Glycosphingolipids are linked to elevated neurotransmission and neurodegeneration in a Drosophila model of Niemann Pick type C

Niemann Pick diseases are lysosomal lipid storage disorders causing early-onset neurodegeneration and childhood death (Niemann, 1914; Pick, 1933; Liu et al., 2020). Niemann Pick type C (NPC) is characterized by lysosomal accumulation of cholesterol and glycosphingolipids (GSLs) (Brady et al., 1966; Newton et al., 2020; Wanikawa et al., 2019). Ninety-five percent of NPC cases are caused by variants in the NPC1 gene (Carstea et al., 1997), encoding an endolysosomal 13-pass transmembrane protein containing a sterol-sensing domain (Iannou, 2000; Long et al., 2020; Millard et al., 2005). Although cholesterol is widely considered to be the primary pathogenic lipid, efforts to disrupt low-density lipoprotein uptake do not impact NPC progressive neurodegeneration (Erickson et al., 2000; Newton et al., 2018). In contrast, miglustat inhibition of glucosylceramide (GlcCer) synthase, the first committed step in the GSL synthesis pathway, clinically stabilizes NPC patient symptoms (Patterson et al., 2007, 2020; Platt et al., 1994). Moreover, miglustat reverses hippocampal glutamatergic neurotransmission hyperexcitability in NPC disease model mice, and miglustat blocks Purkinje neuron death in NPC disease model cats (D'Arcangelo et al., 2016; Stein et al., 2012). These exciting advances suggest that the GSL synthesis pathway could be key to understanding the NPC condition, and that GSL impairments may provide the causal mechanistic link between neurotransmission and neurodegeneration defects.

As in NPC patients and other disease models, the Drosophila NPC model shows neuronal cholesterol accumulation, impaired synaptic transmission, movement defects, neurodegeneration and shortened lifespan (Huang et al., 2005; Phillips et al., 2008). Drosophila has two Npc1 genes, with Npc1a widely expressed throughout the body but Npc1b only expressed within the midgut (Fluegel et al., 2006; Huang et al., 2005). In the Drosophila system, the neuromuscular junction (NMJ) is a well-characterized model glutamatergic synapse (Broadie and Bate, 1995; Chou et al., 2020). Mouse hippocampal glutamatergic synapses manifest dysfunction in the absence of Npc1 (Cariati et al., 2021; D'Arcangelo et al., 2011, 2016; Mitroi et al., 2019). Previous studies show that NPC model mice exhibit an increase in hippocampal synaptic transmission and defects in long-term potentiation (LTP) (D'Arcangelo et al., 2011, 2016; Mitroi et al., 2019), with links to cholesterol accumulation but little study of possible GSL contributions. The Drosophila NMJ provides a powerful genetic system for understanding the lipid storage mechanism of synaptic dysfunction through genetic manipulation of the GSL biosynthetic pathway. Critically, GSLs are established to regulate Drosophila NMJ structure/function (Huang et al., 2018), and GSLs are known to activate apoptosis through inducing autophagy, endoplasmic reticulum stress, lysosomal membrane permeabilization and the release of pro-apoptotic mitochondrial proteins (Garcia-Ruiz et al., 2015; Park and Park, 2020; Rippo et al., 2000; Wheeler et al., 2019). These studies suggest a possible link between GSL and neurotransmission and neurodegeneration defects.

Drosophila mutants in serine palmitoyl transferase (lace gene), the rate-limiting step in sphingolipid synthesis, exhibit elevated NMJ neurotransmission from fewer NMJ synaptic boutons and also display increased apoptosis (Kraut, 2011; West et al., 2018). Likewise, mannosyl glucosylceramide (MacCer), a GSL synthesis pathway intermediate similar to mammalian lactosylceramide (LacCer), regulates NMJ structure and trans-synaptic signaling (Huang et al., 2018; Wandall et al., 2005). Other GSLs are also implicated in conserved pathogenic phenotypes in several models of neurodegenerative diseases. GlcCer accumulation is thought to be the underlying cause of spatial ataxia in Gaucher's disease, caused by loss of glucocerebrosidase 1 (GBA1), the endolysosomal enzyme that de-glycosylates GlcCer (Malekkou et al., 2018). Heterozygous variation in GBA1 is a leading genetic risk factor for Parkinson's disease, characterized by both motor ataxia and progressive neurodegeneration, similar to the NPC state (Stoker et al., 2020). Accumulated GlcCer storage within the lysosome is also thought to lead to an increase in the lumenal pH (Sillence, 2013), and lysosomal de-acidification is well known to result in perforations in the organelle membrane, which cause mitochondrial oxidative stress and activate an intrinsic apoptotic pathway (Vest et al., 2022; Wheeler and Sillence, 2020). We therefore hypothesized that similar GSL trafficking accumulation defects cause elevated glutamatergic neurotransmission and neurodegeneration in NPC.

To test this hypothesis, we used mutants in brainiac (brn), encoding the beta-1,3-N-acetylglucosaminyltransferase that glycosylates MacCer (Wandall et al., 2005), as well as egghead (egh), encoding the beta-4-mannosyltransferase that glycosylates GlcCer to MacCer (Wandall et al., 2005). Using two-electrode voltage-clamp (TEVC) recording at the NMJ, we found that Npc1a and brn mutants have elevated glutamatergic transmission, but observed no change in egh mutants. Neuronal RNA interference (RNAi) knockdown of Npc1a and brn demonstrated a presynaptic requirement. In Npc1a; brn double mutants, we found no further increase in neurotransmission, indicating that the two genes operate within the same pathway. We further found that Npc1a; egh double mutants had elevated transmission comparable to that in Npc1a nulls alone, suggesting that MacCer is not responsible for synaptic dysfunction. Consistently, the GlcCer synthase inhibitor miglustat blocked elevated neurotransmission in Npc1a and brn mutants. MacCer labeling showed no synaptic accumulation in Npc1a nulls, but maintained elevation in Npc1a; brn double mutants. Moreover, Npc1a mutants had increased GlcCer levels, suggesting that GlcCer is likely to be pathogenic. Assaying neurodegeneration with terminal deoxynucleotidyl transferase biotin-dUTP nick end labeling (TUNEL) labeling (Vita et al., 2021) revealed that Npc1a nulls had heightened neuronal death, with synaptic dysfunction preceding neurodegeneration in motor neurons. Consistently, neuronal death was strongly elevated in egh and brn mutants, and similarly increased in Npc1a; brn double mutants. These findings causally link GSL accumulation to elevated neurotransmission and neurodegeneration in the Drosophila NPC disease model.

To test synaptic transmission strength in the Drosophila model of NPC disease, TEVC electrophysiology was used to assay glutamatergic signaling at the well-characterized NMJ (Bhimreddy et al., 2021). Null Npc1a mutants were fed 7-dehydrocholesterol (0.14 mg/g⁻¹) as an ecdysone precursor to circumvent the early larval lethality resulting from the lack of steroid hormone signaling (Huang et al., 2005; Phillips et al., 2008). Alternatively, the 2-286 Gal4 driver was used to express UAS-Npc1a in the endocrine ring gland, the site of ecdysone steroid synthesis (Phillips et al., 2008). The approaches produced identical results. Homozygous Npc1a^57a and trans-heterozygous Npc1a^57a/Npc1a¹ nulls (Fluegel et al., 2006; Huang et al., 2005) were compared to genetic background control (w¹¹¹⁸). The Npc1a^57a; 2-286 Gal4>UAS-Npc1a::YFP line was compared to the 2-286 Gal4 driver alone control (2-286 Gal4/+). Electrical stimulation of the wandering third-instar motor nerve with a glass suction electrode in 1.5 mM [Ca²⁺] was used to drive NMJ glutamatergic transmission onto ventral longitudinal muscle 6 (abdominal segments 3/4), voltage clamped at −60 mV. To assay synaptic transmission strength, ten consecutive excitatory junction current (EJC) traces were collected (0.2 Hz) from each NMJ and then averaged for each single data point (Bhimreddy et al., 2021; Kopke et al., 2017). Representative synaptic electrophysiology recordings and quantified results are shown in Fig. 1.

Repeated nerve stimulations resulted in strong, consistent EJC responses with high-fidelity amplitudes (Fig. 1A). The Drosophila NPC disease model of Npc1a null mutants showed strikingly increased glutamatergic currents compared to matched genetic controls. Synaptic transmission was similarly elevated in both Npc1a^57a and Npc1a^57a/Npc1a¹ nulls compared with the w¹¹¹⁸ genetic background controls (Fig. 1A). The normalized EJC amplitude for the control is 1.00±0.07 (mean±s.e.m.; n=18; Fig. 1B) compared to the homozygous Npc1a^57a mutant elevated response of 1.35±0.12 (n=17; Fig. 1B). Likewise, trans-heterozygous Npc1a^57a/Npc1a¹ mutants showed the same elevation, with an average EJC amplitude of 1.32±0.09 (n=19; Fig. 1A,B). EJC amplitudes of homozygous Npc1a^57a and heterozygous Npc1a^57a/Npc1a¹ were not significantly different (P=0.980), but both were significantly increased from that of the control (P=0.031, P=0.041, respectively; Fig. 1B). Ring gland-rescued Npc1a nulls (Npc1a^57a; 2-286 Gal4>UAS-Npc1a::YFP) closely phenocopied the 7-dehydrocholesterol-fed EJC amplitude elevation (Fig. S1A), with a normalized EJC amplitude of 1.26±0.13 (n=18) compared to that of the 2-286 Gal4/+ control of 1.00±0.059 (n=13, P=0.046; Fig. S1B), indicating that cholesterol supplementation does not impact the increased neurotransmission. Taken together, these findings demonstrate that the Drosophila NPC disease model shows elevated glutamatergic synaptic transmission independent of cholesterol feeding. We therefore next tested whether this functional neurotransmission defect could be explained by a GSL mechanism.

The NPC lipid storage disease state involves GSL mistrafficking and accumulation (te Vruchte et al., 2004; Wheeler et al., 2019; Zervas et al., 2001a,b), and GSL pathway mutants provide a way of testing the neurotransmission requirements (Huang et al., 2018; Newton et al., 2020). There are well-established links between GSL accumulation and elevated glutamatergic signaling in NPC models (D'Archangelo et al., 2016). We therefore hypothesized that GSL pathway mutants would phenocopy our NPC disease model synaptic defects. Previous studies have shown that brn mutants (brn^fs.107) of beta-1,3-N-acetylglucosaminyltransferase that glycosylates MacCer (Wandall et al., 2005) accumulate MacCer at the Drosophila NMJ, which is linked to NMJ overgrowth (Huang et al., 2018). To test parallels in the NPC model, NMJs were co-labeled with anti-horseradish peroxidase (HRP), which recognizes fucosylated N-glycans on the presynaptic neuronal membrane (Jan and Jan, 1982; Kopke et al., 2017), and anti-Discs Large (DLG; also known as Dlg1), which labels the postsynaptic scaffold in the subsynaptic reticulum (Lahey et al., 1994; Kopke et al., 2017). Like brn mutants, Npc1a mutants showed NMJ overgrowth (Fig. S2A). Compared to controls, the mutants had significantly increased NMJ area (P<0.0001; Fig. S2B) and synaptic bouton number (P<0.0001; Fig. S2C). However, there was no change in satellite bouton number (Fig. S2D) and only a small increase in synaptic branch number (P=0.0159; Fig. S2E). There was no change in the presynaptic marker Bruchpilot (Brp) marking synapses (Kittel et al., 2006), suggesting that the NMJ overgrowth is not coupled to neurotransmission function.

To test the connection between GSL pathway and NPC model neurotransmission, we took TEVC recordings of brn mutants as well as egh mutants of the beta4-mannosyltransferase enzyme, which glycosylates GlcCer to MacCer (Fig. 2A). Previous studies have shown that egh⁷ mutants accumulate GlcCer (Wandall et al., 2005). In parallel, we used the GlcCer synthase inhibitor miglustat (D'Archangelo et al., 2016; Patterson et al., 2020) to block ceramide glycosylation (Fig. 2A). Compared to control (w¹¹¹⁸) normalized EJC amplitude (1.00±0.08, n=9), brn^fs.107 mutants had increased synaptic transmission (1.45±0.11, n=9), which was significantly elevated (P=0.0045; Fig. 2B,C) and phenocopied that of Npc1a mutants. In contrast, egh⁷ mutants showed no increase in EJC amplitude compared to that of controls (Fig. 2D,E). We therefore next tested whether the GlcCer synthase inhibitor miglustat affects the Npc1a phenotype. Feeding mutant animals miglustat prevented the elevated synaptic transmission (Fig. 2F). Compared to miglustat-fed control EJC amplitudes of 1.00±0.06 (n=9), miglustat-fed Npc1a nulls had an EJC amplitude of 1.14±0.10 (n=9), which was not significantly different from that of controls (P=0.247; Fig. 2G). Likewise, the miglustat-fed brn^fs.107 mutant EJC amplitude of 1.07±0.09 (n=10) was not significantly different from that of controls (P=0.3494; Fig. S3A,B). Importantly, these results establish that GSL accumulation causes a strong increase in NMJ neurotransmission, and that the GSL synthesis pathway is causative of the increase in NPC glutamatergic synaptic function. We next tested whether Npc1a and brn function within the same cell compartment to increase synaptic transmission.

Previous studies of NPC model synapses revealed presynaptic and postsynaptic defects that can impact neurotransmission strength (Hawes et al., 2010; Mitroi et al., 2019). We therefore next tested the cell type requirements for the increase in synaptic transmission in the Npc1a and brn mutants. We used a targeted Gal4 driver specific to neurons (elav-Gal4) to drive both Npc1a and brn RNAi knockdown, together with TEVC recording at the NMJ to measure changes in glutamatergic neurotransmission. Compared to the transgenic driver alone control (elav-Gal4/w¹¹¹⁸), targeted neuronal knockdown of both Npc1a and brn increased synaptic function (Fig. 3A). Neuron-specific Npc1a RNAi elevated the EJC amplitude to 1.211±0.07 (n=9), normalized to the control 1.00±0.05 (n=14), which was a significant increase (P=0.036; Fig. 3B). Likewise, elav-Gal4-driven brn RNAi elevated synaptic transmission to 1.323±0.09 (n=8), a significant increase compared to that of control (P=0.0043), phenocopying the NPC model (Fig. 3B). These results establish that neuronal knockdown of either Npc1a or brn is sufficient to induce the increase in neurotransmission, indicating that changes in the presynaptic neuron are responsible for the strengthened synaptic transmission. The finding that both Npc1a and brn similarly impact presynaptic neurotransmission supports the hypothesis that Npc1a and brn act in the same pathway regulating synaptic function. We therefore next tested for genetic interactions between the GSL synthesis pathway and the NPC disease model.

Double mutant genetic combinations can be used to test functional interactions in a common versus a parallel mechanism (Haber et al., 2013; Slade and Staveley, 2015). For this test, we first created Npc1a; brn double mutants to test synaptic transmission phenotypes compared to each single mutant and the w¹¹¹⁸ genetic background control. We hypothesized that if the Npc1a; brn double mutants were to display the same elevated neurotransmission as each single mutant alone, then t"he two gene products must act in a common pathway regulating synaptic signaling strength. To further test GSLs upstream of brn and downstream of miglustat inhibition (MacCer and GlcCer) (Huang et al., 2018; Wandall et al., 2005), we next created Npc1a; egh double mutants to determine whether egh loss of function modulates the synaptic transmission elevation in our NPC disease model. We hypothesized that if the Npc1a; egh double mutants exhibit a restored synaptic transmission amplitude, similar to the above miglustat result (Fig. 2F,G), then GSLs downstream of GlcCer would be causative of the increased synaptic transmission in the Npc1a null mutants. In all combinations, we used the same rearing methods as above, including feeding 7-dehydrocholesterol to homozygous Npc1a^57a (Phillips et al., 2008), brn^fs.107 and egh⁷ mutants (Wandall et al., 2005), and the double mutants. As above, we performed TEVC electrophysiological recordings at the third-instar NMJs. Representative traces and quantified data are shown in Fig. 4.

We first tested Npc1a^57a; brn^fs.107 double mutants, to find the same highly elevated synaptic transmission as in each single homozygous Npc1a^57a and brn^fs.107 mutants alone (Fig. 4A). Normalized to the w¹¹¹⁸ genetic background control, the brn^fs.107 mutants showed a mean EJC amplitude of 1.25±0.07 (n=22), which was significantly increased compared to that of matched controls (P=0.027; Fig. 4A,B). Likewise, Npc1a^57a mutants had a normalized EJC amplitude of 1.60±0.08 (n=14), which was also significantly elevated compared to that of w¹¹¹⁸ controls (P<0.0001; Fig. 4A,B). Importantly, Npc1a^57a; brn^fs.107 double mutants showed an intermediate mean EJC amplitude of 1.36±0.12 (n=11), with a similar significant increase compared to that of the controls (P=0.0098; Fig. 4A,B). Moreover, the mean EJC amplitude of the Npc1a^57a; brn^fs.107 double mutants was not significantly increased from that of the Npc1a^57a (P=0.235) or the brn^fs.107 (P=0.824) single mutant. These results indicate that brn and Npc1a function in the same pathway to increase synaptic transmission, as opposed to different pathways that would cause an additive increase in the double mutant EJC amplitude (Fig. 4A,B). Miglustat feeding, which inhibits GlcCer synthase, prevented the increase in synaptic transmission in our NPC disease model (Fig. 2E,F), and the Npc1a^57a; brn^fs.107 double mutants, which have blocked GSL synthesis past MacCer, showed no change in the increased synaptic transmission (Fig. 4A,B); these results together suggest that MacCer or GlcCer may be causative of Npc1a-increased synaptic transmission, and that egh⁷ mutants that inhibit GlcCer to MacCer synthesis may be able to rectify the Npc1a mutant increase in neurotransmission.

We therefore next tested Npc1a^57a; egh⁷ double mutants, to find the same highly elevated neurotransmission as in the single homozygous Npc1a^57a mutant alone (Fig. 4C). Both Npc1a^57a single mutants and Npc1a^57a; egh⁷ double mutants showed indistinguishably increased synaptic transmission compared with the w¹¹¹⁸ genetic background control. Normalized to control (1.00±0.07; n=10), the EJC amplitude of the Npc1a^57a single mutant was 1.408±0.09 (n=9) and that of the Npc1a^57a; egh⁷ double mutant was 1.42±0.12 (n=9), and both were significantly increased from that of w¹¹¹⁸ (P=0.0188 and P=0.0140, respectively; Fig. 4D). Importantly, there was no significant difference between the Npc1a^57a single mutant and Npc1a^57a; egh⁷ double mutant (P=0.9994; Fig. 4D). Because the egh⁷ mutants alone exhibited no change in synaptic transmission strength (1.046±0.10; n=8; P=0.9859; Fig. 4D), and the double mutants showed no change from the single Npc1a^57a mutant (Fig. 4C,D), we can make no firm conclusion regarding the causative influence of GlcCer on the neurotransmission elevation in our NPC model. Therefore, GlcCer and/or MacCer may be the causative, pathogenic GSL underlying the synaptic transmission phenotype, but electrophysiological recording alone has so far been unable to distinguish these possibilities. To address this critical question, we next used a characterized anti-MacCer antibody (Wandall et al., 2005) to visualize and quantify synaptic MacCer levels at the NMJ to determine whether MacCer levels correlate with the synaptic transmission levels in our panel of mutants.

Previous studies have reported the accumulation of LacCer, the mammalian homolog of Drosophila MacCer, in mouse NPC model brains as well as in NPC patient fibroblasts (Colaco et al., 2019; te Vruchte et al., 2004). In light of the importance of MacCer at the Drosophila NMJ synaptic terminal (Huang et al., 2018), we hypothesized that MacCer accumulation would be responsible for the elevated synaptic transmission in our Drosophila NPC model. In order to test for MacCer accumulation defects at the NMJ, we used a characterized anti-MacCer antibody for confocal imaging (Wandall et al., 2005). We tested synaptic MacCer levels in the genetic background controls (w¹¹¹⁸), Npc1a nulls (Npc1a^57a), GSL pathway mutants (brn^fs.107, egh⁷) and Npc1a^57a; brn^fs.107 double mutants. Previous work has shown an increase in MacCer in brn^fs.107/brn^1.6P6 mutants and decrease in MacCer in egh^62d18 mutants at the Drosophila NMJ (Huang et al., 2018). We co-labeled NMJs with anti-MacCer and anti-HRP antibody, which labels the presynaptic neuronal membrane (Jan and Jan, 1982; Kopke et al., 2017). We then measured anti-MacCer fluorescent intensity at HRP-positive NMJ synaptic boutons on muscle 4 (abdominal segments 3,4) (Kopke et al., 2017; Roos et al., 2000). Measurements were made in and around the large type 1b synaptic boutons (Knodel et al., 2014), including presynaptic neuronal and postsynaptic muscle membrane compartments. Representative images and quantifications are shown in Fig. 5.

In w¹¹¹⁸ genetic control NMJs, MacCer localized in punctate accumulations in both presynaptic and postsynaptic regions (Fig. 5A, top row). Similar low-level MacCer labeling was present in both Npc1a^57a and egh⁷ mutants, which were indistinguishable from controls (Fig. 5A, middle row). Normalized to w¹¹¹⁸ controls (1.00±0.06, n=22), Npc1a^57a nulls showed a mean fluorescent intensity of 0.93±0.07 (n=18), which is not different from that of controls (P=0.944; Fig. 5B). Given that MacCer levels appear identical in Npc1a^57a mutants and matched controls, synaptic MacCer is not a good candidate to explain the elevated synaptic transmission in our NPC model. Moreover, we found no difference in MacCer levels at egh⁷ mutant NMJs (1.04±0.11; n=14; P=0.9915; Fig. 5B), which differs from the previously reported decrease in egh^62d18 and egh⁷ mutants in whole-larva thin-layer chromatography analyses (Huang et al., 2018; Wandall et al., 2005). We note that the egh^62d18 allele is a frameshift that deletes most of the egh coding sequence, compared to the egh⁷ single point mutation (Wandall et al., 2005), so egh⁷ may not be strong enough to show the same phenotype. The egh^62d18 mutant appears to be lost based on our extensive efforts to locate this line. Low levels of synaptic MacCer in genetic background controls (Fig. 5A) may make it difficult to quantify the MacCer loss present in egh⁷ whole-larva analyses (Wandall et al., 2005). We expect that egh⁷ mutants do exhibit decreased MacCer levels, consistent with the reported loss of enzymatic function.

brn^fs.107 mutants have previously been shown to exhibit strongly elevated MacCer at the Drosophila NMJ (Huang et al., 2018). Similarly, Npc1a^57a; brn^fs.107 double mutants showed a very striking increase in synaptic MacCer accumulation compared to genetic controls (Fig. 5A, bottom row). The synaptic MacCer elevation in brn^fs.107 mutants and Npc1a^57a; brn^fs.107 double mutants was apparent in larger and far more numerous punctate accumulations in both the presynaptic and postsynaptic regions, but the higher MacCer levels were not restricted to the NMJ. In both genotypes, MacCer was broadly elevated in the muscle extending outside the synaptic domain (Fig. 5A, bottom row; Fig. S4A). We confirmed that brn^fs.107 mutants had an increase in MacCer fluorescence intensity of 1.60±0.28 (n=9) normalized to the w¹¹¹⁸ genetic background control fluorescent level of 1.00±0.07 (n=10), a significant elevation in the mutant (P=0.0465; Fig. S4B). Likewise, compared to the w¹¹¹⁸ control fluorescent level of 1.00±0.06 (n=22), the Npc1a^57a; brn^fs.107 double mutants exhibited a mean synaptic MacCer intensity of 1.44±0.15 (n=18), which is significantly increased compared to that of control (P=0.0079; Fig. 5B). Given that the Npc1a^57a; brn^fs.107 double mutants showed synaptic transmission strength comparable to that of Npc1a^57a mutants (Fig. 4A,B), but the double mutants displayed a striking MacCer accumulation whereas the Npc1a^57a mutants did not (Fig. 5A,B), we conclude that MacCer is not a good candidate to explain the elevated synaptic transmission in our NPC model.

We therefore turned to GlcCer as the best candidate pathogenic lipid for the NPC model phenotypes. However, we lack GlcCer imaging capabilities at the Drosophila NMJ. As a consequence, we performed liquid chromatography–mass spectrometry (LC-MS) to assay the overall GlcCer levels (see Materials and Methods). Levels of HexCer_NS(35;3;2O) were significantly increased in the NPC model (Npc1a^57a nulls), with a measured magnitude of 23,146±1320 (n=5 trials) compared to the w¹¹¹⁸ genetic background control with a normalized level of 14,812±1174 (n=5), which is a significant elevation (P=0.00157; Fig. S5A). Similarly, the levels of HexCer_NS(32;3;2O) were significantly increased in the Npc1a^57a mutants, with a magnitude of 61,054±5541 (n=5) compared to w¹¹¹⁸ controls with a normalized level of 39,561±6460 (n=5, P=0.0407; Fig. S5B). Because Drosophila do not contain any galactosylceramide (Wandall et al., 2005), we can infer that these HexCer species must correspond to GlcCer variants. The observation that GlcCer is accumulated in the Npc1a null Drosophila NPC disease model, but MacCer is not, suggests that GlcCer is most likely to be the pathogenic lipid in this system underlying the above synaptic phenotypes. We next tested whether heightened neurotransmission excitotoxicity provides a mechanistic explanation for the subsequent neurodegeneration in the Drosophila NPC disease model. To test for GSL roles, we assayed neurodegeneration in the Npc1a and GSL pathway mutants.

To test whether GSL-mediated neurotransmission excitotoxicity is linked to neurodegeneration in our NPC disease model, we next used TUNEL labeling to analyze larval neuronal cell death (Vasudevan and Ryoo, 2016; Vita et al., 2021). The wandering third-instar central nervous system components [brain cerebral lobes and ventral nerve cord (VNC)] were co-labeled for the synaptic marker Brp (Kittel et al., 2006) to delineate central neuropils. Characterized cell death occurs normally in the wandering third instar, especially in the brain cerebral lobes (Pinto-Teixeira et al., 2016; Truman and Bate, 1988). We first tested neurodegeneration overall in our NPC disease model by analyzing the TUNEL-positive area in cerebral lobes and VNC. We next tested neurodegeneration specifically in the RP3 motor neurons innervating muscle 6 from the above electrophysiology studies (Broadie et al., 1993), hypothesizing that neurodegeneration due to synaptic hyperexcitability induced excitotoxicity (Peng et al., 2019). Finally, we tested the GSL pathway mutants for neurodegeneration, including both egh⁷ and brn^fs.107 single mutants, and Npc1a^57a; brn^fs.107 double mutants compared to the genetic background control (w¹¹¹⁸). In all four genotypes, we co-labeled for TUNEL and Brp, analyzing TUNEL-positive area in brain cerebral lobes and VNC. Representative images and data quantification are shown in Figs 6 and 7.

Control w¹¹¹⁸ animals exhibit considerable normal TUNEL-labeled neuronal death in the brain cerebral lobes, but very limited neurodegeneration in the VNC (Fig. 6A, top row). In the brain, widespread TUNEL labeling occurs, consistent with neuroblast lineage apoptosis whereby approximately half of all neurons die (Kumar et al., 2009; Pinto-Teixeira et al., 2016). In the VNC, little TUNEL labeling is present around the motor neuron cell bodies along the central midline between the two lateral neuropils (Fig. 6A, top row). In contrast, Npc1a null mutants exhibit clearly increased TUNEL labeling in the brain and particularly in the VNC (Fig. 6A, bottom row). The percentage TUNEL-positive area was quantified for cerebral lobes and the central VNC between the lateral neuropils defined by Brp immunofluorescence (Fig. 6A). Normalized to the w¹¹¹⁸ genetic controls (1.00±0.096; n=37), Npc1a nulls showed 50% higher TUNEL area in the brain cerebral lobes (1.509±0.143; n=34), which is a significant elevation (P=0.0039; Mann–Whitney; Fig. 6B). The elevated cerebral lobe cell death in the NPC model did not appear restricted to any particular brain regions. Normalized to w¹¹¹⁸ genetic controls (1.00±0.158; n=9), Npc1a nulls (Npc1a^57a/Npc1a¹) showed an 8-fold increase in TUNEL area in the central VNC (8.143±2.479; n=7), which is also a significant elevation (n=9; P=0.0018; Mann–Whitney; Fig. 6C). This central VNC neuronal death between the two lateral Brp-positive neuropils (Fig. 6A, bottom right) excludes neuroblast domains and thus represents neurodegeneration of fully differentiated circuitry.

Given the pronounced TUNEL-labeled cell death along the VNC midline in Npc1a mutants, we hypothesized the occurrence of motor neuron death, in line with the elevated NMJ transmission in our NPC model (Fig. 1). To test this hypothesis, the RP3 motor neurons innervating muscle 6 were specifically labeled with CCAP-Gal4 driven UAS-eGFP (green; Fig. S6A,B). We co-labeled for TUNEL (magenta) and Brp to mark the neuropil (blue) in Npc1a^57a null mutants compared to controls (Fig. 6D). Control animals (CCAP-Gal4>UAS-eGFP) again showed essentially no TUNEL-labeled neuronal cell death along the central VNC midline and no indication of RP3 motor neuron death (Fig. 6D, left). In contrast, the NPC disease model null mutants (Npc1a^57a; CCAP-Gal4>UAS-eGFP) showed extensive TUNEL labeling along the VNC midline, albeit with very limited overlap with RP3 motor neurons (Fig. 6D, right). The colocalization between RP3 motor neuron GFP and TUNEL signal was assayed with overlap analyses (Zen program). This quantification indicated a low level of TUNEL-positive RP3 motor neurons in control animals (colocalization ratio, 0.019±0.014; n=11), with a trending increase in the NPC disease model (Npc1a^57a, 0.074±0.039; n=10), but no significant change in RP3 motor neuron TUNEL label (P=0.0728; Mann–Whitney). This trend suggests that RP3 motor neurons are in the earliest stages of neurodegeneration, but the majority of NPC model TUNEL-labeled cell death along the VNC midline clearly did not occur within these motor neurons.

We next tested GSL pathway mutants. Both egh⁷ and brn^fs.107 mutants showed increased TUNEL labeling in cerebral lobes (Fig. 7A) and VNC (Fig. 7B). The Npc1a^57a; brn^fs.107 double mutants showed no further increase in TUNEL labeling (Fig. 7A,B). Quantification in the cerebral lobes normalized to w¹¹¹⁸ controls (1.00±0.0963; n=37) revealed greater TUNEL-positive area in brn^fs.107 mutants (2.135±0.206; n=37; P=0.0029), Npc1a^57a; brn^fs.107 double mutants (2.768±0.4511; n=29; P=0.0013) and especially egh⁷ mutants (4.417±0.439; n=34; P<0.0001; Fig. 7C). Importantly, TUNEL labeling in brn^fs.107 mutants and Npc1a^57a; brn^fs.107 double mutants was not significantly different, suggesting a common pathway. Quantification in the central VNC normalized to controls (1.00±0.1297; n=13) likewise revealed elevated TUNEL in brn^fs.107 (6.769±2.012; n=13; P=0.035) and egh⁷ (4.490±1.165; n=15; P=0.036; Fig. 7D) mutants. Quantification normalized to controls (1.00±0.1297; n=13) revealed similarly elevated TUNEL area in Npc1a^57a; brn^fs.107 mutants (4.854±1.816; n=13; P=0.0449; Fig. 7E). Feeding miglustat to Npc1a^57a, brn^fs.107 and egh⁷ mutants did not alter TUNEL labeling in cerebral lobes or VNC (Fig. S7A,B). Quantification of the miglustat-fed mutants showed maintained higher levels of neurodegeneration (Fig. S7C,D). These results indicate that manipulating GSL levels causes neurodegeneration, phenocopying the cell death exhibited in the NPC model (Npc1a nulls). Overall, our findings link GSLs to the maintenance of normal levels of neurotransmission and neurodegeneration.

This is the first study to test a link between GSL mechanisms and neurotransmission and neurodegeneration defects in the Drosophila NPC disease model. Our previous work in this model demonstrated age-dependent cholesterol accumulation and neurodegeneration in the adult brain, and shortened adult lifespan (Phillips et al., 2008). Others have reported cholesterol accumulation in Npc1a null first instars (Huang et al., 2005) and even earlier defects in Hedgehog signaling in mutant embryos (Bialistoky et al., 2019). The current study fills an important gap in third instars, demonstrating elevated neurotransmission and neurodegeneration in Npc1a nulls. It has been proposed that ganglioside accumulation is a primary driver for NPC phenotypes (Zervas et al., 2001a,b), which is problematic because Drosophila lacks complex gangliosides. However, GSLs may be pathogenic (Wheeler et al., 2019; Zervas et al., 2001a,b), with GSL accumulation prior to cholesterol buildup (Lloyd-Evans et al., 2008), and Drosophila is a good model to test GSL pathway mechanisms. For synaptic transmission, a GSL link is shown by the brn N-acetylglucosaminyltransferase mutant (Wandall et al., 2005) phenocopy of the Npc1a defect, miglustat glucosylceramide synthase inhibitor (Patterson et al., 2007) prevention of Npc1a and brn defects, and the lack of additive defects in Npc1a; brn double mutants. A focus on GlcCer derives from the lack of defect in egh mannosyltransferase mutants (Wandall et al., 2005), lack of MacCer changes in Npc1a mutants (Huang et al., 2018) and GlcCer accumulation. For neurodegeneration, Npc1a, egh, brn and Npc1a; brn double mutants all displayed heightened TUNEL-labeled cell death in the nervous system, with synaptic dysfunction preceding neuronal death, suggesting likely GSL mechanisms in this NPC disease model.

We first found that Npc1a mutants have elevated glutamatergic synaptic transmission. In the mouse NPC model, a similar increase occurs in the Schaffer collateral/commissural pathway composed of glutamatergic pyramidal neurons projecting into the hippocampal CA1 region (D'Arcangelo et al., 2016). Drosophila NMJ electrophysiology allows muscle intracellular recording in individually identified cells (e.g. RP3 motor neuron to muscle 6) (Broadie et al., 1993), in contrast to the mouse hippocampus extracellular recordings for regional changes (Bykhovskaia and Vasin, 2017; D'Arcangelo et al., 2016). Moreover, directly measuring synaptic defects in the motor circuit may be critical to understanding NPC progression, because motor ataxia is a primary NPC symptom (Patterson et al., 2020). Glutamatergic hyperexcitability may also underlie seizures (Wu et al., 2022), another common NPC symptom (Saini et al., 2019). Additionally, hyperexcitability has been linked to intellectual disability in other neurological disorders (Liu et al., 2022) and could account for the intellectual disability in NPC patients (Patterson et al., 2020). The Npc1a NMJ synaptic transmission defect builds on our previous work showing defects in extracellular photoreceptor electroretinogram (ERG) recordings (Phillips et al., 2008). ERG traces show a decrease in synaptic amplitude with advanced neurodegeneration, whereas NMJ recordings show neurotransmission elevation prior to neurodegeneration. In addition to these functional changes, Npc1a mutants display overgrown synaptic architecture with increased NMJ area, synaptic bouton number and terminal branch number. Although NMJ structure can sometimes account for functional defects (Budnik et al., 1990; Menon et al., 2013), Npc1a mutants do not have increased presynaptic active zone number, suggesting that NMJ growth is independent of the strengthened neurotransmission.

We next found that the brn mutant phenocopies the elevated synaptic transmission observed in the Npc1a mutant. These brn mutants accumulate truncated GSLs and lack elongated GSLs (Wandall et al., 2005) and also exhibit defective Drosophila NMJ trans-synaptic signaling (Huang et al., 2018). Consistently, brn mutant punctate MacCer accumulations localize with both presynaptic (HRP) and postsynaptic (DLG) membrane markers at NMJ synaptic boutons, as well as widely in the surrounding extrasynaptic muscle (Huang et al., 2018). GSL changes in many neurodegenerative disorders involve synaptic transmission defects (Garcia-Ruiz et al., 2015). For example, mutants in serine palmitoyl transferase, which converts L-serine and palmitoyl-CoA to 3-oxosphinganine, show striking GSL changes in juvenile amyotrophic lateral sclerosis, a neurodegenerative disease characterized first by synaptic hyperexcitability and then later motor circuit degeneration (Dodge et al., 2015; Johnson et al., 2021; Lone et al., 2022; Menon et al., 2020). This progression appears to be consistent with the brn mutant increase in synaptic transmission. In contrast, we found that egh mutants exhibit no change in neurotransmission strength. Likewise, these same egh mutants exhibit no change in elongated GSLs, showing a slight increase in GlcCer accumulation but no change in MacCer abundance (Wandall et al., 2005). egh^62d18 mutants have aberrant NMJ synaptic architecture, but function was not assayed (Huang et al., 2018), and this allele is sadly now lost. The brn-specific function controlling NMJ neurotransmission strength is consistent with known synaptic GSL roles regulating both endocytosis (Sharma et al., 2005) and receptor localization mechanisms in lipid rafts (Roh et al., 2014).

Miglustat prevented the elevated glutamatergic neurotransmission in Npc1a mutants. Acting as a GlcCer synthase inhibitor (Patterson et al., 2007; Platt et al., 1994), miglustat is the only approved NPC treatment (European Union only), and miglustat is the only drug treatment shown to stabilize the cognitive decline of NPC patients (Patterson et al., 2020). As in the Drosophila NPC model, miglustat feeding prevents the synaptic transmission elevation in the mouse NPC model in the glutamatergic Schaffer collateral/commissural pathway (D'Arcangelo et al., 2016), indicating a conserved GSL mechanism limiting glutamatergic neurotransmission in these two disease models. Moreover, miglustat feeding also prevented brn synaptic transmission elevation, further establishing the pathogenicity of intermediate GSLs and providing evidence for a linked pathway between Npc1a and the GSL biosynthetic pathway. Previous Drosophila studies used the 1-phenyl-2-decanoylamino-3-morpholino-1-propanol inhibitor of GlcCer synthase to show NMJ synaptic architecture changes, with fewer and larger synaptic boutons, but no functional analysis has been reported (Huang et al., 2018). Because miglustat may well have off-target effects, for example inhibiting non-lysosomal glucosyl ceramidase (GBA2), which removes the glucose moiety from GlcCer to convert it back to ceramide in the plasma membrane inner leaflet (Wheeler et al., 2019), it could be useful to test whether PDMP also restores Npc1a mutant synaptic transmission. This could help confirm that miglustat prevention of the Npc1a neurotransmission elevation is due to GlcCer synthase and not to GBA2 inhibition (Wheeler et al., 2019). Regardless, miglustat correction of the Npc1a neurotransmission increase indicates the importance of the GSL synthesis pathway in limiting synaptic function in our NPC disease model.

We next found that neuronally driven Npc1a and brn RNAi knockdown induces the same neurotransmission elevation, suggesting a presynaptic mechanism. Mouse NPC disease models likewise show presynaptic NPC requirements; however, the phenotypes are consistent with impaired neurotransmission, which may occur in a more advanced state of neurodegeneration. Previous studies report a decreased total synaptic vesicle pool, impaired vesicle release and larger vesicles in primary hippocampal neurons in the NPC mouse model (Mitroi et al., 2019; Hawes et al., 2010; Karten et al., 2006). Importantly, the same study that documented impaired synaptic vesicle release confirmed that the glutamatergic Shaffer collaterals to CA1 hippocampal neurons have an excitatory neurotransmission phenotype (Mitroi et al., 2019), so it is possible that the increase in neurotransmission is maintained by the increase in synaptic vesicle size (Mitroi et al., 2019; Karten et al., 2006), or possibly by increases in presynaptic evoked calcium (Lloyd-Evans et al., 2008). It is also important to note that several postsynaptic glutamatergic receptor defects have been reported in the NPC mouse and rat models. Previous studies have suggested impairment in kainate receptors and an imbalance in AMPA receptors (D'Arcangelo et al., 2011), as well as impaired delivery of AMPA receptors during LTP in the hippocampus in the NPC mouse model (Mitroi et al., 2019). More research is needed on presynaptic and postsynaptic NPC1 roles, especially at early time points in disease model progression. Importantly, neuron-targeted brn RNAi phenocopies the Npc1a neurotransmission increase, showing that the GSL requirement is also presynaptic and supporting the conclusion that Npc1a and brn function within the same pathway to increase synaptic transmission.

We next found that Npc1a; brn double mutants have elevated synaptic transmission of the same magnitude as the individual brn and Npc1a single mutants, further indicating that brn and Npc1a likely function within the same pathway to increase presynaptic neurotransmission. Such double mutant genetic assays are a powerful approach to test whether two genes operate in the same pathway (no additional change) or in different, parallel pathways (additive phenotypes) (Haber et al., 2013; Martínez-Laborda et al., 1996). Importantly, our results indicate that brn is not sufficient to alter the elevated synaptic transmission in Npc1a nulls, and thus that accumulation of either MacCer or GlcCer may be responsible for the Npc1a neurotransmission phenotype. Similarly, although miglustat restoration of neurotransmission in CA1 hippocampal neurons of the mouse NPC model has also shown that the GSL synthesis pathway is important in neurotransmission (D'Arcangelo et al., 2016), the synaptic defect has not been linked to accumulation of a specific GSL species. In NPC clinical treatments, miglustat also acts via an uncharacterized GSL correction (Platt et al., 1994; Zervas et al., 2001a,b). In order to determine whether MacCer or GlcCer causes synaptic dysfunction in our NPC model, we next tested the Npc1a; egh double mutant, surprisingly finding no change in the double mutants compared to the Npc1a single mutants. Given that the loss of egh function is insufficient to restore Npc1a synaptic transmission, we cannot make a strong conclusion about the role of GlcCer accumulation in synaptic transmission. We therefore turned to the analysis of synaptic MacCer, which we can image directly, to determine whether MacCer is accumulated as the pathogenic GSL driving the elevated synaptic transmission.

The Drosophila NPC disease model shows no increase in synaptic MacCer levels. Our work confirms previous studies that indicate that brn mutants, in contrast, display elevated MacCer at the NMJ (Huang et al., 2018). This is surprising given that Npc1a and brn operate in the same pathway to increase synaptic transmission, and we had hypothesized a parallel increase in synaptic MacCer levels. Human NPC patients and the feline NPC model both show elevated LacCer, a mammalian dihexosylsceramide counterpart to MacCer (Stein et al., 2012; Vanier, 1983; Wandall et al., 2005). However, synaptic MacCer levels do not correlate with elevated synaptic transmission in the Drosophila NPC model. The Npc1a; brn double mutants do show a dramatic synaptic MacCer elevation, in and around the synapse, but this does not explain the elevated neurotransmission, despite likely MacCer incorporation into important synaptic lipid rafts (Yu et al., 2021). We therefore hypothesize that GlcCer levels are responsible for the Npc1a synaptic defect. We cannot visualize synaptic GlcCer, but we did find that GlcCer levels in whole animals are increased based on LC-MS. This GlcCer elevation is conserved across human patients and mouse disease models. NPC patient liver and spleen have a ∼15- to 35-fold GlcCer increase, a greater elevation than for cholesterol, dihexosylceramides and gangliosides (Vanier, 1983). GlcCer levels are also increased in mouse NPC model brain tissue (Fan et al., 2013). GlcCer is implicated in increased calcium response to glutamatergic signaling (Korkotian et al., 1999) and also in heightened agonist-stimulated calcium release occurring via ryanodine receptors (Lloyd-Evans et al., 2003). We therefore suggest that elevated GlcCer levels cause the increased neurotransmission in our NPC disease model.

We next found that Npc1a larvae exhibit elevated neurodegeneration. Heightened glutamate neurotransmission drives excitotoxicity (Dong et al., 2009; Peng et al., 2019), which we hypothesized is the causal link. In human NPC patients and cat/mouse models, the disease state is characterized by Purkinje cerebellar neuron cell death (Cougnoux et al., 2016; Elrick et al., 2009; Vite et al., 2015), which is phenocopied by Drosophila adult neurodegenerative vacuolization (Phillips et al., 2008). In the larval brain, Npc1a mutants show increased cell death in brain cerebral lobes and VNC. To test links between heightened neurotransmission and VNC neurodegeneration, we assayed TUNEL-labeled cell death within the individually identified RP3 motor neurons used in our electrophysiology experiments (Broadie et al., 1993). Although there is a trend towards increased TUNEL immunoreactivity in RP3 neurons, they remain viable with elevated neurotransmission, indicating that synaptic dysfunction precedes neurodegeneration in this circuit. Our previous adult retina work shows that Npc1a mutants have decreased synaptic transmission alongside retinal degeneration (Phillips et al., 2008), with impaired synaptic function likely reflecting advanced neurodegeneration. In the mouse NPC model, no electrophysiological defects are reported prior to Purkinje cell neurodegeneration (Elrick et al., 2009), suggesting an apparent difference between inhibitory and excitatory glutamatergic neurons. Importantly, glial death has been reported in the mouse NPC model (German et al., 2002); however, we found no TUNEL-positive glial cells surrounding RP3 neurons, suggesting that the dying cells must instead be interneurons. Previous work suggests that there are decreases in cerebellar interneurons in the Npc1^−/− mouse model (Cougnoux et al., 2020), so our future work will focus on studying interneuron loss mechanisms.

We finally found that egh and brn mutants have heightened neurodegeneration, just like Npc1a mutants, in both cerebral lobes and central VNC containing the motor circuitry. In Gaucher's disease, GlcCer accumulation has been linked to progressive spatial ataxia (Malekkou et al., 2018). In Parkinson's disease, dopaminergic neurodegeneration is also linked to GlcCer dysregulation (Belarbi et al., 2020). Likewise, the motor neuron disorder amyotrophic lateral sclerosis is associated with GlcCer accumulation (Johnson et al., 2021). Therefore, elevated egh and brn TUNEL-labeled neuronal death may reflect a common cause for numerous neurodegenerative conditions affecting multiple neuronal classes. The Npc1a; brn double mutants also exhibit comparable TUNEL-labeled neuronal death in cerebral lobes and VNC. Because brn removal does not rescue the Drosophila NPC model neurodegeneration, we hypothesize that earlier GSL accumulation (i.e. GlcCer) is causative of neuronal death. However, neurodegeneration in both Npc1a and GSL mutants is not prevented by miglustat feeding. A previous study reported that miglustat prolongs survival of Purkinje neurons in the feline NPC model (Stein et al., 2012). Other work has documented a decrease in Purkinje cell loss in human NPC patients with miglustat treatment (Pineda et al., 2018). These differences could indicate that miglustat is an insufficiently potent GlcT1 inhibitor to modulate cell death phenotypes. Alternatively, miglustat dosage/delivery in Drosophila may not be comparable to that in humans or cats. Because feeding miglustat to GSL mutants also does not affect neuronal death, phenocopying the NPC model, the drug appears ineffective in modulating GSL-based neurodegeneration. Nevertheless, we conclude that GSL misregulation is clearly linked to neurotransmission upregulation and subsequent neurodegeneration in the Drosophila NPC disease model.

All mutant and transgenic lines were obtained from the Bloomington Drosophila Stock Center (BDSC). All stocks were grown on standard corn meal/molasses food at 25°C. The genetic background control was w¹¹¹⁸. Null Npc1a alleles (Npc1a¹ and Npc1a^57a) were used in both homozygous and trans-heterozygous combinations (Fluegel et al., 2006; Huang et al., 2005). Controls and mutants were fed 0.14 mg/g⁻¹ 7-dehydrocholesterol to circumvent the Npc1a null mutant lethality (Huang et al., 2005; Phillips et al., 2008). The viable egh⁷ mutant is a point mutation at amino acid 308 converting methionine to lysine (M308K) (Wandall et al., 2005). The viable brn^fs.107 mutant is an intragenic deficiency (Goode et al., 1992). All the double mutants were made using standard genetic crosses. The Gal4 transgenic lines were ring gland driver 2-286-Gal4 (Phillips et al., 2008), pan-neuronal driver elav-Gal4 (BDSC, 8760) and RP3 motor neuron driver CCAP-Gal4 (BDSC, 39292). The UAS responder lines were UAS-npc::YFP (BDSC, 41762), UAS-eGFP (BDSC, 5430), UAS-Npc1a RNAi (BDSC, 37504) and UAS-brn RNAi (BDSC, 55386). Transgenes were introduced into mutants using standard recombination techniques.

Recording was done as previously described (Bhimreddy et al., 2021; Kopke et al., 2017). Briefly, wandering third instars were dissected along the dorsal midline, and the body wall was glued (3M Vetbond Tissue Adhesive) to a sylgard-coated slide. The internal organs were removed to reveal the neuromusculature, and the peripheral nerves were cut at the VNC exit points. Dissections were done at 18°C in standard saline: 128 mM NaCl, 4 mM MgCl₂, 1.0 mM CaCl₂, 70 mM sucrose and 5 mM HEPES (pH 7.2). Dissected preparations were imaged with a Zeiss Axioskop microscope using a 40× water-immersion objective. Ventral longitudinal muscle 6 in abdominal segments 3/4 was impaled by two KCl (3 M)-filled microelectrodes (1 mm outer diameter; World Precision Instruments) with ∼15 M Ω input resistance. The cut motor nerves were sucked into a fire-polished glass suction electrode and stimulated using 0.5 ms suprathreshold voltage stimuli at 0.2 Hz (Grass S88 stimulator) to generate EJC recordings. Nerve stimulation-evoked EJC recordings were filtered at 2 kHz. To quantify EJC amplitude, ten consecutive traces were averaged to produce a mean peak value. Data were acquired with Clampex 9.0 and analyzed with Clampfit 9. Miglustat (MedChem Express) in dimethyl sulfoxide (Thermo Fisher Scientific) was fed at concentrations of 0.1-10 ng/ml (Mitroi et al., 2019).

Antibody labeling was done as previously described (Bhimreddy et al., 2021; Kopke et al., 2017). Briefly, wandering third instars were dissected as above, then fixed with 4% paraformaldehyde (PFA). For mouse anti-Brp [1:200; Developmental Studies Hybridoma Bank (DSHB), nc82], rabbit anti-repo (1:1000; gift from Dr Benjamin Altenhein, University of Cologne, Germany; Sieglitz et al., 2013) and mouse anti-DLG (1:200; DSHB, 4F3), samples were fixed in PFA for 10 min. For mouse anti-MacCer (Wandall et al., 2005) and Cy-3 conjugated goat anti-HRP (1:200; The Jackson Laboratory, 123-605-021), samples were fixed in PFA for 60 min (Huang et al., 2018). Samples were washed in phosphate-buffered saline (PBS) with 0.2% Triton X-100 (PBT) and 0.5% bovine serum albumin (BSA), 3× for 10 min each. Samples were incubated with primary antibodies overnight at 4°C, followed by PBT washes 3× for 10 min each. Secondary antibodies used were as follows: Alexa Fluor 488 goat anti-mouse (1:250; DSHB, A11001), Alexa Fluor 488 goat anti-rabbit (1:1000; DSHB, A11008), Alexa Fluor 488 donkey anti-rabbit (1:250; DSHB, A21206), Alexa Fluor 555 donkey anti-mouse (1:250; DSHB, A31570), and Alexa Fluor 633 goat anti-mouse (1:250; DSHB, A21052). Samples were incubated with secondary antibodies for 2 h at room temperature. Samples were washed in PBT 3× for 10 min each, and then mounted in Fluoromount G (Electron Microscopy Sciences). Confocal imaging was done as previously described (Bhimreddy et al., 2021; Kopke et al., 2017). Briefly, all imaging was done on a Zeiss LSM 510 META laser-scanning confocal microscope, with images projected in Zen (Zeiss) and analyzed using ImageJ (NIH open source). NMJ measurements were made with the HRP signal to delineate z-stack areas of maximum projection using the threshold and wand-tracing tools in ImageJ. All imaging setting were kept constant for comparisons within every set of experiments.

Optima LC-MS reagents including methanol (CH₃OH), acetonitrile (CH₃CN), water (H₂O), formic acid, isopropyl alcohol, chloroform (CHCl₃), ammonium acetate, ammonium bicarbonate, ammonium formate (Thermo Fisher Scientific) and tert-butyl methyl ether (MTBE; Sigma Aldrich) were used for LC-MS analyses. Global untargeted lipidomics was performed in the Center for Innovative Technology at Vanderbilt University. Five samples each of w¹¹¹⁸ and Npc1a^57a genotypes were prepared by homogenizing whole third-instar larvae weighing 30-50 mg, rinsing with 50 mM ammonium formate and flash freezing in liquid nitrogen with storage at −80°C. Samples were thawed on ice with 0.5 ml of cold CH₃OH:CH₃CN:H₂O (1:1:2, v:v:v) with 50 mM ammonium bicarbonate (lysis buffer), and sonicated by probe tip for 10 pulses at 50% power, followed by vortex mixing for 10 s. Normalized volumes based on the initial dry larvae weights were aliquoted and diluted to 200 µl with lysis buffer followed by an addition of 800 µl CH₃OH. Samples were vortexed for 30 s and incubated at −80°C overnight for protein precipitation. Samples were then centrifuged at 4°C and 21,000 g for 15 min, and the supernatant was dried using a cold vacuum centrifuge. Dried samples were reconstituted in 100 µl H₂O followed by 100 µl CH₃OH and 10 µl SPLASH® LIPIDOMIX® Mass Spec Standard (Avanti Polar Lipids), and incubated for 10 min at room temperature. Next, liquid–liquid extraction was performed by adding 800 µl MTBE with 30 s of vortex mixing, followed by 10 min incubation on ice and centrifugation at 4°C and 21,000 g for 15 min. The hydrophobic supernatant (800 µl) was isolated and dried using a cold vacuum centrifuge, and extracts were stored at 4°C. Dried lipid fractions were reconstituted in 100 µl CH₃OH:CHCl₃ (9:1, v:v) with exogenous sphingolipid standards (n=2) to assess instrument variability. Equal volumes (25 µl) of each sample were combined to create a pooled quality control (QC). Samples were analyzed using an Agilent 1290 LC system interfaced with an Agilent 6560 Ion Mobility Quadrupole Time-of-Flight (Song et al., 2023). Briefly, injection volume was 5 µl with needle wash enabled in the autosampler, which was held at 4°C. A Hypersil Gold column (1.9 µM, 2.1 mm×100 mm, Thermo Fisher Scientific) at 40°C was used for chromatographic separation. The liquid chromatography gradient was 30 min with a flow rate of 0.25 ml/min. Mobile phase A was 10 mM ammonium acetate and 0.1 formic acid additives in H₂O. Mobile phase B was 10 mM ammonium acetate and 0.1 formic acid additives in CH₃CN:IPA:H₂O (60:36:4, v:v:v). Electrospray ionization (Dual AJS, Agilent) was operated in positive polarity with sheath gas flow at 11.8 l/min, sheath gas temperature of 300°C, capillary at 3500 V, nozzle at 2000 V, octupole radio frequency at 750 peak-to-peak voltage (V_pp), drying gas temperature of 280°C, drying gas flow of 5 l/min and nebulizer at 10 psi. Agilent MassHunter Acquisition software (B.09) was used to operate the instrumentation and to acquire liquid chromatography–ion mobility–tandem mass spectrometry (LC-IM-MS/MS) data in the 75-1700 m/z range. A QC sample was injected seven times before the first sample to condition the liquid chromatography column and then injected after every five samples to assess instrument reliability through principal component analysis. Samples were injected randomly with 10% of samples reinjected for quality control purposes. Calibration solution (Agilent Tuning Mixture) was measured at the beginning and end of analysis for mass error quality assurance. Solvent blanks were used to monitor for spectral contaminants and to generate a tandem mass spectrometry (MS/MS) exclusion list. Fragmentation spectra was acquired for the pooled QC via three iterative, top two, data-dependent MS/MS acquisitions. Progenesis QI software (version 3.0, Nonlinear Dynamics, Newcastle, UK) was used to perform data analysis including retention time alignment, peak picking and deconvolution, abundance normalization and one-factor ANOVA. Lipid annotations were performed with reference to in-house and online databases MS-DIAL (Tsugawa et al., 2015), LipidMatch (Koelmel et al., 2017) and Lipid Annotator (Koelmel et al., 2020) using the classification system previously described (Schrimpe-Rutledge et al., 2016). Both HexCer species described in this study were annotated with level 2 confidence, which are putative annotations supported by precursor mass and MS/MS data.

TUNEL labeling was done according to manufacturer instructions (Roche, 12156792910) to assay neuron apoptosis as previously described (Vita et al., 2021). Briefly, staged larvae were dissected, fixed and permeabilized as above. Preparations were incubated in a thermocycler in 10% PBT and 100 mM sodium citrate (1:100 mix) at 65°C for 30 min, and then washed in PBT 3× for 10 min each. Preparations were incubated in 45 μl labeling solution at 37°C for 30 min. Then, 5 μl of the labeling enzyme was added, and preparations were then incubated at 37°C for 2 h. Preparations were washed 3× for 10 min each in PBT, mounted in Fluoromount, and imaged as above. For TUNEL imaging, z-stacks were collected from brain cerebral lobes, and the central VNC was delineated by Brp labeling. Representative maximum-intensity projection images displayed contain all TUNEL signal from each z-slice. For TUNEL/Brp images, maximum-intensity projection files were analyzed unaware of genotype. For quantification, the whole area was selected, excluding the Brp-positive neuropil. Data were binarized using the threshold tool in ImageJ, and the percentage area of TUNEL labeling was measured. To label only the RP3 motor neurons (Broadie, et al., 1993), the CCAP-Gal4 driver was used to express UAS-eGFP. Images were analyzed unaware of genotype with Zen software for colocalization of TUNEL and eGFP in every z-stack optical section, with colocalization confirmed in orthogonal views.

All statistical analyses were performed using Prism software (GraphPad, version 7.04). All data were subject to D'Agostino-Pearson normality tests. If the data fell within the Gaussian distribution, a ROUT (Q=1%) outlier test was performed. All comparisons between two experimental groups were performed using a parametric unpaired two-tailed Student's t-test or a non-parametric Mann–Whitney test. All comparisons between three or more experimental groups were analyzed with a one-way ANOVA or non-parametric Kruskal–Wallis test. All graphs were made using Prism software to show all individual data points normalized to the control. All data in this study were collected and analyzed continuously. All graphs show the mean±s.e.m., with significance indicated as *P<0.05, **P<0.01, ***P<0.001 and P>0.05 [not significant (ns)].

Special thanks to Carl Thummel (University of Utah Medical Center, Salt Lake City, UT, USA) for the ring gland driver 2-286-Gal4 and Hans Wandall (University of Copenhagen, Denmark) for the anti-MacCer antibody. We are grateful to the BDSC (Indiana University, Bloomington, IN, USA) for essential genetic lines and the DSHB (University of Iowa, Iowa City, IA, USA) for essential antibodies. We are especially grateful to Shannon Leahy for electrophysiology mentoring. This work was supported in part using the resources of the Center for Innovative Technology (CIT) at Vanderbilt University. Profs Katrina L. Leaptrot and Stacy D. Sherrod in the CIT performed the lipidomic sample preparation, untargeted mass spectrometry data acquisition, data analysis and lipid annotations.