Neuronal IFN-beta–induced PI3K/Akt-FoxA1 signalling is essential for generation of FoxA1+Treg cells

Neurons reprogramme encephalitogenic T cells (Tenc) to regulatory T cells (Tregs), either FoxP3+Tregs or FoxA1+Tregs. We reported previously that neuronal ability to generate FoxA1+Tregs was central to preventing neuroinflammation in experimental autoimmune encephalomyelitis (EAE). Mice lacking interferon (IFN)-β were defective in generating FoxA1+Tregs in the brain. Here we show that lack of neuronal IFNβ signalling is associated with the absence of programme death ligand-1 (PDL1), which prevents their ability to reprogramme Tenc cells to FoxA1+Tregs. Passive transfer-EAE via IFNβ-competent Tenc cells to mice lacking IFNβ and active induced-EAE in mice lacking its receptor, IFNAR, in the brain (NesCre:Ifnarfl/fl) result in defective FoxA1+Tregs generation and aggravated neuroinflammation. IFNβ activates neuronal PI3K/Akt signalling and Akt binds to transcription factor FoxA1 that translocates to the nucleus and induces PDL1. Conversely, inhibition of PI3K/Akt, FoxA1 and PDL1 blocked neuronal ability to generate FoxA1+Tregs. We characterize molecular factors central for neuronal ability to reprogramme pathogenic T cells to FoxA1+Tregs preventing neuroinflammation.

R egulatory T cells (T regs ) are essential in controlling inflammation. FoxP3 þ T reg cells have been shown to prevent overactivation of the immune system in systemic and central nervous system (CNS) inflammatory diseases 1,2 . We have recently identified a new T reg cell population, FoxA1 þ T reg cells, which played a pivotal role in regulation of CNS inflammation 3 .
The outcome of CNS inflammation depends on how immune cells, vascular cells and neurons interact 4 . Neurons have evolved several mechanisms to directly interact with T cells, to protect against development of chronic CNS inflammation. Neurons have been shown to promote T-cell apoptosis through Fas/FasL interactions 5 and to produce anti-inflammatory cytokines, such as transforming growth factor-b, which in turn prevents experimental autoimmune encephalomyelitis (EAE) 6,7 . The immunoregulatory activities of neurons and their central role in controlling CNS diseases, partly by regulating CNS inflammation, have recently received profound support 3,[7][8][9] . We have previously shown that neurons directly interact with pathogenic T cells; this subsequently led to the generation of FoxP3 þ T regs with capacity to ameliorate EAE 7 . Moreover, we reported that neurons were capable of converting EAE-inducing encephalitogenic T cells (T enc ) to anti-inflammatory FoxA1 þ T regs (ref. 3). Of note, mice genetically lacking interferon (IFN)-b signalling developed severe relapsing-remitting and demyelinating EAE, which subsequently was inhibited upon passive transfer of neuron-induced FoxA1 þ T regs cells 3 . In support, beneficial IFNb-treatment response in patients with multiple sclerosis (MS) was associated with increased generation of circulating FoxA1 þ T reg cells 3 .
Here we investigate whether neuronal IFNb is essential for their capacity to reprogramme T enc cells to become FoxA1 þ T regs and furthermore study the molecular signalling by which IFNb coordinates this central immunoregulatory function of neurons. We report that IFNb-mediated FoxA1 activation results in expression of neuronal PDL1. IFNb activates neuronal PI3K/Akt signalling. Consequently, Akt binds to transcription factor FoxA1 that translocates to the nucleus and induces PDL1. This is indispensable for neuronal ability to reprogramme T enc cells and generate FoxA1 þ T regs , which in turn are important to combat neuroinflammation.

Results
Lack of FoxA1 þ T regs leads to neuroinflammation in Ifnb À / À mice. Previously with active EAE, we identified FoxA1 þ T regs from CNS-infiltrating T cells in IFNb-competent wild-type (WT) (Ifnb þ / þ ) mice. In the contrary, FoxA1 þ T regs were not found in EAE mice genetically lacking IFNb (Ifnb À / À ) 3 . We have also reported that IFNb signalling in T cells was central for their FoxA1 þ T reg cell fate determination. To exclude the role of intrinsic IFNb signalling in T cells, we utilized IFNb-competent T enc cells (MBP 89-101 reactive T enc cells) to transfer adoptive EAE in Ifnb þ / þ or Ifnb À / À mice. As previously reported for the active EAE 3 , inducing adoptive transfer of EAE led to higher clinical disease score in Ifnb À / À mice (Fig. 1a). This was associated with defective FoxA1 þ T reg (TCR þ FoxA1 þ PDL1 þ ) cell generation in the CNS-infiltrating T cells in Ifnb À / À spinal cord in contrast to Ifnb þ / þ mice, day 30 post adoptive EAE (Fig. 1b); this indicates that despite utilizing the same WT-MBP 89-101 reactive T enc cells in both groups, the determining factor is the host endogenous IFNb. Interestingly, although Ifnb À / À mice developed significantly higher neuroinflammation apparent by elevated total number of infiltrating T cells in the spinal cord even during remission (Fig. 1c), they had significantly lower FoxA1 þ T regs compared with Ifnb þ / þ mice (Fig. 1d).
Of note, FoxA1 þ T regs were often detected not only in perivascular space (Fig. 1b) but also in parenchyma adjacent to neuronal soma (Fig. 1e); this provides neurons with possibility to interact with T cells and induce their FoxA1 expression both via molecules expressed on axons or on neuronal cell bodies, in the areas that T cells usually invade the CNS remote from neuronal bodies, as well as closed to the neuronal bodies upon migrating in the parenchyma, respectively. Moreover, a striking difference was observed in neuronal PDL1 expression in the cerebellum and spinal cord, which was lacking in Ifnb À / À mice (Fig. 1f,g). Of interest, although PDL1 was not detectable in Ifnb À / À mice, FoxA1 was expressed (Fig. 1g). Of interest, the significant increase of FoxA1 þ T regs in Ifnb þ / þ mice (Fig. 1b,d) was associated with neuronal co-expression of FoxA1 and PDL1, whereas lack of IFNb was associated with loss of neuronal PDL1 expression (Fig. 1g,h) and concomitant loss of FoxA1 þ T reg cells in the CNS of Ifnb À / À mice. These results suggested an important role for IFNb signalling in the CNS to regulate the generation of FoxA1 þ T reg cells.
Nes cre :Ifnar fl/fl mice lose ability to generate FoxA1 þ T regs . To address the role of neuronal IFNb-IFNAR signalling in regulation of CNS inflammation associated with FoxA1 þ T reg cell generation, Nes cre :Ifnar fl/fl mice were actively immunized with MOG 35-55 (ref. 10). Quantification of inflammatory cells infiltrating in the spinal cord of mice 35 days post immunization revealed that Nes cre :Ifnar fl/fl mice developed profound neuroinflammation compared with their WT corresponding, Ifnar fl/fl mice (Fig. 2a,b). Similar to mice lacking genomic IFNb, loss of brain IFNAR (IFNa/b receptor) signalling in Nes cre :Ifnar fl/fl mice resulted in the lack of FoxA1 þ T reg -cell generation associated with elevated neuroinflammation ( Fig. 2c-e). Of note, loss of neuronal IFNAR signalling led to the loss of PDL1 expression, while FoxA1 was still expressed by neurons (Fig. 2f,g). Taken together, these results strongly indicated that active neuronal IFNb-IFNAR signalling is central for converting T enc cells to FoxA1 þ T reg cells and hence for controlling neuroinflammation in the CNS.
Neuronal IFNb signalling is essential to generate FoxA1 þ T regs . Although neurons were found to be able to generate FoxA1 þ T regs (ref. 3) and the results above supported an active role for neuronal IFNb signalling, the molecular mechanisms by which neurons exert such a fundamental immunoregulatory property were not known. Here we investigated whether neuronal IFNb was involved in their T-cell-reprograming capacity. To exclude other CNS-resident cell contribution, we established primary neuronal cultures with high purity (Fig. 3a) (that is, mean±s.d. of 98.3±0.28%, n ¼ 3) and compared them with astrocytes and microglial cells in regard to their capacity to generate FoxA1 þ T reg cells upon interacting with T enc cells. As expected, cerebellar granular neurons (CGNs) changed phenotype of activated T enc cells and generated FoxA1 þ T regs , whereas other CNS innate antigen-presenting cells (APCs); microglia and astrocytes did not exert these properties (Fig. 3b). We have previously shown that neurons remain electrically active in culture 7 . To investigate whether neuronal electrical activity is required for the induction of FoxA1 þ T regs , neurons were silenced by tetrodotoxin (TTX). TTX is a potent toxin that specifically binds to voltage-gated sodium channels and blocks the flow of sodium ions through the channel, thereby preventing action potential generation and propagation 11 , and silences neurons 12 . TTX-silenced neurons lost the capacity to induce generation of FoxA1 þ T regs (Fig. 3c).
FoxA1   We next examined the suppressive capacity of these neuronalinduced FoxA1 þ T reg (nFoxA1 þ T reg ) cells in vivo and in vitro. As previously reported 3 , purified nFoxA1 þ T regs could induce significant cell death of activated T enc cells in vitro (Fig. 3d). To confirm their suppressive activities in vivo, purified nFoxA1 þ T regs were adoptively transferred to ears in a murine delayed-type hypersensitivity (DTH) model of tissue inflammation. Ears receiving nFoxA1 þ T regs had significantly less inflammation (Fig. 3e). These results supported suppressive and antiinflammatory properties of FoxA1 þ T regs in vivo and in vitro.
We finally examined the need for endogenous neuronal IFNb in generation of FoxA1 þ T regs . We observed that both CGNs and cortical neurons (CNs) from Ifnb þ / þ neurons were able to generate FoxA1 þ T regs , whereas neurons lacking IFNb (Ifnb À / À ) lost this capacity ( Fig. 3f-h). Treatment of Ifnb À / À neurons with recombinant (r)IFNb to reconstitute their defect, before co-culture with activated T enc cells, restored their ability to generate FoxA1 þ T regs (Fig. 3i). These data indicated that neuronal ability to convert pathogenic T enc cells to FoxA1 þ T reg cells depends on their endogenous IFNb signalling. IFNb share many functional similarities with IFNa, as they share the same receptor, IFNAR; however, they also differ in many of their functions including their different efficiencies as disease treatment. Although it is not well described how IFNb might   Percentage of FoxA1 þ T reg cells were analysed by FACS after co-culture of activated T enc cells with CG neurons, microglial or astrocytes, respectively. Graphs are mean±s.e.m., n ¼ 3. One-way analysis of variance (ANOVA) test was used, ***Po0.001. (c) Percentage of FoxA1 þ T regs were analysed by FACS after co-culture of T enc cells with CGN treated with or without TTX (1 mM) for 24 h before co-culture. Graphs are mean±s.e.m., n ¼ 3, one-way ANOVA test was used, ***Po0.001. (d) Percentage of 7AAD þ (dead) cells was analysed by FACS. Res, responder T cells were activated with plate-bound anti-CD3 (1 mg ml À 1 ) and soluble anti-CD28 (2 mg ml À 1 ) for 24 h, then co-cultured with neurons to generate FoxA1 þ T reg cells (nFoxA1 þ T reg ) for another 24 h. Graphs are mean±s.e.m., n ¼ 3. Unpaired Student's t-test was used, ***Po0.001.
regulate IFNa, it is previously reported that IFNb is required for production of IFNa in fibroblast 13 and we have not detected any compensatory mechanisms in neurons when only IFNb is deleted 9 . Although IFNa might have additional or differential effects independent of IFNb, this has not been observed related to the neuronal activity. Moreover, there are several alleles for Ifna, which result in the production of many active IFNa isoforms. This is imposing much more complexity to investigate the plausible role for IFNa isoforms in relation to the neuronal functions and hence not the focus of the current study.
Generation of FoxA1 þ T regs requires neuronal FoxA1 and PDL1. Previously, we have shown that the soluble cytokine IFNb has an anti-inflammatory role in the CNS, as evidenced by the extensive CNS inflammation, relapses and demyelination in Ifnb À / À mice with EAE 14 . In addition, it was shown that treatment of T cells with exogenous rIFNb was sufficient to induce FoxA1 þ T regs (ref. 3). To understand whether soluble IFNb produced by neurons directly affects T enc cells to change their phenotype to FoxA1 þ T regs , we utilized a transwell system to separate neurons and T cells in co-cultures, allowing free circulation of IFNb. Separation of neurons from T enc cells completely diminished FoxA1 þ T reg cell generation (Fig. 4a), which suggests that cell-tocell contact is necessary for neuronal conversion of pathogenic T enc cells to anti-inflammatory FoxA1 þ T regs . It was established that neuronal signaling molecule PDL1 inhibited cell cycle and induced apoptosis in brain tumours, namely glioblastoma 8 ; however, it was not known whether it could be involved in interaction between neurons and T cells. In T cells, IFNb treatment activates the transcription factor FoxA1 and thereby transcription of its downstream target gene Pdl1 (ref. 3). Earlier, FoxA1 has been reported to play a role in survival of dopaminergic neurons 15 ; however, it was unknown whether IFNb has an impact on neuronal regulation of FoxA1 or whether FoxA1 targets the Pdl1 gene in neurons. We hypothesized that upon IFNb treatment, neurons would activate FoxA1 to induce transcription of Pdl1. The lack of neuronal PDL1 expression both in Ifnb À / À and Nes cre :Ifnar fl/fl mice strongly supported such a scenario. We therefore examined the messenger RNA expression levels of Foxa1 and Pdl1 in CGNs, with or without rIFNb treatment. FoxA1 and Pdl1 mRNA were significantly increased after rIFNb stimuli (Fig. 4b,c). We also found remarkable differences in the basal levels of FoxA1 and PDL1 proteins in Ifnb þ / þ and Ifnb À / À neurons, which were further increased upon rIFNb treatment ( Fig. 4d-h). Moreover, treatment of Ifnb þ / þ neurons with rIFNb led to an increased translocation of FoxA1 from the cytoplasm to the nucleus (Fig. 4f,g). In addition, we identified a striking difference in the subcellular localization of FoxA1 in Ifnb þ / þ versus Ifnb À / À neurons: although FoxA1 was mainly found in the perinuclear space in Ifnb þ / þ neurons and translocated to the nucleus upon rIFNb treatment, in Ifnb À / À neurons FoxA1 was almost exclusively nuclear (Fig. 4g). Of note, the neuronal marker NF200 was also detected in the nucleus, the precise function of which is not well defined; an explanation for this observation might be the fact that neurofilaments are used as ducking or carriage proteins to transport signalling molecules.
We next investigated whether FoxA1 and its downstream target PDL1 were necessary for neuronal capacity to generate FoxA1 þ T regs . PDL1 was found to be much lower in Ifnb À / À neurons and rIFNb significantly upregulated PDL1 protein in both Ifnb þ / þ and Ifnb À / À neurons as previously reported by us 8 . To study the function of neuronal FoxA1 and PDL1, knockdown (KD) of neuronal FoxA1 and PDL1 was successfully achieved using small-interfering RNAs (siRNAs) (Fig. 4i,j,l). Importantly, neuronal FoxA1 KD led to reduced PDL1 expression (Fig. 4i,j), indicating that, as in T cells 3 , Pdl1 is a target gene for FoxA1 in neurons. Moreover, KD of FoxA1 or PDL1 in neurons before co-culture with T enc , significantly reduced their capacity to convert T enc cells to FoxA1 þ T regs (Fig. 4k,m). In support, blocking neuronal PDL1 signalling in the Ifnb þ / þ neurons (using a blocking anti-PDL1 antibody) or blocking PD1 on T enc cells (using a blocking anti-PD1 antibody) resulted in inhibition of FoxA1 þ T reg cell generation (Fig. 4n). Moreover, overexpressing PDL1 (using the pIRESII PDL1 plasmid) on Ifnb À / À neurons restored their ability to generate FoxA1 þ T reg cells (Fig. 4o,p). These results conclusively underscores that defective neuronal FoxA1 and PDL1 could account for the decreased or lack of ability to induce FoxA1 þ T regs , and that IFNb signalling is central for activation of neuronal FoxA1 and its target gene Pdl1, which in turn are indispensable for the function of neurons to interact and convert pathogenic T cells to FoxA1 þ T reg cells.
Next, we examined whether Akt physically binds to FoxA1 in regards to activate the pathway leading to PDL1 induction. In support of this scenario, when we overexpressed FoxA1 with a flag-tag in neuronal cell line N2A, we could immunoprecipitate (IP) FoxA1, leading to pull down of endogenous Akt upon co-IP (Fig. 5i), which was confirmed by IP of endogenous FoxA1 and co-IPing Akt upon rIFNb treatment of N2A and primary neurons (Fig. 5j,k), respectively. Our results provided direct evidence that IFNb-induced PI3K/Akt signalling is essential for the induction of FoxA1 and subsequently PDL1 in neurons.
Neuronal PI3K/Akt is essential to reprogramme T cells. We demonstrated that engagement of PI3K/Akt pathway was required for IFNb-mediated neuronal expressions of FoxA1 and PDL1. To establish whether the PI3K/Akt signalling is required for neurons to convert T enc cells to FoxA1 þ T regs , we blocked neuronal PI3K and Akt activities using either PI3K inhibitor or KD of Akt by siRNA silencing before neurons were co-cultured with encephalitogenic T enc cells. Inhibition of PI3K/Akt signalling in neurons significantly reduced the protein levels of FoxA1 and PDL1 (Fig. 5), and resulted in complete abrogation of their function to convert T enc cells to FoxA1 þ T regs (Fig. 6a-c). Collectively, our data revealed that endogenous IFNb signalling is central for the immunoregulatory function of neurons in which they convert pathogenic T cells to become FoxA1 þ T regs . We showed that IFNb triggers PI3K/Akt activation and Akt subsequently bound to FoxA1. This process in turn activated neuronal FoxA1 to induce PDL1. We also established that IFNb-induced PI3K/Akt pathway and its downstream FoxA1-mediated PDL1 expression were ARTICLE indispensable for the immunoregulatory function of neurons to generate FoxA1 þ T regs (Fig. 6d).

Discussion
The concerted action between immune cells and tissue-specific cells in their microenvironment is crucial for the outcome of inflammation. Autoimmune inflammation is a key element in numerous CNS diseases. The bidirectional interaction between autoreactive (CNS antigen-specific) immune cells with brainresident cells could therefore be central for the fate of autoimmune-mediated inflammatory diseases such as MS. CNS-specific autoimmunity is chiefly related to how autoreactive T and B cells cause damage and/or mutually interact with CNS-resident APCs to consequently maintain chronic activation of autoreactive cells and thereby neuroinflammation [21][22][23][24] . An important role of local neurons in regulation of brain immunity is conceivable; however, it has only recently received more attention. A direct role in immune regulation was reported for neurons to promote T-cell apoptosis 5 . In support, neuronal nuclear factor, nuclear factor-kB was also shown to regulate neuroinflammation 25 . Moreover, we reported several pivotal immunosuppressive properties of neurons, in which neurons interact with autoreactive T enc cells and convert them to regulatory FoxP3 þ T reg via production of transforming growth factor-b and expression of B7.1 (ref. 7), and/or converting them to adapt a FoxA1 þ T reg fate and anti-inflammatory functions 3 . Although the accumulative evidence is reinforcing the importance of neurons in controlling CNS immunity, current understanding of the essential basic molecular mechanisms of neuronal control of neuroinflammation is sparse. Here we investigated the required molecular signalling operative in neurons, which provides them with capacity to convert pathogenic T cells to the newly identified FoxA1 þ T reg cells and their impact on outcome of neuroinflammation. As mice lacking the Ifnb gene were deficient in generating FoxA1 þ T regs in the CNS 3 , we studied the function of endogenous neuronal IFNb when challenged with IFNb-competent T enc cells. Neurons defective in IFNb production were incompetent to convert T enc cells to FoxA1 þ T regs , both in vitro and in vivo, indicating a central function of this cytokine in neuronal immunoregulatory efficacies. In support, neuronal IFNb was shown to be critical for their anti-tumor functions to control glioblastoma 8 . Soluble IFNb was shown to exert anti-tumour activities 16,26 and it was sufficient to convert naive T cells to FoxA1 þ T regs (refs 3,27). However, blocking the direct cell-to-cell interaction between neurons and T cells prevented the conversion of T enc to FoxA1 þ T regs , which revealed that released soluble neuronal IFNb alone was not sufficient to convert T cells. These data suggested that endogenous neuronal IFNb exerted paracrine and autocrine effects to provide neurons with additional signalling key molecules. In support, lack of IFNAR on neurons was leading to the similar loss of function in Nes Cre :Ifnar fl/fl mice, which strongly displayed evidence that released soluble IFNb by neurons 9 requires active autocrine signalling, via its receptor, to synchronize neuronal immunoregulatory functions. Pleiotropic biological activities of IFNb was reported to be mediated by triggering different signalling pathways 16 . We demonstrated that the PI3K/Akt pathway was activated downstream of neuronal IFNb signalling. In agreement, PI3K has been shown to be responsible for several biological activities of IFNa/b in different types of cells even independent of JAK-STAT signalling [28][29][30] . Activation of PI3K and Akt were essential for induction of other downstream signalling molecules in neurons. We showed that Akt binds to FoxA1. As a transcription factor, the role of FoxA1 in neurodevelopmental processes and survival of dopaminergic neurons has been reported 15 , but no role was established related to immunoregulatory function of post-mitotic neurons, in particular related to IFNb signalling. Interestingly, in vivo, FoxA1 could be detected in neurons depleted of IFNb or IFNAR, but lack of active signalling did prevent it from targeting the Pdl1 gene, suggesting that activation of PI3K and Akt could modify FoxA1 activity. Indeed, insulin-like growth factor-I was reported to increase the stability of FoxA1 protein in MCF7 breast cancer cell line via Akt pathway 31 .
Our data suggest that activated Akt binds to FoxA1, which might mediate its translocation to the neuronal nucleus and upon binding to the Pdl1 promoter induces its production. In support, we have previously shown that translocation of FoxA1 to nucleus of T cells resulted in FoxA1-mediated transcription of the transmembrane signalling protein PDL1 (ref. 3). The engagement of PDL1 on T cells with its receptor, programmed death 1 on APCs has been shown to induce anti-proliferative effects 32 . Here we identified that PDL1 is essential for neurons to bind to PD-1 on T enc cells and convert them to FoxA1 þ T regs . IFNb is also produced by CNSresident glial cells 14,33 , which during neuroinflammatory process in EAE also express PDL1 (refs 34-36). Hence, this posts the question of why, unlike neurons, glial cells were incapable of converting T enc cells to FoxA1 þ T regs . It is plausible that these immunoregulatory capabilities of neurons are attributed to the fact that neurons, in contrast to glial cells, do not express major histocompatibility complex class II nor produce the predominant proinflammatory cytokines 7,37,38 , and therefore incapable of conducting a fullactivation signal in T cells, but rather exert tolerogenic signal and hence cause a different outcome for T enc cells. Understanding these Figure 4 | Generation of FoxA1 þ T reg cells requires neuronal FoxA1 and PDL1. (a) Percentage of FoxA1 þ T regs upon co-culture of T encs with CGNs, with or without transwell. Graphs are mean ± s.e.m., n ¼ 3. One-way analysis of variance (ANOVA) test was used, ***Po0.001. (b) Fold change of Foxa1 mRNA and (c) Pdl1 mRNA in CGNs treated with or without rIFNb (100 U ml À 1 ). Graphs are mean±s.e.m., n ¼ 3, two-way ANOVA test was used, *Po0.05 and **Po0.01. (d) Foxa1 and (e) Pdl1 mRNA in CGNs (left) and quantification (right). Graphs are mean ± s.e.m., n ¼ 3, Student's t-test was used, ***Po0.001. (f) Cytosolic (Cyto) and nucleus (Nu) fractions from CN (left). Graph shows normalized optical density (IOD) ratios between cytosolic/nucleus fractions of quantified FoxA1 protein bands (right). Graphs are mean ± s.e.m., n ¼ 2-4. Non-parametric Mann-Whitney test was used for comparison of treated groups with control group. *Po0.05 and **Po0.01. (g) CGNs with or without treatment with rIFNb. NF200 (green), FoxA1 (red) and DAPI (blue). Scale bars, 5 mm. (h) WB of FoxA1, PDL1 and vinculin in CNs extracts with or without treatment with rIFNb. (i) Representative FACS histogram of neuronal FoxA1 and PDL1 after Foxa1 siRNAKD in CGNs and (j) WB of FoxA1, PDL1 and vinculin after Foxa1 siRNAKD in CGNs. (k) Percentage of FoxA1 þ T regs after co-culture of T encs with CGNs after Foxa1 or control siRNAKD in CGNs. Graphs are mean ± s.e.m., n ¼ 3, Student's t-test was used, ***Po0.001. (l) Neuronal PDL1 after Pdl1 or UNC (Universal Negative Control) siRNAKD in CGNs. (m) Percentage of FoxA1 þ T regs upon co-culture of T encs with CGNs after PDL1 or UNC siRNAKD in Ifnb þ / þ CGNs. Graphs are mean ± s.e.m., n ¼ 3, Student's t-test was used, ***Po0.001. (n) Percentage of FoxA1 þ T regs upon co-culture of T encs with CGNs after PDL1 was blocked in Ifnb þ / þ neurons utilizing anti-PDL1 or PD1 was blocked in T encs utilizing anti-PD1 antibodies. Graphs are mean ± s.e.m., n ¼ 3, one-way ANOVA test was used, ***Po0.001. (o) Neuronal PDL1 after transfection of pIRES2-EGFP-PDL1 or pIRES2-EGFP in CGNs. (p) Percentage of FoxA1 þ T regs upon co-culture of T encs with overexpression of PDL1 in Ifnb À / À CGNs. Graphs are mean ± s.e.m., n ¼ 3, one-way ANOVA test was used, ***Po0.001.
pathways and the possible defects are imperative for designing future therapeutic approaches for chronic and progressive neuroinflammatory conditions such as progressive MS.

Methods
Mice. Ifnb À / À mice were backcrossed to C57BL6 or C57BL/10.RIII strains of mice over 20 generations 3 . The WT littermates were used (Ifnb þ / þ ) as controls. Mice were purchased from The Jackson Laboratroy, USA. The mice were bred and kept at conventional animal facilities at the University of Copenhagen. One-to 7-day-old newborn pups were used in neuron-culture experiments. Eight-to 12-week-old Ifnb À / À and WT mice 3 or nes Cre :Ifnar fl/fl (Ifnar gene targeting only in neuroectodermal cells) and their WT littermates (Ifnar fl/fl ) were used for inducing EAE 10 .
All experiments were performed in accordance with the ethical committees in Copenhagen, Denmark and approved by the respective Institutional Review Boards (2013-25-2934-00807). Regarding the age and gender, mice were equally allocated to experimental groups.  Figure 6 | Neuronal ability to generate FoxA1 þ T regs is IFNb-mediated PI3K-Akt-FoxA1 and PDL1 dependent. (a) Representative FACS plots of T enc cells after co-culture with CGNs, with and without Akt siRNA KD or PI3K inhibitor (Wortmannin in DMSO, 200 nm). FoxA1 þ T reg cells were gated on CD4 high PDL1 high cells. (b) FoxA1 expression was gated in CD4 high PDL1 high cells (R1) and in CD4 low PDL1 low cells (R2) and (c) quantified percentage of FoxA1 þ T regs . Graphs are mean±s.e.m., n ¼ 3, one-way analysis of variance (ANOVA) test was used, ***Po0.001. (d) Schematic drawing of how neuronal endogenous IFNb signalling leads to PI3K and Akt phosphorylation. Total and phosphorylated (p)Akt binds to FoxA1, FoxA1-Akt complex translocates to nucleus and binds to Pdl1 promoter, consequently resulting in PDL1 expression. This signal is essential for neurons to convert T enc cells to antiinflammatory FoxA1 þ T reg cells. In contrary, lack of endogenous IFNb signalling in Ifnb À / À neurons leads to insufficient phosphorylated PI3K/Akt signalling, defective FoxA1 mediated PDL1 expression and thereby inability to generate FoxA1 þ T reg cells.
Induction of neuroinflammation EAE. To establish neuroinflammation in the CNS, active or adoptive EAE 14 was induced in Ifnb À / À or their WT mice or in nes Cre :Ifnar fl/fl (ref. 10). Active EAE was induced with MBP 89-101 or MOG  , respectively 10,14 . Each animal was subcutaneously immunized in the base of the tail with 100 ml of a 1:1 emulsion of 150 mg of MOG  or 250 mg of MPB 89-101 (Schafer-Copenhagen) in PBS and complete Freund's adjuvant (CFA) containing 500 mg of Mycobacterium tuberculosis H37Ra (Difco). An intraperitoneal injection of 500 ng of pertussis toxin/PT (Bordetella pertussis; Sigma-Aldrich) dissolved in 100 ml of PBS was given on the day of immunization (day 0). Mice were observed for clinical signs of EAE every day. For adoptive transfer EAE, mice were irradiated (500 rad) and injected in the tail vein with a cell suspension of MBP 89-101 -specific T cells. At day 0 and 2, each animal was given an intraperitoneal injection of 500 ng of pertussis toxin. The reason for utilizing two different models of EAE either in C57BL/6 or C57BL/10RIII backgrounds was to address both chronic EAE induced by MOG  , which is commonly used because of the availability of many transgenic strains in the C57BL/6 genetic background, and to utilize a relapsingremitting EAE model induced by MBP 89-101 , which is the model for relapsingremitting MS in C57BL/10RIII background 10,14 . In addition, such an approach excludes the possible bias introduced by slight differences in the genetic backgrounds (C57BL/6 versus C57BL/10RIII) rather than the function of the target genes, here Ifnb and Ifnar.
CGNs cultures were obtained by decapitating a 7-day-old mouse, whereas CNs were dissected from 1-day-old pup brains. The brains were placed immediately in a 50 ml tube with Hank's balanced salt solution (HBSS) on ice, before dissecting in a Petri-dish containing 1 ml HBSS in a sterile hood. The cerebellum and cortex were dissected out and kept intact, while meninges were removed. A single-cell suspension was prepared by adding trypsin to a final concentration of 200 mg ml À 1 . DNAse and FCS were added to final concentrations of 0.12% and 0.5%, respectively, before incubation for 6 min. Finally, the tissue was mechanically dissociated to single cells using a fire-constricted Pasteur pipette and cells were seeded at a concentration of 4 Â 10 5 cells per ml. CGNs were cultured for 3-5 days and CNs for 21 days before co-culture with T cells.
Glial cell culture. Primary mixed glial cell cultures were established from the forebrain and cerebellum of 1-to 2-day-old B10.RIII (Ifnb þ / þ ) mice. The tissues were carefully dissected out and freed of meninges before being placed in HBSS solution supplemented with 1 mM pyruvate and 11 mM glucose. Tissue was then chopped into smaller pieces using a razor blade and incubated with 1% trypsin (Sigma-Aldrich) for 10 min at 37°C. Thereafter, DNase and FCS were added to a final concentration of 0.12% and 0.5%, respectively, and incubated for 8 min. Finally, the solution was mechanically dissociated to single cells using a fire-constricted Pasteur pipette. The cells were seeded at a concentration of 1.5 Â 10 5 cells per cm 2 in a 1:1 mix of DMEM and F12 medium (Invitrogen Life Technologies) supplemented with 0.16 mM ml À 1 penicillin, 0.03 mM ml À 1 streptomycin and 5% FCS. Medium was changed every third to fourth day. After 7 days in culture, oligodendrocytes were detached from the cell monolayer by shaking on an orbital shaker. After two to three passages, which initially consisted of a mixture of 20-30% Mac-1 þ microglia and 70-80% GFAP þ astrocytes, these cultures were further enriched. After 8 days in culture, the mixed glial cells were washed and vigorously shaken at 900-1,000 r.p.m. for 3 h on an orbital shaker. For microglial cultures, the floating cells were collected, washed and reseeded in 96-well plates (Nunc) at a concentration of 1 Â 10 4 cells per well. After adhering overnight, nonadherent or loosely attached cells were washed away. The adherent cells represented 497% pure microglial cells as determined by Mac-1 þ staining, with o3% being GFAP þ . For astrocyte cultures, the still adherent cells were trypsinized and reseeded in flasks and left to adhere for 30 min. Floating or loosely attached cells were recovered by mild shaking by hand and the adhesion for 1 h. Cells in the supernatant were thereafter collected, washed, and reseeded in 96-well plates (Nunc) at a concentration of 4 Â 10 4 cells per well. These cells were 496% pure astrocytes as determined by GFAP þ staining, o4% were Mac-1 þ cells. Both these cultures were then used for coculture with T cells.
Establishment of T enc cell lines. To generate T-cell lines, 8-to 12-week-old male B10RIII (Ifnb þ / þ ) were immunized in the flank and tail base with 200 ml of a 1:1 emulsion of 250 mg of MBP 89-101 in PBS and CFA containing M. tuberculosis H37Ra (Difco). MOG  T-cell line was generated by immunized with C57BL6 mice with 150 mg of MOG  . Draining lymph nodes were collected 10 days after immunization and a single-cell suspension prepared in PBS by passing through a sieve. Cells were washed and suspended in DMEM with Glutamax-1 (Gibco) supplemented with 10 mM HEPES buffer, 10% heat-inactivated FCS (Sigma), 100 U ml À 1 of penicillin, 100 mg ml À 1 of streptomycin and 50 mM 2-mercaptoethanol, to make complete DMEM (cDMEM). Cells were cultured as 5 Â 10 5 cells in round-bottomed 96-well culture plates (Nunc) in 200 ml of cDMEM, in a humidified 37°C atmosphere containing 5% CO 2 . T cells were stimulated for 4 days with 50 mg of MBP 89-101 . After a resting phase of 8 days in media supplemented with 800 pg ml À 1 of interleukin (IL)-2 (obtained from the supernatant of an IL-2 transfected X63 cell line), T cells were re-stimulated with 20 mg of peptide and irradiated APCs. APC were generated from spleen cells of syngeneic mice, prepared as described above for lymph nodes with an additional 0.84% NH 4 Cl treatment to lyse red blood cells, and irradiated with 3,000 rad before being used at a concentration ten times higher than the T cells. To expand highly specific T-cell lines, stimulation was repeated at intervals of 10-30 days. Between stimulations, the antigen-containing media was removed and the T cells kept in a resting state in cDMEM supplied with IL-2. The media was changed every fourth day. In all experiments, T-cell lines had gone through a total of four to ten stimulation rounds.
Co-culture of neurons with T cells. CGNs and/or CNs were prepared 7,8 and seeded at 4 Â 10 5 cells per ml in 96-or 24-well plates with neuronal media for 3 and 21 days, respectively. T-cell lines were re-stimulated with antigen and APCs for 48 h and are referred to as activated T enc cells. Neurons and activated syngeneic T enc cells were washed twice and co-cultured at a 1:1 ratio for 24 h, unless stated otherwise. MOG  T-cell line and MBP 89-101 T-cell line were co-cultured with neurons from C57BL6 mice and C57BL/10.RIII mice, respectively.
FACS staining and sorting using FACSAria. After washing in FACS buffer (2% FCS in PBS), cells were incubated with anti-Fc receptor Ab (24.G.2, our hybridoma collection) at 10 mg ml À 1 . Thereafter, cells were incubated with biotinylated, fluorescein isothiocyanate (FITC)-or phycoerythrin (PE)-labelled antibodies. For intracellular staining, cells were fixed and permeabilized using BD Cytofix/Cytoperm or using fixation and permeabilization solutions from Human Treg Flow Kit. All antibodies were used at 1-5 mg ml À 1 and were allowed to bind for 20 min on ice. The antibodies used were as follows: APC or Cells were acquired with an FACSAria (BD Biosciences) using the FACSDiva software for acquisition after exclusion of duplets. Dead cells were discriminated in all staining using the LIVE/DEAD Fixable Dead Cell Stain Kit for 405 nm excitation (L34955, Invitrogen). FlowJo 8.8.6 (Tree Star) was used for further analysis.
FoxA1 þ T reg cells were gated by CD4 high PDL1 high and FoxA1 þ expression.
siRNA silencing. Accell SMART pool siRNA was purchased from Dharmacon (catalogue number E-046238-00, Thermo Scientific) and was introduced into neurons according to the manufacturer's protocol. Briefly, SMART pool siRNA combines four different siRNAs to reduce off-target effects. The Accell siRNA is also designed for optimal delivery to hard-to-transfect cells and no transfection reagents were required to introduce the siRNAs. The following SMART pool Accell siRNAs from GE Healthcare were used: Foxa1 (E-046238-00-0005), Cd274 (E-040760-00-0005) and Akt1 (E-040709-00-0005); for detail siRNA sequences, please see Table 1.
In some experiments, CNs were cultured for 7 days and stimulated with rIFNb for 30, 60 or 120 min. Subsequently, the CNs were fractionated into a cytosolic and nucleus fractions 39 . Fractions were loaded on 4-12% gels according to cytosolic protein quantification determined by Pierce BCA protein assay kit. The protein concentrations loaded in the cytosolic fractions were around three times higher than in the nucleus fraction. Antibodies used were as follows: Polyclonal rabbit anti-FoxA1 (1:1,000, Abcam catalogue number ab23738), monoclonal mouse anti GAPDH (1:20.000, Abcam catalogue number ab9484) and polyclonal rabbit anti TBP (1:500, Santa Cruz catalogue number sc-204) were used.
Images have been cropped for presentation. Full-size images are presented in Supplementary Figs 1-4).
intensities in the other separate channels were measured and scored as positive or negative using the automated software analysis.
For Infiltrating cells count, H&E-stained images were converted from NDPI version (NanoZoomer) to TIFF format with 'NDPI tools' plugin on ImageJ. Subsequently, the infiltrated areas were manually detected and a customized macro for counting cells was run for automatic cells count.
Manual counting: For some IF staining images that were manually quantified ImageJ was utilized. Negative controls were utilized as reference. DAPI channel was selected with the ImageJ function find maxima to automatically count the total number of nucleus (cells). Next, visualizing only DAPI and far red channel, TCR either NF200 (far red channel) were manually counted with ImageJ manual counting function, and then, co-expression with red channel (FoxA1) for the double positive and green channel (PDL1) for triple positive were evaluated.
Statistical evaluations. Statistical evaluation was performed using GraphPad Prism (GraphPad Software Inc.). Student's t-test was used for two groups comparison, one-way analysis of variance with post Tukey's multiple comparisons test and two-way analysis of variance with post Sidak's multiple comparisons test were used for more than two groups comparison. Non-parametric Mann-Whitney test was used for EAE score and IF staining quantification data. A value of Po0.05 was considered significant.
Data availability. All relevant data are available from the authors on request.