Whole exome sequencing of an asbestos-induced wild-type murine model of malignant mesothelioma

Fifteen murine MM cell lines were previously established from ascites generated following intra-peritoneal crocidolite asbestos injection into BALB/c ( n = 4) [ 4], CBA ( n = 5) [ 4] and C57BL/6 ( n = 6) mice [ 23, 24] (Table 1). Established cell lines (median passage 14, range 8–31) were used in this study and all cell lines were confirmed free of Mycoplasma spp. infection by polymerase chain reaction. Cells were cultured as previously described [ 25]. Control DNA and RNA samples were obtained from current generation mice sourced from the Animal Resource Centre (Murdoch, WA, Australia) for each mouse strain in accordance with protocols approved by the University of Western Australia Animal Ethics Committee. DNA was extracted using standard isopropanol purification. RNA was extracted using an RNeasy Mini Kit (QIAGEN, Chadstone, Vic, Australia) and cDNA prepared using an OmniScript Reverse Transcription kit (QIAGEN) with oligo dT primers (Promega, Alexandria, NSW, Australia) following the manufacturer’s instructions.

Genomic DNA was fragmented using an S2 sonicator (Covaris, USA). Whole genome libraries were constructed using a NEBNext Ultra DNA Library kit (New England Biosciences, Ipswich, MA) with local modifications (specifically Ion Torrent sequencing adaptors were added) and the exome was enriched using a SureSelectXT Mouse All Exon kit (Agilent, Mulgrave, Vic, Australia) that targets 49.6 Mb of the murine genome (221,784 exons across 24,306 coding genes). Sequencing was performed using a Proton sequencer (Ion Torrent, Thermofisher Scientific) system according to manufacturer’s instructions at the Lotterywest State Biomedical Facility Genomics (Nedlands, WA, Australia). Two samples were pooled and sequenced on a P1 chip for 520 flows to an average depth of 42×, which resulted in 80% of the targeted regions being covered with at least an average depth of 13.2× (See Additional file 1).

Sequence reads were trimmed using Cutadapt (version 1.9) [ 26], aligned to GRCh38/mm10 with BWA-MEM (version 0.7.13-r1126), duplicates marked with Picard (version 1.141, http://​broadinstitute.​github.​io/​picard/​) and coordinates sorted using Samtools (version 1.3) [ 27]. Single nucleotide substitution variants were detected using a dual calling strategy using qSNP [ 28] and the Haplotype caller module of GATK [ 29]. Short indels of ≤50 bp, were called with the Haplotype. Initial read quality filtering for all variants detected included: a minimum of 35 bases in the CIGAR string indicated a match, 3 or fewer mismatches in the sequencing MD field, and a mapping quality greater than 10. High confidence variants were selected based on passing further variant characteristic filtering requirements and were used in all downstream analyses. These filtering requirements were: a minimum coverage of 8 reads in the control data and 12 reads in the tumour data and a maximum of 1000 reads; at least 5 variant supporting reads present where the variant was not within the first or last 5 bases; at least 4 of the 5 reads with individual start positions; the variant was identified in reads of both sequencing directions; the variant was not less than 5 base pairs from a mono-nucleotide run of 7 or more bases in length; and for somatic variants less than 3% variant evidence identified in the control sample. In addition, for indels a minimum of 4 strong variant supporting reads containing a maximum of 3 variant events (including the indel of interest and any other non-reference occurrence) were required. Indels that were located immediately adjacent to homopolymer regions of at least 3 base pairs and for which the inserted or deleted base were identical to the homopolymer base were filtered out as homopolymer-associated errors. Indels required a tumour variant allele frequency of >35% and all coding indels were manually reviewed using IGV [ 30, 31] in order to be considered high confidence. Variants were annotated with gene feature information and transcript or protein consequences using SnpEff (version 4.2) [ 32]. Genes of interest, such as Bap1, Cdkn2a and Nf2 were visually inspected for any possible somatic mutation using IGV.

Mutation rate was calculated from the total number of missense, nonsense and silent SNVs plus coding indels divided by the size of the exome library design. Genes significantly mutated at a rate higher than background were identified using Genome MuSIC [ 33], where a convolution test p-value threshold cutoff of <0.05 was used to determine significance. Mutations that occurred in all cell lines of the same strain at the same position, but were absent from the matched control were excluded as these were considered to be strain-specific polymorphisms.

Somatic CNVs were identified using coverage information for each tumour exome and matching normal sample calculated by GATK. Normalized and log transformed read count ratios were then used to identify regions of amplification and deletion with ExomeCNV [ 34]. Segmentation was performed using DNAcopy [ 35] and segmented data was used as input for the Genetic Identification of Significant Targets in Cancer version 2.0 (GISTIC2.0) program [ 36] The murine reference genome for GISTIC2.0 was kindly provided by Steven Schumacher (Broad Institute, Cambridge, MA). A q-score of less than 0.1 and a confidence interval of 95% was used to determine the significance of regions of CNV across samples. Homozygous deletions of key genes of interest were visually inspected using IGV. To identify whether known tumour suppressor genes, genes in regions of amplification or deletion identified as being significant in the GISTIC2.0 the genes were cross-referenced with the Tumour Suppressor Gene Database 2.0 [ 37].

Pathways of interest in MM were identified from the literature. Gene lists for both the human and murine versions of each pathway were extracted from KEGG [ 38]. All missense, nonsense and frameshift mutations were annotated with corresponding KEGG pathway information for each sample using DAVID Bioinformatics Resources 6.8 [ 39] and cross referenced with the pathway gene lists. The numbers of genes containing missense, nonsense and frameshift mutations in each pathway of interest was determined. Functional annotation clustering on KEGG pathway annotations using DAVID Bioinformatics Resources 6.8 and WebGestalt [ 40] was performed for all genes identified as being significantly mutated above background mutation rate and genes located in regions of significant amplification or deletion, with a Benjamini-Hochberg adjusted p-value threshold of <0.05 from both tools considered significant.

Quantitative real-time polymerase chain reaction (RT-PCR) was used to determine mRNA expression. Primers for genes of interest were identified from the literature or designed using Primer-BLAST (NCBI) and purchased from GeneWorks (Thebarton, SA, Australia) (See Additional file 2). Quantitative RT-PCR was performed using SYBR® Green Universal PCR Master Mix (ThermoFisher Scientific) for 35 cycles of 95 °C for 30s, 60 °C for 30s and 72 °C for 30s. Differences in gene expression were described as fold-changes between tumour and strain-matched non-tumour sample in order to identify expression changes that may be induced by copy number variation. Sanger sequencing was performed as previously described [ 25] to confirm the presence of indels.

Fifteen murine MM cell lines previously established from ascites generated following intra-peritoneal crocidolite asbestos injection into BALB/c ( n = 4), CBA ( n = 5) and C57BL/6 ( n = 6) mice were assessed for somatic mutations and copy number variation (CNV) aberrations (Table 1) [ 4, 23, 24]. Overall there were a greater number of copy number deletion events than amplifications. A total of 14 regions were significantly affected by copy number changes, including 12 regions of deletion and 2 regions of gain (See Additional file 3). The most significant region was a homozygous deletion encompassing the tumour suppressor gene Cdkn2a which was evident in all 15 murine MM tumour cell lines (q = 5 × 10 ⁻¹⁷) (Fig. 1a and b). The breakpoints and length of the region of loss varied between samples, varying from 64.2 to 4025.6 kilobases in length (Table 2). In one sample, AE17, exon 1 of both Cdkn2a isoforms was retained (NM_001040654.1 and NM_009877.2). Quantitative real-time polymerase chain reaction (PCR) confirmed the absence of Cdkn2a expression in messenger RNA in all samples (Fig. 1c).

No other significant amplifications or deletions in genes known to be affected in human MM aside from Cdkn2a were detected in the dataset. However, a number of deleted loci contained known tumour suppressor genes that have been associated with other cancers, as defined by TSGene2.0, such as Epha, Robo1, Arl6, Cadm2, Lipi, Prb1 and Sir1. The number of copy number events did not correlate with time in culture (See Additional file 1).

Copy number aberrations in candidate loci previously implicated in MM were identified in the region of Setd2 in 2 out 15 cell lines and in Lats2 in 3 out of 15 (20%) cell lines. However, we did not observe reduced mRNA expression of these genes (Fig. 1c) and the regions were not considered to be statistically significantly affected across the samples set through GISTIC2.0 analysis. A region on chromosome 15 encompassing the Myc oncogene was observed to be amplified in 9/15 samples (Fig. 1a and b). Gene expression analysis showed a 150-fold increase in expression of Myc in AE17 (Fig. 1c), and a range of 9-to 24-fold increased expression in the other C57BL/6 samples and two CBA samples (AC29, AC31). A homozygous deletion of Trp53 was found in AC29 which was confirmed in the messenger RNA by the absence of product using RT-PCR.

A total of 1286 somatic SNVs and small indels in 1136 genes were detected in the 15 samples. Of these, 899 were missense mutations, 28 were nonsense mutations, 12 were small indels and 347 were silent mutations. On average, there were 84 mutations per sample (range 22 to 438) of which 74% (average 62, range 12 to 327) were protein coding (Fig. 2a; See Additional file 1). Tumours generated in mice of the BALB/c background had a higher average number of mutations ( n = 225) than those from the CBA ( n = 33) and C57BL/6 ( n = 35) strains (Fig. 2a; See Additional file 1). The median overall rate of mutation was 0.7 mutations/Mb (3.9/Mb for BALB/c, 0.6/Mb for CBA and 0.7/Mb for C57BL/6 samples). Mutation frequency among BALB/c samples was variable, ranging from 1.4–8.9 mutations/Mb (See Additional file 1).

In general, C > T and G > A mutations were highly represented in all samples (Fig. 2b). AB13 and AB22 showed marked differences to the other BALB/c samples with an abundance of C > G and A > T mutations, respectively. For the CBA samples, AC24 showed a higher proportion of G > T mutations, while the other CBA samples showed similar proportions of mutation types. Of the C57BL/6 samples, AE19 was dominated by C > T at the expense of G > A mutations (Fig. 2b).

Fifteen genes were mutated in more than one sample and were found to be significantly mutated above what would be expected as background (Table 3). Several of the genes, such as Mmp8, Dock10, Hoxd3 and Fat3, have previously been reported to show mutation or altered expression in cancer and particularly metastasis [ 41– 45]. Two genes, Cacna2d2 and Dbc1 are candidate tumour suppressors in other cancers [ 46, 47].

There was no evidence of any mutation in genes most commonly identified as being mutated in human MM, such as Bap1, Nf2 or Lats2. However, a single missense SNV in Trp53 occurred in one BALB/c sample (Fig. 1a). Reduced expression of Bap1 or Nf2 in the messenger RNA was not observed between tumours and normal samples. However two BALB/c samples, AB12 and AB13, showed a 56- and 136-fold increase of Bap1 in mRNA expression, respectively. The genes Rb1 and Pten were intact.

Three biological pathways were enriched in the list of genes significantly mutated by somatic mutation or copy number alteration when analysed using WebGestalt and DAVID. Of note the Jak-STAT signaling pathway (WebGestalt adjusted p-value = 0.017; DAVID p-value = 0.016) were significantly altered (Table 4). Six pathways known to be dysregulated in human MM were affected by missense mutations in the murine MM tumour cell lines: 5/15 samples harbored missense mutations in genes in the Wnt signaling pathway; 3/15 in the Hedgehog signaling pathway; 3/15 in the Notch signaling pathway; 2/15 in the mTOR signaling pathway and 2/15 in the p53 signaling pathway (Fig. 3). All six pathways contained missense mutations in the BALB/c samples. The CBA samples showed mutations in the Wnt, Hedgehog and MAPK signaling pathways, while the C57BL/6 strain contained only one sample carrying a single missense mutation in the MAPK signaling pathway and no mutations in the other five pathways (Fig. 3). There were no mutations in the Ras signaling pathway, however Lats2, which initiates cell apoptosis through the Hippo pathway, contained copy number deletions in 3/15 samples.