Dynamic neuromuscular remodeling precedes motor-unit loss in a mouse model of ALS

Despite being an early event in ALS, it remains unclear whether the denervation of neuromuscular junctions (NMJ) is simply the first manifestation of a globally degenerating motor neuron. Using in vivo imaging of single axons and their NMJs over a three-month period, we identify that single motor-units are dismantled asynchronously in SOD1G37R mice. We reveal that weeks prior to complete axonal degeneration, the dismantling of axonal branches is accompanied by contemporaneous new axonal sprouting resulting in synapse formation onto nearby NMJs. Denervation events tend to propagate from the first lost NMJ, consistent with a contribution of neuromuscular factors extrinsic to motor neurons, with distal branches being more susceptible. These results show that NMJ denervation in ALS is a complex and dynamic process of continuous denervation and new innervation rather than a manifestation of sudden global motor neuron degeneration.


Introduction
Amyotrophic lateral sclerosis (ALS) is a fatal neurodegenerative disease characterized by upper and lower motor neuron (MN) loss. Loss of neuromuscular junctions (NMJs) is a key pathological event in ALS patients (Fischer et al., 2004;Killian et al., 1994) and in animal models of the disease (Clark et al., 2016;Fischer et al., 2004;Frey et al., 2000;Pun et al., 2006;Tallon et al., 2016). Denervation of NMJs is observed prior to symptom onset (Clark et al., 2016;Fischer et al., 2004;Pun et al., 2006;Tallon et al., 2016;Vinsant et al., 2013) and before any significant MN axonal or cell body loss (Fischer et al., 2004;Vinsant et al., 2013). However, whether early NMJ denervation arises from local branch-specific degeneration ('dying-back hypothesis') or is merely the manifestation of a diseased and globally degenerating motor neuron remains ill-defined.
There is considerable evidence which suggests that a subset of degenerating motor-units (MU, a motor neuron and the muscle fibers it innervates) swiftly lose their NMJs while a distinct, yet unaffected, population compensates via axonal sprouting (Pun et al., 2006;Schaefer et al., 2005), somewhat akin to a nerve injury. Indeed, NMJs innervated by fast-fatigable MNs are lost early, followed by fast-fatigue resistant and slow MNs (Frey et al., 2000;Pun et al., 2006). This is further supported by electromyographic (EMG) recordings showing that the number of MUs declines in parallel with muscle strength (Azzouz et al., 1997;Dantes and McComas, 1991;Hegedus et al., 2007;Hegedus et al., 2008;Kennel et al., 1996;McComas et al., 1971) and that surviving MUs increase in size in animal models and in patients (Dantes and McComas, 1991;Dengler et al., 1990;McComas et al., 1971;Schmied et al., 1999). Taken together, these results strongly suggest that the loss of NMJs within a single MU is a dichotomous and mostly synchronous event and that local synaptic events have a limited impact on the denervation process. Under this monotonic paradigm, degenerating MUs are irrevocably caught in a degenerative cascade while surviving MUs attempt to compensate, consequently undermining the relevance of the NMJ as a disease-modifying therapeutic target.
However, the rescue of neuronal loss in SOD1 mice does not necessarily improve NMJ denervation (Gould et al., 2006;Parone et al., 2013;Suzuki et al., 2007) suggesting that local branch-specific events may play a key role in NMJ denervation. In ALS, local alterations of synaptic function and molecular signaling take place very early at the NMJ (Arbour et al., 2015;De Winter et al., 2006;Jokic et al., 2006;Rocha et al., 2013;Taetzsch et al., 2017;Tremblay et al., 2017), implying that local, MU-independent, structural changes could occur. Finally, ALS is a non-cell autonomous disease where cell types other than motor neurons contribute to the disease pathogenesis (Boillée et al., 2006a;Ilieva et al., 2009). Contribution of various cell types known to interact with NMJs and motor axons has been suggested (Arbour et al., 2015;Chiu et al., 2009;Keller et al., 2009;Lobsiger et al., 2009;Loeffler et al., 2016;Nardo et al., 2016;Turner et al., 2010;Wang et al., 2012), which could contribute to NMJ loss. In this scenario, asynchronous branch-specific synaptic changes would take place over time within single MUs.
Altogether, these divergent paradigms highlight that the precise sequence of events taking place at the NMJ in ALS remains ill-defined. From a therapeutic viewpoint, this distinction is of high importance since it could pinpoint the NMJ as an accessible target that can be exploited within a time window between the onset of structural changes and the global degeneration of the MU, as previously suggested (Cantor et al., 2018;Miyoshi et al., 2017;Pérez-García and Burden, 2012). Here, we used longitudinal repeated in vivo imaging of single motor neuron axonal arbors and their NMJs to directly assess the time course and dynamics of structural changes during ALS progression. We eLife digest Amyotrophic lateral sclerosis (ALS), also known as Lou Gehrig's disease, is a fatal neurodegenerative disorder. It occurs when the neurons that control muscles -the motoneuronsdisconnect from their target muscles and die. This causes the muscles to weaken and waste away. More and more muscles become affected over time until eventually the muscles that control breathing also become paralyzed. Most patients die within two to five years of diagnosis.
Motoneurons consist of a cell body plus a cable-like structure called the axon. The cell body of each motoneuron sits within the spinal cord, and the axon extends out of the spinal cord to the motoneuron's target muscle. Within the muscle the axon divides into branches, each of which connects with multiple muscle fibers. The breakdown of these connections, known as neuromuscular junctions, is one of the first signs of ALS.
Does a motoneuron lose all of its connections with muscle fibers at once, or do the connections break down a few at a time? This distinction is important as it will help to identify the events that lead to muscle paralysis in ALS. To find out, Martineau et al. studied mice that had two genetic mutations: one that causes ALS and another that produces fluorescent molecules in some motoneurons. This allowed the branches of the motoneurons to be tracked over time with a fluorescence microscope.
Martineau et al. found that individual neurons lose their connections to muscle fibers gradually. Moreover, motoneurons grow new branches and form new connections even while losing their old ones. This dual process of pruning and budding lasts for several weeks, until eventually the motoneuron dies.
Developing drugs to stabilize neuromuscular junctions during the period when motoneurons gradually disconnect from muscles could be a promising avenue to explore to improve the quality of life of ALS patients. One advantage of this treatment strategy is that neuromuscular junctions in muscles are easier to access than motoneurons inside the spinal cord. To identify potential drugs, future studies will need to focus on the proteins and signals that cause the neuromuscular junctions to break down.
predict that a MU-dependent mechanism would result in the more or less synchronous degeneration of the whole MU over a brief period, while a contribution of local neuromuscular changes would result in the asynchronous loss of NMJs within a single MU.

Results
In vivo imaging of single MU arbors SOD1 G37R mice (loxSOD1 G37R ) are well known for their slow disease progression, making it an ideal model to follow single MU changes over time (Boillée et al., 2006b;Lobsiger et al., 2009). To visualize and follow single MUs in control and ALS mice, we crossed SOD1 G37R mice to animals expressing a cytoplasmic fluorescent marker in a small random subset of MNs (Thy1-YFP line H, Figure 1A) (Feng et al., 2000). Repeated in vivo imaging was performed on the Tibialis anterior muscle since numerous NMJs can be observed near the surface when exposed following a minimally invasive surgery (WT/YFP: N = 5; SOD1 G37R /YFP: N = 11). Postsynaptic sites were identified in vivo by staining nicotinic receptors (nAChR) with fluorescently labeled a-bungarotoxin (Li et al., 2011;Turney et al., 2012). Except for two cases (animal #23 and #21), all imaging sessions were performed at two-week intervals for up to 5-6 sessions (56-76 days). The same MU arbor (axon and its NMJs) was identified on every imaging session through the unique 'pretzel' shape of its NMJs, the branching pattern of the motor axon, and the position of nearby postsynaptic sites ( Figure 1A, landmarks, D, *, . ). NMJs were classified based on their innervation status over time ( Figure 1-figure supplement 1).
We first confirmed that the same MU arbor and its NMJs were reliably imaged over several weeks without inducing any structural damage. We followed single MUs in WT/YFP animals and found that YFP-labeled axons stayed precisely overlapped with nAChR-rich postsynaptic sites over all imaging sessions performed (36/36 NMJs, 5/5 MU arbors, N = 5; Figure 1C). Importantly, very few changes in axonal branching pattern, postsynaptic nAChRs and presynaptic nerve terminal organization were observed. Consistent with previous reports (Balice-Gordon and Lichtman, 1990;Li et al., 2011;Schaefer et al., 2005;Turney et al., 2012), our results show that NMJs of WT mice remained innervated by the same MU and were very stable over long periods of time.

Asynchronous NMJ dismantlement in SOD1 G37R mice
We next performed imaging in symptomatic SOD1 G37R /YFP animals (all animals at or past disease onset) to investigate whether the pattern of NMJ denervation in ALS mice was synchronous or asynchronous. Of note, SOD1 G37R /YFP mice were phenotypically and histologically indistinguishable from SOD1 G37R mice ( Figure 2-figure supplement 1), showing that YFP expression did not alter the disease course and that YFP-expressing MNs are not more susceptible to the disease.

Local NMJ loss precedes global MU degeneration
A number of visualized motor axons and their remaining NMJs were eventually lost in a synchronous manner (7/19 MU arbors, 5/11 mice; Figure 3A) showing that global MU degeneration ultimately occurs. This phenomenon likely reflects an axonal or central event, such as MN death. Importantly, signs of NMJ denervation preceded most of these MU degenerations ( Figure 3B, 6/7 MU arbors). In several cases (11/19 MU arbors), asynchronous NMJ dismantling started more than a month (! 3 sessions) prior to global degeneration of the MU, again showing that NMJ loss is initially a slow asynchronous process ( Figure 3B). Hence, we identified two phases of NMJ loss within a MU: an initial slow local branch-specific dismantlement, followed by a sudden global axonal degeneration. These different patterns of MU degeneration are unlikely to be due to differences in MU types (Frey et al., 2000;Hegedus et al., 2007;Hegedus et al., 2008;Pun et al., 2006). Indeed, fiber type (indicative of MU type) composition was examined on Tibialis anterior cross sections and revealed that nearly all superficial fibers in SOD1 G37R /YFP were fast-fatigable during the period of imaging (myosin heavy-chain isoforms type IIb or IIx;   MUs expand their axonal arbor but do not reinnervate their lost NMJs Surprisingly, a vast majority of MUs in SOD1 G37R /YFP also sprouted toward nearby nAChR-rich postsynaptic sites, forming new synaptic contacts on muscle fibers that they did not innervate initially (MU expansions or 'heterologous re-innervation', Figure 4A,B; 76/253 NMJs, 14/19 arbors, 9/11 animals). These numerous expansions resulted in a net increase in MU size which masked the contemporaneous NMJ dismantlement ( Figure 4C; 26.0 ± 10.2% loss vs 80.7 ± 32.2% expansions on session 6). Importantly, we found that over half of analyzed MUs both lost and formed new NMJs throughout the imaging process (  where no losses, at least one partial NMJ loss, at least one complete NMJ loss or synchronous MU degeneration were observed (green, yellow, red and black dots, respectively). Gray dots represent session where the MU could not be resolved while 'Failed' identifies session terminated for animal care reasons. Notice how synchronous global MU degenerations were almost always preceded by weeks of asynchronous branch-specific changes, even at later stages of the disease. The distribution of surface muscle fiber types (MU types) is presented in Figure 3-figure supplement 1, showing that these different patterns of degeneration are unlikely due to differences in MU types. DOI: https://doi.org/10.7554/eLife.41973.009 The following figure supplement is available for figure 3:   Interestingly, however, there was an inverse relationship between the propensity of axonal branches to be lost or to expand ( Figure 4D). Indeed, NMJs arising from distal axonal branches were more susceptible to being lost ( Figure 4D, red dots, 38 events, n = 253, N = 11, logistic regression, p=0.010), while MU expansions were more likely to occur on proximal branches (mainly first-order branches) ( Figure 4D, blue dots, 76 events, n = 253, N = 11, logistic regression, p=0.001). Hence, proximal branches preserve their capacity to re-innervate heterologous NMJs despite the ongoing degeneration of distal branches at the single MU level.
Regardless of this extensive compensation, few MUs re-innervated their own lost NMJs ('autologous re-innervation') and most partial losses were not repaired ( Figure 4C, purple; 2/21 and 15/40 possibilities, respectively). Immunolabeling of all motor axons after the last in vivo session showed that some of these asynchronously lost NMJs had been reinnervated by other MUs (Figure 4-figure supplement 2, N = 4), confirming that these vacated postsynaptic sites were not refractory to reinnervation in general. This process led the initial NMJ pool of surviving MUs to gradually recede ( Figure 4C, green) even though their total size remained stable or increased in most cases due to heterologous expansions ( Figure 4C, blue).
Altogether, these results show that rather than being exclusively degenerating or compensating, MUs are highly dynamic during disease progression, with specific branches either expanding or disassembling.

NMJ dismantlement seems to propagate
To gain further insight into the mechanisms underlying NMJ dismantlement, we next examined the spatial and temporal relationship between denervation events taking place within a MU. Indeed, local signals emanating from muscle fibers, glial cells or a damaged axonal branch (e.g. chemorepulsive molecules (De Winter et al., 2006;Maimon et al., 2018) or misfolded proteins (Ayers et al., 2016;Grad et al., 2014) could potentially propagate through the extracellular space and contribute to NMJ loss in ALS. If such local signals in the neuromuscular environment influence NMJ dismantlement, one would predict that these events may tend to propagate from one or several focal points. To test this possibility, we performed a Sholl-like analysis on all MU arbors bearing multiple complete NMJ losses (37 events out of 141 NMJs over 9 MU arbors, N = 6; see Materials and methods). Interestingly, asynchronous NMJ dismantling tended to propagate from the first lost NMJ ( Figure 5A and B; multivariate repeated-measures GLM with logistic distribution, interaction between distance and time: p<0.001). Indeed, adjacent events (<300 mm away) occurred preferentially, but not exclusively, within the next two sessions (T 2 ), while remote events (>300 mm) occurred exclusively three or more sessions later (T 3 +). These results unravel a propagation pattern and suggest that factors extrinsic to motor axons (i.e. the local neuromuscular environment) interact with the intrinsic properties of motor axons (e.g. susceptibility of specific branches), to shape the pattern of NMJ loss in ALS.

Discussion
Our data reveal that single MNs initially lose NMJs asynchronously while they contemporaneously expand their axonal arbors to form new synaptic contacts in a mouse model of inherited ALS (Figure 6). The resulting increase in axonal arbor size masks the underlying NMJ dismantlement, effectively reconciling the 'dying-back hypothesis' with the observed MU enlargement in animal models and ALS patients (Dantes and McComas, 1991;Dengler et al., 1990;Fischer et al., 2004;McComas et al., 1971;Pun et al., 2006;Schaefer et al., 2005). Interestingly, these results are also in line with functional single MU recordings obtained from ALS patients identifying a majority of enlarged MUs and a subset of abnormally small MUs reflecting extensive NMJ dismantlement (Dengler et al., 1990;Schmied et al., 1999).
One of our key findings is that NMJ denervation in a MU is initially slow and takes place over more than a month in a single motor neuron in SOD1 G37R mice. This time course reveals a long timewindow after the onset of NMJ loss during which MNs are not globally degenerating and preserve their capacity to reinnervate NMJs. Peripheral stabilization of NMJs and MUs thus represents an attractive therapeutic target in the context of ALS (Cantor et al., 2018;Miyoshi et al., 2017), particularly when considering the accessibility of the neuromuscular system (i.e. no blood-brain-barrier). The propagating pattern of events and the asynchronous or multidirectional changes within a single MU suggest that neuromuscular factors extrinsic to motor axons (i.e. the local neuromuscular environment) affect or contribute to NMJ loss in ALS. These factors would presumably interact with the intrinsic vulnerability of axonal branches (distal vs proximal, etc.) to shape MU dynamism in ALS. Notably, chemo-repulsive molecules (De Winter et al., 2006;Jokic et al., 2006;Maimon et al., 2018;Moloney et al., 2014), availability of trophic factors (Moloney et al., 2014;Taetzsch et al., 2017), enhancement of NMJ stabilizing signals (Miyoshi et al., 2017;Pérez-García and Burden, 2012), secreted pathological proteins or dysfunctional glial cells (Arbour et al., 2015;Arbour et al., 2017) are mechanisms of therapeutic interest.
Another main finding is that MUs are not segregated as either degenerating or compensating (Pun et al., 2006;Schaefer et al., 2005). Rather, we observed a broad spectrum of MU phenotypes (Figure 4-figure supplement 1), with the majority continuously alternating between expansion and disassembly for weeks. This asynchronous branch-specific behavior contrasts with previous observations made by Schaefer and colleagues using single time point MU imaging in the fast-progressing SOD1 G93A model (Schaefer et al., 2005). They reported that MUs with fragmented (degenerating) branches lacked thin axons (sprouts) while enlarged MUs lacked signs of degeneration. A number of observations can explain this apparent discrepancy. First, our time lapse analysis revealed that only some MUs underwent gains and losses in the same session (Figure 4-figure supplement 1), thus often giving the impression that they are either degenerating or compensating if only a single time point is considered. Second, we seldom observed a degenerating fragmented branch (9/38, Figures 1 A,2A,A,4 for example), which led us to believe that the complete dismantlement of a single axonal branch can easily be missed if identified only by the fragmented axonal morphology. Another possibility would be that different MU types were considered in these studies, whereas the present study only focused on fast-fatigable MUs (MHC IIb and IIx fibers). However, Pun et al., 2006  The following source data is available for figure 5: Source data 1. Spreadsheet of the innervation status of all NMJs in each MU arbor with their distance from first lost NMJ in the arbor (in mm) and timing of subsequent losses relative to the initial losses (T 0 , T 1 , T 2 , T 3 and T 4 ). DOI: https://doi.org/10.7554/eLife. 41973.018 reported no signs of compensatory reinnervation on the surface of the TA (same region as the present study) while Schaefer et al., 2005 imaged a mixture of fast-fatigable and fast-resistant MUs (Sternomastoid, Cleidomastoid and Clavotrapezius), with the former presumably behaving similarly to what we describe here. Hence, we do not believe these discrepancies can be attributed to differences between MU types. Nevertheless, these studies raise the possibility that slower MU types may behave differently in ALS. One might speculate that the fast-resistant and slow MUs may be resistant to the asynchronous dismantlement and only exhibit late synchronous degeneration. Alternatively, these MUs may degenerate similarly to fast-fatigable MUs, but at a later time point, in part due to their small axonal arborization size (see below). Further examination of MU dynamism in a slowtwitch muscle, which could not be achieved here due to technical and ethical considerations, could shed light on these possibilities.
Interestingly, this pattern is reminiscent of post-natal synapse elimination at the NMJ, where the elimination of supernumerary axonal branches is also asynchronous and branch-specific within single MUs (Keller-Peck et al., 2001). This resemblance suggests that similar molecular mechanisms could be involved. A recent study has shown that selective axonal branch loss during post-natal synapse elimination was mediated by branch-specific microtubule destabilization (Brill et al., 2016). Interestingly, pharmacological stabilization of microtubules is beneficial to ALS mice (Fanara et al., 2007), suggesting that microtubule instability could underly the increased MU dynamism described here.
One unexpected finding was that MNs failed to re-innervate their lost NMJs, despite preserving their capacity to sprout and form new synaptic contacts onto other heterologous postsynaptic sites. Indeed, motor axons usually autologously reinnervate their original postsynaptic sites with great fidelity following an axonal lesion (Nguyen et al., 2002). It could be argued that MU expansions represent the autologous re-innervation of NMJs that were dismantled before session 1. However, this scenario seems unlikely considering that some YFP-negative axons reinnervated NMJs lost by the imaged YFP-positive motor axon (Figure 4-figure supplement 2). One possible explanation may be that either the heterologous sprout itself or a molecular signal at the endplate prevents the original axon from reinnervating its target. Alternatively, asynchronous losses occur more frequently on distal branches, which seem intrinsically less likely to form sprouts ( Figure 4D), thus reducing the likelihood of autologous reinnervation occurring.
Consistent with previous results (Tallon et al., 2016), we report that NMJs arising from distal branches in complex arbors (higher branch order) are more susceptible than proximal ones at the single MN level in ALS. These results support the notion that a larger axonal arborization size increases the energetic load on neurons, thus making them more vulnerable to pathological insults in neurodegenerative disease (Le Masson et al., 2014;Pacelli et al., 2015). One could speculate that the constant dynamism and the excessive MU expansions further increases the energetic load of MNs thus adding to their stress during disease progression.

Concluding remarks
Overall, this detailed in vivo analysis reveals a previously unsuspected dynamism at the NMJ in an animal model of ALS. Our findings highlight a long temporal window between early branch-specific dismantling and global motor axon degeneration which could be therapeutically exploited in ALS (Arbour et al., 2017;Moloney et al., 2014).

Repeated in vivo imaging
Procedures for repeated in vivo imaging of the Tibialis anterior muscle were adapted from previous reports (Li et al., 2011;Schaefer et al., 2005;Turney et al., 2012). Mice were anaesthetized with isoflurane (2 -3% in 98 -97% O 2 ) in an induction chamber and maintained under anesthesia using a breathing mask. Mice were placed on their side and a hind limb was immobilized on a custom-made stand. Eye drying was prevented by applying a lubricant (Vaseline). The Tibialis anterior muscle was exposed through an incision made on the external side of the leg (from~5 -10 mm proximal to the knee to~5 mm proximal the ankle). The skin was drawn tight using four hooks (Guthrie Retractors; #17021 -13; FST, Canada) mounted on custom-made blocks around the leg. The muscle was immersed in a physiological solution by regular irrigation with sterile lactated Ringer (B. Braun Medical, CDMV, Canada). Postsynaptic nAChRs were labeled with a non-blocking concentration of Alexa594-conjuguated a-Bungarotoxin (BTX, 5 mg/mL in sterile lactated Ringer for 10 min; Molecular Probes, Fischer Scientific, Canada). a-BTX was applied on the first session and was reapplied once (if necessary) only when the labeling was too dim, usually on session 5 or 6. Muscle contractions and synaptic transmission are known to be preserved at this concentration of a-BTX (Arbour et al., 2015;Li et al., 2011;Turney et al., 2012;Zuo et al., 2004 and unpublished observations). NMJs near the surface were imaged using a 20X water immersion objective (0.4 NA, Nikon, Japan) mounted on an upright epifluorescent microscope (Nikon, Optiphot-2) equipped with a Neo-sCMOS camera (Andor, UK). Fluorescence excitation and emission were filtered using a Brightline Pinkel filter set optimized for CFP/YFP/HcRed (CFP/YFP/HcRed-3X-A-000; Semrock, NY). Images of the whole superficial MU axonal arborization at multiple focal planes were acquired using the Metafluor software (Molecular Devices, CA). After the imaging session, the wound was sutured using 5 -0 or 6 -0 vicryl suture (Johnson and Johnson), followed by tissue glue (GLUture; Abbot Laboratories, WPI, FL). Buprenorphine (3 mg/10 g body weight; Temgesic, CEVA Animal Health Ltd, UK) was administered by subcutaneous injections three times a day for 24 hr following the surgery. Antibiotics (0.1 mg/10 g body weight; Baytril; Bayer.Inc, Canada) were administered subcutaneously every 24 hr for 72 hr. Mice were given a small treat after each surgery (Nutella) to reduce stress and improve recovery. Except for two cases, imaging sessions were performed 2 weeks apart.

In vivo images and data analysis
All available SOD1 animals with YFP-positive surface motor axons, were included in the in vivo analysis (16/44 mice tested). Only mice in which the MU arbor could be reliably followed for at least three sessions, with at least three NMJs near the surface, were included in the analysis (11/16 mice). NMJs which were too deep to be fully resolved were discarded. Images were analyzed using the Fiji software. First, focal planes from the same region were aligned using the 'StackReg' pluggin. MU branches were traced using the 'Simple Neurite Tracer' plugin. The clearest session (generally session 1) was chosen for tracing. All segments of the MU arbor (from each region imaged) were merged in Photoshop (Adobe) and BTX labelling (in the overlap between the regions) was used as a reference for alignment. Tracings were used as a map to identify each NMJ on subsequent imaging sessions. For the purpose of this analysis, two labeled branches that could not be undoubtedly linked to the same main motor axon were considered as independent MNs. Then, the state of innervation of each NMJ was determined and a MU dynamic diagram showing the state of each NMJ within the arborization on each session was created using Visio2013 (Microsoft). NMJs were then classified based on their dynamic changes (Figure 1-figure supplement 1). For figure representations, contrast was linearly adjusted to facilitate observations. For the spatio-temporal analysis (Sholl-like analysis) of asynchronous NMJ dismantling ( Figure 5), distance between NMJs and the initial loss was measured on the full MU tracings or, for a few cases, directly on the raw aligned images. The initial event (center) was defined as the first completely lost NMJ. When two or more NMJs were lost simultaneously, the one which first showed signs of denervation (partial loss) was defined as the center. If they could not be distinguished in this manner, the analysis was carried out with multiple centers (2/9 MU arbors). For one animal (#34), several branches were fairly close to each other, that is in the same field of view (branches A and B, C and D). To more accurately assess the presence of a spreading pattern, a single center per field of view was determined. Branch order was determined using the motor-unit dynamic diagrams. MU expansions were assigned the same branch order as the branch from which they originated from, to determine likelihood of sprouting of axonal branches.
To create the heat map ( Figure 4-figure supplement 1), the % of change was calculated by counting the number of gained or lost NMJs over the number of initial NMJs. Partial gains/losses counted as half an event. When an NMJ could not be observed, the % of change was spread across the sessions where it could have occurred.

Whole-mount NMJ immunolabeling and motor neuron counts
Procedures for whole-mount muscle preparation, spinal cord sectioning and immunostaining were performed as previously described (Tremblay et al., 2017), with some minor adjustments. Mice were transcardially perfused with cold PBS 1X and 4% formaldehyde. Both Tibialis anteriors were then dissected and post-fixed for 20 min in 4% formaldehyde while the whole mouse was further fixed overnight at 4˚C. Lumbar spinal cords were dissected, post-fixed for 2 hr, cryoprotected in a 30% sucrose-PBS solution for 72 hr at 4˚C and frozen in cooled isopentane (À40 to À50˚C).
Observations and image acquisition were performed on a Zeiss LSM 880 confocal microscope with a 20X water immersion objective (N.A. 1.0) or a 63X oil immersion objective (N.A. 1.4). No image manipulations were performed after acquisition, except for linear contrast adjustments for figure presentation. For motor neuron counts, MNs in both ventral horns were counted from 15 to 20 sections per animal, at least 90 mm apart. ChAT-and NeuN-positive cells in the ventral horn were counted as a-MNs while ChAT-positive and NeuN-negative cells were counted as g-MNs as previously described (Lalancette-Hebert et al., 2016;Tremblay et al., 2017). The Allan Brain Atlas Mouse Spinal cord reference set was used as a reference to ensure that all analyzed sections were in the lumbar spinal cord. Results are expressed as the average number of cells counted per ventral horn in each animal.

Statistical analysis
For the Sholl-like analysis, each lost NMJ represented an event in time (time after initial loss) and space (distance from the initial loss) occurring over a given number of trials (total number of NMJs). This type of data (i.e. number of observations over a number of trials as a % of events) follows a logistic distribution rather than a Gaussian distribution. Hence, a multivariate repeated-measures Generalized linear models (GLM) using a logistic distribution was created to test the effects of time and space on the likelihood of observing an event. The time after the initial loss was set as a 'within-NMJ' factor (the repeated measure) while distance from the initial loss was set as a continuous 'between-NMJs' variable. The raw measured distance was used rather than the range bins presented in Figure 5. For the branch order analysis, a similar approach was used. Each lost or gained NMJ represented an event occurring over a given number of trials (number of NMJs within that branch order) in eight conditions (branch orders). Hence, a logistic regression was used to test the effect of the branch order on the likelihood of an NMJ being loss (using the univariate GLM with logistic distribution command). The branch order was set as an ordinal variable (continuous). For these analyses, 'n' (the number of trials, i.e. the sample size) represent the number of NMJs observed while 'N' (the biological replicates) represents the number of animals.p-Values smaller than 0.05 (a = 5%) were considered statistically significant. Analyses were performed in the SPSS 24.0.0.0 (IBM) software.
When two independent groups were compared, unpaired two-tailed t-tests were performed using Welch's correction for unequal variance or a Mann-Withney test when data did not follow the assumption of a Gaussian distribution. Variances were compared using the F-test. When the effect of one variable was compared over time (motor behavior, Figure 2-figure supplement 1), two-way ANOVAs with repeated measures (RM) were used. For the post-test, Holm-Sidak's correction was used. For NMJ innervation (Figure 2-figure supplement 1), the effect of genotype was analyzed using a GLM with a logistic distribution and Holm-Sidak's correction was applied for the post-test as previously described (Tremblay et al., 2017). Importantly, a GLM with a logistic distribution was used for NMJ innervation instead of a t-test (as for motor neuron counts) because NMJ innervation data (# of denervated NMJs/ # of total NMJs for each animal) does not follow a Gaussian distribution. Unless otherwise stated, data are presented as mean ± SEM in the histograms and in the text. For these analyses, 'N' represents the number of biological replicates (animals, i.e. the sample size) while "n' represents the number of observations (number of NMJs unless otherwise stated). p-Values smaller than 0.05 (a = 5%) were considered statistically significant. These analyses were performed in the GraphPad Prism 7.0 software, with the exception of the analysis for the NMJ innervation was made in SPSS 24.0.0.0.
No sample size estimations were performed before the experiments. All available SOD1 animals which met our inclusion/exclusion criteria were included (see in vivo images and data analysis). For the NMJ innervation and motor neuron counts experiments, sample sizes similar to previous experiments were used (Tremblay et al., 2017). No test to detect outliers was performed. Data availability All data generated or analysed during this study are included in the manuscript and supporting files. Source data has been provided for Figure 1