Fundamentally Different Roles of Neuronal TNF Receptors in CNS Pathology: TNFR1 and IKKβ Promote Microglial Responses and Tissue Injury in Demyelination, TNFR2 Protects in Excitotoxicity in Mice


 BACKGROUNDDuring inflammatory demyelination TNF receptor 1 (TNFR1) mediates detrimental proinflammatory effects of soluble TNF, whereas TNFR2 mediates beneficial effects of transmembrane TNF through oligodendrocytes, microglia, and possibly other cell types. This model supports use of selective inhibitors of soluble TNF/TNFR1 as antinflammatory drugs for CNS disease. A potential obstacle is the neuroprotective effect of soluble TNF pretreatment described in cultured neurons, but the in vivo relevance is unknown. METHODSTo address this question we generated mice with neuron-specific depletion of TNFR1, TNFR2 or IKKβ and applied experimental models of inflammatory demyelination and acute and preconditioning glutamate excitotoxicity. We also investigated the molecular and cellular requirements of soluble TNF (and therefore TNFR1) neuroprotection by generating astrocyte-neuron co-cultures with different combinations of wildtype and TNF and TNF receptor knockout cells and measuring NMDA excitotoxicity in vitro.RESULTSNeither neuronal TNFR1 nor TNFR2 protected mice during inflammatory demyelination. In fact, both neuronal TNFR1 and neuronal IKKβ promoted microglial responses and tissue injury, and TNFR1 was further required for oligodendrocyte loss and axonal damage in cuprizone demyelination. In contrast, neuronal TNFR2 increased preconditioning protection in a kainic acid excitotoxicity model in mice, and limited hippocampal neuron death. The neuroprotective effects of neuronal TNFR2 observed in vivo were further investigated in vitro. Here as expected, pretreatment of astrocyte-neuron co-cultures with soluble TNF protected them against NMDA excitotoxicity. However, protection was dependent on astrocyte, not neuronal TNFR1, on astrocyte transmembrane TNF-neuronal TNFR2 interactions, and was reproduced by a TNFR2 agonist. CONCLUSIONSThese results demonstrate that neuronal TNF receptors perform fundamentally different roles in CNS pathology in vivo, with neuronal TNFR1 and IKKβ promoting microglial inflammation and neurotoxicity in demyelination, and neuronal TNFR2 mediating neuroprotection in excitotoxicity. They also reveal that previously-described neuroprotective effects of soluble TNF (and therefore TNFR1) against glutamate excitotoxicity in vitro are indirect, and mediated by astrocyte transmembrane TNF-neuron TNFR2 interactions. These results consolidate the concept that selective inhibition of soluble TNF/TNFR1 with maintenance of TNFR2 function would have anti-inflammatory and neuroprotective properties required for the safe treatment of CNS disease.


Introduction
TNF is a multifunctional cytokine system with diverse roles in CNS, ranging from the regulation of physiological functions such as synaptic transmission and plasticity to the induction of in ammation, excitotoxicity and oxidative tissue damage under disease conditions. TNF also has important bene cial functions in demyelinating diseases where it mediates the recruitment and differentiation of oligodendrocyte precursor cells (OPC) and remyelination [1,2]. The relative contribution of bene cial over deleterious TNF functions to one human demyelinating disease, multiple sclerosis (MS), are clearly illustrated by clinical data, which show that non-selective TNF inhibitors exacerbate disease in MS patients [3], and induce de novo demyelination in patients treated for other diseases [4,5]. Also, TNFRSF1A is a risk gene for MS where it encodes a soluble form of TNF receptor 1 that can block TNF [6,7]. These data suggest that bene cial TNF functions in remyelination and suppression of in ammation outweigh deleterious ones and are important for CNS resistance to demyelination.
Understanding the mechanisms of TNF effect in the CNS is important for the re nement of TNF inhibitors as safe therapeutics for various in ammatory diseases including MS. The TNF system involves two ligands and two receptors [8]. Soluble TNF (solTNF) signals mainly through TNFR1 to mediate deleterious TNF effects [9,10], including an inhibitory effect on remyelination [11]. Transmembrane TNF (tmTNF) signals mainly through TNFR2 and mediates bene cial effects, such as remyelination [12]. This information provides an important mechanistic framework upon which to base the design of novel selective TNF inhibitors for the treatment of CNS in ammatory diseases [13].
One potential obstacle for selective inhibition of solTNF/TNFR1 signaling as an anti-in ammatory therapy in CNS disease is its role in neuroprotection. Pretreatment of cultured neurons with solTNF (TNF preconditioning) protects them against a wide range of metabolic, excitotoxic and oxidative death stimuli and promotes maintenance of calcium homeostasis [14,15]. Whether these effects are exerted directly by TNF signaling in neurons or indirectly by small numbers of glial cells in the cultures, and are relevant in vivo remains unknown. Studies in conventional knockout mice showed worsening of CNS pathology in experimental stroke in TNFR1 knockout (TNFR1KO) mice [16], and in experimental autoimmune encephalomyelitis (EAE) and retinal ischemia in TNFR2KO mice [17][18][19], but the mechanisms underlying these protective effects are poorly understood. The availability of mutant mice carrying conditional TNFR1 and TNFR2 alleles now allow the cell speci city of TNF effects to be elucidated. Elegant studies recently showed that depletion of TNFR2 selectively in microglia [20] or oligodendroglia [21] led to early onset or enhanced disease respectively, in EAE, showing that glial cell-speci c bene cial functions of TNFR2 are su cient to protect the CNS during in ammatory demyelination.
To investigate the relevance of neuronal TNFR for CNS disease, especially their possible roles in neuroprotection, we generated mice with neuron-speci c depletion of TNFR1, TNFR2 or IKKβ, a main downstream mediator of TNFR signaling [22], and used them with models of in ammatory demyelination and glutamate excitotoxity. We also used astrocyte-neuron co-cultures with different combinations of wild type, TNFKO, tmTNFKI and TNFRKO cells and glutamate excitotoxicity models in vitro, to verify and further de ne the molecular and cellular interactions underlying TNFR-mediated neuroprotection. Here we show that neuronal TNFR1 and TNFR2 have distinct in ammatory/neurotoxic and neuroprotective effects, in mouse models of CNS demyelination and excitotoxicity, respectively. Speci cally, neuronal TNFR1 and IKKβ both advance the onset of in ammatory demyelination, TNFR1 is further required for axon damage and OLG loss during cuprizone demyelination, and neuronal TNFR2 increases protection against glutamate excitotoxicity in vitro and in vivo. These results consolidate the concept that selective inhibition of solTNF/TNFR1 with maintenance of TNFR2 function would have anti-in ammatory and neuroprotective properties required for the safe treatment of CNS disease.
TNFR1ff, TNFR2ff, and IKKβff littermate mice were used as respective controls in all in vivo experiments. DNA PCR analysis for the detection of recombination (deletion) events in different tissues of nTNFR1KO mice was performed using 2 independent PCR reactions. Primers sense 5'-CAA GTG CTT GGG GTT CAG GG-3' and antisense 5'-CGT CCT GGA GAA AGG GAA AG-3' (ThermoFisher Scienti c) were used for the detection of wildtype Tnfrsf1a (134bp band) and oxed TNFR1ff alleles (195bp band). Primers sense 5'-CCT-GCA-GAC-ACA-CGG-GGA-AA -3' and antisense 5'-TGA-ACT-CAG-GTT-GCC-AGA-CG-3' (ThermoFisher Scienti c) were used for the detection of recombined (deleted, "de oxed") allele (300bp band). DNA PCR analysis for the detection of recombination (deletion) events in tissues of nTNFR2KO mice was performed using a combination of 3 primers in one PCR reaction; sense 5'-CAC ATG TAT GTA CAC CTG TGT G-3', antisense 5'-CTC TCC TGG GCC TAA TGT AG-3', and antisense 5'-ATT GTC TAC TCA GCA CTG GG-3' (ThermoFisher Scienti c) for the detection of wildtype Tnfrsf1b allele (151 bp band), oxed TNFR2ff (200 bp band), and recombined (deleted, "de oxed") alleles (321 bp band). Detection of Cre recombinase was performed using primers, sense 5 '-ATT ACC GGT CGA TGC AAC GAG T-3' and antisense  5'-CAG GTA TCT CTG ACC AGA GTC A-3' (800bp band). All mice were backcrossed and maintained on the B6 genetic background. Animals were maintained under speci c pathogen-free conditions at the animal facilities of the Hellenic Pasteur Institute and all experimental procedures were approved by national authorities and conformed to ARRIVE guidelines and EU Directive 2010/63/EU for animal experiments. EAE EAE was induced in 10-12-week-old female nTNFR1KO and nTNFR2KO mice, and their respective TNFR1ff and TNFR2ff littermate controls as previously described [31]. EAE was induced by s.c. tail base injection of 37 μg of rat myelin oligodendrocyte glycoprotein 35-55 peptide (MOG) dissolved in 100 μl of saline and emulsi ed in 100 μl CFA (Sigma-Aldrich) supplemented with additional 400 μg of H37Ra Mycobacterium tuberculosis (Sigma-Aldrich) (MOG/CFA). Mice also received an i.p. injection of 200 ng of Bordetella pertussis toxin (PTx) (Sigma-Aldrich) on days 0 and 2. Mice were assessed daily for clinical signs as previously described [31]. Brie y, mice were examined using an EAE scoring scale as follows: grade 0, no clinical symptom; grade 1, limp tail; grade 2, hindlimb weakness; grade 3, hindlimb paralysis; grade 4, forelimb and hindlimb paralysis; grade 5, moribund or dead. Animals with score 4 or above were euthanized. Mice were allowed free access to food and water throughout the experiment.

CPZ-induced demyelination and remyelination
CPZ induced demyelination was induced in 8-10-week-old male nTNFR1KO, nTNFR2KO, and nIKKβKO mice and their respective TNFR1ff, TNFR2ff and IKKβff littermate controls. Mice were fed ad libitum with 0.2% w/w CPZ (Sigma-Aldrich; C9012) in powdered standard mouse chow using feeder devices (Analab, Tecniplast, Italy) for 6 weeks and then returned to normal diet, as described [32]. Consistent with this model, hallmark clinical features of disease were measured in the midline corpus callosum of CPZtreated mice at the end of the second week of CPZ feeding (CPZ2) (microglial response), CPZ3 (microgliosis, axonal damage, and demyelination by LFB staining), CPZ5 (maximal demyelination by immunohistochemical staining), and several time-points after removal of CPZ from the diet at week 6 of feeding, including one week after cessation of CPZ feeding (CPZ6+1) (remyelination), CPZ6+2 and CPZ6+4 (both almost complete remyelination) [11].
Acute and preconditioning KA excitotoxicity models in mice In an acute KA seizure model, 2-6-month-old male nTNFR1KO and nTNFR2KO mice, and their respective TNFR1ff or TNFR2ff littermate controls, were injected i.p with 20 or 24 mg/kg KA according to manufacturer's instructions (Tocris Bioscience). In a modi ed preconditioning KA protocol based on a previously described model [33], mice were injected i.p. with 15 mg/kg KA and subsequently after 24 h i.p. with 20 mg/kg KA. Seizures were scored every 5 min for 90 min, using a clinical seizure scale based on previously described methods [34,35]. Brie y, mice were scored as follows: 0, no behavioral response; 1, immobility; 2, one or more of the following: head bobbing, whiskers movement, hunching, outstretched forelimbs; 3, rearing and one-sided forelimb spasms; 4, repeated rearing and two-sided forelimb spasms; 5, one or more of the following: convulsions with sideways leaning, jumping and wild running followed by convulsions, death. Animals with score 5 were euthanized.
RNA isolation and quantitative RT-PCR Brain was removed and spinal cord was ushed from the vertebral column after transcardial perfusion with ice-cold PBS at sacri ce by carbon dioxide inhalation. Total RNA was extracted from spinal cords and brains using TRIzol reagent (Thermo Fisher Scienti c), according to the manufacturer's instructions. RNA purity and quanti cation were assessed using NanoDrop 2000 spectrophotometer data. RNA was used for one-step quantitative RT-PCR using a QuantiFastTM SYBR® green RT-PCR kit (Qiagen Inc.) and QuantiTect Primer Assays for Ccl2 (Mm_Ccl2_1_SG), Cxcl16 (Mm_Cxcl16_1_SG), H2Ab1 (Mm_H2-Ab1_1_SG), Mbp (Mm_Mbp_1_SG), Olig2 (Mm_Olig2_1_SG), Snap25 (Mm_Snap25_2_SG), Tnf (Mm_Tnf_1_SG), and Gusb (Mm_Gusb_1_SG) (Qiagen Inc). All reactions were performed in duplicate using the LightCycler system (Roche, Mannheim, Germany). At the end of each PCR run, melting curve analysis was performed to verify the integrity and homogeneity of PCR products. Results were analyzed using LightCycler software version 3.5 (Roche Diagnostics). Normalization was done using Gusb as a reference gene. Two methods were used for measuring RNA levels, with the same results. For spinal cord samples from EAE experiments, gene expression levels were calculated using already created standard curves for each gene as previously described [36]. These standard curves were created by plotting threshold cycle values vs the logarithm of serial diluted RNA concentrations. Least-squares methods were used for the determination of A and B values in the equation threshold cycle = A*log(CRNA) + B. The coe cient of determination (R2) was greater than 0.99. A second method was used for CPZ brain samples, except for Snap25, which was calculated as above. The difference between threshold cycle values of the target and reference gene (ΔCT = Ct target gene -Ct reference gene) in each sample was calculated. The relative gene expression was then estimated according to the ΔΔ threshold cycle method in which ΔΔCT = (ΔCT of a target sample) -(ΔCT of the average of reference/ naïve samples). The nal relative gene expression was expressed as 2 -ΔΔCT value.

Histology quanti cation
Tissue sections were viewed with an Olympus BX-50 microscope and images captured with an Olympus DP71 microscope digital camera using cell^A imaging software (Soft Imaging System GmbH). Quantitative histopathological analysis was performed in midline corpus callosum, unless otherwise stated using Image J software. Demyelination was measured as loss of LFB staining in coronal sections of whole corpus callosum using a semi quantitative method, as described [11,12], and loss of CNPase immunostaining in the midline corpus callosum by densitometry. Axonal damage was measured as the numbers of positive APP-stained spheroids per mm 2 tissue, and loss of SMI 31 immunostaining by densitometry. Microglia and astrocyte responses were measured as the area covered by Iba1-and GFAPimmunoreactivity, respectively. AIF levels were measured as the area covered by AIF-immunoreactivity, and CD3 + T lymphocytes were counted as cells/mm 2 . Autophagy induction was measured as the numbers of LC3 puncta/LC3-immunoreactive cell in (lateral) corpus callosum area as previously described [38], using a macro developed in Fiji software. Fifty LC3-immunoreactive cells were analyzed for each condition (CPZ0 and CPZ5 control and nTNFR1KO mice).

Results
Neuronal TNFR1 promotes neuroin ammation and the onset of EAE With the aim of interpreting the opposing neurotoxic and neuroprotective effects of the TNF cytokine system described in the literature, we studied the effects of neuronal TNFR1 and TNFR2 using two CNS demyelination models, experimental autoimmune encephalomyelitis (EAE) and cuprizone-induced demyelination (CPZ). We crossed TNFR1ff and TNFR2ff mice with CamkII-Cre mice to achieve TNFR1 or TNFR2 depletion selectively in glutamatergic excitatory CNS neurons. Neuron-speci c TNFR1KO (nTNFR1KO) and TNFR2KO (nTNFR2KO) mice were born at the expected Mendelian ratio, were viable, fertile and didn't display any spontaneous phenotypes. E ciency and speci city of TNFR1 and TNFR2 deletion in brain and spinal cord tissues was veri ed by DNA analysis which showed recombination speci cally in brain and spinal tissues, and not any other tissue tested (Supplementary Figure 1).
nTNFR1KO, nTNFR2KO, and their respective TNFR1ff and TNFR2ff littermate controls were immunized with MOG to induce EAE, a disease driven by myelin-reactive T cells and other in ltrating immune effector cells. In our MOG-EAE model in B6 mice pathology is characterized by severe in ltration of the meninges and spinal cord parenchyma by immune cells (including CD4 + and CD8 + T cells, B cells, NK1.1 + cells and myeloid cells), microglia and astrocyte activation, demyelination and axonal damage in the white matter at the peak of disease. Pathology partially resolves during the chronic phase of disease [30].
nTNFR1KO, nTNFR2KO and control mice all showed full susceptibility to EAE (Figure 1). nTNFR1KO mice showed a small but signi cant delay in disease onset compared to TNFR1ff controls in 2 of 3 experiments, but thereafter the two groups of mice showed no differences in clinical disease progression up to the last time point studied (day 70) ( Figure 1A). nTNFR2KO mice developed EAE with equal onset and clinical course as TNFR2ff littermate controls up to the last time point studied (day 38) (Figure 1C). The results suggest that neuronal TNFR1 plays a small disease-advancing role at the onset of EAE; otherwise, neuronal TNFR1 and TNFR2 do not have signi cant effects on the clinical course of EAE.
To investigate how neuronal TNFR1 advances EAE onset, RNA was isolated from the spinal cord of nTNFR1KO and TNFR1ff mice at disease peak in the control group and analyzed for the expression of early-disease marker genes [36]. nTNFR1KO and TNFR1ff spinal cord showed equally marked reduction in expression of myelin (Mbp) and neuronal (Snap25) gene markers compared to corresponding naïve mice con rming the onset of EAE ( Figure 1B). However, while TNFR1ff spinal cord showed signi cantly increased expression of the in ammatory gene marker (H2-Ab1) at disease peak compared to naïve, nTNFR1KO showed no increase ( Figure 1B). These results suggest that neuronal TNFR1 promotes an acute neuroin ammatory response in spinal cord tissue involving microglia and astrocytes, and thereby the onset of clinical symptoms in EAE, while neuronal TNFR2 has no obvious effect on disease development.
Neuronal TNFR1 advances the onset of microglia responses and demyelination, and is necessary for OLG loss and axon damage in cuprizone demyelination We next used nTNFR1KO, nTNFR2KO, and control mice in an acute toxicity demyelination model induced by 6-weeks of dietary CPZ [32]. CPZ is a copper chelator that causes mitochondrial dysfunction and oxidative stress, both of which are key features of progressive MS [42]. In B6 mice CPZ induces acute responses of microglia and astrocytes, massive death of mature OLG between weeks 2-5 of CPZ feeding that is thought to be triggered by immune mechanisms [43], and predictable primary demyelination in the midline corpus callosum, followed by almost complete remyelination by 2 weeks after removal of CPZ from the diet [11,32].
CPZ-induced pathology in the corpus callosum of nTNFR1KO mice was overall less severe than in TNFR1ff controls. nTNFR1KO mice showed less demyelination at CPZ3 by LFB staining of myelin ( Figure  2Ai&ii), a time point where myelin detected by the more sensitive CNPase immunostaining method appeared still well preserved in both strains. nTNFR1KO mice showed less OLG and myelin loss at CPZ5 by CNPase immunostaining, compared to TNFR1ff controls (Figure 2Bi&ii). nTNFR1KO mice also showed a reduced microglia response at CPZ3 by Iba1 immunostaining (Figure 2Ci&ii), and no measurable axon damage at any time point by numbers of axonal APP spheroids, compared to controls (Figure 2Di&ii).
Both strains showed full resolution of pathology by 2 weeks after cessation of CPZ feeding (CPZ6+2) indicating that remyelination proceeds normally. These results indicate that the combination of solTNF/TNFR1 signaling and CPZ induces stress in neurons that results in the exacerbation of CPZinduced pathology.
Considering that both TNF signaling and oxidative stress are strong inducers of autophagy [44,45], and that autophagy in microglia is responsible for the degradation and clearance of myelin debris in vitro [46], we also investigated the effect of neuronal TNFR1 on autophagy in CPZ demyelination. We measured autophagy levels in the brain sections from naïve and CPZ-fed nTNFR1KO and control mice by immunostaining for LC3, a key component of autophagosomes. Autophagic induction was signi cantly up-regulated in cells located in the corpus callosum (glial cells) of CPZ5 control mice compared to naïve CPZ0 controls, measured by the numbers of LC3-immunoreactive puncta per LC3-positive cell (Figure 3). On the other hand, levels of autophagic induction were similarly low in CPZ5 and naïve CPZ0 nTNFR1KO mice, as in naïve control mice, showing an absence of up-regulation. These results suggest that neuronal TNFR1 is necessary for stimulating autophagy in glial cells, probably in myelin-phagocytosing microglia and re ecting the increased levels of demyelination and need for clearance of myelin debris in the control animals.
nTNFR2KO and TNFR2ff mice showed no differences in CPZ-induced pathology, measured by LFB staining, or CNPase, Iba1, and APP immunostaining ( Supplementary Figure 2A-D) and both strains showed full resolution of pathology by 2 weeks after cessation of CPZ feeding (CPZ6+2).
These results show that as in EAE, neuronal TNFR1 promotes neuroin ammation and demyelination in the CPZ model. Notably, neuronal TNFR1 contributes to acute microglia responses and is required for damage of both axons and OLG in this model. The effect of neuronal TNFR1 on axon damage might be direct by increasing the oxidative stress initiated by CPZ in neurons themselves, and indirect in both neurons and OLG by increasing microglia responses and the production of pro-in ammatory cytokines and chemokines. As in EAE, neuronal TNFR2 has no effect on disease development in the CPZ demyelination model.

Neuronal IKKβadvances neuroin ammation and the onset of demyelination in cuprizone demyelination
The pro-in ammatory effects of TNFR1 are dominantly mediated through activation of the transcription factor NF-κB, which in turn induces expression of genes encoding cytokines, chemokines and antiapoptosis molecules [22]. Under disease conditions it is induced and is a critical mediator of in ammation. To examine the link between neuronal TNFR1 and NF-κB signaling in CNS demyelinating disease, we crossed mice carrying a conditional oxed allele for IKKβ (IKKβff mice), the main NF-κB activating kinase in the canonical NF-κB pathway [22] with CamkII-Cre mice, to generate nIKKβKO mice [30]. We previously showed that nIKKβKO mice have a small but signi cant delay in EAE onset similar to that described here for nTNFR1KO mice, but unlike nTNFR1KO mice nIKKβKO mice subsequently develop a severe non-remitting disease, indicative of neuroprotective effects of neuronal IKKβ during chronic disease [30].
Here we induced CPZ demyelination in nIKKβKO and IKKβff mice. At the onset of pathology at CPZ3, nIKKβKO mice showed reduced demyelination by LFB staining of myelin (Figure 4Ai&ii), axon damage by numbers of APP-immunoreactive spheroids in axons (Figure 4Di&ii), and neuroin ammation by Iba1 immunostaining of microglia ( Figure 5Ai&ii) and GFAP immunostaining of astrocytes (Figure 5Di&ii), compared to IKKβff controls, effects that resemble those in nTNFR1KO mice. Unlike nTNFR1KO mice, nIKKβKO mice subsequently developed full pathology with loss of myelin and OLG measured at CPZ5 by CNPase immunostaining (Figure 4B). This was further shown by immunostaining for apoptosis-inducing factor (AIF), a mitochondrial oxidoreductase that contributes to apoptosis and has been implicated in CPZ-induced OLG death [43]. AIF-immunoreactivity was increased in the corpus callosum of both IKKβff and nIKKβKO mice at CPZ5, most probably in oligodendrocytes which are cells that undergo apoptosis in this model (Figure 4Ci&ii).
Analysis of whole brain RNA levels for disease marker genes, rst for myelin, showed that naïve (CPZ0) nIKKβKO mice express higher levels of Mbp in brain compared to IKKβff mice (Supplementary Figure 3A). Expression of Mbp and Olig2 dropped during CPZ feeding and increased again in both nIKKβKO and IKKβff mice after CPZ removal, showing myelin recovery independently of neuronal IKKβ (Supplementary Figure 3A, B). Expression of the neuron-speci c gene Snap25, which encodes a protein essential for synaptic function, sharply dropped during CPZ feeding and increased again in both groups after CPZ removal. Interestingly, the rebound increase of Snap25 expression in nIKKβKO mice was consistently much higher than in controls ( Figure 4E), suggesting that neuronal NF-κB might inhibit recovery of neuronal functions following CPZ toxicity. Expression of in ammatory markers Tnf, Ccl2 and Cxcl16 in control IKKβff mice showed increases at two distinct time points, rst at CPZ2 at the initiation of microglia responses (Tnf,Ccl2, Cxcl16), and again at CPZ5 at the peak of demyelination/initiation of remyelination (Ccl2, Cxcl16), or CPZ6+0,5 during remyelination (Tnf) ( Figure 5B, C, E). Neither Tnf nor Ccl2 showed initial increase in expression at CPZ2 in nIKKβKO mice, correlating with the delay in glial cell activation observed in these mice at CPZ3 (Figure 5B, C). Also, levels of Ccl2 and Cxcl16 were signi cantly lower in nIKKβKO mice at the peak of disease at CPZ5 ( Figure 5C, E). These results support the pathological ndings that neuronal IKKβ promotes neuroin ammation induced by CPZ feeding.
T cells, and speci cally IL-17-producing T cells, are reported to participate in CPZ demyelination [47], so we investigated T cell in ltration by counting numbers of CD3-immunoreactive cells in the corpus callosum. In ltrating CD3-immunoreactive T cells were localized in the corpus callosum of control IKKβff mice at CPZ5, and in ltration was resolved after CPZ withdrawal at CPZ6+1 and CPZ6+4 (Supplementary Figure 3Ci&ii). In ltrating CD3-immunoreactive T cells were also localized in the corpus callosum of nIKKβKO mice, but unlike in control IKKβff mice, continued to accumulate after CPZ withdrawal at CPZ6+1, showing late resolution at CPZ6+4 (Supplementary Figure 3Ci&ii).
These data show that neuronal NF-κB activity, like neuronal TNFR1, plays a role in initiating neuroin ammation (microglia and astrocytes), demyelination and axon damage in response to dietary CPZ, and might have additional, different roles to TNFR1 during disease resolution and recovery.
Neuronal TNFR2 increases preconditioning protection against seizures and the survival of hippocampal neurons in kainic acid-induced excitotoxicity Evidence for neuroprotective properties of TNF stems mainly from in vitro studies in which solTNF pretreatment of enriched neuron cultures (TNF preconditioning) protects them against a wide range of metabolic, excitotoxic and oxidative death stimuli and promotes maintenance of calcium homeostasis [14,15]. To investigate the role of neuronal TNFR1 or TNFR2 in glutamate excitotoxicity in vivo, we rst used an acute kainic acid (KA) excitotoxicity model. KA is a non-degradable glutamate analogue that induces epileptic seizures and death of CA3/CA2 hippocampal neurons in susceptible mouse strains. B6 mice are known to exhibit high seizure scores but no excitotoxic neuron death [34]. Also, low doses of KA are known to protect mice against seizures induced by subsequent higher doses of KA [33], allowing us also to examine the effect of neuronal TNFR in preconditioning protection against glutamate excitotoxicity in vivo.
Acute systemic (i.p.) administration of KA (20 or 24 mg/kg) consistently induced seizures in groups of nTNFR1KO, nTNFR2KO and control mice without high mortality. No differences in seizure activity between nTNFR1KO ( Figure 6A, grey circles) and TNFR1ff controls ( Figure 6A, black circles) or nTNFR2KO ( Figure 5B, grey circles) and TNFR2ff controls ( Figure 6B, black circles) were observed in this acute model during the 90 minutes of monitoring, although a trend for less severe scores was noted in nTNFR2KO mice compared to TNFR2ff controls ( Figure 6B).
We next used a KA preconditioning protocol, adapted from one previously described [33]. In TNFR2ff controls low dosage KA (15 mg/kg, i.p.) signi cantly reduced severity of seizures induced by subsequent high dosage KA (20 mg/kg, i.p.) ( Figure 6B, black squares) compared to high dosage KA alone ( Figure 6B, black circles), an effect consistent with KA preconditioning protection. In nTNFR2KO mice however, the preconditioning protection effect of low dosage KA on seizures induced by subsequent high dosage KA ( Figure 6B, grey squares) compared to high dosage KA alone ( Figure 6B, grey circles) was reduced and non-signi cant. The loss of KA preconditioning protection in nTNFR2KO mice appeared to be a combined effect of reduced acute seizure intensity and reduced preconditioning protection, but these results provide the rst indication that neuronal TNFR2 is necessary for effective preconditioning neuroprotection induced by low-dose KA during glutamate excitotoxicity in vivo. A direct protective role for neuronal TNFR2 was further supported by the presence of frequent pyknotic neurons stained by cresyl violet (Nissl), a feature of neuron death, localized in patches in both CA2 and CA3 regions of the hippocampus of nTNFR2KO, but not TNFR2ff mice, 5 days after KA seizures with and without preconditioning ( Figure  6C).
Astrocyte TNFR1 and tmTNF via neuronal TNFR2 are necessary for, and a TNFR2 agonist reproduces, TNF preconditioning protection of cortical neurons against NMDA excitotoxicity in vitro.
To reconcile previously reported neuroprotective effects of solTNF (and therefore TNFR1) against glutamate excitotoxicity in vitro [14,15], with the in vivo ndings of TNFR2 neuroprotection here, we modeled different cellular TNF/TNFR interactions in vitro using astrocyte-neuron co-cultures and measured glutamate excitotoxicity induced by NMDA in neurons. Mouse cortical neurons and astrocytes were isolated from WT mice, or mice de cient in TNF (TNFKO), soluble TNF (tmTNFKI), TNFR1 (TNFR1KO) or TNFR2 (TNFR2KO), and different combinations of these cells were used for co-cultures.
In this system, exposure of WT astrocyte-neuron co-cultures to NMDA (50 μM) supplemented with glycine (10 μM) for 24 h induced neuron death, measured by increased percentage of NeuN-positive neurons with pyknotic nuclei stained by Hoechst (Supplementary Figure 4). Astrocyte viability was not affected by NMDA (data not shown), and neuron death was strongly inhibited by the non-competitive NMDA antagonist MK801 (Supplementary Figure 4). Consistent with previous studies [10,48], pretreatment of WT B6 neuron-astrocyte co-cultures with soluble recombinant hTNF or mTNF (100 ng/ml) for 24 h prior to NMDA challenge, consistently and equally protected neurons by 17-20% against NMDA death, and hTNF was used for all experiments described below (Supplementary Figure 4; Figure 6).
Neuroprotection by hTNF was absent when WT neurons were cultured with TNFKO astrocytes but maintained in WT neurons cultured with tmTNFKI astrocytes, and in TNFKO, tmTNFKI, as well as TNFR1KO neurons cultured with WT astrocytes ( Figure 7A). Importantly, neuroprotection was reduced when TNFR2KO neurons were cultured with WT astrocytes ( Figure 7A). These results suggest that preconditioning protection of cortical neurons induced by exogenous soluble hTNF is mainly dependent on astrocyte tmTNF-neuronal TNFR2 interactions. Under these conditions, neuronal TNFR1 also has a very small neuroprotective effect, detectable in WT astrocyte-TNFR2KO neuron co-cultures ( Figure 7A).
To understand how soluble hTNF triggers this mechanism, we performed experiments using TNFR1KO neurons and astrocytes. hTNF signals e ciently through mouse TNFR1 but not through TNFR2, due to species speci city of mouse TNFR2 [49]. Absence of TNFR1 from astrocytes, not neurons, abolished protection ( Figure 7B). Together the results suggest that soluble hTNF engages astrocyte TNFR1, inducing activation and production of tmTNF. solTNF has been previously shown to increase the expression of tmTNF in cell lines [50]. Astrocyte tmTNF-neuronal TNFR2 interactions then precondition neurons for protection against glutamate excitotoxicity (Diagram 1).
To con rm the neuroprotective role of neuronal TNFR2, we compared pretreatment of WT astrocyteneuron co-cultures at NA-DIV7 with a novel TNFR2 agonist (TNFR2ag 100 ng/ml) [51], or with soluble hTNF (100 ng/ml), for 24 h prior to NMDA challenge. As previously described [40,52], TNFR2ag pretreatment induced strong neuroprotection against NMDA death, equal to that induced by soluble hTNF ( Figure 7C). Together these results suggest that astrocyte tmTNF-neuronal TNFR2 interactions dominantly mediate preconditioning protection of neurons against glutamate excitotoxicity in vitro, and likely in vivo. Neuronal TNFR1 has limited neuroprotective effect in models of glutamate excitotoxicity.

Discussion
The TNF cytokine system is complex and associated with multiple functions in the CNS. It mediates overall neurotoxic and neuroprotective effects during CNS pathology depending on the cellular and molecular speci cities of ligand-receptor interaction, and the underlying pathological mechanisms involved [1,2]. To investigate the speci c contributions of neuronal TNFR1 and TNFR2 to in ammatory and excitotoxic neurodegeneration processes relevant for MS, and to place these in the context of previous knowledge concerning neuroprotective effects of solTNF (and therefore TNFR1) in neuronal cultures in vitro [14,15], we used mice in which TNFR1, TNFR2 or IKKβ was selectively depleted from excitatory glutamatergic neurons for experimental models of CNS demyelination and excitotoxicity, and astrocyte-neuron co-culture systems with different combinations of wild type, TNFKO, tmTNFKI, TNFR1KO and TNFR2KO cells in a glutamate excitotoxicity model.
The TNF system has well-described in ammatory and neurotoxic effects in the CNS. In vivo studies using conventional TNF or TNFR knockout mice, or TNF inhibitors, showed neurotoxic roles for solTNF and TNFR1 in a wide range of experimental models including ischemia-reperfusion retinal damage [17], spinal cord injury [53], Parkinson's disease [54], Alzheimer's disease [55], and MS [18,19]. However, when TNFR1 is absent or solTNF is blocked in the entire organism, it is not possible to interpret the cellular mechanisms involved in neurotoxicity. A study of bone marrow chimeric mice generated using wild type and TNFR1KO bone marrow cells showed the essential TNFR1-responsive cells during EAE induction are CNS-resident cells [56]. Potent in ammatory effects of TNF, directly by activation of microglia, astrocytes, and endothelial cells, and indirectly through the induction of cell necroptosis and further neuroin ammation [57], are likely to dominate the neurotoxic effects of TNF in vivo. This is supported by the nding that NF-κB activation in nestin-expressing cells (astrocytes and neurons), and not OLG, determines neuroin ammation and myelin damage in the CPZ model [58].
Here we show that both neuronal TNFR1 and IKKβ accelerate microglia and astrocyte responses, demyelination and the onset of axon damage in the demyelination models, indicating common signaling pathways. Neuronal TNFR1 also induced the induction of autophagy in glial cells in the demyelinating lesions in the CPZ model, possibly in myelin-phagocytosing microglia, as recently described [46], and re ecting the increased demyelination and demand for clearance of myelin debris in control mice.  [59], and potentiating excitotoxicity in vivo [60], [61,62]. Similar effects are seen with NMDAR [63] [64]. Glutamate receptors and transporters are dysregulated during experimental demyelination in vivo [65], and might be su cient to activate microglia in vivo in the context of disease and induce neuroin ammation. Second, by the induction of oxidative stress in neurons. Exposure of neurons to pathophysiological concentrations of solTNF results in rapid and profound decrease in mitochondrial function and reduced viability, an effect reduced by pretreatment of neurons with an anti-TNFR1 antibody [66]. In the CPZ toxicity model, the function of many enzymes and mitochondria are already compromised by copper chelation, resulting in metabolic stress and the selective loss of OLG and demyelination [43]. Like OLG, neurons are also highly sensitive to oxidative stress. Damage of neuronal axons is a pathological feature of acute CPZ toxicity in young B6 mice and observed here in TNFR1ff control mice by increased numbers of APP-immunoreactive spheroids at the peak of demyelination. The observation that APP spheroids do not accumulate in nTNFR1KO mice indicates that the combination of neuronal TNFR1 and CPZ causes direct damage in neurons and their axons.
The TNF system also has prominent neuroprotective roles in the CNS. One of the rst reported effects of solTNF (and therefore TNFR1) in vitro was neuroprotection. Pretreatment of primary neuron cultures with solTNF protected them against a variety of metabolic-excitotoxic-oxidative insults and promoted maintenance of calcium homeostasis [14,15]. Neuroprotective effects of TNFR1 and TNFR2 were also observed in speci c in vivo experimental paradigms including cerebral stroke [16] and glutamate excitotoxicity [40,52], but the cellular mechanisms of these effects is poorly understood. Recently, the availability of mouse lines carrying conditional TNFR1 [27] or TNFR2 [28] alleles made investigation of cell-speci c roles of the TNFR in CNS possible. An elegant study using mice with myeloid cell-speci c ablation of TNFR2, shows that microglial TNFR2 protects mice against the initiation of EAE [20]. Another study using mice with OLG lineage cell-speci c ablation of TNFR2 shows that OLG TNFR2 ameliorates clinical symptoms particularly in the chronic phase of EAE, reduces myelin and axonal pathology, drives the differentiation of OLG precursor cells into mature OLG, and improves remyelination [21]. The effects of OLG-speci c TNFR2 depletion in EAE largely recapitulate the effects of total TNFR2 KO in CPZ demyelination [12] suggesting that protective effects of TNFR2 in models of in ammatory demyelination are mediated predominantly by microglia and OLG. The nding here that neuronal TNFR2 did not alter the development of EAE or CPZ demyelination further supports this conclusion. Neuronal TNFR2 did however play a role in the inducton of preconditioning protection against seizures in a KA excitotoxicity model.
Also, neuronal TNFR2 protected hippocampal neurons against excitotoxic death in control mice.
Together, our results show that neuronal TNFR1 and IKKβ promote neuroin ammation and the onset of demyelination in mouse models of MS, and that neuronal TNFR1 is additionally required for axon damage, OLG loss and induction of glial cell autophagy in white matter lesions during CPZ demyelination, possibly by increasing neuronal oxidative stress. On the contrary, neuronal TNFR2 had no detectable effects on the clinical or histopathological features of the MS models, but our results provide the rst evidence that TNFR2 directly protects neurons in vivo under conditions of glutamate excitotoxicity that are relevant for a wide range of CNS diseases. Also, the previously reported neuroprotective effects of solTNF (and therefore TNFR1) in vitro were reconciled here with in vivo ndings, with evidence from astrocyte-neuron co-cultures that astrocyte TNFR1 and tmTNF, and neuronal TNFR2 are necessary, and that a TNFR2 agonist is su cient, to mediate TNF preconditioning neuroprotection against glutamate excitotoxicity in vitro. This is consistent with previous ndings with TNFR2 agonists, which protected human dopaminergic neuronal cells against oxidative stress via activation of the PI3K-PKB/Akt pathway, or challenge with 6-OHDA in vitro [67], and enriched mouse cortical neurons against glutamate excitotoxicity [52].

Conclusions
The results of this study show that the neuronal TNFR have fundamentally different roles in CNS pathology in vivo, with signaling through neuronal TNFR1 and IKKβ promoting microglia responses and neurotoxicity in mouse demyelination models, and neuronal TNFR2 being involved in preconditioning protection in excitotoxicity, but not in in ammatory demyelination. Details of the molecular and cellular mechanisms underlying previously described neuroprotective properties of solTNF in vitro were obtained using astrocyte-neuron co-cultures with different combinations of wildtype and knockout cells. Results further supported the in vivo ndings, with solTNF (and therefore TNFR1) indirectly protecting neurons against NMDA excitotoxicity via astrocyte tmTNF-neuronal TNFR2 interactions, and being fully reproduced by a TNFR2 agonist. Together the results consolidate the concept that selective inhibition of solTNF/TNFR1 with maintenance or stimulation of bene cial TNFR2 function represents a promising strategy for suppressing in ammation and maintaining neuroprotection for the treatment of a wide range of CNS diseases. Availability of supporting data The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.