Comprehensive genomic and tumour immune profiling reveals potential therapeutic targets in malignant pleural mesothelioma

Malignant pleural mesothelioma (MPM) has a poor overall survival with few treatment options. Whole genome sequencing (WGS) combined with the immune features of MPM offers the prospect of identifying changes that could inform future clinical trials. We analysed somatic mutations from 229 MPM samples, including previously published data and 58 samples that had undergone WGS within this study. This was combined with RNA-seq analysis to characterize the tumour immune environment. The comprehensive genome analysis identified 12 driver genes, including new candidate genes. Whole genome doubling was a frequent event that correlated with shorter survival. Mutational signature analysis revealed SBS5/40 were dominant in 93% of samples, and defects in homologous recombination repair were infrequent in our cohort. The tumour immune environment contained high M2 macrophage infiltrate linked with MMP2, MMP14, TGFB1 and CCL2 expression, representing an immune suppressive environment. The expression of TGFB1 was associated with overall survival. A small subset of samples (less than 10%) had a higher proportion of CD8 T cells and a high cytolytic score, suggesting a ‘hot’ immune environment independent of the somatic mutations. We propose accounting for genomic and immune microenvironment status may influence therapeutic planning in the future.


Background
Malignant pleural mesothelioma (MPM) is a rapidly lethal cancer of the mesothelial lining involving the pleura that is causally linked with asbestos exposure [1,2]. Genomics studies of MPM have focused mostly on exome sequencing in combination with copy number, gene expression or methylation analysis [3][4][5]. These studies have confirmed that MPM is an unusual cancer predominantly driven by loss of tumour suppressors (BAP1, NF2 and CDKN2A) with a lack of oncogenic gain-of-function events. Treatment options, including chemotherapy, radiotherapy and surgery, remain largely ineffectual. Genome-guided medicine has yielded benefits in other cancer types, but the translation of genomic knowledge for MPM is yet to be realized.
Targeted HDAC, EZH2 and PARP treatments are being pursued based upon BAP1 status (clinical trial: NCT03207347) and its purported role in chromatin remodelling, transcriptional regulation and DNA repair, respectively [6]. In addition, recent combination immunotherapy trial results offer new treatment options [7][8][9]. With the FDA recently approving combination nivolumab and ipilimumab treatment for patients with unresectable MPM [7], a treatment that also induces good responses in patients with the historically intractable sarcomatoid histology. Interestingly, MPM lacks the genomic markers that have been generally associated with immunotherapy responsiveness, such as, high tumour mutation burden [10][11][12], high neoantigen load and mutational signatures associated with mismatch repair or homologous recombination repair deficiency [13]. Therefore, a deep genomic or immunological characterization of MPM samples may reveal alternative features that could influence treatment choices.
We performed WGS on 58 MPM tumours with matched transcriptome sequencing and combined this with publicly available data to provide a cohort of over 200 MPMs [3,4]. This enabled an in-depth characterization of mutational signatures and driver genes, plus an exploration of the tumour immune microenvironment.

Clinical cohort description and sample processing
We profiled 58 pathologically confirmed MPM tumour samples and matched germline DNA from patients with MPM; the data from the 58 MPM samples is referred to as the 'Creaney et al. ' data throughout the manuscript. The source, sample type and associated clinical details relating to the 58 MPM patients are in Additional file 1: Table S1. The 58 MPM samples were comprised of pleura tissue samples (n = 21), pleural effusions (n = 29) and low passage pleural effusion cell lines (n = 8). Samples were collected for genome-based studies with written informed consent from participants and approval of the Dana-Farber Brigham and Women's Cancer Center Institutional Review Board (Boston, MA, USA) (n = 4 patients), University Hospitals of Leicester National Health Service Trust (Leicester, UK) (n = 12 patients) and the Sir Charles Gairdner and Osborne Park Hospital Research Ethics Committee (SCGOPH REC; Perth, Western Australia, Australia) (n = 42 patients). Pleural tumour tissue was snap frozen in liquid nitrogen after surgical removal. Pleural effusions were drained as clinical indicated then transported to the laboratory at room temperature for processing. For 29 samples, tumour cells were enriched directly from the pleural effusions by CD45 depletion following the manufactures directions (Stemcell Technologies, Tullamarine, Victoria, Australia). For 8 samples, an aliquot of the effusion cell pellet was placed into RPMI-1640 media (Gibco) supplemented with 15% foetal calf serum, 200 mM Hepes, 10 mM 2-mercaptoethanol, 1x glutamax (Gibco), 1x nonessential amino acids (Gibco), plus 1x sodium pyruvate (Gibco) and tumour cells enriched through culturing at 37 o C in a 5%CO 2 humidified atmosphere for between 4 and 10 passages as previously described [14]. All cultures were confirmed to be Mycoplasma spp. free by polymerase chain reaction. Nucleic acids were extracted from pleura tissue, effusion, low passage effusion and blood samples using the Qiagen AllPrep Universal kit. MPM and matched normal DNA samples were screened using an Illumina SNP array and tumour purity assessed using the qpure tool [15]; samples with > 40% tumour cell content were selected for WGS. The WGS analysis was conducted at QIMR Berghofer with approval from the QIMR Berghofer Research Ethics Committee (P3521). Approval for this study was granted by SCGOPH REC (RGS0000001517).

Somatic variant calling
Single nucleotide variants (SNV) were detected using a dual calling strategy using qSNP [20] and the GATK Hap-lotypeCaller [21] and indels (1-50 bp) were called with the GATK HaplotypeCaller as previously described [22]. Variants were annotated with the Ensembl v75 gene feature information and transcript or protein consequences using SnpEff (version 4.2). Structural variants were determined using qSV as previously described [23] and annotated for their potential consequence on genes. Positions of breakpoints were annotated using the Ensembl v75 gene model. Copy number was determined using ascatNGS [24]. Samples identified as having undergone a whole genome duplication event had an overall ploidy of > 2.7 and > 50% of the genome estimated as copy number 3 to 6 with two alleles present, i.e. heterozygosity of those segments with copy number gain. We grouped copy number alterations to high gain (copy number ≥ 6) and homozygous deletion (copy number 0). Recurrent copy number changes were detected for chromosomal regions using GISTIC2.0 [25]. Significant genes with focal copy number changes were identified using confidence level of 0.95 and a q-value < 0.05.

Driver gene analysis
Driver genes were identified based on the frequency and functional impact bias of variants. We applied a consensus approach using three tools to detect significantly mutated genes (SMG) affected by single nucleotide variants and indels. We performed SMG meta-analysis by merging Mutation Annotation Format (MAF) files of the samples from three studies: this study (Creaney et al.), TCGA [4] and Bueno et al. [3]. The SMG methods, including OncodriveFML [26], MutPanning [27] (V2.0) and MutSigCV [28], were run using default parameters. OncodriveFML was executed using CADD v1.0 via the web interface (http:// bbglab. irbba rcelo na. org/ oncod rivef ml/ home). MutSigCV and MutPanning were executed using modules in GenePattern (https:// www. genep attern. org). We considered q-value< 0.05 threshold for OncodriveFML and MutSigCV and FDR < 0.05 for Mut-Panning. Genes were considered as significantly mutated if they were identified by more than two tools except for PBRM1 gene, which was detected by MutPanning and has high number of SV breakpoints.

Detection of mutations in the promoter of TERT and estimation of telomere length
We identified TERT promoter mutations using a pileup approach to determine the number of bases that were variant from the reference. The TERT promoter positions assessed were c.−57 T>G (chr5:1295161), c.−124 G>A (chr5:1295228), c.−138 G>A (chr5:1295242), c.−139 G>A (chr5:1295243) and c.−146 G>A (chr5:1295250) which were previously reported as frequently mutated in mesothelioma [29,30] and/or melanoma [31,32]. A mutation was considered present if the mutant allele frequency was > 10%. We estimated the telomere length for tumour samples relative to the matched normal sample by counting telomere motifs in the whole genome data using qmotif [33].

Mutational signatures
Mutational signature analysis was performed using the COSMIC [34] V3 signatures and the R package YAPSA [35] (version 1.14.0). Single base substitutions and INDEL signatures were assigned to the Alexandrov [36] SNV PCAWG with artefacts and INDEL PCAWG signatures respectively, supplied with YAPSA and downloaded August 16th, 2018. A cut-off of 10% was used to determine signature exposure.

Estimation of homologous recombination repair deficiency
To determine if samples were HR deficient two tools were used, HRDetect [37] and HRD-score [38]. HRDetect was run with SNV mutational signatures identified with deconstructSigs [39] and SV mutational signatures from YAPSA [35]. Samples were considered HR deficient if they were predicted by both tools (HRDetect > 0.7 and HRD-sum > 42).

TCGA cohort data analysis
TCGA MPM exome and transcriptome data for 74 previously described [4] cases were downloaded in BAM format, converted to Fastq format, and sequence reads were analysed using the same pipeline as for the Creaney et al. samples. Single nucleotide substitution variants and indels were detected using the same pipeline as the whole genome analysis. Structural variants were not called in the exome data from TCGA. Copy number was determined using sequenza [40].
Two of the cases that underwent WGS in our cohort (the Creaney et al. cohort) were previously exome sequenced and reported by TCGA (MESO1701 corresponds to TGCA-UD-AABZ and MESO2041 corresponds to TCGA-UD-AAC4) [4]. We used qsignature [41] to confirm the WGS and exome sequencing data was from the same individuals. When presenting genome data from TCGA, we excluded the two cases from TCGA and only reported the WGS data. We performed a comparison of the mutations detected by WGS and TCGA for these two cases (Additional file 2: Fig. S1).

HLA typing and neoantigen prediction
Optitype (v1.3.1) [42] was used to compute class I HLA genotypes for paired tumour-control whole genome datasets using default parameters. Somatic variants were annotated for wildtype and mutant peptide sequences with Ensembl variant effect predictor (v86) (VEP) [43]. High confidence somatic coding variants were used to predict neoantigens using pVAC-Seq (v4.0.10) [44] and NetMHCpan [45]. Epitopes with binding affinity Inhibitory Concentration (IC50) ≤ 500 nM were considered to be potential neoantigens that bind to HLA alleles. Expressed neoantigens (IC50 ≤ 500 nM) were identified using qbasepileup [46] run in SNP mode to count the reference and mutant bases at each mutation position in the RNA-seq BAM files. Duplicate and poorly mapped reads were excluded and a mutation was considered to be expressed if there was a minimum of 10 reads with evidence of the mutation.
Gene fusion events from RNA-seq data were predicted using STAR-Fusion [47] (v1.10.0) with default settings. Predicted gene fusion events were filtered to include only those with evidence of the SV in WGS data; the remaining events were annotated using AGFusion [48] with commands '--middlestar' to indicate the fusion position in the fusion peptide sequence and '--noncanonical' to annotate the fusion with information from all possible transcripts. Annotated gene fusion events were used as input for pVACfuse, which is a component of pVACtools [44] to predict neoantigens using NetMHCpan (v4.0) algorithm [45] with MHC-class-I alleles.

Whole transcriptome sequencing and analysis
Sequence libraries were generated from 1 μg intact RNA using the TruSeq stranded mRNA kit (Illumina, San Diego, CA, USA) from tumour samples. Transcriptome paired-end sequencing reads of 100 bp were generated using a Hiseq 2500 instrument (Illumina, San Diego, CA, USA) to a targeted depth of 100 million reads per sample. Adapter sequences were removed using Cutadapt [16] (v1.11) and aligned using STAR [49] (v2.5.2a) to the GRCh37 assembly and Ensembl v75 gene model. Quality control metrics were computed using RNA-SeQC (version 1.1.8), and gene expression as TPM was estimated using RSEM [50] (v1.2.30). Further analysis was carried out using TPM gene expression values to facilitate cross sample comparisons, deconvolution of immune cells and estimation of cytolytic score.

Analysis of the tumour microenvironment (TME)
The proportion of different immune cells in the tumour micro-environment was estimated from RNA-seq data using CIBERSORT [51]. The TPM counts for the Creaney et al. and TCGA [4] datasets were calculated independently and used for deconvolution of the tumour microenvironment on a per sample basis. The CIBESORT algorithm was run with default settings, excluding quantile normalization, for 100 permutations with the published LM22 reference signature matrix to estimate the abundance of 22 immune cells types. The proportion of estimated immune cells for the Creaney et al. and TCGA [4] datasets was visualized together. Cytolytic activity was calculated for each sample as geometric mean of GZMA and PRF1 expression (using TPM, 0.01 offset) as previously described [52].

Statistical and survival analysis
All statistical analyses were performed in the R (v3.6.0) environment. A p value of less than 0.05 estimated using the Wilcoxon test was considered significant. Survival analysis was performed using patient overall survival and a log rank test. To determine an association with survival and the presence of whole genome duplication (WGD) a log rank Mantel-Cox test was performed. Survival analysis for neoantigen load and cytolytic activity was performed by separating samples into quartiles and comparing low and high quartiles. Survival analysis for the expression of specific genes (TGFB1, CCL2, MMP2 and MMP14) was performed by stratifying samples into lower, middle and upper tertiles of log-transformed TPM+1 gene expression values. Kaplan-Meier survival curves and calculation of p value from a log-rank test were performed using the Survival R package (v3.3-1).

Whole genome analysis of malignant pleural mesothelioma
Whole genome sequencing of 58 MPM and matched germline samples (Additional file 1: Table S1) was generated to an average read depth of 66.7 (49.7-83.6) for tumour samples and 33.7 (23.5-43.7) for matched germline samples (Additional file 3: Table S2). Data from the 58 MPM samples is referred to as the 'Creaney et al. ' data throughout the manuscript. MPM samples included pleural tissues, pleural effusions and cell lines derived from short-term effusion cultures. The patients were predominantly male (76%, 44/58), with an average age of diagnosis of 68 years (range 43 to 95 years). Fifty percent of patients were current or previous smokers with an average of 17 ± 20 pack year smoking history, and 85% reported previous asbestos exposure ( Fig. 1A and Additional file 1: Table S1). A median of 3286 (range 594 to 5855) somatic SNV and 274 (range 55 to 484) indels were detected per sample (Additional file 3: Table S2 and Additional file 4: Table S3). The average mutation burden was 1.36 mutations per megabase (median 1.31, range 0.24 to 2.24) ( Fig. 1B and Additional file 3: Table S2), which is consistent with previous studies reporting a burden of < 2 mutations per megabase [3,4]. We detected a total of 9025 somatic structural variants (SVs) (Additional file 5: Table S4) including insertions and deletions, duplications, inversions and translocations, with a median of 125 events per sample (range 5 to 466) (Fig. 1C). The majority (59%) of breakpoints occurred in intergenic regions, with 38% occurring within introns and < 3% in exons of genes. There were 159 predicted in frame gene fusion events; however, none were recurrent between samples. There were 8 genes identified in fusion events in multiple samples; however, each event was unique with no common gene partners or exon usage patterns.
Whole genome duplication (WGD) was a common event, occurring in 29% (17 of 58) of cases (Fig. 1D). WGD was significantly associated with a decrease in overall survival (p-value < 0.01, Log rank Mantel-Cox) (Fig. 1E). In one pleural tissue sample, loss of heterozygosity (LOH) affected almost 80% of the genome (Fig. 1D). Genomic near-haploidization, that is a genome wide LOH, has been previously described for five MPM tumours [4] which were all BAP1, PBRM1 and SETD2 wildtype, with all of the cases containing inactivating point mutations in, or homozygous deletion of, SETDB1. In agreement with this, the sample with genome wide LOH in our cohort was also BAP1, PBRM1 and SETD2 wildtype with a SETDB1 inactivating mutation (frameshift deletion, chr1:150923937 AGAAG>-). The patient was male, diagnosed at 50 years of age, had no recorded exposure to asbestos and had never smoked.

Mutational signatures from WGS
A mutation signature associated with asbestos exposure has not been identified, therefore we hypothesized that the increased unbiased coverage of the genome afforded by WGS would provide more power to detect signatures. When assigning to the known signatures from COSMIC database v3 [34,36] S3). In addition a total of 6 SV signatures were identified (Fig. 2C).
The majority of samples (54 of 58) contained either SBS5 or SBS40 signature at a proportion of > 50%. Both these signatures have been described as flat signatures that are similar to one another and present in multiple cancer types [36], which means their relative assignment may be uncertain. SBS40 and SBS5 have been shown to correlate with age of diagnosis in some cancer types [36]. Although there was a trend (p 0.035), we did not see a strong correlation of SBS40 with age in our cohort (Additional file 2: Fig. S4A). The indel signature ID5, which was present in most samples (38 of 58) at a proportion of > 40%, showed a stronger association with age (Additional file 2: Fig. S4B). Previously, it has been proposed that the mutational processes that underlie the age-correlated signatures SBS40 and ID5 are similar [36]; however, we did not see a correlation in the abundance of these signatures in the mesothelioma cohort. Additionally, the proportion of SBS40 or SBS5 signatures within patients were not associated with asbestos exposure, smoking history or age of diagnosis.
The four samples that did not contain SBS40/5 signatures had signatures that were unique or infrequent within the cohort (Fig. 2A). The first sample (MESO09746) was comprised of SBS8, SBS25 and SBS39. The SBS39 signature was present in 6 other samples, and currently, there is no known aetiology for this signature. SBS8 has been implicated in homologous recombination deficiency (HRD); however, we could not find any mutations in the BRCA genes of MESO09746, and this sample had a low HRD-sum score (Fig. 2D) and HRDetect score (Fig. 2E) suggesting the sample was not HR deficient. SBS25 has been associated with chemotherapy treatment [36], and this sample had undergone chemotherapy treatment. Similarly, MESO10268 was also pretreated and contained SBS31 also associated with platinum compound chemotherapy [36]. The second sample that did not contain SBS40/5 (MESO2041) was comprised of SBS39 and SBS54, the former has no known aetiology, while SBS54 has been described as a sequence artefact possibly due to germline variant contamination. This sample had the lowest number of mutations within the cohort (594 SNVs) which has impacted the ability to confidently resolve signatures (Additional file 2: Fig. S2A). The third sample (MESO11863) contained a germline The SBS3 HR deficient signature has previously been associated with HRD and germline and somatic BRCA1 and BRCA2 mutations and BRCA1 promoter methylation in breast [53], pancreatic [54] and ovarian [23] cancers. SBS3 was present in 9 samples, with low levels (< 30%) in 7 samples (Fig. 2A). Two of these samples (MESO0013 and MESO0003) also contained a high ID6 signature and high HRD-sum and HRDetect scores suggesting these samples are HR deficient. A somatic frame-shift mutation in BRCA1 (NM_007294.4:c.1504_1508del (p.Leu502fs)) with LOH of the other allele accounts for the signature in MESO0013, while MESO0003 contains a somatic SV break within the BRCA2 gene which potentially results in a disruption of the gene. These findings suggest a low prevalence of HR deficiency in a subset of MPM. Another DNA damage signature present at low prevalence (< 30%) in 10 samples was SBS30 which has been associated with base excision repair deficiency due to inactivating mutations in NTHL1 [55]. However, we did not detect NTHL1 mutations in our cohort.

Somatic driver gene analysis and frequently disrupted genes
Driver oncogenes for MPM have not been previously identified. In order to generate increased power for identifying such genes, we combined the protein coding substitution and indel variants in our cohort with published studies from TCGA [4] and Bueno et al. [3] and used three approaches to identify significantly mutated genes (SMG). The driver gene meta-analysis of 229 patients identified 7 driver genes (BAP1, NF2, TP53, SETD2, LATS2, DDX3X and SETDB1) which were significant in two SMG approaches (q-value of < 0.05) ( Fig. 3A and Additional file 6: Table S5), all of which have previously been reported as MPM tumour drivers [3,4]. Another gene, PBRM1, which is known to frequently undergo copy number loss in MPM [56] was found to be significantly mutated by only one approach (Additional file 6: Table S5). Interestingly, the meta-analysis highlighted low frequency mutations in LATS1 and SETD5 genes, which although not significant are family members of the known MPM driver genes (Fig. 3A). A significantly mutated or candidate driver gene was detected in 61.2% (140/229) of the samples (Fig. 3A).
In addition to this, we analysed the Creaney et al. WGS dataset in isolation to incorporate the WGS-derived gene breakage (SV), copy number alterations (CNA) and gene promoter mutations. The chromosome breakpoints increased the mutational frequency of BAP1, NF2, TP53, SETD2, LATS2 and LATS1 (Fig. 3B) and highlighted other recurrently affected genes including CDKN2A (which was also significant by GISTIC (Fig. 1F)) and RBFOX1. We also found mutations in the promoter of the TERT gene in 3 of 58 samples (5.2%) (Fig. 3B) confirming previous reports of frequent mutations in the TERT promoter [29,30,57,58]. The presence of TERT promoter mutations was not associated with a high tumour mutation burden, a low SV number or relative tumour telomere length (Additional file 3: Table S2). In agreement with previous reports [30], we found that TERT mutations were mutually exclusive to BAP1 mutation, as the 3 samples with TERT promoter mutation and an additional sample with TERT amplification did not contain loss of function BAP1 mutations (Fig. 3B).
The RNA-seq data was used to assess the impact of mutation on expression of these genes. There was a significant difference in expression of BAP1 between BAP1 mutated and wild type samples in the WGS pleura (p 0.042) (Additional file 2: Fig. S5), but not in the samples from TCGA (Additional file 2: Fig. S6). One sample (MESO1147) was wild type for BAP1 mutation but contained a low expression of BAP1. There was also a significant association with mutation and low expression for NF2 in samples from cell lines (p 0.042), pleura (p 0.00017) and effusions (p 0.0047). Although only significant in effusion samples (p 0.005) and TCGA data (p 0.038), the majority of samples of both cohorts with a loss of function mutation in CDKN2A also contained lower expression compared to wild type samples.

Neoantigens in MPM tumours
Somatic mutations may create tumour specific neoantigens that could be important in immune recognition. We used the SNV and indel somatic mutations to predict and found a median of 37 neoantigens per sample (range 3-93) in the Creaney et al. WGS data presented in this manuscript (Fig. 4A) and 18 neoantigens per sample (range 1-116) in data from TCGA (Fig. 4A). The number of predicted neoantigens correlated with number of SNV in the Creaney et al. (Fig. 4C) and TCGA data (Fig. 4D). The Creaney et al. cohort was comprised of samples derived from the pleura, effusions or short-term cell cultures, but the neoantigen load was not associated with sample type (Fig. 4E). The neoantigen load in the pleural tissue samples from the Creaney et al. cohort was significantly higher than TCGA (p 0.00023, Fig. 4E). This difference may be associated with differences in tumour purity between the cohorts or different sequencing approaches possibly due to WGS having the potential to produce more even read depth across the genome and may include coding regions not targeted by the exome sequencing. The RNA-seq was used to determine which of the somatic mutations that were predicted to form neoantigens were expressed. The short-term cell line samples contained significantly more expressed neoantigens (p < 0.05, Fig. 4F) possibly due to their high tumour content. Very few neoantigens were predicted from somatic indels (Additional file 2: Fig. S7A and Fig. S7B). In contrast with other cancers [59], there was no association between indel neoantigen load and overall survival in the Creaney et al. or TCGA data (Additional file 2: Fig. S7C and Fig. S7D).
Gene fusion events represent another potential source of neoantigens. We identified a total of 159 candidate gene fusion events in the 58 samples (Additional file 5: Table S4); 53 were caused by inter-chromosomal events (Additional file 2: Fig. S8A) and 106 by intra-chromosomal events (Additional file 2: Fig. S8B); of these, 13 were detected in the RNA-seq (Additional file 2: Fig. S8C). Only 6 of these fusion events were predicted to result in the formation of neoantigens (Additional file 2: Fig. S8D).

Immune cells within the MPM microenvironment
Deconvolution of the RNA-seq data to estimate immune cells using CIBERSORT revealed M2 macrophages, monocytes, CD4 and CD8 T cells in the tumour microenvironment (TME) of TCGA pleura samples (Fig. 5A) and the pleura and effusion samples of Creaney et al. (Fig. 5B). There was a lower proportion of immune cells in the pleural effusion samples of the Creaney et al. data, likely due to the samples undergoing CD45 depletion prior to genomic analysis. Generally, the most prevalent immune cell type was M2 Macrophages. However for seven samples the proportion of CD8 T cells in the TME was high (> 0.2) (Fig. 5A  and B). The proportion of CD8 T cells correlated with cytolytic activity in pleura and effusion samples (Additional file 2: Fig. S9A). The cytolytic activity has been reported as a prognostic factor in a variety of cancer types [52]; however, in MPM, we did not see an association with survival (Additional file 2: Fig. S9B).

TGFB is associated with M2 macrophages and worse survival
The secretion of chemokines and growth factors, in particular TGFB1, into the TME has a role in MPM development and may affect the immune cells within the TME [60]. The expression of chemokines in our data and TCGA revealed that CCL2 was the highest expressed chemokine in all sample types (cell lines, pleural effusion and pleura) (Fig. 5C and D and Additional file 2: Fig.  S10A and Fig. S10B). Similarly, matrix metalloproteases (MMPs) may promote invasion of MPM cell lines [61] and have been associated with immune desert regions in the TME [62]. We found that MMP14 and MMP2 were the highest expressed MMPs (Fig. 5C and D and Additional file 2: Fig. S10A and Fig. S10B).
The expression of CCL2, TGFB1, MMP14 and MMP2 were associated with the presence of M2 macrophages (Pearson correlation of 0.3, 0.7, 0.6 and 0.6 respectively with a p < 0.001) (Fig. 5E), suggesting that they may be contributing to a cold TME enriched with M2 macrophages. A low expression of TGFB1 was associated with a better survival within the TCGA (Fig. 5F) and Creaney et al. datasets (Fig. 5G) (p < 0.05, log rank test), whereas low MMP2 and MMP14 expression was associated with better survival (p = 0.041 and 0.011 respectively, log rank test), but only in the TCGA data (Additional file 2: Fig. S10C).

Expression of immune checkpoint receptors in MPM
Immune checkpoint blockade (ICB) agent anti-PD1 (Nivolumab) and anti-CTLA4 (ipilimumab) have been approved by the FDA for MPM and showed efficacy in the CheckMate 743 trial [7]. Therefore, we performed a retrospective analysis of immune checkpoint receptors to identify potential ICB targets in MPM tumours that may improve the responses to ICB therapy (Additional file 2: Fig. S11). The expression of VSIR which encodes the V-type immunoglobulin domain-containing suppressor of T cell activation (VISTA) was high in most samples as previously reported [4]. Similarly, high expression of CD276 was observed in most samples. The pleura and pleural effusion samples in the Creaney et al. and TCGA datasets clustered into two groups. A small subset of samples (less than 10%) clustered into group 2 and were associated with higher cytolytic scores and high expression of immune checkpoint receptors (Additional file 2: Fig. S11) indicative of a potential T cell inflamed immune 'hot' phenotype, suggesting these may potentially respond to immunotherapy. Previously, loss of BAP1 has been associated with immunogenicity in MPM [63], but here, we found no association of BAP1 status with neoantigen load (Additional file 2: Fig. S11) or the immune hot subgroup.

Discussion
Previous genomic studies of MPM were performed primarily using exome and transcriptome sequencing [3][4][5]64], with limited whole genome studies. Here, we report a large WGS study of MPM. We confirm that MPM is driven by loss of function of tumour suppressor genes [4], and using WGS, we identify SV breakpoints disrupting key driver genes to extend the mutational spectrum of these genes. Finally, we use RNA-seq to explore the TME to create the most complete picture of MPM.
Obtaining MPM surgical tissue samples for genomic studies is challenging, as surgery does not always provide clinical benefits to all MPM patients as demonstrated by the randomized clinical Mesothelioma and Radical Surgery (MARS) trial [65]. Therefore, genomic analysis of surgical samples is generally limited to a subgroup of patients that are operable. Pleural effusions are a common early feature in most patients with epithelioid MPM and provide a 'window' to the underlying tumour; because the fluid is readily accessible, the fluid may capture more of the tumour heterogeneity as it contains cells shed from multiple areas of the tumour and sequential samples can be obtained. Effusion samples may contain a low proportion of tumour cells, which may inhibit their utility for WGS. We found that cell line and pleural effusion samples contain an overall mutation spectrum similar to MPM tissue samples, in keeping with previous findings [66], supporting the use of effusion samples to profile MPM. However, in our cohort, we performed tumour cell enrichment of effusion samples using low passage cell culture or CD45 depletion to improve tumour purity and detection of mutations. It should be noted that both these approaches will impact the proportion of immune cells within the TME, and future studies using single cell approaches to profile effusion samples would reveal greater insight into tumour immune composition.
Mesothelioma is a carcinogen-associated cancer linked to exposure of asbestos [1]. Several other carcinogenassociated cancers have been associated with a high tumour burden and specific mutational signatures [67] (UV-driven melanoma and smoking in lung cancer). In contrast, we found that MPM contains a modest SNV and indel mutation burden, which agrees with previous studies of MPM [3][4][5]. The most frequent signature observed was SBS40, which has not previously been associated with a specific aetiology and has been detected in a variety of cancer types. One could speculate that Signature SBS40 is present in MPM due to the inflammatory effects of asbestos exposure and reactive oxygen species induced DNA damage. However, this needs to be confirmed in more MPM samples or with functional studies. Several cases harboured specific signatures, which may reflect individual tumour characteristics. For example, we found one case with an APOBEC signature. Interestingly, this patient was the only case in the cohort to have a BAP1 germline variant. However, an association with BAP1 germline variants and the APOBEC signature has not been reported, and the lack of the APOBEC signature within samples that contained somatic BAP1 mutations means they are likely not associated.
BAP1 is known to be frequently disrupted in MPM [3,4,68], the WGS and SV breakpoints, enabled a more complete mutation picture, and we found that BAP1 alteration occurs in over two thirds of WGS cases and was significantly associated with lower gene expression in pleura samples. In our analysis, gene expression in TCGA data was not significantly different between BAP1 mutated and wildtype samples, possibly because exome sequence data is unable to identify all mutation events in BAP1 such as CNA and SV. We did not assess other mechanisms of BAP1 inactivation such as methylation that may contribute to low BAP1 gene expression in samples without BAP1 mutation. The prevalence of BAP1 loss adds impetus to the search for treatments that could benefit this subgroup of patients. An in vitro study demonstrated that a BAP1 mutant cell line responded to a combination of PARP inhibition and cisplatin [69] supporting BAP1 as target for therapy, with patient trials underway in MPM [70,71] and other BAP1 altered tumours. These approaches either directly target BAP1 function using HDAC inhibitors and EZH2 inhibitors or target DNA damage repair mechanisms with the use of platinum chemotherapy or PARP inhibitors [6]. Results from an early stage MPM clinical trial of PARP inhibitors reported a manageable toxicity profile and partial response in 3 of 26 patients, all with BAP1 loss [70]. In breast and ovarian cancer, HR deficiency-associated mutational signatures are predictive of mutations in BRCA1/2 and postulated to be associated with PARP inhibitor response [37]. However, in our WGS-driven analysis, we found these signatures were present at a low frequency in a small subset of mesothelioma patients, with only two samples considered as HR deficient. This suggests that PARP inhibition may only be effective in a small subset of MPM, and we await the outcomes of several PARP inhibitor clinical trials currently underway (NCT03531840 and NCT03207347).
In addition to BAP1, we also identified other previously reported MPM driver genes, and novel candidate low frequency MPM driver genes. RBFOX1 was the gene most frequently disrupted by SV events, which supports previous reports [72]. Whether chromothripsis or other mechanisms causing complex rearrangements was responsible for this recurrent rearrangements is not clear, and future work to profile the chromatin landscape through methylation profiling or ATAC-seq would be useful. We also confirm previous reports of recurrent TERT promoter mutations, which rarely co-occur [57] or are mutually exclusive to BAP1 mutation [30]. However, in our cohort, TERT promoter mutations were detected in 5.2% of samples, which is lower than previous studies reporting 15% [58], 11.6% [29] and 10.4% [30]. The difference in prevalence may be due to the MPM samples that underwent WGS in our cohort being predominantly an epithelioid histologic subtype, since TERT promoter mutations have been reported as enriched in sarcomatoid [58] and non-epithelioid MPM [57]. Unlike previous reports in melanoma [32], the TERT promoter mutations were not associated with a high tumour mutation burden, a low SV number or telomere length [32]; however, our analysis was limited as only 3 samples contained a TERT promoter mutation. Two novel candidate genes were detected (LATS1 and SETD5) and are family members acting within the same pathway as previously reported MPM driver genes (LATS2 and SETDB1) and occur in a mutually exclusive manner. MPM has similarities to clear cell renal cell carcinoma (ccRCC), as both harbour frequent loss of function mutations in SETD2, SETD5 and BAP1, and we confirm in this study PBRM1 [73,74]. PBRM1 mutations have previously been detected in two MPM cases in Bueno et al. [3] and Fig. 6 An overview of key mutation processes and the tumour microenvironment in MPM. Inhaled asbestos fibres located at parietal pleura of lung. Asbestos fibres may trigger cell damage and contribute to initiation of mesothelioma cells. Whole genome sequencing of mesothelioma samples revealed 13 candidate driver genes and mutations were enriched with SBS40/5 mutation signature. Whole transcriptome sequencing identified 'hot' a TME marked with the presence of T cells and cytolytic activity in a subset of samples. The majority of MPM favour the growth of the tumour cells by promoting a "cold" TME comprised of M2 Macrophages and TGFB1 expression. Figure created with BioRe nder. com TCGA [4]; using WGS, we also identified somatic disruptive SV and CNA events as a common mechanism to disrupt PBRM1. PBRM1 and BAP1 loss co-occurred, which may be due to both genes being located within 200 kb of each other on chromosome 3. Interestingly, PBRM1 loss-of-function events in ccRCC have been associated with an increased response to immunotherapy checkpoint blockade [75], warranting investigation for MPM. Checkpoint blockade has emerged as a treatment for MPM patients [7,9]. Whether a tumour will respond to immunotherapy is impacted by both the tumour cells and the tumour immune environment. Markers to identify patients that will respond to immunotherapy will be useful in the management of MPM. We identified a small subset of patients that contained a high proportion of CD8 T cells and a strong cytolytic activity score, suggesting an immune 'hot' phenotype that may be more amenable to immunotherapy. Loss of BAP1 has been described as a candidate predictive marker of immunotherapy response in MPM [76] and in peritoneal mesothelioma BAP1 loss has been linked to an inflammatory tumour microenvironment with increased T cell infiltrate [63]. Our findings did not support a link with BAP1 loss and the immune 'hot' phenotype in MPM. Compared to peritoneal mesothelioma, MPM harbour more frequent loss of CDKN2A, and loss of CDKN2A has been linked to resistance to immunotherapy in non-small cell lung cancer [77]. Therefore, the CDKN2A loss may contribute to the lack of association with the immune 'hot' phenotype in MPM. The most abundant immune cell type in MPM samples was M2 macrophages, suggesting an immunosuppressive phenotype [78]. CCL2 is involved in the polarization of monocytes into M2 macrophages in the TME of a variety of cancers [79] including MPM [80], in agreement with this we detected high expression of CCL2. Similarly, we confirmed high TGFB1 expression, which is also involved in M2-like macrophage polarization [81], and an association with a poorer outcome [82]. The clinical role of these observations is the subject of ongoing studies.
Chemokine-related treatment avenues have been suggested as a potential opportunity of therapy and given their high expression may be relevant to MPM. Potential treatment approaches include targeting the CCL2-CCR2 axis [83] or blocking TGF-β [84]. Blocking TGFβ has previously shown a potential survival benefit in MPM [85]. Tumour vaccines targeting neoantigens is another emerging therapy [86]. However, as the mutation profile of MPMs has a high interpatient heterogeneity, neoantigen vaccine approaches may need to be personalized.

Conclusions
In summary, we extend the mutational analysis of MPM using WGS. By combining our data with existing datasets [3,4] we confirmed previously described driver genes and identified candidate new driver genes. Integration of the transcriptome revealed the complexity of predicting patient treatment response and highlighted the importance of the tumour environment. We show that MPM is driven by mutations in tumour suppressor genes and frequently contains the SBS40 mutational signature. The tumours contain a high M2 macrophage infiltrate within the pleura and effusion TME, which is linked to high TGFB1 and CCL2 expression. A small subset of samples displayed an immune 'hot' phenotype with CD8 T cells and high cytolytic activity (Fig. 6). We propose that, depending on the immune microenvironment, patients may respond to immune checkpoint and/or TGFB blockade. Further genomic analysis of MPM immunotherapy clinical trial samples will be key to better understanding the potential molecular biomarkers and ultimately improving outcomes for patients.
Additional file 1: Table S1. Clinical characteristics of the MPM samples that underwent whole genome sequencing Additional file 2: Figure S1. A comparison of somatic mutations detected by WGS and exome sequencing; Figure S2. Single base substitution somatic mutational signature identified in MPM; Figure S3. Somatic short indel mutational signatures identified in MPM; Figure S4. Spearman correlation of age of diagnosis with mutational signatures; Figure S5. Gene expression of MPM genes in tumours with and without mutation in the Creaney at al. cohort; Figure S6. Gene expression of MPM genes in tumours with and without mutation in TCGA cohort; Figure S7. Indel neoantigen load in MPM; Figure S8. Gene fusion events and neoantigen prediction in MPM; Figure S9. Cytolytic activity and survival in MPM; Figure S10. Chemokine, cytokine, interleukin and matrix metalloproteases expression in MPM; Figure S11. Expression of immune checkpoint receptors in MPM. Table S2. Summary of read depth and the number of somatic mutations identified in MPM samples from whole genome sequencing. Table S3. Somatic coding mutations detected in the 58 MPM.

Additional file 4:
Additional file 5: Table S4. Somatic structural rearrangements detected in the 58 MPM samples. Table S5. Significantly mutated gene analysis in MPM.