NK cell development requires Tsc1-dependent negative regulation of IL-15-triggered mTORC1 activation

Activation of metabolic signalling by IL-15 is required for natural killer (NK) cell development. Here we show that Tsc1, a repressor of mTOR, is dispensable for the terminal maturation, survival and function of NK cells but is critical to restrict exhaustive proliferation of immature NK cells and activation downstream of IL-15 during NK cell development. Tsc1 is expressed in immature NK cells and is upregulated by IL-15. Haematopoietic-specific deletion of Tsc1 causes a marked decrease in the number of NK cells and compromises rejection of ‘missing-self' haematopoietic tumours and allogeneic bone marrow. The residual Tsc1-null NK cells display activated, pro-apoptotic phenotype and elevated mTORC1 activity. Deletion of Raptor, a component of mTORC1, largely reverses these defects. Tsc1-deficient NK cells express increased levels of T-bet and downregulate Eomes and CD122, a subunit of IL-15 receptor. These results reveal a role for Tsc1-dependent inhibition of mTORC1 activation during immature NK cell development.

N atural killer (NK) cells have critical roles in innate immunity against 'unwanted' cells, such as transformed tumour cells and virally infected cells, as well as that against class I major histocompatibility complex (MHC-I)mismatched allogeneic bone marrow. NK cell development occurs in the bone marrow of adults, and it is strictly dependent on signalling triggered by the pleiotropic cytokine interleukin-15 . Thus, mice lacking either IL-15 or any one of the IL-15 receptor subunits, including the a, b and g chains, display severely abnormal NK cell development 1,2 . IL-15 signalling is also thought to be involved in the maintenance of NK cell survival and homeostasis in peripheral immune niches. This cytokine supports NK cell development and survival mainly through the activation of a FOXO-dependent transcriptional programme and the prevention of Bim-mediated apoptosis [3][4][5] . IL-15-mediated signalling is also essential for NK cell homeostasis in vivo, primarily due to its regulation of the level of the anti-apoptotic protein . In view of its importance to NK cell physiology, IL-15 signalling must be actively regulated to prevent its aberrant activation, which is likely detrimental to NK cell development, homeostasis and quiescence.
IL-15-responsive NK cell progenitors (NKp), characterized by the expression of IL-2Rb (CD122), a subunit of IL-15 receptor, require the common gamma chain (gc/IL-2Rg/CD132) to transduce activating signals. IL-15 stimulation primarily activates the kinase function of Janus kinase (JAK), which leads to phosphorylation and activation of signal transducer and activator of transcription 3 (STAT3) and STAT5. Thus, humans or mice lacking JAK and STAT5A/B display impaired NK cell development 7,8 . IL-15 dramatically induces the expression of suppressor of cytokine signalling 2 (SOCS2) during its late phase of stimulation, which negatively modulates the JAK-STAT pathway, suggesting that SOCS2 is likely required for NK cell function due to its downregulation of IL-15 signalling 9 .
IL-15 also has the ability to biochemically activate PI3K pathways 10 . Mice simultaneously lacking the PI3K subunits P110 g and d have severe defects in early NK cell development 11,12 . Most recently, we have observed that PDK1 deficiency causes an almost 95% reduction in the number of NK cells 13 , comparable to p110 g and d double-mutant mice 14 strongly suggesting that PI3K regulates NK cell development largely via the recruitment of PDK1. PI3K activity is negatively controlled by phosphatase and tensin homologue deleted on chromosome ten and Src homology 2 domain-containing inositol-5-phosphatase-1. Mice lacking either phosphatase and tensin homologue deleted on chromosome 10 or Src homology 2 domain-containing inositol-5phosphatase-1 exhibit abnormal NK cell development and functionality, demonstrating the crucial roles of these two proteins in actively maintaining NK cell function [15][16][17] .
The distal signalling downstream of PI3K activation activates mammalian target of rapamycin (mTOR), which is a component of two active complexes, mTOR complex 1 (mTORC1) and mTORC2. mTOR kinase has been reported to play a crucial role as a key metabolic checkpoint in NK cell proliferation and activation 18 . The signalling mediated by mTOR regulates early NK cell development mainly via the induction of E4BP4, an NK cell-specific transcription factor, which maintains IL-15 responsiveness by upregulating CD122 expression. Therefore, the activity of IL-15 is amplified via a positive feedback loop 13 ; however, the activation of mTOR must be tightly regulated according to this activity. One of the most indispensable proteins involved in this process is tuberous sclerosis 1 (Tsc1). This peripheral membrane protein has been implicated as a tumour suppressor. Tsc1 forms a complex with Tsc2, which regulates mTORC1 signalling, and this process is mediated by Akt. Human Tsc1 mutations may result in tuberous sclerosis by causing functional impairment of the hamartin-tuberin complex. Interestingly, Tsc1 plays critical roles in immune processes, such as T-cell differentiation 19,20 , peripheral T-cell homeostasis 21 , dendritic cell development 22 and natural killer T (NKT) cell terminal differentiation 23 . Tsc1 is also required for the generation of memory CD8 þ cells, a process that strictly requires IL-15 signalling 24 . However, it remains unknown whether Tsc1 is necessary for restraining IL-15/mTOR signalling during NK cell development, homeostasis and functioning.
To address how IL-15 signalling is negatively regulated in NK cells, in the current study, we examine dynamic changes in the expression of negative regulators of two of the above-mentioned signalling pathways, JAK-STAT and PI3K-mTOR, after IL-15 triggering. Tsc1 is found to be upregulated at the late time point of IL-15 stimulation. Thus, we generate Tsc1-deficient mice to dissect the roles of Tsc1 in IL-15 signalling and NK cell development. We identify a unique role of Tsc1-dependent inhibition of mTORC1 activity in the optimization of IL-15 signalling in the early stage of NK cell development.

IL-15 stimulation upregulates Tsc1 in NK cells.
To investigate how IL-15-mediated signalling is negatively controlled, we measured the messenger RNA levels of several potential suppressors involved in the JAK-STAT pathway and PI3K/mTOR pathway in NK cells stimulated with a high dose of IL-15 for 18 h (h). As expected, Socs2, a negative regulator of the JAK-STAT pathway, was found to be upregulated, as previously reported 9 . Notably, the expression of Tsc1 was increased by over twofold after IL-15 triggering (Fig. 1a). The stimulation of NK cells by a gradient concentration of IL-15 resulted in a dose-dependent increase in Tsc1 expression (Fig. 1b). An in-depth analysis demonstrated that Tsc1 expression was slightly suppressed 3 h after IL-15 stimulation but then gradually increased at the later time points (9-18 h; Fig. 1c). To examine whether the IL-15-induced changes in Tsc1 expression correlates with mTOR activity, the phosphorylation of S6K (pS6K), an indicator of mTORC1 activation, was measured. Tsc1 expression was decreased at the earliest time of IL-15 stimulation (3 h) whereas pS6K was upregulated. At 9-18 h after IL-15 stimulation, however, Tsc1 expression was increased whereas pS6K was downregulated to the baseline level (Fig. 1d). Taken together, these results indicate that Tsc1 likely acts as a negative regulator to prevent prolonged IL-15-induced mTORC1 activation.
To understand how the dynamic regulation of Tsc1 in response to IL-15 was achieved, Tsc1 expression was monitored after treatment with rapamycin, an inhibitor of mTORC1. This treatment significantly counteracted the upregulation of Tsc1 by IL-15 (Fig. 1e). Therefore, the increased expression of Tsc1 is dependent on mTORC1 activation.
Tsc1 is mainly expressed in immature NK cells. To understand the physiological role of Tsc1 in NK cell development, the expression levels of Tsc1 were compared among the three major types of lymphocytes. Compared with T and B lymphocytes, Tsc1 was highly expressed in common lymphoid progenitor (CLP) and NK cells (Fig. 1f,g), mainly NKp, as well as relatively immature NK cells (CD27 À CD11b À ). The Tsc1 level was found to gradually decrease with NK cell maturation (Fig. 1g). The dynamic expression of Tsc1 suggests that this protein might be involved in IL-15-regulated early NK cell differentiation.
Tsc1 deletion affects the number of T and B cells. We first generated haematopoietic-specific Tsc1-deleted mice, Tsc1 fl/fl /Vav1-Cre þ (referred to as Tsc1 À / À mice). The efficiency ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12730 of Tsc1 deletion was confirmed by quantitative PCR (Supplementary Fig. 1a). Given that two previous studies have established that inducible knockdown of Tsc1 can lead to abnormal hematopoietic stem cell (HSC) numbers 25,26 , we initially determined whether Tsc1 deletion affected the generation of HSCs and CLPs in our model. The total numbers of Tsc1-deleted bone marrow cells was much lower than in wildtype (WT) mice ( Supplementary Fig. 1b); the percentage of HSCs in Tsc1-deleted mice were higher, but the absolute numbers of HSCs were slightly lower than in WT mice. Despite this observation, the absolute numbers of CLPs were comparable between the two genotypes ( Supplementary Fig. 1c-e), which allowed us to study lymphocyte development in the Tsc1 À / À mice.
The numbers of T and B cells were first measured. The Tsc1 À / À mice had considerably fewer T cells (including CD1d-tetramerreactive NKT cells) in the spleen ( Supplementary Fig. 2a,b), and the residual Tsc1-deficient T cells had an expanded CD62L lo CD44 hi population with an activated or memory phenotype ( Supplementary Fig. 1c); however, the CD4 þ and CD8 þ T-cell subsets displayed normal selection in the Tsc1 À / À thymus and had comparable numbers of maturation-related markers, such as CD62L and CD69 ( Supplementary Fig. 1d,e), suggesting that Tsc1 deficiency affects T-cell homeostasis but not development, as previously reported 21 . We further revealed that the Tsc1 À / À mice had normal number of pro-B cells, but they had reduced numbers of pre-B, immature and mature B cells, suggesting that the deletion of Tsc1 blocked B-cell development at an early stage ( Supplementary Fig. 3). Taken together, Tsc1 has different roles in the development and homeostasis of adaptive immune cells.
Tsc1-deficient mice have a minimal pool of NK cells. Next, we focused on the roles of Tsc1 in innate NK cell physiology.
Notably, the NK cell numbers were reduced by over 90% in the spleens and bone marrow of the Tsc1 À / À mice (Fig. 2a,b). Remarkable reductions in the numbers of NK cells were also observed in other organs, including the lymph nodes, liver and lungs (Fig. 2a,b). To address whether the diminished NK cell pools in the Tsc1 À / À mice were cell-autonomous defects, and were not due to the possible defect of HSC generation, Lin À CD122 þ NK1.1 À NKp cells isolated from CD45.1 þ WT and CD45.2 þ Tsc1 À / À mice were mixed at a 1:1 ratio, then adoptively transferred into immunodeficient Rag1 À / À gc À mice. NKp cells from the CD45.1 þ mice showed much greater repopulation activity than those from the Tsc1 À / À mice (Fig. 2c). Therefore, the critical requirement of Tsc1 for NK cell development is cell intrinsic.
Tsc1 regulates immature NK cell development. NK cells sequentially acquire differentiation markers with their maturation. Based on these markers, NK cell development is divided into multiple stages. Tsc1 À / À mice had comparable numbers of NKp cells. The proportion of early CD3 À CD122 þ NK1.1 þ CD11b À immature NK cells was much higher in the Tsc1 À / À mice, but their absolute numbers were dramatically reduced (Fig. 2d,e). The proportions and absolute numbers of mature NK cells that are CD3 À CD122 þ NK1.1 þ CD11b þ were significantly decreased in the Tsc1 À / À mice (Fig. 2d,e). Analysis of NK cell receptors demonstrated that the proportion of NK cells expressing Ly49 family members (including Ly49C, D, G and H) and integrin DX5 was minimal in the Tsc1 À / À mice, whereas the proportion of NK cells with immature markers (either CD117 or CD127) was markedly higher (Fig. 2f). These data collectively suggest that NK cell development in the Tsc1 À / À mice was blocked at an early stage.
mice, the ability of NK cells to mediate 'missing-self' rejection was evaluated, as previously described 27,28 . We found that MHC-Imissing splenocytes were largely eliminated by NK cells in control mice; however, NK cell-mediated killing was barely detectable in Tsc1 À / À mice (Fig. 3a). We further examined the NK cellmediated in vivo rejection of RMA-S cells, an NK-sensitive tumour cell line, and observed that the severe defect in the Tsc1 À / À mice was nearly comparable to that in immunocompromised mice (Fig. 3b). Lastly, we analysed whether Tsc1 deletion affected the ability of NK cells to prevent the metastasis of B16 melanoma. Notably, the lung metastasis of B16 cells in Tsc1 À / À mice was much severer than that in control mice, as evidenced by the increases in the lung weights and numbers of tumour colonies in the Tsc1 À / À mice (Fig. 3c). When WT NK cells were adoptively transferred into the Tsc1 À / À mice, the defective phenotype was largely rescued, further confirming that Tsc1 À / À mice have severe defects in NK cell function (Fig. 3c). To exclude the possibility that the NK cell defect was extrinsically caused by altered T-cell homeostasis ( Supplementary Fig. 2), Tsc1 À / À mice were further backcrossed onto Rag1-deficient mice. Tsc1 deficiency in these mice led to severe defect in the anti-metastasis of B16 melanoma, even in the absence of T and B cells (Fig. 3d). Thus, Tsc1 À / À mice effectively lack peripheral NK cells and innate cytotoxic activity in vivo.
Tsc1 restricts the proliferation of immature NK cells. To further determine whether the low number of NK cells in the Tsc1 À / À mice resulted from the reduced capacity for cell proliferation, we assessed the in vivo cycling of NK cells. Unexpectedly, Tsc1 deletion resulted in a twofold increase in the frequency of proliferating cells, as determined by the incorporation of bromodeoxyuridine (BrdU) and by staining with the proliferation marker Ki67 under steady-state conditions ( Fig. 4a,b). Using Fucci-2 reporter mice, in which cells in the S, G2 and M phases of the cell cycle were fluorescent 29 , we observed that 5-9% of Tsc1 À / À NK cells exhibited proliferating activity compared with 3% of WT NK cells (Fig. 4c). Therefore, Tsc1 is a negative regulator of NK cell proliferation.
In agreement with this finding that Tsc1 À / À NK cells proliferate rapidly, freshly isolated Tsc1 À / À NK cells were consistently larger size and showed more cells containing (Fig. 4d). Furthermore, the expression levels of the nutritional markers CD71 and CD98, as well as that of the activation marker CD69, were upregulated in these cells (Fig. 4e,f). Further experiments using mixed bone marrow chimera assays showed that the constitutive activation in the Tsc1 À / À mice was NK cell intrinsic ( Fig. 4g,h). The phenotype was also observed in Tsc1 À / À Rag1 À / À mice ( Supplementary Fig. 4a,b). Thus, Tsc1 regulates NK cell development, likely by restricting the proliferation of immature NK cells.

Constitutive mTORC1 activation disrupts NK cell development.
We next determined the biochemical mechanisms by which Tsc1 controls NK cell quiescence. We first measured the effect of Tsc1 deficiency on mTORC1 activity by examining phosphorylation of the ribosomal protein S6. Freshly isolated Tsc1 À / À NK cells exhibited an increase in the basal activity of this conventional mTORC1 target compared with naïve WT NK cells. However, Tsc1 À / À NK cells had the same level of phosphorylation of Akt at Ser473, which is phosphorylated by mTORC2 (Fig. 5a). To assess whether the increased activity of mTORC1 disrupted the dormancy programme in Tsc1 À / À NK cells, Tsc1 À / À mice were administered with rapamycin for 5 consecutive days. The chemical inhibition of mTORC1 resulted in the partial recovery of the percentage and absolute number of NK cells in the Tsc1 À / À mice (Fig. 5b), indicating that Tsc1 has an inhibitory effect on mTORC1 activity in NK cells.    To genetically confirm the requirement of optimal mTORC1 activity for early NK cell differentiation, we generated Tsc1 À / À mice that were heterozygous for Raptor, a scaffold protein of mTORC1 (Tsc1 fl/fl /Raptor fl/ þ /Vav1-Cre þ mice, referred to as Tsc1 À / À Raptor þ / À mice in brief), or for Rictor, which is a major component of mTORC2 (Tsc1 fl/fl /Rictor fl/ þ /Vav1-Cre þ mice, referred to as Tsc1 À / À Raptor þ / À mice in brief). The genetic knockdown of Raptor markedly increased the percentage and absolute number of NK cells in the Tsc1 À / À mice. However, the inactivation of a single allele of Rictor exacerbated the defective NK cell development in Tsc1 À / À mice, suggesting that mTORC2 plays a role in NK cell development (Fig. 5c). Furthermore, the partial inactivation of mTORC1 activity appeared to attenuate the increased activation in Tsc1 À / À NK cells (Fig. 5d,e). These results clearly demonstrate that the exacerbated mTORC1 activity in Tsc1 À / À mice is detrimental to the early development of immature NK cells.
Tsc1 deficiency decreases the expression of Eomes and CD122. Metabolic activation plays an essential role in T-cell differentiation by regulating the expression of transcription factors 30 . Thus, we sought to determine whether the over-activation of mTORC1 altered the expression profile of transcription factors in Tsc1 À / À mice. Intracellular staining of T-bet and Eomes, two T-box family transcription factors involved in the NK cell development, was performed (Fig. 6a). T-bet was over-expressed in Tsc1 À / À NK cells compared with Tsc1-intact NK cells. By contrast, the amount of Eomes was dramatically reduced in Tsc1-null NK cells, which was more apparent in relatively immature CD27 þ CD11b À NK cells (Fig. 6b,c). This finding seems to be consistent with a previous study showing that T-bet and Eomes could antagonize one another's expression in NK cells 31 . To avoid confounding effects of comparing NK cell development in two different genotypes, we also examined the expression of the two transcription factors in a developmentally matched comparison, which revealed that Tsc1 À / À NK cells exhibited more T-bet, but less Eomes, at all stages (Fig. 6c). Thus, Tsc1-deficient NK cells had a deficiency in the expression of transcription factors involved in NK cell development.
Eomes regulates NK cell developments mainly through transcriptional regulation of CD122, such that Eomes-deficient NK cells have less CD122 (ref. 32). These previous data motivated us to examine CD122 level in Tsc1 À / À NK cells. As expected, CD122 expression was dramatically reduced in Tsc1 À / À NK cells (Fig. 6d). The decreased amount of CD122 in Tsc1 À / À NK cells  (c,d) B16 metastasis assay, the indicated mice were injected intravenously with 2 Â 10 5 B16 cells. The mice were killed 14 day later, and the lung weights and numbers of tumour nodules were counted. To prove NK-cell dependence, 2 Â 10 6 of purified NK cells from WT mice were adoptively transferred to Tsc1 À / À mice ('Tsc1 À / À þ NK' in c). Each symbol represents an individual mouse. Data are shown as means ± s.d. ***Po0.001. Unpaired Student's t-tests (two-tailed) was used to calculate these values. could largely be rescued by the partial inactivation of mTORC1 in Tsc1 À / À Raptor þ / À mice (Fig. 6e). Moreover, knocking-down Raptor expression was also able to significantly correct the aberrant expression of T-bet and Eomes (Fig. 6f,g). Therefore, over-activation of mTORC1 in Tsc1 À / À NK cells leads to an imbalance of T-bet and Eomes. In addition, the decreased Eomes expression most likely correlates with the defective expression of CD122, a subunit of the IL-15 receptor that determines the sensitivity of NK cells to IL-15.
Tsc1-deficient NK cells undergo apoptosis. NK cells usually require high levels of CD122 to respond to IL-15. Deficient IL-15 signalling leads to Bim-dependent NK cell apoptosis 4 . We noted that Tsc1 À / À NK cells exhibited increased caspase activity and Annexin V staining, which are two hallmarks of apoptotic cell death (Fig. 7a). The increased apoptosis might result from the over-activation of mTORC1 because the reduction of Raptor could largely diminish the number of apoptotic NK cells in Tsc1 À / À mice (Fig. 7b). Interestingly, upon IL-15 stimulation, Tsc1 À / À NK cells were more likely to undergo apoptosis, whereas the apoptotic death of Tsc1 À / À Raptor þ / À NK cells was much less pronounced (Fig. 7b). These data suggest that Tsc1 plays an important role in preventing activation-induced NK cell apoptosis.
The balance of pro-and anti-apoptotic members of the Bcl2 family controls the survival of immune cells. Among these members, Bim is considered one of important initiators of NK cell apoptosis 4 . Although the expression of anti-apoptotic Bcl2 was not significantly affected by Tsc1 deficiency, Tsc1 À / À (e,f) Expression of the activation markers CD71, CD98 (e) and CD69 (f, upper, spleen; lower, BM) was assessed by flow cytometry, and the MFI was calculated. Each symbol represents an individual mouse. Data are shown as means ± s.d. (g,h) BM cells from WT or Tsc1 À / À CD45.2 mice were mixed with cells from CD45.1 mice at a 1:1 ratio and were transferred into irradiated Rag1 À / À gc À recipients. The sizes (FSC), granularity (SSC) and the expression of CD71, CD98, CD69 (g) and the proliferation marker Ki67 (h) of splenic NK cells from spike BM derived (CD45.1 þ ) and WT or Tsc1 À / À donor BM-derived (CD45.2 þ ) cells in the reconstituted mixed chimeras were analysed. **Po0.01 and ***Po0.001. Unpaired Student's t-tests (two-tailed) was used to calculate these values.
NK cells showed elevated Bim expression in the resting state (Fig. 7c). To determine whether Tsc1 À / À NK cells undergo Bim-dependent apoptosis, Tsc1 À / À mice were bred with Bim fl/fl mice to yield Tsc1 fl/fl /Bim fl/fl /Vav1-Cre þ mice (referred to as Tsc1 À / À Bim À / À mice). As a result, the additional deletion of Bim partially but significantly increased the frequency and absolute number of NK cells in Tsc1 À / À Bim À / À mice compared with those in Tsc1 À / À mice (Fig. 7d). Moreover, the removal of Bim largely counteracted the increased sensitivity of Tsc1 À / À NK cells to IL-15-induced apoptosis (Fig. 7e). However, the Bim deletion failed to correct the abnormal activation of Tsc1 À / À NK cells ( Supplementary Fig. 5), suggesting that Bim-dependent NK cell death in Tsc1 À / À mice is most likely a consequence of the activation. Taken together, the Bim-dependent apoptotic pathway only partially mediates the death of Tsc1 À / À NK cells, indicating that other apoptosis pathways may exist 4 .
Tsc1 is dispensable for NK cell terminal differentiation. The above findings reveal a very unique role for Tsc1 in early NK cell differentiation. To further investigate whether Tsc1 is required for NK cell terminal differentiation and function, we generated a Tsc1 fl/fl /Ncr1-Cre þ mouse model in which Tsc1 was deleted during the late stage of NK cell development by the enzyme Cre, which is controlled by the promoter of Ncr1, which encodes NKp46 in mice. Cre expression in terminal NK cells lead to an 80% reduction of Tsc1 level (Fig. 8a). The terminal deletion of Tsc1 did not result in a decrease in the percentage and number of NK cells (Fig. 8b). This model facilitated the assessment of the roles of Tsc1 in NK cell functions. Tsc1-deleted NK cells exhibited comparable levels of IFN-g production and CD107a expression following stimulation with haematopoietic RMA-S and YAC-1 cell lines in vitro (Fig. 8c) and by plate-coated antibodies against the ITAM-containing receptor NK1.1 or Ly49D (Fig. 8d). These results were supported by the finding of Tsc1-null NK cell activation by the combination of the cytokines IL-12 and IL-18 (Fig. 8d). Moreover, an extensive in vivo analysis showed that the Tsc1-deleted mice successfully rejected the mismatched b2m À / À splenocytes and killed the RMA-S cells, comparable to the control mice (Fig. 8e,f).
To exclude the possibility that the dispensable activity of Tsc1 observed during the late stage of NK cell differentiation and function was due to the unsuccessful deletion of Tsc1 in the model, we further bred Tsc1 fl/fl Ncr1-Cre þ mice with eYFP STOP reporter mice to yield Tsc1 fl/fl /eYFP STOP /Ncr1-Cre þ mice, in which the Cre-expressing NK cells were mostly eYFP-positive. Tsc1 messenger RNA was hardly detectable in the eYFP þ Tsc1 fl/fl /Ncr1-Cre þ NK cells (Supplementary Fig. 6a). Using this fate-mapping strategy, we failed to observe apparent differences in NK cell functions, such as cytokine production and degranulation, between the two genotypes when NK cells were triggered by several types of stimuli, including tumour cells, platecoated antibodies and a cytokine cocktail (Supplementary Fig. 6). We therefore concluded that Tsc1 is not required for terminal NK cell development and effector functions.

Discussion
NK cells are the most important innate lymphocytes that mediate tumour surveillance. A fundamental immunological question is how the development and homeostasis of NK cells are regulated at the metabolic and transcriptional levels 33,34 . Although IL-15 has been confirmed to be indispensable for these processes 35  and multiple intracellular signalling pathways and transcription factors that promote NK cell development have been extensively studied, knowledge is limited regarding how these signalling networks are negatively regulated. In addition, NK cells are regulated to maintain their quiescence for immune homeostasis, but the manner by which their quiescence is carefully maintained also remains undefined. In this study, using two genetic mouse models, we dissected the critical role of Tsc1 in the maintenance of NK cell quiescence. The disruption of Tsc1 during the early stage of NK cell commitment (using Vav1-Cre) led to a large decrease in the number of NK cells. Intriguingly, inactivation of Tsc1 at the late stage (using Ncr1-Cre) had no obvious effect on terminal NK cell maturation or function. Therefore, our results reveal a unique role for Tsc1 in regulating NK cell development in the early stage by controlling IL-15-stimulated mTORC1 activation. Human patients with Tsc1 deficiency have a higher tumour incidence; in fact, Tsc1 was originally regarded as a tumour repressor 37,38 , and it is intrinsically required for normal cells to suppress the undue activation of mTORC1, which is believed to have increased activity in most tumors 39 . The immune system plays an important role in the surveillance of malignancy. With increased understanding of the comprehensive roles of Tsc1 in immune cells, such as adaptive T and B cells, as well as in innate immune cells, including dendritic cells, it is reasonable to speculate that the increased tumour incidence in Tsc1-deficient patients is likely correlated with compromised antitumor immunity. Among the immune cells, NK cells are generally considered to be the innate lymphocytes responsible for the immunesurveillance of tumour cells. In this study, we discovered that the loss of Tsc1 caused a decrease in the number of NK cells and that this decrease was strongly associated with the lung metastasis of melanoma. Further research is required to determine whether NK cells are defective in patients with Tsc1 mutations, which may explain the high incidence of tumors in Tsc1-null patients. Therefore, Tsc1 may play an extrinsic role in the suppression of tumour growth via regulation of the cellularity of NK cells.
Emerging evidence obtained using mouse genetic models has revealed that IL-15-mediated metabolic activation is critical for NK cell development via the triggering of mTOR activation and the upregulation of nutritional receptors 13,18,40 . The NK cellspecific deletion of mTOR (Ncr1-cre generated by Eric Vivier's laboratory) markedly impaired NK cell differentiation and activation 18 . In addition, our recent results have revealed that the deletion of PDK1, an upstream mTOR kinase, causes severe defects in NK cell development 13 ; however, the manner by which mTOR activation regulates NK cell development remains unknown. Further, it is unclear whether mTORC1 or 2 is involved in IL-15 signalling and how the distal signalling is negatively controlled. In the current study, Tsc1 expression was found to be dynamic following IL-15 stimulation. During the very early stage, IL-15 slightly inhibited Tsc1 expression, accompanied by an increase in mTORC1 activation. Thus, the transient downregulation of Tsc1 likely highlights a novel mechanism that facilitates transcriptional events promoting early NK cell development via a positive feedback loop involving PDK1-Mtor/E4BP4/Eomes/CD122 (ref. 13); however, continuous mTORC1 activation could increase Tsc1 expression during the late stage of IL-15 stimulation, thereby initiating a mechanism of feedback inhibition to suppress the prolonged mTOR activation. Therefore, we have identified a regulatory mechanism of IL-15 signalling that prevents excessive mTORC1 activation during NK cell development.
In T cells, the loss of Tsc1 could cause an increase in mTORC1 activation. Moreover, Tsc1-deficient T cells display decreased activation of mTORC2 (ref. 21). Despite this finding, emerging evidence suggests that increased mTORC1 activity but not decreased mTORC2 activity contributes to most of the observed phenotypes caused by Tsc1 deficiency 21,22 . Our data confirm that the biochemical inhibition of mTORC1 activity by rapamycin largely restores the defects observed during early NK cell differentiation in the absence of Tsc1, indicating that the control of mTORC1 activity by Tsc1 is indispensable for NK cell quiescence. Further evidence obtained from genetic experiments indicates that the removal of one allele of Raptor also largely diminishes the abnormal phenotype of Tsc1-deficient mice. Thus, the Tsc1-dependent control of mTORC1 is vital for the functioning of the innate and adaptive immune systems.
Although convincing evidence suggests that increased mTORC1 activity contributes to the disruption of early NK cell quiescence, we unexpectedly found that inactivation of a main component of mTORC2, Rictor, aggravated the developmental defects in Tsc1-deficient NK cells. This finding has two implications. First, it further confirms that Tsc1 regulates early NK cell differentiation independent of mTORC2 activation. Moreover, Rictor-containing mTORC2 may participate in early NK cell development in an mTORC1-independent manner. Thus, further studies involving the genetic inactivation of mTORC1 and mTORC2 will help to elucidate the multifaceted role of mTOR in NK cell development.
A previous study has reported that Tsc1 is critical for the regulation of T-cell quiescence and function 21 . Tsc1 deletion causes no obvious developmental defects in T cells, with the exception of NKT cells in the thymus; therefore, Tsc1 only affects the quiescence of terminally differentiated T cells. Similarly, Tsc1 is only involved in NKT cell terminal maturation through the regulation of T-box transcription factor expression 41 . Moreover, Tsc1 regulates effector functions; for example, Tsc1-deficient T cells also exhibit hyporesponsiveness to T-cell receptor triggering. In contrast, in this study, we have revealed a distinct role of Tsc1 in NK cell biology. The stage-specific deletion of Tsc1 allowed us to observe the dispensable role of Tsc1 in terminal NK cell differentiation and effector functions. NK cells with terminal deletion of Tsc1 exhibit normal survival and function, strongly suggesting that mTORC1 signalling is not strictly required for other receptors involving NK activation. Furthermore, although IL-15 is mandatory for NK cell survival and homeostasis 42 , terminal deletion of Tsc1 had no visible effect on peripheral NK cell homeostasis. In contrast, early deletion of Tsc1 caused severe defects in NK cell differentiation. These findings indicate that Tsc1 plays a unique role in IL-15-mediated mTORC1 activation, which is critical for NK cell development.
In summary, NK cell development in the early stage is tightly controlled by IL-15 signalling. Negative regulation of this process by Tsc1 is essential for NK cell development. Tsc1 deficiency causes the hyper-activation of mTORC1, which is detrimental to immature NK cell development. Thus, we have identified a unique checkpoint that is crucial for NK cell development. Given the importance of the adoptive transfer of NK cells in tumour immunotherapy, the optimal mTORC1 activation triggered by IL-15 may provide new strategies for the ex vivo expansion of NK cells.

Methods
Mice. The Tsc1 flox/flox mice were a gift from Dr. Hongbin Zhang (Beijing Union Medical College, China). Tsc1 fl/fl mice were backcrossed to C57BL/6 background for eight generations before being bred with the other mice. Haematopoieticor NK cell-specific Tsc1-deficient mice were generated by crossing Tsc1 fl/fl with Vav1-Cre (B6.Cg-Tg(Vav1-cre)A2Kio/J, Jackson lab, stock Number, 008610) or Ncr1-Cre mice generated in our lab 13 . Fucci-2 reporter mice were a gift from Dr. Hai Qi (Tsinghua University) and crossed with Tsc1 fl/fl Vav1-Cre þ mice to obtain Tsc1-deficient Fucci-2 transgenic mice, where cells in S, G2 and M phases of the cell cycle are fluorescent 29 . Rag1 À / À gc À mice was described previously 13 . b2m-deficient mice, C57BL/6 mice, CD45.1, Rptor flox/flox mice, Rictor flox/flox mice and Bim flox/flox mice were purchased from Jackson lab. Both female and male mice with age of 8-12 weeks were used in our experiments. All the mice are C57BL/6 background and maintained under specific pathogen-free animal facilities of Tsinghua University. All procedures involving animals were approved by the Animal Ethics Committee of Tsinghua University. Anti-phospho-AKT (S473, #9271, 1:50), and Anti-phospho-S6 (#4856, 1:200) were obtained from Cell Signaling Technology (Beverly, MA). CD1d-PBS157 tetramer was kindly provided by NIH tetramer facility. For analysis of surface markers, cells were stained in PBS containing 2% (wt/vol) foetal bovine serum with indicated antibodies from eBioscience or BD. The expression level was presented as net mean fluorescence intensity (DMFI), which was determined by subtracting mean fluorescence intensity of isotype control. For detection of phosphorylated signalling proteins, NK cells were fixed with Phosflow Lyse/Fix buffer, followed by permeabilization with Phosflow Perm buffer III (BD) and staining with antibodies.
Real-time PCR. Total RNA from splenocytes, bone marrow cells or FACS-sorted NK, T and B cell subsets was extracted using Trizol Kit (Invitrogen), and reversetranscribed using reverse transcription system (Promega, A3500). Quantitative PCR was conducted using SYBR green-based detection. The expression level of the genes of interest was determined relative to the expression of b-actin.
Detection of NK cell apoptosis. Splenocytes were stained with Annexin V (BD Biosciences), and Caspase activity was measured with fluorescein isothiocyanateconjugated z-VAD-fmk according to the manufacturer's instruction (eBiosences).
In vitro cell activation. Mice were treated with intraperitoneal injection of polyI:C (10 mg mg À 1 ) for 18 h. Poly I:C-activated splenocytes (2 Â 10 6 ) were co-cultured with the same number of different target cells in a total volume of 500 ml in 24-well plate. For antibody stimulation, 24-well plates were coated with indicated antibodies at a concentration of 1 mg ml À 1 overnight. For cytokine stimulation, Poly I:C-activated splenocytes were treated with cytokines (recombinant mouse IL-12 (10 ng ml À 1 ), recombinant mouse IL-18 (10 ng ml À 1 )). BD GolgiStop reagent (BD Biosciences) was incorporated to inhibit intracellular protein transport processes, and meanwhile, eFluor-660-conjugated anti-CD107a antibody or respective control isotype was added at the beginning of incubation to detect lysosome synthesis. Splenocytes stimulated with PMA (50 ng ml À 1 ) plus ionomycin (1 mM) stimulation were used as positive controls. Medium only was used as negative control. Four hours after stimulation, cell mixtures were harvested for detection of intracellular IFN-g. First, the cells were stained for NKp46 (fluorescein isothiocyanate-NKp46) plus PE-Cy5-CD3, and then fixed/permeabilized with BD Cytofix/Cytoperm Buffer (BD Biosciences). These cells were then stained with anti-IFN-g-PE, or PE-conjugated isotype control as negative control. Data were acquired using a BD LSR Fortessa flow cytometer, and analysed with Cell Quest software (BD Biosciences).
In vivo splenocyte rejection assay. Splenocytes from b2m-deficient (b2m À / À ) mice were depleted of red blood cells by Ficoll-Hypaque density gradient centrifugation, and then labelled with 5 mM CFSE (Molecular Probes). At the same time, splenocytes from WT mice were labelled with 0.5 mM CFSE (10-fold ob2m À / À cells). Two types of CFSE-labelled splenocytes were mixed at 1:1 ratio. A mixture of 2 Â 10 6 splenocytes was intravenously injected into mice pre-treated with 200 mg Poly I:C for 18 h. 18 h later, CFSE-positive cells in blood, spleen and lymph nodes were determined by flow cytometry. To exclude the role of NK cells in the recipient mice, RAG1 À / À gc À mice were used. The percentage of b2 m À / À splenocyte rejection was calculated with the following formula: 100 Â [1 À (percentage of residual b2m À / À population in total CFSE þ cells of experimental group/percentage of residual b2m À / À population in total CFSE þ cells of RAG1 À / À gc À mice group)].
In vivo RMA-S clearance assay. Mice treated with 200 mg Poly I: C for 18 h were intraperitoneally injected with a mixture of target cells, NK-sensitive RMA-S cells expressing green fluorescent protein (GFP; 10 6 ) and NK-non-sensitive RMA expressing DsRed (10 6 ). 18 h later the mice were killed, and cells in peritoneal cavity were collected by repeated washing with PBS containing 2 mM EDTA. After centrifugation, cells were resuspended in 1 ml PBS. The relative percentages of RMA-S and RMA cells were measured by flow cytometry. The percentage of RMA-S cell rejection was calculated with the following formula: 100 Â [1 À (percentage of residual GFP þ population in total GFP þ and DsRed þ of experimental group/percentage of residual GFP þ population in total GFP þ and DsRed þ cells of RAG1 À / À gc À mice group)].
B16 melanoma lung metastasis mouse model. B16F10 melanoma cells in the log phase were resuspended in 1 Â Hanks Balanced Salt Solution and intravenously injected into the mice (2 Â 10 5 cells per mouse). fourteen days later the mice were killed. The lung was weighted and the number of lung surface nodules was counted under a dissecting microscope.
Data availability. The data that support the findings of this study are available from the corresponding author upon request.