IL-12 sensing in neurons induces neuroprotective CNS tissue adaptation and attenuates neuroinflammation in mice

Interleukin-12 (IL-12) is a potent driver of type 1 immunity. Paradoxically, in autoimmune conditions, including of the CNS, IL-12 reduces inflammation. The underlying mechanism behind these opposing properties and the involved cellular players remain elusive. Here we map IL-12 receptor (IL-12R) expression to NK and T cells as well as neurons and oligodendrocytes. Conditionally ablating the IL-12R across these cell types in adult mice and assessing their susceptibility to experimental autoimmune encephalomyelitis revealed that the neuroprotective role of IL-12 is mediated by neuroectoderm-derived cells, specifically neurons, and not immune cells. In human brain tissue from donors with multiple sclerosis, we observe an IL-12R distribution comparable to mice, suggesting similar mechanisms in mice and humans. Combining flow cytometry, bulk and single-nucleus RNA sequencing, we reveal an IL-12-induced neuroprotective tissue adaption preventing early neurodegeneration and sustaining trophic factor release during neuroinflammation, thereby maintaining CNS integrity in mice.

Interleukin-12 (IL-12) is a potent driver of type 1 immunity.Paradoxically, in autoimmune conditions, including of the CNS, IL-12 reduces inflammation.The underlying mechanism behind these opposing properties and the involved cellular players remain elusive.Here we map IL-12 receptor (IL-12R) expression to NK and T cells as well as neurons a nd o li go de nd ro cytes.Conditionally ablating the IL-12R across these cell types in adult mice and assessing their susceptibility to experimental autoimmune e nc ep ha lomy elitis revealed that the neuroprotective role of IL-12 is mediated by neuroectoderm-derived cells, specifically neurons, and not immune cells.In human brain tissue from donors with multiple sclerosis, we observe an IL-12R distribution comparable to mice, suggesting similar mechanisms in mice and humans.Combining flow cytometry, bulk and single-nucleus RNA sequencing, we reveal an IL-12-induced neuroprotective tissue adaption preventing early neurodegeneration and sustaining trophic factor release during neuroinflammation, thereby maintaining CNS integrity in mice.

Article
https://doi.org/10.1038/s41593-023-01435-zmice, which lack Il12rb2 in all hematopoietic cells (and endothelial cells) but did not show exacerbated disease compared to littermate control mice (Fig. 1k-n and Extended Data Fig. 1n,o).Together, our findings allow the elimination of immune cells (including CNS-resident microglia 11 ) from the list of potential candidates for the IL-12-mediated protective phenotype in EAE and indicate that the protective features of IL-12 are mediated by neuroectodermal cells.

Neurons and oligodendrocytes sense IL-12 in mouse and human
We next sought to capture and localize potential sensors of IL-12, resident to the murine CNS.To do so, we used immunostaining for β-galactosidase (β-gal) to detect cells expressing IL-12Rβ2 protein within the steady-state CNS of Il12rb2 LacZ/LacZ reporter mice, coupled with multiplexed RNA fluorescence in situ hybridization (RNAscope) for Il12rb1 and Il12rb2 mRNA.IL-12R was expressed in the neuroectoderm, specifically in neurons and oligodendrocytes, already at steady state (Fig. 2a-h and Extended Data Fig. 2a,b), consistent with a recently published single-nucleus RNA-sequencing (snRNA-seq) dataset of adult mouse hippocampus 12 .Both β-gal/IL-12Rβ2 immunoreactivity (Fig. 2a-h and Extended Data Fig. 2a,b) and Il12rb2 mRNA transcripts (Fig. 2i,j) were localized to NeuN-positive and Rbfox3/Map2-positive neuronal cells, respectively.The β-gal-positive signal was also present in CC1 + myelin-forming oligodendrocytes and mature oligodendrocytes (MOLs; Fig. 2g).While Ιl12rb1 mRNA was more strongly expressed in Sox10-positive oligodendrocytes (Fig. 2i,j), neither astrocytes nor microglia expressed the IL-12R subunits, suggesting their inability to sense IL-12 (Fig. 2f,j).Il12rb1/ Il12rb2 gene expression patterns were similar in the inflamed murine CNS (Extended Data Fig. 2c-f) where we further validated our findings by qPCR of populations subjected to fluorescence-activated cell sorting (FACS; Fig. 2k) and fluorescence-activated nuclei sorting (FANS; Fig. 2l).Apart from infiltrating T cells and NK cells, IL-12R subtypes were strongly expressed on oligodendrocytes and neurons, thereby representing the primary radioresistant non-hematopoietic CNS-resident cells expressing the IL-12R complex at steady state and during MOG-induced neuroinflammation.
We next evaluated the presence of the IL-12-sensing molecular machinery in the human CNS by examining snRNA-seq datasets consisting of 66,432 nuclei from white matter (WM) lesional tissue of deceased individuals with progressive MS and age-matched and sex-matched controls who had died from non-neurological causes 13 (Extended Data Fig. 2g-i).Analogous to the mouse, human neurons and oligodendrocytes expressed transcripts encoding IL-12R subtypes (Extended Data Fig. 2h,i).We next interrogated a second independent human dataset (n = 39,579 nuclei) of whole MS tissue sections 14 (comprising cortical gray matter (GM), demyelinated lesions and underlying subcortical WM) of individuals with secondary progressive MS (n = 12) and controls (n = 9; Extended Data Fig. 2j,k).IL12RB2 was expressed across all excitatory neuron clusters (upper, intermediate and deep cortical layers) at similar levels in MS and control samples (Extended Data Fig. 2k).We further confirmed neuronal IL-12Rβ2 protein expression in the cortex of individuals with MS by immunohistochemistry (Extended Data Fig. 2l).Finally, when we exposed primary mouse neurons (Extended Data Fig. 2m,n) or oligodendrocytes (Extended Data Fig. 2o,p) to recombinant IL-12, we observed a rapid increase of STAT4 phosphorylation-a hallmark of IL-12 receptor signaling.Together, these data demonstrate that in both mice and humans CNS-endogenous neuroectodermal cells are capable of sensing and transducing IL-12 signaling intracellularly.

Neuroectodermal IL-12R signaling attenuates neuroinflammation
We then asked whether the ability of the neuroectoderm to sense and respond to IL-12 was functionally linked to IL-12-induced tissue protection in EAE.We crossed Il12rb2 fl/fl mice to a Nestin Cre/+ strain (Fig. 3a).
unique subunits p35 and p19 engaging with IL-12Rβ2 and IL-23R, respectively, while sharing the common p40 subunit, which binds IL-12Rβ1.Dual targeting of IL-12 and IL-23 signaling, however, has not shown efficacy in a phase II clinical trial in MS therapy 7 , possibly due to interfering with two opposing underlying disease mechanisms simultaneously.In fact, higher IL12RB2 expression is correlated with a lower risk of relapse in relapsing-remitting MS patients and better response to MS therapy 8 .Thus, IL-12 may attenuate neuroinflammation in MS not only in mice but also in humans.The mechanism behind IL-12-mediated tissue protection and immune regulation in MS has never been resolved.
To reveal the mechanistic underpinnings of how IL-12 limits immunopathology during CNS autoimmunity, we first identified and then conditionally targeted IL-12R-expressing cell types in neuroinflammation.We found that ablation of the IL-12R in the neuroectoderm (that is, astrocytes, neurons and oligodendrocytes) alone phenocopied the hypersusceptibility to experimental autoimmune encephalomyelitis (EAE) observed in germline Il12rb2 knockout mice.We here describe the discovery of a neuroprotective loop triggered by IL-12-producing CNS-infiltrating monocyte-derived cells (MdCs), which implicated autocrine and paracrine survival and trophic factor release by distinct IL-12-sensing neuronal populations in neuroinflammation.

Immune cells are dispensable for IL-12-mediated tissue protection
To identify the mechanism by which IL-12 limits immunopathology in CNS neuroinflammation, we first generated a conditional strain of IL-12-specific receptor subunit b2: Il12rb2 fl (ref.9).We crossed Il12rb2 fl/fl mice with CMV Cre mice to generate mice lacking Il12rb2 in all tissues, hereafter termed Il12rb2 del/del mice (Fig. 1a).These mice were hypersusceptible to myelin oligodendrocyte glycoprotein (MOG)-induced EAE, as previously reported in mice lacking functional IL-12 (refs.5,6; Fig. 1b,c).While the time to clinical disease onset was comparable between mice with or without Il12rb2 (Fig. 1b), the absence of IL-12 signaling was associated with significantly higher maximal and cumulative EAE scores (Fig. 1c and Extended Data Fig. 1a).
As helper T (T H ) cells are the main drivers of MOG-induced EAE, we asked whether deleting the IL-12R across aβ T cells would recapitulate the exacerbated clinical symptoms seen in Il12rb2 del/del mice.We crossed Il12rb2 fl/fl mice with Cd4 Cre/+ mice (Fig. 1d-g), generating progeny in which Il12rb2 was deleted in T cells.By analyzing IL-12-induced phosphorylation of STAT4 (Extended Data Fig. 1b) and qPCR (Fig. 1g), we confirmed the deletion of Il12rb2 in T cells.Surprisingly, Cd4 Cre/+ Il12rb2 fl/fl mice and Il12rb2 fl/fl littermates developed EAE of similar clinical severity (Fig. 1f and Extended Data Fig. 1c) suggesting that T cells are not responsible for the IL-12-mediated tissue protection.Similarly, Ncr1 Cre/+ Il12rb2 fl/fl mice showed comparable clinical outcomes to the littermate control group suggesting no major contribution of natural killer (NK) cells to the IL-12-induced immunoregulatory circuit (Extended Data Fig. 1d-h).
As testing for equivalence was not conclusive for this dataset, we confirmed that NK cells had no measurable impact on EAE disease development, adding to the controversy of the role of NK cells in CNS autoimmunity 10 .Here, mice treated with an anti-NK1.1 antibody to delete NK cells showed similar clinical symptoms compared to the PBS-treated control group (Extended Data Fig. 1i-m).Together, our findings eliminate the two most obvious targets within the IL-12R-expressing lymphocyte compartment.
We next generated bone marrow chimeras in which Il12rb2 was expressed either in the adult hematopoietic stem cell-derived systemic immune compartment or in radioresistant cells of the CNS (Table 1) (Fig. 1h-j).We found that the loss of IL-12R in immune cells did not affect the clinical outcome of EAE.In contrast, mice lacking Il12rb2 expression in radioresistant cells phenocopied the Il12rb2 del/del strain (Fig. 1h-i) accompanied by significantly more infiltration of CD4 + T cells and MdCs into their CNS (Fig. 1j).This was further validated in Vav1 Cre/+ Il12rb2 fl/fl Article https://doi.org/10.1038/s41593-023-01435-z The Nestin Cre/+ strain permits the simultaneous targeting of both neurons and oligodendrocytes, which we identified as the relevant IL-12 receptor-bearing cell types in the CNS.Taking into account previous studies raising concerns about the targeting promiscuity of the Nestin Cre/+ strain outside the neuroectoderm 15 , we confirmed that ablation of Il12rb2 occurred in neurons and oligodendrocytes (Fig. 3b) but not in splenic and CNS-infiltrating T cells and NK cells (Extended Data Fig. 3a-c).Strikingly, Nestin Cre/+ Il12rb2 fl/fl mice phenocopied the EAE hypersusceptibility of the Il12rb2 full knockout strain (Fig. 3c,d and Extended Data Fig. 3d,e).We excluded potential effects of Cre
We next compared the immune cell landscape between Nestin Cre/+ Il12rb2 fl/fl and Il12rb2 fl/fl littermate control mice at peak disease using flow cytometry (FC) of CNS leukocytes (gated on live CD45 + singlets), coupled with dimensionality reduction (uniform manifold approximation and projection (UMAP)) and FlowSOM clustering (Fig. 3e-i and Extended Data Fig. 3f).We observed a significantly higher CNS infiltration of CD4 + T cells and MdCs (Fig. 3g), as well as a reactive microglia signature (Fig. 3h,i), consistent with the exacerbated EAE phenotype in the absence of IL-12R signaling in cells of the neuroectoderm.Quantitative immunofluorescence confirmed severe inflammation and CNS demyelination in Nestin Cre/+ Il12rb2 fl/fl mice (Fig. 3j).Together, our data indicate that IL-12 signaling by cells of the neuroectoderm drives immunomodulatory and/or tissue-protective processes that attenuate inflammatory demyelinating CNS pathology in EAE.

IL-12 induces neuroprotection and trophic factor release
We next sought to precisely disentangle the molecular underpinnings of this yet-unperceived neuroimmune cross-talk instructed by IL-12.Importantly, for this analysis, we opted for a time point when IL-12 is already present in the CNS but has not (yet) translated in clinical differences between Nestin Cre/+ Il12rb2 fl/fl and Il12rb2 fl/fl control mice.Hence, we measured the concentration of IL-12 in brain and spinal cord lysates during EAE induction and disease (Fig. 3k).We found that IL-12 was barely detectable for almost a week after immunization, but began to increase from the day of clinical onset (10 days post immunization (d.p.i.); Fig. 3k), likely coinciding with the influx of MdCs.In fact, compared to microglia, infiltrating MdCs were the dominant source of IL-12 in the inflamed CNS (Fig. 3l,m).As fragile neuroectodermal cells are typically underrepresented in most single-cell omic approaches, we performed snRNA-seq of FANS-isolated Hoechst + nuclei from the cerebellum, brainstem and cervical spinal cord of Nestin Cre/+ Il12rb2 fl/fl mice and their Il12rb2 fl/fl littermates (n = 4 per group) at disease onset (10 d.p.i.), where clinical symptoms and also CNS protein levels of IL-12 were comparable between the two genotypes (Fig. 3n,o).
After quality control, doublet exclusion and batch integration, snRNA-seq yielded a total of 133,151 single-nucleus transcriptomic profiles, among which 1,453 median genes per nucleus and 2,428 median unique molecular identifiers (UMIs) per nucleus were detected (Fig. 4 and Extended Data Fig. 4a,b).Community detection of the combined dataset yielded 21 clusters that were assigned to diverse neuronal, glial and other cell types on the basis of known lineage marker genes (Fig. 4a,b and Extended Data Fig. 4c).Among these clusters were 5 populations of neurons reflecting the cellular composition of the cerebellum 16 and 5 additional neuronal clusters classified by neurotransmitter type.We also captured 6 clusters of glial cells corresponding to oligodendrocytes, astrocytes, microglia, oligodendrocyte precursor cells and Bergmann glia.Non-neural cells encompassed infiltrating mononuclear phagocytes, vascular and leptomeningeal cells, ependymal cells, choroid plexus epithelial cells and neuroblasts.Among all recovered nuclei, 85% were neurons, and 15% were glial cells (Fig. 4b), consistent with previous descriptions 16,17 .When comparing cluster abundance between the two genotypes, we noted an increased frequency of both resident and infiltrating mononuclear phagocytes in Nestin Cre/+ Il12rb2 fl/fl mice (Fig. 4c).
When we looked at the distribution of Il12rb2 mRNA transcripts across the defined cell clusters, we found that it was most abundant in neuronal populations (Extended Data Fig. 4d), while Il12rb1 transcripts were enriched within MOL subsets, confirming our initial observations.Next, we assessed how the transcriptional profile of the neuroectoderm differed in the presence or absence of IL-12 receptor signaling in EAE by interrogating differentially expressed genes (DEGs).This revealed pronounced transcriptional alterations particularly in granule cells, excitatory neurons, MOL1, cholinergic neurons, microglia, molecular layer interneurons (MLIs) and MOL2, suggesting a dominating action of IL-12 on these populations (Fig. 4d-f, Extended Data Fig. 4e and Supplementary Table 1).
Of note, an enrichment in genes involved in major histocompatibility complex (MHC) class I signaling (B2m, H2-K1, H2-D1, H2-Q7, H2-T23, Tap1, Nlrc5), Stat1 and the complement component C4b, which we refer to as the hyperinflammatory gene signature in Nestin Cre/+ Il12rb2 fl/fl mice, was observed across nearly all cell types (Fig. 4g and Extended Data Fig. 4e).These features were previously referred to as a diseaseassociated signature in oligodendroyctes 21 and indicate how decreased neuroprotection in Nestin Cre/+ Il12rb2 fl/fl mice may foster inflammation.
To test whether the neuroprotective features induced by IL-12 might be crucial for the tissue integrity of the neuroectoderm, we compared our signature to a publicly available dataset from a genome-wide in vivo knockdown screen 22 .Indeed, we noted a decreased CNS essential gene signature corresponding to genes that were found to be relevant for both short-term and long-term survival across oligodendrocytes and neurons in Nestin Cre/+ Il12rb2 fl/fl mice (Fig. 4h and Supplementary Table 2).Notably, ablation of each of the CNS essential genes led to spontaneous neurodegeneration in mice 22 .
To evaluate whether the induction of a neuroprotective gene module and the concomitant suppression of hyperinflammatory features in the presence of IL-12 signaling could be validated across experimental mice, we next performed pseudo-bulk analysis in conjunction with DESeq2 differential expression analysis (Extended Data Fig. 5a,b and Supplementary Table 3).This approach has been reported to reduce false discovery rates (FDRs) and increase external validity in recent benchmarking studies 23 .We confirmed differential expression for 27 genes essential for neuronal survival (gene module: Erbb4, Ntf3, Nectin3, Ntng2, Ptprt) and function in granule cells and MLIs (Extended Data Fig. 5a), as well as 44 features validating the increased signature of inflammation (gene module: C4b, Hif3a, Stat1, B2m, H2-K1, H2-D1, Cxcl10) in Nestin Cre/+ Il12rb2 fl/fl mice across all clusters (Extended Data Fig. 5b).
Neurodegeneration and neuroinflammation can be caused by pathophysiological intercellular communication.To explore how IL-12 affects intercellular communication by IL-12R-expressing neurons and oligodendrocytes, we analyzed ligand-receptor interactions that are altered upon loss of IL-12 using the NicheNet algorithm 24 (Fig. 4i,j and Extended Data Figs.5c and 6a-e).We intentionally defined sender cells by their ability to sense and respond to IL-12 (that is, specific neuronal and oligodendrocyte subpopulations), while recipients included all cells captured by snRNA-seq (IL-12R expressing and not expressing).While DEG analysis can only predict direct IL-12-mediated transcriptional changes, this approach allowed us to decipher IL-12-induced, indirect changes.Two common patterns of intercellular communication shaped by IL-12 arose: neuron-intrinsic protection counteracting neurodegeneration and trophic factor release supporting cells of the neuroectoderm (Fig. 4i,j and Extended Data Figs.5c and 6a-e).
The most pronounced theme observed in the NicheNet analysis of MOL populations (that is, MOL1 and MOL2) was a dysregulated trophic factor environment in the absence of IL-12 receptor signaling (Extended Data Fig. 6a,c,e).Trophic signals shaping the neuron-oligodendrocyte cross-talk that were affected by IL-12 included ADAM17-ERBB4 and neuron-derived NECTIN-FGFR2/FGFR3, NRG2-ERBB3/ERBB4 and FGF10-FGFR2 (Extended Data Fig. 6a).Taken together, our data  suggest IL-12 to be critically involved in neuroprotection and shaping the trophic factor milieu within the inflamed CNS.We observed that the dominant source of trophic factors were neurons (Fig. 4j and Extended Data Figs.5c and 6a,e) highlighting their function as critical homeostatic gatekeepers.Conversely, dysregulation or loss of trophic factor release by neurons in the absence of IL-12 may further propagate the degeneration of neurons, oligodendrocytes and possibly other cell types.

Neurons mediate IL-12-induced tissue protection in experimental autoimmune encephalomyelitis
To test whether IL-12 signaling in neurons alone was sufficient or whether oligodendrocytes are also needed to sense IL-12 to promote tissue integrity during inflammation, we crossed the Il12rb2 conditional allele to the Plp1 CreER strain (Fig. 5a) that allowed for the depletion of Il12rb2, specifically in oligodendrocytes while sparing neurons (Fig. 5b,c and Extended Data Fig. 7a).In contrast to the Nestin Cre/+ Il12rb2 fl/fl mice, which lacked Il12rb2 expression in both populations, Plp1 CreER/+ Il12rb2 fl/fl transgenic mice did not display exacerbated clinical symptoms compared to littermate control mice (Fig. 5d,e and Extended Data Fig. 7b).These data solidify the notion that neurons are the main executors of the IL-12-driven neuroprotective tissue adaptation in EAE.
Our transcriptomic data, however, do not fully exclude the possibility that changes in the neuronal compartment are secondary to the different levels of inflammation observed between Nestin Cre/+ Il12rb2 fl/fl and Il12rb2 fl/fl mice, despite displaying similar clinical scores at the onset of disease.Unfortunately, neuroprotective features cannot be tested at steady state due to the absence of MdC infiltration (and therefore IL-12) in the homeostatic CNS (Fig. 3l,m).To overcome these limitations and to solidify the findings of the snRNA-seq dataset, we performed bulk RNA-seq of primary murine neurons stimulated with recombinant IL-12 (Fig. 5f-j).Principal component analysis (PCA) revealed a dose-dependent response to IL-12 in the neuronal transcriptome (Fig. 5g).We identified 1,878 significantly DEGs (P ≤ 0.01, log ratio ≥ 0.5; Supplementary Table 4); several of the upregulated genes (1,091 in total) have been linked to promoting neurogenesis (Fpr1, Mbd1), neuroprotection (Timp1, Angptl4) and enhancing remyelination (Timp1, Matn2; Fig. 5h).IL-12 suppressed the expression of Mmp12, whose inhibition in neurons attenuates axonal pathology and degeneration 27 and induced the expression of Serpina3n, encoding a granzyme B inhibitor, which prevents direct neuronal killing by T cells and NK cells and induces neuroprotection in EAE 28 (Fig. 5h).Strikingly, IL-12 triggered the neuronal secretion of trophic factors for oligodendrocytes including those encoded by Lif and Il33 (Fig. 5h), confirming our in vivo findings.Modular gene coexpression network analysis revealed ten groups of genes (modules) whose expression was jointly altered by IL-12 (Fig. 5i).One of the predicted modules (M2 or neuroprotective module), which was induced by IL-12, harbored gene hubs (that is, genes strongly coexpressed and/or interacting with other genes or both) linked to neurogenesis and neuroprotection (Fgf2, Vegfa, Egr1), axonal regeneration (Tubb3), oligodendrocyte precursor cell mobilization (Ccl2), endogenous myelin repair (Vcam1, Fgf2) and neuronal metabolism (glycolysis (Tpi1), copper ion transport (Slc16a3); Fig. 5i).Furthermore, IL-12 directly triggered pathways known to exert pro-survival effects in CNS pathology; for instance, hypoxia 29 and the mTORc1 (ref.30) signaling pathway (Fig. 5j).We performed qPCR for selected genes in neuronal cultures from Il12rb2 fl/fl versus Il12rb2 del/del mice (Extended Data Fig. 7c,d) and confirmed that elements of the M2 neuroprotective (that is, Fgf2, Vegfa) module depended on Il12rb2 expression in neurons (Extended Data Fig. 7e).Collectively, these results demonstrated a direct neuroprotective and neurotrophic effect of IL-12 in neurons.

Discussion
The discovery of IL-12 coincided with understanding its powerful effects on NK cells and, only later, T cells as well 31 .IL-12 is firmly linked with type 1 immunity 32 and is widely held as a major driver of inflammation.However, more than a decade later it became clear that IL-23, which shares the p40 subunit with IL-12, is the pivotal factor for the induction of autoimmune tissue pathologies, including neuroinflammation.IL-23 is required for the formation and maintenance of encephalitogenicity in T H cells 33 .
The observation that IL-12-deficient mice develop stronger tissue inflammation across multiple preclinical models of chronic inflammatory diseases (as reviewed in ref. 34) has represented somewhat of a paradox.Cytokines, however, are pleiotropic and specific cytokine receptors are often expressed across different cell types.Cytokine networks involve not only immune cells, but also various cellular entities outside the hematopoietic compartment 35 .A prime example is psoriasiform skin inflammation where IL-12 triggers tissue protection via epithelial cells 9 .Equally, Murphy et al. 36 demonstrated that germline mutants specifically lacking IL-23 (Il23a −/− mice) were completely resistant to joint autoimmune inflammation, whereas the selective absence of IL-12 (Il12a −/− mice) led to hyperinflammation and exacerbated bone pathology.In vivo administration of IL-12 in a model of systemic sclerosis alleviated disease pathology, implying immunoregulatory actions on IL-12-responsive fibroblasts 37 .Therefore, one could speculate that the spectrum of IL-12's actions outside the hematopoietic compartment extends far beyond what we currently know.
Here, we identified two CNS-endogenous neuroectodermal IL-12 sensors: neurons and oligodendrocytes.Whereas conditional deletion of Il12rb2 in immune cells was dispensable for IL-12-mediated tissue protection, the loss of IL-12 sensing in neuroectodermal cells, and more specifically, in neurons, led to hyperinflammation during EAE, phenocopying the Il12rb2 germline mutation that has been reported before 5 .The tissue pathology of Nestin Cre/+ Il12rb2 fl/fl mice was characterized by  2) between Nestin Cre/+ Il12rb2 fl/fl and Il12rb2 fl/fl mice across cell types.i, Schematic of the receptor-ligand interaction analysis (NicheNet; created with BioRender.com).Senders were defined by their ability to sense IL-12 (expression of Il12rb1 and/or Il12rb2 transcripts) and the magnitude of their differential gene expression (cutoff > 35 DEGs).Both autocrine and paracrine ligand-receptor interactions were considered.j, Circos plots showing links between unique ligands from senders (ribbon color indicates subcluster of origin for each ligand; multiple sender cells are indicated by black ribbon) and predicted associated DEGs in granule cells.Transparency indicates interaction strength, and the ribbon thickness is proportional to the ligand's regulatory potential.Heatmap (right) displaying potential receptors expressed in granule cells associated with each predicted ligand.Differential gene expression in e-h was tested using a two-sided Wilcoxon rank-sum test and applying a Benjamini-Hochberg correction and is presented in Supplementary Table 1.OPC, oligodendrocyte progenitor cell; VLMC, vascular and leptomeningeal cell.UBC, unipolar brush cells.severe CNS tissue damage coupled with markedly increased inflammatory infiltrates, which might be a consequence of increased tissue damage 38 .Alternatively, IL-12 might directly interfere with leukocyte attraction through neurons as we for instance detected increased Ccl2 expression in neurons of Nestin Cre/+ Il12rb2 fl/fl mice.In fact, the concept of the neuroectoderm as an active immunomodulator, rather than passive bystander of inflammation, has emerged in diverse forms of T cell-mediated autoimmune brain pathologies 39,40 .Neurons, in particular, can influence CNS immune responses in both protective 41 and detrimental 42 ways.In addition to regulating immune responses, neurons have been demonstrated to be key players in counteracting CNS tissue damage via trophic factor release in the context of diverse brain ailments 43 .We found that IL-12 induced transcriptional adaptation of the neuroectoderm and specifically instructed neuroprotective cues and trophic factor support by neuronal cell populations within the inflamed CNS (Extended Data Fig. 8).Members of the fibroblast growth factor (FGF) family, for instance, were consistently upregulated by IL-12 and have been suggested to stimulate remyelination and enhance neurogenesis 44 among its broad repertoire of CNS-protective functions 45 .
The loss of IL-12, was accompanied by a global hyperinflammatory transcriptomic signature in both neurons and oligodendrocytes, including genes linked to antigen presentation (for example, B2m) and type 1 signature (for example, Stat1) indicating pathogenic processes that further potentiate inflammation 39 .Neuronal MHC class I expression mediates a broad repertoire of functions in the steady-state CNS 46 , while the upregulation of MHC class I and Stat1 renders neurons prone to CD8 + cytotoxic T cell-mediated degeneration 42 and impedes CNS repair 47 , suggesting increased neuronal vulnerability in the absence of IL-12.Conversely, a recent report suggested a detrimental role of IL-12 in Alzheimer's disease-like disease pathology 12 .The opposing effects in acute neuroinflammation versus chronic neurodegeneration are best explained by differences in the pathophysiology of the respective model.These include the cytokine context (inflammatory milieu including IL-23 and GM-CSF in EAE, which is absent in Alzheimer's disease-like APP/PS1 mice 35 ), the sources and amount of IL-12 (high in MdCs in EAE and lower in glial cells in APP/PS1 mice), the nature of tissue damage (multifocal disseminated in EAE and more restricted to the cerebrum in APP/PS1 mice) as well as CNS region-specific transcriptomic changes of neuroectodermal cells.IL-12, therefore, joins the ranks of other molecules with opposing functions in inflammation versus degeneration of the CNS (for example, GM-CSF 48,49 ).
A neuroprotective effect of IL-12 on neurons, uncoupled from the effect of inflammation, however, has also been reported in diverse in vivo and in vitro settings.For instance, administration of IL-12 after EAE induction ameliorates the disease outcome in mice 50,51 .Similarly, IL-12 was shown to exert a therapeutic effect in an in vivo model of spinal cord injury.Local administration of IL-12 into the injured sites induced the expression of neurotrophic factors (BDNF, NT-3) in CNS lesions, thereby enhancing endogenous axonal regeneration and remyelination 52 .IL-12 also promoted regeneration in primary mouse superior cervical ganglion cell cultures, where IL-12 enhanced neurite outgrowth 53 .
In summary, our study sheds light on a 20-year-old conundrum by elucidating the cellular targets and molecular underpinnings of a neuroimmune cross-talk orchestrated by IL-12 in neuroinflammation (Extended Data Fig. 8).The observation that the IL-12 receptor is expressed in both mouse and human neuroectoderm raises the possibility that IL-12 might trigger a similar reparative mechanism also in the human CNS.Similar to the findings in mice, we found that cortical GM excitatory neurons-healthy, but also underlying meningeal inflammation in MS-expressed IL12RB2, possibly suggesting a state of readiness to sense IL-12.Thereby, IL-12 sensing during disease onset might counteract early neurodegenerative pathways as recently described in human MS 54 .The neuroprotective function of IL-12 might contribute to an inflammation-induced emergency break crucial to prevent immunopathology.We hypothesize that such a mechanism applies not only in autoimmunity but also in other inflammatory CNS diseases, for instance, during pathogen-induced meningitis or encephalitis.The fact that IL-12 confers neuroprotection may provide a new avenue of disease intervention in both neuroinflammatory and degenerative diseases.
Age-matched, sex-matched (male and female) and litter-matched 8-to 12-week-old mice were used for all in vivo experiments.For primary neuronal and oligodendrocyte cultures, pups were euthanized between postnatal day 0 (P0) and 4 (P4).For the sample size calculation of mouse experiments, we used an alpha = 0.05 and a power of 0.8 (no blocking for age, sex and litter in the power calculation).We used data from previous experiments for the mean and standard deviation to calculate the standardized effect size using G Power.We determined the required sample size to be able to see a 30% difference in the cumulative score (area under the curve) for disease course and 70% by qPCR/gene expression for assessing the targeting efficacy/specificity of our strains.

Human samples
Brain tissue from three individuals with MS was obtained from the archives of the Institute of Neuropathology at the University Hospital Zurich, Switzerland.Informed consent for autopsy was given by the next of kin in all cases.Case series do not need institutional review board approval according to Swiss legislation.

Experimental autoimmune encephalomyelitis induction
EAE was induced by injection of 200 μg of MOG epitope MOG 35-55 (RP10245, Genscript) emulsified in complete Freund's adjuvant (CFA; 231131, BD).All mice received two subcutaneous injections into the flank and a single intraperitoneal injection of 50-100 ng pertussis toxin (PT; List Biological Laboratories, 179A).The mice received a second injection of PT on 2 d.p.i.Of note, by reducing the concentration of PT, Il12rb2 fl/fl mice developed a mild disease, whereas Il12rb2 del/del mice were still hypersusceptible, but within animal welfare guidelines.EAE clinical scores were defined as follows: no detectable signs of EAE, 0; tail limp at distal end, 0.5; entirely limp tail, 1; limp tail and hind limb weakness (occasional grid test positive), 1.5; unilateral partial hind limb paralysis (one leg constantly falls through the grid), 2; bilateral partial hind limb paralysis, 2.5; complete bilateral hind limb paralysis, 3; complete bilateral hind limb paralysis and partial forelimb paralysis, 3.5; complete paralysis of fore and hind limbs (moribund), 4. In case the clinical symptoms couldn't be assigned to one or the other score, we used in-between scoring intervals of 0.25, which allowed a more accurate description of the clinical symptoms.Withdrawal criteria included clinical score ≥ 3 for more than 7 d, clinical score ≥ 3.5 for more than 3 d, weight loss > 25%, as well as pain symptoms.

Tamoxifen treatment
Tamoxifen (Sigma) was dissolved in ethanol and corn oil to 25 mg ml −1 and administered in 200-μl doses via oral gavage (5 mg per dose).

Generation of bone marrow chimeras
Host animals were lethally irradiated with a split dose of 2 × 550 rad at a 24-h interval before receiving an intravenous injection of 5 × 10 6 donor bone marrow cells.The immune reconstitution period before EAE induction was 6 weeks.Transplantation of Il12rb2-null (Il12rb2 del ) bone marrow into wild-type (Il12rb2 fl ) mice was used to assess the effect of Il12rb2 depletion in hematopoietic cells (that is, peripheral and CNS-invading leukocytes).Transfer of wild-type (Il12rb2 fl ) bone marrow into Il12rb2-null (Il12rb2 del ) mice was used to assess the effect of IL-12 sensing in radioresistant and non-hematopoietic (CNS resident) cell types, that is, neuroectoderm, tissue-resident microglia and tissue-resident memory T (T RM ) cells.Wild-type (Il12rb2 fl ) into wild-type (Il12rb2 fl ) and Il12rb2-null (Il12rb2 del ) mice into Il12rb2-null (Il12rb2 del ) mice served as positive and negative controls, respectively (Table 1).

Natural killer cell depletion
To deplete NK cells in vivo, C57BL/6 mice were injected intraperitoneally with 200 μg of anti-NK1.1 monoclonal antibody (BP0036, clone PK136, BioXcell) at days 0 and 2 after EAE induction, and after every 6 d.The percentage of NK cell depletion was assessed in the CNS and spleen.

Cell isolation from the adult mouse CNS
Mice were euthanized through CO 2 asphyxiation and transcardially perfused with ice-cold PBS.
To enrich for and improve oligodendrocyte viability for downstream FACS sorting and RNA isolation, CNS tissue was subjected to enzymatic digestion using papain (LS003126, Worthington) for 30 min at 37 °C and the enzymatic reaction stopped with ovomucoid trypsin inhibitor (LS003086, Worthington).Myelin debris was again removed by 30% Percoll (P4937, GE Healthcare) gradient centrifugation.

Flow cytometry
Before surface labeling, cells were incubated with purified anti-mouse CD16/CD32 (clone 93, BioLegend) for 10 min on ice to prevent nonspecific binding of primary antibodies.Single-cell suspensions were then directly incubated with the primary surface antibody cocktail in PBS for 20 min at 4 °C.Following a wash step with PBS (350g, at 4 °C) the cells were then incubated in the secondary surface antibody cocktail (fluorochrome-conjugated streptavidin) for 20 min at 4 °C.For intranuclear labeling with Ki67 and Foxp3, cells were fixed and permeabilized utilizing the Foxp3/Transcription Factor Staining Buffer Set (eBioscience) according to the manufacturer's instructions and subsequently incubated in the intranuclear antibody mixture overnight at 4 °C.Anti-mouse fluorochrome-conjugated monoclonal antibodies used in this study were: CD8 (564920, BD Biosciences, clone 53-6.Sample data were acquired on a 5-laser Aurora spectral analyzer (Cytek Biosciences; SpectroFlo v2 Software), FACS Symphony A5 (BD Biosciences; FACS Diva Software v9.1) or LSR II Fortessa (BD Biosciences; FACS Diva Software v9).Flow cytometry data were analyzed using FlowJo (version 10.8.0, Tree Star) and Rstudio (version 4.0.1).Doublet discrimination was performed based on SSC-A/H, FSC-A/H and dead cell exclusion by using a Fixable Viability Kit (Near-IR staining, dilution 1:1,000, BioLegend).

Fluorescence-activated cell sorting
Cell sorting was performed using a BD FACS Aria III (FACSDIVA Software v9).Anti-mouse fluorochrome-conjugated monoclonal antibodies Doublet exclusion was performed based on SSC-A/H, FSC-A/H and dead cell exclusion by using a Fixable Viability Kit (Near-IR staining, dilution 1:500, BioLegend).

Quantitative RT-PCR
Total RNA was isolated from sorted cells and primary cultures using either the RNeasy Micro Kit (74004, Qiagen) or the QuickRNA Microprep Kit (R1051, Zymo Research) according to the manufacturer's instructions.Complementary DNA (cDNA) was synthesized using the M-MLV Reverse Transcriptase (28025013, Invitrogen) and qRT-PCR was performed on a CFX384 Touch Real-Time PCR Detection System (Bio-Rad) using SYBR Green (Bio-Rad).Total RNA was isolated from sorted Hoechst + Olig2 + and Hoechst + NeuN + nuclei using the RNeasy Micro Kit (74004, Qiagen).Nuclear RNA was reverse transcribed to cDNA with the RevertAid H Minus First Strand cDNA Synthesis Kit (K1631, Thermo Fisher Scientific) according to the manufacturer's instructions.Gene expression was calculated as 2 −ΔCt relative to Pol2 as the endogenous control.
Horseradish peroxidase (HRP)-coupled goat anti-rabbit secondary antibodies (ab6721, Abcam; dilution 1:20,000) and donkey anti-mouse secondary antibodies (715-625-151, Jackson ImmunoResearch, Alexa Fluor 680; dilution 1:10,000) were incubated for 30 min at RT, and membranes were washed again five times for 5 min in TBST.Fluorescence signals were captured using the Odyssey CLx Imager.SuperSignal West Pico Chemiluminescent Substrate (34579, Thermo Fisher Scientific) was applied to visualize HRP-labeled antibodies and developed using the Fujifilm Luminescent Image Analyzer LAS-1000 plus & Intelligent Dark Box II (Fujifilm).Western blot membranes were re-probed for pStat4 and Stat4 using a mild stripping protocol from Abcam: briefly, membranes were incubated for 5-10 min with mild stripping buffer (200 mM glycine, 20 mM SDS, 0.01% Tween 20, pH 2.2) followed by 10 min incubation with PBS and 5 min incubation with TBST.This process was repeated twice.Images were processed and analyzed using ImageJ version 1.53t 60 .

CNS tissue homogenization and protein extraction
Murine brain and spinal cord were isolated as described above and snap-frozen in liquid nitrogen for storage at −80 °C.For protein extraction, samples were subjected to mechanical homogenization in Cell Lysate Buffer [(1 mM EGTA, 1 mM EDTA, 1% Triton X-100, TBS buffer pH 7.5 (20 mM Tris, 150 mM NaCl)].All buffers were kept on ice and the final Cell Lysate Buffer was supplemented with cOmplete Mini EDTA-free Protease Inhibitor Cocktail (11836170001, Roche, one tablet per 25 ml) and PhosSTOP (4906845001, Roche, one tablet per 10 ml) immediately before use.The initial homogenization was performed mechanically (two metal beads per sample) using a TissueLyserLT (85600, Qiagen) homogenizer (50 oscillations per second for 10 min), followed by subsequent incubation on ice for 30 min and centrifugation (full speed, 10 min, 4 °C).Protein concentrations of each sample were determined using the Pierce BCA Protein Assay Kit (23225, Thermo Fisher Scientific) according to the manufacturer's protocol using the Photometer Tecan Infinite 200 M (Tecan).

ELISA analysis
Enzyme-linked immunosorbent assay (ELISA) to detect IL-12/IL-23 total p40 (DY499, R&D) was performed according to the manufacturer's instructions.All samples were analyzed in duplicate.Absorption was read at 450 nm and 570 nm (for wavelength correction) on a microplate reader (Infinite 200M, Tecan) and analyzed using the Magellan Software (Tecan).

Immunofluorescence and imaging of murine brain tissue
Mice were euthanized through CO 2 asphyxiation and transcardially perfused with ice-cold PBS.Brains were carefully removed, fixed in 4% PFA (11762.00500,Morphisto) for 24 h at 4 °C, rinsed in PBS and then incubated in 30% sucrose in PBS at 4 °C for 24-72 h.Each brain hemisphere was embedded in Cryo Embedding Medium (41-3011-00, Medite) and frozen on dry ice.Free-floating brain sections were cut at a thickness of 35-45 μm using a Hyrax C60 cryostat (Zeiss).Sections were washed three times with PBS and incubated in 10% normal goat serum (PCN5000, Thermo Fisher Scientific) and 0.5% Triton X-100 (Sigma-Aldrich, T8787-100ML) in PBS (2 h, RT) for blocking and permeabilization.Sections that were labeled with antibodies derived from mouse were additionally blocked for another 45 min at RT Article https://doi.org/10.1038/s41593-023-01435-zwith the M.O.M. Mouse Ig Blocking Reagent (90 μl in 2.5 ml PBS; BMK-2202, Vector Laboratories).Sections were incubated with the primary antibody cocktail overnight at 4 °C in 5% normal goat serum and 0.5% Triton X-100 in PBS.

Quantitative histopathological analysis of spinal cords from mice with experimental autoimmune encephalomyelitis
Mice were euthanized through CO 2 inhalation and transcardially perfused with ice-cold PBS.Spinal columns were carefully dissected out and fixed in 4% PFA (11762.00500,Morphisto) for 24 h at 4 °C, rinsed in PBS followed by decalcification in 0.5 M, pH 8, EDTA solution (A3145,0500, Axonlab) at 4 °C for five consecutive days while rotating.The spinal cords were then dissected into three parts (cervical, thoracic and lumbar) and cryoprotected in 30% sucrose in PBS at 4 °C for 24-72 h.Each spinal cord fraction was embedded coronally in Cryo Embedding Medium (41-3011-00, Medite) and frozen on dry ice.Frozen sections were cut at a 20-25-μm thickness using a Hyrax C60 cryostat (Zeiss).Subsequent blocking, staining steps and mounting were performed as described above.For the quantification of demyelination and inflammation, sequential spinal cord sections cut 200 μm apart were fluorescently labeled with rabbit anti-Iba1 (019-19741, Wako; dilution 1:200) and FluoroMyelin Green Fluorescent Myelin Stain (F34651, Thermo Fisher Scientific).For automated imaging, fluorochrome-labeled sections were acquired on a Zeiss Axio Scan.Z1 Slidescanner (Zen 2 software, blue edition) using a ×20 objective lens.The following fluorescence filters were used: AF405, AF488, AF546 or AF594 and AF647.The semiautomated quantification was performed on 20-30 images of n = 4 mice per group.Acquired images were cropped in Fiji to remove all pixels outside the spinal cord, and cropped images were then processed in Ilastik (v1.3.3) 61.
Ilastik was trained to recognize lesions (DAPI bright, fluoromyelin low), WM (DAPI low, fluoromyelin high), GM (DAPI low, fluoromyelin low) and background (no signal), and to calculate the probability that each pixel belonged to each category.These probabilities were exported to Fiji as 8-bit images using the default algorithm.The areas of lesions and WM were then measured, and the percentage of demyelination was calculated as follows: 100 × area of lesions / (area of lesions + area of WM).We report the average percentage demyelination of each mouse over the total number of sections.

Bulk RNA isolation of primary neuronal cultures
Dissociated cerebellar cells were prepared and maintained as described above.At DIV 14, primary neuronal cells were stimulated with 30 ng ml −1 or 100 ng ml −1 of Recombinant Murine IL-12p70 (210-12, PeproTech) for 18 h.Cerebellar neurons were harvested and lysed in 300 μl RLT buffer of the RNeasy Micro Kit (74004, Qiagen) and total RNA was extracted according to the manufacturers' protocol.

Library preparation, cluster generation and next-generation sequencing
The quality of the isolated RNA was determined with a Fragment Analyzer (Agilent).The TruSeq Stranded mRNA (Illumina) was used in the succeeding steps.Briefly, total RNA samples (100-1,000 ng) were subjected to poly(A) enrichment and then they were reverse transcribed into double-stranded cDNA.The cDNA samples were fragmented, end-repaired and adenylated before ligation of TruSeq adaptors containing unique dual indices for multiplexing.Fragments containing TruSeq adaptors on both ends were selectively enriched with PCR.The quality and quantity of the enriched libraries were validated using the Fragment Analyzer (Agilent).The product is a smear with an average fragment size of approximately 260 bp.The libraries were normalized to 10 nM in Tris-Cl 10 mM, pH8.5 with 0.1% Tween 20.The NovaSeq 6000 (Illumina) was used for cluster generation and sequencing according to standard protocol.Sequencing configuration produced single-end reads of 100 bp.

Bulk transcriptome analysis of primary neuronal cultures
The raw reads were aligned to mouse genome build GRCm39 using the STAR aligner v2.7.8a, with FeatureCounts used to calculate read counts per gene based on GENCODE gene annotation release M26.Counts were counts-per-million normalized, log 2 transformed and only genes were retained if expressed at one count per million in at least two samples.PCA was computed using the BiocGenerics package after variance stabilizing transformation using the DEseq2 package 62 (1.38.3) and visualized using ggplot2.DEGs between conditions were computed Article https://doi.org/10.1038/s41593-023-01435-zusing gene-wise negative binomial generalized linear models as implemented in the edgeR package 63 (v3.40.2) and applying a Benjamini-Hochberg correction to account for multiple testing.Heat maps of DEGs were generated using the pheatmap package.
Differential analysis of modular gene coexpression was performed using the R implementation of CEMiTool using default parameters 64 .For gene-set enrichment analysis and interactome integration, murine gene names were converted to human orthologs using biomaRt.Gene-set enrichment analysis of gene modules was performed using the hallmark gene-set collection from MSigDB 65 .
Interactome data from the STRING database version 11 was integrated after filtering on interactions with a minimum score of 200 (ref.66) to obtain gene coexpression and interaction networks for each module.

Single-nuclei preparation and fluorescence-activated nuclei sorting
Single-nuclei suspensions were prepared as described before 12 .Mouse cerebellum and brainstem (of one brain hemisphere) and cervical spinal cord (C1-C2) were collected from adult male mice (8-12 weeks old) and immediately snap-frozen in liquid nitrogen and stored at −80 °C until further processing.Nuclei were isolated with the EZ PREP lysis buffer (NUC-101, Sigma).Tissue samples were homogenized using a glass dounce tissue grinder (D8938, Sigma; 25 strokes with pastel A, 25 strokes with pastel B) in 2 ml of ice-cold EZ prep lysis buffer and incubated on ice with an additional 2 ml of ice-cold EZ PREP lysis buffer.During incubation, 1 μΜ of Hoechst (H3570, Thermo Fisher Scientific) dye and 40 U μl −1 of RiboLock inhibitors (EO0382, Thermo Fisher Scientific) were added to the homogenate.Following incubation, the homogenate was filtered through a 30-μm FACS tube filter.Nuclei were sorted based on the fluorescence Hoechst signal using a BD FACS Aria sorter III with an 85-μm nozzle configuration at 4 °C, directly into PBS + 4% BSA + RiboLock inhibitors (40 U μl −1 ; EO0382).As CNS nuclei vary strongly in size, no doublet exclusion was performed based on FSC or SSC to avoid bias against nucleus size.Nuclei were then counted based on brightfield image and Trypan Blue staining (15250061) using a Neubauer counting chamber and a Keyence BZX-710 microscope.

Droplet-based single-nucleus RNA sequencing
RNA-seq of single nuclei was performed in two independent batches (batch 1, n = 1 Nestin Cre/+ Il12rb2 fl/fl and n = 1 Il12rb2 fl/fl ; batch 2, n = 3 Nestin Cre/+ Il12rb2 fl/fl and n = 3 Il12rb2 fl/fl ).Immediately after extraction, 17,500 sorted nuclei per sample were loaded onto a Chromium Single Cell 3′ Chip (10x Genomics) and processed for the single-nucleus cDNA library preparation (Chromium Next GEM Single Cell 3′ Reagent Kits v3.1 protocol).Around 50,000 reads per nucleus were sequenced using the Illumina NovaSeq 6000 no. 1 platform according to the manufacturer's instructions without modifications (R1 = 28, i7 = 10, i5 = 10, R2 = 90).Preparation of cDNA libraries and sequencing were performed at the Functional Genomics Center Zurich.Cell Ranger software (v6.0.2) was implemented for library demultiplexing, barcode processing, fastq file generation, gene alignment to the mouse genome (GENCODE reference build GRCm39) and UMI counts.We implemented the include-introns option for counting intronic reads, as the snRNA-seq assay captures unspliced pre-mRNA as well as mature mRNA.For each sample, a Cell Ranger report was obtained with all the available information regarding sequencing and mapping parameters.All samples were merged into a matrix using Cell Ranger (cellranger --aggr function).

Quality control and data pre-processing
Starting from the filtered gene-cell count matrix produced by Cell Ranger's built-in cell calling algorithms, we proceeded with the SCANPY (v1.8.2) workflow in Python 67 .Doublet exclusion was used individually for each sample using the Scrublet 68 package with variable thresholds.Quality-control processing was performed individually for each batch.
Low-quality nuclei were filtered out based on the number of unique genes (<500 and <200 in batch 1 and batch 2, respectively), total UMIs (<6,000) and fraction of mitochondrial counts (<5%).Genes present in less than three nuclei were removed.Counts were log1p transformed normalized to 10,000 counts and scaled for visualization purposes.Individual batches were integrated into a combined SCVI 69 model consisting of two layers and 30 latent dimensions using the 3,000 most variable features.

Dimensionality reduction, clustering visualization and cell-type identification
Dimensionality reduction using UMAP was used within the SCANPY framework.For this, the neighbor graph was constructed from the combined SCVI latent space.Community detection using Leiden was applied for cell-type classification.For assigning clusters to cell types, we defined a list of marker genes per cluster and ranked genes applying the Wilcoxon rank-sum test.The assignment of cell-type identity to clusters was based on known linage markers in line with previously published snRNA-seq 16 studies and atlases 70,71 .

Differential gene expression and downstream analysis
For differential expression between Il12rb2 fl/fl and Nestin Cre/+ Il12rb2 fl/fl mice, all genes per cell type were tested.DEGs were identified using a non-parametric Wilcoxon rank-sum test and Benjamini-Hochberg correction was applied to correct for multiple testing.For pseudo-bulk analysis, the sum of the transcripts for all cells within a given cluster of a sample was computed.Sample-level differential expression was carried out using the DEseq2 package 62 .Gene signatures were manually curated (Supplementary Table 2) and were computed as described before 72 .Bar graphs, volcano plots and lollipop plots were drawn in ggplot2.Heat maps were drawn using the pheatmap package.Dot plots were drawn using the Seurat framework 73 , in SCANPY 67 or using ggplot2.Violin plots and scatterplots were drawn in SCANPY.

Analysis of ligand-receptor interactions
Ligand-receptor interaction analysis was carried out using NicheNet 24 .Potential ligand-receptor pairs were restricted to pairs in which the ligand is expressed in at least 10% of the sender cells and the receptor is present in at least 10% of receivers.Sender cells were defined based on their ability to express Il12rb2 and demonstrate transcriptional changes after genetic deletion using Nestin Cre/+ Il12rb2 fl/fl mice (MOL1, MOL2, excitatory neurons, cholinergic neurons, granule cells and MLIs).For each receiver cell, NicheNet predicted the ligands that best explained DEGs (adjusted P value < 0.05 and abs(log2FC) > 0.5) upon Il12rb2 deletion.Top-ranking predicted ligands and the corresponding target genes for each receiver cell type were visualized in circos plots using the circlize package.Dot plots showing the respective ligand expression across sender cells were generated in ggplot2.Heat maps of top-ranking predicted ligands in sender cells and corresponding receptors in recipients were generated using the pheatmap package.

Reanalysis of single-nucleus RNA sequencing of multiple sclerosis tissue
Publicly available gene expression matrix data from Absinta et al. (accessed at GSE180759; five individuals with progressive MS and three age-matched and sex-matched non-affected, non-dementia controls) 13 were reanalyzed for the expression of IL12RB1 and IL12RB2.We downloaded the available gene expression matrix, and downstream analysis was performed by implementing the Seurat v4 R package 73 .
In brief, the filtered raw count matrix was log-normalized within each nucleus and included n = 66,432 nuclei.The top 2,000 variable genes calculated by Seurat were used to harmonize and integrate sample datasets by individual (FindIntegrationAnchors and Integrate-Data functions).On scaled data, linear dimension reduction (PCA) was performed, and the top 30 principal components were implemented https://doi.org/10.1038/s41593-023-01435-zfor the unsupervised clustering.The assignment of cell type was based on the original clustering annotation provided.
We further downloaded raw snRNA-seq data available (Sequence Read Archive accession number PRJNA544731) from ref. 14, in which nuclei were extracted from entire MS tissue sections (n = 12 cases) and compared to similarly processed control brain tissue (n = 9 cases).We processed the raw data, starting from the library demultiplexing, as described above for our own data.Cell Ranger software was implemented for library demultiplexing, barcode processing, fastq file generation, gene alignment to the human genome (GRCh38-3.0human genome with the intronic reads option) and UMI counts.Each sample was individually filtered based on Seurat guidelines for snRNA-seq, and only nuclei with <5% of mitochondrial contamination and between 200 and 2,500 genes expressed were retained for further analysis, reducing the probability of doublets.
The Seurat function SCTransform (v2) 74 was used to normalize the datasets, FindIntegrationAnchors was used to identify anchors between datasets and IntegrateData was used to run an integrated analysis.DEGs were defined as P < 0.05 (P adjusted for multiple comparisons) in ≥25% of cell populations and showing, on average, a >0.25-fold difference (log scale) between groups of cells.

Fluorescence-activated nuclei sorting for bulk RNA isolation
Nuclei of murine brains and spinal cords were isolated and sorted as described above.For specific labeling of neuronal and oligodendrocyte nuclei, the homogenates were directly stained with Hoechst, NeuN (ab190195, clone EPR12763, Abcam; dilution 1:200) and Olig2 (ab225100, clone EPR2673, Abcam; dilution 1:200).Hoechst + NeuN + and Hoechst + Olig2 + populations were directly sorted into RLT Lysis buffer (74004, Qiagen) and further processed according to the manufacturer's instructions.

High dimensional analysis of flow cytometry data
Raw flow cytometry data were pre-processed in FlowJo (Tree Star).Compensated and cleaned data were exported to Rstudio version 4.0.1.After transformation and percentile normalization, UMAPs 75 were generated using the package 'umap' version 0.2.7.0, and FlowSOM 76 version 2.6.0 metaclusters were merged and annotated based on the median expression profile and their localization on the UMAP.Heat maps were generated using the pheatmap' package version 1.0.12.All remaining plots were generated using the 'ggplot2' package version 3.3.5.

Statistical analysis
Statistical analysis was carried out using Prism 9 (GraphPad Software).Statistical significance of in vivo experiments (clinical scores over time) was determined with a repeated-measures two-way analysis of variance (ANOVA) or mixed-effects model (restricted maximum likelihood) when values were missing, followed by Bonferroni or Sidak multiple-comparison test.To test if the data were normally distributed, the Shapiro-Wilk test or Kolmogorov-Smirnov test were used and the quantile-quantile plots were examined.Group means were compared with two-tailed, unpaired t-tests, Mann-Whitney tests and one-way ANOVA or Kruskal-Wallis tests, corrected for multiple comparisons using the Dunn's or Bonferroni test.
Statistical analysis of CNS infiltration of different cell types between different groups was carried out in Rstudio (version 4.0.1)using t-tests.P values below 0.05 were considered as significant and are indicated by asterisks (*P < 0.05, **P < 0.01, ***P < 0.001; ****P < 0.0001) or numerical values on the respective graphs and figure legends.N represents the number of biologically independent animals.In all cases, data are shown as the mean ± s.e.m., unless otherwise indicated.Cohen's d effect size estimates were calculated using the 'effsize' package.The Wilcoxon rank-sum test was applied to calculate statistical differences among human snRNA-seq cell populations.
Unless otherwise stated, we compare genetically modified and littermate control mice that were weaned and randomly allocated to their housing cages by animal caretakers who were not involved in the experimental planning.The distribution of the genotype was determined by the Mendelian ratio (assumed to be the case for all strains), and selection for the experimental cohort was not influenced by the researcher, thereby randomized.Simultaneously, mice of different genotypes were randomly co-housed-controlling for environmental bias, for example, genotype-specific microbiome.Data collection was randomized using a computer randomization process (R).Scoring, data collection and analysis of all experiments was performed in a blinded manner.Blocking was used to control for sex, age and litter in the experimental design.
No animals or data points were excluded.Animals that met withdrawal criteria were euthanized and scores were only recorded from the surviving animals.To minimize attrition bias, we reported the cumulative clinical score per day.

Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Life sciences study design
All studies must disclose on these points even when the disclosure is negative.

Sample size
For the sample size calculation we used an alpha = 0.05 and a power of 0.8.We used data from previous experiments for the mean and Standard deviation to calculate the standardized effect size using G Power.We determined the required sample size to be able to see a 30% difference in the cumulative score (AUC) for disease course and 70% by qPCR/ gene expression for assessing the targeting efficacy/specificity of our strains.snRNA-seq: More than 10,000 (n=4) per sample were sequenced (10X Genomics).Details are provided in the respective text, figure legends and methods section.
Data exclusions No animals were excluded from downstream analysis.
For spectral flow cytometry, we pregated on single, live CD45+ cells for downstream analysis in R studio.For single nucleus RNA sequencing data analysis, nuclei exclusion was done on the basis of QC and is described in detail in the methods.Initially, snRNA seq data of a NestinCre_Il12rb2fl mouse and Il12rb2fl control has been generated during steady state.However, as initial analyses showed no marked differences between nuclei derived from both genotypes, we excluded steady state data from the manuscript and repeated the sequencing for EAE only.

Replication
All experiments (except targeting experiments) were performed at least twice.The precise number of experiments is indicated in the figure legends.
Bulk RNAseq of primary murine neuronal cultures was performed in n=5 replicates per group.snRNA-seq: We performed three independent sequencing experiments and successfully sequenced 11 samples.The first experiment included two samples: Il12rb2fl_naive and NestinCre_Il12rb2fl_naive. The second experiment included three samples: Il12rb2fl_EAE, NestinCre_Il12rb2fl_EAE and Il12rb2_naive_AUG.Sample Il12rb2_naive_AUG was only an internal control to assess reproducibility and batch effects between the two sequencing experiments and is excluded from the final visualizations displayed in our manuscript.Since initial analyses indicated that genotype differences were predominantly present in early onset EAE, we performed the validation only for EAE samples and excluded the initial steady state data from the manuscript.
The third experiment included six samples: Il12rb2fl/fl EAE (n=3) and NestinCreIl12rb2fl/fl (n=3) EAE.Across the manuscript we only refer to the two batches of EAE samples (steady state samples were not included in the analysis).
Randomization Unless otherwise stated, we compare genetically modified and littermate control mice which have been weaned and randomly allocated to their housing cages by animal caretakers who are not involved in the experimental planning.The distribution of the genotype is determined by the Mendelian ratio (assumed to be the case for all strains) and selection for the experimental cohort was not influenced by the researcher, thereby randomized.Simultaneously, mice of different genotypes are randomly cohoused -controlling for environmental bias e.g.genotype specific microbiome.Data collection was randomized using a computer randomization process (R).

Blinding
Data collection and analysis of all experiments was performed in a blinded manner.

Reporting for specific materials, systems and methods
We require information from authors about some types of materials, experimental systems and methods used in many studies.Here, indicate whether each material, system or method listed is relevant to your study.If you are not sure if a list item applies to your research, read the appropriate section before selecting a response.

Fig. 4 |
Fig. 4 | IL-12 signaling promotes neuronal survival, homeostasis and trophic support to oligodendrocytes in the inflamed murine CNS.Nuclei were isolated from pooled cerebellum, brainstem (dissected from one brain hemisphere) and cervical spinal cord from Il12rb2 fl/fl and Nestin Cre/+ Il12rb2 fl/fl male mice (n = 4 per genotype) at the onset of EAE (10 d.p.i.).Nuclei were lysed, purified by FANS selecting for the Hoechst + fraction and used for snRNA-seq.a, snRNA-seq clustering of 133,151 nuclei by cell type labeled on the basis of known lineage markers and visualized as a UMAP.Each dot corresponds to a single nucleus and each color to a cell-type cluster.b-d, Mean frequencies of cell clusters (b), average fold change of cluster abundance (c) and number of DEGs per cluster (d) between Nestin Cre/+ Il12rb2 fl/fl and Il12rb2 fl/fl mice.e,f, Dot plots showing differential expression of hallmark genes across neurons and oligodendrocytes.Dot size indicates significance and color indicates log fold change between Nestin Cre/+ Il12rb2 fl/fl and Il12rb2 fl/fl mice in the respective cell types.g,h, Average fold change of manually selected gene signatures for hyperinflammation and neuroprotection (Supplementary Table2) between Nestin Cre/+ Il12rb2 fl/fl and

1
Article https://doi.org/10.1038/s41593-023-01435-z it a t o r y _ n e u r o n s G r a n u l e _ c e l l s P u r k in je _ c e l l s C h o l in e r g ic _ n e u r o n s M in e r g ic _ n e u r o n s M ic r o g l ia G r a n u l e _ c e l l s M O L 2 P u r k in je _ c e l l s E x c it a t o r y _ n e u r o n s M L I s Average FC enrichment for CNS essential gene signature Average FC enrichment for hyperinflammatory gene Higher in Nestin Cre/+ Il12rb2 fl/fl Lower in Nestin Cre/+ Il12rb2 fl/fl