Meta-Analysis and Systematic Review of the Genomics of Mucosal Melanoma

Mucosal melanoma originates from melanocytes residing within extracutaneous epithelial tissues of ectodermal origin. This subtype of melanoma is not driven by ultra-violet radiation exposure, except in the conjunctiva (1); however, the etiologic factors driving tumorigenesis have not been established. Mucosal melanoma can develop on any mucosal surface, but primary lesions commonly arise in the sinonasal tract, oral cavity, female genital tract, the anorectum, and urinary tracts (2–5). Incidence of mucosal melanoma in the United States is 2.2 per million, with women having a slightly, but significantly, higher incidence than men (2.8 vs. 1.5 per million; ref. 3), largely due to instances of vulvovaginal melanoma, which represent 18% of mucosal melanoma cases (reviewed in ref. 6). Mucosal melanoma incidence is similar between populations; however, mucosal melanoma represents a greater proportion of cases in people of non-European descent (2–4, 7–9).

Patients with metastatic mucosal melanoma have a significantly shorter median overall survival compared with other melanoma subtypes (10) and this is attributed partly to frequent patient diagnosis with advanced disease and a lack of effective therapies (9, 11, 12). A notable contributor to lack of treatment options is that genomic drivers and cellular pathways hijacked by tumors are yet to be fully characterized.

Several next-generation sequencing (NGS) studies of small cohorts have examined: (i) targeted gene panels (13, 14); (ii) whole exomes (15–19); (iii) whole genomes (1, 18, 20). Other studies used Sanger sequencing of known mutation hotspots (21–58). Given the diversity of genomic changes in cancer cells, important driver events can be missed by small studies. A systematic review of these data is therefore timely, to form the most comprehensive analysis of mucosal melanoma genomics to date.

We acquired and analyzed published data to assess genomic changes in mucosal melanomas from different body sites. The identified changes revealed signaling pathways frequently altered, with different pathways components altered across the compiled tumors. These pathways include: (i) the MAPK, PI3K, NOTCH/β-catenin, Wnt, and cell-cycle pathways, leading to the cancer hallmarks of sustained proliferative signaling and evading growth suppression; (ii) methylation maintenance, chromatin modification, the spliceosome and DNA repair pathways, leading to genome instability and mutation; and (iii) canonical and ALT routes of telomere maintenance, enabling replicative immortality.

A literature review was conducted using PubMed to identify relevant studies published up until March 2019. The search used the search terms: “mucosal melanoma”, “sinonasal melanoma”, “oral cavity melanoma”, “vulva/vaginal melanoma”, and “anorectal melanoma”. Abstracts were used to identify studies including a sequencing component. Selected manuscripts were assessed to determine whether entire exons were sequenced and if complete mutation details were reported.

Data were acquired from each of the studies as detailed in the Supplementary Methods and Supplementary Table S1. Briefly, seven eligible studies were identified that used whole-exome sequencing (WES) and/or whole-genome sequencing (WGS) of fresh frozen (FF) tumors (1, 15–20, 59) and two that used formalin-fixed, paraffin-embedded (FFPE) tissue (15, 60). Sequencing data were accessed through the European Genome-Phenome Archive, cBioPortal, or directly from the studies' supplementary materials. WGS data from Zhou and colleagues was stated to be publicly available; however we were unable to gain access; therefore, only mutation data (20) have been included. Data from MSK-IMPACT is hosted on AACR GENIE and can be visualized on cBioPortal (http://www.cbioportal.org/genie); the corresponding data were downloaded via Synapse (https://www.synapse.org). The targeted hotspot sequencing studies focused on BRAF, NRAS, and KIT were hand-collated for analysis.

Only NGS data from tumors with matched normal were included. Samples from FF tissue were included in the “main” cohort (n = 173; refs. 15–20, 59), while FFPE samples were included as a “validation” cohort (n = 48; refs. 15, 60; Table 1; Supplementary Table S1). We were unable to include samples from Zhou and colleagues (20) in assessments of copy-number aberrations (CNA) or structural variants (SV) due to lack of data availability. Therefore, a total of 72 samples were available for this aspect of data analysis (Table 1).

There were 104 mucosal melanomas in the MSK-IMPACT dataset at the time of data acquisition (Table 1; ref. 13); samples were analyzed by one of three targeted gene panels (Supplementary Table S1). The number of samples included in targeted hotspot analysis was dictated by the exons sequenced within 36 studies assessing KIT (13, 14, 22–24, 27–29, 31–35, 37, 39–58, 61, 62), 25 assessing BRAF (14, 21–35, 37, 39–46), and 22 assessing NRAS (13, 14, 21, 22, 24, 25, 27–29, 31, 33, 34, 36, 38–46; Table 1; Supplementary Table S1).

For complete details of the analyses carried out, see Supplementary Methods. Raw sequencing reads from Hintzche and colleagues (15) and Furney and colleagues (18) were aligned against hg19 and variants identified using our analysis pipeline (63). Data from the other cohorts were taken either from the manuscript supplementary data or from cBioPortal. Variants were annotated using the Ensembl Variant Effect Predictor (64). Mucosal melanoma data from GENIE MSK-IMPACT were downloaded directly from Synapse; these calls were not reannotated by our pipeline. Tumor purity was previously assessed using ascatNGS, a tool that uses CNAs to estimate tumor content (59).

Data visualization of the main cohort was achieved using Maftools (65) version 2.4, run via Bioconductor release 4.11. Targeted hotspot sequencing data was visualized in RStudio using ggplot2. To identify significantly mutated genes, MutSigCV, Oncodrive FML, and OncodriveCLUST were used. GISTIC2.0 was used to identify the targets of focal CNAs (66). Genes discussed were analyzed for mutual exclusivity and cooccurrence using the somaticInteractions tool in Maftools in RStudio. Mutation signatures were analyzed using an in-house tool by comparing the distribution of single base substitutions (SBS) with the COSMIC mutation signatures version 3.0 (67). Genes were assigned to cancer hallmarks using the COSMIC Census (https://cancer.sanger.ac.uk/census) as a guide. Cancer Genome Interpreter (https://cancergenomeinterpreter.org/home) was used to identify genomic biomarkers of drug response.

To harmonize the anatomic sites reported across studies, the primary sites were recategorized to represent an anatomic region, rather than a precise site, where appropriate. The main sites included were sinonasal, oral cavity, esophageal, vulvovaginal, and anorectal (Supplementary Table S1). MSK-IMPACT did not subclassify mucosal melanomas occurring in the head and neck.

Primary sites represented across the three cohort, main (n = 173), validation (n = 48), and targeted panel (n = 104) are summarized in Table 2. A range of sites were represented across the cohorts, most commonly the sinonasal tract, oral cavity, vulvovaginal. and anorectal tract. The studies by Zhou and colleagues (20) and Lyu and colleagues (19) only included sites of oral cavity and sinonasal tract, which skewed assessment toward melanomas of the upper anatomy in the main cohort. The primary sites of samples from the hotspot targeted panel were summarized per manuscript due to inconsistencies by which primary sites were reported between studies (Supplementary Table S1).

Throughout, details of the genomic alterations found in the main cohort (n = 173 for mutations, and n = 72 for CNA and SV) are in regular font, the FFPE validation cohort (n = 48) are in italics, and the targeted sequencing of the MSK-IMPACT panel (n = 104; ref. 13) are underlined when available, see Supplementary Methods and Supplementary Tables S1 and S2. For BRAF, NRAS, and KIT, the number of samples sequenced at targeted hotspots are noted. For CNA, only gene deletions (copy number 0 and 1) and focal amplifications (copy number > 6) are included.

Tumor mutation burden (TMB) was calculated using the total number of nonsynonymous mutations. In the main cohort (n = 173), the mean number of mutations was 84 (range 2–732). The highest mutation burden was identified in an esophageal sample. TMB has been associated with DNA repair dysfunction (68) and accordingly, samples with two or more mutations in DNA repair genes were found to carry higher TMB (Kendall's Tau Test P = 0.00003); however, mutations in TP53 were not significantly associated with TMB (Mann–Whitney test P = 0.98846; Supplementary Fig. S1), contrary to observations in other cancers, including cutaneous melanoma (69).

Mutation signatures were assessed in the main cohort (Supplementary Fig. S2A). Recurrent signatures detected in >50% of samples included SBS1 (66%), associated with deamination of 5-methylcytosine to thymine, and SBS39 (63%) which is of unknown etiology. SBS39 was the dominant signature (defined as > 50%) in 12% of samples. Signatures SBS7a and SBS7b, associated with exposure to ultraviolet radiation, were identified in 72 (42%) samples and was the dominant signature in eight; seven were of the upper anatomy and one with unknown primary site. To examine mutation signatures by tumor site, the proportions of each signature were averaged within each primary site group and summarized in Supplementary Fig. S2B. In the eye (n = 2), signature SBS7a and SBS40 (unknown etiology) represented 24% and 35% of the signatures, respectively. In the oral cavity (n = 98) SBS39 was the major signature (20%) and SBS1, SBS7a/b, and SBS29 (associated with chewing tobacco) were also present. Two extremely rare sites were present: the esophagus and the penis. In the esophageal sample, SBS4 (35%; associated with tobacco smoking) SBS37 (32%; unknown etiology), SBS19 (21%; unknown etiology) and SBS1 (12%) were detected. In the penile sample, SBS3 (25%; associated with defective homologous recombination and therefore impaired DNA damage response) was present; this sample had copy loss of homologous recombination genes MRE11A and RAD51B.

Statistical analysis of the variants in the main cohort yielded the well described mucosal melanoma driver genes as significantly altered (q < 0.05) by more than one tool: KIT (19.1%), NF1 (9.8%), BRAF (9.2%), NRAS (8.7%), SF3B1 (8.1%), and SPRED1 (4.0%), highlighting the importance of these common driver mutations (Fig. 1A; Supplementary Table S2). A tendency toward mutual exclusivity between KIT, NF1, BRAF, and NRAS was identified; however, only the KIT/BRAF combination was significant (Fisher exact test: P = 0.044).

Mucosal melanoma is, however, not a cancer predominantly driven by mutation, with high occurrence of CNAs and SVs (59). Examination of the genes within loci statistically significantly affected by CNAs also implicated well described cancer driver genes (Fig. 1B; Supplementary Fig. S3; Supplementary Table S2). These included copy loss of genes encoding: inhibitory cytoplasmic signaling proteins (NF1, 15.2%; SPRED1, 12.5%; PTEN, 22.2%); a cell-cycle checkpoint protein (CDKN2A, 50%); DNA repair proteins (ATM, 34.7%; CHEK2, 12.5%); and chromatin remodeling proteins (ARID1A, 8.3%; ARID1B, 33.3%), and the amplification of genes encoding: a telomere maintenance protein (TERT, 16.7%), a receptor tyrosine kinase (KIT, 15.3%), cytoplasmic signaling proteins (BRAF, 6.9%; YAP1, 8.3%; GAB2, 18.0%), cell-cycle checkpoint proteins (CDK4, 20.8%; CCND1, 15.3%; MDM2, 16.7%; and SKP2, 15.3%) and a transcription factor (MITF, 12.5%). The reported cooccurrence of BRAF p.V600E mutation and MITF amplification in cutaneous melanoma (70) is not seen in mucosal melanoma (Supplementary Table S2).

These mutations and CNAs are collectively described as “significantly altered genes” (SAG) in the remainder of this manuscript.

Activation of KIT initiates a signaling cascade promoting cell survival, proliferation, and differentiation. It was the most commonly mutated gene, with mutations clustering in hotspots that cause KIT activation by disruption of either the autoinhibitory function of the juxtamembrane region at amino acids 560–586 (main: 7.5%; validation cohort: 0%; MSK-IMPACT: 13.5%) or the kinase domains at amino acids 591–710 (5.8%; 6.3%; 1.9%)/771–992 (5.2%; 0%; 4.8%). Analysis of targeted hotspot mutations (Fig. 2A; Supplementary Table S2) identified a total of 253 mutated samples (exon 9: 0.8%; exon 11: 9.0%; exon 13: 1.9%; exon 17: 2.5%; exon 18: 1.0%). Notably, variants in codons 557 (n = 13), 559 (n = 11), and 560 (n = 13) were observed by hotspot exon sequencing, which were only present in low number in the main cohort (n = 0, n = 1 and n = 1, respectively), underlining the importance of collating large numbers of samples to identify rarer KIT mutations. Amplification of KIT was common (15.3%; 17.3%), and sometimes occurred in conjunction with mutations (5.6%; 14.4%; Fig. 1B). Insertion–deletion events were identified in 24 samples (n = 24/1,728) in the hotspot cohort and these occurred exclusively in exon 11 (Fig. 2A), which results in a truncated protein product.

In normal conditions, NF1 and SPRED1 bind to RAS proteins, blocking phosphorylation of RAF proteins, switching off downstream signaling and diminishing downstream ERK activation (71, 72). In mucosal melanoma, events that result in the loss of inhibitory signaling (NF1, SPRED1) or gain of activation signaling (NRAS, BRAF) of the MAPK pathway, promote cell proliferation.

NRAS is the RAS family member mainly mutated in mucosal melanoma, occurring in well described activating hotspots: codon 12 (3.5%; 4.2%; 8.7% and 44/1134 hotspot sequencing), codon 13 (1.2%; 4.2%; 0%; 16/1134, hotspot sequencing) and codon 61 (4%; 8.3%; 4.8% and 59/1139 hotspot sequencing; Figs. 1A and 2B; Supplementary Table S2). Several uncommon mutations were observed, the majority from targeted hotspot sequencing, near to codon 61, (58, 59, 62, 63 and 65) and in exon 2, including p.Y4C, p.V7M and p.V8M (Fig. 2B). NRAS was rarely amplified (2.7%, 0.96%; Supplementary Table S2), but these amplifications cooccurred with NRAS hotspot mutations (n = 2 codon 12; n = 1 codon 61).

NF1 was altered by loss-of-function (LoF) mutations (8.1%; 6.3%; 14.4%), homozygous deletions (HD; 1.4%; 9.6%) and loss of heterozygosity (LoH: 12.5%; 4.8%; SV: 9.7%). SPRED1 was altered by LoF mutations (4.0%; 2.1%; 0%), copy losses (12.5%; 12.5%) and SVs (12.5%). These data include double hits in one sample for NF1 and three samples for SPRED1. Missense variants in NF1 in the main cohort (p.S574I in splice region of exon 15; p.L844F; p.V1627D) and MSK-IMPACT (p.R403L) are of unknown function (Supplementary Table S2).

BRAF is the main RAF family member mutated in mucosal melanoma, with the majority occurring at the p.V600 activating hotspot (2.9%; 0%; 3.8% and 60/1223, targeted hotspot sequencing; Figs. 1A and 2C; Supplementary Table S2). Unlike in cutaneous melanoma (1, 73), there were also a notable number of mutations at p.D594 (1.7%; 0%; 1.9% and 15/1223, targeted hotspot sequencing); Fig. 2C. Single occurrences of mutations in exons 6, 11, 12 and 17 were also observed (Supplementary Table S2). Two samples had two BRAF mutations (p.T241M and p.L485W; p.L597P and p.K690E), which is a rare event in melanoma (1). BRAF was also amplified in nonmutated samples (6.9%; 0.96%; Supplementary Table S2).

SF3B1 is a core component of the U2 small nuclear ribonucleoprotein, involved in recognition of the splicing branchpoint sequence. The SF3B1 mutation hotspot at codon p.R625 is associated with aberrant pre-mRNA splicing, as cryptic splice sites are no longer hidden by secondary structures in the mutated spliceosome (74). This has effects on many genes, causing down-regulation or altered function of proteins, to promote tumorigenesis (75, 76). Several different mutations at hotspot codon p.R625 were observed: p.R625H (4%; 6.3%; 16.3%), p.R625C (0.6%; 4.2%; 4.8%), p.R625 L (1.7%; 0%; 0.96%), p.R625S (0.6%; 0%; 1.9%) and p.R625G (0.6%; 0%; 0%; Fig. 1A; Supplementary Table S2). The majority were observed in lower anatomy (45/57 total) mucosal melanomas, as previously reported (59). One splice site mutation (c.496–2A>G) and single occurrences of missense variants at other codons (p.T964A; p.E622L; p.K998N), which have unknown functional impact, were also observed (Supplementary Table S2).

Many of the SAG encode proteins that coincide in pathways leading to specific cellular effects frequently perturbed in cancer; these functions have been described as the “Hallmarks of Cancer” (77). Further to the SAGs identified, we found additional genomic alterations in other genes within these pathways that did not reach statistical significance. We therefore sought to map these genes within the hallmarks of cancer from which pathways of reliance and vulnerability for mucosal melanoma emerged.

There are ten hallmarks of cancer (77), of which components of four are notably targeted in mucosal melanoma via SAGs, as described previously. These pathways often have overlapping components, meaning genomic alterations can have multiple functional outcomes. The following sections detail these functional “driver” genomic changes in mucosal melanoma, in the context of each hallmark highlighted (Visual Overview).

Maintenance of proliferative signaling and evasion of growth suppression is achieved through different processes culminating in cellular proliferation. Over-activation of cellular pathways via receptor tyrosine kinase (RTK) and growth factors receptor (GFR) results in aberrant signaling. Furthermore, alterations to components within these signaling networks and within the cell-cycle machinery lead to uncontrolled cellular proliferation (Visual Overview).

Autocrine activation of RTK and GFR can occur via overexpression of the ligand within tumors (78). In mucosal melanoma, we identified amplification of 1q21–22 in 9.7% of samples, which contains EFNA1, EFNA3, and EFNA4, encoding ephrin receptor ligands (Visual Overview; Supplementary Fig. S4B). We also identified amplifications in FGF3 (13.9%; 6.7%), FGF4 (13.9%; 6.7%), and FGF19 (13.9%; 7.7%), located in a cluster on 11q13.3, and encode mitogens that bind to fibroblast GFRs (FGFRs; Visual Overview; Supplementary Fig. S4B). Amplification of these genes have been described in other cancer types, including breast cancer (79), hepatocellular carcinoma (80), lung cancer (81), and pancreatic ductal adenocarcinoma (82). Overexpression of GFRs and RTKs occur commonly in tumors, which can result in amplified signal from ligand binding (1, 83–85). In mucosal melanoma, KIT was a SAG (Fig. 1A and B), and the gene encoding the GFR PDGFRA was amplified (16.6%; 11.5%); these events tended to cooccur in the main cohort (Fisher exact test P < 0.0001; Supplementary Fig. S4B). Lower frequency amplifications of ERBB2 (1.4%; 3.8%), EGFR (2.8%; 1.9%), and FGFR1, FGFR3 and FGFR4 (total: 9.7%; 1.9%) were found. Mutations in CBL (4%), a negative regulator of RTK signaling, were focused in the zinc finger domain (381–420), which links the SH2 and RING finger domains and is vital for the E3 ligase activity that controls RTK modification (Supplementary Fig. S4A). These mutations disrupt normal function of the protein, leading to dysregulated receptor trafficking and signaling (86–88). Similarly, IGF2R, a receptor that attenuates the signal from IGF2, was also frequently lost (33.3%), which can result in increased signaling from IGF2 (ref. 89; Supplementary Fig. S4B). Another mechanism by which receptors are activated is via mutations that disrupt specific domain function (90–92), which, aside from KIT, occurred infrequently (in ERBB2, ERBB4 and FGFR1, FGFR2, FGFR3, and FGFR4; Visual Overview; Supplementary Fig. S4A). Upon activation, these RTKs and GFs can drive signaling pathways promoting cellular proliferation, including MAPK, PI3K, and WNT pathways, which are likely context and tissue dependent (Visual Overview; ref. 78).

Upon activation of RTK, the SAG GAB2 (amplification, 18%) is phosphorylated, allowing binding to targets involved in signal transduction. This amplifies transmission of signaling to downstream effectors, including MAPK and PI3K pathways (93). Members of the RAS family are mutated at known functional hotspots (94–96), resulting in constitutive activation of MAPK and PI3K–AKT pathways. The most commonly mutated RAS family member, NRAS, was a SAG (Fig. 1A). Other members of the RAS family were infrequently altered by activating mutations, including HRAS (2.3%; 2.1%; 1.9%) and KRAS (1.2%; 2.1%; 0.96%; Supplementary Fig. S5A).

Downstream of RAS and BRAF, further members of the MAPK signaling pathway were genomically aberrant, such as MAP2K1 (0.6%, 0%, 0.9%), MAP2K2 (0%, 2.1%, 0%), and MAPK1 (0%, 2.1%, 0%). These were generally missense variants of unknown significance (VUS), present at low frequency (Supplementary Table S2), but given the importance of this pathway in driving cellular proliferation are a potential focus for future research.

An alternative activating pathway downstream from RTK and GFs is the PI3K/AKT pathway, which promotes proliferation via mTOR (Visual Overview). PTEN negatively regulates this pathway and its loss results in the removal of this brake (LoF mutation (97, 98): 2.3%, 2.1%, 2.9%; HD: 5.5%, 2.8%; LoH: 16.7%; 0%; SV: 11.1%). There are three classes of PIK3, with family members having different roles and some tissue-specific expression. In mucosal melanoma, the class I family member PIK3CA had oncogenic and hotspot variants (99), including p.N345D p.C420R, p.E545K, and p.H1047L (0%, 0%, 4.8%), and PIK3CG (5.5%; 3.8%) and the class II family member PIK3C2B were amplified (9.7%; Supplementary Table S2). Of the AKT family of downstream effectors, only AKT3 was amplified (8.3%; 4.8%), which can be overexpressed in cutaneous melanoma (100). Downstream from AKT, mTOR is activated to initiate cellular proliferation, which is inhibited by TSC1 and TSC2, both of which were frequently lost in the main cohort (total: 18%; 0.96%), releasing this inhibition (Supplementary Fig. S5A and S5B).

WNT signaling is activated when a WNT ligand binds to the Frizzled receptor family (FZD), which induces several intra-cellular signal transduction cascades, including canonical (WNT/β-catenin dependent) or non-canonical (β-catenin independent) pathways (Visual Overview). There are 19 members of the WNT ligand family, with different tissue and temporal-specific expression, with roles during development. WNT11 and WNT3A have been associated with cancer progression, via the canonical signaling pathway (101–104), and notable amplifications of 1q42 harboring WNT11 (11.1%) and WNT3A/WNT9A (8.3%) occur in mucosal melanoma (Supplementary Fig. S4B). Upon activation of the FZD receptors, dishevelled (DVL) is activated, causing the removal of inhibitory “destruction” complexes from degrading β-catenin, allowing it to build up in the cytoplasm, enter the nucleus and initiate cellular proliferation pathways (Visual Overview). The destruction complex includes APC, which is frequently disrupted in some cancers (105), resulting in stabilized β-catenin signaling; in mucosal melanoma, APC was rarely disrupted (LoF: 0.6%; 2.1%; 0.96%; LoH: 4.2%, 0.96%). Mutations in the hotspot regions in exon 3 of β-catenin (CTNNB1; 4.6%; 4.2%; 5.8%) result in a lack of phosphorylation, stabilizing β-catenin and constitutive activation of this pathway (as reviewed in ref. 106). Mutations in CTNNB1 were more common in upper anatomy mucosal melanomas (13/16 total). Finally, the genes encoding proteins involved in β-catenin phosphorylation, BTRC (encoding β-TrCP) and/or GSK3B (encoding GSK-3) were lost (18.0%), potentially further strengthening signaling through β-catenin in these tumors (Visual Overview; Supplementary Fig. S5A and S5B).

Other cellular pathways interact with aspects of MAPK and/or WNT signaling, including YAP/TAZ, Hedgehog, and Notch pathways (107–111).

The YAP/TAZ pathway can be activated downstream of growth factors, G-protein receptors, WNT, and cellular stresses (via MAPK; Visual Overview). The usual manner of activation involves dephosphorylation of an inhibitory complex including LATS1 (HD: 2.7%, LoH: 29.1%; HD: 4.8%) and LATS2 (LoF: 0.58%, LoH: 11.1%), resulting in the dephosphorylation of the SAG YAP1 (amplified: 8.3%; 1.9%) and TAZ, allowing translocation to the nucleus; LoF of LATS1/2 increases YAP/TAZ signaling (112). Furthermore, MAP4K4 enhances phosphorylation of LATS1/LATS2, inhibiting YAP/TAZ; MAP4K4 was rarely lost (LoF: 1.7%; 4.2%; LoH: 2.7%), potentially increasing YAP/TAZ signaling (ref. 113; Visual Overview; Supplementary Fig. S5A and S5B).

NOTCH signaling pathways have been implicated as oncogenic in melanomas (114), through activation of MAPK, AKT, and β-catenin pathways (Visual Overview). Rare mutations occurred in NOTCH receptor attenuators, including DLL3 (115) (LoF: 0.6%) and NEURL1 (ref. 116; LoH: 13.8%), which are likely to be functional (115, 116). While there are rare, likely functional, alterations in the NOTCH pathway, it is not a recurrent target of genomic changes in mucosal melanoma (Supplementary Fig. S5A).

Alterations involving cell-cycle progression promotion tended to be CNA, rather than mutation (Supplementary Fig. S6A and S6B) and many components are encoded by SAGs, indicating the importance of this process to tumorigenesis. The cell-cycle can be halted via the DNA damage repair pathway (hallmark: genomic instability and DNA mutation) and LoF of these processes can remove these brakes (Visual Overview). ATR (LoH: 11%) and ATM (LoH: 34.7%; HD: 0.96%) are activated in response to single- or double-stranded DNA breakages, respectively, which activate CHEK1 (LoH: 33%) and the CHEK2 (LoH: 12.5%), respectively, to halt the cell cycle via p53 activation. Many samples had overlapping aberrations between the ATR/CHEK1 and ATM/CHEK2 axis, present in 26% of samples (Visual Overview; Supplementary Fig. S6B), thereby affecting response to both single- and double-stranded DNA breakage. Functional p53 was lost by a variety of genomic events (LoF: 4.6%; 12.5%; 17.3%; LoH: 11.1%; SV: 1.4%), resulting in removal of inhibitory cell-cycle signals, allowing the cell cycle to progress (Supplementary Fig. S6A and S6B). In a mutually exclusive manner, the SAG MDM2, negative regulator of p53, was amplified (16.7%; 5.7%), again removing these inhibitory signals. MYC acts as a stimulatory molecule for entering the cell cycle after signaling from MAPK, P13/AKT, and WNT pathways (117), and was amplified (12.5%; 8.6%) in a mutually exclusive manner to TP53 and MDM2 (Supplementary Fig. S6B).

CDK4, CCND1, and CDKN2A were SAGs. CDK4 is the inhibitory binding partner of CDKN2A p16^INK4A and can also form a complex with CDK6/CyclinD1, which targets RB1 and controls progression of G₁. Together, these genes were targeted by genomic aberrations that drive the cell-cycle progression (Fig. 1A; Supplementary Fig. S6B). Notably, samples with RB1 copy loss, also had loss of CDKN2A (Fisher exact test P = 0.011; LoF: 1.1%; 2.1%; 5% HD: 25%; 19.2%; LoH: 25%; 0.96%; SV: 9.7%) and TP53 (Fisher's exact test P < 0.0001; Supplementary Fig. S6B). Amplifications of MDM2 (20.8%) and CDK4 (9.6%) significantly cooccurred (Fisher's exact test P < 0.0001; Supplementary Fig. S6B) and CDK4 binding partner CDK6 was also infrequently, independently amplified (6.9%; 2.8%). CCND1 was amplified (15.3%, 7.7%), often in conjunction with other genes involved in cell-cycle control (Supplementary Fig. S6B). SKP2 promotes S-phase of the cell cycle, via degradation of target proteins that halt progression, including the CDK inhibitor p27, and increased expression drives proliferation (118); SKP2 was frequently amplified (15.3%; Supplementary Fig. S6B). Together, proteins involved in the cell cycle were altered in 90% of samples.

DNA damage can occur due to exposure to endogenous or exogenous sources. Maintenance systems exist to detect and resolve these defects, including instigating DNA damage repair pathways (119); genes encoding components of which are often mutated in cancer. Different tumor types tend to have pronounced alterations to particular DNA repair pathways (78). As highlighted in the “cell-cycle progression” subsection, the proteins encoded by ATR/CHEK1 and ATM/CHEK2 represents the most altered aspect of DNA repair in mucosal melanoma.

Lower frequency aberrations in single stranded (mismatch, MMR; nucleotide, NER; and base excision, BER) and double stranded (homologous recombination, HR; and non-homologous end joining, NHEJ) DNA repair pathways were present in mucosal melanoma, usually LoH events (Visual Overview). Approximately 80% of samples harbored changes to DNA repair genes; however, unlike other cancers (120), no notably over-represented pathway(s) of DNA repair were specifically targeted (Visual Overview; Supplementary Fig. S7A and S7B).

Components of each of the single-stranded DNA repair systems, including the DNA excision and ligation steps were occasionally lost; however, cumulatively 30% of samples had at least one loss or VUS in an MMR gene, approximately 50% each for NER and BER (Visual Overview; Supplementary Fig. S7A and S7B). There were rare occurrences of functional mutations in these genes: (i) MMR pathway: p.S723F in MSH2, which falls within the ATPase domain and leads to deficient mismatch DNA repair (121); (ii) NER: single occurrences of inactivating mutations in DDB1 (p.F829Ifs*11), CUL4A (p.W569*), XPC (p.E138*), and ERCC3 (Q586Rfs*25); (iii) BER: an in-frame insertion at codon 467 in PARP1, which has been previously shown to be the start of a functionally important domain (122); PARP1 is also important in double-stranded DNA repair (123).

Unrepaired double-stranded breaks can lead to deletions or chromosomal aberrations. Genes involved in HR and NHEJ were infrequently lost; however, in addition to PARP1 as described above, genes encoding members of HR complexes were also affected by LoF mutations or HD, including MRE11 (p.N511Ifs*13), which is part of the Mre11–Rad50–NbS1 complex (124) and RAD51 (HD: 0%, 2.8%; LoH: 16.6%; 0%; Supplementary Fig. S7A and S7B). Cumulatively, 33% of samples had at least one aberration in HR genes and 26% in NHEJ.

Modifications to chromatin and DNA structure in tumors lead to altered epigenetic states and changes to chromatin access by regulatory proteins, contributing to tumorigenesis (125, 126). Chromatin regulatory factors (CRF) control chromatin structure and DNA modifications; genomic alterations to CRFs can have widespread epigenetic effects. There are three main classes of CRF: (i) ATP-dependent; (ii) histone tail modifiers; (iii) DNA methyltransferases and demethylases.

ATP-dependent chromatin remodeling complexes use an ATPase subunit to mobilize nucleosomes along DNA, removing histones and replacing histone variants from nucleosomes. Members include the switch/sucrose nonfermentable (SWI/SNF) complex, ISWI (imitation switch), INO80 (inositol), and CHD (chromodomain helicase DNA-binding) families. Of these, ISWI and INO80 complex components were only very rarely genomically altered in these cohorts (Supplementary Table S2).

LoF of the switch/sucrose nonfermentable (SWI/SNF) complex leads to increased H3K27 methylation and tumor cell-cycle progression (127) and components were notably lost in mucosal melanoma (Supplementary Fig. S8A and S8B), including the SAG ARID1B (HD: 1.4%, 4.8%; LoH: 31.9%, 0%) and SMARCA4 (HD: 2.8% and LoH: 25%); Visual Overview. ATRX contains an SNF/SWI2-related helicase domain and acts with the histone chaperone DAXX to insert histone variant H3.3 (128) and disrupted ATRX (LoF: 3.5%; 6.25%; 11.5%; SV: 4.2%) leads to increased homologous recombination facilitating telomere destabilization, and development of ALT telomere elongation (ref. 129; leading to the cancer hallmark: replicative immortality; Visual Overview; Supplementary Fig. S8A and S8B). ATRX mutations were more common in mucosal melanoma of lower anatomy (18/23 total).

Finally, members of the CHD family were rarely lost in mucosal melanoma (CHD1, CHD5 and CHD8, totals LoF: 2.9%; SV: 2.8%; LoH: 15.2% (Visual Overview; Supplementary Fig. S8A and S8B), which have varying but overlapping effects. CHD1 has been reported as having oncogenic (e.g., in estrogen positive breast cancer) or tumor suppressive (e.g., in pancreatic cancer or colorectal cancer; reviewed in ref. 130). The reported colosses of CHD1 and MAP3K7 (e.g., in pancreatic cancer; ref. 131) were not observed. CHD5 efficiently unwraps nucleosomes, which prevents complete destabilization of the nucleosome (132) and acts as a transcriptional repressor; when function is lost, cell-cycle progression gene repression is relieved (leading to the cancer hallmark: sustained proliferative signaling; ref. 133). CHD8 has been ascribed with many cancer-related functions, including interaction with CCCTC-binding factor resulting in alteration in transcription (via chromatin insulation, DNA methylation, and histone acetylation) of members of the WNT/β-catenin signaling pathway and cell-cycle regulators; loss of CHD8 has been commonly described in cancers such as breast, gastric, and colorectal cancers (Visual Overview; Supplementary Fig. S8A and S8B; refs. 130, 134).

Lysine methyltransferases (KMT) leave methylation marks on histone tails; mutations and copy-number losses were observed in members of this family (KMT2A/2B/2C/2D VUS: 17.3%; 6.2%; 15%; HD: 0%, 2.5%; LoH: 37.5%; 0%; Visual Overview; Supplementary Fig. S8A and S8B). Loss of these genes results in aberrant methylation control leading to overexpression of a wide variety of target genes (135). Other methyltransferases, such as NSD1 and EZH2, were only very rarely altered (Supplementary Table S2) and notably, none of the gain-of-function hotspot variants were present (136, 137). Members of the lysine demethylase (KDM) family were also commonly mutated (8.1%; 12.5%) and lost (47%), which can result in the loss of control of gene silencing by methylation (ref. 138; Visual Overview).

Gene promoter methylation is modified by members of the DNMT and TET family of proteins, which were only rarely altered (Supplementary Table S2). The IDH1 mutation at hotspot p.R132 in the substrate binding site, which results in increased DNA hypermethylation via reduced levels of α-ketoglutarate (139), was rarely present in mucosal melanomas (0.58%; 0%; 0.96%).

Of the four main described spliceosome proteins (U2AF1, SRSF2, ZRSR2, and SF3B1) affected by genomic aberration in cancer (140), only SF3B1 was altered in these cohorts and was a SAG (Fig. 1, Visual Overview).

Cancer cells often achieve replicative immortality through the maintenance or elongation of telomeres. Telomerase (TERT) is often a target of genomic change to stabilize telomeres through the many cycles of replication that cancer cells undergo. The TERT promoter is frequently mutated in cutaneous melanoma (1, 141), but less frequently (9%, 5.8%) in mucosal melanoma. However, TERT was a SAG due to amplification (16.6%; 5.7%; Fig. 1B).

As discussed in the context of ATP-dependent CRFs, ATRX was disrupted in a number of samples, which can result in activation of the alternative (ALT) pathway of telomere elongation. These two mechanisms of telomere stabilization/elongation are mutually exclusive.

Analysis using Cancer Genome Interpreter identified 144 biomarkers with a responsive effect targeting the genes altered in mucosal melanoma (Supplementary Table S2). Selected examples are briefly outlined here, grouped by function.

As a high number of mucosal melanoma carry aberrations in genes encoding components of the cell-cycle pathway, inhibitors targeting this process are of particular interest. Drugs such as palbociclib and ribociclib (CDK4/6 inhibitor) have been trialed in solid tumors and hematologic malignancies (e.g., NCT02187783, NCT01237236) including one in acral melanoma (NCT03454919); however, many patients were reported to have progressive disease. Palbociclib is currently approved in breast cancer as a combination therapy, suggesting similar combination therapies could be explored in mucosal melanoma. MDM2 inhibition may also be therapeutic, as shown in a liposarcoma trial (NCT01877382).

Cetuximab (EGFR inhibitor) can confer a survival benefit in patients with cancers expressing high levels of EGFR (142). While EGFR was amplified infrequently in mucosal melanoma (<2.5%), protein expression levels have not been explored. Early preclinical studies suggested specific ERBB4 mutants in CM were more sensitive to lapatinib (143), however, the ERBB4 mutations in mucosal melanoma differ from those identified in CM. The effect of lapatinib on mucosal melanoma bearing ERBB4 mutations therefore needs to be assessed. Several inhibitors of FGFR, which was amplified in 9.7% of mucosal melanoma, are still in preclinical stages, but warrant assessment in mucosal melanoma.

ATR, ATM, and PALB2 alterations, all of which are present in mucosal melanoma, may be sensitive tumors to PARP inhibition, as shown in preclinical studies (144, 145) and early trials (NCT01585805).

Patients with endometrial cancer with CTNNB1 mutations were found to respond well to everolimus (mTOR kinase inhibitor)/letrozole (aromatase inhibitor) combination therapy (146). A novel tankyrase inhibitor of the Wnt/β-catenin pathway was also shown to decrease signaling in APC-mutant mice (147).

Because of the rare nature of mucosal melanoma, studies examining genomic alterations have been limited to small tumor numbers (1, 13–17, 19–62). Despite this, these studies have provided insight into the major alterations seen in mucosal melanomas, mainly identifying the common drivers. These well-characterized driver genes were noted in the cohorts collated here. We systematically reviewed data from multiple studies to provide an in-depth characterization of the less common genomic changes in this rare melanoma subtype in the framework of the hallmarks of cancer (ref. 77; Visual Overview). The main relevant pathway alterations in mucosal melanoma have roles in sustained proliferative signaling, evading growth suppression, genome instability and mutation, and enabling replicative immortality. There are several overlapping aspects to these changes, which have been noted. These changes illustrate the complexity of the hijacking of normal cellular processes by tumor cells. Interestingly, infrequent recurrent variants in mucosal melanoma were identified in our analysis, which are of clinical relevance, including: (i) PIK3CA mutations for which FDA approval has been given for treatment with alpelisib (α-specific PI3K inhibitor); (ii) the IDH1 hotspot p.R132C variant, which is the target for a new inhibitor, AGI-5198 (AG-120), now in phase I clinical trial (NCT02073994).

The gold standard for tumor genome sequencing projects is WGS of FF tumors, with high tumor content. This is not always feasible; therefore, projects have also used WES or targeted gene panels, and FFPE tumor blocks to derive DNA for analyses. This project systematically brought together studies using each of these methods (1, 13, 15–60) to provide the most comprehensive review of mucosal melanoma genomics to date.

The frequency of common drivers BRAF p.V600 (50% vs. 3–5%), NRAS hotspots (∼20% vs. 9–12%), NF1 LoF (∼15% vs. 23–44%), SPRED1 LoF (∼5% vs. ∼15%), KIT hotpots (7% vs. ∼25%), and SF3B1 hotspot (5% vs. 7–25%) notably differ between cutaneous and mucosal subtypes, respectively. This underlines the fact they are melanomas with different driver mechanisms and generalizations uncovered from cutaneous melanoma may not be applicable to other rare subtypes. For example, while the MAPK pathway is still activated, mucosal melanoma tends to carry atypical mutations in BRAF rather than at the cutaneous melanoma hotspot (p.V600). Instead, variants conferring high kinase activity, such as p.G469A and p.K601E, or with lower kinase activity, including p.L597Q, p.N581S, and p.T599I were identified (148). Some lower activity mutants, such as p.G596R, are still capable of signaling via CRAF activation (149). The differences in the mutation frequencies of BRAF are clinically important, as the majority of mucosal melanomas lack the target for the p.V600 based inhibitors that have revolutionized cutaneous melanoma treatment. While these inhibitors have yet to be systematically investigated in a large cohort of BRAF p.V600–positive mucosal melanomas, a small study (n = 19) on combined acral and mucosal melanomas showed an overall median progression-free survival (PFS) of 7.3 months (95% CI, 3.0–11.6), compared with 17.5 months (95% CI, 0.1–34.9) for cutaneous melanoma; however this was not significant (P = 0.429; ref. 150). Another small study assessed BRAF inhibitors (median PFS: 4.4 months; 95% CI, 0.8–12.7) compared with chemotherapy (median PFS: 1.1 months; 95% CI, 0.1–3.3) in a small mucosal melanoma cohort (n = 12; ref. 151). There potentially could be tissue specific contexts to the presence of this variant that renders mucosal melanoma less sensitive to these inhibitors (78); further genomic and functional studies are required to uncover these intricacies.

Compiling mutations from targeted hotspot sequencing revealed rare KIT mutations that were only sporadically present in the main cohort, demonstrating the utility of performing meta-analysis/systematic review. Functional evidence has shown that the two most common mutations found in mucosal melanoma, p.K642E and p.L576P, are activating (36, 152). Furthermore, evidence suggests KIT mutations can elicit diverse capabilities in activating signaling pathways, meaning they have different effects on cellular processes. For example, p.V559D and p.D816H induced stronger activation of the MAPK, PI3K/mTOR, and p38-MAPK pathways compared with p.K642E, and both p.D816H and p.D816V were able to phosphorylate STAT3 without ligand stimulation (153, 154).

The observation that ATRX and SF3B1 mutations occurred more commonly in lower anatomy mucosal melanoma, and CTNNB1 in the upper anatomy is of potential biological and clinical relevance, requiring further investigation. The presence of SF3B1 variants have been associated with worse survival in mucosal melanoma (155), while CTNNB1 and ATRX have been associated with worse prognosis in other cancer types (156, 157), indicating an association between gene aberration and clinical outcome.

The main cohort described samples from WGS/WES FF projects (15–20, 59); unfortunately, the data from Zhou and colleagues (20) was not publicly available, limiting analyses of the CNAs and SVs present in the main cohort to two studies (18, 59).There are genes not on the MSK-IMPACT panel, or only included in the second/third iteration of the panel, meaning the full cohort was not examined for all genes of interest to mucosal melanoma. Notable differences in variant frequency between the main cohort and FFPE validation cohort were apparent, in line with previously described comparisons (158). Furthermore, aggregating WES and WGS data may introduce some biases due to differences in sequencing depth and coverage; however, we did not observe any notable disparities in the data. Most variants can be detected by both sequencing approaches due to good technique sensitivity (159, 160). These caveats demonstrate the importance of gathering data across all available cohorts, to provide the most comprehensive analysis. All the studies were on bulk DNA, making it difficult to identify heterogeneity, as subclonal aberrations are potentially missed. Single-cell sequencing becoming more commonplace will allow resolution of tumor heterogeneity and may help to further refine treatment approaches based on genomic targets (161).

Investigation of genomic changes in the context of cancer hallmarks reveals that the tumorigenic processes are often driven by overlapping components in the same tumor. These observations could help identify vulnerabilities for targeted treatment approaches. It is also important to acknowledge that genomic alterations are not the only way that tumor cells acquire changes in cellular processes that lead to the Hallmarks of Cancer (77). Now the main genomic changes in mucosal melanoma have been described, future approaches to characterize these tumors should involve a combination of genomic, transcriptomic, and proteomic analyses to better understand the complexities of the cellular effects and potential vulnerabilities conferred by these key changes.

N.K. Hayward reports grants from NHMRC during the conduct of the study. No disclosures were reported by the other authors.

N. Broit: Conceptualization, data curation, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. P.A. Johansson: Data curation, formal analysis, supervision, methodology, writing–review and editing. C.B. Rodgers: Visualization, writing–review and editing. S.T. Walpole: Visualization, writing–review and editing. F. Newell: Formal analysis, methodology, writing–review and editing. N.K. Hayward: Conceptualization, supervision, writing–review and editing. A.L. Pritchard: Conceptualization, data curation, formal analysis, supervision, visualization, methodology, writing–original draft, writing–review and editing.

N. Broit is supported by the Australian Government Research Training Program (RTP) Fee-Offset Scholarship through the University of Queensland, the QIMR Berghofer higher degree candidate scholarship, and the Lynette Wei Hung Wo PhD Top-Up Scholarship; we particularly thank the family and friends of Lynette for establishing this scholarship in her memory, which has provided financial support toward this study and further research into mucosal melanoma biology. N. Broit, P.A. Johansonn, S.T. Walpole, N.K. Hayward are supported by the National Health and Medical Research Council of Australia. A.L. Pritchard is supported by the Highlands and Islands Enterprise (HMS9353763). The authors further wish to acknowledge the following for their contribution to this study. We thank Dr. David Adams at the Wellcome Sanger Institute for his advice and support in data collection and downstream analyses. Sample information from the Hintzsche and colleagues, 2017 study was provided by Carol M. Amato and Robert Van Gulick from the University of Colorado. We are grateful to Dr. Anand Mayakonda for technical assistance in altering the code in Maftools to allow visualization of genomic data. This study makes use of data provided by cBioPortal and GENIE and the Cancer Research UK Genomics Initiative. We particularly wish to thank the patients who selflessly donate their tissue and time to studies in medical research at a personally very difficult time.

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.