A germline mutation of CDKN2A and a novel RPLP1-C19MC fusion detected in a rare melanotic neuroectodermal tumor of infancy: a case report

As many pediatric small round cell tumors have either amplification of chromosomal regions containing proto-oncogenes or deletion of regions containing tumor suppressor genes, our expectation was that the MNTI would have copy number variations (CNV). By hybridizing genomic DNA from the MNTI and from the patient’s blood (the reference signal) to a human whole-genome SNP array (4 × 180 k design with a greater density of probes covering 1500 cancer-associated genes), the CNV data proved to be featureless (Fig. 2, innermost track), indicating that the MNTI was predominantly euploid.

Whole-exome sequencing on genomic DNA from both the MNTI and from the patient’s blood, aimed to identify somatic variants by subtracting the germline variants from the total variants detected in the MNTI. All of the somatic single nucleotide variants (SNVs) found in the tumor were synonymous. The patient had eight germline SNVs, which were predicted to be deleterious by any or all of the SIFT, PolyPhen or Condel algorithms (www.​ensembl.​org/​info/​docs/​tools/​vep/​index.​html), and four germline indels. All but one of the indels had prior annotation in dbSNP, and seven of the SNVs may be regarded as SNPs because they were annotated as such in either dbSNP or HGMD. Of most interest, was the SNV in exon 2 of CDKN2A (Chromosome 9:21971137, T to G) which causes mutation of aspartic acid to alanine at position 74 in p16^INK4A and is listed in COSMIC (COSM4163712) as a rare somatic mutation (Table 1). This variant is not a known founder mutation and, until now, has never been described in the germline. In the alternate reading frame of CDKN2A, the corresponding variant is synonymous causing a silent R88R mutation in p14^ARF. No parental DNA was available to identify whether the CDKN2A (p16^INK4A, D74A) mutation was a new mutation or inherited.

By RNA-Seq, we wished to identify differentially expressed genes and fusion genes that might be relevant as potential drivers of the tumorigenesis. As the tumor had been frozen, we were able to extract high quality RNA that was subjected to ultra-high depth (>200 million reads), paired-end RNA sequencing. Analysis with FusionCatcher [11] yielded 50 candidate fusion genes (Table 2 and Additional file 1: Table S1), indicating that the tumor had genomic instability. Intra-chromosomal events were required for 29 of the fusion genes and 23 of these were predicted to be read-through transcripts. Chromosomal translocations were involved in the formation of the remaining 21 fusion genes and for three of these (H2AFV-RP11-386 M24.4, C19MC-RPLP1 and RBBP4-TRA@) reciprocal fusions were identified suggesting balanced translocation had occurred. The fusion of RPLP1, which encodes the large P1 ribosomal protein, to C19MC, the largest microRNA cluster in the human genome, located on chromosome 19q13.41, appears potentially functional. The ~100 kb cluster of 46 microRNA genes is imprinted, with expression largely confined to the first trimester of gestation [12]. Four collagen genes (COL1A1, COL1A2, COL3A1 and COL18A1) were also re-arranged by chromosomal translocation in the MNTI, with COL1A1 and COL3A1 having multiple fusion partners (Fig. 2).

Since the presumed cell of origin of MNTI is from the human neural crest (7,9,10), we also asked how the transcriptional profile of our patient’s tumor might differ from that of human neural crest cells (hNCC). Rada-Iglesias et al. [13] developed an in vitro model which recapitulates neural crest formation during gestation by inducing differentiation of human embryonic stem cells, first into neuroectodermal spheres, and then into migratory hNCC. We compared our RNA-Seq expression data from the tumor with the single RNA-Seq dataset obtained from the differentiated hNCC (GEO acession: GSE28875). The transcriptional profiles were highly similar, in keeping with an hNCC origin for MNTI, with no significant differences in expression for 24146 transcripts out of 24331. We identified just 185 genes with expression that was significantly different (adjusted P-value ≤ 0.05, Fig. 2, bar graph track). The 25 genes showing the greatest differences in expression (Table 3) encode proteins with diverse functions including: ribosomal proteins (RPS17), serum transport proteins (TTR), transcription factors (POU3F3, TFAP2B, SP8) and extracellular matrix proteins (SPON2, DPT). Amongst them is a subgroup of genes that encode components of muscle (TNNT3, MYL1, MYL2, TNNI2), consistent with the finding of strong staining of stromal cells for smooth muscle actin and muscle actin, suggesting that the MNTI has undergone myogenic differentiation. Of note, TYR, which encodes tyrosinase, the enzyme that catalyzes the initial steps in the biosynthesis of melanin from tyrosine, was highly expressed in the MNTI, in keeping with its melanotic phenotype, but was not expressed in hNCC.

The microRNA (miR) gene MIR4737, was the most highly expressed gene in hNCC relative to the MNTI, in which it was not expressed. Down-regulation of expression of MIR4737 in the MNTI implies that the target genes for this miR will be up-regulated. From TargetScan (www.​targetscan.​org), we identified 465 potential targets of hsa-miR-4737, of which only two, EDN2 and MYOD1, were significantly upregulated in the MNTI. The loss of repression of MYOD1, is also consistent with myogenic differentiation and the expression of muscle component genes in the MNTI.

In order to gain insight into how genes differentially expressed in the MNTI, relative to hNCC, might be functioning in a coordinated manner, we carried out a gene set enrichment analysis (GSEA). A total of 28 gene sets were upregulated in the MNTI (data not shown), of which the set having the highest normalized enrichment score was from genes that are enriched in invasive ductal breast carcinoma [14]. The leading edge set of genes from the MNTI enriched in this set comprised: THBS2, SPON2, HLA-DRA, LRRC15, TYROBP, MMP13, LY96, C1QB and C1QA. The coordinated expression of these genes is suggestive of an epithelial to mesenchymal transition (EMT) mediated by the transcriptional regulator Twist1 [14], even though TWIST1 expression was not significantly different in the MNTI from its expression in hNCC.

We reasoned that the potential loss of regulation of Cdk4/Cdk6 due to mutation of an allele of CDKN2A would be expected to make cells of the MNTI more susceptible to cytotoxicity induced by Cdk4/Cdk6 inhibition. To test this, we treated the cell line derived from the MNTI with palbociclib, a specific small molecule inhibitor of Cdk4/Cdk6 [15]. For comparison, we also treated Ewing sarcoma cell lines with palbociclib that were wild-type (SK-N-MC) or had hemizygous deletion (CHP-100) or were nullizygous (A673) for CDKN2A, confirmed by SNP arrayCGH in our laboratory (results not shown). Although not statistically significant, differences in sensitivity were observed (Fig. 3f), with the MNTI cells proving to be the most sensitive to palbociclib (mean IC₅₀ ± s.d., 6.01 ± 2.01 μM) and Ewing cell lines with CDKN2A deletions (CHP-100 and A673) being the least sensitive (mean IC₅₀ ± s.d., 8.03 ± 0.84 and 7.92 ± 1.27 μM, respectively; MNTI IC₅₀ vs. CHP-100 IC₅₀, P = 0.4, Mann–Whitney test).

Since we had been unable to detect any unequivocal driver mutations, previously described oncogenic fusion genes or CNV that might account for tumorigenesis of the MNTI, we hypothesized that epigenetic mechanisms might be involved. Furthermore, we observed variegation upon deriving the MNTI cell line, with a population of differentiated, melanotic cells co-existing with a less differentiated population at early passages (Fig. 1f). The differentiated cells did not persist into later passages (Fig. 1g), which suggested a phenotypic switch. Accordingly, we tested a selection of 47 small molecule inhibitors targeted against a broad range of proteins involved in epigenetic processes to investigate the importance of epigenetic alterations on the viability and phenotype of the MNTI cell line in culture (Fig. 3). Our rationale for testing this panel was that it would permit candidate enzymes to be identified that were either required for maintaining growth of the tumor or that were responsible for the phenotypic switch to a less differentiated state. Viability was reduced when cells were treated with compounds targeting lysine demethylase enzymes (Fig. 3d). Enzymes affected included: the 2-oxygluterate (2-OG) oxygenase family, inhibited by IOX1, the Jumonji C (JmjC) domain-containing superfamily, inhibited by JIB-04 and Methylstat, and JARID1B/KDM5B, UTX/KDM6A and lysine demethylase 6B, which were all targets of GSK-J4. GSK-J5, a less active isomer of GSK-J4 had little effect. In contrast, bromodomain inhibitors (Fig. 3a), histone deacetylase inhibitors (Fig. 3b) and methyltransferase inhibitors (Fig. 3c) did not cause cell death, with the exception of chaetocin, which also inhibits thioredoxin reductase [16]. The sensitivity of the MNTI cell line to compounds targeting different classes of epigenetic regulators (Fig. 3e), though variable, was less than for inhibitors of lysine demethylase enzymes. Although none of these results achieved statistical significance (pairwise Wilcoxon rank sum tests with Holm correction, all P > 0.05) they do point to a trend. Overall, our data suggest that epigenetic mechanisms and, in particular, the balance between methylation and demethylation of histone lysines are required for maintaining the viability of tumor cells in the MNTI.

Total RNA was extracted using an RNeasy Fibrous Tissue kit (Qiagen, Manchester, UK) from a 12.5 mg piece of the tumor, homogenized with a hand-held TissueRuptor homogenizer (Qiagen). Genomic DNA was extracted using a DNeasy Blood and Tissue kit (Qiagen) from a 19.1 mg piece of the tumor, homogenized with a TissueRuptor and digested overnight with proteinase K (Qiagen). Genomic DNA extracted from the patient’s blood was obtained from the Children’s Cancer and Leukaemia Group Biobank, Newcastle. For whole-exome sequencing, targets were captured using SureSelect (Agilent, Stockport, UK). Samples were prepared and enriched according to the manufacturer’s instructions. The concentration of each library was determined using a QPCR NGS Library Quantification Kit (G4880A, Agilent). Samples were pooled prior to sequencing with each sample at a final concentration of 10 nM.

Aliquots of genomic DNA from the tumor and blood were hybridized to a Cytosure Cancer + small nucleotide polymorphism (SNP) array (Oxford Gene Technology, Begbroke, UK) and read on a SureScan scanner (Agilent). The array was a 4 × 180 k design, comprising a whole genome backbone with a greater density of probes for 1500 cancer-associated genes. Whole-exome sequencing was done on a MiSeq using TruSeq version 3 chemistry to generate 2 × 150 base reads (Illumina, Chesterford, UK). Paired-end RNA-Seq reads were acquired on a HiSeq 2500 (Illumina).

ArrayCGH data were analyzed using CytoSure Interpret software, version 4.3.2 (Oxford Gene Technology). For whole-exome sequencing, reads from Fastq files were mapped to the reference human genome (hg19/b37) with the Burrows-Wheeler Aligner (BWA) package, version 0.6.2. Local realignment of the mapped reads around potential insertion/deletion (indel) sites was carried out with the Genome Analysis Tool Kit, version 1.6 (GATK, Broad Institute). Duplicate reads were marked using Picard, version 1.89 (Broad Institute). Additional BAM file manipulations were performed with Samtools 0.1.18. Base quality scores were recalibrated using GATK’s covariance recalibration. SNP variants were called using VarScan2 and Indel variants were called using GATK SomaticIndelDetector. SNP novelty was determined against dbSNP release 135. Deleterious SNPs were identified using the Variant Effect Predictor, release 77 (Ensembl). Putative fusion genes were identified from RNA-Seq data using FusionCatcher, version 0.99.4c [11]. For analysis of transcript abundance from RNA-Seq data, analyses were carried out using the software tools in the Tuxedo pipeline [41]. Paired-end reads from Fastq files were aligned to the human genome (hg19/b37) with the TopHat-Bowtie2 aligner, versions 2.0.13 and 2.2.5, respectively, and expression of transcripts was quantified with Cufflinks, version 2.2.1, as fragments per kilobase per million mapped reads (FPKM). For statistical analysis of differential expression in RNA-Seq data, the getSig function of CummeRbund was used, with the default alpha of 0.05, to access genes with significantly different expression according to adjusted P-values calculated by Cufflinks [42].

A section of tumor approximately 0.5 cm³ was cut from the surgical resection specimen and was homgenized by hand prior to overnight digestion with Dispase II (Roche, Burgess Hill, UK). Cells were cultured in alpha MEM (Lonza, Slough, UK) with 20 % heat inactivated fetal calf serum, 4 mM glutamine, 100 units/mL penicillin and 100 μg/mL streptomycin in a humidified incubator at 37 °C and 5 % CO₂. After approximately 2 weeks with no visible proliferation, the cell number began to increase. Cells were passaged at a ratio of 1:5 when they reached 80 % confluence using trypsin-EDTA (Sigma, Poole, UK). Immunohistochemical analysis of the cells at passage 6 was performed to characterize the identity of the proliferating cells.

Cells in the exponential growth phase were seeded in 96-well plates at a density of 3000 cells per well in a final volume of 100 μL and were treated, in quadruplet, with a single dose of 47 small molecule inhibitors targeted against proteins involved in epigenetic processes. Viability was measured at days 3, 7 and 10 using PrestoBlue (Life Technologies) according to the manufacturer’s protocol and normalized to vehicle controls. Day 3 viability was determined from triplicate readings and viability on days 7 and 10 from duplicate readings.

Cells were seeded in 96-well plates at a density of 2000 cells per well and were treated with doses of palbociclib (Cambridge Bioscience, Cambridge, UK) in the range of 0.1 to 30 μM. After 72 h, viable cells were assayed by adding CellTiter 96 AQueous One Solution Cell Proliferation MTS reagent (Promega, Southampton, UK) and measuring the absorbance of the formazan product at 490 nm. Log dose–response curves were obtained by fitting a four-parameter logistic function to the data (R 3.3.2, R Foundation for Statistical Computing). Mean IC₅₀ values were derived from at least three independent experiments.