Integrated Genomic Classification of Melanocytic Tumors of the Central Nervous System Using Mutation Analysis, Copy Number Alterations, and DNA Methylation Profiling Free

Melanocytic tumors involving the central nervous system can be diagnostically challenging. Melanocytic tumors can arise in the central nervous system, however, more common are central nervous system melanoma metastases. Correct classification of melanocytic tumors of the central nervous system is important for prediction of their likely clinical behavior and selecting appropriate treatment.

The majority of primary melanocytic tumors of the central nervous system presumably arise from leptomeningeal melanocytes. They are termed primary leptomeningeal melanocytic tumors, (1) and are classified according to histopathologic criteria (2), which include mitotic activity, parenchymal infiltration, and Ki67/MIB1 staining. Applying these criteria, primary leptomeningeal melanocytic tumors are classified as melanocytomas, intermediate-grade melanocytomas, and primary leptomeningeal melanomas, referring to benign, atypical/borderline, and malignant tumors, respectively. However, histopathologic classification of primary leptomeningeal melanocytic tumors is not always straightforward. For example, in some primary leptomeningeal melanocytic tumors, histopathologic criteria overlap and do not permit ready classification. Others show aggressive biological behavior despite lacking obvious high-grade histopathologic features.

Primary leptomeningeal melanocytic tumors and uveal melanomas share a similar gene mutation profile (3–9). They frequently harbor mutually exclusive activating hot-spot mutations in either GNAQ, GNA11, PLCB4, or CYSLTR2 (3–9). Additional mutations can also be found in EIF1AX, SF3B1, and BAP1, which are also generally detected in a mutually exclusive pattern (10–12).

Cutaneous melanomas consistently demonstrate a very high mutational load and a characteristic UV-exposure mutation signature (13–16). A genetic classification into four groups based on mutations activating the MAP kinase pathway has been proposed: BRAF-mutated, RAS-mutated, NF1-mutated, or triple wild-type (15, 16). Furthermore, TERT promoter mutations are highly recurrent in cutaneous melanomas (17, 18).

A rare subset of cutaneous melanomas arising in the dermis is termed blue nevus–like melanoma. These tumors lack BRAF, RAS, NF1, or TERT promoter mutations, instead demonstrating genetic alterations similar to those in primary leptomeningeal tumors and uveal melanomas (19–25).

The aim of our study was to explore molecular signatures of primary leptomeningeal melanocytic tumors that would allow classification into prognostically distinct subgroups.

A cohort of 47 primary leptomeningeal tumors comprising 19 melanocytomas, 22 intermediate-grade melanocytomas, and 6 primary leptomeningeal melanomas (including two retrobulbar melanomas) was analyzed. The control group consisted of 17 central nervous system metastases (from 2 uveal melanomas, 2 melanomas of unknown primary, and 13 cutaneous melanomas); 14 primary uveal melanomas; and 2 blue nevus–like melanomas. Each tumor was obtained from a unique patient. In all primary leptomeningeal melanocytic tumors, a primary melanoma of other site, including a uveal melanoma, had been excluded (clinical history, imaging, and frequently fundoscopy). The samples were retrieved from the Institute of Neuropathology and the Department of Dermatology, Essen, Germany, the Department of Neuropathology, Heidelberg, Germany, the Institute of Neuropathology Bonn, Germany, as well as the Melanoma Institute Australia, Sydney, Australia. Some of these tumors have been previously reported (8, 9, 11, 12, 26), but these reports did not include the detailed genetic analyses reported herein. Tumor slides were reviewed by at least two experienced histopathologists (C. Koelsche, J.A.P. van de Nes, M. Gessi, T. Pietsch, K.G. Griewank, R.A. Scolyer, A. von Deimling, or M.E. Buckland). The study was performed in accordance with the guidelines set forth by the ethics committee of the University of Heidelberg and Duisburg-Essen under the IRB protocol numbers 180073 and 16-6951-BO, respectively.

Histopathologic examination was performed on routinely stained hematoxylin and eosin slides. Primary leptomeningeal tumors were diagnosed based on criteria described by Brat and colleagues (2). Well-differentiated primary leptomeningeal tumors with no or very low mitotic activity (0–1 mitoses per 10 high-power fields), devoid of central nervous system parenchymal infiltration, and Ki67/MIB1 index ≤2% were diagnosed as melanocytomas (n = 19). Tumors characterized by increased mitotic activity (Ki67/MIB1 index 1%–4%) and/or microscopic central nervous system parenchymal invasion, but not sufficiently anaplastic to warrant the designation of melanoma were diagnosed as intermediate-grade melanocytomas (n = 22). Tumors demonstrating higher mitotic activity and anaplasia were diagnosed as primary leptomeningeal melanomas (n = 4). Two tumors diagnosed as retrobulbar melanomas were also included in the primary leptomeningeal melanoma group.

In most cases, IHC was performed on a Ventana Benchmark XT Autostainer using the following markers: S-100 (1:5,000, Dako; Z0311); Melan-A (1:100, Dako; M7196); HMB-45 (1:200, Dako; M0634); BAP1 [as previously described (26), applying a BAP1 rabbit polyclonal antibody raised against a synthetic peptide corresponding to amino acids 430-729 of the BAP1 molecule (clone C-4; Santa Cruz Biotechnology Inc.)]; and the proliferation marker Ki67/MIB1 (1:200, Zytomed; MSK0810). BAP1 expression was scored as positive or negative according to the presence or absence of nuclear staining. The percentage of tumor cells staining with Ki-67/MIB1 was designated as the Ki-67/MIB1 proliferation index.

Ten-μm-thick sections were cut from formalin-fixed, paraffin-embedded tumor tissues. The sections were deparaffinized and manually microdissected according to standard procedures. Genomic DNA was isolated using the QIAamp DNA Mini Kit (Qiagen) according to the manufacturer's instructions.

Two previously published (11, 19) custom amplicon-based sequencing panels were used, one covering 10 genes recurrently mutated in uveal melanoma (Supplementary Table S1), and the other covering 29 genes recurrently mutated in cutaneous melanoma (Supplementary Table S2). Both sequencing panels were applied using the GeneRead Library Prep Kit from QIAGEN according to the manufacturer's instructions. Sequencing was performed on an Illumina MiSeq next-generation sequencer. Adapter ligation and barcoding was performed applying the NEBNext Ultra DNA Library Prep Mastermix Set and NEBNext Multiplex Oligos for Illumina from New England Biolabs.

Sequencing analysis was performed applying CLC Cancer Research Workbench from QIAGEN. The analysis workflow described in brief included adapter trimming and read pair merging before mapping to the human reference genome (hg19). Detection of insertions and deletions as well as single nucleotide variant followed. Additional information regarding potential mutation type, known single nucleotide polymorphisms and conservation scores was obtained by cross-referencing various databases (COSMIC, ClinVar, dbSNP, 1000 Genomes Project, HAPMAP, and PhastCons-Conservation_scores_hg19).

The sequencing results were of high quality, with an average coverage of 13,503-fold for all samples analyzed and a minimum of 30-fold coverage achieved in 98% of the targeted sequence. Mutations were reported if the overall coverage of the mutation site was ≥30 reads, ≥10 reads reported the mutated variant and the frequency of mutated reads was ≥7%. All genetic variants identified are presented in Fig. 1.

Copy number and methylation analysis required 500 ng of isolated DNA and was performed on 91% (n = 43 of 47) primary leptomeningeal melanocytic tumors and 100% of blue nevus–like melanoma, cutaneous melanoma, and uveal melanoma samples (n = 2, 15, and 16, respectively). The primary leptomeningeal melanocytic tumor samples in which these analyses were not performed (due to small sample sizes and insufficient DNA) included two melanocytomas, one intermediate-grade melanocytoma and one retrobulbar melanoma (Fig. 1). The Illumina Infinium 450 k or EPIC array was used to obtain genome-wide methylation data (Illumina), according to the manufacturer's instructions. DNA-methylation data were normalized by performing background correction and dye bias correction. Filtering of probes, that is, removal of probes containing single nucleotide polymorphism and not uniquely matching, was performed as described previously (8). For unsupervised hierarchical clustering, the 10,000 most variably methylated probes (SD > 0.25) across the dataset were selected. Euclidean distance and average linkage was applied for ordering of the probes (y-axis) and 1–Pearson correlation as distance measure and average linkage was used for clustering samples (x-axis).The copy number profile was generated from the array data using the “conumee” R package in Bioconductor (http://www.bioconductor.org/packages/release/bioc/html/conumee.html).

Overall, 39 (45%) tumors occurred in females and 46 (52%) in males; in two cases, the sex was unknown. The median age was 56 years and ranged from 15 to 87 years. The primary leptomeningeal tumor cohort consisted of 47 cases (19 melanocytomas, 22 intermediate-grade melanocytomas, and 6 melanomas). Clinical details are listed in Table 1.

All cases were screened by a targeted amplicon-based next-generation sequencing (NGS) panel covering genes known to be recurrently mutated in uveal melanoma, primary leptomeningeal tumors, and blue nevus–like melanoma (Supplementary Table S2). In 47 primary leptomeningeal tumors, 94% (44/47) tumors harbored activating hotspot mutations in GNAQ (n = 26, 55%), GNA11 (n = 13, 28%), CYSLTR2 (n = 3, 6%), and PLCB4 (n = 2, 4%). All of these mutations were mutually exclusive (Fig. 1; Table 1; Supplementary Table S3). Co-occurring mutations in EIF1AX, SF3B1, and BAP1 were detected in 38% (n = 18), 21% (n = 10), and 6% (n = 3) of tumors, respectively. Concurrent EIF1AX and SF3B1 mutations were identified in three primary leptomeningeal tumors.

In the 16 uveal melanoma samples, 75% (n = 12) had GNAQ and 25% (n = 4) had GNA11 mutations. Additional mutations in BAP1, SF3B1, and EIF1AX were identified in 31% (n = 5), 25% (n = 4), and 13% (n = 2) of samples, respectively.

In the blue nevus–like melanomas available, a GNAQ, a GNA11, and two SF3B1 mutations were identified, as previously reported (20).

Metastases from cutaneous melanomas and melanomas of unknown primary (n = 15) harbored activating mutations in genes of the MAPK pathway in 93% (n = 14) of samples, including BRAF V600 (47%, n = 7), NRAS (33%, n = 5), HRAS, and KIT mutations (each 7%, n = 1). In addition, 80% (n = 12) of tumors harbored TERT promoter hot-spot mutations.

Loss of nuclear BAP1 expression is an adverse prognostic marker in uveal melanoma. Based on the known mutational overlap between primary leptomeningeal tumors and uveal melanomas, we examined BAP1 expression by IHC in cases with sufficient material available. Two of 42 primary leptomeningeal tumors (one melanoma and one intermediate-grade melanocytoma) showed loss of nuclear BAP1 expression. In comparison, 53% (8/15) uveal melanomas demonstrated nuclear BAP1 loss, whereas BAP1 expression was retained in all eight metastases from cutaneous melanomas that were analyzed (Fig. 1).

Copy number analysis was performed on 76 samples from 80 tumors: 43 primary leptomeningeal tumors, 14 primary uveal melanomas, 17 melanoma metastases (from two melanomas of unknown primary, two uveal melanomas and 13 cutaneous melanomas), and two blue nevus–like melanomas (Fig. 1).

Primary leptomeningeal tumors, uveal melanomas, and blue nevus–like melanomas showed similar copy number aberrations. These frequently involved gains of chromosome arm 8q and 6p, monosomy of chromosome 3, and loss of chromosome arm 1p and 6q (Figs. 1–3). Copy number profiles of cutaneous melanomas differed from those of primary leptomeningeal tumors, uveal melanomas, and blue nevus–like melanomas (Figs. 1–3). Cutaneous melanomas demonstrated similar 6p gains and 6q losses, however less frequent chromosome 3 and chromosome arm 1p losses. In addition, cutaneous melanomas frequently carried copy number alterations encompassing a loss of CDKN2A on chromosome arm 9p, a loss of PTEN on chromosome 10 and frequent gains of chromosome 7 involving the BRAF locus, similar to previous reports on cutaneous melanomas (27).

Unsupervised hierarchical clustering using the 10,000 most variably methylated CpG probes clearly separates the study cohort into two main clusters. One cluster includes all cutaneous melanomas. The second cluster is composed of primary leptomeningeal tumors and uveal melanomas and blue nevus–like melanomas, which are indistinguishable at the DNA methylation level. However, tumors with chromosome 3 monosomy, loss of BAP1 protein expression, and/or BAP1 mutations formed a distinct subgroup within this methylation cluster. In contrast, tumors not demonstrating these alterations and frequently harboring EIF1AX and/or SF3B1 mutations segregated together.

To the best of our knowledge, this study represents the most comprehensive genetic analysis of primary leptomeningeal tumors to date. It also represents the first detailed direct genetic comparison of primary leptomeningeal tumors with the genetically similar uveal melanomas and blue nevus–like melanomas. The presented mutation and methylation profiles allowed a clear distinction of primary leptomeningeal tumors from cutaneous melanoma metastases. Our analysis identified tumors clustering according to chromosome 3 and BAP1 status which may be of prognostic relevance.

In our cohort of 47 primary leptomeningeal tumors, 94% (n = 44) of tumors harbored mutually exclusive activating mutations in GNAQ, GNA11, PLCB4, or CYSLTR2 as previously reported (9, 28). These gene mutations were not identified in the analyzed metastases from cutaneous melanoma, consistent with the literature (9, 20). In addition to an activating mutation in one of the four above-mentioned genes, primary leptomeningeal tumors also relatively frequently harbored additional EIF1AX (38%) or SF3B1 (19%) mutations, and only rarely (6%) BAP1 mutations. In 45 of 47 primary leptomeningeal tumors (96%), at least one mutation in GNAQ, GNA11, PLCB4, CYSLTR2; EIF1AX, SF3B1, or BAP1 was detected, suggesting that screening for presence of mutations in these genes can be diagnostically useful in distinguishing these tumors from metastases from cutaneous melanoma.

The mutation profile detected in metastases from cutaneous melanoma allows a clear distinction of these tumors from other pigmented entities, in particular primary leptomeningeal tumors and melanotic schwannomas (8). The most frequent mutations identified in the former were BRAF, NRAS, or TERT promoter mutations. In 14 of 15 (93%) cases, one of these three mutations was present and could clearly differentiate a cutaneous melanoma metastasis from other entities. Analyzing these genes for the presence of mutations could suffice for diagnostic purposes in most cases.

The similar gene mutation, copy number variation (CNV), and methylation profiles we identified in primary leptomeningeal melanocytic tumors, uveal melanomas and blue nevus–like melanomas were remarkable. Unsupervised hierarchical clustering of these tumors based on genome-wide methylation data separated them according to chromosome and mutation status, but not according to diagnosis (uveal melanoma, blue nevus–like melanoma, or primary leptomeningeal tumor). Comparing individual uveal melanomas, blue nevus–like melanoma and primary leptomeningeal tumors harboring the same or very similar gene mutations showed similar CNV and methylation profiles (Figs. 4 and 5; Supplementary Fig. S2). As previously postulated (29, 30), our study suggests that these tumors represent a common genetic tumor group.

This relation may be of critical importance in terms of interpreting the clinical meaning of genetic alterations. As uveal melanoma are the most frequent and best studied tumor type in this group, the relevance of different mutations in terms of tumor behavior and prognosis is well-documented (28, 31, 32). Obtaining similarly sized cohorts of blue nevus–like melanomas or primary leptomeningeal tumors with long-term follow-up data allowing a comparable validation of the impact of similar gene alterations in these tumors appears unlikely, given the rarity of these tumors. Incomplete follow-up is also a limitation of our primary leptomeningeal tumor cohort. Despite these limitations, the practically identical genetic profiles observed in these tumor entities (uveal melanomas, blue nevus–like melanomas, and primary leptomeningeal tumors) raises the possibility that the prognostic genetic markers in uveal melanoma may also be of prognostic relevance in primary leptomeningeal tumors.

EIF1AX, SF3B1, and BAP1 mutations in uveal melanoma are associated with favorable, intermediate, and poor prognosis, respectively (28, 31–33). Although our study identifies EIF1AX, SF3B1, and BAP1 mutations in primary leptomeningeal tumors, the gene mutation profile does not appear to fit well with the histologic diagnosis in many cases (Figs. 1, 4, 5; Supplementary Fig. S2). Tumors diagnosed as melanocytoma or intermediate-grade melanocytoma assumed to have a favorable prognosis were found to harbor SF3B1 or BAP1 mutations (Fig. 1). Genetically comparable tumors were diagnosed as uveal melanoma or blue nevus–like melanoma, which is demonstrated in individual cases in Figs. 4, 5, and Supplementary Fig. S2. Here one can observe that primary leptomeningeal tumors, uveal melanomas, and in one case a blue nevus–like melanoma harboring similar or identical gene mutations [GNA11 and SF3B1 (Fig. 4), GNAQ and EIF1AX (Fig. 5), GNA11 and BAP1 (Supplementary Fig. S1)] also have comparable copy number profiles. In all the presented cases, the primary leptomeningeal tumor was diagnosed either as a melanocytoma or intermediate-grade melanocytoma, both assumed to have a favorable prognosis. In contrast, the genetically almost identical uveal melanoma and blue nevus–like melanoma cases were diagnosed as malignant melanoma.

Despite these discrepancies, there was a trend toward certain mutations being associated with tumors predicted to show poorer prognosis based on their histopathologic features. The percentage of tumors harboring SF3B1 or BAP1 mutations was 11% (2/19) in melanocytoma, 32% (7/22) in intermediate-grade melanocytoma, and 66% (4/6) in primary leptomeningeal melanoma (P = 0.02).

Poor prognosis being associated with BAP1 inactivation in uveal melanoma is well documented (26, 28, 32, 34). The limited published data for blue nevus–like melanoma and primary leptomeningeal tumors suggest a similar finding (11, 20, 21, 24, 25). Detecting losses of BAP1 genetically can prove challenging, particularly in picking up smaller deletions. IHC is a reliable method of detecting BAP1 inactivation (26), but some mutations can functionally inactivate BAP1 without causing protein loss (9, 31). Although chromosome 3 monosomy is frequently interpreted as equivalent to BAP1 protein inactivation, there will be exceptions. We believe that screening these tumors for mutations, CNV and methylation profiles should be performed when possible. However, such cost-intensive and sophisticated approaches will not always be available. Determining BAP1 status by IHC is relatively reliable and in our opinion should become a routine procedure in the pathologic evaluation of primary leptomeningeal tumors, uveal melanomas, and blue nevus–like melanomas.

In uveal melanoma, it has become accepted that tumor behavior and prognosis can be better predicted by including genetic testing than relying solely on histopathologic analysis. We believe that a similar approach has value in primary leptomeningeal tumors. In our study, a definitive separation based on genome-wide methylation, CNV, or mutation signatures was not possible, however tumors were found to group together based on chromosome 3 and BAP1 status (Fig. 2). Primary leptomeningeal tumors harboring chromosome 3 loss and BAP1 alterations (mutations and/or BAP1 IHC loss) should be considered high-risk, that is, potentially malignant. Tumors lacking these alterations being EIF1AX, SF3B1, and BAP1-wild-type or harboring an EIF1AX mutation (SF3B1 and BAP1-wild-type) could be assumed to have a favorable prognosis. SF3B1-mutant tumors should be considered to be intermediate-risk tumors, as both blue nevus–like melanomas and uveal melanomas with these mutations can metastasize (20, 35). In addition, we would suggest including tumors with chromosome 3 loss but no detectable BAP1 alteration (mutation or BAP1 IHC loss) to the intermediate-risk group. Applying this risk stratification scheme to our cohort of primary leptomeningeal tumors, 16/47 (34%) cases would have been assigned to the intermediate-risk group and 3/47 (6%) cases to the potential high-risk group. The majority of tumors 28/47 (60%) demonstrated a genetic profile suggesting a favorable prognosis.

A comparison of the genetic tumor classification with the histopathologic diagnosis [according to Brat and colleagues (2)] demonstrated relatively poor correlation (Supplementary Table S4). This is not entirely surprising, given that the distribution of EIF1AX, SF3B1, and BAP1 mutations did not fit well with the histopathologic diagnoses (Fig. 2; Table 1). The availability of only limited follow-up data for the patients in our cohort (Supplementary Table S3) precludes comparisons of the genetic and the Brat and colleagues diagnostic schemes with respect to clinical outcomes. Histopathologic evaluation remains a key component of the diagnostic process, and tumors with overt cell pleomorphism, high mitotic index, and necrosis should be considered malignant (or high-risk) tumors independently of genetic findings. However, similar to the case in uveal melanoma, we believe that genetic testing may become a useful tool to help predict tumor behavior and patient prognosis in primary leptomeningeal melanocytic tumors in which unequivocal histopathologic features of malignancy are not observed.

The two retrobulbar melanomas included in the study were classified as primary leptomeningeal melanomas rather than uveal melanomas, because in both cases, imaging and fundoscopy showed no evidence of a primary uveal tumor. Ultimately, given the genetic similarities between primary leptomeningeal melanocytic tumors and uveal melanomas, this distinction may not necessarily be clinically important for management of affected patients.

The observed genetic similarity of primary leptomeningeal tumors with uveal melanomas and blue nevus–like melanomas may have therapeutic consequences. Uveal melanomas and blue nevus–like melanomas have shown very poor response rates to immunotherapy with anti CTLA-4 and PD-1 antibody therapies (30, 36–38). These suggest few patients with advanced primary leptomeningeal tumors may profit from these therapies. However, as long as other promising therapies are still lacking, immune checkpoint blockade should still be considered a viable therapeutic option in these patients.

Our study has some limitations. The cohort of primary leptomeningeal tumors is large, but in many cases the follow-up data were missing or incomplete. The comparison with uveal melanoma and blue nevus–like melanoma is informative, but the available sample numbers of the latter tumors was limited due to their rarity. Studies involving larger cohorts with complete follow-up data would be valuable to be able to validate associations of gene alterations with prognosis in primary leptomeningeal tumors.

In summary, our study is to date the most comprehensive genetic analysis of primary leptomeningeal tumors, demonstrating they can be clearly distinguished from cutaneous melanoma metastasis and have a strong genetic relationship with uveal melanoma and blue nevus–like melanoma. Our data suggest that determining copy number and mutation status in primary leptomeningeal tumors may assist in prognostic assessment.

No potential conflicts of interest were disclosed.

Conception and design: K.G. Griewank, C. Koelsche, J.A.P. van de Nes, A. von Deimling

Development of methodology: K.G. Griewank, A. von Deimling

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): K.G. Griewank, C. Koelsche, J.A.P. van de Nes, M. Gessi, A. Sucker, R.A. Scolyer, M.E. Buckland, R. Murali, T. Pietsch, A. von Deimling, D. Schadendorf

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): K.G. Griewank, C. Koelsche, D. Schrimpf, R. Murali, A. von Deimling, D. Schadendorf

Writing, review, and/or revision of the manuscript: K.G. Griewank, C. Koelsche, J.A.P. van de Nes, M. Gessi, R.A. Scolyer, M.E. Buckland, R. Murali, T. Pietsch, A. von Deimling, D. Schadendorf

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): K.G. Griewank, C. Koelsche, J.A.P. van de Nes, I. Möller, A. Sucker, T. Pietsch, A. von Deimling, D. Schadendorf

Study supervision: K.G. Griewank, A. von Deimling

The authors thank Nadine Stadler and Julia Kretz for excellent technical assistance. Assistance from colleagues at Melanoma Institute Australia and the Royal Prince Alfred Hospital is also gratefully acknowledged. The research was supported by a grant from the Dr. Werner-Jackstädt-Stiftung (www.jackstaedt-stiftung.de) and Hiege-Stiftung gegen Hautkrebs. This research was also funded in part through the NIH/NCI Cancer Center Support Grant P30 CA008748. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.