The tumour suppressor CYLD regulates the p53 DNA damage response

The tumour suppressor CYLD is a deubiquitinase previously shown to inhibit NF-κB, MAP kinase and Wnt signalling. However, the tumour suppressing mechanisms of CYLD remain poorly understood. Here we show that loss of CYLD catalytic activity causes impaired DNA damage-induced p53 stabilization and activation in epithelial cells and sensitizes mice to chemical carcinogen-induced intestinal and skin tumorigenesis. Mechanistically, CYLD interacts with and deubiquitinates p53 facilitating its stabilization in response to genotoxic stress. Ubiquitin chain-restriction analysis provides evidence that CYLD removes K48 ubiquitin chains from p53 indirectly by cleaving K63 linkages, suggesting that p53 is decorated with complex K48/K63 chains. Moreover, CYLD deficiency also diminishes CEP-1/p53-dependent DNA damage-induced germ cell apoptosis in the nematode Caenorhabditis elegans. Collectively, our results identify CYLD as a deubiquitinase facilitating DNA damage-induced p53 activation and suggest that regulation of p53 responses to genotoxic stress contributes to the tumour suppressor function of CYLD.

C YLD is a tumour suppressor originally identified as the gene mutated in familial cylindromatosis, a disease characterized by the development of benign skin tumours arising from ectodermal appendages 1 . Subsequently, decreased CYLD expression and mutations of the CYLD genomic locus have been reported in a number of different cancers including colon, hepatocellular and myeloid malignancies, suggesting that CYLD exhibits broad tumour suppressor functions [2][3][4][5] .
CYLD is a deubiquitinating enzyme that selectively hydrolyses K63-and M1-linked ubiquitin chains but exhibits very little activity towards K48-linked ubiquitin chains 6 . CYLD was shown to negatively regulate NF-kB and MAPK activation by removing ubiquitin chains from key signalling molecules including NEMO, TRAF2, TRAF6 and RIPK1 (refs 7-10). In addition, CYLD was proposed to inhibit Wnt signalling by deubiquitinating dishevelled 11 . Furthermore, CYLD was identified as an important regulator of TNF-induced apoptosis 12 and programmed necrosis 13 , presumably by deubiquitinating RIPK1 to allow the formation of death-inducing protein complexes. Moreover, we showed previously that CYLD catalytic activity is important for the induction of RIPK3-mediated necroptosis in vivo 14,15 .
CYLD-deficient mice did not develop spontaneous tumours but showed increased susceptibility in different models of carcinogenesis. Mice lacking CYLD exhibited exacerbated colon carcinogenesis induced by administration of the chemical carcinogen azoxymethane (AOM) followed by repeated cycles of dextran sulphate sodium (DSS) -induced inflammation 16 . In this study, the authors proposed that CYLD deficiency leads to increased NF-kB activation in macrophages resulting in elevated inflammation, which drives enhanced colon carcinogenesis. Moreover, CYLD deficiency resulted in chronic production of tumour-promoting cytokines by tumour-associated macrophages leading to more aggressive tumour growth in a syngeneic model of lung cancer 17 . CYLD-deficient animals also showed increased susceptibility to skin tumorigenesis induced by a single topical application of the chemical carcinogen 7,12dimethylbenz(a)anthracene (DMBA) followed by repeated cycles of inflammation induced by application of 12-O-tetradecanoylphorbol-13-acetate (TPA) (ref. 18). In this study, Massoumi et al. suggested that CYLD deficiency in keratinocytes caused sustained ubiquitination and nuclear translocation of BCL3 in response to ultraviolet light or TPA, which induced elevated expression of Cyclin D1 resulting in increased keratinocyte proliferation and tumour growth. Furthermore, transgenic expression of catalytically inactive mutant CYLD in the epidermis also sensitized mice to DMBA/TPA-induced skin tumorigenesis 19 . However, in this study the authors suggested that increased activation of JNK and not NF-kB was responsible for the enhanced tumorigenesis caused by the expression of catalytically inactive CYLD. Therefore, it appears that CYLD may suppress tumour development by different mechanisms acting either in a cell intrinsic manner in premalignant epithelial cells or by regulating the tumour microenvironment by acting in myeloid cells.
The p53 transcription factor is a key tumour suppressor that is mutated in more than 50% of human cancers. p53 not only maintains genomic stability after cellular stress by controlling the expression of genes regulating cellular senescence, cell cycle progression, cell death and DNA repair but has also been recently implicated in the regulation of cellular metabolism, stem cell maintenance and the tumour microenvironment [20][21][22] . Regulation of p53 protein stability by the ubiquitin/proteasome system is the main mechanism controlling p53 function 23,24 . p53 ubiquitination by Mdm2 and a number of other E3 ubiquitin ligases including COP1, Pirh2, ARF-BP1, MSL2 and Parc mediates the degradation and controls the subsecullar localization of p53 (refs 23,24). In response to cellular stress, HAUSP and a number of other deubiquitinating enzymes, including Otub1, USP10, USP29 and USP42 not only remove ubiquitin chains from p53 but also other proteins regulating p53 ubiquitination including Mdm2, inducing p53 stabilization (refs 23,24).
Here we show that CYLD acts in intestinal and skin epithelial cells to suppress DNA damage-induced tumour development and that this tumour suppressor function of CYLD is mediated at least in part by the regulation of p53 activation. Lack of CYLD catalytic activity results in impaired stabilization of p53, reduced expression of p53 target genes as well as reduced apoptosis in epithelial cells in response to genotoxic stress. Elevated transgenic expression of p53 restores DNA damage-induced cellular responses and partly normalizes the increased tumour susceptibility in mice expressing catalytically inactive CYLD in vivo. In addition, we show that CYLD regulates DNA damage-induced p53 responses in Caenorhabditis elegans suggesting that this function of CYLD is evolutionarily conserved. Mechanistically, we show that CYLD interacts with and deubiquitinates p53 in response to DNA damage. Moreover, we provide evidence that CYLD removes K48-linked ubiquitin chains from p53 indirectly by cleaving K63 chains, suggesting that p53 is decorated with complex ubiquitin chains.

Epithelial CYLD suppresses AOM/DSS-induced colon cancer.
To study the function of CYLD deubiquitinase (DUB) activity in tumorigenesis we employed a conditional knockin mouse model allowing the cell-specific expression of the catalytically inactive CYLDR932X mutant (CYLDD932) (ref. 15). The R932X mutation in mouse CYLD is equivalent to the R936X mutation in human CYLD, which truncates subdomain III of the Histidine box that is essential for CYLD catalytic activity 8 and was found to cause cylindromas 1 (Supplementary Fig. 1a). By crossing mice carrying CYLDD932 floxed (CYLDD932 FL ) alleles with LysM-Cre and Villin-Cre transgenics we generated two mouse lines expressing the catalytically inactive CYLDD932 mutant specifically in myeloid (CYLDD932 mye ) or intestinal epithelial cells (IECs) (CYLDD932 IEC ), respectively (Supplementary Fig. 1b-d).
To address the myeloid cell-specific role of CYLD in inflammation-associated colon cancer, we induced colon carcinogenesis in CYLDD932 mye mice by a single injection of the alkylating agent AOM followed by repeated cycles of inflammation induced by oral administration of DSS 25 (Fig. 1a). These experiments revealed that CYLDD932 mye mice showed similar levels of inflammation and tumorigenesis compared with their CYLDD932 FL littermates after AOM/DSS treatment ( Fig. 1b-d). Therefore, in contrast to the suggestion by Zhang et al. 16 that CYLD deficiency in macrophages was responsible for the increased AOM/DSS-induced colon carcinogenesis in Cyld À / À mice to AOM/DSS-induced colon cancer, loss of CYLD catalytic activity specifically in myeloid cells did not sensitize mice to AOM/DSS-induced inflammation and tumour development.
We then assessed the role of epithelial CYLD by exposing CYLDD932 IEC mice to AOM/DSS colon carcinogenesis. CYLDD932 IEC mice reacted very strongly to AOM/DSS treatment, requiring the use of a milder protocol including only one DSS cycle to ensure survival of the mice (Fig. 1e). Under this protocol, CYLDD932 IEC mice developed pronounced weight loss associated with severe colitis, while their CYLDD932 FL littermates exhibited only mild colitis (Fig. 1f,g). Tumour evaluation on day 60 revealed that CYLDD932 IEC mice harboured increased numbers of tumours in their colons compared with their CYLDD932 FL littermates (Fig. 1h) increased incidence of high-grade intraepithelial neoplasia and carcinoma in situ compared with CYLDD932 FL littermates (Fig. 1i,j). Collectively, these results showed that CYLD catalytic activity functions in epithelial cells to suppress AOM/DSSinduced colon tumorigenesis.
CYLD suppresses DNA damage-induced tumorigenesis. To dissect whether CYLD suppresses colon tumorigenesis by preventing AOM-mediated tumour initiation or DSS-induced tumour promotion, we first studied the response of CYLDD932 IEC mice to DSS (Fig. 2a). When treated with multiple cycles of DSS in the absence of AOM, CYLDD932 IEC mice showed similar weight loss and colitis development compared with their CYLDD932 FL littermates, while none of the mice showed colon tumours when sacrificed on day 180 (Fig. 2b,c). These results showed that epithelial CYLD does not regulate DSSinduced colon injury and inflammation, and suggested that CYLD catalytic activity might be important for the response of epithelial cells to AOM-induced DNA damage.
To assess the role of epithelial CYLD in DNA damage-driven tumorigenesis, we exposed CYLDD932 IEC mice to a model of inflammation-independent intestinal tumour development induced by repeated injections of AOM (Fig. 2d). Endoscopic analysis on week 12 revealed the presence of colon tumours in all CYLDD932 IEC mice, while none of their CYLDD932 FL littermates showed tumours at this stage (Fig. 2e). Macroscopic and histological analysis of colons on week 18 revealed strongly increased AOM-induced tumorigenesis in CYLDD932 IEC mice compared with CYLDD932 FL littermates. CYLDD932 IEC mice showed increased tumour incidence and developed more tumours, which were larger and more advanced with increased incidence of highgrade intraepithelial neoplasia and carcinoma in situ compared with CYLDD932 FL animals ( Fig. 2f-h). Thus, epithelial specific inhibition of CYLD catalytic activity sensitized mice to AOMinduced colon carcinogenesis.
To test whether CYLD regulates carcinogen-induced tumour formation in other epithelia, we generated mice expressing the catalytically inactive CYLDD932 mutant in epidermal keratinocytes (CYLDD932 epi ) by crossing CYLDD932 FL mice with K14-Cre transgenics ( Supplementary Fig. 1e,f). To induce skin tumours, we treated CYLDD932 epi and CYLDD932 FL littermates with eight consecutive weekly topical applications of the carcinogen DMBA on the shaved back skin (Fig. 2i). Evaluation of skin tumour development 12 weeks after the last DMBA application revealed the presence of papillomas in B70% of the CYLDD932 epi mice compared with B30% of their littermate controls. Moreover, about two-thirds of the papillomas found in CYLDD932 epi mice were classified as medium to big, while nearly all papillomas found in CYLDD932 FL mice were small (Fig. 2j,k). Collectively, these results showed that CYLD catalytic activity in epithelial cells suppresses DNA damage-induced tumorigenesis in the intestine and skin.
CYLD promotes DNA damage-induced apoptosis. Cell death amid irreparable genomic lesions is a mechanism preventing DNA damage-driven tumour development. We reasoned that CYLD might prevent chemical carcinogen-induced tumour development by promoting DNA damage-induced cell death. Indeed, we found that 8 h after AOM injection colonic crypts in CYLDD932 IEC mice contained reduced numbers of TUNELpositive epithelial cells compared with CYLDD932 FL littermates, indicating that loss of CYLD DUB activity protected IECs from DNA damage-induced death (Fig. 3a). CYLD was previously identified as an important regulator of cell death induced by RIP kinase 1 (RIPK1) downstream of death receptor and Toll-like receptor signalling 12,13 . Furthermore, we showed that CYLD catalytic activity regulates RIPK3-induced necrosis in intestinal and skin epithelial cells in vivo 14,15 . Since RIPK1 was recently shown to regulate cellular responses to DNA damage 26 , we hypothesized that CYLD might control DNA damage-induced cell death and tumour development by regulating RIP kinasedependent pathways and tested this hypothesis using relevant genetic mouse models. We found that neither Ripk3 À / À nor double FADD IEC-KO /Ripk3 À / À mice, which lack RIPK3 in all cells and FADD specifically in IECs 15 , showed increased susceptibility to AOM/DSS-or AOM-mediated intestinal tumorigenesis ( Supplementary Fig. 2). These results suggest that the tumour suppressing role of CYLD is independent of RIPK3 and FADD/ caspase-8 mediated cell death pathways.
CYLD facilitates DNA damage-induced p53 activation. Since AOM and DMBA are genotoxic agents, we reasoned that CYLD activity might increase DNA damage-induced cell death and therefore suppress tumour development by regulating the DNA damage response in epithelial cells. DNA damage induces the stabilization and activation of the transcription factor p53, which promotes cell cycle arrest to facilitate DNA repair or induces the death of cells carrying irreparable DNA lesions. Moreover, AOM-induced death of IECs depends on p53 signalling 27 . To address whether CYLD controls the p53-dependent death of epithelial cells in response to DNA damage we employed intestinal organoid cultures. Organoids from CYLDD932 IEC , CYLDD932 FL , p53 IEC-KO and p53 FL mice were treated with Camptothecin (CpT), a topoisomerase inhibitor inducing replication fork stalling and the formation of DNA doublestrand breaks 28 or Mitomycin C (MMC), a chemical carcinogen that, similarly to AOM, induces DNA alkylation and subsequent interstrand crosslink (ICL) formation 29 , and cell death was assessed by microscopic evaluation as well as detection of caspase-3 cleavage by immunocytochemical, immunoblot and fluorescence-activated cell sorting (FACS) analyses. As expected, CpT and MMC treatment induced epithelial cell death in wild type but not in p53-deficient organoids (Fig. 3b,c and Supplementary Fig. 3a-b). Consistent with our in vivo results, CYLDD932 mutant organoids were largely protected from CpT-and MMC-induced epithelial cell death compared with CYLDD932 FL organoids ( Fig. 3b-d and Supplementary  Fig. 3a-c). Therefore, lack of CYLD catalytic activity inhibited the p53-dependent death of IECs in response to DNA damage in vivo and in vitro.
To address how CYLD regulates DNA damage-induced p53 responses we first examined whether CYLD regulates p53 stabilization in response to genotoxic stress. We found that CYLDD932 organoids showed strongly impaired stabilization of p53 as well as reduced expression of the p53-dependent genes Cdkn1a, Bax and gadd45 in response to CpT (Fig. 3e,f). To assess whether lack of CYLD DUB activity affected DNA damageinduced p53 responses in other epithelial cells, we analysed CpT-induced p53 activation in primary epidermal keratinocytes. Similarly to IECs, CYLDD932 mutant keratinocytes showed impaired DNA damage-induced p53 stabilization and p53dependent gene expression compared with controls ( Fig. 3g,h). Together, these experiments showed that CYLD DUB activity is required for efficient stabilization and activation of p53 and the induction of cell death in primary epithelial cells exposed to DNA damage.  stabilization of p53 in response to genotoxic stress. To assess whether increased expression of p53 could rescue impaired DNA damage responses in cells expressing mutant CYLD, we crossed the CYLDD932 IEC mice with 'super p53' mice carrying supernumerary copies of the p53 gene in the form of large genomic transgenes 30 . Indeed, increased p53 expression could largely restore DNA damage-induced apoptosis and expression of p53 target genes in CYLDD932 IEC intestinal organoids ( Fig. 4a-c). In addition, CYLDD932 IEC Superp53 mice developed less AOM/DSS-and AOM-induced colon tumours compared with CYLDD932 IEC littermates, showing that increased p53 expression could partially restore the increased susceptibility of CYLDD932 IEC mice to chemical carcinogen-induced colon tumorigenesis (Fig. 4d,e). These results provide additional experimental evidence supporting that CYLD controls cellular responses to DNA damage by promoting the stabilization and activation of p53, and that regulation of p53-dependent responses to genotoxic stress contributes to the tumour suppressing function of CYLD.
Inactivating CYLD mutations are found in several human cancers. Mutations of the Cyld genomic locus have been previously reported in different tumour types 2,3,5 . We analysed sequencing data from 7,042 human tumour samples 31  cancers ( Supplementary Fig. 4a). Determination of the frequency of inactivating mutations in 27,836 genes across cancer (COSMIC database) showed that inactivating mutations were significantly enriched in CYLD and that the CYLD mutation spectrum resembled the spectrum of known tumour suppressors (Supplementary Fig. 4b). In addition, we found a statistically significant association between the presence of CYLD mutation and the accumulation of mutations per tumour ( Supplementary Fig. 4c), suggesting that loss of CYLD catalytic activity may favour the accumulation of somatic mutations in human cancers by impairing DNA damage-induced p53 responses. Moreover, CYLD was predicted to regulate p53 in an unbiased bioinformatics approach aiming to build a functional human protein interaction network by combining protein interaction, gene expression and gene ontology annotations with genome-wide cancer data sets 32 , further supporting that regulation of p53 signalling by CYLD is functionally relevant for human cancer.
CYLD regulates DNA damage-induced p53 responses in C. elegans. To assess the functional conservation of the role of CYLD in regulating p53-dependent cellular responses to DNA damage, we investigated the function of the highly conserved cyld-1 gene in C. elegans. Similar to mammals, nematodes activate a conserved DNA damage checkpoint pathway that induces cell cycle arrest of mitotic germ cells and in meiotic pachytene cells activates the p53 homologue CEP-1 to trigger apoptosis [33][34][35] through transcriptional activation of the BH3 domain only proteins EGL-1 and CED-13 (ref. 36,37). In agreement with previous results 35 , cep-1 knockdown completely protected C. elegans from DNA damage-induced germ cell apoptosis (Fig. 5a). Strikingly, worms treated with RNA interference (RNAi)-mediated knockdown of cyld-1 showed significantly reduced number of apoptotic cells following ionizing radiation (IR)-induced DNA damage (Fig. 5a). Likewise, a cyld-1(tm3768) mutant strain, which harbours a deletion of 496 bp in exon 14 that is predicted to result in the expression of truncated catalytically inactive CYLD-1 ( Supplementary Fig. 5), showed reduced IR-induced germ cell death compared with wild-type worms (Fig. 5b).

AOM injection
Tumour evaluation Indeed, cyld-1(tm3768);gld-1(op236) double-mutant worms displayed an intermediate phenotype between cyld-1(tm3768) and gld-1(op236) single mutants, demonstrating that increased levels of CEP-1/p53 partly restored the sensitivity of cyld-1 mutant worms to DNA damage-induced apoptosis (Fig. 5c). In addition, these findings showed that loss of CYLD-1 catalytic activity partly inhibited the increased sensitivity of gld-1(op236) worms to apoptosis, providing additional evidence that CYLD-1 acts at the level of CEP-1/p53 to control the DNA damage response in C. elegans. Together, these results show that nematode CYLD-1 is required for the efficient activation of CEP-1/p53-dependent responses to DNA damage. Since the NF-kB signalling pathway is absent in C. elegans 39 , these results also provide evidence that the role of CYLD in regulating DNA damage-induced p53 responses is not dependent on its function as a negative regulator of NF-kB activation. In agreement with this, lack of CYLD catalytic activity did not considerably affect the DNA damage-induced IkBa degradation and the transcription of NF-kB target genes in mouse primary epithelial cells, supporting an NF-kBindependent function of CYLD in the regulation of the DNA damage response (Supplementary Fig. 6). enzyme, we hypothesized that CYLD could regulate p53 activation by deubiquitination. We first tested whether CYLD and p53 interact with each other in cells exposed to DNA damage. We found that HA-CYLD expressed in HEK-293 T cells co-immunoprecipitated with endogenous p53 or overexpressed GFP-p53 in reciprocal IPs and that this interaction was enhanced after DNA damage (Fig. 6a,b). Furthermore, GST pull-down assays showed that recombinant His-CYLD bound recombinant GST-p53 but not GST, suggesting that CYLD directly interacts with p53 (Fig. 6c). In addition, endogenous CYLD co-immunoprecipitated with endogenous p53 in primary epidermal keratinocytes and HCT116 cells in response to DNA damage (Fig. 6d,e). Together, these results showed that CYLD interacts with p53 and this interaction is enhanced by genotoxic stress.
CYLD deubiquitinates p53 facilitating its stabilization. To test whether CYLD expression could alter p53 ubiquitination, we first used HEK-293 T cells. Expression of p53 together with HA-Ub in HEK-293 T cells resulted in robust ubiquitination of p53 also in the absence of DNA damage (Fig. 7a). Consistent with a role of CYLD in negatively regulating p53 ubiquitination, co-expression of HA-CYLD strongly reduced the ubiquitination of p53 ( Fig. 7a and Supplementary Fig. 7a,b). In addition, overexpression of HA-CYLD diminished the ubiquitination and increased the stabilization of endogenous p53 in CpT-treated HCT116 cells (Fig. 7b,c). To assess whether CYLD DUB activity is required for the removal of ubiquitin chains from p53, we examined whether expression of two catalytically inactive CYLD mutants (the C-terminal truncated mutant R936X that is the human equivalent to the mouse R932X mutation and the CYLDH871N mutant where the putative catalytic Histidine is changed to Arginine abolishing DUB activity 8 ) affected p53 ubiquitination. As shown in Fig. 7d and Supplementary Fig. 7c,d, in contrast to WT CYLD, the catalytically inactive CYLDR936X (R/X) and CYLDH871N (H/N) mutants did not diminish p53 ubiquitination although they bound p53, demonstrating that CYLD DUB activity is required to reduce p53 ubiquitination. Furthermore, recombinant His-CYLD reduced ubiquitination of Flag-p53 immunoprecipitated from CpT-treated HEK293T cells, showing that CYLD can directly deubiquitinate p53 in a cell-free in vitro assay ( Supplementary Fig.7e). Together, these results suggested that CYLD directly interacts with and deubiquitinates p53 facilitating its optimal stabilization in response to DNA damage.
CYLD removes complex ubiquitin chains from p53. It is generally believed that ubiquitination of p53 with K48-linked chains controls its constant degradation in unstressed cells, while inhibition of the formation as well as DUB-mediated removal of K48 chains is essential for p53 stabilization in response to DNA damage 24 . CYLD selectively hydrolyses K63-and M1-linked ubiquitin chains but exhibits very little activity towards K48linked ubiquitin chains 6 . We therefore wondered whether CYLD could remove K48-linked ubiquitin chains from p53. To address this question, we first assessed the capacity of CYLD to reduce p53 ubiquitination in cells overexpressing HA-tagged Lys-to-Arg ubiquitin mutants that can only form K48-or K63-linked chains. This experiment showed that CYLD diminishes p53 ubiquitination in cells expressing HA-mutant ubiquitin forming K63 only chains, but also in cells expressing HA-mutant ubiquitin forming K48 only chains (Fig. 7e). As an alternative approach, we used K63 or K48 linkage-specific antibodies to assess the type of chains CYLD removes from p53. As shown in Fig. 7f, wild type but not catalytically inactive CYLD diminished both K63-and K48-linked ubiquitin chains from p53 in cells overexpressing wild-type ubiquitin (Fig. 7f). Together, these results provided evidence that CYLD removes also K48-linked ubiquitin chains from p53. To understand how CYLD, a DUB that efficiently hydrolyses K63 and linear chains but shows little activity against K48-linked ubiquitin chains, removes K48 chains from p53 we performed ubiquitin chain restriction (UbiCRest) analysis 40 on ubiquitinated Flag-p53 (Flag-p53(Ub)n) immunoprecipitated from MG132/ CpT-treated HEK-293 T cells. This experiment showed that USP21 (unspecific DUB) completely removed ubiquitin chains from p53, while OTUB1 (K48 linkage-specific DUB) and AMSH (K63-linkage-specific DUB), but not OTULIN (M1-linkagespecific DUB), strongly reduced p53 ubiquitination (Fig. 7g), suggesting that p53 is ubiquitinated with K48 and K63 but not ARTICLE linear chains. Interestingly, immunoblot analysis with K48 linkage-specific antibodies showed that USP21 and OTUB1 completely removed, but also AMSH strongly diminished K48linked ubiquitin chains from p53 (Fig. 7g). Since AMSH is highly specific for K63 chains and cannot hydrolyse K48 linkages, its effect on removing K48 chains from p53 is most likely indirect, either by hydrolysing K63 linkages in a heterotypic K63/K48 ubiquitin chain background, or by cleaving intact K48 chains at the substrate-proximal ubiquitin. CYLD behaved similarly to AMSH in these experiments, suggesting that CYLD most likely also removes K48 chains from p53 indirectly.

Discussion
Genomics studies in human cancers as well as experiments in mouse models have established CYLD as an important tumour suppressor in a variety of malignancies 1-5, [16][17][18][19] . Most studies so far attributed the tumour suppressing properties of CYLD to its function as negative regulator of NF-kB, MAPK and Wnt signalling. Nevertheless, the mechanisms of CYLD-mediated tumour suppression remain poorly understood. CYLD was recently shown to regulate death receptor-induced apoptosis and necroptosis, suggesting that regulation of death receptormediated programmed cell death could also contribute to its tumour suppressing functions. However, our in vivo genetic studies showing that double FADD/RIPK3 deficiency did not sensitize mice to AOM-induced colon tumorigenesis provided evidence that regulation of death receptor-induced programmed cell death is not critical for the tumour suppressing role of CYLD. Our results presented here revealed a novel function of CYLD as a regulator of p53 signalling. The role of CYLD in regulating p53 is evolutionarily conserved as shown by our findings that CYLD deficiency impairs p53-dependent DNA damage-induced germ cell apoptosis in C. elegans. Since the role of ubiquitination (e,f) HEK-293 T cells were transfected with the indicated constructs and treated with MG-132/CpT. p53 was immunoprecipitated and its levels and type of ubiquitination were analysed by immunoblot using the indicated antibodies. (g) Ubiquitinated Flag-p53 (Flag-p53(Ub)n) was immunoprecipitated from MG-132/CpT-treated HEK-293 T cells and subjected to an in vitro deubiquitination assay using the indicated DUBs. Total and K48-linked ubiquitin chains on p53 are shown by immunoblot using specific antibodies.
in the regulation of CEP-1/p53 signalling in C. elegans remains unclear, the precise mechanism by which CYLD controls DNA damage-induced CEP-1/p53 activation in worms remain to be investigated. Loss of CYLD catalytic activity reduced, but did not completely abolish, DNA damage-induced stabilization and activation of p53 in primary epithelial cells, suggesting that CYLD facilitates optimal p53 activation. This modulatory function of CYLD explains why CYLD-deficient mice do not phenocopy the severe phenotype of p53 knockout animals.
The stabilization and activation of p53 in response to cellular stress is regulated by the combined action of a number of E3 ubiquitin ligases and deubiquitinases 23,24 . p53 stability is controlled by K48 ubiquitin chains that target p53 for proteasome-dependent degradation. In addition, Ubc13 has been shown to elicit K63-dependent ubiquitination of p53 regulating its transcriptional activation 41 . Until recently, research on ubiquitin signalling has focused mainly on homogenous ubiquitin chains. However, there is increasing evidence from both in vivo and in vitro studies that single polyUb chains can contain multiple linkage types sequentially or in branched structures 42,43 . The properties of individual linkage types as well as their ability to be hydrolysed by DUBs remain intact when chains are branched 43,44 . Moreover, branched chains were shown to be processed faster by the 26S proteasome compared with homogeneous chains 42,44 . To our knowledge, the presence and possible functional role of complex ubiquitin chains on p53 has not been studied thus far. Our results using mutant ubiquitins and chain-specific ubiquitin antibodies showed that CYLD could remove not only K63 but also K48 chains from p53. Although we cannot exclude that CYLD could also hydrolyse directly K48 chains on p53, this is unlikely as previous experiments showed that CYLD efficiently hydrolyses K63 and linear chains but exhibits very little activity against K48 chains 45 . The results of the UbiCRest experiment showing that AMSH, a DUB that is highly specific for K63 chains could also remove K48 chains from p53 similarly to CYLD, suggest that these enzymes remove K48 chains indirectly. The most likely explanation of these findings is that p53 is decorated with mixed and/or branched K63/K48 ubiquitin chains and that CYLD can remove K48 chains indirectly by catalysing the cleavage of K63 linkages. We did not directly address the specific role of the proteasome in mediating CYLDinduced degradation of p53, as any experiment involving proteasome inhibition would result in p53 stabilization and thus would be inconclusive in proving the specific role of CYLD. However, given that CYLD mutation did not alter p53 mRNA levels ( Supplementary Fig. 8), it is reasonable to conclude that CYLD stabilizes p53 by preventing its K48 chain-dependent proteasomal degradation. Collectively, these results provide a rational mechanism explaining how CYLD can regulate p53 protein stability directly, although it remains unclear whether CYLD targets additional proteins in the p53 pathway. Moreover, these findings provide experimental evidence that p53 is decorated with mixed and/or branched ubiquitin chains containing more than one linkages, suggesting that complex ubiquitin chains play an important role in regulating p53 signalling.
Our findings that CYLD catalytic activity is required for efficient DNA damage-induced activation of p53 suggest that regulation of p53 signalling is a major tumour suppressing mechanism of CYLD that is likely to synergize with its previously reported functions in negatively regulating NF-kB and Wnt signalling. By regulating these three pathways CYLD can affect three independent but interconnected processes that are central for tumorigenesis: genomic stability, inflammation and cell growth. The combined function of CYLD in modulating the activity of these pathways provides a rational mechanism for its important tumour suppressor role in a variety of human malignancies. In conclusion, our results reveal a novel tumour suppressor function of CYLD as a regulator of p53-dependent cellular responses to DNA damage.

Methods
Mice. CYLDD932 FL (ref. 15), Rip3k À / À (ref. 46 50 have been previously described. For all experiments 8-10-week old gender-matched mice were used. Littermates carrying loxP-flanked alleles but not Cre served as control mice. Animals requiring veterinary attention were provided with appropriate care and excluded from the experiments. Mice were maintained at the animal facilities of the Institute for Genetics, University of Cologne, kept under a 12 h light cycle, and given a regular chow diet (Harlan, diet #2918) ad libitum. All animal procedures were conducted in accordance with European, national and institutional guidelines and protocols and were approved by local government authorities (Landesamt für Natur, Umwelt und Verbraucherschutz Nordrhein-Westfalen, Germany).
Carcinogenesis protocols. Two per cent DSS (MP, MW 36000-50000) was provided in drinking water ad libitum. AOM (Sigma) was dissolved in sterile PBS and injected intraperitoneally at a dose of 7.5 or 10 mg kg À 1 body weight. Colitis and tumour formation was monitored by high-resolution mini endoscopy. For DMBA-induced skin cancer, mice were treated topically on the shaved back skin with 25 mg of DMBA (Sigma) in 100 ml acetone once a week during 8 weeks. Animals were sacrificed after 12 weeks of latency. The tumour size scoring was done as follows; small: Z1 mm; medium: Z3 mm; and big: Z5 mm diameter.
High-resolution mini endoscopy. Mice were anaesthetized using intraperitoneal injection of ketamine (Ratiopharm)/Rompun (Bayer) and a high-resolution miniendoscope, denoted Coloview (Karl-Storz, Tuttlingen, Germany), was used to determine the murine endoscopic index of colitis severity as described previously 51 .
Histology and immunohistochemistry. Samples were fixed overnight in 4% paraformaldehyde, embedded in paraffin and cut in 4 mm sections. Paraffin sections were rehydrated and heat-induced antigen retrieval was performed in 10 mM Sodium Citrate, 0.05% Tween-20 pH 6. Primary antibody used for IHC was antiactive Caspase 3 (Cell Signalling, 9661 (1:1,000 dilution). Biotinylated secondary antibodies were purchased from Perkin Elmer and Dako. Stainings were visualized with ABC Kit Vectastain Elite (Vector) and DAB substrate (DAKO). Incubation times with the DAB substrate were equal for all samples. General cell death was evaluated using DeadEnd Fluorometric TUNEL System (Promega). Pictures were taken with a fluorescence microscope (Leica) at the same exposure and intensity settings.
Histopathological evaluation. Intraepithelial neoplasias (IEN) were classified according to their grading in low-or high-grade/carcinoma in situ, this classification was made according to the WHO standard guidelines. The cumulative tumour size was calculated by summing up the size of all tumours found per sample. Tumour grade and size were evaluated blindly by two independent experienced pathologists. Histological sections that were not optimal for a proper evaluation were discarded from the analysis.
Organoid cultures. Mice were killed and intestines were flushed with PBS and cut longitudinally before villi were mechanically removed. Intestinal crypts were collected in PBS containing 2 mM EDTA. Crypts were plated in matrigel (BD) and supplemented with minimal growth medium containing ENR: 100 ng ml À 1 mNoggin, 100 ng ml À 1 R-Spondin, 50 ng ml À 1 mEGF. ENR was added to the medium every 3-4 days. Outgrowing crypts were passaged once a week. Organoids were stimulated in culture with CpT 10 mM (Sigma) or MMC 10 mg ml À 1 (Sigma) for the indicated times. For protein and RNA extraction, organoids were harvested with PBS, disrupted by pipetting and washed again with PBS in order to discard the organoid's lumen content.
Keratinocyte isolation and culture. Primary epidermis keratinocytes were isolated from the skin of mice between days 0 and 3 and cultured in minimal calcium medium (0.05 mM CaCl) supplemented with 4% Chelex-treated fetal calf serum and epidermal growth factor (10 ng ml À 1 ). Keratinocytes were stimulated in culture with CpT 10 mM (Sigma) for the indicated times. Keratinocytes were lysed with RIPA lysis buffer containing proteases and phosphatase inhibitor tablets (Roche) for protein extraction and in Trizol reagent (Invitrogen) for RNA extraction.
Cell death assays. Three days after splitting, intestinal organoids were treated with CpT (Sigma) 10 mM for the indicated time points. To get single-cell suspension, organoids were harvested with Tryple solution (Life Technologies), passed four times through a syringe (23G needle) and incubated at 37°for 5 min. Digestion was stopped with 2%FCS/PBS solution. Cells were then incubated with PBS containing amine-reactive dye (LIVE/DEAD dye, Live Technologies) washed, fixed with 2% PFA and permeabilized with PBS containing 1%BSA, 0.1%TritonX-100 in PBS. Cells were then stained with anti-cleaved caspase-3 (Asp175) antibodies (Cell Signaling; 9661; 1:50 dilution), followed by secondary antibodies coupled to ALEXA Fluor 488 (Molecular Probes; 1:200 dilution). Cleaved caspase 3 only positive cells were analysed using a FACS Calibur (BD).
Pull-down assay. GST fusion proteins were purified from the Escherichia coli BL-21 strain in lysis buffer using glutathione-Sepharose (Amersham Bioscience). Sepharose beads with GST fusion proteins where incubated with the indicated HEK-293 T-transfected cell lysate in binding buffer O/N at 4°C. Beads were washed with binding buffer and proteins were eluted with LDS reducing sample buffer (Bio-Rad) by heating at 70°C.
Ubiquitin chain-restriction analysis on p53. Ubiquitin chain restriction analysis was performed as previously described 40,54 . Briefly, ubiquitinated p53 was immunoprecipitated from transfected HEK-293 T cells, as described above, and incubated with the different deubiquitinases (DUBs) (USP21, 1. RNA interference treatment in C. elegans. Overnight cultures of HT115 bacteria containing specific RNAi constructs were grown in lysogeny broth media containing ampicillin. RNAi expression was induced by adding 1 mM Isopropyl b-D-1-thiogalactopyranoside (IPTG) and incubating the cultures at 37°C for 20 min before seeding the bacteria on NGM agar supplemented with ampicillin and 3 mM IPTG. Worms were used for experiments after 2 generations of RNAi feeding.
Irradiation experiments in C. elegans. For ionizing radiation experiments, synchronized populations of L1 worms were obtained by passing mixed-stage cultures through 11-micrometre filter. Worms were then grown at 20°C on Petri dishes containing NGM agar supplemented with OP50 E. coli. 48 h later late L4 worms were picked on new plates and irradiated using ISOVOLT, Titan E machine (GE) in combination with a 0.5 mm aluminium filter. Germ cell corpses were scored blindy as described 56 . Quantification of corpses formation upon irradiation is shown as mean of germ cell corpses/gonad (n ¼ 20-25)±s.d.
Mutational analysis of human tumour samples. To determine the CYLD mutation frequency across different cancer entities, sequencing files from 7,042 human tumour samples 31 (Sanger Institute Server) were analysed. We used the Human Genome HG19 annotation to link each mutation to its respective gene. Samples with low tumour content or weak sequencing criteria and mutations that were annotated as non-somatic were excluded from further analysis. A sample was classified as Cyld mutant, if it carried at least one somatic mutation in the coding region of the Cyld gene. We analysed primary tumour sites, for which at least two independent Cyld mutant samples were available (recurrent mutation).
To characterize the impact of CYLD mutations on the mutation spectrum size, we counted the number of mutations per sample between Cyld mutant and wild type and compared the average mutation count by using Student's t-test. For all primary tumour sites, which contained recurrent Cyld mutations, at least 175 samples were examined. Therefore samples were large enough to fulfil the prerequisites of Student's t-test as well as in order to estimate the variation within each group of data. In all groups compared by t-test variance was similar enough to apply a homoscedastic testing.
Further, we compared the mutation types in CYLD with the mutation spectra of known tumour suppressors and oncogenes. We determined the frequency of inactivating mutations in 27,836 genes across the entire COSMIC database. For each gene, we tested whether inactivating mutations were significantly enriched or underrepresented in its mutation spectrum, assuming a uniform distribution of inactivating mutations (w 2 -test). Significance values were corrected for multiple testing (Benjamini-Hochberg correction).
Statistical analysis. For the animal experiments in order to determine group size necessary for adequate statistical power, power analysis employing the program G*power 57 was performed using preliminary data sets. Mice of the indicated genotype were assigned at random to groups. Mouse studies were performed in a blinded fashion. Results are shown as mean ± s.d. or ± s.e.m. Statistical significance was determined with the Student's t-test; *Pr0.05, ** Pr0.005, ***Pr0.0005. Groups were large enough to fulfil the prerequisites of Student's t-test as well as to determine that the variance between groups is similar.
Reproducibility of experiments. For the animal experiments; experiments presented in Figs 1b-d,f-h and 2b were repeated twice obtaining similar results. Results shown in Fig. 2f,j correspond to pooled data from 3 independent experiments. Results shown in Fig. 4d, Supplementary Fig. 2b-d and Supplementary Fig. 2f correspond to pooled data from two independent experiments. The histological pictures shown in Figs 1i and 2c,g,k are representative for 36, 17, 30, 46 animals in total, respectively.
Immunofluorescence images presented in Fig. 3a are representative for six animals in total. Microscopy images presented in Supplementary Fig. 3 are representative for two different experiments.
FACS quantification of cleaved caspase-3 experiments shown in Figs 3c and 4b were repeated three times with technical duplicates when possible, obtaining similar results. qPCR analysis shown in Fig. 3f,h are representative from two independent repetitions using different primary cell isolations and technical duplicates. In qPCR analysis in Fig. 4c biological triplicates were used. Results shown in Supplementary  Fig. 6a are representative from two independent repetitions.
Western blot shown in Fig. 3b,e,g and Fig. 4a were repeated at least three times. Interaction experiments shown in Fig. 6a were repeated four times; and experiments shown in Fig. 6b-e were repeated twice obtaining similar results.
Ubiquitination experiments shown in Fig. 7a,d were repeated four times. Experiments presented in Fig. 7e,f were reproduced twice. The ubiquitin chain restriction analysis shown in Fig. 7g was reproduced three times obtaining comparable results.
For the C. elegans data; experiments shown in Fig. 5a were repeated twice and experiments presented in Fig. 5b,c were repeated three times obtaining comparable results.
Data availability. The mutational analysis of human tumour samples presented in Supplementary Fig. 4 was performed on the sequencing data reported by Alexandrov et al. 31 , which is available from ftp://ftp.sanger.ac.uk/pub/cancer/ AlexandrovEtAl. The authors declare that all other data supporting the findings of this study are available within the article and its Supplementary Information files or from the corresponding author upon reasonable request.