Inactivation of Bap1 Cooperates with Losses of Nf2 and Cdkn2a to Drive the Development of Pleural Malignant Mesothelioma in Conditional Mouse Models Free

Pleural malignant mesothelioma is an aggressive, treatment-resistant cancer, which is causally related to asbestos exposure. Early cytogenetic and deletion mapping studies had uncovered several prominent sites of chromosomal loss in human pleural malignant mesothelioma, including 3p21, 9p21, and 22q12, and these recurrent abnormalities often occurred in combination in a given tumor, suggesting a multistep pathogenetic process (1). Alterations in the target tumor suppressor genes at two of these sites, that is, CDKN2A at 9p21 and NF2 at 22q12, were first discovered more than two decades ago (2–6), whereas the identity of the crucial 3p21 gene in malignant mesothelioma remained a mystery until 2011, when BAP1 was identified as the critical malignant mesothelioma tumor suppressor gene at this location (7, 8). In a recent comprehensive study as part of The Cancer Genome Atlas, somatic genetic alterations of BAP1, NF2, and CDKN2A were observed in combination in 25 of 74 (34%) pleural malignant mesotheliomas (9).

The alterations of CDKN2A involve heterozygous or homozygous deletions of all or part of the locus (2). Abnormalities of NF2 in malignant mesothelioma frequently are biallelic, consisting of both an inactivating mutation and loss of the second allele (10). Biallelic losses of CDKN2A have been documented in approximately 40% of primary malignant mesotheliomas (7) and up to 90% of malignant mesothelioma cell lines (2), whereas inactivating NF2 mutations have been reported in 20%–55% of cases (5–7, 10). Using Sanger sequencing, somatic inactivating mutations of BAP1 were initially reported in 20%–25% of sporadic malignant mesotheliomas (7, 8), but more recent studies with other sequencing platforms such as targeted next generation sequencing and multiplex ligation–dependent probe amplification have uncovered BAP1 alterations in approximately 60% of cases, with a preponderance of exonic deletions (11).

The CDKN2A locus encodes the tumor suppressors p16INK4A and p14ARF, which regulate the Rb and p53 cell-cycle pathways, respectively. In malignant mesothelioma, the deleted region of CDKN2A typically includes exon 2 (2), which encodes portions of p16INK and p14ARF, and is thus predicted to affect both Rb and p53 pathways. Reexpression of p16INK4A in malignant mesothelioma cells results in cell-cycle arrest and tumor suppression/regression (12), while reexpression of p14ARF in malignant mesothelioma cells induces G₁ arrest/apoptosis (13). Underscoring the relevance of Cdkn2a to malignant mesothelioma pathogenesis, heterozygous Cdkn2a knockout mice treated with asbestos develop malignant mesothelioma at a significantly accelerated rate compared with asbestos-treated wild-type (WT) littermates (14). In addition, mice deficient for both p16Ink4a and p19Arf exhibit enhanced asbestos-induced malignant mesothelioma formation relative to mice deficient for either 16Ink4a or 19Arf alone (14).

Loss of the NF2 product, merlin, in malignant mesothelioma leads to cell-cycle progression by upregulation of cyclin D1 both transcriptionally (15) and posttranscriptionally via activation of mTORC1 (16). NF2 also inhibits Pak and FAK signaling, which play key roles in cell migration and spreading, respectively, and inactivation of NF2 in malignant mesothelioma cells promotes invasiveness and spreading (17, 18). Nf2^+/− mice treated with asbestos develop malignant mesothelioma at an accelerated rate compared with asbestos-treated WT mice, suggesting that merlin inactivation contributes significantly to malignant mesothelioma development (4, 19, 20). Moreover, mice with heterozygous losses of both Nf2 and Cdkn2a show further acceleration of asbestos-induced malignant mesothelioma, with resulting tumors exhibiting increased cancer stem cells and enhanced tumor spreading capability compared with that observed in mice with losses of only one or the other of these genes (21). Similarly, conditional knockout (CKO) mice harboring homozygous deletions of both Nf2 and p16Ink4a/19Arf in the mesothelial lining of the thoracic cavity developed a high incidence of pleural malignant mesothelioma that showed increased pleural invasion compared with Nf2;Tp53 CKO mice (20).

BRCA1-associated protein-1 (BAP1) was discovered as an ubiquitin hydrolase that associates with the RING finger domain of BRCA1 and enhances BRCA1-mediated inhibition of breast cancer cell growth (22). BAP1 interacts with ASXL family members to form the polycomb group (PcG) repressive deubiquitinase (PR-DUB) complex involved in stem cell pluripotency and other developmental processes (23). BAP1 also interacts with and deubiquitinates the transcriptional regulator host cell factor 1 (HCF1) (24). Importantly, BAP1 has been shown to form complexes with numerous transcription factors and cofactors, including transcription factor YY1 (25). The BAP1 ubiquitin carboxyl hydrolase activates transcription in an enzyme-dependent manner and regulates the expression of a variety of genes involved in numerous processes, including cell proliferation, DNA damage and inflammatory responses, metabolism, and various mechanisms of cell death (24, 26–29). In animal models, asbestos exposure induces a significant increase in the incidence of malignant mesothelioma in heterozygous Bap1-mutant mice as compared with asbestos-exposed WT littermates (30, 31).

Malignant mesothelioma patients with germline BAP1 mutations have an improved long-term survival compared with malignant mesothelioma patients without such heritable variants (32, 33). However, it remains unclear whether somatic BAP1 mutations/deletions are associated with a poor prognosis in sporadic malignant mesothelioma, as is the case in uveal melanoma and renal cell carcinoma (34, 35). While most human malignant mesotheliomas exhibit somatic alterations of BAP1, NF2, and/or CDKN2A, currently it is not known whether inactivation of BAP1 cooperates with loss of NF2 and/or CDKN2A to drive a more aggressive malignant mesothelioma phenotype. Here, we address these questions experimentally using CKO models.

All mouse studies were performed in accordance with a protocol approved by the Fox Chase Cancer Center (FCCC) Institutional Animal Care and Use Committee. Nf2^f/f;Cdkna^f/f mice in FVB/N genetic background (20), a gift from Anton Berns (Netherlands Cancer Institute, Amsterdam, The Netherlands), were maintained in our laboratory in a mixed FVB/N x 129/Sv background. The LoxP sites in the Cdkn2a locus of these mice permit excision of exon 2, which results in inactivation of both p16Ink4a and p19Arf.

Bap1^f/f mice in FVB/N background were developed in our laboratory using zinc finger nuclease (ZFN) technology, with the assistance of the FCCC Transgenic Mouse Facility. Custom ZFNs targeting Bap1 were designed and validated in mammalian cells by Sigma-Aldrich. A pair of ZFNs was identified that binds to and cuts within a site in intron 5 of Bap1 with high efficiency and specificity. We then designed a donor DNA construct containing LoxP sites in introns 6 and 7 (Fig. 1A), such that upon adenovirus-mediated expression of Cre recombinase there is deletion of exon 7 of Bap1. The ZFN mRNAs and donor DNA were injected into the pronucleus of one-cell embryos of FVB/N mice, and embryos were then transferred into pseudo-pregnant females. DNA from pups was analyzed for correct targeting by PCR amplification of the gene and sequencing. Three founder mice with LoxP sites integrated in the Bap1 locus were identified, one of which was used for experiments reported here. Representative examples of genotyping of DNA from Bap1 CKO mice are shown in Fig. 1B.

In addition to Bap1^f/f, Nf2^f/f, and Cdkn2a^f/f mice, these CKO animals were crossed to generate cohorts with the following genotypes: Bap1^f/f;Nf2^f/f, Bap1^f/f;Cdkn2a^f/f, Nf2^f/f;Cdkn2a^f/f and Bap1^f/f;Nf2^f/f;Cdkn2a^f/f. The following primers and annealing temperatures were used for genotyping: Bap1^f/f 1063: 5′-CCC TGA GAC CCA GAA AAT CA-3′, Tm = 54.3°C; Bap1^f/f 1228: 5′-GGG AGG CTC TTT GAA TTG GA-3′, Tm = 54.9°C; Nf2^f/f 1048: 5′-CTT CCC AGA CAA GCA GGG TTC-3′, Tm = 58.0°C; Nf2^f/f 1049: 5′-GAA GGC AGC TTC TTC CTT AAG TC-3′, Tm = 53.2°C; Cdkn2a^f/f 1025: 5′-GCA GTG TTG CAG TTT GAA CCC-3′, Tm = 57.4°C; and Cdkn2a^f/f 1026: 5′-TGT GGC AAC TGA TTC AGT TGG-3′, Tm = 55.8°C.

Ad5CMVCre (Adeno-Cre) virus was obtained from the Viral Vector Core of the University of Iowa (Iowa City, IA). Mice 8–12 weeks of age were injected intrathoracically (20), specifically intrapleurally, with Adeno-Cre virus (50 μL of 3–6 × 10¹⁰ PFU), with approximately equal numbers of mice of each gender in each cohort. Mice were monitored daily and were euthanized upon visible signs of distress, including extreme fatigue, labored breathing, or when mice exhibited a 10% change in body weight. Tissues of all organs of the pleural and peritoneal cavities were collected from sacrificed mice and tumor specimens were subjected to histopathologic assessment. Portions of tumors were also saved in both O.C.T. Compound and RNAlater Solution (Thermo Fisher Scientific) and immediately frozen at −80°C.

The histopathologic procedures used are essentially the same as described previously (23). In summary, after paraffin embedding, sectioning, and deparaffinization, hematoxylin and eosin–stained sections were used for histopathologic evaluation, and unstained sections were used for IHC. For IHC, sections were subjected to heat-induced epitope retrieval using citrate buffer (pH 6.0) for 20 minutes.

Malignant mesotheliomas were diagnosed by two independent pathologists (A.J. Klein-Szanto and K.Q. Cai). To confirm the diagnosis of malignant mesothelioma, IHC was performed for various malignant mesothelioma markers, including mesothelin, detected with Mesothelin antibody PA5-79698 (Invitrogen), and cytokeratin 8, detected with TROMA-1 antibody (DSHB, University of Iowa, Iowa City, IA). In addition, reverse transcription-PCR (RT-PCR) analysis was also performed to analyze malignant mesothelioma markers, including Wt1, Msln (mesothelin), and Krt18/19 (cytokeratin 18/19). RT-PCR analysis for Msln used primers 5′-ATCAAGACATTCCTGGGTGGG-3′ and 5′-CGGTTAAAGCTGGGAGCAGAG-3′. Primers for other malignant mesothelioma markers have been described previously (4). To assess cell proliferation, IHC staining was performed with antibodies for Ki-67 (Dako/Agilent).

Immunoblots were prepared with 30–50 μg of protein lysate per sample, as described previously (21). Anti-Bap1 antibody (A302-243, 1:2,000) was purchased from Bethyl Laboratories. Anti-p19Arf (5-C3-1, sc-32748 1:1,000), anti-Gapdh (6C5, sc-32233, 1:50,000), and anti-β-actin (C4, sc-47778, 1:50,000) were from Santa Cruz Biotechnology; anti-Nf2 (D1D8, #6995S, 1:1,000), anti-total-Akt (#9272S, 1:1,000), and anti-P-Akt, Ser473 (D9E XP, #4060, 1:1,000) were from Cell Signaling Technology and anti-p16 (ab189034 1:1,000) was from Abcam. Immunoblots were imaged using Immobilon Western Chemiluminescent HRP Substrate (EMD Millipore; P90720 catalog no. WBKLS0500).

Primary normal mesothelial cells were isolated from individual CKO mice, 10–12 weeks of age, with each of the following genotypes: Bap1^f/f, Nf2^f/f, Cdkn2a^f/f, Bap1^f/f;Nf2^f/f, Bap1^f/f;Cdkn2a^f/f, Nf2^f/f;Cdkn2a^f/f, and Bap1^f/f;Nf2^f/f;Cdkn2a^f/f. The mesothelial cells were isolated by sacrificing animals and adding trypsin (0.25% Trypsin-EDTA, Gibco, 25200-056) to the peritoneal cavity, according to the method of Bot and colleagues (36). Normal mesothelial cells at passage 2 were then seeded in 6-well plates (Thermo Fisher BioLite 6-well Multidish, catalog no. 130184) and treated with 4 × 10¹⁰ PFU/mL of either control Ad5CMVempty (Adeno-CMV) or Adeno-Cre virus for 1 hour. Cells were then washed in PBS, cultured in normal mesothelial media (34) for 72 hours, and trypsinized. Next, 5,000 mesothelial cells were seeded in each well of a 6-well nonadherent plate (Corning, Costar) in serum-free DMEM/F12 media supplemented with B27, EGF (10 ng/mL), basic FGF (10 ng/mL), and penicillin/streptomycin, as described previously (21). The experiment was performed in triplicate for each cell genotype. Spheroids were photographed using light microscopy after 9 days of culture.

Tumor RNA was isolated in TRIzol and purified with RNeasy columns (Qiagen). RNA-seq was performed with a HiSeq Sequencer and the following reagents: TruSeq Stranded mRNA Library, HiSeq Rapid SR Cluster, and HiSeq Rapid SBS v2 kits (Illumina). Stranded mRNA-seq libraries were prepared according to Illumina's product guide. First-strand cDNA was synthesized using SuperScript II reverse transcriptase (Thermo Fisher Scientific) and random primers at 42°C for 15 minutes, followed by second strand synthesis at 16°C for 1 hour. Adapters with Illumina P5/P7 sequences and indices were ligated to cDNA fragments, and libraries were pooled and loaded onto the sequencer. Fastq files were aligned to the mm10 mouse genome using TopHat2. Raw sequence counts for each gene were produced with HTseq (https://htseq.readthedocs.io) and differentially expressed genes were identified by DESeq2 (37). RNA-seq heatmaps were constructed from log₂(x+1)-transformed counts/million, standardized across rows, using the “heatmap.2” function from the R gplots library. Raw sequencing data have been deposited in the GEO repository (accession number GSE131942).

For functional enrichment analysis, genes identified as differentially expressed by DESeq2 with nominal P < 0.5 were ranked by fold-change and mapped to human genes using R biomaRt (38). These genes were then analyzed using the GSEAPreranked method of Gene Set Enrichment Analysis (GSEA; ref. 39) (with “classic” enrichment statistic) applied to the curated (c2) and Gene Ontology gene sets from the Molecular Signature Database (MSigDB). Confirmation of differentially expressed genes was performed by quantitative RT-PCR analysis; Taqman (Thermo Fisher Scientific) assay IDs used for 13 genes validated by RT-PCR analysis are shown in Supplementary Table S1.

Injection of Adeno-Cre virus into the pleural space of Rosa26 LacZ reporter (R26R) mice has been previously reported to result in efficient β-galactosidase expression in the mesothelial cell lining of the chest cavity (20). Locotemporal expression of Cre recombinase by intrathoracic injection of Adeno-Cre was used to induce mesothelial cell–specific loss of Bap1, Nf2, and/or Cdkn2a in homozygous single-gene CKO mice and in homozygous compound CKO mice. The injection scheme used for the CKO mice is depicted in Fig. 1C. Tumors that arose in most mouse cohorts were mainly malignant mesotheliomas, although other neoplasms were also observed (Table 1).

In the cohorts of mice with homozygous knockout of a single gene only, few malignant mesotheliomas were seen. Among 32 Bap1^f/f CKO mice injected intrathoracically with Adeno-Cre, a slow growing, well-differentiated malignant mesothelioma was identified in a single animal 45 weeks after injection. Interestingly, this malignant mesothelioma acquired loss of Nf2 and p16Ink4a expression during tumor formation (Supplementary Fig. S1A). As expected, this tumor showed Bap1 loss–related deficient deubiquitination activity; and consistent with the acquired loss of Nf2/merlin, downstream Yap was not phosphorylated (Supplementary Fig. S1B). PCR analysis confirmed that losses of Nf2 and Cdkn2a are acquired (Supplementary Fig. S1C). Malignant mesotheliomas were observed in 2 of 15 Nf2^f/f mice and 0 of 12 Cdkn2a^f/f mice injected with Adeno-Cre, results comparable with those reported in an earlier study (20).

Among the intrathoracically injected mice with homozygous inactivation of two genes, malignant mesotheliomas were observed in 7 of 42 (16.6%) Bap1;Nf2 CKO mice, 6 of 27 (22.2%) Bap1;Cdkn2a CKO mice, and 15 of 24 (62.5%) Nf2;Cdkn2a CKO mice. The highest incidence of malignant mesothelioma (22/26, 84.6%) was observed in Bap1;Nf2;Cdkn2a triple-CKO mice. A summary of malignant mesothelioma incidence among Bap1 CKO mice and mice with the various compound CKO genotypes is shown in Fig. 2A.

With regard to histologic subtype, notably no epithelioid malignant mesotheliomas were observed with any of the mouse genotypes. Sarcomatoid malignant mesotheliomas predominated in most cohorts, with the exception being Bap1;Nf2 CKO mice, in which 6 of 7 MMs showed mixed (biphasic) histology.

Malignant mesothelioma latency varied between 21 and 40 weeks among the different compound double-CKO mice. A markedly shorter malignant mesothelioma latency, 12 weeks, was observed in Bap1;Nf2;Cdkn2a triple-CKO mice, a highly significant difference (log-rank test, P < 0.0001) from the 27 weeks observed for Nf2;Cdkn2a CKO mice or for any of the other compound double-CKO (DKO) mice. The malignant mesothelioma survival differences between the three cohorts with different DKO combinations were not statistically significant (Supplementary Table S2). Kaplan–Meier survival curves of intrathoracically injected mice that developed malignant mesothelioma are shown in Fig. 2B. In addition, malignant mesotheliomas from Bap1;Nf2;Cdkn2a CKO mice were consistently high-grade, very invasive, and highly proliferative tumors (Fig. 3A and B), whereas malignant mesothelioma morphology was generally more variable and less hypercellular with the other mouse genotypes. Immunoblotting confirmed the absence of expression of proteins encoded by conditionally knocked out genes Bap1, Nf2, and Cdkn2a (p16Ink4a and p19Arf) in 12 of 12 malignant mesotheliomas tested, and interestingly there was also acquired loss of expression of Bap1 in 4 of 6 malignant mesotheliomas from Nf2;Cdkn2a CKO mice (Fig. 4). We also assessed the status of Akt in these tumors, because Akt activation occurs frequently in human and murine malignant mesotheliomas (4). Akt activation was seen in all malignant mesotheliomas tested, as indicated by expression of phosphorylated (active) Akt (Fig. 4).

Besides malignant mesothelioma, various other tumor types were also seen in mice injected intrathoracically with Adeno-Cre (Table 1). Among mice with homozygous inactivation of Bap1 alone in the thoracic cavity, 2 lung adenocarcinomas and 1 lymphoma were seen. The remaining mice died of other age-related diseases, including bladder obstructions or lymphoproliferative diseases. Among the compound CKO mice, most notable was the cohort with inactivation of both Bap1 and Nf2, which showed a high incidence (28/42, 66.7%) of hepatocellular carcinoma (HCC) and intrahepatic cholangiocarcinoma (ICC; Supplementary Fig. S2A and S2B, respectively), with these tumors generally arising later than malignant mesotheliomas in this cohort (median 35 weeks vs. 21 weeks, respectively). Altogether, 28 Bap1;Nf2 CKO mice developed hepatic cancers, including 7 with HCC alone, 4 with ICC alone, 14 with both HCC and ICC, 1 with HCC and malignant mesothelioma, and 2 with HCC, ICC, and malignant mesothelioma.

To assess the cooperativity of Bap1, Nf2, and Cdkn2a losses in vitro, we next used a spheroid growth assay. Primary normal mesothelial cells were isolated from Bap1^f/f, Nf2^f/f, Cdkn2a^f/f, Bap1^f/f;Nf2^f/f, Bap1^f/f;Cdkn2a^f/f, Nf2^f/f;Cdkn2a^f/f, and Bap1^f/f;Nf2^f/f;Cdkn2a^f/f mice. The mesothelial cells were then exposed to either control Adeno-CMV or Adeno-Cre virus for 1 hour, seeded in nonadherent plates in serum-free stem cell medium (21), and photographed after 9 days. There was little or no evidence of spheroid formation in cultured WT mesothelial cells or mesothelial cells with Adeno-Cre–induced homozygous inactivation of one or various combinations of two tumor suppressor genes (representative examples shown in Fig. 5). In contrast, Adeno-Cre–treated mesothelial cells from Bap1^f/f;Nf2^f/f;Cdkn2a^f/f mice, that is, with inactivation of all three malignant mesothelioma driver genes, showed multiple large spheroids, indicative of stem cell features (Fig. 5).

Given the striking shift in malignant mesothelioma latency between Bap1;Nf2;Cdkn2a CKO mice and Nf2;Cdkn2a CKO mice, as well as the consistently more aggressive nature of malignant mesotheliomas from triple-CKO (TKO) mice, we performed RNA-seq analysis on malignant mesotheliomas from these two cohorts to shed light on the contribution of Bap1 inactivation in this experimental setting. This analysis was carried out on all 6 malignant mesotheliomas from Bap1;Nf2;Cdkn2a CKO mice shown in Fig. 4 versus 2 malignant mesotheliomas from Nf2;Cdkn2a CKO mice in which Bap1 expression was retained (tumors 375 and 411 in Fig. 4). Among 1,425 genes identified as differentially expressed with fold-change of at least 2 and false discovery rate (FDR) <1%, significantly fewer genes (628, 44%) were upregulated than downregulated (P = 8.4 × 10⁻⁶, exact binomial test) in malignant mesotheliomas from Bap1-deficient TKO mice versus Bap1-expressing malignant mesotheliomas from DKO mice. Functional enrichment analysis with GSEA identified several highly significant (family-wise error rate < 0.0005) curated gene sets. Prominent among the positively enriched gene sets were those related to writers of histone H3K27Me3, including PRC2 and SUZ12 (Fig. 6A), as well as genes participating in cell proliferation and transcription regulation involving ubiquitin signaling mediated by the Bap1–HCF1–YY1 complex, EED-related stem cell pluripotency, cytokine signaling, and cell adhesion; 100 of the top pathways differentially affected in malignant mesotheliomas from TKO versus DKO mice are presented in Supplementary Table S3. Considering that Bap1 plays an important role in chromatin remodeling and transcriptional regulation (23–25), and the fact that many of the top differentially expressed genes were significantly enriched for PRC2 target genes (40), we decided to focus further on PRC2 targets here. A heatmap of genes differentially expressed in malignant mesotheliomas from triple-CKO mice versus malignant mesotheliomas from Nf2;Cdkn2a double-CKO mice is shown in Fig. 6B. Thirteen cancer-related genes were validated by RT-PCR analysis, 6 of which (e.g., Ptchd1 and Diras1) are depicted in Fig. 6C and Supplementary Fig. S3. Mmp9 protein was also shown to be overexpressed and activated in malignant mesotheliomas from TKO mice versus those from DKO mice (Fig. 6D).

In this investigation, our aim was to induce mesothelial cell–specific homozygous deletions of three tumor suppressor genes (Bap1, Nf2, and Cdkn2a) that have been implicated as major drivers of human malignant mesothelioma pathogenesis. While the majority of malignant tumors observed were malignant mesotheliomas, additional tumor types also resulted, possibly due to infection of other intrathoracic tissues or by passage of virus via blood vessels to the liver. Notably, HCCs and ICCs were very common later in life in Bap1;Nf2 CKO mice but were observed in only two CKO mice with any of the other genotypes. Notably, liver-specific deletion of Nf2 in the developing or adult mouse has been reported to promote a marked, progressive expansion of progenitor cells throughout the liver, with all surviving mice eventually developing both HCC and ICC (41). Interestingly, HCC and ICC were not seen in our Bap1;Nf2;Cdkn2a CKO cohort, perhaps because these mice succumbed to malignant mesothelioma before hepatic tumors could be detected.

Overall, the malignant mesotheliomas observed in our GEM cohorts generally showed morphologic features, IHC staining, and invasiveness similar to that of human sarcomatoid malignant mesotheliomas, with the exception being those observed in Bap1;Nf2 CKO mice, in which 6 of 7 malignant mesotheliomas showed mixed (biphasic) histology that also was similar histologically to their human biphasic counterparts.

The Kaplan–Meier curves (Fig. 2B) indicate that homozygous inactivation of different combinations of Bap1, Nf2, and Cdkn2a affect malignant mesothelioma latency differently, with losses of all three genes profoundly accelerating tumor development. Specifically, Bap1;Nf2;Cdkn2a CKO mice bearing malignant mesothelioma showed much shorter tumor latency than mice in any compound double-CKO cohort. Moreover, malignant mesotheliomas from triple-CKO mice consistently appeared to be high grade, very invasive, hypercellular tumors, whereas malignant mesotheliomas from mice with other genotypes were more variable phenotypically. All malignant mesotheliomas observed in our study were sarcomatoid or biphasic, similar to what was reported by Jongsma and colleagues for mice with homozygously floxed conditional compound alleles that were injected intrathoracically with Adeno-Cre (20). For example, in their Nf2;Cdkn2a CKO mice, 31 malignant mesotheliomas were sarcomatoid, 13 had mixed histology, and only 1 was epithelioid (20). These investigators had speculated that the predominant epithelioid histology seen in human malignant mesotheliomas, as opposed to mice, may be the result of species-specific differences or route of induction, including the long latency and exposure to asbestos in the human disease counterpart (20). It is also likely that the particular mouse strain can influence the phenotype observed. For example, in our previous asbestos carcinogenesis work, Nf2^+/− mice in a 129Sv/Jae genetic background injected intraperitoneally with asbestos developed roughly equivalent percentages of epithelioid, mixed, and sarcomatoid malignant mesotheliomas (4), whereas asbestos-injected Nf2^+/− mice in a FVB/N background mostly developed sarcomatoid malignant mesotheliomas (21).

Heterozygous knockout (+/−) and heterozygous mutant (+/mut) Bap1 mice injected intraperitoneally with asbestos exhibit a highly significant increase in the incidence of malignant mesotheliomas as compared with asbestos-injected WT littermates, providing in vivo evidence that Bap1 inactivation plays an important role in malignant mesothelioma tumorigenesis (30, 31). Similarly, heterozygous Nf2 mice injected with asbestos develop malignant mesothelioma with a markedly shorter latency than asbestos-treated WT littermates (4, 19). Heterozygous Cdkn2a knockout mice treated with asbestos also develop malignant mesothelioma at an accelerated rate compared with asbestos-treated WT mice (14). Furthermore, asbestos-exposed compound Nf2^+/−;Cdkn2a^+/− mice show further acceleration of asbestos-induced malignant mesothelioma and increased presence of cancer stem cells (21). These carcinogenesis studies, in combination with the gene knockout model studies reported here strongly support a working hypothesis that Bap1, Nf2, and Cdkn2a are key drivers of malignant mesothelioma pathogenesis and that the collective losses of these three tumor suppressors is sufficient to drive rapidly disseminated, lethal disease, similar to the poor clinical outcome of human pleural malignant mesothelioma.

In one study, CKO mice in which only one of these tumor suppressor genes was homozygously deleted developed malignant mesothelioma at a low rate, with spontaneous malignant mesothelioma–like tumors being identified in 5 of 30 Nf2^f/f mice and 0 of 17 Cdkn2a^f/f mice injected intrathoracically with Adeno-Cre (20), supporting the notion that malignant mesothelioma development requires the combined involvement of multiple tumor suppressor gene alterations (42). Similarly, in our Bap1 CKO cohort, malignant mesothelioma was seen in only 1 of 32 mice. The fact that malignant mesothelioma arises spontaneously in a considerable number of CKO mice made deficient for both alleles of two or more tumor suppressor genes, provides compelling support for the role of genetics in malignant mesothelioma causation and the need for accumulated genetic and epigenetic alterations for tumor formation (1). As an example, even in the one Bap1 CKO mouse that did develop malignant mesothelioma, it is intriguing that there was acquired loss of Nf2 and p16Ink4a expression during tumorigenesis.

The in vivo findings presented here with CKO mice closely mimic those found in the human disease counterpart, in which losses of BAP1, CDKN2A, and NF2 are frequently seen in various combinations (7, 43). Their frequent occurrence in malignant mesothelioma and their individual roles both in various cancer predisposition syndromes and in tumor progression strongly suggest that they are key drivers in malignant mesothelioma pathogenesis. Our initial work on BAP1 provided the first demonstration that genetics influences the risk of malignant mesothelioma, a cancer linked to mineral fiber carcinogenesis (8). However, there is evidence that germline mutations of CDKN2A and NF2 may also modulate susceptibility to mineral fiber carcinogens. For example, Betti and colleagues recently described a family in which both malignant mesothelioma and cutaneous melanoma cases were found to have a deleterious germline missense mutation in the CDKN2A gene, and the family member with malignant mesothelioma was determined to be of the low exposure category (44). In addition, germline mutations in NF2 have been reported in patients who developed malignant mesothelioma. In one case, an individual with neurofibromatosis, type II (NF2) developed peritoneal malignant mesothelioma at the relatively young age of 40 (45). Comparative genomic hybridization analysis and IHC staining of the malignant mesothelioma tissue revealed loss of the NF2 gene and protein product, respectively. A second NF2 patient developed a pleural malignant mesothelioma at the age of 75 years (46). Both patients were reported to have occupational asbestos exposure.

Our observation that CKO mice with losses of all three tumor suppressor genes had much higher malignant mesothelioma penetrance and markedly shorter survival than mice with loss of one or two of these genes fits with the model put forward by Vogelstein and Kinzler (47). They proposed that in solid tumors of adults, alterations in a minimum of three mutated driver genes are needed for a cell to evolve into an invasive (malignant) tumor. In our CKO models, loss of one or two tumor suppressors was not consistently sufficient to result in a high incidence of malignant mesotheliomas. The exception was the Nf2;Cdkn2a CKO model, but even in these mice tumor latency was much longer than in the triple-CKO mouse model. One possibility is that in Nf2;Cdkn2a CKO mice, significantly more time is needed to accumulate additional genetic and epigenetic alterations sufficient to result in an invasive cancer. In fact, 4 of 6 MMs tested from Nf2;Cdkn2a CKO mice acquired loss of Bap1 expression (Fig. 4), suggesting that Bap1 loss may be favored in the progression of these tumors. In human malignant mesothelioma, tumors are characterized by the presence of multiple somatic genetic and cell signaling alterations. The fact that large spheroids can be reproducibly formed in vitro in normal mesothelial cells following Adeno-Cre–induced excision of Bap1, Cdkn2a and Nf2, but not by the excision of only one or two of these genes, also provides experimental support for cooperativity between the inactivation of these three tumor suppressor genes in malignant mesothelioma tumorigenesis.

Our RNA-seq analysis revealed that malignant mesotheliomas from Bap1;Nf2;Cdkn2a CKO mice exhibit positively enriched gene sets consistent with cellular processes previously implicated in Bap1 function, such as genes mediated by the Bap1-HCF1-YY1 complex, which play a role in cellular proliferation and transcription regulation involving ubiquitin signaling (25). In addition, in the experimental context used here, we have uncovered a connection between Bap1 and PRC2 that has been less well established than other previously reported Bap1-related cellular activities.

Interestingly, a considerable number of upregulated genes in malignant mesotheliomas from triple-CKO mice have been implicated as oncogenes or proinvasion/metastasis genes, whereas several tumor suppressor genes were downregulated. Lafave and colleagues demonstrated that BAP1 inactivation leads to EZH2-dependent transformation (48). Although their report did not include RNA-seq analysis on human malignant mesothelioma tumors for comparison with our murine data, they demonstrated that BAP1-deficient malignant mesothelioma cells are sensitive to EZH2 pharmacologic inhibition, and our findings add support to their conclusion that such an approach may have clinical efficacy. It is noteworthy that PRC2 subunits such as Ezh2 have been shown to have context-dependent oncogenic or tumor-suppressive roles in different cancers (49).

To investigate whether our CKO mouse model findings are recapitulated in human malignant mesotheliomas with biallelic inactivation of BAP1, CDKN2A, and NF2 versus malignant mesotheliomas with biallelic inactivation of CDKN2A and NF2 only, we analyzed data from The Cancer Genome Atlas (TCGA; Supplementary Materials and Methods; Fig. 6E and F). GSEA analysis of the TCGA data revealed several positively enriched gene sets associated with SUZ12, PRC2, as well as EZH2 mediated trimethylation of H3K27 in human malignant mesotheliomas with biallelic inactivation of all three tumor suppressor genes (Fig. 6F; Supplementary Table S4), similar to the situation in malignant mesotheliomas from our TKO mice.

Mechanistically, our findings suggest that losses of Bap1, Nf2, and Cdkn2a cooperate by disrupting multiple cellular pathways that lead to the transformation of normal mesothelial cells to malignant mesothelioma cells. Thus, Cdkn2a loss would result in insensitivity to antigrowth signals (via p16Ink4a loss) and evasion of apoptosis (p19Arf loss), whereas Nf2 loss would be expected to result in evasion of apoptosis and tissue invasion/metastasis (Hippo, Pak) and TAZ-related malignant mesothelioma cell transformation and transcriptional induction of distinct pro-oncogenic genes, including cytokines (50). On the basis of the upregulated genes identified in malignant mesotheliomas from TKO mice, other possible effects such as Lmo1-induced ECM and integrin-related self-sufficiency in growth signals, Aldh1a2- and Gli1-related maintenance of cancer stem cells, and Mmp9-induced invasiveness (Fig. 6D) may be mechanistically connected with Bap1 inactivation and will be explored in future studies.

Collectively, Bap1^f/f;Nf2^f/f;Cdkn2a^f/f mice injected intrathoracically with Adeno-Cre appear to be useful for modeling an aggressive form of malignant mesothelioma. These malignant mesotheliomas develop at a high incidence after a short latency period. Moreover, our finding that loss of Bap1 contributes to malignant mesothelioma pathogenesis, in part, via loss of PRC2-mediated repression of certain oncogenic target genes, suggests a potential avenue for therapeutic intervention that may be rapidly tested in this CKO model.

No potential conflicts of interest were disclosed.

Conception and design: A.-M. Kukuyan, C.W. Menges, F.J. Rauscher, J.R. Testa

Development of methodology: A.-M. Kukuyan, Y. Kadariya, K.Q. Cai, F.J. Rauscher

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A.-M. Kukuyan, E. Sementino, Y. Kadariya, C.W. Menges, Y. Tan, K.Q. Cai, S. Peri, A.J. Klein-Szanto

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A.-M. Kukuyan, Y. Kadariya, C.W. Menges, M. Cheung, K.Q. Cai, M.J. Slifker, S. Peri, A.J. Klein-Szanto, F.J. Rauscher, J.R. Testa

Writing, review, and/or revision of the manuscript: A.-M. Kukuyan, Y. Kadariya, C.W. Menges, S. Peri, F.J. Rauscher, J.R. Testa

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): E. Sementino, M. Cheung

Study supervision: A.-M. Kukuyan, Y. Kadariya, J.R. Testa

Other (genotyped the mice to acquire the necessary arms for the experiment; checked mice for tumor development and acquired tumor samples used in the RNA-seq; performed spheroid assay; and obtained necessary Western blots): A.-M. Kukuyan

This work was supported by NCI grants CA175691 (to J.R. Testa and F.J. Rauscher III) and CA06927 (to FCCC) and an appropriation from the Commonwealth of Pennsylvania to FCCC. Other support was provided by the Local #14 Mesothelioma Fund of the International Association of Heat and Frost Insulators and Allied Workers. We thank Emmanuelle Nicolas for RT-PCR analyses, Eric Ross for statistical assistance, Alfonso Bellacosa for critical reading of the manuscript, and Raza Zaidi, Xavier Graña, Siddharth Balachandran, and Richard Pomerantz for scientific advice and technical suggestions. The following FCCC core services assisted this project: Laboratory Animal, Transgenic Mouse, Genomics, Cell Culture, DNA Sequencing, Histopathology, and Biostatistics and Bioinformatics Facilities.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.