FoxO1-mediated autophagy is required for NK cell development and innate immunity

Natural killer (NK) cells exert a crucial role in early immune responses as a major innate effector component. However, the underlying mechanisms of NK cell development remain largely elusive. Here we show that robust autophagy appears in the stage of immature NK cells (iNKs), which is required for NK cell development. Autophagy defects result in damaged mitochondria and accumulation of reactive oxygen species (ROS) that leads to apoptosis of NK cells. Autophagy protects NK cell viability during development through removal of damaged mitochondria and intracellular ROS. Phosphorylated Forkhead box O (FoxO)1 is located to the cytoplasm of iNKs and interacts with Atg7, leading to induction of autophagy. FoxO1 deficiency or an inactive FoxO1AAA mutant abrogates autophagy initiation in iNKs and impairs NK cell development and viral clearance. Therefore we conclude that FoxO1-mediated autophagy is required for NK cell development and NK cell-induced innate immunity.

N atural killer (NK) cells, a major component of innate immunity, serve as the first line of defence against transformed tumours and virus-infected cells 1,2 . NK cells were recently defined as a part of the group 1 innate lymphoid cells according to their cytokine secretion pattern 3 . Cytokine secretion and granule-mediated cytotoxicity are the two major effector functions of NK cells, which are critical for early immune responses 2,4 . NK cells also play a pivotal role in orchestrating adaptive immunity 5,6 . Recent studies reported that NK cells obtain antigen specificity and develop into long-lived memory cells under antigen stimulation, displaying their adaptive features of NK cells 6 . Like leukocyte populations, NK cells derive from hematopoietic stem cells (HSCs) in the bone marrow (BM). Each step of NK cell development is finely regulated via signalling by various cytokines and transcription factors. Common lymphoid progenitors (CLPs) derived from multipotent progenitors can differentiate into NK progenitors (NKPs). NKPs express IL-15 receptor b chain (CD122) that allows them to respond to IL-15 (ref. 7). Under IL-15 signalling, NKPs thereafter develop into immature NK (iNKs) and mature NK cells (mNKs) 8 . In addition, transcription factors are needed for NK cell specification. Id2 and Ets1, for example, promote NK cell lineage commitment 7 . E4bp4 directs iNK to mNK transition 9,10 . Gata-3, Eomes, T-bet and TOX are required for the maturation of NK cells [11][12][13][14] . However, it is unclear whether Forkhead box O (FoxO) family transcription factors play a role in NK cell development.
Mammalian FoxO transcription factors, containing FoxO1, FoxO3, FoxO4 and FoxO6, are homologues of the Caenorhabditis elegans FoxO ortholog Daf16 that is crucial for Dauer larval stage formation 15 . The Dauer larva, an alternative developmental stage of nematode worms, arrests development and allows survival in harsh conditions. Most FoxO members harbour an evolutionally conserved role in the modulation of nutrient sensing and stress responses. For instance, FoxO1 plays a critical role in cell cycle arrest, oxidative stress resistance and regulation of metabolism 16 . FoxO1 knockout mice exhibit vascular defects and die at E10.5 (ref. 16). Moreover, FoxO1 is essential for the regulation of homing and survival of naive T cells 15 . FoxO1 deficiency in Treg cells can switch their inhibitory functions to effector functions 17 . In addition, FoxO1 also regulates memory CD8 þ T-cell responses 18 . FoxO1 is also indispensable for early B-cell development and its peripheral functions 19 . Except for the transcriptional activity of FoxO1, cytosolic FoxO1 is able to induce autophagy in human cancer cells upon oxidative stress or serum starvation 20 . Additionally, several previous studies reported that another FoxO family member, FoxO3, is also involved in the autophagy induction in muscle cells 21 . A recent study showed that FoxO3a triggers autophagy that is essential for the life-long maintenance of HSCs 22 .
Macroautophagy (hereafter referred to as autophagy) is an evolutionarily conserved process that degrades long-lived proteins and unwanted organelles to recycle cellular components for survival and homoeostasis 23 . Autophagy participates in a variety of physiological processes, including lymphocyte development, embryonic development, cell reprogramming, tumour suppression and microbial clearance 24,25 . Moreover, autophagy is required for the maintenance of HSCs, T and B cells 23 . Recently, autophagy was shown to be required for plasma cell homoeostasis and humoral immunity 26 . During autophagy, autophagy-related genes, including Atg7, Atg5 and Atg3, are required for autophagosome formation 27 . However, whether autophagy is involved in NK cell development and effector functions are still unknown.
Here we show that robust autophagy appears in iNKs and is required for NK cell development. Phosphorylated FoxO1 is located to the cytoplasm of iNKs and interacts with Atg7, which promotes autophagy induction. FoxO1 deficiency or an inactive FoxO1 AAA mutant abolishes autophagy initiation in iNKs and impairs NK cell development and viral clearance.

Results
Robust autophagy appears in iNKs. To examine whether autophagy occurs during NK cell development, we tested for autophagic activity in different stages of NK cells by using green fluorescent protein (GFP)-fused LC3 (GFP-LC3) transgenic mice, which have been used to monitor autophagic activity. Turnover of GFP-LC3 fluorescence represents robust autophagic flux 28 . We found that the GFP fluorescence signal was remarkably reduced in iNKs (Fig. 1a-c), whereas the GFP signal was apparent in NKPs and mNKs. Four independent experiments obtained similar observations. Bafilomycin A1 (BafA1) serves as an autolysosome inhibitor 28 . With BafA1 treatment, autophagic puncta significantly occurred in iNK cells, but not in NKPs and mNKs (Fig. 1a-c), indicating that autophagic flux is increased in iNKs. These data suggest that robust autophagy occurs in the stage of iNKs.
LC3 conversion and p62 degradation are typical indicators of autophagy 29 . Consistently, we observed that obvious conversion of LC3-I to LC3-II appeared in iNKs (Fig. 1d), and p62, a substrate of autophagosomes, was almost hydrolysed in these cells. Upon BafA1 treatment, the degradation of LC3-II and p62 were abrogated in iNKs, indicating enhanced autophagic flux in these cells. By contrast, conversion of LC3 and degradation of p62 did not appear in NKPs and mNKs. Additionally, the mRNA levels of LC3 and p62 did not apparently change in iNKs ( Supplementary Fig. 1a). We noticed that NKp46 distributed in most iNKs and all mature NK cells, but not in NKPs (Supplementary Fig. 1b), consistent with a previous report 30 . Since an anti-NK1.1 antibody did not work for immunohistochemistry, we then used an anti-NKp46 antibody in place of anti-NK1.1 antibody to detect iNK and mature NK cell populations by immunohistochemistry. Consequently, we observed that iNKs displayed weak GFP-LC3 signals in BM sections (Fig. 1e), whereas mNKs showed strong GFP-LC3 fluorescence. We further analysed NK cell ultrastructures via immuno-eletron microscopy using anti-LC3 antibody. As expected, substantial LC3-positive autophagosome structures occurred in iNKs (Fig. 1f), and over 80% of iNKs displayed greater numbers of autophagosomes. By contrast, there were no overt LC3-positive autophagosome structures in NKPs or mNKs. Taken together, these results suggest that iNKs undergo robust autophagy during NK cell development.
Autophagy is required for NK cell development. To determine the function of autophagy in iNKs, we generated mice that lacked autophagic activity in iNKs by crossing Atg5 flox/flox mice with mice expressing Cre recombinase driven by the NK cell-specific NKp46 promoter (NKp46-Cre) 30 . Given that the NKp46 receptor is expressed in the stages of iNKs and mNKs, Atg5 was successfully deleted in iNKs and mNKs in Atg5 flox/flox ;NKp46-Cre mice ( Fig. 2a; Supplementary Fig. 2a). We observed that littermate Atg5 flox/ þ ;NKp46-Cre mice behaved similar to WT Atg5 þ / þ mice. We therefore used Atg5 flox/ þ ;NKp46-Cre mice as littermate controls. Notably, total lymphocyte numbers in Atg5 flox/flox ;NKp46-Cre mice were comparable to those of Atg5 flox/ þ ;NKp46-Cre control mice. Interestingly, we observed that Atg5 flox/flox ;NKp46-Cre mice had dramatically lowered percentages of iNKs and mNKs both in the BM and spleen (Fig. 2b), indicating reduced numbers of iNKs and mNKs accordingly. By contrast, Cre expression alone did not impact NK cell development and restrictive expression of NKp46   ( Supplementary Fig. 2b,c). Moreover, several surface markers of iNKs and mNKs in Atg5 flox/flox ;NKp46-Cre mice were remarkably reduced by flow cytometry analysis (Fig. 2c), suggesting that Atg5 deficiency impaired the mature transition from NKPs to the stages of iNKs and mNKs. We next stained NK and B cells in BM sections. Expectedly, the number of NKp46 þ cells was dramatically reduced in Atg5 flox/flox ;NKp46-Cre mice compared with Atg5 flox/ þ ;NKp46-Cre mice (Fig. 2d), whereas we did not detect a difference in the number of B cells. These results were further confirmed by flow cytometry analysis (Fig. 2e). The frequencies of other cell types, including HSCs, CLPs, myeloid cells in the BM were not significantly different in Atg5 flox/flox ;NKp46-Cre mice ( Supplementary Fig. 3a). Atg3 and Atg7 are involved in autophagosome formation 31 . Expectedly, silencing of Atg3-or Atg7-impaired NK cell development ( Supplementary Fig. 3b,c). These data indicate that autophagy is required for NK cell development.

Autophagy protects NK cells against ROS-induced apoptosis.
We next wanted to test the physiological role of autophagy in the progression of NK cell development. We noticed that many more apoptotic cells appeared in the BM of Atg5 flox/flox ;NKp46-Cre mice compared with Atg5 flox/ þ ;NKp46-Cre control mice (Fig. 4a). Indeed, the number of AnnexinVpositive NK1.1 þ NK cells was dramatically increased in Atg5 flox/flox ;NKp46-Cre mice according to flow cytometry analysis (Fig. 4b). However, the other cell types we detected in the BM of Atg5 flox/flox ;NKp46-Cre mice did not exhibit overt dead cells (Fig. 4c). Additionally, Atg5 flox/flox ;NKp46-Cre NK cells displayed high activation of caspase 3 and caspase 7 (Fig. 4d), whereas Atg5 flox/ þ ;NKp46-Cre NKs had basal levels. To further verify that NK cells underwent apoptosis in BM, we analysed apoptotic features of NK cells in BM sections. Consistently, most of NKp46 þ NK cells in Atg5 flox/flox ;NKp46-Cre mice exhibited high caspase 3 activation compared with those of Atg5 flox/ þ ;NKp46-Cre mice (Fig. 4e). These results indicate that defective autophagy results in apoptosis of NK cells.
NK cells undergo mTOR-independent autophagy. To explore the underlying pathway of autophagy in iNKs, we next analysed expression profiles of autophagy-related genes. We observed no significant change in the expression of the autophagy-related genes we checked in iNKs (Fig. 5a). We further detected the activation levels of these autophagy-related proteins by analysing their phosphorylation states. Inhibition of mammalian target of rapamycin (mTOR) activation can initiate autophagy. Intriguingly, mTOR was highly phosphorylated in iNKs compared with NKPs ( Fig. 5b). S6K, 4EBP and AKT are downstream substrates for mTOR 35 . Consequently, these mTOR substrates were also highly phosphorylated in iNKs (Fig. 5b). TSC2 is a downstream substrate of AKT (ref. 36). We found that TSC2 was also active in iNKs. However, the activation of AMPKa1 and AMPKb1 was unchanged in iNKs compared with NKPs (Fig. 5b). These results suggest that the mTOR pathway is activated in iNKs. mTOR exists in two distinct complexes, mTORC1 and mTORC2. mTORC1 phosphorylates S6K at Thr389 and mTORC2 phosphorylates AKT at Ser473, which consequently initiate the phosphorylation of AKT at Thr308. Inhibition of mTORC1 can initiate autophagy. While mTORC2 is able to modulate mTORC1 via AKT phosphorylation 35 . Raptor is a major component of mTORC1, while Rictor is a main component of mTORC2. We next wanted to test whether mTOR was involved in the regulation of NK cell autophagy or development. We observed that mTORC1 (Raptor) and mTORC2 (Rictor) were highly expressed in NKPs ( Supplementary Fig. 4c). We knocked down Raptor or Rictor in NKPs, respectively. Silencing of Raptor abrogated NK cell development (Fig. 5c). Moreover, GFP-LC3 signals remained at a high level in Raptor-silenced NK cells (Fig. 5d), suggesting that Raptor depletion failed to induce autophagy in NK cells. Similar results were obtained in Rictorsilenced NKPs (Fig. 5e,f). These data indicate that the mTOR complexes are not implicated in the autophagy initiation, but they are involved in the progression of NK cell development.
Rapamycin has been widely used as an inhibitor of mTOR 37 . We added rapamycin to NKP culture media for an in vitro development assay. We observed that rapamycin treatment abolished NK cell development (Fig. 5g), rather than promoted apparent autophagy in iNKs (Fig. 5h). Altogether, these results suggest that induction of autophagy required for NK cell development occurs independently of mTOR.
FoxO1 is highly phosphorylated in the cytoplasm of iNKs. Several studies have reported that FoxO family members can induce autophagy that is independent of mTOR signalling 20,22 . Given that iNKs undergo mTOR-independent autophagy, we then determined whether FoxO family members initiated autophagy in iNKs. Three FoxO family members, including FoxO1, FoxO3 and FoxO4, are expressed in the hematopoietic system 38 . Interestingly, we found that FoxO1 mRNA was highly transcribed in NKPs, iNKs and mNKs (Fig. 6a,b). However, FoxO3 mRNA level was very low in these NK cells (Fig. 6a,b), and FoxO4 was undetectable. Through immunoblotting, these three FoxO family members were highly expressed in LSKs (Lin À Sca1 þ cKit þ cells) (Fig. 6b), which were used as a positive control. However, the protein level of FoxO1 was remarkably reduced in mNKs (Fig. 6b). The proteasome inhibitor MG132 was able to block the degradation of FoxO1 and augmented ubiquitinated protein levels in mNKs (Fig. 6c). Similarly, a more specific proteasomal inhibitor epoxomicin also blocked the hydrolysis of FoxO1 in mNKs ( Supplementary  Fig. 4d). These observations suggest that FoxO1 is degraded in mNKs by the proteasome-dependent degradation pathway. We observed that FoxO1 mainly resided in the nuclei of NKPs (Fig. 6d). However, FoxO1 was mainly distributed in the cytoplasm of iNKs. By contrast, FoxO1 levels were extremely reduced in mNKs (Fig. 6d). These results were further validated by nuclear and cytosolic fractionation assays (Fig. 6e). Furthermore, phosphorylated FoxO1 levels were elevated in the cytoplasm of iNKs (Fig. 6e).
FoxO1 phosphorylation by AKT causes its translocation from the nucleus to the cytosol 39 , leading to transcriptional inactivation of its target genes. FoxO1 activation can upregulate many target genes, including Il7ra, Bcl2 and Klf4 (ref. 16), meanwhile, it also downregulates several repressed genes such as Tbx21, Ccr7 and Atg7 (refs 17,18). We observed that expression of genes upregulated by FoxO1 was significantly inhibited during the transition of NKPs to iNKs (Fig. 6f), whereas expression of FoxO1-repressed genes was activated in this process. Additionally, FoxO1 was highly phosphorylated in iNKs (Fig. 6g), while FoxO1 phosphorylation was almost undetectable in mNKs. These observations were validated by immunoblotting (Fig. 6h). As expected, FoxO1 levels were remarkably reduced in mNKs. More importantly, highly phosphorylated FoxO1 was localized in the cytoplasm of iNKs (Fig. 6i), which was confirmed by nuclear and cytosolic fractionation assays (Fig. 6e). In sum, we conclude that phosphorylated FoxO1 resides in the cytoplasm of iNKs (as opposed to the nucleus of NKPs). The differential expression of its target genes in iNKs relative to NKPs is consistent with a loss of FoxO1 transcriptional activity.

FoxO1 deficiency or mutation abrogates NK cell development.
To further confirm the role of FoxO1 in NK cell development, we generated mice with NK cell-specific knockout of Foxo1 gene by crossing Foxo1 flox/flox mice with NKp46-Cre mice. FoxO1 was completely deleted in iNKs and mNKs (Fig. 7a). We observed that FoxO1 deficiency impaired NK cell development (Fig. 7b), indicating that FoxO1 plays a critical role in NK cell development. FoxO1 undergoes phosphorylation at three residues (Thr24, Ser256 and Ser319) by AKT 16 . A mutant version of FoxO1, FoxO1 AAA (residues Thr24, Ser256 and Ser319 are mutated to Ala) cannot be phosphorylated and remains in the nucleus, leading to its lasting activation of its target genes 39 . FoxO1 AAA mice were generated by insertion of FoxO1 AAA mutant into Rosa26 locus preceded by a loxPflanked 'STOP' cassette 17 . We generated FoxO1 AAA -mutant mice (FoxO1 flox/flox ;FoxO1 AAA/ þ ;NKp46-Cre) that lacked endogenous expression of WT FoxO1 in NK cells and expressed the FoxO1 AAA -mutant gene under the control of NKp46 promoter 17 . FoxO1 AAA was expressed in the NK cells of these mice (Fig. 7a). Interestingly, the FoxO1 AAA mutant also impaired NK cell development (Fig. 7b). The mutant FoxO1 was restricted to nuclei  (Fig. 7c), while endogenous FoxO1 resided in the cytoplasm of iNKs in FoxO1 flox/flox mice. Consequently, nuclear location of FoxO1 AAA mutant abrogated autophagy induction in iNK cells (Fig. 7c-e). Notably, restoration of WT FoxO1 in FoxO1 flox/flox ;NKp46-Cre mice rescued the autophagic activity of iNKs (Fig. 7d,e), suggesting that FoxO1 is essential for autophagy initiation in iNKs.
The suppression of genes usually activated by FoxO1 in iNKs relative to NKPs was not affected in FoxO1 flox/flox ;NKp46-Cre mice compared with FoxO1 flox/flox or FoxO1 flox/flox ;NKp46-Cre rescued with WT FoxO1 (Supplementary Fig. 5a). Likewise the increased expression of genes usually repressed by FoxO1 in iNKs relative to NKPs was unaffected in these genotypes. However, this differential expression of genes usually regulated by FoxO1, which we presume is due to a loss of FoxO1 nuclear localization in iNKs relative to NKPs was not seen in FoxO1 flox/flox ;FoxO1 AAA/ þ ;NKp46-Cre mice where FoxO1 is restricted to the nucleus (Supplementary Fig. 5a).
Similarly, FoxO1 deficiency resulted in the same defective autophagy phenotype in iNK cells as FoxO1 AAA -mutant mice. Furthermore, FoxO1-deficient or FoxO1 AAA -mutant NK cells exhibited high accumulation of ROS (Fig. 7f), and elevated dead cells (Fig. 7g). Overall, we conclude that FoxO1 is involved in the induction of autophagy during NK cell development.
To explore the effect of FoxO1 on anti-MCMV response, we used MCMV to infect FoxO1-deficient or FoxO1 AAA -mutant mice. We observed that FoxO1 À / À or FoxO1 AAA NK cells almost blocked IFN-g production after MCMV infection (Fig. 7h), whereas FoxO1 flox/flox mice generated large amounts of IFN-g. Moreover, FoxO1 À / À or FoxO1 AAA NK cells displayed impaired cytotoxicity (Fig. 7i), and significantly reduced numbers in the periphery (Fig. 7j). Finally, FoxO1-deficient or FoxO1 AAAmutant mice were more susceptible to MCMV infection (Fig. 7k). These data suggest that FoxO1-mediated autophagy is required for NK cell development and NK cell-mediated innate immunity. FoxO1-binding domain mutants fail to induce autophagy. Atg7 is required for formation of autophagosomes 31 . Cytosolic FoxO1 can bind to Atg7 to induce autophagy in human tumour cells independently of its transcriptional activity 20 . We detected autophagy-related genes in iNKs and found that Atg7 was the only autophagy gene differentially regulated in iNKs (Fig. 8a). High expression of Atg7 in iNKs was validated by immunoblotting (Fig. 8b). We next isolated autophagosomes from iNK cells according to a previous study 40   FoxO1 and Atg7 were located in autophagosomes (Fig. 8c), suggesting that phosphorylated FoxO1 and Atg7 could be implicated in the regulation of autophagy in iNKs. Furthermore, the interaction of FoxO1 with Atg7 was verified by immunoprecipitation in iNKs (Fig. 8d). The C-terminal residues 537-596 are the binding domain for Atg7 (ref. 20), deletion of this binding domain of FoxO1 (DBD-FoxO1) abrogates its interaction with Atg7. We isolated NKPs and iNKs from the BM of FoxO1 flox/flox ;NKp46-Cre mice and infected with lentivirus encoding WT FoxO1 or DBD-FoxO1 for 24 h. With co-immunoprecipitation assays, we noticed that Atg7 associated with FoxO1 in iNK lysates (Fig. 8d). Expectedly, DBD-FoxO1 did not interact with Atg7 in iNK lysates (Fig. 8d). We found that DBD-FoxO1 ablated autophagy induction in iNKs ** ( Fig. 8e,f), whereas WT FoxO1-induced robust autophagy. Consistently, FoxO1-deficient iNKs did not initiate autophagy (Fig. 8e,f). Moreover, Atg7-silenced iNKs also abrogated autophagy induction and NK cell development ( Supplementary  Figs 3c and 5b). These data suggest that the interaction of FoxO1 with Atg7 may be essential for the induction of autophagy in iNKs.
To further confirm the in vivo role of FoxO1 in NK cell development and effector functions, we transplanted NKPs expressing WT FoxO1 or DBD-FoxO1 into Rag1 À / À Il2rg À / À mice (Fig. 9a). Expression levels of FoxO1 and DBD-FoxO1 in transplanted NK cells were comparable (Fig. 9a). Moreover, NKPs isolated from these three mice strains similarly distributed in recipient mouse spleens (Supplementary Fig. 5c). Of note, DBD-FoxO1-transplanted NKPs failed to induce autophagy in iNKs (Fig. 9b), and consequently generated a high level of ROS (Fig. 9c). We noticed that FoxO1-deficient iNKs gave similar results to DBD-FoxO1-transplanted iNKs. Consequently, DBD-FoxO1-transplanted NKPs impaired NK cell development (Fig. 9d). Finally, DBD-FoxO1-transplanted NKs impaired their ability to produce IFN-g (Fig. 9e), and defensive response to MCMV infection (Fig. 9f,g), which was similar to FoxO1 deficiency. Altogether, we conclude that the association of cytosolic FoxO1 with Atg7 likely contributes to the autophagy induction of iNKs that initiates NK cell development and effector functions.

NK cells play a critical role in early immune responses against transformed tumours and virus-infected cells 1 . NK cells develop
from BM-derived HSCs and undergo several developmental stages to give rise to mature lineages 8 . However, the underlying mechanisms of NK cell development remain largely elusive. In this study, we show that autophagy is required for NK cell development. FoxO1 and Atg7 are highly expressed in the cytoplasm of iNKs in BM, and when their association is disrupted robust autophagy fails to be initiated (Fig. 9h). Autophagy defects cause ROS accumulation that leads to apoptosis of NK cells. Autophagy is responsible for removal of intracellular ROS, which protects NK cell viability during NK cell specification. During NK cell development, NK cells express activating receptors and inhibitory receptors, as well as synthesize cytotoxic granules, which make NK cells ready for immune responses 7 . We also demonstrate that FoxO1-mediated autophagy is essential for effector functions of NK cells against viral infection.
Autophagy harbours multiple effects on immunity 23 . Invading microorganisms induce autophagy by competing nutrients or by activating innate immune receptors such as Toll-like receptors (TLRs). Autophagosomes engulf microbes and degrade them by recognition of autophagic adaptors, known as sequestosome 1like receptors 41 . Recently, autophagy has been considered to be implicated in innate and adaptive immune responses 24 . Lack of autophagy affects lymphoid precursor formation 42 . Autophagy deficiency abolishes both negative and positive selection of CD4 þ T cells 43 . Naive T cells require autophagy for their maturation after exiting the thymus 44 . Autophagy is also needed for B1 cell survival 45 . Additionally, plasma cells also require autophagy to sustain their viability during differentiation and to maintain their long-lived humoral immunity 26 . For the first time, we demonstrate that autophagy is essential for NK cell development and effector responses against viral infection.
ROS refer to O 2 -free radicals and non-free radicals' derivatives. ROS are mainly produced from mitochondria accompanying mitochondrial respiration. Moderate ROS levels are required for hematopoiesis during embryonic development and adult hematopoietic homoeostasis 46 . However, high levels of ROS can cause DNA damage, mitochondrial dysfunction and harmful protein modifications 47 . Since mitochondria are both the producers and targets of ROS, dysfunctional mitochondria can generate excessive ROS and release apoptosis-related factors such as cytochrome c and AIF (apoptosis inducing factor), which result in cell death. Increased ROS content triggers stem cell differentiation to short-term repopulating cells 48 . Here we show that ROS are kept low levels in NKPs, iNKs and mNKs in wildtype (WT) mice. Autophagy deficiency causes damaged mitochondria and substantial ROS accumulation in iNKs and mNKs, which lead to apoptosis of NK cells and blockade of NK cell maturation. Autophagy scavengers are able to counteract this effect, suggesting that autophagy is responsible for removal of ROS during NK cell development. Additionally, high ROS levels can inhibit effector functions of NK cells 49 . Thus, removal of high ROS accumulation could also be needed for the immune responses of NK cells.
Blockade of mTOR signalling is the most well-known trigger for autophagy initiation. Inhibition of mTOR leads to the activation of ULK1 complex that initiates autophagy 50 . We found mTOR is activated during NK cell development. Blockade of mTOR by rapamycin impairs, but does not promote, autophagy during NK cell maturation. Additionally, depletion of Ractor or Rictor in iNKs suppresses rather than promotes autophagy induction. Thus we conclude that autophagy induction in iNKs is independent of mTOR signalling. A recent study showed that a high dose of IL-15 induces mTOR activation and elevated bioenergetic metabolism in NK cells 51 . mTOR is an essential component of IL-15-mediated signalling, which is ARTICLE crucial for NK cell commitment. We observed that blocking IL-15 signalling abrogates autophagy induction in iNKs (Supplementary Fig. 5d), suggesting IL-15 signalling might be a upstream stimulus for autophagy initiation. mTOR activation is needed for the signalling transduction in PI3K-AKT pathway 35,52 . Transduction of IL-15 signalling relies on the PI3K-AKT pathway 53 . Actually, we show that the PI3K-AKT pathway is activated during NK cell maturation. Therefore, the active mTOR signalling is needed to activate the PI3K-AKT pathway for NK cell lineage specification.
FoxO transcription factors take part in many physiological processes, including integration of growth factor signalling, oxidative stress and immune responses 16 . FoxO1, a FoxO family member, regulates cell cycle, cell metabolism and apoptosis as a transcription factor 39 . FoxO1 is crucial for the fate determination of T cells 16 . Activation of FoxO1 drives T cells towards Treg cells or CD8 þ memory T cells 17 . Here we found that FoxO1 resides in the autophagosomes of iNKs and its interaction with Atg7 initiates autophagy induction. We noticed that blocking IL-15 signalling abolishes phosphorylation of FoxO1 and consequent translocation to the cytoplasm of iNKs ( Supplementary Fig. 5e), leading to autophagy inhibition. By contrast, cytosolic FoxO1 is rapidly hydrolysed in the cytoplasm of mNKs, which impairs autophagy induction. We still need to further investigate how cytosolic FoxO1 is degraded in mNKs. Thereby, the autophagic activity of FoxO1 is essential for NK cell maturation, which is independent of its transcriptional activation of its target genes.
A recent study reported a transcription-independent role of FoxO1 (ref. 20). Cytosolic FoxO1 dissociates from its deacetylase SIRT2 and induces its acetylation. Acetylated FoxO1 interacts with Atg7 to initiate autophagy in tumour cells 20 , which exerts antitumour activity. Here we show that phosphorylated cytosolic FoxO1 associates with Atg7 to induce autophagy in iNKs, which is essential for NK cell development. Whether the acetylation of FoxO1 is also needed for interaction with Atg7 in iNKs remains to be further explored. Cytosolic FoxO1 is dramatically degraded in mNKs, which may impede the interaction of FoxO1 with Atg7 leading to inhibition of autophagy. We propose that the modifications of FoxO1 may promote its interaction with Atg7, which stabilizes FoxO1 in the cytoplasm of iNKs resulting in autophagy induction. The subcellular localization of FoxO1 depends on post-translational modifications, especially phosphorylation 54 . Localization of FoxO1 depends on a variety of kinases such as AKT 54 . AKT is highly active in iNKs, suggesting that AKT-mediated phosphorylation of FoxO1 could contribute to the translocation of FoxO1 to the cytosol of iNK cells. We found that FoxO1 is uniquely localized in the nuclei of NKPs, and it resides in the cytoplasm of mNKs and iNKs.
Recently, Deng et al. 55 reported that FoxO1 is dispensable for NK cell development, and inactivation of FoxO1 is required for T-bet expression. The reasons for the apparent discrepancy between the results of Deng et al. and the current study are currently unclear. NKp46-Cre mice were generated independently in this study but we did follow the same procedure as described in Deng et al. 55 . In our study, as in Deng et al. 55 we found that the transcriptional activity of FoxO1 is impaired in iNKs ( Supplementary Fig. 5a). However, we failed to detect FoxO1 signals in mature NKs by using an anti-FoxO1 antibody (D7C1H, CST). Consistent results were obtained using two other anti-FoxO1 antibodies from another company (C-9 and H-128, Santa Cruz). In addition, we found FoxO1 underwent phosphorylation and proteasome-dependent degradation in mNKs (Fig. 6c). Deng et al. 55  (1) the C29H4 antibody against FoxO1 could harbour stronger affinity to probe FoxO1 in mNK cells; (2) for flow cytometry assays, staining buffers could impair proteasome function in mNKs abrogating the degradation of FoxO1. In our experimental assays, we used propidium iodide staining to exclude necrosis and late stage apoptotic cells. Additionally, we stained cells with Annexin V to analyse early apoptotic cells for flow cytometry assays. However, Deng et al. did not state the exclusion of dead cells in their assay systems 55 . Flow cytometry was the major assay for cell number calculations in their experiments. Dead cells could potentially impact the number counts of NK cells by flow cytometry analyses. Another possible explanation for this difference is that autophagy induction is a very dynamic response susceptible to different environments and body metabolism 56,57 . Different food, water, living environments and metabolic states of mice might cause different regulations for autophagy induction. We hypothesize that the mice with different feeding conditions might cause distinct autophagic regulations that affect NK cell development. Finally, subtle expression differences between NKp46-Cre lines could be another potential explanation for the differences.
In conclusion, we reveal that FoxO1-mediated autophagy in iNKs is required for NK cell development and NK cell-mediated innate immunity, which may provide a novel strategy to promote NK cell activity by manipulating FoxO1 for future clinical applications.  ;NKp46-Cre mice (CD45.1) were infected with lentivirus encoding WT or DDB-FoxO1 or vector control, and transferred into Rag1 À / À Il2rg À / À mice (CD45.2). FoxO1 in CD3 þ NK1.1 þ CD45.1 þ NK cells was analysed by western blotting. (b) Autophagy of iNKs from the indicated mice was analysed by immunoblotting. (c) ROS accumulation in NK cells expressing different FoxO1 constructions was analysed by DC-FDA staining. (d) After adoptive transfer for 3 weeks, BM CD45.1 þ NK cells from the indicated mice were analysed by flow cytometry. (e) IFN-g is impaired in DBD-FoxO1 NK cells. Intracellular IFN-g of the CD3 À NK1.1 þ CD45.1 þ splenic NK cells from the indicated mice was detected after MCMV infection. (f) Rag1 À / À Il2rg À / À mice transferred with WT, knockout or DBD-FoxO1 NKPs were infected with MCMV for 3 days. Specific lysis of NK cells were analysed as described in Fig. 7i. (g) The indicated mice were infected with MCMV followed by detection of virus titres. n ¼ 6 mice for each group. All data represent at least three separate experiments and calculated data are shown as means±s.d. **Po0.01; ***Po0.001. (h) Work model for FoxO1-induced autophagy during NK cell development. All data represent at least three independent experiments. For c,d,e and g, a two-way analysis of variance post hoc Bonferroni test was used. In vitro cytotoxicity assay. NK cells from mouse spleen were isolated. Yac1 cells (from ATCC) were labelled with 200 mCi 51 Cr at 37°C for 1 h and plated at 1 Â 10 4 cells per well. NK cells were incubated with Yac1 cells at the indicated E/T ratios for 4 h, and the supernatants were harvested for the detection of specific 51 Cr release using MicroBeta counter (PerkinElmer). Specific cytotoxicity was calculated as [(sample release-spontaneous release)/(maximum release-spontaneous release)] Â 100. Maximum release of Yac1 cells was measured by treatment with 2% TritonX-100.

Methods
Electron microscopy. For immuno-EM, 1 Â 10 6 NK cells at different stages form WT mice were isolated. and fixed with 4% paraformaldehyde and 0.05% glutaraldehyde for 2 h on ice and then cryo-sectioned by Tokuyasu method and mounted in copper grids followed by blocking in 1% BSA. Sections were stained with anti-LC3 (30 ng ml À 1 ) antibody (PM036, MBL) and further stained with 10 nM gold linked anti-rabbit IgG. Then the sections were fixed with 2.5% glutaraldehyde and stained with neutral uranyl acetate before coating with 2% methyl cellulose. Images were taken with a FEI Tecnai spirit transmission electron microscope.
RNA interference. shRNAs against Atg3, Atg7, Raptor or Rictor were constructed into pSIN-EF2 vector. Lentiviruses encoding RNA interference sequences were generated by transfecting pSIN-EF2-shRNA and packaging vectors into HEK293T cells, followed by ultracentrifugation concentration. The RNA interference sequences were previously described 25 .
MCMV infection. Mice were intraperitoneally injected with 1 Â 10 5 PFU MCMV (VR-1399, Smith strain from ATCC). After 72 h, mice were sacrificed for the collection of spleens. Intracellular IFN-g in NK cells was analysed by flow cytometry. For determination of virus titres, the spleens from infected mice were homogenated in 2 ml of complete cold medium followed by standard plaque-forming assay.
Statistical analysis. For statistical analysis, two-tailed unpaired Student's t-test and two-way analysis of variance post hoc Bonferroni test were used with Sigma Plot and GraphPad softwares. We calculated medians of fluorescence intensity by using FlowJo software for flow cytometry data, and using Image J software for confocal data. We calculated s.d. by using SigmaPlot for at least four independent experiments.
The uncropped data for blot and gel images can be found in Supplementary Figs 6-9.