Astrocytic SARM1 promotes neuroinflammation and axonal demyelination in experimental autoimmune encephalomyelitis through inhibiting GDNF signaling

Astrocytes are important components of the innate immune response in the central nervous system (CNS), involving in the inflammatory and neurotoxic responses that occur in CNS diseases, such as multiple sclerosis (MS). Recent studies have shown that SARM1 plays a critical role in axonal degeneration and inflammation. However, the detailed role of astrocytic SARM1 in MS remains unclear. Here, we established the MS model of mice - experimental autoimmune encephalomyelitis (EAE) and found that SARM1 was upregulated in astrocytes of the spinal cords of EAE mice. Moreover, conditional knockout of astrocytic SARM1 (SARM1GFAP-CKO mice, SARM1Aldh1L1-CKO mice) delayed EAE with later onset, alleviated the inflammatory infiltration, and inhibited the demyelination and neuronal death. Mechanically, RNA-seq revealed that the expression of glial-derived neurotrophic factor (GDNF) was upregulated in SARM1−/− astrocytes. Western blot and immunostaining further confirmed the upregulation of GDNF in spinal cord astrocytes of SARM1GFAP-CKO EAE mice. Interestingly, the downregulation of GDNF by streptozotocin (STZ, a drug used to downregulate GDNF) treatment worsened the deficits of SARM1GFAP-CKO EAE mice. These findings identify that astrocytic SARM1 promotes neuroinflammation and axonal demyelination in EAE by inhibiting the expression of GDNF, reveal the novel role of SARM1/GDNF signaling in EAE, and provide new therapeutic ideas for the treatment of MS.


INTRODUCTION
Multiple sclerosis (MS) is one of the autoimmune diseases whose clinical symptoms are monocular vision loss caused by optic neuritis, limb weakness or sensory loss caused by rhabdomyositis, and ataxia caused by cerebellar lesions, which seriously endangers the health of young people, especially women [1][2][3][4]. According to the latest diagnostic criteria, MS is divided into four subtypes: 1) clinically isolated syndrome (CIS); 2) relapsing-remitting (RRMS); 3) secondary progression (SPMS); 4) primary progression (PPMS) [4,5]. The most common type is RRMS on which research of clinical drugs has achieved great progress nowadays. However, some patients can develop from RRMS to SPMS, and a small number of patients present with PPMS at first, which represents different pathological mechanisms among different MS types [6]. Therefore, according to the pathogenesis of different MS types, taking the corresponding treatment measures becomes the focus of researches.
Previous opinion thought that peripheral immune cells are activated to infiltrate the central nervous system (CNS), where an inflammatory cascade occurs to attack myelin components, resulting in axonal damage [7,8]. However, some phenomena, such as loss of oligodendrocytes and myelin in the absence of T or B cell infiltration in early lesions [9]; extensive loss of myelin in the cerebral cortex and deep gray matter but little infiltrating immune cells [10][11][12] refute this theory. There is a growing awareness that peripheral immune cell-mediated inflammatory mechanisms can not fully explain the degenerative process. In fact, recent studies suggest that in progressive disease, the innate immune response of resident CNS cells may play a key role in myelin damage and axonal degeneration, especially the immunity of astrocytes and microglia. Astrocytes, the most abundant cell type in the CNS, play a pivotal role in MS [13]. It is a key component of the blood-brain barrier (BBB) function and interacts with BBB endothelial cells, pericytes and neighboring neurons, which controls the permeability of BBB [14]. In MS/EAE, astrocytes are activated. On the one hand, astrocytes secrete chemokines to recruit peripheral leukocytes into the CNS by disrupting the BBB, thereby causing an inflammatory cascade to damage myelin. Moreover, mitochondria and neurons are damaged by astrocytes by reducing neurotrophic factors and increasing NO, glutamate and other active substances [15]. On the other hand, leukocytes are restricted from entering the CNS because astrocytes participate in forming the glial scars [16]. Moreover, in acute and chronic injury, reactive astrocytes play a major role in defending against oxidative stress. Because of the powerful and diverse functions of astrocytes in MS, therapeutic approaches targeting them are expected to have a significant clinical impact on progressive disease.
In addition to inflammation and demyelination, axon damage is also an important part of MS, especially in the progressive stage of degeneration [13,17,18]. The wallerian degeneration (WD) was characterized by cytoskeletal granule disintegration, mitochondrial swelling, and axonal fragmentation. It is a classic model of axonal degeneration [19]. A recent study has shown that SARM1 (sterile alpha and TIR motif containing 1) induces activation of part of the molecular pathway for WD [20]. SARM1 is an intracellular protein that is highly expressed in the nervous system of mammals. It contains a C-terminal Toll/interleukin-1 receptor (TIR) domain, two tandem sterile alpha motif (SAM) domains and an N-terminal region with multiple armadillos repeat motifs (ARMs) [20,21]. SARM1 is activated because of the reduction of nicotinamide mononucleotide adenylyl transferase (NMNAT2), which leads to the decrease of NADase activity in its TIR domain, NAD depletion, NMN accumulation, ATP energy depletion, Ca 2+ influx increase, and finally axon degeneration [21,22]. Interestingly, previous studies have shown that the effect of the inactivation of SARM1 on the immune response depends on the disease model or species. In human peripheral blood leukocytes, the function of SARM1 is to suppress the immune response [23][24][25]. However, in the animal models of CNS infection, the loss of SARM1 leads to a reduction in CNS inflammation, indicating that the function of SARM1 is to enhance the antiviral response of the CNS [26,27]. In addition, it has been reported that SARM1 gene knockout not only can prevent neuronal degeneration and perinatal death but also can inhibit toxic neuropathy, such as amyotrophic lateral sclerosis (ALS) and glaucoma-related degeneration [28]. At present, the mechanism and pathway of SARM1 in neurons are relatively clear. In neurons, the activation of SARM1 stimulates the production of inflammatory cytokines and chemokines, and SARM1 stimulates the Jnk-c-Jun axis through the MAPK pathway, thereby initiating neuroinflammation responses [29]. However, it remains unclear what role the astrocytic SARM1 may play in MS and EAE.
In this study, we found that SARM1 was upregulated in the astrocytes of the spinal cords of EAE mice, and the knockout of astrocytic SARM1 alleviated neuroinflammation, inhibited demyelination and axon damage, and decreased the neuronal death in EAE mice. In addition, through RNA sequencing, we found that the expression of GDNF was upregulated in SARM1 -/astrocytes. Moreover, the level of GDNF protein was significantly increased in SARM1 GFAP -CKO EAE mice, compared with that in SARM1 f/f EAE mice. Since streptozotocin (STZ) is usually used to inhibit the expression of GDNF [30,31], its treatment deteriorated the deficits of SARM1 GFAP -CKO EAE mice, suggesting that astrocytic SARM1 may promote EAE through inhibiting GDNF signaling. Our findings provided some evidence for the mechanisms of demyelination and neuroinflammation in EAE, which may contribute to developing new MS therapies.

SARM1 was upregulated in astrocytes of spinal cords of EAE mice
To explore whether SARM1 is activated and the potential role of glial SARM1 in MS, we first examined the expression pattern of SARM1 in EAE mice. Western blot showed that the expression of SARM1 was significantly upregulated in the cerebellum, cervical, thoracic, and lumbar spinal cords of EAE mice (Fig. 1A, B). Further double immunostaining of SARM1 and glial markers, including GFAP (astrocytes marker) and Iba1 (microglia marker) showed that SARM1 was mainly detected in GFAP + astrocytes (Fig. 1C, D), but not in Iba1 + microglia (Fig. 1E), which was consistent with Fig. 1 SARM1 was upregulated in astrocytes of spinal cords of EAE mice. A Western blot analysis of SARM1 expression in the cerebellum, cervical, thoracic and lumbar spinal cords of control and EAE mice. The higher molecular weight band is marked with an arrow as the band corresponding to SARM1. B Quantitative analysis of the relative SARM1 levels as shown in A (normalized to control mice, n = 4). C Double immunostaining analysis of SARM1 (green) and GFAP (red) in control and EAE mice. D Quantitative analysis of intensity of SARM1 C (n = 3). E Double immunostaining analysis of SARM1 (red) and Iba1 (green) in the spinal cords of control mice and EAE mice. Scale bar, 50 μm. Data were mean ± SEM. Student's t-test, *p < 0.05, **p < 0.01. previous research [20]. Moreover, the intensity of SARM1 in EAE mice was higher than that in the control mice (Fig. 1C, D), suggesting that SARM1 was upregulated in astrocytes of EAE mice.
EAE was relieved with later onset, less inflammatory infiltration, and fewer neuronal death in SARM1 GFAP -CKO mice To further study the role of astrocytic SARM1 in EAE model, the floxed SARM1 allele (SARM1 f/f ) mice were crossed with GFAP-Cre transgenic mice to generate SARM1 GFAP -CKO mice. Indeed, SARM1 was effectively knockout in the cerebellum, cervical, thoracic and lumbar spinal cords of SARM1 GFAP -CKO mice ( Supplementary Fig.  S1A, B), which indicated that SARM1 GFAP -CKO mice were successfully constructed.
Next, female SARM1 f/f mice and SARM1 GFAP -CKO mice aged 8-12 weeks were subjected to treatment with MOG  and pertussis toxin to induce EAE. There was no significant difference in body weight between SARM1 f/f EAE mice and SARM1 GFAP -CKO EAE mice, but the EAE score of SARM1 GFAP -CKO mice was significantly reduced and the peak time of onset was delayed, suggesting that dyskinesia of EAE was alleviated in SARM1 GFAP -CKO mice ( Fig. 2A, B).
Further Nissl staining revealed that there was no difference in the density of Nissl bodies in the control group, suggesting that SARM1 knockout in astrocytes had no effect on the development of spinal cords (Fig. 2C, D). In SARM1 f/f EAE mice, the density of Nissl bodies was significantly decreased, compared with that in the control SARM1 f/f mice, however, the loss of Nissl bodies was significantly reduced in SARM1 GFAP -CKO EAE mice (Fig. 2C, D). Furthermore, immunostaining also showed the loss of NeuN + cells was significantly decreased in the lumbar spinal cords of SARM1 GFAP -CKO EAE mice, compared with that in SARM1 f/f EAE mice, further confirming that the neuroprotective effect of deletion of astrocytic SARM1 in EAE model (Fig. 2E, F). To detect the inflammatory infiltration in EAE mice, HE staining was performed and showed that there was no difference in the density of infiltrating cells in the control group (Fig. 2G, H). However, compared with that in SARM1 f/f EAE mice, the density of infiltrating cells was significantly decreased in SARM1 GFAP -CKO EAE mice (Fig. 2G, H), indicating that the inflammatory infiltration of SARM1 GFAP -CKO EAE mice was alleviated. Similarly, there was no difference in density of CD45 + (a marker of common leukocyte antigen) cells in the control group. However, the density of CD45 + cells in SARM1 GFAP -CKO EAE mice was significantly lower than that in SARM1 f/f EAE mice (Fig. 2I, J). Taken together, the above results suggested that conditional knockout of astrocytic SARM1 delayed onset time, and decreased the loss of neurons and inflammatory infiltration in EAE mice.
The density of astrocytes and microglia, and proliferation of astrocytes were reduced in the spinal cords of SARM1 GFAP -CKO EAE mice Astrocytes and microglia are the key components of MS immunopathology and play an indispensable role in the formation of harmful positive feedback inflammation cycles [32,33]. Therefore, we tested whether the ablation of astrocytic SARM1 affected the response of these inflammatory cells. Double immunostaining showed there was no obvious difference in the density of Iba1 + microglia and GFAP + astrocytes in the control group (Fig. 3A, B). In the EAE group, the density of Iba1 + and GFAP + cells were obviously increased, and their morphology was more hypertrophic than that in the control group. However, compared with that in SARM1 f/f EAE mice, the density of Iba1 + microglia and GFAP + astrocytes were significantly decreased in the lumbar spinal cords of SARM1 GFAP -CKO EAE mice (Fig. 3A, B). Previous studies have shown that Aldh1L1 is a better immunohistochemical astrocyte marker than GFAP because it better labels the cell bodies of astrocytes in both white and grey matter [34]. To further determine the changes of astrocytes, we performed double immunostaining of Aldh1L1 and GFAP and found that astrocytes were activated in control EAE mice, but decreased in SARM1 GFAP -CKO EAE mice (Fig. S2A, B). These results suggested that microglia and astrocytes were activated in EAE mice, while astrocytic SARM1 deletion resulted in less activation of microglia and astrocytes. Furthermore, immunostaining of PH3/Ki67 and GFAP showed that the percentage of PH3 + GFAP + /GFAP + astrocytes (Fig. 3C, D) and the percentage of Ki67 + GFAP + /GFAP + astrocytes (Fig. S2C, D) were significantly decreased in SARM1 GFAP -CKO EAE mice, compared with that in SARM11 f/f EAE mice, which suggested that decrease of astrocyte density might be due to decrease of astrocyte proliferation. Since SARM1 was not expressed in microglia [20], the effect of SARM1 on microglia might be indirect. These results suggested that conditional deletion of astrocytic SARM1 alleviated the inflammatory response of EAE and reduced the proliferation of astrocytes.
The demyelination and axonal injury were alleviated in the spinal cords of SARM1 GFAP -CKO EAE mice In addition to inflammation and neuronal death, demyelination and axonal injury are also pathological features of MS/EAE [4]. We next tested whether SARM1 ablation in astrocytes affected the demyelination and axonal injury in EAE model. Myelin basic protein (MBP) is a common marker of myelin protein. It has been reported that the 21.5 kDa isoform of MBP occurs in the early stage of myelin formation or in areas of remyelination of injured white matter, while the 18.5 kDa isoform was rich in the mature myelin [35]. As shown in Fig. 4A, B, 21.5 kDa isoform of MBP was more significantly affected in SARM1 GFAP -CKO mice, while the 18.5 kDa and 17.2 kDa isoforms were less affected, which revealed that SARM1 deletion might had more effect on myelin formation and regeneration, and less effect on mature myelin in EAE mice. Moreover, there was no difference in the level of MBP protein in SARM1 f/f mice and SARM1 GFAP -CKO mice. However, the protein level of 21.5 kDa isoform of MBP was significantly decreased in SARM1 f/f EAE mice, compared with that in SARM1 f/f mice, whereas the protein level of 21.5 kDa isoform of MBP of SARM1 GFAP -CKO EAE mice was significantly higher than that in SARM1 f/f EAE mice, which indicated myelin formation and regeneration instead of mature myelin might be the target of SARM1 in EAE mice. Furthermore, electron microscopy showed that the demyelinating phenotype in SARM1 GFAP -CKO EAE mice was significantly alleviated, compared with that in SARM1 f/f EAE mice ( Fig. 4C-E). Besides, double immunostaining showed that the intensity of MBP and NF (neurofilament heavy polypeptide) in SARM1 f/f EAE mice was obviously decreased, compared with that in SARM1 f/f mice. Similarly, the intensity of MBP and NF in SARM1 GFAP -CKO EAE mice was significantly higher than that in SARM1 f/f EAE mice (Fig. 4F, G). Finally, immunostaining of GAP43 (another axonal marker) showed that the intensity of GAP43 in SARM1 GFAP -CKO EAE mice was significantly higher than that in SARM1 f/f EAE mice (Fig. 4H, I).
In conclusion, these results suggested that the deletion of astrocytic SARM1 alleviated the demyelination and axonal injury in EAE mice.
EAE was relieved with later onset, fewer neuronal death, demyelination, axonal injury and less inflammatory infiltration in SARM1 Aldh1L1 -CKO mice Because aldehyde dehydrogenase family member 1L1 (Aldh1L1), specifically expressed by astrocytes, but not by oligodendrocytes or neurons [34], is a recognized cell type-specific immunohistochemical marker for astrocytes [36], Aldh1L1-Cre is used as a more specific marker for astrocytes than GFAP in the forebrain and spinal cords [37]. Therefore, SARM1 Aldh1L1 -CKO mice were used to further exclude the interference of SARM1 in neurons and other glial cells and to confirm the function of SARM1 in astrocytes. Western blot showed that SARM1 was efficiently knockout in the cerebellum, cervical, thoracic, and lumbar spinal cords of SARM1 Aldh1L1 -CKO mice (Fig. 5A, B), which indicated that our SARM1 Aldh1L1 -CKO mice were successfully constructed.
In EAE model, we found that the body weight of the control mice and EAE mice has no difference (Fig. 5C) and the clinical score of EAE was significantly lower in SARM1 Aldh1L1 -CKO EAE mice than that in SARM1 f/f EAE mice (Fig. 5D). Next, Nissl staining revealed that there was no difference in the density of Nissl bodies between SARM1 f/f mice and SARM1 Aldh1L1 -CKO mice. However, the loss of Nissl bodies in SARM1 Aldh1L1 -CKO EAE mice was significantly decreased, compared with that in SARM1 f/f EAE mice (Fig. 5E, F).
Furthermore, compared with that in SARM1 f/f EAE mice, the density of infiltrating cells (Fig. 5G, H) and CD45 + cells (Fig. 5I, J) was significantly decreased in SARM1 Aldh1L1 -CKO EAE mice, indicating that ablation of astrocytic SARM1 could alleviate inflammatory infiltration in EAE. Furthermore, the intensity of MBP and NF in SARM1 Aldh1L1 -CKO EAE mice was also significantly higher than that in SARM1 f/f EAE mice (Fig. 5K, L). Taken together, the above results further suggested that deletion of astrocytic SARM1 delayed onset time, relieved demyelination, and decreased neuronal death and inflammatory infiltration in EAE mice.
GDNF was upregulated in astrocytes, and inhibited NF-κB signaling in SARM1 GFAP -CKO EAE mice How can ablation of astrocytic SARM1 lead to the relief of EAE phenotypes? RNA-sequencing showed that GDNF in primary cultured astrocytes from SARM1 GFAP -CKO mice was higher than that from SARM1 f/f mice (Fig. 6A, B). GDNF is a member of the TGFβ superfamily [38], first discovered in rat glial cells, which can mediate the growth, differentiation and migration of neurons and is necessary for the formation of axons and dendrites [39]. Further quantitative real-time PCR (q-PCR), western blot and Fig. 2 EAE was relieved with later onset, less inflammatory infiltration, and fewer neuronal death in SARM1 GFAP -CKO mice. A The weight of SARM1 f/f mice and SARM1 GFAP -CKO mice, and SARM1 f/f EAE mice and SARM1 GFAP -CKO EAE mice ranged from 0 to 21 dpi (n = 5, two-way ANOVA with Bonferroni's post-tests). B The EAE score of SARM1 f/f and SARM1 GFAP -CKO mice ranged from 0 to 21 dpi (n = 5, two-way ANOVA with Bonferroni's post-tests). C Typical images of Nissl staining in the lumbar spinal cords of SARM1 f/f mice and SARM1 GFAP -CKO mice, and SARM1 f/f EAE mice and SARM1 GFAP -CKO EAE mice. D Quantitative analysis of density of Nissl bodies as shown in C (n = 5). E Typical images of NeuN + immunostaining in the lumbar spinal cords of SARM1 f/f and SARM1 GFAP -CKO mice, and SARM1 f/f EAE and SARM1 GFAP -CKO EAE mice. F Quantitative analysis of the density of NeuN + cells as shown in E (n = 5). G Typical images of HE staining in the lumbar spinal cords in of SARM1 f/f mice and SARM1 GFAP -CKO mice, and SARM1 f/f EAE mice and SARM1 GFAP -CKO EAE mice. H Quantitative analysis of the density of infiltrating cells as shown in G (n = 5). I The typical images of CD45 + immunostaining in the lumbar spinal cords of SARM1 f/f mice and SARM1 GFAP -CKO mice, and SARM1 f/f EAE mice and SARM1 GFAP -CKO EAE mice. J Quantitative analysis of the density of CD45 + cells as shown in (I) (n = 5). Scale bar, 50 μm. The data were mean ± SEM. Student's t-test unless otherwise indicated, n.s., not significant (p > 0.05), **p < 0.01, ***p < 0.001. Fig. 3 The density of microglia and astrocytes, and proliferation of astrocytes were reduced in the spinal cords of SARM1 GFAP -CKO EAE mice. A Double immunostaining of Iba1 (green) and GFAP (red) in lumbar spinal cords of SARM1 f/f mice and SARM1 GFAP -CKO mice, and SARM1 f/f EAE mice and SARM1 GFAP -CKO EAE mice. B Quantitative analysis of the density of Iba1 + and GFAP + was quantified as shown in A (n = 5). C Double immunostaining analysis of PH3 (green) and GFAP (red) in lumbar spinal cords of SARM1 f/f mice and SARM1 GFAP -CKO mice, and SARM1 f/f EAE mice and SARM1 GFAP -CKO EAE mice. D The quantitative percentage of PH3 + GFAP + cells in the total number of GFAP + cells as shown in C (n = 5). Scale bar, 50 μm. The data were mean ± SEM. Student's t-test, n.s., not significant (p > 0.05), *p < 0.05, ***p < 0.001.
In addition, it is reported that GDNF promotes the transformation of M1 microglia to M2 microglia by inhibiting NF-κB and promoting the PI3K/AKT signaling pathway [40]. To observe whether upregulation of GDNF in SARM1 GFAP -CKO EAE mice regulates the above-mentioned pathways, we examined the changes of GDNF and NF-κB in the protein level. Interestingly, our results showed that the expression of GDNF was upregulated in SARM1 GFAP -CKO mice, compared with that in SARM1 f/f mice and the expression of GDNF in SARM1 GFAP -CKO EAE mice was significantly higher than that in SARM1 f/f EAE mice. In addition, the expression of GDNF was low at the basal level, but it was significantly increased in EAE mice (Fig. 6H, I), which might be related to the increased secretion of GDNF to promote the survival of neurons under disease conditions [40]. Meanwhile, the protein level of NF-κB in EAE mice was upregulated, while it was significantly lower in SARM1 GFAP -CKO EAE mice than that in SARM1 f/f EAE mice (Fig. 6H, J). Next, CD206 (M2 microglial marker) immunostaining showed that the density of CD206 + cells was increased in the SARM1 GFAP -CKO EAE mice, compared with that in SARM1 f/f EAE mice (Fig. 6K-L). Taken together, these results suggested that GDNF was upregulated in astrocytes, inhibited NF-κB signaling, and promoted the polarization of M2 microglia in SARM1 GFAP -CKO EAE mice, which may be responsible for the neuroprotective effect of astrocytic SARM1 deletion in EAE mice.
Deletion of astrocytic SARM1 performed the neuroprotective effects through GDNF signaling in EAE mice To examine whether SARM1 deletion in astrocytes performed neuroprotective effects through GDNF signaling, we next checked whether inhibition of GDNF worsened EAE phenotypes in SARM1 GFAP -CKO EAE mice. Previous studies have suggested that STZ reduces the expression of GDNF in astrocytes [30]. As expected, primary astrocytes from SARM1 GFAP -CKO mice were cultured and treated with STZ (100 μM) for 24 h, and the expression of GDNF in astrocytes was indeed reduced by STZ treatment (Supplementary Fig. S3A, B). Interestingly, STZ significantly worsened the clinical scores of SARM1 GFAP -CKO EAE mice (Fig. 7A). We found that GDNF was significantly decreased after injected with STZ, while the expression of MBP protein was decreased and the expression of NF-κB protein was increased in SARM1 GFAP -CKO EAE mice (Fig. 7B, C). Moreover, the intensity of GAP43, MBP and NF was significantly decreased in SARM1 GFAP -CKO EAE mice after injected with STZ ( Fig. 7D-G). Meanwhile, we found that the density of NeuN + cells was decreased and the density of CD45 + inflammatory cells was significantly increased in STZ-treated SARM1 GFAP -CKO EAE mice, compared with control-treated SARM1 GFAP -CKO EAE mice (Fig. 7H-K), suggesting that deletion of astrocytic SARM1 performed the neuroprotective effects through GDNF signaling in EAE mice.

DISCUSSION
In this study, we provided evidence for the role of SARM1 in EAE and proposed a working model (Fig. 8). In this model, SARM1 is upregulated in astrocytes of EAE mice, and somehow inhibits the expression of GDNF, which may contribute to neuroinflammation and demyelination by activating NF-κB signaling. Interestingly, in EAE model, conditional knockout of astrocytic SARM1 indeed inhibits neuronal death, neuroinflammation, and improves the behavioral recovery of motor function. SARM1 is usually present in axons in an inactive state under normal conditions, but can be activated when the body is injured [28], which is consistent with our results (Fig. 1A, B). Regarding how SARM1 regulates the phenotype of EAE, mRNA sequencing revealed that GDNF was increased in SARM1 -/astrocytes (Fig. 6A, B). Carty's research team has shown the removal of SARM1 can inhibit the expression of IL-1β depending on the activation of the NLRP3 inflammasome [41]. Previous studies have found that the NLRP3 inflammasome was indeed activated in EAE model [42]. In addition, in rat neuron-glia co-culture, damaged dopaminergic neurons induce astrocytes to release GDNF through IL-1β [43], which indicates that the expression of GDNF in astrocytes is stimulated by inflammation and can provide neurotrophic support for injured neurons under the conditions of CNS inflammation [44,45]. Another example is that IL-1β stimulates GDNF expression in C6 glioma cells through the inhibitor kappa B (IκB), p38 MAP kinase, p44/p42 MAP kinase and JAK-STAT3 pathways [46]. In our RNA sequencing results, we surprisingly found that IL-1β was higher in SARM1 -/astrocytes than that in SARM1 +/+ astrocytes (Fig. 6A, B). Therefore, we suspect that the knockout of SARM1 in astrocytes may increase GDNF by increasing the expression of IL-1β. In the future, further studies need to be performed to test this possibility.
The release of GDNF is expected to have therapeutic potential for neurodegenerative diseases, traumatic and inflammatory brain injuries [46]. In our study, inflammation was reduced in SARM1 GFAP -CKO EAE mice and SARM1 Aldh1L1 -CKO EAE mice, which may be related to the increase of GDNF that can inhibit the NF-κB inflammation signaling pathway (Fig. 6H-J) and increase the density of M2 microglia (Fig. 6K, L). It is reported that AKT plays an important role in the nervous system and inhibits the SARM1mediated axonal injury pathway [47]. We speculate that changes of demyelination and axon damage may be related to AKT. Although there was no direct evidence in our experiments that increase of the intensity of MBP and NF in SARM1 GFAP -CKO EAE mice and SARM1 Aldh1L1 -CKO EAE mice is indeed the reason for improving the recovery of motor function, a large amount of evidence indicates that increase of the intensity MBP and NF is associated with the recovery of motor function [48]. Therefore, we speculate that increased intensity of MBP and NF in SARM1 GFAP -CKO EAE mice and SARM1 Aldh1L1 -CKO EAE mice, compared with that in SARM1 f/f EAE mice may be the reason for the improvement in motor function recovery. Finally, after inhibition of the expression of GDNF by intraperitoneal injection of STZ, the improved symptoms of SARM1 GFAP -CKO EAE mice were suppressed, which further proved that SARM1 regulated EAE through GDNF signaling. How astrocytic SARM1 regulates the expression of GDNF in EAE needs to be further studied in the future.
In summary, our findings reveal an unrecognized function of removal of astrocytic SARM1 to prevent axonal injury, demyelination and neuroinflammation, and reveal its possible regulatory pathways, which will help to develop new therapies for MS.

MATERIALS AND METHODS Animals
The SARM1 allele (SARM1 f/f ) mice were generated as previously described [48]. GFAP-Cre mice (from Jackson Laboratory) were crossed with the Fig. 4 Demyelination and axonal injury were alleviated in the spinal cords of SARM1 GFAP -CKO EAE mice. A Western blot analysis of MBP expression in spinal cords of SARM1 f/f mice and SARM1 GFAP -CKO mice, and SARM1 f/f EAE mice and SARM1 GFAP -CKO EAE mice. The band of higher molecular weight may be 21.5 kDa isoform, while that of lower molecular weight may represent 18.5 kDa and 17.2 kDa isoform. B Quantitative analysis of MBP was shown in (A) (normalized to SARM1 f/f mice, n = 5). C The typical electron microscopic images of lumbar spinal cords in SARM1 f/f mice and SARM1 GFAP -CKO mice, and SARM1 f/f EAE mice and SARM1 GFAP -CKO EAE mice. Scale bars, 2 μm (low power image) and 1 μm (high power image). D Quantitative analysis of percentage of myelinated axons as shown in C (n = 4). E Quantitative analysis of G-ratio as shown in C (n = 50, two-way ANOVA with Bonferroni's post-tests). F Double immunostaining MBP (green) and NF (red) in spinal cords of SARM1 f/f mice and SARM1 GFAP -CKO mice, and SARM1 f/f EAE mice and SARM1 GFAP -CKO EAE mice. G Quantitative analysis of the relative expression of MBP and NF as shown in (F) (n = 5). H Immunostaining of GAP43 (red) in spinal cords of SARM1 f/f mice and SARM1 GFAP -CKO mice, and SARM1 f/f EAE mice and SARM1 GFAP -CKO EAE mice. I Quantitative analysis of the relative GAP43 expression as shown in (H) (n = 5). Scale bar, 50 μm. Data were mean ± SEM. Student's t-test unless otherwise indicated, n.s., not significant (p > 0.05), **p < 0.01, ***p < 0.001.
floxed SARM1 f/f mice to get SARM1 GFAP -CKO mice. SARM1 Aldh1L1 -CKO mice were generated by crossing the floxed SARM1 f/f mice with Aldh1L1-Cre transgenic mice (gifted from Zilong Qiu's lab, Institute of Neuroscience, Chinese Academy of Sciences). These mice were based on the C57BL/6 background and genotyped by PCR. All mice were housed in a specific pathogen-free (SPF) facility at Wenzhou Medical University under controlled temperature (22-25°C) with a 12-h light-dark cycle, with water and food provided ad libitum. All animal experiments strictly follow the guidelines of the Laboratory Animals Ethics Committee of Hangzhou Normal University and Wenzhou Medical University.

Establishment and treatment of EAE model in mice
The choice of sample size was based on experience with previously published EAE models and studies. The EAE model is as described Fig. 5 EAE was relieved with later onset, fewer neuronal death, demyelination, axonal injury and less inflammatory infiltration in SARM1 Aldh1L1 -CKO mice. A Western blot detected the expression of SARM1 in the cerebellum, cervical, thoracic and lumbar spinal cords of SARM1 f/f mice and SARM1 Aldh1L1 -CKO mice. B Quantification of protein levels of SARM1 as shown in A (normalized to SARM1 f/f mice, n = 3). C The weight of SARM1 f/f EAE and SARM1 Aldh1L1 -CKO EAE mice ranged from 0 to 21 dpi (n = 4 mice, two-way ANOVA with Bonferroni's posttests). D The EAE score of SARM1 f/f and SARM1 Aldh1L1 -CKO mice ranged from 0 to 21 dpi (n = 4 mice, two-way ANOVA with Bonferroni's posttests). E Typical images of Nissl staining in lumbar spinal cords of SARM1 f/f and SARM1 Aldh1L1 -CKO mice, and SARM1 f/f EAE and SARM1 Aldh1L1 -CKO EAE mice. F Quantitative analysis of the density of Nissl bodies as shown in E (n = 4). G The typical images of HE staining of lumbar spinal cords in SARM1 f/f and SARM1 Aldh1L1 -CKO mice, and SARM1 f/f EAE and SARM1 Aldh1L1 -CKO EAE mice. H Quantitative analysis of the density of infiltrating cells as shown in G (n = 4). I The typical images of CD45 + immunostaining in lumbar spinal cords of SARM1 f/f and SARM1 Aldh1L1 -CKO mice, and SARM1 f/f EAE and SARM1 Aldh1L1 -CKO EAE mice. J Quantitative analysis of the density of CD45 + cells as shown in I (n = 3). K The typical images of immunostaining of MBP (green) and NF (red) in lumbar spinal cords of SARM1 f/f and SARM1 Aldh1L1 -CKO mice, and SARM1 f/f EAE and SARM1 Aldh1L1 -CKO EAE mice. L Quantitative analysis of the intensity of MBP and NF as shown in K (n = 4). Scale bar, 50 μm. The data were mean ± SEM. Student's t-test unless otherwise indicated, n.s., not significant (p > 0.05), *p < 0.05, **p < 0.01, ***p < 0.001.
previously [49]. In short, female mice weighing 18 g to 20 g were injected with 0.2 mg of emulsified MOG  (1:1) (HY-P1240A, MedChemExpress) in Freund's complete adjuvant containing 8 mg/ml Mycobacterium tuberculosis (strain H37RA; Difco, USA). Two intraperitoneal injections of 300 ng of pertussis toxin (516561, Sigma) were respectively given at 0 and 48 h after immunization. The mice of the control group were injected with PBS in the same manner. Then the score (0.5 = tail weakness, 1 = tail paralysis, 2 = faltering gait with hind limb weakness, 2.5 = unilateral hind limb paralysis, 3 = bilateral hind limb paralysis, 4 = hind limb paralysis with forelimb weakness, 5 = near death) and body weight were recorded from days 0 to 21. STZ (FS0266, FuShen) was dissolved in sodium citrate solution (C1013, Solarbio) and intraperitoneally injected with 150 mg/kg to mice starved for more than 12 h. Every mouse was randomly assigned to the experimental groups. Evaluation of genotype and experimental conditions was blinded.

Hematoxylin-Eosin (HE) staining
In brief, the mice were perfused with 0.1 M PBS and 4% PFA after anesthetization. Then the spinal cords of the mice were immersed in 4% PFA for 24 h and transferred to 15% sucrose solution until they sank. In the same way, the spinal cords were transferred to 30% sucrose solution and dehydrated again. Subsequently, the spinal cords were embedded with OCT and coagulated at -20°C. The spinal cords were cut into 20 μm-thick sections using a freezing microtome (Thermo, USA) and then pasted on the adhesive glass slides. After being stained with hematoxylin for 1 min, the slides were washed in double distilled water, cultured in acidic liquid alcohol for 30 s, stained with eosin for 50 s, dehydrated with 95% ethanol and 100% ethanol for 1 min, and finally soaked in xylene for 10 min, fixed with neutral resin. Images were collected by microscope (Nikon, Tokyo, Japan) at room temperature and quantitatively analyzed by Image J software (Media Cybernetics, Bethesda, MD, USA).

Nissl's staining
Sections of the spinal cords were obtained as previously described. The 20 μm thick sections were soaked in 0.1% cresol violet at room temperature for 6 min, then washed with double distilled water for 5 min, dehydrated with 95% ethanol and 100% ethanol, respectively, and finally soaked in xylene for 10 min and fixed with neutral resin. The images were collected by microscope (Nikon, Tokyo, Japan) at room temperature and analyzed quantitatively by Image J.

Culture of astrocytes
In short, the cerebral cortex and hippocampus of newborn mice (P1-P3) were dissected and chopped in PBS under a microscope. After digestion with 0.25% trypsin (GIBCO) at 37°C for 15 min, the tissue was stopped digesting by adding DMEM medium containing 10% FBS. Then the tissue was mechanically destroyed to obtain single cell suspension. The cells were then seeded in a poly lysine (0.1 mg/ml, sigma Aldrich) coated culture flask. After the cells were cultured in a 37°, 5% CO 2 incubator for 6-10 d, microglia and oligodendrocytes were removed by shaking at 250 rpm for 4-6 h. Astrocytes were then isolated and plated into poly-D-lysine coated Petri dishes or coverslips. The purity of GFAP + cells was over 94%.

RNA sequencing and functional enrichment analysis
The total RNA was obtained from cultured SARM1 +/+ astrocytes and SARM1 -/astrocytes by using the RNeasy Mini kit (Qiagen). The conditions for sequencing RNA samples are A260:A280 ratio>1.8 and A260:A230 ratio>2.0. The method and analysis of sequence have been described previously [48]. In short, genes that meet the criteria of fold change >2 and adjusted p-value <0.05 are considered to be genes with disparity significance, which were analyzed using heatmap and Kyoto Encyclopedia of Genes and Genomes (KEGG) ontology enrichment.

Electron Microscopy and Quantitative Analysis
The method was followed as described previously [49]. Briefly, after anesthesia, mice were immediately dissected on ice and spinal cords were fixed with 2.5% glutaraldehyde overnight. Then the samples were immersed in 2 ml PBS (0.1 M, pH 7.0) for 3 times (10 min each time), fixed for 1.5 h with 1% OsO4 (SPI-CHEM) made from PBS, and then immersed in PBS for 3 times (10 min each time). After that, the samples were dehydrated with ethanol (30%, 50%, 70%, 80%, 90%, and 95%) for 15 min, then dehydrated with 95% acetone for 20 min, and then transferred to 100% acetone (Sinopharm Fig. 7 Inhibition of GDNF by STZ worsened the EAE deficits in SARM1 GFAP -CKO EAE mice. A The EAE score of SARM1 GFAP -CKO mice, SARM1 GFAP -CKO EAE mice and STZ-treated SARM1 GFAP -CKO EAE mice during EAE modeling process (n = 4). B Western blot detected NF-κB, GDNF and MBP in the spinal cords of SARM1 GFAP -CKO mice, SARM1 GFAP -CKO EAE mice and STZ-treated SARM1 GFAP -CKO EAE mice. C Quantification of protein levels of NF-κB, GDNF and MBP as shown in B (n = 4, normalized to β-actin). D Immunostaining of MBP (green) and NF (red) in spinal cords of SARM1 GFAP -CKO mice, and SARM1 GFAP -CKO EAE mice and STZ-treated SARM1 GFAP -CKO EAE mice. E Quantification of the relative intensity of MBP and NF (normalized to SARM1 GFAP -CKO mice, n = 3). F Immunostaining of GAP43 (red) in spinal cords of SARM1 GFAP -CKO mice, and SARM1 GFAP -CKO EAE mice and STZ-treated SARM1 GFAP -CKO EAE mice. G Quantification of the relative intensity of GAP43 as shown in F (normalized to SARM1 GFAP -CKO mice, n = 3). H Typical images of immunostaining of NeuN + cells in spinal cords of SARM1 GFAP -CKO mice, and SARM1 GFAP -CKO EAE mice and STZ-treated SARM1 GFAP -CKO EAE mice. I Quantitative analysis of the density of NeuN + cells as shown in H (n = 3). J Typical images of immunostaining of CD45 + cells in spinal cords of SARM1 GFAP -CKO mice, and SARM1 GFAP -CKO EAE mice and STZ-treated SARM1 GFAP -CKO EAE mice. K Quantitative analysis of the density of CD45 + cells as shown in J (n = 3). Scale bar, 50 μm. Data were mean ± SEM., One-way ANOVA with Bonferroni's post-tests, *p < 0.05, **p < 0.01, ***p < 0.001. Fig. 8 A working model of astrocytic SARM1's function in EAE mice. SARM1 is upregulated in the astrocytes of spinal cords in EAE. Astrocytic SARM1 inhibits the expression of GDNF, which promotes the neuronal death, neuroinflammation and demyelination in EAE through NF-κB signaling.