We have previously shown that injury-induced autophagy promotes axon regeneration through degrading notch receptors (NOTCH)32. Given the emerging evidence supporting the crosstalk between autophagy and ferroptosis33,34,35, we sought to test whether ferroptosis pathway is involved in axon regeneration by examining axon regrowth in available loss-of-function mutants of C. elegans gpx genes, which are homologous to human GPX genes36 Fig. 1A, B). We assayed axon regrowth in vivo using touch sensory PLM neurons, a well-established model for axon regeneration37,38. Touch neuron-specific reporter alleles muIs32 (Pmec-7::GFP) and zdIs5 (Pmec-4::GFP) were used to label the touch sensory neurons for morphological analysis and single-neuron laser axotomy. We performed axotomy at Day 1 young adult stage and measured axon regrowth 24 h post-injury. As GPX4 is known to suppress ferroptosis, an iron-dependent and ROS-reliant cell death39 (Fig. 1A), we expected to see reduced axon regeneration in gpx mutants. However, we observed a high frequency of axonal fusion, with the amputated proximal axons reconnected with the distal segment, in gpx-3(tm2059), gpx-5(tm2024), and gpx-8(tm2108) mutants (Fig. 1C–F). This enhanced axonal fusion frequency was not found in gpx-1(tm2100), gpx-4(tm2111), and gpx-7(2166). Instead, a significant number of gpx-1(tm2100) and gpx-4(tm2111) animals displayed debris-like structures around the regenerative growth cone (Fig. 1E, G). These debris-like structures displayed similar morphology as previously reported axon debris, the removal of which by CED-1-mediated cell engulfment is critical for axon regeneration post injury40. The distinct phenotypes in mutants of different gpx genes might be due to differential GPX activities and lipid oxidation levels in these mutants.
To examine whether gpx genes regulate axonal fusion cell-autonomously, we generated touch neuron-specific GPX transgenes in gpx mutant background. The enhanced axonal fusion frequency in gpx-5 mutants was rescued by touch neuron-specific GPX-5 expression (Fig. 1H), indicating a cell-autonomous function of GPX-5. Furthermore, transgenic expression of GPX-1 or GPX-4 was sufficient to rescue the fusion phenotype in gpx-5 mutants (Fig. 1H), suggesting conserved function among GPX paralogs. The axon debris phenotype in gpx-1 and gpx-4 mutants was also rescued by transgenic GPX-1 and GPX-4 expression in touch neurons (Fig. 1I). These data suggest that gpx genes act within touch neurons to impact the injury response. When human GPX4 was expressed in touch neurons, it was sufficient to suppress ML162-induced axonal fusion (Fig. 1J, K). Similarly, the axonal fusion and debris phenotypes in gpx-5 and gpx-1 mutants respectively was suppressed by human GPX4 (Fig. 1L–O), indicating an evolutionally conserved function of GPX protein.
Having shown that mutations of gpx genes promoted axonal fusion, we next tested ferroptosis-inducing agents (Fig. 2A) for their effects on axonal fusionML162, ML210, and RSL3 are small molecules that inactivate GPX4 to induce ferroptosis41. We found that low doses (ML162, ML210 and RSL3) of ferroptosis-inducing agents significantly improved axonal reconnection, with a much higher frequency of regenerated axons fused to the distal fragments (Fig. 2B). This fusion-promoting effect was observed in both reporter strains with either muIs32 or zdIs5 allele (Fig. 2B–D). ML162-induced axonal fusion was also found in ALM mechanosensory neurons (Fig. 2E, F).
To determine whether the ferroptosis-induced axonal fusion is functional, we examined vesicle trafficking across the fusion site. We expressed mCherry-tagged RAB-3 in touch neurons (Pmec-4::mcherry::rab-3) to label synaptic vesicles17. We observed both anterograde and retrograde motion of RAB-3-labelled vesicles across the connection site in the PLM of ML162 treated animals (Fig. 2G), indicating that ferroptosis signaling-induced axonal fusion is functional and can support intracellular transport. We further assessed functional recovery of the fused PLM axons post axotomy using the touch sensitivity assay42. We performed laser axotomy in Day 1 adult animals (24 h post L4 stage) treated with ML162 and tested touch sensitivity 24 h post injury, immediately followed by assessment of axon regrowth status. In the mock control group, in which animals received mock laser surgery but not axotomy, they responded to nearly 100% of light touches at the tail region. In the axotomy group, animals with axon regrowth but no axonal fusion responded to 6 out of 10 posterior gentle touches on average, while animals with axonal fusion showed a significantly improved response rate compared to those without fusion (Fig. 2H). Taken together, these data suggest that ferroptosis-induced axonal fusion is functional and can restore the neural circuit. We noticed that the effects of ferroptosis agents were dose dependent. When we treated the animals with a higher dose of ML162, we observed a reduced fusion rate compared to the lower dose treatment (Fig. 2I–K). In the high dose group, we also observed debris structures around the terminal of a regenerative axon, which was often near the injury site due to the limited axonal regrowth (Fig. 2I–K). The debris structures were similar to those found in gpx-1(tm2100) and gpx-4(tm2111) mutants (Fig. 1D–G). As low dose of ferroptosis-inducing agents mimics the effects of gpx-3(tm2059), gpx-5(tm2024) and gpx-8(tm2108) mutations, while high dose mimics the effects of gpx-1(tm2100) and gpx-4(tm2111) mutations, it’s possible that loss of GPX-3, GPX-5 or GPX-8 leads to a low level of lipid oxidation, and loss of GPX-1 or GPX-4 causes a higher level of lipid oxidation. In both reporter strains (muIs32 and zdIs5), we found that Ferrostatin-1 (Fer-1), an antioxidant agent that traps LOO• and thus can function as a ferroptosis inhibitor43, suppressed axonal fusion induced by GPX4 inhibitors (ML210, ML162, and RSL3) (Fig. 2L–N). Together, these results suggest that a moderate level of ferroptosis signaling promotes axonal fusion, while excessive ferroptosis signaling leads to axonal debris, possibly due to ferroptosis-induced axon degeneration.
We next sought to understand how ferroptosis signaling regulates injury response. N-Acetylcysteine (NAC) is the synthetic precursor of intracellular GSH and is used to increase GPX enzyme activity (Fig. 3A)44,45,46. We observed no effects of NAC treatment on axonal fusion rate or axon debris phenotype in wildtype animals (Fig. 3B, C). We next tested whether inhibiting ferroptosis with NAC can affect fusion or debris in gpx mutants, which showed either enhanced fusion rate or debris frequency (Fig. 1D–G). In gpx-1(tm2100) animals, which normally exhibited high debris rate and low fusion rate, NAC treatment significantly enhanced fusion rate and reduced the percentage of animals showing axonal debris (Fig. 3D, E). These results support a notion that gpx-1(tm2100) has a high level of ferroptotic signaling and lowering the level with NAC results in a moderate degree of lipid oxidation, a condition favoring axonal fusion. Further supporting this notion, we observed that heterozygous gpx-1(tm2100) mutants had a higher fusion rate and a lower debris frequency compared to the homozygous mutants (Fig. 3D, E). We also accessed the effect of NAC in gpx-5(tm2024), which appeared to have a low lipid oxidation level and normally showed a high fusion rate and a low debris rate. We found that NAC treatment was sufficient to lower the fusion rate but not to affect the already low debris rate in gpx-5(tm2024) (Fig. 3F, G). And heterozygous gpx-5(tm2024) mutants also showed a lower fusion rate compared to the homozygous mutants (Fig. 3F, G). Together, these results suggest that an optimal level of ferroptosis signaling is required for a high axonal fusion efficiency. Ferroptosis suppressor protein 1 (FSP1) is a CoQ10 plasma membrane oxidoreductase and protects cells against glutathione-independent ferroptosis (Fig. 3A)47,48. We found that touch neuron-specific overexpression of WAH-1, the ortholog of human AIFM1 and FSP1 (aka AIFM2), inhibited ML162-induced axonal fusion (Fig. 3H, I). Also, the debris rate in gpx-1(tm2100) and axonal fusion rate in gpx-5(tm2024) were decreased by WAH-1 overexpression (Fig. 3H–K), suggesting that axonal fusion and debris phenotypes are tightly controlled by the level of ferroptosis signaling and lipid peroxidation.
The results above supported the hypothesis that injury responses (axon regrowth, axonal fusion and debris) are convertible by manipulating lipid peroxidation level. To test this hypothesis, we first examined the effects of ferroptosis agents and GPX mutations on lipid peroxidation using BODIPY staining. Consistent with its role in promoting lipid peroxidation, treatment with 0.5 µM of ML162 significantly elevated BODIPY intensity, and increasing the dose of ML162 to 1 µM further enhanced BODIPY intensity (Fig. 3L). Despite the elevated lipid peroxidation level, this dose increasement of ML162 abolished the effects of ML162 on promoting axonal fusion (Fig. 2J), suggesting that axonal fusion requires a moderate level of lipid peroxidation. Similarly, loss of GPX-1 or GPX-5 led to increased BODIPY staining, and treatment with ML162 or Fer-1 enhanced or diminished BODIPY intensity in gpx-5 mutants respectively (Fig. 3M). Using the genetically encoded hydrogen peroxide senser Hyper49, we found that gpx-1 and gpx-5 mutants displayed higher levels of hydrogen peroxide in PLM neurons (Fig. 3N). We next examined the effects of Fer-1 and ML162 in wildtype and gpx mutants. In young adult wildtype control, low dose of ML162 treatment enhanced axonal fusion rate and Fer-1 treatment did not show an effect (Fig. 3O, and Supplementary Table 1). Inhibiting ferroptosis with Fer-1 in gpx-1(tm2100) reduced debris rate and enhanced fusion rate, but ML162 treatment did not affect the injury response pattern in gpx-1(tm2100) (Fig. 3P, Q). Similar pattern was found in gpx-4(tm2111) and gpx-7(tm2166), except for that gpx-7(tm2166) did not normally show debris (Fig. 3O, and Supplementary Table 1). Thus, gpx-1(tm2100), gpx-4(tm2111) and gpx-7(tm2166) animals might have a relatively high level of lipid peroxidation. Fer-1 treatment might reduce lipid peroxidation to a moderate level which favors axonal fusion, whereas ML162 treatment might not further enhance the already high lipid peroxidation level. In the cases of gpx-3(tm2059), gpx-5(tm2024) and gpx-8(tm2108), they normally showed high fusion rate and low debris rate (Fig. 1D–G), suggesting a moderate level of lipid peroxidation. Reducing their lipid peroxidation level with Fer-1 decreased fusion rate but did not affect debris rate. Enhancing the lipid peroxidation with ML162 led to decreased fusion rate and increased debris rate (Fig. 3O, R, S, and Supplementary Table 1). This is consistent with the notion that gpx-3(tm2059), gpx-5(tm2024) and gpx-8(tm2108) mutants have a moderate level of lipid oxidation and that either enhancing or reducing the level disrupts axonal fusion. Taken together, these results suggest that an optimal level of lipid peroxidation is required for axonal fusion and that manipulating the level is sufficient to shift the injury response from one to another.
ROS levels increase during normal physical aging due to an increase in oxidation products and a decline in antioxidants defenses50. Exposure to ROS can induce lipid peroxidation, which occurs when ROS oxidizing agents interact with membrane phospholipids51. Given the role of lipid peroxidation in promoting axonal fusion, we examined injury responses in aged neurons. We found that axonal fusion and debris rates were both increased in Adult Day 10 PLM neurons compared to Day 1 (Fig. S1A–C). An enhanced axonal fusion rate was previously observed in aged neurons18. Using HyPer and roGFP reporters, genetically encoded fluorescent sensor for detecting intracellular hydrogen peroxide and ROS respectively49,52, we confirmed that aged neurons possessed a higher level of ROS (Fig. S1D, E). The enhanced axonal fusion in aged neurons was suppressed by ferroptosis inhibitor Fer-1 and by GPX-1 overexpression. It was also suppressed by ferroptosis inducer ML162 (Fig. S1F). For the enhanced axonal debris rate in aged animals, it was suppressed by Fer-1, overexpression of GPX-1 or GPX-5, as well as NAC, the glutathione precursor and an antioxidant, but not by ML162 (Fig. S1G). The reduction of oxidative levels in aged PLM neurons caused by NAC was confirmed by the HyPer sensor (Fig. S1H). These results support that aged neurons possess a moderate-to-high level of lipid ROS, as confirmed by BODIPY staining (Fig. S1I), resulting in higher fusion and debris rates than young neurons (Fig. S1J). Fer-1 and GPX expression can reduce the lipid ROS level and suppress axonal fusion and debris, while ML162 treatments further enhance lipid ROS level, therefore shifting the cells from a status supporting fusion to a status causing debris.
It has been previously reported that phosphatidylserine (PS) externalization in PLM axon after injury serves as an important “save me” signal for the regrowing axon to recognize and connect16. It was further demonstrated that the level of PS exposure correlates with axonal fusion rate18. PS exposure serves as a conserved “eat me” signal for phagocytic uptake of dying cells53. Lipid peroxidation is known to promote PS exposure54,55, and a recent study reported that ferroptosis can induce PS exposure in a cultured human T cell line56. To determine whether ferroptosis signaling can induce induced PS exposure to promote axonal fusion, we visualized touch neuron-specific PS with a mKate2-tagged Annexin V (Pmec-4::mKate2::AnxV). In this system, Annexin V is expressed in touch neurons under the mec-4 promoter. Annexin V can be then translocated across the plasma membrane and secreted via extracellular vesicles, followed by binding to the exposed PS on the outer membrane57. Consistent with previous reports16, we observed mKate2-Annexin V in the distal fragment of the injured PLM of wildtype animals 1 h post injury, while in intact neurons mKate2 signal was detectable only in the cell body (Fig. 4A, B). AnxV was detectable in intact axons of gpx-1(tm2100), and its level was further elevated and significantly higher than that in wildtype after injury (Fig. 4A, B), indicating that ferroptosis signaling can induce PS exposure. We further tested other gpx mutants and found that AnxV levels were increased in all gpx mutants. gpx-1(tm2100), gpx-4(tm2111) and gpx-7(tm2166), which displayed a high debris rate and a low fusion rate (Fig. 1F, G), showed significantly higher AnxV level in injured axons than wildtype (Fig. 4B). In contrast, gpx-3(tm2059), gpx-5(tm2024) and gpx-8(tm2108), which displayed a high axonal fusion rate and a low debris rate (Fig. 1F, G), showed slightly higher level compared to wildtype (Fig. 4B). Therefore, these results suggest that a low level of PS exposure promotes axonal fusion, while a high level of PS exposure leads to axonal debris.
Importantly, we found that AnxV level in gpx mutants could be altered by ML162 or Fer-1 treatment. In general, a high AnxV level could be decreased by Fer-1 and a moderate or low level of AnxV signal could be elevated by ML162 (Fig. S2A–E; Supplementary Table 2). We next sought to evaluate the relationship between the level of PS exposure and axon injury responses by generating heatmaps using average index values of AnxV levels, axonal fusion and debris from wildtype and gpx mutants under different conditions. Our results highlighted that a moderate level of lipid peroxidation leads to moderate PS exposure and promotes axonal fusion, while a high level of lipid peroxidation induces a high level of PS exposure which is associated with the formation of a debris-like structure (Fig. S2F–H).
The exposed PS after axon injury is recognized and bound by PS receptor PSR-1, one of the molecules involved in apoptotic engulfment pathways that recognize the PS ‘eat-me’ signal and mediate the clearance of dying cells by phagocytes1,16,58. PSR-1 functions upstream to EFF-1, the membrane fusogen that mediates the fusion of the two membranes of regrowing axon and the distal axon fragments16. We next tested whether axonal fusion induced by ferroptosis signaling required the engulfment pathway components and EFF-1. We found that ML162 failed to induce axonal fusion in two independent psr-1 mutants (Fig. 4C). Overexpression of PSR-1 was sufficient to increase axonal fusion rate in young animals treated with ML162 (Fig. 4D, E), as well as untreated aged animals (Fig. S1K, L). Similar to psr-1 mutation, eff-1 mutation completely abolished the effect of ML162 in promoting axonal fusion (Fig. 4F, G). It’s known that TTR-52 can bind PS exposed by apoptotic cells59 and it has been previously reported that TTR-52, CED-6, and CED-7, which are components of a phagocytic pathway, are required for injury-induced axonal fusion16. We next tested if ferroptosis-induced axonal fusion is dependent on these components. We observed that in ced-6, ced-7, and ttr-52 mutants, ML162 treatment led to significantly increased fusion rates, even though the fusion rates were lower than that in WT (Fig. S3A–D). Therefore, these data suggest that ferroptosis signaling-induced axonal fusion is predominantly mediated by PSR-1 and EFF-1.
To investigate PSR-1’s role in enhanced fusion under oxidative conditions, we examined its localization in PLM axons. GFP::PSR-1 was predominantly localized in the cell body in intact neurons. One hour post injury, we observed strong GFP::PSR-1 puncta at the severed axon stumps (Fig. 4H, I; Fig S3A, B), and this localization pattern was not affected by ML162 (Fig. 4J). During the active regeneration phase (12-24 hours post-injury), PSR-1 puncta were constantly found at the tips of regenerative growth cones and at the axonal fusion sites (Fig. 4K, L).
Injury-induced PSR-1 puncta formation prompted us to test if PSR-1 undergoes phase-separated condensation. We first tested if PSR-1 puncta could be affected by 1, 6-Hexanediol (1,6-HD), an aliphatic alcohol that has been routinely used to disrupt multivalent hydrophobic protein-protein interactions to disassemble phase-separated droplets60,61. We found that 1,6-HD significantly reduced PSR-1 localization at the tip of injured axons (Fig. 4M). 1,6-HD treatment also abolished the fusion-promoting effects of ML162 at PSR-1 overexpression background (Fig. 4D, E), suggesting that PSR-1 condensation is required for effective axonal fusion.
We next purified recombinant mCherry::PSR-1 protein (Fig. S4A–C) and performed in vitro droplet formation assays. In the presence of PEG8000, purified PSR-1 formed both droplets and aggregates (Fig. 4N, O). The size of the PSR-1 droplets and aggregates correlated with protein concentration (Fig. S4D, E). Fluorescence recovery after photobleaching (FRAP) to assess the liquidity of condensation systems62 observed a slow and partial recovery from both droplets and aggregates (Fig. 4P, Q), indicating that these condensates exhibit gel-like characteristics63.
As PSR-1 is required for ferroptosis signaling-induced axonal fusion, we next tested if PSR-1 condensates were affected by oxidative environment. We performed droplets formation assay in the presence of 5 mM H2O2 and found that H2O2 promoted PSR-1 aggregation, converting the scattered droplets and aggregates into clustered aggregates (Fig. 4N, O). These clustered aggregates display low percentage recovery of FRAP like the scattered droplets and aggregates (Fig. 4P, Q). We further examined the effect of DTT on PSR-1 condensation in vitro. The presence of DTT in the droplet formation buffer greatly diminished PSR-1 droplets and aggregates (Fig. 4N, O). These results suggest that the redox state might regulate PSR-1 condensation and its function in axonal fusion.
Proteins that form phase separated condensates through multivalent interactions usually fall into two categories: those featuring modular domains and those characterized by intrinsically disordered regions (IDRs)64. The CTD of PSR-1 harbors amino acid sequences indicative of an IDR (Fig. S4A). To test if CTD is mediating PSR-1 condensation, we purified the CTD-deleted PSR-1 protein and performed droplet formation assay (Fig. S4C). Unlike full-length PSR-1, which forms both droplets and aggregates, PSR-1(ΔCTD) formed only circular droplets, which were also sensitive to 1,6-HD (Fig. S4D, E). PSR-1(ΔCTD) localization pattern was not affected by ML162 and reduced by 1,6-HD (Fig. S4F). H2O2 transformed dispersed PSR-1(ΔCTD) droplets into aggregated forms, transitioning from a more fluid-like state to a gel-like consistency, as evidenced by FRAP outcomes (Fig. S4G–J). Complete dissolution of PSR-1(ΔCTD) droplets was observed upon treatment with DTT (Fig. S4G, H). Additionally, in contrast to full-length PSR-1, overexpression of PSR-1(∆CTD) not only failed to enhance ML162-induced axonal fusion but also partially inhibited this process (Fig. S4K). The reduced axonal fusion in PSR-1(∆CTD) transgenic animals was further abolished by 1,6-HD. These observations suggest that the CTD of PSR-1 can modulate PSR-1 condensation and function.
Given the important role of a moderate level of PS exposure induced by ferroptosis signaling in promoting axonal fusion, we asked whether other signals that trigger PS exposure can also promote axonal fusion. PS exposure on the outer plasma membrane has long been considered a feature of apoptotic cells, and recent studies have reported that PS exposure is also detected in other types of cell death, including necroptosis and ferroptosis56,65. We first tested whether axonal fusion rate in gpx mutants can be affected by apoptosis or necroptosis inhibitor, similar to ferroptosis inhibitor. We used pan-caspase inhibitor zVAD-fmk and receptor-interacting serine/threonine kinase 1 (RIPK1) kinase activity inhibitor Necrostatin-1 (Nec-1)66,67 to inhibit apoptosis and necroptosis, respectively. As we showed above, ferroptosis inhibitor Fer-1 treatment on gpx-1(tm2100) animals, which normally displayed a high debris rate and a low fusion rate, was sufficient to decrease the debris rate and increase the fusion rate (Fig. 3P, Q). However, zVAD-fmk or Nec-1 treatment did not significantly alter the injury response in gpx-1(tm2100) (Fig. 5A, B). In gpx-5(tm2024), Fer-1 treatment reduced its fusion rate, while zVAD-fmk or Nec-1 treatment did not have an obvious effect on either the fusion rate of the debris rate (Fig. 5C, D). These results suggest that apoptosis or necroptosis signaling might not be involved in axonal fusion regulation. We also examined the effect of an apoptosis-inducing agent, Bisphenol A (BPA)68, on injury response in wildtype animals. BPA treatment reduced the length of axon regeneration, but axonal fusion rate was not enhanced by any tested concentration of BPA. Instead, debris-like structures were observed in animals treated with high concentrations (25 or 50 µM) of BPA (Fig. 5E, F). This is consistent with a previous report that the core apoptotic pathway genes, including CED-9/BCL-1, CED-4/APAF-1 and cell-killing caspase ced-3, are not required for axonal fusion16. Antimycin A is an inhibitor of electron transport from cytochrome b to cytochrome complex III69. Similar to BPA, antimycin A treatment led to reduced PLM axon regrowth length and enhanced axonal debris rate in a concentration-dependent manner (Fig. 5G, H), but did not promote axonal fusion. These data suggest that axonal fusion is specifically induced by ferroptosis signaling, but not other cell death signaling.
Lipid peroxidation can be induced by cellular ROS, among which hydrogen peroxide (H2O2) is a non-radical ROS and a major member of ROS family21. Given that an optimal level of lipid peroxidation is associated with a moderate level of PS exposure to promote axonal fusion (Fig. S2), we next tested whether H2O2 could promote axonal fusion. We found that treating animals with 3 mM H2O2 led to significantly longer axon regrowth (Fig. 5J). This promotive effect of H2O2 in axon regeneration has been reported previously in zebrafish sensory neurons70,71. However, when we treated the animals with a higher H2O2 concentration (30 mM), the regrowth enhancing effect was abolished (Fig. 5J). 3 mM H2O2 treatment also significantly enhanced axonal fusion rate, while 30 mM H2O2 treatment led to axonal debris (Fig. 5I, K). To determine whether the effect of H2O2 on axonal fusion is mediated by lipid peroxidation, we applied H2O2 and Fer-1 simultaneously and examined if blocking lipid peroxidation could block H2O2-induced axonal fusion. We found that axonal fusion induced by 3 mM H2O2 was partially rescued by 0.5 µM Fer-1 and completely rescued by 1 µM Fer-1 (Fig. 5L), indicating that cellular ROS can promote axonal fusion through lipid peroxidation. We next examined mutants of skn-1 gene, the C. elegans ortholog of the key regulator of antioxidant response pathway Nrf272. Consistent with the role of SKN-1/Nrf2 as a transcription factor that regulates expression of Gpx genes and other ferroptosis genes73, a point mutation skn-1(zj15) led to a slightly enhanced axon fusion rate (Fig. 5M, N). However, ML162-induced axonal fusion was partially diminished by the point mutation skn-1(zj15) and fully abolished by the deletion mutation skn-1(ok2315) (Fig. 5M, N). Furthermore, ML162 treatment in skn-1(ok2315) resulted in enhanced axonal debris formation. These results further support that axonal fusion requires a moderate level of ferroptosis signaling and is inhibited by a high level.
Although axonal fusion after injury has not been directly demonstrated in vertebrates to date, there is extensive evidence showing that axonal fusion-like mechanisms can be induced to promote nerve repair in mammals, including humans1. The chemical fusogen polyethylene glycol (PEG) has been used to promote membrane fusion by removing water from the area, thus forcing membranes into close contact74. PEG has been shown to be highly effective in restoring function in transected nerves and PEG treatment has become a common strategy in pre-clinical trials for robust functional recovery1,8,9,75,76. To test if ferroptosis signaling can induce axonal fusion in mammalian neurons, we applied the ferroptosis inducing agent ML162 to transected and re-sutured mouse sciatic nerves and examined the effect on functional restoration (Fig. 6A). We adapted the defined five-step process used in PEG fusion76. First, the mouse sciatic nerve was transected with flat ends. Second, cut ends were then rinsed with Ca2+-free hypotonic saline to prevent plasmalemmal sealing. Third, the ends were re-joined with microsutures. Fourth, ML162 (0.25 µM) and/or PEG (500 mM) solution was applied to the sutured nerve to induce membrane fusion. Fifth, the nerve was rinsed with Ca2+-containing isotonic solution to remove ML162 and/or PEG and to promote vesicle-mediated membrane repair (Fig. S5A). We compared the control and treated animals before and after surgery using various behavioral tests: the pinprick, foot fault (FF) asymmetry, and toe spreading test (Fig. S5B–D)77,78.
The nociceptive sensitivity in the mice was measured by applying a pinprick to the distal skin territory of the injured sciatic nerve. After a sciatic nerve injury, the resuture-only group showed a gradual recovery of nociceptive sensitivity in the skin of the lateral paw. At 27 days post-surgery, the pinprick score has restored to a level comparable to pre-injury (Fig. 6B, and Fig. S6A). PEG treated group displayed a higher pinprick score than the resuture-only control group at most time points, suggesting an improved functional recovery. Similarly, ML162 treated animals showed higher pinprick scores than the control group. Notably, animals treated with both PEG and ML162 displayed the best pinprick scores among the 4 groups (Fig. 6B, and Fig. S6A).
FF is more sensitive to proximal sensory-motor functions and earlier behavioral recovery following sciatic nerve injury78. We found that the FF score of the control group did not improve until 14 days post injury (Fig. 6C, and Fig. S6B). In contrast, either the PEG or ML162 groups showed a significantly increased FF score in the early stage (from Day 4 to Day 10). Like in the pinprick test, the ML162 and PEG co-treatment group showed better FF scores compared to either ML162 or PEG-treated groups, particularly at early stages (Day 4 to Day 10) (Fig. 6C, and Fig. S6B). Considering that the severed nerves are degenerative within 1–3 days (Fig. 6A)79,80, an enhanced behavior restoration score before Day 4 suggests that axonal fusion might have occurred to prevent axon degeneration. We noticed that the FF scores of the resuture-only control group increased rapidly from Day 14, resulting in a gradually reduced difference between the control group and other treatment groups (Fig. 6C, and Fig. S6B). This rapid restoration of physical function after Day 10 is likely due to the effect of axon regeneration, as the axon regrowth rate is about 1-2 mm per day81,82, and it takes roughly 10–14 days for the injured sciatic nerves to regenerate and reinnervate to the target muscles. Therefore, although this evidence is indirect, the functional restoration observed between Day 4 and Day 10 in the ML162-treated groups suggests that ML162 may induce axonal fusion in mice.
We further measured toe spreading reflex to assess motor recovery after the sciatic nerve injury. We found that the ML162 + PEG co-treatment group showed significantly higher toe motor function restoration at the early stages compared to the rest three groups (Fig. 6D, E; Supplementary Video 1), which displayed rapid behavior recovery after Day 14 (Fig. 6D, E; Fig. S6C), likely due to nerve regeneration and reinnervation. The effect of ML162 + PEG on behavior recovery at early stages again suggest that a proper level of ferroptosis signaling may promote functional recovery through axonal fusion.
Having shown that the ferroptosis inducing drug ML162 can promote functional recovery likely through axonal fusion, we next asked if genetically suppressing GPX4 activity could promote axonal fusion and functional restoration. We crossed our GPX4 cKO strain with the Avil-CreER line83,84, and induced GPX4 conditional knockout (cKO) in sensory ganglia by tamoxifen injection at 8 weeks old. Tamoxifen treated GPX4fx/fx;Avil-CreER mice died within 3 weeks after tamoxifen injection, indicating that GPX4 expression in sensory neurons are essential for survival. Despite the lethal phenotype 3 weeks after GPX4 ablation, we observed improved behavior restoration in GPX4fx/fx;Avil-CreER mice after SNT and resuture as measured by FF asymmetry and Toe spreading test (Fig. 6F–I). This improved physical function in GPX4 cKO mice within 4 to 10 days after nerve transection further supports that manipulating GPX4 level may induce axonal fusion to promote functional recovery.
The positive effect of PEG and ML162 on functional restoration prompted us to test if they had an impact on axon regeneration. We first assessed DRG axon regrowth after the behavior tests were completed (30 days post SNT and resuture) by staining for the regeneration marker SCG1085,86. As expected, SCG10-positive axons were present throughout the sciatic nerve and we did not observe any significant difference among the 4 groups: resuture only control group, PEG treated group, ML162 treated group, and PEG + ML162 co-treatment group (Fig. S7A, B). We next measured DRG axon regeneration 3 days post injury and observed SCG10 positive axons around the injury sites (Fig. 7A), indicating limited axon regeneration at this time point. PEG and/or ML162 treatment did not affect axon regeneration, as SCG10 intensity was at a similar level for all 4 groups (Fig. 7A, B). This further supports that the effect of PEG and ML162 on promoting functional recovery is through axonal fusion but not axonal regrowth.
Previous studies have shown that distal axon fragments of mammalian DRG neurons degenerate within 1-3 days after axon injury, resulting in disruption of neuromuscular junctions (NMJ)79,80. After nerve injury, the regenerative axons will reinnervate muscles, but the reinnervation do not occur until two weeks post injury due to the rate limit of axon regeneration87. If the enhanced functional recovery in ML162 and/or PEG treated mice within the first week post injury is due to axonal fusion, we would expect to see innervated NMJs within the first week. We therefore examined the NMJs structure at 3 days after transection and resuture using anti-neurofilament (NF200) and α-Bungarotoxin (BTX) to label axon and postsynaptic membrane respectively. The resuture only control group showed no innervated NMJs on abductor digiti minimi lateral plantar muscles, indicating that motor neurons have completely degenerated 3 days post-operation (Fig. 7C–E). Very few innervated NMJs were detected in the PEG or ML162 treated groups. However, significantly more innervated NMJs were observed in the ML162 + PEG group and innervated NMJs were detected in the plantar muscles of all individual animals treated with ML162 and PEG (Fig. 7C–E). As expected, we detected full restoration of motor end plate reinnervation at Day 30, as a result of axon regeneration (Fig. S7A–C). The presence of innervated NMJs at Day 3 is consistent with the enhanced functional recovery in this group during the early period (Fig. 6). While we have not ruled out the possibility that ML162 may delay Wallerian degeneration, our results support that the combined treatment of ML162 and PEG prevents axon degeneration and NMJ denervation, possibly by promoting axonal fusion (Fig. 7F).
