Predominance of triple wild-type and IGF2R mutations in mucosal melanomas

The study involved clinically and surgically defined tumors that comprised of 41 MM and 48 CM from the John Wayne Cancer Institute (JWCI) tissue bank archive. All surgeries were performed at Saint John’s Health Center (SJHC), and specimens were identified by experienced melanoma surgical pathologists at SJHC, Dept. of Surgical Pathology. Detailed characteristics of specimens are shown in Additional file 1: Table S1. In MM patients, metastases from CM were ruled out as they had no medical history or evidence of CM or uveal melanoma. This study followed the principles in the Declaration of Helsinki. All human specimens and clinical information for this study, including informed consent, were obtained according to the protocol guidelines approved by the SJHC/JWCI Western Institutional Review Board.

Frozen tissues (n = 62) were homogenized with a sonicator and filtered using QIAshredder (QIAGEN, Valencia, CA), and DNA was extracted using the ZR-Duet DNA/RNA MiniPrep (Zymo Research, Irvine, CA), according to the manufacturer’s protocol. DNA extraction from formalin-fixed paraffin-embedded (FFPE) specimens (n = 27) was performed using ZR FFPE DNA MiniPrep (Zymo Research), as previously described [15]. DNA was quantified using UV spectrophotometer (BioTek, Winooski, VT) and Quant-iT PicoGreen dsDNA Assay Kit (ThermoFisher Scientific, Carlsbad, CA). For specimens contaminated with strong melanin content, OneStep PCR Inhibitor Removal Kit (Zymo research) was used for melanin removal. For tumor purity, we assessed frozen tissues that were dissected by a surgical pathologist from the original tumor surgery whereby a representative tissue was made into a FFPE tumor block. The majority (> 90%) of the cells in the frozen tissues were melanoma cells. For FFPE tumor block analysis, we assessed H&E stained slides and performed micro-dissection of melanoma tumor cells.

An Agilent Haloplex custom target enrichment kit that captured all exons in 89 genes related to melanoma and tumors from mucous membranes [5, 16‐19] was designed using the Agilent SureDesign software (Agilent Technologies, Santa Clara, CA). The genes on the panel are listed in Additional file 2: Table S2. Target-enriched libraries were constructed from genomic DNA (3 μg), following an instruction from HaloPlex Target Enrichment for Illumina Kit (Agilent Technologies) [13]. Only library fragments within 175 to 625 bp were considered for the final quantification, normalization, and pooling. The final multiplexed Haloplex custom target library pool was sequenced on the Illumina HiSeq 2500 (Illumina, San Diego, CA) on rapid mode using paired-end 100 bp reads. Overall, the panel showed high coverage rates (median 349x) for all the target regions.

Raw genomic sequence reads were mapped to the 1000 Genomes (b37) built of the human genome reference using BWA-MEM (version 0.7.5a) with default settings [20]. The resulting alignments were further processed using GATK (version 2.8–1) following the GATK Best Practices recommendations [21‐23]. Single nucleotide variants were identified using MuTect version 1.1.4 [24]. MuTect was run using default parameters in the High Confidence mode along with dbSNP (version 137) and COSMIC (version 67) databases on tumor only samples. Only SNVs that were classified as “KEEP” by MuTect were used for downstream analysis. Genomic annotations were performed using the ANNOVAR annotation pipeline [25]. In order to remove potential germline variants from our set of variants, we further filtered the variants using SnpSift [26]. Only variants that had a population frequency below 1%, mutation allele frequency > 10%, and coverage of >20X were used for further analysis. The final mutation call set had a mean depth of coverage of ~274X. Mutation spectrum and signatures analyses were done using the Bioconductor packages SomaticSignatures [27] and deconstructSigs [28], respectively. A total of 997 mutations that were located within exonic regions were analyzed for mutation spectrum and prevalence of mutations, and 685 non-synonymous mutations were investigated for identification of MM associated genes and pathways. MAPK pathway, one of the most biologically and clinically important pathways in melanoma [5], was considered to be mutated when there was at least one non-synonymous mutation in BRAF, HRAS, KRAS, MAPK2K1, NF1 or NRAS.

IHC was performed as previously described [29, 30], using anti-PD-L1 rabbit monoclonal antibody (1:100 dilution, #ab205921; Abcam). After the antigen retrieval step, the sections were incubated in 10 g/L trichloroisocyanuric acid (176,125; Sigma-Aldrich) solution for 30 min at room temperature to bleach the melanin [31]. Photographs were obtained using a Nikon Eclipse Ti microscope and NIS elements software (Nikon). Total of 39 specimens (21 MM and 18 CM) were available for immunohistochemistry. The expression of PD-L1 was quantified using an H score system, which considers both the intensity and percentage of positive cells [32]. Intensity of PD-L1 on tumor cell membrane was determined between 0 (no staining), 1 (weak), 2 (moderate), and 3 (strong). The score was calculated using the following formula; 1 x (% of 1+ cells) + 2 x (% of 2+ cells) + 3 x (% of 3+ cells) [32] .

Raw genomic sequence data obtained in this study can be accessed from NCBI SRA under Bioproject number PRJNA379027.

TCGA mutation frequency data was downloaded using the RTCGAToolbox Bioconductor package for run date “20,150,821” [33].

Continuous variables were assessed using Student’s t test or Wilcoxon rank-sum test, and categorical variables were assessed using χ² test or Fisher’s exact tests. FDR corrected p-value < 0.05 was considered to perform multiple test for 96 substitutions. Overall survival (OS) was analyzed based on the time since being diagnosed with melanoma using the Kaplan-Meier method and log-rank test. All statistical analyses were performed with JMP, version 11.0 (SAS Institute Inc., Cary, NC) or R (https://​www.​R-project.​org), and a two-sided p-value < 0.05 was regarded as statistically significant.

UV exposure is a major mutagenic factor that drives malignant transformation of melanocytes into CM [34]. However, MM is generally not exposed to UV due to its anatomical locations [1, 3, 8]. Epidemiologic studies have indicated potential risk factors for MM, including tobacco use for oral MM [4, 35]; however, no definitive risk factor has been identified. To investigate the involvement of UV or tobacco exposure in MM, we performed a mutational spectrum analysis for a total of 997 mutations on exons in 89 genes. Within six different substitutions, C > T substitutions, were the most predominant substitution type in both MM and CM (Fig. 1a). Importantly, the C > T substitutions were significantly less in MM (57.8%) than CM (64.8%, Fig. 1a, χ² test, p = 0.002). Despite the lower frequency of C > T substitutions in MM, they were still the most predominant substitution type in MM. Other factors, such as aging, induce C > T substitutions and therefore are the most prevalent substitutions in many cancers [16, 36], including other mucosal origin tumors such as gastric adenocarcinoma and colorectal cancer (CRC) [16, 19] and are not highly specific to UV damage [16, 36].

To further analyze the mutational spectrum in MM, information on the nucleotides immediately 5′ and 3′ to each mutated base (i.e. tri-nucleotide context) were incorporated into our analysis, making 96 possible substitutions [36, 37]. Eleven substitutions types, mostly C > T substitutions, significantly differed between MM and CM (Fig. 1b, t-test, FDR corrected p < 0.05). We categorized these 96 substitutions into 30 different signatures defined by the COSMIC (Catalogue of somatic mutations in cancer, http://​cancer.​sanger.​ac.​uk/​cosmic/​signatures, see methods) database. These 30 different signatures propose specific etiologies or mutational mechanisms that lead to specific signatures. As expected, the UV damage signature (signature 7) was less enriched in MM (Fig. 1c, Wilcoxon, p = 0.001), consisted with the C > T mutational spectrum analysis above. Besides the UV damage signature, eight MM specimens (19.5%) presented signatures for tobacco exposure (signature 4 and 29). Of the eight MM that displayed the tobacco related mutation signature, 5 were head & neck (29% of 17 head & neck cases), 1 was genital (10%), and 2 were anorectal (14%). While patient smoking history was not available for the individuals included in this study and while the precise mechanisms of how tobacco exposure might drive MM remains unknown, this mutational spectrum analysis suggests smoking is a potential pathogenic factor in MM.

Due to lower exposure of mutagenic UV light in MM (Fig. 1a, c), we speculated that mutations are less prevalent in MM than CM. CM, which exhibits the highest tumor mutation burden among different cancer types [36], demonstrated 14.8 mutations/Mb (median, 11.5 mutations per sample) in our cohort, while the number of mutations was significantly less in MM (median 6.5 mutations/Mb (5.0 mutations per sample), Fig. 2a, Wilcoxon, p = 0.001). We further investigated PD-L1 expression in both types, since PD-L1 expression is positively associated with tumor mutation burden [38] and predicts clinical benefit from immunotherapies in different tumors, including CM [39‐41]. In accordance with low tumor mutation burden, PD-L1 expression was significantly less in MM compared to CM (Fig. 2b, Wilcoxon, p = 0.003), suggesting that immune checkpoint inhibitors, such as nivolumab or pembrolizumab, should be carefully considered to apply to MM patients.

To investigate mutations of potential driver genes in MM, we analyzed 685 non-synonymous variants and evaluated prevalence of mutations in each gene. Previous studies have demonstrated that CM has a high prevalence of driver gene mutations, such as BRAF mutations or NRAS mutations [5, 7], while these driver mutations are less frequent in MM [3, 8]. In our cohort, mutations in BRAF, a major driver gene in CM [5], was significantly less frequent in MM than CM (12.2% vs 50%, Tables 1, 2, Fisher’s exact test, p < 0.001). Within BRAF mutations, 75% and 87% of were located in the BRAF^V600 hotspot mutation in our CM cohort and in the TCGA cohort [5], respectively. In contrast, the BRAF^V600 hotspot mutation was not present in MM. Frequency of NRAS mutations, another major driver mutation in CM, was also lower in MM (17.1%) than CM (29.2%) (Table 1). Moreover, other driver mutations (KRAS mutations 2.4%, HRAS mutations 2.4%, TP53 mutations 7.3%) were not prevalent in MM.

Although MM exhibited less mutations in common driver genes for CM, we identified mutations in two genes that were significantly more frequent in MM than CM (Table 2). Mutations in IGF2R, which is involved in the insulin-like growth factor (IGF) pathway [42], was the most frequent mutation in MM (31.7%) and was significantly more prevalent than CM (6.3%, Table 2, Fisher’s exact test, p = 0.002, Fig. 3a). Notably, the frequency of IGF2R mutations were higher than other cancer types from TCGA database (Fig. 3b). In accordance with previous reports [3, 8, 43], KIT mutations were also significantly more frequent in MM (9.8% vs 0%, Fisher’s exact test, p = 0.042) (Table 2). Overall, MM showed mutated genes that are distinct from CM and was significantly more frequent in IGF2R and KIT mutations.

We further classified the 89 samples into the four molecular subtypes, BRAF mutated (47% in melanoma TCGA), RAS mutated (NRAS/KRAS/HRAS mutations, 29%), NF1 mutated (9%), and Triple-WT (subgroup lacking above mutations, 15%), proposed by the recent TCGA melanoma cohort [5]. Notably, MM showed significantly higher prevalence of the Triple-WT subtype than CM (70.7% vs 25.0%, Fisher’s exact test, p < 0.001). In accordance with the high prevalence of Triple-WT in MM, mutations in the MAPK pathway (mutations in any of the following genes; BRAF, HRAS, KRAS, MAPK2K1, NF1 or NRAS) was significantly less frequent in MM than CM (36.6% vs 81.3%, Fisher’s exact test, p < 0.001). To further characterize the Triple-WT in MM, we compared Triple-WT (n = 29) and non-Triple-WT (n = 12) in MM. Despite the difference in driver gene mutations between Triple-WT and non-Triple-WT, there was no significant difference in their UV damage signatures (C > T substitutions, χ² test, p = 0.35; signature 7, Wilcoxon, p = 0.39). Accordingly, the number of mutations were similar in both types (median 6.5 mutations/MB (5.0 mutations per sample) for both types, Wilcoxon, p = 0.78). Furthermore, there was no significant difference in overall survival (OS; log rank test, p = 0.43). Despite the similarity in the above mutational landscapes between Triple-WT and non-Triple-WT in MM, prevalence of Triple-WT was different between anatomical subtypes of MM. Dividing the MM into three major anatomic groups, genital (n = 10), head & neck (n = 17), and anorectal (n = 14) melanoma, frequencies of Triple-WT were 90.0%, 70.6% and 57.1%, respectively, suggesting distinct genetic background underlining MM anatomic subtypes.

We further investigated any potential prognostic markers for MM. DCC mutations were observed only in five patients from MM (12.2%) and was less prevalent in MM than in CM (Table 2). However, the presence of DCC mutations was significantly associated with poor OS in MM (Fig. 4, log-rank test, p = 0.02). In contrast, although DCC mutations were significantly more frequent in CM than MM (41.7% vs 12.2%), DCC mutations in CM was not associated with patients’ prognosis (log-rank test, p = 0.87). These results suggest the potential of DCC mutations as a potential prognostic marker in MM.

Melanomas in the anorectal region, primary anorectal MM or CM that has metastasized to the bowel, are rare diseases [44], and distinguishing these two types is occasionally challenging due to similarity in their anatomical sites [45]. Definite diagnosis for these types is clinically important, since CM has been well investigated for different treatment strategies, such as molecular targeted therapies or immunotherapies [7, 40, 41]. MM demonstrated a distinct mutational spectrum and tumor mutation burden from CM (Fig. 1a, c), thus we speculated that targeted NGS would enable us to distinguish between these types. Similar to Fig. 1a, CM metastasized to the bowel (n = 10) displayed significantly higher prevalence of C > T substitutions compared to anorectal MM (n = 14) (Fig. 5a, 76.2% vs 47.9%, χ² test, p < 0.001). Accordingly, mutations were significantly more frequent in CM metastasized to the bowel than in anorectal MM (Fig. 5b, median 15.5 mutations/Mb (12 mutations per sample) vs 5.2 mutations/Mb (4 mutations per sample), Wilcoxon test, p = 0.002). Overall, mutational analyses were able to distinguish primary anorectal MM from CM metastasized to the bowel.