Article

Structural and Mechanistic Regulation of the Pro- degenerative NAD Hydrolase SARM1

Graphical Abstract

Highlights
d Autoinhibited (3.3 A˚ ) and active (6.8 A˚ ) structures of pro- degenerative NADase SARM1 solved

d Identification of a critical autoinhibitory lock

d Lock mutations convert inactive SARM1 to an active, neurotoxic state

d Enzymatic studies explain SARM1’s functional dependence on local metabolic environment

Authors
Matthew Bratkowski, Tian Xie, Desiree A. Thayer, …, Sean P. Brown, Xiaochen Bai, Shilpa Sambashivan

Correspondence
[email protected]

In Brief
Bratkowski et al. describe cryo-EM structures of autoinhibited and active SARM1, an axon-enriched, injury- activated NADase. Structure-function studies elucidate the mode of autoinhibition and SARM1 activity regulation. The authors provide mechanistic insight into SARM1’s regulation by the local metabolic environment, laying the foundation for future SARM1-based therapies to treat neurological diseases.

Bratkowski et al., 2020, Cell Reports 32, 107999
August 4, 2020 ª 2020 The Authors. https://doi.org/10.1016/j.celrep.2020.107999

OPEN ACCESS

Article
Structural and Mechanistic Regulation
of the Pro-degenerative NAD Hydrolase SARM1
Matthew Bratkowski,1,4 Tian Xie,2,4 Desiree A. Thayer,1 Shradha Lad,1 Prakhyat Mathur,1 Yu-San Yang,1 Gregory Danko,1 Thomas C. Burdett,1 Jean Danao,1 Aaron Cantor,1 Jennifer A. Kozak,3 Sean P. Brown,3 Xiaochen Bai,2
and Shilpa Sambashivan1,5,*
1Biology Department, Nura Bio Inc., South San Francisco, CA 94080, USA
2Department of Biophysics, University of Texas Southwestern Medical Center, Dallas, TX 75390, USA
3Chemistry Department, Nura Bio Inc., South San Francisco, CA 94080, USA
4These authors contributed equally
5Lead Contact
*Correspondence: [email protected] https://doi.org/10.1016/j.celrep.2020.107999

SUMMARY

The NADase SARM1 is a central switch in injury-activated axon degeneration, an early hallmark of many neurological diseases. Here, we present cryo-electron microscopy (cryo-EM) structures of autoinhibited (3.3 A˚ ) and active SARM1 (6.8 A˚ ) and provide mechanistic insight into the tight regulation of SARM1’s function
by the local metabolic environment. Although both states retain an octameric core, the defining feature of the autoinhibited state is a lock between the autoinhibitory Armadillo/HEAT motif (ARM) and catalytic Toll/inter- leukin-1 receptor (TIR) domains, which traps SARM1 in an inactive state. Mutations that break this lock acti- vate SARM1, resulting in catastrophic neuronal death. Notably, the mutants cannot be further activated by the endogenous activator nicotinamide mononucleotide (NMN), and active SARM1 is product inhibited by Nicotinamide (NAM), highlighting SARM1’s functional dependence on key metabolites in the NAD salvage pathway. Our studies provide a molecular understanding of SARM1’s transition from an autoinhibited to an injury-activated state and lay the foundation for future SARM1-based therapies to treat axonopathies.

INTRODUCTION

Wallerian degeneration is a conserved and specific axon destruction pathway that is triggered after injury. Sarm1 (sterile alpha and Toll/interleukin-1 receptor [TIR] motif containing 1) was initially identified as a critical pro-degenerative factor in the pathway and was subsequently shown to have nicotinamide adenine dinucleotide (NAD) hydrolase function (Gerdts et al., 2015; Osterloh et al., 2012). Sarm1 deletion robustly protects axons in models of acute and sub-acute/chronic injury including axotomy, chemotherapy-induced peripheral neuropathy, nerve crush/transection, and traumatic brain injury, supportive of a therapeutically beneficial outcome to inhibiting the target (Geis- ler et al., 2016; Gerdts et al., 2013; Henninger et al., 2016; Tur- kiew et al., 2017).
SARM1 is a multi-domain protein containing 724 amino acids (Figure 1A). It contains a predicted N-terminal autoinhibitory Armadillo/HEAT motif (ARM) domain, followed by tandem SAM1 and SAM2 oligomerization domains and a C-terminal cat- alytic TIR domain (Chuang and Bargmann, 2005; Essuman et al., 2017; Gerdts et al., 2013). In the tertiary structure, the ARM and TIR domains are proposed to be proximal to each other, and truncation of the ARM domain results in active SARM1 (Sum- mers et al., 2016). The NAD hydrolase activity of SARM1 resides in the TIR domain, reminiscent of other enzymatically active, ho-

modimeric bacterial and plant TIR domains (Carty et al., 2006; Essuman et al., 2017, 2018; Horsefield et al., 2019; Wan et al., 2019). While forced homodimerization of the SARM1 TIR domain is sufficient to drive NAD depletion and axon fragmentation, deletion or mutation of the SAM domains prevents these pro- cesses (Gerdts et al., 2013, 2015; Horsefield et al., 2019; Sporny et al., 2019). This has led to the hypothesis that oligomerization of the SAM domains may be a rate-limiting step in SARM1 activa- tion that then drives TIR multimerization.
SARM1-driven axon fragmentation appears to function down- stream of nicotinamide mononucleotide adenylyltransferase 2 (NMNAT2), another major player in the Wallerian degeneration pathway (Gilley et al., 2015; Loreto et al., 2015; Walker et al., 2017; Zhao et al., 2019). NMNAT2 is also an axon-enriched enzyme that synthesizes NAD from precursors of nicotinamide mononucleotide (NMN) and adenosine triphosphate (ATP) (Raf- faelli et al., 2002; Yalowitz et al., 2004). Injury results in rapid deple- tion ofshort-lived NMNAT2 in axons, leading to an increase in NMN and decrease in NAD; both of which contribute to axon degenera- tion (Di Stefano et al., 2015; Gilley and Coleman, 2010). Notably, although Nmnat2—/— mice are perinatal lethal, Sarm1—/— can rescue this phenotype: double knockouts are viable, develop nor- mally into adulthood, and do not demonstrate neuromuscular de- fects associated with axon loss (Gilley et al., 2017). These results suggest that SARM1 likely functions downstream of NMNAT2.

Cell Reports 32, 107999, August 4, 2020 ª 2020 The Authors. 1
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

A Figure 1. SARM150–724 Assembles into an Oc-
tamer
(A)SARM1 domain schematic. Domains are color coded, and amino acid numbers are indicated. MTS, mitochondrial translocation sequence.
(B)Cryo-EM map of SARM1. Prime (0) labels are
used to differentiate different copies of the same domain, and domains are color coded with alter- nating dark (non-prime domains) and light (prime domains) shades: ARM (pink), SAM1 (yellow), SAM2 (brown), and TIR (gray). One monomer is shown as a cartoon representation enclosed by map density (cyan mesh) to better represent the path of the protein chain.

Recent structure-function studies of isolated domains of SARM1 have provided some insight into its assembly and related activity. The SAM domains in isolation oligomerize into an oc- tamer, suggesting that SARM1 itself may exist as a large complex (Horsefield et al., 2019; Sporny et al., 2019). The crystal structure of the TIR domain in isolation revealed the putative active site in the absence of substrate (Horsefield et al., 2019). However, these structures provide little insight into how this multi-domain protein assembles in its different functional states. Finally, a recent report suggested that SARM1 is activated by NMN and cell-permeable NMN analogs (Zhao et al., 2019). However, it is not clear whether this mechanism is specific for NMN or if other NAD pathway me- tabolites are involved in SARM1 regulation. Here, we reveal the structures of near-full-length, autoinhibited SARM1 and ARM- deleted, active SARM1. We elucidate the underlying molecular mechanism of autoinhibition and provide mechanistic insight into subsequent NAD hydrolase activation by NMN and product inhibition by nicotinamide (NAM) laying the foundation for SARM1-based therapies to treat neurological diseases.

RESULTS

The Cryo-EM Structure of Near-Full-Length SARM1 Reveals an Autoinhibited Octamer We purified SARM1 containing all functional domains and lacking only the mitochondrial targeting sequence herein referred to as SARM150–724 (Figure 1A). The protein elutes at a position consis- tent with a pre-assembled, higher-order oligomer by size-exclu- sion chromatography (Figure S1A), in agreement with another recent study (Sporny et al., 2019). We next solved the structure
of SARM150–724 by cryoelectron microscopy (cryo-EM) to 3.3 A˚
resolution (map envelope representation [Figure 1B]; data pro- cessing [Figures S1B–S1G]; and structure statistics [Table S1]).
The structure confirms that SARM150–724 is a homo-octamer that covers a surface area of 439,100 A˚ 2. The oligomerization of
SARM1 is driven by the formation of an octameric, double-layered
ring, composed of SAM1 and SAM2 domains, which creates an aperture with a diameter of ~50 A˚ (Figure 1B). The ARM domains assemble on the perimeters of the SAM rings and leaf outward (Figure 1B). The TIR domains interact only with the ARM domains
on the perimeter (Figure 1B). Continuous, high-quality map den- sity for most of the ARM, SAM1, and SAM2 domains is present, with the exception of a few regions in the ARM domain and the linker between the SAM2 and TIR domains (Figures S2A–S2C, respectively). The TIR domain, however, is not as well resolved,

and modeling is limited, as discussed later (Figure S2D). Overall, the structure confirms that SARM1 forms an oligomer in its native, inactive form.
The SAM1 and SAM2 domains oligomerize into an octamer and serve as the core of the SARM1 structure. They are each composed of five alpha helices (Figure 2A) and assume an overall similar secondary structure (Figure S3A). The SAM1 domains interact in a head-to-tail fashion but form a closed ring instead of the more commonly observed SAM polymeric architecture (Kim et al., 2001; Knight et al., 2011). The SAM rings are stabilized by both homo- and hetero-dimeric domain interactions. The SAM1- SAM10 interface forms between helices H2, H3, and H4 with H50 (Figure 2A, black box; and detailed in Figure 2B). The interface in- volves key hydrogen bond pairs between D439-R4680, E450- K4640, Q437-R4650, D454-G4600 mainchain, and Q436-S4590 (Fig-
ure 2B). A key hydrophobic interaction also occurs between L442 and I4610 near the center of the interface (Figure 2B). The SAM1- SAM2 interface occurs between helices H3 and H5 of SAM1, with H2 of SAM2 and the linker between the domains (Figure 2A, green; and detailed in Figure 2C). It features hydrogen-bond pairs between D441-Y503, K474-Y479, A477-R499, and Q500-T471.
The SAM2-SAM20 domains interface in a head-to-tail manner with helix H50-binding helices H3 and H4 (Figure 2A, blue box; and detailed in Figure 2D). Key interactions include the sidechain of D526 hydrogen bonding to the mainchain nitrogens of L5310 and G5320, a mainchain to sidechain hydrogen bond between C508-H5340, and a hydrophobic interaction between sidechains of L510 and L5310. Notably, mutation of SAM-domain interfaces decreases NAD hydrolase activity, emphasizing that the oligomer- ization of SAM domains is critical for functional SARM1 (Horsefield et al., 2019; Sporny et al., 2019).
The SAM1 and SAM2 domains in our structure oligomerize in essentially the same orientation when attached to the ARM and TIR domains as in isolation from previous crystal structures (Fig- ure S3B) (Horsefield et al., 2019; Sporny et al., 2019). We also solved a cryo-EM structure of SARM1 lacking the ARM domain (SARM1409–724) but containing the tandem SAM and TIR
domains to ~6.8 A˚ resolution (map envelope representation [Fig-
ure S3C]; data processing [Figure S3D–S3G]; and structure statis- tics [Table S1]). The SAM domains were easily docked into the EM map in a similar conformation as in SARM50–724 (Figure S3C). Although fragmented additional density was visible above the SAM2 domain, the TIR domains could not be resolved in the final map, likely because of limited local resolution (Figure S3C). Over- all, the SAM domains form a stable core that allows SARM1 to

Figure 2. The SAM Domains Form a Stable Core
(A)SAM domain interfaces displayed as a cartoon representation. Helices are labeled (H1–H5 or H10– H50) according to the corresponding domain indi- cator. Domain-domain interfaces are indicated by dashed boxes and color coded as black, green, and blue for the SAM1-SAM10, SAM1-SAM2, and SAM2-SAM20 interfaces, respectively.
(B)Detailed view of the SAM1-SAM10 interface. Hydrogen bonds are represented by black, dashed lines.
(C)Zoomed-in view of the SAM1-SAM2 in- teractions.
(D)Detailed view of the SAM2-SAM20 interface.

form an octamer, regardless of whether it is in an autoinhibited or active conformation, contrary to a reported hypothesis that activa- tion drives oligomerization (Horsefield et al., 2019). On the other hand, the TIR domains appear dynamic, and results presented later further support that assumption.
The ARM domains wrap around the SAM1-SAM2 ring and leaf outward (Figure 1B). The ARM domain (residues 50–408) is composed of armadillo repeats composed of 21 alpha helices (Figure 3A). Each ARM domain binds to a SAM1 domain within the same chain: helix H20 of the ARM domain contacts helix H1 of SAM1 (Figure 3A, black box; and detailed in Figure 3B). The in- teractions in the interface include ARM-SAM1 hydrogen bonds between L406-W412, P404-Q423, and Y380-W420 (Figure 3B). Additionally, F476 of SAM1 hydrophobically interacts with P407 of the ARM domain (Figure 3B). The ARM domain helices H12 and H15 bind to helix H10 of a SAM20 domain (Figure 3A, blue box; and detailed in Figure 3C). Key ARM-SAM20 hydrogen bonds include K281-S4800, R243-N4860, and Q239-N4780 (Figure 3C).
Overall, these interactions cause the ARM domains to cover the outer perimeter of the SAM domains, which are exposed in SAM1-SAM2 domain crystal structures (Figure S3B) and are involved in crystal packing interactions (Horsefield et al., 2019; Sporny et al., 2019). Possibly, the ARM domain prevents further oligomerization of the SAM1 and SAM2 domains; SAM domains in other proteins form polymers and also free SAM1-SAM2 protein is reportedly prone to aggregation (Kim et al., 2001; Knight et al., 2011; Sporny et al., 2019). Importantly, the ARM domain also plays a key role in regulating the activity of the TIR domain as dis- cussed below.

The ARM Domain Traps the TIR Domain in an Inactive Conformation
The TIR domain is connected to SAM2 by an unmodeled linker (residues 546–560) and binds solely to the ARM domain on the

outer perimeter of SARM1, positioning it about 25 A˚ away from the neighboring TIR domain (Figure 4A). An interchain
ARM-TIR interaction appears to occur (Figure 4A, black box; and expanded and detailed in Figure S4A) and is discussed as such for simplicity. However, because we cannot model the SAM2-TIR linker,
we note that it is equally likely that an intrachain ARM-TIR inter- action occurs (Figure S4B). Most secondary structural elements of the TIR domain from SARM150–724 are in a similar conforma- tion as those of the crystal structure of the isolated TIR domain (Horsefield et al., 2019), although the BB, DD, and SS loops of the former could not be modeled because of the lack of map density (Figure 4B). Because the loops are critical for activity and axon fragmentation (Horsefield et al., 2019; Summers et al., 2016), it is possible that, in the context of the full-length protein, they only become ordered when TIR-TIR domains interact.
In contrast to the loop regions, the ARM-interacting residues of the TIR domain are well resolved when TIR-TIR domains interact (Figure S2D). The main ARM-TIR0 binding site is formed by inter- actions between TIR0 helix aa0 and ARM H13 (Figure 4A, black box; and detailed in Figure 4C). The interaction appears predom-
inately hydrophobic in nature with carbon-carbon distances be- tween interacting side chains ranging from 3.2 to 4.0 A˚ in length.
TIR0 residue V5820 interacts with ARM residues P256 and W253, and W253 also interacts with TIR0 residue L5860 (Figure 4C). Addi- tionally, TIR0 residue K5810 interacts with ARM residue F255. This interaction is likely hydrophobic instead of a cation-aromatic interaction because the z nitrogen of K581 points away from the aromatic ring of F255 (Figure 4C). Overall, these interactions secure the TIR domain in an inactive, monomeric state.

Disrupting the ARM-TIR Lock Enzymatically Activates SARM1 and Drives Neuronal Cell Death Based on the structure of SARM150–724, we designed and purified recombinant forms of several truncations to functionally validate the importance of the ARM-TIR lock (Figures 5A and S5A). Using
a fluorescent εNAD assay, we confirmed that recombinant
SARM150–724 was enzymatically inactive, whereas maltose-bind- ing protein (MBP)-SARM1409–724, which has the ARM domain

Figure 3. The ARM Domain Interacts with the SAM Domains
(A)Overall ARM-SAM1/SAM2 interface. ARM domain helices (H1–H21) are labeled. The ARM- SAM1 interface is indicated by a black, dashed box, and the ARM-SAM20 interface is represented by a blue, dashed box.
(B)Zoomed-in view of the ARM-SAM1 interface.
(C)Detailed view of ARM-SAM20 interactions.

deleted, is active (Figure S5B). SARM1409–724 with the MBP tag cleaved also assembles into an octamer (Figures S3C and S5C) and retains catalytic activity (Figure S5D). SARM1559–700 is a monomer by gel filtration (Figure S5E) and was inactive in the
εNAD assay at protein concentrations that were in 8- or 40-fold
molar excess above those tested for SARM150–724 and MBP- SARM1409–724 (Figure S5B). In a more sensitive high-performance liquid chromatography (HPLC)-based assay, SARM1559–700 only displayed significant activity at a protein concentration that was 3,000 times greater than that used for MBP-SARM1409–724 (Fig- ure S5F). This result is in line with results from on-bead style as- says reported previously (Essuman et al., 2017; Horsefield et al., 2019) and suggests inefficient activity driven by molecular crowd- ing of TIR monomers, as opposed to the robust and specific activ- ity displayed by octameric SARM1409–724.
To determine whether breaking the ARM-TIR lock results in activation of full-length SARM1, we purified and characterized two key variants—one with two critical residues in the ARM domain mutated (W253A and F255A, called SARM150–724[ARM-mut]) and the other with three critical residues in the TIR domain mutated (V582A, L586A, and K581A, called SARM150–724[TIR-mut]) (Figures 5A and S5G). Notably, SARM150–724(ARM-mut) (expressed as an MBP fusion in Escherichia coli) demonstrated significantly greater NAD hydrolase activity compared with MBP-SARM150–724 (Figure 5B). On the other hand, SARM150–724(TIR-mut) remained inactive (Figure 5B). Our results indicate that mutation of residues W253

transfected with SARM150–724(TIR-mut) (Fig- ure 5C, blue bar), in agreement with our biochemical data. SARM150–724(ARM-mut) also resulted in significantly greater ATP depletion, suggesting reduced cell viability, compared with wild-type SARM150–724 and SARM150–724(TIR-mut)
(Figure 5D) despite overall similar pro-
tein-expression levels (Figure S5H). Thus, breaking the ARM-TIR lock by mu- tation is sufficient to drive general cell death and correlates with the observed increase in NAD hydrolase activity.
Finally, we interrogated the effect of disrupting ARM-TIR in- teractions in mouse primary dorsal root ganglion (DRG) neu- rons. We generated enhanced green fluorescent protein (EGFP) fused mRNA constructs for full-length autoinhibited Sarm11–724 and the active forms of SARM1: Sarm1409–724 and Sarm11–724(ARM-mut). We co-transfected primary DRG cultures with each SARM1 construct plus an mCherry expression plasmid and monitored cell and neurite degeneration over time by live-cell imaging (Figures 5E and 5F, representative time points; Figure S6, full time course). mCherry was used as the primary readout of cell viability instead of EGFP because some SARM1 constructs were acutely toxic even at low expression levels, wherein EGFP was barely visible (Fig- ure S6). mCherry is readily expressed in healthy cells, and the signal rapidly disappears in degenerating cells. We observed increased cell and neurite degeneration over a period of 28 h in the DRG cultures transfected with Sarm11–724(ARM-mut) (Fig- ure 5E, right panel; and Figure 5F, right panel on the left and the pink curve on the right) comparable to the constitutively active ARM-deleted Sarm1409–724 (Figure 5E, middle right panel; and Figure 5F, middle right panel on the left and the red curve on the right) while observing minimal cell death in Sarm11–724 con- trol (Figure 5E, middle left panel; and Figure 5F, middle left panel on the left and the gray curve on the right). Therefore, disrupting the ARM-TIR interaction by mutation also specifically results in axon fragmentation and neuronal loss.

and F255 is sufficient to prevent ARM domain-mediated autoinhibition.
To determine whether changes in the biochemical NAD hy- drolase activity translate to changes in SARM1-mediated cell death, we transfected SARM1 variants into the immortalized osteosarcoma U2OS cell-line. We observed an increase in NAD depletion in cells transfected with SARM150–724(ARM-mut) (Figure 5C, pink bar) but no increase in NAD depletion in cells

NMN-Driven Activation of SARM1 Is Dependent on an Intact ARM-TIR Lock
It has recently been shown that NMN and the cell-permeable NMN analog CZ48 can activate SARM1 and drive non-apoptotic cell-death (Zhao et al., 2019). It is not known if just NMN or other metabolites in the NAD salvage pathway bind to SARM1, if the binding is to the active site or to an allosteric site, or how NMN

A

B

C

Figure 4. The ARM Domain Traps the TIR Domain in an Inactive Conformation
(A)Surface and cartoon representation showing that the TIR domain hangs off of the ARM domain in a monomeric conformation and is about 25 A˚ away from the neighboring TIR domain. The ARM-TIR0 interface is indicated by the black,
dashed box.
(B)Alignment of the TIR domain from the cryo-EM structure with the crystal structure of the TIR domain (PDB: 6O0R). Helices and loop regions are labeled.
(C)Detailed view of the ARM-TIR0 interface shown as a cartoon and stick
representation. The interaction involves hydrophobic interactions between TIR helix aa0 and ARM helix H13.

binding activates the TIR domain (Zhao et al., 2019). We found that treatment with NMN, but no other metabolites in the NAD salvage pathway, activates near full-length, mammalian-ex-

pressed SARM150–724 in an εNAD-based assay (Figure 6A) and an HPLC-based assay (Figure S7A). Bacterially expressed MBP SARM150–724 wild-type was also stimulated by NMN (Fig-
ure S7B). We observed a clear dose-response with NMN treat- ment (Figure 6B). Furthermore, CZ48 also activated SARM1 in cortical cells, resulting in neuronal cell death as measured by cell viability (Figure S7C). In comparison, Sarm1—/— cortical cells were significantly less susceptible to CZ48 (Figure S7C). Impor- tantly, NMN did not further stimulate constitutively active MBP SARM150–724(ARM-mut) (Figure 6C), suggesting that an intact ARM-TIR lock is necessary for NMN activation.
We next wanted to investigate whether the NMN binding pocket overlaps with the NAD substrate binding pocket. NAM has been shown to inhibit the reported NAD hydrolase activity of the TIR domain (Essuman et al., 2017), and docking indicates that NAM likely binds near catalytic residue E642 (Figure S7D), suggesting product inhibition of SARM1 by NAM. We show that NAM inhibits constitutively active SARM1409–724 and NMN-activated SARM150–724 in a quantitative HPLC assay (Fig- ures 6D and 6E). In fact, NAM inhibition is right-shifted upon increasing NAD concentrations (Figures 6D and 6E) and increasing NAM concentrations resulted in increased KM and decreased Vmax (Figure 6F), consistent with NAM being a mixed competitive inhibitor with preference for binding free enzyme over the enzyme-substrate complex. That co-incubation of NMN and NAM with SARM150–724 did not abolish NMN-depen- dent SARM1 activation and NAM-dependent SARM1 inhibition, suggests that NMN likely binds to an allosteric site of SARM1 to activate NAD degradation.
We attempted structural reconstruction of an activated SARM150–724 in complex with NMN, but a highly preferred parti- cle orientation and partial ARM domain deterioration in some particles prevented high-resolution reconstruction. Representa- tive class-average particles looked similar to autoinhibited SARM150–724 and did not display a clear conformational change of the TIR domains (Figure S7E). Potentially, SARM1 activation may involve multimerization of the TIR domains or even a re- structuring of loop regions upon ARM release that allows the active site to adopt an active conformation, but these processes may be transient and difficult to capture accurately for recon- structions. Consistent with this, we did not observe a change in protein oligomerization state of SARM150–724 by gel-filtration analysis (Figure S7F). Therefore, after incubation with NMN, acti- vation remains an area for future research, although our results clearly indicate that the ARM-TIR lock serves as a specific meta- bolic sensor that can tightly couple changes in the local meta- bolic environment, including an increase in NMN to SARM1 activation.

DISCUSSION

Our structural and functional studies provide novel insight into the architecture of SARM1 and its autoinhibition. Our structures reveal that SARM1 exists as a pre-assembled, SAM-domain- dependent octamer in the inactive and active states. This is con- trary to the expectation that SAM-domain oligomerization is a key conformational transition in SARM1 activation. Our cryo- EM reconstruction of SARM150–724 also reveals the structure of

A

C

E

Figure 5. Disrupting the ARM-TIR Lock Activates SARM1
(A)Schematic diagram of SARM1 protein variants used in assays. Mutations made in SARM150–724 are represented as bars above the domain where they reside.
(B)εNAD fluorescence-based assay of SARM150–724 mutants.
(C)NAD levels of cells transfected with SARM1 constructs as analyzed by HPLC.
(D)Cell-titer Glo assay to evaluate cell viability in cells transfected with ARM and TIR domain containing SARM1 constructs.
(E)Representative images of DRG cell clusters co-transfected with wild-type (WT), ARM-deleted, or ARM-mutated SARM1 and mCherry. White arrows and asterisks denote degenerating cell bodies and axons, respectively. The full time course is shown in Figure S6.
(F)Left, whole-well overview of DRG cells as described in (E). Right, quantitation of mCherry signal normalized to initial DRG cell area. For all graphs, data are represented as the means + SD.

D E F

Figure 6. NMN Activates Wild-Type SARM1 but Not ARM-Mut
(A)Analysis of SARM150–724 activation using metabolites in the NAD synthesis and degradation pathways as analyzed by the εNAD fluorescence-based assay. Each reaction used a concentration of 1 mM of metabolite. NMN, nicotinamide mononucleotide; NAM, nicotinamide; ADPR, adenine diphosphate ribose; cADPR, cyclic adenine diphosphate ribose; ATP, adenine triphosphate; PRPP, phosphoribosyl pyrophosphate.
(B)Dose response of NMN activation of SARM150–724 as analyzed by the εNAD fluorescence-based assay.
(C)Activation of SARM1 variants upon treatment with 1 mM of NMN as analyzed by the εNAD fluorescence-based assay.
(D)NAM inhibits constitutively active SARM1409–724, as analyzed by HPLC assay.
(E)NAM also inhibits NMN-stimulated SARM150–724, as analyzed by HPLC assay.
(F)NAM is a mixed competitive inhibitor of constitutively active SARM1409–724 as shown by kinetic analysis. For all graphs, data are represented as the means ± SD.

the previously uncharacterized ARM domain and shows that, although it interacts with all other SARM1 domains, importantly, it directly regulates the activity of the TIR domain.
The ARM domain binds to the TIR domain, and mutating this interface is sufficient to activate the protein. Several pieces of evidence suggest that TIR-TIR interactions are required for activity. First, we and others have observed that the TIR domain alone is active only at very high protein con- centrations (Figure S5F) or conditions that favor molecular crowding and potentially forced oligomerization (Essuman et al., 2017; Horsefield et al., 2019). Furthermore, the isolated TIR domain does not drive downstream axon fragmentation (Summers et al., 2016; Zhao et al., 2019). The TIR domains also become more proximal to each other upon activation with a cell permeant NMN analog, as analyzed by a split lucif- erase reporter assay (Zhao et al., 2019). Therefore, it is logical that the ARM domain prevents TIR-TIR interactions as evi- denced in our autoinhibited structure, but a complete under- standing of how TIR-TIR interactions form continues to elude us because the TIR domains could not be resolved in the active SARM1409–724 cryo-EM structure (Figure S3C). A recent crystal structure (Horsefield et al., 2019) reveals putative TIR- TIR0 packing interactions that the ARM-TIR interactions observed in the SARM150–724 cryo-EM structure would block

(Figure 7A). Intriguingly, the ARM interface residues clash with an interaction between residues V582 and L586 on the aa helix with the same residues on the aa0 helix of its mate in the asymmetric unit (Figure 7B). The ARM-TIR0 interaction would additionally block a different potential TIR-TIR0 interac- tion surface between residues Y568 and H6850 (Figure S7G). Although the significance of packing interactions should be treated with caution, it is compelling that mutation of either of these residues abolishes activity, suggesting that the site is likely physiologically relevant (Horsefield et al., 2019).
We report that the ARM-TIR lock serves as a specific sensor of injury-driven local metabolic dysregulation and disruption of this interaction is sufficient to drive axon fragmentation (Figure 5E). That NMN specifically activates SARM1 also provides a compel- ling explanation for the in vivo observation that Sarm1—/— rescues the perinatal lethality associated with Nmnat2—/—. Addi- tionally, the fact that NAM product inhibits SARM1 (Figure 6F) speaks to the exquisite regulation of this potent axon destruction mode by the NAD salvage pathway.
Neurons in general and axons specifically are energetically demanding. Although the human body has three NAD synthesis pathways, the salvage pathway serves as the main source of NAD in neurons (Liu et al., 2018). It is thus intriguing and yet fitting that SARM1, an axon-enriched NAD hydrolase, is so tightly and

A B

C

Figure 7. Models of TIR-TIR Interactions That May Activate SARM1
(A)Alignment of the TIR domain from the cryo-EM structure (light gray surface) with the TIR domain crystal (TIR, green cartoon; TIR0, cyan cartoon; PDB: 6O0R) indicates that the ARM domain (magenta surface) occupies the same space as the TIR0 domain from the crystal structure. The main interaction site is indicated with a black, dashed box.
(B)A zoom-in view of the ARM-TIR interaction (transparent cartoons and sticks) indicates that ARM residues critical for interaction with the TIR domain would clash with residues from a TIR-TIR0 interaction from the crystal structure.
(C)Hypothetical model of SARM1 activation. SARM1 exists in an autoinhibited conformation where the ARM domains lock the TIR domains in an inactive conformation (top left). Hiding the ARM domains (top right) reveals that the TIR domains are connected to the SAM2 domains via a flexible (and unresolved) linker. A transient high concentration of NMN triggers a conformation change, releasing the ARM domains and likely causing the TIR domains to dimerize—possibly on top of the SAM2 domains—stimulating the NAD hydrolase function of SARM1 (bottom). Note, this is a hypothetical model, and although active TIR domains are displayed as dimers, other oligomerization states are possible.

specifically regulated by the major metabolites in the NAD- salvage pathway. Our results also explain why SARM1 serves as the central switch that drives axons from a latent phase to a degenerative phase assimilating the upstream metabolic changes in the local axonal environment while having the possi- bility of being able to apply breaks on its catastrophic destructive ability by a metabolite in the same pathway. Finally, our studies also can explain some of the uncertainty around the role of modulating nicotinamide phosphoribosyltransferase (NAMPT)—the enzyme that synthesizes NMN—to prevent axon degeneration (Di Stefano et al., 2015). Our data provide an expla- nation for why NAMPT inhibition may be beneficial in the context of neurological injuries where NMNAT2 is depleted. In these in- stances, NAMPT inhibition will lower NMN levels and keep NAM levels higher, both of which will result in reduced SARM1 activity. In vivo, in addition to NMN, additional environmental fac- tors, such as elevated Ca2+ may also have a role in SARM1 acti- vation, which will need to be investigated further.
Interestingly, despite significant efforts, high-resolution struc- tural information for the active hydrolase state of SARM1 (NMN- activated SARM150–724 and ARM-deleted, constitutively active SARM1409–724) remain elusive. Because the TIR domain is connected to the SAM2 domain and because some additional density was observed above this domain in SARM1409–724 recon- struction (Figure S3C), it is tempting to hypothesize that the TIR domains assemble on top of the SAM2 ring. In this hypothetical model, TIRs may oligomerize into dimers, tetramers, or an oc- tamer to become active (Figure 7C). However, if the TIR domains assembled into stable oligomers upon activation, we should have been able to capture those more readily in our cryo-EM studies. On the contrary, the TIR domains were poorly resolved in the active structures despite clear resolution of the SAM domain oc- tameric core. It is conceivable that the NMN-activated hydrolase state is a transient state en route to formation of a larger, pro- degeneration complex. Possibly, other regions in the ARM or SAM domains or other co-factor proteins may additionally facili- tate positioning the TIR domain into an active conformation upon release of autoinhibition in a cellular context. Recently, phos- phorylation of SARM1 at residue S548—which lies in the linker between the SAM2 and TIR domains—was shown to increase ac- tivity, suggesting that post-translation modification may also be involved in regulation (Murata et al., 2018). Overall, our studies provide an understanding of the molecular and mechanistic de- tails of the transition of SARM1 from an autoinhibited to active state. We show that SARM1’s activity is tightly regulated by the key metabolites in the NAD-salvage pathway. SARM1 is allosteri- cally activated upon increased NMN levels, hydrolyzes NAD on the path to axon destruction, and can be catalytically inhibited by the product NAM. Our studies lay the foundation for future therapies directed against axon degeneration in neurological disorders.

STAR+METHODS

Detailed methods are provided in the online version of this paper and include the following:

d KEY RESOURCES TABLE

d RESOURCE AVAILABILITY
B Lead Contact
B Materials availability
B Data and code availability
d EXPERIMENTAL MODEL AND SUBJECT DETAILS
B Bacterial Cultures for Protein Purification
B HEK293F Cultures for Protein Purification
B U2OS Cultures for CellTiter-Glo and NAD HPLC As- says and western blot
B Primary Neuronal Culture
B Primary Cortical Cell Culture
d METHOD DETAILS
B Protein Expression and Purification
B Cryo-EM data collection
B Cryo-EM Image processing
B Model building and refinement
B Gel Filtration Analysis
B Fluorescent NAD analog hydrolase assay
B CellTiter-Glo Assay
B NAD HPLC Assay
B Western Blot Analysis
B Primary Neuronal Culture, mRNA Transfection, and Assay Quantification
B mRNA Construction and Production
B SARM1 NADase HPLC assay
B SARM1 Kinetic Analysis
B Primary Cortical Cell Culture and CZ48 treatment
B Molecular Docking Analysis and Structural Figures
d QUANTIFICATION AND STATISTICAL ANALYSIS

SUPPLEMENTAL INFORMATION

Supplemental Information can be found online at https://doi.org/10.1016/j. celrep.2020.107999.

ACKNOWLEDGMENTS

We thank Steve McKnight (University of Texas [UT] Southwestern), Marc Freeman (Vollum Institute at Oregon Health and Science University), and Michael Coleman (University of Cambridge) for critical reading of the manu- script. We thank Min Fang and Lillian Sutherland from the McKnight lab and Paul Meraner from the Freeman lab for helpful discussions during the project. Single-particle cryo-EM data were collected at the UT Southwestern Cryo- Electron Microscopy Facility that is funded by a Cancer Prevention and Research Institute of Texas (CPRIT) core facility support award (RP170644). We thank D. Nicastro and D. Stoddard for facility access and data acquisition. We thank Craig Yoshioka (Oregon Health and Science University) for assis- tance with cryo-EM sample imaging. X.B. is the Virginia Murchison Linthicum Scholar in Medical Research at UT Southwestern. The research is supported by Nura Bio and in part by the Cancer Prevention and Research Institute of Texas (CPRIT) (RR160082) and the Welch Foundation (I-1944-20180324).

AUTHOR CONTRIBUTIONS

M.B., T.X., D.A.T., Y.-S.Y., T.C.B., X.B., and S.S., designed the experiments for the study. T.X. and X.B. collected EM data and built the structure. A.C. cloned, expressed, and purified SARM150–724 for the structural studies. P.M., S.L., and
M.B. cloned, expressed, and purified all other SARM1 proteins. S.L, P.M., and
M.B. performed εNAD-based biochemical experiments. J.D. conducted the cell-based toxicity assay. D.A.T. and S.L. performed HPLC-based experi- ments. Y.-S.Y, G.D., and T.C.B. designed and executed the neuronal toxicity studies. J.A.K. and S.P.B. conducted molecular docking experiments. M.B.,

T.X., and T.C.B. generated the figures. M.B., S.S., T.X., and X.B wrote the manuscript with contributions from all authors. S.S. supervised the program.

DECLARATION OF INTERESTS

M.B., D.A.T., S.L., P.M., Y.-S.Y., G.D., T.C.B., J.D., A.C., J.A.K., S.P.B., and
S.S. are employees or former employees of Nura Bio and hold Nura Bio stock.

Received: March 6, 2020
Revised: June 9, 2020
Accepted: July 14, 2020
Published: August 4, 2020

REFERENCES

Adams, P.D., Afonine, P.V., Bunko´ czi, G., Chen, V.B., Davis, I.W., Echols, N.,
Headd, J.J., Hung, L.W., Kapral, G.J., Grosse-Kunstleve, R.W., et al. (2010). PHENIX: a comprehensive Python-based system for macromolecular struc- ture solution. Acta Crystallogr. D Biol. Crystallogr. 66, 213–221.
Carty, M., Goodbody, R., Schro¨ der, M., Stack, J., Moynagh, P.N., and Bowie,
A.G. (2006). The human adaptor SARM negatively regulates adaptor protein TRIF-dependent Toll-like receptor signaling. Nat. Immunol. 7, 1074–1081.
Chen, V.B., Arendall, W.B., 3rd, Headd, J.J., Keedy, D.A., Immormino, R.M., Kapral, G.J., Murray, L.W., Richardson, J.S., and Richardson, D.C. (2010). MolProbity: all-atom structure validation for macromolecular crystallography. Acta Crystallogr. D Biol. Crystallogr. 66, 12–21.
Chuang, C.F., and Bargmann, C.I. (2005). A Toll-interleukin 1 repeat protein at the synapse specifies asymmetric odorant receptor expression via ASK1 MAPKKK signaling. Genes Dev. 19, 270–281.
Di Stefano, M., Nascimento-Ferreira, I., Orsomando, G., Mori, V., Gilley, J., Brown, R., Janeckova, L., Vargas, M.E., Worrell, L.A., Loreto, A., et al. (2015). A rise in NAD precursor nicotinamide mononucleotide (NMN) after injury promotes axon degeneration. Cell Death Differ. 22, 731–742.
Emsley, P., Lohkamp, B., Scott, W.G., and Cowtan, K. (2010). Features and development of Coot. Acta Crystallogr. D Biol. Crystallogr. 66, 486–501.
Essuman, K., Summers, D.W., Sasaki, Y., Mao, X., DiAntonio, A., and Mil- brandt, J. (2017). The SARM1 Toll/interleukin-1 receptor domain possesses intrinsic NAD+ cleavage activity that promotes pathological axonal degenera- tion. Neuron 93, 1334–1343.e1335.
Essuman, K., Summers, D.W., Sasaki, Y., Mao, X., Yim, A.K.Y., DiAntonio, A., and Milbrandt, J. (2018). TIR domain proteins are an ancient family of NAD+- consuming enzymes. Curr. Biol. 28, 421–430.e424.
Geisler, S., Doan, R.A., Strickland, A., Huang, X., Milbrandt, J., and DiAntonio,
A. (2016). Prevention of vincristine-induced peripheral neuropathy by genetic deletion of SARM1 in mice. Brain 139, 3092–3108.
Gerdts, J., Summers, D.W., Sasaki, Y., DiAntonio, A., and Milbrandt, J. (2013). Sarm1-mediated axon degeneration requires both SAM and TIR interactions. J. Neurosci. 33, 13569–13580.
Gerdts, J., Brace, E.J., Sasaki, Y., DiAntonio, A., and Milbrandt, J. (2015). SARM1 activation triggers axon degeneration locally via NAD+ destruction. Science 348, 453–457.
Gilley, J., and Coleman, M.P. (2010). Endogenous Nmnat2 is an essential sur- vival factor for maintenance of healthy axons. PLoS Biol. 8, e1000300.
Gilley, J., Orsomando, G., Nascimento-Ferreira, I., and Coleman, M.P. (2015). Absence of SARM1 rescues development and survival of NMNAT2-deficient axons. Cell Rep. 10, 1974–1981.
Gilley, J., Ribchester, R.R., and Coleman, M.P. (2017). Sarm1 Deletion, but Not WldS, Confers Lifelong Rescue in a Mouse Model of Severe Axonopathy. Cell Rep. 21, 10–16.
Henninger, N., Bouley, J., Sikoglu, E.M., An, J., Moore, C.M., King, J.A., Bowser, R., Freeman, M.R., and Brown, R.H., Jr. (2016). Attenuated traumatic axonal injury and improved functional outcome after traumatic brain injury in mice lacking Sarm1. Brain 139, 1094–1105.

Horsefield, S., Burdett, H., Zhang, X., Manik, M.K., Shi, Y., Chen, J., Qi, T., Gil- ley, J., Lai, J.S., Rank, M.X., et al. (2019). NAD+ cleavage activity by animal and plant TIR domains in cell death pathways. Science 365, 793–799.
Kim, C.A., Phillips, M.L., Kim, W., Gingery, M., Tran, H.H., Robinson, M.A., Fa- ham, S., and Bowie, J.U. (2001). Polymerization of the SAM domain of TEL in leukemogenesis and transcriptional repression. EMBO J. 20, 4173–4182.
Knight, M.J., Leettola, C., Gingery, M., Li, H., and Bowie, J.U. (2011). A human sterile alpha motif domain polymerizome. Protein Sci. 20, 1697–1706.
Liu, H.W., Smith, C.B., Schmidt, M.S., Cambronne, X.A., Cohen, M.S., Mi- gaud, M.E., Brenner, C., and Goodman, R.H. (2018). Pharmacological bypass of NAD+ salvage pathway protects neurons from chemotherapy-induced degeneration. Proc. Natl. Acad. Sci. USA 115, 10654–10659.
Loreto, A., Di Stefano, M., Gering, M., and Conforti, L. (2015). Wallerian degen- eration is executed by an NMN-SARM1-dependent late Ca2+ influx but only modestly influenced by mitochondria. Cell Rep. 13, 2539–2552.
Murata, H., Khine, C.C., Nishikawa, A., Yamamoto, K.I., Kinoshita, R., and Sa- kaguchi, M. (2018). c-Jun N-terminal kinase (JNK)-mediated phosphorylation of SARM1 regulates NAD+ cleavage activity to inhibit mitochondrial respira- tion. J. Biol. Chem. 293, 18933–18943.
Osterloh, J.M., Yang, J., Rooney, T.M., Fox, A.N., Adalbert, R., Powell, E.H., Sheehan, A.E., Avery, M.A., Hackett, R., Logan, M.A., et al. (2012). dSarm/ Sarm1 is required for activation of an injury-induced axon death pathway. Sci- ence 337, 481–484.
Raffaelli, N., Sorci, L., Amici, A., Emanuelli, M., Mazzola, F., and Magni, G. (2002). Identification of a novel human nicotinamide mononucleotide adenylyl- transferase. Biochem. Biophys. Res. Commun. 297, 835–840.
Scheres, S.H. (2012a). A Bayesian view on cryo-EM structure determination. J. Mol. Biol. 415, 406–418.
Scheres, S.H. (2012b). RELION: implementation of a Bayesian approach to cryo-EM structure determination. J. Struct. Biol. 180, 519–530.
Scheres, S.H., and Chen, S. (2012). Prevention of overfitting in cryo-EM struc- ture determination. Nat. Methods 9, 853–854.
Sporny, M., Guez-Haddad, J., Lebendiker, M., Ulisse, V., Volf, A., Mim, C., Isu- pov, M.N., and Opatowsky, Y. (2019). Structural evidence for an octameric ring arrangement of SARM1. J. Mol. Biol. 431, 3591–3605.
Studier, F.W. (2005). Protein production by auto-induction in high density shaking cultures. Protein Expr. Purif. 41, 207–234.
Summers, D.W., Gibson, D.A., DiAntonio, A., and Milbrandt, J. (2016). SARM1- specific motifs in the TIR domain enable NAD+ loss and regulate injury- induced SARM1 activation. Proc. Natl. Acad. Sci. USA 113, E6271–E6280.
Tang, G., Peng, L., Baldwin, P.R., Mann, D.S., Jiang, W., Rees, I., and Ludtke,
S.J. (2007). EMAN2: an extensible image processing suite for electron micro- scopy. J. Struct. Biol. 157, 38–46.
Turkiew, E., Falconer, D., Reed, N., and Ho¨ ke, A. (2017). Deletion of Sarm1 gene is neuroprotective in two models of peripheral neuropathy. J. Peripher. Nerv. Syst. 22, 162–171.
Walker, L.J., Summers, D.W., Sasaki, Y., Brace, E.J., Milbrandt, J., and DiA- ntonio, A. (2017). MAPK signaling promotes axonal degeneration by speeding the turnover of the axonal maintenance factor NMNAT2. eLife 6, e22540.
Wan, L., Essuman, K., Anderson, R.G., Sasaki, Y., Monteiro, F., Chung, E.H., Osborne Nishimura, E., DiAntonio, A., Milbrandt, J., Dangl, J.L., and Nishi- mura, M.T. (2019). TIR domains of plant immune receptors are NAD+-cleaving enzymes that promote cell death. Science 365, 799–803.
Waterhouse, A., Bertoni, M., Bienert, S., Studer, G., Tauriello, G., Gumienny, R., Heer, F.T., de Beer, T.A.P., Rempfer, C., Bordoli, L., et al. (2018). SWISS- MODEL: homology modelling of protein structures and complexes. Nucleic Acids Res. 46 (W1), W296–W303.
Williams, S.J., Yin, L., Foley, G., Casey, L.W., Outram, M.A., Ericsson, D.J., Lu, J., Boden, M., Dry, I.B., and Kobe, B. (2016). Structure and function of the TIR domain from the grape NLR protein RPV1. Front. Plant Sci. 7, 1850.
Yalowitz, J.A., Xiao, S., Biju, M.P., Antony, A.C., Cummings, O.W., Deeg, M.A., and Jayaram, H.N. (2004). Characterization of human brain nicotinamide 50-

mononucleotide adenylyltransferase-2 and expression in human pancreas. Biochem. J. 377, 317–326.
Zhang, K. (2016). Gctf: Real-time CTF determination and correction. J. Struct. Biol. 193, 1–12.
Zhao, Z.Y., Xie, X.J., Li, W.H., Liu, J., Chen, Z., Zhang, B., Li, T., Li, S.L., Lu,
J.G., Zhang, L., et al. (2019). A cell-permeant mimetic of NMN activates