SARM1 acts downstream of neuroinflammatory and necroptotic signaling to induce axon degeneration

While there are many different causes of neurodegenerative diseases, targeting common mechanisms promoting neuronal dysfunction and death would be a powerful approach for treating these devastating disorders. Early efforts to identify a common mechanism focused on neuronal apoptosis in hopes of blocking neuronal cell death. While tremendous strides were made in delineating the molecular mechanisms of apoptosis in neurons, targeting this pathway has not generated new treatments for neurodegeneration (Mattson, 2000; Maino et al., 2017). There are multiple reasons for this failure. First, we now know that there are many other cell death pathways that can kill neurons (Fuchs and Steller, 2015). Indeed, necroptosis is now appreciated to be a druggable pathway of programmed necrosis that is a very exciting target in many neurodegenerative diseases (Zhang et al., 2017; Yuan et al., 2019). Second, neurodegenerative diseases activate not only cell death pathways but also axonal death programs (Simon and Watkins, 2018). As synapse and axon loss is an early event in many neurodegenerative disorders (Coleman and Höke, 2020), mechanisms to maintain neuronal connectivity may need to be integrated with approaches for blocking neuronal cell death in order to maintain a functional nervous system. Finally, neurons are not the only cells affected in neurodegenerative diseases. In fact, neuroinflammation and microglial function is a central driver of many neurodegenerative diseases (Hickman et al., 2018; Perry and Holmes, 2014). Using both an in vivo glaucoma model and an in vitro neuronal culture model, we now explore functional and molecular links among neuroinflammatory signaling, necroptosis, and axonal degeneration.

Glaucoma is a neurodegenerative disease of retinal ganglion cells (RGCs) and a major preventable cause of blindness (Weinreb et al., 2014). While increases in intraocular pressure (IOP) are the proximate cause of most cases of glaucoma, neuroinflammation is a major contributor to the pathogenesis of glaucoma. Indeed, glaucoma can develop in the absence of IOP, and neurodegeneration can proceed even after successful restoration of normal pressures (Beidoe and Mousa, 2012). In both patients and rodent models, TNF-α is up-regulated in glaucomatous eyes (Yan et al., 2000; Sawada et al., 2010; Tezel et al., 2001). In experimental models, TNF-α is sufficient to activate microglia, stimulating a positive feedback loop that is capable of inducing RGC axon loss and eventual cell death (Nakazawa et al., 2006; Roh et al., 2012). Here, we used this neuroinflammatory model of glaucoma to test the hypothesis that SARM1 (sterile alpha and TIR motif 1), the central executioner of pathological axon degeneration, contributes to the pathogenesis of glaucoma.

SARM1 is an inducible NAD⁺ cleavage enzyme whose activation triggers metabolic catastrophe and axon degeneration in both classic injury-induced Wallerian degeneration and the dying back axon loss of peripheral neuropathy (Gerdts et al., 2016; Geisler et al., 2016; Essuman et al., 2017). Following optic nerve crush, SARM1 is required for distal RGC axon degeneration, but is not necessary for injury-dependent apoptosis and RGC death (Fernandes et al., 2018). While it is clear that SARM1 functions in RGCs to promote Wallerian degeneration, whether it acts in more disease-relevant models of RGC death is unknown. Overexpression of the fusion protein Wld^s, which includes the NAD⁺ biosynthetic enzyme NMNAT1 (nicotinamide nucleotide adenylyltransferase 1) and so can inhibit SARM1 (Sasaki et al., 2016), has given conflicting results in glaucoma models. Wld^s can protect axons but not cell bodies in a rat glaucoma model (Beirowski et al., 2008), while the coadministration of the NAD⁺ precursor nicotinamide in conjunction with Wld^s (Williams et al., 2017a) or the overexpression of cytosolically targeted NMNAT1 enzyme (Zhu et al., 2013) can protect axons and cell bodies. Whether NMNAT1/Wld^s is primarily augmenting NAD⁺ biosynthesis or inhibiting SARM1 in these paradigms is unknown. Hence, it is important to test the requirement for SARM1 directly in a glaucoma model, in particular because SARM1 is an enzyme and so a druggable target (DiAntonio, 2019). Finally, mechanisms of SARM1 activation are well understood in Wallerian degeneration, where axotomy blocks delivery of labile axon survival factors that inhibit SARM1 (Figley and DiAntonio, 2020). However, whether and how other disease processes activate SARM1 are not known.

Here, we test the role of SARM1 in a neuroinflammatory model of glaucoma, demonstrating that SARM1 is required for RGC axon loss, RGC death, and oligodendrocyte death following intravitreal TNF-α application. We show that oligodendrocyte death is secondary to axon loss in this glaucoma model using a therapeutically translatable adeno-associated virus (AAV) gene therapy approach to express a SARM1 dominant-negative transgene (Geisler et al., 2019b). We explore the mechanisms linking TNF-α signaling to SARM1 in neuronal culture, demonstrating that in a TNF-α–dependent neuronal necroptosis model, SARM1 induces axon degeneration. Our findings demonstrate that in contrast to canonical necroptotic cell death in which MLKL (mixed lineage kinase domain-like pseudokinase) is the final executioner, there exists a specialized “axonal necroptosis” process in which MLKL works through the SARM1 axon degeneration pathway to drive axon demise. Activated MLKL induces loss of the axon survival factors NMNAT2 and SCG10/STMN2, known inhibitors of SARM1 activation that are also lost following axotomy, chemotherapy-induced neurotoxicity, and mitochondrial dysfunction (Geisler et al., 2019a; Loreto et al., 2020; Summers et al., 2020; Gilley et al., 2015). Hence, disparate upstream pro-degenerative signals converge on these axon survival factors to activate SARM1. These studies identify molecular mechanisms linking neuroinflammation, necroptosis, and SARM1 activation, and suggest that local inflammatory signaling in the axon may drive SARM1-dependent axon loss. This unification of three major drivers of neurodegeneration may have broad implications for treatment of neurological diseases.

Glaucoma is a neurodegenerative disease in which increased IOP leads to neuroinflammation, triggering loss of optic nerve axons and subsequent death of RGCs (Tezel and Wax, 2004; Bosco et al., 2011; Williams et al., 2017b, 2017c). SARM1 is the central executioner of axon degeneration in other pathological conditions (Henninger et al., 2016; Geisler et al., 2016), and we therefore investigated whether SARM1 promotes the loss of RGC axons in response to TNF-α injection into the vitreous cavity (Fig. 1 A), an established neuroinflammatory model of glaucoma.

We injected TNF-α or PBS into the vitreous cavity of 5-wk-old WT and SARM1 knockout (KO) mice. We assessed axon number in the optic nerve via EM and RGC number via Brn3a staining of cell bodies in retinal whole mounts at 2 wk and 2 mo after treatment. In this model, axon loss occurs by 2 wk, while RGC loss is complete by 2 mo (Kitaoka et al., 2006, 2009, 2013; Nakazawa et al., 2006; Roh et al., 2012). In agreement with prior findings, optic nerves from TNF-α–treated WT eyes demonstrated an ∼37% reduction in axon number, a degree of degeneration that would likely lead to abnormalities in visual field testing in human patients (Kerrigan-Baumrind et al., 2000). In WT axons, there is also a concomitant ∼26% increase in axon size after 2 wk (Fig. 1, C and D). Analysis of the distribution of axon sizes at 2 wk does not indicate the loss of a distinct cohort of axons, as the distribution of axon sizes is similar between PBS and TNF-α–treated WT axons (Fig. 2 E). In contrast, TNF-α treatment in SARM1 KO mice did not lead to significant differences in optic nerve axon number or axon size. There is no significant further loss of axons 2 mo following TNF-α injection in WT mice; however, there is a continued increase in axon size, driven largely by a population of large axons (>1.6 µm) that are not apparent in PBS-treated or -untreated optic nerve (Fig. 1, B–E). In contrast, SARM1 KO mice remain resistant to the effects of intravitreal TNF-α injection, showing no significant loss of total axon number or increase in axon size at 2 mo following TNF-α treatment (Fig. 1, B–E). Hence, there is not merely a delay in axon degeneration in SARM1 KO mice; instead, SARM1 is required for axon loss in this TNF-α–triggered glaucoma model.

Having assessed axon number, we next quantified RGC survival following intravitreal TNF-α injection. We stained retinas with Brn3a to identify RGCs. While no RGC loss was observed at 2 wk (Kitaoka et al., 2006, 2009), there was significant loss of RGCs in WT retinas 2 mo after TNF-α treatment (Fig. 1, F and G). SARM1 KO mice were fully resistant to TNF-α–induced RGC death, showing no significant decrease in RGC cell number at 2 mo (Fig. 1, F and G). Hence, loss of SARM1 protects both RGC axons and cell bodies in this model of TNF-α–triggered glaucoma. The temporal pattern of axon and cell body loss is consistent with the idea that RGC death is a consequence of axon degeneration.

Having demonstrated that loss of SARM1 protects RGC axons and cell bodies from TNF-α–induced degeneration, we next assessed whether SARM1 influenced other TNF-α–dependent events. Intravitreal TNF-α injection activates microglia that then amplify the inflammatory response (Nakazawa et al., 2006; Roh et al., 2012; Son et al., 2010). To quantify microglial activation, we measured the size of microglial soma, as soma size increases upon microglial activation while processes become less complex (Davis et al., 2017). Intravitreal injection can temporarily activate microglia (Seitz and Tamm, 2014), so we first performed control intravitreal PBS injections. In WT mice, microglial soma size increased within 1 d and returned to a normal soma size with more complex processes by 3 d. In age-matched SARM1 KO mice, microglial soma size was also normal 3 d following intravitreal PBS injection (Fig. 2, A and B). In contrast, intravitreal TNF-α injection led to microglial activation in both WT and SARM1 KO retina that persisted for at least 1 wk (Fig. 2, A and B). Therefore, intravitreal TNF-α stimulates a sustained neuroinflammatory response in the retina that is not SARM1-dependent. Hence, SARM1 is likely acting downstream of this neuroinflammatory response to promote axon loss and RGC death.

While RGC death is central to the pathogenesis of glaucoma, oligodendrocytes in the optic nerve are also lost in models of glaucoma (Nakazawa et al., 2006; Son et al., 2010). Oligodendrocytes both insulate and provide metabolic support to RGC axons in the optic nerve, and so we investigated whether oligodendrocytes are also preserved in the absence of SARM1 in this glaucoma model. We stained for CC1, a marker of nonproliferative mature oligodendrocytes (Payne et al., 2013), in the optic nerves of TNF-α–treated WT and SARM1 KO mice. In WT mice, TNF-α treatment decreased the number of CC1-positive oligodendrocytes by ∼30% after 1 wk. Despite this loss in oligodendrocyte number, there is no change in the G-ratio of surviving axons (Fig. 2 E). Hence, oligodendrocyte loss likely occurs concomitantly with axon loss, rather than inducing aberrant myelination of surviving axons. In SARM1 KO animals, TNF-α treatment induced no significant loss of oligodendrocytes in the optic nerve (Fig. 2, C and D). Hence, SARM1 is required for oligodendrocyte death in response to neuroinflammatory signaling.

The preservation of oligodendrocytes in TNF-α–treated SARM1 KO mice could reflect a cell-autonomous action of SARM1 in oligodendrocytes to drive cell death, or alternatively a noncell autonomous role for axons in maintaining oligodendrocytes. To distinguish between these possibilities, we took advantage of a dominant-negative SARM1 (SARM1.DN) that potently blocks SARM1 function in vivo following viral delivery (Geisler et al., 2019b). SARM1-DN fused to EGFP or WT SARM1 fused to EGFP were cloned into an AAV vector and expressed under control of the neuron-specific human synapsin (hSyn) promoter. A combination of the 7m8 capsid (Dalkara et al., 2013; Büning and Srivastava, 2019) and hSyn promoter enabled robust retinal expression that is limited to neurons (Fig. 2 F). 4 wk after intravitreal injection of AAV2-hSyn-SARM1-DN-EGFP (AAV-SARM1.DN) or control virus (AAV-SARM1.WT), we observed robust EGFP labeling of RGCs and axons (Fig. 2 F; and Fig. S1, A and B). In contrast, we do not observe expression in oligodendrocytes in the optic nerve (Fig. S1, C and D), allowing us to test whether neuron-specific inhibition of SARM1 is sufficient to block TNF-α induced oligodendrocyte death. Expression of WT SARM1 (AAV-SARM1.WT) showed the expected loss of oligodendrocytes 1 wk after intravitreal injection of TNF-α. In contrast, oligodendrocyte loss was significantly attenuated by expression of the SARM1-DN in RGCs (Fig. 2, G and H). These findings indicate that loss of oligodendrocytes is secondary to SARM1-dependent function in RGCs, and suggests that maintenance of healthy axons is required for oligodendrocyte survival in response to neuroinflammatory signals.

Taken together, this series of experiments demonstrates that SARM1 does not block the ability of TNF-α to induce neuroinflammatory stimulation of microglia, but is required for subsequent oligodendrocyte death, axon loss, and RGC cell death. Hence, SARM1 is an essential downstream mediator of neuroinflammatory damage in this model of glaucoma, and so is a candidate therapeutic target for preventing neurodegeneration in glaucoma.

Having explored the cellular consequences of intravitreal TNF-α injection, we next investigated activation of pro-degenerative molecular pathways. TNF-α can trigger apoptosis or necroptosis in different contexts, including in promoting RGC death in models of glaucoma (Fuchs and Steller, 2015; Weinlich et al., 2017; Do et al., 2017; Almasieh et al., 2012). Hence, we used immunoblotting to assess molecular markers of apoptosis and necroptosis in the optic nerve following intravitreal TNF-α injection. We did not detect cleaved caspase 3 via immunoblotting of optic nerve from TNF-α–injected eyes of either WT or SARM1 KO mice (Fig. S2 A), consistent with our previous demonstration that SARM1 does not participate in apoptosis (Gerdts et al., 2013). We next investigated whether intravitreal TNF-α activates necroptosis, an inflammatory form of programmed cell death (Kearney and Martin, 2017; Silke et al., 2015; Heckmann et al., 2019). To assess activation of necroptotic signaling, we assayed levels and phosphorylation of RIPK3 and MLKL, two essential proteins for necroptosis (Ofengeim et al., 2015; Caccamo et al., 2017; Ito et al., 2016; Vitner et al., 2014). At 3 d following intravitreal injection of TNF-α or PBS into WT and SARM1 KO mice, we harvested optic nerves and performed immunoblot analysis. In response to TNF-α injection, there was a significant increase in the levels of phosphorylated MLKL, total MLKL, phosphorylated RIPK3, and total RIPK3 in optic nerves of both WT and SARM1 KO mice (Fig. 2, I–L). Similar induction of necroptotic signaling proteins is detected in other neurodegenerative diseases (Ofengeim et al., 2015; Caccamo et al., 2017; Vitner et al., 2014; Ito et al., 2016; Zhang et al., 2017). Hence, TNF-α activates necroptotic signaling in the optic nerve in a SARM1-independent manner, suggesting that SARM1 may function downstream of necroptosis to promote axon degeneration and cell death.

To explore the relationships among TNF-α, necroptosis, and SARM1, we turned to cultured dorsal root ganglion (DRG) neurons, a pure neuronal population amenable to a mechanistic analysis of axon degeneration (Fig. 3 A). We first developed a TNF-α–dependent in vitro model of necroptosis in DRG neurons. As there is cross-talk between apoptosis and necroptosis (Heckmann et al., 2019), in cultured cells TNF-α can be biased to stimulate necroptosis by simultaneously inhibiting apoptosis with Z-VAD (carbobenzoxy-valyl-alanyl-aspartyl-[O-methyl]-fluoromethylketone) and blocking survival signaling with a SMAC (second mitochondria-derived activator of caspases) mimetic (Gong et al., 2017; Wang et al., 2012). Using this cocktail, we found that TNF-α treatment of DRG neurons induces the phosphorylation of MLKL. This increase in phosphorylated MLKL is blocked by necrostatin-1s, an inhibitor of RIPK1, a key component of the necrosome required for MLKL phosphorylation during necroptosis (Fig. 3 B). Next, we assessed whether TNF-α triggers axon degeneration in cultured DRG neurons. Axons were imaged over time using an Operetta high-content imaging system. Axon morphology was assessed via bright-field imaging, and axonal energetics were examined using tetramethylrhodamine methylester (TMRM), a fluorescent reporter of mitochondrial potential. TNF-α induced axonal swelling by 48 h and progressed to frank degeneration by 72 h, at which point there was a dramatic loss in TMRM signal (Fig. 3 C). This TNF-α–induced axon degeneration is driven by necroptosis, as it is prevented in the presence of necrostatin-1s. Importantly, both the TNF-α–induced axonal degeneration and loss of mitochondrial potential are fully blocked in SARM1 KO neurons (Fig. 3, C–E). Hence, TNF-α and necroptosis require SARM1 to induce axon degeneration. While this cytokine-induced axon degeneration is slower than axotomy-induced axon degeneration, the common requirement for SARM1 demonstrates that neuroinflammatory and necroptotic signaling in the axon share molecular mechanisms with Wallerian degeneration.

While TNF-α can activate a model of necroptosis in DRG neurons, in other cell types it can activate multiple death pathways (Weinlich et al., 2017; Kearney and Martin, 2017; Heckmann et al., 2019), making it more difficult to dissect the mechanistic relationship between necroptosis and SARM1-dependent degeneration. To test directly the hypothesis that SARM1 functions downstream of necroptosis, we used an engineered variant of MLKL that allows for temporal and spatial induction of necroptosis (Gong et al., 2017; Orozco et al., 2014; Fegan et al., 2010; Zhang et al., 2016). The N-terminal domain of MLKL (MLKL.ND) was fused to Fkbp^F36v (Fig. 4 A, labeled with yellow), a protein whose dimerization is induced in the presence of the rapamycin analogue B/B homodimerizer (BB; star). Forced dimerization of MLKL.ND with BB in DRG neurons shows several features of necroptosis within 24 h (Fig. 4 B), including loss of cytosolic GFP consistent with membrane rupture, externalization of phosphatidylserine as shown by annexin V staining, and cell death as marked by nuclear accumulation of ethidium homodimer (EthD), which does not enter live cells (Fig. 4 B). Interestingly, a few neurons with annexin V membrane staining (arrowhead, BB_24hr panel) show intact GFP fluorescence in cytosol and no EthD fluorescence, suggesting that they are still alive. These findings are consistent with the loss of membrane asymmetry induced by ATP depletion or dysfunction of ATP-dependent enzymes preceding membrane rupture during necroptosis (Gong et al., 2017; Hankins et al., 2015). Taken together, these findings show dimerized MLKL.ND causes necroptosis in DRG neurons.

We next tested whether necroptosis induced by dimerization of MLKL.ND triggers axon degeneration and assessed the requirements for signaling proteins. We find that dimerization of MLKL.ND causes axon degeneration and loss of TMRM staining of WT axons within 24 h. Direct dimerization of MLKL.ND should bypass the requirement for upstream RIPK activity, and as predicted the RIPK1 inhibitor necrostatin-1s did not prevent axon degeneration following dimerization of MLKL.ND (Fig. S3, A and B). Dimerized MLKL.ND can interact with endogenous MLKL to mediate downstream necroptotic signaling (Tanzer et al., 2016). We knocked down MLKL with an shRNA targeting the C terminus of MLKL, so that it would not impact the levels of the dimerizable MLKL.ND construct, and validated the efficacy of the shRNA in both DRG neurons and in N2A cells that express high levels of MLKL (Fig. S4). Upon knockdown of endogenous MLKL, the dimerization of MLKL.ND does not trigger axon degeneration (Fig. S5 A). In addition, the dimerization of MLKL.ND increases the level of phosphorylated full-length MLKL, and this is not blocked by necrostatin (Fig. S5, B and C). Hence, in this system, dimerized MLKL.ND recruits endogenous MLKL to mediate downstream axon degeneration.

With this system in hand, we asked if axon degeneration induced by MLKL.ND dimerization requires SARM1. In contrast to WT neurons, axons from SARM1 KO neurons were fully protected from both axon degeneration and loss of mitochondrial potential for at least 48 h after MLKL.ND dimerization (Fig. 4, C and D). We also assessed calcium influx, a common feature of necroptosis and Wallerian degeneration. While the mechanism of calcium influx during necroptosis is not fully understood, there is a consensus that calcium influx is the consequence of membrane permeabilization induced by MLKL (Wang et al., 2014; Su et al., 2014; Cai et al., 2014; Xia et al., 2016). We used live-cell imaging to simultaneously monitor intracellular calcium using GCaMP6 (Chen et al., 2013) and axonal integrity with mRuby3, a photostable fluorescent cytosolic protein (Bajar et al., 2016). In WT axons, dimerization of MLKL.ND triggered calcium influx and axon degeneration that was completely prevented in axons from SARM1 KO neurons (Fig. 4, E and F; and Video 1). This is a surprising finding, since in other systems MLKL induces calcium influx via direct membrane permeabilization. In contrast, in axonal necroptosis, SARM1 becomes activated and triggers calcium influx and axon degeneration. Interestingly, this relationship is selective for the axon, as in the cell body the activation of necroptosis via dimerization of MLKL.ND triggered calcium influx followed by membrane rupture in both WT and SARM1 KO neurons (Video 1). Indeed, there is no significant difference in the extent of cell death induced by necroptosis in WT versus SARM1 KO neurons (Fig. S6). Taken together, these findings suggest the exciting hypothesis that canonical necroptotic signaling drives cell death in sensory neurons, but that a specialized form of axonal necroptosis uses a distinct SARM1-dependent mechanism to trigger calcium influx and axon degeneration.

As a first test of this hypothesis, we investigated whether other components of the SARM1 Wallerian degeneration pathway function downstream of necroptosis for axon degeneration. In classical Wallerian degeneration, overexpression of the axon survival factor cytoNMNAT1 and inhibition of the neuronal stress kinases DLK (dual leucine zipper kinase) and LZK (leucine zipper-bearing kinase) potently inhibit axotomy-induced axon degeneration (Summers et al., 2018). Upon MLKL.ND dimerization-induced necroptosis, cytoNMNAT1 overexpression and treatment with the DLK/LZK inhibitor GNE-3511 both potently block axon degeneration (Fig. S7). These findings support the hypothesis that axonal necroptosis can stimulate the canonical Wallerian degeneration program to drive axon loss. DLK inhibition can block apoptotic neuronal cell death (Ghosh et al., 2011), but as with SARM1 KO, does not block necroptotic cell death induced by MLKL.ND dimerization (Fig. S7). These findings highlight that necroptosis-induced cell death and necroptosis-induced axon degeneration are mechanistically distinct.

While necroptosis induces degeneration via distinct mechanisms in the cell body and the axon, MLKL.ND is dimerized throughout the cell, and so it is not possible to determine whether there is a local axonal necroptotic program or whether the axonal effects are secondary to cell body death. To distinguish between these possibilities, we used microfluidic devices to spatially isolate axons from cell bodies, allowing us to test whether dimerization of MLKL.ND exclusively in axons is sufficient to trigger axon degeneration (Fig. 5 A). Dissociated DRG neurons expressing MLKL.ND were seeded in the somal compartment and extended their axons into the axonal compartment of the microfluidic devices.

To test the hypothesis that necroptosis can occur axon autonomously, we performed three independent experiments. First, addition of the BB dimerizing agent to the axonal side of microfluidic devices was sufficient to induce loss of axonal TMRM staining and local axon degeneration without cell body death. Moreover, this axon degeneration and TMRM loss require SARM1 (Fig. 5, B and C). Hence, local axonal induction of necroptosis in the absence of cell death induces SARM1-dependent axon degeneration. While local axon degeneration is robust in this paradigm, the extent of axon degeneration in microfluidic devices is less dramatic than in normal culture dishes, suggesting that concurrent induction of necroptosis in the soma may contribute to this degenerative event (compare with Fig. 4 C). Second, we observed local calcium influx in axons following axonal-induction of necroptosis. Following axonal application of the BB dimerizer, live-cell imaging revealed a large calcium influx in axons, but not in soma, of WT but not SARM1 KO neurons (Fig. 5 D and,Video 2 shows absence of somal calcium influx over time). Taken together, these two experiments show that axonal activation of necroptosis triggers local, SARM1-dependent loss of mitochondrial potential, calcium influx, and axon degeneration.

These studies show that local necroptosis can stimulate the SARM1 degeneration program, but they rely on an engineered system to activate axonal necroptosis. We wondered whether endogenous mechanisms can activate axonal necroptosis in the absence of cell body necroptosis. To test this hypothesis, we asked whether axon-only application of TNF-α could induce local necroptosis as assayed by phosphorylation of MLKL. For immunoblot analysis, microfluidic devices do not provide enough axonal material. Hence, we made spot cultures of SARM1 KO DRG neurons on standard culture dishes, allowed axons to extend, and then removed all cell bodies. Since SARM1 blocks axotomy-induced axon degeneration, these axons remain intact and metabolically healthy for at least 5 d (Geisler et al., 2019b), thus providing adequate time to examine necroptotic signaling. First, we showed that severed axons treated with DMSO for 40 h do not show increases in phosphorylated MLKL (pMLKL), demonstrating that axotomy alone does not increase pMLKL (Fig. 5 E). We then applied TNF-α to these isolated severed axons, collected samples 40 h after TNF-α application, and found that TNF-α triggers an increase in the levels of phosphorylated MLKL that is blocked by the application of necrostatin. Hence, local TNF-α can trigger the necroptotic cascade in axons (Fig. 5 E). This could have important implications for diseases such as multiple sclerosis, where local neuroinflammatory signaling could directly impact axonal health.

We next investigated the mechanism by which dimerized MLKL.ND activates SARM1 to promote axon degeneration. For axonal maintenance, axon survival factors such as NMNAT2 and SCG10/STMN2 are continuously supplied to the axon to inhibit SARM1 (Gilley et al., 2015; Walker et al., 2017; Shin et al., 2012). Therefore, we tested whether necroptosis decreases NMNAT2 and SCG10 levels in axons. After activation of necroptosis via dimerization of MLKL.ND, axonal NMNAT2 levels significantly decreased over the ensuing 16 h (Fig. 6, A and B). This loss of NMNAT2 only occurs in axons, as there is no change in the levels of somal NMNAT2 (Fig. 6, C and D). Moreover, the levels of axonal SCG10 also decreased after dimerization of MLKL.ND, showing that coregulation of these axon maintenance factors occurs downstream of necroptosis (Fig. 6, C and D). Finally, having demonstrated that endogenous MLKL is required for axon degeneration induced by MLKL.ND dimerization (Fig. S5), we also find that endogenous MLKL is required for the rapid loss of axonal NMNAT2 after dimerization of MLKL.ND (Fig. 6, E and F). Hence, axonal necroptosis triggers the rapid loss of NMNAT2.

SARM1 is an inducible NAD⁺ cleavage enzyme. During Wallerian degeneration, SARM1 cleaves NAD⁺, triggering a metabolic crisis with loss of ATP and subsequent axon degeneration (Fig. 7 A; Sasaki et al., 2020; Essuman et al., 2017; Gerdts et al., 2016). However, SARM1 is also a TLR (Toll-like receptor)-adaptor protein that may participate in traditional innate immune signaling (Mukherjee et al., 2013) separate from its newly found role as an enzyme. Hence, we tested whether this SARM1 NADase activity is required for necroptosis-triggered axon degeneration. First, we assessed the requirement for catalytically active SARM1. The glutamic acid at position 642 is the catalytic residue of the SARM1 NADase, and mutations at position 642 render SARM1 incapable of mediating Wallerian degeneration (Essuman et al., 2017). Here we dimerized MLKL.ND in SARM1 KO neurons expressing GFP, WT SARM1, or the catalytically dead SARM1-E642A. MLKL.ND dimerization triggers axon degeneration only in neurons expressing WT SARM1, demonstrating that catalytically active SARM1 is required for necroptosis-induced axon degeneration (Fig. 7, B and C). Next, we tested whether axon degeneration could be blocked downstream of SARM1 by boosting synthesis of NAD⁺ or ATP. During Wallerian degeneration, addition of the NAD⁺ precursor nicotinamide riboside (NR) while overexpressing NR kinase (NRK1) increases NAD⁺ levels and delays axon degeneration (Sasaki et al., 2016), while incubating neurons with methyl-pyruvate boosts ATP levels (Yang et al., 2015; Cheong et al., 2011). In the presence of NRK1 plus NR or methyl-pyruvate, axon degeneration induced by MLKL.ND dimerization is blocked for at least 24 h. At 48 h, axons are beginning to show minor signs of degeneration (Fig. 7 E). These findings indicate that necroptosis induces a SARM1-dependent metabolic crisis in the axon, and suggest that the mechanisms downstream of SARM1 are likely the same whether SARM1 is activated by axotomy or necroptosis.

Our understanding of the molecular mechanisms of neurodegeneration has expanded dramatically with the realization that neuroinflammation contributes to most neurodegenerative diseases (Hammond et al., 2019), that necroptosis is likely as important as apoptosis in driving neuronal cell death (Weinlich et al., 2017; Yuan et al., 2019), and that SARM1 is the central executioner of an axonal self-destruction program activated in diverse neurodegenerative conditions (Figley and DiAntonio, 2020). Here, we unite these disparate degenerative pathways, identifying molecular and functional links among neuroinflammation, necroptosis, and SARM1-mediated axonal self-destruction. These studies support a model in which local neuroinflammatory signals can activate an axonal form of necroptosis that in turns triggers SARM1 NADase activity and axonal demise. These findings have important mechanistic and therapeutic implications for a broad range of neurodegenerative disorders.

While high IOP is the proximate cause of most cases of glaucoma, it is now appreciated that glaucoma is a neurodegenerative disease with a major neuroinflammatory component (Weinreb et al., 2014). Indeed, glaucoma can occur despite normal IOPs, and many patients with high IOP experience continued visual field loss even after restoration of normal pressure (Beidoe and Mousa, 2012). Hence, there is great interest in developing neuroprotective strategies that could be combined with pressure-lowering medications for the treatment and prevention of glaucoma. Here we identify SARM1 as a therapeutic target for neuroprotection in glaucoma.

Using a neuroinflammatory model of glaucoma induced by intravitreal TNF-α injection, we find that SARM1 is required for RGC axon loss and cell death as well as oligodendrocyte loss in the optic nerve. In this system, microglia are activated within days, oligodendrocyte death and RGC axon loss occurs within 2 wk, and ultimately RGC neurons die by 2 mo. Our data demonstrate that SARM1 is not required for microglia activation. Therefore, we suggest that SARM1 functions downstream of neuroinflammatory signaling to promote axonal self-destruction, which in turn promotes oligodendrocyte and RGC death. A variety of results supports the model that axon loss is the driver of oligodendrocyte and RGC death. First, using viral delivery of a neuronally expressed SARM1 dominant-negative transgene provides noncell autonomous protection to oligodendrocytes following TNF-α treatment. Since this oligodendrocyte loss occurs well before RGC death but around the same time as RGC axon loss, it is likely secondary to SARM1-mediated axonal degeneration. Second, while SARM1 can promote neuronal cell death, it does not promote apoptotic cell death (Gerdts et al., 2013; Summers et al., 2014), the likely mechanism of RGC death in glaucoma (Syc-Mazurek and Libby, 2019). Indeed, inhibiting apoptosis blocked RGC death in the DBA/2 chronic glaucoma model, but failed to protect optic nerve axons (Libby et al., 2005). In contrast, following optic nerve transection, SARM1 KO blocks degeneration of the severed distal axons, but does not protect against cell body death, which instead is mediated by retrograde axonal signals that activate apoptosis (Welsbie et al., 2013; Simon and Watkins, 2018; Syc-Mazurek and Libby, 2019). The different requirements for SARM1 following optic nerve transection and TNF-α application are instructive. Transection directly disrupts axons, thereby triggering retrograde apoptotic signals, while with TNF-α treatment, SARM1-induced axon degeneration is the cause of the axon loss that likely activates the apoptotic signals. Hence, by inhibiting SARM1, axons are protected and secondary apoptotic signaling is prevented.

A role for SARM1 in glaucoma may explain the mechanism of metabolic dysfunction in glaucomatous optic nerves. In DBA/2J, ATP levels in optic nerve are dramatically reduced at 6 mo, well before optic nerve loss (Baltan et al., 2010; Inman and Harun-Or-Rashid, 2017; Libby et al., 2005; Howell et al., 2007). SARM1 is an inducible NAD⁺ cleaving enzyme, and SARM1 activation induces axonal NAD⁺ loss that is followed by ATP loss, likely due to the central role of NAD⁺ in both oxidative phosphorylation and glycolysis (Essuman et al., 2017; Gerdts et al., 2016). If neuroinflammation induces low-level SARM1 activation, this may trigger metabolic vulnerability of glaucomatous optic axons and explain the protective effects of NAD⁺ supplementation (Kitaoka et al., 2009; Williams et al., 2017b) and overexpression of the NAD biosynthetic enzyme NMNAT in many (Howell et al., 2007; Kitaoka et al., 2009; Williams et al., 2017a; Zhu et al., 2013), but not all, glaucoma models (Beirowski et al., 2008).

TNF-α is up-regulated in the vitreous, retina, and optic nerves of glaucoma patients (Yan et al., 2000; Sawada et al., 2010; Tezel et al., 2001; Huang et al., 2010) as well as in response to increased IOP in rodent models (Roh et al., 2012; Nakazawa et al., 2006; Cueva Vargas et al., 2015). Pharmacological and genetic manipulation of TNF-α signaling demonstrates that it promotes microglial activation, yet how that ultimately triggers RGC axon loss and death remains unknown. Here we show that TNF-α application rapidly induces the up-regulation and phosphorylation of RIPK3 and MLKL in optic nerve, consistent with induction of necroptosis. We also show that TNF-α–dependent loss of oligodendrocytes, RGC axons, and RGC cell bodies is SARM1-dependent. These data, in conjunction with prior studies in glaucoma, are consistent with a model in which increases in IOP trigger TNF-α release, which in turn activates microglia, initiating a positive feedback loop that leads to further release of damaging cytokines. These inflammatory factors may then act on RGC axons to trigger an axonal form of necroptosis that activates SARM1 NADase activity. Accordingly, activated SARM1 would impair RGC axonal metabolism and promote axon degeneration and loss of associated oligodendrocytes. Once the axon is disconnected from its target, retrograde apoptotic signals would promote death of RGCs. If this model were correct, then pharmacological or gene therapy approaches to inhibit SARM1 would be neuroprotective in glaucoma, maintaining axonal integrity and blocking downstream death of oligodendrocytes and RGCs. Moreover, such a mechanism is unlikely to be limited to the eye, and may be broadly relevant to neurodegenerative diseases associated with neuroinflammation.

In classical Wallerian degeneration, the mechanism of SARM1-dependent axon degeneration is well understood. In a healthy axon, SARM1 is held inactive via inhibition by the axonal survival factors NMNAT2 and SCG10/STMN2. Both NMNAT2 and SCG10 are labile with very short half-lives, and so they must be constantly replenished in the axon via fast axonal transport. Upon axotomy, no new NMNAT2 or SCG10 can be delivered to the distal axon, and levels of these axon survival factors consequently drop, allowing stimulation of SARM1 NADase activity and triggering a metabolic crisis and axonal degeneration. A similar mechanism is thought to occur in response to neurotoxic chemotherapeutics such as vincristine, which disrupt axonal transport without a frank transection, and so would be expected to reduce NMNAT2 and SCG10 at the most distal portion of the axon. This likely explains the distal dying-back pathology that is a hallmark of peripheral neuropathy. Here, we greatly expand the reach of this model for SARM1 activation, demonstrating that neuroinflammatory and necroptotic signaling in the axon reduces the levels of the axonal survival factors NMNAT2 and SCG10 and triggers SARM1-dependent axonal degeneration. Since neuroinflammation and necroptosis are major contributors to a broad range of neurodegenerative diseases, and since axon degeneration is an early and major contributor to neuropathology in many of these disorders, we hypothesize that the molecular links identified here among neuroinflammation, necroptosis, and SARM1 activity are important mediators of axon loss in these diseases.

Here, we provide evidence for a specialized form of necroptosis that works in the axon through SARM1 to promote axon degeneration. First, we show that TNF-α stimulates necrostatin- and SARM1-dependent axon degeneration, and find that TNF-α application exclusively to axons triggers necroptosis as assayed by phosphorylation of MLKL. Hence, the machinery for necroptosis is present in the axons. Second, we demonstrate that direct activation of necroptosis via dimerization of MLKL.ND works through endogenous MLKL to trigger mechanistically distinct degenerative outcomes in the axon versus cell body. In the cell body, dimerized MLKL.ND induces calcium influx and cell death that is SARM1-independent, which is consistent with the canonical role for MLKL as an inducer of membrane permeabilization, either directly via forming a pore or through activation of TRPM7 channels (Cai et al., 2014; Xia et al., 2016). In the axon, however, dimerized MLKL.ND requires SARM1 to trigger calcium influx and axon degeneration, implying that axonal MLKL is not sufficient to disrupt membranes. This is a very surprising finding given the “pore forming” function of MLKL in canonical necroptosis. Instead, we show that dimerized MLKL.ND and endogenous MLKL promote the loss of the axon survival factors NMNAT2 and SCG10. Once axonal necroptosis lowers levels of these survival factors, the mechanism of axon degeneration is similar to that observed for Wallerian degeneration. Whether triggered by axonal necroptosis or axotomy, axon loss requires SARM1 NADase activity, is blocked by overexpression of cytoplasmic NMNAT1 or inhibition of the DLK/LZK stress kinases, and involves metabolic catastrophe that can be rescued via supplementation with NAD⁺ and ATP precursors.

How might axonal necroptosis trigger loss of the axonal survival factors NMNAT2 and SCG10? The finding that axonal necroptosis coregulates both factors is a clue, as we have previously demonstrated that palmitoylated forms of NMNAT2 and SCG10 colocalize to axonal vesicles, and are coregulated by DLK/LZK working through a JNK neuronal stress pathway (Summers et al., 2018). JNK activity promotes the turnover of both NMNAT2 and SCG10. Interestingly, necroptosis can also stimulate JNK activation (McNamara et al., 2013; Gautheron et al., 2014), and hence this could be a mechanism by which necroptosis stimulates SARM1-dependent axon degeneration.

There is controversy about whether or not necroptosis contributes to traditional Wallerian degeneration. Yang et al. (2015) reported that inhibiting necroptosis does not influence axotomy-induced Wallerian degeneration, but a recent study from Arrázola et al. (2019) claimed that axotomy activates necroptosis and promotes Wallerian degeneration. Our results are consistent with previous findings (Yang et al., 2015), as we do not see activation of necroptosis in response to axotomy, and neither genetic nor pharmacological inhibition of necroptosis blocks axotomy-induced axon degeneration (Fig. S8). However, the recent findings that necroptosis promotes axon degeneration in both excitotoxicity and Parkinson’s models (Hernández et al., 2018; Oñate et al., 2020) are consistent with the findings reported here, although the role of the SARM1 pathway was not explored in those studies.

Here, we identify a central role for SARM1 in promoting axon, RGC, and oligodendrocyte loss in glaucoma. Major efforts are underway to develop pharmacological and gene therapy approaches to inhibit SARM1 (DiAntonio, 2019), and hence SARM1 is a realistic therapeutic target for neuroprotection in glaucoma. In addition, we identify a specialized form of axonal necroptosis that links neuroinflammation to SARM1-dependent axon degeneration. This unification of three major neurodegenerative mechanisms has broad implications for the pathogenesis and treatment of a wide range of neurodegenerative disorders. This finding may be of particular interest in diseases such as multiple sclerosis, where neuroinflammation may directly impact the axon.

Further information about resources and materials is listed in Table S1.

All procedures were performed in accordance with guidelines mandated in the National Institutes of Health Guide for the Care and Use of Laboratory Animals and approved by the Washington University School of Medicine in St. Louis Institutional Animal Care and Use Committee. C57BL/6 mice were purchased from Jackson Laboratory, and SARM1KO mice were a gift from M. Colonna at Washington University in St. Louis (Szretter et al., 2009). CD1 mice (gestation day 11.5) for sensory neuron cultures was purchased from Charles River Laboratories.

2 µl TNF-α (1 ng in 1 µl of PBS with 0.1% BSA) or PBS was injected into the vitreous chamber of one eye of 5-wk-old mice (WT or SARM1 KO) using a syringe pump (Harvard Apparatus 22 syringe pump; Harvard Apparatus). A 33-gauge beveled needle (NanoFil) is used for 5-wk-old mice. Given that the vitreous volume is ∼10 µl (Remtulla and Hallett, 1985; Sharma et al., 2005; Yu and Cringle, 2006), the concentration of TNF-α is ∼0.2 mg/ml. Only one eye was injected with TNF-α or PBS for experiments. The needle tip with 45° angle was slowly inserted into the vitreous body through the sclera of the superior hemisphere of the eye. The speed of the injection is 7.5 µl/min. If eyes are damaged resulting in retinal detachment or bleeding, and/or iris/lens injury, then those eyes were excluded for data analysis because previous studies have shown that eye injury releases neuronal survival factors (Fischer et al., 2001; Leon et al., 2000).

AAV vector expressing EGFP (pAAV-hSyn-EGFP) was obtained from Addgene (gift from B. Roth, School of Medicine at the University of North Carolina at Chapel Hill, Chapel Hill, NC; Addgene 50465). This vector was cut with BamHI and NcoI, and then SARM1.WT or SARM1-K193R/H194A/H685A (SARM1.DN) was inserted between the synapsin promoter and EGFP sequence using in-Fusion HD Cloning kit (Clontech; Geisler et al., 2019b). In the retina, viral expression driven by the synapsin promoter is weaker than that of the commonly used nonneuronal promoter chicken-b-actin promoter with cytomegalovirus enhancer, so we used a capsid mutant, 7m8, to overcome tissue penetration for gene delivery (Yin et al., 2011; Dalkara et al., 2013). This virus is delivered with intravitreal injection like TNF-α injection.

Mouse DRG neurons were dissected from embryonic days 13.5 or 14.5 from CD1 mouse embryos. DRG neurons were incubated with 0.05% trypsin containing 0.02% EDTA at 37°C for 20 min and then resuspended in neurobasal media (Gibco) containing 2% B27 (Invitrogen), 50 ng/ml nerve growth factor (Harlan Laboratories), 1 µM 5-fluoro-2deoxyuridine (Sigma-Aldrich), 1 µM uridine (Sigma-Aldrich), and penicillin/streptomycin (Thermo Fisher Scientific). The dissociated DRG neurons were plated in 24-well or 96-well plates (Corning) coated with poly-D-lysine (0.1 mg/ml; Sigma-Aldrich) and laminin (3 µg/ml; Invitrogen) and then allowed to adhere in humidified tissue culture incubator (5% CO₂) for 15 min. Finally, DRG growth medium was gently added (500 µl for 24-well, 100 µl for 96-well). Lentivirus was transduced at 1 or 2 d in vitro (DIV). At DIV7, assays for axon degeneration and/or cell death were performed by applying BB (or TNF-α) or axotomy.

The following murine MLKL-specific (target sequence: 5′-GAG​ATC​CAG​TTC​AAC​GAT​A-3′) shRNA was engineered using two primers (Integrated DNA Technologies) that were annealed, phosphorylated, and ligated into AgeI and EcoRI restriction sites of pLKO.1 vector: shMLKL (5′-CCG​GTG​GAG​ATC​CAG​TTC​AAC​GAT​ATT​CAA​GAG​ATA​TCG​TTG​AAC​TGG​ATC​TCT​TTT​TTG-3′, and 5′-AAT​TCA​AAA​AAG​AGA​TCC​AGT​TCA​ACG​ATA​TCT​CTT​GAA​TAT​CGT​TGA​ACT​GGA​TCT​CCA-3′). GCaMP6f (Addgene 72303) and mRuby3 (Addgene 74252) were amplified by PCR and ligated into BamHI and EcoRI restriction sites of FCIV lentiviral vector (Araki et al., 2004) using homologous recombination-based InFusion system (Clontech): GCaMP6 (forward: 5′-GAC​TCT​AGA​GGA​TCC​GCC​ACC​ATG​GGT​TCT​CAT​CAT​CAT​CAT-3′, reverse: 5′-CTT​GAT​ATC​GAA​TTC​TCA​CTT​CGC​TGT​CAT​CAT​TTG-3′), mRuby3 (forward: 5′-GAC​TCT​AGA​GGA​TCC​GCC​GCC​ACC​ATG​GTG​TCT​AAG​GGC​GAA​GAG​CTG-3′, reverse: 5′-CTT​GAT​ATC​GAA​TTC​TTA​CTT​GTA​CAG​CTC​GTC​CAT-3′). MLKL.ND (Human ORF Library, CloneID: TOLH-1510032, Tissue Culture Support Center, Washington University in St. Louis) were amplified by PCR and ligated into BsiWI and EcoRI restriction sites of FCIV-Fkbp^F36V-TIR (Gerdts et al., 2015): MLKL.ND (forward: 5′-AAA​CTG​GAA​TCG​TAC​GCG​TCC​CGG​GGC​GGT​GGC​TCA​TCT​GGC​GGA​GGT​ATG​GAA​AAT​TTG​AAG​CAT​ATT-3′; reverse: 5′-GCT​TGA​TAT​CGA​ATT​CCT​ACT​GCC​TCA​AAG​TTT​CCT​TGA​T-3′).

HEK293T cells were used for lentivirus production (Araki et al., 2004). Cells were seeded at 70–80% confluency per 35-mm well the day before transfection. FCIV or shRNA lentivirus constructs (1.2 µg) were cotransfected with vesicular stomatitis virus - G (600 ng) and pSPAX2 (800 ng) using FuGENE 6 (Promega). The lentiviral supernatants were collected at 2 d after transfection, and then the cleared supernatant was concentrated with Lenti-X Concentrator (Clontech) to a final concentration of 1–10 × 10⁷ particles/ml. Lentivirus transduction efficiency was monitored with tagged fluorophore and is ∼100% in DRG neurons. There is no difference in expression level between WT and SARM1 KO DRG neurons.

DRG neurons were cultured in a FluoroDish (WPI) that has a glass bottom, allowing for use of an objective lens with immersion oil for GCaMP6 calcium imaging. At DIV2, lentivirus expressing GCaMP6 for calcium and mRuby3 for axon integrity were transduced to cultured DRG neurons. To maintain 37°C temperature, and 100 ml/min 5% CO₂/95% airflow rate during imaging, a Chamlide TC (Live Cell Instrument) was connected to a Leica DMI4000B microscope under confocal settings using 20× oil immersion objectives (NA 0.6) and Leica DFC7000 T 2.8 MP Color Microscope Camera at RT. The Leica Application Suite X is the software platform to acquire and analyze images. Optical sectioning and laser settings were kept constant across all image data acquisition sessions. Images for GCaMP6 and mRuby3 were acquired every 30 or 60 min until the end of the imaging session.

Retina and optic nerves were fixed with 4% PFA for 2 h at RT. Blocking solution was made with PBS including 10% normal goat serum and 0.3% TX-100. The primary and secondary antibodies were diluted in blocking solution and incubated at 4°C for overnight. Antibodies include mouse anti-Brn3a (1:100, Santa Cruz, Sc-8429), rabbit anti-Iba1 (1:1,000, Wako, 019–19741), mouse anti-CC1 (1:500, Millipore Sigma-Aldrich, OP80), and mouse (BioLegend, 801202) or rabbit (Millipore Sigma-Aldrich, T3952) anti-TUJ1 (1:1,000). The following day, Alexa Fluor–conjugated secondary antibodies were incubated for 2 h at RT (Table S1). Tissue samples were mounted in Vectashield (with DAPI) for visualization with a Leica DMI4000B microscope under confocal settings. For the 16-µm tissue sections of retina or optic nerve (Fig. S1), a cryostat (Leica CM1860) was used.

Lysate buffers contain protease (cOmplete, mini, EDTA-free protease inhibitor cocktail) and phosphatase inhibitor (PhosSTOP). The lysate buffers were preferentially selected for different proteins. While radioimmunoprecipitation assay buffer was preferentially used for detecting MLKL and phosphorylated MLKL proteins, Laemmli buffer was specifically used for detecting NMNAT2 and SCG10. Other proteins have no preference. Lysates were precleared of debris by centrifugation at 10,000 g in a refrigerated microcentrifuge for 10 min. Supernatants were mixed with 5% 2-mercaptoethanol (Millipore Sigma-Aldrich) and then boiled for 10 min. Finally, clarified extracts were mixed with Laemmli buffer (60 mM Tris-HCl, pH 6.8, 50% glycerol, 2% SDS, and 0.1% bromophenol blue) and analyzed by SDS/PAGE and Western blotting. TBS-T (TBS and 0.2% Tween 20) was used to wash immunoblots in all procedures. For blocking buffers, 5% milk was commonly added to TBS-T. However, for detecting phosphorylated MLKL proteins, 5% BSA was specifically used instead of milk. After 1 h block at RT, the membrane was incubated overnight with primary antibody. Primary antibodies were diluted in the blocking buffer as follows: rabbit anti-pMLKL (1:1,000, S345, Abcam, Ab196436), rabbit anti-MLKL (1:500, abcepta, AP14272B), rabbit anti-RIP3 (1:500, Abcam, ab56164), rabbit anti-pRIP3 (T231/S232, 1:500, Cell Signaling, 57220), mouse anti-TUJ1 (1:1,000, BioLegend, 801202), rabbit anti-caspase 3 (1:1,000, Cell Signaling, 9962), rabbit anti-cleaved caspase 3 (1:500, Cell Signaling, 9661), mouse anti-NMNAT2 (1:100, Santa Cruz, Sc-515206), and rabbit anti-SCG10 (1:2,000, Proteintech Group). The following day, HRP-conjugated anti-mouse or anti-rabbit secondary antibodies (1:5,000 or 1:10,000, Jackson ImmunoResearch; 115–035-003 or 111–035-144) were incubated for 1 h. For developing immunoblots, a chemiluminescent method (Millipore Sigma-Aldrich) was used. TUJ1 was used in all experiments as a loading control. All data quantification of immunoblots was performed with ImageJ.

3 d after TNF-α or PBS injection, optic nerves were collected and homogenized with cold radioimmunoprecipitation assay buffer. Homogenization of optic nerves was repeated three times every 20 min on ice.

To acquire axon-only lysates, dense spot cultures were prepared in 24-well plates. Due to the high density of DRG neurons, half media changes are made every 3 d to prevent nutrient loss. On DIV7, a plate was placed first on ice, and then washed with a cold PBS several times. A microscalpel was used to remove cell bodies of DRG neurons (Fig. S8 A and,Fig. 5 E). Axon-only lysate was collected from three or four wells.

In each experiment, sample number (n) indicates the number of animals in vivo, and the number of embryos in vitro. At least five EM pictures per optic nerve were randomly imaged, and then analyzed/averaged from one animal (Fig. 1, C–E; and Fig. 2 E). To count Brn3a-positive neurons (Fig. 1 G) and Iba1-positive microglia (Fig. 2 B), three pictures of the retina near optic nerve head were acquired and analyzed from each animal. For quantification of CC1-positive oligodendrocytes after TNF-α injection (Fig. 2, D and H), three to four cross-sections (16-µm thickness/section) of the proximal optic nerve regions were obtained from a distance of ∼1 mm behind the eye. Those three to four sections from one optic nerve were analyzed, and then averaged as n = 1. When counting the number of CC1-positive oligodendrocytes, the staining that overlaid with nuclear staining was considered specific.

The axon degeneration index (DI) at baseline should be <0.2 to define intact axons. If baseline measurements were >0.2, we did not further process experiments as this is an indication of an unhealthy neuronal culture. In this study, if the axon DI is >0.3 after treatment, then we define those axons as degenerated. As values increase beyond 0.3, the axons tend to have smaller axonal fragments remaining.

To define cell death, we used DAPI (or Hoechst) and EthD (or propidium iodide [PI]). In general, nuclei of dead cells only stain for EthD (or PI), and do not stain for DAPI (or Hoechst). The percentage of cell death is calculated by EthD staining / (EthD staining + DAPI staining) × 100.

To confirm that AAV-SARM1.DN is not expressed in oligodendrocyte (Fig. S1, C and D), single-section images of four different channels of a confocal microscope were acquired for the analysis of pixel intensity. As seen in Fig. S1 C, one line is drawn in a merged image, and then intensity is measured of each pixel for four different channels. The intensity of pixels is plotted in Fig. S1 D. In this example, because the peak of CC1 pixel (red) is far from the peak of SARM1.DN (green), we are able to define that SARM1.DN is not expressed in oligodendrocytes.

All statistics were performed with GraphPad Prism 8 software. The Shapiro–Wilk test was used for normality whether or not data assumed a Gaussian distribution. To determine statistical significance between groups, unpaired two-way Student’s t test, one-way ANOVA with post hoc Tukey test, or two-way ANOVA with post hoc Bonferroni’s multiple comparison test were used (detailed in figure legends). Statistical significance was noted as *, P < 0.05; **, P < 0.01; and ***, P < 0.001. All error bars represent mean ± SEM.

Fig. S1 shows that intravitreal injection of AAV-hSyn-SARM1.DN-EGFP drives expression in RGCs but not oligodendrocytes. Fig. S2 shows that cleaved caspase 3 is not detected in the optic nerve of TNF-α–injected eyes. Fig. S3 shows that necrostatin-1s does not prevent axon degeneration following dimerization of MLKL.ND. Fig. S4 shows validation of shRNA for MLKL in cultured DRG neurons and N2a cells. Fig. S5 shows that dimerized MLKL.ND requires endogenous MLKL to induce axon degeneration. Fig. S6 shows that SARM1 KO does not prevent cell death induced by dimerization of MLKL.ND. Fig. S7 shows that DLK inhibition and cytoNMNAT1 expression block axon degeneration induced by dimerization of MLKL.ND. Fig. S8 shows that necroptosis is not activated by or required for Wallerian degeneration. Table S1 shows reagents used in the study. Video 1 shows that calcium influx by dimerization of MLKL.ND is completely prevented in axons from SARM1 KO neurons. Video 2 shows local calcium influx in axons, not in soma by dimerization of MLKL.ND.

We thank members of the DiAntonio and Milbrandt laboratories for fruitful discussions. In particular, we thank Amy Strickland and Wandy Beatty for tissue processing of optic nerves and EM. AAV capsid 7m8 was a kind gift from J. Flannery and D. Schaffer (University of California, Berkeley, Berkeley, CA).

This work was supported by National Institutes of Health grants R01CA219866, and RO1NS087632 (to J. Milbrandt and A. DiAntonio), and RF1-AG013730 (to J. Milbrandt) and the Hope Center Viral Vector Core at the Washington University School of Medicine in St. Louis.

A. DiAntonio, J. Milbrandt, and Washington University are inventors on patents related to this work. A. DiAntonio and J. Milbrandt are cofounders of Disarm Therapeutics, and A. DiAntonio and J. Milbrandt are members of its scientific advisory board. The remaining author declares no competing financial interests.

Author contributions: K.W. Ko and A. DiAntonio designed the research. K.W. Ko performed all research and data analysis. K.W. Ko, J. Milbrandt, and A. DiAntonio wrote the manuscript. J. Milbrandt and A. DiAntonio supervised experiments.