Young adult (3–12 months) male Thy1-GFP-M hemizygotes on a C57BL/6 background (RRID: IMSR_JAX:007788) served as the control group (WT-GFP; n = 12 mice, 21 axons with 694 boutons, combined axonal length 8.2 mm), and were crossed with SARM1 null mutants (homozygotes) on the same background strain C57BL/6 (RRID: IMSR_JAX:018069) for the experimental group (SARM1KO-GFP; n = 9 mice; 21 axons with 628 boutons, combined axonal length 6.6 mm). The SARM1 mutant mice used in this study were first reported by Kim et al. (2007), are bred as homozygotes, and no gene product (protein) is detected by Western blot analysis of brain from homozygotes. Genotypes were confirmed during breeding by PCR. All animals were bred at the University of Tasmania’s Cambridge Facility Farm (CFF). Littermates were kept in groups of up to 5 mice per cage, housed under a 12-h light/dark cycle at 22 °C, and had ad libitum access to standard chow and water. All experiments involving animals were conducted in compliance with the University of Tasmania Animal Ethics Committee guidelines and approved under the Australian Code for the Care and Use of Animals for Scientific Purposes, 8th edition (National Health and Medical Research Council, 2013).

A modified protocol (Holtmaat and Svoboda, 2009) was used. Adult mice received a subcutaneous injection of analgesia (buprenorphine 0.1 mg/kg) at least 30 min before surgery. Anaesthesia was induced with 5% isoflurane, then maintained at 1.8–2.5% isoflurane in 100% oxygen flow at 0.6–1.0 L/min. Once unconscious, animals were secured in a stereotaxic frame on a heat pad, and the scalp was infiltrated with a local anaesthetic (Bupivacaine 0.25%) before incision. The cranial window was positioned over the primary somatosensory cortex (S1) on the right hemisphere, with the centre approximately 1.5 mm posterior to bregma and 2.5 mm lateral to the midline. This region was chosen due to its accessibility and suitability for imaging, consistent with earlier studies on synaptic dynamics in cortical axons (Canty et al., 2020; Canty et al., 2013a,b). A micro drill with a 0.5 mm burr was used to remove a 3 mm diameter piece of calvarial bone. Sterile artificial cerebrospinal fluid (aCSF) buffer (125 mM NaCl, 5 mM KCl, 10 mM glucose, 10 mM HEPES, 2 mM CaCl2, 2 mM MgSO₄, pH 7.3–7.4; ~300 mOsm) was applied regularly to cool the surface and remove any bone dust during drilling. The surface of the intact dura was irrigated with sterile gel foam (Pfizer) soaked in artificial cerebrospinal fluid (aCSF), and dexamethasone (4 mg/mL) was applied to reduce inflammation, improve clarity, and maintain proper moisture levels (Park et al., 2015). The craniotomy was covered with a round glass coverslip (5 mm in diameter) and then sealed using Loctite 454 glue and Heraeus Paladur dental cement (Canty et al., 2020; Tang et al., 2021). A titanium bar (10 mm, M2 bolt hole) was attached for imaging stability. Mice recovered for approximately 2–3 weeks before undergoing imaging.

In vivo two-photon imaging was performed using a custom-built laser-scanning microscope (Scientifica) equipped with galvo-galvo mirrors, an Olympus 20X water-immersion objective (NA = 1.0, Zeiss Plan-Apo), and a high-sensitivity GaAsP non-descanned photomultiplier detector (Hamamatsu). Femtosecond pulsed infrared excitation was delivered by a mode-locked Ti-sapphire laser (Mai Tai, DeepSee, Spectra-Physics) tuned to 910 nm with group velocity dispersion compensation. Laser power delivered to the back aperture ranged from 20 to 90 mW, depending on the imaging depth, which enabled imaging deeper than 100 μm into the cortex while minimising bleaching of fluorophores or thermal damage. Anaesthesia induction for imaging was initiated and maintained as described earlier. To maintain stability and reduce motion artifacts, the titanium head bar was secured to the stereotaxic stage mounted on an air table to minimise vibrations and ensure stability. The vascular landmark correlation in the window was used to identify the same region in subsequent imaging sessions.

After switching from bright-field to two-photon mode, the labelled neuropil appeared as a dense meshwork of GFP-labelled axons and dendrites. All images for analysis were captured as image stacks (512 × 512 pixels per slice with an XY resolution of 0.2 μm per pixel and a Z-step of 2 μm between slices) using ScanImage 3.8.1 software (Vidrio Technologies, LLC; Pologruto et al., 2003) incorporating the Navigator plugin (Meyers, 2013) in MATLAB. The same images were captured across repeated sessions.

During the first imaging session for each mouse, the GFP-labelled neuropil across the upper layers of the sensory-motor cortex was searched to an average depth of 100 μm, to identify clear axons for imaging. Axons were selected based on three main characteristics: defined terminal endings, number of synapses, and length (at least 150 μm, minimum 14 boutons). Each visible synapse was identified and monitored at 48-h intervals to assess its stability. Since terminal axons of sufficient length were difficult to locate, axons that extended beyond the imaging window (i.e., without visible endings) were also included, to increase sampling and avoid bias toward only short or truncated axons. At the point of injury, care was taken to ensure that the surviving segments of the axon (connected to the cell body) contained at least 14 synapses after the injury (post-axotomy).

Images were captured at the same time of day, 48 h apart, across seven imaging sessions, with a targeted laser lesion applied immediately after image collection in the 4th session. The baseline and post-axotomy imaging sessions are as follows: timepoints 1, 2, 3, and 4 (TP1 to TP4) prior to the lesion (baseline) and post-axotomy timepoints 5, 6, and 7 (TP5 to TP7, Figure 1A).

A red fluorescent dye (1 mM Sulforhodamine-B, 10 mL/kg i.p.) was administered at the start of imaging session 4 to visualize a vascular map of the brain (Kovalchuk et al., 2015), which assisted in targeting axotomy sites away from blood vessels that could rupture. The laser was tuned to 850 nm to image this vascular marker simultaneously with GFP.

A complete set of images was collected before axotomy to establish a baseline for comparison. The microlesion (at least 25 μm away from blood vessels) was created by setting the laser to a wavelength of 850 nm to deliver a pulse at 100% laser power with a peak power of 1.2–1.4 W. A “scan” of 32 × 32 pixels was performed without XY movement, ensuring that the laser was focused on a central spot of approximately 3 μm diameter. The frame rate was 5.92 Hz, so capturing one frame meant the shutter remained open for approximately 170 ms. An estimated 50% of delivered power was lost in the objective and glass (Holtmaat and Svoboda, 2009), so only approximately 0.7 W of laser power (1.4 W max) reached the cortex, and a total energy of 120 mJ (0.7 W × 170 ms) was calculated to be delivered to the lesion site. An energy density of approximately 1.68 MJ/cm² was applied to the 3 μm circular lesion area, effectively severing the axon while minimising damage to adjacent tissues. Lesions were deemed successful when a clear cut in the axon shaft was visible, often (but not always) accompanied by a transient bright “halo” around the lesion site (Figure 2B). We often observed limited, temporary dysmorphology in nearby axons, which typically resolved within 20 to 30 min.

All image stacks were processed using ImageJ software (Fiji, 2.1.0, Schindelin et al., 2012) starting with 16-bit (intensity value 0–65,535) greyscale images from the microscope. The process typically involved structuring raw data, converting images into an 8-bit format (with intensity values ranging from 0 to 255), validating their quality, reducing noise, and normalising them. Maximum projections were used to montage overlapping frames together and verify axon length and morphology. Curation ensured the images were formatted, montaged, and reliable, facilitating subsequent analytical stages. Adobe Photoshop CS2 and XuvStitch 1.8.099 (Xuvtools license 1.0, Emmenlauer et al., 2009) were used to montage maximum projections and z-stacks, respectively. All the images and montages were saved in Tagged Image File Format (TIFF) (Linkert et al., 2010).

Synaptic boutons were classified as terminaux boutons (TB) or en passant boutons (EPB) following De Paola et al. (2006). Unlike earlier reports that described exclusively TB-rich or EPB-rich populations, most cortical axons in the Thy1-GFP-M mice used in this study exhibited a mixture of TBs and EPBs. The literature indicates that TBs exhibit greater plasticity compared to EPBs (Canty et al., 2013a,b; De Paola et al., 2006; Fulopova et al., 2025), and the relative abundance of each type was quantified for each axon, ranging from 10 to 75% TBs. Within each axon, these percentages were averaged across imaging sessions.

Synaptic boutons were annotated using custom MATLAB scripts included with ScanImage (scim_spineAnalysis.m) (Holtmaat and Svoboda, 2009). Each bouton was assigned a unique identifier to track its status (stable, lost, or gained) across consecutive imaging sessions. Axon segment length was measured to normalise bouton counts and account for length variability, with the same segment consistently analysed at each timepoint.

Bouton classification followed established morphological criteria. EPBs were defined as a swelling visible in at least two optical slices and at least twice the GFP labelling signal of the surrounding axon shaft (Figure 1C, light blue arrowheads). TBs were identified as protrusions 1–5 μm in length, present in at least two consecutive z-slices (2 μm apart; Figure 1C, dark red arrowheads). We marked EPBs as either present or absent by manual annotation, and TBs were measured by drawing a line along their length. Structures exceeding 5 μm were classified as short branches (Figure 1C, blue arrow) and excluded from the analysis. Identification of EPB and TB synapses in Thy1-GFP mice is a commonly used approach for investigating synaptic dynamics in the intact brain, with (i) both gains and losses of boutons over time and (ii) electron microscopy reconstruction demonstrating active zones at GFP swellings along the axon or in terminal boutons confirmed in previous studies by this authorship team, and others (Bass et al., 2025; Bass et al., 2017; De Paola et al., 2006; Canty et al., 2013a; Fulopova et al., 2022, 2025; Holtmaat et al., 2008).

Data were exported to R for analysis using a custom script. Synaptic density was calculated as the number of boutons per unit length of axon. Synaptic turnover ratio (TOR) between two consecutive time points (A to B) was calculated as:

We applied a minimal threshold of 0.1 μm to register bouton presence at any length, annotating and correlating across time (gain, loss, and stable). EPBs were classified as present or absent, while TBs had to reach at least 1.0 μm in one session to be included. The final dataset included a unique mouse ID, axon segment number, axon length, number of boutons (including EPB and TB), TOR, and gains and losses at each time point. TB% was calculated as:

Statistical analyses and visualisations were performed in R4.1.0 (R Core Team, 2021) using generalised linear mixed models (GLMMs) fitted with lme4 (v1.1–37; Bates et al., 2015) and the glmmTMB (v1.1.11; Brooks et al., 2017) packages. In contrast, GLMMs were employed to accommodate a wide range of statistical distributions under a hierarchical design. We report p-values (<0.05 as statistically significant) alongside effect sizes (standardised β coefficients) to convey the magnitude of effects. For linear mixed-effects models (LMMs) or GLMMs, marginal and conditional R² quantified variance explained by fixed effects alone is calculated to assess model fit and effect size. Marginal R² represents the proportion of variance explained by the fixed effects alone, versus fixed plus random effects (Nakagawa and Schielzeth, 2013), aiding interpretation in the presence of biological variability (Nebe et al., 2023). To account for the hierarchical design (multiple axons per mouse), all mixed models included random intercepts for unique mouse ID with axons nested within ID: (1| ID/axon). Fixed effects were genotypes (WT-GFP, SARM1KO-GFP), timepoint (48-h intervals), TB%, and the genotype × timepoint interactions.

Model 1: Synaptic density and turnover ratio (Gaussian LMM):

Model 2: Synaptic gains (negative binomial GLMM with length offset):

An offset term of log (Length) was included to normalise synaptic gains by segment size, reflecting that longer axons can host more boutons. Treating length as exposure allows fair composition of gain rates across genotypes and timepoints, independent of segment size. Thus, the model estimates a bouton gain rate per length rather than raw counts.

Model 3: Synaptic losses (poisson GLMM with prior- bouton offset):

This formula explained the variability in synaptic loss (response variable) by reflecting the effects of SARM1 removal and TB percentage at each timepoint, along with previous values of the total number of boutons, and by accounting for axons nested within mouse ID as random effects.

Following axotomy, the disconnected distal segment from the cell body typically exhibited morphological change after a delay of minutes to hours, followed by rapid degeneration. The surviving proximal stump exhibited an immediate injury response characterised by the rapid formation of axonal varicosities (“beading”) and focal swelling along the axon shaft that resolved within 20–30 min, similar to prior reports (Beirowski et al., 2010; Canty et al., 2013b; Gu et al., 2017). No regrowth of the severed stump was observed in either genotype for at least 14 days post-injury.

A total of 20 axons (WT-GFP = 10 axons in 5 mice, and SARM1KO-GFP = 10 axons in 7 mice) were imaged at 48 intervals before and after axotomy, a subset of the axons previously discussed (see Figure 2). Initially, a Welch t-test was performed on the data distribution to compare synaptic density pre-axotomy as well as post-axotomy. This step was essential to show that there were no significant differences between the two genotypes (all p > 0.05), confirming comparable starting conditions for the proximal segments. Once this equivalence was established, the post-axotomy analysis was conducted for the same axons at TP5, TP6 and TP7. We used a similar linear mixed-effects model approach as for baseline data, with the addition of a fixed effect for timepoint relative to injury. We found no significant differences in density post-axotomy (Table 3, Figure 3A).

Mixed-effect modelling showed a significant increase in TOR over time in the post-axotomy proximal axon (timepoint as a main effect over the entire model, p < 0.002). Neither the main effect of genotype, nor the interaction between time and genotype was significant, suggesting that the pattern of TOR change over time did not differ significantly between the WT-GFP and SARM1KO-GFP groups (Figure 3B). No individual time interval or interaction term reached significance in the model (all p > 0.05), and TB% had no detectable influence on TOR (p = 0.52). The stable synaptic density, coupled with modest increases in turnover, is consistent with the typical pattern observed in the control group in which SARM1 is present, aligning with previously reported studies (Bradshaw et al., 2021; Canty et al., 2013b; Chen et al., 2021).

Synaptic remodelling following axotomy was assessed by separately analysing bouton formation and elimination using mixed-effects models. For bouton formation (gains), no significant main effect of time or genotype was observed (Table 2; Figure 3C). In contrast, synaptic losses showed a significant main effect of timepoint (p = 0.016), suggesting time-dependent modulation of synaptic loss (Figure 3D). However, neither the main effect of genotype [β = −0.19, 95% CI (−1.04, 0.66), p = 0.660] nor the genotype × time interaction was significant. No individual time interval or interaction term were significant in the model (all p > 0.05),

Together, these findings indicate that synaptic density is maintained after axotomy in both WT-GFP and SARM1KO-GFP axons. There is an increase in synaptic loss rates, but this is not enough to lower overall density. SARM1 deletion does not alter synaptic density, TOR, losses or gains post-axotomy. TB% does not significantly alter any of these parameters.

Two-photon imaging revealed that baseline cortical axon morphology is comparable in Thy1-GFP wild-type and SARM1KO-GFP cortical axons. This finding suggests that SARM1 deletion does not compromise the structural integrity of this specific population of adult cortical axons under normal conditions. This “morphologically silent” phenotype reflects earlier observations Thy1-YFP/Sarm1−/− mice, which similarly reported unchanged axon caliber, branching, and myelination before injury (Henninger et al., 2016; Marion et al., 2019) and is consistent with in vitro data showing only a context-dependent branching effect (Ketschek et al., 2022).

In terms of the response to an acute laser mediated axonal lesion, our findings align with our previous work (Canty et al., 2013a,b), indicating that laser-mediated axotomy results in maintenance of the surviving axon, and increase in synaptic TOR in the post-lesion period.

Our work in this current study examined a sub-population of axons bearing a mixture of both EPBs and TBs, with the proportion of TBs varying between axons. TBs have been described as having higher turnover rates and greater vulnerability to injury (Canty et al., 2013a; Fulopova et al., 2025). Interestingly, while our results indicate a strong positive effect of TB% on density, there was no significant correlation between TB% and synaptic plasticity.

In the absence of SARM1, baseline pre-axotomy synaptic density is comparable to wildtype axons (0.10 boutons per μm) and synapses remodel in a similar way to wildtype axons. This result mirrors several “quiet brain” datasets in the literature where constitutive SARM1 deletion left resting synapse numbers unchanged in cortex or striatum. It contrasts with the hippocampal CA1 study of Lin et al. which reported a modest increase in spine density in SARM1KO mice only after the pruning phase of adolescence (Lin et al., 2014). This indicates that SARM1 is not essential for maintaining the steady-state synaptic dynamics outside specific developmental periods and also implies the existence of effective homeostatic mechanisms, independent of SARM1, that regulate synaptic output and plasticity within a particular range, in both the baselines and post-axotomy conditions.

When the axons were severed by laser lesion, we observed an increase in turnover in the proximal axon after 48 h. This increase in TOR was driven by an increase in losses, rather than gains, which remained unaltered. TOR remained elevated at our last timepoint, 6 days post-axotomy, indicating a sustained rise in synaptic remodelling, driven by losses. Increased synaptic losses after injury would need to be of a larger magnitude to be detected as a simultaneous decrease in synaptic density or be maintained over an extended period where the increased loss is only small. In the case of the data presented in this manuscript, the increased losses were detected towards the end of the experimental time window (+6 days). Extending the experimental time frame beyond 6 days post-injury would inform the duration of synaptic losses. We hypothesize that this could be interpreted in relation to the degeneration of the distal axon. The severed portion of the axon typically degenerates in a Wallerian-like pattern, fragmentating over a period of 24 h, with debris removal almost complete after 48 h. With the resulting loss of connectivity, there will be changes to local circuitry, which could impact the remaining outputs in the surviving proximal segment of the severed axon. Local circuitry synaptic remodelling might cause the pruning of some synaptic outputs in the surviving axon (increased losses) without necessarily driving an increase in gains, at least in the short term. Regardless of the mechanisms driving the increase in losses but not gains, neither of these processes were altered by the absence of SARM1 signalling in the post-lesion axon.

In previous work where we stimulated the brain with transcranial magnetic stimulation (TMS) coupled with in vivo imaging, similar excitatory axons in the somatosensory cortex of Thy1-GFM-M mice underwent a significant increase in turnover of boutons following a single round of TMS stimulation, with a large effect size of approximately 100%—turnover doubled within 48 h (Fulopova et al., 2025). In those studies, synaptic changes were detected using a similar volume of synaptic data—8 wildtype mice, 25 axon segments, 757 boutons, combined length 5.6 mm, compared to the current study n = 12 wildtype mice, 21 axons, 694 boutons, combined axonal length 8.2 mm. In the context of SARM1 signalling, Sarm1 knockdown in cultured embryonic hippocampal neurons resulted in a 20% increase in spine density, and glutamate stimulation in the presence of a Fura-2 calcium indicator, showed an increase in the calcium signal, suggesting that Sarm1 knockdown increased calcium influx via the NMDAR by altering the synaptic composition or by altering downstream signalling pathways (Lin et al., 2014). In the current study, we have detected a significant time-related effect on turnover ratio (partial-eta-squared = 0.15), however the observed effect of genotype was too small to be statistically significant and of limited biological relevance. Several in vivo studies examining global brain or spinal injury concur with our finding that bouton density remains unchanged by SARM1 deletion during the first week post-axotomy. In a mouse model of TBI, corpus callosum axons were preserved in SARM1 knockouts, and there was no increase cortical synaptic number within 7 days, even though distal white-matter degeneration was almost completely blocked by the removal of SARM1 signalling (Bradshaw et al., 2021). Similarly, Scheff et al. reported that synaptic loss after a TBI peaks early and recovers within a month, a pattern that WT-GFP and SARM1KO-GFP axons replicate in our study, as evidenced by their flat density curves (Scheff et al., 2005). Together with our data, these studies suggest that synapses in the proximal segment are protected against the catastrophic degeneration process that SARM1 regulates in the distal, disconnected segments.

This aligns with research showing that SARM1’s enzymatic NADase activity becomes significant during periods of metabolic or injury-induced stress. For example, in a TDP-43 ALS neurodegeneration model, mice with intact SARM1 signalling exhibited substantial progressive loss of dendritic spines within cortical neurons, whereas SARM1 deletion resulted in maintenance of dendritic spines (White et al., 2019). Similar synaptic protection after traumatic brain injury in the same mouse model has also been reported (Bradshaw et al., 2021). Recent mechanistic reviews of SARM1 signalling now describe SARM1 as a metabolic checkpoint, remaining inactive during regular remodelling but ready to initiate axon degeneration when NAD+ stress exceeds a threshold (Figley et al., 2021; Karnik and Joshi, 2025).

In the proximal surviving stump of the lesioned axons in this study, axonal transport recovered within 24–48 h, with varicosities resolving and no significant stump retraction, consistent with rapid axonal stabilisation after the initial laser-induced injury. Together, these observations support the view that the energetic/NAD+ balance in the surviving stump is restored quickly, and its perturbation is likely insufficient in duration or severity to activate SARM1 in surviving axons 48 h after injury.

Baseline axonal morphology and synaptic density in WT-GFP and SARM1KO-GFP cortical axons were indistinguishable, indicating that SARM1 deletion does not disrupt structural integrity or steady-state synaptic dynamics under normal conditions. Following laser-induced axotomy, synaptic turnover increased due to elevated losses in the surviving axon, yet this response—and subsequent synaptic remodelling—was unaffected by SARM1 deletion, suggesting that any synaptic effects of SARM1 deletion are either subtle, transient or non-existent.

Several limitations of our study should be acknowledged. Firstly, this study was underpowered for detecting subtle synaptic phenotypes. The axotomy time course experiments are both technically demanding and time-consuming and axon numbers were therefore limited. Secondly, our last timepoint for imaging was 6 days post-axotomy. This allowed for 3 separate post-injury intervals to assess synaptic changes and may have missed any rapid changes that then resolved, or any longer-term changes. Similarly, we did not assess the time course of SARM1 activation in wildtype mice. However, there is clear evidence from other studies, that SARM1 signalling is rapidly induced after axonal injury (for example; Ko et al., 2021) and in studies using the same mice, clear attenuation of axonal pathology is described as early as 2 h after a closed head weight drop injury, peaking at 48 h post injury (Henninger et al., 2016). Finally, the experimental axons in this study were located superficially in the primary somatosensory cortex (S1), which processes tactile and sensorimotor information, with some axons likely crossing into the primary motor cortex (M1). It is well established that bouton density and synaptic dynamics can vary markedly across cortical areas, indicating differences in functional organisation and plasticity demands. For example, EPB-rich axons in the prefrontal cortex (PFC) exhibit higher synaptic density and may potentially undergo greater plasticity than those in the M1/SS1 cortex in 12-month-old mice (Fulopova et al., 2022).