Programmed axon degeneration: from mouse to mechanism to medicine

Abstract

Wallerian degeneration is a widespread mechanism of programmed axon degeneration. In the three decades since the discovery of the Wallerian degeneration slow (Wld^S) mouse, research has generated extensive knowledge of the molecular mechanisms underlying Wallerian degeneration, demonstrated its involvement in non-injury disorders and found multiple ways to block it. Recent developments have included: the detection of NMNAT2 mutations that implicate Wallerian degeneration in rare human diseases; the capacity for lifelong rescue of a lethal condition related to Wallerian degeneration in mice; the discovery of ‘druggable’ enzymes, including SARM1 and MYCBP2 (also known as PHR1), in Wallerian pathways; and the elucidation of protein structures to drive further understanding of the underlying mechanisms and drug development. Additionally, new data have indicated the potential of these advances to alleviate a number of common disorders, including chemotherapy-induced and diabetic peripheral neuropathies, traumatic brain injury, and amyotrophic lateral sclerosis.

Wallerian degeneration was originally defined as the degeneration of an axon that takes place distal to an injury, characterized by granular disintegration of the cytoskeleton, mitochondrial swelling and axon fragmentation¹; however, the mechanism is now known to also occur in many non-injury disorders (TABLE 1, FIG. 1). Wallerian degeneration has gained importance in the fields of neurodegenerative and axonal disorders because of its widespread occurrence, its well-characterized and ‘druggable’ mechanism and the ability to alleviate — and sometimes completely prevent — axon loss by blocking it². Our deep mechanistic understanding of Wallerian degeneration stems largely from axon injury studies, but the regulatory genes that have been identified in these studies are now known to also influence axon survival in many other circumstances³. Recent advances (highlighted below) include the identification of a genetic association between Wallerian degeneration and human disease, new rescue effects in animal models, and the discovery of new drug targets with unexpected enzyme activities associated with Wallerian degeneration and progress towards the elucidation of their protein structures.

Table 1 |.

Human diseases associated with the Wallerian degeneration pathway

Human disease	Findings	Comment	Refs
Alzheimer disease	NMNAT2 protein and mRNA are reduced in brains of individuals with Alzheimer disease	This finding suggests that reduced NMNAT2 is associated with, and may be a risk factor for, the development of Alzheimer disease	95
Erythromelalgia and peripheral neuropathy	A homozygous NMNAT2 partial loss-of-function mutation in two individuals with erythromelalgia and peripheral neuropathy	Loss-of-function mutation in NMNAT2 results in development of peripheral neuropathy	89
Fetal akinesia deformation syndrome	Compound heterozygous NMNAT2 severe loss-of-function mutations in two fetuses with fetal akinesia deformation syndrome	Severe loss-of-function mutation in NMNAT2 results in stillbirth, mimicking observations in NMNAT2 null mice	90
ALS	GWAS show association of the SARM1 locus with ALS	A pathogenic role of SARM1 in ALS must involve gain-of-function variants that increase the risk of axonal degeneration	101,102
ALS	An ALS-causing mutation in TARDBP reduces levels of stathmin 2 in iPSC-derived motor neurons; a similar decrease is common in sporadic ALS	Low levels of stathmin 2 increase the likelihood of Wallerian degeneration and could therefore be an important modifier or risk factor for ALS	96,97

ALS, amyotrophic lateral sclerosis; GWAS, genome-wide association studies; iPSC, induced pluripotent stem cell; NMNAT2, nicotinamide mononucleotide adenylyltransferase 2; SARM1, sterile-α and Toll/interleukin 1 receptor (TIR) motif containing protein 1; TDP43, transactive response DNA binding protein 43 kDa.

Fig. 1 |. Activation of Wallerian degeneration in injury and disease.

The figure shows three ways in which the Wallerian degeneration mechanism can be activated. a | In classical Wallerian degeneration paradigms, axon injury prevents the delivery of nicotinamide mononucleotide adenylyltransferase 2 (NMNAT2) from the soma along with all other cargoes¹². Because of its short half-life, NMNAT2 is quickly lost in the distal stump, which triggers sterile-α and Toll/interleukin 1 receptor (TIR) motif containing protein 1 (SARM1-dependent degeneration²⁷. b | Axonal transport impairment (for example, due to a toxin, a protein defect or ageing) limits the supply of all cargoes to distal axons. Delivery of NMNAT2 to distal ends of the axon is particularly affected because of its short half-life. When NMNAT2 falls below the threshold for survival, the distal terminal dies back. c | Mutation of the Nmnat2 gene leads to loss or dysfunction of the protein^13,89,90. Without it, the axon cannot grow, causing a developmental disorder (unless rescued through a Sarm1 deletion, for example). If the mutation causes only a partial loss of function, the axon grows but is more susceptible to degeneration.

The discovery, three decades ago, of a mouse strain with an unusual phenotype began the research that has led us to today’s detailed molecular understanding of Wallerian degeneration (FIG. 2). In these Wallerian degeneration slow (Wld^S) mice, transected axons survive tenfold longer than normal⁴. Until this finding, Wallerian degeneration was generally considered to result from a lack of ‘nourishment’ of the distal axon by the soma in transected axons¹. Alternatively, Lubinska⁵ had proposed that one specific, essential substance, delivered by axonal transport, inhibits Wallerian degeneration. Lubinska had no means of identifying this substance; the existence of a mutant mouse has now made this possible.

Fig. 2 |. Wallerian degeneration timeline.

Prior to discovery of the Wallerian degeneration slow (Wld^S) mouse, there was wide acceptance that axonal transport was required for axon survival and that axon injury could serve as a model for disease, albeit with no molecular understanding of the process⁵. The identification of the axon-protective gene encoded by Wld^S made it possible to address this question and led to the identification of nicotinamide mononucleotide adenylyltransferase 2 (NMNAT2) as the best match for the previously proposed ‘Wallerian degeneration inhibitor’¹². Conservation of the Wallerian mechanism in Drosophila melanogaster has enabled the discovery of other major players in the pathway, including sterile-α and TIR motif containing protein 1 (SARM1)¹⁸. More recently, human genetic studies have started to identify the clinical disorders involving this preventable degeneration pathway, with promising leads in the quest to develop drugs to block it^89,90. ALS, amyotrophic lateral sclerosis; CIPN, chemotherapy-induced peripheral neuropathy; DLK, dual leucine zipper kinase; GWAS, genome-wide association studies; HFD, high-fat diet; NMN, nicotinamide mononucleotide; SOD1, superoxide dismutase 1; TIR, Toll/interleukin 1 receptor.

Structure–function analysis of the protective Wld^S gene product, a unique fusion of nicotinamide mononucleotide adenylyltransferase 1 (NMNAT1; an enzyme involved in the synthesis of the redox and signalling cofactor nicotinamide adenine dinucleotide (NAD)), with part of the ubiquitin ligase UBE4B⁶, suggested a key role for axonal NAD metabolism in axon maintenance⁷. Indeed, most subsequent studies have suggested that the NMNAT enzyme activity of the Wld^S gene product, localized to axons by the fused protein, is required for the axon protection phenotype^8–10. NMNAT2, a related protein that is normally found in the axon and for which WLD^S can substitute, was subsequently found to be essential for axon growth and survival^11–13 and remains the best fit for Lubinska’s natural ‘inhibitor’ of Wallerian degeneration. A key property of NMNAT2 is its short half-life^12,14, which necessitates constant resupply by new protein synthesis and axonal transport. When NMNAT2 is lost from axons, whether owing to injury, mutation, silencing, axonal transport disruption, or other causes, the resulting degeneration can be blocked by WLD^S, indicating that it involves a Wallerian-like mechanism³ (FIG. 1).

Wallerian degeneration slow (Wld^S) mice

A mutant strain of mouse showing a tenfold delay in the onset of Wallerian degeneration after axotomy as well as axon protection in many disease models.

Axonal transport

The ATP-dependent, bidirectional trafficking of axonal proteins, organelles, mRNAs and other cargoes delivering axonal constituents to where they are required, often over large distances.

The discovery of the mouse Wld^S gene also led to the finding that the pathway is conserved in Drosophila melanogaster^15,16 (FIG. 2). Powerful D. melanogaster genetic studies subsequently revealed a set of additional positive and negative regulators of Wallerian degeneration^17–19 that are largely conserved in mammals and have recently provided a strong basis for drug development efforts (FIG. 3). The key roles of D. melanogaster in elucidating this pathway, and the similarities and differences between the pathway in mammals and flies, have recently been reviewed elsewhere²⁰; therefore, they are not the main focus of this article.

Fig. 3 |. A working model of the Wallerian degeneration pathway.

The figure shows current knowledge of the factors that determine nicotinamide mononucleotide adenylyltransferase 2 (NMNAT2) activity in axons, how NMNAT2 loss activates sterile-α and Toll/interleukin 1 receptor (TIR) motif containing protein 1 (SARM1) and events downstream of SARM1 leading to axon degeneration. Various factors disrupt the delivery of functional NMNAT2 to axons. NMNAT2 turnover is mediated by the mitogen-activated protein kinases (MAPK) dual leucine zipper kinase (DLK) and leucine zipper-bearing kinase (LZK), as well as a ubiquitin ligase complex consisting of MYC-binding protein 2 (MYCBP2), s-phase kinase-associated protein 1A (SKP1A) and F-box protein 45 (FBXO45)³⁸. In combination, this reduces NMNAT2 levels and causes downstream activation of SARM1, probably as a result of increased levels of nicotinamide mononucleotide (NMN) but potentially also in other ways³⁵. The combined effect of decreased NMNAT2 and increased activation of SARM1 leads to a great decrease in axonal nicotinamide adenine dinucleotide (NAD), which may itself cause axon degeneration through ATP synthesis failure²⁵. Alternatively, other SARM1 substrates or its calcium mobilizing products could be important for the later stages of Wallerian degeneration. In Drosophila melanogaster, there is also an as-yet undefined role for Axundead (Axed) in Wallerian axon degeneration¹⁹. ADPR, ADP-ribose; cADPR, cyclic ADP-ribose.

This Review describes the major recent advances in our understanding of the Wallerian degeneration mechanism and distinguishes areas of agreement and ongoing debate. It reviews what we have learned from animal models about the roles of Wallerian degeneration in disease and explains why complementary human studies are vital. Finally, it discusses which proteins in the pathway might be targeted for therapeutic intervention and considers outstanding questions that remain to be addressed in order to move the field towards clinical application.

Wallerian mechanisms: recent advances

Mechanisms driving NAD loss.

NAD levels are known to decline in injured axons and nerves^21,22. Part of the explanation for this decrease is the rapid loss of its normal synthetic axonal enzyme, NMNAT2, which has a short half-life and cannot be replenished by axon transport from the soma after injury. Indeed, the presence of a more stable axonally targeted NMNAT, such as WLD^S, stabilizes NAD levels after injury²². However, an important recent discovery was the identification of the NAD-degrading activity of another axonal protein, sterile-α and Toll/interleukin 1 receptor (TIR) motif containing protein 1 (SARM1), as well as its activation as part of the molecular pathway that induces Wallerian degeneration²³ (FIG. 3).

SARM1 is a Toll-like receptor adapter protein that is required for the normal, rapid rate of Wallerian degeneration¹⁸. It has an ‘execution’ role in Wallerian degeneration, analogous to the roles of caspases as executors of apoptosis, although the molecular mechanism is distinct. This function was discovered in D. melanogaster, replicated using Sarm1^−/− mice¹⁸ and later confirmed in a separate mammalian RNA interference study²⁴. However, the mechanism by which it kills axons was not immediately apparent. In 2015, experiments employing forced dimerization of the SARM1 TIR signalling domain demonstrated an unknown NADase activity that generates ADP-ribose (ADPR) and cyclic ADPR (cADPR) from NAD²⁵. It was subsequently shown that murine SARM1 is required for extensive NAD loss after injury²⁵ and that a mutated human SARM1 that is unable to degrade NAD cannot, unlike wild-type human SARM1, support fast Wallerian degeneration when introduced to Sarm1^−/− mouse neurons²³. Isotopic labelling of NAD in mouse dorsal root ganglion (DRG) neuron cultures showed that there is both NAD synthesis failure at around 2 h after axon transection (likely owing to NMNAT2 loss) and a SARM1-dependent, fourfold acceleration of NAD degradation²⁶. The intrinsic NADase activity of SARM1’s own TIR domain was then demonstrated by experiments with cell-free transcribed and translated TIR protein²³.

SARM1 is known to act downstream of NMNAT2 loss in the Wallerian degeneration pathway as its removal in mice protects NMNAT2 null axons without fully preserving NAD levels²⁷. SARM1 activation exacerbates the effects of slowing NAD generation caused by NMNAT2 loss through an increase in NAD degradation, causing a precipitous decline in NAD levels (FIG. 3). The discovery of this enzyme activity has also changed the perception of the suitability of SARM1 as a drug target (discussed below).

Toll-like receptor

A family of receptors on plasma and endosomal membranes that detect molecular patterns associated with infection or cell damage, activating signalling pathways leading to inflammation or cell death.

The contribution of NMN to Wallerian degeneration.

Given the well-known roles of NAD in ATP synthesis and many other cellular activities²⁸, its loss is an obvious and attractive candidate mechanism to explain the axon degeneration that occurs after injury. However, it remains unclear whether the loss of NAD offers a full explanation. Manipulating NAD levels through methods that do not involve altering NMNATs (or their substrates and precursors) has been repeatedly found not to alter Wallerian degeneration^22,29,30. Moreover, inhibitors of nicotinamide phosphoribosyl transferase (NAMPT), the synthetic enzyme that generates NMN, such as FK866, lower NAD by up to 90%; however, rather than killing axons as expected, they partially phenocopy the effects of WLD^S, protecting injured axons^22,29–31. To explain this paradox, a harmful role for the NAD precursor nicotinamide mononucleotide (NMN) was proposed²² NMN accumulates in the axon distal to the injury, at least transiently, during Wallerian degeneration^22,26 and both WLDS and FK866 prevent this accumulation — WLD^S maintains NMN removal from axons by metabolizing it to produce NAD, whereas FK866 stops NMN synthesis²² (FIG. 4). Supporting the idea of NMN toxicity, ectopically expressed bacterial NMN deamidase, which removes NMN without altering NAD (at least after axon injury), strongly protects cut axons in mice^22,26,32 and exogenous NMN or ectopically expressed NMN synthase both restore degeneration to cut murine neurites that have been protected by FK866 (REFS^30,33,34). However, in conditions in which NAD levels are also high, NMN does not activate axon degeneration²⁶.

Fig. 4 |. Activation of SARM1 by NMN.

The figure summarizes evidence supporting activation of sterile-α and Toll/interleukin 1 receptor (TIR) motif containing protein 1 (SARM1) by nicotinamide mononucleotide (NMN). NMN accumulates after axon injury²² because the enzyme that normally removes it, nicotinamide mononucleotide adenylyltransferase 2 (NMNAT2), is quickly degraded¹². If left unchecked, this activates SARM1, leading to axon degeneration²⁷. However, NMN accumulation can be prevented in three ways (dark blue): through inhibition of its synthetic enzyme, nicotinamide phosphoribosyl transferase (NAMPT)^22,29–31; by overexpression, or increased axonal targeting, of any isoform of NMNAT^3,10, its normal processing enzyme in mammals; or by ectopic expression of bacterial NMN deamidase^22,32. Conversely, its membrane-permeable analogue CZ-48 activates SARM1 directly³⁵. Thus, NMN, or a close analogue, appears to be necessary and sufficient for SARM1 activation. ADPR, ADP-ribose; cADPR, cyclic ADP-ribose; NAD, nicotinamide adenine dinucleotide; NAMN, nicotinic acid mononucleotide; Nm, nicotinamide; WLD^S, Wallerian degeneration slow.

Since neither NAD loss nor NMN accumulation by themselves explain all available data, it can be hypothesized that both may be part of the mechanism. For example, the ratio between NAD and NMN levels may be important. Alternatively, the recent finding that NMN, or a synthetic membrane-permeable analogue (CZ-48), can activate SARM1 suggests that NMN could drive further NAD loss³⁵ (FIG. 3). However, such NMN-dependent SARM1 activation has not yet been demonstrated in axons.

Downstream executors of degeneration.

Even once SARM1 is activated, it is still unclear whether the cause of the axon degeneration is NAD loss, an increase in the level of a SARM1 enzyme product or the increased metabolism of an alternative substrate³⁵. If it is NAD loss, it is clear that levels must become very low to have an effect: NAD declines by 50% in transected sciatic nerves in Sarm1^−/− mice within 5 days of injury²⁷ and by 80% in severed DRG neurites in Sarm1^−/− mice within 36 h of injury³⁶; however, in both cases, the neurons survive for a considerably longer period. Changes in calcium mobilization and/or influx provide one alternative mechanism. The SARM1 products ADPR and cADPR are powerful calcium mobilizers²⁸ and, in mice, NMN activates a SARM1-dependent rise in intra-axonal calcium that immediately precedes the demise of the axon³⁴. Interestingly, in D. melanogaster, the protein Axundead (Axed), which (like SARM1 and its D. melanogaster orthologue dSarm) is required for rapid Wallerian degeneration, appears to act downstream of dSarm¹⁹. Axed is likely to act either as an executioner of degeneration or as a negative regulator of a powerful compensatory mechanism. Axed is a BTB/BACK-containing protein with no obvious role in calcium or NAD metabolism, although its ability to protect either in the presence of activated dSarm or in the absence of dNmnat¹⁹ (the only known NAD synthetic enzyme in D. melanogaster) strongly suggests that it influences NAD in some way. Elucidation of its function, as well as the identification and characterization of any mammalian orthologue will be crucial steps in understanding and preventing Wallerian degeneration.

Regulating NMNAT2 stability.

Understanding the regulation of NMNAT2 stability is another rapidly developing area. NMNAT2 had a half-life of less than 40 min in human embryonic kidney cell experiments¹⁴ and was greatly depleted in mouse neurites 4 h after transection¹². How such a labile protein can reach distal axons before being degraded, a journey that takes 2 days in the longest human axons, remains unclear³⁷. The existence of two distinct NMNAT2 pools — vesicular and non-vesicular — is probably part of the explanation¹⁴. It has been shown that, in mice, the distribution between vesicular and non-vesicular Nmnat2 is regulated in part by palmitoylation^14,38. However, studies that hypothesized that the moving, vesicular form of the enzyme may be more stable than non-vesicular NMNAT2 found that the opposite was true: disrupting vesicle targeting lengthened NMNAT2’s half-life and greatly strengthened its capacity to protect axons^14,39. Although candidate palmitoylation and depalmitoylation enzymes were identified⁴⁰, these are probably not good drug targets as they have many substrates. It has also been shown that mitogen-activated protein kinase (MAPK) signalling promotes NMNAT2 turnover as depletion of MKK4 (also known as MAP2K4) and MKK7 or blockage of their upstream regulators dual leucine zipper kinase (DLK) and leucine zipper-bearing kinase (LZK) extends the half-life of the palmitoylated form of NMNAT2 and increases the survival of transected murine neurites^38,41. MAPK signalling is also influenced by SARM1, but whether this influences axon survival remains unclear^38,42.

Turnover of the soluble, more stable NMNAT2 pool¹⁴ is influenced by the MYC-binding protein 2 (MYCBP2)–S-phase kinase-associated protein 1 (SKP1)-F-box protein 45 (FBXO45) ubiquitin ligase complex³⁸, each member of which has been separately implicated in Wallerian degeneration^17,43–46. Ubiquitylation determines NMNAT2 turnover rate¹⁴, but the newly discovered serine/threonine monoubiquitylation activity of MYCBP2 (REF.⁴⁷) raises important questions about the role of classical lysine polyubiquitylation⁴⁵ in this process. Like SARM1 and MAPK signalling, this is another promising area for drug development, especially as the structural basis of serine/threonine ubiquitylation is known⁴⁷.

Calcium mobilization

The release of calcium from intracellular stores such as endoplasmic reticulum and mitochondria, potentially activating calcium-activated proteases, under the control of second messenger molecules (many of which are NAD metabolites).

Next steps.

The recent rapid progress in our understanding of Wallerian degeneration mechanisms has raised many new questions. Prominent among them is ‘how is SARM1 activated?’ The recently discovered activation of SARM1 by NMN appears to provide one possible answer to this question³⁵ and, if confirmed in axons, could explain why NMN promotes Wallerian degeneration^22,32,34 (FIG. 4); however, whether NMN binds directly to SARM1 (and if so where), whether other proteins are involved and whether NAD can block this activation (and if so, what level of NAD is required to do so)²⁶ remain unclear. Emerging information about the structure of SARM1 (REFS^48,49) will be important in answering these questions.

Both in axons and in non-neuronal cells there are indications of other SARM1 activation mechanisms that need to be brought together for a full understanding of how this protein is regulated. On the proximal side of an axon injury, SARM1 is required for proinflammatory responses⁵⁰; however, in this location, there is no NMNAT2 loss¹² nor a rise in NMN, so there must be another activation mechanism at work. SARM1 expression level appears to be another point of regulation as this is altered by some viruses and Toll-like receptor ligands in neuronal and/or non-neuronal cells^51,52. A full understanding of SARM1 regulation also requires a better understanding of the roles of post-translational modifications, such as the reported JNK-dependent phosphorylation⁵³ and of mitochondrial targeting, reported by some to be important for the prodegenerative function of SARM1 (REFS^53,54) (but not by others²⁴). The normal subcellular location of SARM1 also remains to be resolved. While overexpressed and fluorescent protein-tagged SARM1 colocalizes with mitochondria under basal conditions^55,56, the endogenous protein does not colocalize with mitochondria¹⁸, or at least not completely^24,57.

Other questions arise about the evolution of SARM1 and its relationship to NAD hydrolysis by other, evolutionary-divergent TIR domains^49,58,59. The wider involvement of this newly discovered enzyme activity in innate immunity suggests that there may be an ancient mechanism of pathogen defence involving NAD modulation^49,59. It could be hypothesized that axon injury may have co-opted this mechanism later in evolution to remove damaged axons. However, the ultimate effector of this mechanism remains unknown.

Questions also remain about other regulators of NMNAT2. Could local translation of NMNAT2 (REF.⁶⁰) as well as the existence of different NMNAT2 pools^14,38 help explain how this unstable protein reaches distal axons without being degraded? It is not known precisely how MYCBP2, DLK and LZK influence the steady state level of each NMNAT2 pool; specifically, it is unclear which NMNAT2 residues they may phosphorylate or ubiquitylate and whether they are responsible for its polyubiquitylation or monoubiquitylation. Stabilizing, boosting the synthesis of⁶¹ and activating NMNAT2 have all been proposed as important therapeutic strategies for axonopathies, making these key questions for advancement in this area. Finally, identifying proteins that control NMNAT2 expression level, enzyme activity and axonal transport is also important. Given the absolute requirement for NMNAT2 for axon survival and the enhanced level of protection provided by the soluble form of this enzyme³⁹, such modulators are likely to influence axon survival; however, very few of these factors have yet been identified^62,63.

Linking degeneration to disease

Animal models of disease.

The notion that Wallerian degeneration mechanisms are important in disease dates back to the work of Waller¹ (FIG. 2) but could only be tested after mutant Wld^S mice were discovered. Axon degeneration in many animal models of disease can be alleviated by the presence of WLD^S (as previously summarized³), which has prompted a number of studies investigating the effects of manipulating the Wallerian degeneration pathway in other ways in animal models of disease.

Local translation

The synthesis of proteins directly within axons using mRNAs delivered by axonal transport.

Several studies have attempted to prevent distal axonal degeneration in mouse models of peripheral neuropathy by Sarm1 deletion. Peripheral neuropathy caused by the commonly used chemotherapy drugs vincristine or bortezomib was prevented in Sarm1^−/− mice^64,65. Similarly, peripheral neuropathy was prevented in Sarm1^−/− mice given either paclitaxel, another common chemotherapy drug that causes neuropathy, or a high-fat diet (HFD) (a model of metabolic syndrome that results in neuropathy)⁶⁶. In each study, the primary end point was the loss of intraepidermal nerve fibres, the distal endings of unmyelinated sensory axons that tend to degenerate first in most axonal peripheral neuropathies. Thus, it is possible that the distal axonal degeneration seen in these toxic and metabolic syndromes involves molecular mechanisms that are shared with those of Wallerian degeneration.

What about the other types of axonal injury seen in human diseases? For example, traumatic brain injury (TBI) causes stretching and shearing of axons. One would expect that axons in Wld^S and Sarm1^−/− mice would be protected from axon degeneration following TBI, as the mechanism may be similar to traumatic axotomy. Indeed, in two separate studies, Sarm1^−/− mice showed decreased axonal damage in TBI models^67,68, with functional preservation also observed in the milder model⁶⁷. Furthermore, blocking NAMPT with FK866 partially phenocopied Sarm1^−/−, mirroring findings in peripheral nerves⁶⁸. Obviously, traumatic injuries pose a therapeutic challenge as any clinical intervention has to happen after injury. It is unclear how effective a delayed blockade of SARM1 will be in these models; nevertheless, interventions such as the application of FK866, cleavage of SARM1, or transduction of NMNAT1 protein have been found to be effective several hours after injury in animal models^22,25,69. Combined with the observation that delayed overexpression of WLD^S in axons up to 4–5 h after axotomy can prevent Wallerian degeneration⁷⁰, these results suggest that intervention early after TBI could be feasible.

Secondary or primary progressive multiple sclerosis, in which gradual axonal degeneration is important in the progressive loss of neurological function, may be another potential therapeutic target for Wallerian degeneration-blocking drugs. Axonal injury has been shown to be present in several rodent models of multiple sclerosis, with studies showing that behavioural deficits and axonal loss are attenuated in an experimental autoimmune encephalitis (EAE) model of multiple sclerosis in Wld^S mice^71,72. No studies have been published to date investigating the effects of EAE in Sarm1^−/− mice.

Manipulation of the Wallerian degeneration pathway through treatment with nicotinamide, a precursor of NAD, has been shown to be effective in a genetic mouse model of glaucoma in which there is an age-associated decline in retinal NAD levels and axonal degeneration of retinal ganglion cells^73,74. Furthermore, the same approach has been used successfully in murine models of diabetic peripheral neuropathy⁷⁵. Dietary supplementation of nicotinamide riboside (NR) prevented neuropathy both in a mouse model of HFD-induced metabolic syndrome and in mice that received both HFD and low-dose streptozotocin to model type 2 diabetes⁷⁵. However, in these studies, it is unclear whether NR truly acts by blocking the Wallerian degeneration pathway as no mechanistic studies were carried out. Furthermore, the dose required for effect in these models is much higher than equivalent doses feasible in humans: the maximum tested dose of nicotinamide in humans is 1,000 mg twice a day, which showed no efficacy against several metabolomic parameters in obese men despite raising blood NAD levels⁷⁶. Perhaps combining NR or nicotinic acid riboside with the manipulation of enzymes in the NAD synthesis pathway (using NAMPT inhibitors, for example) may be more effective, as observed in a recent vincristine-induced neuropathy model in murine primary cultures³⁰ (FIG. 4). Similarly, in models of increased intraocular pressure glaucoma, both axons^77,78 and NAD levels⁷³ are preserved in Wld^S mice and, when supplemented with additional nicotinamide, axon protection is further increased. A similar treatment also further abrogated the disease phenotype and axonal loss in EAE in mice⁷¹.

The molecular mechanism by which Sarm1 deletion prevents distal axonal degeneration is likely to be cell-autonomous within axons, but the role of non-neuronal cells in peripheral neuropathies needs to be considered. Examples of sites of non-neuronal actions of SARM1 include macrophages⁷⁹ and in D. melanogaster, glia⁸⁰. Overexpressed SARM1 was found to be present in both the cytosol and the mitochondrial matrix of human embryonic kidney cells and endogenous SARM1 was associated with the matrix proapoptotic Nod-like receptor, NLRX1 (REF.⁵⁷). Although apoptosis in non-neuronal cells was dependent on both SARM1 and NLRX1, axonal degeneration in murine primary neurons induced by vinblastine was dependent only on SARM1, indicating perhaps a different role for SARM1 in non-neuronal cells.

Despite these promising results, the findings of animal studies suggest that manipulating SARM1 or other components of the Wallerian degeneration pathway may not prevent all axonal degeneration. Although one of the earliest events in a mouse model of amyotrophic lateral sclerosis (ALS) driven by mutations in superoxide dismutase (SOD1) is distal axonal degeneration and denervation at the neuromuscular junctions⁸¹, when the mutant Sod1 mice were crossed to Wld^S mice, there was minimal impact on disease onset or survival⁸². Similarly, when Sod1^G93A mutant mice were bred with Sarm1^−/− mice, there was no rescue⁸³. Thus, neurodegeneration in this particular model of ALS is unlikely to depend on the molecular players involved in Wallerian degeneration⁸², although there is evidence that Wallerian degeneration influences axon survival in a TDP-43 model of ALS⁸⁴.

The role that SARM1 plays in axon degeneration versus neuronal cell death has also been shown to vary significantly across cell populations and disease models. For example, in an optic nerve crush-induced retinal ganglion cell (RGC) death model of glaucoma, Sarm1^−/− mice showed preservation of distal optic nerve axons but no prevention of DLK–JNK pathway activation in the somanor of RGC death⁸⁵. Similarly, in a prion disease model, Sarm1^−/− mice had an exacerbated neuronal death upon infection, independent of any role of SARM1 in immune cells⁸⁶. This observation mirrors the findings of the initial study that described the development of a Sarm1^−/− mouse line, in which infection with West Nile virus led to increased neuronal death in Sarm1^−/− mice due to impairment of microglial activation⁸⁷. Finally, the expression of mutant proteins that cause ALS in Caenorhabditis elegans motor neurons, induces an innate immune response that propagates neuronal death⁸⁸; this process requires the SARM1 orthologue. Therefore, these studies point to a potentially complex role for SARM1 in neuronal and non-neuronal cells and their interactions in disease states.

Mechanistic basis of disease-modifying effects.

As outlined above, axotomy studies have revealed Wallerian degeneration to be a pathway that can be activated in multiple ways, the most specific of which is the removal of just one protein, NMNAT2. Under these circumstances, SARM1 is absolutely required for degeneration as axons are rescued permanently in Nmnat2::Sarm1 double null mutant mice². Less specific lesions, such as axotomy or general axonal transport impairment, prevent the delivery of all cargoes from the soma. Eventually, the levels of other proteins must fall below the threshold for survival, which we believe may explain why rescue through Sarm1 knockout is only temporary in these models.

With this in mind, there are several scenarios in which blocking Wallerian degeneration in the clinic is likely to be particularly effective. One is in disorders caused or exacerbated by specific disruption of NMNAT2, for example, those driven by its genetic mutation^89,90 or inhibition⁹¹. Another is in disorders in which there is an ongoing modest impairment of the delivery of many cargoes to distal axons but in which only NMNAT2 falls below the threshold for survival; this could include age-related disorders of long or highly branched axons, where the age-related decline in axonal transport⁶² and disease-specific mechanisms together impair the delivery of cargoes to axon terminals. A third scenario is in disorders in which there is a temporary impairment of transport or synthesis that is resolved before the loss of other more stable proteins becomes limiting. For example, the treatment of DRG primary cultures with vincristine for 24 h, followed by washout, causes the degeneration of wild-type neurites from which WLD^S provides permanent protection⁹². A similar outcome was seen when protein synthesis was transiently blocked and then restored in superior cervical ganglion neurons¹². The treatment of chemotherapy-induced peripheral neuropathy (CIPN) is an obvious application of this knowledge, as chemotherapy is temporary, the timing predictable, and rescue has been very widely replicated and extended to in vivo studies^{9,42,64–66,93}.

There are many other ways in which axons may be depleted of functional NMNAT2 in disease, which we suggest may explain why blocking Wallerian degeneration alleviates symptoms in some disease models that are not obviously caused by axonal transport impairment. Axonal NMNAT2 activity is likely to be lowered if there is impaired synthesis at the transcriptional, post-transcriptional or translational level, or by reduced protein stability, enzyme inhibition or posttranslational modifications that influence any of the above. NMNAT2 transcription is regulated in part by cAMP response element-binding protein (CREB)⁶³ and Ras-responsive element-binding protein 1 (RREB1; the mammalian orthologue of D. melanogaster zinc transcription factor Pebbled) is another candidate regulator of NMNAT2 transcription given its influence on Wallerian degeneration⁹⁴. There is huge variation between individuals in NMNAT2 mRNA expression in human post mortem brains, suggesting that there are multiple transcriptional and epigenetic regulators of this protein⁹⁵. Acute protein synthesis blockade causes axon degeneration through the Wallerian pathway¹², suggesting that disorders of protein synthesis may activate it. Normal ageing lowers NMNAT2 axonal transport by 40–50%, potentially contributing to axon loss in age-related diseases⁶². Depletion of stathmin 2 (SCG10), a protein that co-migrates with NMNAT2 in axons and shares several properties with it¹⁴, potentiates Wallerian degeneration and is frequently seen in sporadic ALS^96,97, raising the prospect of a role in that disease. Finally, environmental inhibitors of NMNAT2 enzyme activity⁹¹ are likely to activate a Wallerian mechanism and have been associated with peripheral neuropathy⁹⁸. These many influences on axonal NMNAT2 help to explain how diverse primary causes could activate a common Wallerian degeneration response⁹⁹.

Stathmin 2

A protein that, like NMNAT2, is palmitoylated, has a short half-life and negatively regulates Wallerian degeneration. Its loss is insufficient to activate Wallerian degeneration but accelerates it after injury and its overexpression delays degeneration. The same protein is depleted in many induced pluripotent stem cell-derived motor neurons from sporadic amyotrophic lateral sclerosis patients.

Wallerian degeneration in human disease.

Identifying the pathogenic or disease-modifying effects of Wallerian-regulating proteins directly in human disorders promises to be a powerful indicator of where best to apply Wallerian-blocking drugs (TABLE 1). Animal studies are extremely useful but also have unavoidable caveats, making it essential to complement them with clinical data. For example, EAE is a very limited model of human MS and extrapolating directly from effectiveness in this model to the human disease is risky. Conversely, the failure of WLD^S, or SARM1 ablation, to rescue SOD1 ALS mouse models by no means rules out the potential for therapy in other inherited forms of human ALS or in the much more prevalent sporadic forms.

Although no major neurological disease has so far been associated with mutations in genes involved in Wallerian degeneration, levels of NMNAT2 mRNA and protein are reduced in brains of individuals with Alzheimer disease⁹⁵. Whether this is a cause or consequence of axonal and neuronal degeneration in Alzheimer disease remains unknown. Nevertheless, two recent studies have identified potentially pathogenic mutations in NMNAT2 in two rare neurological conditions^89,90. The first of these describes two sisters who developed childhood onset peripheral neuropathy and erythromelalgia and were found to be homozygous for a T94M mutation in the NMNAT2 gene, likely indicating an autosomal recessive pattern of inheritance⁸⁹. The T94M mutation confers partial loss of function for NMNAT2 (full loss of function of NMNAT2 is likely to be lethal). A companion report describes two fetuses with severe fetal akinesia deformation sequence and multiple developmental neurological abnormalities leading to stillbirth. These two siblings carried biallelic loss-of-function mutations in the NMNAT2 gene, with one (NMNAT2^R232Q) disrupting both NAD⁺ synthase and chaperone activities and the other (NMNAT2^Q135Pfs*44) greatly lowering protein expression. Each variant NMNAT2 was unable to prevent Wallerian degeneration when overexpressed in neurons⁹⁰.

In another study, a 72-year-old patient with a 2-year history of progressive brainstem atrophy was shown to have a heterozygous mutation in NMNAT2 that had the potential to result in haploinsufficiency, although no functional analysis was presented¹⁰⁰. Given the normalcy of the heterozygous parents and siblings of the individuals carrying biallelic loss-of-function mutations of NMNAT2 (REFS^89,90), it is unclear whether NMNAT2 haploinsufficiency necessarily leads to axonopathies and which other environmental and genetic factors influence this. Furthermore, in the report on the patient with progressive brainstem atrophy, whole exome sequencing identified other potential candidate genes¹⁰⁰.

Loss-of-function mutations of SARM1 would be expected to be protective, meaning that exome sequencing of disease cases will not find these mutations. However, evidence that SARM1 loss of function reduces disease risk or severity in humans would indicate involvement of the Wallerian degeneration mechanism in the disease. An ideal cohort in which to evaluate this possibility would be patients undergoing chemotherapy, assessing whether those who do not develop peripheral neuropathy despite high doses of chemotherapy carry more loss-of-function mutations in SARM1.

A genome-wide association study (GWAS) showed linkage to the SARM1 locus in sporadic ALS¹⁰¹, an observation that was later replicated in a separate GWAS study¹⁰². Although the authors in the second GWAS suggested that a cis-expression quantitative trait loci (eQTL) effect on Polymerase delta-interacting protein 2 (POLDIP2) might be the driver of linkage at this locus, a role for SARM1 has not been ruled out. For example, SARM1 expression might respond in a genetically determined way to environmental ALS risk factors that are not represented in the expression quantitative trait loci data. Two recent studies provide a further link between Wallerian degeneration and ALS. ALS-causing mutations in TDP43 reduce levels of the JNK substrate SCG10 in patient fibroblast-derived motor neurons and this is associated with reduced axonal growth^96,97. This phenotype could be rescued by overexpression of SCG10 in human induced pluripotent stem cell-derived motor neurons. Previously, it had been shown that axotomy in mouse DRG neurons leads to a rapid reduction in SCG10 levels and that axonal degeneration could be delayed by its overexpression, indicating that SCG10 is a moderately strong modifier of Wallerian degeneration¹⁰³. These observations do not imply that SCG10 depletion by itself is sufficient to cause axonal degeneration but that, when SCG10 is lowered as part of another pathogenic mechanism, Wallerian-like axon degeneration becomes more likely. Moreover, upregulation of SCG10 levels may be sufficient to protect axons from degeneration.

Following the discovery of NMNAT2 loss-of-function mutations in rare disorders^89,90, it will be important to see what other pathogenic mutations emerge in these, or related, disorders. The human population exome and genome sequence database, the Genome Aggregation Database (gnomAD)¹⁰⁴, compares the observed and expected frequencies of nonsense, frameshift and splicing mutations to estimate a ‘loss-of-function intolerance’ score, likely related to pathogenicity. Within this database, NMNAT2 has a 0.99 probability of loss-of-function intolerance, one of the highest scores, suggesting that many heterozygous alleles can also cause lethality or disease, at least in combination with other genes and/or environmental factors. SARM1 gain of function (for example, resulting in a mutant that is overexpressed, more stable or more easily activated) would also be expected to cause or increase disease risk.

Wallerian gene variants are likely to have wider clinical significance through disease-modifying effects: that is, by influencing risk and severity in both common and rare disorders. Evidence is emerging of a spectrum of axon vulnerability in the human population (FIG. 5) caused by both coding and gene expression variation of Wallerian pathway genes^89,90,95. Data from mouse experiments already indicate that gene expression alterations have disease-modifying effects¹⁰⁵. With reports that NMNAT2 expression level varies widely in post mortem brains⁹⁵, it will be important to determine the consequences of this variation for disease and which genetic and environmental factors bring this about, especially as low Nmnat2 expression is associated with sensory and motor deficiencies in mice¹⁰⁵. Similarly, increased SARM1 expression (for example, in viral infections^51,52,106 or following prolonged low doses of rotenone¹⁰⁷) is likely to increase disease risk. Interestingly, rotenone also causes SARM1-dependent axon degeneration in vitro¹⁰⁸. The effect of other pesticides and environmental toxins on the pathway⁹¹ is another important area for study.

Fig. 5 |. An axon vulnerability spectrum in humans and mice.

The figure shows the range of known axon survival phenotypes in mice and proposed analogous phenotypes in the human population for which evidence is growing. The line graph depicts the likely prevalence in the human population and the gradients underneath depict protein and enzyme activity levels of nicotinamide mononucleotide adenylyltransferase 2 (NMNAT2) and sterile-α and Toll/interleukin 1 receptor (TIR) motif containing protein 1 (SARM1) in axons. The pink and blue boxes show five distinct phenotypes in humans and mice, respectively, all of which have been extensively characterized in mice^6,13,105. In humans, the first two phenotypes (non-viable and pathogenic) have been linked to NMNAT2 mutation^89,90. Causes of the proposed disease-modifying effects indicated towards the right side, corresponding to lower axon vulnerability, may include (1) SARM1 splicing alleles that are present in the Genome Aggregation Database (gnomAD), which are likely to disrupt the function of this pro-degeneration protein and (2) higher than average expression levels of NMNAT2⁹⁵, which are likely to result in gain of function of this pro-survival protein.

Factors increasing axonal demand for NMNAT2 or reducing the soma’s ability to supply it are another risk; for example, ageing lowers NMNAT2 axonal transport⁶² and increases motor unit size (the number of muscle fibres that each motor neuron innervates) by up to 2.4-fold¹⁰⁹, as surviving axons sprout to innervate neuromuscular junctions vacated by degenerating axons. Combined with reduced axonal transport, this is likely to create a ‘double hit’ on the ability to deliver sufficient NMNAT2 to keep distal axons alive. Thus, even minor impairment of the motor neuron soma could underlie the onset of a dying-back disorder such as ALS or ageing associated neuromuscular junction denervation¹¹⁰. Nigrostriatal neurons have even larger axonal arbours than motor neurons, with a combined length for a single neuron estimated at up to 4.5 m (REFS^111,112). Age-related compensatory sprouting in this pathway, similar to that seen following a partial lesion¹¹³, would put a huge strain on the ability of surviving neurons to deliver critical cargoes such as NMNAT2. Thus, it will be important to determine whether such a mechanism could contribute to the dying-back observed in the striatum in Parkinson disease¹¹⁴.

In future studies, it will be important to directly test for Wallerian degeneration mechanisms in human disorders. In axonal transport disorders such as some forms of CIPN, it will be important to ask whether genetic variation in NMNAT2, SARM1 or other Wallerian-regulating proteins influences neuropathy risk. Carefully phenotyped patient populations are vital to firmly link neuropathy to a specific risk factor such as chemotherapy or diabetes; such cohorts exist¹¹⁵ but await careful genotyping. Genetic disruption of axonal transport is another important area for future research. For example, the mutation of tubulin folding cofactor E (TBCE) in early-onset encephalopathy with distal spinal muscular atrophy¹¹⁶ suggests the involvement of Wallerian degeneration as its mouse orthologue is disrupted in a model that was particularly strongly protected by Wld^S, progressive motor neuronopathy (pmn)¹¹⁷.

Criteria for involvement of Wallerian degeneration in disease.

As the field moves forward to clinical translation, it is important to establish strong criteria for the involvement of Wallerian degeneration in a particular disease (TABLE 2). For a long time, in animal models, the gold standard was evidence for significant alleviation of axon loss (and, ideally, symptoms) by Wld^S. To this we can now add alleviation of axon loss and symptoms through the deletion of Sarm1, which in some models is more effective than Wld^S2. Biochemical changes specifically associated with loss of NMNAT2 (such as changes in the NMN-to-NAD ratio²²) or activation of SARM1 (such as a large increase in cADPR levels³⁵) may play an increasing role in defining the presence of Wallerian degeneration, both in animal tissue and in human biopsy material. Meanwhile, significant disease associations with human genetic variants of Wallerian-regulating genes will become another strong indicator of its presence in human disorders. In these cases, it is expected that increased disease risk should correlate with NMNAT2 loss of function or SARM1 gain of function and that protective effects would be associated with alleles having the opposite effects. Disease modelling in human induced pluripotent stem cell lines will also help establish roles for Wallerian degeneration as this allows molecular pathways to be manipulated genetically or pharmacologically to test for rescue effects.

Table 2 |.

Criteria for defining the contribution of Wallerian degeneration to disease

Criterion	Established in (disease)	Limitations	Refs
Protection by WLDS, NMNATs or SARM1 ablation in animal or cell culture models	Many (listed in REF.3) as well as metabolic syndrome and/or type II diabetes, CIPN, TBI, nerve growth failure and ALS	Animal and cell culture models only partially represent corresponding human diseases	2,3,27,64–68,84
Causation or exacerbation by removal and/or depletion of NMNAT2 in animal or cell culture models	Nerve growth failure, small fibre neuropathy (especially thermosensitivity and late onset motor impairment) and CIPN	Animal and cell culture models only partially represent corresponding human diseases	11,13,105
Genetic or protein expression association in humans with NMNAT2, SARM1 or other Wallerian degeneration regulator	Polyneuropathy with erythromelalgia, FADS, brainstem degeneration, ALS and Alzheimer disease	Correlation, not causation, although some supported by functional evidence; GWAS link may reflect another nearby gene	89,90,95–97,100–102
Chemical markers of Wallerian degeneration (raised levels of NMN or cADPR or lowered levels of NAD) in animal or cell culture models	Glaucoma, CIPN, nerve outgrowth failure and type II diabetes	Correlation not causation; animal and cell culture models only partially represent the corresponding human diseases	21,27,73–75

ALS, amyotrophic lateral sclerosis; cADPR, cyclic ADP-ribose; CIPN, chemotherapy-induced peripheral neuropathy; FADS, fetal akinesia deformation sequence; GWAS, genome-wide association study; NMN, nicotinamide mononucleotide; NMNATs, nicotinamide mononucleotide adenylyltransferases; SARM1, sterile-α and Toll/interleukin 1 receptor (TIR) motif containing protein 1; TBI, traumatic brain injury; WLD^s, Wallerian degeneration slow.

Drug development.

As the Wallerian degeneration mechanism and its involvement in disease, including human disease, become clearer, it is timely to consider what will be needed to drive the application of Wallerian degeneration-blocking therapies in the clinic. Safe and effective drugs will need to be trialled in appropriate disorders. Where different mechanisms underlie the same or closely related disorders (for example, peripheral neuropathies), the focus needs to be on the most relevant subtype. Finally, biomarkers are needed to measure efficacy.

One attractive clinical target is CIPN, which allows intervention before significant axonal injury occurs. The possibility of pre-medicating patients with a potentially neuroprotective drug (such as a SARM1 inhibitor or an FK866-like drug) before they receive chemotherapy would also simplify clinical trial design. These studies would be relatively short, as most chemotherapy drugs cause peripheral neuropathy within 6 months of starting treatment. The presence of validated outcome measures for CIPN would also simplify trial design and the recent development of neurofilament light chain levels as a potential biomarker of axonal injury make it even simpler^118,119.

In terms of CNS diseases, TBI and primary and/or secondary progressive multiple sclerosis may be good initial targets. In TBI, one could potentially intervene immediately after an injury and contain the consequences of axonal injury. However, variability in the severity of injury and timely access to patients may make clinical trial design challenging. In multiple sclerosis, despite the development of effective immune-modulating therapies¹²⁰, there are no treatments that prevent secondary axonal degeneration. Given the observations in EAE in Wld^S mice⁷¹, it is reasonable to consider axon-protective strategies that modulate the Wallerian degeneration pathway as a therapeutic target for multiple sclerosis. Clinical trials in this area will, however, be challenging as they will require long-term treatment and monitoring. Nevertheless, the use of biomarkers, such as neurofilament light chain levels, may help.

Given that the preclinical data show the efficacy of targeting several different components of the Wallerian degeneration pathway, which one is likely to be the most suitable for drug development? One of the most attractive targets is SARM1 as this molecule plays a relatively downstream role in Wallerian degeneration, meaning that one drug may be useful in multiple diseases. However, it will be challenging to develop a SARM1 inhibitor that does not affect other NAD-utilizing enzymes in neurons and/or non-neuronal cells, in order to avoid unwanted side effects. An inhibitor that targets the activation of SARM1 by preventing TIR domain dimerization may be a more suitable target. Recent observations that SARM1 forms an octomeric structure and that inhibition of oligomerization is able to prevent the cell death-promoting activity of SARM1, favour this approach^48,49. Another attractive approach is to develop a dominant negative SARM1 gene therapy¹²¹ as this avoids the side effects of inhibiting other NADases; however, the delivery of such a therapy to appropriate neuronal populations remains a big challenge. Finally, anti-sense oligonucleotides have blossomed into FDA-approved therapies for rare neurodegenerative diseases such as amyloid neuropathy and spinal muscular atrophy¹²² and could therefore offer a more suitable delivery path.

Other therapeutic interventions may target other components of the Wallerian pathway. For example, drugs may target NMNAT2 expression level⁶¹ by preventing its degradation or by using a combined approach to prevent NMN synthesis (using FK866, for example) and providing an alternative NAD⁺ precursor such as nicotinic acid riboside³⁰. Alternatively, drugs targeting pathway components that play a more prominent role in specific neuronal populations and disease states (such as DLK and/or LZK in RGCs in glaucoma) may lead to the development of disease-specific therapies.

Concluding remarks

The recent pace of discoveries in Wallerian degeneration research, the choice of suitable drug targets and the attractiveness of CIPN as a primary application, leaves us in little doubt that the next 5–10 years will see significant advances in translation in this field, as well as further developments in the mechanistic understanding. Genetics — whether in mice, flies or humans — will continue to provide the major advances that it has done for three decades. The biggest question is perhaps whether Wallerian-blocking drugs will be limited to rare, inherited disorders, where they are likely to be highly effective, or whether they will find extremely wide application across the neurodegenerative disease treatment spectrum. The prospects for the next three decades appear even more exciting than the last three.