In order to examine whether degradation of CYLD observed in TNF stimulated MEFs was due to proteolytic cleavage, FLAG-CYLD was immunoprecipitated from the reconstituted Cyld^−/− MEFs and blotted with the same antibody. A FLAG-tagged product from CYLD of approximately 25kDa (CYLDp25) was detected upon TNF stimulation (Figure 2a) suggesting that CYLD undergoes cleavage. Furthermore, the 25kDa cleavage product from endogenous CYLD was similarly detected in untransfected wildtype MEFs (Figure 2b). We hypothesised that CYLD protein might be regulated by active CASPASE 8, particularly since computational analysis also indicated a relationship between CASPASE 8 and CYLD gene expression levels (Supplementary Figure 1), especially in lymphoid cells. Consistent with this hypothesis, the CASPASE 8 inhibitor IETD-fmk reduced the level of the CYLDp25 fragment. Co-transfection of HEK 293 cells revealed that over-expression of wild-type CASPASE 8, but not the catalytically inactive mutant CASPASE 8-C360S, causes degradation of CYLD protein (Figure 2c). Interaction between transfected CYLD and CASPASE 8 by co-immunoprecipitation was observed only when the activity of CASPASE 8 was blocked by the pan-caspase inhibitor zVAD-fmk, or by mutation of the CASPASE 8 active site, suggesting that CYLD is a substrate for proteolytic cleavage by CASPASE 8 (Figure 2d). To provide genetic evidence that CASPASE 8 is cleaving CYLD, FLAG-CYLD was stably expressed in Casp8^+/+ and Casp8^−/− MEFs and the cells were stimulated with TNF. The p25 fragment was not detected in the Casp8^−/− MEFs (Figure 2e) indicating that CYLDp25 is a product of the proteolytic cleavage of CYLD by CASPASE 8, under conditions where active CASPASE 8 suppresses necrosis^{7, 12} (CRITERIA #2).

In the MEF model, CASPASE 8 generates a survival signal. Since CASPASE 8 is also known to initiate apoptosis, which antagonizes programmed necrosis, we examined whether CYLD is similarly cleaved by CASPASE 8 during apoptosis in the human Jurkat T cell model. Levels of endogenous CYLD were decreased upon TNF stimulation of SMAC mimetic¹³-treated parental Jurkat T cells (clone A3) but not in CASPASE 8-deficient mutant clone I9.2 (Supplementary Figure 2a). Likewise, endogenous CYLD was cleaved to generate CYLDp25 when CASPASE 8 was activated following TNF stimulation of NEMO-deficient Jurkat T cells (clone 8321¹⁴, Supplementary Figure 2c). CYLD was not cleaved in CASPASE 8-deficient Jurkat T cells treated with TNF and SMAC mimetic (Supplementary Figure 2a, 2d), which failed to undergo apoptosis but instead died by programmed necrosis⁵ (Supplementary Figure 2b). Similar to the MEF model, a correlation between the stabilisation of CYLD and necrosis is also observed in the Jurkat T cell model. Finally, we examined CYLD cleavage in a third model using murine L929 cells. Cleavage of endogenous CYLD (Supplementary Figure 3a) or transfected FLAG-tagged CYLD (Supplementary Figure 3b) could be detected in TNF-treated L929 cells in a caspase-dependent manner. When Caspase 8 activity was inhibited, L929 cells died by necrosis, which was blocked by knockdown of either CYLD or RIPK3 (Supplementary Figure 3c). Processing of CYLD to CYLDp25 was also observed in SMAC mimetic and TNF-treated MCF7 human breast cancer cells (Supplementary Figure 4a), which are deficient in CASPASE 3¹⁵, indicating that CASPASE 3 is not required for CYLD cleavage. Therefore CYLD cleavage by CASPASE 8 is observed whether it is generating a survival or an apoptotic death signal, and CYLD stabilisation coincided with cells undergoing necrosis. To provide further corroboration that CASPASE 8 activity was required for CYLD cleavage, we stably expressed CrmA, a viral inhibitor of CASPASE 8 and a known trigger of programmed necrosis^{2, 4}, in FLAG-CYLD-expressing MEFs. CrmA prevented processing of CYLD to CYLDp25 (Supplementary Figure 4b) whereas an aspartate to alanine mutation that abrogates the ability of CrmA to inhibit CASPASE 8 (CrmA-D303A) was unable to prevent cleavage of CYLD. Likewise, wildtype CrmA expression predisposed MEFs to programmed necrosis upon TNF treatment whereas mutant CrmA did not (Supplementary Figure 4c). Therefore, the genetic data from MEFs and Jurkat T cells indicate that the cleavage of CYLD upon TNF treatment requires CASPASE 8. The data with the pharmacological and viral inhibitors of CASPASE 8 activity suggest that proteolytically active CASPASE 8 is required for processing of CYLD. To test whether CASPASE 8 could directly cleave CYLD, affinity-purified FLAG-tagged CYLD protein was incubated with recombinant CASPASE 8. Purified CYLD protein was processed by recombinant CASPASE 8 to generate the N-terminal CYLDp25 fragment (Figure 2f). Therefore, CASPASE 8 directly cleaves CYLD.

Amino acid sequences of CYLD from several species were aligned to identify a conserved CASPASE 8 cleavage site motif (LxxD) that would produce an N-terminal 25kDa fragment (Figure 3a). Two potential motifs were identified and mutation of D215 was sufficient to prevent proteolysis of CYLD by CASPASE 8. Upon co-expression in HEK 293 cells, CYLD-D215A was more resistant to degradation by CASPASE 8 (Figure 3b) even though both CYLD-WT and CYLD-D215A co-precipitated with CASPASE 8 (Figure 3c) to the same extent. The D215A mutation abrogated the appearance of CYLDp25 in TNF-stimulated reconstituted Cyld^−/− MEFs (Figure 3d) or in Jurkat T cells and L929 cells (Supplementary Figure 2e and 3b, respectively), suggesting that CASPASE 8 cleaves CYLD at D215. Recombinant CASPASE 8 was unable to generate the CYLDp25 fragment from purified CYLD-D215A in vitro (Figure 2f), confirming that CASPASE 8 directly cleaves CYLD after D215.

In both the MEF and Jurkat T cell models, there was a strict correlation between necrosis and stabilisation of CYLD. CYLD was rapidly lost in TNF-treated CASPASE 8-sufficient Jurkat T cells dying by apoptosis but remained unchanged in the CASPASE 8-deficient cells dying by necrosis (Supplementary Figure 2a). Cell death in CASPASE 8-deficient Jurkat T cells was blocked by Necrostatin-1^{2, 16}, confirming that CASPASE 8 represses programmed necrosis (Supplementary Figure 2b). Similarly, blocking CYLD cleavage in MEFs with caspase inhibitors correlated with entry into necrotic death (Figures 1b & 2b). While dramatic down-regulation of CYLD was observed in TNF-treated cells, no detectable change occurred for RIPK1 and RIPK3 in either model (Figure 1d and Supplementary Figure 2a). Therefore, we hypothesised that CYLD is the key substrate cleaved by CASPASE 8 to prevent necrosis. To test this hypothesis, Cyld^−/− MEFs stably reconstituted with CYLD-WT or CYLD-D215A were treated with TNF. CYLD-D215A-expressing MEFs quickly died by necrosis, without the need for a caspase inhibitor (Figure 4a). In contrast, MEFs expressing CYLD-WT only underwent significant amounts of necrosis if treated with TNF in the presence of IETD-fmk (Figure 4a) or zVAD-fmk (Figure 4b). Cell death initiated by TNF in Cyld^−/− MEFs expressing CYLD-D215A was blocked by the RIPK1 inhibitor Necrostatin-1¹⁷ (Figure 4a, compare panel 6 and panel 7), or by specific knockdown of Ripk1 or Ripk3 (Figure 4c), demonstrating that RIPK1 and RIPK3 are downstream of CYLD. Neither RIPK1 nor RIPK3 was required for the processing of CYLD to CYLDp25 in MEFs (Supplementary Figure 4d), indicating that CASPASE 8 inactivation of CYLD occurs upstream of these necrosome components. Recruitment of RIPK1 to the necrosome is strictly dependent on CYLD (Figure 1c) and CASPASE 8 removal of CYLD would be expected to prevent the recruitment of RIPK1. Therefore, stabilising CYLD by mutating the CASPASE 8 cleavage site at D215 is sufficient to permit TNF-induced necrosis even in the presence of CASPASE 8 (CRITERIA #3).

CYLD is a deubiquitinase that removes K63-polyubiquitin chains from RIP1¹⁸; these ubiquitin chains are required for recruitment of signaling molecules such as NEMO^{19, 20} that prevent RIPK1 from functioning as a pro-death molecule^{21, 22}, in part by preventing RIPK1 from associating with downstream death signaling molecules and in part by mediating activation of pro-survival NFκB. Consistent with an inhibitory role for CYLD during IKK activation, Cyld^−/− MEFs display faster kinetics of IκBα phosphorylation when compared to Cyld^−/− MEFs reconstituted with CYLD-WT (Supplementary Figure 5a). However, the kinetics of IκBα phosphorylation appears to be fairly similar between the Cyld^−/− cells reconstituted with CYLD-WT or CYLD-D215A with the D215A cells displaying a modest reduction in IκBα phosphorylation at 60 minutes. In a control experiment, mutation of D215 did not affect the ability of CYLD to bind NEMO in HEK 293 cells (Supplementary Figure 5b). We confirmed that the level of CYLD-WT was reduced following TNF stimulation whereas the CYLD-D215A mutant was not (Supplementary Figure 6a), coincident with survival in the former and necrotic cell death in the latter. Thus while the CYLD D215A mutation had a minimal effect on NF-κB signaling, it had a striking effect on necrosis suggesting that CYLD deubiquitination of downstream molecules such as RIPK1 controls entry into this death pathway. To test this hypothesis, we first examined the fate of the C-terminal p82 fragment containing the deubiquitinase domain resulting from the cleavage. In experiments up to this point, this C-terminal fragment was not detectable suggesting that it may be unstable. Sequence analysis of CYLD indicated a PEST motif from amino acids 397 to 415. Addition of the proteosome inhibitor MG132 to TNF-stimulated MCF7 cells (Figure 5a) or MEFs (Supplementary Figure 6b) lead to the stabilisation of the p82 fragment. Therefore when TNF is promoting survival, CASPASE 8 cleaves CYLD to remove the deubiquitinase domain, which is likely to affect RIPK1 ubiquitination and its interaction with signaling partners. We examined this by immunoprecipitating the pro-survival NEMO complex from TNF-treated CYLD-WT and CYLD-D215A-expressing MEFs because NEMO binds ubiquitinated RIPK1^{19, 20}. Therefore, the amount of RIPK1 in the NEMO complex can be regarded as an indirect measure of RIPK1 ubiquitination. Furthermore, the NEMO-RIPK1 complex has been proposed to be a survival complex that functions by preventing RIPK1 from interacting with downstream death signaling molecules^22–24. More RIPK1 protein co-precipitated with NEMO from CYLD-WT than CYLD-D215A-expressing MEFs suggesting that RIPK1 ubiquitination may be reduced in the D215A cells (Figure 5b). Interestingly, the level of ubiquitinated proteins in the NEMO complex was reduced in CYLD-D215A MEFs at both steady state and upon TNF stimulation consistent with the idea that CYLD stabilisation resulted in enhanced deubiquitinase activity (Figure 5b). The ubiquitination level of RIPK1 was directly examined by RIPK1 immunoprecipitation followed by blotting with anti-ubiquitin. RIPK1 isolated from the CYLD-D215A MEFs exhibited lower levels of ubiquitination after TNF stimulation, which is most discernible at the 2 and 3 hour time points (Figure 5c), suggesting that stabilisation of the CYLD protein leads to accelerated deubiquitination of RIPK1 concurrent with reduced interaction of RIPK1 with the pro-survival NEMO protein. Disruption of the interaction of RIPK1 with NEMO would be predicted to lead to increased association of RIPK1 with the downstream necrotic death apparatus. Indeed, more RIPK1 was recruited to the necrosome complex after 90 minutes of TNF treatment of CYLD-D215A-expressing MEFs (Figure 5d). Therefore, removal of CYLD-WT by CASPASE 8 prolongs the ubiquitination state of RIPK1 and maintains RIPK1 in a pro-survival complex with NEMO. In contrast, the resistance of CYLD-D215A to proteolytic cleavage by CASPASE 8 resulted in less ubiquitination of RIPK1 (Figure 5), which enhanced RIPK1 interaction with the necrosome to initiate death (Figure 4). These observations demonstrate that CASPASE 8 cleavage of CYLD functions as a pro-survival event by preventing CYLD from deubiquitinating RIPK1.

In conclusion, we provide evidence that CYLD is cleaved by CASPASE 8 to promote survival. Significantly, the cleavage fragment containing the deubiquitinase domain is unstable and undergoes degradation by the proteosome. CYLD level diminished in a time-dependent manner after TNF stimulation suggesting that CYLD may be continuously recruited to and degraded by the CASPASE 8 signaling complex to maintain the suppression of necrosis. CASPASE 8-dependent CYLD degradation is also observed in cells undergoing apoptosis, which antagonizes necrosis. Therefore proteolytic degradation of CYLD may serve to disable the necrosis machinery in multiple settings. More significantly, mutation of the cleavage site on CYLD drives MEFs into necrotic cell death with no requirement for caspase inhibition. Cleavage of CYLD by CASPASE 8 is both necessary and sufficient for CASPASE 8 to repress programmed necrosis, indicating that processing of CYLD is the critical determinant of cell survival versus death by necrosis in MEFs. Processing of CYLD by CASPASE 8 does not require RIPK1 or RIPK3, which suggests that this cleavage event occurs early after TNFR1 ligation in order to prevent recruitment of RIPK1 to the necrosome. When the removal of CYLD by CASPASE 8 is prevented by mutation of D215, we observe accelerated deubiquitination of RIPK1 and the dismantling of the pro-survival RIPK1 and NEMO complex, with the concomitant formation of the RIPK1 and FADD necrosome complex. Since CYLD is a tumour-suppressor²⁵ and can be removed by CASPASE 8, this study should motivate a re-examination of the controversial role of CASPASE 8 in tumourigenesis. CYLD is a known regulator of other non-apoptotic functions attributed to active CASPASE 8 such as proliferation, cell migration and metastasis²⁶; therefore, processing of CYLD by CASPASE 8 is likely to have important biological functions that extend beyond the repression of programmed necrosis.

1

2

3