Familial and Somatic BAP1 Mutations Inactivate ASXL1/2-Mediated Allosteric Regulation of BAP1 Deubiquitinase by Targeting Multiple Independent Domains

BAP1 was identified as a ubiquitin hydrolase that binds to the RING finger domain of BRCA1 and enhances BRCA1-mediated inhibition of breast cancer cell growth (1). The N-terminal 240 amino acids of BAP1 show significant homology to a class of thiol proteases, designated ubiquitin C-terminal hydrolases (UCH), one of several classes of deubiquitinating enzymes implicated in proteolytic processing of ubiquitin. BAP1 residues forming the catalytic site (Q85, C91, H169, and D184) are completely conserved. BAP1 contains two protein binding motifs for BRCA1 and BARD1 that form a tumor suppressor heterodimeric complex (2), as well as a binding domain for HCF1, which interacts with histone-modifying complexes during cell division (3). The C-terminus of BAP1 contains two nuclear localization signals and a UCH37-like domain (ULD). BAP1 interacts with ASXL family members to form the polycomb group–repressive deubiquitinase complex involved in stem cell pluripotency and other developmental processes (4, 5).

BAP1 is a nuclear-localized UCH that can cleave ubiquitin substrates in vitro. The UCH homology of BAP1 implies a role for either ubiquitin-mediated, proteasome-dependent degradation or other ubiquitin-mediated regulatory pathways in BRCA1 function (1, 6). Regulated ubiquitination of proteins and subsequent proteasome-dependent proteolysis play a role in almost every cellular growth, differentiation and homeostatic process (6, 7). The pathway is regulated by enzymes at the level of proteolytic deubiquitination and ubiquitin hydrolysis and are classified into two families: ubiquitin-specific proteases (UBP) and UCHs. The UBPs cleave ubiquitin or ubiquitin conjugates from large substrates, whereas UCHs such as BAP1 cleave ubiquitin from ubiquitin-conjugated small substrates.

BAP1 exhibits tumor suppressor activity in cancer cells (1, 2). Frequent somatic mutations/deletions of BAP1 are found in metastasizing uveal melanomas (8), malignant mesothelioma (9, 10), and other cancers (Fig. 1A). Families with heritable mutations of BAP1 were first published simultaneously in two papers in Nature Genetics; one reported two families with a high incidence of mesothelioma, uveal melanoma, and several types of epithelial cancer, such as renal carcinoma (10); the second reported two families with numerous cases of atypical melanocytic tumors and occasional uveal and cutaneous melanomas (11). Contemporaneously, a third group independently reported a family with a germline BAP1 mutation associated with uveal melanoma, lung adenocarcinoma, meningioma, and other cancers (12). Subsequently, numerous reports have expanded on the discovery of germline BAP1 mutations in families with these and other cancers (13, 14), now defined as the BAP1 tumor predisposition syndrome.

Mutations/deletions in BAP1 cause premature protein termination, protein instability, and/or loss of UCH catalytic activity. Homozygous deletion of mouse Bap1 is embryonic lethal, while systemic deletion in adults recapitulates features of human myelodysplastic syndrome (15). Heterozygous Bap1 knockout and knock-in mice survive but develop numerous spontaneous malignant tumors and increased susceptibility to asbestos-induced mesothelioma formation (16). Knock-in mice expressing Bap1 revealed that BAP1 interacts with HCF1, OGT, and polycomb group proteins ASXL1 and ASXL2 in vivo (15).

BAP1 functions as part of a large polycomb-like complex that has an obligate partner for enzyme activity, the ASXL1/2 family members (5). Drosophila PcG Calypso protein is highly homologous to BAP1. Calypso complexes with the PcG protein Asx, and this Polycomb repressive deubiquitinase (PR-DUB) complex binds to PcG target genes. The human homologs of Asx are ASXL1/2/3 (Fig. 1B). The N-terminus of ASXL, which contains the highly conserved ASXH domain (ASX homologous, AB boxes), is required for Calypso/BAP1 protein association. Similar to Drosophila Asx, human ASXL1/2 complexes with BAP1 to deubiquitinate H2A ub1 (5), thus providing evidence that an epigenetic modifier and associated proteins play a special role in modifying chromatin. However, the mechanism by which cancer-related mutations in BAP1 and other PR-DUB genes and their encoded proteins/protein complexes leads to tumorigenesis has not been determined.

Mutations of ASXL1/2/3 genes leading to protein truncations have been found to associate with human cancers and other diseases. De novo nonsense mutations in ASXL1 cause Bohring–Opitz syndrome, a severe malformation disorder (17), and de novo truncating mutations in ASXL3 result in a clinical phenotype with some similarities (18). Somatic mutations in ASXL1 in patients with myeloid disorders/leukemias are associated with adverse outcome (19, 20). ASXL1 is an epigenetic modifier that plays a role in polycomb repressive complex 2 (PRC2)-mediated transcriptional repression in hematopoietic cells, and ASXL1 loss-of-function mutants in myeloid malignancies result in loss of PRC2-mediated gene repression of leukemogenic target genes (19).

In this study, we define molecular mechanisms whereby tumor-derived, discrete in-frame deletions and insertions outside of the BAP1 catalytic domain perturb BAP1–ASXL2 interactions leading to tumor-related loss of BAP1 catalytic activity.

The pC3-BAP1-FL(6His), pC3-BAP1-FL, C91S(6His), pC3.1-Myc-BAP1, FL wild-type (WT), and patient-related BAP1 mutations/deletions [C91W, S172R, D672G, ΔE283-S285, ΔE631-A634, ΔK637-C638(sub of N), ΔR666-H669] plasmids were constructed through PCR-based cloning. The nucleotide locations for these mutations, corresponding to those observed in uveal melanoma patients (8), are as follows: C91W(NT: T388G); S172R(NT: C631G); D672G(NT: A2130G); ΔE283-S285(NT: 960-968 del); ΔE631-A634(NT: 2006-2017 del); ΔK637-C638 sub of N(NT: 2026-2028 del); and ΔR666-H669(NT: 2112-2120 del). Note that the C91W mutation was originally reported as C91G (8), but was later corrected following discussions with staff affiliated with the COSMIC cancer database. The pEZ-M12-BAP1-WT and mutation/deletion plasmids were from AMB. The C91S mutation, originally described by Jensen and colleagues (1), is known to yield a protein with no detectable UCH activity. The pFastBacHTa-BAP1, FL WT, and mutations/deletions from patients [C91W, S172R, D672G, ΔE283-S285, ΔE631-A634, ΔK637-C638(sub of N), ΔR666-H669] constructs were subcloned from pC3.1-Myc-BAP1, FL WT, and mutations/deletions plasmids. The UCH-WT and UCH-C91S mutant plasmids were also subcloned (1). All plasmids were sequenced to confirm the authenticity of the BAP1- and ASLX-mutant inserts. In addition, pGEX-4T-1-BAP1-FL was subcloned from pC3.1-Myc-BAP1 and FL-WT, and pGEX-2TK-BAP-UCH-WT (1–250 aa) was previously reported (1). The pGEX-4T-1-BAP1-ULD and pQE30-BAP1-ULD (601–729) plasmids were constructed through PCR-based cloning. Plasmids with the mutations pQE30-BAP1-ULD, L650P, or F660P were introduced using Quikchange site-directed mutagenesis (Stratagene). The pQE30-BAP1-ULD plasmids with mutations or deletions D72G, ΔE283-S285, ΔE631-A634, ΔK637-C638 (sub of N), or ΔR666-H669 were constructed via PCR-based cloning.

The fragments pC3.1-Flag-ASXL2 (261-1435aa), pC3.1-Flag-ASXL2 (261-649 aa), pQE30-ASXL2 (261-649aa), pFastBac-Flag-ASXL2 (261-1435 aa), pFastBac-Flag-ASXL2 (261-649 aa), pFastBacHTb-ASXL2 (261-1435 aa), pFastBacHTb-ASXL2 (261-649 aa) were subcloned from pBluescriptR-ASXL2 (Open Biosystems) into pC3.1-Flag, pQE30, and pFastBacHTb vectors, respectively. The pQE30-ASXL2-AB (261–381 aa), pQE30-ASXL2-A (261–318 aa), and pQE30-ASXL2-B (319–381 aa) plasmids were constructed through PCR-based cloning. The pQE30-ASXL2-AB mutations (R284P, VDR-AAA, NEF-AAA, ERL-AAA, EKE-AAA) were introduced by Quikchange site-directed mutagenesis (Stratagene).

The ³⁵S-labeled BAP1-FL, ASXL2 (261–1,435 aa), and ASXL2 (261–469 aa) probes were produced by coupled in vitro transcription and translation (IVT) using TNT@T7 quick coupled transcription/translation system (Promega).

The baculovirus (Bv) His-BAP1-FL WT and mutant proteins were expressed individually or coexpressed with Flag-ASXL2 (261-649 aa) protein in Bv-infected Sf9 cells (Wistar Protein Expression and Molecular Screening Facilities). High-titer Bv stocks for each protein were prepared using standard techniques, and Sf9 cells were infected for 48 hours. The cells were harvested, and pellets were lysed for protein preparation or stored at −80°C.

The GST- and His-tagged BAP1 and ASXL2 proteins were expressed in E. coli BL21 (DE3; Stratagene) and SG13009 (S9) (Qiagen), respectively. The bacteria bearing the desired plasmid were propagated with aeration at 37°C in 800 mL of 2YT to an A₆₀₀ absorbance of approximately 0.6. IPTG was added to 1 mmol/L, and growth was continued at either 37°C for 3 hours or 20°C overnight. The cells were then harvested by centrifugation.

The Bv-expressed proteins were purified using either ANTI-FLAG M2 Affinity Gel (Sigma) or Ni-NTA resin (Qiagen) as recommended by the manufacturers. The GST-fusion proteins were purified as described previously (21). The bacterial His-tagged proteins were purified under denaturing conditions (Qiagen) and refolded by dialysis as described previously (21).

GST association assays were performed as described previously (22).

Cleavage of Ub-AMC by wild-type or BAP1 mutants or by coexpressed proteins, for example, BAP1-ASXL2-AB (wild-type and mutant forms) was assayed in reactions containing 5 nmol/L Ub-AMC and the indicated proteins or complexes. Ub-AMC activity was assayed at 25°C in 20 mmol/L HEPES buffer (pH 7.5), 100 mmol/L NaCl, 1 mmol/L EDTA, 5 mmol/L DTT, and 0.05% (w/v) Tween-20. Assays were performed in 384-well plates (PerkinElmer) in a 30 μL reaction volume. Fluorescence was measured at 5-minute intervals at excitation and emission wavelengths of 355 nm and 460 nm, respectively. The release of AMC was monitored by fluorescence spectroscopy at 436 nm. All assays were performed simultaneously in duplicates. The hydrolysis rate was linear for 40–60 minutes and corrected for background signal (no enzyme).

For detecting IVT BAP1 and ASXL2 protein interactions, coimmunoprecipitation (co-IP) was performed by mixing IVT BAP1 and ASXL2 proteins in 1× PBS in 100 μL reaction volume at 37°C for 1 hour. The co-IP was performed in RIPA buffer with 10 μg of αBAP1 (rAb) at 4°C for 1 hour, followed by adding Protein A/G plus-Agarose (Santa Cruz Biotechnology) and continued incubation for an additional 1 hour. The resin was washed with RIPA buffer, then extracted with 5× Laemmli sample buffer. The protein complex was resolved in 4%–12% NuPAGE (Invitrogen) in MOPS running buffer. The gel was then dried and exposed on X-ray film.

For detecting the interaction between BAP1-FL WT/mutant proteins and ASXL2 (261–649 aa) protein, IP-Western blotting was performed. These proteins were expressed from Bv and purified with either Ni-NTA resin (Qiagen) or ANTI-FLAG M2 Affinity Gel [anti-Flag (mAb); Sigma]. The purified His-BAP-FL WT/mutant proteins were added to the resin, in which the Flag-ASXL2 protein was already bound in binding buffer (50 mmol/L Tris-HCl, pH 7.4, 150 mmol/L NaCl, 2 mmol/L EDTA, 0.25% NP40, 2% BSA). The association reaction was incubated at room temperature for 1 hour. The resin was washed with washing buffer [50 mmol/L Tris-HCl, pH 7.4, 150 mmol/L NaCl (500 mmol/L NaCl for the last wash, desalted with 20 mmol/L Tris-HCl), 5 mmol/L EDTA, 0.5% NP40], and extracted with 5× Laemmli sample buffer. The protein complex was resolved in 4%–12% NuPAGE (Invitrogen) in MES running buffer. Western blot analysis was performed using BAP1(rAb) (Rauscher laboratory) and Flag(rAb) (Cell Signaling Technology) antibodies.

To detect the association of BAP1 WT/mutant proteins and ASXL2 (261-649 aa) protein in vivo, HEK 293 cells were cotransiently transfected with both plasmids expressing BAP1 WT/mutant and ASXL2 proteins (Lipofectamine 2000; Invitrogen), as described (23). At 48-hour posttransfection, cells were collected and whole-cell lysates were used for co-IP with ANTI-FLAG M2 Affinity Gel [anti-Flag (mAb); Sigma]. Western blot analyses were performed with BAP(rAb) and Flag(rAb) antibodies.

Gel filtration was carried out as described previously (23). The purified individual BAP1-UCH, BAP1-ULD, and ASXL2-AB proteins or preformed BAP1-UCH-ULD-AB protein complex were analyzed by gel filtration with a Superdex 200 HR 10/30 column (GE Healthcare) equilibrated in BB100 buffer (20 mmol/L Tris-HCl, pH 8.0, 100 mmol/L NaCl, 0.2 mmol/L EDTA, 10% glycerol). The column was set at 4°C; the flow rate was 0.5 mL/minute, and 1-mL fractions were collected. Sixty microliters of the proteins or protein complex from each fraction were mixed with 5× Laemmli sample buffer, heated, and resolved on Nu-PAGE (Invitrogen). The proteins were visualized by Coomassie Blue stain.

The sequence of the UCH domain of human BAP1 (Q92560) was blasted against the Protein Data Bank (PDB), which revealed structure 3a7s (of UCH-L5/UCH37) as the most homologous. Amino acids were changed on 3a7s to those found in BAP1 and the larger loop of BAP1 built using homologous sequence/structures of PDB 2nxp and 3zzw. The structure was energy minimized multiple times using AMBER03 force field (24) in water using YASARA structure (25). This structure was aligned to each of the UCH domains of known protein structures for UCH-L1 (PDB 3kw5), UCH-L3 (PDB 2we6), and UCH-L5 (PDB 3ris) using MUSTANG algorithm (26). Sequences for ASXL were aligned from human ASXL1 (Uniprot Q8IXJ9), mouse ASXL1 (P59598), human ASXL2 (Q76L83), mouse ASXL2 (Q8BZ32), and Drosophila melanogaster ASX (Q9V727) using ClustalW2 (27), revealing a highly conserved domain (composed of box A and B). Structures for this domain were created using QUARK (28) ab initio modeling and energy minimized. Docking locations for ASXL were identified using overlapping 10-aa fragments of the ASXL model to the BAP1-UCH model using AutoDock (29), with 25 confirmations for each run. Using the top binding energies of all docking experiments, the full ASXL model was aligned to BAP1-UCH domain and interactions energy minimized. Models for BAP1-ULD were created using QUARK, selecting for optimal hydrophobic packing, and aligned to the structure of BAP1-UCH using PDB 3ihr (UCH-L5 with its ULD domain) with the MUSTANG algorithm.

The domain structure of BAP1 is unique (Fig. 1A). The UCH domain is similar to other mammalian UCHs (UCH-L1, UCH-L3, and UCH-L5). Mutations in the conserved catalytic site residues (Q85, C91, H169, and D184) completely abolish BAP1 ubiquitin hydrolase activity. Mutations C91W and H169Q in the catalytic site of UCH were identified in uveal melanomas (8). Other mutations/deletions/truncations in BAP1-UCH were found in uveal melanomas, mesothelioma, and breast cancer, and were confirmed as loss-of-function changes leading to loss of enzymatic activity (30). BAP1 interacts with BARD1 and BRCA1 and inhibits the E3 ligase activity of BRCA1/BARD1. Therefore, BAP1 has been proposed to regulate BRCA/BARD1–mediated ubiquitination during DNA damage response and cell cycling (2). The consensus sequence NHNY of BAP1 is critical for its interaction with HCF1, a transcriptional cofactor found in regulatory complexes. BAP1 mediates HCF1 deubiquitination and thereby helps control cell proliferation by regulating HCF1 protein levels and by associating with genes involved in cell-cycle regulation (3). The C-terminus ULD domain associates with the RING finger of BRCA1 involved in forming the BRCA1/BARD1 heterodimer, and this complex coordinately regulates ubiquitination during DNA damage response and cell cycling (1, 2). The ULD domain also interacts with ASXL members to form the PR-DUB involved in stem cell pluripotency and developmental processes (5). Many mutations/deletions/truncations in the ULD domain were identified in human cancers (Fig. 1A; refs. 8, 10). However, the molecular mechanism for these alterations remains elusive.

ASXL1/2/3 are human homologs of Drosophila Asx (Fig. 1B). These Asx family members share many similar domain structures. The N-terminus of ASXL contains a highly conserved ASXH domain, composed of the AB box required for Calypso/BAP1 protein binding. The C-terminus has a highly conserved plant homeo domain (PHD). The associated proteins for the PHD of the Asx family remain unknown. Several mutations in ASXL proteins have been found in myeloid leukemia and other disorders. However, the mechanism by which such mutations cause cancer and other diseases remains unclear.

We investigated the molecular mechanism for the mutations/deletions occurring in either the N-terminus of UCH or in the C-terminus of ULD domain for inactivating BAP1 in cancer, as well as for mutations occurring in ASXL proteins (Fig. 1A and B). ASXL functions as a molecular scaffold to recruit BAP1 to transcription factors, such as YY1, which specifically bind to its target genes. The N-terminus of ASXL, which includes ASXH, binds the ULD domain of BAP1, allowing the UCH catalytic site to be brought in proximity to the target promoter. ASXL then associates with transcription factors that bind to its target promoter. BAP1 ubiquitin hydrolase specifically removes ubiquitin from histones of chromatins, thereby repressing target genes. Our hypothesis is that BAP1-UCH, BAP1-ULD, and ASXL-AB tripartite binding is required for enzyme activity and for deubiquitination of histone H2A. Thus, in this model, BAP1 and associated epigenetic modifiers (complexes) play a specific role in modifying chromatin and regulating gene transcription. To test this hypothesis and to explain how mutations/deletions/truncations occurring in either ULD or UCH domains inactivate BAP1 in cancer, we used computer predicted molecular modeling of these proteins as a guide and biochemically characterized the protein–protein and/or domain–domain interactions of BAP1 and ASXL2 in vivo and in vitro using highly purified recombinant proteins, in coordination with BAP1 ubiquitin enzymatic activity.

Association of BAP1 with ASXL2 was tested in vitro by binding of full-length BAP1 to ASXL2. Immunoprecipitation was performed for both ³⁵S-labeled BAP1 and ASXL2 using αBAP1 (rAb). BAP1 was bound to ASXL2 (261–1,435 aa) and ASXL2 (261–649 aa; Fig. 2A). The N-terminal region of ASXL2 (261–649 aa), containing the predicted AB box, is necessary and sufficient for interaction with BAP1.

We next wanted to recapitulate the association between BAP1 and ASXL2 using recombinant proteins. BAP1 and ASXL2 proteins were produced in Bv by single- or coexpression. Single-expressed BAP1 WT, C91S mutant, and ASXL2 proteins were soluble and easily affinity-purified either with Ni-NTA or ANTI-FLAG M2 Affinity Gels (Fig. 2B). ASXL2 could be readily pulled down by either BAP1 WT or C91S proteins, based on Ni-NTA resin purification of coexpressed recombinant His-tagged proteins.

As BAP1 associates with ASXL2 in vitro and in vivo, BAP1 ubiquitin hydrolase activity was tested using purified recombinant proteins produced from Bv. Ub-AMC was used as substrate for the BAP1 enzymatic assay. BAP1 WT protein alone showed UCH basal deubiquitinase activity, whereas ASXL alone did not (Fig. 2C). When both proteins were mixed, ASXL2 greatly stimulated WT BAP1 activity, whereas a BAP1 mutant affecting the active cysteine (C91S) was enzymatically inactive (Fig. 2C), indicating that ASXL2 not only binds to BAP1 but also stimulates its ubiquitin hydrolase activity.

To define the minimal domain of ASXL2 required to interact with BAP1, we made truncations of ASXL2 based on computer predictions of homologous domains having stable structure fold predictions (Fig. 3A). Truncated proteins were expressed and purified from Bv and E. coli (Bac), followed by analysis for BAP1 binding (Fig. 3B, left). ASXL2 (261–649 aa) and AB box proteins were pulled down by GST-BAP1, while A box or B box of ASXL2 was not adequate to bind BAP1 (Fig. 3B, right). As shown by the negative control, ASXL2 (261–649 aa) did not bind GST. Next, we asked whether the AB box of ASXL2 stimulates BAP1 ubiquitin hydrolase activity as well as ASXL2 (261-649 aa). Ub-AMC assays were performed for BAP1 WT protein and truncated ASXL2 proteins. BAP1 showed basal ubiquitin hydrolase activity (Fig. 3C). The AB box of ASXL2 significantly stimulated BAP1 activity, as did ASXL2 (261–649 aa), whereas box A or B alone had intermediately stimulatory effects on BAP1 activity (Fig. 3C). These data suggest that the predicted AB box of ASXL2 is the minimal domain for binding to BAP1 and stimulating its ubiquitin hydrolase activity, and are also consistent with BAP1′s association with ASXL1 (2–365 aa; ref. 5).

To define potential protein–protein interaction surfaces between BAP1 and ASXL2-AB, we investigated amino acid alignments of ASXH AB boxes of ASXL members from human, mouse, and Drosophila. Conserved amino acids were visualized on a predicted molecular model of the ASXH domain (Fig. 4A and B, cyan), which revealed several potential interaction sites. The AB box is predicted to form five helices. The amino acid residues are very conserved in the AB box among ASXL members for hydrophobic packing of the domain (Fig. 4B, cyan). As confirmed through AutoDock experiments, several conserved surface amino acids were suggested to be critical to binding of the ASXH domain to BAP1 (Fig. 4B, middle structures, red). We suspected that these conserved amino acids of the AB box are important for direct interaction with BAP1. We made mutations for these amino acids by either disrupting the helix structure (R284P in Helix 1) or by changing conserved amino acids to alanines on surfaces of helices (VDR-AAA, NEF-AAA, ERL-AAA, EKE-AAA in Helices 2, 3, and 4, respectively; Fig. 4A and B). All WT and mutant AB box proteins were produced in Bac and purified (Fig. 4C, left). GST–BAP1 association assays with WT and mutant AB box proteins were performed. As expected, WT AB Box was pulled down by GST-BAP1 protein (Fig. 4C, right). Mutant AB box proteins (R284P: Helix 1; EKE-AAA: Helix 4) had intermediate effects on BAP1 binding, whereas VDR-AAA, NEF-AAA, and ERL-AAA (located in Helices 2 and 3) caused marked defects in BAP1 binding (Fig. 4C, right), suggesting these conserved amino acids in helices of the AB box make direct contact with BAP1 and that mutations of these amino acids disrupt its interaction with BAP1. This suggests that Helices 2 and 3 of the AB box of ASXL2 are the major contacted surfaces toward BAP1.

We next investigated whether AB box mutants with impact for BAP1 binding also coordinately lost stimulation of BAP1 ubiquitin hydrolase activity. Ub-AMC assays for BAP1 with AB box WT and mutant proteins were performed. As expected, BAP1 showed basal deubiquitin enzymatic activity and WT AB box stimulated BAP1 enzymatic activity (Fig. 4D). However, mutations (VDR-AAA) and (NEF-AAA) of the AB box that failed to bind BAP1 coordinately lost the ability to stimulate BAP1 activity (Fig. 4C and D). Furthermore, mutations (R284P), (ERL-AAA), and (EKE-AAA) that decreased binding to BAP1 showed less stimulation of BAP1 enzymatic activity as compared with WT AB box (Fig. 4D). These data indicate that ASXL2 stimulation of BAP1 ubiquitin hydrolase activity coordinates well with interactions between ASXL2 and BAP1 and that stimulation of BAP1 activity requires physical association of ASXL2 with BAP1.

To investigate how cancer-related mutations/deletions of BAP1 affect ASXL2 binding and BAP1 ubiquitin hydrolase activity, WT and mutant BAP1 proteins were produced in Bv and purified for ASXL2 binding in vitro. Bv Flag-ASXL2 (261–649 aa) was first bound to αFlag (mAb) resin, and then purified WT and mutant BAP1 proteins were incubated with resin-bound ASXL2. Immunoblotting was performed for associated WT and mutant BAP1. BAP1 mutant S172R and BAP1 ΔE631-A634, ΔK637-C638, ΔR666-H669 significant abolished ASXL2 protein binding in vitro (Fig. 5A and B; Supplementary Fig. S1), indicating that BAP1 harboring in-frame deletions in the ULD domain abolished ASXL binding. In parallel, Ub-AMC assays were performed with ASXL2 and BAP1 WT and mutant proteins. Three different protein concentrations (1×, 2×, 4×) of BAP1 were mixed with ASXL2. ASXL2 greatly stimulated WT BAP1 activity in a dose-dependent manner (Fig. 5C). Although the catalytically dead BAP1 mutant C91W could bind to ASXL2, it showed no enzymatic activity. When present at low concentrations (1×, 2×), mutants with deletions in BAP1-ULD showed no or extremely low activity in the presence of ASXL2. At a high concentration (4×) of BAP1 deletion mutants, bound ASXL2 inhibited BAP1 ΔE283-S285 activity. Although ASXL2 showed significantly decreased binding to BAP1 ΔE631-A634, it slightly inhibited BAP1 activity. BAP1 mutant ΔK637-C638 showed no enzymatic activity in the presence of ASXL2, consistent with data demonstrating that the protein–protein interaction is lost between ASXL2 and BAP1 ΔK637-C638 (Fig. 5A–C). Another BAP1 deletion, ΔR666-H669, showed no binding to ASXL2 and, as expected, BAP1 activity was not affected by the presence of ASXL2, although the BAP1 deletion mutant R666-H669 itself showed some basal activity (Fig. 5).

Structure predictions for BAP1-UCH suggest similar folding (Fig. 6A) as compared with other UCH proteins whose structures have been solved; however, BAP1 contains a larger loop than the others. This suggests that larger substrates should be accessible to the catalytic cleavage site of BAP1, but the large loop is also likely to restrict the access site for ubiquitin. The ULD domain has been shown in UCH-L5 to interact with UCH (31), in which BAP1 also contains a highly homologous ULD. The ULD for BAP1 was thus modeled (Fig. 6B) and amino acids in this domain associated with cancer were visualized (Fig. 6B, yellow). Aligning the ULD and UCH domains of BAP1 to the structure of UCH-L5 (PDB 3ihr), allows for an approximation of the ULD interaction with UCH (Fig. 6C and D). Docking of the ASXH domain to this structure suggests a tripartite consisting of a BAP1-UCH, BAP1-ULD, and ASXH domain complex. We predicted that the BAP1-ULD domain folds back on the UCH domain, and then the ULD domain recruits the ASXH domain, which subsequently stabilizes the UCH domain loop, thereby increasing catalytic activity of BAP1-UCH by opening the site of binding for histones and ubiquitin (Fig. 6C and D). To prove this and reconstitute the tripartite protein complex in vitro, a GST–ULD association assay with ASXL2-AB protein was performed using highly purified recombinant proteins. ASXL2-AB protein was pulled down by GST-ULD protein, and this interaction between ULD and AB was found to be direct and specific (Fig. 7A). Furthermore, GST-UCH also binds ULD, through which the ASXL2-AB box is recruited to GST-UCH. However, GST-UCH itself did not directly associate with ASXL2-AB (Fig. 7B). To test whether UCH, ULD, and AB form a ternary stable complex, the purified proteins were mixed to preform a complex in solution, then subjected to gel filtration. Meanwhile, each individual purified protein was also subjected to gel filtration. As shown in Supplementary Fig. S2 I, the protein complex containing UCH, ULD, and AB was coeluted. The peak fraction of the complex was eluted with apparent molecular mass of 44 KDa. Individual UCH, ULD, and AB proteins eluted more likely as monomers with expected molecular masses (Supplementary Fig. S2 II–IV). The peak fraction of the AB box in the complex shifted dramatically from #17 to #15 compared with AB alone, supporting our hypothesis that BAP1-UCH, BAP1-ULD, and ASXL2-AB form a ternary stable complex in vitro and that ULD recruits ASXL2-AB to access the BAP1-UCH catalytic motif. This complex is sustained during gel filtration.

According to molecular modeling shown in Fig. 6C, BAP1-ULD is predicted to be composed of helices and associated with the ASXL2-AB box. The long helix of ULD is packed close to the UCH. We hypothesized that if this helix were disrupted, it would change the ULD such that its association with AB box and/or UCH domain would be affected. To test this hypothesis, two mutations (L650P, F660P) located in the middle of the long helix were individually introduced into the ULD motif. Then, GST–UCH association assays were performed using highly purified WT and mutant ULD proteins and ASXL2-AB. Surprisingly, L650P and F660P mutants did not affect UCH binding; however, they did abolish AB box binding (Supplementary Fig. S3), suggesting that the interface of the AB box is on the long helix of ULD and that disruption of this helix disassociates binding to the AB box.

We next investigated mutation/deletions affecting BAP1 ULD identical to those observed in cancer patients. These mutation/deletions are predicted to encode BAP1 ULD alterations located on the interfaces either to ASXL-AB or to UCH, thus affecting BAP1 ubiquitin hydrolase activity (Fig. 6C). Therefore, GST–UCH association assays were carried out with purified BAP1 ULD WT and mutant proteins and ASXL2-AB. Mutation/deletions in BAP1 ULD impacted UCH and/or AB box binding (Fig. 7C). BAP1 ULD mutants D672G and ΔK637-C638 slightly decreased UCH binding but significantly decreased AB box association. However, BAP1 ULD mutants ΔE631-A634 and ΔR666-H669 significantly decreased UCH binding and almost completely abolished AB box binding (Fig. 7C), consistent with in vitro binding of full-length BAP1 with ASXL2 (261–649 aa) shown in Fig. 5A and B.

Herein, we have characterized protein–protein interactions of BAP1 and ASXL2 using computer modeling, biochemical approaches, and enzymatic activity analyses. We also investigated the molecular mechanism underlying loss-of-function mutations/deletions affecting BAP1 and the BAP1–ASXL2 complex that contribute to tumor susceptibility and progression in patients with mesothelioma, uveal melanoma, and other cancers. We have drawn the following conclusions from our data. First, the interaction between BAP1 and ASXL2 is direct, and the association of ASXL2 greatly stimulates BAP1 hydrolase activity. Second, the highly conserved AB box of ASXL2 is the minimal domain required for the interaction with BAP1 and its stimulation of BAP1 ubiquitin hydrolase activity. Mutations/deletions of highly conserved amino acids in predicted helices within the AB box abolish its association with BAP1 and its ability to enhance BAP1 enzymatic activity. These conserved amino acids are predicted to be on the surface of helices in the AB box of ASXL2. Third, cancer-derived mutations/deletions in the BAP1-ULD abolish ASXL2 binding and thereby result in loss of ASXL2′s ability to stimulate BAP1 activity. These alterations in BAP1-ULD are reside at protein–protein interfaces among the BAP1-ULD, BAP1-UCH, and ASXL-AB. Finally, BAP1-UCH, BAP1-ULD, and ASXL-AB could be reconstituted to from a ternary stable complex in vitro using recombinant proteins. The mutations structurally disrupted the predicted helix of the ULD domain of BAP1 and abolished ASXL2 binding. Alterations in the ULD domain of BAP1 occurring in cancers failed to recruit the ASXL2-AB box to the UCH catalytic site, such that BAP1 deubiquitin hydrolase activity was dramatically reduced.

A previous study defined interaction domains between Drosophila Calypso and Asx (2–337 aa) and between human BAP1 and ASXL1 (2–365 aa; ref. 5). Here, we took advantage of molecular modeling for BAP1 and ASXL members. We predicted structures for these proteins based on known NMR and/or crystallization of proteins from evolutionarily highly conserved homologous sequences/structures and computer simulation. We then further defined the structure for these conserved domains and docked the locations for protein–protein interaction of the complex. We defined the highly conserved AB box for the ASXL family and predicted its three-dimensional structure, which is composed of five helices bundling to form a stable conformation (Fig. 4A and B). Our data implicate the AB box as the minimal domain required to interact with BAP1. The AB box also stimulates BAP1 deubiquitin activity (Fig. 3A–C). Moreover, conserved amino acids on the surface of helices within the AB box directly interact with BAP1 (Fig. 4A and B), and mutations of these amino acids impacted BAP1 binding and ubiquitin hydrolase activity (Fig. 4C and D).

The structure predictions for BAP1-UCH suggest folding similar to that of other UCH proteins (Fig. 6A). However, BAP1 contains a larger loop than the others, suggesting that larger substrates would be accessible to the catalytic cleavage site, such as ubiquitin. The ULD domain of UCH-L5 interacts with the UCH domain (31), and BAP1 contains a highly homologous ULD domain. Thus, the ULD for BAP1 was modeled (Fig. 6B). Alignment of UCH and ULD of BAP1 to the structure of UCH-L5 allows for an approximation of the ULD interaction with the UCH of BAP1 (Fig. 6C). Docking of the AB box to this structure revealed a tripartite BAP1-UCH, BAP1-ULD, and ASXL2-AB domain complex. We predicted that BAP1-ULD folds back to the UCH and that the ULD recruits the AB box, which subsequently stabilizes the UCH loop, thereby increasing the catalytic activity of BAP1-UCH (Fig. 6C and D). Indeed, this tripartite complex was reconstructed in vitro using purified recombinant proteins (Fig. 7A and B; Supplementary Fig. S2). This experimental evidence supported the predicted molecular modeling for the protein complex. Surprisingly, BAP1 mutations L650P and F660P, within the long helix of ULD (Fig. 6A–C), did not interfere with UCH binding, but did abolish association with the AB box. Both mutations (L650P, F660P) were predicted to kink the long helix of ULD and disrupt UCH binding. The interpretation is that either the AB box directly interacts with the long helix of ULD, or that the ULD conformation changes such that the AB box can no longer bind to ULD (Fig. 6C and D; Supplementary Fig. S3). Furthermore, we discovered a novel loss-of-function mechanism by which BAP1 mutations/deletions occur in cancer. These mutations/deletions affect the interface of BAP1 with either ASXL-AB or UCH. Mutations of BAP1 ULD impact either AB box binding and/or UCH interactions. The tripartite complex stability is thereby adversely affected and UCH activity is markedly decreased, potentially altering BAP1′s tumor suppressive activity. Deubiquitination of target proteins/chromatins by BAP1 is inhibited, resulting in deregulation of target genes (Fig. 7D). Our data provide evidence supporting the computer-predicted molecular modeling of the BAP1–ASXL interaction. The critical target proteins/chromatins for BAP1-ASXL2 repression remain to be determined. However, previous work suggests that ASXL proteins contain domains that likely serve in the recruitment of chromatin modulators and transcriptional effectors to DNA (32). ASXL1 plays a significant role in epigenetic regulation of gene expression by facilitating PRC2-mediated transcriptional repression of known leukemic oncogenes. Therefore, ASXL1 may serve as a scaffold for recruitment of the PRC2 complex to specific loci in hematopoietic cells, although ASXL loss leads to BAP1-independent alterations in chromatin state/gene expression in hematopoietic cells.

On the basis of our studies, we propose a working model for the mechanism of germline and/or somatic loss of BAP1 that increases susceptibility to various cancers. One scenario is that ASXL functions as a molecular scaffold to recruit BAP1 to transcription factors, which specifically bind to its target genes. Then, BAP1 ubiquitin hydrolase specifically removes the ubiquitin from histones of chromatins to regulate these target genes (Fig. 7D). ASXL not only functions as a molecular scaffold for BAP1 but also greatly stimulates its activity. When mutations/deletions occur in BAP1, either they cause enzymatic loss-of-function of BAP1 or abolish BAP1′s association with ASXL. Loss of binding to ASXL would dramatically decrease BAP1 deubiquitination activity, because of inability to bring ASXL to BAP1′s catalytic site (e.g., deletions in the ULD domain: E631-A634 del and K637-C638 del). On the other hand, products of ASXL gene mutations that lose association with BAP1 also lead to BAP1 loss of function. These mutations occur in highly conserved amino acids of the ASXL-AB box, the minimal domain required for BAP1 interaction. Thus, BAP1 ubiquitin hydrolase activity would be greatly reduced. Our study also suggests that ASXL loss would lead to BAP1-dependent alterations in chromatin state/gene expression in human cancers/disorders.

In summary, through an integrated use of advanced computational molecular modeling, molecular biology, and biochemistry strategies, we have provided evidence for a direct protein–protein interaction involving BAP1 and its physically obligate partner ASXL2 for ubiquitin hydrolase activity. We have also elucidated a molecular mechanism for certain naturally occurring loss-of-function BAP1 mutations/deletions in cancer. Drosophila Calypso-Asx and human BAP1-ASXL2 complexes deubiquitinate H2A ub1, and mutations in chromatin-modifying enzymes such as BAP1 are implicated in the epigenetic deregulation that occurs commonly in human cancers (5). It will be important to use epigenomic platforms to elucidate how these disease alleles contribute to tumorigenesis in different contexts. Moreover, many known oncogenes/tumor suppressors contribute, at least in part, to transformation through direct or indirect alterations in the epigenetic state (33). Further epigenomic studies of human malignancies will likely uncover novel routes to malignant transformation, and therapeutic strategies that reverse epigenetic changes might be of benefit in patients with mutations in epigenetic regulators.

J.W. Harbour has ownership interest (including patents) in Castle Biosciences and is a consultant/advisory board member for Castle Biosciences and Immunocore. A.M. Bowcock has ownership interest (including patents) in a U.S. Patent 9,133,523 entitled "Compositions and Methods for Detecting Cancer Metastasis" (James William Harbour and Anne Bowcock; issued on September 15, 2015; licensed to Castle Biosciences 10/2011). No potential conflicts of interest were disclosed by the other authors.

The opinions, findings, conclusions, or recommendations expressed in this work are those of the authors and not necessarily of the directors, officers, or staff of the funding agencies listed here.

Conception and design: H. Peng, J.R. Testa, F.J. Rauscher III

Development of methodology: H. Peng, J. Prokop, F.J. Rauscher III

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): H. Peng, J. Prokop, J. Karar, K. Park, J.W. Harbour, F.J. Rauscher III

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): H. Peng, J. Prokop, F.J. Rauscher III

Writing, review, and/or revision of the manuscript: H. Peng, J. Prokop, J.W. Harbour, A.M. Bowcock, S.B. Malkowicz, M. Cheung, J.R. Testa, F.J. Rauscher III

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): H. Peng, L. Cao, M. Cheung, J.R. Testa

Study supervision: H. Peng, F.J. Rauscher III

Other (development of some of the BAP1 mutation constructs): A.M. Bowcock

This work was supported by NCI grants R01CA175691 (J.R. Testa, F.J. Rauscher), P30CA010815 (F.J. Rauscher), P30CA006927 (Fox Chase Cancer Center), R01CA161870 (J.W. Harbour, A.M. Bowcock), R01CA125970 (to J.W. Harbour), and R01CA163761 (F.J. Rauscher). Support for Shared Resources was provided by P30CA010815 to The Wistar Institute. Also supported by the Jayne Koskinas Ted Giovanis Foundation for Health and Policy (F.J. Rauscher), the Palmira and James Nicolo Family Research Fund (S.B. Malkowicz), Local #14 Mesothelioma Fund of the International Association of Heat and Frost Insulators & Allied Workers (J.R. Testa), Ovarian Cancer Research Fund Alliance (F.J. Rauscher), Samuel Waxman Cancer Research Foundation (F.J. Rauscher), a Susan G. Komen grant KG110708 (F.J. Rauscher), Office of the Assistant Secretary of Defense for Health Affairs, through the Breast Cancer Research Program, under Award numbers W81XWH-17-1-0506, W81XWH-14-1-0235 and W81XWH-11-1-0494 (F.J. Rauscher).

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.