Motoneurons Inhibitory Synapses Homeostatically Respond to Neuronal Activity and Modulate Amyotrophic Lateral Sclerosis Pathogenesis

Abstract

Alterations in excitation/inhibition (E/I) balance and changes in motor neurons (MN) activity may contribute to MN vulnerability in ALS. The balance of pathogenic versus adaptive changes occurring in inhibitory synapses and affecting E/I balance remain unclear. Confocal microscopy of MN from P45 male SOD1G93A mice reveal downregulated GlyR but upregulated GABAR clusters at inhibitory synapses. GlyR and GABAR respond to PSAM and DREADD chemogenetic alterations of MN excitability, with increased activity driving increase in inhibitory clusters. An E3 ligase-conjugated intrabody (GFE3) degrades Gephyrin, decreases GABAR and GlyR clusters, increases net activity, and downregulates disease markers. However, simultaneous decrease of inhibition and increased activity by actPSAM and GFE3 shows no net beneficial effects on disease markers. Thus inhibitory synapses are involved in the early phases of ALS pathogenesis and respond to persistent homeostatic loops, and their suppression delivers a net activity increase, offering potential benefits on disease pathways.

Significance Statement

This study reveals significant changes in inhibitory connections onto motor neurons (MN) during the early stages of ALS, highlighting their involvement in the excitation/inhibition (E/I) imbalance seen at this disease stage. Using advanced chemogenetic tools, we show that inhibitory synapses respond homeostatically to modulations in MN activity. Moreover, decreasing inhibitory receptor components using a functionalized nanobody increases net MN excitability and reduces disease markers. These findings show the dynamic nature of inhibitory synapses in ALS MN, emphasizing their abilities to adapt to activity changes which can in turn influence disease progression.

Introduction

Alterations in motor neuron (MN) excitability are considered a critical factor in determining MN vulnerability in amyotrophic lateral sclerosis (ALS). Hyperexcitability of MN, defined by a lower rheobase or a steeper input/output relationship, has been repeatedly identified in the early presymptomatic transgenic mice carrying human mutant SOD1 genes (Kuo et al., 2004; Martin et al., 2013), as well as in iPSC-derived MN from ALS patients (Burley et al., 2022; Harley et al., 2023). According to the excitotoxicity model, the hyperexcitability of MN is critical to the triggering of cell death cascades (Rothstein, 2009). In the SOD1(G93A) (SOD1) mice, lumbar MN display hyperexcitability at embryonic stage (E17.5; Martin et al., 2013). However, as disease progresses, MN intrinsic excitability declines. Fast-fatigable (and large fatigue-resistant) MN, characterized by their large soma size and by MMP9 expression (MMP9+ MN), are normo-excitable at postnatal days (P) 6–10 but are hypoexcitable at P45–P55, i.e., at degeneration onset of their neuromuscular junction, whereas slow (small soma size and MMP9−) MN, nonvulnerable to the pathogenic cascade, are still hyperexcitable at P6–P10 and reach a normo-excitable state at P45–P55 (Leroy et al., 2014; Martínez-Silva et al., 2018). Human MN from iPSC globally (all MN types pooled) shows an excitability shift, from hyper- to hypoexcitability, as they mature in vitro (Naujock et al., 2016; Catanese et al., 2021).

Along with changes in intrinsic excitability, multiple lines of evidence have shown alterations in excitatory synapses on MN, implying a decreased excitatory drive which, together with hypoexcitability, may drive MN degeneration (Roselli and Caroni, 2015). In fact, ALS patients display a substantial loss of synapses (Aousji et al., 2023) and a disturbance in excitatory synapses (Aly et al., 2023), and the remaining glutamatergic inputs may be disturbed at structural and functional level (Bączyk et al., 2020; Broadhead et al., 2022). Disorganized excitatory synapses and reduced excitability may reduce the excitatory drive and precipitate the final stages of vulnerability due to the loss of activity-dependent neuroprotection (Bading, 2013; Song et al., 2023).

Nevertheless, MN display remarkable homeostatic mechanisms aimed at maintaining a stable input/output profile, and disturbed homeostasis may be involved in driving their vulnerability (Kuo et al., 2020; Huh et al., 2021). To this respect, alterations in intrinsic excitability and in synaptic excitation may drive compensations or overcompensations through inhibitory inputs.

Recent evidence suggests that inhibitory synapses, too, are altered in ALS (Mora and Allodi, 2023). Reduced inhibitory drive may result from the loss of presynaptic terminals (Allodi et al., 2021) as consequence of progressive degeneration of several subpopulations of inhibitory neurons (Montañana-Rosell et al., 2024) or from reduced excitability of inhibitory interneurons (Cavarsan et al., 2023). On the other hand, electrophysiological evidence points toward transient postsynaptic alterations underlying the alteration of inhibition (Nascimento et al., 2024).

It remains unclear whether changes in inhibition result primarily from pathogenic pathways or from compensation (i.e., if inhibition is responsive to feedback loops in ALS MN). Moreover, the net effect of alterations of inhibition on vulnerability and disease pathways are unclear: according to one model, reduced inhibition may worsen excitotoxicity, but, alternatively, it may increase MN firing and enhance neuroprotection (thus representing a homeostatic attempt to restore firing setpoints).

Here we exploited a set of chemogenetic systems to manipulate MN activity in vivo, while monitoring the response of inhibitory synapses, and we used a functionalized nanobody to selectively remove Gephyrin while monitoring their impact on multiple disease pathways.

We demonstrate a distinct alteration of GABA and Glycine receptors occurring in MN and their differential response to excitability changes, and we show that decrease in inhibition ameliorates the burden of disease markers.

Materials and Methods

Animals

Experiments were conducted using the B6SJL-Tg(SOD1*G93A)1Gur/J mouse line from the Jackson Laboratory (#002726; Gurney et al., 1994; Tu et al., 1996). Their healthy littermates are denoted as wildtype (WT). Following the ALS staging previously described, mice at P45 are considered presymptomatic, while those at P110 exhibit hindlimb symptoms and are defined as fully symptomatic (Gurney et al., 1994; Tu et al., 1996; Saxena et al., 2013; Bączyk et al., 2020; Ouali Alami et al., 2020).

For chemogenetic experiments, the B6;129S6-Chat^tm2(cre)Lowl/J line (the Jackson Laboratory, stock #006410; Rossi et al., 2011), expressing Cre recombinase in choline acetyltransferase (ChAT)+ neurons, was utilized. Homozygous females of this line were crossed with SOD1 males, resulting in F1 littermates, among which SOD1-ChAT-Cre double transgenic males were administered intraspinal adeno-associated virus (AAV) injections.

Because of the known difference in progression rate observed in males and females SOD1(G93A) mice, only male mice were used in the present study. Execution and quantification of the experiments were performed under blinding of the mouse genotypes or treatment.

Animal protocols adhered to the standards set by the Tierforschungszentrum (University of Ulm, Germany). The experiments received approval from the Regierungspräsidium Tübingen (licenses #1404, 1440, and 1634).

Intraspinal AAV injection

Intraspinal viral injections were performed following the methodology outlined in Bączyk et al. (2020). Briefly, mice aged between P21 and P23 were anesthetized with buprenorphine (0.1 mg/kg, opioid-analgesic) and meloxicam (1 mg/kg, nonsteroid anti-inflammatory analgesic); throughout the procedure, they were kept anesthetized on a stereotaxic frame with continuous delivery of 2% isoflurane in oxygen. Under a zoom stereo microscope, a skin incision was made, followed by removal of the subcutaneous fascia and gentle lateral displacement of the underlying paraspinal muscles. Afterward, laminectomy was performed at T12, and the vertebral bone flap was removed to expose the spinal cord. Subsequently, a mixture containing the virus suspension and 1% Fast Green dye in PBS++ was prepared in a 1:1 ratio, and 1 μl was delivered into the spinal cord at the following stereotaxic coordinates: y = +0.3 mm and z = −0.5 mm, relative to the dorsal artery, to inject the mixture into the ventral MN pool, maximizing infection of these cells. Following injection, the capillary remained in place for 5 min to allow local diffusion of the viral suspension and prevent backflow. Finally, muscles and skin were sutured and mice were transferred to their cage and monitored. Approximately 3 weeks after injections at P45, mice were perfused and their tissue was collected.

Mice injected with PSAM viral vectors were injected with PSEM308 effector molecule obtained from Apex Scientific, while mice injected with Gs-DREADD viral vector were administered clozapine N-oxide (CNO; obtained from Tocris). Both injections were administered intraperitoneally starting from 17 d after surgery for 8 consecutive days at a dosage of 5 mg/kg in saline. Mice were perfused 3 h after the final administration of the agonist.

Transcardial perfusion and cryoembedding

A terminal anesthesia injection consisting of ketamine (100 mg/kg) and xylazine (16 mg/kg) prepared in saline was given to mice. After the absence of reflexes was confirmed, mice were intracardially perfused. For mice intended for in situ hybridization (ISH) or immunostaining, the following procedure was conducted: first, 50 ml of ice-cold 5 mM EDTA in PBS was pumped into the mouse body to flush out the blood, preventing clotting. Subsequently, 50 ml of 4% PFA in PBS was used for fixation. Following dissection, the spinal cord was incubated in 4% PFA in PBS at 4°C overnight and then rinsed in PBS and incubated in 30% sucrose in PBS for cryoprotection for up to 5 d. Finally, the lumbar portion of the spinal cord was cryo-embedded in OCT and stored at −20°C before being sectioned into 20 μm sections in a cryostat. For samples intended for laser capture microdissection (LCM) and reverse transcription quantitative polymerase chain reaction (RT-qPCR), precautions were taken to ensure RNA integrity: the chemical hood and dissection tools were cleaned with RNAse zap to eliminate RNAses. Intracardial perfusion was then performed with RNAse-free PBS for 2 min, and then the lumbar spinal cord was dissected and directly OCT embedded and stored at −80°C.

Viral vectors

Custom-made AAV9 viruses encoding pAAV-Ef1a-DIO_EGFP-GFE3 (Addgene #79871), pAAV-Ef1a-DIO_EGFP-RandE3 (Addgene#79872), and pAAV-hSyn-DIO-rM3D(Gs)-mCherry (Gs-DREADD; Addgene #50458) were prepared in house at a concentration of 1.6 × 10¹⁴ viral genomes (vg)/ml, 2.6 × 10¹³ vg/ml, and 1.0 × 10¹³ vg/ml, respectively. AAV9 expressing the inhibitory Pharmacologically Selective Activation Module [PSAM; pAAV(9)-pCAG-A7-floxed-PSAM(L141F,Y115F)-GlyR-GFP-WPRE] and Activator PSAM [(pAAV(9)-pCAG-A7-floxed-PSAM(L141F, Y115F) 5HT3-WPRE] were obtained from Vector Biolabs, at the titers 6 × 10¹³ vg/ml and 7 × 10¹³ vg/ml, respectively.

Single-molecule mRNA in situ hybridization

The ISH protocol employed was based on the RNAscope Multiplex Fluorescent Reagent kit v2 from Advanced Cell Diagnostics (ACD) using probes against GlyT2 and against VGAT (Suppl. Table 1), with slight adjustments made. Initially, the sections collected from the spinal cord were washed three times in RNAse-free PBS, each time for an hour, and subsequently transferred onto Superfrost Plus microscope slides. These slides were then stored overnight at −80°C. The next day, the slides underwent two washes in RNAse-free PBS for 2–5 min each at room temperature on a shaker before being placed inside a preheated HybEZ oven at 60°C for 30 min. Following this, the sections were postfixed with prechilled 4% PFA in RNAse-free PBS for 10 min at 4°C. Subsequently, the slides were dehydrated by a single dip in 50% and 70% ethanol in RNAse-free PBS, followed by two dips in 100% ethanol solution for 5 min at room temperature. The sections were then treated with RNAscope hydrogen peroxide for 10 min at room temperature and rinsed three times in RNAse-free H2O to remove the hydrogen peroxide. Next, the slides were incubated in boiling 1× target retrieval solution in RNAse-free H2O for 5 min at a temperature between 95 and 100°C, ensuring the temperature remained within this range. Following this, the slide rack was dipped a few times in RNAse-free H₂O, then again in RNAse-free H₂O, and finally in ethanol to facilitate quick drying of the slides. A hydrophobic barrier was drawn around the sections and allowed to dry for 30 min before proceeding. The protease digestion step was carried out inside a preheated HybEZ oven at 40°C using RNAscope Protease III for 20 min. After a quick double wash in RNAse-free H₂O, the probe hybridization was performed by pipetting the appropriate amount of target probes (VGAT-C3 probe diluted 1:50 in GlyT2-C1 probe) onto the slides, followed by incubation in the oven for 2 h at 40°C. Afterward, the slides underwent two washes with a washing buffer for 5 min at room temperature and then were incubated with amplification-1 reagent at 40°C for 30 min. After two washes for 5 min at RT each, the slides were incubated with amplification-2 reagent at 40°C for 15 min, followed by two washes at RT for 5 min each. The last amplification step involved a 30 min incubation with amplification-3, followed by two washes for 5 min each at RT. Subsequently, sections were incubated in the oven with HRP-C1 for 15 min at 40°C, followed by two washes with a washing buffer for 5 min each at RT. Next, a mixture of TSA Vivid 520 diluted in TSA buffer (1:750) was added to the sections, which were then incubated for 30 min at 40°C. After two 5 min washes at RT, HRP blocker was added to the sections and they were incubated for 15 min at 40°C, followed by two washes at RT for 5 min each. Then, HRP-C3 was added to the sections, which were incubated for 15 min at 40°C, and washed with a washing buffer for 5 min each at RT. Following this, a mixture of TSA Vivid 570 diluted in TSA buffer (1:1,000) was added to the sections, which were then incubated for 30 min at 40°C. After two 5 min washes at RT, HRP blocker was added to the sections, which were incubated for 15 min at 40°C. The final washing step was extended to 10 min.

The coimmunostaining procedure followed immediately after this. The sections were blocked with a blocking buffer (BB; 10% BSA 0.3% Triton X-100 in 1× RNAse-free PBS) at RT for 1 h and then incubated with anti-NeuN antibody (pre-diluted 1:250 in BB) at 4°C overnight. The next day, three 30 min washes in 0.1% Triton X-100 in 1× RNAse-free PBS (PBS-T) were carried out at RT on a shaker before the incubation with secondary antibodies (1:250) for 2 h at RT. Finally, sections were washed three times in PBS-T and once (2–5 min) in wash buffer before applying a few drops of DAPI, which were removed after 30 s. The sections were then mounted with Fluorogold ProLong Antifade mounting medium.

Free-floating immunofluorescence for mouse samples

Immunohistochemistry was performed as previously reported (Bączyk et al., 2020). A blocking buffer was prepared containing 3% BSA/ 0.3% Triton X-100/PBS; or 3% donkey serum/0.3% Triton X-100/PBS for experiments with a goat primary antibody. Free-floating sections from L3-L5 segments were first treated by a blocking solution for a duration of 2 h. Subsequently, they were subjected to a 50 h incubation period at 4°C on a rotary shaker with primary antibodies diluted in the blocking buffer. Following this, the sections underwent a washing process in PBS three times, each lasting for 30 min, before being incubated for 2 h at room temperature on a rotary shaker with secondary antibodies diluted in the blocking buffer. After another round of washing in PBS for three cycles of 30 min each, the sections were mounted, air-dried, and then imaged.

For experiments where goat anti-VAChT was used, a special pretreatment was done. First, the free-floating sections were incubated overnight at 4°C in a mixture of anti-VAChT antibody diluted in 0.3% Triton X-100 /PBS. The day after, the sections were blocked in 3% BSA/0.3% Triton X-100/PBS for 2 h and then incubated with the primary antibodies (all except anti-VAChT) diluted in the blocking buffer. Then the staining continued as usual. Based on our experience, if the goat anti-VAChT is incubated in presence of BSA or horse serum, no section immunolabeling is detected. All the primary and secondary antibodies used are shown in Supplementary Tables 1 and 2.

Mouse samples imaging and analysis

Image processing and analysis were carried out with Fiji (Schindelin et al., 2012).

ISH experiments were acquired with the Keyence BZ-X800E using a 20× air objective (NA 0.75) in a 1,024 × 1,024 pixel 8 bit format, a tile scan of 4 × 4 images was taken and stitched together. Approximately 15 stacks, with z-step size of 1 µm, were collapsed together to form one image; each image was subject to rolling-ball background subtraction (50 pixels diameter) in all channels. Each spinal cord section was then divided into different laminae (1–10), and three regions were used for quantification: region 1 (dorsal horn) consisting of laminae 1–3, region 2 (vantral horn) consisting of laminae 8 and 9 (and partly 7), and region 3 consisting of most parts of lamina 7. The three subtypes of inhibitory interneurons (glycinergic, gabaergic, and mixed) were separately counted in each of these regions.

Confocal imaging was performed with a laser-scanning confocal microscope (Leica DMi8) using either a 20× (overview) or 63× oil objective (NA 1.40; for synapse imaging). Images were acquired in a 1,024 × 1,024 pixel 8 bit format; multichannel acquisition parameters were established according to the emission and excitation spectra, avoiding saturation. MNs at the surface of the section (located in the range 3–10 µm from the nearest section surface) were imaged to avoid issues related to uneven probe penetration or surface cutting artifacts. MN located in the medial motoneuron group were not considered; since MMP9 coimmunostaining was not feasible in all conditions, due to the available imaging channels, in all experiments, MNs of large soma size (>500 µm²) were considered in order to enrich vulnerable MN (77–79% of MN with soma size >500 µm² are MMP9+; Supplementary Fig. 1L). Only large MNs with clear and numerous (three or more) C-boutons were considered, to exclude gamma-MN (which do not display C-boutons; Saxena et al., 2013).

Synapses belonging to a given MN were identified according to multiple criteria: (1) the MN soma must be clearly identifiable, with nucleus (VAChT-negative) and main dendrites visible in the confocal stack; loose pieces of dendrites or cells with unidentifiable nucleus are not considered for analysis; (2) the synapse under consideration must be part of the synaptic ribbon surrounding the MN, whose perimeter is determined by the spatial arrangement of the C-boutons and by the cytoplasmic immunoreactivity for VAChT; isolated synapses or synapses not part of the ribbon continuity are not considered for analysis; (3) the polarity of the synaptic organization must be preserved, so that, on the ideal axis connecting the presynaptic terminal to the MN nucleus, the postsynaptic structures would be located in between the presynaptic structure and the cytoplasmic side. Based on these criteria, the visual examination of the confocal stacks enables the selection of synapses under study based on morphological criteria; because of the large soma size of MN, their perimeter can be reliably visualized (as seen in multiple low-magnification overviews provided, e.g., in Fig. 2A,D,G,J,M or Fig. 4I,K,M). Although some actual MN synapses may be not included due to convolutions of cell surfaces (i.e., false-negative synapses), the very large number of synapses considered per MN minimizes the possible impact of false-negatives. The identification of inhibitory synapses rested on the molecular identity of the presynaptic marker, either GlyT2 or VGAT; since inhibitory synapses in the spinal cord are largely coexpressing GlyT2 and VGAT, the two markers have been used interchangeably and the terminals, identified by either GlyT2 or by VGAT immunoreactivity, are labeled “inhibitory presynaptic terminals.”

Figure 2.

Distinct alterations of GABAR and GlyR in inhibitory synapses on ALS MN. A, D, G, J, M, Representative images of GABAARα2/γ2 (A and J), GlyR (D and M), and Gephyrin in inhibitory synapses (VGAT+ or GlyT2+; G) juxtaposed to a MN identified by VAChT. B, C, Significant increase in GABAR cluster area (µm²; LMM. F_(1,2.83) = 23.12, p = 0.0195, WT: N = 80, SOD1: N = 77, n = 4; fold change = 1.25, 95% CI [1.08, 1.48]; marginal R² = 0.16) but not in the density of GABAR clusters in P45 SOD1 mice (t test on medians, t₍₃₎ = −0.98, p = 0.3988, WT: N = 80, SOD1: N = 77, n = 4; fold change = 1.04, 95% CI [0.92, 1.19]). E, F, No changes in GlyR cluster area [µm²; LMM (log-transformed). F_(1,2.97) = 0, p = 0.9564, WT: N = 82, SOD1: N = 78, n = 4; fold change = 0.98, 95% CI [0.69, 1.41]; marginal R² = 0.00], but significant reduction in the density of GlyR clusters/100 μm in P45 ALS MN (LMM. F_(1,2.97) = 14.35, p = 0.0328, WT: N = 82, SOD1: N = 78, n = 4; fold change = 0.64, 95% CI [0.50, 0.88]; marginal R² = 0.15). H, I, Significant reduction in density of Gephyrin clusters/100 µm (LMM; F_(1,5.92) = 7.98, p = 0.0306; WT: N = 47, SOD1: N = 54, n = 4; fold change = 0.88, 95% CI [0.80, 0.98]; marginal R² = 0.08), but not of Gephyrin cluster size (µm²; LMM; F_(1,2.94) = 1.1, p = 0.3723; WT: N = 47, SOD1: N = 54, n = 4; fold change = 0.90, 95% CI [0.71, 1.24]; marginal R² = 0.01) in inhibitory (VGAT+) synapses in P45 ALS MN. K, L, P110 SOD1 mice showed a decrease in the density of GABAR/100 µm (LMM. F_(1,2.97) = 36.07, p = 0.0095 for fold change = 0.62, 95% CI [0.51, 0.77]; marginal R² = 0.34), but not in the GABAR cluster size [µm²; LMM (log-transformed); F_(1,5.99) = 2.23, p = 0.186 for 2 K; WT: N = 51, SOD1: N = 50, n = 4; fold change = 0.74, 95% CI [0.46, 1.21]; marginal R² = 0.06]. N, O, Similarly P110 mice showed a decrease in the density of GlyR (LMM. F_(1,101.07) = 15.03, p = 0.0002; fold change = 0.79, 95% CI [0.72, 0.89]; marginal R² = 0.12) but not the GlyR cluster size [µm²); LMM (log-transformed). F_(1,6) = 2.91, p = 0.138 for 2N; WT: N = 51, SOD1: N = 50, n = 4; fold change = 0.58, 95% CI [0.27, 1.27]; marginal R² = 0.14]. Blue/pink dot, one neuron; red dot, one animal; red lines join the animal pairs processed together. Scale bars: 10 µm (inset, 1 µm). *p < 0.05, **p < 0.01, ***p < 0.001.

Figure 4.

Gephyrin degradation by functionalized intrabody selectively reduces inhibition and enhances MN activity in P45 ALS mice. A, Experimental design for the intraspinal delivery of AAV9-GFE3. B, Overview image for the injection of GFE3 in the ventral horn of the lumbar spinal cord. C, D, IF staining of VAChT, GFE3-GFP, and Gephyrin shows a reduction in Gephyrin intensity in GFE3+ MN (generalized LMM-log link, gamma distribution. ${\chi }_{(1)}^2=25.84,$ p = <0.0001, Control: N = 52, GFE3+: N = 49, n = 4; fold change = 0.61, 95% CI [0.50, 0.74]; marginal R² = 0.13). E–H, IF staining of GABAARα2/γ2 (E) and GlyR (F) in VGAT+ or GlyT2+ synapses juxtaposed to a MN (VAChT-positive) either GFE3-GFP positive or negative showed (G, H) a significant decrease in the number of GABAR per perimeter (LMM. F_(1,36) = 5.59), p = 0.0236, Control: N = 23, GFE3+: N = 17, n = 3; fold change = 0.56, 95% CI [0.34, 0.92]; marginal R² = 0.11) as well as in the GlyR cluster density (LMM. F_(1,42.25) = 17.67, p = 0.00013, Control: N = 21, GFE3+: N = 23, n = 3; fold change = 0.44, 95% CI [0.34, 0.60]; marginal R² = 0.24) in GFE3+ MN. I, K, IF staining of VAChT, GFE3+, and pCREB shows an increase in pCREB intensity in GFE3+ MN (t test on medians, t₍₇₎ = −2.71, p = 0.03014, Control: N = 432, GFE3+: N = 231, n = 8; fold change = 1.21, 95% CI [1.02, 1.49]). J, L, IF for VAChT, GFE3-GFP, and pPDH shows a decrease in pPDH intensity in GFE3+ MN [LMM (log-transformed). F_(1,3.94) = 20.22, p = 0.0112, Control: N = 255, GFE3+: N = 252, n = 5; fold change = 0.79, 95% CI [0.70, 0.90]; marginal R² = 0.06]. Gray/green dot, one neuron; black dot, one mouse; black lines join the animal pairs processed together. Scale bar: 100 µm overview (B), 20 µm (C, I, J), 10 µm (inset, 1 µm) (E, F). *p < 0.05; **p < 0.01; ***p < 0.001.

For the synaptic quantification, single MNs (selected as described above) were imaged with a 63× (NA 1.4) oil objective. Each image consisted of four 0.35-µm-thick optical sections that were collapsed in a maximum-intensity projection; these stacks were subject to rolling-ball background subtraction (50 pixels diameter) in all channels and then despeckled. Using VAChT signal, the perimeter of the MN was manually outlined and measured; thresholding was done for presynaptic terminals (GlyT2, VGAT, VGluT2, VGluT2) as well as postsynaptic proteins (glycine receptor, GABA_A receptor, Shank2, Neuroligin2, Gephyrin). Then, a watershed filter (Soille, 1990) was used to isolate the presynaptic terminals. Then postsynaptic proteins that are in contact with the inner border of the presynaptic terminals were measured and quantified; postsynaptic terminals touching the preselected postsynaptic proteins were also measured and counted.

For Gephyrin intensity quantification, MNs (selected as described above) were imaged with a 63× (NA 1.4) oil objective. Each image consisted of four 0.35-µm-thick optical sections that were collapsed in a maximum-intensity projection; these stacks were subject to rolling-ball background subtraction (50 pixels diameter) in all channels and then despeckled. Then the cells were outlined, and the intensity of gephyrin was measured in six different places on the cell outline. The average of the measurements was calculated and used as the intensity of gephyrin in that neuron.

For the rest of the IF experiments, where the immunostaining intensity was measured (experiments of pPDH, pCREB, and disease markers), the ventral horn of the spinal cord was imaged with a 20× (NA 1.2) oil objective. Each image consisted of six 1-µm-thick optical sections that were collapsed in a maximum-intensity projection; these stacks were subject to rolling-ball background subtraction (50 pixels diameter) in all channels. Using VAChT signal (which highlights to cytoplasm and excludes the nucleus), the region of interest was manually traced (around the MN, excluding the nucleus), and the mean gray value (MGV) was measured.

In all immunostaining procedures, close control of temperature, incubation time, antibody concentration, and reagents preparation are closely monitored (and include repeated wt samples for quality control) and performed as previously reported (Saxena et al., 2013; Bączyk et al., 2020).

Statistical analysis

Statistical comparisons of cluster size, cluster density, immunofluorescence intensity, and synaptic terminal areas between experimental groups were conducted using mixed-effect models. We specified separate random intercepts for slide and animal, which correctly partitions slide-level variability from between-animal variability and avoids pseudoreplication. Models were fitted with lmer from lme4 (https://cran.r-project.org/web/packages/lme4/index.html) in R 4.4. When a model exhibited a singular fit, we reduced the random-effects structure to the simplest justified form. Model assumptions were evaluated with DHARMa's (https://cran.r-project.org/web/packages/DHARMa/vignettes/DHARMa.html) quantile-deviation, distribution, dispersion, and outlier tests; residual Q–Q plots and histograms were also inspected. If assumptions were violated, we explored the following alternatives: (1) log-transforming the response or (2) fitting a generalized linear mixed model with a gamma distribution and log link using glmmTMB (https://cran.r-project.org/web/packages/glmmTMB/index.html). In the few instances where no mixed-effects specification satisfied diagnostics, median values were computed per animal, and paired t tests at the slide level were applied; residual plots and Shapiro–Wilk tests confirmed that the assumptions for the paired t test (normality of paired differences and absence of influential outliers) were met. For designs with more than two experimental groups, pairwise post hoc comparisons of estimated marginal means were conducted with Tukey's honestly significant difference adjustment using emmeans. Results are reported as model-estimated marginal means ± standard deviation (SD), and the stated “n” always denotes the number of individual animals (biological replicates).

Group effects are reported as fold change, calculated as the ratio of model-based estimated marginal means; for linear mixed models, this represents a re-expression of marginal means and does not affect statistical inference. In addition, Nakagawa's marginal R² is reported as a standardized effect size measure summarizing the proportion of variance explained by fixed effects in linear and generalized linear mixed models.

A significance level of p < 0.05 was deemed statistically significant, asterisks in both figures and their respective legends are *p < 0.05, **p < 0.01, or ***p < 0.001.

Results

Distinct alterations of GABAR and GlyR in inhibitory synapses on ALS MN

Preliminary to the investigation of changes in inhibitory synapses, we confirmed the overall integrity of the inhibitory neurons population at the age of P45: the number of inhibitory interneurons (coexpressing GlyT2 and VGAT) constituted ∼25% of the total number of NeuN+ in dorsal laminae and 40% in intermediate laminae and ventral horn (in agreement with previous literature; Chaudhry et al., 1998; Polgár et al., 2013) and their number in dorsal, intermediate, and ventral laminae was comparable in SOD1(G93A; SOD1) and WT animals at P45 (Fig. 1A–C).

Figure 1.

Reduced activity of vulnerable MN in SOD ALS without loss of inhibitory interneurons. A, Overview images of spinal cord of P45 WT and SOD1 mice showing single-molecule mRNA in situ hybridization identifying GlyT2+ neurons (green) and VGAT+ neurons (red) with a NeuN staining of neurons (magenta) in different laminae (B) of the spinal cord. Scale bar: 100 µm and 40 μm. C, No changes in the numbers of the inhibitory interneurons between WT and SOD1 mice in the ventral horn, in lamina VII, or in the dorsal horn (t test, blue/pink dot, one hemichord; red dot, one animal; n = 5). D, E, IF for phosphorylated pyruvate dehydrogenase (pPDH) and VAChT in SOD1 and WT at P22 shows no changes in pPDH intensity [linear mixed model (LMM)-log-transformed; F_(1,3) = 3.45, p = 0.16, WT: N = 540, SOD1: N = 532, n = 4; fold change = 1.18, 95% CI [0.89, 1.58]; marginal R² = 0.03]. F–G, IF for pPDH, MMP9, and VAChT in SOD1 and WT at P45 shows increased pPDH intensity in VAChT+ MN SOD1 mice (generalized LMM-log link, gamma distribution; ${\chi }_{(1)}^2=13.54,$ p = 0.0002, WT: N = 952, SOD1: N = 1044, n = 6; fold change = 2.17, 95% CI [1.43, 3.27]; marginal R² = 0.23). H, Unchanged pPDH intensity in VAChT+/MMP9− MN [LMM (log-transformed). F_(1,5) = 0.78, p = 0.4181, WT: N = 436, SOD1: N = 354, n = 6; fold change = 1.20, 95% CI [0.75, 1.90]; marginal R² = 0.02] but (I) increase in pPDH in MMP9+ MN (t test on medians, t₍₅₎ = −4.01, p = 0.0102, WT: N = 516, SOD1: N = 589, n = 6; fold change = 2.25, 95% CI [1.25, 11.41]). J, IF for pS6, MMP9, and VAChT in SOD1 and WT at P45 shows (K–M) decreased pS6 intensity in MMP9+ MN (t test on medians, t₍₄₎ = 2.78, p = 0.0500, WT: N = 521, SOD1: N = 566, n = 5; fold change = 0.71, 95% CI [0.55, 1.00]) but no changes in pS6 intensity in MMP9− (t test on medians, t₍₄₎ = 0.43, p = 0.691, WT: N = 176, SOD1: N = 209, n = 5, 1L) or mixed MN in P45 SOD1 mice (t test on medians, t₍₄₎ = 2.13, p = 0.1, WT: N = 355, SOD1: N = 351, n = 5). Blue/pink dot, one neuron; red dot, one animal. Red lines join the animal pairs processed together. Scale bars: 40 μm. *p < 0.05; **p < 0.01; ***p < 0.001.

Next we assessed the overall MN activity by monitoring the level of phosphorylated pyruvate dehydrogenase (pPDH; pyruvate dehydrogenase) whose phosphorylation increases upon decreased neuronal activity (either increased inhibition or decreased excitation; Yang et al., 2024) and of the levels of phosphorylated ribosomal protein S6 (pS6, an independent marker positively correlated with activity; Knight et al., 2012). Whereas pPDH intensity was comparable between WT and SOD1 MN at P22 (Fig. 1D,E), at P45 the SOD1 MN displayed a significant increase in average pPDH (Fig. 1F,G). The increase in pPDH characterized MMP9+ (vulnerable) MN but not MMP9− (nonvulnerable) MN (Fig. 1H,I). These findings were confirmed by the decrease in pS6 levels, also pointing to reduced activity (Fig. 1J,M). In order to compensate for possible differences in depth-dependent staining efficiencies, we assessed the intensity of the VAChT staining and compensated the pPDH intensity for the VAChT intensity. VAChT intensity was comparable in WT and SOD1 MN (overall, MMP9+ or MMP9−; Supplementary Fig. 1A–C) and differences in pPDH remained significant upon VAChT compensation (Supplementary Fig. 1D–F).

Then we set out to explore the structural integrity of inhibitory synapses on MN, reasoning that we anticipated upregulation of GABAR or GlyR1 if the reduced MN activity was due to increased inhibition or their downregulation if the inhibitory synapses were homeostatically reacting to reduced activity. In order to enrich vulnerable MN, we considered here and henceforth only MN with a cross-sectional area of >500 µm², which included ∼77–79% of MMP9+ MN (Supplementary Fig. 1L).

Since the largest majority (∼90%, not shown) of presynaptic terminals displayed immunoreactivity for both GlyT2 and VGAT, in agreement with the established literature (Chaudhry et al., 1998), we interchangeably used GlyT2 or VGAT as presynaptic marker of inhibitory synapses.

Interestingly the size of GABARα2/γ2 clusters (but not their density, defined as number of clusters per MN perimeter) in inhibitory synapses (Fig. 2A,B) was significantly increased in SOD1 MN (Fig. 2A–C), whereas the density of the GlyR1 clusters (but not their size) in inhibitory synapses was significantly decreased (Fig. 2D–F; since GlyR1 are substantially smaller than GABAR clusters, a decrease in size of GlyR1 may make them undetectable, thus not contributing to cluster size metrics). We also evaluated the distribution of the inhibitory synapses scaffold protein Gephyrin. As for GABAR and GlyR, multiple Gephyrin clusters were opposed to individual presynaptic inhibitory terminals (Fig. 2G): a significant decline in Gephyrin clusters density (but not in their size) was detected (Fig. 2G–I). However, when the more advanced stage of disease was considered (P110, corresponding to symptomatic stage; Ouali Alami et al., 2018), we detected a significant decrease in the overall density of both GABAR (Fig. 2J–L) and GlyR (Fig. 2M–O).

On the other hand, we did not detect changes in the size or density of inhibitory terminals in SOD1 MN at P45 (Supplementary Fig. 1M,N), but a significant loss of terminals was seen at P110 (Supplementary Fig. 1O–P). Ensuring an unbiased sampling of the MN populations in the two genotypes, the size of MN taken into consideration was comparable in SOD1 and WT across all experiments at P45 (Suppl. Fig. 1G–I) as well as at P110 (although fewer MNs were available for analysis at this stage; Suppl. Fig. 1J,K).

Taken together, these findings imply an increased representation of GABAR and a decreased representation of GlyR in inhibitory synapses at presymptomatic stage and a more extensive loss of pre- and postsynaptic structures at fully symptomatic stage.

Chemogenetic stimulation of activity enhanced inhibitory receptor density in ALS MN

After identifying alterations in the synaptic GlyR1 and GABAR in ALS MN, we set out to establish their response to changes in MN activity. We reasoned that if loss of inhibitory receptors was a response to reduced MN activity, stimulation of MN firing should induce a homeostatic upregulation of their size, whereas if it was due to excessive excitation, they would respond to chemogenetic stimulation with a further downregulation. We used the PSAM/PSEM chemogenetic system (Magnus et al., 2011; Bączyk et al., 2020; Ouali Alami et al., 2020; Song et al., 2023) as well as a cAMP/PKA coupled Gs-DREADD (an approach fully orthogonal to the PSAM/PSEM system) to either enhance or suppress MN excitability.

The chemogenetic systems were expressed in MN by intraspinal injection of AAV9 in SOD1/ChAT-Cre mice (Fig. 3A,B); daily agonist treatment (PSEM for PSAMs or CNO for Gs-DREADDs) was started 17 d after injection (at P21) and lasted for 8 d. Activation of cation-permeable PSAM (PSAM-act; identified by the binding of fluorescently-labeled bungarotoxin; Magnus et al., 2011) resulted in the decrease in phospho-PDH (Fig. 3C,D), whereas pPDH levels increased in MN in which the anion-permeable PSAM (PSAMinh) was activated (Fig. 3E,F). Gs-DREADD activation resulted in the decrease in pPDH immunoreactivity in the infected MN (Fig. 3G,H), confirming the differential excitability engagement by the three chemogenetic constructs.

Figure 3.

Chemogenetic stimulation of MN excitability enhanced inhibitory receptor numbers at P45. A, Experimental design for the intraspinal delivery of AAV9-PSAMact, AAV9-PSAMinh, or AAV9-DREADD-Gs. B, Overview images for the injection of PSAMs and DREADD-Gs in the ventral horn of the lumbar spinal cord. C–H, IF for pPDH, VAChT, and chemogenetic construct (PSAMact; PSAMinh; DREADD-Gs) shows a reduction in pPDH intensity in PSAMact+ MN (t test, p = 0.0086, Control: N = 359, PSAMact+: N = 141, n = 4), an increase in PSAMinh+ MN [LMM (log-transformed). F_(1,4.81) = 12.71, p = 0.0172, Control: N = 444, PSAMinh+: N = 406, n = 6; fold change = 1.21, 95% CI [1.05, 1.39]; p = 0.017; marginal R² = 0.02], and a decrease in Gs-DREADD+ MN [LMM (log-transformed). F_(1,3.98) = 40.09, p = 0.0032, Control: N = 382, DREADD+: N = 434, n = 5; fold change = 0.57, 95% CI [0.43, 0.77]; marginal R² = 0.22]. I–N, IF for GABAARα2/γ2 in VGAT+ synapses juxtaposed to a MN either expressing the chemogenetic construct (PSAMact; PSAMinh; DREADD-Gs) or negative identified by VAChT. A significant increase in the GABAR cluster density in PSAMact+ [LMM (log-transformed). F_(1,3.93) = 9.76, p = 0.0362, Control: N = 53, PSAMact+: N = 60, n = 5; fold change = 1.39, 95% CI [1.05, 1.84]; marginal R² = 0.10] and DREADD-Gs+ MN (LMM. F_(1,100.48) = 5.69, p = 0.0190, Control: N = 56, DREADD+: N = 51, n = 6; fold change = 1.15, 95% CI [1.02, 1.31]; marginal R² = 0.04) was observed but not in PSAMinh+ MN [LMM (log-transformed). F_(1,105.91) = 2.22, p = 0.139, Control: N = 54, PSAMinh+: N = 59, n = 5; fold change = 1.18, 95% CI [0.95, 1.47]; marginal R² = 0.02]. O–T, IF for GlyR in GlyT2+ synapses juxtaposed to a MN expressing chemogenetic constructs (PSAMact; PSAMinh; DREADD-Gs+) or negative identified by VAChT. The GlyR cluster density increased in PSAMact+ (LMM. F_(1,109.21) = 6.22, p = 0.0141, Control: N = 53, PSAMact+: N = 64, n = 5; fold change = 1.17, 95% CI [1.03, 1.36]; marginal R² = 0.03) and DREADD-Gs+ MN (LMM. F_{(1, 99.05)} = 4.94, p = 0.0285, Control: N = 55, DREADD+: N = 51, n = 6; fold change = 1.12, 95% CI [1.01, 1.24]; marginal R² = 0.03) but not in PSAMinh+ MN (LMM. F_(1,105.13) = 0.86, p = 0.3554, Control: N = 56, PSAMinh+: N = 56, n = 5; fold change = 0.94, 95% CI [0.84, 1.07]; marginal R² = 0.00). Gray/red/green/purple dot, one neuron; black dot, one mouse; black lines join the animal pairs processed together. Scale bar: 100 µm overviews (B), 40 µm (C, E, G), 10 µm (inset, 1 µm) (I–S). *p < 0.05; **p < 0.01; ***p < 0.001.

To assess the effect of chemogenetics on GABAR and GlyR clusters, we compared MN expressing the PSAM or DREADD constructs with uninfected MN in the same spinal cord section (so that the noninfected MN would be exposed to the same pharmacology and tissue processing). Interestingly, we observed an increase in cluster density of both GABAR and GlyR at inhibitory synapses upon upregulating excitability through PSAM-act (Fig. 3I,J,O,P, respectively) as well as through Gs-DREADD (Fig. 3M,N,S,T, respectively). However, no significant changes in receptor density of GABAR or GlyR was seen upon the suppression of excitability by PSAMinh (Fig. 3K,L,Q,R, respectively). Chemogenetic PSAM and Gs-DREADD manipulations did not result in significant changes in the size or number of presynaptic terminals (Suppl. Fig. 2A–F).

Taken together, these findings show that GlyR and GABAR at inhibitory synapses on ALS MN homeostatically respond to changes in MN excitability and are upregulated upon increased activity.

Degradation of Gephyrin by a functionalized intrabody decreases inhibition and reduces disease burden in MN

If the decrease in inhibitory synapses observed in SOD1 MN constitutes a homeostatic response to decreased MN excitability peculiar to this disease stage (Martínez-Silva et al., 2018), we hypothesized that a further decrease in inhibition may be beneficial. On the contrary, if the observed reduction in inhibitory receptor clusters is pathogenic, further reducing the inhibitory drive would result in the worsening of the disease pathways.

We exploited an intrabody directed against Gephyrin fused to an E3 ubiquitin ligase to induce the selective degradation of Gephyrin (previously reported; Gross et al., 2016) and the loss of GlyR1 and GABAR. MN-selective expression of the GFP-tagged anti-Gephyrin FingR-E3 construct (henceforth GFE3) was achieved by intraspinal injection of AAV9 (under EF1a promoter and DIO architecture) in SOD1/ChAT-Cre mice at P21 (Fig. 4A,B) followed by killing at P45.

GFE3⁺ MN displayed reduced Gephyrin immunoreactivity (Fig. 4C,D) and the concomitant decrease in GABAR (Fig. 4E,G) and GlyT2 clusters (Fig. 4F,H). Trends toward reduced presynaptic terminal size were detected upon Gephyrin degradation (Suppl. Fig. 3A,B), indicating a possible small effect on the stability of the presynaptic terminals. Likewise, GFE3 did not affect excitatory synapses in terms of size and number of Shank clusters and of VGluT2+ terminals (Suppl. Fig. 3C–G).

Importantly, MN expressing GFE3 displayed a significant increase in phosphorylated CREB (marker of increased activity; Fig. 4I,K) and at the same time a significant decrease in pPDH (Fig. 4J,L); taken together, these two markers confirm that GFE3 results in increased MN activity.

We then employed a panel of disease markers to evaluate the impact of GFE3 expression on pathogenic pathways. As control of the expression of a ubiquitin ligase, the effect of random-sequence E3-conjugated nanobody (RandE3; Fig. 5A,B) was assessed in independent experiments. Compared with noninfected MN (from the same section), GFE3+ cells displayed a consistent decrease in the burden of misfolded SOD1 (BH10 antibody; Fig. 5C,D), in the level of autophagy marker LC3A (Fig. 5E,F), and in the immunoreactivity for the oxidative stress marker 8OH-DG (Fig. 5G,H). On the other hand, RandE3+ expression in MN did not affect Gephyrin (Fig. 5I,J) or disease markers misfolded SOD1 (Fig. 5K,L) and LC3A (Fig. 5M,N).

Figure 5.

Decrease of inhibition by Gephyrin-targeted intrabody reduces disease burden in P45 ALS MN. A, Experimental design for the intraspinal delivery of AAV9-GFE3 or its control AAV-RandE3. B, Overview images for the expression of GFE3-GFP or RandE3-GFP in the ventral horn of the lumbar spinal cord. C–H, IF for VAChT, GFE3-GFP, and B8H10 (C), LC3A (E), or 8ohDG (G) shows a reduction in the intensity of all three disease markers: B8H10 (t test on medians, t₍₄₎ = 11.81, p = 0.0003, Control: N = 83, GFE3+: N = 57, n = 5; fold change = 0.89, 95% CI [0.87, 0.92]), LC3A (LMM. F_(1,138.21) = 18.76, p < 0.0001, Control: N = 86, GFE3+: N = 59, n = 5; fold change = 0.87, 95% CI [0.82, 0.92]; marginal R² = 0.08), and 8oDG (t test on medians, t₍₄₎ = 8.06, p = 0.0013, Control: N = 214, GFE3+: N = 80, n = 5; fold change = 0.89, 95% CI [0.86, 0.93]) in GFE3+ MN. I–N, IF staining of VAChT, RandE3-GFP, and Gephyrin (I), B8H10 (K), and LC3A (L) shows no changes in the intensity of all three markers: Gephyrin (LMM. F_(1,63.36) = 0.11, p = 0.7379, Control: N = 33, RandE3+: N = 35, n = 4; fold change = 1.02, 95% CI [0.92, 1.15]; marginal R² = 0.00), B8H10 [LMM (log-transformed). F_(1,2.81) = 1.51, p = 0.3125, Control: N = 194, RandE3+: N = 125, n = 4; fold change = 1.07, 95% CI [0.87, 1.31]; marginal R² = 0.00], LC3A (LMM. F_(1,251.13) = 3.48, p = 0.063, Control: N = 136, RandE3+: N = 120, n = 5; fold change = 1.07, 95% CI [1.00, 1.16]; marginal R² = 0.01) in RandE3+ MN. Gray/red/green dot, one neuron; black dot, one mouse; black lines join the animal pairs processed together. Scale bars: 100 µm overviews (B), 20 µm (C–M). *p < 0.05, **p < 0.01, ***p < 0.001.

Taken together, these data imply that decreasing inhibition by selective, targeted Gephyrin degradation results in stimulated MN firing and reduced burden of disease markers, thus delivering a beneficial effect.

Simultaneous decrease in inhibition and stimulation of neuronal activity does not have an additive beneficial outcome

According to the excitotoxic hypothesis of ALS (Rothstein, 2009), increasing MN firing should have detrimental consequences. Although our results may appear to oppose this concept, we reasoned that while restoring MN firing within a physiological range may be protective (Depp et al., 2018), substantially exceeding it may be pathogenic (Bading, 2017).

To test this hypothesis, we performed a simultaneous PSAM-act chemogenetic stimulation of MN and GFE3 suppression of inhibition in the same MN. Spinal cord was injected with a 1:1 mixture of the two AAVs at 21 and, after 17 d, the PSEM agonist was delivered for 8 consecutive days (Fig. 6A). MNs were either GFE3+ (green), PSAM-act+ (red), or expressing both (yellow; Fig. 6B,C,E,G,I).

Figure 6.

Simultaneous decrease of inhibition and stimulation of neuronal activity does not have an additive beneficial outcome. A, Experimental design for the intraspinal codelivery of AAV9-PSAMact and AAV9-GFE3. B, Overview image for the injection of coexpression PSAMact and GFE3-GFP in the ventral horn of the lumbar spinal cord. C, D, IF for VAChT, GFE3, PSAMact, and pPDH shows a reduction in pPDH intensity in GFE3+ MN (p = 0.0030; fold change = 0.78, 95% CI [0.65 0.94]) and in PSAMact+ MN (p = 0.0226; fold change = 0.82, 95% CI [0.68 0.98]) compared with control, while double-positive MN had the most decrease compared with all three groups (p < 0.0001 compared with control, fold change = 0.59, 95% CI [0.48 0.72]; p = 0.0194 compared with GFE3+ MN, fold change = 0.75, 95% CI [0.58, 0.96]; p = 0.0016 compared with PSAMact+ MN, fold change = 0.72, 95% CI [0.57, 0.91]; Control: N = 346, GFE3+: N = 83, PSAMact+: N = 84, double+: N = 71, n = 5). E–J, IF staining of VAChT, GFE3-GFP, PSAMact+, and B8H10 (E) or LC3A (G) or 8ohDG (I) shows a reduction in of misfolded SOD1 in GFE3+ MN (p = 0.0298; fold change = 0.77, 95% CI [0.62 0.97]) and PSAMact+ MN (p = 0.0088; fold change = 0.75, 95% CI [0.60 0.93]) but not in double-transfed MN (fold change = 0.92, 95% CI [0.74 1.14]; Control: N = 252, GFE3+: N = 158, PSAMact+: N = 63, double+: N = 133, n = 5). LC3A was reduced in GFE3+ MN (p < 0.0001; fold change = 0.85, 95% CI [0.77 0 = 0.92]), in PSAMact+ MN (p = 0.0034; fold change = 0.82, 95% CI [0.70 0.95]), but not in double-positive MN (fold change = 0.84, 95% CI [0.76 1.14]) compared with control MN (Control: N = 249, GFE3+: N = 134, PSAMact+: N = 31, double+: N = 83, n = 5). Decrease of 8ohDG was observed in GFE3+ MN (p = 0.0266; fold change = 0.76, 95% CI [0.59 0.97]) and PSAMact+ MN (p = 0.0166; fold change = 0.73, 95% CI [0.57 0.95]) but not in double-positive MN (fold change = 0.88, 95% CI [0.69 1.13]; Control: N = 145, GFE3+: N = 106, PSAMact+: N = 72, double+: N = 121, n = 5). For all experiments, pairwise post hoc comparisons of estimated marginal means were conducted with Tukey's honestly significant difference adjustment using means. Gray/green/red/orange dot, one neuron; black dot, one animal; black lines join the animal pairs processed together. Scale bar: 50 µm for overview images (B) and 20 µm for MNs (C–I). *p < 0.05; **p < 0.01; ***p < 0.001.

In agreement with their anticipated effect, pPDH was decreased by either PSAM-act or GFE3, and double manipulation resulted in a further decrease (Fig. 6C,D). However, we found that whereas GFE3+ or PSAMact+ MN displayed a reduced burden of misfolded SOD, the double GFE3+/PSAM+ MN did not display any additive beneficial effect (Fig. 6E,F). Furthermore, the levels of the autophagy marker LC3A (Fig. 6G,H) and of the oxidative damage marker 8OH-DG (Fig. 6I,J) were decreased by GFE3 or PSAM, but again no additive effect was observed in the GFE3+/PSAM+.

Thus, these findings demonstrate that while single manipulations of MN activity may be beneficial, no major additive effects are detected when multiple manipulations are put in place.

Discussion

In the present work, we demonstrate the alteration of inhibitory synapses at postsynaptic level on MN in the SOD1 mouse model, at a stage predating neuronal degeneration and muscle denervation, and we show that glycinergic and gabaergic receptors are differentially affected. By employing several independent chemogenetic approaches, we show that clustering of receptors at inhibitory synapses remains homeostatically regulated, increasing upon increased neuronal activity. Finally, by employing a selective functionalized intrabody, we observed a decrease in misfolded SOD1 burden, in autophagy overload and in oxidative stress upon further reducing inhibition.

Several recent papers have investigated the involvement of inhibitory synapses in SOD1-ALS, aiming to establish the contribution of inhibition to the alteration of an excitation/inhibition balance in ALS. Electrophysiological evidence obtained in vitro (Chang and Martin, 2011) and more recently in ex vivo spinal cord (Nascimento et al., 2024) suggests an impaired quantal size associated with decreased glycine receptor density occurring at glycinergic synapses in the MN–Renshaw cell microcircuit (but not in other inhibitory inputs) at early presymptomatic stages (P22). However, electrophysiological recordings of MN in vivo (at P50; Nascimento et al., 2024) reveal no net alteration in inhibitory conductance from Renshaw cells. Whereas the ex vivo recordings were performed in the presence of the pan-GABAR antagonist gabazine (therefore isolating glycinergic responses), the in vivo recordings included both GABA and glycine responses. Since Renshaw cells have been reported to also release GABA (most inhibitory synapses display VGAT and GlyT2 colocalization; González-Forero and Alvarez, 2005), our observation of simultaneous downregulation of GlyR and upregulation of the GABAR may partially account for the unchanged overall inhibition. Our data are also in agreement with previous reports of reduced inhibitory synapse numbers on MN (Allodi et al., 2021) and preserved inhibitory neurons at early stages (Montañana-Rosell et al., 2024).

A central point of the debate regards the pre- or postsynaptic origin of the alterations of inhibition observed in SOD-ALS. Counts of inhibitory interneurons are not altered at P45 (our evidence and Montanana-Rosell et al., 2024); however, synaptic remodeling (as suggested for Renshaw cells; Wootz et al., 2013) or early functional alterations of presynaptic terminals (although presynaptic alterations were not detected by electrophysiological recordings; Nascimento et al., 2024) may contribute to alterations of inhibition. Although we have not observed a loss in presynaptic terminals number or size (the use of immunolabeling vs genetic labeling and of watershed algorithms vs thresholding may account for the difference with Allodi et al., 2021), structural characterization of presynaptic inhibitory terminals at higher resolution level may be warranted to address this point.

Our data show that GlyR and GABAR clusters are homeostatically increased by PSAM or DREADD-Gs-mediated stimulation of excitability but remain unchanged after PSAM-mediated suppression (the latter effect possibly due to a biological floor effect and/or technical limitations in resolving clusters at the lower end of the size spectrum). Thus, MNs display substantial adaptation abilities to changes in synaptic excitation or activity loads (also reported for GPCR; Antonucci et al., 2024) and in general agreement with what previously hypothesized (Kuo et al., 2020). Interestingly, no change was detected in the size of the presynaptic terminals (as assessed by GlyT2 or VGAT immunolabeling) under chemogenetic manipulation, possibly indicating a smaller effect size or slower response at presynaptic level.

GlyR and GABAR appear to be differentially regulated in ALS MN, with GlyR downregulated and GABAR upregulated compared with WT. If this was due to acute changes in excitation, such as in the context of excitotoxicity, one may expect a decrease in GABAR, since acute excitotoxic glutamatergic stimulation downregulates GABAR (Garcia et al., 2021). However, chronic changes in MN excitability by decreasing glycinergic inhibition do not result in over neurodegeneration but in the compensatory expression of GABAR (Tadros et al., 2014); thus a similar compensation of GlyR downregulation by GABAR may be occurring in ALS MN. If GABAR follows GlyR alterations, what drives the GlyR alterations in the first place? On the one side, the response to the modulation of activity is compatible with a cell-autonomous homeostatic feedback loop (Wen and Turrigiano, 2024). On the other side, GlyR clustering is increased by its own activation (Nakahata et al., 2017), thus dysfunctional or diminished release of Glycine from presynaptic terminals may also contribute to the declustering of GlyR in postsynaptic structures.

The beneficial effect observed upon gephyrin degradation by GFE3 (Gross et al., 2016) is consistent with the concept of insufficient activity-dependent neuroprotection at this stage of the disease in vulnerable MN (Roselli and Caroni, 2015; Bączyk et al., 2020). Similar improvement in disease burden has been also achieved by direct stimulation of MN excitability through PSAM/PSEM chemogenetic (Saxena et al., 2013; Ouali Alami et al., 2020) or DREADD-mediated PKA stimulation (Bączyk et al., 2020) as well as by optogenetic or pharmacological stimulation of iPSC-derived MN from ALS patients (Naujock et al., 2016; Catanese et al., 2021). Activity-dependent neuroprotection has been observed in multiple systems (Zhang et al., 2009; Depp et al., 2018), including other MN diseases (Simon et al., 2021), and depends on CREB and SRF activity-dependent transcriptional programs, since activity-dependent neuroprotection is lost in SRF^−/− MN (Song et al., 2023). Of note, excitatory synapses are also affected by the disease process, reducing excitation (Baczyk et al., 2020) to MN.

Nevertheless, our findings, although in apparent contradiction to the simplest and most traditional model of excitotoxicity, are not incompatible with the current understanding of synaptic versus extrasynaptic signaling competition (Yan et al., 2020, 2024): in this model, synaptic signaling to the nucleus is shut down by extrasynaptic signaling originated from the interaction of NMDAR with TRPM4 channels. Gephyrin degradation modifies the excitation/inhibition balance in a condition of reduced excitatory drive (Baczyk et al., 2020) and may enhance the activity-dependent neuroprotection (in agreement with the improvement in disease markers). Of note, Gephyrin degradation also produced a small effect on the size of presynaptic terminals, suggesting that over longer timeframes, the effects of GFE3 might also retrogradely affect neurotransmitter vesicles pools or inhibitory boutons structure.

Interestingly, combined Gephyrin degradation and PSAM activation does not result in increased toxicity but also does not result in additive protective effects. This unexpected finding may point toward a ceiling-effect in activity-dependent neuroprotection possibly due to incomplete restoration of synaptic inputs (Bączyk et al., 2020; Aousji et al., 2023). Since MNs display substantial homeostatic capabilities even in disease states (Kuo et al., 2020; Antonucci et al., 2024), these feedbacks may dampen the effectiveness of activity-dependent interventions. Nevertheless, the phospho-PDH levels are still decreased upon Gephyrin degradation or actPSAM intervention, implying that the MN still experience an increased activity upon Gephyrin degradation and do not supracompensate to become less excitable. Thus, there may be an intrinsic limit to the amount of neuroprotection delivered through activity manipulations in MN.

One worth-noting limitation of this study is the possible discrepancies between magnitude of the effects (and statistical significance) and their biological meaning. Although statistically significant, several results have small apparent effect; this may on the one side indicate the relatively minor contribution of inhibition in shaping MN vulnerability. On the other side, it is unclear how much apparently small effects may translate into biological responses, which could be nonlinear and influenced by threshold crossing (i.e., small changes may push cells above or below the threshold for degeneration).

In conclusion, our findings confirm the involvement of inhibitory synapses in the pathophysiology of ALS MN and uncover the homeostatic responses of inhibition to excitability modulation of MN and, on the other side, the responses of disease pathways to manipulations of inhibition. Nevertheless, our findings elicit additional questions: how much of the inhibitory synapses phenotype is due to dysfunction of inhibitory interneurons versus cell-autonomous MNs alterations? Causative and reactive mechanisms may contribute to observed synaptic alterations.