Targeted Genomic Profiling of Acral Melanoma

Abstract

Background

Acral melanoma is a rare type of melanoma that affects world populations irrespective of skin color and has worse survival than other cutaneous melanomas. It has relatively few single nucleotide mutations without the UV signature of cutaneous melanomas, but instead has a genetic landscape characterized by structural rearrangements and amplifications. BRAF mutations are less common than in other cutaneous melanomas, and knowledge about alternative therapeutic targets is incomplete.

Methods

To identify alternative therapeutic targets, we performed targeted deep-sequencing on 122 acral melanomas. We confirmed the loss of the tumor suppressors p16 and NF1 by immunohistochemistry in select cases.

Results

In addition to BRAF (21.3%), NRAS (27.9%), and KIT (11.5%) mutations, we identified a broad array of MAPK pathway activating alterations, including fusions of BRAF (2.5%), NTRK3 (2.5%), ALK (0.8%), and PRKCA (0.8%), which can be targeted by available inhibitors. Inactivation of NF1 occurred in 18 cases (14.8%). Inactivation of the NF1 cooperating factor SPRED1 occurred in eight cases (6.6%) as an alternative mechanism of disrupting the negative regulation of RAS. Amplifications recurrently affected narrow loci containing PAK1 and GAB2 (n = 27, 22.1%), CDK4 (n = 27, 22.1%), CCND1 (n = 24, 19.7%), EP300 (n = 20, 16.4%), YAP1 (n = 15, 12.3%), MDM2 (n = 13, 10.7%), and TERT (n = 13, 10.7%) providing additional and possibly complementary therapeutic targets. Acral melanomas with BRAF^V600E mutations harbored fewer genomic amplifications and were more common in patients with European ancestry.

Conclusion

Our findings support a new, molecularly based subclassification of acral melanoma with potential therapeutic implications: BRAF^V600E mutant acral melanomas with characteristics similar to nonacral melanomas that could benefit from BRAF inhibitor therapy, and non-BRAF^V600E mutant acral melanomas. Acral melanomas without BRAF^V600E mutations harbor a broad array of therapeutically relevant alterations. Expanded molecular profiling would increase the detection of potentially targetable alterations for this subtype of acral melanoma.

Acral melanomas originate from non-hair-bearing skin of the palms, soles, or nail apparatus. Although it has similar absolute incidence rates across racial-ethnic groups worldwide, it is the most common melanoma subtype in African and Asian populations (1,2). Acral melanoma has poor survival (3) and accounts for a considerable share of melanoma-associated morbidity and mortality worldwide. Whereas melanomas on non-acral skin have a high number of UV-induced mutations (4), acral melanomas have few point mutations, instead harboring increased numbers of structural rearrangements and amplifications (5–7).

It is well established that acral melanomas are driven by mutations in BRAF, NRAS, KIT, and amplifications of CCND1, CDK4, MITF, and TERT (8). Individual studies have nominated kinase fusions as drivers in acral melanoma (9,10). Although BRAF mutations are common in cutaneous melanoma and an important therapeutic target (11), their frequency in acral melanoma is considerably lower (∼20% vs ∼50%). BRAF^V600E mutations are common in melanocytic nevi, benign neoplasms that are a risk factor and precursor for melanomas. By contrast, acral melanomas mostly arise from melanoma in situ instead of nevi. Acral melanoma in situ can be subtle with occult disease extending into apparently normal skin, characterized by normal-appearing melanocytes that share pathogenic alterations with adjacent melanoma (8). Thus, the progression of acral melanomas has distinct clinical and genetic features. To provide further insight into the mechanisms of acral melanoma development, we catalogued alterations of genes frequently altered in 122 acral melanomas to further characterize the genetic features of acral melanoma and identify additional therapeutic targets.

Methods

Cases

We searched the archives of Kaiser Permanente Northern California, University of California, San Francisco, Fox Chase Cancer Center, and University of Colorado, Denver for melanoma occurring on the hands and feet. By review of the pathology report and case material, we confirmed the primary melanoma arose on glabrous skin of the palms, soles, or nail unit. We selected the earliest stage of acral melanoma available for each patient (primary, local recurrence, nodal metastasis, distant metastasis). We reviewed a total of 194 cases and excluded those with limiting tumor material (n = 69). We obtained formalin-fixed, paraffin-embedded tissues or DNA extracted from fresh frozen tissue from Kaiser Permanente Northern California (n = 70), University of California, San Francisco (n = 37), Fox Chase Cancer Center (n = 12), and University of Colorado, Denver (n = 6). The study was approved by the Committees on Human Research at each institution and conducted according to the Declaration of Helsinki. Informed consent or a waiver of consent was obtained for each patient. Study data were anonymized to protect subjects’ identities.

Sequencing

DNA extracted from tumor tissue was used for sequencing library preparation. Bait libraries targeted the coding regions and select introns of 293 (version 1), 365 (version 2), or 511 (version 3) genes. Sequencing was performed as paired-end 100-bp reads on a HiSeq2000 or 2500 (Illumina, San Diego, CA).

Data were analyzed using Burrows-Wheeler aligner (12), Picard (13), the Genome Analysis Toolkit (14), FreeBayes (15), Delly (16), Annovar (17), THetA2 (18), and CNVkit (19). We excluded samples with mean target coverage below 50× (n = 3, Supplementary Figure 1, available online).

We reviewed truncating mutations and mutations annotated as pathogenic in ClinVar. For genes in which a pathogenic alteration was identified, we annotated all variants of that gene.

Estimation of ancestry composition was performed using 1000 Genomes data (http://www.internationalgenome.org/) (20) and projection analysis by ADMIXTURE (21).

A subset of tumors that lacked RAS/RAF/KIT activating mutations was profiled by targeted RNA-Seq to identify oncogenic fusions (n = 5). Hybrid selection was performed using SureSelect Human Kinome or All ExonV4 kits (Agilent, Santa Clara, CA, p/n 5190–4801, 5190–4631). Sequencing was performed as paired-end 100-bp reads on a HiSeq2500 instrument (Illumina). Fusion detection was performed with Tophat Fusion (22).

Immunohistochemistry

Tissue microarrays were created using cases that had additional tumor material with 2.5-mm cores. NF1 immunohistochemistry was performed using the monoclonal antibody NFC as previously described (23). p16 immunohistochemistry was performed using the CINtec p16 Histology assay (Ventana, Oro Valley, AZ, p/n 725–4713) with dilution of the working antibody 1:10.

Statistical Analysis

Continuous variables were evaluated between conditions using Wilcoxon rank sum tests. Pairs of binary variables were compared using Fisher’s exact test. Tests for mutual exclusivity between types of events were performed using elementary probability and under the condition of independence. A multivariable Cox regression model was used to estimate unadjusted hazard ratios and 95% confidence intervals for melanoma-specific mortality. Tests with P values less than .05 were considered statistically significant. All statistical tests were two-sided.

Results

BRAF, RAS, and KIT Mutations

In total 122 acral melanomas (115 primary, seven metastatic, 44.3% male, 55.7% female) were sequenced with a median average target coverage of 289 (Supplementary Table 1, available online). Activating mutations in BRAF (n = 26, 21.3%), RAS isoforms (n = 39, 32.0%), and KIT (n = 14, 11.5%) occurred in a mutually exclusive pattern (P < .001; Figure 1). Most BRAF mutations resulted in V600E substitution (n = 21, 80.8%) with infrequent V600K (n = 3), K601E (n = 1), and G469S (n = 1) substitutions. NRAS was the most frequently mutated RAS isoform (n = 34, 27.9%) with most NRAS mutations affecting codon Q61 (n = 26) and the remainder affecting codons G12 or G13 (n = 8). Activating mutations in HRAS or KRAS occurred in less than 5% of cases.

The mutant RAS allele was amplified in four of the 34 (11.8%) tumors with mutant NRAS, all three tumors with mutant KRAS, and one of two (50.0%) tumors with mutant HRAS. Wild-type KRAS was amplified in six cases (4.9%), four of which had no mutation in other RAS isoforms, BRAF, or KIT. Activating KIT mutations were equally distributed between the juxtamembranous and kinase domains of KIT and amplified in 35.7% of KIT mutant cases (Figure 2).

Kinase Fusions

Structural rearrangements resulted in kinase fusions that are known or predicted to activate the MAPK pathway in eight (6.6%) cases. None of these cases had activating mutations in BRAF, RAS genes, or KIT. The BRAF fusion genes (n = 3, 2.5%) retained the BRAF kinase domain with loss of the autoinhibitory domain and included AGK-BRAF and ERC1-BRAF, as previously described (24,25), and a novel CNTNAP2-BRAF fusion (Figure 3A;Supplementary Table 2, available online). Similar BRAF fusions occur in Spitz tumors (26).

Fusion genes involving receptor tyrosine kinases occurred in four cases (3.3%), three involving NTRK3 (2.5%) (Figure 3B) and one involving ALK (0.8%). They all contained an intact kinase domain and consisted of MYO5A-NTRK3 (27), previously described in Spitz tumors, and novel TUBGCP3-NTRK3 and SDHA-ALK fusions. The SDHA-ALK fusion gene joined the first intron of SDHA to intron 17 of ALK, resulting in a predicted out-of-frame transcript and underwent secondary amplification. We suspect the SDHA-ALK fusion contains an alternative translation start site that results in a truncated form of the ALK protein with an intact kinase domain.

An ATP2B4-PRKCA fusion transcript encoding the kinase domain of protein kinase C alpha was identified by RNA-Seq in a tumor (0.8%) with wild-type BRAF, RAS genes, and KIT (Figure 3C). Similar ATP2B4-PRKCA fusions lacking the autoinhibitory domains of PRKCA have been reported in pigmented epithelioid melanocytomas (28,29).

RASopathy-Associated Mutations

The RASopathies are genetic syndromes caused by germline mutations in genes involved in MAPK pathway signaling. They include neurofibromatosis and Legius syndrome, which are caused by loss-of-function mutations in NF1 or SPRED1, respectively (30). NF1 promotes RAS deactivation by stimulating its GTPase activity. Losses of NF1 function resulting from missense mutations, structural rearrangements, or deletions occurred in 16 (13.1%) cases and affected both NF1 alleles in 10 (8.2%) cases. NF1 expression by immunohistochemistry in evaluable cases was extinguished in all three cases with bi-allelic inactivation, two of four cases with monoallelic truncating mutations, three of 12 (25.0%) cases with single copy deletion of NF1, but only two of 65 (3.1%) cases without identifiable genetic alterations of NF1 (Figure 4A;Supplementary Table 3, available online). These results indicate that in some cases, inactivation of NF1 occurred by mechanisms not detected by our assay, which could include cryptic rearrangements or epigenetic modifications. Altogether, NF1 inactivation, evidenced by bi-allelic genetic alterations and/or loss of protein expression, affected 18 (14.8%) cases.

SPRED1 is required for NF1’s negative regulation of RAS activity (30) and has recently been identified as a tumor suppressor in mucosal melanoma (31). Loss of function alterations affecting both alleles of SPRED1 were found in eight (6.6%) cases, none of which harbored NF1 inactivation. Bi-allelic losses of function of NF1 or SPRED1 were mutually exclusive with BRAF^V600E mutations (P = .001), but concomitant activating alterations in KIT, NRAS, or KRAS were often present (Figure 4B). Notably, either NF1 or SPRED1 were lost in six of 14 (42.9%) melanomas with KIT mutations (P = .04), indicating possible cooperativity between these alterations.

RASopathy-associated mutations in PTPN11 (n = 3, 2.5%), MAP2K1 (n = 1, 0.8%), or RIT1 (n = 1 of 84, 1.2%) were identified in 4% of acral melanomas. The three PTPN11 mutations (R265Q, Y279C, and A461T) cause Noonan syndrome. Activating mutations in RIT1 and MAP2K1 also cause Noonan syndrome (30).

Recurrent Amplifications and Deep Deletions

Many acral melanomas had highly rearranged genomes with amplifications and deep deletions, as previously described (Supplementary Figure 2, available online) (5).The long arm of chromosome 11 harbored three distinct amplicons, located at 11q14 containing PAK1 and GAB2 (n = 27, 22.1%); at 11q13 containing CCND1, encoding cyclin D1 (n = 24, 19.7%); and at 11q21 containing YAP1 (n = 15, 12.3%) (Supplementary Figure 3, available online). The regions were co-amplified in many tumors but amplified in isolation in some cases. PAK1 and GAB2 are within 1 megabase of each other and were always co-amplified. PAK1 mediates MAPK and Hippo pathway signaling (32,33), whereas GAB2 overexpression increases PI3K signaling and potentiates melanoma invasion (34). YAP1 is an effector of the Hippo pathway, which regulates contact inhibition of cells, and its amplification and overexpression leads to dysregulated cell growth (35).

Amplifications on chromosome 12 affected CDK4 at 12q13 in 27 (22.1%) cases and MDM2 at 12q15 in 13 (10.7%) cases. MDM2 was co-amplified with CDK4 in more than one-third of CDK4 amplified cases; however, amplification of MDM2 without amplification of CDK4 occurred in four (3.3%) cases.

EP300, on chromosome 22, was amplified in 20 (16.4%) cases. It encodes p300, a transcriptional coactivator with lysine acetyltransferase and E4 ubiquitin ligase activities (36). p300 is a cofactor for MITF (37) and also upregulates p53 transcriptional activity or leads to increased degradation of p53, depending on its cellular context (38,39).

Bi-allelic inactivation of CDKN2A occurred in 15 (12.3%) cases, in 14 cases by homozygous deletion. Because CDKN2A can be silenced by epigenetic mechanisms, we evaluated expression of its protein product p16 by immunohistochemistry for 85 cases. p16 was not detectable in 45.9%, including all cases with bi-allelic loss, approximately one-half of cases with shallow deletions, and one-third of cases without detected alterations of CDKN2A (Supplementary Table 3, available online). These results suggest that, similar to what we observed for NF1, p16 expression can be disrupted by mechanisms not detected by our assay.

Cyclin Dependent Kinase 4/6 Pathway Disruption

CDK4 and CDK6 drive G1 cell-cycle progression by phosphorylating Rb. Both are activated by D-type cyclins and inhibited by p16, p15, and other CDK inhibitors. p16 loss was identified in 36.1% of all cases, either by bi-allelic disruption of CDKN2A or loss of p16 expression by immunohistochemistry (Figure 5). p15 is encoded by CDKN2B, which is located near CDKN2A at 9p21. Loss-of-function single nucleotide variants of CDKN2B were not identified, but deep deletions affecting CDKN2A also involved CKDN2B in all but one case. CCND2, which encodes cyclin D2, was amplified in five cases (4.1%) in a mutually exclusive pattern with CCND1 amplification (P < .001). CDK6 amplification occurred in five cases (4.1%), and in one case CDK4 amplification was also present. Loss-of-function mutations in RB1 were not identified. Altogether, activating alterations of the CDK4/6 pathway were identified in 62.3% of cases, with many cases demonstrating alterations in multiple components of the pathway.

p53 Pathway Disruption

The p53 pathway is considered the “guardian of the genome” and triggers cell cycle arrest or apoptosis in response to DNA damage. Although TP53 is frequently mutated in cancer, melanoma has a comparatively low rate of TP53 mutation. The high number of structural rearrangements and copy number changes in acral melanoma suggests corruption of the guardian function of the p53 pathway. Nevertheless, TP53 point mutations were rare in our cohort with only a single pathogenic mutation (H179Q, 0.8%). p53 is antagonized by MDM2, which not only ubiquitinates p53—thereby targeting it for destruction—but also directly binds p53 to inhibit its transcriptional activity (40). MDM4 is homologous to MDM2 and can also bind to p53 and negatively regulate its transcriptional function (41). MDM4 was amplified in one case (0.8%) in addition to the 13 cases with MDM2 amplification. The ability of the p53 pathway to respond to DNA damage can also be disrupted by loss of ATM, which phosphorylates p53 in response to double-stranded DNA breaks. Bi-allelic inactivation of ATM occurred in six (4.9%) cases, and an additional case demonstrated a single truncating ATM mutation. p14ARF, another gene product of CDKN2A, inhibits MDM2, and thus loss of CDKN2A may contribute to p53 pathway dysfunction. p300 has various effects on p53 activity and increased cytoplasmic p300 leads to decreased p53 levels (38). Altogether, genetic disruption of the p53 pathway was identified in 39.3% of acral melanomas (Figure 6).

TERT Alterations

The majority of melanomas on sun-exposed skin harbor TERT promoter mutations (8). TERT promoter mutations were uncommon in acral melanoma (n = 4 or 5.3% of 75 cases for which TERT promoter could be genotyped). Instead, the TERT locus was amplified in 13 (10.7%) cases. Almost one-half of the TERT amplifications involved a copy number transition within 40 kilobases upstream of TERT, indicating that they may alter promoter or enhancer sequences to increase TERT expression (42,43). An additional four (3.3%) cases harbored copy number transitions within 40 kb upstream of TERT with relative gain of TERT, indicating they too may contain altered promoter or enhancer sequences.

Pathways Recurrently Mutated in Melanoma

Previous genomic studies of melanoma identified mutations affecting various biological pathways (6,44). Although only some members of these pathways were genotyped (Supplementary Table 4, available online), alterations were identified in the histone modification (n = 55, 44.3%), receptor tyrosine kinase (n = 33, 27.0%), WNT signaling (n = 26, 21.3%), PI3K (n = 19, 15.6%), SWI/SNF chromatin remodeling (n = 9, 7.4%), and DNA methylation pathways (n = 4, 3.3%) (Supplementary Figure 4, available online).

Associations Between Clinical and Genetic Features

Melanoma-specific mortality among patients with acral melanoma is associated with tumor thickness and stage at presentation, but not with race or ethnicity (45). Here, in a multivariable Cox survival model predicting melanoma-specific mortality for patients with long-term follow-up information (n = 67) based on BRAF, NRAS, KIT, CDKN2A, CCND1, CDK4, and EP300 mutation status; age at diagnosis; and sex, only CDKN2A bi-allelic inactivation was statistically significantly associated with mortality (hazard ratio = 4.46, 95% confidence interval = 1.11 to 17.87, P = .03).

Subungual melanomas may have a particularly poor prognosis (46,47). KIT mutation was identified in three of six subungual melanomas, at a greater frequency than in nonsubungual acral melanomas (P = .03), whereas BRAF or NRAS mutations were not associated with subungual location.

Two Subtypes of Acral Melanoma

The BRAF^V600E mutations seen in a subset of acral melanomas are typical of melanomas on sun-exposed but not chronically sun-damaged skin (low-CSD melanomas) typically affecting younger individuals of light complexion (8,48). These melanomas often arise from nevi that harbor BRAF^V600E. An adjacent nevus was identified in one of the 21 acral melanomas with BRAF^V600E mutation. We tested whether BRAF^V600E mutations are associated with age, sex, European ancestry, and number of genomic amplifications. BRAF^V600E mutations were associated with female sex (P < .001) and fewer copy number amplifications (mean 1.9 vs 6.8 amplifications per tumor, P < .001). Although patients with BRAF^V600E mutations had higher levels of European ancestry, this did not reach statistical significance (P = .09). Across mutational studies of acral melanoma in Asian and African populations, BRAF^V600E mutations occur in 13.0% compared to 17.2% in our cohort, but this difference was not statistically significant (P = .34) (9,49–51). There was no statistically significant difference in age at diagnosis between acral melanomas with or without BRAF^V600E mutation. NRAS and KIT mutations did not show these associations.

Discussion

Our findings suggest the existence of two types of melanoma originating from acral skin. BRAF^V600E mutant acral melanomas are similar to melanomas occurring in nonchronically sun-damaged skin (low-CSD melanomas) in that they have fewer DNA copy number changes, may arise from precursor nevi, and European ancestry appears to be a risk factor. Acral melanomas without BRAF^V600E mutation comprised 82.8% of cases and are characterized by a broad array of other MAPK activating mutations, a high number of amplifications and rearrangements, and lack of a specific racial predilection. There are several indications that these acral melanomas correspond to what Coleman et al. (52) described as acral lentiginous melanoma. Their characteristic lentiginous growth pattern refers to an in situ component in which increased numbers of single melanocytes are aligned along the basilar epidermis, as opposed to the pagetoid growth pattern of superficial spreading melanoma. Melanomas that arise on chronically sun-damaged skin (high-CSD) also frequently display a lentiginous growth pattern, lack BRAF^V600E mutations, and have loss-of-function mutations in NF1 (53) and activating KIT mutations (54). However, they differ from acral melanomas by having a very high point mutation burden and fewer copy number aberrations (53). Several of the novel alterations identified in our study can be linked to the lentiginous growth pattern. Syndromes due to pathogenic mutations in NF1, SPRED1, PTPN11, or RIT1 are associated with lentigines, café-au-lait macules (30), and high-CSD melanomas (53), all of which have a lentiginous growth pattern, suggesting that these alterations cause lentiginous proliferations of melanocytes.

We identified an extended spectrum of MAPK activating alterations in acral melanomas, including several fusion kinases for which drugs are already in the clinic. Trk inhibitors are showing promising results in early clinical studies in solid tumors with NTRK fusions (55), and tumors with BRAF fusions including melanomas have responded to BRAF or MEK inhibitors (11). In preclinical studies, MEK inhibitors are effective against some RAS/RAF wild-type melanomas with loss of NF1 (56). Thus, loss-of-function mutations in SPRED1 or other RASopathy associated mutations would be expected to similarly predict response to MEK inhibition. Loss of NF1 or SPRED1 in a clinically significant fraction of KIT mutant acral melanomas suggests that combining KIT inhibitors with MEK inhibitors may lead to improved outcomes for these tumors.

Forty percent of acral melanomas in our study lacked targetable MAPK pathway alterations. Although most of these cases harbored activating RAS mutations, a subset had no identifiable MAPK activating alteration. Therapeutic targets beyond the MAPK pathway are of particular interest for these tumors as well as for use in combination with existing therapies to prevent drug resistance. Based on our genetic findings, CDK4/6 inhibitors, inhibitors of the MDM2-p53 interaction, and Hippo pathway inhibitors may be effective in acral melanomas with defects in the corresponding pathways.

Our study was limited because we sequenced only targeted regions of the genome and therefore could not discover novel pathogenic point mutations or gene rearrangements. Additionally, we did not sequence normal DNA from the patients and thus were unable to precisely determine if variants were germline or somatic. To compensate for this, we performed careful annotation and analysis to identify pathogenic or likely pathogenic variants and did not include variants of unknown significance in our analysis.

In conclusion, our study indicates the existence of two distinct melanoma subtypes on acral sites. Acral lentiginous melanomas characterized by the absence of BRAF^V600E mutations and many amplifications and structural rearrangements represent the majority of acral melanomas and display considerable diversity of oncogenic drivers, many potentially actionable, which should motivate comprehensive tumor profiling for patients in need of systemic treatment to identify therapeutic targets. By contrast, acral melanomas with BRAF^V600E mutations are likely similar to low-CSD melanomas and likely can be treated similarly.

Funding

This work was supported by the Melanoma Research Alliance; Dermatology Foundation; the National Cancer Institute at the National Institutes of Health (grant nos. 5P30CA082103, 1R35CA220481); and the National Institute of Arthritis and Musculoskeletal and Skin at the National Institutes of Health (grant no. K24 AR069760).

Notes

Affiliations of authors: University of California, San Francisco, CA (IY, MX, JPN, AHS, AO, PYK, BCB); Kaiser Permanente Division of Research, Oakland, CA (EJ, LS); Neuropathology, University of Heidelberg and German Cancer Research Center (DKFZ), Heidelberg, Germany (DR, AvD); Fox Chase Cancer Center, Philadelphia, PA (HW); University of Colorado, Denver, CO (WAR); Department of Dermatology, Massachusetts General Hospital, and Department of Population Medicine, Harvard Pilgrim Health Care Institute, Harvard Medical School, Boston, MA (MMA).

The funders had no role in the design of the study; the collection, analysis, and interpretation of the data; the writing of the manuscript; and the decision to submit the manuscript for publication. Drs Yeh and Bastian report grant funding to the University of California, San Francisco from Ignyta, Bayer, and Novartis to support a phase 2 clinical trial on kinase fusions in melanoma. Dr North is a consultant for AdviNOW Medical, outside the submitted work. Dr Reuss and Dr von Deimling report having licensed the NF1 antibody to Cell Marque Corporation. All terms are being managed by the University of Heidelberg in accordance with its conflict of interest policies. Dr Bastian is a consultant for Lilly Inc, outside the submitted work. Dr Asgari reports grant funding to Massachusetts General Hospital from Pfizer Inc and Valeant. The other authors have no disclosures.

We thank Monica Sokil, Alexander Gagnon, and Shanshan Liu for excellent technical assistance.

References

Author notes