DNA methylation and gene deletion analysis of brain metastases in melanoma patients identifies mutually exclusive molecular alterations

Abstract

Background

The brain is a common target of metastases for melanoma patients. Little is known about the genetic and epigenetic alterations in melanoma brain metastases (MBMs). Unraveling these molecular alterations is a key step in understanding their aggressive nature and identifying novel therapeutic targets.

Methods

Genome-wide DNA methylation analyses of MBMs (n = 15) and normal brain tissues (n = 91) and simultaneous multigene DNA methylation and gene deletion analyses of metastatic melanoma tissues (99 MBMs and 43 extracranial metastases) were performed. BRAF and NRAS mutations were evaluated in MBMs by targeted sequencing.

Results

MBMs showed significant epigenetic heterogeneity. RARB, RASSF1, ESR1, APC, PTEN, and CDH13 genes were frequently hypermethylated. Deletions were frequently detected in the CDKN2A/B locus. Of MBMs, 46.1% and 28.8% had BRAF and NRAS missense mutations, respectively. Compared with lung and liver metastases, MBMs exhibited higher frequency of CDH13 hypermethylation and CDKN2A/B locus deletion. Mutual exclusivity between hypermethylated genes and CDKN2A/B locus deletion identified 2 clinically relevant molecular subtypes of MBMs. CDKN2A/B deletions were associated with multiple MBMs and frequently hypermethylated genes with shorter time to brain metastasis.

Conclusions

Melanoma cells that colonize the brain harbor numerous genetically and epigenetically altered genes. This study presents an integrated genomic and epigenomic analysis that reveals MBM-specific molecular alterations and mutually exclusive molecular subtypes.

Although melanoma brain metastases (MBMs) are detected in 40% of advanced stage patients antemortem, autopsy reports reveal that up to 75% of melanoma patients develop MBMs.¹ It has been suggested that this high frequency of MBM may be due to a “homing” influence, since melanoma cells (seeds) and the brain (soil) are both derived from the neural crest.^2,3 Once melanoma cells colonize the brain, a rapid decline in patient quality of life is typically observed, and median survival is only 4–6 months after diagnosis.^2,4 Recent advances in targeted systemic therapy agents such as ipilimumab, vemurafenib, and dabrafenib, combined with surgical resection of large metastases and stereotactic radiosurgery, show potential for improving the survival of melanoma patients with MBM.^5,6 Additionally, a number of prognostic indices have been validated in patients with brain metastasis,⁷ including the number and location of metastatic lesions as strong predictors of patient survival.^8,9 However, to effectively implement these therapeutic strategies and further improve melanoma patient management, there is an urgent need to understand the molecular alterations implicated in MBM.

Gene deletion and hypermethylation are well-established molecular alterations involved in cancer progression leading to metastasis.^10,11 The functional impact of these mechanisms is often the loss or downregulation of normal gene expression. When these alterations affect tumor-related genes (TRGs), cells lose key cellular controls. Advances in high-throughput genomic technologies have contributed to the identification of TRGs altered in primaries and metastatic melanoma.¹² However, specific genetic and epigenetic alterations of MBM have only recently begun to be described. Our group and collaborators have identified a comprehensive list of genetically and epigenetically affected TRGs in MBM¹³ and its microenvironment.^14,15 In this study, we expanded on these discoveries by characterizing genetic and epigenetic alterations of TRGs (and their relationship) on MBM.

We aimed to identify MBM-specific gene deletions and hypermethylation patterns that could discriminate clinically relevant prognostic indices and enhance understanding of molecular alterations contributing to the colonization of the brain by melanoma cells.

Materials and Methods

Patients, Tissue Samples, Cell Lines, and DNA Isolation

Paraffin-embedded archival tissue (PEAT) sections of 142 metastatic melanomas (11 liver metastases [MLvMs], 32 lung metastases [MLuMs], and 99 MBMs [Table 1]) were included in the study under protocols approved by the joint institutional review board of the Saint John's Health Center/John Wayne Cancer Institute, the Western Institutional Review Board, and the Sydney Local Health District (Royal Prince Alfred Hospital Zone) Human Ethics Review Committee. The experiments were performed in accordance with the World Medical Association Declaration of Helsinki and the National Institutes of Health Belmont Report. Tissues were de-identified and coded according to recommendations of the Health Insurance Portability and Accountability Act to ensure the confidentiality of the patients. Each melanoma tissue was embedded in 10% neutral buffered formalin at least 72 h prior to histologic sectioning. Genomic DNA was isolated using ZR FFPE DNA MiniPrep (Zymo Research, Irvine, CA) after laser capture or needle microdissection from 8-μm-thick PEAT sections as we described.¹³ Three adult melanocyte cultures (MNCs; Life Technologies) and 5 melanoma lines established from MBM tissues at the John Wayne Cancer Institute were included in this study. Genomic DNA from cell lines was extracted using DNAzol reagent (Life Technologies) as we described.¹³

Table 1.

Metastases from cutaneous melanoma patients

Characteristics	Brain (N = 99)		Other Organs (N = 43)
	n Patients	(%)	n Patients	(%)
Gender
Male	63	64	29	67
Female	36	36	14	33
Age at metastasis diagnostic, y
Median	59		56
Range	25–85		23–79
Primary melanoma location
Head/neck	12	12	7	16
Trunk/abdomen	29	29	18	42
Limbs	31	31	15	35
Occult	7	7		0
Unknown	10	10		0
Primary melanoma histology
Desmoplastic	7	7	0	0
Desmoplastic with neurotropia	5	5	0	0
Lentigo maligna melanoma	6	6	1	2
Nodular melanoma	31	31	15	35
Superficial spreading	26	26	10	23
Superficial spreading with nodular melanoma	7	7	1	2
Occult	7	7	5	12
Unknown	10	10	11	26
No. of cerebral metastases
Single	59	60	NA
Multiple (≥2)	40	40	NA
Location of cerebral metastases
Cerebral cortex
Frontal lobe	36	36	NA
Parietal lobe	21	21	NA
Temporal lobe	14	14	NA
Occipital lobe	7	7	NA
Frontoparietal	5	5	NA
Parietooccipital	4	4	NA
Frontotemporal	2	2	NA
Temperoparietal	2	2	NA
Dura mater	1	1	NA
Cerebellum	7	7	NA

Genome-wide DNA Methylation Profiling

Fourteen MBMs resected from the frontal lobe were analyzed using Human Methylation 450K beadchip arrays (Illumina) as we described.¹³ From these arrays were obtained data from 91 postmortem normal brain tissues generated by the Institute of Psychiatry at the Medical Research Council Social, Genetic and Developmental Psychiatry Centre, the King's College of London¹⁶ (Gene Expression Omnibus; GSE43414).

Methylation-specific Multiplex Ligation Probe Amplification

Gene deletion and hypermethylation of 24 TRGs (Table 2) were evaluated using the methylation-specific multiplex ligation probe amplification mix ME001 (MRC-Holland). The procedure was performed as we described.^17–19 To evaluate DNA methylation level, we applied HhaI (Promega), a methylation-sensitive restriction enzyme. The restriction-resistant probes were amplified by PCR, and the fragments were separated by capillary array electrophoresis in a CEQ8000 DNA sequencer (Beckman Coulter). The area of each probe was normalized by dividing it by the average area of the control probes. The methylation ratio (MR) was calculated by dividing the normalized area value of the digested sample by the normalized area value of the undigested sample. Based on the analysis of DNA extracted from melanoma cell lines, a conservative threshold of MR ≥ 0.2 was considered “methylated.” To evaluate copy number (CN) ratio, the normalized area value of the undigested sample was divided by the normalized area value of the undigested sample of a normal PEAT control. Based on the analysis of DNA extracted from melanoma cell lines and normal peripheral blood leukocytes, a conservative threshold of CN ≤ 0.6 was considered “deleted” (Supplementary Fig. S1).

Table 2.

TRGs included in the multigene DNA methylation and CN analysis

#	Symbol	Gene Name	Chr. pos.
1	APC	Adenomatous polyposis coli	5q21
2	ATM	Ataxia telangiectasia mutated	11q22
3	BRCA1	Breast cancer 1, early onset	17q21
4	BRCA2	Breast cancer 2, early onset	13q12.3
5	CASP8	Caspase 8	2q33
6	CD44	CD44 molecule (Indian blood group)	11p13
7	CDH13	Cadherin 13, H-cadherin (heart)	16q24.2
8	CDKN1B	Cyclin-dependent kinase inhibitor 1B (p27, Kip1)	12p13.1
9	CDKN2A	Cyclin-dependent kinase inhibitor 2A (melanoma, p16, inhibits CDK4)	9p21
10	CDKN2B	Cyclin-dependent kinase inhibitor 2B (p15, inhibits CDK4)	9p21
11	CHFR	Checkpoint with forkhead and ring finger domains	12q24.33
12	DAPK1	Death-associated protein kinase 1	9q34.1
13	ESR1	Estrogen receptor 1	6q25.1
14	FHIT	Fragile histidine triad gene	3p14.2
15	GSTP1	Glutathione S-transferase pi 1	11q13
16	HIC1	Hypermethylated in cancer 1	17p13.3
17	CADM1	Cell adhesion molecule 1	11q23.2
18	MLH1	mutL homolog 1, colon cancer, nonpolyposis type 2 (E. coli)	3p21.3
19	PTEN	Phosphatase and tensin homolog	10q23.3
20	RARB	Retinoic acid receptor, beta	3p24
21	RASSF1	Ras association (RalGDS/AF-6) domain family member 1	3p21.3
22	TIMP3	TIMP metallopeptidase inhibitor 3	22q12.1
23	TP73	Tumor protein p73	1p36.3
24	VHL	von Hippel–Lindau tumor suppressor	3p26–p25

Sequencing of BRAF and NRAS

MBM genomic DNA was amplified with primers specific for exons 1 and 2 of NRAS and exon 15 of BRAF (Supplementary Table S1). After amplification, PCR products were purified using the QIAquick PCR Purification Kit (Qiagen) and subsequently visualized in a 2% agarose gel. Successfully amplified samples were then quantified by UV absorption spectrophotometry and sequenced using an internal primer (Supplementary Table S1) by Eurofins MWG Operon. Sequencing results were analyzed using Chromas Lite v2.1.1 (Technelysium), and missense mutations in NRAS and BRAF were annotated.

Quantitative Real-time PCR Amplification

Total RNA was extracted and quantified as we described.¹³ Real-time reaction was performed as we described.¹³ We performed quantitative cDNA amplification of cyclin-dependent kinase inhibitor 2A (CDKN2A), phosphatase and tensin homolog (PTEN), retinoic acid receptor beta (RARB), and Ras association (RalGDS/AF-6) domain family member 1 (RASSF1), and the normalized expression was calculated by the ΔΔ–comparative quantitation method using β2-microglobulin expression as a reference as we described.¹³ The sequence of primers and probes used in this study are listed in Supplementary Table S1.

Demethylating Agent Treatment

The treatment was performed as we described.¹³ Briefly, melanoma cell lines were incubated in Roswell Park Memorial Institute (RPMI) medium containing 10% fetal bovine serum and 2.5 μM 5-aza-2-deoxycytidine (AZA) for 72 h. For nontreated controls, melanoma cell lines were incubated in RPMI medium containing 10% fetal bovine serum and 0.1% dimethyl sulfoxide.

Protein Extraction and Western Blot Analysis

Protein extraction and western blots were performed as previously described.¹³ Membranes were immunoblotted with primary antibodies against RARB (1:1000; Abcam #ab53161), RASSF1 (1:1000; eBioscience #14-6888-80), and β-actin (1:5000; Sigma-Aldrich #A1978-200UL) overnight at 4°C, followed by peroxidase-conjugated secondary antibody against rabbit IgG for RARB (1:4000; GE Healthcare) and against mouse IgG for RASSF1 and β-actin (1:6000; GE Healthcare). Detection was carried out using the Pierce SuperSignal West Femto Maximum Sensitivity chemiluminescent substrate followed by electrochemiluminescent scanning using MyECL Imager.

Biostatistical Analysis

Cytosine–phosphate–guanine (CpG) differentially methylated sites (DMSs) between MBM and MNC were analyzed using the nonparametric Wilcoxon rank-sum test (W). Multiple comparison correction was performed by the Benjamini–Hochberg false discovery rate procedure (FDR-corrected). Unsupervised hierarchical cluster, principal component analysis (PCA), and terrain maps were generated using MultiExperiment Viewer v4.9 (Dana-Farber Cancer Institute)²⁰ as we described.¹³ The stability of the phenetic tree topology was assessed by bootstrap (Bt) resampling using 1000 iterations. Correlation among DNA methylation profiles was evaluated by Spearman's rho coefficient. Binary DNA methylation and gene deletion differences between groups were compared using Pearson's χ² test. Kaplan–Meier curve with a log-rank significance test and Cox proportional hazard models were used to compare the time to brain metastasis (TTBM) between groups and estimate hazard ratios adjusted for potential confounders (including gender, location, melanoma subtype, age, tumor/lymph node/metastasis stages at the primary melanoma resection, pre-MBM diagnostic treatment, presence of extracranial metastasis at MBM resection, location, number, and size of resected MBM).

Results

Differentially Methylated Tumor-related Genes in Frontal Lobe Melanoma Brain Metastases

To identify aberrantly methylated TRGs in MBMs, the methylation level of 5124 CpG sites located in 180 TRGs (Supplementary Table S2) was analyzed in a cohort of MBMs resected from the frontal cortex and compared with MNCs. We identified 938 DMSs located in 141 TRGs (W; FDR-corrected P > .05). The methylation level of these DMSs significantly separated MNCs and MBMs into 2 clusters (Bt = 100; Fig. 1A). In MBMs, 548 DMSs were hypomethylated and 390 DMSs were hypermethylated (Supplementary Tables S3 and S4). Interestingly, hypomethylated DMSs were frequently located at gene bodies and hypermethylated DMSs at promoter regions of TRGs (Fig. 1B). This DNA methylation signature has been identified in transcriptionally inactive alleles.^21,22 Promoter hypermethylation and gene body hypomethylation were identified in 46 TRGs (Supplementary Table S5), including previously reported melanoma-related genes such as RARB, ESR1, RUNX3, and RASSF1, classically known as tumor suppressor genes.^23–26 Aberrantly methylated TRGs were evaluated in specimens representing different stages of melanoma progression to MBM, as we published.¹³ The value of the most variable probe for TRGs was averaged for MNCs, primary melanoma, lymph node metastasis, and MBM. ESR1, RARB, RASSF1, and RUNX3, among other genes, presented a progressive increment in DNA methylation level, while BCL2, FGR2, and SLC6A11, among other genes, presented a progressive decrease in DNA methylation from MNCs to MBM (Supplementary Fig. S2).

Fig. 1.

DNA methylation and CN analysis of MBMs. (A) Hierarchical cluster analysis of genome-wide DNA methylation data from MBMs (gray bar) and MNCs (green bar). (B) Distribution of MBM hypomethylated (blue) and hypermethylated (red) CpG sites according to gene structures. (C) Histogram of the frequency of correlation coefficient (rho) among MBM and normal brain tissues. (D) Frequency of hypermethylated genes in MBMs. (E) Frequency of deleted genes in MBMs. Error bars for D and E represent the standard error of proportion. (F) Correlation matrix representing Spearman's rho coefficient among TRGs frequently altered in MBMs.

To evaluate interpatient variation in TRG methylation levels, MBM DMSs were analyzed in normal frontal lobe tissues. A significantly higher correlation in TRG methylation levels was observed among normal brain tissues than among MBMs (W; P = 6.0⁻³²; Fig. 1C). This finding indicates the existence of interpatient heterogeneity in the methylation level of TRGs, even in MBMs resected from the same lobe. The higher heterogeneity of MBM processes may have implications for behavioral differences pertaining to metastatic tumor progression.

Aberrantly Methylated Tumor-related Genes in a Large Cohort of Melanoma Brain Metastases

The hypermethylation of classical tumor suppressor genes and the higher interpatient DNA methylation heterogeneity led to an expansion of analysis to include MBMs resected from other operable brain regions. To build upon these findings, a large clinically well-annotated cohort of MBMs (n = 99; Table 1) was evaluated by targeted multigene analysis (Table 2). This analysis was standardized to simultaneously detect gene deletion and hypermethylation of 24 TRGs (Supplementary Fig. S1). Using a conservative threshold (MR ≥ 0.2), aberrant methylation was detected in at least one TRG in 72/99 MBMs (mean = 3.6 genes/MBM). Hierarchical cluster analysis stratified independent clusters of MBMs, suggesting the existence of diverse DNA methylation profiles. One cluster containing 11 MBMs was significantly enriched in hypermethylated genes, reflecting a differential incidence of aberrant methylation (Supplementary Fig. S3A). The most frequently methylated TRGs were RARB, ESR1, RASSF1, PTEN, APC, and CDH13 (Fig. 1D). We have previously demonstrated that hypermethylation of these genes plays a significant role in the initiation and progression of early stage melanomas,^26–28 thereby indicating that these genes continue to play an important role throughout metastatic progression to MBM.

Frequent Deletions of CDKN2A and CDKN2B in Melanoma Brain Metastases

Copy number analysis of the 24 TRGs revealed that gene deletions occurred significantly less frequently than aberrant methylation (W; P = 4.1⁻¹³). Only 49 MBMs presented deletion in at least one TRG (mean = 1.2 genes/MBM). Hierarchical cluster analysis identified one cluster containing 29 MBMs with a high incidence of gene deletions (Supplementary Fig. S3B). Overall, gene deletions almost exclusively affected CDKN2A and CDKN2B, making up over 42% of the deletions detected (Fig. 1E). These genes are encoded in the same chromosomal subband (9p21.3) and are reportedly altered in primary and metastatic melanomas.^29–32 In contrast, other TRGs presented a substantially lower frequency of deletion, none of which reached >7% occurrence. Remarkably, a significantly higher frequency of TRG deletions was observed in MBMs also harboring deletions in CDKN2A or CDKN2B (W; P = 2.61⁻⁴), indicating that these 29 MBMs may have a higher influence from CN alterations in the tumor genome than those 11 MBMs previously identified with frequently hypermethylated genes. Altogether, these results suggest the potential existence of distinct molecular alteration pathways in melanoma progression to MBM.

Co-occurrence of Gene Deletion and Aberrant Methylation on Tumor-related Genes in Melanoma Brain Metastases

Multigene analysis allowed the identification of gene deletion and hypermethylation affecting different alleles of the same TRG. Concurrent deletion and hypermethylation of ESR1 was detected in 4 MBMs; of CDKN2A in 2 MBMs; and of APC, CDH13, TIMP3, and TP73 in one MBM each (Supplementary Fig. S3C). This result suggests a significant requirement for biallelic alterations of these key TRGs in melanoma progression to MBM. Consequently, in this study, the detection of aberrant methylation, gene deletion, or both events affecting one specific gene was defined as a molecular alteration.

BRAF and NRAS Mutation Status in Melanoma Brain Metastases

To further investigate alternative molecular alterations preceding MBM, a screening for missense oncogenic mutations on BRAF and NRAS genes was performed. Mutations in BRAF exon 15 were identified in 46.1% of MBMs and mutations in NRAS exons 1 and 2 in 28.8% of MBMs. Amongst BRAF mutations, 82.9% were missense changes of codon 600 (p.V600E, p.V600K, and p.V600R). Amongst NRAS mutations, 63.2% were missense changes of codon 61 (p.Q61R, p.Q61K, and p.Q61L). As previously described,³³ a clear mutual exclusivity between BRAF and NRAS mutations was identified; no MBMs presented concurrent mutations of these genes (Supplementary Fig. S3D). No significant associations were found between DNA methylation, gene deletion, and mutation status of BRAF and NRAS (χ²; P > .05). A weak nonsignificant tendency was observed between BRAF wild type and CDKN2A gene deletion (χ²; P = .09). The mean number of methylated or deleted genes was not significantly related to the mutation status of BRAF and NRAS. Collectively, the data show that BRAF and NRAS mutation status is not significantly associated with deletion or hypermethylation of this panel of TRGs.

Mutually Exclusive Molecular Alterations in Melanoma Brain Metastases

Integrative DNA methylation and CN analysis identified the existence of molecular alterations in all of the analyzed MBMs (mean = 4.7 genes/MBM). Additionally, significant concurrent and divergent molecular alterations among the TRGs were identified. Molecular alterations affecting CDKN2A and CDKN2B were strongly associated, having a significantly high correlation coefficient (rho = 0.68, P = 9.8⁻¹³; Fig. 1F). At the same time, both genes had a significant inverse correlation with APC (rho = −0.25; P = .02). Interestingly, APC showed significant concurrent molecular alterations with other frequently methylated TRGs, such as RARB (rho = 0.31; P = .001) and ESR1 (rho = 0.28; P = .01). Moreover, RARB had significant concurrent molecular alteration with PTEN (rho = 0.24; P = .02), while ESR1 showed significant concurrent molecular alterations with RASSF1 (rho = 0.30; P = .002) and CDH13 (rho = 0.24; P = .02; Fig. 1F). The significant inverse correlation identified between these frequently altered genes—CDKN2A/CDKN2B and APC/RARB/ESR1/PTEN/RASSF1—depicted the existence of 2 mutually exclusive subtypes of molecular alterations in MBM. The coexistence of these 2 molecular subtypes in a melanoma cell does not confer an advantage. The first subtype included the frequently deleted CDKN2A/B genes, while the second subtype contained the frequently methylated APC, CDH13, ESR1, PTEN, RARB, and RASSF1 genes (Supplementary Fig. S4A). Reflecting the correlation among the TRGs, PCA identified a closer relation between CDKN2A/B genes than among the members of the second group, suggesting that the methylated group may be composed of distinct, but related, combinations of hypermethylated genes that produce the similar result of advancing MBM development. PCA indicated that RARB was closer to PTEN, APC was closer to CDH13, and ESR1 was closer to RASSF1 (Supplementary Fig. S4B). The correlation matrix revealed the differences between 2 subgroups. The first subgroup of MBMs had high incidence of CDKN2A/B deletions (n = 27) and the second subgroup of MBMs had high incidence of hypermethylated genes (n = 72; Supplementary Fig. S4C). The landscape of the statistical distances among MBMs depicted by terrain map analysis further supported the existence of these molecular subtypes by identifying a unique cluster of elevations containing MBMs with CDKN2A/B deletions and 3 clusters of elevations containing MBMs with methylation of APC, CDH13, ESR1, PTEN, RARB, and RASSF1 (Supplementary Fig. S4D). These observations suggest that TRGs can be affected by one of the 2 subtypes of molecular alterations (deletion of CDKN2A/B or hypermethylation of APC, CDH13, ESR1, PTEN, RARB, and RASSF1), but the coexistence of molecular alterations from both groups of genes does not represent a selective advantage for MBM cells.

Impact of Aberrant Methylation and Gene Deletion on Tumor-related Gene Expression

To assess the effect of DNA methylation on gene expression, the mRNA and protein expression of the most frequently hypermethylated TRGs was evaluated. Cell lines with RARB hypermethylated presented significantly lower levels of RARB mRNA expression (Fig. 2A). AZA treatment significantly increased RARB mRNA (Fig. 2B) and protein (Fig. 2C) expression. Similarly, cell lines with RASSF1 hypermethylation had a significantly lower level of RASSF1 mRNA expression (Fig. 2D). AZA treatment, in this case, induced a discrete significant increment in RASSF1 mRNA expression (Fig. 2E); however, the protein level increment was not significant (Fig. 2F). These results suggest that DNA methylation of the targeted regions has a relative effect on mRNA and protein expression and that a generalization of the DNA methylation effect cannot be drawn. In addition, to assess the effect of gene deletion on mRNA expression, 3 melanoma cell lines were classified according to their CDKN2A gene CN. FD cell line presented 2 copies (+/+), M20 cell line presented a hemizygous deletion (+/−), and M211 cell line presented a homozygous deletion (−/−; Supplementary Fig. S1G–I). The expression level of CDKN2A mRNA was significantly lower in M20 than in FD (P = .005) and undetectable in M211 (Fig. 2G). Additionally, 3 cell lines were classified according to PTEN gene CN and methylation status. M16 presented normal PTEN CN (+/+), M211 had a homozygous deletion (−/−), and M20 had normal CN with 1 hypermethylated allele (+/Met). Expression of mRNA of this gene was detected in M16 (+/+) but not in M211(−/−), while in M20, PTEN mRNA expression was significantly lower than in M16 (Fig. 2H). After demethylation treatment, the expression of PTEN was not significantly increased in M20 (Fig. 2I). Collectively, these results suggest that gene deletion as well as hypermethylation can affect mRNA and even protein levels; however, the impact on each TRG varies.

Fig. 2.

Impact of gene deletion and hypermethylation on gene expression. (A and D) Quantitative PCR analysis of RARΒ (A) and RASSF1 (D) mRNA in melanoma lines. (B and E) Quantitative PCR analysis of RARΒ (B) and RASSF1 (E) with (+) and without (−) 2.5 μM AZA for 72 h. (C and F) Western blot analysis of RARΒ (C) and RASSF1 (F) with (+) and without (−) 2.5 μM AZA for 72 h. (G, H) Quantitative PCR analysis of CDKN2A (G) and PTEN (H) in melanoma lines harboring diploid CN (+/+), hemizygous deletion (+/−), homozygous deletion (−/−), or diploid CN plus monoallelic methylation (+/Met) of the respective genes. (I) Western blot analysis of PTEN with (+) and without (−) 2.5 μM AZA for 72 h. Error bars represent SEM.

Melanoma Brain Metastasis–specific Patterns of Tumor-related Gene Alterations

To evaluate whether the methylation and/or deletion patterns of this panel of TRGs were representative of MBM, nonbrain melanoma metastases from frequently affected organs, such as the liver (MLvM; n = 11) and the lungs (MLuM; n = 32), were also analyzed. Overall, MBMs demonstrated a higher frequency of molecular alterations of the 24-TRG panel than non-MBMs (W; P = .001; Fig. 3A). The largest difference was observed in the frequency of gene deletions (Fig. 3B and C). CDKN2A/B deletions were more frequently detected in MBMs than in non-MBMs (both χ²; P = .0002 and P = .01, respectively; Fig. 3D and E). Additionally, CDH13 was significantly more frequently hypermethylated in MBMs than in non-MBMs (χ²; P = .01; Fig. 3F). Independent evaluation of each metastatic site indicated that alterations in RASSF1, ESR1, and TIMP3 presented similar frequencies of molecular alterations among these metastatic melanomas. Molecular alterations of APC were significantly more frequent in MBM and MLvM than in MLuM (W; P = .04), while molecular alterations of RARB were significantly more frequent in MBM and MLuM than in MLvM (W; P = .04; Fig. 3G and Supplementary Fig. S5).

Fig. 3.

Differential molecular alterations between MBMs and non-MBMs. (A) Mean number of molecular alterations, (B) number of hypermethylated genes, and (C) number of deleted genes. Error bars for A–C represent SEM. (D) Frequency of CDKN2A and (E) CDKN2B deletions and (F) CDH13 hypermethylation. Error bars for D and F represent the standard error of proportion. (G) Venn diagram summarizing trends and significant (**P < .01 and *P < .05) associations between organ metastases and molecular alterations of TRGs.

Genomic and Epigenomic Alterations of Melanoma Brain Metastases According to Brain Location

Specialized functions of different parcels of the cerebrum are strongly influenced by specific epigenetic programs of both neuron and glia cells.^21,34 The multitude of process specializations of human cortex suggests that distinct lobes in the cerebrum might generate unique microenvironmental conditions for colonizing melanoma cells. To verify this, genome-wide methylation data generated by Pidsley and collaborators¹⁶ (GSE43414) from 91 postmortem normal brain specimens obtained from 4 distinct brain regions were analyzed. Three regions were obtained from the cerebrum (superior frontal gyrus, n = 17; superior temporal gyrus, n = 24; and middle temporal gyrus, n = 29) and one region was obtained from the cerebellum (n = 21). Hierarchical cluster analysis of these 4 tissue types using the 800 most variable CpG sites showed a divergence between cerebellum and cerebrum, separating all the specimens into 2 main clusters. The cerebrum tissue cluster had 3 subclusters, reflecting the 3 different cortex regions (Fig. 4A). Correlation analysis of these methylation profiles revealed high positive correlation among tissues located in the same brain region and a high negative correlation between cerebellum and cerebrum (Fig. 4B). This evidence was used to hypothesize that as a consequence of divergent microenvironments, genetic and epigenetic differences can differentiate MBMs metastasizing distinct regions of the brain. Molecular alteration comparisons of the 24 TRGs between MBMs resected from the cerebellum and the cerebrum identified a significantly higher frequency of molecular alterations of PTEN in MBM from the cerebellum (χ²; P = .03; Fig. 4C and D). TRG molecular alterations among MBMs from different lobes of the cerebrum were also compared. Lesions resected from interlobar sulci were excluded from this analysis in order to remove specimens that were potentially exposed to multiple microenvironmental influences (n = 13; Table 1). Overall, nonsignificant differences in molecular alterations of these TRGs were detected among MBMs from different cortex lobes. A significantly higher frequency of CDKN2A alterations was detected in MBMs resected from the occipital lobe than from the temporal lobe (χ²; P = .04; Fig. 4E and F). In contrast, ESR1 was more frequently altered in metastases resected from the temporal lobe than from the occipital lobe (χ²; P = .06; Fig. 4G and H).

Fig. 4.

Molecular alterations of MBMs according to the brain compartment. (A) Hierarchical cluster analysis of genome-wide DNA methylation data from normal cerebellum (CER), normal superior frontal gyrus (SFG), normal superior temporal gyrus (STG), and normal middle temporal gyrus (MTG) specimens. (B) Correlation matrix depicting Spearman's rho coefficient among the 4 normal brain structures. (C) Frequency of PTEN alterations in cerebellum and cerebrum MBMs. (D) Frequency of PTEN alterations in MBM from cerebellum (Cer), frontal lobe (Fro), temporal lobe (Tem), parietal lobe (Par), and occipital lobe (Occ). (E) Frequency of CDKN2A alterations in occipital and parietal MBMs. (F) Frequency of CDKN2A alterations among MBMs in cortex lobes. (G) Frequency of ESR1 alterations in occipital and parietal MBMs. (H) Frequency of ESR1 alterations among MBMs in cortex lobes. Error bars for C, E, and G represent the standard error of proportion.

Clinicopathological Characteristics of Mutually Exclusive Molecular Subtypes

The number of MBMs represents an important factor influencing patient survival.^8,9 In this study, with a range of 1–21 MBMs/patient, 60% of the patients presented a single clinically detected metastatic lesion (Table 1). MBMs with CDKN2A deletion were associated with multiple lesions (χ²; P = .048; Fig. 5A), while single metastatic processes presented a higher frequency of RASSF1 methylation (χ²; P = .04; Fig. 5B). The number of methylated genes was not associated with the number of MBMs. Another key parameter for the clinical evolution of cancer patients is TTBM.^35,36 To investigate whether the molecular alterations of MBMs were related to the TTBM, 3 groups of MBMs were selected. One group contained MBMs harboring high incidence of CDKN2A/B gene deletions (n = 24), another group contained MBMs harboring high incidence of methylated genes (n = 30), and the last group contained MBMs with an overall lower number of molecular alterations (n = 34; Fig. 5C). Patients with MBMs harboring methylated genes presented a shorter TTBM (log rank; P = .002; Fig. 5D). However, this association was not independent of pre-MBM treatment and pathological features of the primary melanoma and therefore was not statistically significant in multivariate analyses.

Fig. 5.

Association between molecular alterations in MBMs and the number and time to brain metastasis. (A) Frequency of CDKN2A deletion in single (1) and multiple (≥2) lesions. (B) Frequency of RASSF1 hypermethylation in single (1) and multiple (≥2) lesions. Error bars for A–B represent the standard error of proportion. (C) Hierarchical cluster analysis of molecular alterations shown. Shadows in the topological tree indicate the groups compared in the study. (D) Kaplan–Meier curve for TTBM according to the molecular alteration group.

Discussion

MBM is a major cause of death in melanoma patients. To date, little is known about the molecular alterations that allow melanoma cells to metastasize to the brain. In this study, we identified genetic and epigenetic alterations that can help elucidate the molecular mechanisms of melanoma colonizing brain tissue.

Molecular alterations of a large number of genes are involved in cancer progression to metastasis.¹⁰ Furthermore, significant genetic and epigenetic heterogeneity exists among metastatic lesions, even within the same patient.³⁷ It has been reported that the molecular alterations of diverse genes are selected according to microenvironmental variations of the metastatic niche.³⁸ Our analysis identified interpatient DNA methylation heterogeneity, suggesting that numerous alternative altered pathways may contribute to melanoma cells surviving in the brain. To better understand the molecular alterations that play an important role in MBM, our analysis was expanded to a large cohort of metastatic melanoma tissues (n = 142). As our laboratory and others have previously shown, RARB, RASSF1, and ESR1 are frequently methylated genes in primary melanoma, lymph node metastasis, and cell-free DNA.^25,26,28,39 In this current study, we identified that the methylation of these genes, along with PTEN, APC, and CDH13, is also a common epigenetic alteration of MBM.

In concordance with reports in other tumor types,⁴⁰ our data from MBMs indicate that aberrant methylation is a more frequent molecular alteration than is gene deletion. Deletions were almost exclusively detected in the CDKN2A/B locus. This is a frequent event in primary melanoma and metastatic melanomas and has been associated with a shorter survival of melanoma patients.²⁹ CDKN2A deletions have also been associated with a more aggressive behavior of glioblastomas.⁴¹ Combined with our results, this implies that CDKN2A deletions may play an important role in the advancement of tumor cells in the brain microenvironment.

We observed that homozygous deletion of the assessed region on CDKN2A silenced the expression of the gene, while hemizygous deletion led to significantly reduced mRNA expression. Similarly, hypermethylation of RARB substantially reduced mRNA and protein expression. However, for other assessed TRGs, such as RASSF1 and PTEN, the impact of DNA methylation on mRNA and protein expression was relative. This observation, along with current knowledge about the complexity of transcription and translation regulation, suggests that no generalization can be drawn about the impact of aberrant methylation. Future studies should undergo thorough functional validations to expand on these initial findings.

Recent studies on the epigenetics of the human brain indicate a remarkable role for methylation on physiological functions^21,42 such as learning⁴³ and memory,⁴⁴ as well as mental illnesses.^45,46 Our analysis of different structures of normal brain indicates that specific DNA methylation patterns are found in different brain regions. We hypothesize that melanoma cells may require alternative molecular alterations when metastasizing to diverse brain locations. Our analysis found a significantly higher frequency of PTEN methylation in MBMs resected from the cerebellum than from the cortex. However, in order to elaborate solid conclusions, this result has to be verified in a larger cohort of patients with controlled confounding variables.

A significant mutual exclusivity between CDKN2A/B deletions and TRG hypermethylation defined 2 clinically relevant molecular subtypes. A recent report in more than 100 melanoma cell lines identified a mutual exclusion between RARB hypermethylation and molecular alterations of CDKN2A.³² Our results strongly support the previous study in MBM tissues and expand the spectrum of hypermethylated genes to other TRGs, such as APC, ESR1, PTEN, RASSF1, and CDH13. Importantly, our data indicate a good translation of findings from cell lines to actual tissue samples. These mutually exclusive subtypes presented clinicopathological correlations. MBMs with frequent TRG hypermethylation had a shorter TTBM, while MBMs with CDKN2A/B locus deletion were associated with multiple metastases. Our hypothesis about these observations is that melanoma cells with frequent hypermethylated genes have greater plasticity to adapt to diverse microenvironments. This capability may be important for establishment and proliferation of MBM. However, melanomas with key genetic alterations, such as CDKN2A deletion may require a longer period of time to acquire, establish, and proliferate in the brain tissue. Yet, once this secondary lesion is established, it may be more likely to develop multiple foci. Separate studies testing this hypothesis should be carried out.

This study provides the basis for unraveling molecular alterations and their relations coexisting in melanoma cells that colonized the brain. To better comprehend their consequences, detailed functional experiments should be carried out in subsequent studies.

Supplementary Material

Supplementary material is available online at Neuro-Oncology (http://neuro-oncology.oxfordjournals.org/).

Funding

This work was supported by the National Institutes of Health; the National Cancer Institute (grant no. PO1 CA029605 Project II and Core C to D.H.; RO1 CA167967 to D.H.); the Dr. Miriam and Sheldon G. Adelson Medical Research Foundation (D.H.); and the Ruth and Martin H. Weil Foundation (D.H.).

Conflict of interest statement. None declared.