LUBAC prevents lethal dermatitis by inhibiting cell death induced by TNF, TRAIL and CD95L

The linear ubiquitin chain assembly complex (LUBAC), composed of HOIP, HOIL-1 and SHARPIN, is required for optimal TNF-mediated gene activation and to prevent cell death induced by TNF. Here, we demonstrate that keratinocyte-specific deletion of HOIP or HOIL-1 (E-KO) results in severe dermatitis causing postnatal lethality. We provide genetic and pharmacological evidence that the postnatal lethal dermatitis in HoipE-KO and Hoil-1E-KO mice is caused by TNFR1-induced, caspase-8-mediated apoptosis that occurs independently of the kinase activity of RIPK1. In the absence of TNFR1, however, dermatitis develops in adulthood, triggered by RIPK1-kinase-activity-dependent apoptosis and necroptosis. Strikingly, TRAIL or CD95L can redundantly induce this disease-causing cell death, as combined loss of their respective receptors is required to prevent TNFR1-independent dermatitis. These findings may have implications for the treatment of patients with mutations that perturb linear ubiquitination and potentially also for patients with inflammation-associated disorders that are refractory to inhibition of TNF alone.

A proper balance between cell death, proliferation and differentiation maintains homeostasis of the skin that is critical to retain this organ's vital role as an immunological barrier against pathogens and as a physical barrier to water loss and mechanical insults. If any of these processes are deregulated, pathological conditions, such as skin infections, autoinflammatory and autoimmune disorders or cancer can occur 1 . Members of the tumour necrosis factor (TNF) and TNF receptor (TNFR) superfamilies are essential mediators of cell death and inflammation and play critical roles in innate as well as adaptive immune responses in many tissues and cell types, including in the skin and epidermal keratinocytes 2 .
Engagement of TNFR1 by TNF induces formation of the TNFR1 signalling complex (TNFR1-SC), also referred to as complex I of TNFR1 signalling, an event that triggers gene activation by nuclear factor (NF)-κB and transcription factors activated downstream of mitogen-activated protein kinases 3,4 . TNFR1 signalling can, however, also result in cell death. This is triggered by a secondary cytoplasmic complex, also called complex II, which is formed by the recruitment of Fas-associated protein with a death domain (FADD) and caspase-8 to receptorinteracting serine/threonine-protein kinase 1 (RIPK1). In this platform, caspase-8 is cleaved and thereby activated, inducing apoptotic cell death 5,6 . Alternatively, when caspase-8 is inhibited or either FADD or caspase-8 is absent, RIPK1 recruits RIPK3, which in turn activates mixed lineage kinase domain-like protein (MLKL), resulting in the induction of regulated necrosis, also referred to as necroptosis 7,8 . However, complex II formation and activity is minimised when complex I is properly assembled and activated [9][10][11][12][13][14] .
The linear ubiquitin chain assembly complex (LUBAC) regulates the balance between gene activation and cell death upon engagement of TNFR1 and certain other innate and adaptive immune receptors including Toll-like receptors (TLRs), TNFrelated apoptosis-inducing ligand (TRAIL), NOD-like receptors and T and B cell receptors [15][16][17][18][19][20] . LUBAC, composed of three proteins, Heme-oxidized IRP2 ubiquitin ligase 1 (HOIL-1), Shank-associated RH domain-interacting protein (SHARPIN) and HOIL-1-interacting protein (HOIP), is the only E3 ligase identified so far capable of generating linear ubiquitin linkages de novo [21][22][23][24][25] . We previously showed that LUBAC prevents complex II formation upon TNFR1 stimulation, thereby inhibiting TNFR1-mediated cell death [26][27][28] . Mice deficient for SHAR-PIN, known as chronic proliferative dermatitis mice (cpdm) and referred to as Sharpin cpdm/cpdm mice herein, suffer from severe inflammation in the skin and other organs [29][30][31] , which is caused by excessive TNFR1-mediated death of keratinocytes 22,27,32 . In contrast, deficiency in HOIP or HOIL-1 results in embryonic lethality 26,28,33 . The differences in the phenotypes of mice deficient for the different LUBAC components is due to the fact that in the absence of HOIP or HOIL-1 there is a complete lack of linear ubiquitination in complex I, whereas in the absence of SHARPIN it is merely reduced 28 . Thus, whereas HOIP and HOIL-1 are both essential for LUBAC activity, SHARPIN only contributes to it.
To explore the role of HOIP and HOIL-1 in the control of epidermal cell death and skin homeostasis, we sought to investigate the effect of deleting them in keratinocytes. Surprisingly, we found that keratinocyte-specific deletion of HOIP or HOIL-1 results in a lethal inflammatory skin disease, which is only partially dependent on TNFR1-induced cell death. The TNFR1independent dermatitis is also a consequence of cell death that can, intriguingly, be redundantly triggered by TRAIL or CD95L. These findings identify a vital and previously unrecognised physiological role of HOIP and HOIL-1 in preventing cell death-induced inflammation, importantly beyond TNF as the only endogenous inducer of this cell death.

Results
HOIP and HOIL-1 are essential to maintain skin homeostasis. To understand the role of LUBAC in the skin, we generated mice that lack HOIP or HOIL-1 selectively in epidermal keratinocytes by crossing Hoip-and Hoil-1-floxed mice with mice expressing the Cre recombinase under the control of the human keratin 14 (K14) promoter. The genotype of the mice was confirmed by PCR ( Supplementary Fig. 1a). At the protein level, deletion of HOIP or HOIL-1 in keratinocytes was verified by western blot and immunohistochemistry ( Supplementary Fig. 1b, c). As expected, HOIP deficiency abrogated linear ubiquitination at the TNFR1-SC ( Supplementary Fig. 1d) and reduced TNFR1-mediated NF-κB activation in primary murine keratinocytes (PMKs) without preventing it ( Supplementary Fig. 1e). Mice homozygous for keratinocyte-specific deletion of HOIP or HOIL-1 (Hoip E-KO and Hoil-1 E-KO mice, respectively) were born at the expected Mendelian frequencies and were macroscopically indistinguishable from littermates up to postnatal day (P) 2 (data not shown). From this day onwards, however, both Hoip E-KO and Hoil-1 E-KO mice developed severely damaged and scaly skin, which, invariably, resulted in the death of these mice between P4 and P6 (Fig. 1a). No Hoip or Hoil-1 gene dosage effect was observed as Hoip fl/wt K14Cre + and Hoil-1 fl/wt K14Cre + mice developed normally into adulthood without showing any signs of skin disease (data not shown).
Histological analysis of Hoip E-KO and Hoil-1 E-KO mice at P4 revealed increased epidermal thickness, parakeratosis, hyperkeratosis and keratinocyte differentiation defects (Fig. 1b, c). These pathologies were accompanied by abnormal myeloid cell infiltration and high levels of cell death as demonstrated by increased cleaved caspase-3 and terminal deoxinucleotidyl transferasemediated dUTP-fluorescein nick end labelling (TUNEL) staining ( Fig. 1b, d, e and Supplementary Fig. 1f, g). Together, these observations reveal that HOIP and HOIL-1 are essential to prevent fatal dermatitis characterised by disruption of the normal epidermal structure, inflammation and aberrant keratinocyte death.
Lethal dermatitis is only partially mediated by TNFR1. The inflammatory syndrome of Sharpin cpdm/cpdm mice can be abrogated by the absence of TNF and TNFR1 or by the loss of the kinase activity of RIPK1 22,27,32,34 . We therefore first tested whether genetic ablation of TNFR1 could also prevent the morbidity and mortality in Hoip E-KO and Hoil-1 E-KO mice. Unexpectedly, however, inflammation was only delayed in Tnfr1 KO ;Hoip E-KO and Tnfr1 KO ;Hoil-1 E-KO mice as they progressively developed severe skin lesions resulting in a median survival of 70 days ( Fig. 2a and Supplementary Fig. 2a). Sick Tnfr1 KO ;Hoip E-KO and Tnfr1 KO ;Hoil-1 E-KO mice presented with epidermal disruption and thickening, parakeratosis, hyperkeratosis and inflammation (Fig. 2b, c). Crucially, infiltration by myeloid and lymphoid cells and cell death were significantly augmented in the epidermis of adult Tnfr1 KO ;Hoip E-KO mice compared to control animals (Fig. 2b, d and Supplementary Fig. 2b-d).
Next, we addressed whether ablation of the kinase activity of RIPK1 was sufficient to prevent inflammation in Hoip E-KO mice. Surprisingly, genetic ablation of the kinase activity of RIPK1 was substantially less effective than loss of TNFR1 in preventing dermatitis as Ripk1 D138N ;Hoip E-KO mice died at around P8 from severe skin disease (Fig. 2f, g). Thus lethal dermatitis caused by keratinocyte-specific deficiency in either HOIP or HOIL-1, the two essential components of LUBAC 28 , is only partially dependent on TNFR1 and, in the presence of TNFR1, almost completely independent of the kinase activity of RIPK1.
Increased cell death precedes inflammation. We next investigated the temporal relationship between aberrant cell death and inflammation in Hoip E-KO and Hoil-1 E-KO mice. Abnormally increased cell death in the epidermis of Hoip E-KO and Hoil-1 E-KO mice was already apparent in utero at E18.5 and at birth (P0) (Fig. 3a, b and Supplementary Fig. 3a, b). This implies that lack of linear ubiquitination in keratinocytes results in aberrant cell death in sterile conditions. Hoip E-KO and Hoil-1 E-KO mice displayed abnormally increased immune cell infiltration at P2 but not at birth (Fig. 3a, c and Supplementary Fig. 3c-e). Accordingly, keratinocyte differentiation and epidermal thickness appeared abnormal at P2 but not at E18.5 or P0 ( Fig. 3a and Supplementary  Fig. 3f). Thus excessive cell death precedes the inflammatory response suggesting that keratinocyte death upon loss of HOIP or HOIL-1 may trigger lethal dermatitis.
To understand the mechanism of cell death induction in the skin of Hoip E-KO and Hoil-1 E-KO mice, we analysed the formation of the signalling platform known to trigger cell death downstream of various death receptors 35 by immunoprecipitating the adaptor protein FADD in PMKs in the presence of the caspase inhibitor, Z-VAD-fmk. This revealed that, even without an exogenous stimulus, a FADD/caspase-8/RIPK1-containing complex was readily detectable in HOIP-deficient but not in control PMKs (Fig. 4a). Consistent with apoptotic signalling by such a complex, the HOIP-deficient cells were less viable even in the absence of exogenous stimuli (Fig. 4b). This loss in cell viability was significantly reduced by inhibition of caspases and RIPK1 kinase activity but not by blocking the kinase activity of RIPK3 (Fig. 4b).
Inhibition of TNF or genetic ablation of TNFR1 also increased the viability of Hoip E-KO PMKs (Fig. 4c, d). These results indicate that in PMKs HOIP prevents aberrant RIPK1 kinase-dependent apoptosis triggered by spontaneously produced autocrine TNF.

LUBAC loss in adulthood induces cell death-driven dermatitis.
To assess the impact of acute loss of HOIP in keratinocytes in adult mice, we treated Hoip fl/fl K14CreER Tam mice with 4-hydroxytamoxifen (4-OHT) in a localised area of the skin. This treatment resulted in rapid cell death induction, followed by increased immune cell infiltration (Fig. 5a-c and Supplementary  Fig. 4a). This was accompanied by epidermal thickening, hyperplasia, hyperkeratosis as well as parakeratosis and defects in keratinocyte differentiation ( Fig. 5d and Supplementary Fig. 4b-e). These findings demonstrate that HOIP is required to maintain normal skin architecture and function in adult mice and that also Aberrant apoptosis is responsible for lethal dermatitis. Since aberrantly increased cell death was the first abnormal event we could detect in the epidermis of Hoip E-KO and Hoil-1 E-KO mice and because it is also observed in the absence of TNFR1 in adult mice, we next evaluated genetically whether and, if so, which form(s) of aberrant cell death cause the early and the late dermatitis.
Consistent with the apoptotic cell death observed in vitro, genetic ablation of Ripk3 in Hoil-1 E-KO or the loss of Mlkl in Hoip E-KO mice failed to prevent aberrant cell death and skin inflammation and did not delay the postnatal lethality (Fig. 6b, c, e and Supplementary Fig. 5).
Collectively, these results demonstrate that caspase-8-mediated apoptosis is responsible for the lethal dermatitis in mice lacking HOIP or HOIL-1 in keratinocytes and that RIPK3/MLKL-mediated necroptosis does not contribute to the disease.
RIPK1 kinase is required for TNFR1-independent dermatitis.
Since aberrant cell death is the cause of dermatitis in Tnfr1 KO ; Hoip E-KO and Tnfr1 KO ;Hoil-1 E-KO mice, we aimed to further characterise this type of cell death. In order to evaluate the contribution of necroptosis to the pathology of Tnfr1 KO ;Hoip E-KO mice, we generated Mlkl KO ;Tnfr1 KO ;Hoip E-KO mice. These mice developed less severe skin lesions at day 70 and their survival was significantly prolonged (Fig. 7a-c). Thus, in the absence of TNFR1, necroptosis contributes to skin inflammation.
To investigate the involvement of the kinase activity of RIPK1 in the TNFR1-independent disease caused by LUBAC deficiency in the skin, we next fed Tnfr1 KO ;Hoil-1 E-KO and control mice from E14.5 until day 100 with the RIPK1 inhibitor GSK'547A 41 . As a control, Sharpin cpdm/cpdm and Hoip E-KO mice were also fed with GSK'547A. In line with the genetic analysis of Hoip E-KO mice (Fig. 2f, g) and previous reports with Sharpin cpdm/cpdm mice 34 , this treatment delayed disease development and death of Hoip E-KO mice up to P8 and prevented dermatitis in Sharpin cpdm/ cpdm mice ( Supplementary Fig. 7a, b). Strikingly, pharmacologic inhibition of the kinase activity of RIPK1 rescued the majority of Tnfr1 KO ;Hoil-1 E-KO mice from dermatitis for the duration of the treatment with only three of the ten treated mice developing small punctate scales (Fig. 7c-e and Supplementary Fig. 7c). Thus the lethal dermatitis caused by keratinocyte-specific HOIL-1 that occurs in the absence of TNFR1 is mediated by RIPK1 kinasedependent apoptosis and necroptosis.
TNF, TRAIL and CD95L drive cell death and dermatitis. Finally, we sought to identify the instigator(s) of the TNFR1independent cell death that is responsible for the fatal dermatitis in Hoip E-KO and Hoil-1 E-KO mice. Consistent with our previous findings in other cell types 17,42 , PMKs derived from Tnfr1 KO ; Hoil-1 E-KO mice were more sensitive than control cells to induction of cell death by TRAIL, CD95 (Fas/APO-1) ligand (CD95L) or Polyinosinic:polycytidylic acid (Poly(I:C)) (Fig. 8a). We, therefore genetically ablated the death domain (DD) of CD95 specifically in keratinocytes or deleted TRAIL-R or TLR3 systemically in Tnfr1 KO ;Hoil-1 E-KO mice. However, Cd95 E-DD ;Tnfr1 KO ;Hoil-1 E-KO , Trail-r KO ;Tnfr1 KO ;Hoil-1 E-KO and Tlr3 KO ;Tnfr1 KO ;Hoil-1 E-KO mice all suffered from skin lesions that were indistinguishable in severity from those seen in the Tnfr1 KO ;Hoil-1 E-KO mice ( Supplementary Fig. 8).
Reasoning that these death receptors might be able to drive disease in a redundant manner in Tnfr1 KO ;Hoil-1 E-KO mice, we next assessed the impact of their combined loss on the onset of dermatitis. Co-deletion of TRAIL-R and TLR3 in Tnfr1 KO ;Hoil-1 E-KO mice resulted in slightly milder skin lesions at day 70 (Fig. 8b, c), yet these mice still succumbed to inflammatory skin disease with a median survival of 77 days (Fig. 8d). Strikingly, however, combined loss of TRAIL-R with keratinocyte-specific deletion of the DD of CD95 resulted in prevention of dermatitis in Tnfr1 KO ;Hoil-1 E-KO mice reflected by the significantly reduced severity score of skin lesions and prolonged survival (Fig. 8b-d). We therefore conclude that the cell death-driven inflammatory disease caused by HOIL-1 deficiency in keratinocytes is induced by multiple death receptor-ligand systems, namely the TNF/ TNFR1, TRAIL/TRAIL-R and CD95L/CD95 systems. Thus, cell death induction via either of these receptors is sufficient to cause disease ( Supplementary Fig. 9) in the absence of linear ubiquitination caused by HOIL-1 deficiency.

Discussion
Our study shows that both HOIP and HOIL-1 are essential to maintain skin homeostasis by preventing skin inflammation caused by death receptor-induced cell death. Intriguingly, the dermatitis in Hoip E-KO and Hoil-1 E-KO mice is different from that in Sharpin cpdm/cpdm mice, both with regards to severity and the mechanisms responsible for driving it. Whereas the dermatitis in Hoip E-KO and Hoil-1 E-KO mice is driven by cell death induced by mechanisms beyond TNF/TNFR1, deletion of one copy of the Tnf gene completely prevented dermatitis in Sharpin cpdm/cpdm mice 22 . This uncovers a physiological role for HOIP and HOIL-1, which is more complex regarding LUBAC function than the one we and others previously identified for SHARPIN 22,27,32 . Crucially, this additional complexity relies on cell death-inducing systems other than the TNF/TNFR1 system. Curiously, while inhibition of the kinase activity of RIPK1 restores viability of HOIP-deficient PMKs in vitro, neither pharmacologic inhibition nor genetic impairment of RIPK1's kinase activity in Hoip E-KO mice prevented keratinocyte death and consequent dermatitis. It therefore appears that the regulation of cell death in vivo is more complex than revealed by the study of PMKs ex vivo. It is interesting to note in this context that apoptosis dependent on the kinase activity of RIPK1 was also described to occur upon inhibition or absence of TAK1, the NF-κB essential modulator (NEMO) or IKKα/IKKβ following TNF stimulation 11,43,44 .
We and others previously showed that TNF-induced cell death can cause inflammation and inflammation-associated diseases 22,45 . The results we present here provide additional evidence in support of cell death as an aetiology of inflammation-associated diseases. Importantly, however, we extend this concept to endogenous factors capable of inducing inflammatory cell death beyond TNF. Crucially, we discover that these factors can act in concert with TNF to induce inflammatory cell death. This means that patients with a disease aetiology similar to the one we report here may benefit from a therapy that combines the inhibition of TNF, or TNF-induced death, with that of the kinase activity of RIPK1 and/or of other death-inducing cytokines, most importantly CD95L and TRAIL. It is noteworthy that patients with LUBAC-inactivating germline mutations, such as those with HOIL-1, HOIP and OTULIN mutations, suffer from immunodeficiency and autoinflammation [46][47][48][49][50] . Intriguingly, whereas TNF-inhibitory treatment only temporarily ameliorated the pathology in one of the HOIL-1-deficient patients who received it, the benefit to the OTULIN-deficient patients identified so far was substantial 47,49,50 . Our observation that TNF is not the sole driver of cell death-driven inflammation and the suggestion of potentially effective combinatorial therapeutic options, based on simultaneous genetic interference with different death receptor-ligand systems in our model of LUBAC deficiency in the skin, may have implications for the future treatment of patients harbouring mutations that perturb normal LUBAC activity.
TNF inhibition is effective in treating several autoinflammatory and autoimmune disorders, including rheumatoid arthritis, psoriasis and Crohn's disease [51][52][53] . However, a significant fraction of patients with these diseases fails to respond to anti-TNF treatment. It is tempting to speculate that autoimmune patients with a cell death aetiology of their disease may benefit from combining the inhibition of TNF with that of RIPK1 and/or CD95L and TRAIL. This concept could possibly extend to patients with a cell death-driven autoimmune or otherwise inflammation-associated disease in which TNF inhibition, at least when applied alone, has so far failed, including amyotrophic lateral sclerosis and multiple sclerosis 54 56 , strain AZO-Nn4Cre (K14). The K14CreER Tam57 , Mlkl KO28 , Ripk3 KO58 , Casp8 KO59 , Trail-r KO60 and Ripk1 D138N61 mice have been previously described. Tnfr1 KO , Tlr3 KO , Cd95-DD fl/fl mice (C57BL/6-Fastm1Cgn/J) and Sharpin cpdm/cpdm (C57BL/Ka) mice were purchased from The Jackson Laboratories. To induce deletion of HOIP in epidermal keratinocytes of adult mice, Hoip fl/fl K14CreER Tam mice were treated as previously described 62 . Briefly, a small shaved area of the dorsal neck was treated with 50 μL of 4-OHT 20 mg mL −1 dissolved in ethanol every other day for a total of 1, 2, 3 or 4 treatments, as indicated. As a vehicle treatment, a small dorsal area close to the tail was shaved and treated with ethanol. Hoip fl/wt K14CreER Tam mice were used as tamoxifen controls. Mice were analyzed 2 days after the last treatment or as indicated in the figure legends. Timed matings were performed as previously described 26 . All mice were genotyped by PCR analysis. Colonies were fed ad libitum. All animal experiments were conducted under an appropriate UK project license in accordance with the regulations of UK home office for animal welfare according to ASPA (animal (scientific procedure) Act 1986).
Pharmacological inhibition of RIPK1 kinase activity. Pregnant females were fed with rodent chow containing 100 mg kg −1 day −1 GSK3540547A (GSK'547A) (GlaxoSmithKline LLC) from 14 days post coitum and continued the special diet throughout the nursing period. At weaning age, Sharpin cpdm/cpdm and Tnfr1 KO ; Hoil-1 E-KO mice and littermate controls were continuously treated with GSK'547A for another 100 days.
Immunostaining and quantification. Four-μm-thick formalin-fixed paraffinembedded skin sections were stained following standard protocols. Briefly, sections were boiled in 10 mM sodium citrate buffer (pH 6.0) in a microwave. Slides were blocked in buffer containing Tween 20 0.5% and bovine serum albumin 0.2%. For CD45 staining, slides were boiled in Retrievagen A (BD) and blocked with buffer without Tween. Next, slides were incubated with the primary antibody overnight at 4°C. The following antibodies were used Cambridge Bioscience) at room temperature for 1 h. Where an HRP-conjugated antibody was used, the TSA™ Plus Cyanine 3 System (Perkin Elmer) was applied according to the manufacturer's instructions. Sections were counterstained with 4,6-diamidino-2-phenylindole (DAPI; Roche). For HOIP and HOIL-1 staining, conventional immunohistochemistry (antibody dilution 1/100) was performed on BOND-III (Leica Microsystems) and BenchMark Ultra (Ventana-Roche Medical System) according to a protocol previously described 63 . For TUNEL staining, which was performed in combination with cleaved caspase-3 staining, the ApopTag Red In Situ Apoptosis Detection Kit (Merck Millipore) and streptavidin conjugate Cy2 (Jackson Immuno Research), respectively, were used according to the manufacturer's instructions. Sections were analyzed by fluorescent microscopy.
At least ten different images (×40) per slide were acquired. Quantification was performed by an experimenter who was blinded to the genotype of the samples by using the ImageJ Software on monochrome images as the percentage of cells positive for the specific staining in relation to the total number of cells (DAPIpositive) within the epidermis.
Epidermal thickness quantification. The epidermal thickness was measured in five different positions per microscopic field for at least ten different fields per mouse. Quantification was performed by an experimenter who was blinded to the genotype of the samples by using the ImageJ Software.
Dermatitis scoring criteria. Mice were assessed macroscopically based on two main clinical criteria. Each region of the body, comprising head, neck, back and flank, affected by lesions, was given a score of 1 and the sum of these provided information of how expanded the lesions were. The other criteria were the characteristics of the lesion: punctuated small crusts, coalescent crusts, and ulceration, with a score of 1-3, respectively. The sum of both criteria represented the total severity score of the lesions. Scoring was performed by two independent researchers. Flow cytometry. Cell suspensions obtained from skin samples were fluorescently labelled with Fixable Viability Dye eFluor® 780 (eBioscience). Samples were then stained with antibodies against the following cell surface markers: CD45-APC, CD45-AF700, CD3-PerCP/Cy5.5, CD4-FITC, CD8-PE/Cy7, GR1-FITC, GR1-PE/ Cy7, F4/80-PE, F4/80-BV786, CD11b-Percp/Cy5.5 (Biolegend), and CD19-BV650 (Biolegend). Data were acquired with a LSRFORTESSA X-20 (BD) or Accuri (BD) with subsequent analysis using the FlowJo software.
Statistics. Data were analyzed with the GraphPad Prism 6 software (GraphPad Software) or Microsoft Excel. Data shown in graphs represent the mean values ± s.e.m., as indicated in the figure legends. Preliminary data sets were used to determine the group size necessary for adequate statistical power. Statistical analyses were performed by unpaired two-tailed Student's t test. For multiple grouped comparisons, two-way analysis of variance was applied. Statistical significance in survival curves was determined using a log-rank test. A P value of >0.05 was considered not significant (NS), whereas P ≤ 0.05 was indicated with one asterisk (*), P ≤ 0.01 with double asterisks (**), P ≤ 0.001 with triple asterisks (***) and P ≤ 0.0001 with four asterisks (****). In all cases, comparisons were made between the indicated knockout (KO) mice and the respective littermate controls.

Data availability
All data are available from the authors upon request. Additional information on this manuscript can be found in the Supplementary Information.