Intrinsic apoptosis shapes the tumor spectrum linked to inactivation of the deubiquitinase BAP1

See allHide authors and affiliations

Science  19 Apr 2019:
Vol. 364, Issue 6437, pp. 283-285
DOI: 10.1126/science.aav4902

Tissue specificity of tumor suppression

It is well known that the loss of tumor suppressor genes leads to a limited subset of cancers in specific tissues. But why just those tissues? He et al. found a relatively simple explanation for how this tissue selectivity works for the tumor suppressor BAP1 using a mouse model of BAP1-induced cancer. In most cells, loss of BAP1 caused cell death or apoptosis. But in the tissues that formed tumors, differences in the regulation of genes with antiapoptotic effects allowed the cells to survive, even though BAP1 was lost. At least for this one tumor suppressor, its inactivation would normally cause a cell to die; however, this mechanism is absent in a subset of tissues, allowing the cells to proliferate and cause a tumor.

Science, this issue p. 283


Malignancies arising from mutation of tumor suppressors have unexplained tissue proclivity. For example, BAP1 encodes a widely expressed deubiquitinase for histone H2A, but germline mutations are predominantly associated with uveal melanomas and mesotheliomas. We show that BAP1 inactivation causes apoptosis in mouse embryonic stem cells, fibroblasts, liver, and pancreatic tissue but not in melanocytes and mesothelial cells. Ubiquitin ligase RNF2, which silences genes by monoubiquitinating H2A, promoted apoptosis in BAP1-deficient cells by suppressing expression of the prosurvival genes Bcl2 and Mcl1. In contrast, BAP1 loss in melanocytes had little impact on expression of prosurvival genes, instead inducing Mitf. Thus, BAP1 appears to modulate gene expression by countering H2A ubiquitination, but its loss only promotes tumorigenesis in cells that do not engage an RNF2-dependent apoptotic program.

Heterozygous germline mutations in the deubiquitinase BAP1 cause a human cancer predisposition syndrome with a high incidence of uveal melanoma, mesothelioma, and renal cell carcinoma (15). Consistent with BAP1 being a classic two-hit tumor suppressor gene, patient tumors often exhibit inactivating somatic mutation of the remaining functional BAP1 allele (15). BAP1 deficiency may combine with other oncogenic mutations to promote tumorigenesis because BAP1 is often comutated with BRAF in melanoma (6), GNAQ or GNA11 in uveal melanoma (7), and CDKN2A or NF2 in mesothelioma (8). Mice with heterozygous Bap1 mutations also have an increased tumor incidence, including spontaneous and asbestos-induced mesotheliomas (911). Combined deletion of Bap1, Nf2, and Cdkn2a in the mouse thoracic cavity produces mesothelioma with features of the human disease (12). In other studies, Bap1 deletion from mouse melanocytes synergized with mutant BRAF and ultraviolet light exposure to produce melanoma (13), and Bap1 heterozygosity in nephron progenitor cells combined with Vhl loss to cause renal cell carcinoma (14). Therefore, BAP1 mediates tumor suppression in both mice and humans.

To gain insights into how BAP1 suppresses tumorigenesis, we generated knock-in mice (Rosa26-CreERT2 Bap1iC19A/fl) that express catalytically inactive BAP1 Cys91→Ala (C91A) instead of wild-type BAP1 after tamoxifen treatment (fig. S1A). Bap1C19A/− embryonic stem (ES) cells derived from these mice expressed amounts of BAP1 C91A comparable to those of wild-type BAP1 in Bap1+/− control ES cells (fig. S1B). Tamoxifen-induced expression of BAP1 C91A in adult mice caused splenomegaly, leukocytosis, neutrophilia, anemia, thrombocytopenia, liver damage, and atrophy of the pancreas within 4 weeks (fig. S1, C to H), as does tamoxifen-induced BAP1 deficiency (15, 16). Therefore, BAP1 enzymatic activity is critical for normal hematopoiesis and normal functioning of the liver and pancreas.

Compared with their wild-type or Bap1+/− counterparts, Bap1C91A/− and Bap1−/− liver and pancreas showed increased immunolabeling of cleaved caspase-3, a hallmark of apoptosis (Fig. 1, A and B, and fig. S1, I and J). These data prompted us to explore whether BAP1 is critical for the survival of other cell types. BAP1 loss or inactivation also killed cultured ES cells (Fig. 1C and fig. S2A), primary keratinocytes (fig. S2B), and E1A-immortalized mouse embryo fibroblasts (MEFs) (fig. S2C). The cells appeared to die by apoptosis because caspase-3 and its substrate PARP [poly(ADP-ribose) polymerase] were cleaved (fig. S2D), and cell death was inhibited by the pan-caspase inhibitor emricasan (fig. S2E). We used a genome-wide CRISPR-Cas9 screen (fig. S2F and table S1) to identify Rnf2 as a gene promoting the death of Bap1−/− MEFs (Fig. 1D). RNF2 is one of only two E3 ubiquitin ligases that modify histone H2A with monoubiquitin at residue Lys119 (H2AK119Ub) to suppress gene transcription (17, 18), whereas BAP1 removes H2AK119Ub (19). Thus, increased RNF2-dependent ubiquitination of histone H2A after BAP1 inactivation may promote cell death. RNF2 deletion improved the viability of Bap1−/− ES cells, whereas deletion of RING1, the other E3 that generates H2AK119Ub, provided little benefit (fig. S2, G to J). However, the combined loss of RNF2 and RING1 improved survival of Bap1−/− ES cells more than RNF2 loss alone (fig. S2, I and J). Therefore, RNF2 and RING1 both appear to contribute to the death of Bap1−/− cells.

Fig. 1 BAP1 inactivation induces RNF2-dependent apoptosis.

(A) Rosa26-CreERT2 livers immunolabeled for cleaved caspase-3 (brown) at 4 weeks after tamoxifen treatment. Scale bars, 100 μm. (B) Quantification of the immunolabeling in (A). Each symbol represents a different mouse. Lines indicate the mean. *P < 0.01 by Student’s t test (n = 5 to 7 mice per genotype). (C) Graph indicates the percentage of viable Rosa26-CreERT2 ES cells at 6 days after beginning 4-OHT treatment. Each symbol represents a technical replicate. Lines indicate the mean. (D) Graph indicates guide RNA (gRNA) enrichment in MEFs surviving Bap1 deletion. Each dot represents a distinct gRNA. Ntc, nontargeting control.

If BAP1 deubiquitinates H2AK119Ub made by RNF2, then the enzymatic activity of RNF2 should be required for the death of Bap1−/− cells. Indeed, BAP1 deficiency killed Ring1−/−Rnf2−/− ES cells reconstituted with wild-type RNF2 but not those expressing catalytically inactive RNF2 I53S (I, Ile; S, Ser) (20) (Fig. 2, A and B). To test whether RNF2 might suppress genes required for cell survival in Bap1−/− cells, we used RNA sequencing (RNA-seq) to compare Bap+/+, Bap1−/−, and Bap1−/− Rnf2−/− ES cells. Emricasan was included in the cultures to prevent apoptosis-induced changes in gene expression. BAP1 loss caused many genes to be suppressed or up-regulated in an RNF2-dependent manner (Fig. 2C and fig. S3A), and similar changes were observed in cells expressing catalytically inactive BAP1 C91A (fig. S3B). Expression of the anti-apoptotic genes Bcl2 and Mcl1 was down-regulated in Bap1−/− or Bap1C91A/− cells compared with that in control cells, but expression was normal in Bap1−/− Rnf2−/− cells (Fig. 2D and fig. S3C). Consistent with down-regulation of Bcl2 and Mcl1 driving the death of Bap1−/− cells, ectopic expression of BCL2 and MCL1 ameliorated this death (fig. S3, D and E).

Fig. 2 RNF2 monoubiquitinates histone H2A Lys119 to suppress Bcl2 expression in BAP1-deficient ES cells.

(A) Western blots of Rosa26-CreERT2 Bap1fl/fl Ring1−/− Rnf2−/− ES cells reconstituted with wild-type (WT) RNF2 or RNF2 I53S and at 6 days after 4-OHT treatment. GAPDH, glyceraldehyde phosphate dehydrogenase. (B) Graph indicates the percentage of viable Rosa26-CreERT2 ES cells at 6 days after 4-OHT treatment. Each symbol represents a technical replicate. Lines indicate the mean. (C) Heat map comparison of ES cell transcriptomes. Columns represent three technical replicates of each genotype. (D) Bcl2 expression in ES cells by RNA-seq. Symbols indicate technical replicates. Lines indicate the mean. Rpkm, reads per kilobase million. (E) Graphs indicate ChIP-seq reads in the Bcl2 promoter of ES cells. Blue boxes indicate the location of Bcl2 coding regions. (F) Histone H2AK119Ub at the Bcl2 promoter in ES cells by ChIP-qPCR (quantitative polymerase chain reaction). Symbols represent technical replicates. Lines show the mean. Cells in (D) to (F) were cultured in emricasan to limit apoptosis.

Chromatin immunoprecipitation (ChIP) sequencing of ES cells revealed that BAP1 and RNF2 were bound directly to the Bcl2 promoter (Fig. 2E). We failed to obtain compelling evidence for their binding close to the Mcl1 gene, but as a general rule, BAP1 and RNF2 occupied similar regions across the genome (fig. S3F). Regions with higher BAP1 binding often exhibited lower RNF2 binding. Accordingly, BAP1 loss increased RNF2 recruitment to the Bcl2 promoter (Fig. 2E), as well as to other locations within the genome (fig. S3G). Increased binding of RNF2 to histone complexes was observed in both Bap1−/− and Bap1C91A/− cells (fig. S3H), which suggests that BAP1 does not simply displace RNF2 from chromatin but probably counters ubiquitination events important for RNF2 binding to chromatin. Increased binding of RNF2 to the Bcl2 promoter in Bap1−/− cells coincided with increased amounts of H2AK119Ub in this region (Fig. 2F). Consistent with BAP1 deubiquitinating H2AK119Ub to suppress Bcl2 and Mcl1 expression, abundance of BCL2 and MCL1 proteins was decreased in Bap1−/− ES cells (Fig. 3A). However, amounts of other BCL2 family members, such as BCL-XL, BAK, BAX, BAD, BID, BIM, and PUMA, were unaltered (Fig. 3A and fig. S3I). Deletion of BAK and BAX from Bap1−/− ES cells prevented cleavage of caspase-3 (Fig. 3B) and cell death (Fig. 3C), confirming that BAP1 loss activated the intrinsic apoptosis pathway.

Fig. 3 BAP1 inactivation triggers BAX- or BAK-dependent apoptosis.

(A and B) Western blots of ES cells cultured with (A) and without (B) emricasan. (C) Graph indicates the percentage of viable Rosa26-CreERT2 ES cells at 6 days after 4-OHT treatment. Symbols indicate technical replicates. Lines show the mean. (D) Kaplan-Meier survival curve of mice after tamoxifen treatment. ALT, alanine aminotransferase; U/L, units per liter. (E) and G-CSF (F) in the serum of Rosa26-CreERT2 mice at 4 weeks after tamoxifen treatment. Each symbol represents a different mouse. Lines show the mean. *P < 0.05 by Student’s t test.

We investigated whether lethality caused by Bap1 deletion in adult mice resulted from activation of the intrinsic apoptosis pathway. When Rosa26-CreERT2 Bap1fl/fl mice were intercrossed with Bak−/− Baxfl/fl mice (21), five of nine Rosa26-CreERT2 Bap1fl/fl mice became moribund within 4 weeks of tamoxifen treatment, compared with zero of nine Rosa26-CreERT2 Bap1fl/fl Bak−/− Baxfl/fl mice (Fig. 3D). The improved survival of BAP1-deficient mice in the absence of BAX and BAK coincided with reduced cleavage of caspase-3 in the liver (fig. S4, A and B) and pancreas (fig. S4, C and D); decreased liver damage and pancreatic atrophy (Fig. 3E and fig. S4, E and F); and a lack of anemia, neutrophilia, and thrombocytopenia (fig. S4, G to I). BAX and BAK deficiency also prevented decreased amounts of CXCL5 (C-X-C motif chemokine ligand 5) and increased amounts of G-CSF (granulocyte colony-stimulating factor), interferon gamma, and CCL12 (C-C motif chemokine ligand 12) in Bap1−/− serum (Fig. 3F and fig. S4, J to L). These data indicate that BAX/BAK-dependent apoptosis is a major driver of the hematopoietic, liver, and pancreatic defects caused by BAP1 loss.

A prosurvival function for BAP1 was unexpected given that its loss promotes tumorigenesis. Mutations in BAP1 occur along with mutations of other oncogenes or tumor suppressor genes in uveal melanoma and mesothelioma (68). Therefore, we tested whether such mutations might prevent induction of apoptosis after BAP1 loss and reveal other BAP1-regulated genes. Mutations in GNAQ (G protein subunit alpha q) and BAP1 often coexist in uveal melanoma (22), with GNAQ Q209L or GNAQ Q209P (Q, Gln; L, Leu; P, Pro) causing constitutive activation of the protein kinase ERK1/2 (23) (fig. S5A). However, neither GNAQ Q209L nor GNAQ Q209P protected Bap1−/− ES cells from cell death (fig. S5B). Loss of the tumor suppressor Cdkn2a, which is frequently mutated alongside BAP1 in mesothelioma, also failed to block the death of Bap1−/− ES cells (fig. S5, C and D). Therefore, it seems unlikely that other mutations enable cells to tolerate BAP1 loss.

We investigated whether loss of BAP1 triggers cell death in tissues that are more susceptible to tumorigenesis in human families with mutations in BAP1, such as melanocytes and mesothelial cells. Primary melanocytes from neonatal mouse epidermis were cultured in 4-hydroxytamoxifen (4-OHT) for 4 days to induce Bap1 deletion (fig. S6A) and then characterized by single-cell RNA-seq after an additional 3 days in culture. Melanocytes expressing Mitf (melanocyte-inducing transcription factor), Tyr (tyrosinase), Dct (dopachrome tautomerase), Mlana (melan-A), Tyrp1 (tyrosinase-related protein 1), Slc45a2 (solute carrier family 45, member 2), and Ednrb (endothelin receptor type B) accounted for ~1% of the cells in control cultures, whereas Bap1−/− cultures contained ~3% melanocytes (fig. S6B). Moreover, after 4 weeks in culture, melanocytes were approximately three times more prevalent in Bap1−/− or Bap1C91A/ cultures than in control cultures (Fig. 4, A and B). Thus, BAP1 inactivation appears not to trigger apoptosis in melanocytes but instead facilitates melanocyte differentiation or proliferation or both. Accordingly, Bap1−/− and Bap1C91A/ melanocytes expressed BCL2 and MCL1 normally (fig. S6C). In addition, the melanocytes exhibited minimal binding of RNF2 to the Bcl2 promoter (fig. S6D). Loss of BAP1 or its inactivation in primary omentum–derived cell cultures also did not alter the prevalence of mesothelial cells (identified by their expression of podoplanin but not platelet-derived growth factor receptor alpha) (fig. S6E). Thus, loss of BAP1 appears not to trigger the death of mesothelial cells or melanocytes. Bap1−/− or Bap1C91A/ mesothelial cells, like Bap1−/− or Bap1C91A/ melanocytes, contained normal amounts of MCL1 and BCL2 (fig. S6F) and had little RNF2 bound to the Bcl2 promoter (fig. S6G). Therefore, cell type is an important determinant of whether BAP1 promotes the expression of prosurvival genes. Consistent with this notion, there was little similarity in the gene expression changes induced by BAP1 loss in ES cells compared with those induced in melanocytes (fig. S6H) or mesothelial cells (fig. S6I).

Fig. 4 BAP1 inactivation enhances melanocyte numbers.

(A) Light micrographs of Rosa26-CreERT2 melanocytes after 4 weeks in culture. Scale bars, 500 μm. (B) Quantification of melanocyte numbers in (A). Each symbol represents cells from a different mouse. Lines show the mean. *P < 0.01 by Student’s t test. (C) Mitf expression in melanocytes by RNA-seq. Each symbol represents cells pooled from 6 or 7 neonates. Lines show the mean. (D) Western blots of the melanocytes in (A). (E) BAP1 and MITF expression in human uveal melanomas (TCGA data).

We were intrigued by the fact that Bap1−/− or Bap1C91A/ melanocytes consistently outnumbered control melanocytes after 4 weeks in culture. RNA-seq indicated that Bap1−/− melanocytes had up-regulated expression of genes associated with pigmentation and melanocyte differentiation (fig. S6J). For example, the melanoma oncogene Mitf, which encodes the melanocyte-inducing transcription factor (MITF) (24), had an approximately fourfold increase in expression in Bap1−/− melanocytes compared with control melanocytes (Fig. 4C). MITF protein was also more abundant in the Bap1−/− melanocytes (Fig. 4D). To determine if this gene regulation was also relevant to human cancers, we examined human uveal melanomas in The Cancer Genome Atlas (TCGA) database for BAP1 and MITF expression. Most uveal melanomas lack BAP1 and MITF because of monosomy 3, but analysis of tumors with wild-type MITF revealed a strong inverse correlation between the expression of BAP1 and MITF (Fig. 4E). Overall, our data indicate that many cell types do not survive inactivation of BAP1 owing to RNF2-mediated suppression of Bcl2 family anti-apoptosis genes. However, this gene regulation appears to be muted or absent in cell types such as melanocytes and mesothelial cells, perhaps because of tissue differences in the factors that regulate where RNF2 binds in the genome. One consequence of these tissue differences is that only cell types surviving BAP1 inactivation would be expected to give rise to tumors. A cautionary note is that, unlike our engineered mice that completely lack BAP1 in a majority of cells, human families with heterozygous BAP1 germline mutations lose the remaining allele in a single cell.

Other tumor suppressors, such as BRCA1, RB, and WT1, also suppress cell death in a tissue-specific manner (2527), albeit how they affect the cell death machinery is less clear. Therefore, it is possible that cell type–specific intolerance toward the loss of a tumor suppressor might help to more broadly explain why mutation of particular tumor suppressor genes confers susceptibility to particular tumor types.

Supplementary Materials

Materials and Methods

Figs. S1 to S6

Table S1

References (2839)

References and Notes

Acknowledgments: We thank R. Judson and B. Bastian (UCSF) for suggestions, R. Asuncion and L. Ta for animal husbandry, and the Genentech genetic analysis and pathology cores for technical assistance. All mouse protocols were approved by the Genentech Institutional Animal Care and Use Committee (IACUC). Funding: Work was funded by Genentech. Author contributions: M.H. designed and performed experiments with assistance from M.S.C., D.L.D., A.D., and B.H.; S.K. and R.R. performed bioinformatic analyses; J.D.W. analyzed histological data; S.C., Y.-J.C., and Z.M. performed library generation and next-generation sequencing; S.L. performed immunohistochemistry, and labeling was quantified by J.E.-A. R.C. and M.R.-G. derived ES cells; K.N. and V.M.D. contributed to experimental design; M.H. and K.N. wrote the manuscript with input from all authors. Competing interests: All authors are or were employees of Genentech. Data and materials availability: RNA-seq and ChIP-seq data are available through the GEO database (BAP1 and RNF2 ChIP-seq, GSE120302; ES cell RNA-seq, GSE120413; melanocyte RNA-seq, GSE120414; mesothelial cell RNA-seq, GSE120415; epidermis single cell RNA-seq, GSE120421). Mice and antibody reagents generated by Genentech are available from V.M.D. under a material agreement with Genentech.

Stay Connected to Science

Navigate This Article