Combined activation of MAP kinase pathway and β-catenin signaling cause deep penetrating nevi

Deep penetrating nevus (DPN) is characterized by enlarged, pigmented melanocytes that extend through the dermis. DPN can be difficult to distinguish from melanoma but rarely displays aggressive biological behavior. Here, we identify a combination of mutations of the β-catenin and mitogen-activated protein kinase pathways as characteristic of DPN. Mutations of the β-catenin pathway change the phenotype of a common nevus with BRAF mutation into that of DPN, with increased pigmentation, cell volume and nuclear cyclin D1 levels. Our results suggest that constitutive β-catenin pathway activation promotes tumorigenesis by overriding dependencies on the microenvironment that constrain proliferation of common nevi. In melanoma that arose from DPN we find additional oncogenic alterations. We identify DPN as an intermediate stage in the step-wise progression from nevus to melanoma. In summary, we delineate specific genetic alterations and their sequential order, information that can assist in the diagnostic classification and grading of these distinctive neoplasms.

C ommon nevi are benign melanocytic neoplasms located superficially in the skin. Nevus cells become smaller and less pigmented as their distance from the surface epithelium increases. This characteristic is termed 'maturation' and is an important diagnostic feature to distinguish benign nevi from melanoma 1,2 . In 1989 Helwig and colleagues described a neoplasmdeep penetrating nevus (DPN)as an exception to this rule in which the constituent melanocytes maintain their cell size and pigmentation throughout the dermis 3 . Before their initial description, these microscopic characteristcs resulted in the misdiagnosis of almost one-third of DPN as melanoma. While the establishment of DPN as a diagnostic entity has reduced the number of cases falsely classified as melanoma, tumors with the characteristic phenotypic features can present diagnostic challenges, and occasional tumors classified as DPN metastasize with fatal outcome 4 .
While the majority of common nevi are clonal proliferations of BRAF V600E mutant melanocytes 5,6 , the genetic drivers of DPN are largely unknown. DPN have overlapping phenotypic features with blue nevi in that they are mainly dermal based and contain numerous melanophages 7 . However, they lack the GNAQ and GNA11 mutations that are found in the majority of blue nevi 8 . DPN also have overlapping phenotypic features with Spitz nevi in that they have large melanocytes with abundant cytoplasm and have previously been shown to harbor HRAS mutations in 6% of cases 8  DNA from 18 DPN and 10 common nevi, using a platform that includes exons of several hundred cancer genes and introns involved in oncogenic rearrangements. We sequenced additional nevi with overlapping features of DPN and blue nevi. The resulting sequence data was analyzed for single-nucleotide variants, insertion/deletions (indels), structural variants and copy number changes. We identify MAPK activating mutations in combination with β-catenin activating mutations as a characteristic feature of DPN. We demonstrate that CTNNB1 mutations arise after BRAF mutation in DPN arising from common acquired nevi and confer the characteristic phenotypic features of DPN. In DPN-like melanomas, we identify the presence of additional genetic alterations indicating that DPN are genetically intermediate between benign nevi and DPN-like melanomas.
Results β-catenin and MAPK pathway mutations define DPN. Cases with typical histological features of DPN were genetically distinct from common nevi and nevi whose features overlapped between DPN and blue nevus. 17 out of 18 (94%) DPN had activating mutations of the β-catenin pathway, mostly affecting CTNNB1 (16 cases) with rare APC mutations ( Fig. 1a and Supplementary Table 1), whereas no common nevi harbored β-catenin activating mutations (P = 0.007214, binomial test). By contrast, nevi with overlapping features of DPN and blue nevi were devoid of CTNNB1 and APC mutations (P = 0.03, binomial test), instead harboring GNAQ activating mutations in 5 out of 7 (71%) cases (P = 0.02, binomial test). Thus, the latter are better classified as blue nevi based on their genetic profile. Only one DPN harbored GNAQ Q209L without mutation in CTNNB1 or APC. This tumor also had an IDH1 R132C mutation that was not present in the other GNAQ mutant tumors in our series. Perhaps this mutation led to epigenetic changes that altered histopathological features contributing to the apparent morphological misclassification of this lesion that genetically represents a blue nevus.
16 out of 18 DPN had mutations of the mitogen-activated protein kinase (MAPK) pathway in addition to the activating mutations of the β-catenin pathway, affecting BRAF, MAP2K1 and HRAS in a mutually exclusive pattern. 9 out of 18 DPN (50%) had BRAF mutations: BRAF V600E in six cases, BRAF A598_T599insI , BRAF K601E and AKAP9-BRAF fusion in one case each. One DPN harbored an HRAS Q61R mutation. MAP2K1 alterations were present in 6 of 18 DPN (33%) consisting of small indels near codons 102 to 107, within helix C of MEK1, which is displaced by the activation loop in the active conformation 12,13 . In one case, a small deletion affected the negative regulatory region (Fig. 1c) 13 . Similar MAP2K1 mutations occur in Langerhans cell histiocytosis and hairy cell leukemia, also in a mutually exclusive pattern with BRAF V600E14, 15 .
Three of the five MAP2K1 indels have been previously observed in melanoma or other cancers, and two have been experimentally demonstrated to constitutively activate MAP2K1 14,16,17 . We expressed 4 of the observed MAP2K1 mutations in 293FT cells and found that they activate MAP kinase signaling, which could be inhibited by the MEK inhibitor trametinib (Fig. 1d). Notably, the MAP2K1 mutations present in DPN are distinct from those at positions P124 and E203, which often occur in combination with BRAF or NRAS mutations in melanoma 18,19 .
By contrast, common nevi harbored BRAF V600E in 7/10, NRAS Q61R in 3/10, with no additional identified mutations. Thus DPN are genetically defined by the combination of MAPK activating and β-catenin activating mutations and are distinct from common and blue nevi.
To further investigate the role of CTNNB1 mutation in DPN, we used immunohistochemistry to evaluate β-catenin expression level and localization in DPN and common nevus controls. In common nevi, nuclear β-catenin was expressed in superficial melanocytes situated near the epidermis, whereas the smaller melanocytes in the deeper portions of the nevus had considerably lower expression levels, consistent with prior reports (Fig. 2) 20 .  Only deep dermal melanocytes situated near hair follicles also had increased levels of nuclear β-catenin, suggesting that proximity to epithelial structures -epidermis or hair follicle epithelium -may increase β-catenin signaling ( Supplementary Fig. 1). In DPN, β-catenin staining was more intense and by contrast did not diminish with distance from epithelia (Fig. 2, Supplementary  Fig. 2). We also measured the expression of AXIN2, a marker of β-catenin transcription 21 , by RNA in situ hybridization. In DPN, melanocytes demonstrated constant levels of AXIN2 expression throughout the neoplasm, whereas in common nevi AXIN2 expression decreased with distance from epithelium (Supplementary Table 2 and Supplementary Fig. 3).
Cyclin D1 is a direct transcriptional target of β-catenin 22 and is frequently amplified in melanoma 23 . Melanocytes in DPN demonstrated strong and uniform nuclear expression of cyclin D1 by immunohistochemistry. In contrast, immunoreactivity for cyclin D1 in common nevi was limited to melanocytes near epithelia, mimicking the expression pattern of CTNNB1 (Fig. 2, Supplementary Figs. 1, 2).
CTNNB1 mutations and confer phenotypic characteristics of DPN. Some DPN arise juxtaposed with a common nevus [24][25][26] . We hypothesized genetic progression of a BRAF V600E melanocyte within a common nevus by acquisition of a β-catenin activating mutation results in an adjacent DPN. In 66 out of 68 (97%) bi-phenotypic cases the DPN displayed increased nuclear βcatenin and cyclin D1 compared to the adjacent common nevus. β-catenin and cyclin D1 levels in the DPN remained uniform, whereas they decreased in the adjacent common nevus with distance from the epithelium (Fig. 3). We genotyped BRAF, MAP2K1, and CTNNB1 in the DPN of 11 biphasic lesions. Activating CTNNB1 mutations were identified in 10 out of 11 DPN within common nevi, usually in combination with BRAF V600E (Supplementary Table 3). In 2 cases we genotyped the adjacent common nevus component and found identical BRAF V600E mutations as in the adjacent DPN but no CTNNB1 mutations. These results indicate that common nevi can evolve to DPN by acquiring β-catenin activating mutations.
We transduced immortalized mouse melanocytes (melan-a) with BRAF V600E either alone or in combination with CTNNB1 S33F . Melanocytes expressing BRAF V600E and CTNNB1 S33F were larger, more pigmented, and expressed higher levels of cyclin D1 than melanocytes expressing BRAF V600E alone (Fig. 4), recapitulating the characteristic features of DPN. These experimental findings indicate that CTNNB1 mutations are sufficient to induce the phenotypic changes characteristic of DPN in BRAF mutant melanocytes.

DPN-like melanomas harbor additional oncogenic mutations.
Rare DPN metastasize with lethal outcomes 4, 27 . It is not known whether DPN that metastasize are genetically different from those that do not. We genotyped two metastases that originated from previously diagnosed DPN and identified that in addition to the characteristic combination of MAPK activating (NRAS Q61R or MAP2K1 I103D/E203K ) and CTNNB1 mutation (Fig. 5 28 . One also harbored TERT promoter (C228T) and TP53 T86M mutations.
Some melanomas have features of DPN such as melanocytes with pigmented cytoplasm and numerous stromal melanophages. To determine whether DPN-like melanomas share genetic characteristics with DPN, we analyzed 6 cases. One melanoma harbored a GNAQ R183Q activating mutation and truncating BAP1 mutation with loss of heterozygosity and thus was re-classified as a blue nevus-like melanoma 29 . The five remaining melanomas all harbored MAPK pathway activating mutations in BRAF or NRAS and three had activating mutations in the β-catenin pathway (Fig. 5). In all cases, copy number alterations and additional oncogenic alterations were identified, including TERT promoter mutation or biallelic loss of CDKN2A.

Discussion
In summary, our study shows that DPN result from the combined mutational activation of the MAP kinase and β-catenin pathways. Some DPN evolve from pre-existing nevi by acquiring a β-catenin pathway mutation secondary to the initiating MAP kinase pathway mutation. Prior work demonstrates that β-catenin signaling  30 , which could contribute to the deeply invasive phenotype of DPN. The differences in β-catenin pathway activation between common nevi and DPN reveal another possible mechanism by which β-catenin pathway activation contributes to neoplastic progression. In DPN, melanocytes remain large and pigmented throughout the entire neoplasm, whereas in common nevi without mutational activation of the β-catenin pathway these cytomorphological features are constrained to areas in proximity to epithelia, and melanocytes become smaller and less pigmented with increasing distance from epithelia. Epithelia secrete WNTs 31 , which may explain the decrease in β-catenin signaling with distance from epithelia. A dependence on WNT signaling may prevent common nevi from invading the deeper dermis. Constitutive β-catenin pathway activation in DPN may override this limitation, allowing these tumors to 'deeply penetrate'. Our data further indicates that β-catenin and MAPK pathway mutations alone are insufficient to fully transform melanocytes, and that additional mutations such as immortalizing mutations of TERT and loss of CDKN2A are required for DPN to progress to melanoma, (Fig. 6) consistent with prior work in mouse models 32,33 . Our results identify DPN as an intermediate melanocytic neoplasm, with a progression stage positioned between benign nevus and DPN-like melanoma.

Methods
Clinical Case Selection and Histopathologic Classification. We searched the archives at the University of California Dermatopathology Unit and the personal consultation files of AdlF for cases annotated as having features of DPN. The study was approved by the Committee on Human Research of the University of California, San Francisco, with a waiver of patient consent. Consent to publish patient photos was obtained. The histopathology of all cases were reviewed by at least 2 expert dermatopathologists (AdlF, BCB, IY, ED) and categorized as either DPN, nevus with overlapping features of DPN and blue nevus, 'combined nevi' with common nevus and DPN cytomorphology, or melanoma with features of DPN (markedly pigmented with defective maturation and occasionally deep invasion). Samples were excluded if the amount of tumor did not fall into one of the above predefined histopathologic subtypes or the tumor volume was deemed insufficient for further analysis. The number of samples in our cohort for sequencing was limited by the availability of DPN with enough residual material in our archives.  Supplementary Data 1-6. The target intervals cover mostly exonic but also some intronic and untranslated regions of 293 (version 1), 538 (version 2), and 480 (version 3) target genes. The target genes were curated to comprise common cancer genes with particular relevance to melanoma.
Captured libraries were sequenced as paired-end 100 bp reads on a HiSeq 2000 or HiSeq 2500 instrument (Illumina). Sequence reads were mapped to the reference human genome (hg19) using the Burrows-Wheeler aligner (BWA) 34 . Recalibration of reads was performed using the Genome Analysis Toolkit (GATK) 35 . Mutation calling was performed with FreeBayes 36 . Coverage and sequencing statistics were determined using Picard CalculateHsMetrics and Picard CollectInsertSizeMetrics 37 (Supplementary Table 5). Variant annotation was performed with Annovar 38 and variants with frequency in the 1000 genomes or exome sequencing project datasets of > 0.001 were excluded from further analysis. For fusion detection, read pairs with one or more reads unaligned, insert sizes greater than 1000 bp, or with soft clipping of at least one read were extracted and re-aligned using BWA-SW 39 and used as input to CREST 40  Statistical analysis. We used the binomial test with the null hypothesis that β-catenin activating mutations are equally distributed between DPN and common nevi, and DPN and nevi with features of DPN and blue nevus. We also tested the null hypothesis that GNAQ mutations distinguish nevi with features of DPN and blue nevus from DPN.
Drug studies. 293FT cells were transfected with MAP2K1 wild-type and mutant expression constructs using Jetprime (Polyplus-transfection, Illkirch) and 2 µg plasmid DNA as recommended by the manufacturer. Twenty four hours after transfection, the cells were treated for 4 h with 100 nM of trametinib (Selleckchem).
RNA in situ hybridization. Formalin-fixed, paraffin-embedded tissue sections were pretreated with 13 min of boiling followed by 12 min of protease digestion (RNAscope Protease Plus). Hybridization for hs-Axin2 (catalog #400249) was performed in parallel with DapB (catalog # 310043) and PPIB (catalog # 313901) as negative and positive controls respectively. Scoring for each case was performed in melanocytes near the epidermis and at a distance from epithelium (epidermis and adnexal epithelium). A semi-quantitative scoring method was used for each area (Supplementary Table 6