A20 Ubiquitin Ligase–Mediated Polyubiquitination of RIP1 Inhibits Caspase-8 Cleavage and TRAIL-Induced Apoptosis in Glioblastoma

The TNF-related apoptosis-inducing ligand (TRAIL) (1, 2) executes the innate and adaptive immune responses in the process of tumor immunosurveillance (3). The anticancer activity of TRAIL is attributable to its ability to induce apoptosis through the binding of death receptors 4 and 5 (DR4, DR5) and the recruitment of intracellular apoptosis-initiating caspase-8 through Fas-associated death domain (FADD) for the assembly of a death-inducing signaling complex (DISC) (4). Recombinant TRAIL as well as agonistic DR4 and DR5 antibodies targeting this apoptotic pathway have been generated as potential therapies for the treatment of cancer (5). In clinical trials, however, investigators have proven that cancers are resistant to TRAIL pathway-targeted therapies (6–8), thus suggesting that cancers escape from TRAIL-mediated immunosurveillance and are resistant to TRAIL-targeted therapies.

The dimerization and cleavage of caspase-8 in the DISC are the critical upstream events in TNF family ligand-induced apoptosis (9–12), and ubiquitination of proteins in the DISC regulates these biochemical processes (13). Ubiquitin is covalently attached to lysine residues of the substrate proteins through the catalytic reactions mediated by the ubiquitin-activating enzyme (E1), conjugating enzyme (E2), and ligase (E3) and is removed by deubiquitinating enzymes (14). Ubiquitin has 7 lysine (K) residues and an N-terminal methionine (M1), each of which can be linked to the C-terminal glycine residue of another ubiquitin to form polyubiquitin chains (15); this regulates TNF-α–induced signaling (13). Upon TNF-α binding, TNF receptor 1 (TNFR1) recruits receptor-interacting protein 1 (RIP1), cellular inhibitor of apoptosis protein 1 and 2 (cIAP1 and cIAP2), and TNFR-associated factor 2 (TRAF2) for the assembly of TNFR1-associated complex I (16). TRAF2, cIAP1, and cIAP2 are E3 ligases that activate NF-κB through the attachment of polyubiquitin chains to RIP1 (17) and binding of the polyubiquitin chain to IκB kinase γ (IKKγ) (18). TRAF2 and RIP1 then detach from TNFR1 and recruit FADD and caspase-8 for the assembly of the cytoplasmic complex II (19), where the deubiquitinating cylindromatosis removes the polyubiquitin chains from RIP1 to promote caspase-8 cleavage for TNF-α–induced apoptosis (20).

In contrast to TNFR1, DR4 and DR5 recruit FADD and caspase-8 in the assembly of a plasma membrane-bound DISC, where caspase-8 becomes dimerized and cleaved, initiating apoptosis (21, 22). Cullin 3 (CUL3), an E3 ligase, adds K48- and K63-linked polyubiquitin chains to caspase-8 and facilitates its dimerization and cleavage in the DISC, where A20 deubiquitinating enzyme removes the polyubiquitin chains from caspase-8 (23). A20 (TNF-α–induced protein 3; TNFAIP3) is well known for its anti-inflammatory activities (24) through its N-terminal ovarian tumor domain (OTU), which acts as a deubiquitinating enzyme and removes K63-linked polyubiquitin chains from RIP1, TRAF6, and RIP2, thus restricting TNFR1, Toll-like receptor, and nucleotide-binding oligomerization domain-induced NF-κB signaling (25–27). A20 also contains a C-terminal Zinc finger (Znf) domain of an E3 ligase (28), but the function of the Znf E3 ligase has yet to be established.

Here, we show that the Znf domain of the A20 E3 ligase mediates RIP1 K63-linked polyubiquitination; this polyubiquitin chain binds to the caspase-8 protease p18 domain, which blocks caspase-8 dimerization and cleavage and thus inhibits TRAIL-induced apoptosis in human glioblastoma. Glioblastoma is the most common brain cancer and has no curative treatment. Recent studies have identified tumor-initiating cells (29) and shown that the tumor-initiating cells retain the original tumor genomic features (30), possess self-renewal and tumorigenic capacity (31), and are responsible for the tumor resistance to treatments (32). In this study, we further establish that A20 E3 ligase-mediated RIP1 ubiquitination inhibits caspase-8-initiated and TRAIL-induced apoptosis in the tumor-initiating cells isolated from glioblastomas surgically removed from patients.

In exploring the role of the ubiquitin enzymes A20, cIAP1/2, CUL3, and TRAF2 in glioblastoma, we first analyzed the expression of these proteins in glioblastoma tissues. Immunoblotting revealed that A20 was highly expressed in the tumors as compared with normal brain tissue, whereas CUL3, cIAP2, and TRAF2 were expressed consistently in the normal brain and tumor tissues and cIAP1 was seen in some tumors (Fig. 1A). To examine whether these enzymes regulate TRAIL signaling in glioblastoma cells, we analyzed their expression in TRAIL-sensitive (LN18, LN71, T98G, U343MG) and resistant glioblastoma cell lines (LN443, U87MG, U118MG, U138MG). The TRAIL sensitivity of these cell lines was defined in our earlier study (33) and confirmed by colony formation assay (Supplementary Fig. S1A). A20 was highly expressed in the resistant cell lines but barely detected in the sensitive cell lines and normal human astrocytes (34); in contrast, CUL3, cIAP1, cIAP2, RIP1, and TRAF2 were consistently expressed in the resistant and sensitive cell lines (Supplementary Fig. S1B).

The DISC was then isolated from each cell line through immunoprecipitation after the cells were treated with mixed Flag-TRAIL and Flag antibody. For the unstimulated control, the cells were lysed first and treated with Flag-TRAIL and Flag antibody. Taking this approach, we have shown that DR5 but not DR4 is the functional receptor that is expressed and interacts with Flag-TRAIL in glioblastoma cells (33). Immunoblotting detected DR5, A20, TRAF2, and RIP1 in the DISC in the resistant lines but only DR5 and TRAF2 in the sensitive lines (Fig. 1B; Supplementary Fig. S2A). FADD and caspase-8 were recruited to the DISC in both resistant and sensitive cell lines; however, caspase-8 was cleaved only in the DISC isolated from the sensitive cells. In contrast, cIAP1 was not seen in the DISC from any of these cell lines, whereas cIAP2 was detected only in the DISC of sensitive cell lines. CUL3 was detected in the DISC of H460, a TRAIL-sensitive lung cancer line as reported (23), but not in the DISC isolated from glioblastoma cell lines (Supplementary Fig. S2B). TRAIL stimulated caspase-8 ubiquitination in H460, consistent with a previous report (23), but not in glioblastoma cell lines (Supplementary Fig. S2C).

To our surprise, A20, RIP1, and TRAF2 were seen in the unstimulated controls in the resistant cell lines whereas only TRAF2 was seen in the sensitive cell lines (Fig. 1B), suggesting that these proteins might interact with DR5 and form a complex before treatment with TRAIL. To test this, we performed size exclusion analysis and immunoblotting (19) and identified DR5, A20, RIP1, and TRAF2 in the approximately 669-kDa fractions from resistant lines but only DR5 and TRAF2 in sensitive lines (Fig. 1C; Supplementary Fig. S3A). These proteins remained in the high-molecular-weight fractions in the cells after TRAIL treatment. In contrast, caspase-8 isoforms were eluted in the fractions corresponding to their monomeric molecular weights before treatment with TRAIL but quickly shifted to high-molecular-weight fractions under treatment with TRAIL. These results suggest the presence of a DR5-associated signaling complex that we named the preligand assembly complex under physiologic conditions. To confirm this, we isolated the preligand assembly complex from the pooled high- and low-molecular-weight fractions through immunoprecipitation using Flag-TRAIL and Flag antibody and detected DR5, A20 and RIP1 in the high- but not the low-molecular-weight fractions (Fig. 1D).

Because DR5 is a type I transmembrane protein, we conducted subcellular fractionation to determine whether DR5-associated preligand assembly complex is plasma membrane-bound. Immunoblotting detected DR5 mainly in the membrane fractions and RIP1, A20, and FADD in both the membrane and cytosolic fractions of TRAIL-resistant cells (Fig. 1E). The DR5-associated complex was then isolated through immunoprecipitation via the use of a DR5 antibody. Immunoblotting identified DR5, A20, RIP1, and TRAF2 in the membrane but not cytosolic fractions in the resistant but only DR5 and TRAF2 in the sensitive cells (Fig. 1F; Supplementary Fig. S3B). These data indicate that although RIP1, A20, and TRAF2 are present in the membrane and cytosolic fractions, the preligand assembly complex is formed as a membrane bound complex composed of DR5, A20, RIP1, and TRAF2 in resistant cells but only DR5 and TRAF2 in sensitive cells. TRAIL stimulates the recruitment of FADD and caspase-8 to the plasma membrane-bound preligand assembly complex for the assembly of a plasma membrane-bound DISC.

To determine whether A20, TRAF2, and RIP1 inhibit caspase-8 cleavage, we conducted knockdown experiments by transfecting TRAIL-resistant cell lines with siRNA specific to A20, RIP1, and TRAF2. Identical results were obtained with the use of 2 different siRNA sequences specific to each gene, indicating no off-target effects of the sequences; thus, the data were presented with one of the siRNA sequences targeting each gene. The siRNA-transfected LN443 cells were treated with TRAIL. Significant apoptosis was observed in the cells transfected with A20 and RIP1 but not TRAF2 siRNA, as shown by cell death (Fig. 2A), caspase-8 enzymatic activity (Fig. 2B), phase contrast microscopy (Fig. 2C), and Annexin V assay (Supplementary Fig. S4A). These assays showed that the transfection of A20 or RIP1 siRNA alone did not cause apoptosis. Furthermore, A20 siRNA transfection did not affect the sensitivity of LN443 cells to TNF-α, Fas ligand (FasL), or cisplatin (Supplementary Fig. S4B).

Immunoblotting confirmed A20 and TRAF2 knockdown in the transfectants and revealed the cleavage of caspase-8 and RIP1 in the A20 siRNA-transfected but not the TRAF2 siRNA-transfected LN443 cells (Fig. 2D). Knockdown of A20 or RIP1 but not TRAF2 restored caspase-8 cleavage in other resistant cell lines (Supplementary Fig. S5A). The DISC was isolated and immunoblotting identified caspase-8 cleavage in the DISC in the A20 and RIP1 siRNA transfected LN443 cells (Fig. 2E). The caspase-8 cleavage was abolished by z-IEDT, a caspase-8 inhibitor (Fig. 2F). To determine the long-term effects of A20 knockdown, we introduced a short hairpin RNA (shRNA) specific to A20 through lentiviral transduction into LN443 cells. The A20 shRNA transduction alone did not inhibit the colony formation of the cells (Supplementary Fig. S5B). Noncleaved RIP1 was detected in the preligand assembly complex, whereas neither noncleaved nor cleaved RIP1 were observed in the DISC in the A20 shRNA-transduced cells (Supplementary Fig. S5C), suggesting that RIP1 is detached from the DISC in the absence of A20. These results indicate that both A20 and RIP1 are required for the inhibition of caspase-8 cleavage in the DISC.

A20 has an N-terminal OTU domain that acts as a de-ubiquitinating enzyme and a C-terminal Znf domain that acts as an E3 ligase (28). To test whether the OTU or Znf domains inhibit caspase-8 cleavage, we generated inactive OTU (C103A) and Znf4 (C624A, C627A) A20 mutants (mt; Supplementary Fig. S5D). A20 wild-type and inactive OTU and Znf4 mt plasmids were introduced through lentiviral transduction into LN71, an A20-deficient TRAIL-sensitive cell line. Stable clones were established, the expression of A20 wild-type (wt) and mt proteins was verified by immunoblotting, and the stable clones were selected for expressing wt and mt A20 proteins at levels similar to the endogenous A20 in the resistant LN443 cells (Supplementary Fig. S6A). The clones were treated with TRAIL and cell death (Fig. 3A), caspase-8 enzymatic activity (Fig. 3B), and annexin V staining (Supplementary Fig. S6B) demonstrated that the expression of A20 wt and OTU but not Znf4 mt inhibited TRAIL-induced apoptosis. Expression of A20 wt, OTU, or Znf4 mt alone did not cause apoptotic cell death (Supplementary Fig. S6C). The preligand assembly complex and DISC were isolated from the clones, and A20 wt and mt proteins were observed in the complexes. RIP1 was enriched in the TRAIL-resistant A20 wt- and OTU mt-expressing clones more than in the TRAIL-sensitive empty vector and Znf mt-expressing clones (Fig. 3C). The clones were treated with TRAIL and RIP1 cleavage was detected in the empty vector and Znf mt but not the A20 wt- or OTU mt-expressing clones (Fig. 3D). Together, these gain-of-function studies suggest that Znf4 is responsible for caspase-8 inhibition in the DISC.

The finding that both A20 Znf4 and RIP1 are required for caspase-8 inhibition suggests that A20 E3 ligase may inhibit caspase-8 through RIP1 ubiquitination. An in vitro ubiquitination assay has shown that A20 E3 ligase adds a K48-linked polyubiquitin chain to RIP1 (28); however, an in vivo assay has revealed that RIP1 is ubiquitinated through a K63-linked polyubiquitin chain (17). To evaluate the role of the A20 Znf domain, we repeated the in vitro ubiquitination assay using the A20 OTU mt containing the Znf domain in the presence of the K48-linked polyubiquitin chain-specific E2 enzyme UBCH5A and the K63-linked polyubiquitin chain-specific E2 enzyme UBC13 (28). Flag-RIP1 and His-Myc-A20 OTU mt proteins were added to an in vitro ubiquitination assay consisting of ATP, biotin-ubiquitin, E1, and UBCH5A or UBC13 (Supplementary Fig. S7A).

After the reaction, proteins were separated and immunoblotted for avidin-bound biotin-polyubiquitin chains in the presence of UBCH5A or UBC13 (Fig. 4A). To confirm this in vivo, HEK293T cells were co-transfected with His-Myc-A20 OTU mt, Flag-RIP1, and HA-UB mts that contain only one lysine at K48 or K63 or have a single point mutation of R48 or R63 (Supplementary Fig. S7B; ref. 17). The transfected cells were lysed in 1% SDS denaturing buffer to dissociate proteins and then diluted 10 times in non–SDS-containing buffer. Flag-RIP1 was isolated through immunoprecipitation by the Flag antibody. Immunoblotting using the HA antibody identified K48- and K63-linked polyubiquitin chain-conjugated Flag-RIP1 (Fig. 4B). These assays suggest that the Znf domain can mediate RIP1 ubiquitination through either the K48- or K63-linked polyubiquitin chain.

To determine whether RIP1 is ubiquitinated in glioblastoma cells, we treated glioblastoma cell lines with MG132, a 26S proteasome inhibitor. The K48-linked polyubiquitin chain targets substrates to the 26S proteasome for degradation; however, MG132 treatment did not affect the levels of RIP1 protein in the absence or presence of TRAIL (Supplementary Fig. S7C), suggesting that it is less likely that RIP1 is ubiquitinated by a K48-linked polyubiquitin chain. To examine this further, U87MG cells were transfected with HA-UB mts, treated with TRAIL and lysed in a denaturing buffer. RIP1 was isolated through immunoprecipitation. Immunoblotting using an HA antibody detected more polyubiquitin chain-conjugated RIP1 in the K63 and R48 mt than in the K48 and R63 ubiquitin mt transfected cells (Fig. 4C), suggesting that RIP1 is ubiquitinated by K63-linked polyubiquitin chains in glioblastoma cells.

To determine the site of RIP1 ubiquitination, the preligand assembly complex and DISC were isolated from TRAIL-resistant cells. Immunoblotting identified DR5 and RIP1 in the preligand assembly complex and DISC. High-molecular-weight RIP1 species were enriched in the DISC compared with the preligand assembly complex (Fig. 4D), suggesting that TRAIL treatment enhances RIP1 polyubiquitination in the DISC. To verify this, we carried out a double immunoprecipitation and isolated RIP1 from the preligand assembly complex and DISC by using a RIP1 antibody. Again, ubiquitin-conjugated RIP1 was detected more strongly in the DISC than in the preligand assembly complex (Fig. 4E). Furthermore, immunoblotting using an antibody specific to the K63-linked polyubiquitin chain identified more polyubiquitin chain-conjugated RIP1 in the DISC than in the preligand assembly complex (Fig. 4F). To identify the A20 domains responsible for RIP1 K63-linked polyubiquitin, we isolated the preligand assembly complex and DISC from A20 wt-, OTU mt-, and Znf4 mt-expressing clones. K63-linked polyubiquitin chain-conjugated RIP1 was identified in the DISC of the A20 wt and OTU mt but not the Znf4 mt expressing clones (Fig. 4G). These data suggest that TRAIL stimulates A20 Znf domain E3 ligase-mediated RIP1 ubiquitination through K63-linked polyubiquitin chains in glioblastoma cells.

We then examined whether ubiquitinated RIP1 binds to caspase-8 in resistant cells. Caspase-8 was isolated from untreated or TRAIL-treated LN443 cells with a caspase-8 antibody and greater molecular weight RIP1 species coimmunoprecipitated with caspase-8 in TRAIL-treated cells (Fig. 5A). The cells were then subjected to subcellular fractionation; caspase-8 was isolated from the membrane and cytosol fractions and immunoblotting detected enrichment of poly-ubiquitin species in the membrane fraction (Supplementary Fig. S8A). These data suggest that TRAIL stimulates the interaction of ubiquitinated RIP1 and caspase-8.

Caspase-8 consists of 2 death effector domains and a protease domain composed of a p18 and p12 subunit. To determine whether ubiquitinated RIP1 binds to the caspase-8 protease domain and inhibits its cleavage, we performed a ubiquitin-binding assay using Fv-caspase-8 (35), in which the death effector domains were replaced with Fv, a derivative of the FK506 binding protein (Supplementary Fig. S8B). LN443 cells were subjected to subcellular fractionation after treatment with TRAIL. RIP1 was isolated from the cytosolic and membrane fractions and then incubated with recombinant Fv-caspase-8. Immunoblotting identified the binding of Fv-caspase-8 to the K63-linked polyubiquitin chain-conjugated RIP1 in the membrane fractions (Fig. 5B).

To further define caspase-8 as an ubiquitin binding protein, we carried out a series of caspase-8 pull-down assays. Fv-caspase-8 was bound to protein G-beads through the incubation of Fv-caspase-8, caspase-8 antibody, and protein G-beads. The Fv-caspase-8-bound beads were incubated with recombinant monoubiquitin, M1-linked linear, and K48- and K63-linked polyubiquitin chains. Unbound ubiquitin proteins were washed off the beads and bound ubiquitin proteins were eluted. Fv-caspase-8 pulled-down K63-linked polyubiquitin chains (Fig. 5C). The experiment was repeated with the Fv-caspase-8 p18 catalytic site mt (35, 36); the results show that the caspase-8 p18 catalytic site is not required for the interaction of caspase-8 and K63-linked polyubiquitin chain (Supplementary Fig. S8C). To confirm this, we used a recombinant caspase-8 protease p18 subunit in a pull-down assay. The protease p18 subunit was bound to protein G-beads with a caspase-8 antibody and caspase-8 p18-labeled beads were incubated with monoubiquitin and polyubiquitin chains. The pull-down assay showed that the protease p18 subunit-labeled beads mainly pulled down K63-linked polyubiquitin chains. To determine whether the K63-linked polyubiquitin chain directly binds to Fv-caspase-8, we generated a nickel column bound with His-tagged K63-linked polyubiquitin chains. Fv-caspase-8 was added to the His-K63-linked polyubiquitin chain-labeled nickel column and the nickel column pull-down assay identified the binding of Fv-caspase-8 to the His-K63-linked polyubiquitin chains (Fig. 5D).

To define the role of the K63-linked polyubiquitin chain in caspase-8 dimerization and cleavage, we used an in vitro Fv-caspase-8 dimerization and cleavage assay (35) in which Fv-caspase-8 is dimerized and cleaved by adding a synthetic Fv ligand, AP20187 (Supplementary Fig. S8D). Monoubiquitin and polyubiquitin chains were added to the Fv-caspase-8 cleavage assay. After the reaction, the proteins were separated and examined by immunoblotting with the use of caspase-8 and ubiquitin antibody. The results showed that the cleavage of Fv-caspase-8 in the presence of AP20187 was significantly inhibited by the K63-linked polyubiquitin chains and slightly inhibited by the linear polyubiquitin chain (Fig. 5E, Supplementary Fig. S8E). Taken together, these results suggest that binding of K63-linked polyubiquitin chain to the caspase-8 protease domain inhibits its dimerization and cleavage.

To validate whether A20-mediated ubiquitination of RIP1 occurs in vivo, we examined human glioblastoma tumor tissues surgically removed from patients (Supplementary Fig. S9A) and glioblastoma tumor-initiating cells enriched by CD133 sorting as previously reported (29, 32) (Supplementary Fig. S9B). The CD133⁺ cells were used soon after isolation to avoid prolonged culturing effects. Immunoblotting showed that A20 and RIP1 were expressed in the CD133⁺ cells at levels similar to those seen in the matched parental tissues (Supplementary Fig. S9C), as recently reported (37). A size-exclusion assay detected DR5, RIP1, A20, and TRAF2 in the high-molecular-weight fractions of the tissues (Fig. 6A) and matched CD133⁺ cells (Fig. 6B).

Subcellular fractionation further revealed DR5, RIP1, A20, and TRAF2 enrichment in the membrane fractions of the tissues (Fig. 6C) and matched CD133⁺ cells (Fig. 6D). To further confirm this in the tumor initiating cells, we isolated the preligand assembly complex from the pooled high- and low-molecular-weight fractions of the CD133⁺ cells through immunoprecipitation using Flag-TRAIL and Flag antibody and detected A20 and RIP1 in the high- but not the low-molecular-weight fractions (Fig. 6E). These results suggest that a membrane-bound DR5-associated preligand assembly complex is formed in glioblastoma tumor tissues and tumor-initiating cells.

To further evaluate RIP1 ubiquitination in CD133⁺ cells, we propagated the cells in neurosphere culture conditions, which maintains the original cancer genomic features (30). The preligand assembly complex and DISC were isolated from the CD133⁺ cells and DR5, RIP1, and A20 were detected in both the preligand assembly complex and DISC whereas FADD and caspase-8 were recruited to the DISC, where caspase-8 was not cleaved (Fig. 6F). K63-linked polyubiquitin chain-conjugated RIP1 was also identified in the DISC but not the preligand assembly complex (Fig. 6G). Taken together, these results validate the preligand assembly complex and DISC models, as established from studies of TRAIL-resistant cell lines, in glioblastoma tissues and derived tumor-initiating cells and suggest that glioblastomas in patients are most likely resistant to TRAIL treatment.

To confirm that the CD133⁺ population represents tumor-initiating cells (32), we established the self-renewal ability by neurosphere formation, the differentiation ability by cell differentiation assay, and the tumorigenic ability by mouse brain xenograft formation (Supplementary Fig. S10A–C). Once identified as the tumor-initiating cells, CD133⁺ cells were treated with 100 ng/mL TRAIL for 24 hours. Approximately 20% cell death was detected in EH 091112 and 100113 cells, but no cell death was seen in EH 091217 and 091106 cells (Fig. 7A).

Next, we transfected A20 siRNA in CD133⁺ cells and confirmed A20 knockdown by immunoblotting (Fig. 7B). This showed that transfection of the A20 siRNA alone had no significant effects on cell survival (Fig. 7A) and self-renewal (Supplementary Fig. S11A). The A20 siRNA-transfected CD133⁺ cells were treated with 100 ng/mL TRAIL, and TRAIL-induced apoptosis was observed, marked by a significant increase in the enzymatic activities of caspase-8 (Fig. 7C) and caspase-3/-7 (Supplementary Fig. S11B). To confirm that the Znf domain of E3 ligase inhibits caspase-8-initiated apoptosis, we examined the negative dominant effect of the inactive OTU and Znf4 mts on TRAIL-induced apoptosis in CD133⁺ cells. A20 wt, OTU mt, and Znf4 mt were introduced into CD133⁺ cells through lentiviral transduction. Expression of A20 Znf4 mt but not A20 wt or OTU mt enhanced TRAIL-induced apoptosis, as shown by cell death and caspase activity (Fig. 7D). Together, these results suggest that the Znf E3 ligase is responsible for caspase-8 inhibition and TRAIL resistance in the tumor-initiating cells of human glioblastomas.

Recent advances have generated novel cancer therapeutics targeting the TRAIL apoptotic pathway; however, the results of clinical trials have suggested that cancers are resistant to these treatments. The results presented here reveal a molecular mechanism by which A20 E3 ligase-mediated RIP1 polyubiquitination inhibits caspase-8 dimerization and cleavage and TRAIL-induced apoptosis in glioblastoma cells through two signaling complexes (Fig. 7E). A20 and RIP1 are both highly expressed in glioblastomas, in which they interact with the transmembrane DR5 and form a plasma membrane-bound preligand assembly complex under physiologic conditions. TRAIL treatment leads to the DR5-mediated recruitment of FADD and caspase-8 to the preligand assembly complex for the formation of the DISC, where the A20 E3 ligase mediates K63-linked RIP1 polyubiquitination. The K63-linked polyubiquitin chain of RIP1 binds to the caspase-8 protease domain, blocks caspase-8 dimerization and cleavage, and inhibits TRAIL-induced apoptosis in glioblastoma cells.

This 2-complex model is compatible with recent reports that ubiquitination regulates the TRAIL pathway. TRAIL-induced apoptosis requires caspase-8 polyubiquitination in that caspase-8 is a known substrate of the CUL3 E3 ligase and caspase-8 ubiquitination facilitates caspase-8 cleavage in TRAIL-sensitive cell lines (23). As a complement to this model, our study establishes a molecular mechanism of TRAIL resistance. Here, we identify caspase-8 as an ubiquitin binding protein and show that a RIP1 K63-linked polyubiquitin chain binds to the caspase-8 protease domain to inhibit its dimerization and cleavage. The findings that A20 and RIP1 are highly expressed and present in both the preligand assembly complex and DISC in resistant cells are in line with the recent report that the content and status of proteins in a cell determine whether or not cell death or cell survival occurs under TRAIL treatment (38).

The results reported here establish a novel model that reconciles the disparity in earlier studies. A size exclusion assay has identified TNFR1 in fractions corresponding to its monomeric molecular weight, leading to the notion that the TNFR1-associated complex I is formed after TNF-α stimulation to enact TNF-α–induced NF-κB signaling, whereas the cytoplasmic complex II is formed through the recruitment of FADD and caspase-8 after complex I is detached from TNFR1 to enact TNF-α–induced apoptosis (20). We report here a distinct, 2-complex TRAIL signaling model in which the preligand assembly complex is spontaneously formed under physiologic conditions. TRAIL treatment then stimulates the recruitment of FADD and caspase-8 to the preligand assembly complex for the formation of the DISC, leading to either cell death or survival depending on the composition and status of proteins in the complexes. A20 and RIP1 are associated with DR5 in the spontaneously formed preligand assembly complex in TRAIL-resistant cells, and TRAIL stimulates A20 E3 ligase-dependent RIP1 polyubiquitination and caspase-8 binding through a K63-linked polyubiquitin chain that inhibits caspase-8 dimerization and cleavage to ultimately prevents initiation of TRAIL-induced apoptosis. Once apoptosis is inhibited by A20-mediated RIP1 polyubiquitination, subsequent recruitment of IKKg (17), cellular FADD-like interleukin-1β-converting enzyme-inhibitory protein (c-FLIP), and phosphoprotein enriched in diabetes to the DISC leads to the activation of NF-κB (33, 39) and extracellular signal-regulated kinase 1/2 (40, 41) to promote cell growth (42).

The results presented here establish the role of A20 E3 ligase in the inhibition of TRAIL-induced apoptosis. A20 has an N-terminal OTU deubiquitinating enzyme and a C-terminal E3 ligase (28). An in vitro ubiquitination assay suggests that A20 E3 ligase mediates RIP1 K48-linked ubiquitination for its degradation and thus inhibits TNF-α–induced NF-κB signaling (28). In contrast, in vivo studies indicate that RIP1 is ubiquitinated through a K63-linked polyubiquitin (17) and that IKKg binds to either K63 or M1-linked polyubiquitin chains for NF-κB activation (18). Further studies have confirmed that the A20 can remove K63-linked polyubiquitin chain from RIP1, RIP2, and TRAF6 (25–27). However, A20 has been shown to inhibit TNF-α–induced apoptosis in mouse hepatocytes (24), IKKg-deficient Jurkat cells (43), and glioblastoma-derived CD133⁺ cells (37) through unknown mechanisms. We show here that the A20 E3 ligase is able to mediate RIP1 ubiquitination through K48- and K63-linked polyubiquitin chains in vitro and in vivo in the presence of a polyubiquitin chain-specific E2. In glioblastoma cells, however, A20 E3 ligase mediates RIP1 ubiquitination through a K63-linked polyubiquitin chain that binds to the caspase-8 protease and inhibits TRAIL-induced apoptosis.

Somatic mutations in TNFAIP3, the gene encoding the A20 protein, have been identified in lymphomas (44, 45). The mutations are clustered in the A20 Znf6 and Znf7 domains and result in loss of A20 function. Reconstitution of wild-type TNFAIP3 in the TNFAIP3-mutated lymphoma cells induces cell death (44). However, it is unclear how the transfected TNFAIP3 triggers apoptosis in the lymphoma cells. These results suggest that TNFAIP3 may act as a tumor suppressor gene in lymphomas (46). In contrast to the lymphomas, genomic analysis of glioblastomas by The Cancer Genome Atlas (47) has failed to identify TNFAIP3 mutations. However, the REpository for Molecular BRAin Neoplasia DaTa (REMBRANDT), sponsored by the National Cancer Institute, indicates TNFAIP3 mRNA overexpression in human glioblastomas, keeping in line with our finding that A20 protein is highly expressed in these tumors and inhibits TRAIL-induced apoptosis in the tumor-initiating cells. It appears that A20 plays a different role in lymphocytic tumors as opposed to solid tumors, which is consistent with the original report that it restricts NF-κB signaling in lymphocytes but protects hepatocytes from apoptotic insult in mice (24). In conclusion, the data presented here suggest that A20 E3 ligase acts as an oncogene that inhibits TRAIL-induced apoptosis. Targeting of the A20-mediated RIP1 ubiquitination process may therefore lead to the development of combination therapies that can eliminate TRAIL resistance in tumor-initiating cells and enhance the therapeutic efficacy of TRAIL-targeted therapies in human glioblastomas.

The glioblastoma and normal brain tissues used in Figure 1 were kindly provided by the London (Ontario) Brain Tumor Tissue Bank (London Health Sciences Center, London, Ontario, Canada) and the normal brain tissues were sampled from epilepsy lobectomy specimen. The glioblastoma tissues used in Figures 6 and 7 were collected from Emory University Hospital in accordance with protocols approved by the Emory University Institutional Review Boards. CD133 cells were sorted from dissociated tumors using CD133 antibody-labeled magnetic microbeads (Miltenyi Biotec) on the basis of previous reports (32) and labeled with EH (Emory Hospital) and numbers. Glioblastoma cell lines (33) and normal human astrocytes were reported (34), and no authentication was done by the authors since.

The pcDNA3.1/myc-his A20 wt, OTU, and Znf4 mt were generated through site-directed mutagenesis (Genscript). Lentiviral vector pLenti6.3/V5-DEST was inserted with A20 wt, OTU, and Znf4 mt and introduced into LN71 cells through lentiviral transduction in the presence of polybrene (2 μg/mL). The transfectants were grown in blasticidin-containing selection medium, and single-cell clones were expanded and examined by immunoblotting for the stable expression of the A20 protein.

The siRNA specific to TNFAIP3 (CCGAGCTGTTCCACTTGTTAA; CAGATGTATGGCTAACCGGAA), RIP1, TRAF2, and scrambled siRNA (QIAGEN) were transfected using HiPerfect Transfection (QIAGEN) for 72 hours. The transfectants were treated or untreated with TRAIL (PeproTech, Inc.) and examined by Cell Titer-Glo Luminescent Cell Viability assay for cell death and Caspase-Glo8 and Glo3/7 kits (Promega) for the caspase activities. Lentiviral scrambled shRNA (SHC002) and A20 shRNA (NM_006290.2-635s1c1:5′-CCGGCACTGGAAGAAATACACATATCTCGAGATATGTGTATTTCTTCCAGTGTTTTTG-3′) were from Sigma Mission RNAi.

Cell lines and CD133⁺ cells (10⁸), treated or untreated with 100 ng/mL TRAIL for 5 minutes, were lysed in CHAPS-containing lysis buffer (14 mM CHAPS, 150 mM NaCl, and 20 mM Tris-Hcl; pH 7.4) plus complete protease inhibitors (Sigma-Aldrich) and 1 mM phenylmethylsulfonylfluoride. Lysates were filtered through a 0.45-micron filter and loaded onto superdex-200 HR16/60 column. Proteins were eluted at 1 mL/min. Fractions (1 mL) were analyzed by western blotting for DR5, RIP1, A20, TRAF2, and caspase-8. The apparent molecular weight was evaluated after column calibration with standard proteins (GE Healthcare): thyroglobulin (669 kDa), ferritin (440 kDa), adolase (158 kDa), Conalbumin (75 kDa), and ovalbumin (43 kDa).

Subcellular fractionation experiments were performed using the ProteoExtract subcellular proteome extraction kit (Calbiochem) with 5 × 10⁶ cells per sample. For RIP ubiquitination, the cytosol and membrane fractions were treated with 1% SDS and boiled for 10 minutes to separate proteins in the complexes and then diluted 10× to 0.1% SDS. RIP1 was immunoprecipitated with the use of 2 μg of RIP1 polyclonal antibody (Santa Cruz Biotechnology, Inc.) for 4 hours that was washed 5 times: 2 times for 10 minutes in lysis buffer plus 1M NaCl and 3 times for 10 minutes in lysis buffer.

To isolate the DISC, 1 × 10⁷ cells were incubated with mixed 500 ng/mL Flag-TRAIL and 1,500 ng/mL Flag antibody for 15 to 90 minutes at 37°C or 3 hours at 4°C for the detection of CUL3. After wash, the cells were lysed for 30 min on ice in DISC immunoprecipitation lysis buffer (30 mM Tris, pH 7.5, 150 mM NaCl, 10% glycerol, 1% Triton X-100). Cell lysates were incubated with protein G-linked agarose overnight at 4°C. The beads were washed and eluted with 150 ng/μL of 33 Flag peptide (Sigma-Aldrich) (33). To isolate the preligand assembly complex, the cells were lysed first and then incubated with mixed Flag-TRAIL and Flag antibody as described previously. The preligand assembly complex, DISC, size exclusion samples, subcellular fractions and lysates from cell lines, CD133⁺ cells, and tissues were examined by immunoblotting as described (33) with the use of antibodies specific to CUL3 (BD Biosciences), DR5 (ProSi Inc.), TRAF2 (H249), ubiquitin (P4D1), myc (Santa Cruz Biotechnology, Inc.), RIP1, FADD (BD Biosciences), K63-linked polyubiquitin chain (Biomol International and Millipore), caspase-8 (MBL), DFF45 (StressGen), HA (Covance), and His tag (Novagen).

Flag-RIP1 and myc-A20 wt proteins were generated through TNT Quick Coupled Transcription/Translation Systems (Promega). After reaction, a Flag-RIP1 was mixed with Flag antibody-labeled agarose beads and eluted using Flag peptide (Sigma-Aldrich) and concentrated through Amicon Ultra-0.5 Centrifugal Filters (Millipore). Myc-A20 wt protein was isolated through immunoprecipitation using myc-agarose beads. In vitro ubiquitination was performed in a 20-μL reaction volume containing 2 μg of N-terminal biotinylated ubiquitin, 5 μg of ubiquitin, 200 ng of E1, 400 ng of UBC13 (Boston Biochem) or UBCH5A (Calbiochem), 2 μL of 10× reaction buffer, and 1× Mg-ATP (Boston Biochem). After a 1-hour incubation at 30°C, reactions were terminated by adding SDS loading buffer and examined by immunoblotting. For in vivo ubiquitination, U87MG cells were transfected with plasmids encoding HA-UB and mts for 24 hours and treated or untreated with 100 ng/mL TRAIL for 1.5 hours. Cell lysates were heated at 95°C for 10 minutes in 1% SDS to dissociate proteins and diluted ten times in non-SDS-containing buffer. RIP1 was isolated by RIP1 antibody (Santa Cruz Biotechnology, Inc.) through immunoprecipitation and examined by immunoblotting using HA (Covance) and RIP1 antibody (BD Biosciences).

In vivo caspase-8 binding to ubiquitin was examined as follows: LN443 cells were treated or not with 100 ng/mL TRAIL for 1.5 and 3 hours and subjected to subcellular fractionation. RIP1 was isolated from the cytosol and membrane fraction under denaturing conditions, then incubated with Fv-caspase-8 for 1 hour and examined by immunoblotting using antibodies to caspase-8 and K63-linked polyubiquitin chain. In vitro caspase-8 binding to ubiquitin was examined in the following 2 experiments. In the first experiment, Fv-caspase-8-labeled or caspase-8 protease p18 domain-labeled agarose beads were incubated with recombinant ubiquitin (Boston Biochem) for 1 hour at 4°C in ubiquitin binding buffer (50 mM Tris, 150 mM NaCl, 0.2% Triton x-100, 1 mM EDTA, 0.5 mM DTT). After washes, the agarose beads were precipitated by centrifugation. In the second experiment, Fv-caspase-8 or caspase-8 protease p18 subunit was incubated with His-K63-linked polyubiquitin chain-bound Nickel agarose beads (QIAGEN). In both experiments, unbound proteins were washed off the beads with binding buffer, and bound proteins were eluted by 1% SDS-containing binding buffer and examined by immunoblotting using ubiquitin and caspase-8 antibody. The caspase-8 dimerization and cleavage assay was performed as described (35, 36).

All values are expressed as mean ± SD. Statistical significance was assessed by unpaired Student t test, one-way ANOVA followed by Dunnett test, and 2-way ANOVA followed by Bonferroni test.

No potential conflicts of interest were disclosed.

We thank Dr. Keith Wilkinson for his constructive suggestions of ubiquitination experiments and Zhaobin Zhang for his technical assistance in animal studies.

This work was supported in part by NIH grant CA129687 to C. Hao. C. Hao was a Georgia Cancer Coalition Distinguished Scholar.