Promoting regeneration while blocking cell death preserves motor neuron function in a model of ALS

Abstract

Amyotrophic lateral sclerosis (ALS) is a devastating and fatal neurodegenerative disease of motor neurons with very few treatment options. We had previously found that motor neuron degeneration in a mouse model of ALS can be delayed by deleting the axon damage sensor MAP3K12 or dual leucine zipper kinase (DLK). However, DLK is also involved in axon regeneration, prompting us to ask whether combining DLK deletion with a way to promote axon regeneration would result in greater motor neuron protection. To achieve this, we used a mouse line that constitutively expresses ATF3, a master regulator of regeneration in neurons. Although there is precedence for each individual strategy in the SOD1^G93A mouse model of ALS, these have not previously been combined. By several lines of evidence including motor neuron electrophysiology, histology and behaviour, we observed a powerful synergy when combining DLK deletion with ATF3 expression. The combinatorial strategy resulted in significant protection of motor neurons with fewer undergoing cell death, reduced axon degeneration and preservation of motor function and connectivity to muscle. This study provides a demonstration of the power of combinatorial therapy to treat neurodegenerative disease.

Introduction

Amyotrophic lateral sclerosis (ALS) is a fatal neurodegenerative disease typically characterized by progressive motor symptoms with respiratory compromise late in the disease course. Although ALS is an uncommon disease, with a prevalence of 5.2 per 100 000 in the US, it represents a significant cause of morbidity to patients and their families.¹ There has been only one drug approved for ALS by the FDA in the last 22 years, bringing the total number of approved medications for this disease to just two,² and these have limited efficacy. Given the severity of the disease and lack of effective treatment options, further research is needed. One therapeutic approach for ALS that is likely to be the path forward for future treatments is the combination of therapies, because any one treatment alone is unlikely to reverse the multiple pathologies associated with disease.^3,4

Although the majority of ALS cases occur sporadically, 5–10% of ALS cases are familial (fALS), caused by specific gene mutations. Mutations in the gene coding for Cu/Zn superoxide dismutase type-1 (SOD1) are responsible for roughly 20% of fALS cases.⁵ The G93A mutation in SOD1 is one of the best studied fALS mutations and the first to be investigated as a transgenic mouse model. In the SOD1^G93A transgenic mouse model, human mutant SOD1^G93A is overexpressed (with intact expression of endogenous mouse SOD1) and mice develop phenotypes similar to human ALS, including adult-onset motor neuron degeneration and progressive paralysis.⁶

In previous work, we showed that motor neuron degeneration in this mouse model could be delayed by deleting the axon damage sensor dual leucine zipper kinase (DLK).⁷ DLK is a highly conserved multifunctional signalling molecule. During development, it plays critical roles in establishing neuronal circuitry including selective neuronal pruning,⁸ axonal growth and cell migration.⁹ In the adult, DLK is a sensor of a wide range of axonal injuries occurring in trauma, neurodegenerative diseases and metabolic or toxic insults, and a key regulator of the fate of injured neurons.

DLK damage sensing can drive neuronal degeneration, including axon and synapse loss and neuronal death.¹⁰ Accordingly, DLK deletion can promote neuron survival, including that of spinal motor neurons in the SOD1^G93A mouse model of ALS.^7,11,12 Conversely, yet still consistent with its role as a damage sensor, DLK is also required for regeneration of axotomized sensory and motor neurons in worms and flies^13–16 and for neuromuscular junction (NMJ) reinnervation by motor neurons after a sciatic nerve crush in mice.¹⁷ One mechanism by which DLK promotes regeneration is through a transcriptional switch downstream of DLK kinase signalling.^18–20

This DLK-dependent transcriptional switch is part of a larger transcriptomic transformation undergone by injured peripheral neurons on their path to recovery. These neurons downregulate genes associated with somatosensory function and upregulate regeneration-associated genes (RAGs).^18,21,22 One critical RAG that is partly DLK-dependent is ATF3 or activating transcription factor 3,^11,19 which is required for a majority of transcriptional changes.²¹Atf3 null mice exhibit reduced axon regeneration after facial nerve transection²³ and ATF3 overexpression promotes axon regeneration in sensory neurons.²⁴ Interestingly, constitutive ATF3 expression in neurons was reported to improve physiological outcomes in the SOD1^G93A mouse.²⁵

Because DLK signalling promotes regeneration and partly mediates ATF3 expression in injured neurons, blocking DLK to prevent neuronal death likely also reduces regeneration inherent in the axon injury response. We hypothesized that DLK inhibition to treat neurodegeneration would be improved by combining it with ATF3 expression—this should promote regeneration while blocking neuronal death. We tested this in the SOD1^G93A mouse by conditionally deleting DLK and constitutively expressing neuronal ATF3. We compared this combination to each individual manipulation for a wide variety of measures including motor neuron electrophysiology, histology, behaviour and survival. Across all end points, the combination more powerfully preserved motor neuron function than either alone, demonstrating the limitations of simply blocking neuronal death or just promoting regeneration and the power of combinatorial strategies to preserve neuronal function.

Materials and methods

Mice

All animal care and experimental procedures were performed in accordance with animal study proposals approved by the National Institute of Child Health and Human Disease Animal Care and Use Committee, animal protocol numbers ASP-17-003 and 20-003. Map3k12^fl/fl homozygous mice,^7,26 obtained by MTA from Genentech, were bred to a heterozygous ROSA-CAG-Cre^ER transgenic line (JAX #004453) to produce mice with tamoxifen-dependent inducible DLK deletion (Map3k12^fl/fl; Cre^ER+/wt mice, which we refer to as ‘DLK^cKO’). The heterozygous presence of the Cre^ER allele determined DLK deletion as all mice in this study were Map3k12^fl/fl and treated with tamoxifen (more details below). The DLK^cKO mice were crossed to a hemizygous transgenic line overexpressing ATF3 under the Thy-1.2 promoter (Dr Clifford Woolf lab, cryopreserved line JAX #910352 transgenic line #4).²⁷ In this study, hemizygous ATF3 transgene carriers are referred to as ‘ATF3^OE’. In parallel, hemizygous SOD1^G93A males (JAX #004435; congenic with C57Bl6/J) were crossed to Map3kK12^fl/fl mice whose progeny were also interbred. The resulting Map3k12^fl/fl; SOD1^G93A males were crossed to Map3k12^fl/fl; CAG-Cre^ER+/wt; ATF3^OE females, resulting in our study cohort. Refer to Table 1 for a description of the genotype groups in this study. Similar numbers of male and female mice were used in all parts of the study. SOD1 copy number was verified by droplet digital PCR (ddPCR) for animals in this study and as standard practice for all male SOD1-positive breeders in the colony. Genotype groups were balanced for sex and litter. Cage arrangement was randomized for females and to the extent possible for males. Sample size was based on analogous study design in Le Pichon et al.⁷

Husbandry/animal care/tamoxifen

The animals were kept in their parent cages until 3 weeks of age, after which they were weaned into cages of up to five mice of the same sex. Males with persistent fighting were singly housed. Regular diet (NIH-07 Rodent Chow) and soft Dietgel (ClearH₂O product code: 72-07-5022X) were provided for the first 3 days post-weaning, with regular diet only for the following 4 days. From 4 to 7 weeks of age, all mice were administered a diet containing 40 mg/kg tamoxifen (Invigo #TD130860). Mice were fed a regular diet from 7 weeks onwards. A handful of food pellets was placed on the floor of the cage starting at 16 weeks and soft DietGel was given to supplement the pellets beginning at 18 weeks of age. Water was available ad libidum. A 12-h light/dark cycle was maintained, with lights on from 6:00 a.m. to 5:59 p.m.

Data collection and analysis

All behavioural, electrophysiological and histological tests and analyses were performed by experimenters who were blind to mouse genotype. For all measurements, we analysed males and females separately, but did not detect significant sex differences; therefore, all the data shown in this study combine both sexes.

SOD1 copy number verification, ddPCR

To ensure all mice in the study had comparable SOD1^G93A transgene copy number, as loss of transgene copies can reduce severity of disease, ddPCR for human SOD1^G93A was performed on tail DNA from all mice in the study. Any mouse with a SOD1^G93A copy number significantly lower than the average was examined for evidence of reduced disease severity, such as longer lifespan or improved histopathology, as together these would be criteria for exclusion. We did not detect any such outliers. Tail clips were collected from mice at weaning and processed with DNeasy Blood & Tissue Kit (Qiagen #69506) according to the manufacturer’s instructions. Sample concentrations were determined using NanoDrop2000 (Thermo Scientific ND-2000) and diluted in DNase free water to obtain a working concentration of 1 ng/μl. SOD1^G93A copy number in these DNA samples was measured using Digital Droplet PCR (QX200 AutoDG System, Bio-Rad). Diluted samples were then loaded into a 96-well plate with each well containing a diluted sample, Super Mix for Probes (Bio-Rad #1863023), SOD1^G93A probe (Bio-Rad #12001950, UniqueAssayID: qHsaCIP0026883), RPP30 control probe (Bio-Rad #10031228, UniqueAssayID: qMmCEP0054985), HaeIII restriction enzyme and DNase free water. The plate was sealed before vortexing, then loaded into the automated droplet generator (QX200) along with a cool block and pipette tips. After droplets were generated, the plate was immediately resealed and loaded in a PCR cycler (Bio-Rad T100 Thermal Cycler). While the PCR cycler was running, the plate layout was programmed in QuantaSoft. Once the PCR was finished, the plate was loaded into the plate reader. Wells with 9000 or more reads were kept for analysis while those with fewer reads were re-run. In the QuantaSoft software ‘2D Amplitude’ and ‘Copy Number’ were selected for analysis. Results were saved and exported as a CSV.

ALS Therapy Development Institute neurological exam

Mice in the survival cohort were assessed once a week for their hindlimb function, gait and righting latency, as per the ALS Therapy Development Institute (TDI) protocol for SOD1^G93A mice.²⁸ They were assessed once a week on these measures from 7 to 17 weeks of age, and then three times a week from 18 weeks of age until euthanasia. Control mice were separated from their cage mates after 18 weeks of age and were no longer tested. In addition to the ALS TDI exam, animals in the survival cohort were tested for forelimb grip strength and inverted grid latency. The battery of tests is described below in the order in which they were performed.

Forelimb grip strength

All mice in the survival cohort competed a forelimb grip strength test once a week from 7 to 18 weeks of age, using Bioseb’s Grip Strength Test and software Bio-Cis version 1.4.0.54. Their grip strength was tested in the order they completed their ALS TDI neurological exams so that each had approximately equal recovery time. The mouse was held close to the base of its tail, positioned so that only its forelimbs were grasping the wire grid, and pulled down along the grid angle until its grip was released. The test was run five times in a row per animal and the final score was an average of the five trials.

Inverted grid latency

Each mouse in the survival cohort was placed on a large wire grid approximately 0.75 m above and parallel to the testing surface. The grid was then rotated 180 degrees, forcing the mouse to hang on to the grid, upside down, with its four paws. The mouse was timed for how long it could hold on to the grid; if it stayed on the grid for 60 s, it completed the test with the maximum score of 60 s. If the mouse failed to hold on for 60 s, it was given up to two more trials to make it to 60 s. All of the latencies to fall were recorded.

Hindlimb function

Each mouse in the survival cohort completed a tail suspension test in which it was held by the base of its tail approximately 3 cm over the wire top of its home cage for an (inconsistent) amount of time. The tail suspension test was repeated three times in a row, after which its left and right hindlimbs were scored individually. A score of ‘0’ was assigned if the hindlimb extended fully in a normal splay. A score of ‘1’ was given if the hindlimb showed an abnormal splay, clasping, slight collapse towards lateral midline or exhibited a tremor even if the hindlimb was fully extended. A score of ‘2’ was recorded if the hindlimb showed moderate collapse towards the lateral midline and partial paralysis, and a score of ‘3’ if the hindlimb presented with full paralysis.

Righting latency

If a mouse in the survival cohort displayed paresis in the hindlimb or gait exams, it was placed on its left side on a paper mat, and its latency to right itself was recorded. The test was repeated for the right side. The mouse was considered righted when it returned to standing on all four paws.

Lifespan

Animals in the survival cohort met euthanasia criteria for this study if they were unable to right themselves (see ‘Righting Latency’ section) within 15 s on either side. Righting latency was checked every other day until an animal was no longer able to right itself instantaneously, after which their righting latency was checked daily for euthanasia qualification.

Compound muscle action potential: tibialis anterior muscle

Each mouse in a cohort of mice separate from the survival cohort was anaesthetized at 2.0% isoflurane and maintained at 1.5% isoflurane for the duration of the recording. During the experiment, the mice were kept on a heating pad and Altalube hydrating ointment (59390-198-50) was placed over the animals’ eyes. The left hind limb and surrounding area were shaved to remove excess fur and a total of five electrodes (MLA1213) were placed subcutaneously in the hind limb and surrounding area. The stimulating anode was placed near the sciatic notch and the stimulating cathode was placed on the dorsolateral side of the knee. The recording electrode was placed in the middle of the tibialis anterior (TA) muscle and a reference electrode was placed near the tendon of the TA muscle. A ground electrode was placed at the base of the animals’ tail. Recording was performed using LabChart software and took place for 60 s with a 10 V 2 mA stimulus pulse every 3 s generated by a Stimulator HC (FE155) connected to an Animal Bio Amp (FE126). Each animal received 18–20 stimulation pulses and 18 consecutive traces were averaged to produce the final amplitude for each mouse at each time point. Amplitudes analysed in this study are the total amplitude from maximum response peak to minimum response peak.

Tissue preparation

Animals were anesthetized with intraperitoneal injections of 1.2% Avertin. Mice were then perfused, first with 7–10 ml 1× PBS followed by 7–10 ml 4% paraformaldehyde (PFA). TA muscle was post-fixed in 4% PFA for 20 min before transfer to 1× PBS + 0.01% sodium azide. Spinal cord and brain were post-fixed for 24 h in 4% PFA before transfer to 1× PBS + 0.01% sodium azide. For sciatic nerve, animals were anesthetized with intraperitoneal injections of 1.2% Avertin and perfused with 7–10 ml of 1× PBS followed by 4% PFA 2.5% glutaraldehyde. The entire animal was skinned and post-fixed in a 4% PFA 2.5% glutaraldehyde solution for 24 h. Sciatic nerves were then dissected out and placed in 4% PFA 2.5% glutaraldehyde until processing.

RNAscope

Spinal cord tissue was dissected as described above. The lumbar region was isolated and placed in a 30% sucrose solution overnight before being embedded in OCT compound (Tissue-Tek) then frozen on dry ice. Blocks were sectioned into 16-µm thick coronal slices onto positively charged slides using a Lecia CM3050 S Research Cryostat. Multiplexed in situ hybridization was performed according to the manufacturer’s instructions for fixed frozen sections (ACD: 323100, 323120), with the addition of an extended initial bake time of 60 min. Probe targets (Atf3, Chat, Sod1 and Tubb3) were visualized using Opal dyes 520, 570 and 690 (Akoya).

Slides were imaged using a Zeiss confocal LSM800. Images were stitched using Zeiss ZEN software (blue edition). ImageJ was used to generate maximum intensity projections and adjust brightness and contrast.

Spinal cord amino cupric silver stain per cent area quantifications

Amino cupric silver staining was performed by NeuroScience Associates (NSA) on 30 µm spinal cord sections. Image analysis on the resulting silver-stained sections was performed in house using Arivis 4D Vision software by an investigator blind to genotype. Five lumbar spinal cord sections across 3.6 mm of lumbar spinal cord were analysed per animal. A custom region of interest was first placed on each section to identify the ventral horn region. The images were then run through a denoising filter to remove background signal. Once filtered, an RGB threshold was applied to each image to identify positive amino cupric silver staining. A secondary filter then removed any signal detected outside the ventral horn region. The area of each positive signal that was completely within the custom region defining the ventral horn was then exported in an Excel sheet along with the total area of the custom drawn region. Positive signal areas were summed, divided by the total ventral horn region area and multiplied by 100 to obtain the total per cent area of the ventral horn region that was amino cupric silver stain positive in the tissue section. The five lumbar sections that were analysed per animal were then averaged to give the final ‘Ventral Horn % area silver stained’ score for each animal that was plotted in Fig. 5C.

Spinal cord Iba1 per cent area quantifications

Iba1 immunostaining was performed by NSA on 30-µm thick spinal cord sections. Image analysis on the immunostained sections was performed in house using Arivis 4D Vision software by an investigator blind to genotype. Briefly, five lumbar spinal cord sections across 3.6 mm of lumbar spinal cord were analysed per animal. Two circular regions of interested were added manually to the left and right ventral horn regions of each spinal cord image to identify the ventral horn region (Supplementary Fig. 7). The images were filtered to remove positive signal outside of these regions. An intensity filter was then applied to find positive Iba1 signal within each ventral horn region of interest. The areas of each positive signal detected within each ventral horn region were then exported into an Excel spreadsheet along with the area of each custom ventral horn region of interest. For each image, the areas of all positive signals were summed and divided by the total area of the ventral horn region (areas of the left and right circular custom regions), then multiplied by 100 to obtain a per cent area. The five sections analysed per animal were then averaged to produce one score per animal that is depicted in Fig. 5D.

Spinal cord motor neuron counts

Solochrome and choline acetyltransferase (ChAT) immunostaining were performed by NSA on 30 µm spinal cord sections. Motor neurons were hand-counted from five lumbar solochrome sections per mouse by an investigator who was blind to sample genotype; final counts were the total number of motor neurons from both the left and right ventral horn regions across the five sections analysed. There was 720 µm between each section counted, covering 3.6 mm of lumbar spinal cord from approximately L2 to L6. Due to inconsistency of ChAT staining to clearly reveal motor neuron cell bodies, this stain was instead used as a guide for the region in which to count prominently stained solochrome-stained cell bodies with characteristic motor neuron polygonal shape and diameters above 17 µm. This cell count may overrepresent the true number of motor neurons, because solochrome is not specific to motor neurons. Using the solochrome stain to quantify motor neuron number also helped avoid bias toward false negatives where ChAT levels are low (see Supplementary Fig. 1, Chat is downregulated in SOD1^G93A mice).

Neuromuscular junction quantifications

TA muscles were dissected out of 1× PBS/4% PFA perfused mice, post-fixed in 4% PFA for 20 min before transfer to 1× PBS + 0.01% sodium azide. Left TA muscles were shipped to Clarapath (Clarapath.com) in 1× PBS + 0.01% sodium azide. Clarapath produced 20-µm thick longitudinal sections through the muscles. Immunofluorescence staining was performed on these sections using rabbit anti-VACHT (Synaptic Systems #139 103) and TMR-α-bungarotoxin (ThermoFisher #T1175). Sections were imaged with wide-field fluorescence microscopy. The total number of TMR-α-bungarotoxin positive and VACHT positive counts per muscle were recorded along with the per cent VACHAT/TMR-α-bungarotoxin as a function of sample area.

Sciatic nerve axonal area quantifications

Fixed sciatic nerves (4%PFA, 2.5% glutaraldehyde) were delivered to the NICHD Microscopy and Imaging Core (MIC) for thin sectioning and myelin staining. Slides were collected from the MIC and imaged on a Zeiss Axiocam 506 colour camera. Brightfield images of myelin-stained cross-sections of mid-thigh sciatic nerve were loaded into ImageJ for initial editing by an investigator blind to genotype. For each sciatic nerve, four different regions of interest of equal area were selected randomly throughout the tibial nerve for analysis. These four regions of interest were then subject to ImageJ’s Enhance Local Contrast (CLAHE) function and saved as individual RGB tiff files. Each tiff was then loaded into Arivis Vision4D for axonal area quantifications. Briefly, each image was colour-inverted and subjected to a normalization function. Images were then run through a threshold segmenter that selected dark objects of a specified intensity. These segments were then filtered by area (smaller than 60 µm² and larger than 0.5 µm²) as well as roundness (greater than 0.3). All images were then briefly checked by eye to ensure that no non-axonal compartments were captured, and any missed axons were added by hand. Roughly 150–200 axons were counted per image and their corresponding projected areas were saved as Excel files. Axonal segments were then binned by size to determine the total numbers of small (>6 µm²), medium (6–12 µm²) and large (>12 µm²) axons in each image.

Statistical analysis

All statistical analyses were done using Graphpad Prism software v.9.3.1 except for repeated measures analysis in Fig. 1A, which was done using JMP version 16.2.0. All reported results in figures show individual data-points and/or mean ± standard error of the mean (SEM) error bars.

Data availability

Raw data from the study are available upon request.

Results

Combinatorial strategy to delete DLK and overexpress ATF3 preserves motor neuron connectivity and function until late-stage disease

Progressive motor neuron degeneration has been well described in the SOD1^G93A mouse model of ALS.²⁹ One of the best characterized early events in motor neuron degeneration is the loss of their synaptic terminals at the neuromuscular junction (NMJ).^30,31 At a later stage, death of motor neuron cell bodies can also be detected.^32,33 This progressive degeneration is thought to affect motor neurons in order of vulnerability, with the most sensitive being those motor neurons that form fast-firing motor units and innervate fast-twitch muscle fibres.^29,31 The loss of motor neuron–muscle connectivity can be measured by recording compound muscle action potentials (CMAPs) elicited by nerve stimulation.³⁴ This technique is advantageous in that it permits longitudinal in vivo recording of muscle contractility in the same animals over time.

To examine the combined effects of DLK deletion and ATF3 overexpression on motor neuron function in the SOD1^G93A mouse model of ALS, we generated and analysed mice from eight different genotype groups for combinations of the following transgenic alleles: SOD1^G93A, DLK flox, CAG-Cre^ER and ATF3 including all possible control groups (Table 1). The SOD1 mutation is carried on a transgene where SOD1^G93A is overexpressed (Supplementary Fig. 1A). Elimination of DLK is embryonic lethal; therefore, we used a Cre^ERT strategy to delete it in adults. All mice in the study were DLK^fl/fl, and conditional DLK deletion (DLK^cKO) was determined by the presence of the ROSA-CAG-Cre^ER allele, which enabled tamoxifen-inducible DLK deletion in a majority of cells, including skeletal motor neurons (Supplementary Fig. 1B and C). ATF3 was expressed under the Thy1 promoter in a subset of neurons including skeletal motor neurons (Supplementary Fig. 2). Expression of all transgenes in the study overlapped in a majority of skeletal motor neurons.

To assess the connectivity of motor neurons to muscle, we performed bimonthly longitudinal recordings of CMAPs in the TA muscle, a predominantly fast-firing muscle type that is considered vulnerable.^32,35 In the TA muscle of SOD1^G93A controls, CMAP amplitude was rapidly lost between 7 and 11 weeks of age, dropping to plateau slightly below 20 mV by 11 weeks of age, and in line with a previous report³⁴ (Fig. 1A, black curve).

Strikingly, SOD1^G93A mice with motor neurons in which DLK was deleted and ATF3 overexpressed (see Supplementary Figs 1 and 2 for transgene expression patterns) maintained CMAP amplitudes until very late into disease progression (Fig. 1A and Supplementary Fig. 1, purple). Although SOD1 mice carrying the ATF3 transgene (blue) also had significantly improved CMAP amplitudes, the effect size in the SOD1^G93A mice with the combinatorial manipulation (purple) was greater. These mice (purple) had significantly improved CMAP amplitudes compared to SOD1^G93A mice (black) at all time points beyond 7 weeks of age. Their amplitudes were similar to wild-type controls until 13 weeks of age, and remained around 40 mV before declining at 19 weeks of age.

Most interestingly, the combinatorial manipulation of DLK deletion with ATF3 overexpression was synergistic. CMAP amplitudes in SOD1^G93A mice lacking DLK were no different from control SOD1^G93A animals (Fig. 1A and Supplementary Fig. 1, red versus black). Although DLK deletion alone did not preserve CMAP amplitudes, its combination with ATF3 overexpression was significantly more protective than for ATF3 overexpression alone (Fig. 1A purple versus blue). SOD1^G93A mice with ATF3 overexpression alone exhibited a striking preservation of CMAP amplitudes (blue), but its combination with DLK deletion (purple) had a significantly stronger effect (15–17 weeks old, Fig. 1A, Table 2 and Supplementary Fig. 1). All four healthy control groups maintained CMAP amplitudes above 45 mV throughout the study (Fig. 1A).

To appreciate the histological and behavioural correlates of this preservation of CMAP amplitude by the combinatorial treatment, we examined histological end points in motor neurons and muscle, and performed motor phenotyping.

Histological and behavioural correlates of CMAP amplitude preservation in the combinatorial treatment

Delayed loss of neuromuscular junction innervation

NMJ innervation was evaluated in the TA muscle early in the disease course (9 weeks old), at a time where we detected a loss of CMAP amplitudes in the control SOD1^G93A mice and the SOD1^G93A mice lacking DLK but not in either SOD1^G93A group with ATF3 overexpression. The extent of NMJ innervation mirrored the difference observed by CMAP, with an average of 75% motor neuron synaptic connections at the muscle endplate being retained in SOD1^G93A mice with ATF3 expression and the combination of ATF3 overexpression with DLK deletion (P = 0.0018 and 0.0032, respectively, by one-way ANOVA; Fig. 1C), while SOD1^G93A controls and SOD1^G93A mice lacking DLK had lost 50% of their NMJ connections in TA muscle.

Delayed loss of hindlimb grip strength

To correlate the CMAP measurements with motor function, we measured hindlimb grip strength later in the disease course (15–19 weeks of age). Hindlimb grip strength in the SOD1^G93A mice with DLK deletion combined with ATF3 overexpression was significantly stronger than the other SOD1^G93A groups at 17 weeks (P < 0.01 by one-way ANOVA; Fig. 1D), a time in the disease course at which SOD1^G93A mice have essentially lost hindlimb muscle function.

Combinatorial strategy preserves large axon calibre

We hypothesized we would also detect a delay in axon degeneration, so we examined axon morphology in the sciatic nerve at 17 weeks (Fig. 2A). The tibial branch of the sciatic nerve was analysed bcause it innervates the TA muscle recorded from the CMAP experiment. We examined the area of each axon for this mixed population of motor and sensory neurons. SOD1^G93A mice with DLK deletion and those with the combined manipulations had a significantly higher proportion of large axons compared to the SOD1^G93A controls (Fig. 2B–E, P = 0.0370 and P = 0.0002, respectively, by one-way ANOVA), suggesting that the combinatorial effects of DLK deletion with ATF3 overexpression not only preserved NMJ terminals but also delayed the axon shrinkage that precedes motor neuron axon degeneration.

Combinatorial strategy preserves tibialis anterior muscle weight

As an additional correlate of neuromuscular health, we collected TA muscle weights at 19 weeks old, late in disease (Fig. 3A). Based on Seijffers et al.,²⁵ we expected the muscle weights of animals expressing the ATF3 transgene to be increased up to 2-fold. Because the ATF3 transgene is controlled by the Thy1 promoter with expression in neurons and not muscle,³⁶ this preservation of muscle weight is likely a direct result of enduring innervation by motor neurons.²⁵ Both groups of SOD1^G93A mice carrying the ATF3 transgene exhibited a 1.5-fold greater retention of muscle mass even at the late time point of 19 weeks old (Fig. 3B; average TA mass and P-values for groups compared to 18.88 mg for SOD1^G93A: SOD1^G93A; ATF3^OE = 28.40 mg, P = 0.0042, SOD1^G93A; ATF3^OE; DLK^cKO = 32.94 mg, P < 0.0001). TA muscle mass correlated well with CMAP recordings in TA muscle (Fig. 3C, R² = 0.55).

Deletion of DLK delays motor neuron death and ATF3 overexpression amplifies this effect

To determine whether protective effects of DLK deletion combined with ATF3 overexpression could also be detected in the spinal cord, we examined a separate cohort of 14-week-old mice. At this stage, SOD1^G93A mice are overtly symptomatic, and motor neuron cell body loss, neurite degeneration and neuroinflammation histologically apparent.

To quantify the total number of motor neurons in the lumbar spinal cord, we counted large cells in solochrome-stained sections in ventral regions defined by ChAT-positive staining in neighbouring sections (see the ‘Materials and methods’ section). The combination of DLK deletion with ATF3 overexpression significantly preserved motor neuron cell bodies in SOD1^G93A mice (Fig. 4), revealing a synergy between these manipulations. Highlighting the degree of rescue, the number of motor neurons in this group was comparable to healthy controls lacking the SOD1 transgene (P = 0.267 by Tukey’s multiple comparisons, Fig. 3C, white versus purple). Interestingly, ATF3 overexpression alone was not sufficient to preserve motor neurons (P = 0.9929, black versus blue), suggesting that the distal protection observed in this group is not via an increase of motor neuron survival. By contrast, SOD1^G93A mice lacking DLK had significantly more motor neurons than the group overexpressing ATF3 and the SOD1^G93A control group (P = 0.0056 and P = 0.0224, respectively, by Tukey’s multiple comparisons), despite this group lacking any distal protection. There were no differences in motor neuron numbers among the four healthy control genotype groups (Supplementary Fig. 4), showing that neither DLK^cKO, ATF3 overexpression nor their combination alters the total number of motor neurons (Supplementary Fig. 4C).

Combinatorial treatment reduces neuropathology in the spinal cord

To further assess neuropathology in the spinal cord, we stained the set of 14-week-old tissues with amino cupric silver as a marker of neurodegeneration and anti-Iba1 as a measure for microgliosis. The combination of ATF3 overexpression and DLK deletion significantly reduced the level of neurodegeneration as well as neuroinflammation in SOD1^G93A mice (P = 0.0109 and P = 0.0019, respectively, for combinatorial group compared to SOD1^G93A controls; Fig. 5). SOD1^G93A mice lacking DLK exhibited less neurodegeneration compared to SOD1^G93A mice that overexpressed ATF3 (P = 0.0012; Fig. 5A and C) indicating that blocking DLK was the main driver of neuroprotection. Here again, we found evidence of a synergy between manipulations since in the SOD1^G93A mice the ATF3 transgene alone had no effect, but its combination with DLK deletion achieved a significant degree of protection (P = 0.0026 blue versus purple). No differences in either neurodegeneration or inflammation were detected between the four healthy control groups (Supplementary Fig. 5).

Behavioural and survival assessment: modest effects of the combinatorial manipulation

Because expression of all the transgenes we examined in these mice overlapped in Chat-positive skeletal motor neurons (Supplemental Figs. 1 and 2), we hypothesized that larger effects might be observed for measures focused specifically on these cells compared to tissue- or organism-level measures. Nevertheless, we also assessed survival and symptomology over the lifespan of these animals. Mice were subject to neurological scoring (see the ‘Materials and methods’ section) and grip strength testing (Fig. 6).

None of the genetic manipulations in our study significantly increased the lifespan of SOD1^G93A transgenic mice (Fig. 6A). Both SOD1^G93A genotype groups with conditional DLK deletion (with or without ATF3 overexpression) had a slightly longer median survival (by 2.5 and 5 days, respectively) in comparison with SOD1^G93A controls (158 days; Fig. 6A). This trend was generally concordant with a previous study in which the effect of conditional DLK deletion provided a 7-day median survival benefit.⁷ SOD1^G93A mice overexpressing ATF3 had a relatively shorter median survival (152 days). In a previous study,²⁵ ATF3 overexpression increased survival by a mean of 7.7 days; the difference between our results was perhaps due to the use of different ATF3 transgenic lines in our respective studies (see Methods).

The onset of muscle weakness is reflected by an ALS-TDI score of 2 (Fig. 6B). SOD1^G93A mice with the combination of ATF3 overexpression and DLK deletion showed a trend towards later onset of paresis, but this did not reach significance. In addition to the ALS-TDI neuroscore exam, this cohort was assessed for the ability to hang from an inverted grid (Fig. 6C) as well as forelimb grip strength (Fig. 6D and Supplementary Fig. 6). Inverted grid testing did not reveal any differences between genotype groups (Fig. 6C). The grip strength of SOD1^G93A mice with DLK deletion combined with ATF3 overexpression was significantly stronger at 10, 11 and 12 weeks when compared with SOD1^G93A controls (mixed-effects analysis, P < 0.0001, P = 0.0002, P = 0.0358 for 10, 11 and 12 weeks, respectively). SOD1^G93A mice in which ATF3 was overexpressed also had small but significant improvements to forelimb grip strength during this time frame (Supplementary Fig. 6).

Overall, the genetic manipulations we tested provided only modest symptom improvement. However, among them, the combination of DLK deletion with ATF3 overexpression showed the greatest trend towards delaying symptomology.

Discussion

Our study revealed that conditional deletion of DLK signalling with targeted upregulation of ATF3 in a large subset of motor neurons of the SOD1^G93A mouse model acts in a combinatorial fashion to preserve skeletal motor neuron function. This combination resulted in retention of close to normal CMAP amplitudes throughout much of disease progression, increased survival of motor neuron cell bodies in the spinal cord, reduced axon degeneration, preservation of NMJ synapses and grip strength and reduced muscle atrophy. Synapses at the NMJ were rescued by ATF3 overexpression alone and not by conditional deletion of DLK. By contrast, ATF3 overexpression did not prevent degeneration while DLK deletion reduced motor neuron cell death and axon degeneration. Combining both manipulations provided the greatest benefit to motor neuron survival and function, as highlighted by the CMAP recordings, presumably because DLK deletion preserved a larger proportion of surviving motor neurons in which ATF3 overexpression could act to promote regeneration. Interestingly, the combination of DLK deletion with ATF3 overexpression better prevented neuronal cell death than DLK deletion alone, suggesting that the death signals to which DLK signalling contributes are further repressed in the presence of the transcriptional program under ATF3. This highlights how the distal end of the neuron and the cell body are two compartments whose survival are independently regulated, presumably by distinct signalling mechanisms, that can also influence each other.

Some of the data in this study did not completely replicate previous findings for the effect of neuronal ATF3 overexpression in SOD1^G93A mice.²⁵ This previous study reported that neuronal ATF3 overexpression in this mouse model extended lifespan and improved motor neuron survival at a 90-day time point. We did not replicate either finding at a similar time point (14 weeks, or 100 days). It is possible that we did not rederive the same exact ATF3 overexpression line because seven different Atf3 transgenic mouse lines were originally created with varying transgene expression patterns and levels.²⁷ Nevertheless, as evidenced by the strong effect on preservation of NMJ innervation (Fig. 1C), muscle mass (Fig. 3A and B) and the partial CMAP rescue (Fig. 1A) observed in SOD1^G93A mice overexpressing neuronal ATF3, the ATF3 transgene otherwise functioned as expected. The effect of DLK^cKO in this study on SOD1^G93A mouse lifespan and motor neuron survival was also weaker than previously reported.⁷ Potential factors include differences in animal housing facilities and mouse background due to rederivation, crossing and genetic drift. Nevertheless, similar trends for neuroprotection in the DLK^cKO were observed for end points that did not reach statistical significance.

Importantly, and regardless of the magnitude of effect sizes for each individual transgene, synergy between DLK deletion and ATF3 overexpression was observed for every end point examined that directly related to motor neuron and muscle anatomy or function. Skeletal motor neuron survival, neuroinflammation, extent of neurodegeneration, NMJ innervation, muscle contraction elicited by nerve stimulation and grip strength were all significantly improved. All these benefits are consistent with a potential for improving quality of life during the symptomatic phase of ALS. The progressive physical impairments that characterize ALS have clear repercussions on patient emotional well-being and mental health. Indeed, incidences of pain, anxiety and depression are all increased in ALS patients.^37–39 There is clear value in pursuing treatments that are predicted to slow physical decline and preserve quality of life even if they do not increase life expectancy.

Our combinatorial experiment revealed a striking dissociation between the preservation of nerve stimulation-elicited muscle contraction and organism survival. The close to normal CMAP amplitudes in triple transgenic mice until within 4 weeks of the animals reaching euthanasia criteria were surprising. One potential explanation is that end-stage physical decline in the SOD1^G93A model is not uniquely due to loss of motor neuron function. Indeed, mutant SOD1^G93A likely causes dysfunction and disease pathology in many cell types in addition to the neurons in which DLK deletion and ATF3 overexpression were achieved.^40–43 This result could also reflect that motor neuron axons in these mice are able to transmit an exogenous nerve stimulation to muscle, but the motor neurons are not healthy enough to generate their own activity. Alternatively, it is also possible that an acceleration of motor neuron loss occurs in the final phase of the disease course regardless of genotype group. Even if it does not lead to increased survival, increased preservation of motor neuron connectivity to muscle would be a significant advance for the treatment of ALS.

A key aspect of disease pathogenesis in all types of ALS is the loss of functional connections between motor neurons and muscle. Our work demonstrates that preserving motor neuron connectivity to muscle is possible into very late stages of disease progression. This protection was achieved by a combinatorial approach to slow neurodegeneration by blocking DLK while promoting regeneration via ATF3 expression. It will be of interest to understand what molecular changes are mediating the long-lasting benefits we detected, especially in the group combining ATF3^OE and DLK^cKO. One speculation is that the forced ATF3 expression results in prolonged upregulation of many RAGs promoting long-lasting distal connections, and that DLK deletion results in the delay of a pro-apoptotic cascade. How these genotypes synergize when combined will be extremely interesting to investigate in future molecular studies.

This type of combinatorial approach could in theory be translated into therapy by developing safe inhibitors for DLK⁴⁴ or alternative methods to block neuronal death and providing ATF3 in the form of gene therapy. Our study also highlights how treating a complex neurodegenerative disease like ALS will necessitate the combination of multiple therapeutic strategies^3,4 to effectively treat the many aspects of disease pathology.

Acknowledgements

We would like to thank the NINDS animal technicians (Porter Neuroscience building 35) as well as Dr Heather Narver for animal care and welfare checks; the NICHD Microscopy and Imaging Core (Dr Vincent Schram, Lynne Holtzclaw and Chip Dye) for dissection, staining and imaging assistance; the NIMH Rodent Behavior Core (Kevin Cravedi, Alice Graham and Dr Yohann DuHoffman) for assistance with behaviour equipment and analysis; Dr Timothy Petros (NICHD) for generously providing additional mouse cage space; Dr Casper Hoogenraad (Genentech), Dr Mark Hoon (NIDCR) and Dr Alexander Chesler (NCCIH) for helpful discussion and comments on the manuscript.

Funding

This work was funded intramurally (NICHD ZIA-HD008966).

Competing interests

The authors report no competing interests.

Supplementary material

Supplementary material is available at Brain online.

References

Author notes