Microglia deploy TAM receptors to kill motor neurons in a mouse model of amyotrophic lateral sclerosis Authors: Youtong Huang, Ananya Mavinkurve, Bristy Sabikunnahar, Beth Stevens, Greg Lemke Published: 2026 ABSTRACT Activation of microglia is a prominent feature of amyotrophic lateral sclerosis (ALS), a neurodegenerative disease that leads to the death of motor neurons. A key component of this activation is elevated expression of the TAM receptor tyrosine kinases Axl and Mer (gene name Mertk). Here we show that germline and microglial-restricted inactivation of the Axl and Mertk genes in the SOD1G93A mouse model of ALS leads to an extension of lifespan, which is tied to the preservation of cholinergic motor neurons and neuromuscular synapses. Also elevated on SOD1G93A neuronal surfaces is the essential TAM co-ligand phosphatidylserine, a potent ‘eat-me’ signal through which apoptotic cells are engulfed by microglia. Correspondingly, we find that microglial lysosomes are filled with the remains of cholinergic neurons in the SOD1G93A spinal cord, whereas this accumulation is markedly reduced in the SOD1G93AAxl-/-Mertk-/- cord. Together, these results suggest that microglia phagocytically kill living neurons, and thereby hasten death in ALS. Introduction Amyotrophic lateral sclerosis is a progressive, invariably fatal neurodegenerative disease that results from the death of motor neurons1, with a global incidence estimated at ~1.7 per 100,000 person-years of follow-up2. ALS patients exhibit upper- and lower-motor-neuron dysfunction and loss, with progressive weakness and eventual paralysis of the muscles controlling limb movement, swallowing, speaking, and respiration. There is a strong genetic contribution to the development of ALS, but most forms are sporadic rather than inherited3. Familial cases of ALS account for ~10% of diagnoses, with pathogenic mutations in the C9orf72, SOD1, FUS, and TARDBP genes representing, in order, the most common causes of the inherited disease1,3. Although ALS research has historically focused on the health and survival of motor neurons, dysfunction in other central nervous system (CNS) cells contributes to pathogenesis. Among the most significant of these are microglia, the tissue-specific macrophages of the CNS4–6. Recent transcriptomic analyses have demonstrated that microglia are strongly activated in human ALS7, and in mouse models of the disease, microglial proliferation and inflammation - as assessed by elevated expression of interleukin (IL)−1β, tumor necrosis factor (TNF)α, and other markers - have been consistently described5,6,8. Two members of the TAM family of receptor tyrosine kinases (RTKs), comprised of Tyro3, Axl, and Mer9,10, are especially prominent markers of activated microglia and other macrophages. All non-activated microglia express readily detectable levels of Mer but very low levels of Axl11,12. Restricted subsets of dendritic cells express Tyro313, but microglia do not11,12. Upon infection by viruses or bacteria, or after exposure to inflammatory cytokines, or in the inflammatory environments that characterize Alzheimer’s and Parkinson’s disease, Axl levels are dramatically elevated in microglia, macrophages, and dendritic cells11,12,14–16, and up-regulated Axl is now universally recognized as an Alzheimer’s disease (AD) biomarker17–19. In each of these settings, Axl and Mer, acting in concert, serve two critical functions: they are required for the detection and phagocytic engulfment of apoptotic cells11,14,20,21, and concomitantly, they act as intrinsic feedback inhibitors of the innate immune response to dampen inflammation10,16. Recent analyses have suggested that components of the TAM system may also be up-regulated in microglia in sporadic ALS and in an inherited form of the disease caused by mutation of the C9orf72 gene22. We investigated this question in both ALS and a mouse model of the disease, and documented pronounced engagement of the TAM system in both settings. When we asked how TAM deletion might influence the course of disease in the mouse model, we observed an extension of lifespan following germline- or microglial-restricted ablation of Mertk and Axl. Further analyses indicated that this lifespan extension was correlated with diminished phagocytic engulfment of diseased cholinergic neurons by microglia. Results TAM system expression during disease We used immunohistochemical (IHC) analyses to examine the expression of AXL and MER protein in ionized calcium binding adapter-positive (IBA1+) microglia in postmortem sections of the lumbar spinal cord of six sporadic ALS patients and three age-matched human controls (Fig. 1a). (See Supplementary Table 1 for patient evaluation.) We quantified a modest 3-fold increase in microglial MER in ALS relative to non-diseased control sections, but a 16-fold increase in microglial AXL in the same sections (Fig. 1b).Fig. 1TAM system engagement in ALS and SOD1G93A mice.a Representative cross sections of the human lumbar spinal cord from age-matched controls (left) and ALS patients (right) stained with IBA1, AXL, and MER antibodies. b Quantification of MER (left) and AXL expression (right) in IBA1+ microglia within lumbar ventral horn from human spinal cord of n = 3 control and n = 6 ALS patients. c Representative cross sections of SOD1G93A lumbar ventral horn at disease stage 3, stained with Iba1, Axl, Mer, and ChAT antibodies. d Representative flow cytometric analyses of Axl expression in WT (left, black) and SOD1G93A (right, red) microglia at day 160. e Quantification of Mer (left) and Axl expression (right) in CD45int CD11b+ GR1- CD206- Cx3cr1+ cells from spinal cords of n = 4 WT and SOD1G93A mice. f Representative sections of human lumbar spinal cord from age-matched controls (left) and ALS patients (right) stained with IBA1 (microglia or µglia) and ChAT (motor neurons or MN) antibodies, and probed for GAS6 mRNA using RNAscope. g Quantification of fluorescence intensity of GAS6 mRNA in nuclei of indicated cells of the lateral motor column of the lumbar ventral horn from n = 3 control and n = 3 ALS patients. Increase in GAS6 in ChAT+ cells between ALS and controls is statistically significant (p = 0.0042). h Representative sections of WT (top row) and SOD1G93A lumbar spinal cord at disease stage 3 (bottom), stained with Gas6, NeuN, and Iba1 antibodies, repeated independently ≥3 times. i Representative sections of ventral horn of WT (top row) and SOD1G93A lumbar spinal cord injected with PSVue (surface PtdSer) intrathecally at score 3 (bottom) and stained with antibodies to ChAT and NeuN. j Representative cross-section of SOD1G93A lumbar spinal cord at score 3 injected with PSVue and pSIVA (surface PtdSer), and stained with Neurotrace, and antibodies to cCasp3 (apoptotic cells) and Iba1. The arrow indicates a single apoptotic cell. All large cells with externalized PtdSer are cCasp3-. Experiment with PSVue was independently repeated four times yielding similar results. P values calculated using two-sided Welch’s t-test (b, e, g comparison in ChAT+ cells between ALS and control), and one-way ANOVA with Dunnett’s multiple comparison test (g comparison between ChAT+ cells, Iba1+ microglia, and other cell types). In all figures, p values are only indicated for statistically significant differences. Data are mean values ± SD. Source data are provided as a source data file. Scale bars: 50 μm. TAM system engagement in ALS and SOD1G93A mice. a Representative cross sections of the human lumbar spinal cord from age-matched controls (left) and ALS patients (right) stained with IBA1, AXL, and MER antibodies. b Quantification of MER (left) and AXL expression (right) in IBA1+ microglia within lumbar ventral horn from human spinal cord of n = 3 control and n = 6 ALS patients. c Representative cross sections of SOD1G93A lumbar ventral horn at disease stage 3, stained with Iba1, Axl, Mer, and ChAT antibodies. d Representative flow cytometric analyses of Axl expression in WT (left, black) and SOD1G93A (right, red) microglia at day 160. e Quantification of Mer (left) and Axl expression (right) in CD45int CD11b+ GR1- CD206- Cx3cr1+ cells from spinal cords of n = 4 WT and SOD1G93A mice. f Representative sections of human lumbar spinal cord from age-matched controls (left) and ALS patients (right) stained with IBA1 (microglia or µglia) and ChAT (motor neurons or MN) antibodies, and probed for GAS6 mRNA using RNAscope. g Quantification of fluorescence intensity of GAS6 mRNA in nuclei of indicated cells of the lateral motor column of the lumbar ventral horn from n = 3 control and n = 3 ALS patients. Increase in GAS6 in ChAT+ cells between ALS and controls is statistically significant (p = 0.0042). h Representative sections of WT (top row) and SOD1G93A lumbar spinal cord at disease stage 3 (bottom), stained with Gas6, NeuN, and Iba1 antibodies, repeated independently ≥3 times. i Representative sections of ventral horn of WT (top row) and SOD1G93A lumbar spinal cord injected with PSVue (surface PtdSer) intrathecally at score 3 (bottom) and stained with antibodies to ChAT and NeuN. j Representative cross-section of SOD1G93A lumbar spinal cord at score 3 injected with PSVue and pSIVA (surface PtdSer), and stained with Neurotrace, and antibodies to cCasp3 (apoptotic cells) and Iba1. The arrow indicates a single apoptotic cell. All large cells with externalized PtdSer are cCasp3-. Experiment with PSVue was independently repeated four times yielding similar results. P values calculated using two-sided Welch’s t-test (b, e, g comparison in ChAT+ cells between ALS and control), and one-way ANOVA with Dunnett’s multiple comparison test (g comparison between ChAT+ cells, Iba1+ microglia, and other cell types). In all figures, p values are only indicated for statistically significant differences. Data are mean values ± SD. Source data are provided as a source data file. Scale bars: 50 μm. We also measured Axl and Mer expression in the SOD1G93A mouse model of ALS23. These mice overexpress a glycine93-to-alanine variant of human Cu/Zn superoxide dismutase 1 (SOD1), which causes a dominantly inherited form of familial ALS24. Although there are now multiple animal models of ALS, SOD1G93A mice were the first transgenic mouse model to be produced23,25 and are to date the most studied. They accurately recapitulate and reliably develop many features of ALS, including motor neuron death, muscle weakness, and paralysis. We staged phenotypic progression in these mice using consensus criteria relating to hindlimb paralysis26,27. Score 0 mice were pre-symptomatic; score 1 mice (‘symptomatic’) exhibited an acute hindlimb angle and tremor upon suspension but a normal gait; score 2 mice (‘disease onset’) exhibited hindlimb collapse toward the midline of at least one hindlimb upon suspension and an altered gait; score 3 mice (‘hindlimb paralysis’) additionally exhibited paralysis of at least one hindlimb but were able to right themselves; and clinical endpoint mice exhibited paralysis of both hindlimbs and were unable to right themselves within 20 seconds. On western blots of extracts of total spinal cord (Supplementary Fig. 1a) or of cervical, thoracic, or lumbar cord (Supplementary Fig. 1b), the expression of Axl was seen to increase markedly as disease progressed in SOD1G93A mice. Mer was readily detected in pre-symptomatic mice, but also increased toward the disease endpoint. This pattern of TAM receptor regulation during disease is very similar to the pattern we previously documented in the brains of the APP/PS1 mouse model of AD12 and in the spinal cord of the Thy1-SNCAA53T (SNCAA53T) mouse model of Parkinson’s disease (PD)11. Within the lumbar cord of SOD1G93A mice at score 3, IHC analyses indicated that Iba1+ microglia were, as for human ALS, the most prominent sites of Axl and Mer expression (Fig. 1c). In contrast, neither choline acetyltransferase (ChAT)+ neurons (Fig. 1c) nor activated GFAP+ astrocytes (Supplementary Fig. 1c) expressed either Axl or Mer. Given this, we performed flow cytometric analyses on dissociated spinal cords after removing the leptomeninges, gating on CD45int CD11b+ GR1- CD206- Cx3cr1+ cells for resident microglia (Supplementary Fig. 1d), and again demonstrated that SOD1G93A parenchymal microglia elevate Axl at late disease stage (Fig. 1d, e). An expansion of the total microglial population in the SOD1G93A spinal cord - as monitored by the number of Iba1+ cells (Supplementary Fig. 1e), or GFP+ cells in SOD1G93A mice crossed into the Cx3cr1GFP/+ microglial reporter line11,28 (Supplementary Fig. 1f) - was also observed at disease score 2 or postnatal day 120, as previously documented by others29. As noted above, the pronounced up-regulation of microglial Axl we detect in the setting of both ALS and the SOD1G93A mouse model is consistent with a large body of literature in other neurodegenerative diseases. We also examined the expression of Gas6, the ligand that binds and activates all TAM receptors10,30. The other TAM ligand, Protein S (Pros1), binds and activates Mer and Tyro3 but not Axl10,30. Many earlier studies have shown that the Gas6 gene is widely expressed in the wild-type (WT) mouse brain12,31,32, and Gas6 protein expression in microglia and macrophages has been found to be elevated as a consequence of Axl up-regulation12,14. We used RNAscope in situ hybridization (see Methods) to examine and quantify the expression of human GAS6 mRNA in several cell types of the lumbar spinal cord in both ALS and age-matched controls. These analyses revealed that this mRNA was again widely distributed, but was most prominently expressed in large ChAT+ neurons (Fig. 1f, g), with a significant increase in ALS versus control neurons (Fig. 1g). In the WT mouse lumbar spinal cord, IHC analyses revealed that Gas6 protein was prominently expressed by NeuN+ neurons, and was strongly up-regulated in both neurons and microglia in the SOD1G93A cord at score 3 (Fig. 1h). These results are in alignment with recent single-nucleus transcriptomic and chromatin accessibility analyses that strongly implicate TAM signaling from motor neurons (Gas6) to microglia (Axl and Mer) in the development of late-stage disease in SOD1G93A mice33. Finally, we assessed the expression of an essential co-ligand for TAM receptor signaling – the phospholipid phosphatidylserine (PtdSer)34. This phospholipid is entirely confined to the cytoplasmic leaflet of the plasma membrane bilayer of healthy cells, but is translocated onto the extracellular surface of diseased, infected, apoptotic, or depolarized cells through the action of a series of phospholipid scramblases21,35,36 We used intrathecally administered PSVue, a fluorescent small molecule that binds to PtdSer37, to show that externalized PtdSer was undetectable on the surface of any cell in the WT mouse spinal cord, but was prominently expressed on the surface of many cells in the SOD1G93A cord late in disease (Fig. 1i). These included the large ChAT+ and ChAT+NeuN+ motor neurons that are preferentially vulnerable in ALS38. We found that the PSVue+ cells, which were Neurotrace+, were also stained after intrathecal injection of pSIVA39, a fluorescent Annexin XII-derived protein that also binds externalized PtdSer (Fig. 1j). Notably, most PtdSer+ cells were not stained with antibodies to cleaved caspase 3, a definitive marker for apoptotic cells (Fig. 1j). Thus, the vast majority of motor neurons in the late-stage SOD1G93A cord that display externalized PtdSer are alive. The fact that PtdSer is externalized on live cells in the SOD1G93A cord is central to the interpretation of the phenotypes described below, since full TAM receptor activation only occurs when the amino-terminal ‘Gla’ domains of the ligands Gas6 and Protein S bind to this phospholipid34. Two additional properties of PtdSer are important with regard to these phenotypes: first, PtdSer is the most potent and universal of the ‘eat-me’ signals by which apoptotic cells are recognized by microglia and other phagocytes21; and second, its externalization targets select cells for engulfment or dismemberment even if these cells are still alive11,40–42. TAM regulation of disease in mice The above results demonstrate that the critical components of TAM signaling are present in the spinal cord in ALS and its SOD1G93A mouse model. We therefore assessed TAM action in the development of disease in the ALS mouse model by first crossing germline Axl-/- and Mertk-/- mouse mutants43 into SOD1G93A mice on a C57BL/6 background, and asking how long it took the compound mutants to reach each disease score compared with SOD1G93A alone. We analyzed large populations (21-67 mice, depending on genotype) of males and females, ~50% each, since prior studies have established equivalent disease development as a function of sex in SOD1G93A C57BL/6 mice44,45. These analyses revealed effects of the Axl and Mertk mutations that displayed a striking divergence during the course of the disease. SOD1G93AAxl-/-Mertk-/- mice reached disease score 1 (hindlimb tremor and weakness) ~30 days earlier than their SOD1G93A counterparts (Fig. 2a). This is not surprising, since Axl-/-Mertk-/- double mutants are already very sick: they develop broad-spectrum autoimmunity46, splenomegaloly14,46, liver disease47, reduced male fertility43, retinal degeneration48, and chronic immune inflammation throughout the body16,49 prior to any SOD1 mutation. The SOD1G93AAxl-/-Mertk-/- mice also reached disease score 2 (disease onset) at an earlier average age than SOD1G93A mice, but the difference was less pronounced (Fig. 2b). When we examined the age to reach score 3 (hindlimb paralysis), there was a trend for the positions of the two genotypes to be reversed (Fig. 2c). And when we examined the age to reach clinical endpoint, at which humane practice required the mice to be sacrificed, we saw that this occurred on average three weeks later in SOD1G93AAxl-/-Mertk-/- mice compared to SOD1G93A alone (Fig. 2d). Thus, removing Axl and Mer from the SOD1G93A mice made them sick sooner, but then allowed these sick mice to live appreciably longer.Fig. 2Effect of germline Axl and Mertk mutations on disease phenotype in SOD1G93A mice.a–d Time, in postnatal days, to reach score 1 (a), score 2 (b), score 3 (c), and clinical endpoint (d), expressed as percent of total population, for SOD1G93A mice (black curves) and SOD1G93AAxl-/-Mertk-/- mice (green curves). N = 67, 59, 31, and 27 for SOD1G93A mice, and 47, 53, 28, and 21 for SOD1G93AAxl-/-Mertk-/- mice, for score 1, 2, 3, and the endpoint, respectively. P values calculated using the log-rank test. e Quantification of choline acetyltransferase-expressing (ChAT+) motor neurons in the lateral motor column of the ventral horn of 25 μm hemi-sections of the lumbar (L3-5) spinal cord, at the indicated postnatal days (d) in mice of the indicated genotypes. Data points are numbers of ChAT+ neurons per mouse averaged from ≥ 3 L3-5 ventral horn hemi-sections per mouse from n = 5 mice for both SOD1G93A and SOD1G93AAxl-/-Mertk-/- groups at day 80, and from n = 3 mice for SOD1G93A, n = 5 for SOD1G93AAxl-/-Mertk-/-, n = 5 for WT, and n = 3 for Axl-/-Mertk-/- groups at day 160. P values calculated using one-way ANOVA with Dunnett’s multiple comparison test. Data are mean values ± SD. Source data are provided as a source data file. Effect of germline Axl and Mertk mutations on disease phenotype in SOD1G93A mice. a–d Time, in postnatal days, to reach score 1 (a), score 2 (b), score 3 (c), and clinical endpoint (d), expressed as percent of total population, for SOD1G93A mice (black curves) and SOD1G93AAxl-/-Mertk-/- mice (green curves). N = 67, 59, 31, and 27 for SOD1G93A mice, and 47, 53, 28, and 21 for SOD1G93AAxl-/-Mertk-/- mice, for score 1, 2, 3, and the endpoint, respectively. P values calculated using the log-rank test. e Quantification of choline acetyltransferase-expressing (ChAT+) motor neurons in the lateral motor column of the ventral horn of 25 μm hemi-sections of the lumbar (L3-5) spinal cord, at the indicated postnatal days (d) in mice of the indicated genotypes. Data points are numbers of ChAT+ neurons per mouse averaged from ≥ 3 L3-5 ventral horn hemi-sections per mouse from n = 5 mice for both SOD1G93A and SOD1G93AAxl-/-Mertk-/- groups at day 80, and from n = 3 mice for SOD1G93A, n = 5 for SOD1G93AAxl-/-Mertk-/-, n = 5 for WT, and n = 3 for Axl-/-Mertk-/- groups at day 160. P values calculated using one-way ANOVA with Dunnett’s multiple comparison test. Data are mean values ± SD. Source data are provided as a source data file. Since degeneration of ventral horn motor neurons is a key pathological feature of ALS, we used antibodies to ChAT to quantify the numbers of these cells in lumbar hemi-sections at 80 and 160 days in the SOD1G93A and SOD1G93AAxl-/-Mertk-/- mice. We chose these ages because there was a large difference between genotypes to reach score 1 at day 80, with SOD1G93AAxl-/-Mertk-/- leading (Fig. 2a, dotted line), while there was an even larger difference between genotypes to reach clinical endpoint at day 160, with SOD1G93AAxl-/-Mertk-/- trailing (Fig. 2d, dotted line). We measured no difference in the number of ChAT+ cells in the ventral cord between genotypes at day 80 (Fig. 2e), indicating that the accelerated time at which SOD1G93AAxl-/-Mertk-/- mice first displayed symptoms was not due to an accelerated loss of these neurons. In contrast, at day 160, SOD1G93AAxl-/-Mertk-/- mice had approximately three times more motor neurons remaining in the lumbar spinal cord than their SOD1G93A counterparts (Fig. 2e). We further stratified ChAT+ neurons based on their expression of matrix metalloproteinase 9 (MMP9) and NeuN, since in the mouse NeuN-MMP9-, NeuN+MMP9-, and NeuN+MMP9+ cells mark slow or gamma (mostly ALS-resistant), fast-fatigue-resistant, and fast-fatigable (the most ALS-vulnerable) motor neurons, respectively38,50,51. At day 160, we found that the number of each of these motor neuron subtypes reflected that seen for the summed ChAT+ population (Supplementary Fig. 2a). Thus, the extended lifespan of SOD1G93AAxl-/-Mertk-/- mice relative to SOD1G93A was tied to the extended survival of all lower motor neuron subtypes in the lumbar spinal cord. We detected two additional features of Axl and Mertk mutations in the context of SOD1G93A. The first was a reduced level of microgliosis in the lumbar ventral spinal cord, as assessed by expression of Iba1 (Supplementary Fig. 2b) and the lysosomal marker CD68 (Supplementary Fig. 2c). The up-regulation of these proteins is indicative of phagocytic activation in microglia. At day 160, Iba1 levels in the ventral horn of the spinal cord were reduced ~47% in SOD1G93AAxl-/-Mertk-/- mice relative to SOD1G93A (Supplementary Fig. 2b), and CD68 levels were reduced ~57% in the same comparison (Supplementary Fig. 2c). We did not detect any substantial change in the expression of mRNAs encoding potentially neurotoxic cytokines or inflammatory regulators between SOD1G93AAxl-/-Mertk-/- and SOD1G93A spinal cords at day 80 or 120, (Supplementary Fig. 2d). The day 120 and day 160 SOD1G93A spinal cords were clearly inflamed relative to non-diseased mice, particularly with respect to the expression of Il1b and Tnfa, but no inflammatory regulator showed a statistically significant difference between the SOD1G93A and SOD1G93AAxl-/-Mertk-/- cords at day 120. A modest but significant decrease between these genotypes was observed for Tnfa and iNOS mRNAs at day 160 (TNF-α and iNOS are neurotoxic in many settings) that was balanced by an equally modest decrease in mRNA encoding TGFβ, which is often neuroprotective (Supplementary Fig. 2d). These findings are in agreement with our prior single-cell RNA sequencing analyses of WT versus Axl-/-Mertk-/- microglia in APP/PS1 mice12. We also observed reduced expression of a higher molecular weight aggregate of mutant SOD1, a presumed homodimer, in SOD1G93AAxl-/-Mertk-/- mice. Expression of this dimer, which is observed in SOD1G93A mice and in familial ALS patients with SOD1 mutations52,53 (Supplementary Fig. 2e), was reduced 4- to 5-fold at day 80, 120, and 160 in SOD1G93AAxl-/-Mertk-/- relative to SOD1G93A mice (Supplementary Fig. 2e, f). Gas6 protein levels in the ventral horn of the day 160 lumbar spinal cord were maintained in NeuN+ neurons, but Gas6 protein was markedly reduced in Iba1+ microglia in SOD1G93AAxl-/-Mertk-/- mice (Supplementary Fig. 2g), consistent with many prior analyses indicating that Gas6 protein expression in all tissue macrophages is dependent on Axl expression12,14. The density of GFAP+ cells in the ventral horn of the day 160 spinal cord, and the expression of GFAP and Aldh1l1 per cell were unchanged between SOD1G93A and SOD1G93AAxl-/-Mertk-/- mice (Supplementary Fig. 2h). TAM regulation of neuromuscular junctions We next asked whether the effects of TAM mutation on motor neuron numbers in the mouse spinal cord were reflected at the neuromuscular junctions (NMJs) formed by these neurons in different hindlimb muscles. We visualized post-synaptic densities in muscle flat mounts at day 160 with fluorescent alpha-bungarotoxin (αBTX), which binds tightly to the acetylcholine receptors that are highly clustered in these structures, and visualized presynaptic terminals with antibodies to synaptophysin. Representative co-staining with these reagents in the quadriceps and gastrocnemius muscles in SOD1G93A, SOD1G93AAxl-/-Mertk-/-, and WT mice at day 160 are shown in Fig. 3a. We assessed overlap between synaptophysin and αBTX (example in Fig. 3b) and used an unbiased FIJI-based analysis (see Methods) to assign NMJs to one of three categories: complete, in which 90–100% of the post-synaptic αBTX signal area was covered with pre-synaptic synaptophysin; partial, in which 10-90% of αBTX signal overlapped with synaptophysin; and empty, in which <10% of the αBTX cluster was covered with pre-synaptic synaptophysin. Consistent with the large loss of ChAT+ cells that we quantified in the ventral horn of the lumbar spinal cord of day 160 SOD1G93A mice (Fig. 2e), 90% of NMJs in the quadriceps and 75% of those in the extensor digitorum longus (EDL) of these mice were scored as empty (Fig. 3c, black bars). None of the NMJs in the quadriceps and only 10% in the EDL were complete (Fig. 3c). In contrast, 85% and 80% of the NMJs in the WT quadriceps and EDL, respectively, were scored as complete at day 160 in WT (Fig. 3c, white bars). Crossing the Axl-/-Mertk-/- double mutation into the SOD1G93A mice yielded an obvious phenotypic rescue: now, ~45% of the NMJs in quadriceps and ~35% of the NMJs in the EDL at day 160, respectively, were scored as complete (Fig. 3c, green bars).Fig. 3TAM mutation rescues neuromuscular junction (NMJ) loss and muscle dystrophy in late-stage SOD1G93A mice.a Representative flatmount staining of day 160 quadriceps (quad, top row) and gastrocnemius (gastroc, bottom row) muscles with anti-synaptophysin (syn, green) as a pre-synaptic marker and alpha bungarotoxin (αBTX, magenta) as a post-synaptic marker. b Left, Representative superimposed syn and αBTX images for a single NMJ (white signal is overlap); and right, masks of this image generated by Fiji. c Fraction of 160 day NMJs in the quadriceps (top) and extensor digitorum longus (EDL, bottom) muscles scored as empty (E, no synapse), partial (P), or complete (C, normal synapse) in mice of the indicated genotypes. N = 7 and 5 for SOD1G93A, n = 3 and 3 for SOD1G93AAxl-/-Mertk-/-, and n = 5 and 3 for WT mice for upper and bottom panels, respectively. Each data point is the average of 3-4 fields of view of the indicated muscle type from a separate mouse. P values calculated using two-way ANOVA with Tukey’s multiple comparison test. d Representative immunostaining of quadriceps cross-sections from day 160 mice of the indicated genotypes with anti-laminin. e Quantification of results in (d). Data points are muscle diameter averaged from n = 5 for SOD1G93A, n = 3 for SOD1G93AAxl-/-Mertk-/- and n = 4 for WT mice. P values calculated using two-way ANOVA with Dunnett’s test. Data are mean values ± SD. Source data are provided as a source data file. Scale bars: a 50 μm; b 10 μm; c 100 μm. TAM mutation rescues neuromuscular junction (NMJ) loss and muscle dystrophy in late-stage SOD1G93A mice. a Representative flatmount staining of day 160 quadriceps (quad, top row) and gastrocnemius (gastroc, bottom row) muscles with anti-synaptophysin (syn, green) as a pre-synaptic marker and alpha bungarotoxin (αBTX, magenta) as a post-synaptic marker. b Left, Representative superimposed syn and αBTX images for a single NMJ (white signal is overlap); and right, masks of this image generated by Fiji. c Fraction of 160 day NMJs in the quadriceps (top) and extensor digitorum longus (EDL, bottom) muscles scored as empty (E, no synapse), partial (P), or complete (C, normal synapse) in mice of the indicated genotypes. N = 7 and 5 for SOD1G93A, n = 3 and 3 for SOD1G93AAxl-/-Mertk-/-, and n = 5 and 3 for WT mice for upper and bottom panels, respectively. Each data point is the average of 3-4 fields of view of the indicated muscle type from a separate mouse. P values calculated using two-way ANOVA with Tukey’s multiple comparison test. d Representative immunostaining of quadriceps cross-sections from day 160 mice of the indicated genotypes with anti-laminin. e Quantification of results in (d). Data points are muscle diameter averaged from n = 5 for SOD1G93A, n = 3 for SOD1G93AAxl-/-Mertk-/- and n = 4 for WT mice. P values calculated using two-way ANOVA with Dunnett’s test. Data are mean values ± SD. Source data are provided as a source data file. Scale bars: a 50 μm; b 10 μm; c 100 μm. We also stained cross-sections of the day 160 quadriceps with an antibody to laminin, a component of the extracellular matrix surrounding muscle fibers, in order to quantify fiber diameter as a measure of muscle dystrophy in each of the genotypes (Fig. 3d). Skeletal muscle fibrosis, marked by laminin accumulation, is commonly seen in ALS subsequent to pathological denervation54. Fiber diameter was reduced from ~57 μm in WT to ~27 μm in SOD1G93A mice (Fig. 3e, left). It was restored to ~52 μm in SOD1G93AAxl-/-Mertk-/- mice (Fig. 3e, left). Correspondingly, we detected a trend toward WT laminin staining in cross sections of quadriceps of SOD1G93AAxl-/-Mertk-/- mice compared to SOD1G93A at day 160 (Fig. 3e, right), consistent with a partial rescue of muscle wasting in the absence of Axl and Mer. Microglia-restricted TAM inactivation in SOD1 mice The above phenotypes were observed when germline Axl and Mertk mutations43 were crossed into the SOD1G93A ALS model, but Axl and Mer are expressed in many cells in addition to microglia. Moreover, the Mertk mutation we have used43,55 has heretofore been shown to alter the expression of genes that are closely linked to Mertk in the mouse genome56–58. To more selectively inactivate Mertk and Axl in adult microglia, we exploited conditional floxed alleles of these genes11,59 together with a tamoxifen-inducible Cx3cr1CreER driver that is active in all microglia11,40,60. This driver is also active in peripheral monocytes that infiltrate the CNS in neurodegenerative disease61 and in border-associated macrophages62. We have previously validated both sets of reagents, but we nonetheless confirmed that GFP-expressing cells in the SOD1G93ACx3cr1GFP/+ spinal cord were Iba1+Mer+ microglia (Supplementary Fig. 3a), and that Axl and Mer expression was lost in these cells for >3 months after tamoxifen treatment of SOD1G93ACx3cr1CreER/+Axlf/fMertkf/f mice at day 40 (Supplementary Fig. 3b). As expected, when we injected SOD1G93ACx3cr1CreER/+Axlf/fMertkf/f mice with corn oil control at day 100 (tamoxifen is diluted in corn oil), we observed a time course to reach clinical endpoint that resembled that of SOD1G93A alone (Fig. 4a). 80% of SOD1G93A mice had reached clinical endpoint by day 160 (Fig. 2d), and 60% of oil-treated SOD1G93ACx3cr1CreER/+Axlf/fMertkf/f mice had reached this same stage by the same time (Fig. 4a). However, when we injected SOD1G93ACx3cr1CreER/+Axlf/fMertkf/f mice with tamoxifen (tmx) at day 100 to inactivate TAM receptor genes in microglia, we observed a time course to reach clinical endpoint that resembled that of SOD1G93AAxl-/-Mertk-/- mice (Fig. 4a). Only ~5% of the latter had reached clinical endpoint at day 160 (Fig. 2d), and no mice in the day 100 tmx-injected SOD1G93ACx3cr1CreER/+Axlf/fMertkf/f population had reached endpoint by day 160 (Fig. 4a). Thus, germline inactivation of the Axl and Mertk genes and conditional inactivation of these genes in Cx3cr1+ cells in adult individuals yielded comparable extensions of lifespan.Fig. 4Selective inactivation of Axl and Mertk genes in Cx3cr1+ cells phenocopies germline Axl-/-Mertk-/- effects in SOD1G93A mice.a, Time, in postnatal days, at which the indicated percent of the SOD1G93ACx3cr1CreER/+Axlf/fMertkf/f population injected at day 100 with corn oil (control, black curve) or tamoxifen (tmx, green curve) reached clinical endpoint. b Number of ChAT+ (left), ChAT+ NeuN+ (middle), and ChAT+ NeuN+ MMP9+ (right) motor neurons in the lateral motor column of the ventral horn 25 μm hemi-sections of the lumbar (L3-5) spinal cord at day 160 in mice of the indicated treatment. N = 5 per group. c Representative images of flat mounts of the tibialis anterior (four left panels) or quadriceps (four right panels) from SOD1G93ACx3cr1CreER/+Axlf/fMertkf/f mice injected at day 40 with tamoxifen (four top panels) or corn oil control (four bottom panels) and analyzed at day 160 with antibodies to synaptophysin (pre-synaptic terminals) and fluorescent alpha bungarotoxin (post-synaptic acetylcholine receptors). d Fraction of empty (E), partial (P), and complete (C) synapses, as determined by synpatophysin/bungarotoxin co-localization defined in Fig. 3, in SOD1G93ACx3cr1CreER/+Axlf/fMertkf/f mice injected at day 40 with tamoxifen or corn oil control and analyzed at day 160. N = 3 (oil) and 4 (tamoxifen). P values calculated using log-rank test (a), two-sided Mann-Whitney test (b), and two-way ANOVA with Tukey’s multiple comparison test (d). Data are mean values ± SD. Source data are provided as a source data file. Scale bar (c): 50 μm. Selective inactivation of Axl and Mertk genes in Cx3cr1+ cells phenocopies germline Axl-/-Mertk-/- effects in SOD1G93A mice. a, Time, in postnatal days, at which the indicated percent of the SOD1G93ACx3cr1CreER/+Axlf/fMertkf/f population injected at day 100 with corn oil (control, black curve) or tamoxifen (tmx, green curve) reached clinical endpoint. b Number of ChAT+ (left), ChAT+ NeuN+ (middle), and ChAT+ NeuN+ MMP9+ (right) motor neurons in the lateral motor column of the ventral horn 25 μm hemi-sections of the lumbar (L3-5) spinal cord at day 160 in mice of the indicated treatment. N = 5 per group. c Representative images of flat mounts of the tibialis anterior (four left panels) or quadriceps (four right panels) from SOD1G93ACx3cr1CreER/+Axlf/fMertkf/f mice injected at day 40 with tamoxifen (four top panels) or corn oil control (four bottom panels) and analyzed at day 160 with antibodies to synaptophysin (pre-synaptic terminals) and fluorescent alpha bungarotoxin (post-synaptic acetylcholine receptors). d Fraction of empty (E), partial (P), and complete (C) synapses, as determined by synpatophysin/bungarotoxin co-localization defined in Fig. 3, in SOD1G93ACx3cr1CreER/+Axlf/fMertkf/f mice injected at day 40 with tamoxifen or corn oil control and analyzed at day 160. N = 3 (oil) and 4 (tamoxifen). P values calculated using log-rank test (a), two-sided Mann-Whitney test (b), and two-way ANOVA with Tukey’s multiple comparison test (d). Data are mean values ± SD. Source data are provided as a source data file. Scale bar (c): 50 μm. We also examined SOD1G93ACx3cr1CreER/+Axlf/fMertkf/f mice that were injected with tmx or corn oil on day 40 with respect to ChAT+ motor neuron number (Fig. 4b) and the preservation of neuromuscular junctions (Fig. 4c, d) at day 160. The number of both ChAT+ and ChAT+NeuN+ motor neurons in the ventral horn of the lumbar spinal cord was significantly higher in tmx- as opposed to corn oil-injected mice (Fig. 4b). Muscle flat mounts of the tibialis anterior and quadriceps displayed the same representation of αBTX+ post-synaptic densities in tmx- and corn oil-injected mice, but very few synaptophysin+ pre-synaptic motor end plates in the latter (Fig. 4c). Quantification of this effect into complete, partial, and empty NMJs in the quadriceps (defined as for Fig. 2c) revealed that ~90% of NMJs were empty and only ~2% were complete in corn oil-injected mice, whereas ~59% were empty and ~18% were complete in tmx-injected mice (Fig. 4d). Thus, the conditional removal of Axl and Mer from microglia at day 40 in SOD1G93ACx3cr1CreER/+Axlf/fMertkf/f mice led to a significant amelioration of the loss of NMJs seen at day 160. TAM-mediated phagocytosis of diseased neurons We asked whether the above phenotypes might reflect TAM action in microglial phagocytosis. We have previously shown that when assayed in vitro14 using apoptotic cells as targets, microglia are among the most vigorous phagocytes in the body11, and that Mer and Axl mediate ~50% of this phagocytic activity11,12. In addition, we and others have provided evidence that microglia also normally phagocytose large parts or even the entirety of still-living cells63. For example, when CNS cells are infected with adenoviral vectors, they externalize PtdSer on their surface and are eventually eliminated from the brain through TAM-dependent, PtdSer-dependent microglial engulfment64. Live two-photon imaging has demonstrated that this clearance is associated with complete microglial dismemberment of the live transfected cells in as short as 7 h64. Similarly, we have previously shown that a fraction of cells that are normally cleared from neurogenic regions of the adult mouse brain—the subventricular zone near the lateral ventricle and subgranular zone of the hippocampus—are cells that appear to be eaten alive through TAM-mediated microglial phagocytosis11,40. Using another phagocytic system centered on the fractalkine receptor, microglia have also been seen to engulf viable oligodendrocyte progenitor cells in the neonatal mouse corpus callosum65. Disabling microglial phagocytosis in each of these settings results in the survival—and differentiation—of cells that otherwise would have been killed. Others have argued that microglial phagocytosis of live cells may be a common event in both development and the response to tissue trauma and infection41,63. We therefore quantified the level of ChAT-positive material contained within the CD68-positive lysosomes of Iba1-positive microglia located in the lumbar ventral horn of the spinal cord of SOD1G93A versus SOD1G93AAxl-/-Mertk-/- mice at day 160, since motor neurons account for >90% of the ChAT+ cells in this region66. (ChAT is not a definitive marker for skeletal and visceral motor neurons in the ventral lumbar spinal cord, since select interneurons in this region are also cholinergic66). Representative Imaris surface renderings of immunostained sections of the lumbar spinal cord of these mice are shown in Fig. 5a, and quantification of engulfed material is shown in Fig. 5b. We measured a ~ 10-fold reduction in ChAT+ neuronal cytoplasm within SOD1G93AAxl-/-Mertk-/- compared to SOD1G93A microglial lysosomes. Since we have previously shown that combined Axl and Mertk mutation leads to changes in microglial morphology in the APP/PS1 model of AD12, we also normalized these results to total microglial volume, and again observed a large reduction in cholinergic neuron engulfment upon TAM mutation (Supplementary Fig. 4a). As expected, the ‘phagocytic index’ for microglia, as quantified by the fraction of Iba1+ volume occupied by CD68+ lysosomes, was also substantially reduced upon TAM mutation (Supplementary Fig. 4b).Fig. 5TAM-dependent engulfment of diseased cholinergic neurons.a Representative Imaris surface renderings of microglia from lumbar spinal cord sections of SOD1G93A (left) and SOD1G93AAxl-/-Mertk-/- (right) mice stained with antibodies to the microglial marker Iba1 (green), the lysosomal marker CD68 (yellow), and the cholinergic neuron marker ChAT (magenta). b Quantification of engulfment of neuron material, expressed as ChAT+ signal within CD68+ lysosomes within Iba1+ microglia, in the lumbar spinal cord of SOD1G93A mice (black, N = 31 microglia in n = 3 mice) and SOD1G93AAxl-/-Mertk-/- mice (green, N = 36 microglia in n = 3 mice) sampled from 3 to 4 fields of view per spinal cord section across 3–6 sections per mouse, normalized to SOD1G93A. Left panel is results from all microglia, and right panel is average over three mice. P values calculated using two-sided unpaired t-test (b). Data are mean values ± SD. Source data are provided as a source data file. Scale bar a: 3 μm. TAM-dependent engulfment of diseased cholinergic neurons. a Representative Imaris surface renderings of microglia from lumbar spinal cord sections of SOD1G93A (left) and SOD1G93AAxl-/-Mertk-/- (right) mice stained with antibodies to the microglial marker Iba1 (green), the lysosomal marker CD68 (yellow), and the cholinergic neuron marker ChAT (magenta). b Quantification of engulfment of neuron material, expressed as ChAT+ signal within CD68+ lysosomes within Iba1+ microglia, in the lumbar spinal cord of SOD1G93A mice (black, N = 31 microglia in n = 3 mice) and SOD1G93AAxl-/-Mertk-/- mice (green, N = 36 microglia in n = 3 mice) sampled from 3 to 4 fields of view per spinal cord section across 3–6 sections per mouse, normalized to SOD1G93A. Left panel is results from all microglia, and right panel is average over three mice. P values calculated using two-sided unpaired t-test (b). Data are mean values ± SD. Source data are provided as a source data file. Scale bar a: 3 μm. Discussion Our analyses indicate that microglial TAM receptor signaling affects the course of disease development in the SOD1G93A model of ALS in an unexpected manner: the removal of Axl and Mer from microglia extends lifespan. This extension was correlated with increased numbers of motor neurons in the spinal cord and increased numbers of anatomically intact neuromuscular junctions in the SOD1G93A periphery. These observations indicate that the TAM receptors normally control a microglia activity that kills motor neurons in the setting of SOD1G93A overexpression. Although it remains possible that TAM activity drives the expression of a lethal agent or cytokine in microglia—or normally inhibits the expression of a neuroprotective agent—we have not obtained evidence for such TAM regulation in the current study or in earlier work with other mouse models of neurodegeneration, in which we used single-cell RNA sequencing to compare the transcriptomes of WT and Axl-/-Mertk-/- microglia11,12. Our results instead favor the possibility that microglia are killing SOD1G93A motor neurons by eating them. As noted above, neurons that are stressed by repeated depolarization and ensuing Ca2+ overload36 externalize PtdSer on their surface due to the activation of Ca2+-dependent phospholipid scramblases. Moreover, the PtdSer+ spinal motor neurons that are lost in ALS have previously been shown to preferentially express high levels of the best-studied of these scramblases – TMEM16F67. Excitotoxicity and calcium dysregulation have long been recognized as important contributors to motor neuron loss in both sporadic and familial ALS68, and Riluzole, one of two FDA-approved drugs for the treatment of ALS, is a glutamate receptor antagonist that reduces excitotoxicity69. MMP9+ fast-fatigable alpha motor neurons that express Ca2+-permeable glutamate receptors are especially susceptible to aberrantly elevated intracellular calcium and higher rates of firing due to a relative deficit in Ca2+-binding proteins such as calbindin D28K and parvalbumin70,71. We found that many cholinergic neurons in the ventral horn of SOD1G93A spinal cord express abundant scramblase-externalized PtdSer, their viability notwithstanding. This makes them targets for microglial attack, dismemberment, and engulfment64. These events are reflected in the appearance of ChAT-positive neuronal material within microglial lysosomes of the SOD1G93A spinal cord. This material is reduced in microglial lysosomes in SOD1G93AAxl-/-Mertk-/- mice, which is correlated with their extended lifespan. The SOD1G93A line we have used23 massively over-expresses mutant SOD1 (~ 20-fold per haploid genome), which is legitimately cited as a feature that introduces pathological artifacts and limits its translational predictive value for human ALS. For our purposes, however, we only require that a model result in the death of motor neurons in the spinal cord, which the SOD1G93A over-expressors unambiguously and reliably do. A limitation of our study is that we have not quantified the phagocytosis of SOD1G93A motor neuron material by WT versus Axl-/-Mertk-/- microglia in in vitro experiments, or visualized microglial engulfment or dismemberment of motor neurons in real time in vivo. These represent important avenues for future study. It should be noted that we have previously documented a TAM-regulated extension of lifespan very similar to that described above in the SNCAA53T mouse model of PD. We found that these mice, which display prominent overexpression of an A53T mutant α-synuclein in motor neurons of the spinal cord72, had a median survival time of 255 days in our facility, while SNCAA53TAxl-/-Mertk-/- mice had a median survival time of 310 days11. Together, these observations suggest that in pathological settings, microglia may interpret externalized PtdSer as a ‘kill-me’ signal for sick cells. Although this might be viewed as a compensatory protective mechanism that removes irreparably damaged motor neurons and prevents the spread of pathology along circuits, it also eliminates synapses that drive muscle contraction. In this way, microglial predation may expedite the course of neurodegenerative diseases via the phagocytic killing of diseased neurons. Our finding that inactivation of the TAM system in ALS and PD mouse models extends lifespan might be interpreted as suggesting avenues for treating neurodegenerative disease via pharmacological inhibition of TAM. However, the use of TAM-directed agents—e.g., Axl and Mer small-molecule tyrosine kinase inhibitors that are in development for the treatment of various cancers73—as potential therapies, without addressing the fundamental underlying causes of ALS or PD that are making motor neurons sick in the first place, does not seem tenable to us. Conversely, the generality of direct TAM-mediated phagocytic killing by macrophages, and its potential application to the treatment of human diseases, have recently been convincingly demonstrated. We have previously shown that the Gla domains of Gas6 and Pros1, the amino-terminal segments of these proteins that bind PtdSer, can be deleted without affecting the TAM receptor binding affinity of the truncated ligands34. Although these ‘Gla-less’ forms of Gas6 and Pros1 are incapable of binding PtdSer or normally activating the TAM tyrosine kinases34, their activity can be restored by replacing the deleted Gla domain with a recombinant single-chain variable fragment (scFv) of an antibody (or a nanobody), and then providing the antigen target for this antibody74,75. In a set of applications, scFvs directed against either CD19, which is expressed by B lymphocytes and all B lineage cells, or TYRP1, which is expressed by all melanocytes and many melanomas, were fused to a truncated, Gla-less Pros174. Injection of engineered scFvCD19-Pros1 protein demonstrated phagocytic killing (depletion) of live B cells and the alleviation of autoimmunity in MRLlpr/lpr mice, and injection of engineered scFvTYRP1-Pros1 protein demonstrated phagocytic killing (depletion) of live cancer cells and a reduction in tumor burden in mice harboring B16.F10 melanoma cells74. Methods All animal procedures were conducted in compliance with the ethical guidelines and according to protocols approved by the Salk Institute Animal Care and Use Committee (Protocol No. 17-00009) and Boston Children’s Hospital (Protocol No. 00002118). Mice C57BL/6 J wild-type (WT), Cx3cr1GFP/+ (JAX number: 005582), Cx3cr1CreER/+ (JAX number: 021160), and B6.Cg-Tg(SOD1*G93A)1Gur/J (SOD1G93A mice on B6 background, JAX number: 004435) mice were obtained from the Jackson Laboratory. The Axl-/-Mertk-/- and Axlf/f Mertkf/f strains have been described previously11,43,59. SOD1G93A hemizygous mice23 were crossed to Axl-/-Mertk-/-, Cx3cr1GFP/+, and Cx3cr1CreER/+Axlf/f Mertkf/f lines to generate hemizygous SOD1G93AAxl-/-Mertk-/-, SOD1G93ACx3cr1GFP/+, and SOD1G93ACx3cr1CreERT/+Axlf/f Mertkf/f lines. All lines were backcrossed for >10 generations and maintained on a C57BL/6 background. Animals were group-housed under a 12-hour light:12-hour dark cycle. Ad libitum access to food and water was provided, and environmental conditions were maintained in accordance with guidance from the Association for Assessment and Accreditation of Laboratory Animal Care. All mice were used at the ages specified in the experimental procedures outlined in the results and in the figure legends. Unless otherwise stated, animals were compared to sex-matched littermates. Male and female mice were included in all experiments and were randomly allocated to experimental groups unless otherwise noted. Sex-based analyses were not performed because they were not the primary focus of the current study. SOD1G93ACx3cr1CreER/+Axlf/f Mertkf/f mice were randomly chosen to be in the experimental or control group. Beginning at day 40, they were injected intraperitoneally (i.p.) with 150 mg/kg tamoxifen (Sigma) for the experimental group or corn oil (Sigma) vehicle alone in two successive injections 48 h apart as described previously11. Conditional deletion of Mer and Axl was analyzed immunohistochemically and by western blot at day 80, 120, and 160 to confirm lack of expression of Axl and Mer protein. Disease stage scoring based on motor deficits SOD1G93A mice disease stage was scored from 0 to 4, with 0 being no overt symptoms and 4 being clinical end-stage, based on their motor deficits, modified from published studies27. Score 0 (pre-symptomatic) mice exhibit normal gait and hindlimb splay angle during tail suspension test; score 1 (‘symptomatic’) mice exhibit an acute hindlimb angle often with weak tremor upon suspension but have an otherwise normal gait; score 2 (‘disease onset’) mice exhibit hindlimb collapse and changes in gait including limping and hindering in walking; score 3 (‘hindlimb paralysis’) mice have extreme weakness in both hindlimbs and not using hindlimbs in walking but are still able to right themselves unassisted within 20 s; score 4 (‘clinical endpoint’) mice can no longer right themselves to sternum within 20 s after being placed on their side. Motor deficits were scored blind to their genotype or treatment on two consecutive days. If a mouse showed the same motor deficits on both days, score advancement was confirmed. Antibodies and reagents Primary antibodies used for immunohistochemistry (IHC) on mouse spinal cord sections or muscle flat-mounts and western blotting (WB) were as follows: anti-Iba1 (Wako 019-19741, polyclonal rabbit, IHC 1:250-500), anti-Iba1 (Novus NB100-1028, polyclonal goat, IHC 1:200), anti-CD68 (BioRad MCA1957, clone FA-11, monoclonal rat, IHC 1:200), anti-NeuN (Chemicon MAB377, Clone A60, monoclonal mouse, IHC 1:500), anti-MMP9 (Abcam ab38898, polyclonal rabbit, IHC 1:1000), anti-GFAP (Invitrogen 13-0300, monoclonal rat, IHC, 1:500), anti-Aldh1l1 (Abcam 87117, polyclonal rabbit, IHC, 1:500), anti-ChAT (Chemicon AB144P, polyclonal goat, IHC 1:500), anti-Mer (R&D AF591, polyclonal goat; IHC 1:100 and WB 1:1000), anti-Mer (eBioscience, clone: DS5MMER, monoclonal rat; IHC: 1:200 and WB 1:1000), anti-Axl (R&D AF854, polyclonal goat; WB 1:1000, IHC 1:100), anti-Gas6 (R&D AF986, polyclonal goat; WB 1:1000, IHC 1:50), anti-GAPDH (Millipore, MAB374, monoclonal mouse clone 6C5; WB 1:10000) as loading control, anti-synaptophysin (Abcam, AB16659, polyclonal rabbit; IHC 1:500 or 1:200 for flat mount), anti-Laminin (Sigma, L-9393, polyclonal rabbit, IHC 1:500), anti-mutant human SOD1 (MediMabs, B8H10, polyclonal mouse, WB 1:1000), and cleaved-Casp3 (Cell Signaling, Asp175, #9661, IHC 1:200). For human postmortem spinal cord sections, primary antibodies used for IHC were as follows: guinea pig anti-human IBA1 (Cedarlane labs 234308(SY), 1:500), mouse anti-human AXL (R&D MAB154, 1:100), rabbit anti-human MER (Thermo scientific MA5-31991, 1:200), and anti-ChAT (Chemicon AB144P, polyclonal goat, 1:100). Secondary antibodies for WB were all used at 1:10000 and were IRDye 680RD anti-mouse and anti-goat IgG secondary antibodies for Li-Cor detection and quantification analysis, HRP-conjugated anti-goat (705-035-003) and anti-mouse (115-036-062) from Jackson immunoresearch. Secondary antibodies for IHC were all used at 1:500 and were fluorophore-conjugated anti-rat (712-545-153 or 712-165-153 from Jackson ImmunoResearch), anti-goat (A-11056 from Life Technologies, or 705-166-147 from Jackson ImmunoResearch), anti-rabbit (A-11071 or A-21206 from Life Technologies), and anti-mouse (A-11029 from Life Technologies, 715-166-150 or 715-176-150 from Jackson ImmunoResearch). For flow cytometry, the antibodies used are as follows: anti-CD16/CD32 (BD Biosciences, 553142, 2.4G2, 1:50), Brilliant Ultra Violet (BUV)395 anti-Mer (Invitrogen, 363-5751-82, DS5MMER, 1:50), BUV395 Rat IgG2a κ isotype control (BD Biosciences, 563556, R35-95, 1:50), BUV395 anti-CD11b (BD Biosciences, 563553, M1/70, 1:100), BUV496 anti-TCR β chain (BD Biosciences, 749915, H57-597, 1:100), BUV563 anti-I-A/I-E (BD Biosciences, 748846, M5/114.15.2, 1:50), BUV805 anti-Ly-6G and Ly-6C (BD Biosciences, 741920, RB6-8C5, 1:50), Brilliant Violet(BV)421 anti-CD206 (MMR) (BioLegend, 141717, C068C2, 1:100), BV510 anti-CX3CR1 (BioLegend, 149025, SA011F11, 1:100), BV785 anti-CD11c (BioLegend, 117335, N418, 1:100), Alexa Fluor(AF) 488 anti-CD11b (BioLegend, 101217, M1/70, 1:100), AF700 anti-CD45 (BioLegend, 147716, I3/2.3, 1:200), PE anti-CD45 (BioLegend, 103106, 30-F11, 1:200), PE-Cyanine7 anti-Axl (Invitrogen, 25-1084-82, MAXL8DS, 1:50), PE-Cyanine7 Rat IgG2a κ isotype control (Invitrogen, 25-4321-82, eBR2a, 1:50), and live-dead stain (Invitrogen eBioscience Fixable Viability Dye eFluor780, 65-0865-18). Key additional reagents and their usage were as follows: PSVue-550 (Molecular targeting, P-1005, following manufacturer’s instructions for activation of PSVue-550), PSIVA-IANBD (Abcam Ab129817), Neurotrace 435/455 (Molecular Probe N21479), alpha-bungarotoxin (Biotium 00004, IHC 1:1000), tamoxifen (Sigma T5648) and corn oil (Sigma C8267), as described previously76, Trueblack plus (Biotium #23014), BD Horizon Brilliant Stain Buffer Plus (BD Biosciences, 566385), and ultraComp eBeads Plus Compensation Beads (Invitrogen, 01-3333-42) for flow cytometry. Euthanasia and anesthesia Routine perfusion procedures were performed as described previously12,14. Briefly, adult mice (80–180 days) were anesthetized with a final concentration of 100 mg/kg Ketamine/10 mg/kg Xylazine (i.p.) or Avertin (240 mg/kg, i.p.) in accordance with IACUC guidelines approved by the Salk Institute Animal Care and Use Committee (Protocol No. 17-00009) and Boston Children’s Hospital (Protocol No. 00002118). Mice were then transcardially perfused with 20U/ml heparin in PBS followed by freshly prepared 4% PFA in PBS. RNA isolation and RT-qPCR RNA from snap-frozen whole spinal cord tissue was isolated with TRIzol (Thermo Fisher Scientific 15596026) according to the manufacturer’s instructions. The RT Transcriptor First Strand cDNA Synthesis Kit (Roche) with anchored oligo(dT) primers (Roche) was used for reverse transcription. Quantitative PCR was run in a 384-well plate format on a QuantStudio Q5 as previously described12. Primer sequences used were: Il1b forward primer 5’-3’ CCT CTC CAG CCA AGC TTC C; and reverse primer 5’-3’ CTCA TCA GGA CAG CCC AGG T. Il6 forward primer 5’-3’ AGA CAA AGC CAG AGT CCT TCA GA; and reverse primer 5’-3’ GCC ACT CCT TCT GTG ACT CCA. Tnfa forward primer 5’-3’ GCC ACC ACG CTC TTC TGT CT; and reverse primer 5’-3’ CAG CTG CTC CTC CAC TTG GT. Ifna forward primer 5’-3’ CCC ACA GCC CAG AGA GTG AC; and reverse primer 5’-3’ GCC CTC TTG TTC CCG AGG T. Ifng forward primer 5’-3’ CAA TCA GGC CAT CAG CAA CA; and reverse primer 5’-3’ AAC AGC TGG TGG ACC ACT CG. Inos forward primer 5’-3’ GCA AGC ACC TTG GAA GAG GA; and reverse primer 5’-3’ AGG CCA AAC ACA GCA TAC CTG. Il4 forward primer 5’-3’ AGG AGC CAT ATC CAC GGA TGC GA; and reverse primer 5’-3’ TGT TCT TCG TTG CTG TGA GGA CGT. Tgfb forward primer 5’-3’ AGT GTG GAG CAA CAT GTG GAA; and reverse primer 5’-3’ CAG CCA CTC AGG CGT ATC AG. Cox2 forward primer 5’-3’ TGG GAGGCG AAG TGG GTT TT; and reverse primer 5’-3’ TGG TAG GCT GTG GAT CTT GC. Il10 forward primer 5’-3’ TGA AGA CCC TCA GGA TGC G; and reverse primer 5’-3’ TTC ACC TGC TCC ACT GCC TT. Relative expression of genes of interest was normalized to GAPDH primer pairs: forward sequence 5’-3’ AGG TCG GTG TGA ACG GAT TTG; reverse sequence 5’-3’ GGG GTC GTT GAT GGC AAC. Immunohistochemistry Mouse spinal cords were micro-dissected and post-fixed in 4% PFA in PBS overnight at 4 °C and subsequently infiltrated in 30% sucrose in PBS for 1 day at 4 °C. The spinal cords were further split into cervical, thoracic, and lumbar segments based on counting the ventral roots and frozen in tissue-freezing medium. Cross-sections of spinal cords consisting of lumbar L3-5 were cryo-sectioned at 25 µm, air-dried overnight, and subsequently processed for staining. Antigen retrieval with citrate buffer was performed if needed by heating cord sections to 80 °C for 3 min in pre-warmed citrate buffer and cooling to room temperature. Non-specific binding was blocked by 1 h incubation in blocking buffer (1xPBS containing 0.3% Triton-X100, 0.1% Tween-20, 5% donkey serum and 2% IgG-free BSA). Sections were incubated overnight at 4 °C with primary antibody diluted in blocking buffer, then washed in PBS 0.1% Tween-20, and incubated for 2 h at 22 °C–24 °C in the dark with fluorophore-coupled secondary antibodies and subsequently in Hoechst for 5 min diluted in blocking buffer. Sections were washed, sealed with Fluoromount-G (SouthernBiotech) and stored at 4 °C. Human post-mortem analyses Post-mortem human spinal cords consisting of lumbar segments were obtained by the Stevens lab from J. Trojanowski of the Center for Neurodegenerative Disease Research at the University of Pennsylvania. These were cryoprotected and cryosectioned at 20 µm and either immunostained as previously described12 or subjected to RNAscope assays followed by immunostaining as described below. For human post-mortem immunostaining, spinal cord sections were thawed to room temperature, subjected to citrate buffer antigen retrieval, and blocked by 1 h incubation with blocking buffer (5% horse serum, 2% IgG-free BSA, 0.3% Triton X-100 in 1x PBS). Sections were incubated overnight at 4 °C with human-specific primary antibody diluted in blocking buffer, then washed in PBS treated with TrueBlack Plus Lipofuscin Autofluorescence Quencher (diluted 1:20 in 70% Ethanol) for 90 s followed by incubation with fluorophore-coupled secondary antibodies diluted in blocking buffer without Triton X-100 at room temperature in the dark. Sections were washed in PBS before mounting. Sex- and gender-based analyses were not performed. Because of the low number of female and male patients, no statistical comparison between sex was feasible. Informed consent was obtained by the University of Pennsylvania from the next of kin at the time of death. It is the policy of the University of Pennsylvania that its Institutional Review Board does not review studies involving the collection and use of post-mortem tissue. Fluorescence in situ hybridization (RNAscope) Fluorescence in situ Hybridization was performed using the RNAscope™ Multiplex Fluorescent V2 Assay (Cat. No. 323100) from Advanced Cell Diagnostics (ACD) to visualize the transcript-level expression of microglial TAM receptors and their ligands in the ventral horn of human lumbar spinal cord. Post-mortem human spinal cord cross sections (20 μm) were freshly collected from OCT-embedded tissue blocks onto Superfrost Plus microscope slides (Fisher Scientific, 22-037-246). Slides were dried in room temperature within 1–2 h and stored in −80 °C. Upon heating at 60 °C, tissue sections were sequentially treated with hydrogen peroxide (Ref no. 322335), RNAscope Target Retrieval reagent (Ref. No. 322000), and 100% molecular-grade ethanol. Sections were then digested with Protease IV (Ref. no. 322336) for 15 min, followed by conjugation with customized target probes for 2 h at 40 °C in the HybEZ Hybridization System. Control slides were treated with Human-specific RNAscope 4-plex positive and negative control probes (Cat. No. 321831). To enhance the detection of the conjugated probes, sections were treated with amplifiers from the detection kit and Opal dyes (1:500, diluted in TSA buffer) corresponding to the target probes. Note that given the known impact of autofluorescence in human tissue fluorescent imaging, care was taken to select the strongest signal-to-noise fluorophore for RNAscope. To determine the cell-type specificity of the transcript signals, IHC on human sections were continued to be conducted after the RNAscope assay. Briefly, non-specific binding was blocked by 1 h incubation with blocking buffer (5% horse serum, 2% IgG-free BSA, 0.3% Triton X-100 in 1x PBS). Sections were incubated overnight at 4 °C with human-specific primary antibody diluted in blocking buffer, then washed in PBS treated with TrueBlack Plus Lipofuscin Autofluorescence Quencher (diluted 1:20 in 70% Ethanol) for 90 s. (Such treatment on human tissue sections sufficiently blocked autofluorescence without affecting RNAscope signals and enhanced signal-to-noise ratio in every channel but the 488 channel, which was left open in the assay without any antibodies or probes. Slides were rinsed in PBS and incubated for 2 h at room temperature in the dark with fluorophore-coupled secondary antibodies diluted in blocking buffer without Triton X-100. Sections were washed in PBS before mounting. Mouse muscle collection and processing for NMJ analyses Quadriceps, tibialis anterior (TA), extensor digitorum longus (EDL), gastrocnemius (gastroc), and soleus were micro-dissected from both hindlimbs after transcardiac PFA perfusion. Following sucrose cryopreservation, individual muscles were split in half crosswise and embedded in tissue freezing medium to obtain serial 10 μm cross-cryosections for analysis of muscle fiber diameter and fibrosis. The other half was processed for flat mounts for NMJ analysis to quantify denervation in each muscle group. Muscle denervation was quantified as described previously77,78. Muscle fibers were first rinsed in 1xPBS three times for 1 h at RT and permeabilized in pre-chilled 100% MeOH at −20 °C for 10 min then rinsed in 1xPBS three times for 5 min at RT. The muscle fibers were then incubated with 0.1 M glycine/1xPBS for 2 h at RT and subsequently blocked overnight with 5% donkey serum, 0.1% DMSO, 0.1% Tween-20 in glycine/PBS buffer. To label innervating presynaptic motor endplates, muscle fibers were then incubated with anti-synaptophysin antibody (1:200) overnight at 4 °C. Muscles were washed three times for 1 h in 0.5% Triton X-100 in 1xPBS and stained with fluorophore-conjugated alpha-bungarotoxin to label post-synaptic acetyl-choline receptors and second antibody donkey anti-rabbit overnight at 4 °C Celsius. Muscles were washed three times for 1 h in 0.5% Triton X-100 in 1xPBS followed by wash in 1xPBS for 1 h before mounting with Fluoromount-G (SouthernBiotech). Slides were dried at RT overnight with weights on top of cover glass to flatten muscles before storage at 4 °C. Intrathecal injection of PSVue Zinc-activated and non-activated PSVue-550 were freshly prepared according to the manufacturer’s user manual and published studies79–81. Mice were anesthetized using isoflurane (2–4% exposure for induction followed by 1.5% throughout the procedure). Anesthetized mice were injected by 2 μl of PSVue-550 or a 1:1 mixture of PSVue and undiluted pSIVA intrathecally, and transcardially perfused 2–3 h after PSVue injection with PBS followed by 4% PFA. Microscopy One-photon laser scanning confocal images were acquired with a Zeiss LSM 710 or Zeiss LSM 880 confocal microscope using Plan-Apochromat 20×0.8-NA air-matched or 63×1.4-NA oil objectives (laser lines: 405, 488, 594, and 633 nm). Image size was 1024 × 1024 pixels. Stack thickness was typically 20μm for mouse spinal cord sections and 100 µm for NMJ analysis. Muscle cross sections for fiber diameter were imaged using Olympus VS-120 Virtual Slide Scanning Microscope using 20x objective. Exposure acquisition was set according to the WT or non-mutant samples in all experiments. Images were obtained and processed via the Zen 2.5 Black and Blue editions. Imaging data analysis For lumbar motor neuron subtype quantification, maximum intensity projection images (212 x 212 μm, 1024 × 1024 pixel resolution) consisting of hemi-sections of the ventral L3-L5 spinal cord were analyzed using Fiji (ver1.0). 6–10 hemisections were imaged per cord/spinal region per mouse. Motor neuron numbers per ventral hemi-section of L3-5 were first determined using ChAT and Hoechst, and counted manually. Triple colocalization of MMP9, NeuN, and ChAT was quantified using the Cell Count plug-in using Fiji, with the investigators blind to genotype or treatment. For quantification of Iba1 and CD68 mean fluorescence intensity, the area of the ventral horn was measured. Thresholds were set to exclude non-specific background immunoreactivity and maintained throughout the dataset. The area of CD68 and Iba1 above the intensity threshold was measured, and % area was determined using the area of the respective ventral horn as the denominator. For quantification of astrocyte numbers and activation based on GFAP and Aldh1l1 immunostaining, cell nuclei were detected by DAPI thresholding in selected regions in the ventral horn of lumbar spinal cord on QuPath (v0.6.0.) and expanded uniformly by 1 µm to cover areas in the cell soma. GFAP threshold was set and maintained throughout the dataset analysis. The number of nuclei positive for GFAP and the mean fluorescence intensity of Aldh1l1 and GFAP within positive nuclei were analyzed on a per-image basis and summarized for each mouse. For quantification of fiber diameter and mean fluorescence intensity using laminin immunostaining, the threshold of laminin was set to exclude non-specific background immunoreactivity and maintained throughout the dataset. Fiber diameter was measured by the length of the longest diameter of each fiber and was averaged across each section of the muscle bundle. 20–30 fibers were typically measured in each muscle section, and at least two were analyzed to compensate for heterogeneity in fiber size, and 12–20 cross-sections were analyzed in each mouse. For NMJ quantification, maximum intensity projection images from muscle flat-mount stacks were segmented into individual NMJs based on the morphology of alpha BTX. Thresholds were set to exclude nonspecific background immunoreactivity for both synaptophysin and alpha-BTX channel. Using the aNMJ-morph macro in Fiji82, % NMJ innervation was determined by dividing the total number of areas of overlap between synaptophysin and BTX signals by the number of areas of alpha-BTX signal. The NMJs were binned into three categories: ‘empty’ (with less than 10% occupied alpha BTX area), ‘partial’ (with 10–90% occupied alpha BTX area), and ‘complete’ (more than 90% occupied alpha BTX area). Structural reconstruction and co-localization engulfment analyses were performed using Imaris software (v9.9.1). Confocal LSM880 63x z-stack images (0.5 µm step size, 0.07 µm pixel size) were acquired from 3 to 4 fields of view per spinal cord section across 3-6 sections per animal, consisting of both lumbar L3-5 ventral horn lateral motor columns. Individual channel images based on Iba1 and CD68 relative fluorescent intensity were processed in 3D surface rendering on Imaris software, and volumes and ChAT intensity were calculated, masked, and normalized to determine the percentage of co-localization. All experiments were performed and analyzed blind to the treatment group and genotype. For relative RNAscope signal intensity analysis in human postmortem disease versus control spinal cord cross sections, maximum intensity projection images (512 x 512 μm, 2048 × 2048 pixel resolution) consisting of the human lumbar spinal cord ventral horn were analyzed with QuPath v0.6.0. DAPI was used to positively detect cell nuclei, which were classified into subgroups based on the relative expression of cellular markers (IBA1 for microglia, CHAT for cholinergic neurons, and neither for others), and the morphology and size of the cells were determined. Finally, the summed fluorescence intensity of RNAscope signals within each nucleus was calculated for cell type and summarized per patient case normalized either to cell type or to the mean of the healthy control cases. Isolation and purification of mouse spinal cord immune cells for flow cytometry Spinal cord sample preparation and isolation were adapted from the CNS tissue dissociation protocol optimized for flow cytometry. Briefly, mice were first anesthetized with avertin (240 mg/kg, i.p.) and then intracardially injected with anti-CD45-PE antibodies (clone 30-F11, 0.2 mg/kg), which were allowed to circulate for 3 min to label circulating immune cells for downstream exclusion. This was then followed by ice-cold HBSS transcardiac perfusion. The spinal cords were promptly dissected out, submerged in ice-cold RPMI-H to remove meninges, and minced into a fine paste on a pre-chilled petri dish on ice. The minced tissues were digested with collagenase-P (0.5 mg/ml) and DNase-1(250 U/ml) at room temperature for 60 min. The following steps were all performed at 4 °C or on ice unless otherwise specified. Digested tissues were triturated in FACS buffer, filtered through 70 µm strainers, and subjected to purification and myelin removal with 25% BSA in HBSS. The purified cell pellets were resuspended in FACS buffer, counted, and plated for staining. The cells were first blocked for 10 min with FcR (1:50) and Brilliant Stain Buffer (20%) in HBSS containing 2 mM EDTA, and immunostained with their corresponding antibody-fluorophore conjugates for 20 min at 4 °C. The cells were fixed with eBioscience IC Fixation buffer for 20 min at RT, and washed and resuspended in FACS buffer at 4 °C. The samples were run on Cytek Aurora 5 L spectrum flow cytometer with the multi-autofluorescence module applied on autofluorescence cell population, and further analyzed and graphed on FlowJo software (v10.10.0). Western blotting Spinal cords and whole brains were dissected and snap-frozen in liquid nitrogen. Frozen tissues were homogenized with a handheld motorized homogenizer for 30 s and lysed using RIPA buffer and phosphatase protease inhibitor (Roche, Sigma) for 30 min on ice. Supernatants were stored at −80 °C in aliquots after samples were spun down at 12,000 rpm for 5 min. Immunoblotting was performed as described previously12,14,76. Membranes were imaged on Odyssey Infrared Imager (Li-Cor) and quantified using ImageStudio software (Li-Cor, RRID:SCR_015795) with Gapdh as loading control. Statistics Numeric data analysis and statistics were performed with Microsoft Excel (ver15.36) and GraphPad Prism (version 8.0, 9.0, and 10.0) software. Group sample sizes were chosen based on previous studies and/or power analyses. Experiments were performed with at least two independent replicates whenever applicable. Datasets displayed normal distribution and equal standard deviations unless indicated by the unequal variance test. Statistical tests used were always two-sided. P values from post hoc multiple-comparison tests were adjusted for multiple comparisons. Exact P values are noted whenever possible with 95% confidence interval. All data are presented as the mean ± 1 standard deviation with n indicating the number of mice analyzed, unless otherwise stated in the figure legend. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary Information Reporting Summary Transparent Peer Review file Supplementary Information Reporting Summary Transparent Peer Review file Source data Source Data Source Data