Linear ubiquitin chain assembly complex coordinates late thymic T-cell differentiation and regulatory T-cell homeostasis

The linear ubiquitin chain assembly complex (LUBAC) is essential for innate immunity in mice and humans, yet its role in adaptive immunity is unclear. Here we show that the LUBAC components HOIP, HOIL-1 and SHARPIN have essential roles in late thymocyte differentiation, FOXP3+ regulatory T (Treg)-cell development and Treg cell homeostasis. LUBAC activity is not required to prevent TNF-induced apoptosis or necroptosis but is necessary for the transcriptional programme of the penultimate stage of thymocyte differentiation. Treg cell-specific ablation of HOIP causes severe Treg cell deficiency and lethal immune pathology, revealing an ongoing requirement of LUBAC activity for Treg cell homeostasis. These data reveal stage-specific requirements for LUBAC in coordinating the signals required for T-cell differentiation.

T he thymus orchestrates the differentiation of haematopoietic precursors into diverse T-cell sub-lineages. These lineages include conventional T-cell receptor (TCR)ab CD4 þ and CD8 þ T cells, Forkhead box-P3 þ (FOXP3 þ ) regulatory T (Treg) cells, natural killer T (NKT) cells, TCRgd T cells and CD8aa T cells. A major determinant of cell fate is the specificity of the newly rearranged TCR for major histocompatibility complex (MHC) or MHC-like molecules presenting self-constituents, yet this stimulus alone is not sufficient to elaborate the many different T-cell types. T-cell differentiation is also influenced by cytokine receptors, members of the tumour necrosis factor receptor (TNFR) superfamily, chemokine receptors and adhesion molecules. Yet, precisely how these various cues are integrated to coordinate T-cell differentiation is unclear.
Positive selection rescues double-positive (DP) thymocytes from death-by-neglect and initiates the largest transcriptional re-programming in T-cell differentiation 1 . The upregulation of the C-C chemokine receptor type 7 (CCR7) mediates the migration of thymocytes from the cortex to the medulla as they differentiate into CD4 þ or CD8 þ single-positive (SP) cells. During residency in the medulla 2 , SP thymocytes undergo further maturation that involves a switch in TCR responses from apoptosis to proliferation and acquisition of the capacity to emigrate from the thymus 3 . Few of the stimuli that drive this maturation are known, although the nuclear factor-kB (NF-kB) pathway and interleukin (IL)-7 receptor signalling are important [3][4][5] .
Treg cells are a potent immune modulatory subset of CD4 þ T cells that emerge during the late stage of thymocyte differentiation 6 . The integration of cues from the TCR, members of the TNFR superfamily and cytokine receptors (mainly the IL-2 receptor) culminate in the expression of the key transcription factor, FOXP3 (refs 7,8). The NF-kB signalling pathway is critical for Treg cell differentiation, in particular, c-REL is necessary to consolidate FOXP3 expression to enable Treg cell proliferation 6,7 . In the periphery, Treg cells continue to rely on TCR and co-stimulatory inputs for their proliferation and differentiation into the various effector states that are required for proper immune regulation [9][10][11] .
The linear ubiquitin chain assembly complex (LUBAC) is composed of at least three proteins: ring finger protein 31 (RNF31/HOIP), RanBP-type and C3HC4-type zinc finger containing 1 (RBCK1/HOIL-1) and SHANK-associated RH domain interacting protein (SHARPIN/SIPL1) 12 . LUBAC can regulate diverse cell signalling pathways by catalysing the addition of linear ubiquitin chains to substrates. Innate and adaptive immune responses depend on LUBAC activity downstream of TNFR1, NOD2, TLR, NLRP3, TCR and B-cell receptor ligation 13,14 . These signals involve the linear ubiquitination of NEMO to reinforce canonical NF-kB signalling, although it is likely to be that other LUBAC substrates exist. Loss of LUBAC activity drives cells into apoptosis or necroptosis following exposure to TNF, lymphotoxin a or genotoxic stress [15][16][17][18][19] . All three LUBAC components are required for maximal linear ubiquitination; however, not all components are equal. Although HOIP deficiency alone completely ablates LUBAC activity 18,19 , SHARPIN-deficient cells still display substantial linear ubiquitination, because HOIL/HOIP complexes are able to sustain significant LUBAC function [17][18][19] . Consistent with these observations, HOIP-deficient mice are embryonic lethal 18 , whereas the SHARPIN-deficient mice from the chronic proliferative dermatitis mutation (cpdm) strain (hereafter referred to as Sharpin cpdm mice) are born viable, but succumb to severe dermatitis at 12-14 weeks of age 20,21 .
Patients with loss-of-function mutations in RBCK1 (encoding HOIL-1) or RNF31 (encoding HOIP) exhibit impaired NF-kB responses, defects in B-cell activation and hyper-responsiveness of monocytes to IL-1b, the latter presumably driving autoinflammatory disease 22,23 . These patients also had evidence of T-cell defects, including low thymic output and decreased TCRab þ CD4 þ and CD8 þ T cells, which exhibit poor proliferative responses to mitogens and antigens 22,23 , but whether these defects represent T-cell intrinsic defects is unclear.
In this study, we examine the requirement for each LUBAC component in T-cell and Treg cell lineages. The data reveal that LUBAC components play pivotal roles in late thymocyte differentiation of conventional T cells, non-conventional T cells and Treg cell homeostasis. LUBAC activity is necessary for the transcriptional programming of late thymocyte differentiation. Consistent with the distinct requirements for HOIL and HOIP versus SHARPIN in linear ubiquitination, the T-cell defects observed are more severe with HOIL-1 or HOIP deficiency compared with Sharpin deficiency. These data highlight previously unappreciated roles for LUBAC in T-cell biology.

LUBAC activity is required for thymic T-cell differentiation.
To determine whether T-cell differentiation requires LUBAC activity, we used loss-of-function genetic models for each of the three known components. We used a Cd4 Cre transgene to induce conditional excision of loxP-flanked alleles of Rnf31 (the gene encoding HOIP) or Rbck1 (the gene encoding HOIL-1) to create mice with T-lineage-specific deletion (hereafter termed Hoip DCd4 or Hoil DCd4 , respectively). The role of the third LUBAC component was investigated using Sharpin cpdm mice, which lack SHARPIN in all cells and were analysed before the development of extensive skin pathology.
The peripheral immune organs of Hoip DCd4 and Hoil DCd4 mice were almost completely devoid of CD8 þ and CD4 þ abTCR þ T cells (Fig. 1a,b). Although the proportions of FOXP3 þ Treg cells among abTCR þ CD4 þ T cells were normal in Hoip DCd4 and Hoil DCd4 mice, their number was greatly reduced, in line with the overall T-cell deficiency (Fig. 1a,b). The residual abTCR þ T cells in Hoip DCd4 and Hoil DCd4 mice were predominantly CD44 hi CD62L lo (Fig. 1c,d), suggestive of an activated/effector phenotype that is often associated with 'homeostatic' expansion during lymphopenia 24 . By contrast, the numbers, proportions and activation status of conventional CD8 þ and CD4 þ T cells in Sharpin cpdm mice were comparable to controls (Fig. 1a-d). Consistent with recent reports 25,26 , we observed that Treg cells were reduced in Sharpin cpdm mice, although to a much lesser extent compared with the Hoip DCd4 and Hoil DCd4 mice (Fig. 1a,b). These data indicate that the LUBAC components HOIL-1 and HOIP are necessary for conventional CD4 þ and CD8 þ T-cell differentiation, and that SHARPIN is dispensable.
This differential requirement for LUBAC components extended to non-conventional T cells. CD1d-dependent NKT cells are potent immune modulatory T cells that can be detected by staining with a-galactosylceramide-loaded CD1d tetramers. NKT cells were virtually undetectable in the spleen and lymph nodes of Hoip DCd4 and Hoil DCd4 mice, yet they could be recovered from Sharpin cpdm mice (albeit in reduced numbers compared to controls) (Fig. 1e,f and data not shown). Normal numbers of gdTCR þ T cells were found in Sharpin cpdm mice (and, as expected, in Hoip DCd4 and Hoil DCd4 mice, as Cd4 Cre only becomes active after divergence of the abTCR and gdTCR T cell lineages; Supplementary Fig. 1).
We tracked the origin of these T-cell defects to the thymus. The proportions of CD4 þ and CD8 þ SP thymocytes were significantly reduced in Hoip DCd4 and Hoil DCd4 mice but were found to be normal in Sharpin cpdm mice (Fig. 2a,b and Supplementary Fig. 2). By contrast, the proportions and numbers of FOXP3 þ Treg cells among CD4SP thymocytes were greatly diminished in all three strains, as were the CD25 þ FOXP3 À and CD25 À FOXP3 þ thymic Treg cell precursors (Fig. 2a,b). These data demonstrate that all three LUBAC components are required for the earliest checkpoint in Treg cell differentiation. NKT cells were almost undetectable in the thymus of Hoip DCd4 and Hoil DCd4 mice but were present in normal numbers in Sharpin cpdm mice (Fig. 2c,d), demonstrating that this lineage has a dependency on LUBAC similar to that of conventional abTCR T cells.

Hoil ΔCd4
Hoil ΔCd4 were comparable to those seen in controls (data not shown). These data suggest that the reduction in HELIOS high FOXP3 À CCR7 þ CD4SP cells in mice lacking LUBAC components was not associated with altered TCR signal strength but rather impaired induction of differentiation programmes parallel to or downstream of high-avidity TCR signals.
Interactions between maturing thymocytes and thymic epithelial cells are essential for the induction of a normal thymic medulla, primarily via the provision of ligands for members of the TNFR superfamily that are expressed by the epithelium 30 .
To investigate the possibility that defects in the thymic microenvironment might contribute to the block in SP thymocyte differentiation observed in Hoip DCd4 and Hoil DCd4 mice, we created irradiation chimeras reconstituted with 50:50 mixtures of haematopoietic precursors from the mutant mice with CD45.1 þ congenically marked wild-type (WT) mice. Although the double-negative (DN) and DP thymocyte precursor populations showed B40-50% representation of the CD45.2 þ Hoip DCd4 or Hoil DCd4 compartments, there was a specific loss of CD45.2 þ cells at the CD8SP, CD4SP CCR7 þ 'wave 2' and FOXP3 þ Treg cell stages (Fig. 3g). Virtually, no HOIP-or HOIL-1-deficient T-lineage cells were detected in the periphery of these chimeras (Fig. 3g). These data demonstrate that the requirement for LUBAC activity is T-cell intrinsic.
We then tested whether LUBAC deficiency was causing apoptosis of SP thymocytes. LUBAC activity might be required to prevent induction of pro-apoptotic BH3-only proteins or to upregulate pro-survival BCL-2 proteins downstream of TCR signalling 31 , or in response to DNA damage 15 . Therefore, we tested whether the complete ablation of the mitochondrial pathway of apoptosis would rescue T-cell differentiation in Hoip DCd4 mice. The multi-BH domain pro-apoptotic BCL-2 family proteins BAX and BAK are essential for the mitochondrial outer membrane permeabilization that executes this pathway of apoptosis 32 . Extensive redundancy between BAX and BAK, and the early lethality observed in Bax À / À Bak À / À mice necessitated conditional deletion of Bax using Cd4 Cre on a Bak À / À background to induce T-cell-specific ablation of the mitochondrial pathway of apoptosis. As expected, the DN block observed in Bax À / À Bak À / À haematopoietic chimeras 33 was bypassed in Bax DCd4 Bak À / À mice, yet expansion of CD4SP, CD8SP and mature DN thymocytes, and increased percentages of peripheral T cells were observed (Fig. 3h). The compound loss of BAX and BAK in Hoip DCd4 mice did not restore late-stage thymocyte differentiation or peripheral T cells ( Fig. 3h and data not shown). These data indicate that LUBAC was neither required to antagonize thymocyte deletion nor to transduce pro-survival signals to block the mitochondrial apoptotic pathway.

NF-jB signalling is partially impaired by LUBAC deficiency.
Previous studies found that LUBAC is critical for optimal NF-kB activation downstream of immune receptor signalling in B-cell lymphomas and Jurkat T cells by associating with the CARD11/BCL-10/MALT1 (CBM) complex 34,35 . Therefore, we compared the kinetics and extent of activation of the NF-kB pathway following CD3/CD28 stimulation of thymocytes from Hoil DCd4 , Sharpin cpdm and control mice. The degradation of inhibitor of kBa (IkBa) is a hallmark of NF-kB activation and was apparent within 0.5 h of CD3/CD28 stimulation of control thymocytes (Fig. 4a). However, in CD3/CD28-stimulated thymocytes from Hoil DCd4 mice, IkBa degradation was delayed (Fig. 4a), suggesting that loss of this LUBAC component caused defects in NF-kB activation. By contrast, the kinetics of p38 mitogen-activated protein kinase phosphorylation following CD3/CD28 stimulation of Hoil DCd4 and Sharpin cpdm thymocytes was comparable to controls (Fig. 4a). Although these data show that LUBAC is involved in transducing TCR-dependent NF-kB signals in thymocytes, this defect is unlikely to explain the block in late-stage thymocyte differentiation observed in Hoil DCd4 and Hoip DCd4 mice. Loss of CARD11, BCL10 or MALT1 completely blocks NF-kB activation following TCR stimulation of thymocytes, yet these defects do not impair conventional T-cell development (for example, see ref. 36). Collectively, these findings suggest a requirement for LUBAC in thymocyte differentiation beyond NF-kB activation downstream of TCR and the CBM complex. Therefore, we tested whether LUBAC was also required in thymocytes for NF-kB activation downstream of stimulation of TNFR family members. TNF stimulation of WT thymocytes induced phosphorylation of p65 (RELA) within 5 min and degradation of IkBa within 15 min (Fig. 4b). HOIL-1 or HOIP deficiency reduced and delayed TNF-induced p65 phosphorylation and IkBa degradation (Fig. 4b). Although TNF-stimulated thymocytes from Hoil DCd4 and Hoip DCd4 mice exhibited similar kinetics of p38 mitogen-activated protein kinase phosphorylation, the overall levels appeared to be lower than in control cells (Fig. 4b). Collectively, these data reveal a requirement for LUBAC activity in optimal NF-kB activation downstream of both TCR and TNFR ligation.
introduced a Cre-inducible allele of mutant Ikbkb that encodes a constitutively active form of IKK2 (IKKca) when expressed 37 . Although the proportions of CD4SP and CD8SP remained low in Hoip DCd4 IKKca mice, the proportions of mature CD4 þ CD24 low CD62 high cells and FOXP3 þ cells were restored to levels observed in control mice (Fig. 4c). Nevertheless, Hoip DCd4 IKKca mice had severe T-cell deficiency in the periphery (Fig. 4d). This outcome indicates that the reinforcement of NF-kB signalling in LUBAC-deficient T cells only partially rescues the block in late-stage T-cell differentiation, and that other cell survival or differentiation programmes must also rely on LUBAC.
LUBAC does not antagonize TNF-induced killing of thymocytes. Impaired LUBAC function can switch pro-survival TNFR1 signalling into caspase-8-dependent apoptosis or caspaseindependent, RIPK1/RIPK3/MLKL-mediated necroptosis 17,38,39 . TNF is produced constitutively in the thymic medulla by epithelial cells and dendritic cells 8 , prompting the hypothesis that LUBAC deficiency might lead to the death of medullary SP and Treg cells. Surprisingly, we found that TNF treatment of thymocytes from Hoip DCd4 , Hoil DCd4 and Sharpin cpdm mice did not induce greater cell death than observed in control thymocytes (Fig. 5a). Moreover, treatment of cells with TNF plus a small molecular mimetic of second mitochondria-derived activator of caspases (SMACs) did not induce additional death of Hoip DCd4 , Hoil DCd4 and Sharpin cpdm thymocytes. Similarly, when TNF/SMAC mimetic-induced cell death was blocked by the caspase inhibitor, QVD-OPh, to engage the alternative cell death mechanism, necroptosis, the viability of thymocytes from Hoip DCd4 , Hoil DCd4 and Sharpin cpdm mice was comparable to WT thymocytes (Fig. 5a). Necroptosis is dependent on the activities of RIPK1, RIPK3 and the pseudo-kinase MLKL. Blocking necroptosis using the RIPK1 inhibitor necrostatin-1 did not alter the survival of thymocytes from Hoip DCd4 , Hoil DCd4 or Sharpin cpdm mice in vitro. These data do not support the notion that LUBAC deficiency sensitizes thymocytes to TNF-induced cell death.
We also tested this hypothesis in vivo by analysing whether the loss of thymic Treg cells in Sharpin cpdm mice could be rescued by genetic ablation of TNF or critical cell death inducers. Tnf À / À Sharpin cpdm mice are protected from multi-organ inflammation 19 , but the fivefold reduction in thymic Treg cells caused by SHARPIN deficiency was not restored by loss of TNF (Fig. 5b,c). However, lymphotoxin a can serve as an alternative ligand for TNFR1 and is also extensively expressed in the thymic medulla. Formation of the death-inducing complex II following TNFR1 ligation can initiate caspase-8-dependent apoptosis or, when caspases are inhibited, RIPK3-and MLKL-dependent necroptosis 40 . Although normal differentiation of thymic Treg cells was observed in Mlkl À / À and Mlkl À / À Casp8 À / À mice, the reduction of thymic Treg cells caused by SHARPIN deficiency was not corrected in Sharpin cpdm Mlkl À / À Casp8 À / À mice (Fig. 5b,c). Likewise, loss of Ripk3 and Casp8 haploinsufficiency failed to rescue the thymic Treg cell defects in Sharpin cpdm mice (Fig. 5b,c)  ARTICLE To test whether the block in conventional thymocyte differentiation observed in Hoip DCd4 mice was caused by complex II-mediated apoptotic cell death, we generated Hoip DCd4 Casp8 DCd4 Mlkl À / À mice. Genetic ablation of caspase-8-mediated apoptosis and MLKL-mediated necroptosis failed to restore normal SP thymocyte differentiation or peripheral T-cell numbers in Hoip DCd4 mice (Fig. 5d). Collectively, these data demonstrate that LUBAC is not solely required to prevent apoptotic or necroptotic cell death in medullary thymocytes, but rather must be necessary for a process that is critical for the differentiation of maturing thymocytes.
LUBAC is required for transcriptional programming of T cells.
To identify the impact of LUBAC deficiency on the transcriptional programme of thymocyte differentiation immediately following positive selection, we fluorescence-activated cell sorting (FACS) purified lineage-depleted CD69 þ MHC I low and CD69 þ MHC I high thymocytes (Fractions 3 and 4; ref. 1) from WT, Hoil DCd4 and Sharpin cpdm mice, and subjected them to RNA sequencing. These populations were selected because: (1) this transition is associated with consolidation of the large transcriptional changes that follow positive selection 1 ; (2) the subset composition of these fractions, defined by CD4 and CD8 expression, was comparable among mice of the different genotypes ( Supplementary Fig. 3); and (3) this transition immediately precedes the loss of mature SP observed in the Hoil DCd4 and Hoip DCd4 mice (Fig. 3a).
HOIL deficiency altered the transcriptome immediately following positive selection (124 differentially expressed genes in CD69 þ MHC I low thymocytes) and this effect was amplified in   Percentages (b) and absolute numbers (c) of CD4 þ FOXP3 þ cells in the thymus of WT (n ¼ 18), Sharpin cpdm (n ¼ 11), Mlkl À / À (n ¼ 4), Mlkl À / À Casp8 À / À (n ¼ 4), Sharpin cpdm Casp8 À / À Mlkl À / À (n ¼ 4), Rip3k À / À Casp8 þ / À (n ¼ 4), Sharpin cpdm Rip3k À / À Casp8 þ / À (n ¼ 4), Tnf À / À (n ¼ 4) and Sharpin cpdm Tnf À / À (n ¼ 5) mice. Shpn cpdm refers to Sharpin cpdm mice. CD69 þ MHC I high thymocytes (724 differentially expressed genes; Fig. 6a). Surprisingly, thymocytes from Sharpin cpdm mice had more substantial alteration of the transcriptome, perhaps reflecting the consequences of moderate LUBAC defects throughout T-cell differentiation (compared with the conditional deletion of HOIL-1 at the DP stage) and minor differences in the genetic background (still largely C57BL/Ka versus WT C57BL/6). To analyse the transcriptional changes associated with the block in SP thymocyte differentiation, we performed a heat-map analysis of the most significantly upregulated (25) or downregulated (50) genes in CD69 þ MHC I low and CD69 þ MHC I high thymocytes from Hoil DCd4 compared with WT mice (Fig. 6b). Reduced expression of many core NF-kB target genes was a common feature of thymocytes from Hoil DCd4 and Sharpin cpdm mice. These included genes involved in negative feedback (for example, Nfkbia (or IkBa), Nfkbie (or IkBe), Birc3 (or cIAP2) and Tnfaip3 (or A20)) and T-cell differentiation (particularly of FOXP3 þ Treg cells; for example, Gadd45b, Il2ra, Tnfrsf18, Tnfrsf4 (or OX40) and Relb; Fig. 6b). To determine whether there was a defect in the induction of NF-kB target genes in the LUBAC-deficient strains during the differentiation from CD69 þ MHC I low into CD69 þ MHC I high thymocytes, we first identified 154 NF-kB target genes that were significantly up-or downregulated during this transition in WT cells. Barcode enrichment plots show that, despite the dampened transcription of NF-kB target genes, the magnitude and direction of changes induced in these transcripts during the post-positive selection stages was maintained in HOIL-and SHARPIN-deficient cells ( Supplementary Fig. 4). These findings support our earlier data showing that, although LUBAC-deficient thymocytes have impaired NF-kB signalling, this defect does not by itself explain the block in thymocyte differentiation observed in Hoil DCd4 and Hoip DCd4 mice. We therefore focused on transcriptional changes that were observed in thymocytes from Hoil DCd4 mice (where differentiation was blocked), but not Sharpin cpdm mice (where conventional thymocyte differentiation proceeds normally). This filter revealed that genes involved in cytokine signalling were prominent; HOIL-deficient thymocytes failed to upregulate Il7r and downregulate Cish (encoding the SOCS family member, CIS, an inhibitor of IL-2 signalling), Il2rb and Cxcr4 (Fig. 6c,d). Consistent with these data, HOIL-deficient thymocytes appeared to have a specific defect in an IL-2-sensitive glycolytic transcriptional programme, with heightened expression of Bcl6 and reduced Myc and Slc2a3 (a regulator of glycolysis), an apparent parallel with a recent study in mature T-cell differentiation 41 . The interferon signalling pathway was also selectively impaired in HOIL-deficient thymocytes, with reduced transcription of Stat1, Irf1, Irf7 and Irf9 in CD69 þ MHC I low cells (Fig. 6c). Previous studies demonstrating that type I interferon is a feature of late thymocyte differentiation 5 , and that Irf1 is important for thymocyte differentiation 42,43 , suggest that these changes may also be critical for the impaired thymocyte development seen in the HOIL-and HOIP-deficient mice.
In summary, post-positive selection thymocytes from the LUBAC-deficient strains shared baseline defects in the transcription of NF-kB targets that may explain the observed Treg cell deficiency. Furthermore, the specific loss of transcripts in a number of essential signal transduction pathways from HOIL-1-deficient cells is likely to account for their block in late-stage thymic T-cell differentiation.
LUBAC is essential for Treg cell homeostasis. The thymus of Hoip DCd4 , Hoil DCd4 and Sharpin cpdm mice all had equivalent reductions in 'wave 2' cells, Treg cell precursors and mature Treg cells, yet only SHARPIN-deficient mice had substantial numbers of peripheral Treg cells (Figs 1 and 2). To determine whether the absence of Treg cells in Hoip DCd4 mice was merely secondary to the loss of conventional T cells in the periphery resulting in reduced levels of IL-2 (an essential cytokine for the maintenance of Treg cells) or whether this phenotype reflected an ongoing requirement for LUBAC activity for Treg cell homeostasis, we created male Foxp3 Cre/y ; Rnf31 fl/fl and female Foxp3 Cre/Cre ; Rnf31 fl/fl mice (both termed Hoip DFoxp3 hereafter). Mice of the control genotypes remained healthy and survived beyond 60 days of age, but all Hoip DFoxp3 mice developed a severe wasting disease and died around weaning (Fig. 7a-c). Hoip DFoxp3 mice exhibited severe immune pathology, including lymphadenopathy, lymphocytic perivascular infiltration and tissue destruction of the lung, liver and exocrine pancreas, hyper-IgE production and abnormally high numbers of activated CD4 þ and CD8 þ T cells (CD44 high CD62L low KI-67 þ ; Fig. 7d-h and Supplementary Fig. 5a-c). These features are all hallmarks of the Foxp3-deficient scurfy mouse phenotype 44 .
Although we observed normal proportions of thymic FOXP3 þ Treg cells among CD4SP in Hoip DFoxp3 mice, there was a marked decrease in the numbers and proportions of CD4 þ Treg cells in the spleen and lymph nodes ( Fig. 7i and Supplementary  Fig. 5d,e). To determine whether there was a cell intrinsic requirement for HOIP in peripheral Treg cells, we took advantage of the fact that the Foxp3 YFP À Cre knock-in allele is X-linked 45 to create chimeras in heterozygous Foxp3 Cre/ þ females through X-inactivation. Although we could recover a substantial fraction of YFP þ Treg cells from the periphery of Foxp3 Cre/ þ Hoip fl/ þ control mice, very few YFP þ Treg cells were detected in Foxp3 Cre/ þ Hoip fl/fl females ( Supplementary Fig. 5f,g). Treg cell loss in Hoip DFoxp3 mice could not be rescued by repeated anti-TNF treatment of neonatal mice commencing from day 5 (data not shown), suggesting that this defect was not caused by excessive TNF-induced cell death. These findings establish that ongoing LUBAC activity is required for the maintenance of mature Treg cells, with a greater reliance on HOIP activity than SHARPIN.
Nevertheless, a deeper examination of the peripheral Treg cell compartment of Sharpin cpdm mice revealed substantial homeostatic perturbation, with markedly increased numbers of proliferating (Ki-67 þ ) Treg cells, elevated expression of CTLA-4 (a key Treg cell effector molecule 46 ), increased proportions of effector Treg cells (CD44 high CD62L low ) and reduced expression of the pro-survival protein, BCL-2 (Fig. 7j).

Discussion
These data reveal an essential, cell intrinsic role for LUBAC in multiple aspects of T-cell differentiation. HOIP and HOIL-1 were required for the differentiation of conventional abT cells, FOXP3 þ Treg cells and NKT cells in the thymus (in Hoip DCd4 and Hoil DCd4 mice) and maintenance of Treg cells in the periphery (in Hoip DFoxp3 mice). These findings suggest that the T-cell deficiency observed in patients with loss-of-function mutations affecting HOIL and HOIP is a primary defect 22,23 . The thymic phenotypes caused by loss of HOIL or HOIP are reminiscent of those observed with T-cell-specific deletion of NEMO 47 or TAK1 (refs 5,48,49) and align well with the interactions described between LUBAC and these components of NF-kB signalling 13,14,50 . However, constitutive activation of the NF-kB pathway only partially rescued these defects in HOIP-deficient thymocytes, suggesting additional roles for LUBAC in thymocyte differentiation.
By contrast, SHARPIN deficiency had a relatively mild impact on T-cell differentiation; SP maturation was normal but HELIOS upregulation in CD4SP was impaired, and numbers of Treg    precursors and differentiated thymic FOXP3 þ CD25 þ Treg cells markedly reduced. These distinct phenotypes are likely to reflect the relative roles of HOIP and SHARPIN in linear ubiquitination; loss of HOIP completely ablates linear ubiquitination following TNF stimulation, yet SHARPIN deficiency only partially impairs LUBAC activity (refs 17-19, our unpublished data). Thus, HOIL-1/HOIP complexes would sustain sufficient LUBAC function to support conventional thymocyte and NKT cell differentiation, yet optimal LUBAC activity (including SHARPIN) is necessary for the Treg cell sub-lineage, which is more heavily reliant on NF-kB signals 6 . LUBAC is likely to coordinate signals from several stimuli essential for T-cell differentiation. Although a role for LUBAC in mediating B-cell receptor-and TCR-driven NF-kB signals via interactions with the CBM has been described 34,35 , the normal T-cell differentiation observed in CARD11-, BCL10-or MALT1-deficient mice (for example, see ref. 36) and our data showing that constitutive active IKK2 cannot restore peripheral T cells suggest that there are other LUBAC-dependent signals downstream of the TCR that are required for SP thymocyte maturation. Indeed, a role for SHARPIN has been implicated in JNK and ERK activation downstream of TCR signals 25,26 . Another possibility is that LUBAC is required to transduce signals from other cell surface receptors critical for SP maturation, such as members of the TNFR superfamily. This scenario may be particularly pertinent to the defects observed in HELIOS upregulation and thymic Treg differentiation. GITR, TNFR2 and OX40 play important, redundant roles in the intra-thymic differentiation of Treg cells that have received high affinity TCR signals 8 . Our data indicating that LUBAC has an important role in TNFR signalling in thymocytes support the notion that TNFR superfamily signals might also be important for the final stages of conventional T-cell differentiation in the thymus.
In this context, an important LUBAC function in several cell types is the inhibition of death receptor-mediated apoptosis or necroptosis. SHARPIN deficiency can predispose cells to caspase-8-dependent apoptosis or necroptosis, the latter via a pathway involving RIP3K and MLKL in cells receiving TNF signals 17 . Several lines of evidence suggest that induction of these cell death pathways does not account for the T-cell developmental defects we observed: (1) simultaneous genetic ablation of both of these cell death pathways did not rescue the impaired generation of thymic Treg cells in Sharpin cpdm mice or the block in conventional thymocyte differentiation in Hoip DCd4 mice; (2) HOIL-1-, HOIP-and SHARPIN-deficient thymocytes were not predisposed to TNF-induced cell death; (3) pharmacologic inhibition of apoptosis or necroptosis did not alter thymocyte viability; and (4) TNF blockade in vivo did not rescue thymocyte differentiation in Hoip DCd4 mice or the loss of Treg cells in Hoip DFoxp3 mice. These data contrast recent findings that TNF deficiency could restore late thymocyte differentiation in IKK-deficient mice 51 or TAK1-deficient mice 5 and suggest roles for LUBAC beyond inhibiting TNF-induced cell death. In addition, the failure of combined BAX/BAK deletion to rescue SP thymocyte maturation in Hoip DCd4 mice provides evidence that LUBAC activity is not required to antagonize thymocyte deletion triggered by BH3-only proteins or to mediate cytokinederived survival programmes (such as IL-7 or IL-2), stimuli that affect the mitochondrial pathway of apoptosis. Although we cannot exclude that alternative cell death pathways might be activated in LUBAC-deficient thymocytes, it is likely to be that LUBAC transduces other signals necessary for the transcriptional programmes guiding conventional T-cell and Treg cell differentiation in the thymus.
The partial rescue of thymocyte maturation observed in Hoip DCd4 IKKca mice suggests that LUBAC-mediated NF-kB activation is important, but not sufficient to drive the final stages of thymocyte differentiation. Our transcriptional analysis comparing postpositive selection thymocytes from Hoip DCd4 and Sharpin cpdm mice suggests a prominent role for pathways regulating cytokine responsiveness and metabolic fitness. Further studies will establish how LUBAC activity influences these pathways, but it is likely to be that this role also extends to peripheral Treg cell homeostasis. The deletion of HOIP following the thymic differentiation of FOXP3 þ cells caused near-complete loss of peripheral Treg cells, establishing a cell intrinsic requirement for ongoing LUBAC activity in this lineage. The kinetics of the ensuing immunopathology was much swifter compared with that observed in mice lacking the pro-survival BCL-2 family member, MCL-1, specifically in Treg cells 52 , but was similar to that seen in Foxp3-deficient scurfy mutant mice 44 . This finding indicates an acute requirement for continued LUBAC-dependent signalling in Treg cells following their export to the periphery.
We conclude that LUBAC is essential for coordinating multiple signals required for the differentiation and homeostasis of conventional and non-conventional T-cell types required for adaptive immunity and tolerance.
Generation of Rbck1 floxed mice. The targeting construct was designed to introduce loxP sites on either side of a 1.5 kb genomic fragment containing Rbck1 promoter and exons 1 and 2 (exon 2 contains the ATG start codon), as well as a FRT-flanked PGK-hygromycin resistance cassette for screening purposes (Supplementary Fig. 6). The targeting construct was electroporated into C57BL/6-derived Bruce-4 embryonic stem (ES) cells 56 . Homologous recombination events were identified by Southern blotting and blotting with a hygro-specific probe was used to confirm single-construct integration. A correctly targeted ES cell clone was injected into blastocysts, resulting in the gene-targeted mouse strain. The hygromycin-resistance cassette was deleted by crossing the resultant Hoil1-floxed heterozygous mice with C57BL/6-flpe-transgenic mice and the flpe transgene was subsequently eliminated by crossing offspring to C57BL/6 mice. All of the mice analysed were devoid of hygro and flpe. WT and floxed alleles were discriminated by a PCR with primers Hoil-Fwd Monomers of biotinylated mouse CD1d-PBS44 (a-GalCer analogue with a C24:1 acyl chain were tetramerized with streptavidin conjugated to phycoerythrin (BD Pharmingen). Staining for CCR7 was done for 60 min at 37°C in pre-warmed FACS buffer (PBS containing 1% vol/vol heat-inactivated bovine serum and 5 mM EDTA). All other surface stains were incubated for 30 min at 4°C. Intracellular staining for FOXP3, Ki67, BCL2 or Helios was performed after fixation and permeabilization using the reagents from the eBiosciences FOXP3 staining kit. Sample data were acquired on an LSRII or Fortessa flow cytometer (BD Biosciences) and analysed using FlowJo software (TreeStar).
Measurement of serum IgE. Serum IgE (1:10 dilution) was measured by ELISA using 2 mg ml À 1 rat anti-mouse Ig antibodies (Southern Biotech, clone 23G3) as a capture reagent and developed with 1:500 mouse Ig isotype-specific goat antimouse IgE (Fc specific) conjugated to horseradish peroxidase (Nordic MUBio) 57 . Capture reagent was applied overnight at 4°C, sera was applied for 4 h at room temperature and ELISA plates were developed in the dark for 45 min at room temperature. IgE isotype monoclonal anti-dinitrophenyl antibody produced in mouse, IgE isotype (Sigma, MO, clone SPE-7) were used as standards.
RNA sequencing. Thymocytes were prepared from three mice of each genotype (WT, Hoil DCd4 and Sharpin cpdm ), and CD69 þ MHC I low and CD69 þ MHC I high cells were FACS purified on a MoFlo cell sorter (Beckman Coulter), with a dump channel to gate out PI þ , CD25 þ , CD44 þ , NK1.1 þ , B220 þ , MHC II þ , Gr1 þ , Mac-1 þ and dTCR þ cells. Sorted cells were preserved in RNAlater (ThermoFisher Scientific) and frozen at À 80°C, then RNA was isolated with RNeasyPlus Mini kit (Qiagen). Messenger RNA reverse transcription and complementary DNA libraries were prepared using the TruSeq RNA Sample preparation kit (Illumina) following the manufacturer's instructions. Indexed sample libraries were subjected to 75 base single-end sequencing using the 75 cycle high-output kit v2 chemistry for the NextSeq 500 sequencing instrument (Illumina).
Bioinformatics analysis. Sequencing reads were mapped to the mouse genome (mm10) using the subread aligner 58 implemented in the Rsubread software package. Gene-level read counts were obtained using featureCounts 59 and its inbuild mm10 annotation, which includes Entrez gene ID, chromosome and gene length information, corresponding to the NCBI RefSeq annotations. Gene annotation was obtained from the NCBI Mus musculus gene info file (downloaded 25 September 2015). Statistical analysis used the edgeR 60 and limma 61 software packages. Genes were filtered as not expressed if they failed to show at least 0.5 count per million reads in at least 3 samples. As there are female and male mice in this experiment, genes from chromosome Y and the gene Xist were filtered out so as to correct for gender effect. Predicted genes and the genes without official gene symbols were also filtered out. TMM scale normalization 62 was applied and read counts were transformed to log2 counts per million with a prior count of 1 using the edgeR cpm function. Linear models were used to test for expression differences between different genotypes. Empirical array quality weights were estimated to allow for differences in quality between the RNA samples 63 . Each mouse was treated as a random block, allowing for correlation between CD69 þ MHC I low and CD69 þ MHC I high cells from the same mouse 64 . Differential expression between genotypes were assessed using empirical Bayes moderated t-statistics, allowing for an abundance trend in the s.e. and for robust estimation of Bayesian hyperparameters 65 . The Benjamini and Hochberg method 66 was used to adjust the P-values so as to control the false discovery rate.
A list of NF-kB target genes was obtained from http://www.bu.edu/nf-kb/generesources/target-genes. A number of steps were required to convert the various gene identifiers to mouse Entrez Gene IDs. Where possible, gene aliases were converted to current official human gene symbols using the Bioconductor annotation package, org.Hs.eg.db. Otherwise, human RefSeq accession numbers were converted to human symbols using org.Hs.eg.db and mouse RefSeq accession numbers were converted to mouse Entrez Gene IDs using org.Mm.eg.db. Finally, human symbols were mapped to mouse Entrez Gene IDs using the Jackson Laboratory mouse-human orthologue table downloaded December 2012 (ref. 67) and the NCBI mouse-human homologue table downloaded August 2013. NF-kB target genes that were differentially expressed at 5% false discovery rate between CD69 þ MHC I high versus CD69 þ MHC I low cells in WT mice were then used for gene-set testing between CD69 þ MHC I high versus CD69 þ MHC I low in Sharpin and between CD69 þ MHC I high versus CD69 þ MHC I low in HOIL-1. Gene-set testing was conducted using limma's roast function 68 , with 9,999 residual rotations and the same linear model settings as for the differential expression analysis. Barcode enrichment plots were produced using limma's barcode plot function.
Statistical analysis. Statistical comparisons were made using one-way analysis of variance with a Tukey's post-hoc test for multiple comparisons with Prism v.6.0 (GraphPad). P-values o0.05 were considered to indicate a statistically significant difference.
Data availability. Sequence data that support the findings of this study have been deposited in GEO with the primary accession code GSE74552. All additional data supporting the findings of this study are available within this article and its Supplementary Information files or from the corresponding author on a reasonable request.