylated in 48% of primary melanomas and 73% of metastatic melanomas in series 1 (Table 2). CDH11 and PPP1R3C showed the absence of methylation in common nevi; only 5 and 14% of dysplastic nevi harbored methylation for CDH11 and PPP1R3C, respectively. Methylation frequencies in primary and metastatic melanomas were 41 and 47% for CDH11, and 52 and 60% for PPP1R3C, respectively. GNMT harbored promoter methylation in 4% of common and 11% of dysplastic nevi, but yet again higher methylation frequencies were found in primary (46%) and metastatic (47%) melanomas. For MAPK13, methylation was observed in 18% of common and 26% of dysplastic nevi, with higher methylation frequencies in primary (62%) and metastatic (67%) melanomas. The methylation patterns of the five genes, showing progressive increase in methylation frequency in different stages of melanocytic neoplasia, are depicted in Figure 3a. We noted significantly higher promoter methylation frequencies in particular following the transition to melanoma. Methylation of each of the five genes was detected only 12 times (4%) in 55 common nevi and 32 times (11%) in 57 dysplastic nevi, significantly lower when compared to 196 times (50%) in 79 primary melanomas (P<0.001) and 44 times (59%) in 15 metastatic melanomas (Figure 3b). Complete absence of methylation for all five genes was found in 44 of 55 (80%) common nevi, 39 of 57 (68%) dysplastic nevi, 18 of 79 (23%) primary melanomas, and 2 of 15 (13%) metastatic melanomas (Figure 3c). There was no significant correlation between the grade of atypia of dysplastic nevi and promoter methylation frequency.
Table 2Methylation frequency of five candidate genes in biopsy samples of series 1 (55 common nevi, 57 dysplastic nevi, 79 primary melanomas, and 15 metastatic melanomas), together with the specificity and sensitivity of each gene for the distinction of primary melanoma samples from dysplastic nevus samples of series 1
Figure 3Pattern and frequency of CLDN11, CDH11, PPP1R3C, MAPK13, and GNMT promoter methylation in a large series of common nevi, dysplastic nevi, primary melanomas, and metastatic melanomas (series 1). (a) Heatmap depiction of CLDN11, CDH11, PPP1R3C, MAPK13, and GNMT promoter methylation status in 55 benign nevi, 57 dysplastic nevi, 79 primary melanomas, and 15 metastatic melanomas from series 1. White, unmethylated; red, methylated. (b) Percentage of methylation events (methylated genes) found in each sample within the groups of common nevus, dysplastic nevus, primary melanoma, and metastatic melanoma. **P<0.01 and ***P<0.001 by two-sided Fisher’s exact test. (c) Overview of total number of genes that were methylated in each sample within the groups of common nevus, dysplastic nevus, primary melanoma, and metastatic melanoma. NS, nonsignificant.
Promoter hypermethylation is known to increase with age, and the mean age of the melanoma patients was higher (63 years) than of the dysplastic nevus patients (46 years). The differences in methylation frequencies of CLDN11, CDH11, PPP1R3C, MAPK13, and GNMT between melanomas and dysplastic nevi were comparable in the subset of patients younger than 50 years and those older than 50 years (Supplementary Table S4a online). From this it can be concluded that the observed methylation differences between melanoma and dysplastic nevus cannot be attributed to age. For gender, similar results were obtained (Supplementary Table S5a online). CLDN11, CDH11, PPP1R3C, MAPK13, and GNMT promoter methylation has been demonstrated to be associated with transcriptional silencing in tumor cell lines (Table 1). Additional pathway analysis specified that the products of these five genes and their predicted interactors are part of gene signaling networks involved in cell–cell adhesion, cell junction assembly, and adherens junction organization (Supplementary Figure S2 online).
A diagnostic algorithm combining epigenetic markers to distinguish melanoma and dysplastic nevus
The diagnostic specificities and sensitivities of CLDN11, CDH11, PPP1R3C, MAPK13, and GNMT for the distinction of primary melanomas from dysplastic nevi of series 1 are given in Table 2. After having determined these characteristics of the five methylation markers individually, we continued by examining which combinations of markers could aid in the differential diagnosis of primary melanoma and dysplastic nevus. Diagnostic algorithms were created using the set of 57 dysplastic nevi and 79 primary melanomas from series 1, and the accuracy of these models was subsequently tested in a second sample set consisting of 72 dysplastic nevus and 82 primary melanoma biopsy samples, designated series 2. In series 1, the three most differentially methylated genes were CLDN11, CDH11, and PPP1R3C. A first, simple three-gene diagnostic model was created that considers methylation of either of these three genes as indicative of melanoma. This model yielded a specificity of 89% and sensitivity of 67% in series 2 used as test set (Figure 4a).
Figure 4Testing and validation of diagnostic models that incorporate CLDN11, CDH11, PPP1R3C, MAPK13, and GNMT promoter methylation. (a) Diagnostic value was tested in 57 dysplastic nevi and 79 primary melanomas from series 1 (training set), yielding diagnostic scores in the form of a three-gene model and a two-step model, followed by validation of these models in 72 dysplastic nevi and 82 primary melanomas from series 2 (test set). (b) Receiver operating characteristic curve of the two-step diagnostic model. (c) Percentage of methylation events within the dysplastic nevi and primary melanomas of series 2. ***P<0.001 by two-sided Fisher’s exact test. (d) Promoter methylation status of the five genes in 72 dysplastic nevi and 82 primary melanomas from series 2. White, unmethylated; red, methylated.
As CLDN11 methylation occurred exclusively in melanoma in series 1 used as training set, this epigenetic event had the highest discriminatory value with a specificity of 100% and sensitivity of 48% (Table 2). On the basis of a logistic regression model, a diagnostic algorithm was constructed consisting of two discrete steps. First, CLDN11 methylation was evaluated, and, if present, a lesion was classified as melanoma. Second, methylation of CDH11, PPP1R3C, MAPK13, and GNMT was taken in consideration for samples with no CLDN11 methylation (Figure 4a). Adding methylation information of the other four genes generated additional diagnostic value by increasing the sensitivity of the model to detect melanoma. This model has a receiver operating characteristic curve with an area under the curve of 0.806 (Figure 4b). The condition that a lesion is classified as melanoma if CLDN11 or at least two other genes are methylated yields a specificity of 89% and sensitivity of 66% in series 2 (Figure 4a). Promoter methylation frequencies and patterns of the five interrogated genes in series 2, visualized in Figure 4c and d and reported in Supplementary Table S6a online, were generally similar to that of series 1. In series 2, CLDN11 methylation was detected in 6% of dysplastic nevus samples and 52% of primary melanoma samples. Validation of methylation frequencies for the five genes in sample series 2 demonstrated that, in addition to CLDN11, also methylation of CDH11, PPP1R3C, and GNMT have high specificity for melanoma.
Discussion
Dysplastic nevi are melanocytic neoplasms with cytonuclear and architectural atypia and stromal alterations and are generally considered to constitute an intermediate stage between common nevi and melanoma. These lesions have an increased risk of developing into melanoma, and it can be difficult to clinically distinguish dysplastic nevus from early-stage melanoma (
). In this study, we examined promoter CpG island methylation of 12 genes, previously identified in a genome-wide methylation screen, in metastatic and primary melanoma, dysplastic nevus, and common nevus biopsy specimens. We observed progressive promoter CpG island hypermethylation of these genes, with the methylation frequencies increasing from common nevus to metastatic melanoma. Dysplastic nevi, although less commonly than melanoma, demonstrate promoter hypermethylation of genes with tumor-suppressive functions, including CDH11. This shows that dysplastic nevi may already resemble their malignant counterparts at the epigenetic level and suggests that epigenetic instability can occur early, in premalignant stages of melanocytic neoplasia. In addition, it reinforces the notion of dysplastic nevus as an intermediate step in melanoma progression.
Promoter CpG island methylation analysis in a first, smaller series of samples using BMCA suggested that promoters of the CLDN11, CDH11, PPP1R3C, MAPK13, and GNMT genes could be selectively methylated in melanoma. This would render these methylation events suitable epigenetic biomarkers to improve the diagnosis of melanoma and to allow distinction with dysplastic nevus. Using MSP in a large series of 251 melanocytic neoplasm samples, we were able to show the discriminatory value of detecting promoter methylation of these five genes. Methylation detection by the MSP technique is especially suited in this setting, as it yields reproducible results and can directly be applied to FFPE-based samples in the clinic. On the basis of a logistic regression model analysis, we developed a diagnostic score that incorporates different gene methylation features, consisting of assessment of CLDN11 methylation first, followed by determination of the methylation frequency of CDH11, PPP1R3C, MAPK13, and GNMT in DNA isolated from a biopsy sample. Testing of the diagnostic accuracy of this score in another independent series consisting of 82 primary melanoma and 72 dysplastic nevus samples revealed a receiver operating characteristic area under the curve of 0.806 in this independent test set. A simpler three-gene model that incorporates only CLDN11, CDH11, and PPP1R3C as marker of melanoma has a specificity of 89% and sensitivity of 67% in the validation sample set. In our analysis, we have pursued methylation events that are present in melanoma and do not occur in dysplastic nevus. This has resulted in a panel of epigenetic markers with very high specificity for melanoma, but with moderate sensitivity. In particular for screening purposes, a sensitive test would be preferred over a more specific test. By varying the parameters of the logistic regression model or addition of genes that are more frequently methylated in melanoma, such as HOXA9, an increase in sensitivity can be achieved, but at the expense of specificity. The binary results of methylation detection may confer an advantage over comparative genomic hybridization, fluorescent in situ hybridization, or combined immunohistochemical detection of melanoma markers, where interpretation of results can have higher interobserver variability (
). Comparative genomic hybridization and fluorescent in situ hybridization, genomic methodologies used primarily in the research setting, may yield higher diagnostic accuracy than methylation detection of a few genes.
Remarkably, methylation of the CLDN11 gene was completely specific for melanoma, that is, methylation affected 48% of primary melanoma and 73% of metastatic samples, whereas it was absent in common and dysplastic nevi in the first large sample series. In the second sample series, used to validate the diagnostic algorithm, CLDN11 was found to be methylated in 6% of dysplastic nevus samples. Detection of CLDN11 methylation might in particular be used clinically to distinguish malignant from benign melanocytic lesions. Its methylation was shown previously to be associated with transcriptional silencing (
). CLDN11 encodes a member of the claudin family, components of tight junctions that maintain a physical barrier and polarity of cells. CLDN11 hypermethylation was previously reported in bladder and gastric cancer, where epigenetic silencing increased cell motility and invasiveness (
). Interestingly, in mouse skin tumorigenesis, changes were found in the distribution pattern of claudin tight-junction proteins, where epidermal expression of claudin proteins including Cldn11 decreased during tumor progression (
). We hypothesize that in melanoma development loss of CLDN11 expression through promoter hypermethylation facilitates invasive behavior by disrupting intercellular cohesion provided by tight-junction structures. In addition, methylation of CDH11, PPP1R3C, and GNMT occurred in more than half of melanomas but rarely in dysplastic nevi. The results for CDH11 are in line with previous studies showing that methylation of this cadherin gene that inhibits tumor growth and metastasis preferentially occurs in lymph node metastases of melanoma patients (
). Tumor-suppressive properties have also been reported for PPP1R3C and GNMT.
Thus far, promoter methylation studies in melanoma often used a limited number of clinical specimens and lacked examination of non-malignant samples, thereby making it impossible to distinguish cancer-specific from tissue-specific methylation events (
). To our knowledge, this is a previously unreported study that makes use of a large series of clinical specimens to show that promoter methylation of several genes, including tumor suppressor genes, is present to a small extent in dysplastic nevi. Using the methylation pattern of five genes, we propose a diagnostic algorithm to distinguish melanoma from benign melanocytic lesions. The presence of CLDN11, CDH11, PPP1R3C, and GNMT methylation in a suspicious melanocytic lesion might be regarded as an indicator of malignancy. Taken together, the findings presented in this study provide insight in the epigenetic changes that occur in melanoma development and can aid in the molecular diagnosis of melanocytic lesions.
Materials and Methods
Patient samples
To prioritize the 12 candidate genes, we analyzed an independent set of fresh-frozen or boonfix-fixed, paraffin-embedded (BFPE) tissues from patients diagnosed with common nevus (n=10), dysplastic nevus (n=20), and primary melanoma (n=15) at Leiden University Medical Center (LUMC), the Netherlands. For confirmation of methylation frequencies and testing of diagnostic discriminatory value of the final five genes, we examined FFPE tissues from patients diagnosed with common nevus (n=62), dysplastic nevus (n=72), primary melanoma (n=101), and metastatic melanoma (n=16) at the Maastricht University Medical Center (MUMC), the Netherlands, and University Hospitals of the University of Leuven (KUL), Belgium (series 1). For validation of the diagnostic algorithm, we examined fresh-frozen, BFPE, and FFPE tissues from patients diagnosed with dysplastic nevus (n=74) and primary melanoma (n=82) at LUMC, MUMC, and KUL (series 2). The grade of atypia (mild and moderate-severe) of dysplastic nevi was determined by an experienced dermatopathologist (VJW), based on criteria formulated by
); biopsy samples contained at least 50% melanocytic cells, as analyzed on hematoxylin and eosin-stained sections. Patient consent for experiments was not required because French laws consider human tissue left over from surgery as discarded material.
DNA isolation and bisulfite conversion
Genomic DNA from fresh-frozen and BFPE tissues was extracted with the Genomic-tip kit (Qiagen, Venlo, The Netherlands) and the RecoverAll Nucleic Acid kit (Ambion, Carlsbad, CA), respectively; DNA from FFPE tissues was extracted by macrodissection with the QIAamp DNA Micro Kit (Qiagen) or with the Maxwell 16 FFPE Plus LEV DNA Purification kit (Promega, Leiden, The Netherlands). Bisulfite conversion was performed using either the EZ DNA methylation kit (Zymo Research, Orange, CA; BMCA) or the EpiTect Bisulfite Kit (Qiagen; MSP analysis).
Bisulfite melting curve analysis
Bisulfite primers were designed, and the sensitivity of primer sets was tested as previously described (
; Supplementary Table S8 online). Methylation could be accurately detected if 10% of the analyzed DNA was methylated. Melting curves were generated for each biopsy sample; a sample was considered methylated if the amplicon had at least 10% methylated DNA, with the melting curve pattern of a 1:9 methylated to unmethylated DNA mixture serving as scoring standard.
Methylation-specific PCR
MSP analysis using MSP primers on bisulfite-treated DNA was performed as described (
; Supplementary Table S8 online). To facilitate MSP analysis on DNA retrieved from FFPE tissue, DNA was first amplified with flanking PCR primers used as template for the PCR. All PCRs were performed with controls for unmethylated alleles (DNA from human umbilical vein endothelial cells, methylated alleles (normal lymphocyte DNA treated in vitro with Sssl methyltransferase (IVD)), and a control without DNA. To ensure reproducibility, MSP reactions have been performed in duplicate or triplicate starting from DNA amplification with flanking PCR primers. The reproducibility was >93% for all primer sets. Nonconcordant MSP results were analyzed a third time, and two out of three concordance was used as the end result. Bands with approximately equal intensity for unmethylated and methylated DNA were scored as positive for methylation. Faint methylated bands were considered negative for methylation. Unclear results were analyzed a second time. In most cases, methylation levels were clearly negative (no M-band detected) or positive (strong M-band detected).
Statistical analysis
Fisher’s exact test was applied to measure the association between two sample groups within a 2 × 2 contingency table; a two-sided P-value <0.05 was considered significant. Logistic regression analysis was used to test the diagnostic value of the five genes in the training data set. The logistic regression did not converge because of the strong diagnostic effect of CLDN11; therefore, we opted for a two-step model: data were filtered for samples without CLDN11 methylation, followed by binary logistic regression analysis with a single covariate counting how many of CDH11, PPP1R3C, MAPK13, and GNMT were methylated. In the resulting diagnostic score, a patient sample was classified as melanoma if either CLDN11 was methylated or if at least 1, 2, 3, or 4 of CDH11, PPP1R3C, MAPK13, and GNMT were methylated, depending on the chosen cutoff. A receiver operating characteristic curve was plotted for this diagnostic score. For this assessment, an independent test set was used in order to prevent any optimism bias due to overfit. Statistical analyses were performed using SPSS 20 (IBM, Armonk, NY) and R (R Core Team (2013)—http://R-project.org/).
Acknowledgments
RvD is supported by a Melanoma Research Alliance young investigator award. This work was supported by Profileringsfonds Maastricht University Medical Center (PF=278). We thank Kathleen Daenen and Kim van Straeten for excellent technical assistance.
The human cadherin 11 is a pro-apoptotic tumor suppressor modulating cell stemness through Wnt/beta-catenin signaling and silenced in common carcinomas.
36580-5&id=gr1.jpg){kind=link}
36580-5&id=gr2.jpg){kind=link}
36580-5&id=gr3.jpg){kind=link}
36580-5&id=gr4.jpg){kind=link}