Novel MIR143-NOTCH Fusions in Benign and Malignant Glomus Tumors

Abstract

Glomus tumors (GT) have been classified among tumors of perivascular smooth muscle differentiation, together with myopericytoma, myofibroma/tosis, and angioleiomyoma, based on their morphologic overlap. However, no molecular studies have been carried out to date to investigate their genetic phenotype and to confirm their shared pathogenesis. RNA sequencing was performed in three index cases (GT1, malignant GT; GT2, benign GT and M1, multifocal myopericytoma), followed by FusionSeq data analysis, a modular computational tool developed to discover gene fusions from paired-end RNA-seq data. A gene fusion involving MIR143 in band 5q32 was identified in both GTs with either NOTCH2 in 1p13 in GT1 or NOTCH1 in 9q34 in GT2, but none in M1. After being validated by FISH and RT-PCR, these abnormalities were screened on 33 GTs, 6 myopericytomas, 9 myofibroma/toses, 18 angioleiomyomas and in a control group of 5 sino-nasal hemangiopericytomas. Overall NOTCH2 gene rearrangements were identified in 52% of GT, including all malignant cases and one NF1-related GT. No additional cases showed NOTCH1 rearrangement. As NOTCH3 shares similar functions with NOTCH2 in regulating vascular smooth muscle development, the study group was also investigated for abnormalities in this gene by FISH. Indeed, NOTCH3 rearrangements were identified in 9% of GTs, all present in benign soft tissue GT, one case being fused to MIR143. Only 1/18 angioleiomyomas showed NOTCH2 gene rearrangement, while all the myopericytomas and myofibroma/toses were negative. In summary we describe novel NOTCH1-3 rearrangements in benign and malignant, visceral and soft tissue GTs.

INTRODUCTION

Pericytes are specialized vascular smooth muscle cells that play an important role in supporting and maintaining the capillary structure. Pericytic tumors comprise a histologic continuum of neoplasms with perivascular myoid differentiation. Until recently their classification has been somewhat controversial and historically were lumped together with other tumors of similar morphology, such as hemangiopericytoma, which subsequently was reclassified together with solitary fibrous tumors as showing fibroblastic rather then true pericytic lineage. The 2013 WHO classification of soft tissue tumors includes glomus tumors, myopericytoma, myofibroma and angioleiomyoma as members of the pericytic family of neoplasms (Fletcher et al., 2013a). Despite their histologic overlap and lesions with hybrid features a unifying concept supported by shared genetic abnormalities has not been yet established. In this study we investigated a subset of pericytic tumors by RNA sequencing for novel gene discovery with potential role in the pathogenesis of tumors of perivascular myoid lineage. Our hypothesis is that a better understanding of their genetic abnormalities may clarify the relationship among the various pericytic tumors and improve the current classification based on morphologic features alone.

MATERIAL AND METHODS

Patient Selection and Tumor Characteristics

The Pathology files of two participating Institutions and the personal consultations of the corresponding author were searched for cases of glomus tumor (GT), myopericytoma, myofibroma, myofibromatosis and angioleiomyoma, with adequate material available for molecular work-up. Hematoxylin and eosin (H&E) stained slides from all cases were reviewed by two pathologists (CRA and JMM). Immunostains for muscle markers (smooth muscle actin [SMA], common muscle actin [CMA], and desmin) to support the above diagnosis were performed (pre-diluted antibodies from Ventana Medical Systems, Inc, Tucson, AZ) or available for review in all cases. Cases with hybrid morphology were classified based on the predominant growth pattern. Grade of malignancy was determined using the following criteria, including marked nuclear pleomorphism and mitotic activity or the presence of atypical mitotic figures (Fletcher et al., 2013b).

Three index cases with available frozen tissue were subjected to RNA sequencing: GT1, a malignant gastrointestinal GT (Fig. 4A–C); GT2 a benign soft tissue glomus tumor arising in the neck (Fig. 2A); and M1, a multifocal soft tissue myopericytoma of lower extremity (Fig. 4E–H), all characterized by classic morphology and immunophenotype. The genetic abnormalities identified in the discovery step were validated and then screened in a larger cohort of cases, spanning all members of the pericytic tumor family, as well as a wide variety of anatomic locations and grade of malignancy. The study group included 33 GTs, 6 myopericytomas, 9 myofibroma/infantile myofibromatosis, and 18 angioleiomyomas. The clinicopathologic features are presented in Tables 1&2. Also included in the analysis was a control group of five sino-nasal hemangiopericytomas for potential associations. The study was approved by the Institutional Review Board at each institution (IRB# 02-060 MSKCC and IRB# 1007011157 WCMC).

Fig. 4. Morphologic spectrum of pericytic tumors.

(A) Index malignant gastrointestinal glomus tumor (GT1) showing transmural involvement; high power showing focal benign component with classic morphology (B), as well as areas of sarcomatous growth with areas of geographic necrosis (C);

(D) Angioleiomyoma of the knee soft tissue area in a 32 year-old male, showing mature smooth muscle bundles proliferating out small vessel walls; this example was the only one of 18 examples tested showing a NOTCH2 gene rearrangement by FISH;

(E) The index soft tissue myopericytoma (M1) showing multifocal presentation within subcutis by coronal STIR MRI and gross appearance (F); microscopically the tumor had a multinodular pattern, including intra-vascular growth (G) and high power showed ovale to short spindle cells in a haphazard, patternless pattern around small capillary vessels (H).

(I) Digital glomus tumor in a patient with NF1 (GT18) showing dermal proliferation of perivascular cuboidal and bland oval cells (J), highlighted by SMA (K) and showing unbalanced rearrangement of NOTCH2 with deletion of telomeric part (Green signal) by FISH (L);

(M) Malignant glomus tumor showing an abrupt transition from a benign monotonous appearance to a highly pleomorphic component (GT6) in the kidney; a different example in the stomach (GT7), showing focal areas of benign GT (N), while most of the peritoneal spread was composed of an undifferentiated spindle cell sarcoma morphology (O). The latter component showed low level of amplification of NOTCH2 centromeric parts (P, Red signal), with loss of the telomeric part (Green).

Fig. 2. MIR143-NOTCH1 gene fusion in a benign glomus tumor of the neck soft tissue (GT2).

(A) Typical morphologic appearance of a glomus tumor with uniform cuboidal cells with pale eosinophilic cytoplasm and round, bland nuclei, with a distinctive angiocentric growth around small blood vessels (H&E, 200x).

(B) FISH analysis showing an unbalanced NOTCH1 rearrangement, with loss of the telomeric part (green signal) (tri-color assay, Orange/Green flanking NOTCH1, Red for C′-ABL used as control, centromeric to NOTCH1 at 9q34).

(C) The top fusion candidate selected by FusionSeq was confirmed by RT-PCR showing the MIR143 exon 1 being fused to exon 27 of NOTCH1.

Table 1.

Glomus Tumors Showing NOTCH Rearrangements by FISH

GT#	Age/Gender	Location	Benign/Malignant	NOTCH FISH	MIR143 FISH
1 *	77/F	Small bowel	malignant	NOTCH2	+
2*	54/F	Neck ST	benign	NOTCH1	+α
3	68/M	Thigh	benign	NOTCH2	+
4	48/M	Thigh	benign	NOTCH2	+
5	52/M	Arm	benign	NOTCH2	+
6	36/M	Kidney	malignant	NOTCH2¥	−
7	32/M	Gastric, omental implants	malignant	NOTCH2¥	+
8	16/M	Leg	benign	NOTCH2	−
9	52/M	Knee	benign	NOTCH2	+
10	52/M	Stomach	benign	NOTCH2	+
11	69/M	Buttock	benign	NOTCH2	+
12	57/M	Forearm	benign	NOTCH2	+
13	67/M	Buttock	benign	NOTCH2	+
14	41/M	Main-stem bronchus	benign	NOTCH2	+
15	64/M	Leg	benign	NOTCH2	+
16	49/F	G–E junction	malignant	NOTCH2	−
17	74/M	Foot	malignant	NOCTH2	−
18Ω	47/M	Finger	benign	NOTCH2	−
19	41/M	Forearm	benign	NOTCH3	+
20	31/M	Knee	benign	NOTCH3	−
21	57/M	Knee	benign	NOTCH3	−

GT, glomus tumor;

^*index cases studied by RNA-seq;

^αconfirmed by RNA-seq and RT-PCR, but by FISH rearranged in <10% cells by FISH; ST, soft tissue;

^¥NOTCH2 break-apart signal in the benign component, while the malignant area showed low level of amplification of centromeric part with loss of the telomeric region;

^ΩNF1-developed GT, metachronous neurofibroma and MPNST negative for NOTCH2 rearrangements.

Table 2.

Glomus Tumors Negative for Structural Rearrangements in NOTCH1-3 and MIR143

GT#	Age/gender	Location	Histologic Grade
22	51/F	Middle finger	Benign
23	42/F	Finger	Benign
24	76/M	Thumb	Benign
25	37/M	Finger	Benign
26	43/F	Thumb	Benign
27	38/F	Digital, subungual	Benign
28	43/M	Digital, subungual	Benign
29	26/M	Knee	Benign
30	52/M	kidney	Benign
31	17/F	Lower extremity, multifocal (familial)	Benign
32	68/M	Multifocal liver, spleen	Benign
33	84/F	Lung	Benign

Among the GTs, there were 24 arising in the soft tissue, including 14 in the non-acral extremity (11 in the lower extremity and buttock, three in the upper extremity), eight in the digits, one in the soft tissue of the foot with secondary destruction of the adjacent metatarsal bone and one in the neck area. The remaining nine tumors occurred in visceral locations, including four gastrointestinal (stomach, two; GE junction and small bowel, one each), two renal, two pulmonary and one multifocal, involving both spleen and liver. There were 28 benign and 5 malignant tumors. The malignant lesions occurred mainly in visceral location, including three in the GI tract (gastro-esophageal junction, stomach, small bowel)(Fig. 4N,O), one in the kidney (Fig. 4M) and one in the soft tissue. The referred diagnoses for these malignant lesions varied significantly, including high grade undifferentiated sarcoma (case GT7, Table 1), small blue round cell tumor/atypical Ewing sarcoma (GT16), and epithelioid gastrointestinal stromal tumor (GT1). Their phenotype thus varied from a spindled and pleomorphic sarcomatoid neoplasm to tumors showing a more monotonous but undifferentiated small blue round cell morphology. The presence of actin reactivity, often focal and weak, was typically disregarded and interpreted as a non-specific finding. In retrospect all these tumor had a co-existing benign component, most commonly blending in with the more sarcomatous/pleomorphic areas.

RNA Sequencing

Total RNA was prepared for RNA sequencing in accordance with the standard Illumina mRNA sample preparation protocol (Illumina). Briefly, mRNA was isolated with oligo(dT) magnetic beads from total RNA (10 μg) extracted from case. The mRNA was fragmented by incubation at 94°C for 2.5 min in fragmentation buffer (Illumina). To reduce the inclusion of artifactual chimeric transcripts due to random priming of transcript fragments into the sequencing library because of inefficient A-tailing reactions that lead to self ligation of blunt-ended template molecules (Quail et al., 2008), an additional gel size-selection step was introduced prior to the adapter ligation step. Size-ranges captured were 300–350 bp during the first size-selection step and then 400–450 bp for the second size-selection step after the ligation of the adapters. The adaptor-ligated library was then enriched by PCR for 15 cycles and purified. The library was sized and quantified using DNA1000 kit (Agilent) on an Agilent 2100 Bioanalyzer according to the manufacturer’s instructions. Paired-end RNA-sequencing at read lengths of 50 or 51 bp was performed with the HiSeq 2000 (Illumina). Across all samples a total of about 268 million paired-end reads were generated, corresponding to about 27 billion bases.

Analysis of RNA Sequencing Results with FusionSeq

All reads were independently aligned with the CASAVA 1.8 software provided by Illumina against the human genome sequence (hg19) and a splice junction library, simultaneously. The splice junction library was generated by considering all possible junctions between exons of each transcript. We considered the University of California, Santa Cruz (UCSC) Known Genes annotation set (Hsu et al., 2006) to generate this library via RSEQtools, a computational method for processing RNA-seq data (Habegger et al., 2011). The mapped reads were converted into Mapped Read Format (Habegger et al., 2011) and analyzed with FusionSeq (Sboner et al., 2010) to identify potential fusion transcripts. FusionSeq is a computational method successfully applied to paired-end RNA-seq experiments for the identification of chimeric transcripts (Tanas et al., 2011; Pierron et al., 2012; Mosquera et al., 2013). Briefly, paired-end reads mapped to different genes are first used to identify potential chimeric candidates. A cascade of filters, each taking into account different sources of noise in RNA-sequencing experiments, was then applied to remove spurious fusion transcript candidates. Once a confident list of fusion candidates was generated, they were ranked with several statistics to prioritize the experimental validation. In these cases, we used the DASPER score (Difference between the observed and Analytically calculated expected SPER): a higher DASPER score indicated a greater likelihood that the fusion candidate was authentic and did not occur randomly. See Sboner et al. (2010) for further details about FusionSeq.

Fluorescence In Situ Hybridization (FISH)

FISH on interphase nuclei from paraffin-embedded 4-micron sections was performed applying custom probes using bacterial artificial chromosomes (BAC), covering and flanking genes that were identified as potential fusion partners in the RNA-seq experiment. BAC clones were chosen according to USCS genome browser (http://genome.uscs.edu), see Supplementary Table 1. The BAC clones were obtained from BACPAC sources of Children’s Hospital of Oakland Research Institute (CHORI) (Oakland, CA) (http://bacpac.chori.org). DNA from individual BACs was isolated according to the manufacturer’s instructions, labeled with different fluorochromes in a nick translation reaction, denatured, and hybridized to pretreated slides. Slides were then incubated, washed, and mounted with DAPI in an antifade solution, as previously described (Antonescu et al., 2010). The genomic location of each BAC set was verified by hybridizing them to normal metaphase chromosomes. Two hundred successive nuclei were examined using a Zeiss fluorescence microscope (Zeiss Axioplan, Oberkochen, Germany), controlled by Isis 5 software (Metasystems). A positive score was interpreted when at least 20% of the nuclei showed a break-apart signal. Nuclei with incomplete set of signals were omitted from the score.

Reverse Transcription Polymerase Chain Reaction (RT-PCR)

An aliquot of the RNA extracted above from frozen tissue (Trizol Reagent; Invitrogen; Grand Island, NY) was used to confirm the novel fusion transcript identified by FusionSeq. RNA quality was determined by Eukaryote Total RNA Nano Assay and cDNA quality was tested for PGK housekeeping gene (247 bp amplified product). Three microgram of total RNA was used for cDNA synthesis by SuperScript ® III First-Strand Synthesis Kit (Invitrogen, Carlsbad, CA). RT-PCR was performed using the Advantage-2 PCR kit (Clontech, Mountain View, CA) for 33 cycles at a 64.5°C annealing temperature, using the following primers: MIR143HG Exon1.3 fwd 5′-CAAACAGGCTGGCTCCCGTCTC-3′; NOTCH2 Exon27 rev 5′-CCGTGTTCTTGAAGCAGTGGTC-3′; NOTCH1 Exon28 rev 5′-CGAAGAACAGAAGCACAAAGGC-3′; NOTCH3 Exon30 rev 5′-GGTCAGTCCGTGCCCCAAG-3′. The PCR products were confirmed by agarose gel electrophoresis with ethidium bromide staining and sequenced using the Sanger method.

Long-Range PCR

Genomic DNA was extracted from frozen tissue using the Phenol/Chloroform assay and quality was confirmed by electrophoresis. 0.5 μg genomic DNA was amplified with the Advantage 2 PCR Kit (Clontech) using the following primers: MIR143HG Intron1.11 fwd 5′-GGTGGGGGTGTCATAGAAGTCTG-3′; NOTCH2 Intron26 rev 5′-GAGATGGGGGTAAAACAGAAGAGTG-3′; NOTCH3 Exon30 rev 5′-GGTCAGTCCGTGCCCCAAG-3′. The PCR product was confirmed by agarose gel electrophoresis with ethidium bromide staining, and then sequenced by Sanger method.

Western Blotting

Total protein lysates were extracted from frozen tissue in GT1 as well as a group of control tumors, including GIST, angiosarcoma, as previously described (Agaram et al., 2007). Electrophoresis and Immunoblotting were done on the total protein extract (30μg) following standard protocols. Notch2 and β-actin were detected by anti-Notch2 (Cell Signaling Technology, Danvers, MA, #5732; 1:1,000 dilution) and by Anti-beta-actin (Cell Signaling, #4970; 1:1,500 dilution). The secondary antibodies used was goat anti-rabbit (Santa Cruz Biotechnology, Dallas, TX; 1:10,000 dilution).

Real-Time PCR for NOTCH2, miR-143/miR-145

Real-Time PCR was done using both TaqMan (NOTCH2) and SYBR Green (miR143HG) systems.1 μg of total RNA was used for cDNA synthesis, using the TaqMan Reverse Transcription Reagents (Invitrogen, # N8080234) and miScript II RT Kit (Qiagen, Valencia, CA, #218160). Real-Time PCR was performed using Invitrogen ViiA 7 for 40 cycles at a 60°C (TaqMan) and 55°C (Qiagen), using TaqMan Universal PCR Master Mix (Invitrogen, Cat.4304437) or miScript SYBR Green PCR Kit (Qiagen, #218073), with the following primers: NOTCH2 (Invitrogen, #HS01050702); NOTCH2-3′End Exon34 (Invitrogen, #AIAAZJ2); GAPDH (Invitrogen, # HS99999905); miR143 (Qiagen, # ms00003514); miR-143* (Qiagen, # ms00008687); miR-145 (Qiagen, # ms00003528); miR-145* (Qiagen, # ms00008708) and Run6 (Qiagen, #ms00033740). The miRNA nomenclature and abbreviations used includes: MIR143, human microRNA gene; miR-143, human mature microRNA; miR-143*, complementary strand. The miRNA expression values were calculated based on 2^ (−ΔΔCt) values, using Run6 as miRNA control, and represent a relative quantification of the Real-Time PCR signal of the target transcript of the sample of interest (i.e. GT1) to that of a control sample (G1-Normal tissue).

Micro-RNA Sequencing

Total RNA was extracted from frozen tumor tissue using Trizol reagent according to the manufacturer’s instructions (Invitrogen). Small RNA cDNA libraries were prepared from 22 mesenchymal tumors, including 3 MIR143-NOTCH fusion positive glomus tumors (GT1, GT2, GT19), one myopericytoma (M1), one infantile myofibromatosis, 3 leiomyomas, 12 leiomyosarcomas, and 2 GISTs, as previously described (Hafner et al., 2010; Italiano et al., 2012). In 20 μl reactions, 2 μg total RNA was ligated to 100 pmol adenylated 3′ adapter containing a unique pentamer barcode at the 5′ end using 1 μg Rnl2(1-249)K227Q (purified from E. coli containing pET16b-Rnl2(1-249)K227Q [Addgene, Cambridge, MA]), in 50mM Tris-HCl, pH 7.6; 10mM MgCl2; 10mM 2-mercaptoethanol; 0.1 mg/ml acetylated bovine serum albumin (BSA; Sigma-Aldrich, St. Louis, MO) and 15% DMSO for 16 hours on ice. After ligation, up to 20 samples bearing unique barcodes were pooled and purified on a 15% denaturing polyacrylamide gel. RNAs of 45 and 50 nucleotides were excised from the gel, eluted, and ligated to 100 pmol 5′ oligoribonucleotide adapter (GUUCAGAGUUCUACAGUCCGACGAUC) as described above for the 3′ adaptors, except that reactions contained 0.2mM ATP and RNL1 instead of RNL2(1-249) K227Q and were incubated for 1 h at 37°C. Ligated small RNAs were purified on a 12% polyacrylamide gel, reverse transcribed using SuperScript III Reverse Transcriptase (Invitrogen), and amplified by PCR. The forward primer was AATGATACGGCGACCACCGACAGGTTCAGAGTTCTACAGTCCGA; reverse transcription and reverse primer was CAAGCAGAAGACGGCATACGA. On average 1,265,133 (range 332,816–2,543,130) sequence reads of miRNAs were obtained per sample.

RESULTS

FusionSeq Identifies Novel Fusions Involving MIR143 with either NOTCH1 or NOTCH2 in the Two Index Glomus Tumors Investigated by RNA Sequencing

FusionSeq identified a MIR143-NOTCH2 fusion as the top candidate in GT1, a malignant gastrointestinal GT (Fig. 4A–C). Alignment of the reads suggested a fusion of MIR143 exon 1 with exon 27 of NOTCH2, fusion transcript sequence, which was then confirmed by RT-PCR (Fig. 1A,B). Furthermore, FISH analysis showed break-apart signals in both NOTCH2 and MIR143 genes (Fig. 1C,D). Long range DNA PCR, showed the fusion of intron 1 (13,278 bp) of MIR143 with 444 bp of intron 26 of NOTCH2 (Fig 1E).

Fig. 1. MIR143-NOTCH2 gene fusion in a malignant gastrointestinal glomus tumor (GT1).

(A) Schematic representation of the MIR143-NOTCH2 fusion indicating the loci that are joint together; MIR143 exon 3 contains the miRNA precursor cluster, composed of pre-miR-143 and pre-miR-145 (the stem-loop structures, indicated with red and blue stars, respectively);

(B) Experimental validation of the fusion by RT-PCR shows the junction sequence between exon 1 of MIR143 and exon 27 of NOTCH2;

(C,D) FISH studies confirming break-apart signals in both NOTCH2 and MIR143 (Red, centromeric; Green, telomeric).

(E) Long Range DNA PCR showing fusion of 13,278 bp of MIR143 intron 1 to the 444 bp of NOTCH2 intron 26;

(F) Western blotting using NOTCH2 ICD antibody showing strong expression of a different size band (red arrow) in keeping with truncated NOTCH2 protein in GT1, compared to wild-type NICD protein seen in the control tumors angiosarcoma (AS) and gastrointestinal stromal tumor (GIST).

In addition to the main fusion candidate MIR143-NOTCH2, a second fusion candidate, NOTCH2-CEP128, was identified by FusionSeq, composed of NOTCH2 exon 26 fused to exon 7 of CEP128. The RT-PCR confirmed a fusion transcript composed of NOTCH2 exon 26 fused to the complimentary strand of CEP128 exon 7, with an intervening small fragment of CEP128 intron 7 (Suplem Fig 1A,B), by using the following primers: NOTCH2 ex 26 fwd 5′-CTGCTCCTCCCCACTTCC-3′ and CEP128 Ex5 fwd 5′-GGAACAATCAATCGACCAACTCC-3′. A CEP128 break-apart signal was also validated by FISH (Suplem Fig 1C) in almost 100% of the cells tested. In summary, the above RT-PCR and DNA PCR results confirmed a single DNA intronic break within intron 26 of NOTCH2. The subsequent fusion transcripts composed of 3′NOTCH2 exon 27 fused to 5′MIR143 exon 1 and 5′NOTCH2 exon 26 fused to the complimentary strand of 3′CEP128 exon 7, suggest the possibility of a three-way translocation. However, the FISH results point to a much more complex and unbalanced event, with losses of telomeric ends of NOTCH2 and the centromeric portion of CEP128. As no additional glomus tumors have been found to carry CEP128 gene rearrangement, this complex translocation event found in GT1 is most likely highly unstable and non-recurrent.

The second case tested by RNA-seq, GT2, a benign soft tissue GT from the neck region (Fig. 2A) showed the presence of a MIR143-NOTCH1 as the top candidate on FusionSeq. FISH analysis detected an unbalanced NOTCH1 break-apart with loss of the telomeric part in most cells examined (Fig 2B), while the FISH for MIR143 identified only a small number of cells with break-apart signal (<10%) to be definitive for a positive result. The RT-PCR confirmed the fusion of MIR143 exon 1 with exon 27 of NOTCH1 (Fig. 2C).

No fusion candidates were identified for M1, the multifocal soft tissue myopericytoma (Fig. 4E–H), by FusionSeq analysis.

Recurrent NOTCH2 Rearrangements are Present in both Benign and Malignant Glomus Tumors

Upon screening the entire cohort, NOTCH2 gene rearrangements were present in 17 of 33 (52%) of the GTs tested. In 12 of the 17 (71%) cases NOTCH2 was fused with MIR143 by FISH.

Of the 24 soft tissue glomus tumors 11 (46%) showed NOTCH2 rearrangements, with most of the positive cases occurring in the non-acral soft tissues (9/15, 60%). Only two NOTCH2-rearranged GTs occurred in an acral location, one in the digit of an NF1-patient (Fig. 4I–L) and the other one in a malignant GT of the foot involving both soft tissues and bone. All the remaining somatic 7 digital GTs were negative for NOTCH2 rearrangements. No other GTs showed rearrangements of NOTCH1 upon screening by FISH.

All 5 malignant GTs, regardless of location, visceral or soft tissue, showed rearrangements of NOTCH2, with 3 of them being fused to MIR143. In two of these cases the malignant sarcomatous area was adjacent to or intermixed with a benign GT component (Fig. 4M–O), thus FISH studies were applied separately in the two components (GT6, GT7). In both cases the benign area showed a NOTCH2 break-apart signal, while in the malignant zone there was in addition low level of amplification of centromeric part with loss of the telomeric region (Fig. 4P) in keeping with an unbalanced translocation event.

NOTCH3 Gene Rearrangements are Present in a Subset of Benign Soft Tissue Glomus Tumors

As both NOTCH2 and NOTCH3 have been implicated to function synergistically in regulating vascular smooth muscle development, we sought to test for possible NOTCH3 gene structural abnormalities by FISH in this cohort of pericytic tumors that were negative for NOTCH1/2 rearrangements. As such, we identified three positive GTs for NOTCH3 break-apart, all three being benign histologically and originating in the soft tissue of extremities (knee, 2; forearm, 1). In one of these cases (GT19, Fig. 3A), NOTCH3 was fused to MIR143 by FISH (Fig. 3C,D) and RT-PCR using RNA extracted from frozen tissue confirmed the fusion of MIR143 exon 1 to exon 29 of NOTCH3 (data not shown). Long-range DNA PCR showed the fusion of intron 1 (11,844 bp) of MIR143 to 66 bp of NOTCH3 exon 29 (Fig. 3B).

Fig. 3. MIR143-NOTCH3 fusion in a benign glomus tumor of the forearm (GT19).

(A) Histologic appearance of a benign glomus tumor;

(B) Long-range DNA PCR showed the fusion of 11,844 bp of MIR143 intron 1 to 66 bp of NOTCH3 exon 29.

(C,D) FISH analysis detected break-apart signals for MIR143 and NOTCH3, respectively (Red centromeric; Green, telomeric).

Rare NOTCH2 Rearrangements were Identified in Angioleiomyoma but not in other Subtypes of Pericytic Lesions

Only one of the 17 angioleiomyomas tested showed a rearrangement in NOTCH2 and none in the other genes investigated by FISH. This case occurred in the knee soft tissue of a 32 year-old male and had a typical morphologic appearance indistinguishable from all the others (Fig 4D). No MIR143 break-apart signal was noted in this case.

None of the other members of the pericytic family included (myopericytoma, myofibroma/tosis) showed any structural abnormalities in NOTCH1-3 or MIR143. Similarly the five sino-nasal hemangiopericytomas included in the control group were negative for rearrangements in all genes tested.

Activation of 3′NOTCH2 by Fusion to the Strong Promoter of MIR143

GT1 and M1 index cases were investigated by the Affymetrix U133A gene chip and the mRNA expression was compared to a previously published sarcoma dataset, spanning a large variety of morphologic types, translocation-associated or complex genomics sarcomas (Segal et al., 2003; Hajdu et al., 2010). GT1 showed remarkably high levels of NOTCH2 mRNA expression, compared to M1 and all other types of soft tissue sarcomas (Fig. 5A). Furthermore, Real-Time PCR using primers for either C-terminal or ectodomain of NOTCH2, confirmed the Affymetrix transcriptional data (Fig. 5B), showing high expression of the 3′NOTCH2 mRNA in GT1, while the mRNA expression off the N-terminal of NOTCH2 was lower than the control group and matched normal tissue (Fig 5C). This result is in keeping with a differential upregulation of the 3′end of NOTCH2 represented in the fusion transcript. This finding was further confirmed by Western blotting using an antibody for the intracellular domain (ICD) of NOTCH2, which is maintained in the predicted fusion protein, showing a strong expression of NOTCH2 ICD in GT1 of different size compared to control cases (Fig. 1F).

Fig. 5. MIR143-NOTCH2 fusion results in overexpression of 3′-NOTCH2 mRNA, triggered by the a strong MIR143 promoter, which is highly expressed in smooth muscle lineage.

(A) Affymetrix U133A gene expression showing high levels of NOTCH2 mRNA expression in GT1 compared to M1 and other sarcoma types on X-axis (SS, synovial sarcoma; MLS, myxoid liposarcoma; MFH, malignant fibrous histiocytoma/undifferentiated pleomorphic sarcoma; LS, dedifferentiated liposarcoma; LMS, leiomyosarcoma; FS, adult type fibrosarcoma; CCS, clear cell sarcoma; AS, angiosarcoma); the Y-axis indicates the normalized expression of NOTCH2 mRNA (B) Real-Time PCR using 3′-NOTCH2 primers confirms the U133A high mRNA expression in GT1 compared to M1 and other tumors (GIST, LM, LMS, AS), while (C) Real-Time PCR with primers for the NOTCH2 ectodomain (outside the break) show low mRNA expression in GT1 compared to matched normal or other tumors. (D) Real-Time PCR for miR-143 expression show high levels in MIR143-fusion positive GT1 as well as in other smooth muscle neoplasms lacking MIR143 structural abnormalities; Y-axis for (B–D) represents the relative expression. (E) miRNA sequencing confirms the highly abundant expression of the miR143/miR145 genomic cluster across different smooth muscle tumors regardless of MIR143 rearrangement status (X-axis: GT#1,2,19, MIR143-fusion positive GTs; M1, myopericytoma; Myo1, infantile myofibromatosis; LM, leiomyoma; LMS, leiomyosarcoma; GIST, gastrointestinal stromal tumor; NF, normal fat; WDLS, well-differentiated liposarcoma; DDLS, dedifferentiated liposarcoma; Y-axis, relative frequency is obtained by the ratio of miRNA read counts by total miRNA reads per sample).

We then investigated the expression on MIR143/145 genomic cluster (Fig. 1) in GT1, M1 and other related smooth muscle lesions. We started with miRNA Q-PCR that investigated the mature miR-143 sequence and the complementary strand to the mature sequence miR-143* and also the downstream miR-145 in GT1, M1, other smooth muscle tumors (LM, leiomyoma; LMS; leiomyosarcoma), gastrointestinal stromal tumors (GIST) and angiosarcomas (AS). The expression of both miR-143 and miR-143* and miR145 sequence was markedly upregulated in all smooth muscle tumors, including MIR143-fusion positive G1, as well as leiomyoma, leiomyosarcoma, but not in GIST and angiosarcoma (Fig. 5D).

Additional miRNA profiling in three MIR143-fusion positive GTs, M1 and a subset of smooth muscle tumors (leiomyoma, leiomyosarcoma) using deep sequencing of small RNA libraries confirmed that miR-143 and miR-145 constituted the most abundant miRNA cistron (encompassing 50% of total miRNA expression) in all smooth muscle tumor types, regardless of the MIR143 rearrangement status, compared to other sarcoma types, including angiosarcoma, liposarcoma, and normal tissues such as adipose tissue (Fig. 5E).

DISCUSSION

Due to their morphologic overlap, it has been hypothesized that various pericytic tumors represent a histologic spectrum among a family of neoplasms of perivascular smooth muscle cell derivation (Granter et al., 1998). Despite their wide recognition, no genetic abnormalities have yet been established, to support their present joined classification based on morphologic grounds. In fact a more advanced understanding of their pathogenesis has been established in syndromic rather than sporadic cases. Multiple familial glomus tumors (a.k.a. glomuvenous malformations) show an autosomal dominant inheritance with variable expressivity and incomplete penetrance, being caused by inactivating mutations in the glomulin (GLMN) gene, located in 1p22.1, which is predominantly expressed in vascular smooth muscle cells (Boon et al., 1999; Brouillard et al., 2000). Furthermore, an association between digital glomus tumor and neurofibromatosis has been reported, with a biallelic inactivation of NF1 proposed as the mechanism of glomus tumor formation in this setting (Brems et al., 2009). A small rate of BRAF and KRAS mutations have been detected in sporadic soft tissue glomus tumors, which appears to be within the expected range of mutations described in other tumor types (Chakrapani et al., 2012)

Although morphologically rather distinct, ‘pericytoma’ is a rare but apparently discrete pathologic entity grouped under the spectrum of myopericytic neoplasms due to their perivascular growth pattern and similar immunophenotype (Fletcher et al., 2013a) (Dahlen et al., 2004). A recurrent t(7;12) translocation has been reported in this subset of tumors, however, none of the 6 classic myopericytomas included in the present study were positive for GLI1 gene rearrangements (data not shown), suggesting a different genetic subgroup of tumors.

The dynamic expression of miR-143/miR-145 from its mir-143 genomic region between differentiated and proliferative phenotypes of vascular smooth muscle cells (VSMCs) suggested that their stage-dependent expression may elicit a critical switch for VSMC phenotypic modulation (Cordes et al., 2009). In a miR-143/145-deficient mouse model, the VSMC were locked in a ‘synthetic’ state, which incapacitated their contractile phenotype and favored neointimal lesion development (Boettger et al., 2009). In contrast, overexpression of miR-145 increased expression of VSMC differentiation marker genes, such as SMA, calponin, and SM-MHC, which in turn were downregulated by treatment of VSMCs with a miR-145 inhibitor (Cheng et al., 2009). These data indicate a prominent role for miR-143 and miR-145 in smooth muscle function and are in keeping with our results of MIR143 being the most abundantly expressed miRNA cistron in different smooth muscle tumors tested, regardless of the MIR143 gene rearrangement status. This novel finding most likely indicates a strong promoter of MIR143 within the smooth muscle lineage and its ability upon translocation to drive NOTCH overexpression.

Accumulating evidence from murine models suggests that Notch2 and Notch3 function together to regulate vascular smooth muscle development and smooth muscle differentiation (Wang et al., 2012a). When Notch2 and Notch3 genes are simultaneously disrupted (combined mutations Notch2^−/−; Notch3^−/−), mice die in utero at mid-gestation due to severe cardiovascular abnormalities secondary to lack of smooth muscle differentiation (Wang et al., 2012b). Although assembly of the vascular network occurs normally, however, smooth muscle cells surrounding the vessels are grossly deficient leading to vascular collapse. Furthermore, Notch2 hypomorphic mice have a loss of smooth muscle markers as early as E10.5 (Wang et al., 2012b).

In adults the Notch3 receptor is highly enriched in VSMCs (Joutel et al., 2000; Villa et al., 2001), suggesting that Notch3 plays a critical role in maintaining the phenotypic stability of VSMCs. NOTCH3 loss of function mutations within the ectoplasmic domain are the genetic hallmark of CADASIL disease, which induces degeneration of cerebral vascular smooth muscle cells, with subsequent cerebral autosomal dominant arteriopathy, sub-cortical infarcts and leukoencephalopathy (Joutel et al., 1996; Joutel et al., 1997). It is not clear whether CADASIL pathology occurs as an indirect consequence of the abnormal accumulation of the NOTCH3 protein, as a direct consequence of perturbed NOTCH signal regulation, or due to a combination of both. A small subset of patients with Alagille syndrome secondary to NOTCH2 mutations are also predisposed to multiple vascular pathologies affecting blood vessels derived smooth muscle, including stenosis of the peripheral pulmonary vascular tree and intracranial aneurysms (McElhinney et al., 2002; High et al., 2007). Together, these observations demonstrate that Notch signaling plays an important role in multiple regions of the developing and adult vasculature.

NOTCH1 and NOTCH2 gene rearrangements have been recently identified in a small subset of ER-negative breast carcinoma cell lines (Robinson et al., 2011), and most likely non-recurrent events in one prostatic carcinomas with neuroendocrine phenotype (Lapuk et al., 2012) and one colorectal carcinoma (Wu et al., 2012). Similar to our results, fusion transcripts retained exons that encode the NOTCH intracellular domain (NICD), which is responsible for inducing the transcriptional program following NOTCH activation (Robinson et al., 2011). In this study the index cell lines showed dependence on NOTCH signaling for proliferation and survival as well as marked reduction in proliferation after treatment with γ-secretase inhibitor DAPT.

In a recent systematic analysis of inherited glomuvenous malformations (GVMs) using a sensitive allele-specific pairwise SNP-chip method, a recurrent so-called ‘acquired uniparental isodisomy’ (aUPID) involving chromosome arm 1p was identified, in an A- and T-rich, high-DNA-flexibility region (Amyere et al., 2013). The 1p13.1-1p12 acquired breakpoint was identified in 70% of familial cases studied, in addition to the homozygous glomulin (GLMN) mutations, suggesting that somatic second hits may be required for the formation of GVMs and can explain the variable phenotype and incomplete penetrance observed (Amyere et al., 2013). The 1p12 locus abnormalities are in keeping with the NOTCH2 gene rearrangements seen in the majority of the sporadic glomus tumors in our study. NOTCH2 is located very close to the centromere (alpha satellite) and heterochromatin 1q12 (beta satellite), both structures rich in tandem repeat sequences, including ALU family sequences. This may explain why NOTCH2 is a break-prone site, positioned at a sensitive region for chromosomal 1 organization and structure stability.

In summary we are reporting novel MIR143-NOTCH fusions in more than half of GTs, regardless of anatomic location or degree of malignancy. Despite the different NOTCH gene partners involved (NOTCH1-3), the pattern of fusion is remarkably conserved, with the first exon on MIR143 being fused to most of the NICD domain of NOTCH (Supplementary Fig. 2). The significant overexpression of NICD at both mRNA and protein level suggests that the most likely mechanism of MIR143-NOTCH tumorigenesis is through oncogenic activation of NOTCH driven by the very strong MIR143 promoter as indicated by extremely high miR-143 expression in the smooth muscle cell lineage. The resulting NOTCH1-3 truncated protein would be nearly identical to NICD and potentially sensitive to NOTCH inhibitors (i.e. γ-secretase inhibitors), which seem attractive therapeutic options in malignant or advanced glomus tumors. The high incidence of NOTCH2 gene rearrangements detected by FISH in malignant GTs suggests that this can be a used as a reliable molecular diagnostic test in challenging cases. These results argue that glomus tumors are genetically distinct than most myopericytic tumors and sinonasal hemangiopericytoma-like tumors (a.k.a. glomangiopericytomas), despite their perivascular growth pattern and shared immunophenotype.

Supplementary Material

Supp Fig S1

Supplementary Fig 1. NOTCH2-CEP128 fusion. (A) IGV view and schematic representation of the NOTCH2-CEP128 fusion indicating the loci that are joint together; (B) Experimental validation of the fusion by RT-PCR showing the junction sequence between exon 26 of NOTCH2 and the complimentary strand of last 11 bp portion of CEP128 intron 7, followed by the complimentary strand of entire exon 7 of CEP128; (C) FISH showing unbalanced rearrangement of CEP128 in almost 100% of cells, with loss of the centromeric part (red signal).

Supp Fig S2

Supplementary Fig 2. Schematic diagram of the NOTCH1-3 domains. Red line indicates the breakpoint in each NOTCH member, being fused to MIR143.

Supplementary Table 1. Custom BAC probes used for FISH analysis (GRCh37/hg19)