Potent activation of SARM1 by NMN analogue VMN underlies vacor neurotoxicity

SARM1 is intensively studied for its role in promoting axon degeneration in injury and disease. We identify VMN, a metabolite of the neurotoxin vacor, as a potent SARM1 activator, an action likely to underlie vacor neurotoxicity in humans. This study provides novel tools to study SARM1 regulation, supports drug discovery, further links programmed axon death to human disease and identifies a new model where axons are permanently rescued.

We first confirmed that VMN and VAD are generated in vacor-treated dorsal root ganglion (DRG) mouse neurons (Fig. S1b) and that these and a second neuron type, superior cervical ganglia (SCG) mouse neurons, exhibit rapid, dose-dependent neurite degeneration ( Fig. 1a-d; Fig. S1c-f). Consistent with the proposed toxic role for vacor metabolites, both nicotinamide (NAM), which competes with vacor as a preferred substrate for NAMPT, and the NAMPT inhibitor FK866 (Fig. S1a), delayed vacorinduced neurite degeneration ( Fig. S1g-j). Such competition may underlie the use of NAM as a treatment to patients who ingested vacor 16 .
As hypothesised, vacor failed to induce degeneration of Sarm1 -/-DRG and SCG neurites ( Fig. 1a-d). This protection was extremely strong with neurites surviving indefinitely even after multiple vacor doses (Fig. S2a,b). Sarm1 -/neurites were not only protected from degeneration, but they also continued to grow normally, even with repeated dosing (Fig. 1e,f). This mirrors the permanent rescue and continued growth previously reported in Nmnat2 null axons 7,15 , suggesting complete efficacy in both toxic and inherited types of neuropathy. It also shows vacor neurotoxicity is predominantly SARM1-dependent. SARM1 levels remained relatively stable following vacor treatment (Fig. S2c-f). We instead confirmed that SARM1 NADase activity is the critical function required for vacor toxicity. Exogenous expression of wild-type human SARM1 (hSARM1) in Sarm1 -/-SCG neurons restored vacor sensitivity, whereas expression of E642A hSARM1, that lacks NADase activity, did not ( Fig. 1g-i; Fig. S2g,h), similar to findings in axotomy 5 . Interestingly, unlike axotomy, activation of the pathway by vacor also caused SARM1-dependent cell death (Fig. 1i), similar to that previously reported with constitutively active SARM1 20,21 . To explore this further, we cultured DRG neurons in microfluidic chambers to allow independent manipulation of the axon and soma compartments (Fig. S3a). Local addition of vacor to either compartment directly activated local death of cell bodies and/or neurites. However, while vacor-induced cell death causes eventual secondary neurite degeneration, the reverse was not true: no cell loss was observed after local induction of distal neurite degeneration by vacor.
Activation of SARM1 depletes NAD and generates cADPR in neurons 5,8,9,21 . Based on the previously reported inhibition of NMNAT2 by VAD 19 , we initially hypothesised that a rise in NMN, the physiological NMNAT2 substrate and a known activator of SARM1 10,12,14 , stimulates SARM1 activity to trigger vacor-dependent axon death.
Surprisingly, while SARM1 activation was confirmed by a dramatic, SARM1dependent decline in NAD (Fig. 2a,b) and cADPR accumulation (Fig. 2c,d) following vacor administration, NMN did not rise and even fell slightly (Fig. 2a,b). Furthermore, NMN and NAD levels and ratio were not altered in our vacor-treated Sarm1 -/neurons ( Fig. 2a,b), suggesting that the reported inhibition of NAMPT and NMNAT2 by vacor/VAD 19 does not occur in this context at this drug concentration. Crucially, NMN deamidase, an enzyme that strongly preserves axotomised axons by preventing NMN accumulation ( Fig. 3a; Fig. S4a) [10][11][12] , was also unable to protect against vacor toxicity ( Fig. 3a,b), even though recombinant NMN deamidase retains its ability to convert NMN to NaMN in the presence of vacor, VMN and VAD (Fig. S4b). Also WLD S , another enzyme that limits NMN accumulation 10 , failed to rescue neurons against vacor neurotoxicity (Fig. S4c,d). These data suggest that, in this specific context at least, NMN is not responsible for SARM1 activation.
Given its structural similarity to the endogenous SARM1 activator NMN (Fig. S5a), we hypothesised that VMN might instead directly activate SARM1 leading to cell and axon death. Crucially, we found that VMN potently activates the NADase activity of recombinant hSARM1, even more so than NMN, having a lower Ka and resulting in a greater induction (Fig. 3c). Conversely, VAD only had a weak inhibitory effect on hSARM1 activity (Fig. S5c). Intriguingly, recombinant hSARM1 NADase activity dropped at higher VMN concentrations (Fig. 3c). Notwithstanding the doses of vacor used in this study result in VMN levels in neurons within the activation range of SARM1, this inhibitory action could reveal critical information on how SARM1 activity is regulated. Analysis via best fitting (see equation in Methods) suggests that our data are compatible with two distinct binding sites of VMN on SARM1 and led us to establish binding constants for activation (Ka) and inhibition (Ki) (Fig. 3c).
While NMN activates SARM1 14 and promotes programmed axon death when NMNAT2 is depleted 10-13 , exogenous NMN does not induce degeneration of uninjured neurites 10 , probably because it is rapidly converted to NAD when NMNAT2 is present, thereby limiting its accumulation. However, we now show that exogenous application of its analogue, VMN, does induce SARM1-dependent death of uninjured DRG neurites (Fig. 3d,e). This difference likely reflects a combination of VMN being both a more potent activator of SARM1 and it accumulating more because it is not efficiently metabolised by NMNAT2 19 .
VMN-dependent activation of SARM1 being the effector of vacor toxicity is also supported by a number of other findings. First, unlike NMN, VMN is not a substrate of NMN deamidase (Fig. S5d) thus explaining its inability to protect against vacor neurotoxicity (Fig. 3a,b). In addition, VMN is not a substrate of the nuclear isoform NMNAT1 and is a poor substrate of mitochondrial NMNAT3 19 , so VMN accumulation in cell bodies (Fig. S5e) and subsequent SARM1 activation, as detected by a rise of cADPR in this compartment (Fig. 2d), provides a clear rationale for why vacor administration also causes cell death.
Overall, this study provides clear evidence that direct SARM1 activation by vacor metabolite VMN underlies vacor neurotoxicity. Crucially, the structural and functional similarity between VMN and NMN provides additional support for a key physiological role of NMN in the regulation of SARM1 activity and axon degeneration. Given its greater potency than NMN, we anticipate that VMN and vacor will be important tools for drug discovery projects. Vacor is currently the most effective chemical to directly activate SARM1 and cause neuronal death, eliminating the need for complex axotomy experiments or the use of drugs with non-specific actions. Furthermore, understanding where VMN binds to SARM1, why it activates it more potently than NMN and expanding on VMN activation/inhibition functions should support drug development projects in several ways. One intriguing possibility is that VMN interacts with SARM1 in the auto-inhibitory ARM domain leading to SARM1 activation, and in the catalytic TIR domain, causing inhibition 5,20 . Finally, obtaining full-length SARM1 structures in physiological and active conformations will be essential for rational drug design but, despite recent important developments [22][23][24][25] , this has proven challenging. However, the possible interaction of VMN with SARM1 could help achieve this.
These data further implicate programmed axon death to human disease. It appears that this degeneration pathway can be aberrantly activated in humans not only by genetic mutation of NMNAT2 3,4 , but also by SARM1 activation in severe neurotoxicity within hours of vacor ingestion 17 . Although vacor use is banned, this study raises important questions of whether other pesticides or environmental chemicals in use today also activate programmed axon death. Finally, as vacor causes pancreatic bcell destruction and diabetes in humans 16 , these findings could have broader implications, uncovering a role for SARM1 in the survival of insulin producing cells and aiding research on diabetes.
8      hSARM1 for microinjection experiments in (Fig. 1g-i). The absence of vacor-induced toxicity in neurons expressing E642A hSARM1 is therefore due to loss of enzymatic activity rather than the lack of protein expression.     conversion yields obtained via this protocol were 70% for VMN and 50% or less for VAD that required two chromatographic steps for purification.

METHODS
All studies conformed to the institution's ethical requirements in accordance with the 1986 Animals (Scientific Procedures) Act.

Microinjection and quantification of % of healthy neurites and viable neurons
DIV5 dissociated Sarm1 -/-SCG neurons were microinjected using a Zeiss Axiovert S100 microscope with an Eppendorf FemtoJet microinjector and Eppendorf TransferMan® micromanipulator. Plasmids were diluted in 0.5X PBS (without CaCl2and MgCl2) and filtered using a Spin-X filter (Costar). The mix was injected directly into the nuclei of SCG neurons using Eppendorf Femtotips. Injected plasmids were allowed to express for 2 days before vacor or vehicle treatment and axotomy.

Determination of NMN, NAD, cADPR, vacor, VMN and VAD tissue levels
Following treatment with vacor or vehicle, DIV7 wild-type and Sarm1 −/− DRG ganglia were separated from their neurites with a scalpel. Neurite and ganglia (containing short proximal neurite stumps as well as cell bodies) fractions were washed in ice-cold PBS and rapidly frozen in dry ice and stored at −80 °C until processed. Tissues were ground in liquid N2 and extracted in HClO4 by sonication followed by neutralisation with K2CO3.
NMN and NAD were subsequently analysed by spectrofluorometric HPLC analysis after derivatization with acetophenone 28 . Vacor, VMN and VAD were determined in DRG whole explant cultures by ion pair C18-HPLC chromatography, as previously described 19 . cADPR levels were determined in DRG whole explant cultures using a cycling assay, as previously described 29 . A single analysis of vacor, VMN, VAD and cADPR levels was performed in DRG neurite and ganglia fractions independently, which was not further repeated due to the low basal levels of these metabolites and the amount of cellular material needed for this type of analysis. Metabolites levels were normalised to protein levels quantified with the Bio-Rad Protein Assay (Bio-Rad) on formiate-resuspended pellets from the aforementioned HClO4 extraction.

VMN and VAD synthesis and purification
VMN and VAD were synthesized as previously reported 19 with minor changes. Vacor was either phosphoribosylated in vitro by murine NAMPT (mNAMPT) into VMN or phosphoribosylated by mNAMPT and adenylated by murine NMNAT2 (mNMNAT2) into VAD. The scheme of these reactions and the reaction mixtures are shown in (Fig.   S6a,b). Inorganic yeast pyrophosphatase (PPase), phosphoribosyl 1-pyrophosphate (PRPP) and adenosine triphosphate (ATP) were all from (Merck). Recombinant mNAMPT and mNMNAT2 were purified as previously described 30,31 . Following incubation, reaction mixtures (A) for VMN and (A+B) for VAD (Fig. S6b) were stopped by rapid cooling on ice and kept refrigerated until injection for purification. Next, the VMN and VAD obtained were purified by FPLC under volatile solvents and lyophilised.
Briefly, a preparative IEC chromatography was carried out on AKTA Purifier onto the anion exchanger resin Source 15Q (GE HealthCare, 20 ml volume). The column was equilibrated at room temperature (~25 °C), at 5 ml/min. After injection of the two mixtures above, a linear gradient elution was applied by mixing the two volatile buffers as indicated. The eluate was monitored at wavelengths of 260 nm and 340 nm (optimal for vacor nucleotides), and both VMN and VAD peaks were collected (Fig. S6c,d).
Next, an additional chromatography was performed by HPLC onto a RP C18 column (Varian, 250 x 4.6 mm, 5μ particles) heated at 50 °C to obtain a pure VAD stock. The equilibration in such case was carried out at 1 ml/min in ammonium formiate buffer; this was followed by multiple injections of the previously collected IEC pool of VAD (~1 ml each), and elution by a linear gradient of increasing acetonitrile in the buffer. The eluate was monitored again at wavelengths 260 nm and 340 nm and the VAD peak was collected (Fig. S6d). After lyophilisation, dry pellets were stored at -80 °C. The resulting VMN and VAD lyophilised powders were 100% pure. Their quantitation after resuspension in H2O was performed spectrophotometrically (ɛ340nm of 17.8 mM -1 cm -1 ).

Recombinant hSARM1 purification
Recombinant, full-length C-terminal Flag-tagged human SARM1 (hSARM1-Flag) was Beads were collected on a magnetic rack and washed 3x with 500 µl KHM buffer and 1x with PBS (with protease inhibitors) and then resuspended in PBS containing 1 mg/ml BSA. hSARM1-Flag concentration in the bead suspension was determined relative to an hSARM1 standard (purified from Drosophila Schneider 2 cells) by immunoblotting using a rabbit polyclonal antibody raised against human SAM-TIR. and VAD detection 19 . The products formed (NAM, ADPR and cADPR) were quantified from the area of separated peaks. Rates were calculated under NAD consumption ≤ 20% from the linearly accumulating products ADPR and cADPR (Fig. S5b). One unit (U) of activity represents the enzyme amount that forms 1 μmol/min of the products above under these assay conditions. Data from three independent experiments were averaged to calculate hSARM1 activity fold activation (Fig. 3c) and then studied by fitting through the Excel software package to the rate equation below that represents a Michaelis-Menten equation adapted to be valid for readily reversible effectors, i.e. including activators and inhibitors binding via distinct sites to one protein target. Maximum velocity (Vmax) represents the rate in the absence of the effector (basal activity) corrected by the two parameters between parentheses, the first for activation and the second for inhibition, taking into account opposite and independent contributions exerted by the single but dual effector I. X is the "relative activity fold" factor at varying concentrations of I, Y is the residual activity fraction at saturating

NMN deamidase activity
Recombinant E. coli NMN deamidase was obtained as previously described 32 . Activity was measured in buffer HEPES/NaOH 50 mM, pH 7.5, in the presence of 4 milliU/ml enzyme, 0.5 mg/ml BSA, and 250 μM NMN or VMN. Vacor, VMN and VAD, all at the concentration of 250 μM, were also assayed in presence of the substrate NMN.
Reactions were incubated at 37 °C, then stopped and analysed by HPLC using the two methods described above 19,28 . The NMN deamidase rates were calculated after separation and quantification of the NaMN product formed from NMN, and finally reported as relative percentages of controls in the presence of NMN alone.

Western blot
Following treatment with vacor, DRG ganglia were separated from their neurites with a scalpel. Neurites and ganglia were collected, washed in ice-cold PBS containing protease inhibitors (Roche), and lysed directly in 15 µl 2x Laemmli buffer containing intensity was determined by densitometry using ImageJ.

Statistical analysis
Statistical testing of data was performed using Prism (GraphPad Software, La Jolla, USA). The appropriate tests used and the n numbers of each individual experiment are described in the figure legends. A p-value < 0.05 was considered significant.