Recurrent PAX3-MAML3 Fusion in Biphenotypic Sinonasal Sarcoma

Abstract

Biphenotypic sinonasal sarcoma (SNS) is a newly described tumor of the nasal and paranasal areas. Herein, we report the novel recurring chromosomal translocation t(2;4)(q35;q31.1) in SNS. The translocation results in the formation of the fusion protein PAX3-MAML3, which is a potent transcriptional activator of PAX3 response elements. The SNS phenotype is characterized by aberrant expression of genes involved in neuroectodermal and myogenic differentiation, which closely simulates the developmental roles of PAX3.

Biphenotypic sinonasal sarcoma or low grade sinonasal sarcoma with neural and myogenic features (SNS) is a recently described tumor that preferentially occurs in the sinonasal area of middle age individuals, more often females¹. Clinically, SNS is an infiltrative tumor that often recurs locally¹. Cytogenetic analysis of two tumors revealed the novel chromosomal translocation t(2;4)(q35;q31.1) (Fig. 1a and Supplementary Fig. 1a). Transcriptome analysis of one of these cases identified a novel fusion transcript in which exons 1–7 of PAX3 were fused to exons 2–5 of MAML3 (Fig. 1a). FISH and RT-PCR studies confirmed the rearrangement of the PAX3 locus in 24 of 25 SNS tumors (96%) and the PAX3-MAML3 fusion gene in 19 of these (79%) (Fig. 1c, Supplementary Fig. 1 and Table 1). Five of the remaining SNS tumors exhibited rearrangement of the PAX3 locus without MAML3 involvement, and a single tumor showed rearrangement of the MAML3 locus without PAX3 involvement. These tumors were screened for possible alternate fusion genes but none were found (see Supplementary information). No reciprocal transcript MAML3-PAX3 was identified. The PAX3-MAML3 fusion was not detected in 118 other tumors, including rhabdomyosarcomas, melanomas, and benign and malignant nerve sheath tumors, or in 18 normal tissues (including 13 normal sinonasal tissues) (Supplementary Table 2).

Figure 1.

Structure and transactivation potential of the PAX3-MAML3 fusion. (a) and (b) The t(2;4) translocation in SNS fuses exons 1–7 of PAX3 to exons 2–5 of MAML3 to create a novel PAX3-MAML3 fusion protein that retains the DNA binding domains of PAX3 but lacks the Notch binding site of MAML3. Small arrows along the chromosomes indicate the transcription orientation of PAX3, MAML3 and PAX3-MAML3. No reciprocal fusion MAML3-PAX3 was found. (c) FISH bring together strategy confirms the juxtaposition of 5’ PAX3 (red) to the 3’ MAML3 (green) loci. The location of these probes is shown in 1a. (d) Transient transcription assays demonstrating the potent transactivation potential of PAX3-MAML3 on the 5xPAX3-Luc reporter plasmid. The data are means of triplicate samples and are shown as fold change over control. The results are representative of three independent experiments. PD, paired domain; HD, homeodomain; TAD, transactivation domain; ** p < 0.005 relative to control in unpaired t-tests.

The fusion sequence predicted a novel PAX3-MAML3 chimeric protein in SNS tumors consisting of the highly conserved paired-box DNA binding domain and the paired-type homeodomain of PAX3 fused to the transactivation domain of MAML3 (Fig. 1b). We hypothesized that the novel fusion protein would be a potent transactivator of PAX3 response elements. To test this hypothesis, the PAX3-MAML3 cDNA was generated from SNS tumor mRNA and subcloned into a mammalian expression vector with a FLAG-epitope tag. The resulting 150 kD FLAG-PAX3-MAML3 fusion protein localized to cell nuclei and bound PAX3 binding sites in vitro (Supplementary Fig. 2). PAX3-MAML3 consistently activated PAX3-driven reporter plasmids by 40-fold or more, whereas wildtype PAX3 activated these reporters approximately 8-fold (Fig. 1d). The PAX3-MAML3 protein had transactivation potential on par with that of PAX3-FOXO1, the fusion protein found in most alveolar rhabdomyosarcomas (ARMS)².

During development, PAX3 plays important roles in the differentiation and migration of neural crest-derived cells^{3, 4}. Specifically, PAX3 functions at a nodal point in melanocytic, neuronal and skeletal muscle differentiation programs, promoting lineage commitment and blocking terminal differentiation^{3, 5–7}. PAX3 germline mutations are implicated in several human disorders (Waardenburg syndrome and craniofacial-deafness-hand syndrome) and in the Splotch phenotype in mice^8–11. To better understand the unusual phenotype of SNS and its possible relationship with the developmental roles of PAX3, gene expression profiling and pathway enrichment analysis were performed on 8 SNS tumors and compared to 33 other tumors. Hierarchical clustering separated SNS from all other tumors (Fig. 2, GEO accession no. GSE52323) and showed altered expression of several genes and signaling networks involved in neural crest, skeletal system and general embryonic development (Supplementary Tables 1 and 3; Supplementary Fig. 3 and 4), including the neurogenic factors NTRK3, ALX1-4, DBX1, GREM1 and NUROG2, the myogenic genes MYOCD, MYOD1, and general developmental genes such as BMP5, FGFR2, POU3F3 and POUF4F1, among others. qPCR analysis and immunohistochemistry confirmed the altered overexpression of several of these genes, including high levels of NTRK3, a receptor tyrosine kinase that plays important roles in neuronal development and oncogenesis^{12, 13}, and absent expression of MLANA, a protein involved in melanosome biogenesis and a downstream effector of the PAX3-regulated transcription factor MITF ¹⁴, also not expressed in SNS. These studies also showed expression of smooth muscle actin (SMA) and muscle specific actin (MSA), weakly and sporadic expression of desmin and MyoD1, and consistent negativity for myogenin. (Supplementary Table 1 and Supplementary Fig. 5). Interestingly, while most SNS expressed high levels of MYOD1 mRNA¹⁵, the protein product of this myogenic transcription factor was detected in only a subset of cases (Supplementary Tables 1 and 3, and Supplementary Fig. 4). This intriguing finding may be explained by post-transcriptional events or the sensitivity of the immunohistochemical analysis for low expression levels of this protein. Taken together, these observations suggest that the SNS phenotype simulates the developmental roles of PAX3, and that the differentiation program of this tumor may be possibly modulated by PAX3-MAML3.

Figure 2.

Unsupervised hierarchical clustering analysis showing the gene expression signature of 8 SNS and 33 other mesenchymal tumors based on the expression level of 516 genes (See Supplementary information). Only the top 150 upregulated and 50 downregulated genes are shown. Genes that were further validated by qPCR and/or immunohistochemistry are shown on the right. SNS shows a unique profiled characterized by the overexpression of several genes involved in neuronal differentiation and downregulation of genes involved in melanocytic differentiation (for detailed information about these genes, see Supplementary Table 3). (SNS, sinonasal sarcoma; ARMS, PAX3-FOXO1+ alveolar rhabdomyosarcoma; ERMS, embryonal rhabdomyosarcoma; BFH, dermatofibroma/benign fibrous histiocytoma; DFSP, dermatofibrosarcoma protuberans; FTS, fibroma of tendon sheath; GIST, gastrointestinal stromal tumor; MPNST, malignant peripheral nerve sheath tumor; SS, SS18-SSX1/2+ synovial sarcoma; detailed information about these tumors is provided in Supplementary Table 2).

MAML3 is a member of the mastermind-like (MAML) family of transcriptional co-activators for the Notch signaling pathway, which is conserved in metazoans and involved in a variety of pivotal cellular processes, including cell proliferation, differentiation and death^{16, 17}. A MAML2 fusion gene occurs in most mucoepidermoid carcinomas, the most common malignant salivary gland tumor¹⁸. In this tumor, exon 1 of the CREB-binding transcriptional coactivator CRTC1 is fused in-frame with MAML2 exons 2–5. The CRTC1-MAML2 oncoprotein retains the CREB-binding site of CRTC1 and the transactivation domain and acidic domains of MAML2. Interestingly, CRTC1-MAML2 is able to induce Notch signaling despite the deletion of the Notch/CSL binding domain in MAML2¹⁷. However, recent studies indicate that the oncogenic effects of CRTC1-MAML2 also involve cAMP/CREB and AREG/EGFR signaling pathways¹⁹. In contrast to mucoepidermoid carcinoma, we did not observe significant changes in several Notch target genes in SNS (HES1, ratio 0.71, p=0.15; HES7, ratio 0.91; p=0.81; HEY1, ratio 0.94, p=0.83; HEYL, ratio 0.77, p=0.22; MYC, ratio 0.67; p=0.14). Further, in transient transcription assays, PAX3-MAML3 did not affect the basal activity of the Notch reporter, Hes5p-Luc, but modestly repressed NICD1-driven transcription (Supplementary Fig. 2d). Similar results were seen with another Notch reporter, 4xCSL-Luc (data not shown). At this point it is unclear how the transactivation domain of MAML3 modulates the oncogenic function of PAX3-MAML3 but major effects on Notch signaling were not observed. Future studies are needed to address this question.

Since nearly all SNS tumors examined showed either the PAX3-MAML3 fusion or PAX3 locus rearrangement, alterations in PAX3 activity may be crucial for the genesis and propagation of SNS. Further, considering the overall structural similarity between PAX3-MAML3 and PAX3-FOXO1 fusion genes, the study of SNS may shed important light onto our understanding of ARMS etiology and PAX3-induced oncogenesis. Indeed, comparing the transcriptional profiles of SNS and ARMS, we have encountered several common denominators but also some important differences (Supplementary Table 4). For example, the expression of MET, GREM1, DAPK1 and TCF7L2 were similar between these two tumors. On the other hand, some direct targets of PAX3-FOXO1 or genes that are commonly highly expressed in ARMS, including FGFR4, MYCN and ALK, are downregulated in SNS. Likewise, ARMS - but not SNS - expressed high levels of several skeletal myogenesis-related genes (e.g. MYOG and DES), FGF8, DCX, BVES and CNR1, among others. In contrast, when compared to ARMS, SNS had higher levels of several neurogenic and cytokine-related genes, metalloproteases, a poorly understood member of the PIK family PIK3C2G, and MYOCD (myocardin). The functional characteristics of the PAX3 fusion partners, MAML3 or FOXO1, likely contribute to these differences but intrinsic characteristics of the cell of origin cannot be disregarded.

Two decades ago Maulbecker and Gruss suggested that “spindle cell sarcomas, particularly those that occur early in the development, should be tested for elevated PAX gene expression”⁴. In spite of the fact that SNS is only occasionally seen in younger patients, our findings nicely fulfill that insightful prediction. Interestingly, in 2003 Gil et al. reported a malignant peripheral nerve sheath tumor histologically identical to SNS arising in the skull base with the similar chromosomal translocation t(2;4)(q35;q31), very likely representing the same entity²⁰.

In summary, we show in this study that PAX3-MAML3 is the characteristic fusion gene found in SNS and a potent transactivator of PAX3 response elements. SNS aberrantly express genes involved in neuroectodermal and myogenic differentiation, closely simulating the developmental roles of PAX3. Finally, our findings may also lead to a better recognition of this novel sarcoma among other spindle cell sarcomas and possibly further contribute to our understanding of other cancers, including ARMS.

Online Methods

Supplementary note

Tumors were retrospectively identified at the Mayo Clinic Tissue Registry and consultation files of the authors (A.M.O. and J.E.L.).

Patients and samples

Frozen tumor sample was obtained from a single specimen characterized at the cytogenetics level (case 1). Formalin-fixed paraffin embedded (FFPE) tumor blocks and histologic sections of biphenotypic sinonasal sarcomas (SNS) biopsied or resected between 1956 and 2013 were retrieved from all 25 tumors, including a second example that was also characterized at the cytogenetics level (Supplementary Table 1). All SNS were originally diagnosed as fibrosarcomas, malignant peripheral nerve sheath tumors and synovial sarcoma. FFPE material from 145 non-related tumors and normal tissues were also retrieved (Supplementary Table 2). The Mayo Clinic Institutional Review Board approved this study.

Transcriptome sequencing and data analyses

Total RNA was extracted from frozen material from a t(2;4) positive primary SNS (case 1) using Qiagen Mini RNeasy Kit (Qiagen, Inc., Valencia, CA). RNA concentration was measured using Qubit® 2.0 Fluorometer (Life Technologies Corporation, Carlsbad, CA). One microgram of total RNA was subjected to TruSeq RNA sample library preparation using the kit from Illumina (Illumina Inc., San Diego, CA). Paired-end 50-base transcriptome sequencing was performed using HiSeq 2000 sequencer (Illumina) at the Mayo Clinic Medical Genome Facility. The FASTQ read files for the sample were used for sequencing data analysis. Briefly, the reads were mapped to the Human Genome Reference Build 37 and exon junctions using BWA²¹. The exon junction database was constructed using the gene definition file downloaded from the UCSC golden path database. A junction usually contains 49 bases from the end of an exon and 49 bases from the beginning of the next exon. If the next exon was shorter than 49 bases, the exon after was used. An exhaustive combination of all exon junctions in a gene with single direction (from the beginning of the transcript) was constructed. The potential fusion transcripts were detected using the SnowShoes-FTD algorithm developed for paired-end mRNA-Seq data²². A total of 412 million 50-base reads were generated from the index case (Supplementary Table 1, case 1), among which 325 million reads (89.9%) were mapped to genome and 74.5 million (18.1%) were mapped to exon junctions. Among mapped reads, 350.5 million reads mapped to known genes curated by UCSC golden path genomic database. In order to estimate the percentage of genes “expressed”, gene counts were log 2 transformed (Log2 (gene count + 1) to avoid Log transform of zero). Similar to the bimodal distributions observed from the cDNA microarray data, the histogram of the log 2 transformed gene counts from this RNA-Seq data also displayed a bimodal distribution, where the distribution around the second mode (larger expression values) represents the “expressed” genes, while the distribution around the first mode are genes whose expression were only detected by chance. We estimated that approximately 68% of the 23,398 genes were “expressed”. The only two candidate fusion sequences identified were: PAX3-MAML3 and GPR128-TFG. The genes involved in the first of these fusion sequences perfectly matched the location of the chromosomal breakpoints and were further investigated. The fusion transcript GPR128-TFG was only found in the index case and its biologic significance is unknown.

Reverse transcriptase polymerase chain reaction (RT-PCR)

RNA was extracted from a frozen tumor specimen and FFPE sections from 25 SNS and 136 additional controls (118 tumors and 18 normal tissues) using High Pure FFPE RNA Micro Kit (Roche Applied Science, Indianapolis, IN) or TRIzol Reagent (Life Technologies) after deparaffinization. RNA concentration was determined by Qubit® 2.0 Fluorometer (Life Technologies). RNA extracted from the tumors and controls was converted into cDNA with random hexamers using Transcriptor First Strand cDNA Synthesis Kit (Roche). Housekeeping gene GAPDH or PGK was amplified as cDNA quality control. PCR reactions were carried out using Platinum Taq DNA Polymerase (Life Technologies) with the following conditions: 94°C for 2 min; for 40 cycles, 94°C for 30 seconds, 62°C for 30 seconds, and 72°C for 30 seconds; 72°C for 7 min. PCR primers used for PAX3-MAML3 fusion validation and fusion screening are described in Supplementary Table 5. PCR primers used for the possible reciprocal fusion gene MAML3-PAX3 and possible alternate fusion genes PAX3-FOXO1, PAX3-MAML1, PAX3-MAML2, PAX3-NCOA1, PAX3-NCOA2 and PAX7-MAML3 are described in Supplementary Table 5. PCR products were separated in 3.0% agarose gels, extracted with QIAquick Gel Extraction Kit (Qiagen), and sequenced with 3730xl DNA Analyzer (Applied Biosystems, Foster City, CA). The sequences of PCR products were analyzed using the BLAST program from NCBI (National Center for Biotechnology Information, http://blast.ncbi.nlm.nih.gov/Blast.cgi).

Fluorescence in situ hybridization (FISH)

Bacterial artificial chromosome (BAC) clones were purchased from Children's Hospital Oakland Research Institute (CHORI, Oakland, CA). BAC clones flanking two sides of the PAX3 locus were RP11-1033L11, RP11-92G23, RP11-774C7 and RP11-71J24 (centromeric side probe size = 493 kb), and RP11-1012L13, RP11-157M20, RP11- 884B7 and RP11-81I8 (telometric side probe size = 489 kb). BAC clones flanking two sides of the MAML3 locus were RP11-121K15, RP11-53D6 and RP11-946J16 (centromeric side probe size = 489 kb), and RP11-625H13, RP11-876B4, RP11-542P2 and RP11-5K16 (telomeric side probe size = 533 kb). BAC clones flanking two sides of MAML1 locus were CTD-3243D15, CTD-2335L20, RP11-1058J18 and RP11-983G15 (centromeric side probe size = 291 kb), and RP11-1148N14, CTD-3221P10, RP11-465D22, RP11-669B15 and RP11-282I19 (telomeric side probe size = 616 kb). BAC clones flanking two sides of MAML2 locus were CTD-544I7, CTD-2252L1, RP11-1123F20, RP11-936C10 and RP11-7D4 (centromeric probe size = 623 kb), and RP11-1056O10, CTD-2325K3 and RP11-8N17 (telomeric probe size = 372 kb). BAC clones flanking two sides of PAX7 locus were CTD-2342P14, RP11-14K12 and RP11-632C1 (telomeric probe size = 411 kb), and RP11-998A17, RP4-540O3 and RP11-22M1 (centromeric probe size = 429 kb). For FOXO1 locus we used a commercially available FISH break apart probe (Abbott Molecular, Abbott Park, IL). The preparation and validation of direct-labeled FISH probes, and interphase molecular cytogenetic studies were performed using a previously described FISH protocol²³ on 4µm-thick FFPE sections of 25 cases of SNS and 28 controls. The probe mixtures were custom designed to allow for three probe strategies to be employed: a PAX3 break-apart probe strategy, a MAML3 break-apart probe strategy, and a PAX3-MAML3 single fusion (bring-together) probe strategy composing a probe telomeric to the PAX3 locus and a probe centromeric to the MAML3 locus. Greater than 10% of tumor nuclei separation was needed to reveal a signal pattern indicative of locus rearrangement by FISH.

Cloning of the PAX3-MAML3 cDNA

RNA from SNS case 1 was reverse transcribed to cDNA using SuperScript III (Life Technologies). Primers were designed to amplify the predicted PAX3-MAML3 fusion gene and incorporate a FLAG-epitope tag on the N-terminus, as well as BamHI and NotI restriction enzyme sites for cloning. Primer sequences are described in Supplementary Table 5. The resulting cDNA was subcloned into pcDNA3 and fully sequenced to confirm the fidelity of the RT-PCR and cloning processes.

Western blotting and Immunofluorescence

Human embryonic kidney (HEK) 293T cells were transiently transfected with pcDNA3-FLAG-PAX3-MAML3 or pcDNA-HA-PAX3 using Lipofectamine (Life Technologies). Cell lysates were prepared after 48 hours, resolved by SDS-PAGE and transferred to Immobilon P membranes (EMD Millipore, Billerica, MA). Membranes were blotted with PAX3 antibody (Life Technologies; rabbit polyclonal 38-1801, 1:1000 dilution). For immunofluorescence assays, murine mesenchymal C2C12 cells were plated on cover slips and transfected with pcDNA3 or pcDNA3-FLAG-PAX3-MAML3. After 48 hours, cells were fixed with 4% paraformaldehyde, permeabilized, and sequentially incubated with anti-FLAG (Sigma-Aldrich Corp., St. Louis, MO; mouse monoclonal F1804, 1:100 dilution) and Alexa Fluor 488 goat anti-mouse antibodies (Life Technologies; A11001, 1:800 dilution). Cells were also incubated with Rhodamine Phalloidin (Life Technologies; R415) and DAPI (Vector Laboratories, Inc., Burlingame, CA; H-1200) to detect cytoplasmic filamentous actin (F-actin) and nuclear DNA, respectively. Images were collected on a Zeiss LSM510 confocal microscope.

Luciferase Assays

C2C12 cells was transiently transfected with the indicated expression plasmids (pcDNA3-FLAG-PAX3-MAML3, pcDNA-PAX3, pcDNA-PAX3-FOXO1 or pcDNA-NICD1), a PAX3 firefly luciferase (Luc) reporter construct (pGL4.27-5xPAX3-BS-Luc) or Hes5p-Luc (Addgene, Cambridge, MA) and with pRL-Null (Promega Corporation, Maddison WI; which was added to all experimental points as a control for transfection efficiency) using Lipofectamine (Life Technologies). Cells were lysed 48 hours later. Firefly and renilla (RL) luciferase activities were measured with Dual Luciferase Assay according to the manufacturer’s instructions (Promega). Values of luciferases activity in each experiment represent the mean of triplicate samples. All experiments were repeated a minimum of three times.

Electrophoretic Mobility Shift Assays (EMSAs)

EMSAs were performed by incubating lysates of monkey kidney fibroblast COS-7 cells transfected with pcDNA3, pcDNA3-HA-PAX3 or pcDNA3-FLAG-PAX3-MAML3 with radiolabeled, double-stranded DNA probes containing PAX3 binding sequences. Complexes were incubated with or without 100-fold excess of unlabeled wildtype or mutant double-stranded DNA probes as indicated. Probes are described in Supplementary Table 5. Control IgG (Southern Biotech, Birmingham, AL; mouse IgG2b, 0104-01), anti-HA (Covance Inc., Princeton, NJ; mouse monoclonal MMS-101R, 1:1000 dilution) or anti-FLAG (Sigma) was added to reactions as indicated²⁴.

Gene expression profiling and pathway analyses

Forty-one samples, including 8 SNS (7 with PAX3-MAML3 and 1 with PAX3 rearrangement only) and 33 different tumors [dermatofibroma (n=3), fibroma of tendon sheath (n=3), melanoma (n=3), neurofibroma (n=3), schawannoma (n=3), malignant peripheral nerve sheath tumor (n=3), PAX3-FOXO1+ alveolar rhabdomyosarcoma (n=3), embryonal rhabdomyosarcoma (n=3), gastrointestinal stromal tumor (n=3), SS18-SSX1/2+ synovial sarcoma (n=3), and dermatofibrosarcoma protuberans (n=3)] were analyzed.

Total RNA was extracted from FFPE materials using miRNeasy FFPE Kit (Qiagen). RNA concentration was measured by Nanodrop-1000 instrument (Thermo Fisher Scientific, Wilmington, DE). RNA quality was evaluated by amplifying cDNA of housekeeping genes GAPDH or PGK. 200ng of RNA from each sample was subjected to Human WG-DASL Assay with Human HT12 v4.0 BeadChips (Illumina), which contains 47,000 probes. Microarray data analyses were performed using Illumina GenomeStudio (Illumina) and Parteck Genomics Suite software (Partek Inc., St. Louis, Missouri). Briefly, the preprocessing of the probe level data was performed using the Genome Studio using quantile normalization with no background subtraction. The “non-expressed” probes were defined as those with detection p value > 0.05 in all 41 samples. There were 20,818 genes that remained after this filtering and the log 2 transformed gene expression values were further analyzed using the Partek Genomic Suit software tools. The differentially expressed genes between the SNS and other tumors were identified using ANOVA to calculate the p values, multiple comparison error corrected p values (step-up correction method from Partek), and fold changes. Pathway enrichment analysis was performed using MetaCore software (Thomson Reuters, New York, NY). The 516 genes with an adjusted p value ≤ 0.05 were used as “focus genes”, and the 20,818 “expressed” genes were used as the “reference genes” for the MetaCore pathway analyses. The hierarchical clustering shown in Fig. 2 shows the top 200 differentially expressed genes.

Relative quantitative real time PCR (qPCR)

Confirmatory qPCR was performed on genes with well-established roles in neural, muscle and melanocytic development and differentiation. Total RNA was extracted from FFPE materials using High Pure FFPE Micro RNA kit (Roche) and was treated with DNase I (Life Technologies). RNA concentration was measured by Nanodrop-1000 instrument (Thermo Fisher Scientific). cDNA was synthesized with random hexamers and 1µg of total RNA using iScript cDNA Synthesis Kit (Bio-Rad Laboratories, Hercules, CA). qPCR was performed on Roche’s Light Cycler 480 (LC480) Instrument with 96-well plates. Human Universal Reference cDNA (Clontech Laboratories, Inc., Mountain View, CA) was used as the calibrator. Hydrolysis probes for the target genes NTRK3, MYOD1, MLAN, MYOG and reference gene PGK1 were UPL probe # 71, # 70, # 56, # 20 and # 69 (Roche), respectively. qPCR primers are described in Supplementary Table 5. Each 20µl of qPCR reaction included 1x LC480 Probes Master (Roche), 3.0 µM of each primer, 2.0 µM of probe, and 5.0 µl of 1:50 diluted cDNA. qPCR conditions were: 95°C for 10 minutes, followed by 50 cycles of 95°C for 10 seconds, 60°C for 30 seconds, 72°C for 5 seconds, and 40°C for 15 seconds. Relative expression ratio was calculated using Calibrator Normalized Relative Quantification Method (Roche). qPCR for NTRK3, MYOD1, MLANA and MYOG target genes was carried out in two separate experiments with triplicate wells in each experiment.

Immunohistochemistry

Immunohistochemical analysis was performed using the Leica Bond III Stainer (Leica, Buffalo, IL). The tissue slides were dewaxed and retrieved on-line using the following reagents Bond Dewax (Leica) and Epitope Retrieval 2 (Leica). Tissue slides were retrieved for 20 minutes. Primary antibodies (with dilutions and incubation times) used were anti-NTRK3 (Cell Signaling Technology Inc, Danvers, MA; mouse monoclonal C44H15, 1:400, 30 min), anti-MyoD1 (Dako North America Inc, Carpinteria, CA; mouse monoclonal 5.8A, 1:500, 15 min), anti-Melan A (Dako, monoclonal mouse A103, 1:50, 16 min), anti-PAX3 (Invitrogen, polyclonal, 1:1400, 30 min), anti-ALX1 (Sigma-Aldrich, polyclonal, 1:300; 15 min), anti-HOXB7 (Sigma-Aldrich, polyclonal, 1:100, 15 min), anti-HES1 (Cell Signaling Technology, rabbit monoclonal D6PZU, 1:2000, 15 min), anti-Myogenin (Dako, mouse monoclonal F5D, 1:1200, 30 min), anti-S100 (Dako, rabbit polyclonal, 1:4000, 32 min), anti-desmin (Leica, monoclonal mouse DER11, 1:100, 32 min), anti-muscle specific actin (MSA; Dako, monoclonal mouse HHF35, 1:100, 30 min), anti-smooth muscle actin (SMA; Dako, monoclonal mouse 1A4, 1:3000, 15 min) and anti-MITF (Ventana Medical Systems Inc., Tucson, AZ; mouse monoclonal C5/D5, pre-diluted from factory, 16 min). The detection system used was Polymer Refine Detection System (Leica). This system includes the hydrogen peroxidase block, secondary antibody polymer, DAB and hematoxylin. Slides were rinsed for 5 minutes in water and dehydrated in increasing concentrations of ethyl alcohol and xylene. A coverslip was permanently mounted over the specimen with xylene-based medium.